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>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
54 #include "inode-map.h"
56 struct btrfs_iget_args {
58 struct btrfs_root *root;
61 static const struct inode_operations btrfs_dir_inode_operations;
62 static const struct inode_operations btrfs_symlink_inode_operations;
63 static const struct inode_operations btrfs_dir_ro_inode_operations;
64 static const struct inode_operations btrfs_special_inode_operations;
65 static const struct inode_operations btrfs_file_inode_operations;
66 static const struct address_space_operations btrfs_aops;
67 static const struct address_space_operations btrfs_symlink_aops;
68 static const struct file_operations btrfs_dir_file_operations;
69 static struct extent_io_ops btrfs_extent_io_ops;
71 static struct kmem_cache *btrfs_inode_cachep;
72 struct kmem_cache *btrfs_trans_handle_cachep;
73 struct kmem_cache *btrfs_transaction_cachep;
74 struct kmem_cache *btrfs_path_cachep;
75 struct kmem_cache *btrfs_free_space_cachep;
78 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
79 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
80 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
81 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
82 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
83 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
84 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
85 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
88 static int btrfs_setsize(struct inode *inode, loff_t newsize);
89 static int btrfs_truncate(struct inode *inode);
90 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
91 static noinline int cow_file_range(struct inode *inode,
92 struct page *locked_page,
93 u64 start, u64 end, int *page_started,
94 unsigned long *nr_written, int unlock);
96 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
97 struct inode *inode, struct inode *dir)
101 err = btrfs_init_acl(trans, inode, dir);
103 err = btrfs_xattr_security_init(trans, inode, dir);
108 * this does all the hard work for inserting an inline extent into
109 * the btree. The caller should have done a btrfs_drop_extents so that
110 * no overlapping inline items exist in the btree
112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
113 struct btrfs_root *root, struct inode *inode,
114 u64 start, size_t size, size_t compressed_size,
116 struct page **compressed_pages)
118 struct btrfs_key key;
119 struct btrfs_path *path;
120 struct extent_buffer *leaf;
121 struct page *page = NULL;
124 struct btrfs_file_extent_item *ei;
127 size_t cur_size = size;
129 unsigned long offset;
131 if (compressed_size && compressed_pages)
132 cur_size = compressed_size;
134 path = btrfs_alloc_path();
138 path->leave_spinning = 1;
139 btrfs_set_trans_block_group(trans, inode);
141 key.objectid = inode->i_ino;
143 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
144 datasize = btrfs_file_extent_calc_inline_size(cur_size);
146 inode_add_bytes(inode, size);
147 ret = btrfs_insert_empty_item(trans, root, path, &key,
154 leaf = path->nodes[0];
155 ei = btrfs_item_ptr(leaf, path->slots[0],
156 struct btrfs_file_extent_item);
157 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
158 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
159 btrfs_set_file_extent_encryption(leaf, ei, 0);
160 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
161 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
162 ptr = btrfs_file_extent_inline_start(ei);
164 if (compress_type != BTRFS_COMPRESS_NONE) {
167 while (compressed_size > 0) {
168 cpage = compressed_pages[i];
169 cur_size = min_t(unsigned long, compressed_size,
172 kaddr = kmap_atomic(cpage, KM_USER0);
173 write_extent_buffer(leaf, kaddr, ptr, cur_size);
174 kunmap_atomic(kaddr, KM_USER0);
178 compressed_size -= cur_size;
180 btrfs_set_file_extent_compression(leaf, ei,
183 page = find_get_page(inode->i_mapping,
184 start >> PAGE_CACHE_SHIFT);
185 btrfs_set_file_extent_compression(leaf, ei, 0);
186 kaddr = kmap_atomic(page, KM_USER0);
187 offset = start & (PAGE_CACHE_SIZE - 1);
188 write_extent_buffer(leaf, kaddr + offset, ptr, size);
189 kunmap_atomic(kaddr, KM_USER0);
190 page_cache_release(page);
192 btrfs_mark_buffer_dirty(leaf);
193 btrfs_free_path(path);
196 * we're an inline extent, so nobody can
197 * extend the file past i_size without locking
198 * a page we already have locked.
200 * We must do any isize and inode updates
201 * before we unlock the pages. Otherwise we
202 * could end up racing with unlink.
204 BTRFS_I(inode)->disk_i_size = inode->i_size;
205 btrfs_update_inode(trans, root, inode);
209 btrfs_free_path(path);
215 * conditionally insert an inline extent into the file. This
216 * does the checks required to make sure the data is small enough
217 * to fit as an inline extent.
219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
220 struct btrfs_root *root,
221 struct inode *inode, u64 start, u64 end,
222 size_t compressed_size, int compress_type,
223 struct page **compressed_pages)
225 u64 isize = i_size_read(inode);
226 u64 actual_end = min(end + 1, isize);
227 u64 inline_len = actual_end - start;
228 u64 aligned_end = (end + root->sectorsize - 1) &
229 ~((u64)root->sectorsize - 1);
231 u64 data_len = inline_len;
235 data_len = compressed_size;
238 actual_end >= PAGE_CACHE_SIZE ||
239 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241 (actual_end & (root->sectorsize - 1)) == 0) ||
243 data_len > root->fs_info->max_inline) {
247 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
251 if (isize > actual_end)
252 inline_len = min_t(u64, isize, actual_end);
253 ret = insert_inline_extent(trans, root, inode, start,
254 inline_len, compressed_size,
255 compress_type, compressed_pages);
257 btrfs_delalloc_release_metadata(inode, end + 1 - start);
258 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
262 struct async_extent {
267 unsigned long nr_pages;
269 struct list_head list;
274 struct btrfs_root *root;
275 struct page *locked_page;
278 struct list_head extents;
279 struct btrfs_work work;
282 static noinline int add_async_extent(struct async_cow *cow,
283 u64 start, u64 ram_size,
286 unsigned long nr_pages,
289 struct async_extent *async_extent;
291 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
292 BUG_ON(!async_extent);
293 async_extent->start = start;
294 async_extent->ram_size = ram_size;
295 async_extent->compressed_size = compressed_size;
296 async_extent->pages = pages;
297 async_extent->nr_pages = nr_pages;
298 async_extent->compress_type = compress_type;
299 list_add_tail(&async_extent->list, &cow->extents);
304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
319 static noinline int compress_file_range(struct inode *inode,
320 struct page *locked_page,
322 struct async_cow *async_cow,
325 struct btrfs_root *root = BTRFS_I(inode)->root;
326 struct btrfs_trans_handle *trans;
328 u64 blocksize = root->sectorsize;
330 u64 isize = i_size_read(inode);
332 struct page **pages = NULL;
333 unsigned long nr_pages;
334 unsigned long nr_pages_ret = 0;
335 unsigned long total_compressed = 0;
336 unsigned long total_in = 0;
337 unsigned long max_compressed = 128 * 1024;
338 unsigned long max_uncompressed = 128 * 1024;
341 int compress_type = root->fs_info->compress_type;
343 actual_end = min_t(u64, isize, end + 1);
346 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
347 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
350 * we don't want to send crud past the end of i_size through
351 * compression, that's just a waste of CPU time. So, if the
352 * end of the file is before the start of our current
353 * requested range of bytes, we bail out to the uncompressed
354 * cleanup code that can deal with all of this.
356 * It isn't really the fastest way to fix things, but this is a
357 * very uncommon corner.
359 if (actual_end <= start)
360 goto cleanup_and_bail_uncompressed;
362 total_compressed = actual_end - start;
364 /* we want to make sure that amount of ram required to uncompress
365 * an extent is reasonable, so we limit the total size in ram
366 * of a compressed extent to 128k. This is a crucial number
367 * because it also controls how easily we can spread reads across
368 * cpus for decompression.
370 * We also want to make sure the amount of IO required to do
371 * a random read is reasonably small, so we limit the size of
372 * a compressed extent to 128k.
374 total_compressed = min(total_compressed, max_uncompressed);
375 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
376 num_bytes = max(blocksize, num_bytes);
381 * we do compression for mount -o compress and when the
382 * inode has not been flagged as nocompress. This flag can
383 * change at any time if we discover bad compression ratios.
385 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
386 (btrfs_test_opt(root, COMPRESS) ||
387 (BTRFS_I(inode)->force_compress) ||
388 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
390 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
393 if (BTRFS_I(inode)->force_compress)
394 compress_type = BTRFS_I(inode)->force_compress;
396 ret = btrfs_compress_pages(compress_type,
397 inode->i_mapping, start,
398 total_compressed, pages,
399 nr_pages, &nr_pages_ret,
405 unsigned long offset = total_compressed &
406 (PAGE_CACHE_SIZE - 1);
407 struct page *page = pages[nr_pages_ret - 1];
410 /* zero the tail end of the last page, we might be
411 * sending it down to disk
414 kaddr = kmap_atomic(page, KM_USER0);
415 memset(kaddr + offset, 0,
416 PAGE_CACHE_SIZE - offset);
417 kunmap_atomic(kaddr, KM_USER0);
423 trans = btrfs_join_transaction(root, 1);
424 BUG_ON(IS_ERR(trans));
425 btrfs_set_trans_block_group(trans, inode);
426 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
428 /* lets try to make an inline extent */
429 if (ret || total_in < (actual_end - start)) {
430 /* we didn't compress the entire range, try
431 * to make an uncompressed inline extent.
433 ret = cow_file_range_inline(trans, root, inode,
434 start, end, 0, 0, NULL);
436 /* try making a compressed inline extent */
437 ret = cow_file_range_inline(trans, root, inode,
440 compress_type, pages);
444 * inline extent creation worked, we don't need
445 * to create any more async work items. Unlock
446 * and free up our temp pages.
448 extent_clear_unlock_delalloc(inode,
449 &BTRFS_I(inode)->io_tree,
451 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
452 EXTENT_CLEAR_DELALLOC |
453 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
455 btrfs_end_transaction(trans, root);
458 btrfs_end_transaction(trans, root);
463 * we aren't doing an inline extent round the compressed size
464 * up to a block size boundary so the allocator does sane
467 total_compressed = (total_compressed + blocksize - 1) &
471 * one last check to make sure the compression is really a
472 * win, compare the page count read with the blocks on disk
474 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
475 ~(PAGE_CACHE_SIZE - 1);
476 if (total_compressed >= total_in) {
479 num_bytes = total_in;
482 if (!will_compress && pages) {
484 * the compression code ran but failed to make things smaller,
485 * free any pages it allocated and our page pointer array
487 for (i = 0; i < nr_pages_ret; i++) {
488 WARN_ON(pages[i]->mapping);
489 page_cache_release(pages[i]);
493 total_compressed = 0;
496 /* flag the file so we don't compress in the future */
497 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
498 !(BTRFS_I(inode)->force_compress)) {
499 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
505 /* the async work queues will take care of doing actual
506 * allocation on disk for these compressed pages,
507 * and will submit them to the elevator.
509 add_async_extent(async_cow, start, num_bytes,
510 total_compressed, pages, nr_pages_ret,
513 if (start + num_bytes < end) {
520 cleanup_and_bail_uncompressed:
522 * No compression, but we still need to write the pages in
523 * the file we've been given so far. redirty the locked
524 * page if it corresponds to our extent and set things up
525 * for the async work queue to run cow_file_range to do
526 * the normal delalloc dance
528 if (page_offset(locked_page) >= start &&
529 page_offset(locked_page) <= end) {
530 __set_page_dirty_nobuffers(locked_page);
531 /* unlocked later on in the async handlers */
533 add_async_extent(async_cow, start, end - start + 1,
534 0, NULL, 0, BTRFS_COMPRESS_NONE);
542 for (i = 0; i < nr_pages_ret; i++) {
543 WARN_ON(pages[i]->mapping);
544 page_cache_release(pages[i]);
552 * phase two of compressed writeback. This is the ordered portion
553 * of the code, which only gets called in the order the work was
554 * queued. We walk all the async extents created by compress_file_range
555 * and send them down to the disk.
557 static noinline int submit_compressed_extents(struct inode *inode,
558 struct async_cow *async_cow)
560 struct async_extent *async_extent;
562 struct btrfs_trans_handle *trans;
563 struct btrfs_key ins;
564 struct extent_map *em;
565 struct btrfs_root *root = BTRFS_I(inode)->root;
566 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
567 struct extent_io_tree *io_tree;
570 if (list_empty(&async_cow->extents))
574 while (!list_empty(&async_cow->extents)) {
575 async_extent = list_entry(async_cow->extents.next,
576 struct async_extent, list);
577 list_del(&async_extent->list);
579 io_tree = &BTRFS_I(inode)->io_tree;
582 /* did the compression code fall back to uncompressed IO? */
583 if (!async_extent->pages) {
584 int page_started = 0;
585 unsigned long nr_written = 0;
587 lock_extent(io_tree, async_extent->start,
588 async_extent->start +
589 async_extent->ram_size - 1, GFP_NOFS);
591 /* allocate blocks */
592 ret = cow_file_range(inode, async_cow->locked_page,
594 async_extent->start +
595 async_extent->ram_size - 1,
596 &page_started, &nr_written, 0);
599 * if page_started, cow_file_range inserted an
600 * inline extent and took care of all the unlocking
601 * and IO for us. Otherwise, we need to submit
602 * all those pages down to the drive.
604 if (!page_started && !ret)
605 extent_write_locked_range(io_tree,
606 inode, async_extent->start,
607 async_extent->start +
608 async_extent->ram_size - 1,
616 lock_extent(io_tree, async_extent->start,
617 async_extent->start + async_extent->ram_size - 1,
620 trans = btrfs_join_transaction(root, 1);
621 BUG_ON(IS_ERR(trans));
622 ret = btrfs_reserve_extent(trans, root,
623 async_extent->compressed_size,
624 async_extent->compressed_size,
627 btrfs_end_transaction(trans, root);
631 for (i = 0; i < async_extent->nr_pages; i++) {
632 WARN_ON(async_extent->pages[i]->mapping);
633 page_cache_release(async_extent->pages[i]);
635 kfree(async_extent->pages);
636 async_extent->nr_pages = 0;
637 async_extent->pages = NULL;
638 unlock_extent(io_tree, async_extent->start,
639 async_extent->start +
640 async_extent->ram_size - 1, GFP_NOFS);
645 * here we're doing allocation and writeback of the
648 btrfs_drop_extent_cache(inode, async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1, 0);
652 em = alloc_extent_map(GFP_NOFS);
654 em->start = async_extent->start;
655 em->len = async_extent->ram_size;
656 em->orig_start = em->start;
658 em->block_start = ins.objectid;
659 em->block_len = ins.offset;
660 em->bdev = root->fs_info->fs_devices->latest_bdev;
661 em->compress_type = async_extent->compress_type;
662 set_bit(EXTENT_FLAG_PINNED, &em->flags);
663 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
666 write_lock(&em_tree->lock);
667 ret = add_extent_mapping(em_tree, em);
668 write_unlock(&em_tree->lock);
669 if (ret != -EEXIST) {
673 btrfs_drop_extent_cache(inode, async_extent->start,
674 async_extent->start +
675 async_extent->ram_size - 1, 0);
678 ret = btrfs_add_ordered_extent_compress(inode,
681 async_extent->ram_size,
683 BTRFS_ORDERED_COMPRESSED,
684 async_extent->compress_type);
688 * clear dirty, set writeback and unlock the pages.
690 extent_clear_unlock_delalloc(inode,
691 &BTRFS_I(inode)->io_tree,
693 async_extent->start +
694 async_extent->ram_size - 1,
695 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
696 EXTENT_CLEAR_UNLOCK |
697 EXTENT_CLEAR_DELALLOC |
698 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
700 ret = btrfs_submit_compressed_write(inode,
702 async_extent->ram_size,
704 ins.offset, async_extent->pages,
705 async_extent->nr_pages);
708 alloc_hint = ins.objectid + ins.offset;
716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
719 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
720 struct extent_map *em;
723 read_lock(&em_tree->lock);
724 em = search_extent_mapping(em_tree, start, num_bytes);
727 * if block start isn't an actual block number then find the
728 * first block in this inode and use that as a hint. If that
729 * block is also bogus then just don't worry about it.
731 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
733 em = search_extent_mapping(em_tree, 0, 0);
734 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
735 alloc_hint = em->block_start;
739 alloc_hint = em->block_start;
743 read_unlock(&em_tree->lock);
749 * when extent_io.c finds a delayed allocation range in the file,
750 * the call backs end up in this code. The basic idea is to
751 * allocate extents on disk for the range, and create ordered data structs
752 * in ram to track those extents.
754 * locked_page is the page that writepage had locked already. We use
755 * it to make sure we don't do extra locks or unlocks.
757 * *page_started is set to one if we unlock locked_page and do everything
758 * required to start IO on it. It may be clean and already done with
761 static noinline int cow_file_range(struct inode *inode,
762 struct page *locked_page,
763 u64 start, u64 end, int *page_started,
764 unsigned long *nr_written,
767 struct btrfs_root *root = BTRFS_I(inode)->root;
768 struct btrfs_trans_handle *trans;
771 unsigned long ram_size;
774 u64 blocksize = root->sectorsize;
775 struct btrfs_key ins;
776 struct extent_map *em;
777 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
780 BUG_ON(root == root->fs_info->tree_root);
781 trans = btrfs_join_transaction(root, 1);
782 BUG_ON(IS_ERR(trans));
783 btrfs_set_trans_block_group(trans, inode);
784 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
786 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
787 num_bytes = max(blocksize, num_bytes);
788 disk_num_bytes = num_bytes;
792 /* lets try to make an inline extent */
793 ret = cow_file_range_inline(trans, root, inode,
794 start, end, 0, 0, NULL);
796 extent_clear_unlock_delalloc(inode,
797 &BTRFS_I(inode)->io_tree,
799 EXTENT_CLEAR_UNLOCK_PAGE |
800 EXTENT_CLEAR_UNLOCK |
801 EXTENT_CLEAR_DELALLOC |
803 EXTENT_SET_WRITEBACK |
804 EXTENT_END_WRITEBACK);
806 *nr_written = *nr_written +
807 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
814 BUG_ON(disk_num_bytes >
815 btrfs_super_total_bytes(&root->fs_info->super_copy));
817 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
818 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
820 while (disk_num_bytes > 0) {
823 cur_alloc_size = disk_num_bytes;
824 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
825 root->sectorsize, 0, alloc_hint,
829 em = alloc_extent_map(GFP_NOFS);
832 em->orig_start = em->start;
833 ram_size = ins.offset;
834 em->len = ins.offset;
836 em->block_start = ins.objectid;
837 em->block_len = ins.offset;
838 em->bdev = root->fs_info->fs_devices->latest_bdev;
839 set_bit(EXTENT_FLAG_PINNED, &em->flags);
842 write_lock(&em_tree->lock);
843 ret = add_extent_mapping(em_tree, em);
844 write_unlock(&em_tree->lock);
845 if (ret != -EEXIST) {
849 btrfs_drop_extent_cache(inode, start,
850 start + ram_size - 1, 0);
853 cur_alloc_size = ins.offset;
854 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
855 ram_size, cur_alloc_size, 0);
858 if (root->root_key.objectid ==
859 BTRFS_DATA_RELOC_TREE_OBJECTID) {
860 ret = btrfs_reloc_clone_csums(inode, start,
865 if (disk_num_bytes < cur_alloc_size)
868 /* we're not doing compressed IO, don't unlock the first
869 * page (which the caller expects to stay locked), don't
870 * clear any dirty bits and don't set any writeback bits
872 * Do set the Private2 bit so we know this page was properly
873 * setup for writepage
875 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
876 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
879 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
880 start, start + ram_size - 1,
882 disk_num_bytes -= cur_alloc_size;
883 num_bytes -= cur_alloc_size;
884 alloc_hint = ins.objectid + ins.offset;
885 start += cur_alloc_size;
889 btrfs_end_transaction(trans, root);
895 * work queue call back to started compression on a file and pages
897 static noinline void async_cow_start(struct btrfs_work *work)
899 struct async_cow *async_cow;
901 async_cow = container_of(work, struct async_cow, work);
903 compress_file_range(async_cow->inode, async_cow->locked_page,
904 async_cow->start, async_cow->end, async_cow,
907 async_cow->inode = NULL;
911 * work queue call back to submit previously compressed pages
913 static noinline void async_cow_submit(struct btrfs_work *work)
915 struct async_cow *async_cow;
916 struct btrfs_root *root;
917 unsigned long nr_pages;
919 async_cow = container_of(work, struct async_cow, work);
921 root = async_cow->root;
922 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
925 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
927 if (atomic_read(&root->fs_info->async_delalloc_pages) <
929 waitqueue_active(&root->fs_info->async_submit_wait))
930 wake_up(&root->fs_info->async_submit_wait);
932 if (async_cow->inode)
933 submit_compressed_extents(async_cow->inode, async_cow);
936 static noinline void async_cow_free(struct btrfs_work *work)
938 struct async_cow *async_cow;
939 async_cow = container_of(work, struct async_cow, work);
943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
944 u64 start, u64 end, int *page_started,
945 unsigned long *nr_written)
947 struct async_cow *async_cow;
948 struct btrfs_root *root = BTRFS_I(inode)->root;
949 unsigned long nr_pages;
951 int limit = 10 * 1024 * 1042;
953 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
954 1, 0, NULL, GFP_NOFS);
955 while (start < end) {
956 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
957 async_cow->inode = inode;
958 async_cow->root = root;
959 async_cow->locked_page = locked_page;
960 async_cow->start = start;
962 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
965 cur_end = min(end, start + 512 * 1024 - 1);
967 async_cow->end = cur_end;
968 INIT_LIST_HEAD(&async_cow->extents);
970 async_cow->work.func = async_cow_start;
971 async_cow->work.ordered_func = async_cow_submit;
972 async_cow->work.ordered_free = async_cow_free;
973 async_cow->work.flags = 0;
975 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
977 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
979 btrfs_queue_worker(&root->fs_info->delalloc_workers,
982 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
983 wait_event(root->fs_info->async_submit_wait,
984 (atomic_read(&root->fs_info->async_delalloc_pages) <
988 while (atomic_read(&root->fs_info->async_submit_draining) &&
989 atomic_read(&root->fs_info->async_delalloc_pages)) {
990 wait_event(root->fs_info->async_submit_wait,
991 (atomic_read(&root->fs_info->async_delalloc_pages) ==
995 *nr_written += nr_pages;
1002 static noinline int csum_exist_in_range(struct btrfs_root *root,
1003 u64 bytenr, u64 num_bytes)
1006 struct btrfs_ordered_sum *sums;
1009 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1010 bytenr + num_bytes - 1, &list);
1011 if (ret == 0 && list_empty(&list))
1014 while (!list_empty(&list)) {
1015 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1016 list_del(&sums->list);
1023 * when nowcow writeback call back. This checks for snapshots or COW copies
1024 * of the extents that exist in the file, and COWs the file as required.
1026 * If no cow copies or snapshots exist, we write directly to the existing
1029 static noinline int run_delalloc_nocow(struct inode *inode,
1030 struct page *locked_page,
1031 u64 start, u64 end, int *page_started, int force,
1032 unsigned long *nr_written)
1034 struct btrfs_root *root = BTRFS_I(inode)->root;
1035 struct btrfs_trans_handle *trans;
1036 struct extent_buffer *leaf;
1037 struct btrfs_path *path;
1038 struct btrfs_file_extent_item *fi;
1039 struct btrfs_key found_key;
1051 bool nolock = false;
1053 path = btrfs_alloc_path();
1055 if (root == root->fs_info->tree_root) {
1057 trans = btrfs_join_transaction_nolock(root, 1);
1059 trans = btrfs_join_transaction(root, 1);
1061 BUG_ON(IS_ERR(trans));
1063 cow_start = (u64)-1;
1066 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1069 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1070 leaf = path->nodes[0];
1071 btrfs_item_key_to_cpu(leaf, &found_key,
1072 path->slots[0] - 1);
1073 if (found_key.objectid == inode->i_ino &&
1074 found_key.type == BTRFS_EXTENT_DATA_KEY)
1079 leaf = path->nodes[0];
1080 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081 ret = btrfs_next_leaf(root, path);
1086 leaf = path->nodes[0];
1092 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094 if (found_key.objectid > inode->i_ino ||
1095 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1096 found_key.offset > end)
1099 if (found_key.offset > cur_offset) {
1100 extent_end = found_key.offset;
1105 fi = btrfs_item_ptr(leaf, path->slots[0],
1106 struct btrfs_file_extent_item);
1107 extent_type = btrfs_file_extent_type(leaf, fi);
1109 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1110 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1111 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1112 extent_offset = btrfs_file_extent_offset(leaf, fi);
1113 extent_end = found_key.offset +
1114 btrfs_file_extent_num_bytes(leaf, fi);
1115 if (extent_end <= start) {
1119 if (disk_bytenr == 0)
1121 if (btrfs_file_extent_compression(leaf, fi) ||
1122 btrfs_file_extent_encryption(leaf, fi) ||
1123 btrfs_file_extent_other_encoding(leaf, fi))
1125 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127 if (btrfs_extent_readonly(root, disk_bytenr))
1129 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1131 extent_offset, disk_bytenr))
1133 disk_bytenr += extent_offset;
1134 disk_bytenr += cur_offset - found_key.offset;
1135 num_bytes = min(end + 1, extent_end) - cur_offset;
1137 * force cow if csum exists in the range.
1138 * this ensure that csum for a given extent are
1139 * either valid or do not exist.
1141 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1144 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1145 extent_end = found_key.offset +
1146 btrfs_file_extent_inline_len(leaf, fi);
1147 extent_end = ALIGN(extent_end, root->sectorsize);
1152 if (extent_end <= start) {
1157 if (cow_start == (u64)-1)
1158 cow_start = cur_offset;
1159 cur_offset = extent_end;
1160 if (cur_offset > end)
1166 btrfs_release_path(root, path);
1167 if (cow_start != (u64)-1) {
1168 ret = cow_file_range(inode, locked_page, cow_start,
1169 found_key.offset - 1, page_started,
1172 cow_start = (u64)-1;
1175 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1176 struct extent_map *em;
1177 struct extent_map_tree *em_tree;
1178 em_tree = &BTRFS_I(inode)->extent_tree;
1179 em = alloc_extent_map(GFP_NOFS);
1181 em->start = cur_offset;
1182 em->orig_start = em->start;
1183 em->len = num_bytes;
1184 em->block_len = num_bytes;
1185 em->block_start = disk_bytenr;
1186 em->bdev = root->fs_info->fs_devices->latest_bdev;
1187 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189 write_lock(&em_tree->lock);
1190 ret = add_extent_mapping(em_tree, em);
1191 write_unlock(&em_tree->lock);
1192 if (ret != -EEXIST) {
1193 free_extent_map(em);
1196 btrfs_drop_extent_cache(inode, em->start,
1197 em->start + em->len - 1, 0);
1199 type = BTRFS_ORDERED_PREALLOC;
1201 type = BTRFS_ORDERED_NOCOW;
1204 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1205 num_bytes, num_bytes, type);
1208 if (root->root_key.objectid ==
1209 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1210 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1215 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1216 cur_offset, cur_offset + num_bytes - 1,
1217 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1218 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1219 EXTENT_SET_PRIVATE2);
1220 cur_offset = extent_end;
1221 if (cur_offset > end)
1224 btrfs_release_path(root, path);
1226 if (cur_offset <= end && cow_start == (u64)-1)
1227 cow_start = cur_offset;
1228 if (cow_start != (u64)-1) {
1229 ret = cow_file_range(inode, locked_page, cow_start, end,
1230 page_started, nr_written, 1);
1235 ret = btrfs_end_transaction_nolock(trans, root);
1238 ret = btrfs_end_transaction(trans, root);
1241 btrfs_free_path(path);
1246 * extent_io.c call back to do delayed allocation processing
1248 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1249 u64 start, u64 end, int *page_started,
1250 unsigned long *nr_written)
1253 struct btrfs_root *root = BTRFS_I(inode)->root;
1255 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1256 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257 page_started, 1, nr_written);
1258 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1259 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260 page_started, 0, nr_written);
1261 else if (!btrfs_test_opt(root, COMPRESS) &&
1262 !(BTRFS_I(inode)->force_compress) &&
1263 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1264 ret = cow_file_range(inode, locked_page, start, end,
1265 page_started, nr_written, 1);
1267 ret = cow_file_range_async(inode, locked_page, start, end,
1268 page_started, nr_written);
1272 static int btrfs_split_extent_hook(struct inode *inode,
1273 struct extent_state *orig, u64 split)
1275 /* not delalloc, ignore it */
1276 if (!(orig->state & EXTENT_DELALLOC))
1279 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1284 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285 * extents so we can keep track of new extents that are just merged onto old
1286 * extents, such as when we are doing sequential writes, so we can properly
1287 * account for the metadata space we'll need.
1289 static int btrfs_merge_extent_hook(struct inode *inode,
1290 struct extent_state *new,
1291 struct extent_state *other)
1293 /* not delalloc, ignore it */
1294 if (!(other->state & EXTENT_DELALLOC))
1297 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1302 * extent_io.c set_bit_hook, used to track delayed allocation
1303 * bytes in this file, and to maintain the list of inodes that
1304 * have pending delalloc work to be done.
1306 static int btrfs_set_bit_hook(struct inode *inode,
1307 struct extent_state *state, int *bits)
1311 * set_bit and clear bit hooks normally require _irqsave/restore
1312 * but in this case, we are only testeing for the DELALLOC
1313 * bit, which is only set or cleared with irqs on
1315 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1316 struct btrfs_root *root = BTRFS_I(inode)->root;
1317 u64 len = state->end + 1 - state->start;
1318 int do_list = (root->root_key.objectid !=
1319 BTRFS_ROOT_TREE_OBJECTID);
1321 if (*bits & EXTENT_FIRST_DELALLOC)
1322 *bits &= ~EXTENT_FIRST_DELALLOC;
1324 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326 spin_lock(&root->fs_info->delalloc_lock);
1327 BTRFS_I(inode)->delalloc_bytes += len;
1328 root->fs_info->delalloc_bytes += len;
1329 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1330 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1331 &root->fs_info->delalloc_inodes);
1333 spin_unlock(&root->fs_info->delalloc_lock);
1339 * extent_io.c clear_bit_hook, see set_bit_hook for why
1341 static int btrfs_clear_bit_hook(struct inode *inode,
1342 struct extent_state *state, int *bits)
1345 * set_bit and clear bit hooks normally require _irqsave/restore
1346 * but in this case, we are only testeing for the DELALLOC
1347 * bit, which is only set or cleared with irqs on
1349 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1350 struct btrfs_root *root = BTRFS_I(inode)->root;
1351 u64 len = state->end + 1 - state->start;
1352 int do_list = (root->root_key.objectid !=
1353 BTRFS_ROOT_TREE_OBJECTID);
1355 if (*bits & EXTENT_FIRST_DELALLOC)
1356 *bits &= ~EXTENT_FIRST_DELALLOC;
1357 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1358 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360 if (*bits & EXTENT_DO_ACCOUNTING)
1361 btrfs_delalloc_release_metadata(inode, len);
1363 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365 btrfs_free_reserved_data_space(inode, len);
1367 spin_lock(&root->fs_info->delalloc_lock);
1368 root->fs_info->delalloc_bytes -= len;
1369 BTRFS_I(inode)->delalloc_bytes -= len;
1371 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1372 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1373 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375 spin_unlock(&root->fs_info->delalloc_lock);
1381 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382 * we don't create bios that span stripes or chunks
1384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1385 size_t size, struct bio *bio,
1386 unsigned long bio_flags)
1388 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1389 struct btrfs_mapping_tree *map_tree;
1390 u64 logical = (u64)bio->bi_sector << 9;
1395 if (bio_flags & EXTENT_BIO_COMPRESSED)
1398 length = bio->bi_size;
1399 map_tree = &root->fs_info->mapping_tree;
1400 map_length = length;
1401 ret = btrfs_map_block(map_tree, READ, logical,
1402 &map_length, NULL, 0);
1404 if (map_length < length + size)
1410 * in order to insert checksums into the metadata in large chunks,
1411 * we wait until bio submission time. All the pages in the bio are
1412 * checksummed and sums are attached onto the ordered extent record.
1414 * At IO completion time the cums attached on the ordered extent record
1415 * are inserted into the btree
1417 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1418 struct bio *bio, int mirror_num,
1419 unsigned long bio_flags,
1422 struct btrfs_root *root = BTRFS_I(inode)->root;
1425 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1431 * in order to insert checksums into the metadata in large chunks,
1432 * we wait until bio submission time. All the pages in the bio are
1433 * checksummed and sums are attached onto the ordered extent record.
1435 * At IO completion time the cums attached on the ordered extent record
1436 * are inserted into the btree
1438 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1439 int mirror_num, unsigned long bio_flags,
1442 struct btrfs_root *root = BTRFS_I(inode)->root;
1443 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1447 * extent_io.c submission hook. This does the right thing for csum calculation
1448 * on write, or reading the csums from the tree before a read
1450 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1451 int mirror_num, unsigned long bio_flags,
1454 struct btrfs_root *root = BTRFS_I(inode)->root;
1458 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460 if (root == root->fs_info->tree_root)
1461 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1466 if (!(rw & REQ_WRITE)) {
1467 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1468 return btrfs_submit_compressed_read(inode, bio,
1469 mirror_num, bio_flags);
1470 } else if (!skip_sum) {
1471 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1476 } else if (!skip_sum) {
1477 /* csum items have already been cloned */
1478 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1480 /* we're doing a write, do the async checksumming */
1481 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1482 inode, rw, bio, mirror_num,
1483 bio_flags, bio_offset,
1484 __btrfs_submit_bio_start,
1485 __btrfs_submit_bio_done);
1489 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1493 * given a list of ordered sums record them in the inode. This happens
1494 * at IO completion time based on sums calculated at bio submission time.
1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1497 struct inode *inode, u64 file_offset,
1498 struct list_head *list)
1500 struct btrfs_ordered_sum *sum;
1502 btrfs_set_trans_block_group(trans, inode);
1504 list_for_each_entry(sum, list, list) {
1505 btrfs_csum_file_blocks(trans,
1506 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1511 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1512 struct extent_state **cached_state)
1514 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1516 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1517 cached_state, GFP_NOFS);
1520 /* see btrfs_writepage_start_hook for details on why this is required */
1521 struct btrfs_writepage_fixup {
1523 struct btrfs_work work;
1526 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1528 struct btrfs_writepage_fixup *fixup;
1529 struct btrfs_ordered_extent *ordered;
1530 struct extent_state *cached_state = NULL;
1532 struct inode *inode;
1536 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1540 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1541 ClearPageChecked(page);
1545 inode = page->mapping->host;
1546 page_start = page_offset(page);
1547 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1549 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1550 &cached_state, GFP_NOFS);
1552 /* already ordered? We're done */
1553 if (PagePrivate2(page))
1556 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1558 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1559 page_end, &cached_state, GFP_NOFS);
1561 btrfs_start_ordered_extent(inode, ordered, 1);
1566 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1567 ClearPageChecked(page);
1569 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1570 &cached_state, GFP_NOFS);
1573 page_cache_release(page);
1578 * There are a few paths in the higher layers of the kernel that directly
1579 * set the page dirty bit without asking the filesystem if it is a
1580 * good idea. This causes problems because we want to make sure COW
1581 * properly happens and the data=ordered rules are followed.
1583 * In our case any range that doesn't have the ORDERED bit set
1584 * hasn't been properly setup for IO. We kick off an async process
1585 * to fix it up. The async helper will wait for ordered extents, set
1586 * the delalloc bit and make it safe to write the page.
1588 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1590 struct inode *inode = page->mapping->host;
1591 struct btrfs_writepage_fixup *fixup;
1592 struct btrfs_root *root = BTRFS_I(inode)->root;
1594 /* this page is properly in the ordered list */
1595 if (TestClearPagePrivate2(page))
1598 if (PageChecked(page))
1601 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1605 SetPageChecked(page);
1606 page_cache_get(page);
1607 fixup->work.func = btrfs_writepage_fixup_worker;
1609 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1613 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1614 struct inode *inode, u64 file_pos,
1615 u64 disk_bytenr, u64 disk_num_bytes,
1616 u64 num_bytes, u64 ram_bytes,
1617 u8 compression, u8 encryption,
1618 u16 other_encoding, int extent_type)
1620 struct btrfs_root *root = BTRFS_I(inode)->root;
1621 struct btrfs_file_extent_item *fi;
1622 struct btrfs_path *path;
1623 struct extent_buffer *leaf;
1624 struct btrfs_key ins;
1628 path = btrfs_alloc_path();
1631 path->leave_spinning = 1;
1634 * we may be replacing one extent in the tree with another.
1635 * The new extent is pinned in the extent map, and we don't want
1636 * to drop it from the cache until it is completely in the btree.
1638 * So, tell btrfs_drop_extents to leave this extent in the cache.
1639 * the caller is expected to unpin it and allow it to be merged
1642 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1646 ins.objectid = inode->i_ino;
1647 ins.offset = file_pos;
1648 ins.type = BTRFS_EXTENT_DATA_KEY;
1649 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1651 leaf = path->nodes[0];
1652 fi = btrfs_item_ptr(leaf, path->slots[0],
1653 struct btrfs_file_extent_item);
1654 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1655 btrfs_set_file_extent_type(leaf, fi, extent_type);
1656 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1657 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1658 btrfs_set_file_extent_offset(leaf, fi, 0);
1659 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1660 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1661 btrfs_set_file_extent_compression(leaf, fi, compression);
1662 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1663 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1665 btrfs_unlock_up_safe(path, 1);
1666 btrfs_set_lock_blocking(leaf);
1668 btrfs_mark_buffer_dirty(leaf);
1670 inode_add_bytes(inode, num_bytes);
1672 ins.objectid = disk_bytenr;
1673 ins.offset = disk_num_bytes;
1674 ins.type = BTRFS_EXTENT_ITEM_KEY;
1675 ret = btrfs_alloc_reserved_file_extent(trans, root,
1676 root->root_key.objectid,
1677 inode->i_ino, file_pos, &ins);
1679 btrfs_free_path(path);
1685 * helper function for btrfs_finish_ordered_io, this
1686 * just reads in some of the csum leaves to prime them into ram
1687 * before we start the transaction. It limits the amount of btree
1688 * reads required while inside the transaction.
1690 /* as ordered data IO finishes, this gets called so we can finish
1691 * an ordered extent if the range of bytes in the file it covers are
1694 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1696 struct btrfs_root *root = BTRFS_I(inode)->root;
1697 struct btrfs_trans_handle *trans = NULL;
1698 struct btrfs_ordered_extent *ordered_extent = NULL;
1699 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1700 struct extent_state *cached_state = NULL;
1701 int compress_type = 0;
1703 bool nolock = false;
1705 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1709 BUG_ON(!ordered_extent);
1711 nolock = (root == root->fs_info->tree_root);
1713 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1714 BUG_ON(!list_empty(&ordered_extent->list));
1715 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1718 trans = btrfs_join_transaction_nolock(root, 1);
1720 trans = btrfs_join_transaction(root, 1);
1721 BUG_ON(IS_ERR(trans));
1722 btrfs_set_trans_block_group(trans, inode);
1723 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1724 ret = btrfs_update_inode(trans, root, inode);
1730 lock_extent_bits(io_tree, ordered_extent->file_offset,
1731 ordered_extent->file_offset + ordered_extent->len - 1,
1732 0, &cached_state, GFP_NOFS);
1735 trans = btrfs_join_transaction_nolock(root, 1);
1737 trans = btrfs_join_transaction(root, 1);
1738 BUG_ON(IS_ERR(trans));
1739 btrfs_set_trans_block_group(trans, inode);
1740 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1742 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1743 compress_type = ordered_extent->compress_type;
1744 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1745 BUG_ON(compress_type);
1746 ret = btrfs_mark_extent_written(trans, inode,
1747 ordered_extent->file_offset,
1748 ordered_extent->file_offset +
1749 ordered_extent->len);
1752 BUG_ON(root == root->fs_info->tree_root);
1753 ret = insert_reserved_file_extent(trans, inode,
1754 ordered_extent->file_offset,
1755 ordered_extent->start,
1756 ordered_extent->disk_len,
1757 ordered_extent->len,
1758 ordered_extent->len,
1759 compress_type, 0, 0,
1760 BTRFS_FILE_EXTENT_REG);
1761 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1762 ordered_extent->file_offset,
1763 ordered_extent->len);
1766 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1767 ordered_extent->file_offset +
1768 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1770 add_pending_csums(trans, inode, ordered_extent->file_offset,
1771 &ordered_extent->list);
1773 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1775 ret = btrfs_update_inode(trans, root, inode);
1782 btrfs_end_transaction_nolock(trans, root);
1784 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1786 btrfs_end_transaction(trans, root);
1790 btrfs_put_ordered_extent(ordered_extent);
1791 /* once for the tree */
1792 btrfs_put_ordered_extent(ordered_extent);
1797 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1798 struct extent_state *state, int uptodate)
1800 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1802 ClearPagePrivate2(page);
1803 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1807 * When IO fails, either with EIO or csum verification fails, we
1808 * try other mirrors that might have a good copy of the data. This
1809 * io_failure_record is used to record state as we go through all the
1810 * mirrors. If another mirror has good data, the page is set up to date
1811 * and things continue. If a good mirror can't be found, the original
1812 * bio end_io callback is called to indicate things have failed.
1814 struct io_failure_record {
1819 unsigned long bio_flags;
1823 static int btrfs_io_failed_hook(struct bio *failed_bio,
1824 struct page *page, u64 start, u64 end,
1825 struct extent_state *state)
1827 struct io_failure_record *failrec = NULL;
1829 struct extent_map *em;
1830 struct inode *inode = page->mapping->host;
1831 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1832 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1839 ret = get_state_private(failure_tree, start, &private);
1841 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1844 failrec->start = start;
1845 failrec->len = end - start + 1;
1846 failrec->last_mirror = 0;
1847 failrec->bio_flags = 0;
1849 read_lock(&em_tree->lock);
1850 em = lookup_extent_mapping(em_tree, start, failrec->len);
1851 if (em->start > start || em->start + em->len < start) {
1852 free_extent_map(em);
1855 read_unlock(&em_tree->lock);
1857 if (!em || IS_ERR(em)) {
1861 logical = start - em->start;
1862 logical = em->block_start + logical;
1863 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1864 logical = em->block_start;
1865 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1866 extent_set_compress_type(&failrec->bio_flags,
1869 failrec->logical = logical;
1870 free_extent_map(em);
1871 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1872 EXTENT_DIRTY, GFP_NOFS);
1873 set_state_private(failure_tree, start,
1874 (u64)(unsigned long)failrec);
1876 failrec = (struct io_failure_record *)(unsigned long)private;
1878 num_copies = btrfs_num_copies(
1879 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1880 failrec->logical, failrec->len);
1881 failrec->last_mirror++;
1883 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1884 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1887 if (state && state->start != failrec->start)
1889 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1891 if (!state || failrec->last_mirror > num_copies) {
1892 set_state_private(failure_tree, failrec->start, 0);
1893 clear_extent_bits(failure_tree, failrec->start,
1894 failrec->start + failrec->len - 1,
1895 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1899 bio = bio_alloc(GFP_NOFS, 1);
1900 bio->bi_private = state;
1901 bio->bi_end_io = failed_bio->bi_end_io;
1902 bio->bi_sector = failrec->logical >> 9;
1903 bio->bi_bdev = failed_bio->bi_bdev;
1906 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1907 if (failed_bio->bi_rw & REQ_WRITE)
1912 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1913 failrec->last_mirror,
1914 failrec->bio_flags, 0);
1919 * each time an IO finishes, we do a fast check in the IO failure tree
1920 * to see if we need to process or clean up an io_failure_record
1922 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1925 u64 private_failure;
1926 struct io_failure_record *failure;
1930 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1931 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1932 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1933 start, &private_failure);
1935 failure = (struct io_failure_record *)(unsigned long)
1937 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1939 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1941 failure->start + failure->len - 1,
1942 EXTENT_DIRTY | EXTENT_LOCKED,
1951 * when reads are done, we need to check csums to verify the data is correct
1952 * if there's a match, we allow the bio to finish. If not, we go through
1953 * the io_failure_record routines to find good copies
1955 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1956 struct extent_state *state)
1958 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1959 struct inode *inode = page->mapping->host;
1960 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1962 u64 private = ~(u32)0;
1964 struct btrfs_root *root = BTRFS_I(inode)->root;
1967 if (PageChecked(page)) {
1968 ClearPageChecked(page);
1972 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1975 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1976 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1977 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1982 if (state && state->start == start) {
1983 private = state->private;
1986 ret = get_state_private(io_tree, start, &private);
1988 kaddr = kmap_atomic(page, KM_USER0);
1992 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1993 btrfs_csum_final(csum, (char *)&csum);
1994 if (csum != private)
1997 kunmap_atomic(kaddr, KM_USER0);
1999 /* if the io failure tree for this inode is non-empty,
2000 * check to see if we've recovered from a failed IO
2002 btrfs_clean_io_failures(inode, start);
2006 if (printk_ratelimit()) {
2007 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2008 "private %llu\n", page->mapping->host->i_ino,
2009 (unsigned long long)start, csum,
2010 (unsigned long long)private);
2012 memset(kaddr + offset, 1, end - start + 1);
2013 flush_dcache_page(page);
2014 kunmap_atomic(kaddr, KM_USER0);
2020 struct delayed_iput {
2021 struct list_head list;
2022 struct inode *inode;
2025 void btrfs_add_delayed_iput(struct inode *inode)
2027 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2028 struct delayed_iput *delayed;
2030 if (atomic_add_unless(&inode->i_count, -1, 1))
2033 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2034 delayed->inode = inode;
2036 spin_lock(&fs_info->delayed_iput_lock);
2037 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2038 spin_unlock(&fs_info->delayed_iput_lock);
2041 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2044 struct btrfs_fs_info *fs_info = root->fs_info;
2045 struct delayed_iput *delayed;
2048 spin_lock(&fs_info->delayed_iput_lock);
2049 empty = list_empty(&fs_info->delayed_iputs);
2050 spin_unlock(&fs_info->delayed_iput_lock);
2054 down_read(&root->fs_info->cleanup_work_sem);
2055 spin_lock(&fs_info->delayed_iput_lock);
2056 list_splice_init(&fs_info->delayed_iputs, &list);
2057 spin_unlock(&fs_info->delayed_iput_lock);
2059 while (!list_empty(&list)) {
2060 delayed = list_entry(list.next, struct delayed_iput, list);
2061 list_del(&delayed->list);
2062 iput(delayed->inode);
2065 up_read(&root->fs_info->cleanup_work_sem);
2069 * calculate extra metadata reservation when snapshotting a subvolume
2070 * contains orphan files.
2072 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2073 struct btrfs_pending_snapshot *pending,
2074 u64 *bytes_to_reserve)
2076 struct btrfs_root *root;
2077 struct btrfs_block_rsv *block_rsv;
2081 root = pending->root;
2082 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2085 block_rsv = root->orphan_block_rsv;
2087 /* orphan block reservation for the snapshot */
2088 num_bytes = block_rsv->size;
2091 * after the snapshot is created, COWing tree blocks may use more
2092 * space than it frees. So we should make sure there is enough
2095 index = trans->transid & 0x1;
2096 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2097 num_bytes += block_rsv->size -
2098 (block_rsv->reserved + block_rsv->freed[index]);
2101 *bytes_to_reserve += num_bytes;
2104 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2105 struct btrfs_pending_snapshot *pending)
2107 struct btrfs_root *root = pending->root;
2108 struct btrfs_root *snap = pending->snap;
2109 struct btrfs_block_rsv *block_rsv;
2114 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2117 /* refill source subvolume's orphan block reservation */
2118 block_rsv = root->orphan_block_rsv;
2119 index = trans->transid & 0x1;
2120 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2121 num_bytes = block_rsv->size -
2122 (block_rsv->reserved + block_rsv->freed[index]);
2123 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2124 root->orphan_block_rsv,
2129 /* setup orphan block reservation for the snapshot */
2130 block_rsv = btrfs_alloc_block_rsv(snap);
2133 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2134 snap->orphan_block_rsv = block_rsv;
2136 num_bytes = root->orphan_block_rsv->size;
2137 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2138 block_rsv, num_bytes);
2142 /* insert orphan item for the snapshot */
2143 WARN_ON(!root->orphan_item_inserted);
2144 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2145 snap->root_key.objectid);
2147 snap->orphan_item_inserted = 1;
2151 enum btrfs_orphan_cleanup_state {
2152 ORPHAN_CLEANUP_STARTED = 1,
2153 ORPHAN_CLEANUP_DONE = 2,
2157 * This is called in transaction commmit time. If there are no orphan
2158 * files in the subvolume, it removes orphan item and frees block_rsv
2161 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2162 struct btrfs_root *root)
2166 if (!list_empty(&root->orphan_list) ||
2167 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2170 if (root->orphan_item_inserted &&
2171 btrfs_root_refs(&root->root_item) > 0) {
2172 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2173 root->root_key.objectid);
2175 root->orphan_item_inserted = 0;
2178 if (root->orphan_block_rsv) {
2179 WARN_ON(root->orphan_block_rsv->size > 0);
2180 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2181 root->orphan_block_rsv = NULL;
2186 * This creates an orphan entry for the given inode in case something goes
2187 * wrong in the middle of an unlink/truncate.
2189 * NOTE: caller of this function should reserve 5 units of metadata for
2192 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2194 struct btrfs_root *root = BTRFS_I(inode)->root;
2195 struct btrfs_block_rsv *block_rsv = NULL;
2200 if (!root->orphan_block_rsv) {
2201 block_rsv = btrfs_alloc_block_rsv(root);
2205 spin_lock(&root->orphan_lock);
2206 if (!root->orphan_block_rsv) {
2207 root->orphan_block_rsv = block_rsv;
2208 } else if (block_rsv) {
2209 btrfs_free_block_rsv(root, block_rsv);
2213 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2214 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2217 * For proper ENOSPC handling, we should do orphan
2218 * cleanup when mounting. But this introduces backward
2219 * compatibility issue.
2221 if (!xchg(&root->orphan_item_inserted, 1))
2229 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2230 BTRFS_I(inode)->orphan_meta_reserved = 1;
2233 spin_unlock(&root->orphan_lock);
2236 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2238 /* grab metadata reservation from transaction handle */
2240 ret = btrfs_orphan_reserve_metadata(trans, inode);
2244 /* insert an orphan item to track this unlinked/truncated file */
2246 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2250 /* insert an orphan item to track subvolume contains orphan files */
2252 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2253 root->root_key.objectid);
2260 * We have done the truncate/delete so we can go ahead and remove the orphan
2261 * item for this particular inode.
2263 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2265 struct btrfs_root *root = BTRFS_I(inode)->root;
2266 int delete_item = 0;
2267 int release_rsv = 0;
2270 spin_lock(&root->orphan_lock);
2271 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2272 list_del_init(&BTRFS_I(inode)->i_orphan);
2276 if (BTRFS_I(inode)->orphan_meta_reserved) {
2277 BTRFS_I(inode)->orphan_meta_reserved = 0;
2280 spin_unlock(&root->orphan_lock);
2282 if (trans && delete_item) {
2283 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2288 btrfs_orphan_release_metadata(inode);
2294 * this cleans up any orphans that may be left on the list from the last use
2297 int btrfs_orphan_cleanup(struct btrfs_root *root)
2299 struct btrfs_path *path;
2300 struct extent_buffer *leaf;
2301 struct btrfs_key key, found_key;
2302 struct btrfs_trans_handle *trans;
2303 struct inode *inode;
2304 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2306 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2309 path = btrfs_alloc_path();
2316 key.objectid = BTRFS_ORPHAN_OBJECTID;
2317 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2318 key.offset = (u64)-1;
2321 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2326 * if ret == 0 means we found what we were searching for, which
2327 * is weird, but possible, so only screw with path if we didnt
2328 * find the key and see if we have stuff that matches
2332 if (path->slots[0] == 0)
2337 /* pull out the item */
2338 leaf = path->nodes[0];
2339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2341 /* make sure the item matches what we want */
2342 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2344 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2347 /* release the path since we're done with it */
2348 btrfs_release_path(root, path);
2351 * this is where we are basically btrfs_lookup, without the
2352 * crossing root thing. we store the inode number in the
2353 * offset of the orphan item.
2355 found_key.objectid = found_key.offset;
2356 found_key.type = BTRFS_INODE_ITEM_KEY;
2357 found_key.offset = 0;
2358 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2359 if (IS_ERR(inode)) {
2360 ret = PTR_ERR(inode);
2365 * add this inode to the orphan list so btrfs_orphan_del does
2366 * the proper thing when we hit it
2368 spin_lock(&root->orphan_lock);
2369 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2370 spin_unlock(&root->orphan_lock);
2373 * if this is a bad inode, means we actually succeeded in
2374 * removing the inode, but not the orphan record, which means
2375 * we need to manually delete the orphan since iput will just
2376 * do a destroy_inode
2378 if (is_bad_inode(inode)) {
2379 trans = btrfs_start_transaction(root, 0);
2380 if (IS_ERR(trans)) {
2381 ret = PTR_ERR(trans);
2384 btrfs_orphan_del(trans, inode);
2385 btrfs_end_transaction(trans, root);
2390 /* if we have links, this was a truncate, lets do that */
2391 if (inode->i_nlink) {
2392 if (!S_ISREG(inode->i_mode)) {
2398 ret = btrfs_truncate(inode);
2403 /* this will do delete_inode and everything for us */
2408 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2410 if (root->orphan_block_rsv)
2411 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2414 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2415 trans = btrfs_join_transaction(root, 1);
2417 btrfs_end_transaction(trans, root);
2421 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2423 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2427 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2428 btrfs_free_path(path);
2433 * very simple check to peek ahead in the leaf looking for xattrs. If we
2434 * don't find any xattrs, we know there can't be any acls.
2436 * slot is the slot the inode is in, objectid is the objectid of the inode
2438 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2439 int slot, u64 objectid)
2441 u32 nritems = btrfs_header_nritems(leaf);
2442 struct btrfs_key found_key;
2446 while (slot < nritems) {
2447 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2449 /* we found a different objectid, there must not be acls */
2450 if (found_key.objectid != objectid)
2453 /* we found an xattr, assume we've got an acl */
2454 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2458 * we found a key greater than an xattr key, there can't
2459 * be any acls later on
2461 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2468 * it goes inode, inode backrefs, xattrs, extents,
2469 * so if there are a ton of hard links to an inode there can
2470 * be a lot of backrefs. Don't waste time searching too hard,
2471 * this is just an optimization
2476 /* we hit the end of the leaf before we found an xattr or
2477 * something larger than an xattr. We have to assume the inode
2484 * read an inode from the btree into the in-memory inode
2486 static void btrfs_read_locked_inode(struct inode *inode)
2488 struct btrfs_path *path;
2489 struct extent_buffer *leaf;
2490 struct btrfs_inode_item *inode_item;
2491 struct btrfs_timespec *tspec;
2492 struct btrfs_root *root = BTRFS_I(inode)->root;
2493 struct btrfs_key location;
2495 u64 alloc_group_block;
2499 path = btrfs_alloc_path();
2501 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2503 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2507 leaf = path->nodes[0];
2508 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2509 struct btrfs_inode_item);
2511 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2512 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2513 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2514 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2515 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2517 tspec = btrfs_inode_atime(inode_item);
2518 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2519 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2521 tspec = btrfs_inode_mtime(inode_item);
2522 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2523 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2525 tspec = btrfs_inode_ctime(inode_item);
2526 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2527 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2529 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2530 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2531 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2532 inode->i_generation = BTRFS_I(inode)->generation;
2534 rdev = btrfs_inode_rdev(leaf, inode_item);
2536 BTRFS_I(inode)->index_cnt = (u64)-1;
2537 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2539 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2542 * try to precache a NULL acl entry for files that don't have
2543 * any xattrs or acls
2545 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2547 cache_no_acl(inode);
2549 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2550 alloc_group_block, 0);
2551 btrfs_free_path(path);
2554 switch (inode->i_mode & S_IFMT) {
2556 inode->i_mapping->a_ops = &btrfs_aops;
2557 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2558 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2559 inode->i_fop = &btrfs_file_operations;
2560 inode->i_op = &btrfs_file_inode_operations;
2563 inode->i_fop = &btrfs_dir_file_operations;
2564 if (root == root->fs_info->tree_root)
2565 inode->i_op = &btrfs_dir_ro_inode_operations;
2567 inode->i_op = &btrfs_dir_inode_operations;
2570 inode->i_op = &btrfs_symlink_inode_operations;
2571 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2572 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2575 inode->i_op = &btrfs_special_inode_operations;
2576 init_special_inode(inode, inode->i_mode, rdev);
2580 btrfs_update_iflags(inode);
2584 btrfs_free_path(path);
2585 make_bad_inode(inode);
2589 * given a leaf and an inode, copy the inode fields into the leaf
2591 static void fill_inode_item(struct btrfs_trans_handle *trans,
2592 struct extent_buffer *leaf,
2593 struct btrfs_inode_item *item,
2594 struct inode *inode)
2596 if (!leaf->map_token)
2597 map_private_extent_buffer(leaf, (unsigned long)item,
2598 sizeof(struct btrfs_inode_item),
2599 &leaf->map_token, &leaf->kaddr,
2600 &leaf->map_start, &leaf->map_len,
2603 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2604 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2605 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2606 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2607 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2609 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2610 inode->i_atime.tv_sec);
2611 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2612 inode->i_atime.tv_nsec);
2614 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2615 inode->i_mtime.tv_sec);
2616 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2617 inode->i_mtime.tv_nsec);
2619 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2620 inode->i_ctime.tv_sec);
2621 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2622 inode->i_ctime.tv_nsec);
2624 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2625 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2626 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2627 btrfs_set_inode_transid(leaf, item, trans->transid);
2628 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2629 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2630 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2632 if (leaf->map_token) {
2633 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2634 leaf->map_token = NULL;
2639 * copy everything in the in-memory inode into the btree.
2641 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2642 struct btrfs_root *root, struct inode *inode)
2644 struct btrfs_inode_item *inode_item;
2645 struct btrfs_path *path;
2646 struct extent_buffer *leaf;
2649 path = btrfs_alloc_path();
2651 path->leave_spinning = 1;
2652 ret = btrfs_lookup_inode(trans, root, path,
2653 &BTRFS_I(inode)->location, 1);
2660 btrfs_unlock_up_safe(path, 1);
2661 leaf = path->nodes[0];
2662 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2663 struct btrfs_inode_item);
2665 fill_inode_item(trans, leaf, inode_item, inode);
2666 btrfs_mark_buffer_dirty(leaf);
2667 btrfs_set_inode_last_trans(trans, inode);
2670 btrfs_free_path(path);
2676 * unlink helper that gets used here in inode.c and in the tree logging
2677 * recovery code. It remove a link in a directory with a given name, and
2678 * also drops the back refs in the inode to the directory
2680 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2681 struct btrfs_root *root,
2682 struct inode *dir, struct inode *inode,
2683 const char *name, int name_len)
2685 struct btrfs_path *path;
2687 struct extent_buffer *leaf;
2688 struct btrfs_dir_item *di;
2689 struct btrfs_key key;
2692 path = btrfs_alloc_path();
2698 path->leave_spinning = 1;
2699 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2700 name, name_len, -1);
2709 leaf = path->nodes[0];
2710 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2711 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2714 btrfs_release_path(root, path);
2716 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2718 dir->i_ino, &index);
2720 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2721 "inode %lu parent %lu\n", name_len, name,
2722 inode->i_ino, dir->i_ino);
2726 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2727 index, name, name_len, -1);
2736 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2737 btrfs_release_path(root, path);
2739 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2741 BUG_ON(ret != 0 && ret != -ENOENT);
2743 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2748 btrfs_free_path(path);
2752 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2753 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2754 btrfs_update_inode(trans, root, dir);
2759 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2760 struct btrfs_root *root,
2761 struct inode *dir, struct inode *inode,
2762 const char *name, int name_len)
2765 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2767 btrfs_drop_nlink(inode);
2768 ret = btrfs_update_inode(trans, root, inode);
2774 /* helper to check if there is any shared block in the path */
2775 static int check_path_shared(struct btrfs_root *root,
2776 struct btrfs_path *path)
2778 struct extent_buffer *eb;
2782 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2785 if (!path->nodes[level])
2787 eb = path->nodes[level];
2788 if (!btrfs_block_can_be_shared(root, eb))
2790 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2799 * helper to start transaction for unlink and rmdir.
2801 * unlink and rmdir are special in btrfs, they do not always free space.
2802 * so in enospc case, we should make sure they will free space before
2803 * allowing them to use the global metadata reservation.
2805 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2806 struct dentry *dentry)
2808 struct btrfs_trans_handle *trans;
2809 struct btrfs_root *root = BTRFS_I(dir)->root;
2810 struct btrfs_path *path;
2811 struct btrfs_inode_ref *ref;
2812 struct btrfs_dir_item *di;
2813 struct inode *inode = dentry->d_inode;
2819 trans = btrfs_start_transaction(root, 10);
2820 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2823 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2824 return ERR_PTR(-ENOSPC);
2826 /* check if there is someone else holds reference */
2827 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2828 return ERR_PTR(-ENOSPC);
2830 if (atomic_read(&inode->i_count) > 2)
2831 return ERR_PTR(-ENOSPC);
2833 if (xchg(&root->fs_info->enospc_unlink, 1))
2834 return ERR_PTR(-ENOSPC);
2836 path = btrfs_alloc_path();
2838 root->fs_info->enospc_unlink = 0;
2839 return ERR_PTR(-ENOMEM);
2842 trans = btrfs_start_transaction(root, 0);
2843 if (IS_ERR(trans)) {
2844 btrfs_free_path(path);
2845 root->fs_info->enospc_unlink = 0;
2849 path->skip_locking = 1;
2850 path->search_commit_root = 1;
2852 ret = btrfs_lookup_inode(trans, root, path,
2853 &BTRFS_I(dir)->location, 0);
2859 if (check_path_shared(root, path))
2864 btrfs_release_path(root, path);
2866 ret = btrfs_lookup_inode(trans, root, path,
2867 &BTRFS_I(inode)->location, 0);
2873 if (check_path_shared(root, path))
2878 btrfs_release_path(root, path);
2880 if (ret == 0 && S_ISREG(inode->i_mode)) {
2881 ret = btrfs_lookup_file_extent(trans, root, path,
2882 inode->i_ino, (u64)-1, 0);
2888 if (check_path_shared(root, path))
2890 btrfs_release_path(root, path);
2898 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2899 dentry->d_name.name, dentry->d_name.len, 0);
2905 if (check_path_shared(root, path))
2911 btrfs_release_path(root, path);
2913 ref = btrfs_lookup_inode_ref(trans, root, path,
2914 dentry->d_name.name, dentry->d_name.len,
2915 inode->i_ino, dir->i_ino, 0);
2921 if (check_path_shared(root, path))
2923 index = btrfs_inode_ref_index(path->nodes[0], ref);
2924 btrfs_release_path(root, path);
2926 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2927 dentry->d_name.name, dentry->d_name.len, 0);
2932 BUG_ON(ret == -ENOENT);
2933 if (check_path_shared(root, path))
2938 btrfs_free_path(path);
2940 btrfs_end_transaction(trans, root);
2941 root->fs_info->enospc_unlink = 0;
2942 return ERR_PTR(err);
2945 trans->block_rsv = &root->fs_info->global_block_rsv;
2949 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2950 struct btrfs_root *root)
2952 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2953 BUG_ON(!root->fs_info->enospc_unlink);
2954 root->fs_info->enospc_unlink = 0;
2956 btrfs_end_transaction_throttle(trans, root);
2959 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2961 struct btrfs_root *root = BTRFS_I(dir)->root;
2962 struct btrfs_trans_handle *trans;
2963 struct inode *inode = dentry->d_inode;
2965 unsigned long nr = 0;
2967 trans = __unlink_start_trans(dir, dentry);
2969 return PTR_ERR(trans);
2971 btrfs_set_trans_block_group(trans, dir);
2973 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2975 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2976 dentry->d_name.name, dentry->d_name.len);
2979 if (inode->i_nlink == 0) {
2980 ret = btrfs_orphan_add(trans, inode);
2984 nr = trans->blocks_used;
2985 __unlink_end_trans(trans, root);
2986 btrfs_btree_balance_dirty(root, nr);
2990 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2991 struct btrfs_root *root,
2992 struct inode *dir, u64 objectid,
2993 const char *name, int name_len)
2995 struct btrfs_path *path;
2996 struct extent_buffer *leaf;
2997 struct btrfs_dir_item *di;
2998 struct btrfs_key key;
3002 path = btrfs_alloc_path();
3006 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3007 name, name_len, -1);
3008 BUG_ON(!di || IS_ERR(di));
3010 leaf = path->nodes[0];
3011 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3012 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3013 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3015 btrfs_release_path(root, path);
3017 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3018 objectid, root->root_key.objectid,
3019 dir->i_ino, &index, name, name_len);
3021 BUG_ON(ret != -ENOENT);
3022 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3024 BUG_ON(!di || IS_ERR(di));
3026 leaf = path->nodes[0];
3027 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3028 btrfs_release_path(root, path);
3032 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3033 index, name, name_len, -1);
3034 BUG_ON(!di || IS_ERR(di));
3036 leaf = path->nodes[0];
3037 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3038 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3039 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3041 btrfs_release_path(root, path);
3043 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3044 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3045 ret = btrfs_update_inode(trans, root, dir);
3048 btrfs_free_path(path);
3052 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3054 struct inode *inode = dentry->d_inode;
3056 struct btrfs_root *root = BTRFS_I(dir)->root;
3057 struct btrfs_trans_handle *trans;
3058 unsigned long nr = 0;
3060 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3061 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3064 trans = __unlink_start_trans(dir, dentry);
3066 return PTR_ERR(trans);
3068 btrfs_set_trans_block_group(trans, dir);
3070 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3071 err = btrfs_unlink_subvol(trans, root, dir,
3072 BTRFS_I(inode)->location.objectid,
3073 dentry->d_name.name,
3074 dentry->d_name.len);
3078 err = btrfs_orphan_add(trans, inode);
3082 /* now the directory is empty */
3083 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3084 dentry->d_name.name, dentry->d_name.len);
3086 btrfs_i_size_write(inode, 0);
3088 nr = trans->blocks_used;
3089 __unlink_end_trans(trans, root);
3090 btrfs_btree_balance_dirty(root, nr);
3097 * when truncating bytes in a file, it is possible to avoid reading
3098 * the leaves that contain only checksum items. This can be the
3099 * majority of the IO required to delete a large file, but it must
3100 * be done carefully.
3102 * The keys in the level just above the leaves are checked to make sure
3103 * the lowest key in a given leaf is a csum key, and starts at an offset
3104 * after the new size.
3106 * Then the key for the next leaf is checked to make sure it also has
3107 * a checksum item for the same file. If it does, we know our target leaf
3108 * contains only checksum items, and it can be safely freed without reading
3111 * This is just an optimization targeted at large files. It may do
3112 * nothing. It will return 0 unless things went badly.
3114 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3115 struct btrfs_root *root,
3116 struct btrfs_path *path,
3117 struct inode *inode, u64 new_size)
3119 struct btrfs_key key;
3122 struct btrfs_key found_key;
3123 struct btrfs_key other_key;
3124 struct btrfs_leaf_ref *ref;
3128 path->lowest_level = 1;
3129 key.objectid = inode->i_ino;
3130 key.type = BTRFS_CSUM_ITEM_KEY;
3131 key.offset = new_size;
3133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3137 if (path->nodes[1] == NULL) {
3142 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3143 nritems = btrfs_header_nritems(path->nodes[1]);
3148 if (path->slots[1] >= nritems)
3151 /* did we find a key greater than anything we want to delete? */
3152 if (found_key.objectid > inode->i_ino ||
3153 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3156 /* we check the next key in the node to make sure the leave contains
3157 * only checksum items. This comparison doesn't work if our
3158 * leaf is the last one in the node
3160 if (path->slots[1] + 1 >= nritems) {
3162 /* search forward from the last key in the node, this
3163 * will bring us into the next node in the tree
3165 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3167 /* unlikely, but we inc below, so check to be safe */
3168 if (found_key.offset == (u64)-1)
3171 /* search_forward needs a path with locks held, do the
3172 * search again for the original key. It is possible
3173 * this will race with a balance and return a path that
3174 * we could modify, but this drop is just an optimization
3175 * and is allowed to miss some leaves.
3177 btrfs_release_path(root, path);
3180 /* setup a max key for search_forward */
3181 other_key.offset = (u64)-1;
3182 other_key.type = key.type;
3183 other_key.objectid = key.objectid;
3185 path->keep_locks = 1;
3186 ret = btrfs_search_forward(root, &found_key, &other_key,
3188 path->keep_locks = 0;
3189 if (ret || found_key.objectid != key.objectid ||
3190 found_key.type != key.type) {
3195 key.offset = found_key.offset;
3196 btrfs_release_path(root, path);
3201 /* we know there's one more slot after us in the tree,
3202 * read that key so we can verify it is also a checksum item
3204 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3206 if (found_key.objectid < inode->i_ino)
3209 if (found_key.type != key.type || found_key.offset < new_size)
3213 * if the key for the next leaf isn't a csum key from this objectid,
3214 * we can't be sure there aren't good items inside this leaf.
3217 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3220 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3221 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3223 * it is safe to delete this leaf, it contains only
3224 * csum items from this inode at an offset >= new_size
3226 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3229 if (root->ref_cows && leaf_gen < trans->transid) {
3230 ref = btrfs_alloc_leaf_ref(root, 0);
3232 ref->root_gen = root->root_key.offset;
3233 ref->bytenr = leaf_start;
3235 ref->generation = leaf_gen;
3238 btrfs_sort_leaf_ref(ref);
3240 ret = btrfs_add_leaf_ref(root, ref, 0);
3242 btrfs_free_leaf_ref(root, ref);
3248 btrfs_release_path(root, path);
3250 if (other_key.objectid == inode->i_ino &&
3251 other_key.type == key.type && other_key.offset > key.offset) {
3252 key.offset = other_key.offset;
3258 /* fixup any changes we've made to the path */
3259 path->lowest_level = 0;
3260 path->keep_locks = 0;
3261 btrfs_release_path(root, path);
3268 * this can truncate away extent items, csum items and directory items.
3269 * It starts at a high offset and removes keys until it can't find
3270 * any higher than new_size
3272 * csum items that cross the new i_size are truncated to the new size
3275 * min_type is the minimum key type to truncate down to. If set to 0, this
3276 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3278 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3279 struct btrfs_root *root,
3280 struct inode *inode,
3281 u64 new_size, u32 min_type)
3283 struct btrfs_path *path;
3284 struct extent_buffer *leaf;
3285 struct btrfs_file_extent_item *fi;
3286 struct btrfs_key key;
3287 struct btrfs_key found_key;
3288 u64 extent_start = 0;
3289 u64 extent_num_bytes = 0;
3290 u64 extent_offset = 0;
3292 u64 mask = root->sectorsize - 1;
3293 u32 found_type = (u8)-1;
3296 int pending_del_nr = 0;
3297 int pending_del_slot = 0;
3298 int extent_type = -1;
3303 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3305 if (root->ref_cows || root == root->fs_info->tree_root)
3306 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3308 path = btrfs_alloc_path();
3312 key.objectid = inode->i_ino;
3313 key.offset = (u64)-1;
3317 path->leave_spinning = 1;
3318 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3325 /* there are no items in the tree for us to truncate, we're
3328 if (path->slots[0] == 0)
3335 leaf = path->nodes[0];
3336 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3337 found_type = btrfs_key_type(&found_key);
3340 if (found_key.objectid != inode->i_ino)
3343 if (found_type < min_type)
3346 item_end = found_key.offset;
3347 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3348 fi = btrfs_item_ptr(leaf, path->slots[0],
3349 struct btrfs_file_extent_item);
3350 extent_type = btrfs_file_extent_type(leaf, fi);
3351 encoding = btrfs_file_extent_compression(leaf, fi);
3352 encoding |= btrfs_file_extent_encryption(leaf, fi);
3353 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3355 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3357 btrfs_file_extent_num_bytes(leaf, fi);
3358 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3359 item_end += btrfs_file_extent_inline_len(leaf,
3364 if (found_type > min_type) {
3367 if (item_end < new_size)
3369 if (found_key.offset >= new_size)
3375 /* FIXME, shrink the extent if the ref count is only 1 */
3376 if (found_type != BTRFS_EXTENT_DATA_KEY)
3379 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3381 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3382 if (!del_item && !encoding) {
3383 u64 orig_num_bytes =
3384 btrfs_file_extent_num_bytes(leaf, fi);
3385 extent_num_bytes = new_size -
3386 found_key.offset + root->sectorsize - 1;
3387 extent_num_bytes = extent_num_bytes &
3388 ~((u64)root->sectorsize - 1);
3389 btrfs_set_file_extent_num_bytes(leaf, fi,
3391 num_dec = (orig_num_bytes -
3393 if (root->ref_cows && extent_start != 0)
3394 inode_sub_bytes(inode, num_dec);
3395 btrfs_mark_buffer_dirty(leaf);
3398 btrfs_file_extent_disk_num_bytes(leaf,
3400 extent_offset = found_key.offset -
3401 btrfs_file_extent_offset(leaf, fi);
3403 /* FIXME blocksize != 4096 */
3404 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3405 if (extent_start != 0) {
3408 inode_sub_bytes(inode, num_dec);
3411 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3413 * we can't truncate inline items that have had
3417 btrfs_file_extent_compression(leaf, fi) == 0 &&
3418 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3419 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3420 u32 size = new_size - found_key.offset;
3422 if (root->ref_cows) {
3423 inode_sub_bytes(inode, item_end + 1 -
3427 btrfs_file_extent_calc_inline_size(size);
3428 ret = btrfs_truncate_item(trans, root, path,
3431 } else if (root->ref_cows) {
3432 inode_sub_bytes(inode, item_end + 1 -
3438 if (!pending_del_nr) {
3439 /* no pending yet, add ourselves */
3440 pending_del_slot = path->slots[0];
3442 } else if (pending_del_nr &&
3443 path->slots[0] + 1 == pending_del_slot) {
3444 /* hop on the pending chunk */
3446 pending_del_slot = path->slots[0];
3453 if (found_extent && (root->ref_cows ||
3454 root == root->fs_info->tree_root)) {
3455 btrfs_set_path_blocking(path);
3456 ret = btrfs_free_extent(trans, root, extent_start,
3457 extent_num_bytes, 0,
3458 btrfs_header_owner(leaf),
3459 inode->i_ino, extent_offset);
3463 if (found_type == BTRFS_INODE_ITEM_KEY)
3466 if (path->slots[0] == 0 ||
3467 path->slots[0] != pending_del_slot) {
3468 if (root->ref_cows) {
3472 if (pending_del_nr) {
3473 ret = btrfs_del_items(trans, root, path,
3479 btrfs_release_path(root, path);
3486 if (pending_del_nr) {
3487 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3491 btrfs_free_path(path);
3496 * taken from block_truncate_page, but does cow as it zeros out
3497 * any bytes left in the last page in the file.
3499 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3501 struct inode *inode = mapping->host;
3502 struct btrfs_root *root = BTRFS_I(inode)->root;
3503 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3504 struct btrfs_ordered_extent *ordered;
3505 struct extent_state *cached_state = NULL;
3507 u32 blocksize = root->sectorsize;
3508 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3509 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3515 if ((offset & (blocksize - 1)) == 0)
3517 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3523 page = grab_cache_page(mapping, index);
3525 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3529 page_start = page_offset(page);
3530 page_end = page_start + PAGE_CACHE_SIZE - 1;
3532 if (!PageUptodate(page)) {
3533 ret = btrfs_readpage(NULL, page);
3535 if (page->mapping != mapping) {
3537 page_cache_release(page);
3540 if (!PageUptodate(page)) {
3545 wait_on_page_writeback(page);
3547 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3549 set_page_extent_mapped(page);
3551 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3553 unlock_extent_cached(io_tree, page_start, page_end,
3554 &cached_state, GFP_NOFS);
3556 page_cache_release(page);
3557 btrfs_start_ordered_extent(inode, ordered, 1);
3558 btrfs_put_ordered_extent(ordered);
3562 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3563 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3564 0, 0, &cached_state, GFP_NOFS);
3566 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3569 unlock_extent_cached(io_tree, page_start, page_end,
3570 &cached_state, GFP_NOFS);
3575 if (offset != PAGE_CACHE_SIZE) {
3577 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3578 flush_dcache_page(page);
3581 ClearPageChecked(page);
3582 set_page_dirty(page);
3583 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3588 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3590 page_cache_release(page);
3596 * This function puts in dummy file extents for the area we're creating a hole
3597 * for. So if we are truncating this file to a larger size we need to insert
3598 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3599 * the range between oldsize and size
3601 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3603 struct btrfs_trans_handle *trans;
3604 struct btrfs_root *root = BTRFS_I(inode)->root;
3605 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3606 struct extent_map *em = NULL;
3607 struct extent_state *cached_state = NULL;
3608 u64 mask = root->sectorsize - 1;
3609 u64 hole_start = (oldsize + mask) & ~mask;
3610 u64 block_end = (size + mask) & ~mask;
3616 if (size <= hole_start)
3620 struct btrfs_ordered_extent *ordered;
3621 btrfs_wait_ordered_range(inode, hole_start,
3622 block_end - hole_start);
3623 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3624 &cached_state, GFP_NOFS);
3625 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3628 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3629 &cached_state, GFP_NOFS);
3630 btrfs_put_ordered_extent(ordered);
3633 cur_offset = hole_start;
3635 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3636 block_end - cur_offset, 0);
3637 BUG_ON(IS_ERR(em) || !em);
3638 last_byte = min(extent_map_end(em), block_end);
3639 last_byte = (last_byte + mask) & ~mask;
3640 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3642 hole_size = last_byte - cur_offset;
3644 trans = btrfs_start_transaction(root, 2);
3645 if (IS_ERR(trans)) {
3646 err = PTR_ERR(trans);
3649 btrfs_set_trans_block_group(trans, inode);
3651 err = btrfs_drop_extents(trans, inode, cur_offset,
3652 cur_offset + hole_size,
3657 err = btrfs_insert_file_extent(trans, root,
3658 inode->i_ino, cur_offset, 0,
3659 0, hole_size, 0, hole_size,
3664 btrfs_drop_extent_cache(inode, hole_start,
3667 btrfs_end_transaction(trans, root);
3669 free_extent_map(em);
3671 cur_offset = last_byte;
3672 if (cur_offset >= block_end)
3676 free_extent_map(em);
3677 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3682 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3684 loff_t oldsize = i_size_read(inode);
3687 if (newsize == oldsize)
3690 if (newsize > oldsize) {
3691 i_size_write(inode, newsize);
3692 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3693 truncate_pagecache(inode, oldsize, newsize);
3694 ret = btrfs_cont_expand(inode, oldsize, newsize);
3696 btrfs_setsize(inode, oldsize);
3700 mark_inode_dirty(inode);
3704 * We're truncating a file that used to have good data down to
3705 * zero. Make sure it gets into the ordered flush list so that
3706 * any new writes get down to disk quickly.
3709 BTRFS_I(inode)->ordered_data_close = 1;
3711 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3712 truncate_setsize(inode, newsize);
3713 ret = btrfs_truncate(inode);
3719 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3721 struct inode *inode = dentry->d_inode;
3722 struct btrfs_root *root = BTRFS_I(inode)->root;
3725 if (btrfs_root_readonly(root))
3728 err = inode_change_ok(inode, attr);
3732 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3733 err = btrfs_setsize(inode, attr->ia_size);
3738 if (attr->ia_valid) {
3739 setattr_copy(inode, attr);
3740 mark_inode_dirty(inode);
3742 if (attr->ia_valid & ATTR_MODE)
3743 err = btrfs_acl_chmod(inode);
3749 void btrfs_evict_inode(struct inode *inode)
3751 struct btrfs_trans_handle *trans;
3752 struct btrfs_root *root = BTRFS_I(inode)->root;
3756 trace_btrfs_inode_evict(inode);
3758 truncate_inode_pages(&inode->i_data, 0);
3759 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3760 root == root->fs_info->tree_root))
3763 if (is_bad_inode(inode)) {
3764 btrfs_orphan_del(NULL, inode);
3767 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3768 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3770 if (root->fs_info->log_root_recovering) {
3771 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3775 if (inode->i_nlink > 0) {
3776 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3780 btrfs_i_size_write(inode, 0);
3783 trans = btrfs_start_transaction(root, 0);
3784 BUG_ON(IS_ERR(trans));
3785 btrfs_set_trans_block_group(trans, inode);
3786 trans->block_rsv = root->orphan_block_rsv;
3788 ret = btrfs_block_rsv_check(trans, root,
3789 root->orphan_block_rsv, 0, 5);
3791 BUG_ON(ret != -EAGAIN);
3792 ret = btrfs_commit_transaction(trans, root);
3797 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3801 nr = trans->blocks_used;
3802 btrfs_end_transaction(trans, root);
3804 btrfs_btree_balance_dirty(root, nr);
3809 ret = btrfs_orphan_del(trans, inode);
3813 if (!(root == root->fs_info->tree_root ||
3814 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3815 btrfs_return_ino(root, inode->i_ino);
3817 nr = trans->blocks_used;
3818 btrfs_end_transaction(trans, root);
3819 btrfs_btree_balance_dirty(root, nr);
3821 end_writeback(inode);
3826 * this returns the key found in the dir entry in the location pointer.
3827 * If no dir entries were found, location->objectid is 0.
3829 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3830 struct btrfs_key *location)
3832 const char *name = dentry->d_name.name;
3833 int namelen = dentry->d_name.len;
3834 struct btrfs_dir_item *di;
3835 struct btrfs_path *path;
3836 struct btrfs_root *root = BTRFS_I(dir)->root;
3839 path = btrfs_alloc_path();
3842 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3847 if (!di || IS_ERR(di))
3850 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3852 btrfs_free_path(path);
3855 location->objectid = 0;
3860 * when we hit a tree root in a directory, the btrfs part of the inode
3861 * needs to be changed to reflect the root directory of the tree root. This
3862 * is kind of like crossing a mount point.
3864 static int fixup_tree_root_location(struct btrfs_root *root,
3866 struct dentry *dentry,
3867 struct btrfs_key *location,
3868 struct btrfs_root **sub_root)
3870 struct btrfs_path *path;
3871 struct btrfs_root *new_root;
3872 struct btrfs_root_ref *ref;
3873 struct extent_buffer *leaf;
3877 path = btrfs_alloc_path();
3884 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3885 BTRFS_I(dir)->root->root_key.objectid,
3886 location->objectid);
3893 leaf = path->nodes[0];
3894 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3895 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3896 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3899 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3900 (unsigned long)(ref + 1),
3901 dentry->d_name.len);
3905 btrfs_release_path(root->fs_info->tree_root, path);
3907 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3908 if (IS_ERR(new_root)) {
3909 err = PTR_ERR(new_root);
3913 if (btrfs_root_refs(&new_root->root_item) == 0) {
3918 *sub_root = new_root;
3919 location->objectid = btrfs_root_dirid(&new_root->root_item);
3920 location->type = BTRFS_INODE_ITEM_KEY;
3921 location->offset = 0;
3924 btrfs_free_path(path);
3928 static void inode_tree_add(struct inode *inode)
3930 struct btrfs_root *root = BTRFS_I(inode)->root;
3931 struct btrfs_inode *entry;
3933 struct rb_node *parent;
3935 p = &root->inode_tree.rb_node;
3938 if (inode_unhashed(inode))
3941 spin_lock(&root->inode_lock);
3944 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3946 if (inode->i_ino < entry->vfs_inode.i_ino)
3947 p = &parent->rb_left;
3948 else if (inode->i_ino > entry->vfs_inode.i_ino)
3949 p = &parent->rb_right;
3951 WARN_ON(!(entry->vfs_inode.i_state &
3952 (I_WILL_FREE | I_FREEING)));
3953 rb_erase(parent, &root->inode_tree);
3954 RB_CLEAR_NODE(parent);
3955 spin_unlock(&root->inode_lock);
3959 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3960 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3961 spin_unlock(&root->inode_lock);
3964 static void inode_tree_del(struct inode *inode)
3966 struct btrfs_root *root = BTRFS_I(inode)->root;
3969 spin_lock(&root->inode_lock);
3970 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3971 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3972 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3973 empty = RB_EMPTY_ROOT(&root->inode_tree);
3975 spin_unlock(&root->inode_lock);
3978 * Free space cache has inodes in the tree root, but the tree root has a
3979 * root_refs of 0, so this could end up dropping the tree root as a
3980 * snapshot, so we need the extra !root->fs_info->tree_root check to
3981 * make sure we don't drop it.
3983 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3984 root != root->fs_info->tree_root) {
3985 synchronize_srcu(&root->fs_info->subvol_srcu);
3986 spin_lock(&root->inode_lock);
3987 empty = RB_EMPTY_ROOT(&root->inode_tree);
3988 spin_unlock(&root->inode_lock);
3990 btrfs_add_dead_root(root);
3994 int btrfs_invalidate_inodes(struct btrfs_root *root)
3996 struct rb_node *node;
3997 struct rb_node *prev;
3998 struct btrfs_inode *entry;
3999 struct inode *inode;
4002 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4004 spin_lock(&root->inode_lock);
4006 node = root->inode_tree.rb_node;
4010 entry = rb_entry(node, struct btrfs_inode, rb_node);
4012 if (objectid < entry->vfs_inode.i_ino)
4013 node = node->rb_left;
4014 else if (objectid > entry->vfs_inode.i_ino)
4015 node = node->rb_right;
4021 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4022 if (objectid <= entry->vfs_inode.i_ino) {
4026 prev = rb_next(prev);
4030 entry = rb_entry(node, struct btrfs_inode, rb_node);
4031 objectid = entry->vfs_inode.i_ino + 1;
4032 inode = igrab(&entry->vfs_inode);
4034 spin_unlock(&root->inode_lock);
4035 if (atomic_read(&inode->i_count) > 1)
4036 d_prune_aliases(inode);
4038 * btrfs_drop_inode will have it removed from
4039 * the inode cache when its usage count
4044 spin_lock(&root->inode_lock);
4048 if (cond_resched_lock(&root->inode_lock))
4051 node = rb_next(node);
4053 spin_unlock(&root->inode_lock);
4057 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4059 struct btrfs_iget_args *args = p;
4060 inode->i_ino = args->ino;
4061 BTRFS_I(inode)->root = args->root;
4062 btrfs_set_inode_space_info(args->root, inode);
4066 static int btrfs_find_actor(struct inode *inode, void *opaque)
4068 struct btrfs_iget_args *args = opaque;
4069 return args->ino == inode->i_ino &&
4070 args->root == BTRFS_I(inode)->root;
4073 static struct inode *btrfs_iget_locked(struct super_block *s,
4075 struct btrfs_root *root)
4077 struct inode *inode;
4078 struct btrfs_iget_args args;
4079 args.ino = objectid;
4082 inode = iget5_locked(s, objectid, btrfs_find_actor,
4083 btrfs_init_locked_inode,
4088 /* Get an inode object given its location and corresponding root.
4089 * Returns in *is_new if the inode was read from disk
4091 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4092 struct btrfs_root *root, int *new)
4094 struct inode *inode;
4096 inode = btrfs_iget_locked(s, location->objectid, root);
4098 return ERR_PTR(-ENOMEM);
4100 if (inode->i_state & I_NEW) {
4101 BTRFS_I(inode)->root = root;
4102 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4103 btrfs_read_locked_inode(inode);
4104 inode_tree_add(inode);
4105 unlock_new_inode(inode);
4113 static struct inode *new_simple_dir(struct super_block *s,
4114 struct btrfs_key *key,
4115 struct btrfs_root *root)
4117 struct inode *inode = new_inode(s);
4120 return ERR_PTR(-ENOMEM);
4122 BTRFS_I(inode)->root = root;
4123 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4124 BTRFS_I(inode)->dummy_inode = 1;
4126 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4127 inode->i_op = &simple_dir_inode_operations;
4128 inode->i_fop = &simple_dir_operations;
4129 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4130 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4135 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4137 struct inode *inode;
4138 struct btrfs_root *root = BTRFS_I(dir)->root;
4139 struct btrfs_root *sub_root = root;
4140 struct btrfs_key location;
4144 if (dentry->d_name.len > BTRFS_NAME_LEN)
4145 return ERR_PTR(-ENAMETOOLONG);
4147 ret = btrfs_inode_by_name(dir, dentry, &location);
4150 return ERR_PTR(ret);
4152 if (location.objectid == 0)
4155 if (location.type == BTRFS_INODE_ITEM_KEY) {
4156 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4160 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4162 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4163 ret = fixup_tree_root_location(root, dir, dentry,
4164 &location, &sub_root);
4167 inode = ERR_PTR(ret);
4169 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4171 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4173 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4175 if (!IS_ERR(inode) && root != sub_root) {
4176 down_read(&root->fs_info->cleanup_work_sem);
4177 if (!(inode->i_sb->s_flags & MS_RDONLY))
4178 ret = btrfs_orphan_cleanup(sub_root);
4179 up_read(&root->fs_info->cleanup_work_sem);
4181 inode = ERR_PTR(ret);
4187 static int btrfs_dentry_delete(const struct dentry *dentry)
4189 struct btrfs_root *root;
4191 if (!dentry->d_inode && !IS_ROOT(dentry))
4192 dentry = dentry->d_parent;
4194 if (dentry->d_inode) {
4195 root = BTRFS_I(dentry->d_inode)->root;
4196 if (btrfs_root_refs(&root->root_item) == 0)
4202 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4203 struct nameidata *nd)
4205 struct inode *inode;
4207 inode = btrfs_lookup_dentry(dir, dentry);
4209 return ERR_CAST(inode);
4211 return d_splice_alias(inode, dentry);
4214 static unsigned char btrfs_filetype_table[] = {
4215 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4218 static int btrfs_real_readdir(struct file *filp, void *dirent,
4221 struct inode *inode = filp->f_dentry->d_inode;
4222 struct btrfs_root *root = BTRFS_I(inode)->root;
4223 struct btrfs_item *item;
4224 struct btrfs_dir_item *di;
4225 struct btrfs_key key;
4226 struct btrfs_key found_key;
4227 struct btrfs_path *path;
4229 struct extent_buffer *leaf;
4231 unsigned char d_type;
4236 int key_type = BTRFS_DIR_INDEX_KEY;
4241 /* FIXME, use a real flag for deciding about the key type */
4242 if (root->fs_info->tree_root == root)
4243 key_type = BTRFS_DIR_ITEM_KEY;
4245 /* special case for "." */
4246 if (filp->f_pos == 0) {
4247 over = filldir(dirent, ".", 1,
4254 /* special case for .., just use the back ref */
4255 if (filp->f_pos == 1) {
4256 u64 pino = parent_ino(filp->f_path.dentry);
4257 over = filldir(dirent, "..", 2,
4263 path = btrfs_alloc_path();
4266 btrfs_set_key_type(&key, key_type);
4267 key.offset = filp->f_pos;
4268 key.objectid = inode->i_ino;
4270 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4275 leaf = path->nodes[0];
4276 slot = path->slots[0];
4277 if (slot >= btrfs_header_nritems(leaf)) {
4278 ret = btrfs_next_leaf(root, path);
4286 item = btrfs_item_nr(leaf, slot);
4287 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4289 if (found_key.objectid != key.objectid)
4291 if (btrfs_key_type(&found_key) != key_type)
4293 if (found_key.offset < filp->f_pos)
4296 filp->f_pos = found_key.offset;
4298 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4300 di_total = btrfs_item_size(leaf, item);
4302 while (di_cur < di_total) {
4303 struct btrfs_key location;
4305 if (verify_dir_item(root, leaf, di))
4308 name_len = btrfs_dir_name_len(leaf, di);
4309 if (name_len <= sizeof(tmp_name)) {
4310 name_ptr = tmp_name;
4312 name_ptr = kmalloc(name_len, GFP_NOFS);
4318 read_extent_buffer(leaf, name_ptr,
4319 (unsigned long)(di + 1), name_len);
4321 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4322 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4324 /* is this a reference to our own snapshot? If so
4327 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4328 location.objectid == root->root_key.objectid) {
4332 over = filldir(dirent, name_ptr, name_len,
4333 found_key.offset, location.objectid,
4337 if (name_ptr != tmp_name)
4342 di_len = btrfs_dir_name_len(leaf, di) +
4343 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4345 di = (struct btrfs_dir_item *)((char *)di + di_len);
4351 /* Reached end of directory/root. Bump pos past the last item. */
4352 if (key_type == BTRFS_DIR_INDEX_KEY)
4354 * 32-bit glibc will use getdents64, but then strtol -
4355 * so the last number we can serve is this.
4357 filp->f_pos = 0x7fffffff;
4363 btrfs_free_path(path);
4367 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4369 struct btrfs_root *root = BTRFS_I(inode)->root;
4370 struct btrfs_trans_handle *trans;
4372 bool nolock = false;
4374 if (BTRFS_I(inode)->dummy_inode)
4378 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4380 if (wbc->sync_mode == WB_SYNC_ALL) {
4382 trans = btrfs_join_transaction_nolock(root, 1);
4384 trans = btrfs_join_transaction(root, 1);
4386 return PTR_ERR(trans);
4387 btrfs_set_trans_block_group(trans, inode);
4389 ret = btrfs_end_transaction_nolock(trans, root);
4391 ret = btrfs_commit_transaction(trans, root);
4397 * This is somewhat expensive, updating the tree every time the
4398 * inode changes. But, it is most likely to find the inode in cache.
4399 * FIXME, needs more benchmarking...there are no reasons other than performance
4400 * to keep or drop this code.
4402 void btrfs_dirty_inode(struct inode *inode)
4404 struct btrfs_root *root = BTRFS_I(inode)->root;
4405 struct btrfs_trans_handle *trans;
4408 if (BTRFS_I(inode)->dummy_inode)
4411 trans = btrfs_join_transaction(root, 1);
4412 BUG_ON(IS_ERR(trans));
4413 btrfs_set_trans_block_group(trans, inode);
4415 ret = btrfs_update_inode(trans, root, inode);
4416 if (ret && ret == -ENOSPC) {
4417 /* whoops, lets try again with the full transaction */
4418 btrfs_end_transaction(trans, root);
4419 trans = btrfs_start_transaction(root, 1);
4420 if (IS_ERR(trans)) {
4421 if (printk_ratelimit()) {
4422 printk(KERN_ERR "btrfs: fail to "
4423 "dirty inode %lu error %ld\n",
4424 inode->i_ino, PTR_ERR(trans));
4428 btrfs_set_trans_block_group(trans, inode);
4430 ret = btrfs_update_inode(trans, root, inode);
4432 if (printk_ratelimit()) {
4433 printk(KERN_ERR "btrfs: fail to "
4434 "dirty inode %lu error %d\n",
4439 btrfs_end_transaction(trans, root);
4443 * find the highest existing sequence number in a directory
4444 * and then set the in-memory index_cnt variable to reflect
4445 * free sequence numbers
4447 static int btrfs_set_inode_index_count(struct inode *inode)
4449 struct btrfs_root *root = BTRFS_I(inode)->root;
4450 struct btrfs_key key, found_key;
4451 struct btrfs_path *path;
4452 struct extent_buffer *leaf;
4455 key.objectid = inode->i_ino;
4456 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4457 key.offset = (u64)-1;
4459 path = btrfs_alloc_path();
4463 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4466 /* FIXME: we should be able to handle this */
4472 * MAGIC NUMBER EXPLANATION:
4473 * since we search a directory based on f_pos we have to start at 2
4474 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4475 * else has to start at 2
4477 if (path->slots[0] == 0) {
4478 BTRFS_I(inode)->index_cnt = 2;
4484 leaf = path->nodes[0];
4485 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4487 if (found_key.objectid != inode->i_ino ||
4488 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4489 BTRFS_I(inode)->index_cnt = 2;
4493 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4495 btrfs_free_path(path);
4500 * helper to find a free sequence number in a given directory. This current
4501 * code is very simple, later versions will do smarter things in the btree
4503 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4507 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4508 ret = btrfs_set_inode_index_count(dir);
4513 *index = BTRFS_I(dir)->index_cnt;
4514 BTRFS_I(dir)->index_cnt++;
4519 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4520 struct btrfs_root *root,
4522 const char *name, int name_len,
4523 u64 ref_objectid, u64 objectid,
4524 u64 alloc_hint, int mode, u64 *index)
4526 struct inode *inode;
4527 struct btrfs_inode_item *inode_item;
4528 struct btrfs_key *location;
4529 struct btrfs_path *path;
4530 struct btrfs_inode_ref *ref;
4531 struct btrfs_key key[2];
4537 path = btrfs_alloc_path();
4540 inode = new_inode(root->fs_info->sb);
4542 btrfs_free_path(path);
4543 return ERR_PTR(-ENOMEM);
4547 * we have to initialize this early, so we can reclaim the inode
4548 * number if we fail afterwards in this function.
4550 inode->i_ino = objectid;
4553 trace_btrfs_inode_request(dir);
4555 ret = btrfs_set_inode_index(dir, index);
4557 btrfs_free_path(path);
4559 return ERR_PTR(ret);
4563 * index_cnt is ignored for everything but a dir,
4564 * btrfs_get_inode_index_count has an explanation for the magic
4567 BTRFS_I(inode)->index_cnt = 2;
4568 BTRFS_I(inode)->root = root;
4569 BTRFS_I(inode)->generation = trans->transid;
4570 inode->i_generation = BTRFS_I(inode)->generation;
4571 btrfs_set_inode_space_info(root, inode);
4577 BTRFS_I(inode)->block_group =
4578 btrfs_find_block_group(root, 0, alloc_hint, owner);
4580 key[0].objectid = objectid;
4581 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4584 key[1].objectid = objectid;
4585 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4586 key[1].offset = ref_objectid;
4588 sizes[0] = sizeof(struct btrfs_inode_item);
4589 sizes[1] = name_len + sizeof(*ref);
4591 path->leave_spinning = 1;
4592 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4596 inode_init_owner(inode, dir, mode);
4597 inode_set_bytes(inode, 0);
4598 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4599 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4600 struct btrfs_inode_item);
4601 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4603 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4604 struct btrfs_inode_ref);
4605 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4606 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4607 ptr = (unsigned long)(ref + 1);
4608 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4610 btrfs_mark_buffer_dirty(path->nodes[0]);
4611 btrfs_free_path(path);
4613 location = &BTRFS_I(inode)->location;
4614 location->objectid = objectid;
4615 location->offset = 0;
4616 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4618 btrfs_inherit_iflags(inode, dir);
4620 if ((mode & S_IFREG)) {
4621 if (btrfs_test_opt(root, NODATASUM))
4622 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4623 if (btrfs_test_opt(root, NODATACOW) ||
4624 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4625 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4628 insert_inode_hash(inode);
4629 inode_tree_add(inode);
4631 trace_btrfs_inode_new(inode);
4636 BTRFS_I(dir)->index_cnt--;
4637 btrfs_free_path(path);
4639 return ERR_PTR(ret);
4642 static inline u8 btrfs_inode_type(struct inode *inode)
4644 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4648 * utility function to add 'inode' into 'parent_inode' with
4649 * a give name and a given sequence number.
4650 * if 'add_backref' is true, also insert a backref from the
4651 * inode to the parent directory.
4653 int btrfs_add_link(struct btrfs_trans_handle *trans,
4654 struct inode *parent_inode, struct inode *inode,
4655 const char *name, int name_len, int add_backref, u64 index)
4658 struct btrfs_key key;
4659 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4661 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4662 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4664 key.objectid = inode->i_ino;
4665 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4669 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4670 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4671 key.objectid, root->root_key.objectid,
4672 parent_inode->i_ino,
4673 index, name, name_len);
4674 } else if (add_backref) {
4675 ret = btrfs_insert_inode_ref(trans, root,
4676 name, name_len, inode->i_ino,
4677 parent_inode->i_ino, index);
4681 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4682 parent_inode->i_ino, &key,
4683 btrfs_inode_type(inode), index);
4686 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4688 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4689 ret = btrfs_update_inode(trans, root, parent_inode);
4694 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4695 struct inode *dir, struct dentry *dentry,
4696 struct inode *inode, int backref, u64 index)
4698 int err = btrfs_add_link(trans, dir, inode,
4699 dentry->d_name.name, dentry->d_name.len,
4702 d_instantiate(dentry, inode);
4710 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4711 int mode, dev_t rdev)
4713 struct btrfs_trans_handle *trans;
4714 struct btrfs_root *root = BTRFS_I(dir)->root;
4715 struct inode *inode = NULL;
4719 unsigned long nr = 0;
4722 if (!new_valid_dev(rdev))
4726 * 2 for inode item and ref
4728 * 1 for xattr if selinux is on
4730 trans = btrfs_start_transaction(root, 5);
4732 return PTR_ERR(trans);
4734 btrfs_set_trans_block_group(trans, dir);
4736 err = btrfs_find_free_ino(root, &objectid);
4740 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4741 dentry->d_name.len, dir->i_ino, objectid,
4742 BTRFS_I(dir)->block_group, mode, &index);
4743 err = PTR_ERR(inode);
4747 err = btrfs_init_inode_security(trans, inode, dir);
4753 btrfs_set_trans_block_group(trans, inode);
4754 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4758 inode->i_op = &btrfs_special_inode_operations;
4759 init_special_inode(inode, inode->i_mode, rdev);
4760 btrfs_update_inode(trans, root, inode);
4762 btrfs_update_inode_block_group(trans, inode);
4763 btrfs_update_inode_block_group(trans, dir);
4765 nr = trans->blocks_used;
4766 btrfs_end_transaction_throttle(trans, root);
4767 btrfs_btree_balance_dirty(root, nr);
4769 inode_dec_link_count(inode);
4775 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4776 int mode, struct nameidata *nd)
4778 struct btrfs_trans_handle *trans;
4779 struct btrfs_root *root = BTRFS_I(dir)->root;
4780 struct inode *inode = NULL;
4783 unsigned long nr = 0;
4788 * 2 for inode item and ref
4790 * 1 for xattr if selinux is on
4792 trans = btrfs_start_transaction(root, 5);
4794 return PTR_ERR(trans);
4796 btrfs_set_trans_block_group(trans, dir);
4798 err = btrfs_find_free_ino(root, &objectid);
4802 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4803 dentry->d_name.len, dir->i_ino, objectid,
4804 BTRFS_I(dir)->block_group, mode, &index);
4805 err = PTR_ERR(inode);
4809 err = btrfs_init_inode_security(trans, inode, dir);
4815 btrfs_set_trans_block_group(trans, inode);
4816 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4820 inode->i_mapping->a_ops = &btrfs_aops;
4821 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4822 inode->i_fop = &btrfs_file_operations;
4823 inode->i_op = &btrfs_file_inode_operations;
4824 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4826 btrfs_update_inode_block_group(trans, inode);
4827 btrfs_update_inode_block_group(trans, dir);
4829 nr = trans->blocks_used;
4830 btrfs_end_transaction_throttle(trans, root);
4832 inode_dec_link_count(inode);
4835 btrfs_btree_balance_dirty(root, nr);
4839 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4840 struct dentry *dentry)
4842 struct btrfs_trans_handle *trans;
4843 struct btrfs_root *root = BTRFS_I(dir)->root;
4844 struct inode *inode = old_dentry->d_inode;
4846 unsigned long nr = 0;
4850 if (inode->i_nlink == 0)
4853 /* do not allow sys_link's with other subvols of the same device */
4854 if (root->objectid != BTRFS_I(inode)->root->objectid)
4857 if (inode->i_nlink == ~0U)
4860 err = btrfs_set_inode_index(dir, &index);
4865 * 2 items for inode and inode ref
4866 * 2 items for dir items
4867 * 1 item for parent inode
4869 trans = btrfs_start_transaction(root, 5);
4870 if (IS_ERR(trans)) {
4871 err = PTR_ERR(trans);
4875 btrfs_inc_nlink(inode);
4876 inode->i_ctime = CURRENT_TIME;
4878 btrfs_set_trans_block_group(trans, dir);
4881 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4886 struct dentry *parent = dget_parent(dentry);
4887 btrfs_update_inode_block_group(trans, dir);
4888 err = btrfs_update_inode(trans, root, inode);
4890 btrfs_log_new_name(trans, inode, NULL, parent);
4894 nr = trans->blocks_used;
4895 btrfs_end_transaction_throttle(trans, root);
4898 inode_dec_link_count(inode);
4901 btrfs_btree_balance_dirty(root, nr);
4905 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4907 struct inode *inode = NULL;
4908 struct btrfs_trans_handle *trans;
4909 struct btrfs_root *root = BTRFS_I(dir)->root;
4911 int drop_on_err = 0;
4914 unsigned long nr = 1;
4917 * 2 items for inode and ref
4918 * 2 items for dir items
4919 * 1 for xattr if selinux is on
4921 trans = btrfs_start_transaction(root, 5);
4923 return PTR_ERR(trans);
4924 btrfs_set_trans_block_group(trans, dir);
4926 err = btrfs_find_free_ino(root, &objectid);
4930 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4931 dentry->d_name.len, dir->i_ino, objectid,
4932 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4934 if (IS_ERR(inode)) {
4935 err = PTR_ERR(inode);
4941 err = btrfs_init_inode_security(trans, inode, dir);
4945 inode->i_op = &btrfs_dir_inode_operations;
4946 inode->i_fop = &btrfs_dir_file_operations;
4947 btrfs_set_trans_block_group(trans, inode);
4949 btrfs_i_size_write(inode, 0);
4950 err = btrfs_update_inode(trans, root, inode);
4954 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4955 dentry->d_name.len, 0, index);
4959 d_instantiate(dentry, inode);
4961 btrfs_update_inode_block_group(trans, inode);
4962 btrfs_update_inode_block_group(trans, dir);
4965 nr = trans->blocks_used;
4966 btrfs_end_transaction_throttle(trans, root);
4969 btrfs_btree_balance_dirty(root, nr);
4973 /* helper for btfs_get_extent. Given an existing extent in the tree,
4974 * and an extent that you want to insert, deal with overlap and insert
4975 * the new extent into the tree.
4977 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4978 struct extent_map *existing,
4979 struct extent_map *em,
4980 u64 map_start, u64 map_len)
4984 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4985 start_diff = map_start - em->start;
4986 em->start = map_start;
4988 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4989 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4990 em->block_start += start_diff;
4991 em->block_len -= start_diff;
4993 return add_extent_mapping(em_tree, em);
4996 static noinline int uncompress_inline(struct btrfs_path *path,
4997 struct inode *inode, struct page *page,
4998 size_t pg_offset, u64 extent_offset,
4999 struct btrfs_file_extent_item *item)
5002 struct extent_buffer *leaf = path->nodes[0];
5005 unsigned long inline_size;
5009 WARN_ON(pg_offset != 0);
5010 compress_type = btrfs_file_extent_compression(leaf, item);
5011 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5012 inline_size = btrfs_file_extent_inline_item_len(leaf,
5013 btrfs_item_nr(leaf, path->slots[0]));
5014 tmp = kmalloc(inline_size, GFP_NOFS);
5015 ptr = btrfs_file_extent_inline_start(item);
5017 read_extent_buffer(leaf, tmp, ptr, inline_size);
5019 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5020 ret = btrfs_decompress(compress_type, tmp, page,
5021 extent_offset, inline_size, max_size);
5023 char *kaddr = kmap_atomic(page, KM_USER0);
5024 unsigned long copy_size = min_t(u64,
5025 PAGE_CACHE_SIZE - pg_offset,
5026 max_size - extent_offset);
5027 memset(kaddr + pg_offset, 0, copy_size);
5028 kunmap_atomic(kaddr, KM_USER0);
5035 * a bit scary, this does extent mapping from logical file offset to the disk.
5036 * the ugly parts come from merging extents from the disk with the in-ram
5037 * representation. This gets more complex because of the data=ordered code,
5038 * where the in-ram extents might be locked pending data=ordered completion.
5040 * This also copies inline extents directly into the page.
5043 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5044 size_t pg_offset, u64 start, u64 len,
5050 u64 extent_start = 0;
5052 u64 objectid = inode->i_ino;
5054 struct btrfs_path *path = NULL;
5055 struct btrfs_root *root = BTRFS_I(inode)->root;
5056 struct btrfs_file_extent_item *item;
5057 struct extent_buffer *leaf;
5058 struct btrfs_key found_key;
5059 struct extent_map *em = NULL;
5060 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5061 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5062 struct btrfs_trans_handle *trans = NULL;
5066 read_lock(&em_tree->lock);
5067 em = lookup_extent_mapping(em_tree, start, len);
5069 em->bdev = root->fs_info->fs_devices->latest_bdev;
5070 read_unlock(&em_tree->lock);
5073 if (em->start > start || em->start + em->len <= start)
5074 free_extent_map(em);
5075 else if (em->block_start == EXTENT_MAP_INLINE && page)
5076 free_extent_map(em);
5080 em = alloc_extent_map(GFP_NOFS);
5085 em->bdev = root->fs_info->fs_devices->latest_bdev;
5086 em->start = EXTENT_MAP_HOLE;
5087 em->orig_start = EXTENT_MAP_HOLE;
5089 em->block_len = (u64)-1;
5092 path = btrfs_alloc_path();
5096 ret = btrfs_lookup_file_extent(trans, root, path,
5097 objectid, start, trans != NULL);
5104 if (path->slots[0] == 0)
5109 leaf = path->nodes[0];
5110 item = btrfs_item_ptr(leaf, path->slots[0],
5111 struct btrfs_file_extent_item);
5112 /* are we inside the extent that was found? */
5113 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5114 found_type = btrfs_key_type(&found_key);
5115 if (found_key.objectid != objectid ||
5116 found_type != BTRFS_EXTENT_DATA_KEY) {
5120 found_type = btrfs_file_extent_type(leaf, item);
5121 extent_start = found_key.offset;
5122 compress_type = btrfs_file_extent_compression(leaf, item);
5123 if (found_type == BTRFS_FILE_EXTENT_REG ||
5124 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5125 extent_end = extent_start +
5126 btrfs_file_extent_num_bytes(leaf, item);
5127 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5129 size = btrfs_file_extent_inline_len(leaf, item);
5130 extent_end = (extent_start + size + root->sectorsize - 1) &
5131 ~((u64)root->sectorsize - 1);
5134 if (start >= extent_end) {
5136 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5137 ret = btrfs_next_leaf(root, path);
5144 leaf = path->nodes[0];
5146 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5147 if (found_key.objectid != objectid ||
5148 found_key.type != BTRFS_EXTENT_DATA_KEY)
5150 if (start + len <= found_key.offset)
5153 em->len = found_key.offset - start;
5157 if (found_type == BTRFS_FILE_EXTENT_REG ||
5158 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5159 em->start = extent_start;
5160 em->len = extent_end - extent_start;
5161 em->orig_start = extent_start -
5162 btrfs_file_extent_offset(leaf, item);
5163 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5165 em->block_start = EXTENT_MAP_HOLE;
5168 if (compress_type != BTRFS_COMPRESS_NONE) {
5169 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5170 em->compress_type = compress_type;
5171 em->block_start = bytenr;
5172 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5175 bytenr += btrfs_file_extent_offset(leaf, item);
5176 em->block_start = bytenr;
5177 em->block_len = em->len;
5178 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5179 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5182 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5186 size_t extent_offset;
5189 em->block_start = EXTENT_MAP_INLINE;
5190 if (!page || create) {
5191 em->start = extent_start;
5192 em->len = extent_end - extent_start;
5196 size = btrfs_file_extent_inline_len(leaf, item);
5197 extent_offset = page_offset(page) + pg_offset - extent_start;
5198 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5199 size - extent_offset);
5200 em->start = extent_start + extent_offset;
5201 em->len = (copy_size + root->sectorsize - 1) &
5202 ~((u64)root->sectorsize - 1);
5203 em->orig_start = EXTENT_MAP_INLINE;
5204 if (compress_type) {
5205 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5206 em->compress_type = compress_type;
5208 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5209 if (create == 0 && !PageUptodate(page)) {
5210 if (btrfs_file_extent_compression(leaf, item) !=
5211 BTRFS_COMPRESS_NONE) {
5212 ret = uncompress_inline(path, inode, page,
5214 extent_offset, item);
5218 read_extent_buffer(leaf, map + pg_offset, ptr,
5220 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5221 memset(map + pg_offset + copy_size, 0,
5222 PAGE_CACHE_SIZE - pg_offset -
5227 flush_dcache_page(page);
5228 } else if (create && PageUptodate(page)) {
5232 free_extent_map(em);
5234 btrfs_release_path(root, path);
5235 trans = btrfs_join_transaction(root, 1);
5237 return ERR_CAST(trans);
5241 write_extent_buffer(leaf, map + pg_offset, ptr,
5244 btrfs_mark_buffer_dirty(leaf);
5246 set_extent_uptodate(io_tree, em->start,
5247 extent_map_end(em) - 1, NULL, GFP_NOFS);
5250 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5257 em->block_start = EXTENT_MAP_HOLE;
5258 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5260 btrfs_release_path(root, path);
5261 if (em->start > start || extent_map_end(em) <= start) {
5262 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5263 "[%llu %llu]\n", (unsigned long long)em->start,
5264 (unsigned long long)em->len,
5265 (unsigned long long)start,
5266 (unsigned long long)len);
5272 write_lock(&em_tree->lock);
5273 ret = add_extent_mapping(em_tree, em);
5274 /* it is possible that someone inserted the extent into the tree
5275 * while we had the lock dropped. It is also possible that
5276 * an overlapping map exists in the tree
5278 if (ret == -EEXIST) {
5279 struct extent_map *existing;
5283 existing = lookup_extent_mapping(em_tree, start, len);
5284 if (existing && (existing->start > start ||
5285 existing->start + existing->len <= start)) {
5286 free_extent_map(existing);
5290 existing = lookup_extent_mapping(em_tree, em->start,
5293 err = merge_extent_mapping(em_tree, existing,
5296 free_extent_map(existing);
5298 free_extent_map(em);
5303 free_extent_map(em);
5307 free_extent_map(em);
5312 write_unlock(&em_tree->lock);
5315 trace_btrfs_get_extent(root, em);
5318 btrfs_free_path(path);
5320 ret = btrfs_end_transaction(trans, root);
5325 free_extent_map(em);
5326 return ERR_PTR(err);
5331 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5332 size_t pg_offset, u64 start, u64 len,
5335 struct extent_map *em;
5336 struct extent_map *hole_em = NULL;
5337 u64 range_start = start;
5343 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5348 * if our em maps to a hole, there might
5349 * actually be delalloc bytes behind it
5351 if (em->block_start != EXTENT_MAP_HOLE)
5357 /* check to see if we've wrapped (len == -1 or similar) */
5366 /* ok, we didn't find anything, lets look for delalloc */
5367 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5368 end, len, EXTENT_DELALLOC, 1);
5369 found_end = range_start + found;
5370 if (found_end < range_start)
5371 found_end = (u64)-1;
5374 * we didn't find anything useful, return
5375 * the original results from get_extent()
5377 if (range_start > end || found_end <= start) {
5383 /* adjust the range_start to make sure it doesn't
5384 * go backwards from the start they passed in
5386 range_start = max(start,range_start);
5387 found = found_end - range_start;
5390 u64 hole_start = start;
5393 em = alloc_extent_map(GFP_NOFS);
5399 * when btrfs_get_extent can't find anything it
5400 * returns one huge hole
5402 * make sure what it found really fits our range, and
5403 * adjust to make sure it is based on the start from
5407 u64 calc_end = extent_map_end(hole_em);
5409 if (calc_end <= start || (hole_em->start > end)) {
5410 free_extent_map(hole_em);
5413 hole_start = max(hole_em->start, start);
5414 hole_len = calc_end - hole_start;
5418 if (hole_em && range_start > hole_start) {
5419 /* our hole starts before our delalloc, so we
5420 * have to return just the parts of the hole
5421 * that go until the delalloc starts
5423 em->len = min(hole_len,
5424 range_start - hole_start);
5425 em->start = hole_start;
5426 em->orig_start = hole_start;
5428 * don't adjust block start at all,
5429 * it is fixed at EXTENT_MAP_HOLE
5431 em->block_start = hole_em->block_start;
5432 em->block_len = hole_len;
5434 em->start = range_start;
5436 em->orig_start = range_start;
5437 em->block_start = EXTENT_MAP_DELALLOC;
5438 em->block_len = found;
5440 } else if (hole_em) {
5445 free_extent_map(hole_em);
5447 free_extent_map(em);
5448 return ERR_PTR(err);
5453 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5454 struct extent_map *em,
5457 struct btrfs_root *root = BTRFS_I(inode)->root;
5458 struct btrfs_trans_handle *trans;
5459 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5460 struct btrfs_key ins;
5463 bool insert = false;
5466 * Ok if the extent map we looked up is a hole and is for the exact
5467 * range we want, there is no reason to allocate a new one, however if
5468 * it is not right then we need to free this one and drop the cache for
5471 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5473 free_extent_map(em);
5476 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5479 trans = btrfs_join_transaction(root, 0);
5481 return ERR_CAST(trans);
5483 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5485 alloc_hint = get_extent_allocation_hint(inode, start, len);
5486 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5487 alloc_hint, (u64)-1, &ins, 1);
5494 em = alloc_extent_map(GFP_NOFS);
5496 em = ERR_PTR(-ENOMEM);
5502 em->orig_start = em->start;
5503 em->len = ins.offset;
5505 em->block_start = ins.objectid;
5506 em->block_len = ins.offset;
5507 em->bdev = root->fs_info->fs_devices->latest_bdev;
5510 * We need to do this because if we're using the original em we searched
5511 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5514 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5517 write_lock(&em_tree->lock);
5518 ret = add_extent_mapping(em_tree, em);
5519 write_unlock(&em_tree->lock);
5522 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5525 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5526 ins.offset, ins.offset, 0);
5528 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5532 btrfs_end_transaction(trans, root);
5537 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5538 * block must be cow'd
5540 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5541 struct inode *inode, u64 offset, u64 len)
5543 struct btrfs_path *path;
5545 struct extent_buffer *leaf;
5546 struct btrfs_root *root = BTRFS_I(inode)->root;
5547 struct btrfs_file_extent_item *fi;
5548 struct btrfs_key key;
5556 path = btrfs_alloc_path();
5560 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5565 slot = path->slots[0];
5568 /* can't find the item, must cow */
5575 leaf = path->nodes[0];
5576 btrfs_item_key_to_cpu(leaf, &key, slot);
5577 if (key.objectid != inode->i_ino ||
5578 key.type != BTRFS_EXTENT_DATA_KEY) {
5579 /* not our file or wrong item type, must cow */
5583 if (key.offset > offset) {
5584 /* Wrong offset, must cow */
5588 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5589 found_type = btrfs_file_extent_type(leaf, fi);
5590 if (found_type != BTRFS_FILE_EXTENT_REG &&
5591 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5592 /* not a regular extent, must cow */
5595 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5596 backref_offset = btrfs_file_extent_offset(leaf, fi);
5598 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5599 if (extent_end < offset + len) {
5600 /* extent doesn't include our full range, must cow */
5604 if (btrfs_extent_readonly(root, disk_bytenr))
5608 * look for other files referencing this extent, if we
5609 * find any we must cow
5611 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5612 key.offset - backref_offset, disk_bytenr))
5616 * adjust disk_bytenr and num_bytes to cover just the bytes
5617 * in this extent we are about to write. If there
5618 * are any csums in that range we have to cow in order
5619 * to keep the csums correct
5621 disk_bytenr += backref_offset;
5622 disk_bytenr += offset - key.offset;
5623 num_bytes = min(offset + len, extent_end) - offset;
5624 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5627 * all of the above have passed, it is safe to overwrite this extent
5632 btrfs_free_path(path);
5636 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5637 struct buffer_head *bh_result, int create)
5639 struct extent_map *em;
5640 struct btrfs_root *root = BTRFS_I(inode)->root;
5641 u64 start = iblock << inode->i_blkbits;
5642 u64 len = bh_result->b_size;
5643 struct btrfs_trans_handle *trans;
5645 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5650 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5651 * io. INLINE is special, and we could probably kludge it in here, but
5652 * it's still buffered so for safety lets just fall back to the generic
5655 * For COMPRESSED we _have_ to read the entire extent in so we can
5656 * decompress it, so there will be buffering required no matter what we
5657 * do, so go ahead and fallback to buffered.
5659 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5660 * to buffered IO. Don't blame me, this is the price we pay for using
5663 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5664 em->block_start == EXTENT_MAP_INLINE) {
5665 free_extent_map(em);
5669 /* Just a good old fashioned hole, return */
5670 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5671 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5672 free_extent_map(em);
5673 /* DIO will do one hole at a time, so just unlock a sector */
5674 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5675 start + root->sectorsize - 1, GFP_NOFS);
5680 * We don't allocate a new extent in the following cases
5682 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5684 * 2) The extent is marked as PREALLOC. We're good to go here and can
5685 * just use the extent.
5689 len = em->len - (start - em->start);
5693 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5694 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5695 em->block_start != EXTENT_MAP_HOLE)) {
5700 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5701 type = BTRFS_ORDERED_PREALLOC;
5703 type = BTRFS_ORDERED_NOCOW;
5704 len = min(len, em->len - (start - em->start));
5705 block_start = em->block_start + (start - em->start);
5708 * we're not going to log anything, but we do need
5709 * to make sure the current transaction stays open
5710 * while we look for nocow cross refs
5712 trans = btrfs_join_transaction(root, 0);
5716 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5717 ret = btrfs_add_ordered_extent_dio(inode, start,
5718 block_start, len, len, type);
5719 btrfs_end_transaction(trans, root);
5721 free_extent_map(em);
5726 btrfs_end_transaction(trans, root);
5730 * this will cow the extent, reset the len in case we changed
5733 len = bh_result->b_size;
5734 em = btrfs_new_extent_direct(inode, em, start, len);
5737 len = min(len, em->len - (start - em->start));
5739 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5740 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5743 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5745 bh_result->b_size = len;
5746 bh_result->b_bdev = em->bdev;
5747 set_buffer_mapped(bh_result);
5748 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5749 set_buffer_new(bh_result);
5751 free_extent_map(em);
5756 struct btrfs_dio_private {
5757 struct inode *inode;
5764 /* number of bios pending for this dio */
5765 atomic_t pending_bios;
5770 struct bio *orig_bio;
5773 static void btrfs_endio_direct_read(struct bio *bio, int err)
5775 struct btrfs_dio_private *dip = bio->bi_private;
5776 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5777 struct bio_vec *bvec = bio->bi_io_vec;
5778 struct inode *inode = dip->inode;
5779 struct btrfs_root *root = BTRFS_I(inode)->root;
5781 u32 *private = dip->csums;
5783 start = dip->logical_offset;
5785 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5786 struct page *page = bvec->bv_page;
5789 unsigned long flags;
5791 local_irq_save(flags);
5792 kaddr = kmap_atomic(page, KM_IRQ0);
5793 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5794 csum, bvec->bv_len);
5795 btrfs_csum_final(csum, (char *)&csum);
5796 kunmap_atomic(kaddr, KM_IRQ0);
5797 local_irq_restore(flags);
5799 flush_dcache_page(bvec->bv_page);
5800 if (csum != *private) {
5801 printk(KERN_ERR "btrfs csum failed ino %lu off"
5802 " %llu csum %u private %u\n",
5803 inode->i_ino, (unsigned long long)start,
5809 start += bvec->bv_len;
5812 } while (bvec <= bvec_end);
5814 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5815 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5816 bio->bi_private = dip->private;
5821 /* If we had a csum failure make sure to clear the uptodate flag */
5823 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5824 dio_end_io(bio, err);
5827 static void btrfs_endio_direct_write(struct bio *bio, int err)
5829 struct btrfs_dio_private *dip = bio->bi_private;
5830 struct inode *inode = dip->inode;
5831 struct btrfs_root *root = BTRFS_I(inode)->root;
5832 struct btrfs_trans_handle *trans;
5833 struct btrfs_ordered_extent *ordered = NULL;
5834 struct extent_state *cached_state = NULL;
5835 u64 ordered_offset = dip->logical_offset;
5836 u64 ordered_bytes = dip->bytes;
5842 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5850 trans = btrfs_join_transaction(root, 1);
5851 if (IS_ERR(trans)) {
5855 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5857 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5858 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5860 ret = btrfs_update_inode(trans, root, inode);
5865 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5866 ordered->file_offset + ordered->len - 1, 0,
5867 &cached_state, GFP_NOFS);
5869 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5870 ret = btrfs_mark_extent_written(trans, inode,
5871 ordered->file_offset,
5872 ordered->file_offset +
5879 ret = insert_reserved_file_extent(trans, inode,
5880 ordered->file_offset,
5886 BTRFS_FILE_EXTENT_REG);
5887 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5888 ordered->file_offset, ordered->len);
5896 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5897 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5899 btrfs_update_inode(trans, root, inode);
5902 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5903 ordered->file_offset + ordered->len - 1,
5904 &cached_state, GFP_NOFS);
5906 btrfs_delalloc_release_metadata(inode, ordered->len);
5907 btrfs_end_transaction(trans, root);
5908 ordered_offset = ordered->file_offset + ordered->len;
5909 btrfs_put_ordered_extent(ordered);
5910 btrfs_put_ordered_extent(ordered);
5914 * our bio might span multiple ordered extents. If we haven't
5915 * completed the accounting for the whole dio, go back and try again
5917 if (ordered_offset < dip->logical_offset + dip->bytes) {
5918 ordered_bytes = dip->logical_offset + dip->bytes -
5923 bio->bi_private = dip->private;
5928 /* If we had an error make sure to clear the uptodate flag */
5930 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5931 dio_end_io(bio, err);
5934 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5935 struct bio *bio, int mirror_num,
5936 unsigned long bio_flags, u64 offset)
5939 struct btrfs_root *root = BTRFS_I(inode)->root;
5940 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5945 static void btrfs_end_dio_bio(struct bio *bio, int err)
5947 struct btrfs_dio_private *dip = bio->bi_private;
5950 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5951 "sector %#Lx len %u err no %d\n",
5952 dip->inode->i_ino, bio->bi_rw,
5953 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5957 * before atomic variable goto zero, we must make sure
5958 * dip->errors is perceived to be set.
5960 smp_mb__before_atomic_dec();
5963 /* if there are more bios still pending for this dio, just exit */
5964 if (!atomic_dec_and_test(&dip->pending_bios))
5968 bio_io_error(dip->orig_bio);
5970 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5971 bio_endio(dip->orig_bio, 0);
5977 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5978 u64 first_sector, gfp_t gfp_flags)
5980 int nr_vecs = bio_get_nr_vecs(bdev);
5981 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5984 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5985 int rw, u64 file_offset, int skip_sum,
5986 u32 *csums, int async_submit)
5988 int write = rw & REQ_WRITE;
5989 struct btrfs_root *root = BTRFS_I(inode)->root;
5993 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6000 if (write && async_submit) {
6001 ret = btrfs_wq_submit_bio(root->fs_info,
6002 inode, rw, bio, 0, 0,
6004 __btrfs_submit_bio_start_direct_io,
6005 __btrfs_submit_bio_done);
6009 * If we aren't doing async submit, calculate the csum of the
6012 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6015 } else if (!skip_sum) {
6016 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6017 file_offset, csums);
6023 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6029 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6032 struct inode *inode = dip->inode;
6033 struct btrfs_root *root = BTRFS_I(inode)->root;
6034 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6036 struct bio *orig_bio = dip->orig_bio;
6037 struct bio_vec *bvec = orig_bio->bi_io_vec;
6038 u64 start_sector = orig_bio->bi_sector;
6039 u64 file_offset = dip->logical_offset;
6043 u32 *csums = dip->csums;
6045 int async_submit = 0;
6046 int write = rw & REQ_WRITE;
6048 map_length = orig_bio->bi_size;
6049 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6050 &map_length, NULL, 0);
6056 if (map_length >= orig_bio->bi_size) {
6062 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6065 bio->bi_private = dip;
6066 bio->bi_end_io = btrfs_end_dio_bio;
6067 atomic_inc(&dip->pending_bios);
6069 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6070 if (unlikely(map_length < submit_len + bvec->bv_len ||
6071 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6072 bvec->bv_offset) < bvec->bv_len)) {
6074 * inc the count before we submit the bio so
6075 * we know the end IO handler won't happen before
6076 * we inc the count. Otherwise, the dip might get freed
6077 * before we're done setting it up
6079 atomic_inc(&dip->pending_bios);
6080 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6081 file_offset, skip_sum,
6082 csums, async_submit);
6085 atomic_dec(&dip->pending_bios);
6089 /* Write's use the ordered csums */
6090 if (!write && !skip_sum)
6091 csums = csums + nr_pages;
6092 start_sector += submit_len >> 9;
6093 file_offset += submit_len;
6098 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6099 start_sector, GFP_NOFS);
6102 bio->bi_private = dip;
6103 bio->bi_end_io = btrfs_end_dio_bio;
6105 map_length = orig_bio->bi_size;
6106 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6107 &map_length, NULL, 0);
6113 submit_len += bvec->bv_len;
6120 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6121 csums, async_submit);
6129 * before atomic variable goto zero, we must
6130 * make sure dip->errors is perceived to be set.
6132 smp_mb__before_atomic_dec();
6133 if (atomic_dec_and_test(&dip->pending_bios))
6134 bio_io_error(dip->orig_bio);
6136 /* bio_end_io() will handle error, so we needn't return it */
6140 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6143 struct btrfs_root *root = BTRFS_I(inode)->root;
6144 struct btrfs_dio_private *dip;
6145 struct bio_vec *bvec = bio->bi_io_vec;
6147 int write = rw & REQ_WRITE;
6150 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6152 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6159 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6160 if (!write && !skip_sum) {
6161 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6169 dip->private = bio->bi_private;
6171 dip->logical_offset = file_offset;
6175 dip->bytes += bvec->bv_len;
6177 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6179 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6180 bio->bi_private = dip;
6182 dip->orig_bio = bio;
6183 atomic_set(&dip->pending_bios, 0);
6186 bio->bi_end_io = btrfs_endio_direct_write;
6188 bio->bi_end_io = btrfs_endio_direct_read;
6190 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6195 * If this is a write, we need to clean up the reserved space and kill
6196 * the ordered extent.
6199 struct btrfs_ordered_extent *ordered;
6200 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6201 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6202 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6203 btrfs_free_reserved_extent(root, ordered->start,
6205 btrfs_put_ordered_extent(ordered);
6206 btrfs_put_ordered_extent(ordered);
6208 bio_endio(bio, ret);
6211 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6212 const struct iovec *iov, loff_t offset,
6213 unsigned long nr_segs)
6219 unsigned blocksize_mask = root->sectorsize - 1;
6220 ssize_t retval = -EINVAL;
6221 loff_t end = offset;
6223 if (offset & blocksize_mask)
6226 /* Check the memory alignment. Blocks cannot straddle pages */
6227 for (seg = 0; seg < nr_segs; seg++) {
6228 addr = (unsigned long)iov[seg].iov_base;
6229 size = iov[seg].iov_len;
6231 if ((addr & blocksize_mask) || (size & blocksize_mask))
6234 /* If this is a write we don't need to check anymore */
6239 * Check to make sure we don't have duplicate iov_base's in this
6240 * iovec, if so return EINVAL, otherwise we'll get csum errors
6241 * when reading back.
6243 for (i = seg + 1; i < nr_segs; i++) {
6244 if (iov[seg].iov_base == iov[i].iov_base)
6252 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6253 const struct iovec *iov, loff_t offset,
6254 unsigned long nr_segs)
6256 struct file *file = iocb->ki_filp;
6257 struct inode *inode = file->f_mapping->host;
6258 struct btrfs_ordered_extent *ordered;
6259 struct extent_state *cached_state = NULL;
6260 u64 lockstart, lockend;
6262 int writing = rw & WRITE;
6264 size_t count = iov_length(iov, nr_segs);
6266 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6272 lockend = offset + count - 1;
6275 ret = btrfs_delalloc_reserve_space(inode, count);
6281 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6282 0, &cached_state, GFP_NOFS);
6284 * We're concerned with the entire range that we're going to be
6285 * doing DIO to, so we need to make sure theres no ordered
6286 * extents in this range.
6288 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6289 lockend - lockstart + 1);
6292 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6293 &cached_state, GFP_NOFS);
6294 btrfs_start_ordered_extent(inode, ordered, 1);
6295 btrfs_put_ordered_extent(ordered);
6300 * we don't use btrfs_set_extent_delalloc because we don't want
6301 * the dirty or uptodate bits
6304 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6305 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6306 EXTENT_DELALLOC, 0, NULL, &cached_state,
6309 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6310 lockend, EXTENT_LOCKED | write_bits,
6311 1, 0, &cached_state, GFP_NOFS);
6316 free_extent_state(cached_state);
6317 cached_state = NULL;
6319 ret = __blockdev_direct_IO(rw, iocb, inode,
6320 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6321 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6322 btrfs_submit_direct, 0);
6324 if (ret < 0 && ret != -EIOCBQUEUED) {
6325 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6326 offset + iov_length(iov, nr_segs) - 1,
6327 EXTENT_LOCKED | write_bits, 1, 0,
6328 &cached_state, GFP_NOFS);
6329 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6331 * We're falling back to buffered, unlock the section we didn't
6334 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6335 offset + iov_length(iov, nr_segs) - 1,
6336 EXTENT_LOCKED | write_bits, 1, 0,
6337 &cached_state, GFP_NOFS);
6340 free_extent_state(cached_state);
6344 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6345 __u64 start, __u64 len)
6347 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6350 int btrfs_readpage(struct file *file, struct page *page)
6352 struct extent_io_tree *tree;
6353 tree = &BTRFS_I(page->mapping->host)->io_tree;
6354 return extent_read_full_page(tree, page, btrfs_get_extent);
6357 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6359 struct extent_io_tree *tree;
6362 if (current->flags & PF_MEMALLOC) {
6363 redirty_page_for_writepage(wbc, page);
6367 tree = &BTRFS_I(page->mapping->host)->io_tree;
6368 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6371 int btrfs_writepages(struct address_space *mapping,
6372 struct writeback_control *wbc)
6374 struct extent_io_tree *tree;
6376 tree = &BTRFS_I(mapping->host)->io_tree;
6377 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6381 btrfs_readpages(struct file *file, struct address_space *mapping,
6382 struct list_head *pages, unsigned nr_pages)
6384 struct extent_io_tree *tree;
6385 tree = &BTRFS_I(mapping->host)->io_tree;
6386 return extent_readpages(tree, mapping, pages, nr_pages,
6389 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6391 struct extent_io_tree *tree;
6392 struct extent_map_tree *map;
6395 tree = &BTRFS_I(page->mapping->host)->io_tree;
6396 map = &BTRFS_I(page->mapping->host)->extent_tree;
6397 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6399 ClearPagePrivate(page);
6400 set_page_private(page, 0);
6401 page_cache_release(page);
6406 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6408 if (PageWriteback(page) || PageDirty(page))
6410 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6413 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6415 struct extent_io_tree *tree;
6416 struct btrfs_ordered_extent *ordered;
6417 struct extent_state *cached_state = NULL;
6418 u64 page_start = page_offset(page);
6419 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6423 * we have the page locked, so new writeback can't start,
6424 * and the dirty bit won't be cleared while we are here.
6426 * Wait for IO on this page so that we can safely clear
6427 * the PagePrivate2 bit and do ordered accounting
6429 wait_on_page_writeback(page);
6431 tree = &BTRFS_I(page->mapping->host)->io_tree;
6433 btrfs_releasepage(page, GFP_NOFS);
6436 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6438 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6442 * IO on this page will never be started, so we need
6443 * to account for any ordered extents now
6445 clear_extent_bit(tree, page_start, page_end,
6446 EXTENT_DIRTY | EXTENT_DELALLOC |
6447 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6448 &cached_state, GFP_NOFS);
6450 * whoever cleared the private bit is responsible
6451 * for the finish_ordered_io
6453 if (TestClearPagePrivate2(page)) {
6454 btrfs_finish_ordered_io(page->mapping->host,
6455 page_start, page_end);
6457 btrfs_put_ordered_extent(ordered);
6458 cached_state = NULL;
6459 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6462 clear_extent_bit(tree, page_start, page_end,
6463 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6464 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6465 __btrfs_releasepage(page, GFP_NOFS);
6467 ClearPageChecked(page);
6468 if (PagePrivate(page)) {
6469 ClearPagePrivate(page);
6470 set_page_private(page, 0);
6471 page_cache_release(page);
6476 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6477 * called from a page fault handler when a page is first dirtied. Hence we must
6478 * be careful to check for EOF conditions here. We set the page up correctly
6479 * for a written page which means we get ENOSPC checking when writing into
6480 * holes and correct delalloc and unwritten extent mapping on filesystems that
6481 * support these features.
6483 * We are not allowed to take the i_mutex here so we have to play games to
6484 * protect against truncate races as the page could now be beyond EOF. Because
6485 * vmtruncate() writes the inode size before removing pages, once we have the
6486 * page lock we can determine safely if the page is beyond EOF. If it is not
6487 * beyond EOF, then the page is guaranteed safe against truncation until we
6490 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6492 struct page *page = vmf->page;
6493 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6494 struct btrfs_root *root = BTRFS_I(inode)->root;
6495 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6496 struct btrfs_ordered_extent *ordered;
6497 struct extent_state *cached_state = NULL;
6499 unsigned long zero_start;
6505 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6509 else /* -ENOSPC, -EIO, etc */
6510 ret = VM_FAULT_SIGBUS;
6514 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6517 size = i_size_read(inode);
6518 page_start = page_offset(page);
6519 page_end = page_start + PAGE_CACHE_SIZE - 1;
6521 if ((page->mapping != inode->i_mapping) ||
6522 (page_start >= size)) {
6523 /* page got truncated out from underneath us */
6526 wait_on_page_writeback(page);
6528 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6530 set_page_extent_mapped(page);
6533 * we can't set the delalloc bits if there are pending ordered
6534 * extents. Drop our locks and wait for them to finish
6536 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6538 unlock_extent_cached(io_tree, page_start, page_end,
6539 &cached_state, GFP_NOFS);
6541 btrfs_start_ordered_extent(inode, ordered, 1);
6542 btrfs_put_ordered_extent(ordered);
6547 * XXX - page_mkwrite gets called every time the page is dirtied, even
6548 * if it was already dirty, so for space accounting reasons we need to
6549 * clear any delalloc bits for the range we are fixing to save. There
6550 * is probably a better way to do this, but for now keep consistent with
6551 * prepare_pages in the normal write path.
6553 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6554 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6555 0, 0, &cached_state, GFP_NOFS);
6557 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6560 unlock_extent_cached(io_tree, page_start, page_end,
6561 &cached_state, GFP_NOFS);
6562 ret = VM_FAULT_SIGBUS;
6567 /* page is wholly or partially inside EOF */
6568 if (page_start + PAGE_CACHE_SIZE > size)
6569 zero_start = size & ~PAGE_CACHE_MASK;
6571 zero_start = PAGE_CACHE_SIZE;
6573 if (zero_start != PAGE_CACHE_SIZE) {
6575 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6576 flush_dcache_page(page);
6579 ClearPageChecked(page);
6580 set_page_dirty(page);
6581 SetPageUptodate(page);
6583 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6584 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6586 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6590 return VM_FAULT_LOCKED;
6592 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6597 static int btrfs_truncate(struct inode *inode)
6599 struct btrfs_root *root = BTRFS_I(inode)->root;
6602 struct btrfs_trans_handle *trans;
6604 u64 mask = root->sectorsize - 1;
6606 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6610 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6611 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6613 trans = btrfs_start_transaction(root, 5);
6615 return PTR_ERR(trans);
6617 btrfs_set_trans_block_group(trans, inode);
6619 ret = btrfs_orphan_add(trans, inode);
6621 btrfs_end_transaction(trans, root);
6625 nr = trans->blocks_used;
6626 btrfs_end_transaction(trans, root);
6627 btrfs_btree_balance_dirty(root, nr);
6629 /* Now start a transaction for the truncate */
6630 trans = btrfs_start_transaction(root, 0);
6632 return PTR_ERR(trans);
6633 btrfs_set_trans_block_group(trans, inode);
6634 trans->block_rsv = root->orphan_block_rsv;
6637 * setattr is responsible for setting the ordered_data_close flag,
6638 * but that is only tested during the last file release. That
6639 * could happen well after the next commit, leaving a great big
6640 * window where new writes may get lost if someone chooses to write
6641 * to this file after truncating to zero
6643 * The inode doesn't have any dirty data here, and so if we commit
6644 * this is a noop. If someone immediately starts writing to the inode
6645 * it is very likely we'll catch some of their writes in this
6646 * transaction, and the commit will find this file on the ordered
6647 * data list with good things to send down.
6649 * This is a best effort solution, there is still a window where
6650 * using truncate to replace the contents of the file will
6651 * end up with a zero length file after a crash.
6653 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6654 btrfs_add_ordered_operation(trans, root, inode);
6658 trans = btrfs_start_transaction(root, 0);
6660 return PTR_ERR(trans);
6661 btrfs_set_trans_block_group(trans, inode);
6662 trans->block_rsv = root->orphan_block_rsv;
6665 ret = btrfs_block_rsv_check(trans, root,
6666 root->orphan_block_rsv, 0, 5);
6667 if (ret == -EAGAIN) {
6668 ret = btrfs_commit_transaction(trans, root);
6678 ret = btrfs_truncate_inode_items(trans, root, inode,
6680 BTRFS_EXTENT_DATA_KEY);
6681 if (ret != -EAGAIN) {
6686 ret = btrfs_update_inode(trans, root, inode);
6692 nr = trans->blocks_used;
6693 btrfs_end_transaction(trans, root);
6695 btrfs_btree_balance_dirty(root, nr);
6698 if (ret == 0 && inode->i_nlink > 0) {
6699 ret = btrfs_orphan_del(trans, inode);
6702 } else if (ret && inode->i_nlink > 0) {
6704 * Failed to do the truncate, remove us from the in memory
6707 ret = btrfs_orphan_del(NULL, inode);
6710 ret = btrfs_update_inode(trans, root, inode);
6714 nr = trans->blocks_used;
6715 ret = btrfs_end_transaction_throttle(trans, root);
6718 btrfs_btree_balance_dirty(root, nr);
6724 * create a new subvolume directory/inode (helper for the ioctl).
6726 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6727 struct btrfs_root *new_root,
6728 u64 new_dirid, u64 alloc_hint)
6730 struct inode *inode;
6734 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6735 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6737 return PTR_ERR(inode);
6738 inode->i_op = &btrfs_dir_inode_operations;
6739 inode->i_fop = &btrfs_dir_file_operations;
6742 btrfs_i_size_write(inode, 0);
6744 err = btrfs_update_inode(trans, new_root, inode);
6751 /* helper function for file defrag and space balancing. This
6752 * forces readahead on a given range of bytes in an inode
6754 unsigned long btrfs_force_ra(struct address_space *mapping,
6755 struct file_ra_state *ra, struct file *file,
6756 pgoff_t offset, pgoff_t last_index)
6758 pgoff_t req_size = last_index - offset + 1;
6760 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6761 return offset + req_size;
6764 struct inode *btrfs_alloc_inode(struct super_block *sb)
6766 struct btrfs_inode *ei;
6767 struct inode *inode;
6769 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6774 ei->space_info = NULL;
6778 ei->last_sub_trans = 0;
6779 ei->logged_trans = 0;
6780 ei->delalloc_bytes = 0;
6781 ei->reserved_bytes = 0;
6782 ei->disk_i_size = 0;
6784 ei->index_cnt = (u64)-1;
6785 ei->last_unlink_trans = 0;
6787 atomic_set(&ei->outstanding_extents, 0);
6788 atomic_set(&ei->reserved_extents, 0);
6790 ei->ordered_data_close = 0;
6791 ei->orphan_meta_reserved = 0;
6792 ei->dummy_inode = 0;
6793 ei->force_compress = BTRFS_COMPRESS_NONE;
6795 inode = &ei->vfs_inode;
6796 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6797 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6798 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6799 mutex_init(&ei->log_mutex);
6800 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6801 INIT_LIST_HEAD(&ei->i_orphan);
6802 INIT_LIST_HEAD(&ei->delalloc_inodes);
6803 INIT_LIST_HEAD(&ei->ordered_operations);
6804 RB_CLEAR_NODE(&ei->rb_node);
6809 static void btrfs_i_callback(struct rcu_head *head)
6811 struct inode *inode = container_of(head, struct inode, i_rcu);
6812 INIT_LIST_HEAD(&inode->i_dentry);
6813 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6816 void btrfs_destroy_inode(struct inode *inode)
6818 struct btrfs_ordered_extent *ordered;
6819 struct btrfs_root *root = BTRFS_I(inode)->root;
6821 WARN_ON(!list_empty(&inode->i_dentry));
6822 WARN_ON(inode->i_data.nrpages);
6823 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6824 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6827 * This can happen where we create an inode, but somebody else also
6828 * created the same inode and we need to destroy the one we already
6835 * Make sure we're properly removed from the ordered operation
6839 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6840 spin_lock(&root->fs_info->ordered_extent_lock);
6841 list_del_init(&BTRFS_I(inode)->ordered_operations);
6842 spin_unlock(&root->fs_info->ordered_extent_lock);
6845 if (root == root->fs_info->tree_root) {
6846 struct btrfs_block_group_cache *block_group;
6848 block_group = btrfs_lookup_block_group(root->fs_info,
6849 BTRFS_I(inode)->block_group);
6850 if (block_group && block_group->inode == inode) {
6851 spin_lock(&block_group->lock);
6852 block_group->inode = NULL;
6853 spin_unlock(&block_group->lock);
6854 btrfs_put_block_group(block_group);
6855 } else if (block_group) {
6856 btrfs_put_block_group(block_group);
6860 spin_lock(&root->orphan_lock);
6861 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6862 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6864 list_del_init(&BTRFS_I(inode)->i_orphan);
6866 spin_unlock(&root->orphan_lock);
6869 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6873 printk(KERN_ERR "btrfs found ordered "
6874 "extent %llu %llu on inode cleanup\n",
6875 (unsigned long long)ordered->file_offset,
6876 (unsigned long long)ordered->len);
6877 btrfs_remove_ordered_extent(inode, ordered);
6878 btrfs_put_ordered_extent(ordered);
6879 btrfs_put_ordered_extent(ordered);
6882 inode_tree_del(inode);
6883 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6885 call_rcu(&inode->i_rcu, btrfs_i_callback);
6888 int btrfs_drop_inode(struct inode *inode)
6890 struct btrfs_root *root = BTRFS_I(inode)->root;
6892 if (btrfs_root_refs(&root->root_item) == 0 &&
6893 root != root->fs_info->tree_root)
6896 return generic_drop_inode(inode);
6899 static void init_once(void *foo)
6901 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6903 inode_init_once(&ei->vfs_inode);
6906 void btrfs_destroy_cachep(void)
6908 if (btrfs_inode_cachep)
6909 kmem_cache_destroy(btrfs_inode_cachep);
6910 if (btrfs_trans_handle_cachep)
6911 kmem_cache_destroy(btrfs_trans_handle_cachep);
6912 if (btrfs_transaction_cachep)
6913 kmem_cache_destroy(btrfs_transaction_cachep);
6914 if (btrfs_path_cachep)
6915 kmem_cache_destroy(btrfs_path_cachep);
6916 if (btrfs_free_space_cachep)
6917 kmem_cache_destroy(btrfs_free_space_cachep);
6920 int btrfs_init_cachep(void)
6922 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6923 sizeof(struct btrfs_inode), 0,
6924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6925 if (!btrfs_inode_cachep)
6928 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6929 sizeof(struct btrfs_trans_handle), 0,
6930 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6931 if (!btrfs_trans_handle_cachep)
6934 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6935 sizeof(struct btrfs_transaction), 0,
6936 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6937 if (!btrfs_transaction_cachep)
6940 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6941 sizeof(struct btrfs_path), 0,
6942 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6943 if (!btrfs_path_cachep)
6946 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6947 sizeof(struct btrfs_free_space), 0,
6948 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6949 if (!btrfs_free_space_cachep)
6954 btrfs_destroy_cachep();
6958 static int btrfs_getattr(struct vfsmount *mnt,
6959 struct dentry *dentry, struct kstat *stat)
6961 struct inode *inode = dentry->d_inode;
6962 generic_fillattr(inode, stat);
6963 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6964 stat->blksize = PAGE_CACHE_SIZE;
6965 stat->blocks = (inode_get_bytes(inode) +
6966 BTRFS_I(inode)->delalloc_bytes) >> 9;
6971 * If a file is moved, it will inherit the cow and compression flags of the new
6974 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6976 struct btrfs_inode *b_dir = BTRFS_I(dir);
6977 struct btrfs_inode *b_inode = BTRFS_I(inode);
6979 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6980 b_inode->flags |= BTRFS_INODE_NODATACOW;
6982 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6984 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6985 b_inode->flags |= BTRFS_INODE_COMPRESS;
6987 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6990 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6991 struct inode *new_dir, struct dentry *new_dentry)
6993 struct btrfs_trans_handle *trans;
6994 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6995 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6996 struct inode *new_inode = new_dentry->d_inode;
6997 struct inode *old_inode = old_dentry->d_inode;
6998 struct timespec ctime = CURRENT_TIME;
7003 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7006 /* we only allow rename subvolume link between subvolumes */
7007 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7010 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7011 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
7014 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7015 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7018 * we're using rename to replace one file with another.
7019 * and the replacement file is large. Start IO on it now so
7020 * we don't add too much work to the end of the transaction
7022 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7023 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7024 filemap_flush(old_inode->i_mapping);
7026 /* close the racy window with snapshot create/destroy ioctl */
7027 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7028 down_read(&root->fs_info->subvol_sem);
7030 * We want to reserve the absolute worst case amount of items. So if
7031 * both inodes are subvols and we need to unlink them then that would
7032 * require 4 item modifications, but if they are both normal inodes it
7033 * would require 5 item modifications, so we'll assume their normal
7034 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7035 * should cover the worst case number of items we'll modify.
7037 trans = btrfs_start_transaction(root, 20);
7038 if (IS_ERR(trans)) {
7039 ret = PTR_ERR(trans);
7043 btrfs_set_trans_block_group(trans, new_dir);
7046 btrfs_record_root_in_trans(trans, dest);
7048 ret = btrfs_set_inode_index(new_dir, &index);
7052 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7053 /* force full log commit if subvolume involved. */
7054 root->fs_info->last_trans_log_full_commit = trans->transid;
7056 ret = btrfs_insert_inode_ref(trans, dest,
7057 new_dentry->d_name.name,
7058 new_dentry->d_name.len,
7060 new_dir->i_ino, index);
7064 * this is an ugly little race, but the rename is required
7065 * to make sure that if we crash, the inode is either at the
7066 * old name or the new one. pinning the log transaction lets
7067 * us make sure we don't allow a log commit to come in after
7068 * we unlink the name but before we add the new name back in.
7070 btrfs_pin_log_trans(root);
7073 * make sure the inode gets flushed if it is replacing
7076 if (new_inode && new_inode->i_size &&
7077 old_inode && S_ISREG(old_inode->i_mode)) {
7078 btrfs_add_ordered_operation(trans, root, old_inode);
7081 old_dir->i_ctime = old_dir->i_mtime = ctime;
7082 new_dir->i_ctime = new_dir->i_mtime = ctime;
7083 old_inode->i_ctime = ctime;
7085 if (old_dentry->d_parent != new_dentry->d_parent)
7086 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7088 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7089 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7090 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7091 old_dentry->d_name.name,
7092 old_dentry->d_name.len);
7094 ret = __btrfs_unlink_inode(trans, root, old_dir,
7095 old_dentry->d_inode,
7096 old_dentry->d_name.name,
7097 old_dentry->d_name.len);
7099 ret = btrfs_update_inode(trans, root, old_inode);
7104 new_inode->i_ctime = CURRENT_TIME;
7105 if (unlikely(new_inode->i_ino ==
7106 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7107 root_objectid = BTRFS_I(new_inode)->location.objectid;
7108 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7110 new_dentry->d_name.name,
7111 new_dentry->d_name.len);
7112 BUG_ON(new_inode->i_nlink == 0);
7114 ret = btrfs_unlink_inode(trans, dest, new_dir,
7115 new_dentry->d_inode,
7116 new_dentry->d_name.name,
7117 new_dentry->d_name.len);
7120 if (new_inode->i_nlink == 0) {
7121 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7126 fixup_inode_flags(new_dir, old_inode);
7128 ret = btrfs_add_link(trans, new_dir, old_inode,
7129 new_dentry->d_name.name,
7130 new_dentry->d_name.len, 0, index);
7133 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7134 struct dentry *parent = dget_parent(new_dentry);
7135 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7137 btrfs_end_log_trans(root);
7140 btrfs_end_transaction_throttle(trans, root);
7142 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7143 up_read(&root->fs_info->subvol_sem);
7149 * some fairly slow code that needs optimization. This walks the list
7150 * of all the inodes with pending delalloc and forces them to disk.
7152 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7154 struct list_head *head = &root->fs_info->delalloc_inodes;
7155 struct btrfs_inode *binode;
7156 struct inode *inode;
7158 if (root->fs_info->sb->s_flags & MS_RDONLY)
7161 spin_lock(&root->fs_info->delalloc_lock);
7162 while (!list_empty(head)) {
7163 binode = list_entry(head->next, struct btrfs_inode,
7165 inode = igrab(&binode->vfs_inode);
7167 list_del_init(&binode->delalloc_inodes);
7168 spin_unlock(&root->fs_info->delalloc_lock);
7170 filemap_flush(inode->i_mapping);
7172 btrfs_add_delayed_iput(inode);
7177 spin_lock(&root->fs_info->delalloc_lock);
7179 spin_unlock(&root->fs_info->delalloc_lock);
7181 /* the filemap_flush will queue IO into the worker threads, but
7182 * we have to make sure the IO is actually started and that
7183 * ordered extents get created before we return
7185 atomic_inc(&root->fs_info->async_submit_draining);
7186 while (atomic_read(&root->fs_info->nr_async_submits) ||
7187 atomic_read(&root->fs_info->async_delalloc_pages)) {
7188 wait_event(root->fs_info->async_submit_wait,
7189 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7190 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7192 atomic_dec(&root->fs_info->async_submit_draining);
7196 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7199 struct btrfs_inode *binode;
7200 struct inode *inode = NULL;
7202 spin_lock(&root->fs_info->delalloc_lock);
7203 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7204 binode = list_entry(root->fs_info->delalloc_inodes.next,
7205 struct btrfs_inode, delalloc_inodes);
7206 inode = igrab(&binode->vfs_inode);
7208 list_move_tail(&binode->delalloc_inodes,
7209 &root->fs_info->delalloc_inodes);
7213 list_del_init(&binode->delalloc_inodes);
7214 cond_resched_lock(&root->fs_info->delalloc_lock);
7216 spin_unlock(&root->fs_info->delalloc_lock);
7220 filemap_write_and_wait(inode->i_mapping);
7222 * We have to do this because compression doesn't
7223 * actually set PG_writeback until it submits the pages
7224 * for IO, which happens in an async thread, so we could
7225 * race and not actually wait for any writeback pages
7226 * because they've not been submitted yet. Technically
7227 * this could still be the case for the ordered stuff
7228 * since the async thread may not have started to do its
7229 * work yet. If this becomes the case then we need to
7230 * figure out a way to make sure that in writepage we
7231 * wait for any async pages to be submitted before
7232 * returning so that fdatawait does what its supposed to
7235 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7237 filemap_flush(inode->i_mapping);
7240 btrfs_add_delayed_iput(inode);
7248 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7249 const char *symname)
7251 struct btrfs_trans_handle *trans;
7252 struct btrfs_root *root = BTRFS_I(dir)->root;
7253 struct btrfs_path *path;
7254 struct btrfs_key key;
7255 struct inode *inode = NULL;
7263 struct btrfs_file_extent_item *ei;
7264 struct extent_buffer *leaf;
7265 unsigned long nr = 0;
7267 name_len = strlen(symname) + 1;
7268 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7269 return -ENAMETOOLONG;
7272 * 2 items for inode item and ref
7273 * 2 items for dir items
7274 * 1 item for xattr if selinux is on
7276 trans = btrfs_start_transaction(root, 5);
7278 return PTR_ERR(trans);
7280 btrfs_set_trans_block_group(trans, dir);
7282 err = btrfs_find_free_ino(root, &objectid);
7286 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7287 dentry->d_name.len, dir->i_ino, objectid,
7288 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7290 err = PTR_ERR(inode);
7294 err = btrfs_init_inode_security(trans, inode, dir);
7300 btrfs_set_trans_block_group(trans, inode);
7301 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7305 inode->i_mapping->a_ops = &btrfs_aops;
7306 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7307 inode->i_fop = &btrfs_file_operations;
7308 inode->i_op = &btrfs_file_inode_operations;
7309 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7311 btrfs_update_inode_block_group(trans, inode);
7312 btrfs_update_inode_block_group(trans, dir);
7316 path = btrfs_alloc_path();
7318 key.objectid = inode->i_ino;
7320 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7321 datasize = btrfs_file_extent_calc_inline_size(name_len);
7322 err = btrfs_insert_empty_item(trans, root, path, &key,
7328 leaf = path->nodes[0];
7329 ei = btrfs_item_ptr(leaf, path->slots[0],
7330 struct btrfs_file_extent_item);
7331 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7332 btrfs_set_file_extent_type(leaf, ei,
7333 BTRFS_FILE_EXTENT_INLINE);
7334 btrfs_set_file_extent_encryption(leaf, ei, 0);
7335 btrfs_set_file_extent_compression(leaf, ei, 0);
7336 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7337 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7339 ptr = btrfs_file_extent_inline_start(ei);
7340 write_extent_buffer(leaf, symname, ptr, name_len);
7341 btrfs_mark_buffer_dirty(leaf);
7342 btrfs_free_path(path);
7344 inode->i_op = &btrfs_symlink_inode_operations;
7345 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7346 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7347 inode_set_bytes(inode, name_len);
7348 btrfs_i_size_write(inode, name_len - 1);
7349 err = btrfs_update_inode(trans, root, inode);
7354 nr = trans->blocks_used;
7355 btrfs_end_transaction_throttle(trans, root);
7357 inode_dec_link_count(inode);
7360 btrfs_btree_balance_dirty(root, nr);
7364 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7365 u64 start, u64 num_bytes, u64 min_size,
7366 loff_t actual_len, u64 *alloc_hint,
7367 struct btrfs_trans_handle *trans)
7369 struct btrfs_root *root = BTRFS_I(inode)->root;
7370 struct btrfs_key ins;
7371 u64 cur_offset = start;
7374 bool own_trans = true;
7378 while (num_bytes > 0) {
7380 trans = btrfs_start_transaction(root, 3);
7381 if (IS_ERR(trans)) {
7382 ret = PTR_ERR(trans);
7387 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7388 0, *alloc_hint, (u64)-1, &ins, 1);
7391 btrfs_end_transaction(trans, root);
7395 ret = insert_reserved_file_extent(trans, inode,
7396 cur_offset, ins.objectid,
7397 ins.offset, ins.offset,
7398 ins.offset, 0, 0, 0,
7399 BTRFS_FILE_EXTENT_PREALLOC);
7401 btrfs_drop_extent_cache(inode, cur_offset,
7402 cur_offset + ins.offset -1, 0);
7404 num_bytes -= ins.offset;
7405 cur_offset += ins.offset;
7406 *alloc_hint = ins.objectid + ins.offset;
7408 inode->i_ctime = CURRENT_TIME;
7409 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7410 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7411 (actual_len > inode->i_size) &&
7412 (cur_offset > inode->i_size)) {
7413 if (cur_offset > actual_len)
7414 i_size = actual_len;
7416 i_size = cur_offset;
7417 i_size_write(inode, i_size);
7418 btrfs_ordered_update_i_size(inode, i_size, NULL);
7421 ret = btrfs_update_inode(trans, root, inode);
7425 btrfs_end_transaction(trans, root);
7430 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7431 u64 start, u64 num_bytes, u64 min_size,
7432 loff_t actual_len, u64 *alloc_hint)
7434 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7435 min_size, actual_len, alloc_hint,
7439 int btrfs_prealloc_file_range_trans(struct inode *inode,
7440 struct btrfs_trans_handle *trans, int mode,
7441 u64 start, u64 num_bytes, u64 min_size,
7442 loff_t actual_len, u64 *alloc_hint)
7444 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7445 min_size, actual_len, alloc_hint, trans);
7448 static int btrfs_set_page_dirty(struct page *page)
7450 return __set_page_dirty_nobuffers(page);
7453 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7455 struct btrfs_root *root = BTRFS_I(inode)->root;
7457 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7459 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7461 return generic_permission(inode, mask, flags, btrfs_check_acl);
7464 static const struct inode_operations btrfs_dir_inode_operations = {
7465 .getattr = btrfs_getattr,
7466 .lookup = btrfs_lookup,
7467 .create = btrfs_create,
7468 .unlink = btrfs_unlink,
7470 .mkdir = btrfs_mkdir,
7471 .rmdir = btrfs_rmdir,
7472 .rename = btrfs_rename,
7473 .symlink = btrfs_symlink,
7474 .setattr = btrfs_setattr,
7475 .mknod = btrfs_mknod,
7476 .setxattr = btrfs_setxattr,
7477 .getxattr = btrfs_getxattr,
7478 .listxattr = btrfs_listxattr,
7479 .removexattr = btrfs_removexattr,
7480 .permission = btrfs_permission,
7482 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7483 .lookup = btrfs_lookup,
7484 .permission = btrfs_permission,
7487 static const struct file_operations btrfs_dir_file_operations = {
7488 .llseek = generic_file_llseek,
7489 .read = generic_read_dir,
7490 .readdir = btrfs_real_readdir,
7491 .unlocked_ioctl = btrfs_ioctl,
7492 #ifdef CONFIG_COMPAT
7493 .compat_ioctl = btrfs_ioctl,
7495 .release = btrfs_release_file,
7496 .fsync = btrfs_sync_file,
7499 static struct extent_io_ops btrfs_extent_io_ops = {
7500 .fill_delalloc = run_delalloc_range,
7501 .submit_bio_hook = btrfs_submit_bio_hook,
7502 .merge_bio_hook = btrfs_merge_bio_hook,
7503 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7504 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7505 .writepage_start_hook = btrfs_writepage_start_hook,
7506 .readpage_io_failed_hook = btrfs_io_failed_hook,
7507 .set_bit_hook = btrfs_set_bit_hook,
7508 .clear_bit_hook = btrfs_clear_bit_hook,
7509 .merge_extent_hook = btrfs_merge_extent_hook,
7510 .split_extent_hook = btrfs_split_extent_hook,
7514 * btrfs doesn't support the bmap operation because swapfiles
7515 * use bmap to make a mapping of extents in the file. They assume
7516 * these extents won't change over the life of the file and they
7517 * use the bmap result to do IO directly to the drive.
7519 * the btrfs bmap call would return logical addresses that aren't
7520 * suitable for IO and they also will change frequently as COW
7521 * operations happen. So, swapfile + btrfs == corruption.
7523 * For now we're avoiding this by dropping bmap.
7525 static const struct address_space_operations btrfs_aops = {
7526 .readpage = btrfs_readpage,
7527 .writepage = btrfs_writepage,
7528 .writepages = btrfs_writepages,
7529 .readpages = btrfs_readpages,
7530 .sync_page = block_sync_page,
7531 .direct_IO = btrfs_direct_IO,
7532 .invalidatepage = btrfs_invalidatepage,
7533 .releasepage = btrfs_releasepage,
7534 .set_page_dirty = btrfs_set_page_dirty,
7535 .error_remove_page = generic_error_remove_page,
7538 static const struct address_space_operations btrfs_symlink_aops = {
7539 .readpage = btrfs_readpage,
7540 .writepage = btrfs_writepage,
7541 .invalidatepage = btrfs_invalidatepage,
7542 .releasepage = btrfs_releasepage,
7545 static const struct inode_operations btrfs_file_inode_operations = {
7546 .getattr = btrfs_getattr,
7547 .setattr = btrfs_setattr,
7548 .setxattr = btrfs_setxattr,
7549 .getxattr = btrfs_getxattr,
7550 .listxattr = btrfs_listxattr,
7551 .removexattr = btrfs_removexattr,
7552 .permission = btrfs_permission,
7553 .fiemap = btrfs_fiemap,
7555 static const struct inode_operations btrfs_special_inode_operations = {
7556 .getattr = btrfs_getattr,
7557 .setattr = btrfs_setattr,
7558 .permission = btrfs_permission,
7559 .setxattr = btrfs_setxattr,
7560 .getxattr = btrfs_getxattr,
7561 .listxattr = btrfs_listxattr,
7562 .removexattr = btrfs_removexattr,
7564 static const struct inode_operations btrfs_symlink_inode_operations = {
7565 .readlink = generic_readlink,
7566 .follow_link = page_follow_link_light,
7567 .put_link = page_put_link,
7568 .getattr = btrfs_getattr,
7569 .permission = btrfs_permission,
7570 .setxattr = btrfs_setxattr,
7571 .getxattr = btrfs_getxattr,
7572 .listxattr = btrfs_listxattr,
7573 .removexattr = btrfs_removexattr,
7576 const struct dentry_operations btrfs_dentry_operations = {
7577 .d_delete = btrfs_dentry_delete,