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_SET_WRITEBACK | EXTENT_END_WRITEBACK);
438 * we aren't doing an inline extent round the compressed size
439 * up to a block size boundary so the allocator does sane
442 total_compressed = (total_compressed + blocksize - 1) &
446 * one last check to make sure the compression is really a
447 * win, compare the page count read with the blocks on disk
449 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
450 ~(PAGE_CACHE_SIZE - 1);
451 if (total_compressed >= total_in) {
454 disk_num_bytes = total_compressed;
455 num_bytes = total_in;
458 if (!will_compress && pages) {
460 * the compression code ran but failed to make things smaller,
461 * free any pages it allocated and our page pointer array
463 for (i = 0; i < nr_pages_ret; i++) {
464 WARN_ON(pages[i]->mapping);
465 page_cache_release(pages[i]);
469 total_compressed = 0;
472 /* flag the file so we don't compress in the future */
473 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
478 /* the async work queues will take care of doing actual
479 * allocation on disk for these compressed pages,
480 * and will submit them to the elevator.
482 add_async_extent(async_cow, start, num_bytes,
483 total_compressed, pages, nr_pages_ret);
485 if (start + num_bytes < end && start + num_bytes < actual_end) {
492 cleanup_and_bail_uncompressed:
494 * No compression, but we still need to write the pages in
495 * the file we've been given so far. redirty the locked
496 * page if it corresponds to our extent and set things up
497 * for the async work queue to run cow_file_range to do
498 * the normal delalloc dance
500 if (page_offset(locked_page) >= start &&
501 page_offset(locked_page) <= end) {
502 __set_page_dirty_nobuffers(locked_page);
503 /* unlocked later on in the async handlers */
505 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
513 for (i = 0; i < nr_pages_ret; i++) {
514 WARN_ON(pages[i]->mapping);
515 page_cache_release(pages[i]);
523 * phase two of compressed writeback. This is the ordered portion
524 * of the code, which only gets called in the order the work was
525 * queued. We walk all the async extents created by compress_file_range
526 * and send them down to the disk.
528 static noinline int submit_compressed_extents(struct inode *inode,
529 struct async_cow *async_cow)
531 struct async_extent *async_extent;
533 struct btrfs_trans_handle *trans;
534 struct btrfs_key ins;
535 struct extent_map *em;
536 struct btrfs_root *root = BTRFS_I(inode)->root;
537 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
538 struct extent_io_tree *io_tree;
541 if (list_empty(&async_cow->extents))
544 trans = btrfs_join_transaction(root, 1);
546 while (!list_empty(&async_cow->extents)) {
547 async_extent = list_entry(async_cow->extents.next,
548 struct async_extent, list);
549 list_del(&async_extent->list);
551 io_tree = &BTRFS_I(inode)->io_tree;
553 /* did the compression code fall back to uncompressed IO? */
554 if (!async_extent->pages) {
555 int page_started = 0;
556 unsigned long nr_written = 0;
558 lock_extent(io_tree, async_extent->start,
559 async_extent->start +
560 async_extent->ram_size - 1, GFP_NOFS);
562 /* allocate blocks */
563 cow_file_range(inode, async_cow->locked_page,
565 async_extent->start +
566 async_extent->ram_size - 1,
567 &page_started, &nr_written, 0);
570 * if page_started, cow_file_range inserted an
571 * inline extent and took care of all the unlocking
572 * and IO for us. Otherwise, we need to submit
573 * all those pages down to the drive.
576 extent_write_locked_range(io_tree,
577 inode, async_extent->start,
578 async_extent->start +
579 async_extent->ram_size - 1,
587 lock_extent(io_tree, async_extent->start,
588 async_extent->start + async_extent->ram_size - 1,
591 * here we're doing allocation and writeback of the
594 btrfs_drop_extent_cache(inode, async_extent->start,
595 async_extent->start +
596 async_extent->ram_size - 1, 0);
598 ret = btrfs_reserve_extent(trans, root,
599 async_extent->compressed_size,
600 async_extent->compressed_size,
604 em = alloc_extent_map(GFP_NOFS);
605 em->start = async_extent->start;
606 em->len = async_extent->ram_size;
607 em->orig_start = em->start;
609 em->block_start = ins.objectid;
610 em->block_len = ins.offset;
611 em->bdev = root->fs_info->fs_devices->latest_bdev;
612 set_bit(EXTENT_FLAG_PINNED, &em->flags);
613 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
616 write_lock(&em_tree->lock);
617 ret = add_extent_mapping(em_tree, em);
618 write_unlock(&em_tree->lock);
619 if (ret != -EEXIST) {
623 btrfs_drop_extent_cache(inode, async_extent->start,
624 async_extent->start +
625 async_extent->ram_size - 1, 0);
628 ret = btrfs_add_ordered_extent(inode, async_extent->start,
630 async_extent->ram_size,
632 BTRFS_ORDERED_COMPRESSED);
635 btrfs_end_transaction(trans, root);
638 * clear dirty, set writeback and unlock the pages.
640 extent_clear_unlock_delalloc(inode,
641 &BTRFS_I(inode)->io_tree,
643 async_extent->start +
644 async_extent->ram_size - 1,
645 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
646 EXTENT_CLEAR_UNLOCK |
647 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
649 ret = btrfs_submit_compressed_write(inode,
651 async_extent->ram_size,
653 ins.offset, async_extent->pages,
654 async_extent->nr_pages);
657 trans = btrfs_join_transaction(root, 1);
658 alloc_hint = ins.objectid + ins.offset;
663 btrfs_end_transaction(trans, root);
668 * when extent_io.c finds a delayed allocation range in the file,
669 * the call backs end up in this code. The basic idea is to
670 * allocate extents on disk for the range, and create ordered data structs
671 * in ram to track those extents.
673 * locked_page is the page that writepage had locked already. We use
674 * it to make sure we don't do extra locks or unlocks.
676 * *page_started is set to one if we unlock locked_page and do everything
677 * required to start IO on it. It may be clean and already done with
680 static noinline int cow_file_range(struct inode *inode,
681 struct page *locked_page,
682 u64 start, u64 end, int *page_started,
683 unsigned long *nr_written,
686 struct btrfs_root *root = BTRFS_I(inode)->root;
687 struct btrfs_trans_handle *trans;
690 unsigned long ram_size;
693 u64 blocksize = root->sectorsize;
695 u64 isize = i_size_read(inode);
696 struct btrfs_key ins;
697 struct extent_map *em;
698 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
701 trans = btrfs_join_transaction(root, 1);
703 btrfs_set_trans_block_group(trans, inode);
705 actual_end = min_t(u64, isize, end + 1);
707 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
708 num_bytes = max(blocksize, num_bytes);
709 disk_num_bytes = num_bytes;
713 /* lets try to make an inline extent */
714 ret = cow_file_range_inline(trans, root, inode,
715 start, end, 0, NULL);
717 extent_clear_unlock_delalloc(inode,
718 &BTRFS_I(inode)->io_tree,
720 EXTENT_CLEAR_UNLOCK_PAGE |
721 EXTENT_CLEAR_UNLOCK |
722 EXTENT_CLEAR_DELALLOC |
724 EXTENT_SET_WRITEBACK |
725 EXTENT_END_WRITEBACK);
726 *nr_written = *nr_written +
727 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
734 BUG_ON(disk_num_bytes >
735 btrfs_super_total_bytes(&root->fs_info->super_copy));
738 read_lock(&BTRFS_I(inode)->extent_tree.lock);
739 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
742 alloc_hint = em->block_start;
745 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
746 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
748 while (disk_num_bytes > 0) {
751 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
752 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
753 root->sectorsize, 0, alloc_hint,
757 em = alloc_extent_map(GFP_NOFS);
759 em->orig_start = em->start;
760 ram_size = ins.offset;
761 em->len = ins.offset;
763 em->block_start = ins.objectid;
764 em->block_len = ins.offset;
765 em->bdev = root->fs_info->fs_devices->latest_bdev;
766 set_bit(EXTENT_FLAG_PINNED, &em->flags);
769 write_lock(&em_tree->lock);
770 ret = add_extent_mapping(em_tree, em);
771 write_unlock(&em_tree->lock);
772 if (ret != -EEXIST) {
776 btrfs_drop_extent_cache(inode, start,
777 start + ram_size - 1, 0);
780 cur_alloc_size = ins.offset;
781 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
782 ram_size, cur_alloc_size, 0);
785 if (root->root_key.objectid ==
786 BTRFS_DATA_RELOC_TREE_OBJECTID) {
787 ret = btrfs_reloc_clone_csums(inode, start,
792 if (disk_num_bytes < cur_alloc_size)
795 /* we're not doing compressed IO, don't unlock the first
796 * page (which the caller expects to stay locked), don't
797 * clear any dirty bits and don't set any writeback bits
799 * Do set the Private2 bit so we know this page was properly
800 * setup for writepage
802 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
803 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
806 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
807 start, start + ram_size - 1,
809 disk_num_bytes -= cur_alloc_size;
810 num_bytes -= cur_alloc_size;
811 alloc_hint = ins.objectid + ins.offset;
812 start += cur_alloc_size;
816 btrfs_end_transaction(trans, root);
822 * work queue call back to started compression on a file and pages
824 static noinline void async_cow_start(struct btrfs_work *work)
826 struct async_cow *async_cow;
828 async_cow = container_of(work, struct async_cow, work);
830 compress_file_range(async_cow->inode, async_cow->locked_page,
831 async_cow->start, async_cow->end, async_cow,
834 async_cow->inode = NULL;
838 * work queue call back to submit previously compressed pages
840 static noinline void async_cow_submit(struct btrfs_work *work)
842 struct async_cow *async_cow;
843 struct btrfs_root *root;
844 unsigned long nr_pages;
846 async_cow = container_of(work, struct async_cow, work);
848 root = async_cow->root;
849 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
852 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
854 if (atomic_read(&root->fs_info->async_delalloc_pages) <
856 waitqueue_active(&root->fs_info->async_submit_wait))
857 wake_up(&root->fs_info->async_submit_wait);
859 if (async_cow->inode)
860 submit_compressed_extents(async_cow->inode, async_cow);
863 static noinline void async_cow_free(struct btrfs_work *work)
865 struct async_cow *async_cow;
866 async_cow = container_of(work, struct async_cow, work);
870 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
871 u64 start, u64 end, int *page_started,
872 unsigned long *nr_written)
874 struct async_cow *async_cow;
875 struct btrfs_root *root = BTRFS_I(inode)->root;
876 unsigned long nr_pages;
878 int limit = 10 * 1024 * 1042;
880 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
881 EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS);
882 while (start < end) {
883 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
884 async_cow->inode = inode;
885 async_cow->root = root;
886 async_cow->locked_page = locked_page;
887 async_cow->start = start;
889 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
892 cur_end = min(end, start + 512 * 1024 - 1);
894 async_cow->end = cur_end;
895 INIT_LIST_HEAD(&async_cow->extents);
897 async_cow->work.func = async_cow_start;
898 async_cow->work.ordered_func = async_cow_submit;
899 async_cow->work.ordered_free = async_cow_free;
900 async_cow->work.flags = 0;
902 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
904 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
906 btrfs_queue_worker(&root->fs_info->delalloc_workers,
909 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
910 wait_event(root->fs_info->async_submit_wait,
911 (atomic_read(&root->fs_info->async_delalloc_pages) <
915 while (atomic_read(&root->fs_info->async_submit_draining) &&
916 atomic_read(&root->fs_info->async_delalloc_pages)) {
917 wait_event(root->fs_info->async_submit_wait,
918 (atomic_read(&root->fs_info->async_delalloc_pages) ==
922 *nr_written += nr_pages;
929 static noinline int csum_exist_in_range(struct btrfs_root *root,
930 u64 bytenr, u64 num_bytes)
933 struct btrfs_ordered_sum *sums;
936 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
937 bytenr + num_bytes - 1, &list);
938 if (ret == 0 && list_empty(&list))
941 while (!list_empty(&list)) {
942 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
943 list_del(&sums->list);
950 * when nowcow writeback call back. This checks for snapshots or COW copies
951 * of the extents that exist in the file, and COWs the file as required.
953 * If no cow copies or snapshots exist, we write directly to the existing
956 static noinline int run_delalloc_nocow(struct inode *inode,
957 struct page *locked_page,
958 u64 start, u64 end, int *page_started, int force,
959 unsigned long *nr_written)
961 struct btrfs_root *root = BTRFS_I(inode)->root;
962 struct btrfs_trans_handle *trans;
963 struct extent_buffer *leaf;
964 struct btrfs_path *path;
965 struct btrfs_file_extent_item *fi;
966 struct btrfs_key found_key;
979 path = btrfs_alloc_path();
981 trans = btrfs_join_transaction(root, 1);
987 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
990 if (ret > 0 && path->slots[0] > 0 && check_prev) {
991 leaf = path->nodes[0];
992 btrfs_item_key_to_cpu(leaf, &found_key,
994 if (found_key.objectid == inode->i_ino &&
995 found_key.type == BTRFS_EXTENT_DATA_KEY)
1000 leaf = path->nodes[0];
1001 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1002 ret = btrfs_next_leaf(root, path);
1007 leaf = path->nodes[0];
1013 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1015 if (found_key.objectid > inode->i_ino ||
1016 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1017 found_key.offset > end)
1020 if (found_key.offset > cur_offset) {
1021 extent_end = found_key.offset;
1025 fi = btrfs_item_ptr(leaf, path->slots[0],
1026 struct btrfs_file_extent_item);
1027 extent_type = btrfs_file_extent_type(leaf, fi);
1029 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1030 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1031 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1032 extent_offset = btrfs_file_extent_offset(leaf, fi);
1033 extent_end = found_key.offset +
1034 btrfs_file_extent_num_bytes(leaf, fi);
1035 if (extent_end <= start) {
1039 if (disk_bytenr == 0)
1041 if (btrfs_file_extent_compression(leaf, fi) ||
1042 btrfs_file_extent_encryption(leaf, fi) ||
1043 btrfs_file_extent_other_encoding(leaf, fi))
1045 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1047 if (btrfs_extent_readonly(root, disk_bytenr))
1049 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1051 extent_offset, disk_bytenr))
1053 disk_bytenr += extent_offset;
1054 disk_bytenr += cur_offset - found_key.offset;
1055 num_bytes = min(end + 1, extent_end) - cur_offset;
1057 * force cow if csum exists in the range.
1058 * this ensure that csum for a given extent are
1059 * either valid or do not exist.
1061 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1064 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1065 extent_end = found_key.offset +
1066 btrfs_file_extent_inline_len(leaf, fi);
1067 extent_end = ALIGN(extent_end, root->sectorsize);
1072 if (extent_end <= start) {
1077 if (cow_start == (u64)-1)
1078 cow_start = cur_offset;
1079 cur_offset = extent_end;
1080 if (cur_offset > end)
1086 btrfs_release_path(root, path);
1087 if (cow_start != (u64)-1) {
1088 ret = cow_file_range(inode, locked_page, cow_start,
1089 found_key.offset - 1, page_started,
1092 cow_start = (u64)-1;
1095 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1096 struct extent_map *em;
1097 struct extent_map_tree *em_tree;
1098 em_tree = &BTRFS_I(inode)->extent_tree;
1099 em = alloc_extent_map(GFP_NOFS);
1100 em->start = cur_offset;
1101 em->orig_start = em->start;
1102 em->len = num_bytes;
1103 em->block_len = num_bytes;
1104 em->block_start = disk_bytenr;
1105 em->bdev = root->fs_info->fs_devices->latest_bdev;
1106 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1108 write_lock(&em_tree->lock);
1109 ret = add_extent_mapping(em_tree, em);
1110 write_unlock(&em_tree->lock);
1111 if (ret != -EEXIST) {
1112 free_extent_map(em);
1115 btrfs_drop_extent_cache(inode, em->start,
1116 em->start + em->len - 1, 0);
1118 type = BTRFS_ORDERED_PREALLOC;
1120 type = BTRFS_ORDERED_NOCOW;
1123 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1124 num_bytes, num_bytes, type);
1127 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1128 cur_offset, cur_offset + num_bytes - 1,
1129 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1130 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1131 EXTENT_SET_PRIVATE2);
1132 cur_offset = extent_end;
1133 if (cur_offset > end)
1136 btrfs_release_path(root, path);
1138 if (cur_offset <= end && cow_start == (u64)-1)
1139 cow_start = cur_offset;
1140 if (cow_start != (u64)-1) {
1141 ret = cow_file_range(inode, locked_page, cow_start, end,
1142 page_started, nr_written, 1);
1146 ret = btrfs_end_transaction(trans, root);
1148 btrfs_free_path(path);
1153 * extent_io.c call back to do delayed allocation processing
1155 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1156 u64 start, u64 end, int *page_started,
1157 unsigned long *nr_written)
1160 struct btrfs_root *root = BTRFS_I(inode)->root;
1162 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1163 ret = run_delalloc_nocow(inode, locked_page, start, end,
1164 page_started, 1, nr_written);
1165 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1166 ret = run_delalloc_nocow(inode, locked_page, start, end,
1167 page_started, 0, nr_written);
1168 else if (!btrfs_test_opt(root, COMPRESS))
1169 ret = cow_file_range(inode, locked_page, start, end,
1170 page_started, nr_written, 1);
1172 ret = cow_file_range_async(inode, locked_page, start, end,
1173 page_started, nr_written);
1177 static int btrfs_split_extent_hook(struct inode *inode,
1178 struct extent_state *orig, u64 split)
1180 struct btrfs_root *root = BTRFS_I(inode)->root;
1183 if (!(orig->state & EXTENT_DELALLOC))
1186 size = orig->end - orig->start + 1;
1187 if (size > root->fs_info->max_extent) {
1191 new_size = orig->end - split + 1;
1192 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1193 root->fs_info->max_extent);
1196 * if we break a large extent up then leave delalloc_extents be,
1197 * since we've already accounted for the large extent.
1199 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1200 root->fs_info->max_extent) < num_extents)
1204 BTRFS_I(inode)->delalloc_extents++;
1210 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1211 * extents so we can keep track of new extents that are just merged onto old
1212 * extents, such as when we are doing sequential writes, so we can properly
1213 * account for the metadata space we'll need.
1215 static int btrfs_merge_extent_hook(struct inode *inode,
1216 struct extent_state *new,
1217 struct extent_state *other)
1219 struct btrfs_root *root = BTRFS_I(inode)->root;
1220 u64 new_size, old_size;
1223 /* not delalloc, ignore it */
1224 if (!(other->state & EXTENT_DELALLOC))
1227 old_size = other->end - other->start + 1;
1228 if (new->start < other->start)
1229 new_size = other->end - new->start + 1;
1231 new_size = new->end - other->start + 1;
1233 /* we're not bigger than the max, unreserve the space and go */
1234 if (new_size <= root->fs_info->max_extent) {
1235 BTRFS_I(inode)->delalloc_extents--;
1240 * If we grew by another max_extent, just return, we want to keep that
1243 num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1244 root->fs_info->max_extent);
1245 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1246 root->fs_info->max_extent) > num_extents)
1249 BTRFS_I(inode)->delalloc_extents--;
1255 * extent_io.c set_bit_hook, used to track delayed allocation
1256 * bytes in this file, and to maintain the list of inodes that
1257 * have pending delalloc work to be done.
1259 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1260 unsigned long old, unsigned long bits)
1264 * set_bit and clear bit hooks normally require _irqsave/restore
1265 * but in this case, we are only testeing for the DELALLOC
1266 * bit, which is only set or cleared with irqs on
1268 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1269 struct btrfs_root *root = BTRFS_I(inode)->root;
1271 BTRFS_I(inode)->delalloc_extents++;
1272 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1273 spin_lock(&root->fs_info->delalloc_lock);
1274 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1275 root->fs_info->delalloc_bytes += end - start + 1;
1276 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1277 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1278 &root->fs_info->delalloc_inodes);
1280 spin_unlock(&root->fs_info->delalloc_lock);
1286 * extent_io.c clear_bit_hook, see set_bit_hook for why
1288 static int btrfs_clear_bit_hook(struct inode *inode,
1289 struct extent_state *state, unsigned long bits)
1292 * set_bit and clear bit hooks normally require _irqsave/restore
1293 * but in this case, we are only testeing for the DELALLOC
1294 * bit, which is only set or cleared with irqs on
1296 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1297 struct btrfs_root *root = BTRFS_I(inode)->root;
1299 BTRFS_I(inode)->delalloc_extents--;
1300 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1302 spin_lock(&root->fs_info->delalloc_lock);
1303 if (state->end - state->start + 1 >
1304 root->fs_info->delalloc_bytes) {
1305 printk(KERN_INFO "btrfs warning: delalloc account "
1307 (unsigned long long)
1308 state->end - state->start + 1,
1309 (unsigned long long)
1310 root->fs_info->delalloc_bytes);
1311 btrfs_delalloc_free_space(root, inode, (u64)-1);
1312 root->fs_info->delalloc_bytes = 0;
1313 BTRFS_I(inode)->delalloc_bytes = 0;
1315 btrfs_delalloc_free_space(root, inode,
1318 root->fs_info->delalloc_bytes -= state->end -
1320 BTRFS_I(inode)->delalloc_bytes -= state->end -
1323 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1324 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1325 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1327 spin_unlock(&root->fs_info->delalloc_lock);
1333 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1334 * we don't create bios that span stripes or chunks
1336 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1337 size_t size, struct bio *bio,
1338 unsigned long bio_flags)
1340 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1341 struct btrfs_mapping_tree *map_tree;
1342 u64 logical = (u64)bio->bi_sector << 9;
1347 if (bio_flags & EXTENT_BIO_COMPRESSED)
1350 length = bio->bi_size;
1351 map_tree = &root->fs_info->mapping_tree;
1352 map_length = length;
1353 ret = btrfs_map_block(map_tree, READ, logical,
1354 &map_length, NULL, 0);
1356 if (map_length < length + size)
1362 * in order to insert checksums into the metadata in large chunks,
1363 * we wait until bio submission time. All the pages in the bio are
1364 * checksummed and sums are attached onto the ordered extent record.
1366 * At IO completion time the cums attached on the ordered extent record
1367 * are inserted into the btree
1369 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1370 struct bio *bio, int mirror_num,
1371 unsigned long bio_flags)
1373 struct btrfs_root *root = BTRFS_I(inode)->root;
1376 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1382 * in order to insert checksums into the metadata in large chunks,
1383 * we wait until bio submission time. All the pages in the bio are
1384 * checksummed and sums are attached onto the ordered extent record.
1386 * At IO completion time the cums attached on the ordered extent record
1387 * are inserted into the btree
1389 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1390 int mirror_num, unsigned long bio_flags)
1392 struct btrfs_root *root = BTRFS_I(inode)->root;
1393 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1397 * extent_io.c submission hook. This does the right thing for csum calculation
1398 * on write, or reading the csums from the tree before a read
1400 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1401 int mirror_num, unsigned long bio_flags)
1403 struct btrfs_root *root = BTRFS_I(inode)->root;
1407 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1409 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1412 if (!(rw & (1 << BIO_RW))) {
1413 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1414 return btrfs_submit_compressed_read(inode, bio,
1415 mirror_num, bio_flags);
1416 } else if (!skip_sum)
1417 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1419 } else if (!skip_sum) {
1420 /* csum items have already been cloned */
1421 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1423 /* we're doing a write, do the async checksumming */
1424 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1425 inode, rw, bio, mirror_num,
1426 bio_flags, __btrfs_submit_bio_start,
1427 __btrfs_submit_bio_done);
1431 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1435 * given a list of ordered sums record them in the inode. This happens
1436 * at IO completion time based on sums calculated at bio submission time.
1438 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1439 struct inode *inode, u64 file_offset,
1440 struct list_head *list)
1442 struct btrfs_ordered_sum *sum;
1444 btrfs_set_trans_block_group(trans, inode);
1446 list_for_each_entry(sum, list, list) {
1447 btrfs_csum_file_blocks(trans,
1448 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1453 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1455 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1457 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1461 /* see btrfs_writepage_start_hook for details on why this is required */
1462 struct btrfs_writepage_fixup {
1464 struct btrfs_work work;
1467 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1469 struct btrfs_writepage_fixup *fixup;
1470 struct btrfs_ordered_extent *ordered;
1472 struct inode *inode;
1476 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1480 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1481 ClearPageChecked(page);
1485 inode = page->mapping->host;
1486 page_start = page_offset(page);
1487 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1489 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1491 /* already ordered? We're done */
1492 if (PagePrivate2(page))
1495 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1497 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1498 page_end, GFP_NOFS);
1500 btrfs_start_ordered_extent(inode, ordered, 1);
1504 btrfs_set_extent_delalloc(inode, page_start, page_end);
1505 ClearPageChecked(page);
1507 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1510 page_cache_release(page);
1514 * There are a few paths in the higher layers of the kernel that directly
1515 * set the page dirty bit without asking the filesystem if it is a
1516 * good idea. This causes problems because we want to make sure COW
1517 * properly happens and the data=ordered rules are followed.
1519 * In our case any range that doesn't have the ORDERED bit set
1520 * hasn't been properly setup for IO. We kick off an async process
1521 * to fix it up. The async helper will wait for ordered extents, set
1522 * the delalloc bit and make it safe to write the page.
1524 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1526 struct inode *inode = page->mapping->host;
1527 struct btrfs_writepage_fixup *fixup;
1528 struct btrfs_root *root = BTRFS_I(inode)->root;
1530 /* this page is properly in the ordered list */
1531 if (TestClearPagePrivate2(page))
1534 if (PageChecked(page))
1537 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1541 SetPageChecked(page);
1542 page_cache_get(page);
1543 fixup->work.func = btrfs_writepage_fixup_worker;
1545 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1549 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1550 struct inode *inode, u64 file_pos,
1551 u64 disk_bytenr, u64 disk_num_bytes,
1552 u64 num_bytes, u64 ram_bytes,
1554 u8 compression, u8 encryption,
1555 u16 other_encoding, int extent_type)
1557 struct btrfs_root *root = BTRFS_I(inode)->root;
1558 struct btrfs_file_extent_item *fi;
1559 struct btrfs_path *path;
1560 struct extent_buffer *leaf;
1561 struct btrfs_key ins;
1565 path = btrfs_alloc_path();
1568 path->leave_spinning = 1;
1571 * we may be replacing one extent in the tree with another.
1572 * The new extent is pinned in the extent map, and we don't want
1573 * to drop it from the cache until it is completely in the btree.
1575 * So, tell btrfs_drop_extents to leave this extent in the cache.
1576 * the caller is expected to unpin it and allow it to be merged
1579 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1580 file_pos + num_bytes, locked_end,
1581 file_pos, &hint, 0);
1584 ins.objectid = inode->i_ino;
1585 ins.offset = file_pos;
1586 ins.type = BTRFS_EXTENT_DATA_KEY;
1587 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1589 leaf = path->nodes[0];
1590 fi = btrfs_item_ptr(leaf, path->slots[0],
1591 struct btrfs_file_extent_item);
1592 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1593 btrfs_set_file_extent_type(leaf, fi, extent_type);
1594 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1595 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1596 btrfs_set_file_extent_offset(leaf, fi, 0);
1597 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1598 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1599 btrfs_set_file_extent_compression(leaf, fi, compression);
1600 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1601 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1603 btrfs_unlock_up_safe(path, 1);
1604 btrfs_set_lock_blocking(leaf);
1606 btrfs_mark_buffer_dirty(leaf);
1608 inode_add_bytes(inode, num_bytes);
1610 ins.objectid = disk_bytenr;
1611 ins.offset = disk_num_bytes;
1612 ins.type = BTRFS_EXTENT_ITEM_KEY;
1613 ret = btrfs_alloc_reserved_file_extent(trans, root,
1614 root->root_key.objectid,
1615 inode->i_ino, file_pos, &ins);
1617 btrfs_free_path(path);
1623 * helper function for btrfs_finish_ordered_io, this
1624 * just reads in some of the csum leaves to prime them into ram
1625 * before we start the transaction. It limits the amount of btree
1626 * reads required while inside the transaction.
1628 static noinline void reada_csum(struct btrfs_root *root,
1629 struct btrfs_path *path,
1630 struct btrfs_ordered_extent *ordered_extent)
1632 struct btrfs_ordered_sum *sum;
1635 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1637 bytenr = sum->sums[0].bytenr;
1640 * we don't care about the results, the point of this search is
1641 * just to get the btree leaves into ram
1643 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1646 /* as ordered data IO finishes, this gets called so we can finish
1647 * an ordered extent if the range of bytes in the file it covers are
1650 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1652 struct btrfs_root *root = BTRFS_I(inode)->root;
1653 struct btrfs_trans_handle *trans;
1654 struct btrfs_ordered_extent *ordered_extent = NULL;
1655 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1656 struct btrfs_path *path;
1660 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1665 * before we join the transaction, try to do some of our IO.
1666 * This will limit the amount of IO that we have to do with
1667 * the transaction running. We're unlikely to need to do any
1668 * IO if the file extents are new, the disk_i_size checks
1669 * covers the most common case.
1671 if (start < BTRFS_I(inode)->disk_i_size) {
1672 path = btrfs_alloc_path();
1674 ret = btrfs_lookup_file_extent(NULL, root, path,
1677 ordered_extent = btrfs_lookup_ordered_extent(inode,
1679 if (!list_empty(&ordered_extent->list)) {
1680 btrfs_release_path(root, path);
1681 reada_csum(root, path, ordered_extent);
1683 btrfs_free_path(path);
1687 trans = btrfs_join_transaction(root, 1);
1689 if (!ordered_extent)
1690 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1691 BUG_ON(!ordered_extent);
1692 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1695 lock_extent(io_tree, ordered_extent->file_offset,
1696 ordered_extent->file_offset + ordered_extent->len - 1,
1699 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1701 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1703 ret = btrfs_mark_extent_written(trans, root, inode,
1704 ordered_extent->file_offset,
1705 ordered_extent->file_offset +
1706 ordered_extent->len);
1709 ret = insert_reserved_file_extent(trans, inode,
1710 ordered_extent->file_offset,
1711 ordered_extent->start,
1712 ordered_extent->disk_len,
1713 ordered_extent->len,
1714 ordered_extent->len,
1715 ordered_extent->file_offset +
1716 ordered_extent->len,
1718 BTRFS_FILE_EXTENT_REG);
1719 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1720 ordered_extent->file_offset,
1721 ordered_extent->len);
1724 unlock_extent(io_tree, ordered_extent->file_offset,
1725 ordered_extent->file_offset + ordered_extent->len - 1,
1728 add_pending_csums(trans, inode, ordered_extent->file_offset,
1729 &ordered_extent->list);
1731 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1732 btrfs_ordered_update_i_size(inode, ordered_extent);
1733 btrfs_update_inode(trans, root, inode);
1734 btrfs_remove_ordered_extent(inode, ordered_extent);
1735 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1738 btrfs_put_ordered_extent(ordered_extent);
1739 /* once for the tree */
1740 btrfs_put_ordered_extent(ordered_extent);
1742 btrfs_end_transaction(trans, root);
1746 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1747 struct extent_state *state, int uptodate)
1749 ClearPagePrivate2(page);
1750 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1754 * When IO fails, either with EIO or csum verification fails, we
1755 * try other mirrors that might have a good copy of the data. This
1756 * io_failure_record is used to record state as we go through all the
1757 * mirrors. If another mirror has good data, the page is set up to date
1758 * and things continue. If a good mirror can't be found, the original
1759 * bio end_io callback is called to indicate things have failed.
1761 struct io_failure_record {
1766 unsigned long bio_flags;
1770 static int btrfs_io_failed_hook(struct bio *failed_bio,
1771 struct page *page, u64 start, u64 end,
1772 struct extent_state *state)
1774 struct io_failure_record *failrec = NULL;
1776 struct extent_map *em;
1777 struct inode *inode = page->mapping->host;
1778 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1779 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1786 ret = get_state_private(failure_tree, start, &private);
1788 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1791 failrec->start = start;
1792 failrec->len = end - start + 1;
1793 failrec->last_mirror = 0;
1794 failrec->bio_flags = 0;
1796 read_lock(&em_tree->lock);
1797 em = lookup_extent_mapping(em_tree, start, failrec->len);
1798 if (em->start > start || em->start + em->len < start) {
1799 free_extent_map(em);
1802 read_unlock(&em_tree->lock);
1804 if (!em || IS_ERR(em)) {
1808 logical = start - em->start;
1809 logical = em->block_start + logical;
1810 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1811 logical = em->block_start;
1812 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1814 failrec->logical = logical;
1815 free_extent_map(em);
1816 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1817 EXTENT_DIRTY, GFP_NOFS);
1818 set_state_private(failure_tree, start,
1819 (u64)(unsigned long)failrec);
1821 failrec = (struct io_failure_record *)(unsigned long)private;
1823 num_copies = btrfs_num_copies(
1824 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1825 failrec->logical, failrec->len);
1826 failrec->last_mirror++;
1828 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1829 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1832 if (state && state->start != failrec->start)
1834 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1836 if (!state || failrec->last_mirror > num_copies) {
1837 set_state_private(failure_tree, failrec->start, 0);
1838 clear_extent_bits(failure_tree, failrec->start,
1839 failrec->start + failrec->len - 1,
1840 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1844 bio = bio_alloc(GFP_NOFS, 1);
1845 bio->bi_private = state;
1846 bio->bi_end_io = failed_bio->bi_end_io;
1847 bio->bi_sector = failrec->logical >> 9;
1848 bio->bi_bdev = failed_bio->bi_bdev;
1851 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1852 if (failed_bio->bi_rw & (1 << BIO_RW))
1857 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1858 failrec->last_mirror,
1859 failrec->bio_flags);
1864 * each time an IO finishes, we do a fast check in the IO failure tree
1865 * to see if we need to process or clean up an io_failure_record
1867 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1870 u64 private_failure;
1871 struct io_failure_record *failure;
1875 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1876 (u64)-1, 1, EXTENT_DIRTY)) {
1877 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1878 start, &private_failure);
1880 failure = (struct io_failure_record *)(unsigned long)
1882 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1884 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1886 failure->start + failure->len - 1,
1887 EXTENT_DIRTY | EXTENT_LOCKED,
1896 * when reads are done, we need to check csums to verify the data is correct
1897 * if there's a match, we allow the bio to finish. If not, we go through
1898 * the io_failure_record routines to find good copies
1900 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1901 struct extent_state *state)
1903 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1904 struct inode *inode = page->mapping->host;
1905 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1907 u64 private = ~(u32)0;
1909 struct btrfs_root *root = BTRFS_I(inode)->root;
1912 if (PageChecked(page)) {
1913 ClearPageChecked(page);
1917 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1920 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1921 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1922 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1927 if (state && state->start == start) {
1928 private = state->private;
1931 ret = get_state_private(io_tree, start, &private);
1933 kaddr = kmap_atomic(page, KM_USER0);
1937 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1938 btrfs_csum_final(csum, (char *)&csum);
1939 if (csum != private)
1942 kunmap_atomic(kaddr, KM_USER0);
1944 /* if the io failure tree for this inode is non-empty,
1945 * check to see if we've recovered from a failed IO
1947 btrfs_clean_io_failures(inode, start);
1951 if (printk_ratelimit()) {
1952 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1953 "private %llu\n", page->mapping->host->i_ino,
1954 (unsigned long long)start, csum,
1955 (unsigned long long)private);
1957 memset(kaddr + offset, 1, end - start + 1);
1958 flush_dcache_page(page);
1959 kunmap_atomic(kaddr, KM_USER0);
1966 * This creates an orphan entry for the given inode in case something goes
1967 * wrong in the middle of an unlink/truncate.
1969 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1971 struct btrfs_root *root = BTRFS_I(inode)->root;
1974 spin_lock(&root->list_lock);
1976 /* already on the orphan list, we're good */
1977 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1978 spin_unlock(&root->list_lock);
1982 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1984 spin_unlock(&root->list_lock);
1987 * insert an orphan item to track this unlinked/truncated file
1989 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1995 * We have done the truncate/delete so we can go ahead and remove the orphan
1996 * item for this particular inode.
1998 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2000 struct btrfs_root *root = BTRFS_I(inode)->root;
2003 spin_lock(&root->list_lock);
2005 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2006 spin_unlock(&root->list_lock);
2010 list_del_init(&BTRFS_I(inode)->i_orphan);
2012 spin_unlock(&root->list_lock);
2016 spin_unlock(&root->list_lock);
2018 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2024 * this cleans up any orphans that may be left on the list from the last use
2027 void btrfs_orphan_cleanup(struct btrfs_root *root)
2029 struct btrfs_path *path;
2030 struct extent_buffer *leaf;
2031 struct btrfs_item *item;
2032 struct btrfs_key key, found_key;
2033 struct btrfs_trans_handle *trans;
2034 struct inode *inode;
2035 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2037 path = btrfs_alloc_path();
2042 key.objectid = BTRFS_ORPHAN_OBJECTID;
2043 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2044 key.offset = (u64)-1;
2048 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2050 printk(KERN_ERR "Error searching slot for orphan: %d"
2056 * if ret == 0 means we found what we were searching for, which
2057 * is weird, but possible, so only screw with path if we didnt
2058 * find the key and see if we have stuff that matches
2061 if (path->slots[0] == 0)
2066 /* pull out the item */
2067 leaf = path->nodes[0];
2068 item = btrfs_item_nr(leaf, path->slots[0]);
2069 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2071 /* make sure the item matches what we want */
2072 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2074 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2077 /* release the path since we're done with it */
2078 btrfs_release_path(root, path);
2081 * this is where we are basically btrfs_lookup, without the
2082 * crossing root thing. we store the inode number in the
2083 * offset of the orphan item.
2085 found_key.objectid = found_key.offset;
2086 found_key.type = BTRFS_INODE_ITEM_KEY;
2087 found_key.offset = 0;
2088 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
2093 * add this inode to the orphan list so btrfs_orphan_del does
2094 * the proper thing when we hit it
2096 spin_lock(&root->list_lock);
2097 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2098 spin_unlock(&root->list_lock);
2101 * if this is a bad inode, means we actually succeeded in
2102 * removing the inode, but not the orphan record, which means
2103 * we need to manually delete the orphan since iput will just
2104 * do a destroy_inode
2106 if (is_bad_inode(inode)) {
2107 trans = btrfs_start_transaction(root, 1);
2108 btrfs_orphan_del(trans, inode);
2109 btrfs_end_transaction(trans, root);
2114 /* if we have links, this was a truncate, lets do that */
2115 if (inode->i_nlink) {
2117 btrfs_truncate(inode);
2122 /* this will do delete_inode and everything for us */
2127 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2129 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2131 btrfs_free_path(path);
2135 * very simple check to peek ahead in the leaf looking for xattrs. If we
2136 * don't find any xattrs, we know there can't be any acls.
2138 * slot is the slot the inode is in, objectid is the objectid of the inode
2140 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2141 int slot, u64 objectid)
2143 u32 nritems = btrfs_header_nritems(leaf);
2144 struct btrfs_key found_key;
2148 while (slot < nritems) {
2149 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2151 /* we found a different objectid, there must not be acls */
2152 if (found_key.objectid != objectid)
2155 /* we found an xattr, assume we've got an acl */
2156 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2160 * we found a key greater than an xattr key, there can't
2161 * be any acls later on
2163 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2170 * it goes inode, inode backrefs, xattrs, extents,
2171 * so if there are a ton of hard links to an inode there can
2172 * be a lot of backrefs. Don't waste time searching too hard,
2173 * this is just an optimization
2178 /* we hit the end of the leaf before we found an xattr or
2179 * something larger than an xattr. We have to assume the inode
2186 * read an inode from the btree into the in-memory inode
2188 static void btrfs_read_locked_inode(struct inode *inode)
2190 struct btrfs_path *path;
2191 struct extent_buffer *leaf;
2192 struct btrfs_inode_item *inode_item;
2193 struct btrfs_timespec *tspec;
2194 struct btrfs_root *root = BTRFS_I(inode)->root;
2195 struct btrfs_key location;
2197 u64 alloc_group_block;
2201 path = btrfs_alloc_path();
2203 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2205 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2209 leaf = path->nodes[0];
2210 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2211 struct btrfs_inode_item);
2213 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2214 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2215 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2216 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2217 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2219 tspec = btrfs_inode_atime(inode_item);
2220 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2221 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2223 tspec = btrfs_inode_mtime(inode_item);
2224 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2225 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2227 tspec = btrfs_inode_ctime(inode_item);
2228 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2229 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2231 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2232 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2233 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2234 inode->i_generation = BTRFS_I(inode)->generation;
2236 rdev = btrfs_inode_rdev(leaf, inode_item);
2238 BTRFS_I(inode)->index_cnt = (u64)-1;
2239 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2241 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2244 * try to precache a NULL acl entry for files that don't have
2245 * any xattrs or acls
2247 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2249 cache_no_acl(inode);
2251 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2252 alloc_group_block, 0);
2253 btrfs_free_path(path);
2256 switch (inode->i_mode & S_IFMT) {
2258 inode->i_mapping->a_ops = &btrfs_aops;
2259 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2260 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2261 inode->i_fop = &btrfs_file_operations;
2262 inode->i_op = &btrfs_file_inode_operations;
2265 inode->i_fop = &btrfs_dir_file_operations;
2266 if (root == root->fs_info->tree_root)
2267 inode->i_op = &btrfs_dir_ro_inode_operations;
2269 inode->i_op = &btrfs_dir_inode_operations;
2272 inode->i_op = &btrfs_symlink_inode_operations;
2273 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2274 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2277 inode->i_op = &btrfs_special_inode_operations;
2278 init_special_inode(inode, inode->i_mode, rdev);
2282 btrfs_update_iflags(inode);
2286 btrfs_free_path(path);
2287 make_bad_inode(inode);
2291 * given a leaf and an inode, copy the inode fields into the leaf
2293 static void fill_inode_item(struct btrfs_trans_handle *trans,
2294 struct extent_buffer *leaf,
2295 struct btrfs_inode_item *item,
2296 struct inode *inode)
2298 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2299 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2300 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2301 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2302 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2304 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2305 inode->i_atime.tv_sec);
2306 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2307 inode->i_atime.tv_nsec);
2309 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2310 inode->i_mtime.tv_sec);
2311 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2312 inode->i_mtime.tv_nsec);
2314 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2315 inode->i_ctime.tv_sec);
2316 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2317 inode->i_ctime.tv_nsec);
2319 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2320 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2321 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2322 btrfs_set_inode_transid(leaf, item, trans->transid);
2323 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2324 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2325 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2329 * copy everything in the in-memory inode into the btree.
2331 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2332 struct btrfs_root *root, struct inode *inode)
2334 struct btrfs_inode_item *inode_item;
2335 struct btrfs_path *path;
2336 struct extent_buffer *leaf;
2339 path = btrfs_alloc_path();
2341 path->leave_spinning = 1;
2342 ret = btrfs_lookup_inode(trans, root, path,
2343 &BTRFS_I(inode)->location, 1);
2350 btrfs_unlock_up_safe(path, 1);
2351 leaf = path->nodes[0];
2352 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2353 struct btrfs_inode_item);
2355 fill_inode_item(trans, leaf, inode_item, inode);
2356 btrfs_mark_buffer_dirty(leaf);
2357 btrfs_set_inode_last_trans(trans, inode);
2360 btrfs_free_path(path);
2366 * unlink helper that gets used here in inode.c and in the tree logging
2367 * recovery code. It remove a link in a directory with a given name, and
2368 * also drops the back refs in the inode to the directory
2370 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2371 struct btrfs_root *root,
2372 struct inode *dir, struct inode *inode,
2373 const char *name, int name_len)
2375 struct btrfs_path *path;
2377 struct extent_buffer *leaf;
2378 struct btrfs_dir_item *di;
2379 struct btrfs_key key;
2382 path = btrfs_alloc_path();
2388 path->leave_spinning = 1;
2389 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2390 name, name_len, -1);
2399 leaf = path->nodes[0];
2400 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2401 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2404 btrfs_release_path(root, path);
2406 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2408 dir->i_ino, &index);
2410 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2411 "inode %lu parent %lu\n", name_len, name,
2412 inode->i_ino, dir->i_ino);
2416 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2417 index, name, name_len, -1);
2426 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2427 btrfs_release_path(root, path);
2429 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2431 BUG_ON(ret != 0 && ret != -ENOENT);
2433 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2437 btrfs_free_path(path);
2441 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2442 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2443 btrfs_update_inode(trans, root, dir);
2444 btrfs_drop_nlink(inode);
2445 ret = btrfs_update_inode(trans, root, inode);
2450 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2452 struct btrfs_root *root;
2453 struct btrfs_trans_handle *trans;
2454 struct inode *inode = dentry->d_inode;
2456 unsigned long nr = 0;
2458 root = BTRFS_I(dir)->root;
2460 trans = btrfs_start_transaction(root, 1);
2462 btrfs_set_trans_block_group(trans, dir);
2464 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2466 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2467 dentry->d_name.name, dentry->d_name.len);
2469 if (inode->i_nlink == 0)
2470 ret = btrfs_orphan_add(trans, inode);
2472 nr = trans->blocks_used;
2474 btrfs_end_transaction_throttle(trans, root);
2475 btrfs_btree_balance_dirty(root, nr);
2479 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2480 struct btrfs_root *root,
2481 struct inode *dir, u64 objectid,
2482 const char *name, int name_len)
2484 struct btrfs_path *path;
2485 struct extent_buffer *leaf;
2486 struct btrfs_dir_item *di;
2487 struct btrfs_key key;
2491 path = btrfs_alloc_path();
2495 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2496 name, name_len, -1);
2497 BUG_ON(!di || IS_ERR(di));
2499 leaf = path->nodes[0];
2500 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2501 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2502 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2504 btrfs_release_path(root, path);
2506 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2507 objectid, root->root_key.objectid,
2508 dir->i_ino, &index, name, name_len);
2510 BUG_ON(ret != -ENOENT);
2511 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2513 BUG_ON(!di || IS_ERR(di));
2515 leaf = path->nodes[0];
2516 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2517 btrfs_release_path(root, path);
2521 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2522 index, name, name_len, -1);
2523 BUG_ON(!di || IS_ERR(di));
2525 leaf = path->nodes[0];
2526 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2527 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2528 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2530 btrfs_release_path(root, path);
2532 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2533 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2534 ret = btrfs_update_inode(trans, root, dir);
2536 dir->i_sb->s_dirt = 1;
2538 btrfs_free_path(path);
2542 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2544 struct inode *inode = dentry->d_inode;
2547 struct btrfs_root *root = BTRFS_I(dir)->root;
2548 struct btrfs_trans_handle *trans;
2549 unsigned long nr = 0;
2551 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2552 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2555 trans = btrfs_start_transaction(root, 1);
2556 btrfs_set_trans_block_group(trans, dir);
2558 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2559 err = btrfs_unlink_subvol(trans, root, dir,
2560 BTRFS_I(inode)->location.objectid,
2561 dentry->d_name.name,
2562 dentry->d_name.len);
2566 err = btrfs_orphan_add(trans, inode);
2570 /* now the directory is empty */
2571 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2572 dentry->d_name.name, dentry->d_name.len);
2574 btrfs_i_size_write(inode, 0);
2576 nr = trans->blocks_used;
2577 ret = btrfs_end_transaction_throttle(trans, root);
2578 btrfs_btree_balance_dirty(root, nr);
2587 * when truncating bytes in a file, it is possible to avoid reading
2588 * the leaves that contain only checksum items. This can be the
2589 * majority of the IO required to delete a large file, but it must
2590 * be done carefully.
2592 * The keys in the level just above the leaves are checked to make sure
2593 * the lowest key in a given leaf is a csum key, and starts at an offset
2594 * after the new size.
2596 * Then the key for the next leaf is checked to make sure it also has
2597 * a checksum item for the same file. If it does, we know our target leaf
2598 * contains only checksum items, and it can be safely freed without reading
2601 * This is just an optimization targeted at large files. It may do
2602 * nothing. It will return 0 unless things went badly.
2604 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2605 struct btrfs_root *root,
2606 struct btrfs_path *path,
2607 struct inode *inode, u64 new_size)
2609 struct btrfs_key key;
2612 struct btrfs_key found_key;
2613 struct btrfs_key other_key;
2614 struct btrfs_leaf_ref *ref;
2618 path->lowest_level = 1;
2619 key.objectid = inode->i_ino;
2620 key.type = BTRFS_CSUM_ITEM_KEY;
2621 key.offset = new_size;
2623 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2627 if (path->nodes[1] == NULL) {
2632 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2633 nritems = btrfs_header_nritems(path->nodes[1]);
2638 if (path->slots[1] >= nritems)
2641 /* did we find a key greater than anything we want to delete? */
2642 if (found_key.objectid > inode->i_ino ||
2643 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2646 /* we check the next key in the node to make sure the leave contains
2647 * only checksum items. This comparison doesn't work if our
2648 * leaf is the last one in the node
2650 if (path->slots[1] + 1 >= nritems) {
2652 /* search forward from the last key in the node, this
2653 * will bring us into the next node in the tree
2655 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2657 /* unlikely, but we inc below, so check to be safe */
2658 if (found_key.offset == (u64)-1)
2661 /* search_forward needs a path with locks held, do the
2662 * search again for the original key. It is possible
2663 * this will race with a balance and return a path that
2664 * we could modify, but this drop is just an optimization
2665 * and is allowed to miss some leaves.
2667 btrfs_release_path(root, path);
2670 /* setup a max key for search_forward */
2671 other_key.offset = (u64)-1;
2672 other_key.type = key.type;
2673 other_key.objectid = key.objectid;
2675 path->keep_locks = 1;
2676 ret = btrfs_search_forward(root, &found_key, &other_key,
2678 path->keep_locks = 0;
2679 if (ret || found_key.objectid != key.objectid ||
2680 found_key.type != key.type) {
2685 key.offset = found_key.offset;
2686 btrfs_release_path(root, path);
2691 /* we know there's one more slot after us in the tree,
2692 * read that key so we can verify it is also a checksum item
2694 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2696 if (found_key.objectid < inode->i_ino)
2699 if (found_key.type != key.type || found_key.offset < new_size)
2703 * if the key for the next leaf isn't a csum key from this objectid,
2704 * we can't be sure there aren't good items inside this leaf.
2707 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2710 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2711 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2713 * it is safe to delete this leaf, it contains only
2714 * csum items from this inode at an offset >= new_size
2716 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2719 if (root->ref_cows && leaf_gen < trans->transid) {
2720 ref = btrfs_alloc_leaf_ref(root, 0);
2722 ref->root_gen = root->root_key.offset;
2723 ref->bytenr = leaf_start;
2725 ref->generation = leaf_gen;
2728 btrfs_sort_leaf_ref(ref);
2730 ret = btrfs_add_leaf_ref(root, ref, 0);
2732 btrfs_free_leaf_ref(root, ref);
2738 btrfs_release_path(root, path);
2740 if (other_key.objectid == inode->i_ino &&
2741 other_key.type == key.type && other_key.offset > key.offset) {
2742 key.offset = other_key.offset;
2748 /* fixup any changes we've made to the path */
2749 path->lowest_level = 0;
2750 path->keep_locks = 0;
2751 btrfs_release_path(root, path);
2758 * this can truncate away extent items, csum items and directory items.
2759 * It starts at a high offset and removes keys until it can't find
2760 * any higher than new_size
2762 * csum items that cross the new i_size are truncated to the new size
2765 * min_type is the minimum key type to truncate down to. If set to 0, this
2766 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2768 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2769 struct btrfs_root *root,
2770 struct inode *inode,
2771 u64 new_size, u32 min_type)
2774 struct btrfs_path *path;
2775 struct btrfs_key key;
2776 struct btrfs_key found_key;
2777 u32 found_type = (u8)-1;
2778 struct extent_buffer *leaf;
2779 struct btrfs_file_extent_item *fi;
2780 u64 extent_start = 0;
2781 u64 extent_num_bytes = 0;
2782 u64 extent_offset = 0;
2786 int pending_del_nr = 0;
2787 int pending_del_slot = 0;
2788 int extent_type = -1;
2790 u64 mask = root->sectorsize - 1;
2793 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2794 path = btrfs_alloc_path();
2798 /* FIXME, add redo link to tree so we don't leak on crash */
2799 key.objectid = inode->i_ino;
2800 key.offset = (u64)-1;
2804 path->leave_spinning = 1;
2805 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2810 /* there are no items in the tree for us to truncate, we're
2813 if (path->slots[0] == 0) {
2822 leaf = path->nodes[0];
2823 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2824 found_type = btrfs_key_type(&found_key);
2827 if (found_key.objectid != inode->i_ino)
2830 if (found_type < min_type)
2833 item_end = found_key.offset;
2834 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2835 fi = btrfs_item_ptr(leaf, path->slots[0],
2836 struct btrfs_file_extent_item);
2837 extent_type = btrfs_file_extent_type(leaf, fi);
2838 encoding = btrfs_file_extent_compression(leaf, fi);
2839 encoding |= btrfs_file_extent_encryption(leaf, fi);
2840 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2842 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2844 btrfs_file_extent_num_bytes(leaf, fi);
2845 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2846 item_end += btrfs_file_extent_inline_len(leaf,
2851 if (item_end < new_size) {
2852 if (found_type == BTRFS_DIR_ITEM_KEY)
2853 found_type = BTRFS_INODE_ITEM_KEY;
2854 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2855 found_type = BTRFS_EXTENT_DATA_KEY;
2856 else if (found_type == BTRFS_EXTENT_DATA_KEY)
2857 found_type = BTRFS_XATTR_ITEM_KEY;
2858 else if (found_type == BTRFS_XATTR_ITEM_KEY)
2859 found_type = BTRFS_INODE_REF_KEY;
2860 else if (found_type)
2864 btrfs_set_key_type(&key, found_type);
2867 if (found_key.offset >= new_size)
2873 /* FIXME, shrink the extent if the ref count is only 1 */
2874 if (found_type != BTRFS_EXTENT_DATA_KEY)
2877 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2879 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2880 if (!del_item && !encoding) {
2881 u64 orig_num_bytes =
2882 btrfs_file_extent_num_bytes(leaf, fi);
2883 extent_num_bytes = new_size -
2884 found_key.offset + root->sectorsize - 1;
2885 extent_num_bytes = extent_num_bytes &
2886 ~((u64)root->sectorsize - 1);
2887 btrfs_set_file_extent_num_bytes(leaf, fi,
2889 num_dec = (orig_num_bytes -
2891 if (root->ref_cows && extent_start != 0)
2892 inode_sub_bytes(inode, num_dec);
2893 btrfs_mark_buffer_dirty(leaf);
2896 btrfs_file_extent_disk_num_bytes(leaf,
2898 extent_offset = found_key.offset -
2899 btrfs_file_extent_offset(leaf, fi);
2901 /* FIXME blocksize != 4096 */
2902 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2903 if (extent_start != 0) {
2906 inode_sub_bytes(inode, num_dec);
2909 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2911 * we can't truncate inline items that have had
2915 btrfs_file_extent_compression(leaf, fi) == 0 &&
2916 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2917 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2918 u32 size = new_size - found_key.offset;
2920 if (root->ref_cows) {
2921 inode_sub_bytes(inode, item_end + 1 -
2925 btrfs_file_extent_calc_inline_size(size);
2926 ret = btrfs_truncate_item(trans, root, path,
2929 } else if (root->ref_cows) {
2930 inode_sub_bytes(inode, item_end + 1 -
2936 if (!pending_del_nr) {
2937 /* no pending yet, add ourselves */
2938 pending_del_slot = path->slots[0];
2940 } else if (pending_del_nr &&
2941 path->slots[0] + 1 == pending_del_slot) {
2942 /* hop on the pending chunk */
2944 pending_del_slot = path->slots[0];
2951 if (found_extent && root->ref_cows) {
2952 btrfs_set_path_blocking(path);
2953 ret = btrfs_free_extent(trans, root, extent_start,
2954 extent_num_bytes, 0,
2955 btrfs_header_owner(leaf),
2956 inode->i_ino, extent_offset);
2960 if (path->slots[0] == 0) {
2963 btrfs_release_path(root, path);
2964 if (found_type == BTRFS_INODE_ITEM_KEY)
2970 if (pending_del_nr &&
2971 path->slots[0] + 1 != pending_del_slot) {
2972 struct btrfs_key debug;
2974 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2976 ret = btrfs_del_items(trans, root, path,
2981 btrfs_release_path(root, path);
2982 if (found_type == BTRFS_INODE_ITEM_KEY)
2989 if (pending_del_nr) {
2990 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2993 btrfs_free_path(path);
2998 * taken from block_truncate_page, but does cow as it zeros out
2999 * any bytes left in the last page in the file.
3001 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3003 struct inode *inode = mapping->host;
3004 struct btrfs_root *root = BTRFS_I(inode)->root;
3005 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3006 struct btrfs_ordered_extent *ordered;
3008 u32 blocksize = root->sectorsize;
3009 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3010 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3016 if ((offset & (blocksize - 1)) == 0)
3021 page = grab_cache_page(mapping, index);
3025 page_start = page_offset(page);
3026 page_end = page_start + PAGE_CACHE_SIZE - 1;
3028 if (!PageUptodate(page)) {
3029 ret = btrfs_readpage(NULL, page);
3031 if (page->mapping != mapping) {
3033 page_cache_release(page);
3036 if (!PageUptodate(page)) {
3041 wait_on_page_writeback(page);
3043 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3044 set_page_extent_mapped(page);
3046 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3048 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3050 page_cache_release(page);
3051 btrfs_start_ordered_extent(inode, ordered, 1);
3052 btrfs_put_ordered_extent(ordered);
3056 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3058 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3063 if (offset != PAGE_CACHE_SIZE) {
3065 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3066 flush_dcache_page(page);
3069 ClearPageChecked(page);
3070 set_page_dirty(page);
3071 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3075 page_cache_release(page);
3080 int btrfs_cont_expand(struct inode *inode, loff_t size)
3082 struct btrfs_trans_handle *trans;
3083 struct btrfs_root *root = BTRFS_I(inode)->root;
3084 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3085 struct extent_map *em;
3086 u64 mask = root->sectorsize - 1;
3087 u64 hole_start = (inode->i_size + mask) & ~mask;
3088 u64 block_end = (size + mask) & ~mask;
3094 if (size <= hole_start)
3097 btrfs_truncate_page(inode->i_mapping, inode->i_size);
3100 struct btrfs_ordered_extent *ordered;
3101 btrfs_wait_ordered_range(inode, hole_start,
3102 block_end - hole_start);
3103 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3104 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3107 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3108 btrfs_put_ordered_extent(ordered);
3111 trans = btrfs_start_transaction(root, 1);
3112 btrfs_set_trans_block_group(trans, inode);
3114 cur_offset = hole_start;
3116 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3117 block_end - cur_offset, 0);
3118 BUG_ON(IS_ERR(em) || !em);
3119 last_byte = min(extent_map_end(em), block_end);
3120 last_byte = (last_byte + mask) & ~mask;
3121 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
3123 hole_size = last_byte - cur_offset;
3124 err = btrfs_drop_extents(trans, root, inode,
3126 cur_offset + hole_size,
3128 cur_offset, &hint_byte, 1);
3132 err = btrfs_reserve_metadata_space(root, 1);
3136 err = btrfs_insert_file_extent(trans, root,
3137 inode->i_ino, cur_offset, 0,
3138 0, hole_size, 0, hole_size,
3140 btrfs_drop_extent_cache(inode, hole_start,
3142 btrfs_unreserve_metadata_space(root, 1);
3144 free_extent_map(em);
3145 cur_offset = last_byte;
3146 if (err || cur_offset >= block_end)
3150 btrfs_end_transaction(trans, root);
3151 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3155 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3157 struct inode *inode = dentry->d_inode;
3160 err = inode_change_ok(inode, attr);
3164 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3165 if (attr->ia_size > inode->i_size) {
3166 err = btrfs_cont_expand(inode, attr->ia_size);
3169 } else if (inode->i_size > 0 &&
3170 attr->ia_size == 0) {
3172 /* we're truncating a file that used to have good
3173 * data down to zero. Make sure it gets into
3174 * the ordered flush list so that any new writes
3175 * get down to disk quickly.
3177 BTRFS_I(inode)->ordered_data_close = 1;
3181 err = inode_setattr(inode, attr);
3183 if (!err && ((attr->ia_valid & ATTR_MODE)))
3184 err = btrfs_acl_chmod(inode);
3188 void btrfs_delete_inode(struct inode *inode)
3190 struct btrfs_trans_handle *trans;
3191 struct btrfs_root *root = BTRFS_I(inode)->root;
3195 truncate_inode_pages(&inode->i_data, 0);
3196 if (is_bad_inode(inode)) {
3197 btrfs_orphan_del(NULL, inode);
3200 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3202 if (inode->i_nlink > 0) {
3203 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3207 btrfs_i_size_write(inode, 0);
3208 trans = btrfs_join_transaction(root, 1);
3210 btrfs_set_trans_block_group(trans, inode);
3211 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3213 btrfs_orphan_del(NULL, inode);
3214 goto no_delete_lock;
3217 btrfs_orphan_del(trans, inode);
3219 nr = trans->blocks_used;
3222 btrfs_end_transaction(trans, root);
3223 btrfs_btree_balance_dirty(root, nr);
3227 nr = trans->blocks_used;
3228 btrfs_end_transaction(trans, root);
3229 btrfs_btree_balance_dirty(root, nr);
3235 * this returns the key found in the dir entry in the location pointer.
3236 * If no dir entries were found, location->objectid is 0.
3238 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3239 struct btrfs_key *location)
3241 const char *name = dentry->d_name.name;
3242 int namelen = dentry->d_name.len;
3243 struct btrfs_dir_item *di;
3244 struct btrfs_path *path;
3245 struct btrfs_root *root = BTRFS_I(dir)->root;
3248 path = btrfs_alloc_path();
3251 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3256 if (!di || IS_ERR(di))
3259 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3261 btrfs_free_path(path);
3264 location->objectid = 0;
3269 * when we hit a tree root in a directory, the btrfs part of the inode
3270 * needs to be changed to reflect the root directory of the tree root. This
3271 * is kind of like crossing a mount point.
3273 static int fixup_tree_root_location(struct btrfs_root *root,
3275 struct dentry *dentry,
3276 struct btrfs_key *location,
3277 struct btrfs_root **sub_root)
3279 struct btrfs_path *path;
3280 struct btrfs_root *new_root;
3281 struct btrfs_root_ref *ref;
3282 struct extent_buffer *leaf;
3286 path = btrfs_alloc_path();
3293 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3294 BTRFS_I(dir)->root->root_key.objectid,
3295 location->objectid);
3302 leaf = path->nodes[0];
3303 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3304 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3305 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3308 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3309 (unsigned long)(ref + 1),
3310 dentry->d_name.len);
3314 btrfs_release_path(root->fs_info->tree_root, path);
3316 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3317 if (IS_ERR(new_root)) {
3318 err = PTR_ERR(new_root);
3322 if (btrfs_root_refs(&new_root->root_item) == 0) {
3327 *sub_root = new_root;
3328 location->objectid = btrfs_root_dirid(&new_root->root_item);
3329 location->type = BTRFS_INODE_ITEM_KEY;
3330 location->offset = 0;
3333 btrfs_free_path(path);
3337 static void inode_tree_add(struct inode *inode)
3339 struct btrfs_root *root = BTRFS_I(inode)->root;
3340 struct btrfs_inode *entry;
3342 struct rb_node *parent;
3344 p = &root->inode_tree.rb_node;
3347 if (hlist_unhashed(&inode->i_hash))
3350 spin_lock(&root->inode_lock);
3353 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3355 if (inode->i_ino < entry->vfs_inode.i_ino)
3356 p = &parent->rb_left;
3357 else if (inode->i_ino > entry->vfs_inode.i_ino)
3358 p = &parent->rb_right;
3360 WARN_ON(!(entry->vfs_inode.i_state &
3361 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3362 rb_erase(parent, &root->inode_tree);
3363 RB_CLEAR_NODE(parent);
3364 spin_unlock(&root->inode_lock);
3368 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3369 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3370 spin_unlock(&root->inode_lock);
3373 static void inode_tree_del(struct inode *inode)
3375 struct btrfs_root *root = BTRFS_I(inode)->root;
3378 spin_lock(&root->inode_lock);
3379 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3380 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3381 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3382 empty = RB_EMPTY_ROOT(&root->inode_tree);
3384 spin_unlock(&root->inode_lock);
3386 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3387 synchronize_srcu(&root->fs_info->subvol_srcu);
3388 spin_lock(&root->inode_lock);
3389 empty = RB_EMPTY_ROOT(&root->inode_tree);
3390 spin_unlock(&root->inode_lock);
3392 btrfs_add_dead_root(root);
3396 int btrfs_invalidate_inodes(struct btrfs_root *root)
3398 struct rb_node *node;
3399 struct rb_node *prev;
3400 struct btrfs_inode *entry;
3401 struct inode *inode;
3404 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3406 spin_lock(&root->inode_lock);
3408 node = root->inode_tree.rb_node;
3412 entry = rb_entry(node, struct btrfs_inode, rb_node);
3414 if (objectid < entry->vfs_inode.i_ino)
3415 node = node->rb_left;
3416 else if (objectid > entry->vfs_inode.i_ino)
3417 node = node->rb_right;
3423 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3424 if (objectid <= entry->vfs_inode.i_ino) {
3428 prev = rb_next(prev);
3432 entry = rb_entry(node, struct btrfs_inode, rb_node);
3433 objectid = entry->vfs_inode.i_ino + 1;
3434 inode = igrab(&entry->vfs_inode);
3436 spin_unlock(&root->inode_lock);
3437 if (atomic_read(&inode->i_count) > 1)
3438 d_prune_aliases(inode);
3440 * btrfs_drop_inode will remove it from
3441 * the inode cache when its usage count
3446 spin_lock(&root->inode_lock);
3450 if (cond_resched_lock(&root->inode_lock))
3453 node = rb_next(node);
3455 spin_unlock(&root->inode_lock);
3459 static noinline void init_btrfs_i(struct inode *inode)
3461 struct btrfs_inode *bi = BTRFS_I(inode);
3466 bi->logged_trans = 0;
3467 bi->delalloc_bytes = 0;
3468 bi->reserved_bytes = 0;
3469 bi->disk_i_size = 0;
3471 bi->index_cnt = (u64)-1;
3472 bi->last_unlink_trans = 0;
3473 bi->ordered_data_close = 0;
3474 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3475 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3476 inode->i_mapping, GFP_NOFS);
3477 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3478 inode->i_mapping, GFP_NOFS);
3479 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3480 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3481 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3482 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3483 mutex_init(&BTRFS_I(inode)->extent_mutex);
3484 mutex_init(&BTRFS_I(inode)->log_mutex);
3487 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3489 struct btrfs_iget_args *args = p;
3490 inode->i_ino = args->ino;
3491 init_btrfs_i(inode);
3492 BTRFS_I(inode)->root = args->root;
3493 btrfs_set_inode_space_info(args->root, inode);
3497 static int btrfs_find_actor(struct inode *inode, void *opaque)
3499 struct btrfs_iget_args *args = opaque;
3500 return args->ino == inode->i_ino &&
3501 args->root == BTRFS_I(inode)->root;
3504 static struct inode *btrfs_iget_locked(struct super_block *s,
3506 struct btrfs_root *root)
3508 struct inode *inode;
3509 struct btrfs_iget_args args;
3510 args.ino = objectid;
3513 inode = iget5_locked(s, objectid, btrfs_find_actor,
3514 btrfs_init_locked_inode,
3519 /* Get an inode object given its location and corresponding root.
3520 * Returns in *is_new if the inode was read from disk
3522 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3523 struct btrfs_root *root)
3525 struct inode *inode;
3527 inode = btrfs_iget_locked(s, location->objectid, root);
3529 return ERR_PTR(-ENOMEM);
3531 if (inode->i_state & I_NEW) {
3532 BTRFS_I(inode)->root = root;
3533 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3534 btrfs_read_locked_inode(inode);
3536 inode_tree_add(inode);
3537 unlock_new_inode(inode);
3543 static struct inode *new_simple_dir(struct super_block *s,
3544 struct btrfs_key *key,
3545 struct btrfs_root *root)
3547 struct inode *inode = new_inode(s);
3550 return ERR_PTR(-ENOMEM);
3552 init_btrfs_i(inode);
3554 BTRFS_I(inode)->root = root;
3555 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3556 BTRFS_I(inode)->dummy_inode = 1;
3558 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3559 inode->i_op = &simple_dir_inode_operations;
3560 inode->i_fop = &simple_dir_operations;
3561 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3562 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3567 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3569 struct inode *inode;
3570 struct btrfs_root *root = BTRFS_I(dir)->root;
3571 struct btrfs_root *sub_root = root;
3572 struct btrfs_key location;
3576 dentry->d_op = &btrfs_dentry_operations;
3578 if (dentry->d_name.len > BTRFS_NAME_LEN)
3579 return ERR_PTR(-ENAMETOOLONG);
3581 ret = btrfs_inode_by_name(dir, dentry, &location);
3584 return ERR_PTR(ret);
3586 if (location.objectid == 0)
3589 if (location.type == BTRFS_INODE_ITEM_KEY) {
3590 inode = btrfs_iget(dir->i_sb, &location, root);
3594 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3596 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3597 ret = fixup_tree_root_location(root, dir, dentry,
3598 &location, &sub_root);
3601 inode = ERR_PTR(ret);
3603 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3605 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3607 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3612 static int btrfs_dentry_delete(struct dentry *dentry)
3614 struct btrfs_root *root;
3616 if (!dentry->d_inode)
3619 root = BTRFS_I(dentry->d_inode)->root;
3620 if (btrfs_root_refs(&root->root_item) == 0)
3625 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3626 struct nameidata *nd)
3628 struct inode *inode;
3630 inode = btrfs_lookup_dentry(dir, dentry);
3632 return ERR_CAST(inode);
3634 return d_splice_alias(inode, dentry);
3637 static unsigned char btrfs_filetype_table[] = {
3638 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3641 static int btrfs_real_readdir(struct file *filp, void *dirent,
3644 struct inode *inode = filp->f_dentry->d_inode;
3645 struct btrfs_root *root = BTRFS_I(inode)->root;
3646 struct btrfs_item *item;
3647 struct btrfs_dir_item *di;
3648 struct btrfs_key key;
3649 struct btrfs_key found_key;
3650 struct btrfs_path *path;
3653 struct extent_buffer *leaf;
3656 unsigned char d_type;
3661 int key_type = BTRFS_DIR_INDEX_KEY;
3666 /* FIXME, use a real flag for deciding about the key type */
3667 if (root->fs_info->tree_root == root)
3668 key_type = BTRFS_DIR_ITEM_KEY;
3670 /* special case for "." */
3671 if (filp->f_pos == 0) {
3672 over = filldir(dirent, ".", 1,
3679 /* special case for .., just use the back ref */
3680 if (filp->f_pos == 1) {
3681 u64 pino = parent_ino(filp->f_path.dentry);
3682 over = filldir(dirent, "..", 2,
3688 path = btrfs_alloc_path();
3691 btrfs_set_key_type(&key, key_type);
3692 key.offset = filp->f_pos;
3693 key.objectid = inode->i_ino;
3695 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3701 leaf = path->nodes[0];
3702 nritems = btrfs_header_nritems(leaf);
3703 slot = path->slots[0];
3704 if (advance || slot >= nritems) {
3705 if (slot >= nritems - 1) {
3706 ret = btrfs_next_leaf(root, path);
3709 leaf = path->nodes[0];
3710 nritems = btrfs_header_nritems(leaf);
3711 slot = path->slots[0];
3719 item = btrfs_item_nr(leaf, slot);
3720 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3722 if (found_key.objectid != key.objectid)
3724 if (btrfs_key_type(&found_key) != key_type)
3726 if (found_key.offset < filp->f_pos)
3729 filp->f_pos = found_key.offset;
3731 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3733 di_total = btrfs_item_size(leaf, item);
3735 while (di_cur < di_total) {
3736 struct btrfs_key location;
3738 name_len = btrfs_dir_name_len(leaf, di);
3739 if (name_len <= sizeof(tmp_name)) {
3740 name_ptr = tmp_name;
3742 name_ptr = kmalloc(name_len, GFP_NOFS);
3748 read_extent_buffer(leaf, name_ptr,
3749 (unsigned long)(di + 1), name_len);
3751 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3752 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3754 /* is this a reference to our own snapshot? If so
3757 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3758 location.objectid == root->root_key.objectid) {
3762 over = filldir(dirent, name_ptr, name_len,
3763 found_key.offset, location.objectid,
3767 if (name_ptr != tmp_name)
3772 di_len = btrfs_dir_name_len(leaf, di) +
3773 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3775 di = (struct btrfs_dir_item *)((char *)di + di_len);
3779 /* Reached end of directory/root. Bump pos past the last item. */
3780 if (key_type == BTRFS_DIR_INDEX_KEY)
3781 filp->f_pos = INT_LIMIT(off_t);
3787 btrfs_free_path(path);
3791 int btrfs_write_inode(struct inode *inode, int wait)
3793 struct btrfs_root *root = BTRFS_I(inode)->root;
3794 struct btrfs_trans_handle *trans;
3797 if (root->fs_info->btree_inode == inode)
3801 trans = btrfs_join_transaction(root, 1);
3802 btrfs_set_trans_block_group(trans, inode);
3803 ret = btrfs_commit_transaction(trans, root);
3809 * This is somewhat expensive, updating the tree every time the
3810 * inode changes. But, it is most likely to find the inode in cache.
3811 * FIXME, needs more benchmarking...there are no reasons other than performance
3812 * to keep or drop this code.
3814 void btrfs_dirty_inode(struct inode *inode)
3816 struct btrfs_root *root = BTRFS_I(inode)->root;
3817 struct btrfs_trans_handle *trans;
3819 trans = btrfs_join_transaction(root, 1);
3820 btrfs_set_trans_block_group(trans, inode);
3821 btrfs_update_inode(trans, root, inode);
3822 btrfs_end_transaction(trans, root);
3826 * find the highest existing sequence number in a directory
3827 * and then set the in-memory index_cnt variable to reflect
3828 * free sequence numbers
3830 static int btrfs_set_inode_index_count(struct inode *inode)
3832 struct btrfs_root *root = BTRFS_I(inode)->root;
3833 struct btrfs_key key, found_key;
3834 struct btrfs_path *path;
3835 struct extent_buffer *leaf;
3838 key.objectid = inode->i_ino;
3839 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3840 key.offset = (u64)-1;
3842 path = btrfs_alloc_path();
3846 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3849 /* FIXME: we should be able to handle this */
3855 * MAGIC NUMBER EXPLANATION:
3856 * since we search a directory based on f_pos we have to start at 2
3857 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3858 * else has to start at 2
3860 if (path->slots[0] == 0) {
3861 BTRFS_I(inode)->index_cnt = 2;
3867 leaf = path->nodes[0];
3868 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3870 if (found_key.objectid != inode->i_ino ||
3871 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3872 BTRFS_I(inode)->index_cnt = 2;
3876 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3878 btrfs_free_path(path);
3883 * helper to find a free sequence number in a given directory. This current
3884 * code is very simple, later versions will do smarter things in the btree
3886 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3890 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3891 ret = btrfs_set_inode_index_count(dir);
3896 *index = BTRFS_I(dir)->index_cnt;
3897 BTRFS_I(dir)->index_cnt++;
3902 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3903 struct btrfs_root *root,
3905 const char *name, int name_len,
3906 u64 ref_objectid, u64 objectid,
3907 u64 alloc_hint, int mode, u64 *index)
3909 struct inode *inode;
3910 struct btrfs_inode_item *inode_item;
3911 struct btrfs_key *location;
3912 struct btrfs_path *path;
3913 struct btrfs_inode_ref *ref;
3914 struct btrfs_key key[2];
3920 path = btrfs_alloc_path();
3923 inode = new_inode(root->fs_info->sb);
3925 return ERR_PTR(-ENOMEM);
3928 ret = btrfs_set_inode_index(dir, index);
3931 return ERR_PTR(ret);
3935 * index_cnt is ignored for everything but a dir,
3936 * btrfs_get_inode_index_count has an explanation for the magic
3939 init_btrfs_i(inode);
3940 BTRFS_I(inode)->index_cnt = 2;
3941 BTRFS_I(inode)->root = root;
3942 BTRFS_I(inode)->generation = trans->transid;
3943 btrfs_set_inode_space_info(root, inode);
3949 BTRFS_I(inode)->block_group =
3950 btrfs_find_block_group(root, 0, alloc_hint, owner);
3952 key[0].objectid = objectid;
3953 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3956 key[1].objectid = objectid;
3957 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3958 key[1].offset = ref_objectid;
3960 sizes[0] = sizeof(struct btrfs_inode_item);
3961 sizes[1] = name_len + sizeof(*ref);
3963 path->leave_spinning = 1;
3964 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3968 inode->i_uid = current_fsuid();
3970 if (dir && (dir->i_mode & S_ISGID)) {
3971 inode->i_gid = dir->i_gid;
3975 inode->i_gid = current_fsgid();
3977 inode->i_mode = mode;
3978 inode->i_ino = objectid;
3979 inode_set_bytes(inode, 0);
3980 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3981 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3982 struct btrfs_inode_item);
3983 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3985 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3986 struct btrfs_inode_ref);
3987 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3988 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3989 ptr = (unsigned long)(ref + 1);
3990 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3992 btrfs_mark_buffer_dirty(path->nodes[0]);
3993 btrfs_free_path(path);
3995 location = &BTRFS_I(inode)->location;
3996 location->objectid = objectid;
3997 location->offset = 0;
3998 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4000 btrfs_inherit_iflags(inode, dir);
4002 if ((mode & S_IFREG)) {
4003 if (btrfs_test_opt(root, NODATASUM))
4004 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4005 if (btrfs_test_opt(root, NODATACOW))
4006 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4009 insert_inode_hash(inode);
4010 inode_tree_add(inode);
4014 BTRFS_I(dir)->index_cnt--;
4015 btrfs_free_path(path);
4017 return ERR_PTR(ret);
4020 static inline u8 btrfs_inode_type(struct inode *inode)
4022 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4026 * utility function to add 'inode' into 'parent_inode' with
4027 * a give name and a given sequence number.
4028 * if 'add_backref' is true, also insert a backref from the
4029 * inode to the parent directory.
4031 int btrfs_add_link(struct btrfs_trans_handle *trans,
4032 struct inode *parent_inode, struct inode *inode,
4033 const char *name, int name_len, int add_backref, u64 index)
4036 struct btrfs_key key;
4037 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4039 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4040 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4042 key.objectid = inode->i_ino;
4043 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4047 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4048 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4049 key.objectid, root->root_key.objectid,
4050 parent_inode->i_ino,
4051 index, name, name_len);
4052 } else if (add_backref) {
4053 ret = btrfs_insert_inode_ref(trans, root,
4054 name, name_len, inode->i_ino,
4055 parent_inode->i_ino, index);
4059 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4060 parent_inode->i_ino, &key,
4061 btrfs_inode_type(inode), index);
4064 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4066 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4067 ret = btrfs_update_inode(trans, root, parent_inode);
4072 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4073 struct dentry *dentry, struct inode *inode,
4074 int backref, u64 index)
4076 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4077 inode, dentry->d_name.name,
4078 dentry->d_name.len, backref, index);
4080 d_instantiate(dentry, inode);
4088 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4089 int mode, dev_t rdev)
4091 struct btrfs_trans_handle *trans;
4092 struct btrfs_root *root = BTRFS_I(dir)->root;
4093 struct inode *inode = NULL;
4097 unsigned long nr = 0;
4100 if (!new_valid_dev(rdev))
4104 * 2 for inode item and ref
4106 * 1 for xattr if selinux is on
4108 err = btrfs_reserve_metadata_space(root, 5);
4112 trans = btrfs_start_transaction(root, 1);
4115 btrfs_set_trans_block_group(trans, dir);
4117 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4123 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4125 dentry->d_parent->d_inode->i_ino, objectid,
4126 BTRFS_I(dir)->block_group, mode, &index);
4127 err = PTR_ERR(inode);
4131 err = btrfs_init_inode_security(inode, dir);
4137 btrfs_set_trans_block_group(trans, inode);
4138 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4142 inode->i_op = &btrfs_special_inode_operations;
4143 init_special_inode(inode, inode->i_mode, rdev);
4144 btrfs_update_inode(trans, root, inode);
4146 btrfs_update_inode_block_group(trans, inode);
4147 btrfs_update_inode_block_group(trans, dir);
4149 nr = trans->blocks_used;
4150 btrfs_end_transaction_throttle(trans, root);
4152 btrfs_unreserve_metadata_space(root, 5);
4154 inode_dec_link_count(inode);
4157 btrfs_btree_balance_dirty(root, nr);
4161 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4162 int mode, struct nameidata *nd)
4164 struct btrfs_trans_handle *trans;
4165 struct btrfs_root *root = BTRFS_I(dir)->root;
4166 struct inode *inode = NULL;
4169 unsigned long nr = 0;
4174 * 2 for inode item and ref
4176 * 1 for xattr if selinux is on
4178 err = btrfs_reserve_metadata_space(root, 5);
4182 trans = btrfs_start_transaction(root, 1);
4185 btrfs_set_trans_block_group(trans, dir);
4187 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4193 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4195 dentry->d_parent->d_inode->i_ino,
4196 objectid, BTRFS_I(dir)->block_group, mode,
4198 err = PTR_ERR(inode);
4202 err = btrfs_init_inode_security(inode, dir);
4208 btrfs_set_trans_block_group(trans, inode);
4209 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4213 inode->i_mapping->a_ops = &btrfs_aops;
4214 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4215 inode->i_fop = &btrfs_file_operations;
4216 inode->i_op = &btrfs_file_inode_operations;
4217 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4219 btrfs_update_inode_block_group(trans, inode);
4220 btrfs_update_inode_block_group(trans, dir);
4222 nr = trans->blocks_used;
4223 btrfs_end_transaction_throttle(trans, root);
4225 btrfs_unreserve_metadata_space(root, 5);
4227 inode_dec_link_count(inode);
4230 btrfs_btree_balance_dirty(root, nr);
4234 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4235 struct dentry *dentry)
4237 struct btrfs_trans_handle *trans;
4238 struct btrfs_root *root = BTRFS_I(dir)->root;
4239 struct inode *inode = old_dentry->d_inode;
4241 unsigned long nr = 0;
4245 if (inode->i_nlink == 0)
4249 * 1 item for inode ref
4250 * 2 items for dir items
4252 err = btrfs_reserve_metadata_space(root, 3);
4256 btrfs_inc_nlink(inode);
4258 err = btrfs_set_inode_index(dir, &index);
4262 trans = btrfs_start_transaction(root, 1);
4264 btrfs_set_trans_block_group(trans, dir);
4265 atomic_inc(&inode->i_count);
4267 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4272 btrfs_update_inode_block_group(trans, dir);
4273 err = btrfs_update_inode(trans, root, inode);
4275 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4278 nr = trans->blocks_used;
4279 btrfs_end_transaction_throttle(trans, root);
4281 btrfs_unreserve_metadata_space(root, 3);
4283 inode_dec_link_count(inode);
4286 btrfs_btree_balance_dirty(root, nr);
4290 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4292 struct inode *inode = NULL;
4293 struct btrfs_trans_handle *trans;
4294 struct btrfs_root *root = BTRFS_I(dir)->root;
4296 int drop_on_err = 0;
4299 unsigned long nr = 1;
4302 * 2 items for inode and ref
4303 * 2 items for dir items
4304 * 1 for xattr if selinux is on
4306 err = btrfs_reserve_metadata_space(root, 5);
4310 trans = btrfs_start_transaction(root, 1);
4315 btrfs_set_trans_block_group(trans, dir);
4317 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4323 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4325 dentry->d_parent->d_inode->i_ino, objectid,
4326 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4328 if (IS_ERR(inode)) {
4329 err = PTR_ERR(inode);
4335 err = btrfs_init_inode_security(inode, dir);
4339 inode->i_op = &btrfs_dir_inode_operations;
4340 inode->i_fop = &btrfs_dir_file_operations;
4341 btrfs_set_trans_block_group(trans, inode);
4343 btrfs_i_size_write(inode, 0);
4344 err = btrfs_update_inode(trans, root, inode);
4348 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4349 inode, dentry->d_name.name,
4350 dentry->d_name.len, 0, index);
4354 d_instantiate(dentry, inode);
4356 btrfs_update_inode_block_group(trans, inode);
4357 btrfs_update_inode_block_group(trans, dir);
4360 nr = trans->blocks_used;
4361 btrfs_end_transaction_throttle(trans, root);
4364 btrfs_unreserve_metadata_space(root, 5);
4367 btrfs_btree_balance_dirty(root, nr);
4371 /* helper for btfs_get_extent. Given an existing extent in the tree,
4372 * and an extent that you want to insert, deal with overlap and insert
4373 * the new extent into the tree.
4375 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4376 struct extent_map *existing,
4377 struct extent_map *em,
4378 u64 map_start, u64 map_len)
4382 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4383 start_diff = map_start - em->start;
4384 em->start = map_start;
4386 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4387 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4388 em->block_start += start_diff;
4389 em->block_len -= start_diff;
4391 return add_extent_mapping(em_tree, em);
4394 static noinline int uncompress_inline(struct btrfs_path *path,
4395 struct inode *inode, struct page *page,
4396 size_t pg_offset, u64 extent_offset,
4397 struct btrfs_file_extent_item *item)
4400 struct extent_buffer *leaf = path->nodes[0];
4403 unsigned long inline_size;
4406 WARN_ON(pg_offset != 0);
4407 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4408 inline_size = btrfs_file_extent_inline_item_len(leaf,
4409 btrfs_item_nr(leaf, path->slots[0]));
4410 tmp = kmalloc(inline_size, GFP_NOFS);
4411 ptr = btrfs_file_extent_inline_start(item);
4413 read_extent_buffer(leaf, tmp, ptr, inline_size);
4415 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4416 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4417 inline_size, max_size);
4419 char *kaddr = kmap_atomic(page, KM_USER0);
4420 unsigned long copy_size = min_t(u64,
4421 PAGE_CACHE_SIZE - pg_offset,
4422 max_size - extent_offset);
4423 memset(kaddr + pg_offset, 0, copy_size);
4424 kunmap_atomic(kaddr, KM_USER0);
4431 * a bit scary, this does extent mapping from logical file offset to the disk.
4432 * the ugly parts come from merging extents from the disk with the in-ram
4433 * representation. This gets more complex because of the data=ordered code,
4434 * where the in-ram extents might be locked pending data=ordered completion.
4436 * This also copies inline extents directly into the page.
4439 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4440 size_t pg_offset, u64 start, u64 len,
4446 u64 extent_start = 0;
4448 u64 objectid = inode->i_ino;
4450 struct btrfs_path *path = NULL;
4451 struct btrfs_root *root = BTRFS_I(inode)->root;
4452 struct btrfs_file_extent_item *item;
4453 struct extent_buffer *leaf;
4454 struct btrfs_key found_key;
4455 struct extent_map *em = NULL;
4456 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4457 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4458 struct btrfs_trans_handle *trans = NULL;
4462 read_lock(&em_tree->lock);
4463 em = lookup_extent_mapping(em_tree, start, len);
4465 em->bdev = root->fs_info->fs_devices->latest_bdev;
4466 read_unlock(&em_tree->lock);
4469 if (em->start > start || em->start + em->len <= start)
4470 free_extent_map(em);
4471 else if (em->block_start == EXTENT_MAP_INLINE && page)
4472 free_extent_map(em);
4476 em = alloc_extent_map(GFP_NOFS);
4481 em->bdev = root->fs_info->fs_devices->latest_bdev;
4482 em->start = EXTENT_MAP_HOLE;
4483 em->orig_start = EXTENT_MAP_HOLE;
4485 em->block_len = (u64)-1;
4488 path = btrfs_alloc_path();
4492 ret = btrfs_lookup_file_extent(trans, root, path,
4493 objectid, start, trans != NULL);
4500 if (path->slots[0] == 0)
4505 leaf = path->nodes[0];
4506 item = btrfs_item_ptr(leaf, path->slots[0],
4507 struct btrfs_file_extent_item);
4508 /* are we inside the extent that was found? */
4509 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4510 found_type = btrfs_key_type(&found_key);
4511 if (found_key.objectid != objectid ||
4512 found_type != BTRFS_EXTENT_DATA_KEY) {
4516 found_type = btrfs_file_extent_type(leaf, item);
4517 extent_start = found_key.offset;
4518 compressed = btrfs_file_extent_compression(leaf, item);
4519 if (found_type == BTRFS_FILE_EXTENT_REG ||
4520 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4521 extent_end = extent_start +
4522 btrfs_file_extent_num_bytes(leaf, item);
4523 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4525 size = btrfs_file_extent_inline_len(leaf, item);
4526 extent_end = (extent_start + size + root->sectorsize - 1) &
4527 ~((u64)root->sectorsize - 1);
4530 if (start >= extent_end) {
4532 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4533 ret = btrfs_next_leaf(root, path);
4540 leaf = path->nodes[0];
4542 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4543 if (found_key.objectid != objectid ||
4544 found_key.type != BTRFS_EXTENT_DATA_KEY)
4546 if (start + len <= found_key.offset)
4549 em->len = found_key.offset - start;
4553 if (found_type == BTRFS_FILE_EXTENT_REG ||
4554 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4555 em->start = extent_start;
4556 em->len = extent_end - extent_start;
4557 em->orig_start = extent_start -
4558 btrfs_file_extent_offset(leaf, item);
4559 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4561 em->block_start = EXTENT_MAP_HOLE;
4565 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4566 em->block_start = bytenr;
4567 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4570 bytenr += btrfs_file_extent_offset(leaf, item);
4571 em->block_start = bytenr;
4572 em->block_len = em->len;
4573 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4574 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4577 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4581 size_t extent_offset;
4584 em->block_start = EXTENT_MAP_INLINE;
4585 if (!page || create) {
4586 em->start = extent_start;
4587 em->len = extent_end - extent_start;
4591 size = btrfs_file_extent_inline_len(leaf, item);
4592 extent_offset = page_offset(page) + pg_offset - extent_start;
4593 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4594 size - extent_offset);
4595 em->start = extent_start + extent_offset;
4596 em->len = (copy_size + root->sectorsize - 1) &
4597 ~((u64)root->sectorsize - 1);
4598 em->orig_start = EXTENT_MAP_INLINE;
4600 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4601 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4602 if (create == 0 && !PageUptodate(page)) {
4603 if (btrfs_file_extent_compression(leaf, item) ==
4604 BTRFS_COMPRESS_ZLIB) {
4605 ret = uncompress_inline(path, inode, page,
4607 extent_offset, item);
4611 read_extent_buffer(leaf, map + pg_offset, ptr,
4613 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4614 memset(map + pg_offset + copy_size, 0,
4615 PAGE_CACHE_SIZE - pg_offset -
4620 flush_dcache_page(page);
4621 } else if (create && PageUptodate(page)) {
4624 free_extent_map(em);
4626 btrfs_release_path(root, path);
4627 trans = btrfs_join_transaction(root, 1);
4631 write_extent_buffer(leaf, map + pg_offset, ptr,
4634 btrfs_mark_buffer_dirty(leaf);
4636 set_extent_uptodate(io_tree, em->start,
4637 extent_map_end(em) - 1, GFP_NOFS);
4640 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4647 em->block_start = EXTENT_MAP_HOLE;
4648 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4650 btrfs_release_path(root, path);
4651 if (em->start > start || extent_map_end(em) <= start) {
4652 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4653 "[%llu %llu]\n", (unsigned long long)em->start,
4654 (unsigned long long)em->len,
4655 (unsigned long long)start,
4656 (unsigned long long)len);
4662 write_lock(&em_tree->lock);
4663 ret = add_extent_mapping(em_tree, em);
4664 /* it is possible that someone inserted the extent into the tree
4665 * while we had the lock dropped. It is also possible that
4666 * an overlapping map exists in the tree
4668 if (ret == -EEXIST) {
4669 struct extent_map *existing;
4673 existing = lookup_extent_mapping(em_tree, start, len);
4674 if (existing && (existing->start > start ||
4675 existing->start + existing->len <= start)) {
4676 free_extent_map(existing);
4680 existing = lookup_extent_mapping(em_tree, em->start,
4683 err = merge_extent_mapping(em_tree, existing,
4686 free_extent_map(existing);
4688 free_extent_map(em);
4693 free_extent_map(em);
4697 free_extent_map(em);
4702 write_unlock(&em_tree->lock);
4705 btrfs_free_path(path);
4707 ret = btrfs_end_transaction(trans, root);
4712 free_extent_map(em);
4713 return ERR_PTR(err);
4718 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4719 const struct iovec *iov, loff_t offset,
4720 unsigned long nr_segs)
4725 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4726 __u64 start, __u64 len)
4728 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4731 int btrfs_readpage(struct file *file, struct page *page)
4733 struct extent_io_tree *tree;
4734 tree = &BTRFS_I(page->mapping->host)->io_tree;
4735 return extent_read_full_page(tree, page, btrfs_get_extent);
4738 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4740 struct extent_io_tree *tree;
4743 if (current->flags & PF_MEMALLOC) {
4744 redirty_page_for_writepage(wbc, page);
4748 tree = &BTRFS_I(page->mapping->host)->io_tree;
4749 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4752 int btrfs_writepages(struct address_space *mapping,
4753 struct writeback_control *wbc)
4755 struct extent_io_tree *tree;
4757 tree = &BTRFS_I(mapping->host)->io_tree;
4758 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4762 btrfs_readpages(struct file *file, struct address_space *mapping,
4763 struct list_head *pages, unsigned nr_pages)
4765 struct extent_io_tree *tree;
4766 tree = &BTRFS_I(mapping->host)->io_tree;
4767 return extent_readpages(tree, mapping, pages, nr_pages,
4770 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4772 struct extent_io_tree *tree;
4773 struct extent_map_tree *map;
4776 tree = &BTRFS_I(page->mapping->host)->io_tree;
4777 map = &BTRFS_I(page->mapping->host)->extent_tree;
4778 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4780 ClearPagePrivate(page);
4781 set_page_private(page, 0);
4782 page_cache_release(page);
4787 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4789 if (PageWriteback(page) || PageDirty(page))
4791 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4794 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4796 struct extent_io_tree *tree;
4797 struct btrfs_ordered_extent *ordered;
4798 u64 page_start = page_offset(page);
4799 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4803 * we have the page locked, so new writeback can't start,
4804 * and the dirty bit won't be cleared while we are here.
4806 * Wait for IO on this page so that we can safely clear
4807 * the PagePrivate2 bit and do ordered accounting
4809 wait_on_page_writeback(page);
4811 tree = &BTRFS_I(page->mapping->host)->io_tree;
4813 btrfs_releasepage(page, GFP_NOFS);
4816 lock_extent(tree, page_start, page_end, GFP_NOFS);
4817 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4821 * IO on this page will never be started, so we need
4822 * to account for any ordered extents now
4824 clear_extent_bit(tree, page_start, page_end,
4825 EXTENT_DIRTY | EXTENT_DELALLOC |
4826 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
4828 * whoever cleared the private bit is responsible
4829 * for the finish_ordered_io
4831 if (TestClearPagePrivate2(page)) {
4832 btrfs_finish_ordered_io(page->mapping->host,
4833 page_start, page_end);
4835 btrfs_put_ordered_extent(ordered);
4836 lock_extent(tree, page_start, page_end, GFP_NOFS);
4838 clear_extent_bit(tree, page_start, page_end,
4839 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
4840 1, 1, NULL, GFP_NOFS);
4841 __btrfs_releasepage(page, GFP_NOFS);
4843 ClearPageChecked(page);
4844 if (PagePrivate(page)) {
4845 ClearPagePrivate(page);
4846 set_page_private(page, 0);
4847 page_cache_release(page);
4852 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4853 * called from a page fault handler when a page is first dirtied. Hence we must
4854 * be careful to check for EOF conditions here. We set the page up correctly
4855 * for a written page which means we get ENOSPC checking when writing into
4856 * holes and correct delalloc and unwritten extent mapping on filesystems that
4857 * support these features.
4859 * We are not allowed to take the i_mutex here so we have to play games to
4860 * protect against truncate races as the page could now be beyond EOF. Because
4861 * vmtruncate() writes the inode size before removing pages, once we have the
4862 * page lock we can determine safely if the page is beyond EOF. If it is not
4863 * beyond EOF, then the page is guaranteed safe against truncation until we
4866 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4868 struct page *page = vmf->page;
4869 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4870 struct btrfs_root *root = BTRFS_I(inode)->root;
4871 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4872 struct btrfs_ordered_extent *ordered;
4874 unsigned long zero_start;
4880 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4884 else /* -ENOSPC, -EIO, etc */
4885 ret = VM_FAULT_SIGBUS;
4889 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
4891 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4892 ret = VM_FAULT_SIGBUS;
4896 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4899 size = i_size_read(inode);
4900 page_start = page_offset(page);
4901 page_end = page_start + PAGE_CACHE_SIZE - 1;
4903 if ((page->mapping != inode->i_mapping) ||
4904 (page_start >= size)) {
4905 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4906 /* page got truncated out from underneath us */
4909 wait_on_page_writeback(page);
4911 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4912 set_page_extent_mapped(page);
4915 * we can't set the delalloc bits if there are pending ordered
4916 * extents. Drop our locks and wait for them to finish
4918 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4920 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4922 btrfs_start_ordered_extent(inode, ordered, 1);
4923 btrfs_put_ordered_extent(ordered);
4928 * XXX - page_mkwrite gets called every time the page is dirtied, even
4929 * if it was already dirty, so for space accounting reasons we need to
4930 * clear any delalloc bits for the range we are fixing to save. There
4931 * is probably a better way to do this, but for now keep consistent with
4932 * prepare_pages in the normal write path.
4934 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
4935 EXTENT_DIRTY | EXTENT_DELALLOC, GFP_NOFS);
4937 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
4939 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4940 ret = VM_FAULT_SIGBUS;
4941 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4946 /* page is wholly or partially inside EOF */
4947 if (page_start + PAGE_CACHE_SIZE > size)
4948 zero_start = size & ~PAGE_CACHE_MASK;
4950 zero_start = PAGE_CACHE_SIZE;
4952 if (zero_start != PAGE_CACHE_SIZE) {
4954 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4955 flush_dcache_page(page);
4958 ClearPageChecked(page);
4959 set_page_dirty(page);
4960 SetPageUptodate(page);
4962 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4963 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4966 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
4968 return VM_FAULT_LOCKED;
4974 static void btrfs_truncate(struct inode *inode)
4976 struct btrfs_root *root = BTRFS_I(inode)->root;
4978 struct btrfs_trans_handle *trans;
4980 u64 mask = root->sectorsize - 1;
4982 if (!S_ISREG(inode->i_mode))
4984 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4987 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4988 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4990 trans = btrfs_start_transaction(root, 1);
4993 * setattr is responsible for setting the ordered_data_close flag,
4994 * but that is only tested during the last file release. That
4995 * could happen well after the next commit, leaving a great big
4996 * window where new writes may get lost if someone chooses to write
4997 * to this file after truncating to zero
4999 * The inode doesn't have any dirty data here, and so if we commit
5000 * this is a noop. If someone immediately starts writing to the inode
5001 * it is very likely we'll catch some of their writes in this
5002 * transaction, and the commit will find this file on the ordered
5003 * data list with good things to send down.
5005 * This is a best effort solution, there is still a window where
5006 * using truncate to replace the contents of the file will
5007 * end up with a zero length file after a crash.
5009 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5010 btrfs_add_ordered_operation(trans, root, inode);
5012 btrfs_set_trans_block_group(trans, inode);
5013 btrfs_i_size_write(inode, inode->i_size);
5015 ret = btrfs_orphan_add(trans, inode);
5018 /* FIXME, add redo link to tree so we don't leak on crash */
5019 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
5020 BTRFS_EXTENT_DATA_KEY);
5021 btrfs_update_inode(trans, root, inode);
5023 ret = btrfs_orphan_del(trans, inode);
5027 nr = trans->blocks_used;
5028 ret = btrfs_end_transaction_throttle(trans, root);
5030 btrfs_btree_balance_dirty(root, nr);
5034 * create a new subvolume directory/inode (helper for the ioctl).
5036 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5037 struct btrfs_root *new_root,
5038 u64 new_dirid, u64 alloc_hint)
5040 struct inode *inode;
5044 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5045 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5047 return PTR_ERR(inode);
5048 inode->i_op = &btrfs_dir_inode_operations;
5049 inode->i_fop = &btrfs_dir_file_operations;
5052 btrfs_i_size_write(inode, 0);
5054 err = btrfs_update_inode(trans, new_root, inode);
5061 /* helper function for file defrag and space balancing. This
5062 * forces readahead on a given range of bytes in an inode
5064 unsigned long btrfs_force_ra(struct address_space *mapping,
5065 struct file_ra_state *ra, struct file *file,
5066 pgoff_t offset, pgoff_t last_index)
5068 pgoff_t req_size = last_index - offset + 1;
5070 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5071 return offset + req_size;
5074 struct inode *btrfs_alloc_inode(struct super_block *sb)
5076 struct btrfs_inode *ei;
5078 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5082 ei->logged_trans = 0;
5083 ei->delalloc_extents = 0;
5084 ei->delalloc_reserved_extents = 0;
5085 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5086 INIT_LIST_HEAD(&ei->i_orphan);
5087 INIT_LIST_HEAD(&ei->ordered_operations);
5088 return &ei->vfs_inode;
5091 void btrfs_destroy_inode(struct inode *inode)
5093 struct btrfs_ordered_extent *ordered;
5094 struct btrfs_root *root = BTRFS_I(inode)->root;
5096 WARN_ON(!list_empty(&inode->i_dentry));
5097 WARN_ON(inode->i_data.nrpages);
5100 * Make sure we're properly removed from the ordered operation
5104 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5105 spin_lock(&root->fs_info->ordered_extent_lock);
5106 list_del_init(&BTRFS_I(inode)->ordered_operations);
5107 spin_unlock(&root->fs_info->ordered_extent_lock);
5110 spin_lock(&root->list_lock);
5111 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5112 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
5113 " list\n", inode->i_ino);
5116 spin_unlock(&root->list_lock);
5119 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5123 printk(KERN_ERR "btrfs found ordered "
5124 "extent %llu %llu on inode cleanup\n",
5125 (unsigned long long)ordered->file_offset,
5126 (unsigned long long)ordered->len);
5127 btrfs_remove_ordered_extent(inode, ordered);
5128 btrfs_put_ordered_extent(ordered);
5129 btrfs_put_ordered_extent(ordered);
5132 inode_tree_del(inode);
5133 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5134 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5137 void btrfs_drop_inode(struct inode *inode)
5139 struct btrfs_root *root = BTRFS_I(inode)->root;
5141 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5142 generic_delete_inode(inode);
5144 generic_drop_inode(inode);
5147 static void init_once(void *foo)
5149 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5151 inode_init_once(&ei->vfs_inode);
5154 void btrfs_destroy_cachep(void)
5156 if (btrfs_inode_cachep)
5157 kmem_cache_destroy(btrfs_inode_cachep);
5158 if (btrfs_trans_handle_cachep)
5159 kmem_cache_destroy(btrfs_trans_handle_cachep);
5160 if (btrfs_transaction_cachep)
5161 kmem_cache_destroy(btrfs_transaction_cachep);
5162 if (btrfs_path_cachep)
5163 kmem_cache_destroy(btrfs_path_cachep);
5166 int btrfs_init_cachep(void)
5168 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5169 sizeof(struct btrfs_inode), 0,
5170 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5171 if (!btrfs_inode_cachep)
5174 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5175 sizeof(struct btrfs_trans_handle), 0,
5176 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5177 if (!btrfs_trans_handle_cachep)
5180 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5181 sizeof(struct btrfs_transaction), 0,
5182 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5183 if (!btrfs_transaction_cachep)
5186 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5187 sizeof(struct btrfs_path), 0,
5188 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5189 if (!btrfs_path_cachep)
5194 btrfs_destroy_cachep();
5198 static int btrfs_getattr(struct vfsmount *mnt,
5199 struct dentry *dentry, struct kstat *stat)
5201 struct inode *inode = dentry->d_inode;
5202 generic_fillattr(inode, stat);
5203 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5204 stat->blksize = PAGE_CACHE_SIZE;
5205 stat->blocks = (inode_get_bytes(inode) +
5206 BTRFS_I(inode)->delalloc_bytes) >> 9;
5210 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5211 struct inode *new_dir, struct dentry *new_dentry)
5213 struct btrfs_trans_handle *trans;
5214 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5215 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5216 struct inode *new_inode = new_dentry->d_inode;
5217 struct inode *old_inode = old_dentry->d_inode;
5218 struct timespec ctime = CURRENT_TIME;
5223 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5226 /* we only allow rename subvolume link between subvolumes */
5227 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5230 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5231 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5234 if (S_ISDIR(old_inode->i_mode) && new_inode &&
5235 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5239 * 2 items for dir items
5240 * 1 item for orphan entry
5243 ret = btrfs_reserve_metadata_space(root, 4);
5248 * we're using rename to replace one file with another.
5249 * and the replacement file is large. Start IO on it now so
5250 * we don't add too much work to the end of the transaction
5252 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5253 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5254 filemap_flush(old_inode->i_mapping);
5256 /* close the racy window with snapshot create/destroy ioctl */
5257 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5258 down_read(&root->fs_info->subvol_sem);
5260 trans = btrfs_start_transaction(root, 1);
5261 btrfs_set_trans_block_group(trans, new_dir);
5264 btrfs_record_root_in_trans(trans, dest);
5266 ret = btrfs_set_inode_index(new_dir, &index);
5270 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5271 /* force full log commit if subvolume involved. */
5272 root->fs_info->last_trans_log_full_commit = trans->transid;
5274 ret = btrfs_insert_inode_ref(trans, dest,
5275 new_dentry->d_name.name,
5276 new_dentry->d_name.len,
5278 new_dir->i_ino, index);
5282 * this is an ugly little race, but the rename is required
5283 * to make sure that if we crash, the inode is either at the
5284 * old name or the new one. pinning the log transaction lets
5285 * us make sure we don't allow a log commit to come in after
5286 * we unlink the name but before we add the new name back in.
5288 btrfs_pin_log_trans(root);
5291 * make sure the inode gets flushed if it is replacing
5294 if (new_inode && new_inode->i_size &&
5295 old_inode && S_ISREG(old_inode->i_mode)) {
5296 btrfs_add_ordered_operation(trans, root, old_inode);
5299 old_dir->i_ctime = old_dir->i_mtime = ctime;
5300 new_dir->i_ctime = new_dir->i_mtime = ctime;
5301 old_inode->i_ctime = ctime;
5303 if (old_dentry->d_parent != new_dentry->d_parent)
5304 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5306 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5307 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5308 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5309 old_dentry->d_name.name,
5310 old_dentry->d_name.len);
5312 btrfs_inc_nlink(old_dentry->d_inode);
5313 ret = btrfs_unlink_inode(trans, root, old_dir,
5314 old_dentry->d_inode,
5315 old_dentry->d_name.name,
5316 old_dentry->d_name.len);
5321 new_inode->i_ctime = CURRENT_TIME;
5322 if (unlikely(new_inode->i_ino ==
5323 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5324 root_objectid = BTRFS_I(new_inode)->location.objectid;
5325 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5327 new_dentry->d_name.name,
5328 new_dentry->d_name.len);
5329 BUG_ON(new_inode->i_nlink == 0);
5331 ret = btrfs_unlink_inode(trans, dest, new_dir,
5332 new_dentry->d_inode,
5333 new_dentry->d_name.name,
5334 new_dentry->d_name.len);
5337 if (new_inode->i_nlink == 0) {
5338 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5343 ret = btrfs_add_link(trans, new_dir, old_inode,
5344 new_dentry->d_name.name,
5345 new_dentry->d_name.len, 0, index);
5348 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5349 btrfs_log_new_name(trans, old_inode, old_dir,
5350 new_dentry->d_parent);
5351 btrfs_end_log_trans(root);
5354 btrfs_end_transaction_throttle(trans, root);
5356 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5357 up_read(&root->fs_info->subvol_sem);
5359 btrfs_unreserve_metadata_space(root, 4);
5364 * some fairly slow code that needs optimization. This walks the list
5365 * of all the inodes with pending delalloc and forces them to disk.
5367 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
5369 struct list_head *head = &root->fs_info->delalloc_inodes;
5370 struct btrfs_inode *binode;
5371 struct inode *inode;
5373 if (root->fs_info->sb->s_flags & MS_RDONLY)
5376 spin_lock(&root->fs_info->delalloc_lock);
5377 while (!list_empty(head)) {
5378 binode = list_entry(head->next, struct btrfs_inode,
5380 inode = igrab(&binode->vfs_inode);
5382 list_del_init(&binode->delalloc_inodes);
5383 spin_unlock(&root->fs_info->delalloc_lock);
5385 filemap_flush(inode->i_mapping);
5389 spin_lock(&root->fs_info->delalloc_lock);
5391 spin_unlock(&root->fs_info->delalloc_lock);
5393 /* the filemap_flush will queue IO into the worker threads, but
5394 * we have to make sure the IO is actually started and that
5395 * ordered extents get created before we return
5397 atomic_inc(&root->fs_info->async_submit_draining);
5398 while (atomic_read(&root->fs_info->nr_async_submits) ||
5399 atomic_read(&root->fs_info->async_delalloc_pages)) {
5400 wait_event(root->fs_info->async_submit_wait,
5401 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5402 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5404 atomic_dec(&root->fs_info->async_submit_draining);
5408 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5409 const char *symname)
5411 struct btrfs_trans_handle *trans;
5412 struct btrfs_root *root = BTRFS_I(dir)->root;
5413 struct btrfs_path *path;
5414 struct btrfs_key key;
5415 struct inode *inode = NULL;
5423 struct btrfs_file_extent_item *ei;
5424 struct extent_buffer *leaf;
5425 unsigned long nr = 0;
5427 name_len = strlen(symname) + 1;
5428 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5429 return -ENAMETOOLONG;
5432 * 2 items for inode item and ref
5433 * 2 items for dir items
5434 * 1 item for xattr if selinux is on
5436 err = btrfs_reserve_metadata_space(root, 5);
5440 trans = btrfs_start_transaction(root, 1);
5443 btrfs_set_trans_block_group(trans, dir);
5445 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5451 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5453 dentry->d_parent->d_inode->i_ino, objectid,
5454 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5456 err = PTR_ERR(inode);
5460 err = btrfs_init_inode_security(inode, dir);
5466 btrfs_set_trans_block_group(trans, inode);
5467 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5471 inode->i_mapping->a_ops = &btrfs_aops;
5472 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5473 inode->i_fop = &btrfs_file_operations;
5474 inode->i_op = &btrfs_file_inode_operations;
5475 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5477 btrfs_update_inode_block_group(trans, inode);
5478 btrfs_update_inode_block_group(trans, dir);
5482 path = btrfs_alloc_path();
5484 key.objectid = inode->i_ino;
5486 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5487 datasize = btrfs_file_extent_calc_inline_size(name_len);
5488 err = btrfs_insert_empty_item(trans, root, path, &key,
5494 leaf = path->nodes[0];
5495 ei = btrfs_item_ptr(leaf, path->slots[0],
5496 struct btrfs_file_extent_item);
5497 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5498 btrfs_set_file_extent_type(leaf, ei,
5499 BTRFS_FILE_EXTENT_INLINE);
5500 btrfs_set_file_extent_encryption(leaf, ei, 0);
5501 btrfs_set_file_extent_compression(leaf, ei, 0);
5502 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5503 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5505 ptr = btrfs_file_extent_inline_start(ei);
5506 write_extent_buffer(leaf, symname, ptr, name_len);
5507 btrfs_mark_buffer_dirty(leaf);
5508 btrfs_free_path(path);
5510 inode->i_op = &btrfs_symlink_inode_operations;
5511 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5512 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5513 inode_set_bytes(inode, name_len);
5514 btrfs_i_size_write(inode, name_len - 1);
5515 err = btrfs_update_inode(trans, root, inode);
5520 nr = trans->blocks_used;
5521 btrfs_end_transaction_throttle(trans, root);
5523 btrfs_unreserve_metadata_space(root, 5);
5525 inode_dec_link_count(inode);
5528 btrfs_btree_balance_dirty(root, nr);
5532 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5533 struct inode *inode, u64 start, u64 end,
5534 u64 locked_end, u64 alloc_hint, int mode)
5536 struct btrfs_root *root = BTRFS_I(inode)->root;
5537 struct btrfs_key ins;
5539 u64 cur_offset = start;
5540 u64 num_bytes = end - start;
5543 while (num_bytes > 0) {
5544 alloc_size = min(num_bytes, root->fs_info->max_extent);
5546 ret = btrfs_reserve_metadata_space(root, 1);
5550 ret = btrfs_reserve_extent(trans, root, alloc_size,
5551 root->sectorsize, 0, alloc_hint,
5557 ret = insert_reserved_file_extent(trans, inode,
5558 cur_offset, ins.objectid,
5559 ins.offset, ins.offset,
5560 ins.offset, locked_end,
5562 BTRFS_FILE_EXTENT_PREALLOC);
5564 btrfs_drop_extent_cache(inode, cur_offset,
5565 cur_offset + ins.offset -1, 0);
5566 num_bytes -= ins.offset;
5567 cur_offset += ins.offset;
5568 alloc_hint = ins.objectid + ins.offset;
5569 btrfs_unreserve_metadata_space(root, 1);
5572 if (cur_offset > start) {
5573 inode->i_ctime = CURRENT_TIME;
5574 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5575 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5576 cur_offset > i_size_read(inode))
5577 btrfs_i_size_write(inode, cur_offset);
5578 ret = btrfs_update_inode(trans, root, inode);
5585 static long btrfs_fallocate(struct inode *inode, int mode,
5586 loff_t offset, loff_t len)
5594 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5595 struct extent_map *em;
5596 struct btrfs_trans_handle *trans;
5597 struct btrfs_root *root;
5600 alloc_start = offset & ~mask;
5601 alloc_end = (offset + len + mask) & ~mask;
5604 * wait for ordered IO before we have any locks. We'll loop again
5605 * below with the locks held.
5607 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5609 mutex_lock(&inode->i_mutex);
5610 if (alloc_start > inode->i_size) {
5611 ret = btrfs_cont_expand(inode, alloc_start);
5616 root = BTRFS_I(inode)->root;
5618 ret = btrfs_check_data_free_space(root, inode,
5619 alloc_end - alloc_start);
5623 locked_end = alloc_end - 1;
5625 struct btrfs_ordered_extent *ordered;
5627 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5633 /* the extent lock is ordered inside the running
5636 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5638 ordered = btrfs_lookup_first_ordered_extent(inode,
5641 ordered->file_offset + ordered->len > alloc_start &&
5642 ordered->file_offset < alloc_end) {
5643 btrfs_put_ordered_extent(ordered);
5644 unlock_extent(&BTRFS_I(inode)->io_tree,
5645 alloc_start, locked_end, GFP_NOFS);
5646 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5649 * we can't wait on the range with the transaction
5650 * running or with the extent lock held
5652 btrfs_wait_ordered_range(inode, alloc_start,
5653 alloc_end - alloc_start);
5656 btrfs_put_ordered_extent(ordered);
5661 cur_offset = alloc_start;
5663 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5664 alloc_end - cur_offset, 0);
5665 BUG_ON(IS_ERR(em) || !em);
5666 last_byte = min(extent_map_end(em), alloc_end);
5667 last_byte = (last_byte + mask) & ~mask;
5668 if (em->block_start == EXTENT_MAP_HOLE) {
5669 ret = prealloc_file_range(trans, inode, cur_offset,
5670 last_byte, locked_end + 1,
5673 free_extent_map(em);
5677 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5678 alloc_hint = em->block_start;
5679 free_extent_map(em);
5681 cur_offset = last_byte;
5682 if (cur_offset >= alloc_end) {
5687 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5690 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5692 btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
5694 mutex_unlock(&inode->i_mutex);
5698 static int btrfs_set_page_dirty(struct page *page)
5700 return __set_page_dirty_nobuffers(page);
5703 static int btrfs_permission(struct inode *inode, int mask)
5705 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5707 return generic_permission(inode, mask, btrfs_check_acl);
5710 static struct inode_operations btrfs_dir_inode_operations = {
5711 .getattr = btrfs_getattr,
5712 .lookup = btrfs_lookup,
5713 .create = btrfs_create,
5714 .unlink = btrfs_unlink,
5716 .mkdir = btrfs_mkdir,
5717 .rmdir = btrfs_rmdir,
5718 .rename = btrfs_rename,
5719 .symlink = btrfs_symlink,
5720 .setattr = btrfs_setattr,
5721 .mknod = btrfs_mknod,
5722 .setxattr = btrfs_setxattr,
5723 .getxattr = btrfs_getxattr,
5724 .listxattr = btrfs_listxattr,
5725 .removexattr = btrfs_removexattr,
5726 .permission = btrfs_permission,
5728 static struct inode_operations btrfs_dir_ro_inode_operations = {
5729 .lookup = btrfs_lookup,
5730 .permission = btrfs_permission,
5733 static struct file_operations btrfs_dir_file_operations = {
5734 .llseek = generic_file_llseek,
5735 .read = generic_read_dir,
5736 .readdir = btrfs_real_readdir,
5737 .unlocked_ioctl = btrfs_ioctl,
5738 #ifdef CONFIG_COMPAT
5739 .compat_ioctl = btrfs_ioctl,
5741 .release = btrfs_release_file,
5742 .fsync = btrfs_sync_file,
5745 static struct extent_io_ops btrfs_extent_io_ops = {
5746 .fill_delalloc = run_delalloc_range,
5747 .submit_bio_hook = btrfs_submit_bio_hook,
5748 .merge_bio_hook = btrfs_merge_bio_hook,
5749 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5750 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5751 .writepage_start_hook = btrfs_writepage_start_hook,
5752 .readpage_io_failed_hook = btrfs_io_failed_hook,
5753 .set_bit_hook = btrfs_set_bit_hook,
5754 .clear_bit_hook = btrfs_clear_bit_hook,
5755 .merge_extent_hook = btrfs_merge_extent_hook,
5756 .split_extent_hook = btrfs_split_extent_hook,
5760 * btrfs doesn't support the bmap operation because swapfiles
5761 * use bmap to make a mapping of extents in the file. They assume
5762 * these extents won't change over the life of the file and they
5763 * use the bmap result to do IO directly to the drive.
5765 * the btrfs bmap call would return logical addresses that aren't
5766 * suitable for IO and they also will change frequently as COW
5767 * operations happen. So, swapfile + btrfs == corruption.
5769 * For now we're avoiding this by dropping bmap.
5771 static struct address_space_operations btrfs_aops = {
5772 .readpage = btrfs_readpage,
5773 .writepage = btrfs_writepage,
5774 .writepages = btrfs_writepages,
5775 .readpages = btrfs_readpages,
5776 .sync_page = block_sync_page,
5777 .direct_IO = btrfs_direct_IO,
5778 .invalidatepage = btrfs_invalidatepage,
5779 .releasepage = btrfs_releasepage,
5780 .set_page_dirty = btrfs_set_page_dirty,
5783 static struct address_space_operations btrfs_symlink_aops = {
5784 .readpage = btrfs_readpage,
5785 .writepage = btrfs_writepage,
5786 .invalidatepage = btrfs_invalidatepage,
5787 .releasepage = btrfs_releasepage,
5790 static struct inode_operations btrfs_file_inode_operations = {
5791 .truncate = btrfs_truncate,
5792 .getattr = btrfs_getattr,
5793 .setattr = btrfs_setattr,
5794 .setxattr = btrfs_setxattr,
5795 .getxattr = btrfs_getxattr,
5796 .listxattr = btrfs_listxattr,
5797 .removexattr = btrfs_removexattr,
5798 .permission = btrfs_permission,
5799 .fallocate = btrfs_fallocate,
5800 .fiemap = btrfs_fiemap,
5802 static struct inode_operations btrfs_special_inode_operations = {
5803 .getattr = btrfs_getattr,
5804 .setattr = btrfs_setattr,
5805 .permission = btrfs_permission,
5806 .setxattr = btrfs_setxattr,
5807 .getxattr = btrfs_getxattr,
5808 .listxattr = btrfs_listxattr,
5809 .removexattr = btrfs_removexattr,
5811 static struct inode_operations btrfs_symlink_inode_operations = {
5812 .readlink = generic_readlink,
5813 .follow_link = page_follow_link_light,
5814 .put_link = page_put_link,
5815 .permission = btrfs_permission,
5816 .setxattr = btrfs_setxattr,
5817 .getxattr = btrfs_getxattr,
5818 .listxattr = btrfs_listxattr,
5819 .removexattr = btrfs_removexattr,
5822 struct dentry_operations btrfs_dentry_operations = {
5823 .d_delete = btrfs_dentry_delete,