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>
39 #include <linux/posix_acl.h>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
49 struct btrfs_iget_args {
51 struct btrfs_root *root;
54 static struct inode_operations btrfs_dir_inode_operations;
55 static struct inode_operations btrfs_symlink_inode_operations;
56 static struct inode_operations btrfs_dir_ro_inode_operations;
57 static struct inode_operations btrfs_special_inode_operations;
58 static struct inode_operations btrfs_file_inode_operations;
59 static struct address_space_operations btrfs_aops;
60 static struct address_space_operations btrfs_symlink_aops;
61 static struct file_operations btrfs_dir_file_operations;
62 static struct extent_io_ops btrfs_extent_io_ops;
64 static struct kmem_cache *btrfs_inode_cachep;
65 struct kmem_cache *btrfs_trans_handle_cachep;
66 struct kmem_cache *btrfs_transaction_cachep;
67 struct kmem_cache *btrfs_bit_radix_cachep;
68 struct kmem_cache *btrfs_path_cachep;
71 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
72 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
73 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
74 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
75 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
76 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
77 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
78 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
81 static void btrfs_truncate(struct inode *inode);
83 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
92 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
93 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
94 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
102 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
104 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
108 static int cow_file_range(struct inode *inode, u64 start, u64 end)
110 struct btrfs_root *root = BTRFS_I(inode)->root;
111 struct btrfs_trans_handle *trans;
115 u64 blocksize = root->sectorsize;
117 struct btrfs_key ins;
118 struct extent_map *em;
119 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
122 trans = btrfs_join_transaction(root, 1);
124 btrfs_set_trans_block_group(trans, inode);
126 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
127 num_bytes = max(blocksize, num_bytes);
128 orig_num_bytes = num_bytes;
130 if (alloc_hint == EXTENT_MAP_INLINE)
133 BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
134 mutex_lock(&BTRFS_I(inode)->extent_mutex);
135 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
136 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
138 while(num_bytes > 0) {
139 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
140 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
141 root->sectorsize, 0, 0,
147 em = alloc_extent_map(GFP_NOFS);
149 em->len = ins.offset;
150 em->block_start = ins.objectid;
151 em->bdev = root->fs_info->fs_devices->latest_bdev;
152 mutex_lock(&BTRFS_I(inode)->extent_mutex);
153 set_bit(EXTENT_FLAG_PINNED, &em->flags);
155 spin_lock(&em_tree->lock);
156 ret = add_extent_mapping(em_tree, em);
157 spin_unlock(&em_tree->lock);
158 if (ret != -EEXIST) {
162 btrfs_drop_extent_cache(inode, start,
163 start + ins.offset - 1);
165 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
167 cur_alloc_size = ins.offset;
168 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
171 if (num_bytes < cur_alloc_size) {
172 printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
176 num_bytes -= cur_alloc_size;
177 alloc_hint = ins.objectid + ins.offset;
178 start += cur_alloc_size;
181 btrfs_end_transaction(trans, root);
185 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
192 struct btrfs_root *root = BTRFS_I(inode)->root;
193 struct btrfs_block_group_cache *block_group;
194 struct btrfs_trans_handle *trans;
195 struct extent_buffer *leaf;
197 struct btrfs_path *path;
198 struct btrfs_file_extent_item *item;
201 struct btrfs_key found_key;
203 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
204 path = btrfs_alloc_path();
206 trans = btrfs_join_transaction(root, 1);
209 ret = btrfs_lookup_file_extent(NULL, root, path,
210 inode->i_ino, start, 0);
217 if (path->slots[0] == 0)
222 leaf = path->nodes[0];
223 item = btrfs_item_ptr(leaf, path->slots[0],
224 struct btrfs_file_extent_item);
226 /* are we inside the extent that was found? */
227 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
228 found_type = btrfs_key_type(&found_key);
229 if (found_key.objectid != inode->i_ino ||
230 found_type != BTRFS_EXTENT_DATA_KEY)
233 found_type = btrfs_file_extent_type(leaf, item);
234 extent_start = found_key.offset;
235 if (found_type == BTRFS_FILE_EXTENT_REG) {
236 u64 extent_num_bytes;
238 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
239 extent_end = extent_start + extent_num_bytes;
242 if (loops && start != extent_start)
245 if (start < extent_start || start >= extent_end)
248 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
252 if (btrfs_cross_ref_exists(trans, root, &found_key, bytenr))
255 * we may be called by the resizer, make sure we're inside
256 * the limits of the FS
258 block_group = btrfs_lookup_block_group(root->fs_info,
260 if (!block_group || block_group->ro)
263 bytenr += btrfs_file_extent_offset(leaf, item);
264 extent_num_bytes = min(end + 1, extent_end) - start;
265 ret = btrfs_add_ordered_extent(inode, start, bytenr,
266 extent_num_bytes, 1);
272 btrfs_release_path(root, path);
280 btrfs_end_transaction(trans, root);
281 btrfs_free_path(path);
282 return cow_file_range(inode, start, end);
286 btrfs_end_transaction(trans, root);
287 btrfs_free_path(path);
291 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
293 struct btrfs_root *root = BTRFS_I(inode)->root;
296 if (btrfs_test_opt(root, NODATACOW) ||
297 btrfs_test_flag(inode, NODATACOW))
298 ret = run_delalloc_nocow(inode, start, end);
300 ret = cow_file_range(inode, start, end);
305 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
306 unsigned long old, unsigned long bits)
309 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
310 struct btrfs_root *root = BTRFS_I(inode)->root;
311 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
312 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
313 root->fs_info->delalloc_bytes += end - start + 1;
314 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
315 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
316 &root->fs_info->delalloc_inodes);
318 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
323 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
324 unsigned long old, unsigned long bits)
326 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
327 struct btrfs_root *root = BTRFS_I(inode)->root;
330 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
331 if (end - start + 1 > root->fs_info->delalloc_bytes) {
332 printk("warning: delalloc account %Lu %Lu\n",
333 end - start + 1, root->fs_info->delalloc_bytes);
334 root->fs_info->delalloc_bytes = 0;
335 BTRFS_I(inode)->delalloc_bytes = 0;
337 root->fs_info->delalloc_bytes -= end - start + 1;
338 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
340 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
341 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
342 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
344 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
349 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
350 size_t size, struct bio *bio)
352 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
353 struct btrfs_mapping_tree *map_tree;
354 u64 logical = bio->bi_sector << 9;
359 length = bio->bi_size;
360 map_tree = &root->fs_info->mapping_tree;
362 ret = btrfs_map_block(map_tree, READ, logical,
363 &map_length, NULL, 0);
365 if (map_length < length + size) {
371 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
374 struct btrfs_root *root = BTRFS_I(inode)->root;
377 ret = btrfs_csum_one_bio(root, inode, bio);
380 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
383 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
386 struct btrfs_root *root = BTRFS_I(inode)->root;
389 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
392 if (btrfs_test_opt(root, NODATASUM) ||
393 btrfs_test_flag(inode, NODATASUM)) {
397 if (!(rw & (1 << BIO_RW))) {
398 btrfs_lookup_bio_sums(root, inode, bio);
401 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
402 inode, rw, bio, mirror_num,
403 __btrfs_submit_bio_hook);
405 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
408 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
409 struct inode *inode, u64 file_offset,
410 struct list_head *list)
412 struct list_head *cur;
413 struct btrfs_ordered_sum *sum;
415 btrfs_set_trans_block_group(trans, inode);
416 list_for_each(cur, list) {
417 sum = list_entry(cur, struct btrfs_ordered_sum, list);
418 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
424 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
426 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
430 struct btrfs_writepage_fixup {
432 struct btrfs_work work;
435 /* see btrfs_writepage_start_hook for details on why this is required */
436 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
438 struct btrfs_writepage_fixup *fixup;
439 struct btrfs_ordered_extent *ordered;
445 fixup = container_of(work, struct btrfs_writepage_fixup, work);
449 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
450 ClearPageChecked(page);
454 inode = page->mapping->host;
455 page_start = page_offset(page);
456 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
458 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
460 /* already ordered? We're done */
461 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
462 EXTENT_ORDERED, 0)) {
466 ordered = btrfs_lookup_ordered_extent(inode, page_start);
468 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
471 btrfs_start_ordered_extent(inode, ordered, 1);
475 btrfs_set_extent_delalloc(inode, page_start, page_end);
476 ClearPageChecked(page);
478 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
481 page_cache_release(page);
485 * There are a few paths in the higher layers of the kernel that directly
486 * set the page dirty bit without asking the filesystem if it is a
487 * good idea. This causes problems because we want to make sure COW
488 * properly happens and the data=ordered rules are followed.
490 * In our case any range that doesn't have the EXTENT_ORDERED bit set
491 * hasn't been properly setup for IO. We kick off an async process
492 * to fix it up. The async helper will wait for ordered extents, set
493 * the delalloc bit and make it safe to write the page.
495 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
497 struct inode *inode = page->mapping->host;
498 struct btrfs_writepage_fixup *fixup;
499 struct btrfs_root *root = BTRFS_I(inode)->root;
502 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
507 if (PageChecked(page))
510 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
514 SetPageChecked(page);
515 page_cache_get(page);
516 fixup->work.func = btrfs_writepage_fixup_worker;
518 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
522 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
524 struct btrfs_root *root = BTRFS_I(inode)->root;
525 struct btrfs_trans_handle *trans;
526 struct btrfs_ordered_extent *ordered_extent;
527 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
529 struct list_head list;
530 struct btrfs_key ins;
533 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
537 trans = btrfs_join_transaction(root, 1);
539 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
540 BUG_ON(!ordered_extent);
541 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
544 lock_extent(io_tree, ordered_extent->file_offset,
545 ordered_extent->file_offset + ordered_extent->len - 1,
548 INIT_LIST_HEAD(&list);
550 ins.objectid = ordered_extent->start;
551 ins.offset = ordered_extent->len;
552 ins.type = BTRFS_EXTENT_ITEM_KEY;
554 ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
555 trans->transid, inode->i_ino,
556 ordered_extent->file_offset, &ins);
559 mutex_lock(&BTRFS_I(inode)->extent_mutex);
561 ret = btrfs_drop_extents(trans, root, inode,
562 ordered_extent->file_offset,
563 ordered_extent->file_offset +
565 ordered_extent->file_offset, &alloc_hint);
567 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
568 ordered_extent->file_offset,
569 ordered_extent->start,
571 ordered_extent->len, 0);
574 btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
575 ordered_extent->file_offset +
576 ordered_extent->len - 1);
577 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
579 inode->i_blocks += ordered_extent->len >> 9;
580 unlock_extent(io_tree, ordered_extent->file_offset,
581 ordered_extent->file_offset + ordered_extent->len - 1,
584 add_pending_csums(trans, inode, ordered_extent->file_offset,
585 &ordered_extent->list);
587 btrfs_ordered_update_i_size(inode, ordered_extent);
588 btrfs_remove_ordered_extent(inode, ordered_extent);
591 btrfs_put_ordered_extent(ordered_extent);
592 /* once for the tree */
593 btrfs_put_ordered_extent(ordered_extent);
595 btrfs_update_inode(trans, root, inode);
596 btrfs_end_transaction(trans, root);
600 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
601 struct extent_state *state, int uptodate)
603 return btrfs_finish_ordered_io(page->mapping->host, start, end);
606 struct io_failure_record {
614 int btrfs_io_failed_hook(struct bio *failed_bio,
615 struct page *page, u64 start, u64 end,
616 struct extent_state *state)
618 struct io_failure_record *failrec = NULL;
620 struct extent_map *em;
621 struct inode *inode = page->mapping->host;
622 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
623 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
630 ret = get_state_private(failure_tree, start, &private);
632 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
635 failrec->start = start;
636 failrec->len = end - start + 1;
637 failrec->last_mirror = 0;
639 spin_lock(&em_tree->lock);
640 em = lookup_extent_mapping(em_tree, start, failrec->len);
641 if (em->start > start || em->start + em->len < start) {
645 spin_unlock(&em_tree->lock);
647 if (!em || IS_ERR(em)) {
651 logical = start - em->start;
652 logical = em->block_start + logical;
653 failrec->logical = logical;
655 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
656 EXTENT_DIRTY, GFP_NOFS);
657 set_state_private(failure_tree, start,
658 (u64)(unsigned long)failrec);
660 failrec = (struct io_failure_record *)(unsigned long)private;
662 num_copies = btrfs_num_copies(
663 &BTRFS_I(inode)->root->fs_info->mapping_tree,
664 failrec->logical, failrec->len);
665 failrec->last_mirror++;
667 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
668 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
671 if (state && state->start != failrec->start)
673 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
675 if (!state || failrec->last_mirror > num_copies) {
676 set_state_private(failure_tree, failrec->start, 0);
677 clear_extent_bits(failure_tree, failrec->start,
678 failrec->start + failrec->len - 1,
679 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
683 bio = bio_alloc(GFP_NOFS, 1);
684 bio->bi_private = state;
685 bio->bi_end_io = failed_bio->bi_end_io;
686 bio->bi_sector = failrec->logical >> 9;
687 bio->bi_bdev = failed_bio->bi_bdev;
689 bio_add_page(bio, page, failrec->len, start - page_offset(page));
690 if (failed_bio->bi_rw & (1 << BIO_RW))
695 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
696 failrec->last_mirror);
700 int btrfs_clean_io_failures(struct inode *inode, u64 start)
704 struct io_failure_record *failure;
708 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
709 (u64)-1, 1, EXTENT_DIRTY)) {
710 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
711 start, &private_failure);
713 failure = (struct io_failure_record *)(unsigned long)
715 set_state_private(&BTRFS_I(inode)->io_failure_tree,
717 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
719 failure->start + failure->len - 1,
720 EXTENT_DIRTY | EXTENT_LOCKED,
728 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
729 struct extent_state *state)
731 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
732 struct inode *inode = page->mapping->host;
733 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
735 u64 private = ~(u32)0;
737 struct btrfs_root *root = BTRFS_I(inode)->root;
741 if (btrfs_test_opt(root, NODATASUM) ||
742 btrfs_test_flag(inode, NODATASUM))
744 if (state && state->start == start) {
745 private = state->private;
748 ret = get_state_private(io_tree, start, &private);
750 local_irq_save(flags);
751 kaddr = kmap_atomic(page, KM_IRQ0);
755 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
756 btrfs_csum_final(csum, (char *)&csum);
757 if (csum != private) {
760 kunmap_atomic(kaddr, KM_IRQ0);
761 local_irq_restore(flags);
763 /* if the io failure tree for this inode is non-empty,
764 * check to see if we've recovered from a failed IO
766 btrfs_clean_io_failures(inode, start);
770 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
771 page->mapping->host->i_ino, (unsigned long long)start, csum,
773 memset(kaddr + offset, 1, end - start + 1);
774 flush_dcache_page(page);
775 kunmap_atomic(kaddr, KM_IRQ0);
776 local_irq_restore(flags);
783 * This creates an orphan entry for the given inode in case something goes
784 * wrong in the middle of an unlink/truncate.
786 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
788 struct btrfs_root *root = BTRFS_I(inode)->root;
791 spin_lock(&root->list_lock);
793 /* already on the orphan list, we're good */
794 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
795 spin_unlock(&root->list_lock);
799 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
801 spin_unlock(&root->list_lock);
804 * insert an orphan item to track this unlinked/truncated file
806 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
812 * We have done the truncate/delete so we can go ahead and remove the orphan
813 * item for this particular inode.
815 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
817 struct btrfs_root *root = BTRFS_I(inode)->root;
820 spin_lock(&root->list_lock);
822 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
823 spin_unlock(&root->list_lock);
827 list_del_init(&BTRFS_I(inode)->i_orphan);
829 spin_unlock(&root->list_lock);
833 spin_unlock(&root->list_lock);
835 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
841 * this cleans up any orphans that may be left on the list from the last use
844 void btrfs_orphan_cleanup(struct btrfs_root *root)
846 struct btrfs_path *path;
847 struct extent_buffer *leaf;
848 struct btrfs_item *item;
849 struct btrfs_key key, found_key;
850 struct btrfs_trans_handle *trans;
852 int ret = 0, nr_unlink = 0, nr_truncate = 0;
854 /* don't do orphan cleanup if the fs is readonly. */
855 if (root->inode->i_sb->s_flags & MS_RDONLY)
858 path = btrfs_alloc_path();
863 key.objectid = BTRFS_ORPHAN_OBJECTID;
864 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
865 key.offset = (u64)-1;
867 trans = btrfs_start_transaction(root, 1);
868 btrfs_set_trans_block_group(trans, root->inode);
871 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
873 printk(KERN_ERR "Error searching slot for orphan: %d"
879 * if ret == 0 means we found what we were searching for, which
880 * is weird, but possible, so only screw with path if we didnt
881 * find the key and see if we have stuff that matches
884 if (path->slots[0] == 0)
889 /* pull out the item */
890 leaf = path->nodes[0];
891 item = btrfs_item_nr(leaf, path->slots[0]);
892 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
894 /* make sure the item matches what we want */
895 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
897 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
900 /* release the path since we're done with it */
901 btrfs_release_path(root, path);
904 * this is where we are basically btrfs_lookup, without the
905 * crossing root thing. we store the inode number in the
906 * offset of the orphan item.
908 inode = btrfs_iget_locked(root->inode->i_sb,
909 found_key.offset, root);
913 if (inode->i_state & I_NEW) {
914 BTRFS_I(inode)->root = root;
916 /* have to set the location manually */
917 BTRFS_I(inode)->location.objectid = inode->i_ino;
918 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
919 BTRFS_I(inode)->location.offset = 0;
921 btrfs_read_locked_inode(inode);
922 unlock_new_inode(inode);
926 * add this inode to the orphan list so btrfs_orphan_del does
927 * the proper thing when we hit it
929 spin_lock(&root->list_lock);
930 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
931 spin_unlock(&root->list_lock);
934 * if this is a bad inode, means we actually succeeded in
935 * removing the inode, but not the orphan record, which means
936 * we need to manually delete the orphan since iput will just
939 if (is_bad_inode(inode)) {
940 btrfs_orphan_del(trans, inode);
945 /* if we have links, this was a truncate, lets do that */
946 if (inode->i_nlink) {
948 btrfs_truncate(inode);
953 /* this will do delete_inode and everything for us */
958 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
960 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
962 btrfs_free_path(path);
963 btrfs_end_transaction(trans, root);
966 void btrfs_read_locked_inode(struct inode *inode)
968 struct btrfs_path *path;
969 struct extent_buffer *leaf;
970 struct btrfs_inode_item *inode_item;
971 struct btrfs_timespec *tspec;
972 struct btrfs_root *root = BTRFS_I(inode)->root;
973 struct btrfs_key location;
974 u64 alloc_group_block;
978 path = btrfs_alloc_path();
980 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
982 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
986 leaf = path->nodes[0];
987 inode_item = btrfs_item_ptr(leaf, path->slots[0],
988 struct btrfs_inode_item);
990 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
991 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
992 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
993 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
994 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
996 tspec = btrfs_inode_atime(inode_item);
997 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
998 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1000 tspec = btrfs_inode_mtime(inode_item);
1001 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1002 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1004 tspec = btrfs_inode_ctime(inode_item);
1005 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1006 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1008 inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1009 inode->i_generation = btrfs_inode_generation(leaf, inode_item);
1011 rdev = btrfs_inode_rdev(leaf, inode_item);
1013 BTRFS_I(inode)->index_cnt = (u64)-1;
1015 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1016 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1018 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1019 if (!BTRFS_I(inode)->block_group) {
1020 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1022 BTRFS_BLOCK_GROUP_METADATA, 0);
1024 btrfs_free_path(path);
1027 switch (inode->i_mode & S_IFMT) {
1029 inode->i_mapping->a_ops = &btrfs_aops;
1030 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1031 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1032 inode->i_fop = &btrfs_file_operations;
1033 inode->i_op = &btrfs_file_inode_operations;
1036 inode->i_fop = &btrfs_dir_file_operations;
1037 if (root == root->fs_info->tree_root)
1038 inode->i_op = &btrfs_dir_ro_inode_operations;
1040 inode->i_op = &btrfs_dir_inode_operations;
1043 inode->i_op = &btrfs_symlink_inode_operations;
1044 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1045 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1048 init_special_inode(inode, inode->i_mode, rdev);
1054 btrfs_free_path(path);
1055 make_bad_inode(inode);
1058 static void fill_inode_item(struct extent_buffer *leaf,
1059 struct btrfs_inode_item *item,
1060 struct inode *inode)
1062 btrfs_set_inode_uid(leaf, item, inode->i_uid);
1063 btrfs_set_inode_gid(leaf, item, inode->i_gid);
1064 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1065 btrfs_set_inode_mode(leaf, item, inode->i_mode);
1066 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1068 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1069 inode->i_atime.tv_sec);
1070 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1071 inode->i_atime.tv_nsec);
1073 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1074 inode->i_mtime.tv_sec);
1075 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1076 inode->i_mtime.tv_nsec);
1078 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1079 inode->i_ctime.tv_sec);
1080 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1081 inode->i_ctime.tv_nsec);
1083 btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1084 btrfs_set_inode_generation(leaf, item, inode->i_generation);
1085 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1086 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1087 btrfs_set_inode_block_group(leaf, item,
1088 BTRFS_I(inode)->block_group->key.objectid);
1091 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1092 struct btrfs_root *root,
1093 struct inode *inode)
1095 struct btrfs_inode_item *inode_item;
1096 struct btrfs_path *path;
1097 struct extent_buffer *leaf;
1100 path = btrfs_alloc_path();
1102 ret = btrfs_lookup_inode(trans, root, path,
1103 &BTRFS_I(inode)->location, 1);
1110 leaf = path->nodes[0];
1111 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1112 struct btrfs_inode_item);
1114 fill_inode_item(leaf, inode_item, inode);
1115 btrfs_mark_buffer_dirty(leaf);
1116 btrfs_set_inode_last_trans(trans, inode);
1119 btrfs_free_path(path);
1124 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
1125 struct btrfs_root *root,
1127 struct dentry *dentry)
1129 struct btrfs_path *path;
1130 const char *name = dentry->d_name.name;
1131 int name_len = dentry->d_name.len;
1133 struct extent_buffer *leaf;
1134 struct btrfs_dir_item *di;
1135 struct btrfs_key key;
1138 path = btrfs_alloc_path();
1144 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1145 name, name_len, -1);
1154 leaf = path->nodes[0];
1155 btrfs_dir_item_key_to_cpu(leaf, di, &key);
1156 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1159 btrfs_release_path(root, path);
1161 ret = btrfs_del_inode_ref(trans, root, name, name_len,
1162 dentry->d_inode->i_ino,
1163 dentry->d_parent->d_inode->i_ino, &index);
1165 printk("failed to delete reference to %.*s, "
1166 "inode %lu parent %lu\n", name_len, name,
1167 dentry->d_inode->i_ino,
1168 dentry->d_parent->d_inode->i_ino);
1172 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1173 index, name, name_len, -1);
1182 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1183 btrfs_release_path(root, path);
1185 dentry->d_inode->i_ctime = dir->i_ctime;
1187 btrfs_free_path(path);
1189 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1190 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1191 btrfs_update_inode(trans, root, dir);
1192 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1193 dentry->d_inode->i_nlink--;
1195 drop_nlink(dentry->d_inode);
1197 ret = btrfs_update_inode(trans, root, dentry->d_inode);
1198 dir->i_sb->s_dirt = 1;
1203 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1205 struct btrfs_root *root;
1206 struct btrfs_trans_handle *trans;
1207 struct inode *inode = dentry->d_inode;
1209 unsigned long nr = 0;
1211 root = BTRFS_I(dir)->root;
1213 ret = btrfs_check_free_space(root, 1, 1);
1217 trans = btrfs_start_transaction(root, 1);
1219 btrfs_set_trans_block_group(trans, dir);
1220 ret = btrfs_unlink_trans(trans, root, dir, dentry);
1222 if (inode->i_nlink == 0)
1223 ret = btrfs_orphan_add(trans, inode);
1225 nr = trans->blocks_used;
1227 btrfs_end_transaction_throttle(trans, root);
1229 btrfs_btree_balance_dirty(root, nr);
1233 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1235 struct inode *inode = dentry->d_inode;
1238 struct btrfs_root *root = BTRFS_I(dir)->root;
1239 struct btrfs_trans_handle *trans;
1240 unsigned long nr = 0;
1242 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1246 ret = btrfs_check_free_space(root, 1, 1);
1250 trans = btrfs_start_transaction(root, 1);
1251 btrfs_set_trans_block_group(trans, dir);
1253 err = btrfs_orphan_add(trans, inode);
1257 /* now the directory is empty */
1258 err = btrfs_unlink_trans(trans, root, dir, dentry);
1260 btrfs_i_size_write(inode, 0);
1264 nr = trans->blocks_used;
1265 ret = btrfs_end_transaction_throttle(trans, root);
1267 btrfs_btree_balance_dirty(root, nr);
1275 * this can truncate away extent items, csum items and directory items.
1276 * It starts at a high offset and removes keys until it can't find
1277 * any higher than i_size.
1279 * csum items that cross the new i_size are truncated to the new size
1282 * min_type is the minimum key type to truncate down to. If set to 0, this
1283 * will kill all the items on this inode, including the INODE_ITEM_KEY.
1285 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1286 struct btrfs_root *root,
1287 struct inode *inode,
1291 struct btrfs_path *path;
1292 struct btrfs_key key;
1293 struct btrfs_key found_key;
1295 struct extent_buffer *leaf;
1296 struct btrfs_file_extent_item *fi;
1297 u64 extent_start = 0;
1298 u64 extent_num_bytes = 0;
1304 int pending_del_nr = 0;
1305 int pending_del_slot = 0;
1306 int extent_type = -1;
1307 u64 mask = root->sectorsize - 1;
1309 btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1310 path = btrfs_alloc_path();
1314 /* FIXME, add redo link to tree so we don't leak on crash */
1315 key.objectid = inode->i_ino;
1316 key.offset = (u64)-1;
1319 btrfs_init_path(path);
1321 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1326 BUG_ON(path->slots[0] == 0);
1332 leaf = path->nodes[0];
1333 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1334 found_type = btrfs_key_type(&found_key);
1336 if (found_key.objectid != inode->i_ino)
1339 if (found_type < min_type)
1342 item_end = found_key.offset;
1343 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1344 fi = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_file_extent_item);
1346 extent_type = btrfs_file_extent_type(leaf, fi);
1347 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1349 btrfs_file_extent_num_bytes(leaf, fi);
1350 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1351 struct btrfs_item *item = btrfs_item_nr(leaf,
1353 item_end += btrfs_file_extent_inline_len(leaf,
1358 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1359 ret = btrfs_csum_truncate(trans, root, path,
1363 if (item_end < inode->i_size) {
1364 if (found_type == BTRFS_DIR_ITEM_KEY) {
1365 found_type = BTRFS_INODE_ITEM_KEY;
1366 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1367 found_type = BTRFS_CSUM_ITEM_KEY;
1368 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1369 found_type = BTRFS_XATTR_ITEM_KEY;
1370 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1371 found_type = BTRFS_INODE_REF_KEY;
1372 } else if (found_type) {
1377 btrfs_set_key_type(&key, found_type);
1380 if (found_key.offset >= inode->i_size)
1386 /* FIXME, shrink the extent if the ref count is only 1 */
1387 if (found_type != BTRFS_EXTENT_DATA_KEY)
1390 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1392 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1394 u64 orig_num_bytes =
1395 btrfs_file_extent_num_bytes(leaf, fi);
1396 extent_num_bytes = inode->i_size -
1397 found_key.offset + root->sectorsize - 1;
1398 extent_num_bytes = extent_num_bytes &
1399 ~((u64)root->sectorsize - 1);
1400 btrfs_set_file_extent_num_bytes(leaf, fi,
1402 num_dec = (orig_num_bytes -
1404 if (extent_start != 0)
1405 dec_i_blocks(inode, num_dec);
1406 btrfs_mark_buffer_dirty(leaf);
1409 btrfs_file_extent_disk_num_bytes(leaf,
1411 /* FIXME blocksize != 4096 */
1412 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1413 if (extent_start != 0) {
1415 dec_i_blocks(inode, num_dec);
1417 root_gen = btrfs_header_generation(leaf);
1418 root_owner = btrfs_header_owner(leaf);
1420 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1422 u32 newsize = inode->i_size - found_key.offset;
1423 dec_i_blocks(inode, item_end + 1 -
1424 found_key.offset - newsize);
1426 btrfs_file_extent_calc_inline_size(newsize);
1427 ret = btrfs_truncate_item(trans, root, path,
1431 dec_i_blocks(inode, item_end + 1 -
1437 if (!pending_del_nr) {
1438 /* no pending yet, add ourselves */
1439 pending_del_slot = path->slots[0];
1441 } else if (pending_del_nr &&
1442 path->slots[0] + 1 == pending_del_slot) {
1443 /* hop on the pending chunk */
1445 pending_del_slot = path->slots[0];
1447 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1453 ret = btrfs_free_extent(trans, root, extent_start,
1456 root_gen, inode->i_ino,
1457 found_key.offset, 0);
1461 if (path->slots[0] == 0) {
1464 btrfs_release_path(root, path);
1469 if (pending_del_nr &&
1470 path->slots[0] + 1 != pending_del_slot) {
1471 struct btrfs_key debug;
1473 btrfs_item_key_to_cpu(path->nodes[0], &debug,
1475 ret = btrfs_del_items(trans, root, path,
1480 btrfs_release_path(root, path);
1486 if (pending_del_nr) {
1487 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1490 btrfs_free_path(path);
1491 inode->i_sb->s_dirt = 1;
1496 * taken from block_truncate_page, but does cow as it zeros out
1497 * any bytes left in the last page in the file.
1499 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1501 struct inode *inode = mapping->host;
1502 struct btrfs_root *root = BTRFS_I(inode)->root;
1503 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1504 struct btrfs_ordered_extent *ordered;
1506 u32 blocksize = root->sectorsize;
1507 pgoff_t index = from >> PAGE_CACHE_SHIFT;
1508 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1514 if ((offset & (blocksize - 1)) == 0)
1519 page = grab_cache_page(mapping, index);
1523 page_start = page_offset(page);
1524 page_end = page_start + PAGE_CACHE_SIZE - 1;
1526 if (!PageUptodate(page)) {
1527 ret = btrfs_readpage(NULL, page);
1529 if (page->mapping != mapping) {
1531 page_cache_release(page);
1534 if (!PageUptodate(page)) {
1539 wait_on_page_writeback(page);
1541 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1542 set_page_extent_mapped(page);
1544 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1546 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1548 page_cache_release(page);
1549 btrfs_start_ordered_extent(inode, ordered, 1);
1550 btrfs_put_ordered_extent(ordered);
1554 btrfs_set_extent_delalloc(inode, page_start, page_end);
1556 if (offset != PAGE_CACHE_SIZE) {
1558 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1559 flush_dcache_page(page);
1562 ClearPageChecked(page);
1563 set_page_dirty(page);
1564 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1568 page_cache_release(page);
1573 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1575 struct inode *inode = dentry->d_inode;
1578 err = inode_change_ok(inode, attr);
1582 if (S_ISREG(inode->i_mode) &&
1583 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1584 struct btrfs_trans_handle *trans;
1585 struct btrfs_root *root = BTRFS_I(inode)->root;
1586 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1588 u64 mask = root->sectorsize - 1;
1589 u64 hole_start = (inode->i_size + mask) & ~mask;
1590 u64 block_end = (attr->ia_size + mask) & ~mask;
1594 if (attr->ia_size <= hole_start)
1597 err = btrfs_check_free_space(root, 1, 0);
1601 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1603 hole_size = block_end - hole_start;
1605 struct btrfs_ordered_extent *ordered;
1606 btrfs_wait_ordered_range(inode, hole_start, hole_size);
1608 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1609 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
1611 unlock_extent(io_tree, hole_start,
1612 block_end - 1, GFP_NOFS);
1613 btrfs_put_ordered_extent(ordered);
1619 trans = btrfs_start_transaction(root, 1);
1620 btrfs_set_trans_block_group(trans, inode);
1621 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1622 err = btrfs_drop_extents(trans, root, inode,
1623 hole_start, block_end, hole_start,
1626 if (alloc_hint != EXTENT_MAP_INLINE) {
1627 err = btrfs_insert_file_extent(trans, root,
1631 btrfs_drop_extent_cache(inode, hole_start,
1633 btrfs_check_file(root, inode);
1635 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1636 btrfs_end_transaction(trans, root);
1637 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1642 err = inode_setattr(inode, attr);
1644 if (!err && ((attr->ia_valid & ATTR_MODE)))
1645 err = btrfs_acl_chmod(inode);
1650 void btrfs_delete_inode(struct inode *inode)
1652 struct btrfs_trans_handle *trans;
1653 struct btrfs_root *root = BTRFS_I(inode)->root;
1657 truncate_inode_pages(&inode->i_data, 0);
1658 if (is_bad_inode(inode)) {
1659 btrfs_orphan_del(NULL, inode);
1662 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1664 btrfs_i_size_write(inode, 0);
1665 trans = btrfs_start_transaction(root, 1);
1667 btrfs_set_trans_block_group(trans, inode);
1668 ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1670 btrfs_orphan_del(NULL, inode);
1671 goto no_delete_lock;
1674 btrfs_orphan_del(trans, inode);
1676 nr = trans->blocks_used;
1679 btrfs_end_transaction(trans, root);
1680 btrfs_btree_balance_dirty(root, nr);
1684 nr = trans->blocks_used;
1685 btrfs_end_transaction(trans, root);
1686 btrfs_btree_balance_dirty(root, nr);
1692 * this returns the key found in the dir entry in the location pointer.
1693 * If no dir entries were found, location->objectid is 0.
1695 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1696 struct btrfs_key *location)
1698 const char *name = dentry->d_name.name;
1699 int namelen = dentry->d_name.len;
1700 struct btrfs_dir_item *di;
1701 struct btrfs_path *path;
1702 struct btrfs_root *root = BTRFS_I(dir)->root;
1705 if (namelen == 1 && strcmp(name, ".") == 0) {
1706 location->objectid = dir->i_ino;
1707 location->type = BTRFS_INODE_ITEM_KEY;
1708 location->offset = 0;
1711 path = btrfs_alloc_path();
1714 if (namelen == 2 && strcmp(name, "..") == 0) {
1715 struct btrfs_key key;
1716 struct extent_buffer *leaf;
1719 key.objectid = dir->i_ino;
1720 key.offset = (u64)-1;
1721 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1722 if (ret < 0 || path->slots[0] == 0)
1724 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1727 leaf = path->nodes[0];
1728 slot = path->slots[0] - 1;
1730 btrfs_item_key_to_cpu(leaf, &key, slot);
1731 if (key.objectid != dir->i_ino ||
1732 key.type != BTRFS_INODE_REF_KEY) {
1735 location->objectid = key.offset;
1736 location->type = BTRFS_INODE_ITEM_KEY;
1737 location->offset = 0;
1741 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1745 if (!di || IS_ERR(di)) {
1748 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1750 btrfs_free_path(path);
1753 location->objectid = 0;
1758 * when we hit a tree root in a directory, the btrfs part of the inode
1759 * needs to be changed to reflect the root directory of the tree root. This
1760 * is kind of like crossing a mount point.
1762 static int fixup_tree_root_location(struct btrfs_root *root,
1763 struct btrfs_key *location,
1764 struct btrfs_root **sub_root,
1765 struct dentry *dentry)
1767 struct btrfs_root_item *ri;
1769 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1771 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1774 *sub_root = btrfs_read_fs_root(root->fs_info, location,
1775 dentry->d_name.name,
1776 dentry->d_name.len);
1777 if (IS_ERR(*sub_root))
1778 return PTR_ERR(*sub_root);
1780 ri = &(*sub_root)->root_item;
1781 location->objectid = btrfs_root_dirid(ri);
1782 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1783 location->offset = 0;
1788 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1790 struct btrfs_iget_args *args = p;
1791 inode->i_ino = args->ino;
1792 BTRFS_I(inode)->root = args->root;
1793 BTRFS_I(inode)->delalloc_bytes = 0;
1794 inode->i_mapping->writeback_index = 0;
1795 BTRFS_I(inode)->disk_i_size = 0;
1796 BTRFS_I(inode)->index_cnt = (u64)-1;
1797 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1798 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1799 inode->i_mapping, GFP_NOFS);
1800 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1801 inode->i_mapping, GFP_NOFS);
1802 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1803 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1804 mutex_init(&BTRFS_I(inode)->csum_mutex);
1805 mutex_init(&BTRFS_I(inode)->extent_mutex);
1809 static int btrfs_find_actor(struct inode *inode, void *opaque)
1811 struct btrfs_iget_args *args = opaque;
1812 return (args->ino == inode->i_ino &&
1813 args->root == BTRFS_I(inode)->root);
1816 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1819 struct btrfs_iget_args args;
1820 args.ino = objectid;
1821 args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1826 return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1829 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1830 struct btrfs_root *root)
1832 struct inode *inode;
1833 struct btrfs_iget_args args;
1834 args.ino = objectid;
1837 inode = iget5_locked(s, objectid, btrfs_find_actor,
1838 btrfs_init_locked_inode,
1843 /* Get an inode object given its location and corresponding root.
1844 * Returns in *is_new if the inode was read from disk
1846 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
1847 struct btrfs_root *root, int *is_new)
1849 struct inode *inode;
1851 inode = btrfs_iget_locked(s, location->objectid, root);
1853 return ERR_PTR(-EACCES);
1855 if (inode->i_state & I_NEW) {
1856 BTRFS_I(inode)->root = root;
1857 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
1858 btrfs_read_locked_inode(inode);
1859 unlock_new_inode(inode);
1870 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1871 struct nameidata *nd)
1873 struct inode * inode;
1874 struct btrfs_inode *bi = BTRFS_I(dir);
1875 struct btrfs_root *root = bi->root;
1876 struct btrfs_root *sub_root = root;
1877 struct btrfs_key location;
1878 int ret, new, do_orphan = 0;
1880 if (dentry->d_name.len > BTRFS_NAME_LEN)
1881 return ERR_PTR(-ENAMETOOLONG);
1883 ret = btrfs_inode_by_name(dir, dentry, &location);
1886 return ERR_PTR(ret);
1889 if (location.objectid) {
1890 ret = fixup_tree_root_location(root, &location, &sub_root,
1893 return ERR_PTR(ret);
1895 return ERR_PTR(-ENOENT);
1896 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
1898 return ERR_CAST(inode);
1900 /* the inode and parent dir are two different roots */
1901 if (new && root != sub_root) {
1903 sub_root->inode = inode;
1908 if (unlikely(do_orphan))
1909 btrfs_orphan_cleanup(sub_root);
1911 return d_splice_alias(inode, dentry);
1914 static unsigned char btrfs_filetype_table[] = {
1915 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1918 static int btrfs_real_readdir(struct file *filp, void *dirent,
1921 struct inode *inode = filp->f_dentry->d_inode;
1922 struct btrfs_root *root = BTRFS_I(inode)->root;
1923 struct btrfs_item *item;
1924 struct btrfs_dir_item *di;
1925 struct btrfs_key key;
1926 struct btrfs_key found_key;
1927 struct btrfs_path *path;
1930 struct extent_buffer *leaf;
1933 unsigned char d_type;
1938 int key_type = BTRFS_DIR_INDEX_KEY;
1943 /* FIXME, use a real flag for deciding about the key type */
1944 if (root->fs_info->tree_root == root)
1945 key_type = BTRFS_DIR_ITEM_KEY;
1947 /* special case for "." */
1948 if (filp->f_pos == 0) {
1949 over = filldir(dirent, ".", 1,
1957 key.objectid = inode->i_ino;
1958 path = btrfs_alloc_path();
1961 /* special case for .., just use the back ref */
1962 if (filp->f_pos == 1) {
1963 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1964 key.offset = (u64)-1;
1965 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1966 if (ret < 0 || path->slots[0] == 0) {
1967 btrfs_release_path(root, path);
1968 goto read_dir_items;
1971 leaf = path->nodes[0];
1972 slot = path->slots[0] - 1;
1973 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1974 btrfs_release_path(root, path);
1975 if (found_key.objectid != key.objectid ||
1976 found_key.type != BTRFS_INODE_REF_KEY)
1977 goto read_dir_items;
1978 over = filldir(dirent, "..", 2,
1979 2, found_key.offset, DT_DIR);
1986 btrfs_set_key_type(&key, key_type);
1987 key.offset = filp->f_pos;
1989 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1994 leaf = path->nodes[0];
1995 nritems = btrfs_header_nritems(leaf);
1996 slot = path->slots[0];
1997 if (advance || slot >= nritems) {
1998 if (slot >= nritems -1) {
1999 ret = btrfs_next_leaf(root, path);
2002 leaf = path->nodes[0];
2003 nritems = btrfs_header_nritems(leaf);
2004 slot = path->slots[0];
2011 item = btrfs_item_nr(leaf, slot);
2012 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2014 if (found_key.objectid != key.objectid)
2016 if (btrfs_key_type(&found_key) != key_type)
2018 if (found_key.offset < filp->f_pos)
2021 filp->f_pos = found_key.offset;
2023 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2025 di_total = btrfs_item_size(leaf, item);
2026 while(di_cur < di_total) {
2027 struct btrfs_key location;
2029 name_len = btrfs_dir_name_len(leaf, di);
2030 if (name_len < 32) {
2031 name_ptr = tmp_name;
2033 name_ptr = kmalloc(name_len, GFP_NOFS);
2036 read_extent_buffer(leaf, name_ptr,
2037 (unsigned long)(di + 1), name_len);
2039 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2040 btrfs_dir_item_key_to_cpu(leaf, di, &location);
2041 over = filldir(dirent, name_ptr, name_len,
2046 if (name_ptr != tmp_name)
2051 di_len = btrfs_dir_name_len(leaf, di) +
2052 btrfs_dir_data_len(leaf, di) +sizeof(*di);
2054 di = (struct btrfs_dir_item *)((char *)di + di_len);
2057 if (key_type == BTRFS_DIR_INDEX_KEY)
2058 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2064 btrfs_free_path(path);
2068 /* Kernels earlier than 2.6.28 still have the NFS deadlock where nfsd
2069 will call the file system's ->lookup() method from within its
2070 filldir callback, which in turn was called from the file system's
2071 ->readdir() method. And will deadlock for many file systems. */
2072 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
2074 struct nfshack_dirent {
2078 unsigned int d_type;
2082 struct nfshack_readdir {
2089 static int btrfs_nfshack_filldir(void *__buf, const char *name, int namlen,
2090 loff_t offset, u64 ino, unsigned int d_type)
2092 struct nfshack_readdir *buf = __buf;
2093 struct nfshack_dirent *de = (void *)(buf->dirent + buf->used);
2094 unsigned int reclen;
2096 reclen = ALIGN(sizeof(struct nfshack_dirent) + namlen, sizeof(u64));
2097 if (buf->used + reclen > PAGE_SIZE)
2100 de->namlen = namlen;
2101 de->offset = offset;
2103 de->d_type = d_type;
2104 memcpy(de->name, name, namlen);
2105 buf->used += reclen;
2110 static int btrfs_nfshack_readdir(struct file *file, void *dirent,
2113 struct nfshack_readdir buf;
2114 struct nfshack_dirent *de;
2119 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
2123 offset = file->f_pos;
2126 unsigned int reclen;
2130 err = btrfs_real_readdir(file, &buf, btrfs_nfshack_filldir);
2139 de = (struct nfshack_dirent *)buf.dirent;
2141 offset = de->offset;
2143 if (filldir(dirent, de->name, de->namlen, de->offset,
2144 de->ino, de->d_type))
2146 offset = file->f_pos;
2148 reclen = ALIGN(sizeof(*de) + de->namlen,
2151 de = (struct nfshack_dirent *)((char *)de + reclen);
2156 free_page((unsigned long)buf.dirent);
2157 file->f_pos = offset;
2163 int btrfs_write_inode(struct inode *inode, int wait)
2165 struct btrfs_root *root = BTRFS_I(inode)->root;
2166 struct btrfs_trans_handle *trans;
2169 if (root->fs_info->closing > 1)
2173 trans = btrfs_join_transaction(root, 1);
2174 btrfs_set_trans_block_group(trans, inode);
2175 ret = btrfs_commit_transaction(trans, root);
2181 * This is somewhat expensive, updating the tree every time the
2182 * inode changes. But, it is most likely to find the inode in cache.
2183 * FIXME, needs more benchmarking...there are no reasons other than performance
2184 * to keep or drop this code.
2186 void btrfs_dirty_inode(struct inode *inode)
2188 struct btrfs_root *root = BTRFS_I(inode)->root;
2189 struct btrfs_trans_handle *trans;
2191 trans = btrfs_join_transaction(root, 1);
2192 btrfs_set_trans_block_group(trans, inode);
2193 btrfs_update_inode(trans, root, inode);
2194 btrfs_end_transaction(trans, root);
2197 static int btrfs_set_inode_index_count(struct inode *inode)
2199 struct btrfs_root *root = BTRFS_I(inode)->root;
2200 struct btrfs_key key, found_key;
2201 struct btrfs_path *path;
2202 struct extent_buffer *leaf;
2205 key.objectid = inode->i_ino;
2206 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2207 key.offset = (u64)-1;
2209 path = btrfs_alloc_path();
2213 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2216 /* FIXME: we should be able to handle this */
2222 * MAGIC NUMBER EXPLANATION:
2223 * since we search a directory based on f_pos we have to start at 2
2224 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2225 * else has to start at 2
2227 if (path->slots[0] == 0) {
2228 BTRFS_I(inode)->index_cnt = 2;
2234 leaf = path->nodes[0];
2235 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2237 if (found_key.objectid != inode->i_ino ||
2238 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2239 BTRFS_I(inode)->index_cnt = 2;
2243 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2245 btrfs_free_path(path);
2249 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2254 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2255 ret = btrfs_set_inode_index_count(dir);
2260 *index = BTRFS_I(dir)->index_cnt;
2261 BTRFS_I(dir)->index_cnt++;
2266 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2267 struct btrfs_root *root,
2269 const char *name, int name_len,
2272 struct btrfs_block_group_cache *group,
2273 int mode, u64 *index)
2275 struct inode *inode;
2276 struct btrfs_inode_item *inode_item;
2277 struct btrfs_block_group_cache *new_inode_group;
2278 struct btrfs_key *location;
2279 struct btrfs_path *path;
2280 struct btrfs_inode_ref *ref;
2281 struct btrfs_key key[2];
2287 path = btrfs_alloc_path();
2290 inode = new_inode(root->fs_info->sb);
2292 return ERR_PTR(-ENOMEM);
2295 ret = btrfs_set_inode_index(dir, inode, index);
2297 return ERR_PTR(ret);
2300 * index_cnt is ignored for everything but a dir,
2301 * btrfs_get_inode_index_count has an explanation for the magic
2304 BTRFS_I(inode)->index_cnt = 2;
2306 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2307 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2308 inode->i_mapping, GFP_NOFS);
2309 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2310 inode->i_mapping, GFP_NOFS);
2311 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2312 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2313 mutex_init(&BTRFS_I(inode)->csum_mutex);
2314 mutex_init(&BTRFS_I(inode)->extent_mutex);
2315 BTRFS_I(inode)->delalloc_bytes = 0;
2316 inode->i_mapping->writeback_index = 0;
2317 BTRFS_I(inode)->disk_i_size = 0;
2318 BTRFS_I(inode)->root = root;
2324 new_inode_group = btrfs_find_block_group(root, group, 0,
2325 BTRFS_BLOCK_GROUP_METADATA, owner);
2326 if (!new_inode_group) {
2327 printk("find_block group failed\n");
2328 new_inode_group = group;
2330 BTRFS_I(inode)->block_group = new_inode_group;
2331 BTRFS_I(inode)->flags = 0;
2333 key[0].objectid = objectid;
2334 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2337 key[1].objectid = objectid;
2338 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2339 key[1].offset = ref_objectid;
2341 sizes[0] = sizeof(struct btrfs_inode_item);
2342 sizes[1] = name_len + sizeof(*ref);
2344 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2348 if (objectid > root->highest_inode)
2349 root->highest_inode = objectid;
2351 inode->i_uid = current->fsuid;
2352 inode->i_gid = current->fsgid;
2353 inode->i_mode = mode;
2354 inode->i_ino = objectid;
2355 inode->i_blocks = 0;
2356 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2357 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2358 struct btrfs_inode_item);
2359 fill_inode_item(path->nodes[0], inode_item, inode);
2361 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2362 struct btrfs_inode_ref);
2363 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2364 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2365 ptr = (unsigned long)(ref + 1);
2366 write_extent_buffer(path->nodes[0], name, ptr, name_len);
2368 btrfs_mark_buffer_dirty(path->nodes[0]);
2369 btrfs_free_path(path);
2371 location = &BTRFS_I(inode)->location;
2372 location->objectid = objectid;
2373 location->offset = 0;
2374 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2376 insert_inode_hash(inode);
2380 BTRFS_I(dir)->index_cnt--;
2381 btrfs_free_path(path);
2382 return ERR_PTR(ret);
2385 static inline u8 btrfs_inode_type(struct inode *inode)
2387 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2390 static int btrfs_add_link(struct btrfs_trans_handle *trans,
2391 struct dentry *dentry, struct inode *inode,
2392 int add_backref, u64 index)
2395 struct btrfs_key key;
2396 struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
2397 struct inode *parent_inode = dentry->d_parent->d_inode;
2399 key.objectid = inode->i_ino;
2400 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2403 ret = btrfs_insert_dir_item(trans, root,
2404 dentry->d_name.name, dentry->d_name.len,
2405 dentry->d_parent->d_inode->i_ino,
2406 &key, btrfs_inode_type(inode),
2410 ret = btrfs_insert_inode_ref(trans, root,
2411 dentry->d_name.name,
2414 parent_inode->i_ino,
2417 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2418 dentry->d_name.len * 2);
2419 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2420 ret = btrfs_update_inode(trans, root,
2421 dentry->d_parent->d_inode);
2426 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2427 struct dentry *dentry, struct inode *inode,
2428 int backref, u64 index)
2430 int err = btrfs_add_link(trans, dentry, inode, backref, index);
2432 d_instantiate(dentry, inode);
2440 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2441 int mode, dev_t rdev)
2443 struct btrfs_trans_handle *trans;
2444 struct btrfs_root *root = BTRFS_I(dir)->root;
2445 struct inode *inode = NULL;
2449 unsigned long nr = 0;
2452 if (!new_valid_dev(rdev))
2455 err = btrfs_check_free_space(root, 1, 0);
2459 trans = btrfs_start_transaction(root, 1);
2460 btrfs_set_trans_block_group(trans, dir);
2462 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2468 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2470 dentry->d_parent->d_inode->i_ino, objectid,
2471 BTRFS_I(dir)->block_group, mode, &index);
2472 err = PTR_ERR(inode);
2476 err = btrfs_init_acl(inode, dir);
2482 btrfs_set_trans_block_group(trans, inode);
2483 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2487 inode->i_op = &btrfs_special_inode_operations;
2488 init_special_inode(inode, inode->i_mode, rdev);
2489 btrfs_update_inode(trans, root, inode);
2491 dir->i_sb->s_dirt = 1;
2492 btrfs_update_inode_block_group(trans, inode);
2493 btrfs_update_inode_block_group(trans, dir);
2495 nr = trans->blocks_used;
2496 btrfs_end_transaction_throttle(trans, root);
2499 inode_dec_link_count(inode);
2502 btrfs_btree_balance_dirty(root, nr);
2506 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2507 int mode, struct nameidata *nd)
2509 struct btrfs_trans_handle *trans;
2510 struct btrfs_root *root = BTRFS_I(dir)->root;
2511 struct inode *inode = NULL;
2514 unsigned long nr = 0;
2518 err = btrfs_check_free_space(root, 1, 0);
2521 trans = btrfs_start_transaction(root, 1);
2522 btrfs_set_trans_block_group(trans, dir);
2524 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2530 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2532 dentry->d_parent->d_inode->i_ino,
2533 objectid, BTRFS_I(dir)->block_group, mode,
2535 err = PTR_ERR(inode);
2539 err = btrfs_init_acl(inode, dir);
2545 btrfs_set_trans_block_group(trans, inode);
2546 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2550 inode->i_mapping->a_ops = &btrfs_aops;
2551 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2552 inode->i_fop = &btrfs_file_operations;
2553 inode->i_op = &btrfs_file_inode_operations;
2554 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2555 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2556 inode->i_mapping, GFP_NOFS);
2557 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2558 inode->i_mapping, GFP_NOFS);
2559 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2560 mutex_init(&BTRFS_I(inode)->csum_mutex);
2561 mutex_init(&BTRFS_I(inode)->extent_mutex);
2562 BTRFS_I(inode)->delalloc_bytes = 0;
2563 BTRFS_I(inode)->disk_i_size = 0;
2564 inode->i_mapping->writeback_index = 0;
2565 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2566 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2568 dir->i_sb->s_dirt = 1;
2569 btrfs_update_inode_block_group(trans, inode);
2570 btrfs_update_inode_block_group(trans, dir);
2572 nr = trans->blocks_used;
2573 btrfs_end_transaction_throttle(trans, root);
2576 inode_dec_link_count(inode);
2579 btrfs_btree_balance_dirty(root, nr);
2583 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2584 struct dentry *dentry)
2586 struct btrfs_trans_handle *trans;
2587 struct btrfs_root *root = BTRFS_I(dir)->root;
2588 struct inode *inode = old_dentry->d_inode;
2590 unsigned long nr = 0;
2594 if (inode->i_nlink == 0)
2597 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2602 err = btrfs_check_free_space(root, 1, 0);
2605 err = btrfs_set_inode_index(dir, inode, &index);
2609 trans = btrfs_start_transaction(root, 1);
2611 btrfs_set_trans_block_group(trans, dir);
2612 atomic_inc(&inode->i_count);
2614 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2619 dir->i_sb->s_dirt = 1;
2620 btrfs_update_inode_block_group(trans, dir);
2621 err = btrfs_update_inode(trans, root, inode);
2626 nr = trans->blocks_used;
2627 btrfs_end_transaction_throttle(trans, root);
2630 inode_dec_link_count(inode);
2633 btrfs_btree_balance_dirty(root, nr);
2637 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2639 struct inode *inode = NULL;
2640 struct btrfs_trans_handle *trans;
2641 struct btrfs_root *root = BTRFS_I(dir)->root;
2643 int drop_on_err = 0;
2646 unsigned long nr = 1;
2648 err = btrfs_check_free_space(root, 1, 0);
2652 trans = btrfs_start_transaction(root, 1);
2653 btrfs_set_trans_block_group(trans, dir);
2655 if (IS_ERR(trans)) {
2656 err = PTR_ERR(trans);
2660 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2666 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2668 dentry->d_parent->d_inode->i_ino, objectid,
2669 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2671 if (IS_ERR(inode)) {
2672 err = PTR_ERR(inode);
2678 err = btrfs_init_acl(inode, dir);
2682 inode->i_op = &btrfs_dir_inode_operations;
2683 inode->i_fop = &btrfs_dir_file_operations;
2684 btrfs_set_trans_block_group(trans, inode);
2686 btrfs_i_size_write(inode, 0);
2687 err = btrfs_update_inode(trans, root, inode);
2691 err = btrfs_add_link(trans, dentry, inode, 0, index);
2695 d_instantiate(dentry, inode);
2697 dir->i_sb->s_dirt = 1;
2698 btrfs_update_inode_block_group(trans, inode);
2699 btrfs_update_inode_block_group(trans, dir);
2702 nr = trans->blocks_used;
2703 btrfs_end_transaction_throttle(trans, root);
2708 btrfs_btree_balance_dirty(root, nr);
2712 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2713 struct extent_map *existing,
2714 struct extent_map *em,
2715 u64 map_start, u64 map_len)
2719 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2720 start_diff = map_start - em->start;
2721 em->start = map_start;
2723 if (em->block_start < EXTENT_MAP_LAST_BYTE)
2724 em->block_start += start_diff;
2725 return add_extent_mapping(em_tree, em);
2728 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2729 size_t pg_offset, u64 start, u64 len,
2735 u64 extent_start = 0;
2737 u64 objectid = inode->i_ino;
2739 struct btrfs_path *path = NULL;
2740 struct btrfs_root *root = BTRFS_I(inode)->root;
2741 struct btrfs_file_extent_item *item;
2742 struct extent_buffer *leaf;
2743 struct btrfs_key found_key;
2744 struct extent_map *em = NULL;
2745 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2746 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2747 struct btrfs_trans_handle *trans = NULL;
2750 spin_lock(&em_tree->lock);
2751 em = lookup_extent_mapping(em_tree, start, len);
2753 em->bdev = root->fs_info->fs_devices->latest_bdev;
2754 spin_unlock(&em_tree->lock);
2757 if (em->start > start || em->start + em->len <= start)
2758 free_extent_map(em);
2759 else if (em->block_start == EXTENT_MAP_INLINE && page)
2760 free_extent_map(em);
2764 em = alloc_extent_map(GFP_NOFS);
2769 em->bdev = root->fs_info->fs_devices->latest_bdev;
2770 em->start = EXTENT_MAP_HOLE;
2774 path = btrfs_alloc_path();
2778 ret = btrfs_lookup_file_extent(trans, root, path,
2779 objectid, start, trans != NULL);
2786 if (path->slots[0] == 0)
2791 leaf = path->nodes[0];
2792 item = btrfs_item_ptr(leaf, path->slots[0],
2793 struct btrfs_file_extent_item);
2794 /* are we inside the extent that was found? */
2795 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2796 found_type = btrfs_key_type(&found_key);
2797 if (found_key.objectid != objectid ||
2798 found_type != BTRFS_EXTENT_DATA_KEY) {
2802 found_type = btrfs_file_extent_type(leaf, item);
2803 extent_start = found_key.offset;
2804 if (found_type == BTRFS_FILE_EXTENT_REG) {
2805 extent_end = extent_start +
2806 btrfs_file_extent_num_bytes(leaf, item);
2808 if (start < extent_start || start >= extent_end) {
2810 if (start < extent_start) {
2811 if (start + len <= extent_start)
2813 em->len = extent_end - extent_start;
2819 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2821 em->start = extent_start;
2822 em->len = extent_end - extent_start;
2823 em->block_start = EXTENT_MAP_HOLE;
2826 bytenr += btrfs_file_extent_offset(leaf, item);
2827 em->block_start = bytenr;
2828 em->start = extent_start;
2829 em->len = extent_end - extent_start;
2831 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2836 size_t extent_offset;
2839 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2841 extent_end = (extent_start + size + root->sectorsize - 1) &
2842 ~((u64)root->sectorsize - 1);
2843 if (start < extent_start || start >= extent_end) {
2845 if (start < extent_start) {
2846 if (start + len <= extent_start)
2848 em->len = extent_end - extent_start;
2854 em->block_start = EXTENT_MAP_INLINE;
2857 em->start = extent_start;
2862 page_start = page_offset(page) + pg_offset;
2863 extent_offset = page_start - extent_start;
2864 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2865 size - extent_offset);
2866 em->start = extent_start + extent_offset;
2867 em->len = (copy_size + root->sectorsize - 1) &
2868 ~((u64)root->sectorsize - 1);
2870 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2871 if (create == 0 && !PageUptodate(page)) {
2872 read_extent_buffer(leaf, map + pg_offset, ptr,
2874 flush_dcache_page(page);
2875 } else if (create && PageUptodate(page)) {
2878 free_extent_map(em);
2880 btrfs_release_path(root, path);
2881 trans = btrfs_join_transaction(root, 1);
2884 write_extent_buffer(leaf, map + pg_offset, ptr,
2886 btrfs_mark_buffer_dirty(leaf);
2889 set_extent_uptodate(io_tree, em->start,
2890 extent_map_end(em) - 1, GFP_NOFS);
2893 printk("unkknown found_type %d\n", found_type);
2900 em->block_start = EXTENT_MAP_HOLE;
2902 btrfs_release_path(root, path);
2903 if (em->start > start || extent_map_end(em) <= start) {
2904 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2910 spin_lock(&em_tree->lock);
2911 ret = add_extent_mapping(em_tree, em);
2912 /* it is possible that someone inserted the extent into the tree
2913 * while we had the lock dropped. It is also possible that
2914 * an overlapping map exists in the tree
2916 if (ret == -EEXIST) {
2917 struct extent_map *existing;
2921 existing = lookup_extent_mapping(em_tree, start, len);
2922 if (existing && (existing->start > start ||
2923 existing->start + existing->len <= start)) {
2924 free_extent_map(existing);
2928 existing = lookup_extent_mapping(em_tree, em->start,
2931 err = merge_extent_mapping(em_tree, existing,
2934 free_extent_map(existing);
2936 free_extent_map(em);
2941 printk("failing to insert %Lu %Lu\n",
2943 free_extent_map(em);
2947 free_extent_map(em);
2952 spin_unlock(&em_tree->lock);
2955 btrfs_free_path(path);
2957 ret = btrfs_end_transaction(trans, root);
2963 free_extent_map(em);
2965 return ERR_PTR(err);
2970 #if 0 /* waiting for O_DIRECT reads */
2971 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2972 struct buffer_head *bh_result, int create)
2974 struct extent_map *em;
2975 u64 start = (u64)iblock << inode->i_blkbits;
2976 struct btrfs_multi_bio *multi = NULL;
2977 struct btrfs_root *root = BTRFS_I(inode)->root;
2983 em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2985 if (!em || IS_ERR(em))
2988 if (em->start > start || em->start + em->len <= start) {
2992 if (em->block_start == EXTENT_MAP_INLINE) {
2997 len = em->start + em->len - start;
2998 len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
3000 if (em->block_start == EXTENT_MAP_HOLE ||
3001 em->block_start == EXTENT_MAP_DELALLOC) {
3002 bh_result->b_size = len;
3006 logical = start - em->start;
3007 logical = em->block_start + logical;
3010 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
3011 logical, &map_length, &multi, 0);
3013 bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
3014 bh_result->b_size = min(map_length, len);
3016 bh_result->b_bdev = multi->stripes[0].dev->bdev;
3017 set_buffer_mapped(bh_result);
3020 free_extent_map(em);
3025 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
3026 const struct iovec *iov, loff_t offset,
3027 unsigned long nr_segs)
3031 struct file *file = iocb->ki_filp;
3032 struct inode *inode = file->f_mapping->host;
3037 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3038 offset, nr_segs, btrfs_get_block, NULL);
3042 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
3044 return extent_bmap(mapping, iblock, btrfs_get_extent);
3047 int btrfs_readpage(struct file *file, struct page *page)
3049 struct extent_io_tree *tree;
3050 tree = &BTRFS_I(page->mapping->host)->io_tree;
3051 return extent_read_full_page(tree, page, btrfs_get_extent);
3054 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3056 struct extent_io_tree *tree;
3059 if (current->flags & PF_MEMALLOC) {
3060 redirty_page_for_writepage(wbc, page);
3064 tree = &BTRFS_I(page->mapping->host)->io_tree;
3065 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3068 int btrfs_writepages(struct address_space *mapping,
3069 struct writeback_control *wbc)
3071 struct extent_io_tree *tree;
3072 tree = &BTRFS_I(mapping->host)->io_tree;
3073 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3077 btrfs_readpages(struct file *file, struct address_space *mapping,
3078 struct list_head *pages, unsigned nr_pages)
3080 struct extent_io_tree *tree;
3081 tree = &BTRFS_I(mapping->host)->io_tree;
3082 return extent_readpages(tree, mapping, pages, nr_pages,
3085 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3087 struct extent_io_tree *tree;
3088 struct extent_map_tree *map;
3091 tree = &BTRFS_I(page->mapping->host)->io_tree;
3092 map = &BTRFS_I(page->mapping->host)->extent_tree;
3093 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3095 ClearPagePrivate(page);
3096 set_page_private(page, 0);
3097 page_cache_release(page);
3102 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3104 return __btrfs_releasepage(page, gfp_flags);
3107 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3109 struct extent_io_tree *tree;
3110 struct btrfs_ordered_extent *ordered;
3111 u64 page_start = page_offset(page);
3112 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3114 wait_on_page_writeback(page);
3115 tree = &BTRFS_I(page->mapping->host)->io_tree;
3117 btrfs_releasepage(page, GFP_NOFS);
3121 lock_extent(tree, page_start, page_end, GFP_NOFS);
3122 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3126 * IO on this page will never be started, so we need
3127 * to account for any ordered extents now
3129 clear_extent_bit(tree, page_start, page_end,
3130 EXTENT_DIRTY | EXTENT_DELALLOC |
3131 EXTENT_LOCKED, 1, 0, GFP_NOFS);
3132 btrfs_finish_ordered_io(page->mapping->host,
3133 page_start, page_end);
3134 btrfs_put_ordered_extent(ordered);
3135 lock_extent(tree, page_start, page_end, GFP_NOFS);
3137 clear_extent_bit(tree, page_start, page_end,
3138 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3141 __btrfs_releasepage(page, GFP_NOFS);
3143 ClearPageChecked(page);
3144 if (PagePrivate(page)) {
3145 ClearPagePrivate(page);
3146 set_page_private(page, 0);
3147 page_cache_release(page);
3152 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3153 * called from a page fault handler when a page is first dirtied. Hence we must
3154 * be careful to check for EOF conditions here. We set the page up correctly
3155 * for a written page which means we get ENOSPC checking when writing into
3156 * holes and correct delalloc and unwritten extent mapping on filesystems that
3157 * support these features.
3159 * We are not allowed to take the i_mutex here so we have to play games to
3160 * protect against truncate races as the page could now be beyond EOF. Because
3161 * vmtruncate() writes the inode size before removing pages, once we have the
3162 * page lock we can determine safely if the page is beyond EOF. If it is not
3163 * beyond EOF, then the page is guaranteed safe against truncation until we
3166 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3168 struct inode *inode = fdentry(vma->vm_file)->d_inode;
3169 struct btrfs_root *root = BTRFS_I(inode)->root;
3170 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3171 struct btrfs_ordered_extent *ordered;
3173 unsigned long zero_start;
3179 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3186 size = i_size_read(inode);
3187 page_start = page_offset(page);
3188 page_end = page_start + PAGE_CACHE_SIZE - 1;
3190 if ((page->mapping != inode->i_mapping) ||
3191 (page_start >= size)) {
3192 /* page got truncated out from underneath us */
3195 wait_on_page_writeback(page);
3197 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3198 set_page_extent_mapped(page);
3201 * we can't set the delalloc bits if there are pending ordered
3202 * extents. Drop our locks and wait for them to finish
3204 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3206 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3208 btrfs_start_ordered_extent(inode, ordered, 1);
3209 btrfs_put_ordered_extent(ordered);
3213 btrfs_set_extent_delalloc(inode, page_start, page_end);
3216 /* page is wholly or partially inside EOF */
3217 if (page_start + PAGE_CACHE_SIZE > size)
3218 zero_start = size & ~PAGE_CACHE_MASK;
3220 zero_start = PAGE_CACHE_SIZE;
3222 if (zero_start != PAGE_CACHE_SIZE) {
3224 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3225 flush_dcache_page(page);
3228 ClearPageChecked(page);
3229 set_page_dirty(page);
3230 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3238 static void btrfs_truncate(struct inode *inode)
3240 struct btrfs_root *root = BTRFS_I(inode)->root;
3242 struct btrfs_trans_handle *trans;
3244 u64 mask = root->sectorsize - 1;
3246 if (!S_ISREG(inode->i_mode))
3248 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3251 btrfs_truncate_page(inode->i_mapping, inode->i_size);
3252 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3254 trans = btrfs_start_transaction(root, 1);
3255 btrfs_set_trans_block_group(trans, inode);
3256 btrfs_i_size_write(inode, inode->i_size);
3258 ret = btrfs_orphan_add(trans, inode);
3261 /* FIXME, add redo link to tree so we don't leak on crash */
3262 ret = btrfs_truncate_in_trans(trans, root, inode,
3263 BTRFS_EXTENT_DATA_KEY);
3264 btrfs_update_inode(trans, root, inode);
3266 ret = btrfs_orphan_del(trans, inode);
3270 nr = trans->blocks_used;
3271 ret = btrfs_end_transaction_throttle(trans, root);
3273 btrfs_btree_balance_dirty(root, nr);
3277 * Invalidate a single dcache entry at the root of the filesystem.
3278 * Needed after creation of snapshot or subvolume.
3280 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3283 struct dentry *alias, *entry;
3286 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3290 /* change me if btrfs ever gets a d_hash operation */
3291 qstr.hash = full_name_hash(qstr.name, qstr.len);
3292 entry = d_lookup(alias, &qstr);
3295 d_invalidate(entry);
3301 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3302 struct btrfs_trans_handle *trans, u64 new_dirid,
3303 struct btrfs_block_group_cache *block_group)
3305 struct inode *inode;
3308 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3309 new_dirid, block_group, S_IFDIR | 0700, &index);
3311 return PTR_ERR(inode);
3312 inode->i_op = &btrfs_dir_inode_operations;
3313 inode->i_fop = &btrfs_dir_file_operations;
3314 new_root->inode = inode;
3317 btrfs_i_size_write(inode, 0);
3319 return btrfs_update_inode(trans, new_root, inode);
3322 unsigned long btrfs_force_ra(struct address_space *mapping,
3323 struct file_ra_state *ra, struct file *file,
3324 pgoff_t offset, pgoff_t last_index)
3326 pgoff_t req_size = last_index - offset + 1;
3328 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3329 offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3332 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3333 return offset + req_size;
3337 struct inode *btrfs_alloc_inode(struct super_block *sb)
3339 struct btrfs_inode *ei;
3341 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3345 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3346 ei->i_acl = BTRFS_ACL_NOT_CACHED;
3347 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3348 INIT_LIST_HEAD(&ei->i_orphan);
3349 return &ei->vfs_inode;
3352 void btrfs_destroy_inode(struct inode *inode)
3354 struct btrfs_ordered_extent *ordered;
3355 WARN_ON(!list_empty(&inode->i_dentry));
3356 WARN_ON(inode->i_data.nrpages);
3358 if (BTRFS_I(inode)->i_acl &&
3359 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3360 posix_acl_release(BTRFS_I(inode)->i_acl);
3361 if (BTRFS_I(inode)->i_default_acl &&
3362 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3363 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3365 spin_lock(&BTRFS_I(inode)->root->list_lock);
3366 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3367 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3368 " list\n", inode->i_ino);
3371 spin_unlock(&BTRFS_I(inode)->root->list_lock);
3374 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3378 printk("found ordered extent %Lu %Lu\n",
3379 ordered->file_offset, ordered->len);
3380 btrfs_remove_ordered_extent(inode, ordered);
3381 btrfs_put_ordered_extent(ordered);
3382 btrfs_put_ordered_extent(ordered);
3385 btrfs_drop_extent_cache(inode, 0, (u64)-1);
3386 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3389 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3390 static void init_once(void *foo)
3391 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3392 static void init_once(struct kmem_cache * cachep, void *foo)
3394 static void init_once(void * foo, struct kmem_cache * cachep,
3395 unsigned long flags)
3398 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3400 inode_init_once(&ei->vfs_inode);
3403 void btrfs_destroy_cachep(void)
3405 if (btrfs_inode_cachep)
3406 kmem_cache_destroy(btrfs_inode_cachep);
3407 if (btrfs_trans_handle_cachep)
3408 kmem_cache_destroy(btrfs_trans_handle_cachep);
3409 if (btrfs_transaction_cachep)
3410 kmem_cache_destroy(btrfs_transaction_cachep);
3411 if (btrfs_bit_radix_cachep)
3412 kmem_cache_destroy(btrfs_bit_radix_cachep);
3413 if (btrfs_path_cachep)
3414 kmem_cache_destroy(btrfs_path_cachep);
3417 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3418 unsigned long extra_flags,
3419 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3420 void (*ctor)(void *)
3421 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3422 void (*ctor)(struct kmem_cache *, void *)
3424 void (*ctor)(void *, struct kmem_cache *,
3429 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3430 SLAB_MEM_SPREAD | extra_flags), ctor
3431 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3437 int btrfs_init_cachep(void)
3439 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3440 sizeof(struct btrfs_inode),
3442 if (!btrfs_inode_cachep)
3444 btrfs_trans_handle_cachep =
3445 btrfs_cache_create("btrfs_trans_handle_cache",
3446 sizeof(struct btrfs_trans_handle),
3448 if (!btrfs_trans_handle_cachep)
3450 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3451 sizeof(struct btrfs_transaction),
3453 if (!btrfs_transaction_cachep)
3455 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3456 sizeof(struct btrfs_path),
3458 if (!btrfs_path_cachep)
3460 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3461 SLAB_DESTROY_BY_RCU, NULL);
3462 if (!btrfs_bit_radix_cachep)
3466 btrfs_destroy_cachep();
3470 static int btrfs_getattr(struct vfsmount *mnt,
3471 struct dentry *dentry, struct kstat *stat)
3473 struct inode *inode = dentry->d_inode;
3474 generic_fillattr(inode, stat);
3475 stat->blksize = PAGE_CACHE_SIZE;
3476 stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3480 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3481 struct inode * new_dir,struct dentry *new_dentry)
3483 struct btrfs_trans_handle *trans;
3484 struct btrfs_root *root = BTRFS_I(old_dir)->root;
3485 struct inode *new_inode = new_dentry->d_inode;
3486 struct inode *old_inode = old_dentry->d_inode;
3487 struct timespec ctime = CURRENT_TIME;
3491 if (S_ISDIR(old_inode->i_mode) && new_inode &&
3492 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3496 ret = btrfs_check_free_space(root, 1, 0);
3500 trans = btrfs_start_transaction(root, 1);
3502 btrfs_set_trans_block_group(trans, new_dir);
3504 old_dentry->d_inode->i_nlink++;
3505 old_dir->i_ctime = old_dir->i_mtime = ctime;
3506 new_dir->i_ctime = new_dir->i_mtime = ctime;
3507 old_inode->i_ctime = ctime;
3509 ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3514 new_inode->i_ctime = CURRENT_TIME;
3515 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3518 if (new_inode->i_nlink == 0) {
3519 ret = btrfs_orphan_add(trans, new_inode);
3524 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3528 ret = btrfs_add_link(trans, new_dentry, old_inode, 1, index);
3533 btrfs_end_transaction_throttle(trans, root);
3538 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3540 struct list_head *head = &root->fs_info->delalloc_inodes;
3541 struct btrfs_inode *binode;
3542 unsigned long flags;
3544 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3545 while(!list_empty(head)) {
3546 binode = list_entry(head->next, struct btrfs_inode,
3548 atomic_inc(&binode->vfs_inode.i_count);
3549 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3550 filemap_write_and_wait(binode->vfs_inode.i_mapping);
3551 iput(&binode->vfs_inode);
3552 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3554 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3558 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3559 const char *symname)
3561 struct btrfs_trans_handle *trans;
3562 struct btrfs_root *root = BTRFS_I(dir)->root;
3563 struct btrfs_path *path;
3564 struct btrfs_key key;
3565 struct inode *inode = NULL;
3573 struct btrfs_file_extent_item *ei;
3574 struct extent_buffer *leaf;
3575 unsigned long nr = 0;
3577 name_len = strlen(symname) + 1;
3578 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3579 return -ENAMETOOLONG;
3581 err = btrfs_check_free_space(root, 1, 0);
3585 trans = btrfs_start_transaction(root, 1);
3586 btrfs_set_trans_block_group(trans, dir);
3588 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3594 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3596 dentry->d_parent->d_inode->i_ino, objectid,
3597 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3599 err = PTR_ERR(inode);
3603 err = btrfs_init_acl(inode, dir);
3609 btrfs_set_trans_block_group(trans, inode);
3610 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3614 inode->i_mapping->a_ops = &btrfs_aops;
3615 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3616 inode->i_fop = &btrfs_file_operations;
3617 inode->i_op = &btrfs_file_inode_operations;
3618 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3619 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3620 inode->i_mapping, GFP_NOFS);
3621 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3622 inode->i_mapping, GFP_NOFS);
3623 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3624 mutex_init(&BTRFS_I(inode)->csum_mutex);
3625 mutex_init(&BTRFS_I(inode)->extent_mutex);
3626 BTRFS_I(inode)->delalloc_bytes = 0;
3627 BTRFS_I(inode)->disk_i_size = 0;
3628 inode->i_mapping->writeback_index = 0;
3629 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3630 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3632 dir->i_sb->s_dirt = 1;
3633 btrfs_update_inode_block_group(trans, inode);
3634 btrfs_update_inode_block_group(trans, dir);
3638 path = btrfs_alloc_path();
3640 key.objectid = inode->i_ino;
3642 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3643 datasize = btrfs_file_extent_calc_inline_size(name_len);
3644 err = btrfs_insert_empty_item(trans, root, path, &key,
3650 leaf = path->nodes[0];
3651 ei = btrfs_item_ptr(leaf, path->slots[0],
3652 struct btrfs_file_extent_item);
3653 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3654 btrfs_set_file_extent_type(leaf, ei,
3655 BTRFS_FILE_EXTENT_INLINE);
3656 ptr = btrfs_file_extent_inline_start(ei);
3657 write_extent_buffer(leaf, symname, ptr, name_len);
3658 btrfs_mark_buffer_dirty(leaf);
3659 btrfs_free_path(path);
3661 inode->i_op = &btrfs_symlink_inode_operations;
3662 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3663 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3664 btrfs_i_size_write(inode, name_len - 1);
3665 err = btrfs_update_inode(trans, root, inode);
3670 nr = trans->blocks_used;
3671 btrfs_end_transaction_throttle(trans, root);
3674 inode_dec_link_count(inode);
3677 btrfs_btree_balance_dirty(root, nr);
3681 static int btrfs_set_page_dirty(struct page *page)
3683 return __set_page_dirty_nobuffers(page);
3686 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3687 static int btrfs_permission(struct inode *inode, int mask)
3689 static int btrfs_permission(struct inode *inode, int mask,
3690 struct nameidata *nd)
3693 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3695 return generic_permission(inode, mask, btrfs_check_acl);
3698 static struct inode_operations btrfs_dir_inode_operations = {
3699 .lookup = btrfs_lookup,
3700 .create = btrfs_create,
3701 .unlink = btrfs_unlink,
3703 .mkdir = btrfs_mkdir,
3704 .rmdir = btrfs_rmdir,
3705 .rename = btrfs_rename,
3706 .symlink = btrfs_symlink,
3707 .setattr = btrfs_setattr,
3708 .mknod = btrfs_mknod,
3709 .setxattr = generic_setxattr,
3710 .getxattr = generic_getxattr,
3711 .listxattr = btrfs_listxattr,
3712 .removexattr = generic_removexattr,
3713 .permission = btrfs_permission,
3715 static struct inode_operations btrfs_dir_ro_inode_operations = {
3716 .lookup = btrfs_lookup,
3717 .permission = btrfs_permission,
3719 static struct file_operations btrfs_dir_file_operations = {
3720 .llseek = generic_file_llseek,
3721 .read = generic_read_dir,
3722 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
3723 .readdir = btrfs_nfshack_readdir,
3724 #else /* NFSd readdir/lookup deadlock is fixed */
3725 .readdir = btrfs_real_readdir,
3727 .unlocked_ioctl = btrfs_ioctl,
3728 #ifdef CONFIG_COMPAT
3729 .compat_ioctl = btrfs_ioctl,
3731 .release = btrfs_release_file,
3734 static struct extent_io_ops btrfs_extent_io_ops = {
3735 .fill_delalloc = run_delalloc_range,
3736 .submit_bio_hook = btrfs_submit_bio_hook,
3737 .merge_bio_hook = btrfs_merge_bio_hook,
3738 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3739 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3740 .writepage_start_hook = btrfs_writepage_start_hook,
3741 .readpage_io_failed_hook = btrfs_io_failed_hook,
3742 .set_bit_hook = btrfs_set_bit_hook,
3743 .clear_bit_hook = btrfs_clear_bit_hook,
3746 static struct address_space_operations btrfs_aops = {
3747 .readpage = btrfs_readpage,
3748 .writepage = btrfs_writepage,
3749 .writepages = btrfs_writepages,
3750 .readpages = btrfs_readpages,
3751 .sync_page = block_sync_page,
3753 .direct_IO = btrfs_direct_IO,
3754 .invalidatepage = btrfs_invalidatepage,
3755 .releasepage = btrfs_releasepage,
3756 .set_page_dirty = btrfs_set_page_dirty,
3759 static struct address_space_operations btrfs_symlink_aops = {
3760 .readpage = btrfs_readpage,
3761 .writepage = btrfs_writepage,
3762 .invalidatepage = btrfs_invalidatepage,
3763 .releasepage = btrfs_releasepage,
3766 static struct inode_operations btrfs_file_inode_operations = {
3767 .truncate = btrfs_truncate,
3768 .getattr = btrfs_getattr,
3769 .setattr = btrfs_setattr,
3770 .setxattr = generic_setxattr,
3771 .getxattr = generic_getxattr,
3772 .listxattr = btrfs_listxattr,
3773 .removexattr = generic_removexattr,
3774 .permission = btrfs_permission,
3776 static struct inode_operations btrfs_special_inode_operations = {
3777 .getattr = btrfs_getattr,
3778 .setattr = btrfs_setattr,
3779 .permission = btrfs_permission,
3780 .setxattr = generic_setxattr,
3781 .getxattr = generic_getxattr,
3782 .listxattr = btrfs_listxattr,
3783 .removexattr = generic_removexattr,
3785 static struct inode_operations btrfs_symlink_inode_operations = {
3786 .readlink = generic_readlink,
3787 .follow_link = page_follow_link_light,
3788 .put_link = page_put_link,
3789 .permission = btrfs_permission,