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/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
50 #include "compression.h"
53 struct btrfs_iget_args {
55 struct btrfs_root *root;
58 static struct inode_operations btrfs_dir_inode_operations;
59 static struct inode_operations btrfs_symlink_inode_operations;
60 static struct inode_operations btrfs_dir_ro_inode_operations;
61 static struct inode_operations btrfs_special_inode_operations;
62 static struct inode_operations btrfs_file_inode_operations;
63 static struct address_space_operations btrfs_aops;
64 static struct address_space_operations btrfs_symlink_aops;
65 static struct file_operations btrfs_dir_file_operations;
66 static struct extent_io_ops btrfs_extent_io_ops;
68 static struct kmem_cache *btrfs_inode_cachep;
69 struct kmem_cache *btrfs_trans_handle_cachep;
70 struct kmem_cache *btrfs_transaction_cachep;
71 struct kmem_cache *btrfs_path_cachep;
74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
75 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
76 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
77 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
78 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
79 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
80 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
81 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
84 static void btrfs_truncate(struct inode *inode);
85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
86 static noinline int cow_file_range(struct inode *inode,
87 struct page *locked_page,
88 u64 start, u64 end, int *page_started,
89 unsigned long *nr_written, int unlock);
91 static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
95 err = btrfs_init_acl(inode, dir);
97 err = btrfs_xattr_security_init(inode, dir);
102 * this does all the hard work for inserting an inline extent into
103 * the btree. The caller should have done a btrfs_drop_extents so that
104 * no overlapping inline items exist in the btree
106 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root, struct inode *inode,
108 u64 start, size_t size, size_t compressed_size,
109 struct page **compressed_pages)
111 struct btrfs_key key;
112 struct btrfs_path *path;
113 struct extent_buffer *leaf;
114 struct page *page = NULL;
117 struct btrfs_file_extent_item *ei;
120 size_t cur_size = size;
122 unsigned long offset;
123 int use_compress = 0;
125 if (compressed_size && compressed_pages) {
127 cur_size = compressed_size;
130 path = btrfs_alloc_path();
134 path->leave_spinning = 1;
135 btrfs_set_trans_block_group(trans, inode);
137 key.objectid = inode->i_ino;
139 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
140 datasize = btrfs_file_extent_calc_inline_size(cur_size);
142 inode_add_bytes(inode, size);
143 ret = btrfs_insert_empty_item(trans, root, path, &key,
150 leaf = path->nodes[0];
151 ei = btrfs_item_ptr(leaf, path->slots[0],
152 struct btrfs_file_extent_item);
153 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
154 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
155 btrfs_set_file_extent_encryption(leaf, ei, 0);
156 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
157 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
158 ptr = btrfs_file_extent_inline_start(ei);
163 while (compressed_size > 0) {
164 cpage = compressed_pages[i];
165 cur_size = min_t(unsigned long, compressed_size,
168 kaddr = kmap_atomic(cpage, KM_USER0);
169 write_extent_buffer(leaf, kaddr, ptr, cur_size);
170 kunmap_atomic(kaddr, KM_USER0);
174 compressed_size -= cur_size;
176 btrfs_set_file_extent_compression(leaf, ei,
177 BTRFS_COMPRESS_ZLIB);
179 page = find_get_page(inode->i_mapping,
180 start >> PAGE_CACHE_SHIFT);
181 btrfs_set_file_extent_compression(leaf, ei, 0);
182 kaddr = kmap_atomic(page, KM_USER0);
183 offset = start & (PAGE_CACHE_SIZE - 1);
184 write_extent_buffer(leaf, kaddr + offset, ptr, size);
185 kunmap_atomic(kaddr, KM_USER0);
186 page_cache_release(page);
188 btrfs_mark_buffer_dirty(leaf);
189 btrfs_free_path(path);
191 BTRFS_I(inode)->disk_i_size = inode->i_size;
192 btrfs_update_inode(trans, root, inode);
195 btrfs_free_path(path);
201 * conditionally insert an inline extent into the file. This
202 * does the checks required to make sure the data is small enough
203 * to fit as an inline extent.
205 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
206 struct btrfs_root *root,
207 struct inode *inode, u64 start, u64 end,
208 size_t compressed_size,
209 struct page **compressed_pages)
211 u64 isize = i_size_read(inode);
212 u64 actual_end = min(end + 1, isize);
213 u64 inline_len = actual_end - start;
214 u64 aligned_end = (end + root->sectorsize - 1) &
215 ~((u64)root->sectorsize - 1);
217 u64 data_len = inline_len;
221 data_len = compressed_size;
224 actual_end >= PAGE_CACHE_SIZE ||
225 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227 (actual_end & (root->sectorsize - 1)) == 0) ||
229 data_len > root->fs_info->max_inline) {
233 ret = btrfs_drop_extents(trans, root, inode, start,
234 aligned_end, aligned_end, start,
238 if (isize > actual_end)
239 inline_len = min_t(u64, isize, actual_end);
240 ret = insert_inline_extent(trans, root, inode, start,
241 inline_len, compressed_size,
244 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
248 struct async_extent {
253 unsigned long nr_pages;
254 struct list_head list;
259 struct btrfs_root *root;
260 struct page *locked_page;
263 struct list_head extents;
264 struct btrfs_work work;
267 static noinline int add_async_extent(struct async_cow *cow,
268 u64 start, u64 ram_size,
271 unsigned long nr_pages)
273 struct async_extent *async_extent;
275 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
276 async_extent->start = start;
277 async_extent->ram_size = ram_size;
278 async_extent->compressed_size = compressed_size;
279 async_extent->pages = pages;
280 async_extent->nr_pages = nr_pages;
281 list_add_tail(&async_extent->list, &cow->extents);
286 * we create compressed extents in two phases. The first
287 * phase compresses a range of pages that have already been
288 * locked (both pages and state bits are locked).
290 * This is done inside an ordered work queue, and the compression
291 * is spread across many cpus. The actual IO submission is step
292 * two, and the ordered work queue takes care of making sure that
293 * happens in the same order things were put onto the queue by
294 * writepages and friends.
296 * If this code finds it can't get good compression, it puts an
297 * entry onto the work queue to write the uncompressed bytes. This
298 * makes sure that both compressed inodes and uncompressed inodes
299 * are written in the same order that pdflush sent them down.
301 static noinline int compress_file_range(struct inode *inode,
302 struct page *locked_page,
304 struct async_cow *async_cow,
307 struct btrfs_root *root = BTRFS_I(inode)->root;
308 struct btrfs_trans_handle *trans;
312 u64 blocksize = root->sectorsize;
314 u64 isize = i_size_read(inode);
316 struct page **pages = NULL;
317 unsigned long nr_pages;
318 unsigned long nr_pages_ret = 0;
319 unsigned long total_compressed = 0;
320 unsigned long total_in = 0;
321 unsigned long max_compressed = 128 * 1024;
322 unsigned long max_uncompressed = 128 * 1024;
328 actual_end = min_t(u64, isize, end + 1);
331 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
332 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
335 * we don't want to send crud past the end of i_size through
336 * compression, that's just a waste of CPU time. So, if the
337 * end of the file is before the start of our current
338 * requested range of bytes, we bail out to the uncompressed
339 * cleanup code that can deal with all of this.
341 * It isn't really the fastest way to fix things, but this is a
342 * very uncommon corner.
344 if (actual_end <= start)
345 goto cleanup_and_bail_uncompressed;
347 total_compressed = actual_end - start;
349 /* we want to make sure that amount of ram required to uncompress
350 * an extent is reasonable, so we limit the total size in ram
351 * of a compressed extent to 128k. This is a crucial number
352 * because it also controls how easily we can spread reads across
353 * cpus for decompression.
355 * We also want to make sure the amount of IO required to do
356 * a random read is reasonably small, so we limit the size of
357 * a compressed extent to 128k.
359 total_compressed = min(total_compressed, max_uncompressed);
360 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
361 num_bytes = max(blocksize, num_bytes);
362 disk_num_bytes = num_bytes;
367 * we do compression for mount -o compress and when the
368 * inode has not been flagged as nocompress. This flag can
369 * change at any time if we discover bad compression ratios.
371 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
372 btrfs_test_opt(root, COMPRESS)) {
374 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
376 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
377 total_compressed, pages,
378 nr_pages, &nr_pages_ret,
384 unsigned long offset = total_compressed &
385 (PAGE_CACHE_SIZE - 1);
386 struct page *page = pages[nr_pages_ret - 1];
389 /* zero the tail end of the last page, we might be
390 * sending it down to disk
393 kaddr = kmap_atomic(page, KM_USER0);
394 memset(kaddr + offset, 0,
395 PAGE_CACHE_SIZE - offset);
396 kunmap_atomic(kaddr, KM_USER0);
402 trans = btrfs_join_transaction(root, 1);
404 btrfs_set_trans_block_group(trans, inode);
406 /* lets try to make an inline extent */
407 if (ret || total_in < (actual_end - start)) {
408 /* we didn't compress the entire range, try
409 * to make an uncompressed inline extent.
411 ret = cow_file_range_inline(trans, root, inode,
412 start, end, 0, NULL);
414 /* try making a compressed inline extent */
415 ret = cow_file_range_inline(trans, root, inode,
417 total_compressed, pages);
419 btrfs_end_transaction(trans, root);
422 * inline extent creation worked, we don't need
423 * to create any more async work items. Unlock
424 * and free up our temp pages.
426 extent_clear_unlock_delalloc(inode,
427 &BTRFS_I(inode)->io_tree,
429 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
430 EXTENT_CLEAR_DELALLOC |
431 EXTENT_CLEAR_ACCOUNTING |
432 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
440 * we aren't doing an inline extent round the compressed size
441 * up to a block size boundary so the allocator does sane
444 total_compressed = (total_compressed + blocksize - 1) &
448 * one last check to make sure the compression is really a
449 * win, compare the page count read with the blocks on disk
451 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
452 ~(PAGE_CACHE_SIZE - 1);
453 if (total_compressed >= total_in) {
456 disk_num_bytes = total_compressed;
457 num_bytes = total_in;
460 if (!will_compress && pages) {
462 * the compression code ran but failed to make things smaller,
463 * free any pages it allocated and our page pointer array
465 for (i = 0; i < nr_pages_ret; i++) {
466 WARN_ON(pages[i]->mapping);
467 page_cache_release(pages[i]);
471 total_compressed = 0;
474 /* flag the file so we don't compress in the future */
475 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
480 /* the async work queues will take care of doing actual
481 * allocation on disk for these compressed pages,
482 * and will submit them to the elevator.
484 add_async_extent(async_cow, start, num_bytes,
485 total_compressed, pages, nr_pages_ret);
487 if (start + num_bytes < end && start + num_bytes < actual_end) {
494 cleanup_and_bail_uncompressed:
496 * No compression, but we still need to write the pages in
497 * the file we've been given so far. redirty the locked
498 * page if it corresponds to our extent and set things up
499 * for the async work queue to run cow_file_range to do
500 * the normal delalloc dance
502 if (page_offset(locked_page) >= start &&
503 page_offset(locked_page) <= end) {
504 __set_page_dirty_nobuffers(locked_page);
505 /* unlocked later on in the async handlers */
507 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
515 for (i = 0; i < nr_pages_ret; i++) {
516 WARN_ON(pages[i]->mapping);
517 page_cache_release(pages[i]);
525 * phase two of compressed writeback. This is the ordered portion
526 * of the code, which only gets called in the order the work was
527 * queued. We walk all the async extents created by compress_file_range
528 * and send them down to the disk.
530 static noinline int submit_compressed_extents(struct inode *inode,
531 struct async_cow *async_cow)
533 struct async_extent *async_extent;
535 struct btrfs_trans_handle *trans;
536 struct btrfs_key ins;
537 struct extent_map *em;
538 struct btrfs_root *root = BTRFS_I(inode)->root;
539 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
540 struct extent_io_tree *io_tree;
543 if (list_empty(&async_cow->extents))
546 trans = btrfs_join_transaction(root, 1);
548 while (!list_empty(&async_cow->extents)) {
549 async_extent = list_entry(async_cow->extents.next,
550 struct async_extent, list);
551 list_del(&async_extent->list);
553 io_tree = &BTRFS_I(inode)->io_tree;
555 /* did the compression code fall back to uncompressed IO? */
556 if (!async_extent->pages) {
557 int page_started = 0;
558 unsigned long nr_written = 0;
560 lock_extent(io_tree, async_extent->start,
561 async_extent->start +
562 async_extent->ram_size - 1, GFP_NOFS);
564 /* allocate blocks */
565 cow_file_range(inode, async_cow->locked_page,
567 async_extent->start +
568 async_extent->ram_size - 1,
569 &page_started, &nr_written, 0);
572 * if page_started, cow_file_range inserted an
573 * inline extent and took care of all the unlocking
574 * and IO for us. Otherwise, we need to submit
575 * all those pages down to the drive.
578 extent_write_locked_range(io_tree,
579 inode, async_extent->start,
580 async_extent->start +
581 async_extent->ram_size - 1,
589 lock_extent(io_tree, async_extent->start,
590 async_extent->start + async_extent->ram_size - 1,
593 * here we're doing allocation and writeback of the
596 btrfs_drop_extent_cache(inode, async_extent->start,
597 async_extent->start +
598 async_extent->ram_size - 1, 0);
600 ret = btrfs_reserve_extent(trans, root,
601 async_extent->compressed_size,
602 async_extent->compressed_size,
606 em = alloc_extent_map(GFP_NOFS);
607 em->start = async_extent->start;
608 em->len = async_extent->ram_size;
609 em->orig_start = em->start;
611 em->block_start = ins.objectid;
612 em->block_len = ins.offset;
613 em->bdev = root->fs_info->fs_devices->latest_bdev;
614 set_bit(EXTENT_FLAG_PINNED, &em->flags);
615 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
618 write_lock(&em_tree->lock);
619 ret = add_extent_mapping(em_tree, em);
620 write_unlock(&em_tree->lock);
621 if (ret != -EEXIST) {
625 btrfs_drop_extent_cache(inode, async_extent->start,
626 async_extent->start +
627 async_extent->ram_size - 1, 0);
630 ret = btrfs_add_ordered_extent(inode, async_extent->start,
632 async_extent->ram_size,
634 BTRFS_ORDERED_COMPRESSED);
637 btrfs_end_transaction(trans, root);
640 * clear dirty, set writeback and unlock the pages.
642 extent_clear_unlock_delalloc(inode,
643 &BTRFS_I(inode)->io_tree,
645 async_extent->start +
646 async_extent->ram_size - 1,
647 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
648 EXTENT_CLEAR_UNLOCK |
649 EXTENT_CLEAR_DELALLOC |
650 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
652 ret = btrfs_submit_compressed_write(inode,
654 async_extent->ram_size,
656 ins.offset, async_extent->pages,
657 async_extent->nr_pages);
660 trans = btrfs_join_transaction(root, 1);
661 alloc_hint = ins.objectid + ins.offset;
666 btrfs_end_transaction(trans, root);
671 * when extent_io.c finds a delayed allocation range in the file,
672 * the call backs end up in this code. The basic idea is to
673 * allocate extents on disk for the range, and create ordered data structs
674 * in ram to track those extents.
676 * locked_page is the page that writepage had locked already. We use
677 * it to make sure we don't do extra locks or unlocks.
679 * *page_started is set to one if we unlock locked_page and do everything
680 * required to start IO on it. It may be clean and already done with
683 static noinline int cow_file_range(struct inode *inode,
684 struct page *locked_page,
685 u64 start, u64 end, int *page_started,
686 unsigned long *nr_written,
689 struct btrfs_root *root = BTRFS_I(inode)->root;
690 struct btrfs_trans_handle *trans;
693 unsigned long ram_size;
696 u64 blocksize = root->sectorsize;
698 u64 isize = i_size_read(inode);
699 struct btrfs_key ins;
700 struct extent_map *em;
701 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
704 trans = btrfs_join_transaction(root, 1);
706 btrfs_set_trans_block_group(trans, inode);
708 actual_end = min_t(u64, isize, end + 1);
710 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
711 num_bytes = max(blocksize, num_bytes);
712 disk_num_bytes = num_bytes;
716 /* lets try to make an inline extent */
717 ret = cow_file_range_inline(trans, root, inode,
718 start, end, 0, NULL);
720 extent_clear_unlock_delalloc(inode,
721 &BTRFS_I(inode)->io_tree,
723 EXTENT_CLEAR_UNLOCK_PAGE |
724 EXTENT_CLEAR_UNLOCK |
725 EXTENT_CLEAR_DELALLOC |
726 EXTENT_CLEAR_ACCOUNTING |
728 EXTENT_SET_WRITEBACK |
729 EXTENT_END_WRITEBACK);
730 *nr_written = *nr_written +
731 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
738 BUG_ON(disk_num_bytes >
739 btrfs_super_total_bytes(&root->fs_info->super_copy));
742 read_lock(&BTRFS_I(inode)->extent_tree.lock);
743 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
746 alloc_hint = em->block_start;
749 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
750 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
752 while (disk_num_bytes > 0) {
755 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
756 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
757 root->sectorsize, 0, alloc_hint,
761 em = alloc_extent_map(GFP_NOFS);
763 em->orig_start = em->start;
764 ram_size = ins.offset;
765 em->len = ins.offset;
767 em->block_start = ins.objectid;
768 em->block_len = ins.offset;
769 em->bdev = root->fs_info->fs_devices->latest_bdev;
770 set_bit(EXTENT_FLAG_PINNED, &em->flags);
773 write_lock(&em_tree->lock);
774 ret = add_extent_mapping(em_tree, em);
775 write_unlock(&em_tree->lock);
776 if (ret != -EEXIST) {
780 btrfs_drop_extent_cache(inode, start,
781 start + ram_size - 1, 0);
784 cur_alloc_size = ins.offset;
785 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
786 ram_size, cur_alloc_size, 0);
789 if (root->root_key.objectid ==
790 BTRFS_DATA_RELOC_TREE_OBJECTID) {
791 ret = btrfs_reloc_clone_csums(inode, start,
796 if (disk_num_bytes < cur_alloc_size)
799 /* we're not doing compressed IO, don't unlock the first
800 * page (which the caller expects to stay locked), don't
801 * clear any dirty bits and don't set any writeback bits
803 * Do set the Private2 bit so we know this page was properly
804 * setup for writepage
806 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
807 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
810 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
811 start, start + ram_size - 1,
813 disk_num_bytes -= cur_alloc_size;
814 num_bytes -= cur_alloc_size;
815 alloc_hint = ins.objectid + ins.offset;
816 start += cur_alloc_size;
820 btrfs_end_transaction(trans, root);
826 * work queue call back to started compression on a file and pages
828 static noinline void async_cow_start(struct btrfs_work *work)
830 struct async_cow *async_cow;
832 async_cow = container_of(work, struct async_cow, work);
834 compress_file_range(async_cow->inode, async_cow->locked_page,
835 async_cow->start, async_cow->end, async_cow,
838 async_cow->inode = NULL;
842 * work queue call back to submit previously compressed pages
844 static noinline void async_cow_submit(struct btrfs_work *work)
846 struct async_cow *async_cow;
847 struct btrfs_root *root;
848 unsigned long nr_pages;
850 async_cow = container_of(work, struct async_cow, work);
852 root = async_cow->root;
853 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
856 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
858 if (atomic_read(&root->fs_info->async_delalloc_pages) <
860 waitqueue_active(&root->fs_info->async_submit_wait))
861 wake_up(&root->fs_info->async_submit_wait);
863 if (async_cow->inode)
864 submit_compressed_extents(async_cow->inode, async_cow);
867 static noinline void async_cow_free(struct btrfs_work *work)
869 struct async_cow *async_cow;
870 async_cow = container_of(work, struct async_cow, work);
874 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
875 u64 start, u64 end, int *page_started,
876 unsigned long *nr_written)
878 struct async_cow *async_cow;
879 struct btrfs_root *root = BTRFS_I(inode)->root;
880 unsigned long nr_pages;
882 int limit = 10 * 1024 * 1042;
884 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
885 1, 0, NULL, GFP_NOFS);
886 while (start < end) {
887 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
888 async_cow->inode = inode;
889 async_cow->root = root;
890 async_cow->locked_page = locked_page;
891 async_cow->start = start;
893 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
896 cur_end = min(end, start + 512 * 1024 - 1);
898 async_cow->end = cur_end;
899 INIT_LIST_HEAD(&async_cow->extents);
901 async_cow->work.func = async_cow_start;
902 async_cow->work.ordered_func = async_cow_submit;
903 async_cow->work.ordered_free = async_cow_free;
904 async_cow->work.flags = 0;
906 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
908 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
910 btrfs_queue_worker(&root->fs_info->delalloc_workers,
913 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
914 wait_event(root->fs_info->async_submit_wait,
915 (atomic_read(&root->fs_info->async_delalloc_pages) <
919 while (atomic_read(&root->fs_info->async_submit_draining) &&
920 atomic_read(&root->fs_info->async_delalloc_pages)) {
921 wait_event(root->fs_info->async_submit_wait,
922 (atomic_read(&root->fs_info->async_delalloc_pages) ==
926 *nr_written += nr_pages;
933 static noinline int csum_exist_in_range(struct btrfs_root *root,
934 u64 bytenr, u64 num_bytes)
937 struct btrfs_ordered_sum *sums;
940 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
941 bytenr + num_bytes - 1, &list);
942 if (ret == 0 && list_empty(&list))
945 while (!list_empty(&list)) {
946 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
947 list_del(&sums->list);
954 * when nowcow writeback call back. This checks for snapshots or COW copies
955 * of the extents that exist in the file, and COWs the file as required.
957 * If no cow copies or snapshots exist, we write directly to the existing
960 static noinline int run_delalloc_nocow(struct inode *inode,
961 struct page *locked_page,
962 u64 start, u64 end, int *page_started, int force,
963 unsigned long *nr_written)
965 struct btrfs_root *root = BTRFS_I(inode)->root;
966 struct btrfs_trans_handle *trans;
967 struct extent_buffer *leaf;
968 struct btrfs_path *path;
969 struct btrfs_file_extent_item *fi;
970 struct btrfs_key found_key;
983 path = btrfs_alloc_path();
985 trans = btrfs_join_transaction(root, 1);
991 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
994 if (ret > 0 && path->slots[0] > 0 && check_prev) {
995 leaf = path->nodes[0];
996 btrfs_item_key_to_cpu(leaf, &found_key,
998 if (found_key.objectid == inode->i_ino &&
999 found_key.type == BTRFS_EXTENT_DATA_KEY)
1004 leaf = path->nodes[0];
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1011 leaf = path->nodes[0];
1017 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1019 if (found_key.objectid > inode->i_ino ||
1020 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1021 found_key.offset > end)
1024 if (found_key.offset > cur_offset) {
1025 extent_end = found_key.offset;
1029 fi = btrfs_item_ptr(leaf, path->slots[0],
1030 struct btrfs_file_extent_item);
1031 extent_type = btrfs_file_extent_type(leaf, fi);
1033 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1034 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1035 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1036 extent_offset = btrfs_file_extent_offset(leaf, fi);
1037 extent_end = found_key.offset +
1038 btrfs_file_extent_num_bytes(leaf, fi);
1039 if (extent_end <= start) {
1043 if (disk_bytenr == 0)
1045 if (btrfs_file_extent_compression(leaf, fi) ||
1046 btrfs_file_extent_encryption(leaf, fi) ||
1047 btrfs_file_extent_other_encoding(leaf, fi))
1049 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1051 if (btrfs_extent_readonly(root, disk_bytenr))
1053 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1055 extent_offset, disk_bytenr))
1057 disk_bytenr += extent_offset;
1058 disk_bytenr += cur_offset - found_key.offset;
1059 num_bytes = min(end + 1, extent_end) - cur_offset;
1061 * force cow if csum exists in the range.
1062 * this ensure that csum for a given extent are
1063 * either valid or do not exist.
1065 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1068 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1069 extent_end = found_key.offset +
1070 btrfs_file_extent_inline_len(leaf, fi);
1071 extent_end = ALIGN(extent_end, root->sectorsize);
1076 if (extent_end <= start) {
1081 if (cow_start == (u64)-1)
1082 cow_start = cur_offset;
1083 cur_offset = extent_end;
1084 if (cur_offset > end)
1090 btrfs_release_path(root, path);
1091 if (cow_start != (u64)-1) {
1092 ret = cow_file_range(inode, locked_page, cow_start,
1093 found_key.offset - 1, page_started,
1096 cow_start = (u64)-1;
1099 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1100 struct extent_map *em;
1101 struct extent_map_tree *em_tree;
1102 em_tree = &BTRFS_I(inode)->extent_tree;
1103 em = alloc_extent_map(GFP_NOFS);
1104 em->start = cur_offset;
1105 em->orig_start = em->start;
1106 em->len = num_bytes;
1107 em->block_len = num_bytes;
1108 em->block_start = disk_bytenr;
1109 em->bdev = root->fs_info->fs_devices->latest_bdev;
1110 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1112 write_lock(&em_tree->lock);
1113 ret = add_extent_mapping(em_tree, em);
1114 write_unlock(&em_tree->lock);
1115 if (ret != -EEXIST) {
1116 free_extent_map(em);
1119 btrfs_drop_extent_cache(inode, em->start,
1120 em->start + em->len - 1, 0);
1122 type = BTRFS_ORDERED_PREALLOC;
1124 type = BTRFS_ORDERED_NOCOW;
1127 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1128 num_bytes, num_bytes, type);
1131 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1132 cur_offset, cur_offset + num_bytes - 1,
1133 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1134 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1135 EXTENT_SET_PRIVATE2);
1136 cur_offset = extent_end;
1137 if (cur_offset > end)
1140 btrfs_release_path(root, path);
1142 if (cur_offset <= end && cow_start == (u64)-1)
1143 cow_start = cur_offset;
1144 if (cow_start != (u64)-1) {
1145 ret = cow_file_range(inode, locked_page, cow_start, end,
1146 page_started, nr_written, 1);
1150 ret = btrfs_end_transaction(trans, root);
1152 btrfs_free_path(path);
1157 * extent_io.c call back to do delayed allocation processing
1159 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1160 u64 start, u64 end, int *page_started,
1161 unsigned long *nr_written)
1164 struct btrfs_root *root = BTRFS_I(inode)->root;
1166 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1167 ret = run_delalloc_nocow(inode, locked_page, start, end,
1168 page_started, 1, nr_written);
1169 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1170 ret = run_delalloc_nocow(inode, locked_page, start, end,
1171 page_started, 0, nr_written);
1172 else if (!btrfs_test_opt(root, COMPRESS))
1173 ret = cow_file_range(inode, locked_page, start, end,
1174 page_started, nr_written, 1);
1176 ret = cow_file_range_async(inode, locked_page, start, end,
1177 page_started, nr_written);
1181 static int btrfs_split_extent_hook(struct inode *inode,
1182 struct extent_state *orig, u64 split)
1184 struct btrfs_root *root = BTRFS_I(inode)->root;
1187 if (!(orig->state & EXTENT_DELALLOC))
1190 size = orig->end - orig->start + 1;
1191 if (size > root->fs_info->max_extent) {
1195 new_size = orig->end - split + 1;
1196 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1197 root->fs_info->max_extent);
1200 * if we break a large extent up then leave oustanding_extents
1201 * be, since we've already accounted for the large extent.
1203 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1204 root->fs_info->max_extent) < num_extents)
1208 spin_lock(&BTRFS_I(inode)->accounting_lock);
1209 BTRFS_I(inode)->outstanding_extents++;
1210 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1216 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1217 * extents so we can keep track of new extents that are just merged onto old
1218 * extents, such as when we are doing sequential writes, so we can properly
1219 * account for the metadata space we'll need.
1221 static int btrfs_merge_extent_hook(struct inode *inode,
1222 struct extent_state *new,
1223 struct extent_state *other)
1225 struct btrfs_root *root = BTRFS_I(inode)->root;
1226 u64 new_size, old_size;
1229 /* not delalloc, ignore it */
1230 if (!(other->state & EXTENT_DELALLOC))
1233 old_size = other->end - other->start + 1;
1234 if (new->start < other->start)
1235 new_size = other->end - new->start + 1;
1237 new_size = new->end - other->start + 1;
1239 /* we're not bigger than the max, unreserve the space and go */
1240 if (new_size <= root->fs_info->max_extent) {
1241 spin_lock(&BTRFS_I(inode)->accounting_lock);
1242 BTRFS_I(inode)->outstanding_extents--;
1243 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1248 * If we grew by another max_extent, just return, we want to keep that
1251 num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1252 root->fs_info->max_extent);
1253 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1254 root->fs_info->max_extent) > num_extents)
1257 spin_lock(&BTRFS_I(inode)->accounting_lock);
1258 BTRFS_I(inode)->outstanding_extents--;
1259 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1265 * extent_io.c set_bit_hook, used to track delayed allocation
1266 * bytes in this file, and to maintain the list of inodes that
1267 * have pending delalloc work to be done.
1269 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1270 unsigned long old, unsigned long bits)
1274 * set_bit and clear bit hooks normally require _irqsave/restore
1275 * but in this case, we are only testeing for the DELALLOC
1276 * bit, which is only set or cleared with irqs on
1278 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1279 struct btrfs_root *root = BTRFS_I(inode)->root;
1281 spin_lock(&BTRFS_I(inode)->accounting_lock);
1282 BTRFS_I(inode)->outstanding_extents++;
1283 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1284 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1285 spin_lock(&root->fs_info->delalloc_lock);
1286 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1287 root->fs_info->delalloc_bytes += end - start + 1;
1288 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1289 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1290 &root->fs_info->delalloc_inodes);
1292 spin_unlock(&root->fs_info->delalloc_lock);
1298 * extent_io.c clear_bit_hook, see set_bit_hook for why
1300 static int btrfs_clear_bit_hook(struct inode *inode,
1301 struct extent_state *state, unsigned long bits)
1304 * set_bit and clear bit hooks normally require _irqsave/restore
1305 * but in this case, we are only testeing for the DELALLOC
1306 * bit, which is only set or cleared with irqs on
1308 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1309 struct btrfs_root *root = BTRFS_I(inode)->root;
1311 if (bits & EXTENT_DO_ACCOUNTING) {
1312 spin_lock(&BTRFS_I(inode)->accounting_lock);
1313 BTRFS_I(inode)->outstanding_extents--;
1314 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1315 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1318 spin_lock(&root->fs_info->delalloc_lock);
1319 if (state->end - state->start + 1 >
1320 root->fs_info->delalloc_bytes) {
1321 printk(KERN_INFO "btrfs warning: delalloc account "
1323 (unsigned long long)
1324 state->end - state->start + 1,
1325 (unsigned long long)
1326 root->fs_info->delalloc_bytes);
1327 btrfs_delalloc_free_space(root, inode, (u64)-1);
1328 root->fs_info->delalloc_bytes = 0;
1329 BTRFS_I(inode)->delalloc_bytes = 0;
1331 btrfs_delalloc_free_space(root, inode,
1334 root->fs_info->delalloc_bytes -= state->end -
1336 BTRFS_I(inode)->delalloc_bytes -= state->end -
1339 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1340 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1341 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1343 spin_unlock(&root->fs_info->delalloc_lock);
1349 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1350 * we don't create bios that span stripes or chunks
1352 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1353 size_t size, struct bio *bio,
1354 unsigned long bio_flags)
1356 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1357 struct btrfs_mapping_tree *map_tree;
1358 u64 logical = (u64)bio->bi_sector << 9;
1363 if (bio_flags & EXTENT_BIO_COMPRESSED)
1366 length = bio->bi_size;
1367 map_tree = &root->fs_info->mapping_tree;
1368 map_length = length;
1369 ret = btrfs_map_block(map_tree, READ, logical,
1370 &map_length, NULL, 0);
1372 if (map_length < length + size)
1378 * in order to insert checksums into the metadata in large chunks,
1379 * we wait until bio submission time. All the pages in the bio are
1380 * checksummed and sums are attached onto the ordered extent record.
1382 * At IO completion time the cums attached on the ordered extent record
1383 * are inserted into the btree
1385 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1386 struct bio *bio, int mirror_num,
1387 unsigned long bio_flags)
1389 struct btrfs_root *root = BTRFS_I(inode)->root;
1392 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1398 * in order to insert checksums into the metadata in large chunks,
1399 * we wait until bio submission time. All the pages in the bio are
1400 * checksummed and sums are attached onto the ordered extent record.
1402 * At IO completion time the cums attached on the ordered extent record
1403 * are inserted into the btree
1405 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1406 int mirror_num, unsigned long bio_flags)
1408 struct btrfs_root *root = BTRFS_I(inode)->root;
1409 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1413 * extent_io.c submission hook. This does the right thing for csum calculation
1414 * on write, or reading the csums from the tree before a read
1416 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1417 int mirror_num, unsigned long bio_flags)
1419 struct btrfs_root *root = BTRFS_I(inode)->root;
1423 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1425 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1428 if (!(rw & (1 << BIO_RW))) {
1429 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1430 return btrfs_submit_compressed_read(inode, bio,
1431 mirror_num, bio_flags);
1432 } else if (!skip_sum)
1433 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1435 } else if (!skip_sum) {
1436 /* csum items have already been cloned */
1437 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1439 /* we're doing a write, do the async checksumming */
1440 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1441 inode, rw, bio, mirror_num,
1442 bio_flags, __btrfs_submit_bio_start,
1443 __btrfs_submit_bio_done);
1447 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1451 * given a list of ordered sums record them in the inode. This happens
1452 * at IO completion time based on sums calculated at bio submission time.
1454 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1455 struct inode *inode, u64 file_offset,
1456 struct list_head *list)
1458 struct btrfs_ordered_sum *sum;
1460 btrfs_set_trans_block_group(trans, inode);
1462 list_for_each_entry(sum, list, list) {
1463 btrfs_csum_file_blocks(trans,
1464 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1469 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1471 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1473 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1477 /* see btrfs_writepage_start_hook for details on why this is required */
1478 struct btrfs_writepage_fixup {
1480 struct btrfs_work work;
1483 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1485 struct btrfs_writepage_fixup *fixup;
1486 struct btrfs_ordered_extent *ordered;
1488 struct inode *inode;
1492 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1496 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1497 ClearPageChecked(page);
1501 inode = page->mapping->host;
1502 page_start = page_offset(page);
1503 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1505 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1507 /* already ordered? We're done */
1508 if (PagePrivate2(page))
1511 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1513 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1514 page_end, GFP_NOFS);
1516 btrfs_start_ordered_extent(inode, ordered, 1);
1520 btrfs_set_extent_delalloc(inode, page_start, page_end);
1521 ClearPageChecked(page);
1523 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1526 page_cache_release(page);
1530 * There are a few paths in the higher layers of the kernel that directly
1531 * set the page dirty bit without asking the filesystem if it is a
1532 * good idea. This causes problems because we want to make sure COW
1533 * properly happens and the data=ordered rules are followed.
1535 * In our case any range that doesn't have the ORDERED bit set
1536 * hasn't been properly setup for IO. We kick off an async process
1537 * to fix it up. The async helper will wait for ordered extents, set
1538 * the delalloc bit and make it safe to write the page.
1540 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1542 struct inode *inode = page->mapping->host;
1543 struct btrfs_writepage_fixup *fixup;
1544 struct btrfs_root *root = BTRFS_I(inode)->root;
1546 /* this page is properly in the ordered list */
1547 if (TestClearPagePrivate2(page))
1550 if (PageChecked(page))
1553 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1557 SetPageChecked(page);
1558 page_cache_get(page);
1559 fixup->work.func = btrfs_writepage_fixup_worker;
1561 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1565 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1566 struct inode *inode, u64 file_pos,
1567 u64 disk_bytenr, u64 disk_num_bytes,
1568 u64 num_bytes, u64 ram_bytes,
1570 u8 compression, u8 encryption,
1571 u16 other_encoding, int extent_type)
1573 struct btrfs_root *root = BTRFS_I(inode)->root;
1574 struct btrfs_file_extent_item *fi;
1575 struct btrfs_path *path;
1576 struct extent_buffer *leaf;
1577 struct btrfs_key ins;
1581 path = btrfs_alloc_path();
1584 path->leave_spinning = 1;
1587 * we may be replacing one extent in the tree with another.
1588 * The new extent is pinned in the extent map, and we don't want
1589 * to drop it from the cache until it is completely in the btree.
1591 * So, tell btrfs_drop_extents to leave this extent in the cache.
1592 * the caller is expected to unpin it and allow it to be merged
1595 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1596 file_pos + num_bytes, locked_end,
1597 file_pos, &hint, 0);
1600 ins.objectid = inode->i_ino;
1601 ins.offset = file_pos;
1602 ins.type = BTRFS_EXTENT_DATA_KEY;
1603 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1605 leaf = path->nodes[0];
1606 fi = btrfs_item_ptr(leaf, path->slots[0],
1607 struct btrfs_file_extent_item);
1608 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1609 btrfs_set_file_extent_type(leaf, fi, extent_type);
1610 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1611 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1612 btrfs_set_file_extent_offset(leaf, fi, 0);
1613 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1614 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1615 btrfs_set_file_extent_compression(leaf, fi, compression);
1616 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1617 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1619 btrfs_unlock_up_safe(path, 1);
1620 btrfs_set_lock_blocking(leaf);
1622 btrfs_mark_buffer_dirty(leaf);
1624 inode_add_bytes(inode, num_bytes);
1626 ins.objectid = disk_bytenr;
1627 ins.offset = disk_num_bytes;
1628 ins.type = BTRFS_EXTENT_ITEM_KEY;
1629 ret = btrfs_alloc_reserved_file_extent(trans, root,
1630 root->root_key.objectid,
1631 inode->i_ino, file_pos, &ins);
1633 btrfs_free_path(path);
1639 * helper function for btrfs_finish_ordered_io, this
1640 * just reads in some of the csum leaves to prime them into ram
1641 * before we start the transaction. It limits the amount of btree
1642 * reads required while inside the transaction.
1644 static noinline void reada_csum(struct btrfs_root *root,
1645 struct btrfs_path *path,
1646 struct btrfs_ordered_extent *ordered_extent)
1648 struct btrfs_ordered_sum *sum;
1651 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1653 bytenr = sum->sums[0].bytenr;
1656 * we don't care about the results, the point of this search is
1657 * just to get the btree leaves into ram
1659 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1662 /* as ordered data IO finishes, this gets called so we can finish
1663 * an ordered extent if the range of bytes in the file it covers are
1666 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1668 struct btrfs_root *root = BTRFS_I(inode)->root;
1669 struct btrfs_trans_handle *trans;
1670 struct btrfs_ordered_extent *ordered_extent = NULL;
1671 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1672 struct btrfs_path *path;
1676 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1681 * before we join the transaction, try to do some of our IO.
1682 * This will limit the amount of IO that we have to do with
1683 * the transaction running. We're unlikely to need to do any
1684 * IO if the file extents are new, the disk_i_size checks
1685 * covers the most common case.
1687 if (start < BTRFS_I(inode)->disk_i_size) {
1688 path = btrfs_alloc_path();
1690 ret = btrfs_lookup_file_extent(NULL, root, path,
1693 ordered_extent = btrfs_lookup_ordered_extent(inode,
1695 if (!list_empty(&ordered_extent->list)) {
1696 btrfs_release_path(root, path);
1697 reada_csum(root, path, ordered_extent);
1699 btrfs_free_path(path);
1703 trans = btrfs_join_transaction(root, 1);
1705 if (!ordered_extent)
1706 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1707 BUG_ON(!ordered_extent);
1708 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1711 lock_extent(io_tree, ordered_extent->file_offset,
1712 ordered_extent->file_offset + ordered_extent->len - 1,
1715 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1717 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1719 ret = btrfs_mark_extent_written(trans, root, inode,
1720 ordered_extent->file_offset,
1721 ordered_extent->file_offset +
1722 ordered_extent->len);
1725 ret = insert_reserved_file_extent(trans, inode,
1726 ordered_extent->file_offset,
1727 ordered_extent->start,
1728 ordered_extent->disk_len,
1729 ordered_extent->len,
1730 ordered_extent->len,
1731 ordered_extent->file_offset +
1732 ordered_extent->len,
1734 BTRFS_FILE_EXTENT_REG);
1735 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1736 ordered_extent->file_offset,
1737 ordered_extent->len);
1740 unlock_extent(io_tree, ordered_extent->file_offset,
1741 ordered_extent->file_offset + ordered_extent->len - 1,
1744 add_pending_csums(trans, inode, ordered_extent->file_offset,
1745 &ordered_extent->list);
1747 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1748 btrfs_ordered_update_i_size(inode, ordered_extent);
1749 btrfs_update_inode(trans, root, inode);
1750 btrfs_remove_ordered_extent(inode, ordered_extent);
1751 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1754 btrfs_put_ordered_extent(ordered_extent);
1755 /* once for the tree */
1756 btrfs_put_ordered_extent(ordered_extent);
1758 btrfs_end_transaction(trans, root);
1762 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1763 struct extent_state *state, int uptodate)
1765 ClearPagePrivate2(page);
1766 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1770 * When IO fails, either with EIO or csum verification fails, we
1771 * try other mirrors that might have a good copy of the data. This
1772 * io_failure_record is used to record state as we go through all the
1773 * mirrors. If another mirror has good data, the page is set up to date
1774 * and things continue. If a good mirror can't be found, the original
1775 * bio end_io callback is called to indicate things have failed.
1777 struct io_failure_record {
1782 unsigned long bio_flags;
1786 static int btrfs_io_failed_hook(struct bio *failed_bio,
1787 struct page *page, u64 start, u64 end,
1788 struct extent_state *state)
1790 struct io_failure_record *failrec = NULL;
1792 struct extent_map *em;
1793 struct inode *inode = page->mapping->host;
1794 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1795 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1802 ret = get_state_private(failure_tree, start, &private);
1804 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1807 failrec->start = start;
1808 failrec->len = end - start + 1;
1809 failrec->last_mirror = 0;
1810 failrec->bio_flags = 0;
1812 read_lock(&em_tree->lock);
1813 em = lookup_extent_mapping(em_tree, start, failrec->len);
1814 if (em->start > start || em->start + em->len < start) {
1815 free_extent_map(em);
1818 read_unlock(&em_tree->lock);
1820 if (!em || IS_ERR(em)) {
1824 logical = start - em->start;
1825 logical = em->block_start + logical;
1826 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1827 logical = em->block_start;
1828 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1830 failrec->logical = logical;
1831 free_extent_map(em);
1832 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1833 EXTENT_DIRTY, GFP_NOFS);
1834 set_state_private(failure_tree, start,
1835 (u64)(unsigned long)failrec);
1837 failrec = (struct io_failure_record *)(unsigned long)private;
1839 num_copies = btrfs_num_copies(
1840 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1841 failrec->logical, failrec->len);
1842 failrec->last_mirror++;
1844 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1845 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1848 if (state && state->start != failrec->start)
1850 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1852 if (!state || failrec->last_mirror > num_copies) {
1853 set_state_private(failure_tree, failrec->start, 0);
1854 clear_extent_bits(failure_tree, failrec->start,
1855 failrec->start + failrec->len - 1,
1856 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1860 bio = bio_alloc(GFP_NOFS, 1);
1861 bio->bi_private = state;
1862 bio->bi_end_io = failed_bio->bi_end_io;
1863 bio->bi_sector = failrec->logical >> 9;
1864 bio->bi_bdev = failed_bio->bi_bdev;
1867 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1868 if (failed_bio->bi_rw & (1 << BIO_RW))
1873 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1874 failrec->last_mirror,
1875 failrec->bio_flags);
1880 * each time an IO finishes, we do a fast check in the IO failure tree
1881 * to see if we need to process or clean up an io_failure_record
1883 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1886 u64 private_failure;
1887 struct io_failure_record *failure;
1891 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1892 (u64)-1, 1, EXTENT_DIRTY)) {
1893 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1894 start, &private_failure);
1896 failure = (struct io_failure_record *)(unsigned long)
1898 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1900 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1902 failure->start + failure->len - 1,
1903 EXTENT_DIRTY | EXTENT_LOCKED,
1912 * when reads are done, we need to check csums to verify the data is correct
1913 * if there's a match, we allow the bio to finish. If not, we go through
1914 * the io_failure_record routines to find good copies
1916 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1917 struct extent_state *state)
1919 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1920 struct inode *inode = page->mapping->host;
1921 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1923 u64 private = ~(u32)0;
1925 struct btrfs_root *root = BTRFS_I(inode)->root;
1928 if (PageChecked(page)) {
1929 ClearPageChecked(page);
1933 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1936 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1937 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1938 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1943 if (state && state->start == start) {
1944 private = state->private;
1947 ret = get_state_private(io_tree, start, &private);
1949 kaddr = kmap_atomic(page, KM_USER0);
1953 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1954 btrfs_csum_final(csum, (char *)&csum);
1955 if (csum != private)
1958 kunmap_atomic(kaddr, KM_USER0);
1960 /* if the io failure tree for this inode is non-empty,
1961 * check to see if we've recovered from a failed IO
1963 btrfs_clean_io_failures(inode, start);
1967 if (printk_ratelimit()) {
1968 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1969 "private %llu\n", page->mapping->host->i_ino,
1970 (unsigned long long)start, csum,
1971 (unsigned long long)private);
1973 memset(kaddr + offset, 1, end - start + 1);
1974 flush_dcache_page(page);
1975 kunmap_atomic(kaddr, KM_USER0);
1982 * This creates an orphan entry for the given inode in case something goes
1983 * wrong in the middle of an unlink/truncate.
1985 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1987 struct btrfs_root *root = BTRFS_I(inode)->root;
1990 spin_lock(&root->list_lock);
1992 /* already on the orphan list, we're good */
1993 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1994 spin_unlock(&root->list_lock);
1998 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2000 spin_unlock(&root->list_lock);
2003 * insert an orphan item to track this unlinked/truncated file
2005 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2011 * We have done the truncate/delete so we can go ahead and remove the orphan
2012 * item for this particular inode.
2014 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2016 struct btrfs_root *root = BTRFS_I(inode)->root;
2019 spin_lock(&root->list_lock);
2021 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2022 spin_unlock(&root->list_lock);
2026 list_del_init(&BTRFS_I(inode)->i_orphan);
2028 spin_unlock(&root->list_lock);
2032 spin_unlock(&root->list_lock);
2034 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2040 * this cleans up any orphans that may be left on the list from the last use
2043 void btrfs_orphan_cleanup(struct btrfs_root *root)
2045 struct btrfs_path *path;
2046 struct extent_buffer *leaf;
2047 struct btrfs_item *item;
2048 struct btrfs_key key, found_key;
2049 struct btrfs_trans_handle *trans;
2050 struct inode *inode;
2051 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2053 path = btrfs_alloc_path();
2058 key.objectid = BTRFS_ORPHAN_OBJECTID;
2059 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2060 key.offset = (u64)-1;
2064 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2066 printk(KERN_ERR "Error searching slot for orphan: %d"
2072 * if ret == 0 means we found what we were searching for, which
2073 * is weird, but possible, so only screw with path if we didnt
2074 * find the key and see if we have stuff that matches
2077 if (path->slots[0] == 0)
2082 /* pull out the item */
2083 leaf = path->nodes[0];
2084 item = btrfs_item_nr(leaf, path->slots[0]);
2085 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2087 /* make sure the item matches what we want */
2088 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2090 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2093 /* release the path since we're done with it */
2094 btrfs_release_path(root, path);
2097 * this is where we are basically btrfs_lookup, without the
2098 * crossing root thing. we store the inode number in the
2099 * offset of the orphan item.
2101 found_key.objectid = found_key.offset;
2102 found_key.type = BTRFS_INODE_ITEM_KEY;
2103 found_key.offset = 0;
2104 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
2109 * add this inode to the orphan list so btrfs_orphan_del does
2110 * the proper thing when we hit it
2112 spin_lock(&root->list_lock);
2113 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2114 spin_unlock(&root->list_lock);
2117 * if this is a bad inode, means we actually succeeded in
2118 * removing the inode, but not the orphan record, which means
2119 * we need to manually delete the orphan since iput will just
2120 * do a destroy_inode
2122 if (is_bad_inode(inode)) {
2123 trans = btrfs_start_transaction(root, 1);
2124 btrfs_orphan_del(trans, inode);
2125 btrfs_end_transaction(trans, root);
2130 /* if we have links, this was a truncate, lets do that */
2131 if (inode->i_nlink) {
2133 btrfs_truncate(inode);
2138 /* this will do delete_inode and everything for us */
2143 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2145 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2147 btrfs_free_path(path);
2151 * very simple check to peek ahead in the leaf looking for xattrs. If we
2152 * don't find any xattrs, we know there can't be any acls.
2154 * slot is the slot the inode is in, objectid is the objectid of the inode
2156 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2157 int slot, u64 objectid)
2159 u32 nritems = btrfs_header_nritems(leaf);
2160 struct btrfs_key found_key;
2164 while (slot < nritems) {
2165 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2167 /* we found a different objectid, there must not be acls */
2168 if (found_key.objectid != objectid)
2171 /* we found an xattr, assume we've got an acl */
2172 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2176 * we found a key greater than an xattr key, there can't
2177 * be any acls later on
2179 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2186 * it goes inode, inode backrefs, xattrs, extents,
2187 * so if there are a ton of hard links to an inode there can
2188 * be a lot of backrefs. Don't waste time searching too hard,
2189 * this is just an optimization
2194 /* we hit the end of the leaf before we found an xattr or
2195 * something larger than an xattr. We have to assume the inode
2202 * read an inode from the btree into the in-memory inode
2204 static void btrfs_read_locked_inode(struct inode *inode)
2206 struct btrfs_path *path;
2207 struct extent_buffer *leaf;
2208 struct btrfs_inode_item *inode_item;
2209 struct btrfs_timespec *tspec;
2210 struct btrfs_root *root = BTRFS_I(inode)->root;
2211 struct btrfs_key location;
2213 u64 alloc_group_block;
2217 path = btrfs_alloc_path();
2219 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2221 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2225 leaf = path->nodes[0];
2226 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2227 struct btrfs_inode_item);
2229 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2230 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2231 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2232 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2233 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2235 tspec = btrfs_inode_atime(inode_item);
2236 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2237 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2239 tspec = btrfs_inode_mtime(inode_item);
2240 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2241 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2243 tspec = btrfs_inode_ctime(inode_item);
2244 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2245 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2247 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2248 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2249 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2250 inode->i_generation = BTRFS_I(inode)->generation;
2252 rdev = btrfs_inode_rdev(leaf, inode_item);
2254 BTRFS_I(inode)->index_cnt = (u64)-1;
2255 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2257 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2260 * try to precache a NULL acl entry for files that don't have
2261 * any xattrs or acls
2263 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2265 cache_no_acl(inode);
2267 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2268 alloc_group_block, 0);
2269 btrfs_free_path(path);
2272 switch (inode->i_mode & S_IFMT) {
2274 inode->i_mapping->a_ops = &btrfs_aops;
2275 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2276 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2277 inode->i_fop = &btrfs_file_operations;
2278 inode->i_op = &btrfs_file_inode_operations;
2281 inode->i_fop = &btrfs_dir_file_operations;
2282 if (root == root->fs_info->tree_root)
2283 inode->i_op = &btrfs_dir_ro_inode_operations;
2285 inode->i_op = &btrfs_dir_inode_operations;
2288 inode->i_op = &btrfs_symlink_inode_operations;
2289 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2290 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2293 inode->i_op = &btrfs_special_inode_operations;
2294 init_special_inode(inode, inode->i_mode, rdev);
2298 btrfs_update_iflags(inode);
2302 btrfs_free_path(path);
2303 make_bad_inode(inode);
2307 * given a leaf and an inode, copy the inode fields into the leaf
2309 static void fill_inode_item(struct btrfs_trans_handle *trans,
2310 struct extent_buffer *leaf,
2311 struct btrfs_inode_item *item,
2312 struct inode *inode)
2314 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2315 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2316 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2317 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2318 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2320 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2321 inode->i_atime.tv_sec);
2322 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2323 inode->i_atime.tv_nsec);
2325 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2326 inode->i_mtime.tv_sec);
2327 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2328 inode->i_mtime.tv_nsec);
2330 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2331 inode->i_ctime.tv_sec);
2332 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2333 inode->i_ctime.tv_nsec);
2335 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2336 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2337 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2338 btrfs_set_inode_transid(leaf, item, trans->transid);
2339 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2340 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2341 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2345 * copy everything in the in-memory inode into the btree.
2347 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2348 struct btrfs_root *root, struct inode *inode)
2350 struct btrfs_inode_item *inode_item;
2351 struct btrfs_path *path;
2352 struct extent_buffer *leaf;
2355 path = btrfs_alloc_path();
2357 path->leave_spinning = 1;
2358 ret = btrfs_lookup_inode(trans, root, path,
2359 &BTRFS_I(inode)->location, 1);
2366 btrfs_unlock_up_safe(path, 1);
2367 leaf = path->nodes[0];
2368 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2369 struct btrfs_inode_item);
2371 fill_inode_item(trans, leaf, inode_item, inode);
2372 btrfs_mark_buffer_dirty(leaf);
2373 btrfs_set_inode_last_trans(trans, inode);
2376 btrfs_free_path(path);
2382 * unlink helper that gets used here in inode.c and in the tree logging
2383 * recovery code. It remove a link in a directory with a given name, and
2384 * also drops the back refs in the inode to the directory
2386 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2387 struct btrfs_root *root,
2388 struct inode *dir, struct inode *inode,
2389 const char *name, int name_len)
2391 struct btrfs_path *path;
2393 struct extent_buffer *leaf;
2394 struct btrfs_dir_item *di;
2395 struct btrfs_key key;
2398 path = btrfs_alloc_path();
2404 path->leave_spinning = 1;
2405 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2406 name, name_len, -1);
2415 leaf = path->nodes[0];
2416 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2417 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2420 btrfs_release_path(root, path);
2422 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2424 dir->i_ino, &index);
2426 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2427 "inode %lu parent %lu\n", name_len, name,
2428 inode->i_ino, dir->i_ino);
2432 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2433 index, name, name_len, -1);
2442 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2443 btrfs_release_path(root, path);
2445 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2447 BUG_ON(ret != 0 && ret != -ENOENT);
2449 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2453 btrfs_free_path(path);
2457 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2458 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2459 btrfs_update_inode(trans, root, dir);
2460 btrfs_drop_nlink(inode);
2461 ret = btrfs_update_inode(trans, root, inode);
2466 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2468 struct btrfs_root *root;
2469 struct btrfs_trans_handle *trans;
2470 struct inode *inode = dentry->d_inode;
2472 unsigned long nr = 0;
2474 root = BTRFS_I(dir)->root;
2476 trans = btrfs_start_transaction(root, 1);
2478 btrfs_set_trans_block_group(trans, dir);
2480 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2482 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2483 dentry->d_name.name, dentry->d_name.len);
2485 if (inode->i_nlink == 0)
2486 ret = btrfs_orphan_add(trans, inode);
2488 nr = trans->blocks_used;
2490 btrfs_end_transaction_throttle(trans, root);
2491 btrfs_btree_balance_dirty(root, nr);
2495 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2496 struct btrfs_root *root,
2497 struct inode *dir, u64 objectid,
2498 const char *name, int name_len)
2500 struct btrfs_path *path;
2501 struct extent_buffer *leaf;
2502 struct btrfs_dir_item *di;
2503 struct btrfs_key key;
2507 path = btrfs_alloc_path();
2511 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2512 name, name_len, -1);
2513 BUG_ON(!di || IS_ERR(di));
2515 leaf = path->nodes[0];
2516 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2517 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2518 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2520 btrfs_release_path(root, path);
2522 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2523 objectid, root->root_key.objectid,
2524 dir->i_ino, &index, name, name_len);
2526 BUG_ON(ret != -ENOENT);
2527 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2529 BUG_ON(!di || IS_ERR(di));
2531 leaf = path->nodes[0];
2532 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2533 btrfs_release_path(root, path);
2537 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2538 index, name, name_len, -1);
2539 BUG_ON(!di || IS_ERR(di));
2541 leaf = path->nodes[0];
2542 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2543 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2544 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2546 btrfs_release_path(root, path);
2548 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2549 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2550 ret = btrfs_update_inode(trans, root, dir);
2552 dir->i_sb->s_dirt = 1;
2554 btrfs_free_path(path);
2558 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2560 struct inode *inode = dentry->d_inode;
2563 struct btrfs_root *root = BTRFS_I(dir)->root;
2564 struct btrfs_trans_handle *trans;
2565 unsigned long nr = 0;
2567 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2568 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2571 trans = btrfs_start_transaction(root, 1);
2572 btrfs_set_trans_block_group(trans, dir);
2574 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2575 err = btrfs_unlink_subvol(trans, root, dir,
2576 BTRFS_I(inode)->location.objectid,
2577 dentry->d_name.name,
2578 dentry->d_name.len);
2582 err = btrfs_orphan_add(trans, inode);
2586 /* now the directory is empty */
2587 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2588 dentry->d_name.name, dentry->d_name.len);
2590 btrfs_i_size_write(inode, 0);
2592 nr = trans->blocks_used;
2593 ret = btrfs_end_transaction_throttle(trans, root);
2594 btrfs_btree_balance_dirty(root, nr);
2603 * when truncating bytes in a file, it is possible to avoid reading
2604 * the leaves that contain only checksum items. This can be the
2605 * majority of the IO required to delete a large file, but it must
2606 * be done carefully.
2608 * The keys in the level just above the leaves are checked to make sure
2609 * the lowest key in a given leaf is a csum key, and starts at an offset
2610 * after the new size.
2612 * Then the key for the next leaf is checked to make sure it also has
2613 * a checksum item for the same file. If it does, we know our target leaf
2614 * contains only checksum items, and it can be safely freed without reading
2617 * This is just an optimization targeted at large files. It may do
2618 * nothing. It will return 0 unless things went badly.
2620 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2621 struct btrfs_root *root,
2622 struct btrfs_path *path,
2623 struct inode *inode, u64 new_size)
2625 struct btrfs_key key;
2628 struct btrfs_key found_key;
2629 struct btrfs_key other_key;
2630 struct btrfs_leaf_ref *ref;
2634 path->lowest_level = 1;
2635 key.objectid = inode->i_ino;
2636 key.type = BTRFS_CSUM_ITEM_KEY;
2637 key.offset = new_size;
2639 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2643 if (path->nodes[1] == NULL) {
2648 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2649 nritems = btrfs_header_nritems(path->nodes[1]);
2654 if (path->slots[1] >= nritems)
2657 /* did we find a key greater than anything we want to delete? */
2658 if (found_key.objectid > inode->i_ino ||
2659 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2662 /* we check the next key in the node to make sure the leave contains
2663 * only checksum items. This comparison doesn't work if our
2664 * leaf is the last one in the node
2666 if (path->slots[1] + 1 >= nritems) {
2668 /* search forward from the last key in the node, this
2669 * will bring us into the next node in the tree
2671 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2673 /* unlikely, but we inc below, so check to be safe */
2674 if (found_key.offset == (u64)-1)
2677 /* search_forward needs a path with locks held, do the
2678 * search again for the original key. It is possible
2679 * this will race with a balance and return a path that
2680 * we could modify, but this drop is just an optimization
2681 * and is allowed to miss some leaves.
2683 btrfs_release_path(root, path);
2686 /* setup a max key for search_forward */
2687 other_key.offset = (u64)-1;
2688 other_key.type = key.type;
2689 other_key.objectid = key.objectid;
2691 path->keep_locks = 1;
2692 ret = btrfs_search_forward(root, &found_key, &other_key,
2694 path->keep_locks = 0;
2695 if (ret || found_key.objectid != key.objectid ||
2696 found_key.type != key.type) {
2701 key.offset = found_key.offset;
2702 btrfs_release_path(root, path);
2707 /* we know there's one more slot after us in the tree,
2708 * read that key so we can verify it is also a checksum item
2710 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2712 if (found_key.objectid < inode->i_ino)
2715 if (found_key.type != key.type || found_key.offset < new_size)
2719 * if the key for the next leaf isn't a csum key from this objectid,
2720 * we can't be sure there aren't good items inside this leaf.
2723 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2726 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2727 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2729 * it is safe to delete this leaf, it contains only
2730 * csum items from this inode at an offset >= new_size
2732 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2735 if (root->ref_cows && leaf_gen < trans->transid) {
2736 ref = btrfs_alloc_leaf_ref(root, 0);
2738 ref->root_gen = root->root_key.offset;
2739 ref->bytenr = leaf_start;
2741 ref->generation = leaf_gen;
2744 btrfs_sort_leaf_ref(ref);
2746 ret = btrfs_add_leaf_ref(root, ref, 0);
2748 btrfs_free_leaf_ref(root, ref);
2754 btrfs_release_path(root, path);
2756 if (other_key.objectid == inode->i_ino &&
2757 other_key.type == key.type && other_key.offset > key.offset) {
2758 key.offset = other_key.offset;
2764 /* fixup any changes we've made to the path */
2765 path->lowest_level = 0;
2766 path->keep_locks = 0;
2767 btrfs_release_path(root, path);
2774 * this can truncate away extent items, csum items and directory items.
2775 * It starts at a high offset and removes keys until it can't find
2776 * any higher than new_size
2778 * csum items that cross the new i_size are truncated to the new size
2781 * min_type is the minimum key type to truncate down to. If set to 0, this
2782 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2784 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2785 struct btrfs_root *root,
2786 struct inode *inode,
2787 u64 new_size, u32 min_type)
2790 struct btrfs_path *path;
2791 struct btrfs_key key;
2792 struct btrfs_key found_key;
2793 u32 found_type = (u8)-1;
2794 struct extent_buffer *leaf;
2795 struct btrfs_file_extent_item *fi;
2796 u64 extent_start = 0;
2797 u64 extent_num_bytes = 0;
2798 u64 extent_offset = 0;
2802 int pending_del_nr = 0;
2803 int pending_del_slot = 0;
2804 int extent_type = -1;
2806 u64 mask = root->sectorsize - 1;
2809 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2810 path = btrfs_alloc_path();
2814 /* FIXME, add redo link to tree so we don't leak on crash */
2815 key.objectid = inode->i_ino;
2816 key.offset = (u64)-1;
2820 path->leave_spinning = 1;
2821 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2826 /* there are no items in the tree for us to truncate, we're
2829 if (path->slots[0] == 0) {
2838 leaf = path->nodes[0];
2839 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2840 found_type = btrfs_key_type(&found_key);
2843 if (found_key.objectid != inode->i_ino)
2846 if (found_type < min_type)
2849 item_end = found_key.offset;
2850 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2851 fi = btrfs_item_ptr(leaf, path->slots[0],
2852 struct btrfs_file_extent_item);
2853 extent_type = btrfs_file_extent_type(leaf, fi);
2854 encoding = btrfs_file_extent_compression(leaf, fi);
2855 encoding |= btrfs_file_extent_encryption(leaf, fi);
2856 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2858 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2860 btrfs_file_extent_num_bytes(leaf, fi);
2861 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2862 item_end += btrfs_file_extent_inline_len(leaf,
2867 if (item_end < new_size) {
2868 if (found_type == BTRFS_DIR_ITEM_KEY)
2869 found_type = BTRFS_INODE_ITEM_KEY;
2870 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2871 found_type = BTRFS_EXTENT_DATA_KEY;
2872 else if (found_type == BTRFS_EXTENT_DATA_KEY)
2873 found_type = BTRFS_XATTR_ITEM_KEY;
2874 else if (found_type == BTRFS_XATTR_ITEM_KEY)
2875 found_type = BTRFS_INODE_REF_KEY;
2876 else if (found_type)
2880 btrfs_set_key_type(&key, found_type);
2883 if (found_key.offset >= new_size)
2889 /* FIXME, shrink the extent if the ref count is only 1 */
2890 if (found_type != BTRFS_EXTENT_DATA_KEY)
2893 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2895 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2896 if (!del_item && !encoding) {
2897 u64 orig_num_bytes =
2898 btrfs_file_extent_num_bytes(leaf, fi);
2899 extent_num_bytes = new_size -
2900 found_key.offset + root->sectorsize - 1;
2901 extent_num_bytes = extent_num_bytes &
2902 ~((u64)root->sectorsize - 1);
2903 btrfs_set_file_extent_num_bytes(leaf, fi,
2905 num_dec = (orig_num_bytes -
2907 if (root->ref_cows && extent_start != 0)
2908 inode_sub_bytes(inode, num_dec);
2909 btrfs_mark_buffer_dirty(leaf);
2912 btrfs_file_extent_disk_num_bytes(leaf,
2914 extent_offset = found_key.offset -
2915 btrfs_file_extent_offset(leaf, fi);
2917 /* FIXME blocksize != 4096 */
2918 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2919 if (extent_start != 0) {
2922 inode_sub_bytes(inode, num_dec);
2925 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2927 * we can't truncate inline items that have had
2931 btrfs_file_extent_compression(leaf, fi) == 0 &&
2932 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2933 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2934 u32 size = new_size - found_key.offset;
2936 if (root->ref_cows) {
2937 inode_sub_bytes(inode, item_end + 1 -
2941 btrfs_file_extent_calc_inline_size(size);
2942 ret = btrfs_truncate_item(trans, root, path,
2945 } else if (root->ref_cows) {
2946 inode_sub_bytes(inode, item_end + 1 -
2952 if (!pending_del_nr) {
2953 /* no pending yet, add ourselves */
2954 pending_del_slot = path->slots[0];
2956 } else if (pending_del_nr &&
2957 path->slots[0] + 1 == pending_del_slot) {
2958 /* hop on the pending chunk */
2960 pending_del_slot = path->slots[0];
2967 if (found_extent && root->ref_cows) {
2968 btrfs_set_path_blocking(path);
2969 ret = btrfs_free_extent(trans, root, extent_start,
2970 extent_num_bytes, 0,
2971 btrfs_header_owner(leaf),
2972 inode->i_ino, extent_offset);
2976 if (path->slots[0] == 0) {
2979 btrfs_release_path(root, path);
2980 if (found_type == BTRFS_INODE_ITEM_KEY)
2986 if (pending_del_nr &&
2987 path->slots[0] + 1 != pending_del_slot) {
2988 struct btrfs_key debug;
2990 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2992 ret = btrfs_del_items(trans, root, path,
2997 btrfs_release_path(root, path);
2998 if (found_type == BTRFS_INODE_ITEM_KEY)
3005 if (pending_del_nr) {
3006 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3009 btrfs_free_path(path);
3014 * taken from block_truncate_page, but does cow as it zeros out
3015 * any bytes left in the last page in the file.
3017 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3019 struct inode *inode = mapping->host;
3020 struct btrfs_root *root = BTRFS_I(inode)->root;
3021 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3022 struct btrfs_ordered_extent *ordered;
3024 u32 blocksize = root->sectorsize;
3025 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3026 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3032 if ((offset & (blocksize - 1)) == 0)
3037 page = grab_cache_page(mapping, index);
3041 page_start = page_offset(page);
3042 page_end = page_start + PAGE_CACHE_SIZE - 1;
3044 if (!PageUptodate(page)) {
3045 ret = btrfs_readpage(NULL, page);
3047 if (page->mapping != mapping) {
3049 page_cache_release(page);
3052 if (!PageUptodate(page)) {
3057 wait_on_page_writeback(page);
3059 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3060 set_page_extent_mapped(page);
3062 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3064 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3066 page_cache_release(page);
3067 btrfs_start_ordered_extent(inode, ordered, 1);
3068 btrfs_put_ordered_extent(ordered);
3072 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3074 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3079 if (offset != PAGE_CACHE_SIZE) {
3081 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3082 flush_dcache_page(page);
3085 ClearPageChecked(page);
3086 set_page_dirty(page);
3087 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3091 page_cache_release(page);
3096 int btrfs_cont_expand(struct inode *inode, loff_t size)
3098 struct btrfs_trans_handle *trans;
3099 struct btrfs_root *root = BTRFS_I(inode)->root;
3100 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3101 struct extent_map *em;
3102 u64 mask = root->sectorsize - 1;
3103 u64 hole_start = (inode->i_size + mask) & ~mask;
3104 u64 block_end = (size + mask) & ~mask;
3110 if (size <= hole_start)
3113 btrfs_truncate_page(inode->i_mapping, inode->i_size);
3116 struct btrfs_ordered_extent *ordered;
3117 btrfs_wait_ordered_range(inode, hole_start,
3118 block_end - hole_start);
3119 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3120 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3123 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3124 btrfs_put_ordered_extent(ordered);
3127 trans = btrfs_start_transaction(root, 1);
3128 btrfs_set_trans_block_group(trans, inode);
3130 cur_offset = hole_start;
3132 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3133 block_end - cur_offset, 0);
3134 BUG_ON(IS_ERR(em) || !em);
3135 last_byte = min(extent_map_end(em), block_end);
3136 last_byte = (last_byte + mask) & ~mask;
3137 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
3139 hole_size = last_byte - cur_offset;
3140 err = btrfs_drop_extents(trans, root, inode,
3142 cur_offset + hole_size,
3144 cur_offset, &hint_byte, 1);
3148 err = btrfs_reserve_metadata_space(root, 1);
3152 err = btrfs_insert_file_extent(trans, root,
3153 inode->i_ino, cur_offset, 0,
3154 0, hole_size, 0, hole_size,
3156 btrfs_drop_extent_cache(inode, hole_start,
3158 btrfs_unreserve_metadata_space(root, 1);
3160 free_extent_map(em);
3161 cur_offset = last_byte;
3162 if (err || cur_offset >= block_end)
3166 btrfs_end_transaction(trans, root);
3167 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3171 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3173 struct inode *inode = dentry->d_inode;
3176 err = inode_change_ok(inode, attr);
3180 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3181 if (attr->ia_size > inode->i_size) {
3182 err = btrfs_cont_expand(inode, attr->ia_size);
3185 } else if (inode->i_size > 0 &&
3186 attr->ia_size == 0) {
3188 /* we're truncating a file that used to have good
3189 * data down to zero. Make sure it gets into
3190 * the ordered flush list so that any new writes
3191 * get down to disk quickly.
3193 BTRFS_I(inode)->ordered_data_close = 1;
3197 err = inode_setattr(inode, attr);
3199 if (!err && ((attr->ia_valid & ATTR_MODE)))
3200 err = btrfs_acl_chmod(inode);
3204 void btrfs_delete_inode(struct inode *inode)
3206 struct btrfs_trans_handle *trans;
3207 struct btrfs_root *root = BTRFS_I(inode)->root;
3211 truncate_inode_pages(&inode->i_data, 0);
3212 if (is_bad_inode(inode)) {
3213 btrfs_orphan_del(NULL, inode);
3216 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3218 if (inode->i_nlink > 0) {
3219 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3223 btrfs_i_size_write(inode, 0);
3224 trans = btrfs_join_transaction(root, 1);
3226 btrfs_set_trans_block_group(trans, inode);
3227 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3229 btrfs_orphan_del(NULL, inode);
3230 goto no_delete_lock;
3233 btrfs_orphan_del(trans, inode);
3235 nr = trans->blocks_used;
3238 btrfs_end_transaction(trans, root);
3239 btrfs_btree_balance_dirty(root, nr);
3243 nr = trans->blocks_used;
3244 btrfs_end_transaction(trans, root);
3245 btrfs_btree_balance_dirty(root, nr);
3251 * this returns the key found in the dir entry in the location pointer.
3252 * If no dir entries were found, location->objectid is 0.
3254 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3255 struct btrfs_key *location)
3257 const char *name = dentry->d_name.name;
3258 int namelen = dentry->d_name.len;
3259 struct btrfs_dir_item *di;
3260 struct btrfs_path *path;
3261 struct btrfs_root *root = BTRFS_I(dir)->root;
3264 path = btrfs_alloc_path();
3267 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3272 if (!di || IS_ERR(di))
3275 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3277 btrfs_free_path(path);
3280 location->objectid = 0;
3285 * when we hit a tree root in a directory, the btrfs part of the inode
3286 * needs to be changed to reflect the root directory of the tree root. This
3287 * is kind of like crossing a mount point.
3289 static int fixup_tree_root_location(struct btrfs_root *root,
3291 struct dentry *dentry,
3292 struct btrfs_key *location,
3293 struct btrfs_root **sub_root)
3295 struct btrfs_path *path;
3296 struct btrfs_root *new_root;
3297 struct btrfs_root_ref *ref;
3298 struct extent_buffer *leaf;
3302 path = btrfs_alloc_path();
3309 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3310 BTRFS_I(dir)->root->root_key.objectid,
3311 location->objectid);
3318 leaf = path->nodes[0];
3319 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3320 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3321 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3324 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3325 (unsigned long)(ref + 1),
3326 dentry->d_name.len);
3330 btrfs_release_path(root->fs_info->tree_root, path);
3332 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3333 if (IS_ERR(new_root)) {
3334 err = PTR_ERR(new_root);
3338 if (btrfs_root_refs(&new_root->root_item) == 0) {
3343 *sub_root = new_root;
3344 location->objectid = btrfs_root_dirid(&new_root->root_item);
3345 location->type = BTRFS_INODE_ITEM_KEY;
3346 location->offset = 0;
3349 btrfs_free_path(path);
3353 static void inode_tree_add(struct inode *inode)
3355 struct btrfs_root *root = BTRFS_I(inode)->root;
3356 struct btrfs_inode *entry;
3358 struct rb_node *parent;
3360 p = &root->inode_tree.rb_node;
3363 if (hlist_unhashed(&inode->i_hash))
3366 spin_lock(&root->inode_lock);
3369 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3371 if (inode->i_ino < entry->vfs_inode.i_ino)
3372 p = &parent->rb_left;
3373 else if (inode->i_ino > entry->vfs_inode.i_ino)
3374 p = &parent->rb_right;
3376 WARN_ON(!(entry->vfs_inode.i_state &
3377 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3378 rb_erase(parent, &root->inode_tree);
3379 RB_CLEAR_NODE(parent);
3380 spin_unlock(&root->inode_lock);
3384 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3385 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3386 spin_unlock(&root->inode_lock);
3389 static void inode_tree_del(struct inode *inode)
3391 struct btrfs_root *root = BTRFS_I(inode)->root;
3394 spin_lock(&root->inode_lock);
3395 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3396 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3397 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3398 empty = RB_EMPTY_ROOT(&root->inode_tree);
3400 spin_unlock(&root->inode_lock);
3402 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3403 synchronize_srcu(&root->fs_info->subvol_srcu);
3404 spin_lock(&root->inode_lock);
3405 empty = RB_EMPTY_ROOT(&root->inode_tree);
3406 spin_unlock(&root->inode_lock);
3408 btrfs_add_dead_root(root);
3412 int btrfs_invalidate_inodes(struct btrfs_root *root)
3414 struct rb_node *node;
3415 struct rb_node *prev;
3416 struct btrfs_inode *entry;
3417 struct inode *inode;
3420 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3422 spin_lock(&root->inode_lock);
3424 node = root->inode_tree.rb_node;
3428 entry = rb_entry(node, struct btrfs_inode, rb_node);
3430 if (objectid < entry->vfs_inode.i_ino)
3431 node = node->rb_left;
3432 else if (objectid > entry->vfs_inode.i_ino)
3433 node = node->rb_right;
3439 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3440 if (objectid <= entry->vfs_inode.i_ino) {
3444 prev = rb_next(prev);
3448 entry = rb_entry(node, struct btrfs_inode, rb_node);
3449 objectid = entry->vfs_inode.i_ino + 1;
3450 inode = igrab(&entry->vfs_inode);
3452 spin_unlock(&root->inode_lock);
3453 if (atomic_read(&inode->i_count) > 1)
3454 d_prune_aliases(inode);
3456 * btrfs_drop_inode will remove it from
3457 * the inode cache when its usage count
3462 spin_lock(&root->inode_lock);
3466 if (cond_resched_lock(&root->inode_lock))
3469 node = rb_next(node);
3471 spin_unlock(&root->inode_lock);
3475 static noinline void init_btrfs_i(struct inode *inode)
3477 struct btrfs_inode *bi = BTRFS_I(inode);
3482 bi->logged_trans = 0;
3483 bi->delalloc_bytes = 0;
3484 bi->reserved_bytes = 0;
3485 bi->disk_i_size = 0;
3487 bi->index_cnt = (u64)-1;
3488 bi->last_unlink_trans = 0;
3489 bi->ordered_data_close = 0;
3490 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3491 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3492 inode->i_mapping, GFP_NOFS);
3493 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3494 inode->i_mapping, GFP_NOFS);
3495 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3496 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3497 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3498 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3499 mutex_init(&BTRFS_I(inode)->extent_mutex);
3500 mutex_init(&BTRFS_I(inode)->log_mutex);
3503 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3505 struct btrfs_iget_args *args = p;
3506 inode->i_ino = args->ino;
3507 init_btrfs_i(inode);
3508 BTRFS_I(inode)->root = args->root;
3509 btrfs_set_inode_space_info(args->root, inode);
3513 static int btrfs_find_actor(struct inode *inode, void *opaque)
3515 struct btrfs_iget_args *args = opaque;
3516 return args->ino == inode->i_ino &&
3517 args->root == BTRFS_I(inode)->root;
3520 static struct inode *btrfs_iget_locked(struct super_block *s,
3522 struct btrfs_root *root)
3524 struct inode *inode;
3525 struct btrfs_iget_args args;
3526 args.ino = objectid;
3529 inode = iget5_locked(s, objectid, btrfs_find_actor,
3530 btrfs_init_locked_inode,
3535 /* Get an inode object given its location and corresponding root.
3536 * Returns in *is_new if the inode was read from disk
3538 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3539 struct btrfs_root *root)
3541 struct inode *inode;
3543 inode = btrfs_iget_locked(s, location->objectid, root);
3545 return ERR_PTR(-ENOMEM);
3547 if (inode->i_state & I_NEW) {
3548 BTRFS_I(inode)->root = root;
3549 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3550 btrfs_read_locked_inode(inode);
3552 inode_tree_add(inode);
3553 unlock_new_inode(inode);
3559 static struct inode *new_simple_dir(struct super_block *s,
3560 struct btrfs_key *key,
3561 struct btrfs_root *root)
3563 struct inode *inode = new_inode(s);
3566 return ERR_PTR(-ENOMEM);
3568 init_btrfs_i(inode);
3570 BTRFS_I(inode)->root = root;
3571 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3572 BTRFS_I(inode)->dummy_inode = 1;
3574 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3575 inode->i_op = &simple_dir_inode_operations;
3576 inode->i_fop = &simple_dir_operations;
3577 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3578 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3583 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3585 struct inode *inode;
3586 struct btrfs_root *root = BTRFS_I(dir)->root;
3587 struct btrfs_root *sub_root = root;
3588 struct btrfs_key location;
3592 dentry->d_op = &btrfs_dentry_operations;
3594 if (dentry->d_name.len > BTRFS_NAME_LEN)
3595 return ERR_PTR(-ENAMETOOLONG);
3597 ret = btrfs_inode_by_name(dir, dentry, &location);
3600 return ERR_PTR(ret);
3602 if (location.objectid == 0)
3605 if (location.type == BTRFS_INODE_ITEM_KEY) {
3606 inode = btrfs_iget(dir->i_sb, &location, root);
3610 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3612 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3613 ret = fixup_tree_root_location(root, dir, dentry,
3614 &location, &sub_root);
3617 inode = ERR_PTR(ret);
3619 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3621 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3623 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3628 static int btrfs_dentry_delete(struct dentry *dentry)
3630 struct btrfs_root *root;
3632 if (!dentry->d_inode && !IS_ROOT(dentry))
3633 dentry = dentry->d_parent;
3635 if (dentry->d_inode) {
3636 root = BTRFS_I(dentry->d_inode)->root;
3637 if (btrfs_root_refs(&root->root_item) == 0)
3643 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3644 struct nameidata *nd)
3646 struct inode *inode;
3648 inode = btrfs_lookup_dentry(dir, dentry);
3650 return ERR_CAST(inode);
3652 return d_splice_alias(inode, dentry);
3655 static unsigned char btrfs_filetype_table[] = {
3656 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3659 static int btrfs_real_readdir(struct file *filp, void *dirent,
3662 struct inode *inode = filp->f_dentry->d_inode;
3663 struct btrfs_root *root = BTRFS_I(inode)->root;
3664 struct btrfs_item *item;
3665 struct btrfs_dir_item *di;
3666 struct btrfs_key key;
3667 struct btrfs_key found_key;
3668 struct btrfs_path *path;
3671 struct extent_buffer *leaf;
3674 unsigned char d_type;
3679 int key_type = BTRFS_DIR_INDEX_KEY;
3684 /* FIXME, use a real flag for deciding about the key type */
3685 if (root->fs_info->tree_root == root)
3686 key_type = BTRFS_DIR_ITEM_KEY;
3688 /* special case for "." */
3689 if (filp->f_pos == 0) {
3690 over = filldir(dirent, ".", 1,
3697 /* special case for .., just use the back ref */
3698 if (filp->f_pos == 1) {
3699 u64 pino = parent_ino(filp->f_path.dentry);
3700 over = filldir(dirent, "..", 2,
3706 path = btrfs_alloc_path();
3709 btrfs_set_key_type(&key, key_type);
3710 key.offset = filp->f_pos;
3711 key.objectid = inode->i_ino;
3713 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3719 leaf = path->nodes[0];
3720 nritems = btrfs_header_nritems(leaf);
3721 slot = path->slots[0];
3722 if (advance || slot >= nritems) {
3723 if (slot >= nritems - 1) {
3724 ret = btrfs_next_leaf(root, path);
3727 leaf = path->nodes[0];
3728 nritems = btrfs_header_nritems(leaf);
3729 slot = path->slots[0];
3737 item = btrfs_item_nr(leaf, slot);
3738 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3740 if (found_key.objectid != key.objectid)
3742 if (btrfs_key_type(&found_key) != key_type)
3744 if (found_key.offset < filp->f_pos)
3747 filp->f_pos = found_key.offset;
3749 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3751 di_total = btrfs_item_size(leaf, item);
3753 while (di_cur < di_total) {
3754 struct btrfs_key location;
3756 name_len = btrfs_dir_name_len(leaf, di);
3757 if (name_len <= sizeof(tmp_name)) {
3758 name_ptr = tmp_name;
3760 name_ptr = kmalloc(name_len, GFP_NOFS);
3766 read_extent_buffer(leaf, name_ptr,
3767 (unsigned long)(di + 1), name_len);
3769 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3770 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3772 /* is this a reference to our own snapshot? If so
3775 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3776 location.objectid == root->root_key.objectid) {
3780 over = filldir(dirent, name_ptr, name_len,
3781 found_key.offset, location.objectid,
3785 if (name_ptr != tmp_name)
3790 di_len = btrfs_dir_name_len(leaf, di) +
3791 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3793 di = (struct btrfs_dir_item *)((char *)di + di_len);
3797 /* Reached end of directory/root. Bump pos past the last item. */
3798 if (key_type == BTRFS_DIR_INDEX_KEY)
3799 filp->f_pos = INT_LIMIT(off_t);
3805 btrfs_free_path(path);
3809 int btrfs_write_inode(struct inode *inode, int wait)
3811 struct btrfs_root *root = BTRFS_I(inode)->root;
3812 struct btrfs_trans_handle *trans;
3815 if (root->fs_info->btree_inode == inode)
3819 trans = btrfs_join_transaction(root, 1);
3820 btrfs_set_trans_block_group(trans, inode);
3821 ret = btrfs_commit_transaction(trans, root);
3827 * This is somewhat expensive, updating the tree every time the
3828 * inode changes. But, it is most likely to find the inode in cache.
3829 * FIXME, needs more benchmarking...there are no reasons other than performance
3830 * to keep or drop this code.
3832 void btrfs_dirty_inode(struct inode *inode)
3834 struct btrfs_root *root = BTRFS_I(inode)->root;
3835 struct btrfs_trans_handle *trans;
3837 trans = btrfs_join_transaction(root, 1);
3838 btrfs_set_trans_block_group(trans, inode);
3839 btrfs_update_inode(trans, root, inode);
3840 btrfs_end_transaction(trans, root);
3844 * find the highest existing sequence number in a directory
3845 * and then set the in-memory index_cnt variable to reflect
3846 * free sequence numbers
3848 static int btrfs_set_inode_index_count(struct inode *inode)
3850 struct btrfs_root *root = BTRFS_I(inode)->root;
3851 struct btrfs_key key, found_key;
3852 struct btrfs_path *path;
3853 struct extent_buffer *leaf;
3856 key.objectid = inode->i_ino;
3857 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3858 key.offset = (u64)-1;
3860 path = btrfs_alloc_path();
3864 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3867 /* FIXME: we should be able to handle this */
3873 * MAGIC NUMBER EXPLANATION:
3874 * since we search a directory based on f_pos we have to start at 2
3875 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3876 * else has to start at 2
3878 if (path->slots[0] == 0) {
3879 BTRFS_I(inode)->index_cnt = 2;
3885 leaf = path->nodes[0];
3886 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3888 if (found_key.objectid != inode->i_ino ||
3889 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3890 BTRFS_I(inode)->index_cnt = 2;
3894 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3896 btrfs_free_path(path);
3901 * helper to find a free sequence number in a given directory. This current
3902 * code is very simple, later versions will do smarter things in the btree
3904 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3908 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3909 ret = btrfs_set_inode_index_count(dir);
3914 *index = BTRFS_I(dir)->index_cnt;
3915 BTRFS_I(dir)->index_cnt++;
3920 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3921 struct btrfs_root *root,
3923 const char *name, int name_len,
3924 u64 ref_objectid, u64 objectid,
3925 u64 alloc_hint, int mode, u64 *index)
3927 struct inode *inode;
3928 struct btrfs_inode_item *inode_item;
3929 struct btrfs_key *location;
3930 struct btrfs_path *path;
3931 struct btrfs_inode_ref *ref;
3932 struct btrfs_key key[2];
3938 path = btrfs_alloc_path();
3941 inode = new_inode(root->fs_info->sb);
3943 return ERR_PTR(-ENOMEM);
3946 ret = btrfs_set_inode_index(dir, index);
3949 return ERR_PTR(ret);
3953 * index_cnt is ignored for everything but a dir,
3954 * btrfs_get_inode_index_count has an explanation for the magic
3957 init_btrfs_i(inode);
3958 BTRFS_I(inode)->index_cnt = 2;
3959 BTRFS_I(inode)->root = root;
3960 BTRFS_I(inode)->generation = trans->transid;
3961 btrfs_set_inode_space_info(root, inode);
3967 BTRFS_I(inode)->block_group =
3968 btrfs_find_block_group(root, 0, alloc_hint, owner);
3970 key[0].objectid = objectid;
3971 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3974 key[1].objectid = objectid;
3975 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3976 key[1].offset = ref_objectid;
3978 sizes[0] = sizeof(struct btrfs_inode_item);
3979 sizes[1] = name_len + sizeof(*ref);
3981 path->leave_spinning = 1;
3982 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3986 inode->i_uid = current_fsuid();
3988 if (dir && (dir->i_mode & S_ISGID)) {
3989 inode->i_gid = dir->i_gid;
3993 inode->i_gid = current_fsgid();
3995 inode->i_mode = mode;
3996 inode->i_ino = objectid;
3997 inode_set_bytes(inode, 0);
3998 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3999 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4000 struct btrfs_inode_item);
4001 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4003 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4004 struct btrfs_inode_ref);
4005 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4006 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4007 ptr = (unsigned long)(ref + 1);
4008 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4010 btrfs_mark_buffer_dirty(path->nodes[0]);
4011 btrfs_free_path(path);
4013 location = &BTRFS_I(inode)->location;
4014 location->objectid = objectid;
4015 location->offset = 0;
4016 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4018 btrfs_inherit_iflags(inode, dir);
4020 if ((mode & S_IFREG)) {
4021 if (btrfs_test_opt(root, NODATASUM))
4022 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4023 if (btrfs_test_opt(root, NODATACOW))
4024 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4027 insert_inode_hash(inode);
4028 inode_tree_add(inode);
4032 BTRFS_I(dir)->index_cnt--;
4033 btrfs_free_path(path);
4035 return ERR_PTR(ret);
4038 static inline u8 btrfs_inode_type(struct inode *inode)
4040 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4044 * utility function to add 'inode' into 'parent_inode' with
4045 * a give name and a given sequence number.
4046 * if 'add_backref' is true, also insert a backref from the
4047 * inode to the parent directory.
4049 int btrfs_add_link(struct btrfs_trans_handle *trans,
4050 struct inode *parent_inode, struct inode *inode,
4051 const char *name, int name_len, int add_backref, u64 index)
4054 struct btrfs_key key;
4055 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4057 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4058 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4060 key.objectid = inode->i_ino;
4061 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4065 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4066 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4067 key.objectid, root->root_key.objectid,
4068 parent_inode->i_ino,
4069 index, name, name_len);
4070 } else if (add_backref) {
4071 ret = btrfs_insert_inode_ref(trans, root,
4072 name, name_len, inode->i_ino,
4073 parent_inode->i_ino, index);
4077 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4078 parent_inode->i_ino, &key,
4079 btrfs_inode_type(inode), index);
4082 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4084 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4085 ret = btrfs_update_inode(trans, root, parent_inode);
4090 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4091 struct dentry *dentry, struct inode *inode,
4092 int backref, u64 index)
4094 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4095 inode, dentry->d_name.name,
4096 dentry->d_name.len, backref, index);
4098 d_instantiate(dentry, inode);
4106 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4107 int mode, dev_t rdev)
4109 struct btrfs_trans_handle *trans;
4110 struct btrfs_root *root = BTRFS_I(dir)->root;
4111 struct inode *inode = NULL;
4115 unsigned long nr = 0;
4118 if (!new_valid_dev(rdev))
4122 * 2 for inode item and ref
4124 * 1 for xattr if selinux is on
4126 err = btrfs_reserve_metadata_space(root, 5);
4130 trans = btrfs_start_transaction(root, 1);
4133 btrfs_set_trans_block_group(trans, dir);
4135 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4141 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4143 dentry->d_parent->d_inode->i_ino, objectid,
4144 BTRFS_I(dir)->block_group, mode, &index);
4145 err = PTR_ERR(inode);
4149 err = btrfs_init_inode_security(inode, dir);
4155 btrfs_set_trans_block_group(trans, inode);
4156 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4160 inode->i_op = &btrfs_special_inode_operations;
4161 init_special_inode(inode, inode->i_mode, rdev);
4162 btrfs_update_inode(trans, root, inode);
4164 btrfs_update_inode_block_group(trans, inode);
4165 btrfs_update_inode_block_group(trans, dir);
4167 nr = trans->blocks_used;
4168 btrfs_end_transaction_throttle(trans, root);
4170 btrfs_unreserve_metadata_space(root, 5);
4172 inode_dec_link_count(inode);
4175 btrfs_btree_balance_dirty(root, nr);
4179 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4180 int mode, struct nameidata *nd)
4182 struct btrfs_trans_handle *trans;
4183 struct btrfs_root *root = BTRFS_I(dir)->root;
4184 struct inode *inode = NULL;
4187 unsigned long nr = 0;
4192 * 2 for inode item and ref
4194 * 1 for xattr if selinux is on
4196 err = btrfs_reserve_metadata_space(root, 5);
4200 trans = btrfs_start_transaction(root, 1);
4203 btrfs_set_trans_block_group(trans, dir);
4205 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4211 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4213 dentry->d_parent->d_inode->i_ino,
4214 objectid, BTRFS_I(dir)->block_group, mode,
4216 err = PTR_ERR(inode);
4220 err = btrfs_init_inode_security(inode, dir);
4226 btrfs_set_trans_block_group(trans, inode);
4227 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4231 inode->i_mapping->a_ops = &btrfs_aops;
4232 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4233 inode->i_fop = &btrfs_file_operations;
4234 inode->i_op = &btrfs_file_inode_operations;
4235 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4237 btrfs_update_inode_block_group(trans, inode);
4238 btrfs_update_inode_block_group(trans, dir);
4240 nr = trans->blocks_used;
4241 btrfs_end_transaction_throttle(trans, root);
4243 btrfs_unreserve_metadata_space(root, 5);
4245 inode_dec_link_count(inode);
4248 btrfs_btree_balance_dirty(root, nr);
4252 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4253 struct dentry *dentry)
4255 struct btrfs_trans_handle *trans;
4256 struct btrfs_root *root = BTRFS_I(dir)->root;
4257 struct inode *inode = old_dentry->d_inode;
4259 unsigned long nr = 0;
4263 if (inode->i_nlink == 0)
4267 * 1 item for inode ref
4268 * 2 items for dir items
4270 err = btrfs_reserve_metadata_space(root, 3);
4274 btrfs_inc_nlink(inode);
4276 err = btrfs_set_inode_index(dir, &index);
4280 trans = btrfs_start_transaction(root, 1);
4282 btrfs_set_trans_block_group(trans, dir);
4283 atomic_inc(&inode->i_count);
4285 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4290 btrfs_update_inode_block_group(trans, dir);
4291 err = btrfs_update_inode(trans, root, inode);
4293 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4296 nr = trans->blocks_used;
4297 btrfs_end_transaction_throttle(trans, root);
4299 btrfs_unreserve_metadata_space(root, 3);
4301 inode_dec_link_count(inode);
4304 btrfs_btree_balance_dirty(root, nr);
4308 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4310 struct inode *inode = NULL;
4311 struct btrfs_trans_handle *trans;
4312 struct btrfs_root *root = BTRFS_I(dir)->root;
4314 int drop_on_err = 0;
4317 unsigned long nr = 1;
4320 * 2 items for inode and ref
4321 * 2 items for dir items
4322 * 1 for xattr if selinux is on
4324 err = btrfs_reserve_metadata_space(root, 5);
4328 trans = btrfs_start_transaction(root, 1);
4333 btrfs_set_trans_block_group(trans, dir);
4335 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4341 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4343 dentry->d_parent->d_inode->i_ino, objectid,
4344 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4346 if (IS_ERR(inode)) {
4347 err = PTR_ERR(inode);
4353 err = btrfs_init_inode_security(inode, dir);
4357 inode->i_op = &btrfs_dir_inode_operations;
4358 inode->i_fop = &btrfs_dir_file_operations;
4359 btrfs_set_trans_block_group(trans, inode);
4361 btrfs_i_size_write(inode, 0);
4362 err = btrfs_update_inode(trans, root, inode);
4366 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4367 inode, dentry->d_name.name,
4368 dentry->d_name.len, 0, index);
4372 d_instantiate(dentry, inode);
4374 btrfs_update_inode_block_group(trans, inode);
4375 btrfs_update_inode_block_group(trans, dir);
4378 nr = trans->blocks_used;
4379 btrfs_end_transaction_throttle(trans, root);
4382 btrfs_unreserve_metadata_space(root, 5);
4385 btrfs_btree_balance_dirty(root, nr);
4389 /* helper for btfs_get_extent. Given an existing extent in the tree,
4390 * and an extent that you want to insert, deal with overlap and insert
4391 * the new extent into the tree.
4393 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4394 struct extent_map *existing,
4395 struct extent_map *em,
4396 u64 map_start, u64 map_len)
4400 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4401 start_diff = map_start - em->start;
4402 em->start = map_start;
4404 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4405 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4406 em->block_start += start_diff;
4407 em->block_len -= start_diff;
4409 return add_extent_mapping(em_tree, em);
4412 static noinline int uncompress_inline(struct btrfs_path *path,
4413 struct inode *inode, struct page *page,
4414 size_t pg_offset, u64 extent_offset,
4415 struct btrfs_file_extent_item *item)
4418 struct extent_buffer *leaf = path->nodes[0];
4421 unsigned long inline_size;
4424 WARN_ON(pg_offset != 0);
4425 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4426 inline_size = btrfs_file_extent_inline_item_len(leaf,
4427 btrfs_item_nr(leaf, path->slots[0]));
4428 tmp = kmalloc(inline_size, GFP_NOFS);
4429 ptr = btrfs_file_extent_inline_start(item);
4431 read_extent_buffer(leaf, tmp, ptr, inline_size);
4433 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4434 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4435 inline_size, max_size);
4437 char *kaddr = kmap_atomic(page, KM_USER0);
4438 unsigned long copy_size = min_t(u64,
4439 PAGE_CACHE_SIZE - pg_offset,
4440 max_size - extent_offset);
4441 memset(kaddr + pg_offset, 0, copy_size);
4442 kunmap_atomic(kaddr, KM_USER0);
4449 * a bit scary, this does extent mapping from logical file offset to the disk.
4450 * the ugly parts come from merging extents from the disk with the in-ram
4451 * representation. This gets more complex because of the data=ordered code,
4452 * where the in-ram extents might be locked pending data=ordered completion.
4454 * This also copies inline extents directly into the page.
4457 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4458 size_t pg_offset, u64 start, u64 len,
4464 u64 extent_start = 0;
4466 u64 objectid = inode->i_ino;
4468 struct btrfs_path *path = NULL;
4469 struct btrfs_root *root = BTRFS_I(inode)->root;
4470 struct btrfs_file_extent_item *item;
4471 struct extent_buffer *leaf;
4472 struct btrfs_key found_key;
4473 struct extent_map *em = NULL;
4474 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4475 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4476 struct btrfs_trans_handle *trans = NULL;
4480 read_lock(&em_tree->lock);
4481 em = lookup_extent_mapping(em_tree, start, len);
4483 em->bdev = root->fs_info->fs_devices->latest_bdev;
4484 read_unlock(&em_tree->lock);
4487 if (em->start > start || em->start + em->len <= start)
4488 free_extent_map(em);
4489 else if (em->block_start == EXTENT_MAP_INLINE && page)
4490 free_extent_map(em);
4494 em = alloc_extent_map(GFP_NOFS);
4499 em->bdev = root->fs_info->fs_devices->latest_bdev;
4500 em->start = EXTENT_MAP_HOLE;
4501 em->orig_start = EXTENT_MAP_HOLE;
4503 em->block_len = (u64)-1;
4506 path = btrfs_alloc_path();
4510 ret = btrfs_lookup_file_extent(trans, root, path,
4511 objectid, start, trans != NULL);
4518 if (path->slots[0] == 0)
4523 leaf = path->nodes[0];
4524 item = btrfs_item_ptr(leaf, path->slots[0],
4525 struct btrfs_file_extent_item);
4526 /* are we inside the extent that was found? */
4527 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4528 found_type = btrfs_key_type(&found_key);
4529 if (found_key.objectid != objectid ||
4530 found_type != BTRFS_EXTENT_DATA_KEY) {
4534 found_type = btrfs_file_extent_type(leaf, item);
4535 extent_start = found_key.offset;
4536 compressed = btrfs_file_extent_compression(leaf, item);
4537 if (found_type == BTRFS_FILE_EXTENT_REG ||
4538 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4539 extent_end = extent_start +
4540 btrfs_file_extent_num_bytes(leaf, item);
4541 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4543 size = btrfs_file_extent_inline_len(leaf, item);
4544 extent_end = (extent_start + size + root->sectorsize - 1) &
4545 ~((u64)root->sectorsize - 1);
4548 if (start >= extent_end) {
4550 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4551 ret = btrfs_next_leaf(root, path);
4558 leaf = path->nodes[0];
4560 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4561 if (found_key.objectid != objectid ||
4562 found_key.type != BTRFS_EXTENT_DATA_KEY)
4564 if (start + len <= found_key.offset)
4567 em->len = found_key.offset - start;
4571 if (found_type == BTRFS_FILE_EXTENT_REG ||
4572 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4573 em->start = extent_start;
4574 em->len = extent_end - extent_start;
4575 em->orig_start = extent_start -
4576 btrfs_file_extent_offset(leaf, item);
4577 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4579 em->block_start = EXTENT_MAP_HOLE;
4583 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4584 em->block_start = bytenr;
4585 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4588 bytenr += btrfs_file_extent_offset(leaf, item);
4589 em->block_start = bytenr;
4590 em->block_len = em->len;
4591 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4592 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4595 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4599 size_t extent_offset;
4602 em->block_start = EXTENT_MAP_INLINE;
4603 if (!page || create) {
4604 em->start = extent_start;
4605 em->len = extent_end - extent_start;
4609 size = btrfs_file_extent_inline_len(leaf, item);
4610 extent_offset = page_offset(page) + pg_offset - extent_start;
4611 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4612 size - extent_offset);
4613 em->start = extent_start + extent_offset;
4614 em->len = (copy_size + root->sectorsize - 1) &
4615 ~((u64)root->sectorsize - 1);
4616 em->orig_start = EXTENT_MAP_INLINE;
4618 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4619 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4620 if (create == 0 && !PageUptodate(page)) {
4621 if (btrfs_file_extent_compression(leaf, item) ==
4622 BTRFS_COMPRESS_ZLIB) {
4623 ret = uncompress_inline(path, inode, page,
4625 extent_offset, item);
4629 read_extent_buffer(leaf, map + pg_offset, ptr,
4631 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4632 memset(map + pg_offset + copy_size, 0,
4633 PAGE_CACHE_SIZE - pg_offset -
4638 flush_dcache_page(page);
4639 } else if (create && PageUptodate(page)) {
4642 free_extent_map(em);
4644 btrfs_release_path(root, path);
4645 trans = btrfs_join_transaction(root, 1);
4649 write_extent_buffer(leaf, map + pg_offset, ptr,
4652 btrfs_mark_buffer_dirty(leaf);
4654 set_extent_uptodate(io_tree, em->start,
4655 extent_map_end(em) - 1, GFP_NOFS);
4658 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4665 em->block_start = EXTENT_MAP_HOLE;
4666 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4668 btrfs_release_path(root, path);
4669 if (em->start > start || extent_map_end(em) <= start) {
4670 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4671 "[%llu %llu]\n", (unsigned long long)em->start,
4672 (unsigned long long)em->len,
4673 (unsigned long long)start,
4674 (unsigned long long)len);
4680 write_lock(&em_tree->lock);
4681 ret = add_extent_mapping(em_tree, em);
4682 /* it is possible that someone inserted the extent into the tree
4683 * while we had the lock dropped. It is also possible that
4684 * an overlapping map exists in the tree
4686 if (ret == -EEXIST) {
4687 struct extent_map *existing;
4691 existing = lookup_extent_mapping(em_tree, start, len);
4692 if (existing && (existing->start > start ||
4693 existing->start + existing->len <= start)) {
4694 free_extent_map(existing);
4698 existing = lookup_extent_mapping(em_tree, em->start,
4701 err = merge_extent_mapping(em_tree, existing,
4704 free_extent_map(existing);
4706 free_extent_map(em);
4711 free_extent_map(em);
4715 free_extent_map(em);
4720 write_unlock(&em_tree->lock);
4723 btrfs_free_path(path);
4725 ret = btrfs_end_transaction(trans, root);
4730 free_extent_map(em);
4731 return ERR_PTR(err);
4736 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4737 const struct iovec *iov, loff_t offset,
4738 unsigned long nr_segs)
4743 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4744 __u64 start, __u64 len)
4746 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4749 int btrfs_readpage(struct file *file, struct page *page)
4751 struct extent_io_tree *tree;
4752 tree = &BTRFS_I(page->mapping->host)->io_tree;
4753 return extent_read_full_page(tree, page, btrfs_get_extent);
4756 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4758 struct extent_io_tree *tree;
4761 if (current->flags & PF_MEMALLOC) {
4762 redirty_page_for_writepage(wbc, page);
4766 tree = &BTRFS_I(page->mapping->host)->io_tree;
4767 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4770 int btrfs_writepages(struct address_space *mapping,
4771 struct writeback_control *wbc)
4773 struct extent_io_tree *tree;
4775 tree = &BTRFS_I(mapping->host)->io_tree;
4776 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4780 btrfs_readpages(struct file *file, struct address_space *mapping,
4781 struct list_head *pages, unsigned nr_pages)
4783 struct extent_io_tree *tree;
4784 tree = &BTRFS_I(mapping->host)->io_tree;
4785 return extent_readpages(tree, mapping, pages, nr_pages,
4788 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4790 struct extent_io_tree *tree;
4791 struct extent_map_tree *map;
4794 tree = &BTRFS_I(page->mapping->host)->io_tree;
4795 map = &BTRFS_I(page->mapping->host)->extent_tree;
4796 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4798 ClearPagePrivate(page);
4799 set_page_private(page, 0);
4800 page_cache_release(page);
4805 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4807 if (PageWriteback(page) || PageDirty(page))
4809 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4812 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4814 struct extent_io_tree *tree;
4815 struct btrfs_ordered_extent *ordered;
4816 u64 page_start = page_offset(page);
4817 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4821 * we have the page locked, so new writeback can't start,
4822 * and the dirty bit won't be cleared while we are here.
4824 * Wait for IO on this page so that we can safely clear
4825 * the PagePrivate2 bit and do ordered accounting
4827 wait_on_page_writeback(page);
4829 tree = &BTRFS_I(page->mapping->host)->io_tree;
4831 btrfs_releasepage(page, GFP_NOFS);
4834 lock_extent(tree, page_start, page_end, GFP_NOFS);
4835 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4839 * IO on this page will never be started, so we need
4840 * to account for any ordered extents now
4842 clear_extent_bit(tree, page_start, page_end,
4843 EXTENT_DIRTY | EXTENT_DELALLOC |
4844 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
4847 * whoever cleared the private bit is responsible
4848 * for the finish_ordered_io
4850 if (TestClearPagePrivate2(page)) {
4851 btrfs_finish_ordered_io(page->mapping->host,
4852 page_start, page_end);
4854 btrfs_put_ordered_extent(ordered);
4855 lock_extent(tree, page_start, page_end, GFP_NOFS);
4857 clear_extent_bit(tree, page_start, page_end,
4858 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4859 EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
4860 __btrfs_releasepage(page, GFP_NOFS);
4862 ClearPageChecked(page);
4863 if (PagePrivate(page)) {
4864 ClearPagePrivate(page);
4865 set_page_private(page, 0);
4866 page_cache_release(page);
4871 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4872 * called from a page fault handler when a page is first dirtied. Hence we must
4873 * be careful to check for EOF conditions here. We set the page up correctly
4874 * for a written page which means we get ENOSPC checking when writing into
4875 * holes and correct delalloc and unwritten extent mapping on filesystems that
4876 * support these features.
4878 * We are not allowed to take the i_mutex here so we have to play games to
4879 * protect against truncate races as the page could now be beyond EOF. Because
4880 * vmtruncate() writes the inode size before removing pages, once we have the
4881 * page lock we can determine safely if the page is beyond EOF. If it is not
4882 * beyond EOF, then the page is guaranteed safe against truncation until we
4885 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4887 struct page *page = vmf->page;
4888 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4889 struct btrfs_root *root = BTRFS_I(inode)->root;
4890 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4891 struct btrfs_ordered_extent *ordered;
4893 unsigned long zero_start;
4899 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4903 else /* -ENOSPC, -EIO, etc */
4904 ret = VM_FAULT_SIGBUS;
4908 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
4910 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4911 ret = VM_FAULT_SIGBUS;
4915 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4918 size = i_size_read(inode);
4919 page_start = page_offset(page);
4920 page_end = page_start + PAGE_CACHE_SIZE - 1;
4922 if ((page->mapping != inode->i_mapping) ||
4923 (page_start >= size)) {
4924 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4925 /* page got truncated out from underneath us */
4928 wait_on_page_writeback(page);
4930 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4931 set_page_extent_mapped(page);
4934 * we can't set the delalloc bits if there are pending ordered
4935 * extents. Drop our locks and wait for them to finish
4937 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4939 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4941 btrfs_start_ordered_extent(inode, ordered, 1);
4942 btrfs_put_ordered_extent(ordered);
4947 * XXX - page_mkwrite gets called every time the page is dirtied, even
4948 * if it was already dirty, so for space accounting reasons we need to
4949 * clear any delalloc bits for the range we are fixing to save. There
4950 * is probably a better way to do this, but for now keep consistent with
4951 * prepare_pages in the normal write path.
4953 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
4954 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
4957 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
4959 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4960 ret = VM_FAULT_SIGBUS;
4961 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4966 /* page is wholly or partially inside EOF */
4967 if (page_start + PAGE_CACHE_SIZE > size)
4968 zero_start = size & ~PAGE_CACHE_MASK;
4970 zero_start = PAGE_CACHE_SIZE;
4972 if (zero_start != PAGE_CACHE_SIZE) {
4974 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4975 flush_dcache_page(page);
4978 ClearPageChecked(page);
4979 set_page_dirty(page);
4980 SetPageUptodate(page);
4982 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4983 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4986 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
4988 return VM_FAULT_LOCKED;
4994 static void btrfs_truncate(struct inode *inode)
4996 struct btrfs_root *root = BTRFS_I(inode)->root;
4998 struct btrfs_trans_handle *trans;
5000 u64 mask = root->sectorsize - 1;
5002 if (!S_ISREG(inode->i_mode))
5004 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
5007 btrfs_truncate_page(inode->i_mapping, inode->i_size);
5008 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5010 trans = btrfs_start_transaction(root, 1);
5013 * setattr is responsible for setting the ordered_data_close flag,
5014 * but that is only tested during the last file release. That
5015 * could happen well after the next commit, leaving a great big
5016 * window where new writes may get lost if someone chooses to write
5017 * to this file after truncating to zero
5019 * The inode doesn't have any dirty data here, and so if we commit
5020 * this is a noop. If someone immediately starts writing to the inode
5021 * it is very likely we'll catch some of their writes in this
5022 * transaction, and the commit will find this file on the ordered
5023 * data list with good things to send down.
5025 * This is a best effort solution, there is still a window where
5026 * using truncate to replace the contents of the file will
5027 * end up with a zero length file after a crash.
5029 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5030 btrfs_add_ordered_operation(trans, root, inode);
5032 btrfs_set_trans_block_group(trans, inode);
5033 btrfs_i_size_write(inode, inode->i_size);
5035 ret = btrfs_orphan_add(trans, inode);
5038 /* FIXME, add redo link to tree so we don't leak on crash */
5039 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
5040 BTRFS_EXTENT_DATA_KEY);
5041 btrfs_update_inode(trans, root, inode);
5043 ret = btrfs_orphan_del(trans, inode);
5047 nr = trans->blocks_used;
5048 ret = btrfs_end_transaction_throttle(trans, root);
5050 btrfs_btree_balance_dirty(root, nr);
5054 * create a new subvolume directory/inode (helper for the ioctl).
5056 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5057 struct btrfs_root *new_root,
5058 u64 new_dirid, u64 alloc_hint)
5060 struct inode *inode;
5064 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5065 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5067 return PTR_ERR(inode);
5068 inode->i_op = &btrfs_dir_inode_operations;
5069 inode->i_fop = &btrfs_dir_file_operations;
5072 btrfs_i_size_write(inode, 0);
5074 err = btrfs_update_inode(trans, new_root, inode);
5081 /* helper function for file defrag and space balancing. This
5082 * forces readahead on a given range of bytes in an inode
5084 unsigned long btrfs_force_ra(struct address_space *mapping,
5085 struct file_ra_state *ra, struct file *file,
5086 pgoff_t offset, pgoff_t last_index)
5088 pgoff_t req_size = last_index - offset + 1;
5090 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5091 return offset + req_size;
5094 struct inode *btrfs_alloc_inode(struct super_block *sb)
5096 struct btrfs_inode *ei;
5098 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5102 ei->logged_trans = 0;
5103 ei->outstanding_extents = 0;
5104 ei->reserved_extents = 0;
5105 spin_lock_init(&ei->accounting_lock);
5106 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5107 INIT_LIST_HEAD(&ei->i_orphan);
5108 INIT_LIST_HEAD(&ei->ordered_operations);
5109 return &ei->vfs_inode;
5112 void btrfs_destroy_inode(struct inode *inode)
5114 struct btrfs_ordered_extent *ordered;
5115 struct btrfs_root *root = BTRFS_I(inode)->root;
5117 WARN_ON(!list_empty(&inode->i_dentry));
5118 WARN_ON(inode->i_data.nrpages);
5121 * Make sure we're properly removed from the ordered operation
5125 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5126 spin_lock(&root->fs_info->ordered_extent_lock);
5127 list_del_init(&BTRFS_I(inode)->ordered_operations);
5128 spin_unlock(&root->fs_info->ordered_extent_lock);
5131 spin_lock(&root->list_lock);
5132 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5133 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
5134 " list\n", inode->i_ino);
5137 spin_unlock(&root->list_lock);
5140 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5144 printk(KERN_ERR "btrfs found ordered "
5145 "extent %llu %llu on inode cleanup\n",
5146 (unsigned long long)ordered->file_offset,
5147 (unsigned long long)ordered->len);
5148 btrfs_remove_ordered_extent(inode, ordered);
5149 btrfs_put_ordered_extent(ordered);
5150 btrfs_put_ordered_extent(ordered);
5153 inode_tree_del(inode);
5154 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5155 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5158 void btrfs_drop_inode(struct inode *inode)
5160 struct btrfs_root *root = BTRFS_I(inode)->root;
5162 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5163 generic_delete_inode(inode);
5165 generic_drop_inode(inode);
5168 static void init_once(void *foo)
5170 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5172 inode_init_once(&ei->vfs_inode);
5175 void btrfs_destroy_cachep(void)
5177 if (btrfs_inode_cachep)
5178 kmem_cache_destroy(btrfs_inode_cachep);
5179 if (btrfs_trans_handle_cachep)
5180 kmem_cache_destroy(btrfs_trans_handle_cachep);
5181 if (btrfs_transaction_cachep)
5182 kmem_cache_destroy(btrfs_transaction_cachep);
5183 if (btrfs_path_cachep)
5184 kmem_cache_destroy(btrfs_path_cachep);
5187 int btrfs_init_cachep(void)
5189 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5190 sizeof(struct btrfs_inode), 0,
5191 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5192 if (!btrfs_inode_cachep)
5195 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5196 sizeof(struct btrfs_trans_handle), 0,
5197 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5198 if (!btrfs_trans_handle_cachep)
5201 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5202 sizeof(struct btrfs_transaction), 0,
5203 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5204 if (!btrfs_transaction_cachep)
5207 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5208 sizeof(struct btrfs_path), 0,
5209 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5210 if (!btrfs_path_cachep)
5215 btrfs_destroy_cachep();
5219 static int btrfs_getattr(struct vfsmount *mnt,
5220 struct dentry *dentry, struct kstat *stat)
5222 struct inode *inode = dentry->d_inode;
5223 generic_fillattr(inode, stat);
5224 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5225 stat->blksize = PAGE_CACHE_SIZE;
5226 stat->blocks = (inode_get_bytes(inode) +
5227 BTRFS_I(inode)->delalloc_bytes) >> 9;
5231 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5232 struct inode *new_dir, struct dentry *new_dentry)
5234 struct btrfs_trans_handle *trans;
5235 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5236 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5237 struct inode *new_inode = new_dentry->d_inode;
5238 struct inode *old_inode = old_dentry->d_inode;
5239 struct timespec ctime = CURRENT_TIME;
5244 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5247 /* we only allow rename subvolume link between subvolumes */
5248 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5251 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5252 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5255 if (S_ISDIR(old_inode->i_mode) && new_inode &&
5256 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5260 * 2 items for dir items
5261 * 1 item for orphan entry
5264 ret = btrfs_reserve_metadata_space(root, 4);
5269 * we're using rename to replace one file with another.
5270 * and the replacement file is large. Start IO on it now so
5271 * we don't add too much work to the end of the transaction
5273 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5274 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5275 filemap_flush(old_inode->i_mapping);
5277 /* close the racy window with snapshot create/destroy ioctl */
5278 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5279 down_read(&root->fs_info->subvol_sem);
5281 trans = btrfs_start_transaction(root, 1);
5282 btrfs_set_trans_block_group(trans, new_dir);
5285 btrfs_record_root_in_trans(trans, dest);
5287 ret = btrfs_set_inode_index(new_dir, &index);
5291 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5292 /* force full log commit if subvolume involved. */
5293 root->fs_info->last_trans_log_full_commit = trans->transid;
5295 ret = btrfs_insert_inode_ref(trans, dest,
5296 new_dentry->d_name.name,
5297 new_dentry->d_name.len,
5299 new_dir->i_ino, index);
5303 * this is an ugly little race, but the rename is required
5304 * to make sure that if we crash, the inode is either at the
5305 * old name or the new one. pinning the log transaction lets
5306 * us make sure we don't allow a log commit to come in after
5307 * we unlink the name but before we add the new name back in.
5309 btrfs_pin_log_trans(root);
5312 * make sure the inode gets flushed if it is replacing
5315 if (new_inode && new_inode->i_size &&
5316 old_inode && S_ISREG(old_inode->i_mode)) {
5317 btrfs_add_ordered_operation(trans, root, old_inode);
5320 old_dir->i_ctime = old_dir->i_mtime = ctime;
5321 new_dir->i_ctime = new_dir->i_mtime = ctime;
5322 old_inode->i_ctime = ctime;
5324 if (old_dentry->d_parent != new_dentry->d_parent)
5325 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5327 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5328 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5329 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5330 old_dentry->d_name.name,
5331 old_dentry->d_name.len);
5333 btrfs_inc_nlink(old_dentry->d_inode);
5334 ret = btrfs_unlink_inode(trans, root, old_dir,
5335 old_dentry->d_inode,
5336 old_dentry->d_name.name,
5337 old_dentry->d_name.len);
5342 new_inode->i_ctime = CURRENT_TIME;
5343 if (unlikely(new_inode->i_ino ==
5344 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5345 root_objectid = BTRFS_I(new_inode)->location.objectid;
5346 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5348 new_dentry->d_name.name,
5349 new_dentry->d_name.len);
5350 BUG_ON(new_inode->i_nlink == 0);
5352 ret = btrfs_unlink_inode(trans, dest, new_dir,
5353 new_dentry->d_inode,
5354 new_dentry->d_name.name,
5355 new_dentry->d_name.len);
5358 if (new_inode->i_nlink == 0) {
5359 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5364 ret = btrfs_add_link(trans, new_dir, old_inode,
5365 new_dentry->d_name.name,
5366 new_dentry->d_name.len, 0, index);
5369 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5370 btrfs_log_new_name(trans, old_inode, old_dir,
5371 new_dentry->d_parent);
5372 btrfs_end_log_trans(root);
5375 btrfs_end_transaction_throttle(trans, root);
5377 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5378 up_read(&root->fs_info->subvol_sem);
5380 btrfs_unreserve_metadata_space(root, 4);
5385 * some fairly slow code that needs optimization. This walks the list
5386 * of all the inodes with pending delalloc and forces them to disk.
5388 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
5390 struct list_head *head = &root->fs_info->delalloc_inodes;
5391 struct btrfs_inode *binode;
5392 struct inode *inode;
5394 if (root->fs_info->sb->s_flags & MS_RDONLY)
5397 spin_lock(&root->fs_info->delalloc_lock);
5398 while (!list_empty(head)) {
5399 binode = list_entry(head->next, struct btrfs_inode,
5401 inode = igrab(&binode->vfs_inode);
5403 list_del_init(&binode->delalloc_inodes);
5404 spin_unlock(&root->fs_info->delalloc_lock);
5406 filemap_flush(inode->i_mapping);
5410 spin_lock(&root->fs_info->delalloc_lock);
5412 spin_unlock(&root->fs_info->delalloc_lock);
5414 /* the filemap_flush will queue IO into the worker threads, but
5415 * we have to make sure the IO is actually started and that
5416 * ordered extents get created before we return
5418 atomic_inc(&root->fs_info->async_submit_draining);
5419 while (atomic_read(&root->fs_info->nr_async_submits) ||
5420 atomic_read(&root->fs_info->async_delalloc_pages)) {
5421 wait_event(root->fs_info->async_submit_wait,
5422 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5423 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5425 atomic_dec(&root->fs_info->async_submit_draining);
5429 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5430 const char *symname)
5432 struct btrfs_trans_handle *trans;
5433 struct btrfs_root *root = BTRFS_I(dir)->root;
5434 struct btrfs_path *path;
5435 struct btrfs_key key;
5436 struct inode *inode = NULL;
5444 struct btrfs_file_extent_item *ei;
5445 struct extent_buffer *leaf;
5446 unsigned long nr = 0;
5448 name_len = strlen(symname) + 1;
5449 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5450 return -ENAMETOOLONG;
5453 * 2 items for inode item and ref
5454 * 2 items for dir items
5455 * 1 item for xattr if selinux is on
5457 err = btrfs_reserve_metadata_space(root, 5);
5461 trans = btrfs_start_transaction(root, 1);
5464 btrfs_set_trans_block_group(trans, dir);
5466 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5472 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5474 dentry->d_parent->d_inode->i_ino, objectid,
5475 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5477 err = PTR_ERR(inode);
5481 err = btrfs_init_inode_security(inode, dir);
5487 btrfs_set_trans_block_group(trans, inode);
5488 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5492 inode->i_mapping->a_ops = &btrfs_aops;
5493 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5494 inode->i_fop = &btrfs_file_operations;
5495 inode->i_op = &btrfs_file_inode_operations;
5496 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5498 btrfs_update_inode_block_group(trans, inode);
5499 btrfs_update_inode_block_group(trans, dir);
5503 path = btrfs_alloc_path();
5505 key.objectid = inode->i_ino;
5507 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5508 datasize = btrfs_file_extent_calc_inline_size(name_len);
5509 err = btrfs_insert_empty_item(trans, root, path, &key,
5515 leaf = path->nodes[0];
5516 ei = btrfs_item_ptr(leaf, path->slots[0],
5517 struct btrfs_file_extent_item);
5518 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5519 btrfs_set_file_extent_type(leaf, ei,
5520 BTRFS_FILE_EXTENT_INLINE);
5521 btrfs_set_file_extent_encryption(leaf, ei, 0);
5522 btrfs_set_file_extent_compression(leaf, ei, 0);
5523 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5524 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5526 ptr = btrfs_file_extent_inline_start(ei);
5527 write_extent_buffer(leaf, symname, ptr, name_len);
5528 btrfs_mark_buffer_dirty(leaf);
5529 btrfs_free_path(path);
5531 inode->i_op = &btrfs_symlink_inode_operations;
5532 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5533 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5534 inode_set_bytes(inode, name_len);
5535 btrfs_i_size_write(inode, name_len - 1);
5536 err = btrfs_update_inode(trans, root, inode);
5541 nr = trans->blocks_used;
5542 btrfs_end_transaction_throttle(trans, root);
5544 btrfs_unreserve_metadata_space(root, 5);
5546 inode_dec_link_count(inode);
5549 btrfs_btree_balance_dirty(root, nr);
5553 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5554 struct inode *inode, u64 start, u64 end,
5555 u64 locked_end, u64 alloc_hint, int mode)
5557 struct btrfs_root *root = BTRFS_I(inode)->root;
5558 struct btrfs_key ins;
5560 u64 cur_offset = start;
5561 u64 num_bytes = end - start;
5564 while (num_bytes > 0) {
5565 alloc_size = min(num_bytes, root->fs_info->max_extent);
5567 ret = btrfs_reserve_metadata_space(root, 1);
5571 ret = btrfs_reserve_extent(trans, root, alloc_size,
5572 root->sectorsize, 0, alloc_hint,
5578 ret = insert_reserved_file_extent(trans, inode,
5579 cur_offset, ins.objectid,
5580 ins.offset, ins.offset,
5581 ins.offset, locked_end,
5583 BTRFS_FILE_EXTENT_PREALLOC);
5585 btrfs_drop_extent_cache(inode, cur_offset,
5586 cur_offset + ins.offset -1, 0);
5587 num_bytes -= ins.offset;
5588 cur_offset += ins.offset;
5589 alloc_hint = ins.objectid + ins.offset;
5590 btrfs_unreserve_metadata_space(root, 1);
5593 if (cur_offset > start) {
5594 inode->i_ctime = CURRENT_TIME;
5595 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5596 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5597 cur_offset > i_size_read(inode))
5598 btrfs_i_size_write(inode, cur_offset);
5599 ret = btrfs_update_inode(trans, root, inode);
5606 static long btrfs_fallocate(struct inode *inode, int mode,
5607 loff_t offset, loff_t len)
5615 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5616 struct extent_map *em;
5617 struct btrfs_trans_handle *trans;
5618 struct btrfs_root *root;
5621 alloc_start = offset & ~mask;
5622 alloc_end = (offset + len + mask) & ~mask;
5625 * wait for ordered IO before we have any locks. We'll loop again
5626 * below with the locks held.
5628 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5630 mutex_lock(&inode->i_mutex);
5631 if (alloc_start > inode->i_size) {
5632 ret = btrfs_cont_expand(inode, alloc_start);
5637 root = BTRFS_I(inode)->root;
5639 ret = btrfs_check_data_free_space(root, inode,
5640 alloc_end - alloc_start);
5644 locked_end = alloc_end - 1;
5646 struct btrfs_ordered_extent *ordered;
5648 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5654 /* the extent lock is ordered inside the running
5657 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5659 ordered = btrfs_lookup_first_ordered_extent(inode,
5662 ordered->file_offset + ordered->len > alloc_start &&
5663 ordered->file_offset < alloc_end) {
5664 btrfs_put_ordered_extent(ordered);
5665 unlock_extent(&BTRFS_I(inode)->io_tree,
5666 alloc_start, locked_end, GFP_NOFS);
5667 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5670 * we can't wait on the range with the transaction
5671 * running or with the extent lock held
5673 btrfs_wait_ordered_range(inode, alloc_start,
5674 alloc_end - alloc_start);
5677 btrfs_put_ordered_extent(ordered);
5682 cur_offset = alloc_start;
5684 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5685 alloc_end - cur_offset, 0);
5686 BUG_ON(IS_ERR(em) || !em);
5687 last_byte = min(extent_map_end(em), alloc_end);
5688 last_byte = (last_byte + mask) & ~mask;
5689 if (em->block_start == EXTENT_MAP_HOLE) {
5690 ret = prealloc_file_range(trans, inode, cur_offset,
5691 last_byte, locked_end + 1,
5694 free_extent_map(em);
5698 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5699 alloc_hint = em->block_start;
5700 free_extent_map(em);
5702 cur_offset = last_byte;
5703 if (cur_offset >= alloc_end) {
5708 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5711 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5713 btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
5715 mutex_unlock(&inode->i_mutex);
5719 static int btrfs_set_page_dirty(struct page *page)
5721 return __set_page_dirty_nobuffers(page);
5724 static int btrfs_permission(struct inode *inode, int mask)
5726 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5728 return generic_permission(inode, mask, btrfs_check_acl);
5731 static struct inode_operations btrfs_dir_inode_operations = {
5732 .getattr = btrfs_getattr,
5733 .lookup = btrfs_lookup,
5734 .create = btrfs_create,
5735 .unlink = btrfs_unlink,
5737 .mkdir = btrfs_mkdir,
5738 .rmdir = btrfs_rmdir,
5739 .rename = btrfs_rename,
5740 .symlink = btrfs_symlink,
5741 .setattr = btrfs_setattr,
5742 .mknod = btrfs_mknod,
5743 .setxattr = btrfs_setxattr,
5744 .getxattr = btrfs_getxattr,
5745 .listxattr = btrfs_listxattr,
5746 .removexattr = btrfs_removexattr,
5747 .permission = btrfs_permission,
5749 static struct inode_operations btrfs_dir_ro_inode_operations = {
5750 .lookup = btrfs_lookup,
5751 .permission = btrfs_permission,
5754 static struct file_operations btrfs_dir_file_operations = {
5755 .llseek = generic_file_llseek,
5756 .read = generic_read_dir,
5757 .readdir = btrfs_real_readdir,
5758 .unlocked_ioctl = btrfs_ioctl,
5759 #ifdef CONFIG_COMPAT
5760 .compat_ioctl = btrfs_ioctl,
5762 .release = btrfs_release_file,
5763 .fsync = btrfs_sync_file,
5766 static struct extent_io_ops btrfs_extent_io_ops = {
5767 .fill_delalloc = run_delalloc_range,
5768 .submit_bio_hook = btrfs_submit_bio_hook,
5769 .merge_bio_hook = btrfs_merge_bio_hook,
5770 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5771 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5772 .writepage_start_hook = btrfs_writepage_start_hook,
5773 .readpage_io_failed_hook = btrfs_io_failed_hook,
5774 .set_bit_hook = btrfs_set_bit_hook,
5775 .clear_bit_hook = btrfs_clear_bit_hook,
5776 .merge_extent_hook = btrfs_merge_extent_hook,
5777 .split_extent_hook = btrfs_split_extent_hook,
5781 * btrfs doesn't support the bmap operation because swapfiles
5782 * use bmap to make a mapping of extents in the file. They assume
5783 * these extents won't change over the life of the file and they
5784 * use the bmap result to do IO directly to the drive.
5786 * the btrfs bmap call would return logical addresses that aren't
5787 * suitable for IO and they also will change frequently as COW
5788 * operations happen. So, swapfile + btrfs == corruption.
5790 * For now we're avoiding this by dropping bmap.
5792 static struct address_space_operations btrfs_aops = {
5793 .readpage = btrfs_readpage,
5794 .writepage = btrfs_writepage,
5795 .writepages = btrfs_writepages,
5796 .readpages = btrfs_readpages,
5797 .sync_page = block_sync_page,
5798 .direct_IO = btrfs_direct_IO,
5799 .invalidatepage = btrfs_invalidatepage,
5800 .releasepage = btrfs_releasepage,
5801 .set_page_dirty = btrfs_set_page_dirty,
5804 static struct address_space_operations btrfs_symlink_aops = {
5805 .readpage = btrfs_readpage,
5806 .writepage = btrfs_writepage,
5807 .invalidatepage = btrfs_invalidatepage,
5808 .releasepage = btrfs_releasepage,
5811 static struct inode_operations btrfs_file_inode_operations = {
5812 .truncate = btrfs_truncate,
5813 .getattr = btrfs_getattr,
5814 .setattr = btrfs_setattr,
5815 .setxattr = btrfs_setxattr,
5816 .getxattr = btrfs_getxattr,
5817 .listxattr = btrfs_listxattr,
5818 .removexattr = btrfs_removexattr,
5819 .permission = btrfs_permission,
5820 .fallocate = btrfs_fallocate,
5821 .fiemap = btrfs_fiemap,
5823 static struct inode_operations btrfs_special_inode_operations = {
5824 .getattr = btrfs_getattr,
5825 .setattr = btrfs_setattr,
5826 .permission = btrfs_permission,
5827 .setxattr = btrfs_setxattr,
5828 .getxattr = btrfs_getxattr,
5829 .listxattr = btrfs_listxattr,
5830 .removexattr = btrfs_removexattr,
5832 static struct inode_operations btrfs_symlink_inode_operations = {
5833 .readlink = generic_readlink,
5834 .follow_link = page_follow_link_light,
5835 .put_link = page_put_link,
5836 .permission = btrfs_permission,
5837 .setxattr = btrfs_setxattr,
5838 .getxattr = btrfs_getxattr,
5839 .listxattr = btrfs_listxattr,
5840 .removexattr = btrfs_removexattr,
5843 const struct dentry_operations btrfs_dentry_operations = {
5844 .d_delete = btrfs_dentry_delete,