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"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96 struct inode *inode, struct inode *dir,
97 const struct qstr *qstr)
101 err = btrfs_init_acl(trans, inode, dir);
103 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
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,
115 struct page **compressed_pages)
117 struct btrfs_key key;
118 struct btrfs_path *path;
119 struct extent_buffer *leaf;
120 struct page *page = NULL;
123 struct btrfs_file_extent_item *ei;
126 size_t cur_size = size;
128 unsigned long offset;
129 int compress_type = BTRFS_COMPRESS_NONE;
131 if (compressed_size && compressed_pages) {
132 compress_type = root->fs_info->compress_type;
133 cur_size = compressed_size;
136 path = btrfs_alloc_path();
140 path->leave_spinning = 1;
141 btrfs_set_trans_block_group(trans, inode);
143 key.objectid = inode->i_ino;
145 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
146 datasize = btrfs_file_extent_calc_inline_size(cur_size);
148 inode_add_bytes(inode, size);
149 ret = btrfs_insert_empty_item(trans, root, path, &key,
156 leaf = path->nodes[0];
157 ei = btrfs_item_ptr(leaf, path->slots[0],
158 struct btrfs_file_extent_item);
159 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
160 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
161 btrfs_set_file_extent_encryption(leaf, ei, 0);
162 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
163 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
164 ptr = btrfs_file_extent_inline_start(ei);
166 if (compress_type != BTRFS_COMPRESS_NONE) {
169 while (compressed_size > 0) {
170 cpage = compressed_pages[i];
171 cur_size = min_t(unsigned long, compressed_size,
174 kaddr = kmap_atomic(cpage, KM_USER0);
175 write_extent_buffer(leaf, kaddr, ptr, cur_size);
176 kunmap_atomic(kaddr, KM_USER0);
180 compressed_size -= cur_size;
182 btrfs_set_file_extent_compression(leaf, ei,
185 page = find_get_page(inode->i_mapping,
186 start >> PAGE_CACHE_SHIFT);
187 btrfs_set_file_extent_compression(leaf, ei, 0);
188 kaddr = kmap_atomic(page, KM_USER0);
189 offset = start & (PAGE_CACHE_SIZE - 1);
190 write_extent_buffer(leaf, kaddr + offset, ptr, size);
191 kunmap_atomic(kaddr, KM_USER0);
192 page_cache_release(page);
194 btrfs_mark_buffer_dirty(leaf);
195 btrfs_free_path(path);
198 * we're an inline extent, so nobody can
199 * extend the file past i_size without locking
200 * a page we already have locked.
202 * We must do any isize and inode updates
203 * before we unlock the pages. Otherwise we
204 * could end up racing with unlink.
206 BTRFS_I(inode)->disk_i_size = inode->i_size;
207 btrfs_update_inode(trans, root, inode);
211 btrfs_free_path(path);
217 * conditionally insert an inline extent into the file. This
218 * does the checks required to make sure the data is small enough
219 * to fit as an inline extent.
221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
222 struct btrfs_root *root,
223 struct inode *inode, u64 start, u64 end,
224 size_t compressed_size,
225 struct page **compressed_pages)
227 u64 isize = i_size_read(inode);
228 u64 actual_end = min(end + 1, isize);
229 u64 inline_len = actual_end - start;
230 u64 aligned_end = (end + root->sectorsize - 1) &
231 ~((u64)root->sectorsize - 1);
233 u64 data_len = inline_len;
237 data_len = compressed_size;
240 actual_end >= PAGE_CACHE_SIZE ||
241 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
243 (actual_end & (root->sectorsize - 1)) == 0) ||
245 data_len > root->fs_info->max_inline) {
249 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
253 if (isize > actual_end)
254 inline_len = min_t(u64, isize, actual_end);
255 ret = insert_inline_extent(trans, root, inode, start,
256 inline_len, compressed_size,
259 btrfs_delalloc_release_metadata(inode, end + 1 - start);
260 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
264 struct async_extent {
269 unsigned long nr_pages;
271 struct list_head list;
276 struct btrfs_root *root;
277 struct page *locked_page;
280 struct list_head extents;
281 struct btrfs_work work;
284 static noinline int add_async_extent(struct async_cow *cow,
285 u64 start, u64 ram_size,
288 unsigned long nr_pages,
291 struct async_extent *async_extent;
293 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
294 BUG_ON(!async_extent);
295 async_extent->start = start;
296 async_extent->ram_size = ram_size;
297 async_extent->compressed_size = compressed_size;
298 async_extent->pages = pages;
299 async_extent->nr_pages = nr_pages;
300 async_extent->compress_type = compress_type;
301 list_add_tail(&async_extent->list, &cow->extents);
306 * we create compressed extents in two phases. The first
307 * phase compresses a range of pages that have already been
308 * locked (both pages and state bits are locked).
310 * This is done inside an ordered work queue, and the compression
311 * is spread across many cpus. The actual IO submission is step
312 * two, and the ordered work queue takes care of making sure that
313 * happens in the same order things were put onto the queue by
314 * writepages and friends.
316 * If this code finds it can't get good compression, it puts an
317 * entry onto the work queue to write the uncompressed bytes. This
318 * makes sure that both compressed inodes and uncompressed inodes
319 * are written in the same order that pdflush sent them down.
321 static noinline int compress_file_range(struct inode *inode,
322 struct page *locked_page,
324 struct async_cow *async_cow,
327 struct btrfs_root *root = BTRFS_I(inode)->root;
328 struct btrfs_trans_handle *trans;
330 u64 blocksize = root->sectorsize;
332 u64 isize = i_size_read(inode);
334 struct page **pages = NULL;
335 unsigned long nr_pages;
336 unsigned long nr_pages_ret = 0;
337 unsigned long total_compressed = 0;
338 unsigned long total_in = 0;
339 unsigned long max_compressed = 128 * 1024;
340 unsigned long max_uncompressed = 128 * 1024;
343 int compress_type = root->fs_info->compress_type;
345 actual_end = min_t(u64, isize, end + 1);
348 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
349 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
352 * we don't want to send crud past the end of i_size through
353 * compression, that's just a waste of CPU time. So, if the
354 * end of the file is before the start of our current
355 * requested range of bytes, we bail out to the uncompressed
356 * cleanup code that can deal with all of this.
358 * It isn't really the fastest way to fix things, but this is a
359 * very uncommon corner.
361 if (actual_end <= start)
362 goto cleanup_and_bail_uncompressed;
364 total_compressed = actual_end - start;
366 /* we want to make sure that amount of ram required to uncompress
367 * an extent is reasonable, so we limit the total size in ram
368 * of a compressed extent to 128k. This is a crucial number
369 * because it also controls how easily we can spread reads across
370 * cpus for decompression.
372 * We also want to make sure the amount of IO required to do
373 * a random read is reasonably small, so we limit the size of
374 * a compressed extent to 128k.
376 total_compressed = min(total_compressed, max_uncompressed);
377 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
378 num_bytes = max(blocksize, num_bytes);
383 * we do compression for mount -o compress and when the
384 * inode has not been flagged as nocompress. This flag can
385 * change at any time if we discover bad compression ratios.
387 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
388 (btrfs_test_opt(root, COMPRESS) ||
389 (BTRFS_I(inode)->force_compress) ||
390 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
392 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
395 if (BTRFS_I(inode)->force_compress)
396 compress_type = BTRFS_I(inode)->force_compress;
398 ret = btrfs_compress_pages(compress_type,
399 inode->i_mapping, start,
400 total_compressed, pages,
401 nr_pages, &nr_pages_ret,
407 unsigned long offset = total_compressed &
408 (PAGE_CACHE_SIZE - 1);
409 struct page *page = pages[nr_pages_ret - 1];
412 /* zero the tail end of the last page, we might be
413 * sending it down to disk
416 kaddr = kmap_atomic(page, KM_USER0);
417 memset(kaddr + offset, 0,
418 PAGE_CACHE_SIZE - offset);
419 kunmap_atomic(kaddr, KM_USER0);
425 trans = btrfs_join_transaction(root, 1);
426 BUG_ON(IS_ERR(trans));
427 btrfs_set_trans_block_group(trans, inode);
428 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
430 /* lets try to make an inline extent */
431 if (ret || total_in < (actual_end - start)) {
432 /* we didn't compress the entire range, try
433 * to make an uncompressed inline extent.
435 ret = cow_file_range_inline(trans, root, inode,
436 start, end, 0, NULL);
438 /* try making a compressed inline extent */
439 ret = cow_file_range_inline(trans, root, inode,
441 total_compressed, pages);
445 * inline extent creation worked, we don't need
446 * to create any more async work items. Unlock
447 * and free up our temp pages.
449 extent_clear_unlock_delalloc(inode,
450 &BTRFS_I(inode)->io_tree,
452 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
453 EXTENT_CLEAR_DELALLOC |
454 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
456 btrfs_end_transaction(trans, root);
459 btrfs_end_transaction(trans, root);
464 * we aren't doing an inline extent round the compressed size
465 * up to a block size boundary so the allocator does sane
468 total_compressed = (total_compressed + blocksize - 1) &
472 * one last check to make sure the compression is really a
473 * win, compare the page count read with the blocks on disk
475 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
476 ~(PAGE_CACHE_SIZE - 1);
477 if (total_compressed >= total_in) {
480 num_bytes = total_in;
483 if (!will_compress && pages) {
485 * the compression code ran but failed to make things smaller,
486 * free any pages it allocated and our page pointer array
488 for (i = 0; i < nr_pages_ret; i++) {
489 WARN_ON(pages[i]->mapping);
490 page_cache_release(pages[i]);
494 total_compressed = 0;
497 /* flag the file so we don't compress in the future */
498 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
499 !(BTRFS_I(inode)->force_compress)) {
500 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
506 /* the async work queues will take care of doing actual
507 * allocation on disk for these compressed pages,
508 * and will submit them to the elevator.
510 add_async_extent(async_cow, start, num_bytes,
511 total_compressed, pages, nr_pages_ret,
514 if (start + num_bytes < end) {
521 cleanup_and_bail_uncompressed:
523 * No compression, but we still need to write the pages in
524 * the file we've been given so far. redirty the locked
525 * page if it corresponds to our extent and set things up
526 * for the async work queue to run cow_file_range to do
527 * the normal delalloc dance
529 if (page_offset(locked_page) >= start &&
530 page_offset(locked_page) <= end) {
531 __set_page_dirty_nobuffers(locked_page);
532 /* unlocked later on in the async handlers */
534 add_async_extent(async_cow, start, end - start + 1,
535 0, NULL, 0, BTRFS_COMPRESS_NONE);
543 for (i = 0; i < nr_pages_ret; i++) {
544 WARN_ON(pages[i]->mapping);
545 page_cache_release(pages[i]);
553 * phase two of compressed writeback. This is the ordered portion
554 * of the code, which only gets called in the order the work was
555 * queued. We walk all the async extents created by compress_file_range
556 * and send them down to the disk.
558 static noinline int submit_compressed_extents(struct inode *inode,
559 struct async_cow *async_cow)
561 struct async_extent *async_extent;
563 struct btrfs_trans_handle *trans;
564 struct btrfs_key ins;
565 struct extent_map *em;
566 struct btrfs_root *root = BTRFS_I(inode)->root;
567 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
568 struct extent_io_tree *io_tree;
571 if (list_empty(&async_cow->extents))
575 while (!list_empty(&async_cow->extents)) {
576 async_extent = list_entry(async_cow->extents.next,
577 struct async_extent, list);
578 list_del(&async_extent->list);
580 io_tree = &BTRFS_I(inode)->io_tree;
583 /* did the compression code fall back to uncompressed IO? */
584 if (!async_extent->pages) {
585 int page_started = 0;
586 unsigned long nr_written = 0;
588 lock_extent(io_tree, async_extent->start,
589 async_extent->start +
590 async_extent->ram_size - 1, GFP_NOFS);
592 /* allocate blocks */
593 ret = cow_file_range(inode, async_cow->locked_page,
595 async_extent->start +
596 async_extent->ram_size - 1,
597 &page_started, &nr_written, 0);
600 * if page_started, cow_file_range inserted an
601 * inline extent and took care of all the unlocking
602 * and IO for us. Otherwise, we need to submit
603 * all those pages down to the drive.
605 if (!page_started && !ret)
606 extent_write_locked_range(io_tree,
607 inode, async_extent->start,
608 async_extent->start +
609 async_extent->ram_size - 1,
617 lock_extent(io_tree, async_extent->start,
618 async_extent->start + async_extent->ram_size - 1,
621 trans = btrfs_join_transaction(root, 1);
622 BUG_ON(IS_ERR(trans));
623 ret = btrfs_reserve_extent(trans, root,
624 async_extent->compressed_size,
625 async_extent->compressed_size,
628 btrfs_end_transaction(trans, root);
632 for (i = 0; i < async_extent->nr_pages; i++) {
633 WARN_ON(async_extent->pages[i]->mapping);
634 page_cache_release(async_extent->pages[i]);
636 kfree(async_extent->pages);
637 async_extent->nr_pages = 0;
638 async_extent->pages = NULL;
639 unlock_extent(io_tree, async_extent->start,
640 async_extent->start +
641 async_extent->ram_size - 1, GFP_NOFS);
646 * here we're doing allocation and writeback of the
649 btrfs_drop_extent_cache(inode, async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1, 0);
653 em = alloc_extent_map(GFP_NOFS);
655 em->start = async_extent->start;
656 em->len = async_extent->ram_size;
657 em->orig_start = em->start;
659 em->block_start = ins.objectid;
660 em->block_len = ins.offset;
661 em->bdev = root->fs_info->fs_devices->latest_bdev;
662 em->compress_type = async_extent->compress_type;
663 set_bit(EXTENT_FLAG_PINNED, &em->flags);
664 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
667 write_lock(&em_tree->lock);
668 ret = add_extent_mapping(em_tree, em);
669 write_unlock(&em_tree->lock);
670 if (ret != -EEXIST) {
674 btrfs_drop_extent_cache(inode, async_extent->start,
675 async_extent->start +
676 async_extent->ram_size - 1, 0);
679 ret = btrfs_add_ordered_extent_compress(inode,
682 async_extent->ram_size,
684 BTRFS_ORDERED_COMPRESSED,
685 async_extent->compress_type);
689 * clear dirty, set writeback and unlock the pages.
691 extent_clear_unlock_delalloc(inode,
692 &BTRFS_I(inode)->io_tree,
694 async_extent->start +
695 async_extent->ram_size - 1,
696 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
697 EXTENT_CLEAR_UNLOCK |
698 EXTENT_CLEAR_DELALLOC |
699 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
701 ret = btrfs_submit_compressed_write(inode,
703 async_extent->ram_size,
705 ins.offset, async_extent->pages,
706 async_extent->nr_pages);
709 alloc_hint = ins.objectid + ins.offset;
717 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
720 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
721 struct extent_map *em;
724 read_lock(&em_tree->lock);
725 em = search_extent_mapping(em_tree, start, num_bytes);
728 * if block start isn't an actual block number then find the
729 * first block in this inode and use that as a hint. If that
730 * block is also bogus then just don't worry about it.
732 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
734 em = search_extent_mapping(em_tree, 0, 0);
735 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
736 alloc_hint = em->block_start;
740 alloc_hint = em->block_start;
744 read_unlock(&em_tree->lock);
750 * when extent_io.c finds a delayed allocation range in the file,
751 * the call backs end up in this code. The basic idea is to
752 * allocate extents on disk for the range, and create ordered data structs
753 * in ram to track those extents.
755 * locked_page is the page that writepage had locked already. We use
756 * it to make sure we don't do extra locks or unlocks.
758 * *page_started is set to one if we unlock locked_page and do everything
759 * required to start IO on it. It may be clean and already done with
762 static noinline int cow_file_range(struct inode *inode,
763 struct page *locked_page,
764 u64 start, u64 end, int *page_started,
765 unsigned long *nr_written,
768 struct btrfs_root *root = BTRFS_I(inode)->root;
769 struct btrfs_trans_handle *trans;
772 unsigned long ram_size;
775 u64 blocksize = root->sectorsize;
776 struct btrfs_key ins;
777 struct extent_map *em;
778 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
781 BUG_ON(root == root->fs_info->tree_root);
782 trans = btrfs_join_transaction(root, 1);
783 BUG_ON(IS_ERR(trans));
784 btrfs_set_trans_block_group(trans, inode);
785 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
787 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
788 num_bytes = max(blocksize, num_bytes);
789 disk_num_bytes = num_bytes;
793 /* lets try to make an inline extent */
794 ret = cow_file_range_inline(trans, root, inode,
795 start, end, 0, NULL);
797 extent_clear_unlock_delalloc(inode,
798 &BTRFS_I(inode)->io_tree,
800 EXTENT_CLEAR_UNLOCK_PAGE |
801 EXTENT_CLEAR_UNLOCK |
802 EXTENT_CLEAR_DELALLOC |
804 EXTENT_SET_WRITEBACK |
805 EXTENT_END_WRITEBACK);
807 *nr_written = *nr_written +
808 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
815 BUG_ON(disk_num_bytes >
816 btrfs_super_total_bytes(&root->fs_info->super_copy));
818 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
819 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
821 while (disk_num_bytes > 0) {
824 cur_alloc_size = disk_num_bytes;
825 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
826 root->sectorsize, 0, alloc_hint,
830 em = alloc_extent_map(GFP_NOFS);
833 em->orig_start = em->start;
834 ram_size = ins.offset;
835 em->len = ins.offset;
837 em->block_start = ins.objectid;
838 em->block_len = ins.offset;
839 em->bdev = root->fs_info->fs_devices->latest_bdev;
840 set_bit(EXTENT_FLAG_PINNED, &em->flags);
843 write_lock(&em_tree->lock);
844 ret = add_extent_mapping(em_tree, em);
845 write_unlock(&em_tree->lock);
846 if (ret != -EEXIST) {
850 btrfs_drop_extent_cache(inode, start,
851 start + ram_size - 1, 0);
854 cur_alloc_size = ins.offset;
855 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
856 ram_size, cur_alloc_size, 0);
859 if (root->root_key.objectid ==
860 BTRFS_DATA_RELOC_TREE_OBJECTID) {
861 ret = btrfs_reloc_clone_csums(inode, start,
866 if (disk_num_bytes < cur_alloc_size)
869 /* we're not doing compressed IO, don't unlock the first
870 * page (which the caller expects to stay locked), don't
871 * clear any dirty bits and don't set any writeback bits
873 * Do set the Private2 bit so we know this page was properly
874 * setup for writepage
876 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
877 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
880 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
881 start, start + ram_size - 1,
883 disk_num_bytes -= cur_alloc_size;
884 num_bytes -= cur_alloc_size;
885 alloc_hint = ins.objectid + ins.offset;
886 start += cur_alloc_size;
890 btrfs_end_transaction(trans, root);
896 * work queue call back to started compression on a file and pages
898 static noinline void async_cow_start(struct btrfs_work *work)
900 struct async_cow *async_cow;
902 async_cow = container_of(work, struct async_cow, work);
904 compress_file_range(async_cow->inode, async_cow->locked_page,
905 async_cow->start, async_cow->end, async_cow,
908 async_cow->inode = NULL;
912 * work queue call back to submit previously compressed pages
914 static noinline void async_cow_submit(struct btrfs_work *work)
916 struct async_cow *async_cow;
917 struct btrfs_root *root;
918 unsigned long nr_pages;
920 async_cow = container_of(work, struct async_cow, work);
922 root = async_cow->root;
923 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
926 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
928 if (atomic_read(&root->fs_info->async_delalloc_pages) <
930 waitqueue_active(&root->fs_info->async_submit_wait))
931 wake_up(&root->fs_info->async_submit_wait);
933 if (async_cow->inode)
934 submit_compressed_extents(async_cow->inode, async_cow);
937 static noinline void async_cow_free(struct btrfs_work *work)
939 struct async_cow *async_cow;
940 async_cow = container_of(work, struct async_cow, work);
944 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
945 u64 start, u64 end, int *page_started,
946 unsigned long *nr_written)
948 struct async_cow *async_cow;
949 struct btrfs_root *root = BTRFS_I(inode)->root;
950 unsigned long nr_pages;
952 int limit = 10 * 1024 * 1042;
954 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
955 1, 0, NULL, GFP_NOFS);
956 while (start < end) {
957 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
958 async_cow->inode = inode;
959 async_cow->root = root;
960 async_cow->locked_page = locked_page;
961 async_cow->start = start;
963 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
966 cur_end = min(end, start + 512 * 1024 - 1);
968 async_cow->end = cur_end;
969 INIT_LIST_HEAD(&async_cow->extents);
971 async_cow->work.func = async_cow_start;
972 async_cow->work.ordered_func = async_cow_submit;
973 async_cow->work.ordered_free = async_cow_free;
974 async_cow->work.flags = 0;
976 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
978 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
980 btrfs_queue_worker(&root->fs_info->delalloc_workers,
983 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
984 wait_event(root->fs_info->async_submit_wait,
985 (atomic_read(&root->fs_info->async_delalloc_pages) <
989 while (atomic_read(&root->fs_info->async_submit_draining) &&
990 atomic_read(&root->fs_info->async_delalloc_pages)) {
991 wait_event(root->fs_info->async_submit_wait,
992 (atomic_read(&root->fs_info->async_delalloc_pages) ==
996 *nr_written += nr_pages;
1003 static noinline int csum_exist_in_range(struct btrfs_root *root,
1004 u64 bytenr, u64 num_bytes)
1007 struct btrfs_ordered_sum *sums;
1010 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1011 bytenr + num_bytes - 1, &list);
1012 if (ret == 0 && list_empty(&list))
1015 while (!list_empty(&list)) {
1016 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1017 list_del(&sums->list);
1024 * when nowcow writeback call back. This checks for snapshots or COW copies
1025 * of the extents that exist in the file, and COWs the file as required.
1027 * If no cow copies or snapshots exist, we write directly to the existing
1030 static noinline int run_delalloc_nocow(struct inode *inode,
1031 struct page *locked_page,
1032 u64 start, u64 end, int *page_started, int force,
1033 unsigned long *nr_written)
1035 struct btrfs_root *root = BTRFS_I(inode)->root;
1036 struct btrfs_trans_handle *trans;
1037 struct extent_buffer *leaf;
1038 struct btrfs_path *path;
1039 struct btrfs_file_extent_item *fi;
1040 struct btrfs_key found_key;
1052 bool nolock = false;
1054 path = btrfs_alloc_path();
1056 if (root == root->fs_info->tree_root) {
1058 trans = btrfs_join_transaction_nolock(root, 1);
1060 trans = btrfs_join_transaction(root, 1);
1062 BUG_ON(IS_ERR(trans));
1064 cow_start = (u64)-1;
1067 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1070 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1071 leaf = path->nodes[0];
1072 btrfs_item_key_to_cpu(leaf, &found_key,
1073 path->slots[0] - 1);
1074 if (found_key.objectid == inode->i_ino &&
1075 found_key.type == BTRFS_EXTENT_DATA_KEY)
1080 leaf = path->nodes[0];
1081 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1082 ret = btrfs_next_leaf(root, path);
1087 leaf = path->nodes[0];
1093 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1095 if (found_key.objectid > inode->i_ino ||
1096 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1097 found_key.offset > end)
1100 if (found_key.offset > cur_offset) {
1101 extent_end = found_key.offset;
1106 fi = btrfs_item_ptr(leaf, path->slots[0],
1107 struct btrfs_file_extent_item);
1108 extent_type = btrfs_file_extent_type(leaf, fi);
1110 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1111 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1112 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1113 extent_offset = btrfs_file_extent_offset(leaf, fi);
1114 extent_end = found_key.offset +
1115 btrfs_file_extent_num_bytes(leaf, fi);
1116 if (extent_end <= start) {
1120 if (disk_bytenr == 0)
1122 if (btrfs_file_extent_compression(leaf, fi) ||
1123 btrfs_file_extent_encryption(leaf, fi) ||
1124 btrfs_file_extent_other_encoding(leaf, fi))
1126 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1128 if (btrfs_extent_readonly(root, disk_bytenr))
1130 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1132 extent_offset, disk_bytenr))
1134 disk_bytenr += extent_offset;
1135 disk_bytenr += cur_offset - found_key.offset;
1136 num_bytes = min(end + 1, extent_end) - cur_offset;
1138 * force cow if csum exists in the range.
1139 * this ensure that csum for a given extent are
1140 * either valid or do not exist.
1142 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1145 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1146 extent_end = found_key.offset +
1147 btrfs_file_extent_inline_len(leaf, fi);
1148 extent_end = ALIGN(extent_end, root->sectorsize);
1153 if (extent_end <= start) {
1158 if (cow_start == (u64)-1)
1159 cow_start = cur_offset;
1160 cur_offset = extent_end;
1161 if (cur_offset > end)
1167 btrfs_release_path(root, path);
1168 if (cow_start != (u64)-1) {
1169 ret = cow_file_range(inode, locked_page, cow_start,
1170 found_key.offset - 1, page_started,
1173 cow_start = (u64)-1;
1176 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1177 struct extent_map *em;
1178 struct extent_map_tree *em_tree;
1179 em_tree = &BTRFS_I(inode)->extent_tree;
1180 em = alloc_extent_map(GFP_NOFS);
1182 em->start = cur_offset;
1183 em->orig_start = em->start;
1184 em->len = num_bytes;
1185 em->block_len = num_bytes;
1186 em->block_start = disk_bytenr;
1187 em->bdev = root->fs_info->fs_devices->latest_bdev;
1188 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1190 write_lock(&em_tree->lock);
1191 ret = add_extent_mapping(em_tree, em);
1192 write_unlock(&em_tree->lock);
1193 if (ret != -EEXIST) {
1194 free_extent_map(em);
1197 btrfs_drop_extent_cache(inode, em->start,
1198 em->start + em->len - 1, 0);
1200 type = BTRFS_ORDERED_PREALLOC;
1202 type = BTRFS_ORDERED_NOCOW;
1205 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1206 num_bytes, num_bytes, type);
1209 if (root->root_key.objectid ==
1210 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1211 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1216 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1217 cur_offset, cur_offset + num_bytes - 1,
1218 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1219 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1220 EXTENT_SET_PRIVATE2);
1221 cur_offset = extent_end;
1222 if (cur_offset > end)
1225 btrfs_release_path(root, path);
1227 if (cur_offset <= end && cow_start == (u64)-1)
1228 cow_start = cur_offset;
1229 if (cow_start != (u64)-1) {
1230 ret = cow_file_range(inode, locked_page, cow_start, end,
1231 page_started, nr_written, 1);
1236 ret = btrfs_end_transaction_nolock(trans, root);
1239 ret = btrfs_end_transaction(trans, root);
1242 btrfs_free_path(path);
1247 * extent_io.c call back to do delayed allocation processing
1249 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1250 u64 start, u64 end, int *page_started,
1251 unsigned long *nr_written)
1254 struct btrfs_root *root = BTRFS_I(inode)->root;
1256 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1257 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258 page_started, 1, nr_written);
1259 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1260 ret = run_delalloc_nocow(inode, locked_page, start, end,
1261 page_started, 0, nr_written);
1262 else if (!btrfs_test_opt(root, COMPRESS) &&
1263 !(BTRFS_I(inode)->force_compress) &&
1264 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1265 ret = cow_file_range(inode, locked_page, start, end,
1266 page_started, nr_written, 1);
1268 ret = cow_file_range_async(inode, locked_page, start, end,
1269 page_started, nr_written);
1273 static int btrfs_split_extent_hook(struct inode *inode,
1274 struct extent_state *orig, u64 split)
1276 /* not delalloc, ignore it */
1277 if (!(orig->state & EXTENT_DELALLOC))
1280 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1285 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1286 * extents so we can keep track of new extents that are just merged onto old
1287 * extents, such as when we are doing sequential writes, so we can properly
1288 * account for the metadata space we'll need.
1290 static int btrfs_merge_extent_hook(struct inode *inode,
1291 struct extent_state *new,
1292 struct extent_state *other)
1294 /* not delalloc, ignore it */
1295 if (!(other->state & EXTENT_DELALLOC))
1298 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1303 * extent_io.c set_bit_hook, used to track delayed allocation
1304 * bytes in this file, and to maintain the list of inodes that
1305 * have pending delalloc work to be done.
1307 static int btrfs_set_bit_hook(struct inode *inode,
1308 struct extent_state *state, int *bits)
1312 * set_bit and clear bit hooks normally require _irqsave/restore
1313 * but in this case, we are only testeing for the DELALLOC
1314 * bit, which is only set or cleared with irqs on
1316 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1317 struct btrfs_root *root = BTRFS_I(inode)->root;
1318 u64 len = state->end + 1 - state->start;
1319 int do_list = (root->root_key.objectid !=
1320 BTRFS_ROOT_TREE_OBJECTID);
1322 if (*bits & EXTENT_FIRST_DELALLOC)
1323 *bits &= ~EXTENT_FIRST_DELALLOC;
1325 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1327 spin_lock(&root->fs_info->delalloc_lock);
1328 BTRFS_I(inode)->delalloc_bytes += len;
1329 root->fs_info->delalloc_bytes += len;
1330 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332 &root->fs_info->delalloc_inodes);
1334 spin_unlock(&root->fs_info->delalloc_lock);
1340 * extent_io.c clear_bit_hook, see set_bit_hook for why
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343 struct extent_state *state, int *bits)
1346 * set_bit and clear bit hooks normally require _irqsave/restore
1347 * but in this case, we are only testeing for the DELALLOC
1348 * bit, which is only set or cleared with irqs on
1350 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1351 struct btrfs_root *root = BTRFS_I(inode)->root;
1352 u64 len = state->end + 1 - state->start;
1353 int do_list = (root->root_key.objectid !=
1354 BTRFS_ROOT_TREE_OBJECTID);
1356 if (*bits & EXTENT_FIRST_DELALLOC)
1357 *bits &= ~EXTENT_FIRST_DELALLOC;
1358 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1359 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1361 if (*bits & EXTENT_DO_ACCOUNTING)
1362 btrfs_delalloc_release_metadata(inode, len);
1364 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1366 btrfs_free_reserved_data_space(inode, len);
1368 spin_lock(&root->fs_info->delalloc_lock);
1369 root->fs_info->delalloc_bytes -= len;
1370 BTRFS_I(inode)->delalloc_bytes -= len;
1372 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1373 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1374 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1376 spin_unlock(&root->fs_info->delalloc_lock);
1382 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1383 * we don't create bios that span stripes or chunks
1385 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1386 size_t size, struct bio *bio,
1387 unsigned long bio_flags)
1389 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1390 struct btrfs_mapping_tree *map_tree;
1391 u64 logical = (u64)bio->bi_sector << 9;
1396 if (bio_flags & EXTENT_BIO_COMPRESSED)
1399 length = bio->bi_size;
1400 map_tree = &root->fs_info->mapping_tree;
1401 map_length = length;
1402 ret = btrfs_map_block(map_tree, READ, logical,
1403 &map_length, NULL, 0);
1405 if (map_length < length + size)
1411 * in order to insert checksums into the metadata in large chunks,
1412 * we wait until bio submission time. All the pages in the bio are
1413 * checksummed and sums are attached onto the ordered extent record.
1415 * At IO completion time the cums attached on the ordered extent record
1416 * are inserted into the btree
1418 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1419 struct bio *bio, int mirror_num,
1420 unsigned long bio_flags,
1423 struct btrfs_root *root = BTRFS_I(inode)->root;
1426 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1432 * in order to insert checksums into the metadata in large chunks,
1433 * we wait until bio submission time. All the pages in the bio are
1434 * checksummed and sums are attached onto the ordered extent record.
1436 * At IO completion time the cums attached on the ordered extent record
1437 * are inserted into the btree
1439 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1440 int mirror_num, unsigned long bio_flags,
1443 struct btrfs_root *root = BTRFS_I(inode)->root;
1444 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1448 * extent_io.c submission hook. This does the right thing for csum calculation
1449 * on write, or reading the csums from the tree before a read
1451 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1452 int mirror_num, unsigned long bio_flags,
1455 struct btrfs_root *root = BTRFS_I(inode)->root;
1459 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1461 if (root == root->fs_info->tree_root)
1462 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1464 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1467 if (!(rw & REQ_WRITE)) {
1468 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1469 return btrfs_submit_compressed_read(inode, bio,
1470 mirror_num, bio_flags);
1471 } else if (!skip_sum) {
1472 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1477 } else if (!skip_sum) {
1478 /* csum items have already been cloned */
1479 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1481 /* we're doing a write, do the async checksumming */
1482 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483 inode, rw, bio, mirror_num,
1484 bio_flags, bio_offset,
1485 __btrfs_submit_bio_start,
1486 __btrfs_submit_bio_done);
1490 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1494 * given a list of ordered sums record them in the inode. This happens
1495 * at IO completion time based on sums calculated at bio submission time.
1497 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1498 struct inode *inode, u64 file_offset,
1499 struct list_head *list)
1501 struct btrfs_ordered_sum *sum;
1503 btrfs_set_trans_block_group(trans, inode);
1505 list_for_each_entry(sum, list, list) {
1506 btrfs_csum_file_blocks(trans,
1507 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1512 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1513 struct extent_state **cached_state)
1515 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1517 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1518 cached_state, GFP_NOFS);
1521 /* see btrfs_writepage_start_hook for details on why this is required */
1522 struct btrfs_writepage_fixup {
1524 struct btrfs_work work;
1527 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1529 struct btrfs_writepage_fixup *fixup;
1530 struct btrfs_ordered_extent *ordered;
1531 struct extent_state *cached_state = NULL;
1533 struct inode *inode;
1537 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1541 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1542 ClearPageChecked(page);
1546 inode = page->mapping->host;
1547 page_start = page_offset(page);
1548 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1550 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1551 &cached_state, GFP_NOFS);
1553 /* already ordered? We're done */
1554 if (PagePrivate2(page))
1557 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1559 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1560 page_end, &cached_state, GFP_NOFS);
1562 btrfs_start_ordered_extent(inode, ordered, 1);
1567 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1568 ClearPageChecked(page);
1570 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1571 &cached_state, GFP_NOFS);
1574 page_cache_release(page);
1579 * There are a few paths in the higher layers of the kernel that directly
1580 * set the page dirty bit without asking the filesystem if it is a
1581 * good idea. This causes problems because we want to make sure COW
1582 * properly happens and the data=ordered rules are followed.
1584 * In our case any range that doesn't have the ORDERED bit set
1585 * hasn't been properly setup for IO. We kick off an async process
1586 * to fix it up. The async helper will wait for ordered extents, set
1587 * the delalloc bit and make it safe to write the page.
1589 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1591 struct inode *inode = page->mapping->host;
1592 struct btrfs_writepage_fixup *fixup;
1593 struct btrfs_root *root = BTRFS_I(inode)->root;
1595 /* this page is properly in the ordered list */
1596 if (TestClearPagePrivate2(page))
1599 if (PageChecked(page))
1602 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1606 SetPageChecked(page);
1607 page_cache_get(page);
1608 fixup->work.func = btrfs_writepage_fixup_worker;
1610 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1614 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1615 struct inode *inode, u64 file_pos,
1616 u64 disk_bytenr, u64 disk_num_bytes,
1617 u64 num_bytes, u64 ram_bytes,
1618 u8 compression, u8 encryption,
1619 u16 other_encoding, int extent_type)
1621 struct btrfs_root *root = BTRFS_I(inode)->root;
1622 struct btrfs_file_extent_item *fi;
1623 struct btrfs_path *path;
1624 struct extent_buffer *leaf;
1625 struct btrfs_key ins;
1629 path = btrfs_alloc_path();
1632 path->leave_spinning = 1;
1635 * we may be replacing one extent in the tree with another.
1636 * The new extent is pinned in the extent map, and we don't want
1637 * to drop it from the cache until it is completely in the btree.
1639 * So, tell btrfs_drop_extents to leave this extent in the cache.
1640 * the caller is expected to unpin it and allow it to be merged
1643 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1647 ins.objectid = inode->i_ino;
1648 ins.offset = file_pos;
1649 ins.type = BTRFS_EXTENT_DATA_KEY;
1650 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1652 leaf = path->nodes[0];
1653 fi = btrfs_item_ptr(leaf, path->slots[0],
1654 struct btrfs_file_extent_item);
1655 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1656 btrfs_set_file_extent_type(leaf, fi, extent_type);
1657 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1658 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1659 btrfs_set_file_extent_offset(leaf, fi, 0);
1660 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1661 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1662 btrfs_set_file_extent_compression(leaf, fi, compression);
1663 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1664 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1666 btrfs_unlock_up_safe(path, 1);
1667 btrfs_set_lock_blocking(leaf);
1669 btrfs_mark_buffer_dirty(leaf);
1671 inode_add_bytes(inode, num_bytes);
1673 ins.objectid = disk_bytenr;
1674 ins.offset = disk_num_bytes;
1675 ins.type = BTRFS_EXTENT_ITEM_KEY;
1676 ret = btrfs_alloc_reserved_file_extent(trans, root,
1677 root->root_key.objectid,
1678 inode->i_ino, file_pos, &ins);
1680 btrfs_free_path(path);
1686 * helper function for btrfs_finish_ordered_io, this
1687 * just reads in some of the csum leaves to prime them into ram
1688 * before we start the transaction. It limits the amount of btree
1689 * reads required while inside the transaction.
1691 /* as ordered data IO finishes, this gets called so we can finish
1692 * an ordered extent if the range of bytes in the file it covers are
1695 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1697 struct btrfs_root *root = BTRFS_I(inode)->root;
1698 struct btrfs_trans_handle *trans = NULL;
1699 struct btrfs_ordered_extent *ordered_extent = NULL;
1700 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1701 struct extent_state *cached_state = NULL;
1702 int compress_type = 0;
1704 bool nolock = false;
1706 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1710 BUG_ON(!ordered_extent);
1712 nolock = (root == root->fs_info->tree_root);
1714 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1715 BUG_ON(!list_empty(&ordered_extent->list));
1716 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1719 trans = btrfs_join_transaction_nolock(root, 1);
1721 trans = btrfs_join_transaction(root, 1);
1722 BUG_ON(IS_ERR(trans));
1723 btrfs_set_trans_block_group(trans, inode);
1724 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1725 ret = btrfs_update_inode(trans, root, inode);
1731 lock_extent_bits(io_tree, ordered_extent->file_offset,
1732 ordered_extent->file_offset + ordered_extent->len - 1,
1733 0, &cached_state, GFP_NOFS);
1736 trans = btrfs_join_transaction_nolock(root, 1);
1738 trans = btrfs_join_transaction(root, 1);
1739 BUG_ON(IS_ERR(trans));
1740 btrfs_set_trans_block_group(trans, inode);
1741 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1743 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1744 compress_type = ordered_extent->compress_type;
1745 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1746 BUG_ON(compress_type);
1747 ret = btrfs_mark_extent_written(trans, inode,
1748 ordered_extent->file_offset,
1749 ordered_extent->file_offset +
1750 ordered_extent->len);
1753 BUG_ON(root == root->fs_info->tree_root);
1754 ret = insert_reserved_file_extent(trans, inode,
1755 ordered_extent->file_offset,
1756 ordered_extent->start,
1757 ordered_extent->disk_len,
1758 ordered_extent->len,
1759 ordered_extent->len,
1760 compress_type, 0, 0,
1761 BTRFS_FILE_EXTENT_REG);
1762 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1763 ordered_extent->file_offset,
1764 ordered_extent->len);
1767 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1768 ordered_extent->file_offset +
1769 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1771 add_pending_csums(trans, inode, ordered_extent->file_offset,
1772 &ordered_extent->list);
1774 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1775 ret = btrfs_update_inode(trans, root, inode);
1780 btrfs_end_transaction_nolock(trans, root);
1782 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1784 btrfs_end_transaction(trans, root);
1788 btrfs_put_ordered_extent(ordered_extent);
1789 /* once for the tree */
1790 btrfs_put_ordered_extent(ordered_extent);
1795 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1796 struct extent_state *state, int uptodate)
1798 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1800 ClearPagePrivate2(page);
1801 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1805 * When IO fails, either with EIO or csum verification fails, we
1806 * try other mirrors that might have a good copy of the data. This
1807 * io_failure_record is used to record state as we go through all the
1808 * mirrors. If another mirror has good data, the page is set up to date
1809 * and things continue. If a good mirror can't be found, the original
1810 * bio end_io callback is called to indicate things have failed.
1812 struct io_failure_record {
1817 unsigned long bio_flags;
1821 static int btrfs_io_failed_hook(struct bio *failed_bio,
1822 struct page *page, u64 start, u64 end,
1823 struct extent_state *state)
1825 struct io_failure_record *failrec = NULL;
1827 struct extent_map *em;
1828 struct inode *inode = page->mapping->host;
1829 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1830 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1837 ret = get_state_private(failure_tree, start, &private);
1839 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1842 failrec->start = start;
1843 failrec->len = end - start + 1;
1844 failrec->last_mirror = 0;
1845 failrec->bio_flags = 0;
1847 read_lock(&em_tree->lock);
1848 em = lookup_extent_mapping(em_tree, start, failrec->len);
1849 if (em->start > start || em->start + em->len < start) {
1850 free_extent_map(em);
1853 read_unlock(&em_tree->lock);
1855 if (!em || IS_ERR(em)) {
1859 logical = start - em->start;
1860 logical = em->block_start + logical;
1861 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1862 logical = em->block_start;
1863 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1864 extent_set_compress_type(&failrec->bio_flags,
1867 failrec->logical = logical;
1868 free_extent_map(em);
1869 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1870 EXTENT_DIRTY, GFP_NOFS);
1871 set_state_private(failure_tree, start,
1872 (u64)(unsigned long)failrec);
1874 failrec = (struct io_failure_record *)(unsigned long)private;
1876 num_copies = btrfs_num_copies(
1877 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1878 failrec->logical, failrec->len);
1879 failrec->last_mirror++;
1881 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1882 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1885 if (state && state->start != failrec->start)
1887 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1889 if (!state || failrec->last_mirror > num_copies) {
1890 set_state_private(failure_tree, failrec->start, 0);
1891 clear_extent_bits(failure_tree, failrec->start,
1892 failrec->start + failrec->len - 1,
1893 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1897 bio = bio_alloc(GFP_NOFS, 1);
1898 bio->bi_private = state;
1899 bio->bi_end_io = failed_bio->bi_end_io;
1900 bio->bi_sector = failrec->logical >> 9;
1901 bio->bi_bdev = failed_bio->bi_bdev;
1904 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1905 if (failed_bio->bi_rw & REQ_WRITE)
1910 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1911 failrec->last_mirror,
1912 failrec->bio_flags, 0);
1917 * each time an IO finishes, we do a fast check in the IO failure tree
1918 * to see if we need to process or clean up an io_failure_record
1920 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1923 u64 private_failure;
1924 struct io_failure_record *failure;
1928 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1929 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1930 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1931 start, &private_failure);
1933 failure = (struct io_failure_record *)(unsigned long)
1935 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1937 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1939 failure->start + failure->len - 1,
1940 EXTENT_DIRTY | EXTENT_LOCKED,
1949 * when reads are done, we need to check csums to verify the data is correct
1950 * if there's a match, we allow the bio to finish. If not, we go through
1951 * the io_failure_record routines to find good copies
1953 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1954 struct extent_state *state)
1956 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1957 struct inode *inode = page->mapping->host;
1958 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1960 u64 private = ~(u32)0;
1962 struct btrfs_root *root = BTRFS_I(inode)->root;
1965 if (PageChecked(page)) {
1966 ClearPageChecked(page);
1970 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1973 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1974 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1975 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1980 if (state && state->start == start) {
1981 private = state->private;
1984 ret = get_state_private(io_tree, start, &private);
1986 kaddr = kmap_atomic(page, KM_USER0);
1990 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1991 btrfs_csum_final(csum, (char *)&csum);
1992 if (csum != private)
1995 kunmap_atomic(kaddr, KM_USER0);
1997 /* if the io failure tree for this inode is non-empty,
1998 * check to see if we've recovered from a failed IO
2000 btrfs_clean_io_failures(inode, start);
2004 if (printk_ratelimit()) {
2005 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2006 "private %llu\n", page->mapping->host->i_ino,
2007 (unsigned long long)start, csum,
2008 (unsigned long long)private);
2010 memset(kaddr + offset, 1, end - start + 1);
2011 flush_dcache_page(page);
2012 kunmap_atomic(kaddr, KM_USER0);
2018 struct delayed_iput {
2019 struct list_head list;
2020 struct inode *inode;
2023 void btrfs_add_delayed_iput(struct inode *inode)
2025 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2026 struct delayed_iput *delayed;
2028 if (atomic_add_unless(&inode->i_count, -1, 1))
2031 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2032 delayed->inode = inode;
2034 spin_lock(&fs_info->delayed_iput_lock);
2035 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2036 spin_unlock(&fs_info->delayed_iput_lock);
2039 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2042 struct btrfs_fs_info *fs_info = root->fs_info;
2043 struct delayed_iput *delayed;
2046 spin_lock(&fs_info->delayed_iput_lock);
2047 empty = list_empty(&fs_info->delayed_iputs);
2048 spin_unlock(&fs_info->delayed_iput_lock);
2052 down_read(&root->fs_info->cleanup_work_sem);
2053 spin_lock(&fs_info->delayed_iput_lock);
2054 list_splice_init(&fs_info->delayed_iputs, &list);
2055 spin_unlock(&fs_info->delayed_iput_lock);
2057 while (!list_empty(&list)) {
2058 delayed = list_entry(list.next, struct delayed_iput, list);
2059 list_del(&delayed->list);
2060 iput(delayed->inode);
2063 up_read(&root->fs_info->cleanup_work_sem);
2067 * calculate extra metadata reservation when snapshotting a subvolume
2068 * contains orphan files.
2070 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2071 struct btrfs_pending_snapshot *pending,
2072 u64 *bytes_to_reserve)
2074 struct btrfs_root *root;
2075 struct btrfs_block_rsv *block_rsv;
2079 root = pending->root;
2080 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2083 block_rsv = root->orphan_block_rsv;
2085 /* orphan block reservation for the snapshot */
2086 num_bytes = block_rsv->size;
2089 * after the snapshot is created, COWing tree blocks may use more
2090 * space than it frees. So we should make sure there is enough
2093 index = trans->transid & 0x1;
2094 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2095 num_bytes += block_rsv->size -
2096 (block_rsv->reserved + block_rsv->freed[index]);
2099 *bytes_to_reserve += num_bytes;
2102 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2103 struct btrfs_pending_snapshot *pending)
2105 struct btrfs_root *root = pending->root;
2106 struct btrfs_root *snap = pending->snap;
2107 struct btrfs_block_rsv *block_rsv;
2112 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2115 /* refill source subvolume's orphan block reservation */
2116 block_rsv = root->orphan_block_rsv;
2117 index = trans->transid & 0x1;
2118 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2119 num_bytes = block_rsv->size -
2120 (block_rsv->reserved + block_rsv->freed[index]);
2121 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2122 root->orphan_block_rsv,
2127 /* setup orphan block reservation for the snapshot */
2128 block_rsv = btrfs_alloc_block_rsv(snap);
2131 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2132 snap->orphan_block_rsv = block_rsv;
2134 num_bytes = root->orphan_block_rsv->size;
2135 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2136 block_rsv, num_bytes);
2140 /* insert orphan item for the snapshot */
2141 WARN_ON(!root->orphan_item_inserted);
2142 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2143 snap->root_key.objectid);
2145 snap->orphan_item_inserted = 1;
2149 enum btrfs_orphan_cleanup_state {
2150 ORPHAN_CLEANUP_STARTED = 1,
2151 ORPHAN_CLEANUP_DONE = 2,
2155 * This is called in transaction commmit time. If there are no orphan
2156 * files in the subvolume, it removes orphan item and frees block_rsv
2159 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2160 struct btrfs_root *root)
2164 if (!list_empty(&root->orphan_list) ||
2165 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2168 if (root->orphan_item_inserted &&
2169 btrfs_root_refs(&root->root_item) > 0) {
2170 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2171 root->root_key.objectid);
2173 root->orphan_item_inserted = 0;
2176 if (root->orphan_block_rsv) {
2177 WARN_ON(root->orphan_block_rsv->size > 0);
2178 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2179 root->orphan_block_rsv = NULL;
2184 * This creates an orphan entry for the given inode in case something goes
2185 * wrong in the middle of an unlink/truncate.
2187 * NOTE: caller of this function should reserve 5 units of metadata for
2190 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2192 struct btrfs_root *root = BTRFS_I(inode)->root;
2193 struct btrfs_block_rsv *block_rsv = NULL;
2198 if (!root->orphan_block_rsv) {
2199 block_rsv = btrfs_alloc_block_rsv(root);
2203 spin_lock(&root->orphan_lock);
2204 if (!root->orphan_block_rsv) {
2205 root->orphan_block_rsv = block_rsv;
2206 } else if (block_rsv) {
2207 btrfs_free_block_rsv(root, block_rsv);
2211 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2212 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2215 * For proper ENOSPC handling, we should do orphan
2216 * cleanup when mounting. But this introduces backward
2217 * compatibility issue.
2219 if (!xchg(&root->orphan_item_inserted, 1))
2226 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
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);
2540 if (location.objectid == BTRFS_FREE_SPACE_OBJECTID)
2541 inode->i_mapping->flags &= ~__GFP_FS;
2544 * try to precache a NULL acl entry for files that don't have
2545 * any xattrs or acls
2547 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2549 cache_no_acl(inode);
2551 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2552 alloc_group_block, 0);
2553 btrfs_free_path(path);
2556 switch (inode->i_mode & S_IFMT) {
2558 inode->i_mapping->a_ops = &btrfs_aops;
2559 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2560 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2561 inode->i_fop = &btrfs_file_operations;
2562 inode->i_op = &btrfs_file_inode_operations;
2565 inode->i_fop = &btrfs_dir_file_operations;
2566 if (root == root->fs_info->tree_root)
2567 inode->i_op = &btrfs_dir_ro_inode_operations;
2569 inode->i_op = &btrfs_dir_inode_operations;
2572 inode->i_op = &btrfs_symlink_inode_operations;
2573 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2574 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2577 inode->i_op = &btrfs_special_inode_operations;
2578 init_special_inode(inode, inode->i_mode, rdev);
2582 btrfs_update_iflags(inode);
2586 btrfs_free_path(path);
2587 make_bad_inode(inode);
2591 * given a leaf and an inode, copy the inode fields into the leaf
2593 static void fill_inode_item(struct btrfs_trans_handle *trans,
2594 struct extent_buffer *leaf,
2595 struct btrfs_inode_item *item,
2596 struct inode *inode)
2598 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2599 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2600 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2601 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2602 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2604 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2605 inode->i_atime.tv_sec);
2606 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2607 inode->i_atime.tv_nsec);
2609 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2610 inode->i_mtime.tv_sec);
2611 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2612 inode->i_mtime.tv_nsec);
2614 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2615 inode->i_ctime.tv_sec);
2616 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2617 inode->i_ctime.tv_nsec);
2619 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2620 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2621 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2622 btrfs_set_inode_transid(leaf, item, trans->transid);
2623 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2624 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2625 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2629 * copy everything in the in-memory inode into the btree.
2631 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2632 struct btrfs_root *root, struct inode *inode)
2634 struct btrfs_inode_item *inode_item;
2635 struct btrfs_path *path;
2636 struct extent_buffer *leaf;
2639 path = btrfs_alloc_path();
2641 path->leave_spinning = 1;
2642 ret = btrfs_lookup_inode(trans, root, path,
2643 &BTRFS_I(inode)->location, 1);
2650 btrfs_unlock_up_safe(path, 1);
2651 leaf = path->nodes[0];
2652 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2653 struct btrfs_inode_item);
2655 fill_inode_item(trans, leaf, inode_item, inode);
2656 btrfs_mark_buffer_dirty(leaf);
2657 btrfs_set_inode_last_trans(trans, inode);
2660 btrfs_free_path(path);
2666 * unlink helper that gets used here in inode.c and in the tree logging
2667 * recovery code. It remove a link in a directory with a given name, and
2668 * also drops the back refs in the inode to the directory
2670 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2671 struct btrfs_root *root,
2672 struct inode *dir, struct inode *inode,
2673 const char *name, int name_len)
2675 struct btrfs_path *path;
2677 struct extent_buffer *leaf;
2678 struct btrfs_dir_item *di;
2679 struct btrfs_key key;
2682 path = btrfs_alloc_path();
2688 path->leave_spinning = 1;
2689 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2690 name, name_len, -1);
2699 leaf = path->nodes[0];
2700 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2701 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2704 btrfs_release_path(root, path);
2706 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2708 dir->i_ino, &index);
2710 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2711 "inode %lu parent %lu\n", name_len, name,
2712 inode->i_ino, dir->i_ino);
2716 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2717 index, name, name_len, -1);
2726 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2727 btrfs_release_path(root, path);
2729 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2731 BUG_ON(ret != 0 && ret != -ENOENT);
2733 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2738 btrfs_free_path(path);
2742 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2743 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2744 btrfs_update_inode(trans, root, dir);
2749 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2750 struct btrfs_root *root,
2751 struct inode *dir, struct inode *inode,
2752 const char *name, int name_len)
2755 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2757 btrfs_drop_nlink(inode);
2758 ret = btrfs_update_inode(trans, root, inode);
2764 /* helper to check if there is any shared block in the path */
2765 static int check_path_shared(struct btrfs_root *root,
2766 struct btrfs_path *path)
2768 struct extent_buffer *eb;
2772 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2775 if (!path->nodes[level])
2777 eb = path->nodes[level];
2778 if (!btrfs_block_can_be_shared(root, eb))
2780 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2789 * helper to start transaction for unlink and rmdir.
2791 * unlink and rmdir are special in btrfs, they do not always free space.
2792 * so in enospc case, we should make sure they will free space before
2793 * allowing them to use the global metadata reservation.
2795 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2796 struct dentry *dentry)
2798 struct btrfs_trans_handle *trans;
2799 struct btrfs_root *root = BTRFS_I(dir)->root;
2800 struct btrfs_path *path;
2801 struct btrfs_inode_ref *ref;
2802 struct btrfs_dir_item *di;
2803 struct inode *inode = dentry->d_inode;
2809 trans = btrfs_start_transaction(root, 10);
2810 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2813 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2814 return ERR_PTR(-ENOSPC);
2816 /* check if there is someone else holds reference */
2817 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2818 return ERR_PTR(-ENOSPC);
2820 if (atomic_read(&inode->i_count) > 2)
2821 return ERR_PTR(-ENOSPC);
2823 if (xchg(&root->fs_info->enospc_unlink, 1))
2824 return ERR_PTR(-ENOSPC);
2826 path = btrfs_alloc_path();
2828 root->fs_info->enospc_unlink = 0;
2829 return ERR_PTR(-ENOMEM);
2832 trans = btrfs_start_transaction(root, 0);
2833 if (IS_ERR(trans)) {
2834 btrfs_free_path(path);
2835 root->fs_info->enospc_unlink = 0;
2839 path->skip_locking = 1;
2840 path->search_commit_root = 1;
2842 ret = btrfs_lookup_inode(trans, root, path,
2843 &BTRFS_I(dir)->location, 0);
2849 if (check_path_shared(root, path))
2854 btrfs_release_path(root, path);
2856 ret = btrfs_lookup_inode(trans, root, path,
2857 &BTRFS_I(inode)->location, 0);
2863 if (check_path_shared(root, path))
2868 btrfs_release_path(root, path);
2870 if (ret == 0 && S_ISREG(inode->i_mode)) {
2871 ret = btrfs_lookup_file_extent(trans, root, path,
2872 inode->i_ino, (u64)-1, 0);
2878 if (check_path_shared(root, path))
2880 btrfs_release_path(root, path);
2888 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2889 dentry->d_name.name, dentry->d_name.len, 0);
2895 if (check_path_shared(root, path))
2901 btrfs_release_path(root, path);
2903 ref = btrfs_lookup_inode_ref(trans, root, path,
2904 dentry->d_name.name, dentry->d_name.len,
2905 inode->i_ino, dir->i_ino, 0);
2911 if (check_path_shared(root, path))
2913 index = btrfs_inode_ref_index(path->nodes[0], ref);
2914 btrfs_release_path(root, path);
2916 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2917 dentry->d_name.name, dentry->d_name.len, 0);
2922 BUG_ON(ret == -ENOENT);
2923 if (check_path_shared(root, path))
2928 btrfs_free_path(path);
2930 btrfs_end_transaction(trans, root);
2931 root->fs_info->enospc_unlink = 0;
2932 return ERR_PTR(err);
2935 trans->block_rsv = &root->fs_info->global_block_rsv;
2939 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2940 struct btrfs_root *root)
2942 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2943 BUG_ON(!root->fs_info->enospc_unlink);
2944 root->fs_info->enospc_unlink = 0;
2946 btrfs_end_transaction_throttle(trans, root);
2949 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2951 struct btrfs_root *root = BTRFS_I(dir)->root;
2952 struct btrfs_trans_handle *trans;
2953 struct inode *inode = dentry->d_inode;
2955 unsigned long nr = 0;
2957 trans = __unlink_start_trans(dir, dentry);
2959 return PTR_ERR(trans);
2961 btrfs_set_trans_block_group(trans, dir);
2963 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2965 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2966 dentry->d_name.name, dentry->d_name.len);
2969 if (inode->i_nlink == 0) {
2970 ret = btrfs_orphan_add(trans, inode);
2974 nr = trans->blocks_used;
2975 __unlink_end_trans(trans, root);
2976 btrfs_btree_balance_dirty(root, nr);
2980 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2981 struct btrfs_root *root,
2982 struct inode *dir, u64 objectid,
2983 const char *name, int name_len)
2985 struct btrfs_path *path;
2986 struct extent_buffer *leaf;
2987 struct btrfs_dir_item *di;
2988 struct btrfs_key key;
2992 path = btrfs_alloc_path();
2996 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2997 name, name_len, -1);
2998 BUG_ON(!di || IS_ERR(di));
3000 leaf = path->nodes[0];
3001 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3002 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3003 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3005 btrfs_release_path(root, path);
3007 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3008 objectid, root->root_key.objectid,
3009 dir->i_ino, &index, name, name_len);
3011 BUG_ON(ret != -ENOENT);
3012 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3014 BUG_ON(!di || IS_ERR(di));
3016 leaf = path->nodes[0];
3017 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3018 btrfs_release_path(root, path);
3022 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3023 index, name, name_len, -1);
3024 BUG_ON(!di || IS_ERR(di));
3026 leaf = path->nodes[0];
3027 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3028 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3029 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3031 btrfs_release_path(root, path);
3033 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3034 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3035 ret = btrfs_update_inode(trans, root, dir);
3038 btrfs_free_path(path);
3042 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3044 struct inode *inode = dentry->d_inode;
3046 struct btrfs_root *root = BTRFS_I(dir)->root;
3047 struct btrfs_trans_handle *trans;
3048 unsigned long nr = 0;
3050 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3051 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3054 trans = __unlink_start_trans(dir, dentry);
3056 return PTR_ERR(trans);
3058 btrfs_set_trans_block_group(trans, dir);
3060 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3061 err = btrfs_unlink_subvol(trans, root, dir,
3062 BTRFS_I(inode)->location.objectid,
3063 dentry->d_name.name,
3064 dentry->d_name.len);
3068 err = btrfs_orphan_add(trans, inode);
3072 /* now the directory is empty */
3073 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3074 dentry->d_name.name, dentry->d_name.len);
3076 btrfs_i_size_write(inode, 0);
3078 nr = trans->blocks_used;
3079 __unlink_end_trans(trans, root);
3080 btrfs_btree_balance_dirty(root, nr);
3087 * when truncating bytes in a file, it is possible to avoid reading
3088 * the leaves that contain only checksum items. This can be the
3089 * majority of the IO required to delete a large file, but it must
3090 * be done carefully.
3092 * The keys in the level just above the leaves are checked to make sure
3093 * the lowest key in a given leaf is a csum key, and starts at an offset
3094 * after the new size.
3096 * Then the key for the next leaf is checked to make sure it also has
3097 * a checksum item for the same file. If it does, we know our target leaf
3098 * contains only checksum items, and it can be safely freed without reading
3101 * This is just an optimization targeted at large files. It may do
3102 * nothing. It will return 0 unless things went badly.
3104 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3105 struct btrfs_root *root,
3106 struct btrfs_path *path,
3107 struct inode *inode, u64 new_size)
3109 struct btrfs_key key;
3112 struct btrfs_key found_key;
3113 struct btrfs_key other_key;
3114 struct btrfs_leaf_ref *ref;
3118 path->lowest_level = 1;
3119 key.objectid = inode->i_ino;
3120 key.type = BTRFS_CSUM_ITEM_KEY;
3121 key.offset = new_size;
3123 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3127 if (path->nodes[1] == NULL) {
3132 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3133 nritems = btrfs_header_nritems(path->nodes[1]);
3138 if (path->slots[1] >= nritems)
3141 /* did we find a key greater than anything we want to delete? */
3142 if (found_key.objectid > inode->i_ino ||
3143 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3146 /* we check the next key in the node to make sure the leave contains
3147 * only checksum items. This comparison doesn't work if our
3148 * leaf is the last one in the node
3150 if (path->slots[1] + 1 >= nritems) {
3152 /* search forward from the last key in the node, this
3153 * will bring us into the next node in the tree
3155 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3157 /* unlikely, but we inc below, so check to be safe */
3158 if (found_key.offset == (u64)-1)
3161 /* search_forward needs a path with locks held, do the
3162 * search again for the original key. It is possible
3163 * this will race with a balance and return a path that
3164 * we could modify, but this drop is just an optimization
3165 * and is allowed to miss some leaves.
3167 btrfs_release_path(root, path);
3170 /* setup a max key for search_forward */
3171 other_key.offset = (u64)-1;
3172 other_key.type = key.type;
3173 other_key.objectid = key.objectid;
3175 path->keep_locks = 1;
3176 ret = btrfs_search_forward(root, &found_key, &other_key,
3178 path->keep_locks = 0;
3179 if (ret || found_key.objectid != key.objectid ||
3180 found_key.type != key.type) {
3185 key.offset = found_key.offset;
3186 btrfs_release_path(root, path);
3191 /* we know there's one more slot after us in the tree,
3192 * read that key so we can verify it is also a checksum item
3194 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3196 if (found_key.objectid < inode->i_ino)
3199 if (found_key.type != key.type || found_key.offset < new_size)
3203 * if the key for the next leaf isn't a csum key from this objectid,
3204 * we can't be sure there aren't good items inside this leaf.
3207 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3210 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3211 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3213 * it is safe to delete this leaf, it contains only
3214 * csum items from this inode at an offset >= new_size
3216 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3219 if (root->ref_cows && leaf_gen < trans->transid) {
3220 ref = btrfs_alloc_leaf_ref(root, 0);
3222 ref->root_gen = root->root_key.offset;
3223 ref->bytenr = leaf_start;
3225 ref->generation = leaf_gen;
3228 btrfs_sort_leaf_ref(ref);
3230 ret = btrfs_add_leaf_ref(root, ref, 0);
3232 btrfs_free_leaf_ref(root, ref);
3238 btrfs_release_path(root, path);
3240 if (other_key.objectid == inode->i_ino &&
3241 other_key.type == key.type && other_key.offset > key.offset) {
3242 key.offset = other_key.offset;
3248 /* fixup any changes we've made to the path */
3249 path->lowest_level = 0;
3250 path->keep_locks = 0;
3251 btrfs_release_path(root, path);
3258 * this can truncate away extent items, csum items and directory items.
3259 * It starts at a high offset and removes keys until it can't find
3260 * any higher than new_size
3262 * csum items that cross the new i_size are truncated to the new size
3265 * min_type is the minimum key type to truncate down to. If set to 0, this
3266 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3268 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3269 struct btrfs_root *root,
3270 struct inode *inode,
3271 u64 new_size, u32 min_type)
3273 struct btrfs_path *path;
3274 struct extent_buffer *leaf;
3275 struct btrfs_file_extent_item *fi;
3276 struct btrfs_key key;
3277 struct btrfs_key found_key;
3278 u64 extent_start = 0;
3279 u64 extent_num_bytes = 0;
3280 u64 extent_offset = 0;
3282 u64 mask = root->sectorsize - 1;
3283 u32 found_type = (u8)-1;
3286 int pending_del_nr = 0;
3287 int pending_del_slot = 0;
3288 int extent_type = -1;
3293 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3295 if (root->ref_cows || root == root->fs_info->tree_root)
3296 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3298 path = btrfs_alloc_path();
3302 key.objectid = inode->i_ino;
3303 key.offset = (u64)-1;
3307 path->leave_spinning = 1;
3308 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3315 /* there are no items in the tree for us to truncate, we're
3318 if (path->slots[0] == 0)
3325 leaf = path->nodes[0];
3326 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3327 found_type = btrfs_key_type(&found_key);
3330 if (found_key.objectid != inode->i_ino)
3333 if (found_type < min_type)
3336 item_end = found_key.offset;
3337 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3338 fi = btrfs_item_ptr(leaf, path->slots[0],
3339 struct btrfs_file_extent_item);
3340 extent_type = btrfs_file_extent_type(leaf, fi);
3341 encoding = btrfs_file_extent_compression(leaf, fi);
3342 encoding |= btrfs_file_extent_encryption(leaf, fi);
3343 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3345 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3347 btrfs_file_extent_num_bytes(leaf, fi);
3348 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3349 item_end += btrfs_file_extent_inline_len(leaf,
3354 if (found_type > min_type) {
3357 if (item_end < new_size)
3359 if (found_key.offset >= new_size)
3365 /* FIXME, shrink the extent if the ref count is only 1 */
3366 if (found_type != BTRFS_EXTENT_DATA_KEY)
3369 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3371 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3372 if (!del_item && !encoding) {
3373 u64 orig_num_bytes =
3374 btrfs_file_extent_num_bytes(leaf, fi);
3375 extent_num_bytes = new_size -
3376 found_key.offset + root->sectorsize - 1;
3377 extent_num_bytes = extent_num_bytes &
3378 ~((u64)root->sectorsize - 1);
3379 btrfs_set_file_extent_num_bytes(leaf, fi,
3381 num_dec = (orig_num_bytes -
3383 if (root->ref_cows && extent_start != 0)
3384 inode_sub_bytes(inode, num_dec);
3385 btrfs_mark_buffer_dirty(leaf);
3388 btrfs_file_extent_disk_num_bytes(leaf,
3390 extent_offset = found_key.offset -
3391 btrfs_file_extent_offset(leaf, fi);
3393 /* FIXME blocksize != 4096 */
3394 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3395 if (extent_start != 0) {
3398 inode_sub_bytes(inode, num_dec);
3401 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3403 * we can't truncate inline items that have had
3407 btrfs_file_extent_compression(leaf, fi) == 0 &&
3408 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3409 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3410 u32 size = new_size - found_key.offset;
3412 if (root->ref_cows) {
3413 inode_sub_bytes(inode, item_end + 1 -
3417 btrfs_file_extent_calc_inline_size(size);
3418 ret = btrfs_truncate_item(trans, root, path,
3421 } else if (root->ref_cows) {
3422 inode_sub_bytes(inode, item_end + 1 -
3428 if (!pending_del_nr) {
3429 /* no pending yet, add ourselves */
3430 pending_del_slot = path->slots[0];
3432 } else if (pending_del_nr &&
3433 path->slots[0] + 1 == pending_del_slot) {
3434 /* hop on the pending chunk */
3436 pending_del_slot = path->slots[0];
3443 if (found_extent && (root->ref_cows ||
3444 root == root->fs_info->tree_root)) {
3445 btrfs_set_path_blocking(path);
3446 ret = btrfs_free_extent(trans, root, extent_start,
3447 extent_num_bytes, 0,
3448 btrfs_header_owner(leaf),
3449 inode->i_ino, extent_offset);
3453 if (found_type == BTRFS_INODE_ITEM_KEY)
3456 if (path->slots[0] == 0 ||
3457 path->slots[0] != pending_del_slot) {
3458 if (root->ref_cows) {
3462 if (pending_del_nr) {
3463 ret = btrfs_del_items(trans, root, path,
3469 btrfs_release_path(root, path);
3476 if (pending_del_nr) {
3477 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3481 btrfs_free_path(path);
3486 * taken from block_truncate_page, but does cow as it zeros out
3487 * any bytes left in the last page in the file.
3489 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3491 struct inode *inode = mapping->host;
3492 struct btrfs_root *root = BTRFS_I(inode)->root;
3493 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3494 struct btrfs_ordered_extent *ordered;
3495 struct extent_state *cached_state = NULL;
3497 u32 blocksize = root->sectorsize;
3498 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3499 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3505 if ((offset & (blocksize - 1)) == 0)
3507 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3513 page = grab_cache_page(mapping, index);
3515 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3519 page_start = page_offset(page);
3520 page_end = page_start + PAGE_CACHE_SIZE - 1;
3522 if (!PageUptodate(page)) {
3523 ret = btrfs_readpage(NULL, page);
3525 if (page->mapping != mapping) {
3527 page_cache_release(page);
3530 if (!PageUptodate(page)) {
3535 wait_on_page_writeback(page);
3537 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3539 set_page_extent_mapped(page);
3541 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3543 unlock_extent_cached(io_tree, page_start, page_end,
3544 &cached_state, GFP_NOFS);
3546 page_cache_release(page);
3547 btrfs_start_ordered_extent(inode, ordered, 1);
3548 btrfs_put_ordered_extent(ordered);
3552 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3553 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3554 0, 0, &cached_state, GFP_NOFS);
3556 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3559 unlock_extent_cached(io_tree, page_start, page_end,
3560 &cached_state, GFP_NOFS);
3565 if (offset != PAGE_CACHE_SIZE) {
3567 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3568 flush_dcache_page(page);
3571 ClearPageChecked(page);
3572 set_page_dirty(page);
3573 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3578 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3580 page_cache_release(page);
3586 * This function puts in dummy file extents for the area we're creating a hole
3587 * for. So if we are truncating this file to a larger size we need to insert
3588 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3589 * the range between oldsize and size
3591 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3593 struct btrfs_trans_handle *trans;
3594 struct btrfs_root *root = BTRFS_I(inode)->root;
3595 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3596 struct extent_map *em = NULL;
3597 struct extent_state *cached_state = NULL;
3598 u64 mask = root->sectorsize - 1;
3599 u64 hole_start = (oldsize + mask) & ~mask;
3600 u64 block_end = (size + mask) & ~mask;
3606 if (size <= hole_start)
3610 struct btrfs_ordered_extent *ordered;
3611 btrfs_wait_ordered_range(inode, hole_start,
3612 block_end - hole_start);
3613 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3614 &cached_state, GFP_NOFS);
3615 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3618 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3619 &cached_state, GFP_NOFS);
3620 btrfs_put_ordered_extent(ordered);
3623 cur_offset = hole_start;
3625 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3626 block_end - cur_offset, 0);
3627 BUG_ON(IS_ERR(em) || !em);
3628 last_byte = min(extent_map_end(em), block_end);
3629 last_byte = (last_byte + mask) & ~mask;
3630 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3632 hole_size = last_byte - cur_offset;
3634 trans = btrfs_start_transaction(root, 2);
3635 if (IS_ERR(trans)) {
3636 err = PTR_ERR(trans);
3639 btrfs_set_trans_block_group(trans, inode);
3641 err = btrfs_drop_extents(trans, inode, cur_offset,
3642 cur_offset + hole_size,
3647 err = btrfs_insert_file_extent(trans, root,
3648 inode->i_ino, cur_offset, 0,
3649 0, hole_size, 0, hole_size,
3654 btrfs_drop_extent_cache(inode, hole_start,
3657 btrfs_end_transaction(trans, root);
3659 free_extent_map(em);
3661 cur_offset = last_byte;
3662 if (cur_offset >= block_end)
3666 free_extent_map(em);
3667 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3672 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3674 loff_t oldsize = i_size_read(inode);
3677 if (newsize == oldsize)
3680 if (newsize > oldsize) {
3681 i_size_write(inode, newsize);
3682 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3683 truncate_pagecache(inode, oldsize, newsize);
3684 ret = btrfs_cont_expand(inode, oldsize, newsize);
3686 btrfs_setsize(inode, oldsize);
3690 mark_inode_dirty(inode);
3694 * We're truncating a file that used to have good data down to
3695 * zero. Make sure it gets into the ordered flush list so that
3696 * any new writes get down to disk quickly.
3699 BTRFS_I(inode)->ordered_data_close = 1;
3701 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3702 truncate_setsize(inode, newsize);
3703 ret = btrfs_truncate(inode);
3709 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3711 struct inode *inode = dentry->d_inode;
3712 struct btrfs_root *root = BTRFS_I(inode)->root;
3715 if (btrfs_root_readonly(root))
3718 err = inode_change_ok(inode, attr);
3722 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3723 err = btrfs_setsize(inode, attr->ia_size);
3728 if (attr->ia_valid) {
3729 setattr_copy(inode, attr);
3730 mark_inode_dirty(inode);
3732 if (attr->ia_valid & ATTR_MODE)
3733 err = btrfs_acl_chmod(inode);
3739 void btrfs_evict_inode(struct inode *inode)
3741 struct btrfs_trans_handle *trans;
3742 struct btrfs_root *root = BTRFS_I(inode)->root;
3746 trace_btrfs_inode_evict(inode);
3748 truncate_inode_pages(&inode->i_data, 0);
3749 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3750 root == root->fs_info->tree_root))
3753 if (is_bad_inode(inode)) {
3754 btrfs_orphan_del(NULL, inode);
3757 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3758 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3760 if (root->fs_info->log_root_recovering) {
3761 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3765 if (inode->i_nlink > 0) {
3766 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3770 btrfs_i_size_write(inode, 0);
3773 trans = btrfs_start_transaction(root, 0);
3774 BUG_ON(IS_ERR(trans));
3775 btrfs_set_trans_block_group(trans, inode);
3776 trans->block_rsv = root->orphan_block_rsv;
3778 ret = btrfs_block_rsv_check(trans, root,
3779 root->orphan_block_rsv, 0, 5);
3781 BUG_ON(ret != -EAGAIN);
3782 ret = btrfs_commit_transaction(trans, root);
3787 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3791 nr = trans->blocks_used;
3792 btrfs_end_transaction(trans, root);
3794 btrfs_btree_balance_dirty(root, nr);
3799 ret = btrfs_orphan_del(trans, inode);
3803 nr = trans->blocks_used;
3804 btrfs_end_transaction(trans, root);
3805 btrfs_btree_balance_dirty(root, nr);
3807 end_writeback(inode);
3812 * this returns the key found in the dir entry in the location pointer.
3813 * If no dir entries were found, location->objectid is 0.
3815 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3816 struct btrfs_key *location)
3818 const char *name = dentry->d_name.name;
3819 int namelen = dentry->d_name.len;
3820 struct btrfs_dir_item *di;
3821 struct btrfs_path *path;
3822 struct btrfs_root *root = BTRFS_I(dir)->root;
3825 path = btrfs_alloc_path();
3828 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3833 if (!di || IS_ERR(di))
3836 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3838 btrfs_free_path(path);
3841 location->objectid = 0;
3846 * when we hit a tree root in a directory, the btrfs part of the inode
3847 * needs to be changed to reflect the root directory of the tree root. This
3848 * is kind of like crossing a mount point.
3850 static int fixup_tree_root_location(struct btrfs_root *root,
3852 struct dentry *dentry,
3853 struct btrfs_key *location,
3854 struct btrfs_root **sub_root)
3856 struct btrfs_path *path;
3857 struct btrfs_root *new_root;
3858 struct btrfs_root_ref *ref;
3859 struct extent_buffer *leaf;
3863 path = btrfs_alloc_path();
3870 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3871 BTRFS_I(dir)->root->root_key.objectid,
3872 location->objectid);
3879 leaf = path->nodes[0];
3880 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3881 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3882 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3885 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3886 (unsigned long)(ref + 1),
3887 dentry->d_name.len);
3891 btrfs_release_path(root->fs_info->tree_root, path);
3893 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3894 if (IS_ERR(new_root)) {
3895 err = PTR_ERR(new_root);
3899 if (btrfs_root_refs(&new_root->root_item) == 0) {
3904 *sub_root = new_root;
3905 location->objectid = btrfs_root_dirid(&new_root->root_item);
3906 location->type = BTRFS_INODE_ITEM_KEY;
3907 location->offset = 0;
3910 btrfs_free_path(path);
3914 static void inode_tree_add(struct inode *inode)
3916 struct btrfs_root *root = BTRFS_I(inode)->root;
3917 struct btrfs_inode *entry;
3919 struct rb_node *parent;
3921 p = &root->inode_tree.rb_node;
3924 if (inode_unhashed(inode))
3927 spin_lock(&root->inode_lock);
3930 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3932 if (inode->i_ino < entry->vfs_inode.i_ino)
3933 p = &parent->rb_left;
3934 else if (inode->i_ino > entry->vfs_inode.i_ino)
3935 p = &parent->rb_right;
3937 WARN_ON(!(entry->vfs_inode.i_state &
3938 (I_WILL_FREE | I_FREEING)));
3939 rb_erase(parent, &root->inode_tree);
3940 RB_CLEAR_NODE(parent);
3941 spin_unlock(&root->inode_lock);
3945 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3946 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3947 spin_unlock(&root->inode_lock);
3950 static void inode_tree_del(struct inode *inode)
3952 struct btrfs_root *root = BTRFS_I(inode)->root;
3955 spin_lock(&root->inode_lock);
3956 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3957 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3958 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3959 empty = RB_EMPTY_ROOT(&root->inode_tree);
3961 spin_unlock(&root->inode_lock);
3964 * Free space cache has inodes in the tree root, but the tree root has a
3965 * root_refs of 0, so this could end up dropping the tree root as a
3966 * snapshot, so we need the extra !root->fs_info->tree_root check to
3967 * make sure we don't drop it.
3969 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3970 root != root->fs_info->tree_root) {
3971 synchronize_srcu(&root->fs_info->subvol_srcu);
3972 spin_lock(&root->inode_lock);
3973 empty = RB_EMPTY_ROOT(&root->inode_tree);
3974 spin_unlock(&root->inode_lock);
3976 btrfs_add_dead_root(root);
3980 int btrfs_invalidate_inodes(struct btrfs_root *root)
3982 struct rb_node *node;
3983 struct rb_node *prev;
3984 struct btrfs_inode *entry;
3985 struct inode *inode;
3988 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3990 spin_lock(&root->inode_lock);
3992 node = root->inode_tree.rb_node;
3996 entry = rb_entry(node, struct btrfs_inode, rb_node);
3998 if (objectid < entry->vfs_inode.i_ino)
3999 node = node->rb_left;
4000 else if (objectid > entry->vfs_inode.i_ino)
4001 node = node->rb_right;
4007 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4008 if (objectid <= entry->vfs_inode.i_ino) {
4012 prev = rb_next(prev);
4016 entry = rb_entry(node, struct btrfs_inode, rb_node);
4017 objectid = entry->vfs_inode.i_ino + 1;
4018 inode = igrab(&entry->vfs_inode);
4020 spin_unlock(&root->inode_lock);
4021 if (atomic_read(&inode->i_count) > 1)
4022 d_prune_aliases(inode);
4024 * btrfs_drop_inode will have it removed from
4025 * the inode cache when its usage count
4030 spin_lock(&root->inode_lock);
4034 if (cond_resched_lock(&root->inode_lock))
4037 node = rb_next(node);
4039 spin_unlock(&root->inode_lock);
4043 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4045 struct btrfs_iget_args *args = p;
4046 inode->i_ino = args->ino;
4047 BTRFS_I(inode)->root = args->root;
4048 btrfs_set_inode_space_info(args->root, inode);
4052 static int btrfs_find_actor(struct inode *inode, void *opaque)
4054 struct btrfs_iget_args *args = opaque;
4055 return args->ino == inode->i_ino &&
4056 args->root == BTRFS_I(inode)->root;
4059 static struct inode *btrfs_iget_locked(struct super_block *s,
4061 struct btrfs_root *root)
4063 struct inode *inode;
4064 struct btrfs_iget_args args;
4065 args.ino = objectid;
4068 inode = iget5_locked(s, objectid, btrfs_find_actor,
4069 btrfs_init_locked_inode,
4074 /* Get an inode object given its location and corresponding root.
4075 * Returns in *is_new if the inode was read from disk
4077 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4078 struct btrfs_root *root, int *new)
4080 struct inode *inode;
4082 inode = btrfs_iget_locked(s, location->objectid, root);
4084 return ERR_PTR(-ENOMEM);
4086 if (inode->i_state & I_NEW) {
4087 BTRFS_I(inode)->root = root;
4088 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4089 btrfs_read_locked_inode(inode);
4090 inode_tree_add(inode);
4091 unlock_new_inode(inode);
4099 static struct inode *new_simple_dir(struct super_block *s,
4100 struct btrfs_key *key,
4101 struct btrfs_root *root)
4103 struct inode *inode = new_inode(s);
4106 return ERR_PTR(-ENOMEM);
4108 BTRFS_I(inode)->root = root;
4109 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4110 BTRFS_I(inode)->dummy_inode = 1;
4112 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4113 inode->i_op = &simple_dir_inode_operations;
4114 inode->i_fop = &simple_dir_operations;
4115 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4116 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4121 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4123 struct inode *inode;
4124 struct btrfs_root *root = BTRFS_I(dir)->root;
4125 struct btrfs_root *sub_root = root;
4126 struct btrfs_key location;
4130 if (dentry->d_name.len > BTRFS_NAME_LEN)
4131 return ERR_PTR(-ENAMETOOLONG);
4133 ret = btrfs_inode_by_name(dir, dentry, &location);
4136 return ERR_PTR(ret);
4138 if (location.objectid == 0)
4141 if (location.type == BTRFS_INODE_ITEM_KEY) {
4142 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4146 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4148 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4149 ret = fixup_tree_root_location(root, dir, dentry,
4150 &location, &sub_root);
4153 inode = ERR_PTR(ret);
4155 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4157 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4159 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4161 if (!IS_ERR(inode) && root != sub_root) {
4162 down_read(&root->fs_info->cleanup_work_sem);
4163 if (!(inode->i_sb->s_flags & MS_RDONLY))
4164 ret = btrfs_orphan_cleanup(sub_root);
4165 up_read(&root->fs_info->cleanup_work_sem);
4167 inode = ERR_PTR(ret);
4173 static int btrfs_dentry_delete(const struct dentry *dentry)
4175 struct btrfs_root *root;
4177 if (!dentry->d_inode && !IS_ROOT(dentry))
4178 dentry = dentry->d_parent;
4180 if (dentry->d_inode) {
4181 root = BTRFS_I(dentry->d_inode)->root;
4182 if (btrfs_root_refs(&root->root_item) == 0)
4188 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4189 struct nameidata *nd)
4191 struct inode *inode;
4193 inode = btrfs_lookup_dentry(dir, dentry);
4195 return ERR_CAST(inode);
4197 return d_splice_alias(inode, dentry);
4200 static unsigned char btrfs_filetype_table[] = {
4201 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4204 static int btrfs_real_readdir(struct file *filp, void *dirent,
4207 struct inode *inode = filp->f_dentry->d_inode;
4208 struct btrfs_root *root = BTRFS_I(inode)->root;
4209 struct btrfs_item *item;
4210 struct btrfs_dir_item *di;
4211 struct btrfs_key key;
4212 struct btrfs_key found_key;
4213 struct btrfs_path *path;
4216 struct extent_buffer *leaf;
4219 unsigned char d_type;
4224 int key_type = BTRFS_DIR_INDEX_KEY;
4229 /* FIXME, use a real flag for deciding about the key type */
4230 if (root->fs_info->tree_root == root)
4231 key_type = BTRFS_DIR_ITEM_KEY;
4233 /* special case for "." */
4234 if (filp->f_pos == 0) {
4235 over = filldir(dirent, ".", 1,
4242 /* special case for .., just use the back ref */
4243 if (filp->f_pos == 1) {
4244 u64 pino = parent_ino(filp->f_path.dentry);
4245 over = filldir(dirent, "..", 2,
4251 path = btrfs_alloc_path();
4254 btrfs_set_key_type(&key, key_type);
4255 key.offset = filp->f_pos;
4256 key.objectid = inode->i_ino;
4258 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4264 leaf = path->nodes[0];
4265 nritems = btrfs_header_nritems(leaf);
4266 slot = path->slots[0];
4267 if (advance || slot >= nritems) {
4268 if (slot >= nritems - 1) {
4269 ret = btrfs_next_leaf(root, path);
4272 leaf = path->nodes[0];
4273 nritems = btrfs_header_nritems(leaf);
4274 slot = path->slots[0];
4282 item = btrfs_item_nr(leaf, slot);
4283 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4285 if (found_key.objectid != key.objectid)
4287 if (btrfs_key_type(&found_key) != key_type)
4289 if (found_key.offset < filp->f_pos)
4292 filp->f_pos = found_key.offset;
4294 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4296 di_total = btrfs_item_size(leaf, item);
4298 while (di_cur < di_total) {
4299 struct btrfs_key location;
4301 if (verify_dir_item(root, leaf, di))
4304 name_len = btrfs_dir_name_len(leaf, di);
4305 if (name_len <= sizeof(tmp_name)) {
4306 name_ptr = tmp_name;
4308 name_ptr = kmalloc(name_len, GFP_NOFS);
4314 read_extent_buffer(leaf, name_ptr,
4315 (unsigned long)(di + 1), name_len);
4317 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4318 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4320 /* is this a reference to our own snapshot? If so
4323 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4324 location.objectid == root->root_key.objectid) {
4328 over = filldir(dirent, name_ptr, name_len,
4329 found_key.offset, location.objectid,
4333 if (name_ptr != tmp_name)
4338 di_len = btrfs_dir_name_len(leaf, di) +
4339 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4341 di = (struct btrfs_dir_item *)((char *)di + di_len);
4345 /* Reached end of directory/root. Bump pos past the last item. */
4346 if (key_type == BTRFS_DIR_INDEX_KEY)
4348 * 32-bit glibc will use getdents64, but then strtol -
4349 * so the last number we can serve is this.
4351 filp->f_pos = 0x7fffffff;
4357 btrfs_free_path(path);
4361 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4363 struct btrfs_root *root = BTRFS_I(inode)->root;
4364 struct btrfs_trans_handle *trans;
4366 bool nolock = false;
4368 if (BTRFS_I(inode)->dummy_inode)
4372 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4374 if (wbc->sync_mode == WB_SYNC_ALL) {
4376 trans = btrfs_join_transaction_nolock(root, 1);
4378 trans = btrfs_join_transaction(root, 1);
4380 return PTR_ERR(trans);
4381 btrfs_set_trans_block_group(trans, inode);
4383 ret = btrfs_end_transaction_nolock(trans, root);
4385 ret = btrfs_commit_transaction(trans, root);
4391 * This is somewhat expensive, updating the tree every time the
4392 * inode changes. But, it is most likely to find the inode in cache.
4393 * FIXME, needs more benchmarking...there are no reasons other than performance
4394 * to keep or drop this code.
4396 void btrfs_dirty_inode(struct inode *inode)
4398 struct btrfs_root *root = BTRFS_I(inode)->root;
4399 struct btrfs_trans_handle *trans;
4402 if (BTRFS_I(inode)->dummy_inode)
4405 trans = btrfs_join_transaction(root, 1);
4406 BUG_ON(IS_ERR(trans));
4407 btrfs_set_trans_block_group(trans, inode);
4409 ret = btrfs_update_inode(trans, root, inode);
4410 if (ret && ret == -ENOSPC) {
4411 /* whoops, lets try again with the full transaction */
4412 btrfs_end_transaction(trans, root);
4413 trans = btrfs_start_transaction(root, 1);
4414 if (IS_ERR(trans)) {
4415 if (printk_ratelimit()) {
4416 printk(KERN_ERR "btrfs: fail to "
4417 "dirty inode %lu error %ld\n",
4418 inode->i_ino, PTR_ERR(trans));
4422 btrfs_set_trans_block_group(trans, inode);
4424 ret = btrfs_update_inode(trans, root, inode);
4426 if (printk_ratelimit()) {
4427 printk(KERN_ERR "btrfs: fail to "
4428 "dirty inode %lu error %d\n",
4433 btrfs_end_transaction(trans, root);
4437 * find the highest existing sequence number in a directory
4438 * and then set the in-memory index_cnt variable to reflect
4439 * free sequence numbers
4441 static int btrfs_set_inode_index_count(struct inode *inode)
4443 struct btrfs_root *root = BTRFS_I(inode)->root;
4444 struct btrfs_key key, found_key;
4445 struct btrfs_path *path;
4446 struct extent_buffer *leaf;
4449 key.objectid = inode->i_ino;
4450 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4451 key.offset = (u64)-1;
4453 path = btrfs_alloc_path();
4457 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4460 /* FIXME: we should be able to handle this */
4466 * MAGIC NUMBER EXPLANATION:
4467 * since we search a directory based on f_pos we have to start at 2
4468 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4469 * else has to start at 2
4471 if (path->slots[0] == 0) {
4472 BTRFS_I(inode)->index_cnt = 2;
4478 leaf = path->nodes[0];
4479 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4481 if (found_key.objectid != inode->i_ino ||
4482 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4483 BTRFS_I(inode)->index_cnt = 2;
4487 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4489 btrfs_free_path(path);
4494 * helper to find a free sequence number in a given directory. This current
4495 * code is very simple, later versions will do smarter things in the btree
4497 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4501 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4502 ret = btrfs_set_inode_index_count(dir);
4507 *index = BTRFS_I(dir)->index_cnt;
4508 BTRFS_I(dir)->index_cnt++;
4513 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4514 struct btrfs_root *root,
4516 const char *name, int name_len,
4517 u64 ref_objectid, u64 objectid,
4518 u64 alloc_hint, int mode, u64 *index)
4520 struct inode *inode;
4521 struct btrfs_inode_item *inode_item;
4522 struct btrfs_key *location;
4523 struct btrfs_path *path;
4524 struct btrfs_inode_ref *ref;
4525 struct btrfs_key key[2];
4531 path = btrfs_alloc_path();
4534 inode = new_inode(root->fs_info->sb);
4536 return ERR_PTR(-ENOMEM);
4539 trace_btrfs_inode_request(dir);
4541 ret = btrfs_set_inode_index(dir, index);
4544 return ERR_PTR(ret);
4548 * index_cnt is ignored for everything but a dir,
4549 * btrfs_get_inode_index_count has an explanation for the magic
4552 BTRFS_I(inode)->index_cnt = 2;
4553 BTRFS_I(inode)->root = root;
4554 BTRFS_I(inode)->generation = trans->transid;
4555 inode->i_generation = BTRFS_I(inode)->generation;
4556 btrfs_set_inode_space_info(root, inode);
4562 BTRFS_I(inode)->block_group =
4563 btrfs_find_block_group(root, 0, alloc_hint, owner);
4565 key[0].objectid = objectid;
4566 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4569 key[1].objectid = objectid;
4570 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4571 key[1].offset = ref_objectid;
4573 sizes[0] = sizeof(struct btrfs_inode_item);
4574 sizes[1] = name_len + sizeof(*ref);
4576 path->leave_spinning = 1;
4577 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4581 inode_init_owner(inode, dir, mode);
4582 inode->i_ino = objectid;
4583 inode_set_bytes(inode, 0);
4584 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4585 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4586 struct btrfs_inode_item);
4587 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4589 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4590 struct btrfs_inode_ref);
4591 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4592 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4593 ptr = (unsigned long)(ref + 1);
4594 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4596 btrfs_mark_buffer_dirty(path->nodes[0]);
4597 btrfs_free_path(path);
4599 location = &BTRFS_I(inode)->location;
4600 location->objectid = objectid;
4601 location->offset = 0;
4602 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4604 btrfs_inherit_iflags(inode, dir);
4606 if ((mode & S_IFREG)) {
4607 if (btrfs_test_opt(root, NODATASUM))
4608 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4609 if (btrfs_test_opt(root, NODATACOW) ||
4610 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4611 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4614 insert_inode_hash(inode);
4615 inode_tree_add(inode);
4617 trace_btrfs_inode_new(inode);
4622 BTRFS_I(dir)->index_cnt--;
4623 btrfs_free_path(path);
4625 return ERR_PTR(ret);
4628 static inline u8 btrfs_inode_type(struct inode *inode)
4630 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4634 * utility function to add 'inode' into 'parent_inode' with
4635 * a give name and a given sequence number.
4636 * if 'add_backref' is true, also insert a backref from the
4637 * inode to the parent directory.
4639 int btrfs_add_link(struct btrfs_trans_handle *trans,
4640 struct inode *parent_inode, struct inode *inode,
4641 const char *name, int name_len, int add_backref, u64 index)
4644 struct btrfs_key key;
4645 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4647 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4648 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4650 key.objectid = inode->i_ino;
4651 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4655 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4656 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4657 key.objectid, root->root_key.objectid,
4658 parent_inode->i_ino,
4659 index, name, name_len);
4660 } else if (add_backref) {
4661 ret = btrfs_insert_inode_ref(trans, root,
4662 name, name_len, inode->i_ino,
4663 parent_inode->i_ino, index);
4667 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4668 parent_inode->i_ino, &key,
4669 btrfs_inode_type(inode), index);
4672 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4674 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4675 ret = btrfs_update_inode(trans, root, parent_inode);
4680 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4681 struct inode *dir, struct dentry *dentry,
4682 struct inode *inode, int backref, u64 index)
4684 int err = btrfs_add_link(trans, dir, inode,
4685 dentry->d_name.name, dentry->d_name.len,
4688 d_instantiate(dentry, inode);
4696 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4697 int mode, dev_t rdev)
4699 struct btrfs_trans_handle *trans;
4700 struct btrfs_root *root = BTRFS_I(dir)->root;
4701 struct inode *inode = NULL;
4705 unsigned long nr = 0;
4708 if (!new_valid_dev(rdev))
4711 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4716 * 2 for inode item and ref
4718 * 1 for xattr if selinux is on
4720 trans = btrfs_start_transaction(root, 5);
4722 return PTR_ERR(trans);
4724 btrfs_set_trans_block_group(trans, dir);
4726 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4727 dentry->d_name.len, dir->i_ino, objectid,
4728 BTRFS_I(dir)->block_group, mode, &index);
4729 err = PTR_ERR(inode);
4733 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4739 btrfs_set_trans_block_group(trans, inode);
4740 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4744 inode->i_op = &btrfs_special_inode_operations;
4745 init_special_inode(inode, inode->i_mode, rdev);
4746 btrfs_update_inode(trans, root, inode);
4748 btrfs_update_inode_block_group(trans, inode);
4749 btrfs_update_inode_block_group(trans, dir);
4751 nr = trans->blocks_used;
4752 btrfs_end_transaction_throttle(trans, root);
4753 btrfs_btree_balance_dirty(root, nr);
4755 inode_dec_link_count(inode);
4761 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4762 int mode, struct nameidata *nd)
4764 struct btrfs_trans_handle *trans;
4765 struct btrfs_root *root = BTRFS_I(dir)->root;
4766 struct inode *inode = NULL;
4769 unsigned long nr = 0;
4773 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4777 * 2 for inode item and ref
4779 * 1 for xattr if selinux is on
4781 trans = btrfs_start_transaction(root, 5);
4783 return PTR_ERR(trans);
4785 btrfs_set_trans_block_group(trans, dir);
4787 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4788 dentry->d_name.len, dir->i_ino, objectid,
4789 BTRFS_I(dir)->block_group, mode, &index);
4790 err = PTR_ERR(inode);
4794 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4800 btrfs_set_trans_block_group(trans, inode);
4801 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4805 inode->i_mapping->a_ops = &btrfs_aops;
4806 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4807 inode->i_fop = &btrfs_file_operations;
4808 inode->i_op = &btrfs_file_inode_operations;
4809 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4811 btrfs_update_inode_block_group(trans, inode);
4812 btrfs_update_inode_block_group(trans, dir);
4814 nr = trans->blocks_used;
4815 btrfs_end_transaction_throttle(trans, root);
4817 inode_dec_link_count(inode);
4820 btrfs_btree_balance_dirty(root, nr);
4824 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4825 struct dentry *dentry)
4827 struct btrfs_trans_handle *trans;
4828 struct btrfs_root *root = BTRFS_I(dir)->root;
4829 struct inode *inode = old_dentry->d_inode;
4831 unsigned long nr = 0;
4835 /* do not allow sys_link's with other subvols of the same device */
4836 if (root->objectid != BTRFS_I(inode)->root->objectid)
4839 if (inode->i_nlink == ~0U)
4842 btrfs_inc_nlink(inode);
4843 inode->i_ctime = CURRENT_TIME;
4845 err = btrfs_set_inode_index(dir, &index);
4850 * 2 items for inode and inode ref
4851 * 2 items for dir items
4852 * 1 item for parent inode
4854 trans = btrfs_start_transaction(root, 5);
4855 if (IS_ERR(trans)) {
4856 err = PTR_ERR(trans);
4860 btrfs_set_trans_block_group(trans, dir);
4863 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4868 struct dentry *parent = dget_parent(dentry);
4869 btrfs_update_inode_block_group(trans, dir);
4870 err = btrfs_update_inode(trans, root, inode);
4872 btrfs_log_new_name(trans, inode, NULL, parent);
4876 nr = trans->blocks_used;
4877 btrfs_end_transaction_throttle(trans, root);
4880 inode_dec_link_count(inode);
4883 btrfs_btree_balance_dirty(root, nr);
4887 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4889 struct inode *inode = NULL;
4890 struct btrfs_trans_handle *trans;
4891 struct btrfs_root *root = BTRFS_I(dir)->root;
4893 int drop_on_err = 0;
4896 unsigned long nr = 1;
4898 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4903 * 2 items for inode and ref
4904 * 2 items for dir items
4905 * 1 for xattr if selinux is on
4907 trans = btrfs_start_transaction(root, 5);
4909 return PTR_ERR(trans);
4910 btrfs_set_trans_block_group(trans, dir);
4912 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4913 dentry->d_name.len, dir->i_ino, objectid,
4914 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4916 if (IS_ERR(inode)) {
4917 err = PTR_ERR(inode);
4923 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4927 inode->i_op = &btrfs_dir_inode_operations;
4928 inode->i_fop = &btrfs_dir_file_operations;
4929 btrfs_set_trans_block_group(trans, inode);
4931 btrfs_i_size_write(inode, 0);
4932 err = btrfs_update_inode(trans, root, inode);
4936 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4937 dentry->d_name.len, 0, index);
4941 d_instantiate(dentry, inode);
4943 btrfs_update_inode_block_group(trans, inode);
4944 btrfs_update_inode_block_group(trans, dir);
4947 nr = trans->blocks_used;
4948 btrfs_end_transaction_throttle(trans, root);
4951 btrfs_btree_balance_dirty(root, nr);
4955 /* helper for btfs_get_extent. Given an existing extent in the tree,
4956 * and an extent that you want to insert, deal with overlap and insert
4957 * the new extent into the tree.
4959 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4960 struct extent_map *existing,
4961 struct extent_map *em,
4962 u64 map_start, u64 map_len)
4966 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4967 start_diff = map_start - em->start;
4968 em->start = map_start;
4970 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4971 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4972 em->block_start += start_diff;
4973 em->block_len -= start_diff;
4975 return add_extent_mapping(em_tree, em);
4978 static noinline int uncompress_inline(struct btrfs_path *path,
4979 struct inode *inode, struct page *page,
4980 size_t pg_offset, u64 extent_offset,
4981 struct btrfs_file_extent_item *item)
4984 struct extent_buffer *leaf = path->nodes[0];
4987 unsigned long inline_size;
4991 WARN_ON(pg_offset != 0);
4992 compress_type = btrfs_file_extent_compression(leaf, item);
4993 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4994 inline_size = btrfs_file_extent_inline_item_len(leaf,
4995 btrfs_item_nr(leaf, path->slots[0]));
4996 tmp = kmalloc(inline_size, GFP_NOFS);
4997 ptr = btrfs_file_extent_inline_start(item);
4999 read_extent_buffer(leaf, tmp, ptr, inline_size);
5001 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5002 ret = btrfs_decompress(compress_type, tmp, page,
5003 extent_offset, inline_size, max_size);
5005 char *kaddr = kmap_atomic(page, KM_USER0);
5006 unsigned long copy_size = min_t(u64,
5007 PAGE_CACHE_SIZE - pg_offset,
5008 max_size - extent_offset);
5009 memset(kaddr + pg_offset, 0, copy_size);
5010 kunmap_atomic(kaddr, KM_USER0);
5017 * a bit scary, this does extent mapping from logical file offset to the disk.
5018 * the ugly parts come from merging extents from the disk with the in-ram
5019 * representation. This gets more complex because of the data=ordered code,
5020 * where the in-ram extents might be locked pending data=ordered completion.
5022 * This also copies inline extents directly into the page.
5025 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5026 size_t pg_offset, u64 start, u64 len,
5032 u64 extent_start = 0;
5034 u64 objectid = inode->i_ino;
5036 struct btrfs_path *path = NULL;
5037 struct btrfs_root *root = BTRFS_I(inode)->root;
5038 struct btrfs_file_extent_item *item;
5039 struct extent_buffer *leaf;
5040 struct btrfs_key found_key;
5041 struct extent_map *em = NULL;
5042 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5043 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5044 struct btrfs_trans_handle *trans = NULL;
5048 read_lock(&em_tree->lock);
5049 em = lookup_extent_mapping(em_tree, start, len);
5051 em->bdev = root->fs_info->fs_devices->latest_bdev;
5052 read_unlock(&em_tree->lock);
5055 if (em->start > start || em->start + em->len <= start)
5056 free_extent_map(em);
5057 else if (em->block_start == EXTENT_MAP_INLINE && page)
5058 free_extent_map(em);
5062 em = alloc_extent_map(GFP_NOFS);
5067 em->bdev = root->fs_info->fs_devices->latest_bdev;
5068 em->start = EXTENT_MAP_HOLE;
5069 em->orig_start = EXTENT_MAP_HOLE;
5071 em->block_len = (u64)-1;
5074 path = btrfs_alloc_path();
5078 ret = btrfs_lookup_file_extent(trans, root, path,
5079 objectid, start, trans != NULL);
5086 if (path->slots[0] == 0)
5091 leaf = path->nodes[0];
5092 item = btrfs_item_ptr(leaf, path->slots[0],
5093 struct btrfs_file_extent_item);
5094 /* are we inside the extent that was found? */
5095 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5096 found_type = btrfs_key_type(&found_key);
5097 if (found_key.objectid != objectid ||
5098 found_type != BTRFS_EXTENT_DATA_KEY) {
5102 found_type = btrfs_file_extent_type(leaf, item);
5103 extent_start = found_key.offset;
5104 compress_type = btrfs_file_extent_compression(leaf, item);
5105 if (found_type == BTRFS_FILE_EXTENT_REG ||
5106 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5107 extent_end = extent_start +
5108 btrfs_file_extent_num_bytes(leaf, item);
5109 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5111 size = btrfs_file_extent_inline_len(leaf, item);
5112 extent_end = (extent_start + size + root->sectorsize - 1) &
5113 ~((u64)root->sectorsize - 1);
5116 if (start >= extent_end) {
5118 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5119 ret = btrfs_next_leaf(root, path);
5126 leaf = path->nodes[0];
5128 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5129 if (found_key.objectid != objectid ||
5130 found_key.type != BTRFS_EXTENT_DATA_KEY)
5132 if (start + len <= found_key.offset)
5135 em->len = found_key.offset - start;
5139 if (found_type == BTRFS_FILE_EXTENT_REG ||
5140 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5141 em->start = extent_start;
5142 em->len = extent_end - extent_start;
5143 em->orig_start = extent_start -
5144 btrfs_file_extent_offset(leaf, item);
5145 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5147 em->block_start = EXTENT_MAP_HOLE;
5150 if (compress_type != BTRFS_COMPRESS_NONE) {
5151 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5152 em->compress_type = compress_type;
5153 em->block_start = bytenr;
5154 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5157 bytenr += btrfs_file_extent_offset(leaf, item);
5158 em->block_start = bytenr;
5159 em->block_len = em->len;
5160 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5161 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5164 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5168 size_t extent_offset;
5171 em->block_start = EXTENT_MAP_INLINE;
5172 if (!page || create) {
5173 em->start = extent_start;
5174 em->len = extent_end - extent_start;
5178 size = btrfs_file_extent_inline_len(leaf, item);
5179 extent_offset = page_offset(page) + pg_offset - extent_start;
5180 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5181 size - extent_offset);
5182 em->start = extent_start + extent_offset;
5183 em->len = (copy_size + root->sectorsize - 1) &
5184 ~((u64)root->sectorsize - 1);
5185 em->orig_start = EXTENT_MAP_INLINE;
5186 if (compress_type) {
5187 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5188 em->compress_type = compress_type;
5190 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5191 if (create == 0 && !PageUptodate(page)) {
5192 if (btrfs_file_extent_compression(leaf, item) !=
5193 BTRFS_COMPRESS_NONE) {
5194 ret = uncompress_inline(path, inode, page,
5196 extent_offset, item);
5200 read_extent_buffer(leaf, map + pg_offset, ptr,
5202 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5203 memset(map + pg_offset + copy_size, 0,
5204 PAGE_CACHE_SIZE - pg_offset -
5209 flush_dcache_page(page);
5210 } else if (create && PageUptodate(page)) {
5214 free_extent_map(em);
5216 btrfs_release_path(root, path);
5217 trans = btrfs_join_transaction(root, 1);
5219 return ERR_CAST(trans);
5223 write_extent_buffer(leaf, map + pg_offset, ptr,
5226 btrfs_mark_buffer_dirty(leaf);
5228 set_extent_uptodate(io_tree, em->start,
5229 extent_map_end(em) - 1, GFP_NOFS);
5232 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5239 em->block_start = EXTENT_MAP_HOLE;
5240 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5242 btrfs_release_path(root, path);
5243 if (em->start > start || extent_map_end(em) <= start) {
5244 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5245 "[%llu %llu]\n", (unsigned long long)em->start,
5246 (unsigned long long)em->len,
5247 (unsigned long long)start,
5248 (unsigned long long)len);
5254 write_lock(&em_tree->lock);
5255 ret = add_extent_mapping(em_tree, em);
5256 /* it is possible that someone inserted the extent into the tree
5257 * while we had the lock dropped. It is also possible that
5258 * an overlapping map exists in the tree
5260 if (ret == -EEXIST) {
5261 struct extent_map *existing;
5265 existing = lookup_extent_mapping(em_tree, start, len);
5266 if (existing && (existing->start > start ||
5267 existing->start + existing->len <= start)) {
5268 free_extent_map(existing);
5272 existing = lookup_extent_mapping(em_tree, em->start,
5275 err = merge_extent_mapping(em_tree, existing,
5278 free_extent_map(existing);
5280 free_extent_map(em);
5285 free_extent_map(em);
5289 free_extent_map(em);
5294 write_unlock(&em_tree->lock);
5297 trace_btrfs_get_extent(root, em);
5300 btrfs_free_path(path);
5302 ret = btrfs_end_transaction(trans, root);
5307 free_extent_map(em);
5308 return ERR_PTR(err);
5313 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5314 size_t pg_offset, u64 start, u64 len,
5317 struct extent_map *em;
5318 struct extent_map *hole_em = NULL;
5319 u64 range_start = start;
5325 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5330 * if our em maps to a hole, there might
5331 * actually be delalloc bytes behind it
5333 if (em->block_start != EXTENT_MAP_HOLE)
5339 /* check to see if we've wrapped (len == -1 or similar) */
5348 /* ok, we didn't find anything, lets look for delalloc */
5349 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5350 end, len, EXTENT_DELALLOC, 1);
5351 found_end = range_start + found;
5352 if (found_end < range_start)
5353 found_end = (u64)-1;
5356 * we didn't find anything useful, return
5357 * the original results from get_extent()
5359 if (range_start > end || found_end <= start) {
5365 /* adjust the range_start to make sure it doesn't
5366 * go backwards from the start they passed in
5368 range_start = max(start,range_start);
5369 found = found_end - range_start;
5372 u64 hole_start = start;
5375 em = alloc_extent_map(GFP_NOFS);
5381 * when btrfs_get_extent can't find anything it
5382 * returns one huge hole
5384 * make sure what it found really fits our range, and
5385 * adjust to make sure it is based on the start from
5389 u64 calc_end = extent_map_end(hole_em);
5391 if (calc_end <= start || (hole_em->start > end)) {
5392 free_extent_map(hole_em);
5395 hole_start = max(hole_em->start, start);
5396 hole_len = calc_end - hole_start;
5400 if (hole_em && range_start > hole_start) {
5401 /* our hole starts before our delalloc, so we
5402 * have to return just the parts of the hole
5403 * that go until the delalloc starts
5405 em->len = min(hole_len,
5406 range_start - hole_start);
5407 em->start = hole_start;
5408 em->orig_start = hole_start;
5410 * don't adjust block start at all,
5411 * it is fixed at EXTENT_MAP_HOLE
5413 em->block_start = hole_em->block_start;
5414 em->block_len = hole_len;
5416 em->start = range_start;
5418 em->orig_start = range_start;
5419 em->block_start = EXTENT_MAP_DELALLOC;
5420 em->block_len = found;
5422 } else if (hole_em) {
5427 free_extent_map(hole_em);
5429 free_extent_map(em);
5430 return ERR_PTR(err);
5435 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5438 struct btrfs_root *root = BTRFS_I(inode)->root;
5439 struct btrfs_trans_handle *trans;
5440 struct extent_map *em;
5441 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5442 struct btrfs_key ins;
5446 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5448 trans = btrfs_join_transaction(root, 0);
5450 return ERR_CAST(trans);
5452 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5454 alloc_hint = get_extent_allocation_hint(inode, start, len);
5455 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5456 alloc_hint, (u64)-1, &ins, 1);
5462 em = alloc_extent_map(GFP_NOFS);
5464 em = ERR_PTR(-ENOMEM);
5469 em->orig_start = em->start;
5470 em->len = ins.offset;
5472 em->block_start = ins.objectid;
5473 em->block_len = ins.offset;
5474 em->bdev = root->fs_info->fs_devices->latest_bdev;
5475 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5478 write_lock(&em_tree->lock);
5479 ret = add_extent_mapping(em_tree, em);
5480 write_unlock(&em_tree->lock);
5483 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5486 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5487 ins.offset, ins.offset, 0);
5489 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5493 btrfs_end_transaction(trans, root);
5498 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5499 * block must be cow'd
5501 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5502 struct inode *inode, u64 offset, u64 len)
5504 struct btrfs_path *path;
5506 struct extent_buffer *leaf;
5507 struct btrfs_root *root = BTRFS_I(inode)->root;
5508 struct btrfs_file_extent_item *fi;
5509 struct btrfs_key key;
5517 path = btrfs_alloc_path();
5521 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5526 slot = path->slots[0];
5529 /* can't find the item, must cow */
5536 leaf = path->nodes[0];
5537 btrfs_item_key_to_cpu(leaf, &key, slot);
5538 if (key.objectid != inode->i_ino ||
5539 key.type != BTRFS_EXTENT_DATA_KEY) {
5540 /* not our file or wrong item type, must cow */
5544 if (key.offset > offset) {
5545 /* Wrong offset, must cow */
5549 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5550 found_type = btrfs_file_extent_type(leaf, fi);
5551 if (found_type != BTRFS_FILE_EXTENT_REG &&
5552 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5553 /* not a regular extent, must cow */
5556 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5557 backref_offset = btrfs_file_extent_offset(leaf, fi);
5559 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5560 if (extent_end < offset + len) {
5561 /* extent doesn't include our full range, must cow */
5565 if (btrfs_extent_readonly(root, disk_bytenr))
5569 * look for other files referencing this extent, if we
5570 * find any we must cow
5572 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5573 key.offset - backref_offset, disk_bytenr))
5577 * adjust disk_bytenr and num_bytes to cover just the bytes
5578 * in this extent we are about to write. If there
5579 * are any csums in that range we have to cow in order
5580 * to keep the csums correct
5582 disk_bytenr += backref_offset;
5583 disk_bytenr += offset - key.offset;
5584 num_bytes = min(offset + len, extent_end) - offset;
5585 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5588 * all of the above have passed, it is safe to overwrite this extent
5593 btrfs_free_path(path);
5597 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5598 struct buffer_head *bh_result, int create)
5600 struct extent_map *em;
5601 struct btrfs_root *root = BTRFS_I(inode)->root;
5602 u64 start = iblock << inode->i_blkbits;
5603 u64 len = bh_result->b_size;
5604 struct btrfs_trans_handle *trans;
5606 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5611 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5612 * io. INLINE is special, and we could probably kludge it in here, but
5613 * it's still buffered so for safety lets just fall back to the generic
5616 * For COMPRESSED we _have_ to read the entire extent in so we can
5617 * decompress it, so there will be buffering required no matter what we
5618 * do, so go ahead and fallback to buffered.
5620 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5621 * to buffered IO. Don't blame me, this is the price we pay for using
5624 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5625 em->block_start == EXTENT_MAP_INLINE) {
5626 free_extent_map(em);
5630 /* Just a good old fashioned hole, return */
5631 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5632 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5633 free_extent_map(em);
5634 /* DIO will do one hole at a time, so just unlock a sector */
5635 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5636 start + root->sectorsize - 1, GFP_NOFS);
5641 * We don't allocate a new extent in the following cases
5643 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5645 * 2) The extent is marked as PREALLOC. We're good to go here and can
5646 * just use the extent.
5650 len = em->len - (start - em->start);
5654 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5655 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5656 em->block_start != EXTENT_MAP_HOLE)) {
5661 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5662 type = BTRFS_ORDERED_PREALLOC;
5664 type = BTRFS_ORDERED_NOCOW;
5665 len = min(len, em->len - (start - em->start));
5666 block_start = em->block_start + (start - em->start);
5669 * we're not going to log anything, but we do need
5670 * to make sure the current transaction stays open
5671 * while we look for nocow cross refs
5673 trans = btrfs_join_transaction(root, 0);
5677 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5678 ret = btrfs_add_ordered_extent_dio(inode, start,
5679 block_start, len, len, type);
5680 btrfs_end_transaction(trans, root);
5682 free_extent_map(em);
5687 btrfs_end_transaction(trans, root);
5691 * this will cow the extent, reset the len in case we changed
5694 len = bh_result->b_size;
5695 free_extent_map(em);
5696 em = btrfs_new_extent_direct(inode, start, len);
5699 len = min(len, em->len - (start - em->start));
5701 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5702 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5705 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5707 bh_result->b_size = len;
5708 bh_result->b_bdev = em->bdev;
5709 set_buffer_mapped(bh_result);
5710 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5711 set_buffer_new(bh_result);
5713 free_extent_map(em);
5718 struct btrfs_dio_private {
5719 struct inode *inode;
5726 /* number of bios pending for this dio */
5727 atomic_t pending_bios;
5732 struct bio *orig_bio;
5735 static void btrfs_endio_direct_read(struct bio *bio, int err)
5737 struct btrfs_dio_private *dip = bio->bi_private;
5738 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5739 struct bio_vec *bvec = bio->bi_io_vec;
5740 struct inode *inode = dip->inode;
5741 struct btrfs_root *root = BTRFS_I(inode)->root;
5743 u32 *private = dip->csums;
5745 start = dip->logical_offset;
5747 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5748 struct page *page = bvec->bv_page;
5751 unsigned long flags;
5753 local_irq_save(flags);
5754 kaddr = kmap_atomic(page, KM_IRQ0);
5755 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5756 csum, bvec->bv_len);
5757 btrfs_csum_final(csum, (char *)&csum);
5758 kunmap_atomic(kaddr, KM_IRQ0);
5759 local_irq_restore(flags);
5761 flush_dcache_page(bvec->bv_page);
5762 if (csum != *private) {
5763 printk(KERN_ERR "btrfs csum failed ino %lu off"
5764 " %llu csum %u private %u\n",
5765 inode->i_ino, (unsigned long long)start,
5771 start += bvec->bv_len;
5774 } while (bvec <= bvec_end);
5776 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5777 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5778 bio->bi_private = dip->private;
5783 /* If we had a csum failure make sure to clear the uptodate flag */
5785 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5786 dio_end_io(bio, err);
5789 static void btrfs_endio_direct_write(struct bio *bio, int err)
5791 struct btrfs_dio_private *dip = bio->bi_private;
5792 struct inode *inode = dip->inode;
5793 struct btrfs_root *root = BTRFS_I(inode)->root;
5794 struct btrfs_trans_handle *trans;
5795 struct btrfs_ordered_extent *ordered = NULL;
5796 struct extent_state *cached_state = NULL;
5797 u64 ordered_offset = dip->logical_offset;
5798 u64 ordered_bytes = dip->bytes;
5804 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5812 trans = btrfs_join_transaction(root, 1);
5813 if (IS_ERR(trans)) {
5817 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5819 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5820 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5822 ret = btrfs_update_inode(trans, root, inode);
5827 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5828 ordered->file_offset + ordered->len - 1, 0,
5829 &cached_state, GFP_NOFS);
5831 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5832 ret = btrfs_mark_extent_written(trans, inode,
5833 ordered->file_offset,
5834 ordered->file_offset +
5841 ret = insert_reserved_file_extent(trans, inode,
5842 ordered->file_offset,
5848 BTRFS_FILE_EXTENT_REG);
5849 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5850 ordered->file_offset, ordered->len);
5858 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5859 btrfs_ordered_update_i_size(inode, 0, ordered);
5860 btrfs_update_inode(trans, root, inode);
5862 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5863 ordered->file_offset + ordered->len - 1,
5864 &cached_state, GFP_NOFS);
5866 btrfs_delalloc_release_metadata(inode, ordered->len);
5867 btrfs_end_transaction(trans, root);
5868 ordered_offset = ordered->file_offset + ordered->len;
5869 btrfs_put_ordered_extent(ordered);
5870 btrfs_put_ordered_extent(ordered);
5874 * our bio might span multiple ordered extents. If we haven't
5875 * completed the accounting for the whole dio, go back and try again
5877 if (ordered_offset < dip->logical_offset + dip->bytes) {
5878 ordered_bytes = dip->logical_offset + dip->bytes -
5883 bio->bi_private = dip->private;
5888 /* If we had an error make sure to clear the uptodate flag */
5890 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5891 dio_end_io(bio, err);
5894 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5895 struct bio *bio, int mirror_num,
5896 unsigned long bio_flags, u64 offset)
5899 struct btrfs_root *root = BTRFS_I(inode)->root;
5900 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5905 static void btrfs_end_dio_bio(struct bio *bio, int err)
5907 struct btrfs_dio_private *dip = bio->bi_private;
5910 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5911 "sector %#Lx len %u err no %d\n",
5912 dip->inode->i_ino, bio->bi_rw,
5913 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5917 * before atomic variable goto zero, we must make sure
5918 * dip->errors is perceived to be set.
5920 smp_mb__before_atomic_dec();
5923 /* if there are more bios still pending for this dio, just exit */
5924 if (!atomic_dec_and_test(&dip->pending_bios))
5928 bio_io_error(dip->orig_bio);
5930 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5931 bio_endio(dip->orig_bio, 0);
5937 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5938 u64 first_sector, gfp_t gfp_flags)
5940 int nr_vecs = bio_get_nr_vecs(bdev);
5941 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5944 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5945 int rw, u64 file_offset, int skip_sum,
5948 int write = rw & REQ_WRITE;
5949 struct btrfs_root *root = BTRFS_I(inode)->root;
5953 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5957 if (write && !skip_sum) {
5958 ret = btrfs_wq_submit_bio(root->fs_info,
5959 inode, rw, bio, 0, 0,
5961 __btrfs_submit_bio_start_direct_io,
5962 __btrfs_submit_bio_done);
5964 } else if (!skip_sum) {
5965 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5966 file_offset, csums);
5971 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5977 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5980 struct inode *inode = dip->inode;
5981 struct btrfs_root *root = BTRFS_I(inode)->root;
5982 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5984 struct bio *orig_bio = dip->orig_bio;
5985 struct bio_vec *bvec = orig_bio->bi_io_vec;
5986 u64 start_sector = orig_bio->bi_sector;
5987 u64 file_offset = dip->logical_offset;
5991 u32 *csums = dip->csums;
5993 int write = rw & REQ_WRITE;
5995 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5998 bio->bi_private = dip;
5999 bio->bi_end_io = btrfs_end_dio_bio;
6000 atomic_inc(&dip->pending_bios);
6002 map_length = orig_bio->bi_size;
6003 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6004 &map_length, NULL, 0);
6010 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6011 if (unlikely(map_length < submit_len + bvec->bv_len ||
6012 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6013 bvec->bv_offset) < bvec->bv_len)) {
6015 * inc the count before we submit the bio so
6016 * we know the end IO handler won't happen before
6017 * we inc the count. Otherwise, the dip might get freed
6018 * before we're done setting it up
6020 atomic_inc(&dip->pending_bios);
6021 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6022 file_offset, skip_sum,
6026 atomic_dec(&dip->pending_bios);
6030 /* Write's use the ordered csums */
6031 if (!write && !skip_sum)
6032 csums = csums + nr_pages;
6033 start_sector += submit_len >> 9;
6034 file_offset += submit_len;
6039 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6040 start_sector, GFP_NOFS);
6043 bio->bi_private = dip;
6044 bio->bi_end_io = btrfs_end_dio_bio;
6046 map_length = orig_bio->bi_size;
6047 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6048 &map_length, NULL, 0);
6054 submit_len += bvec->bv_len;
6060 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6069 * before atomic variable goto zero, we must
6070 * make sure dip->errors is perceived to be set.
6072 smp_mb__before_atomic_dec();
6073 if (atomic_dec_and_test(&dip->pending_bios))
6074 bio_io_error(dip->orig_bio);
6076 /* bio_end_io() will handle error, so we needn't return it */
6080 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6083 struct btrfs_root *root = BTRFS_I(inode)->root;
6084 struct btrfs_dio_private *dip;
6085 struct bio_vec *bvec = bio->bi_io_vec;
6087 int write = rw & REQ_WRITE;
6090 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6092 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6099 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6100 if (!write && !skip_sum) {
6101 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6109 dip->private = bio->bi_private;
6111 dip->logical_offset = file_offset;
6115 dip->bytes += bvec->bv_len;
6117 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6119 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6120 bio->bi_private = dip;
6122 dip->orig_bio = bio;
6123 atomic_set(&dip->pending_bios, 0);
6126 bio->bi_end_io = btrfs_endio_direct_write;
6128 bio->bi_end_io = btrfs_endio_direct_read;
6130 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6135 * If this is a write, we need to clean up the reserved space and kill
6136 * the ordered extent.
6139 struct btrfs_ordered_extent *ordered;
6140 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6141 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6142 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6143 btrfs_free_reserved_extent(root, ordered->start,
6145 btrfs_put_ordered_extent(ordered);
6146 btrfs_put_ordered_extent(ordered);
6148 bio_endio(bio, ret);
6151 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6152 const struct iovec *iov, loff_t offset,
6153 unsigned long nr_segs)
6158 unsigned blocksize_mask = root->sectorsize - 1;
6159 ssize_t retval = -EINVAL;
6160 loff_t end = offset;
6162 if (offset & blocksize_mask)
6165 /* Check the memory alignment. Blocks cannot straddle pages */
6166 for (seg = 0; seg < nr_segs; seg++) {
6167 addr = (unsigned long)iov[seg].iov_base;
6168 size = iov[seg].iov_len;
6170 if ((addr & blocksize_mask) || (size & blocksize_mask))
6177 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6178 const struct iovec *iov, loff_t offset,
6179 unsigned long nr_segs)
6181 struct file *file = iocb->ki_filp;
6182 struct inode *inode = file->f_mapping->host;
6183 struct btrfs_ordered_extent *ordered;
6184 struct extent_state *cached_state = NULL;
6185 u64 lockstart, lockend;
6187 int writing = rw & WRITE;
6189 size_t count = iov_length(iov, nr_segs);
6191 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6197 lockend = offset + count - 1;
6200 ret = btrfs_delalloc_reserve_space(inode, count);
6206 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6207 0, &cached_state, GFP_NOFS);
6209 * We're concerned with the entire range that we're going to be
6210 * doing DIO to, so we need to make sure theres no ordered
6211 * extents in this range.
6213 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6214 lockend - lockstart + 1);
6217 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6218 &cached_state, GFP_NOFS);
6219 btrfs_start_ordered_extent(inode, ordered, 1);
6220 btrfs_put_ordered_extent(ordered);
6225 * we don't use btrfs_set_extent_delalloc because we don't want
6226 * the dirty or uptodate bits
6229 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6230 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6231 EXTENT_DELALLOC, 0, NULL, &cached_state,
6234 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6235 lockend, EXTENT_LOCKED | write_bits,
6236 1, 0, &cached_state, GFP_NOFS);
6241 free_extent_state(cached_state);
6242 cached_state = NULL;
6244 ret = __blockdev_direct_IO(rw, iocb, inode,
6245 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6246 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6247 btrfs_submit_direct, 0);
6249 if (ret < 0 && ret != -EIOCBQUEUED) {
6250 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6251 offset + iov_length(iov, nr_segs) - 1,
6252 EXTENT_LOCKED | write_bits, 1, 0,
6253 &cached_state, GFP_NOFS);
6254 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6256 * We're falling back to buffered, unlock the section we didn't
6259 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6260 offset + iov_length(iov, nr_segs) - 1,
6261 EXTENT_LOCKED | write_bits, 1, 0,
6262 &cached_state, GFP_NOFS);
6265 free_extent_state(cached_state);
6269 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6270 __u64 start, __u64 len)
6272 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6275 int btrfs_readpage(struct file *file, struct page *page)
6277 struct extent_io_tree *tree;
6278 tree = &BTRFS_I(page->mapping->host)->io_tree;
6279 return extent_read_full_page(tree, page, btrfs_get_extent);
6282 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6284 struct extent_io_tree *tree;
6287 if (current->flags & PF_MEMALLOC) {
6288 redirty_page_for_writepage(wbc, page);
6292 tree = &BTRFS_I(page->mapping->host)->io_tree;
6293 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6296 int btrfs_writepages(struct address_space *mapping,
6297 struct writeback_control *wbc)
6299 struct extent_io_tree *tree;
6301 tree = &BTRFS_I(mapping->host)->io_tree;
6302 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6306 btrfs_readpages(struct file *file, struct address_space *mapping,
6307 struct list_head *pages, unsigned nr_pages)
6309 struct extent_io_tree *tree;
6310 tree = &BTRFS_I(mapping->host)->io_tree;
6311 return extent_readpages(tree, mapping, pages, nr_pages,
6314 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6316 struct extent_io_tree *tree;
6317 struct extent_map_tree *map;
6320 tree = &BTRFS_I(page->mapping->host)->io_tree;
6321 map = &BTRFS_I(page->mapping->host)->extent_tree;
6322 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6324 ClearPagePrivate(page);
6325 set_page_private(page, 0);
6326 page_cache_release(page);
6331 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6333 if (PageWriteback(page) || PageDirty(page))
6335 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6338 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6340 struct extent_io_tree *tree;
6341 struct btrfs_ordered_extent *ordered;
6342 struct extent_state *cached_state = NULL;
6343 u64 page_start = page_offset(page);
6344 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6348 * we have the page locked, so new writeback can't start,
6349 * and the dirty bit won't be cleared while we are here.
6351 * Wait for IO on this page so that we can safely clear
6352 * the PagePrivate2 bit and do ordered accounting
6354 wait_on_page_writeback(page);
6356 tree = &BTRFS_I(page->mapping->host)->io_tree;
6358 btrfs_releasepage(page, GFP_NOFS);
6361 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6363 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6367 * IO on this page will never be started, so we need
6368 * to account for any ordered extents now
6370 clear_extent_bit(tree, page_start, page_end,
6371 EXTENT_DIRTY | EXTENT_DELALLOC |
6372 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6373 &cached_state, GFP_NOFS);
6375 * whoever cleared the private bit is responsible
6376 * for the finish_ordered_io
6378 if (TestClearPagePrivate2(page)) {
6379 btrfs_finish_ordered_io(page->mapping->host,
6380 page_start, page_end);
6382 btrfs_put_ordered_extent(ordered);
6383 cached_state = NULL;
6384 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6387 clear_extent_bit(tree, page_start, page_end,
6388 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6389 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6390 __btrfs_releasepage(page, GFP_NOFS);
6392 ClearPageChecked(page);
6393 if (PagePrivate(page)) {
6394 ClearPagePrivate(page);
6395 set_page_private(page, 0);
6396 page_cache_release(page);
6401 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6402 * called from a page fault handler when a page is first dirtied. Hence we must
6403 * be careful to check for EOF conditions here. We set the page up correctly
6404 * for a written page which means we get ENOSPC checking when writing into
6405 * holes and correct delalloc and unwritten extent mapping on filesystems that
6406 * support these features.
6408 * We are not allowed to take the i_mutex here so we have to play games to
6409 * protect against truncate races as the page could now be beyond EOF. Because
6410 * vmtruncate() writes the inode size before removing pages, once we have the
6411 * page lock we can determine safely if the page is beyond EOF. If it is not
6412 * beyond EOF, then the page is guaranteed safe against truncation until we
6415 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6417 struct page *page = vmf->page;
6418 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6419 struct btrfs_root *root = BTRFS_I(inode)->root;
6420 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6421 struct btrfs_ordered_extent *ordered;
6422 struct extent_state *cached_state = NULL;
6424 unsigned long zero_start;
6430 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6434 else /* -ENOSPC, -EIO, etc */
6435 ret = VM_FAULT_SIGBUS;
6439 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6442 size = i_size_read(inode);
6443 page_start = page_offset(page);
6444 page_end = page_start + PAGE_CACHE_SIZE - 1;
6446 if ((page->mapping != inode->i_mapping) ||
6447 (page_start >= size)) {
6448 /* page got truncated out from underneath us */
6451 wait_on_page_writeback(page);
6453 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6455 set_page_extent_mapped(page);
6458 * we can't set the delalloc bits if there are pending ordered
6459 * extents. Drop our locks and wait for them to finish
6461 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6463 unlock_extent_cached(io_tree, page_start, page_end,
6464 &cached_state, GFP_NOFS);
6466 btrfs_start_ordered_extent(inode, ordered, 1);
6467 btrfs_put_ordered_extent(ordered);
6472 * XXX - page_mkwrite gets called every time the page is dirtied, even
6473 * if it was already dirty, so for space accounting reasons we need to
6474 * clear any delalloc bits for the range we are fixing to save. There
6475 * is probably a better way to do this, but for now keep consistent with
6476 * prepare_pages in the normal write path.
6478 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6479 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6480 0, 0, &cached_state, GFP_NOFS);
6482 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6485 unlock_extent_cached(io_tree, page_start, page_end,
6486 &cached_state, GFP_NOFS);
6487 ret = VM_FAULT_SIGBUS;
6492 /* page is wholly or partially inside EOF */
6493 if (page_start + PAGE_CACHE_SIZE > size)
6494 zero_start = size & ~PAGE_CACHE_MASK;
6496 zero_start = PAGE_CACHE_SIZE;
6498 if (zero_start != PAGE_CACHE_SIZE) {
6500 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6501 flush_dcache_page(page);
6504 ClearPageChecked(page);
6505 set_page_dirty(page);
6506 SetPageUptodate(page);
6508 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6509 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6511 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6515 return VM_FAULT_LOCKED;
6517 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6522 static int btrfs_truncate(struct inode *inode)
6524 struct btrfs_root *root = BTRFS_I(inode)->root;
6527 struct btrfs_trans_handle *trans;
6529 u64 mask = root->sectorsize - 1;
6531 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6535 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6536 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6538 trans = btrfs_start_transaction(root, 5);
6540 return PTR_ERR(trans);
6542 btrfs_set_trans_block_group(trans, inode);
6544 ret = btrfs_orphan_add(trans, inode);
6546 btrfs_end_transaction(trans, root);
6550 nr = trans->blocks_used;
6551 btrfs_end_transaction(trans, root);
6552 btrfs_btree_balance_dirty(root, nr);
6554 /* Now start a transaction for the truncate */
6555 trans = btrfs_start_transaction(root, 0);
6557 return PTR_ERR(trans);
6558 btrfs_set_trans_block_group(trans, inode);
6559 trans->block_rsv = root->orphan_block_rsv;
6562 * setattr is responsible for setting the ordered_data_close flag,
6563 * but that is only tested during the last file release. That
6564 * could happen well after the next commit, leaving a great big
6565 * window where new writes may get lost if someone chooses to write
6566 * to this file after truncating to zero
6568 * The inode doesn't have any dirty data here, and so if we commit
6569 * this is a noop. If someone immediately starts writing to the inode
6570 * it is very likely we'll catch some of their writes in this
6571 * transaction, and the commit will find this file on the ordered
6572 * data list with good things to send down.
6574 * This is a best effort solution, there is still a window where
6575 * using truncate to replace the contents of the file will
6576 * end up with a zero length file after a crash.
6578 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6579 btrfs_add_ordered_operation(trans, root, inode);
6583 trans = btrfs_start_transaction(root, 0);
6585 return PTR_ERR(trans);
6586 btrfs_set_trans_block_group(trans, inode);
6587 trans->block_rsv = root->orphan_block_rsv;
6590 ret = btrfs_block_rsv_check(trans, root,
6591 root->orphan_block_rsv, 0, 5);
6592 if (ret == -EAGAIN) {
6593 ret = btrfs_commit_transaction(trans, root);
6603 ret = btrfs_truncate_inode_items(trans, root, inode,
6605 BTRFS_EXTENT_DATA_KEY);
6606 if (ret != -EAGAIN) {
6611 ret = btrfs_update_inode(trans, root, inode);
6617 nr = trans->blocks_used;
6618 btrfs_end_transaction(trans, root);
6620 btrfs_btree_balance_dirty(root, nr);
6623 if (ret == 0 && inode->i_nlink > 0) {
6624 ret = btrfs_orphan_del(trans, inode);
6627 } else if (ret && inode->i_nlink > 0) {
6629 * Failed to do the truncate, remove us from the in memory
6632 ret = btrfs_orphan_del(NULL, inode);
6635 ret = btrfs_update_inode(trans, root, inode);
6639 nr = trans->blocks_used;
6640 ret = btrfs_end_transaction_throttle(trans, root);
6643 btrfs_btree_balance_dirty(root, nr);
6649 * create a new subvolume directory/inode (helper for the ioctl).
6651 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6652 struct btrfs_root *new_root,
6653 u64 new_dirid, u64 alloc_hint)
6655 struct inode *inode;
6659 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6660 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6662 return PTR_ERR(inode);
6663 inode->i_op = &btrfs_dir_inode_operations;
6664 inode->i_fop = &btrfs_dir_file_operations;
6667 btrfs_i_size_write(inode, 0);
6669 err = btrfs_update_inode(trans, new_root, inode);
6676 /* helper function for file defrag and space balancing. This
6677 * forces readahead on a given range of bytes in an inode
6679 unsigned long btrfs_force_ra(struct address_space *mapping,
6680 struct file_ra_state *ra, struct file *file,
6681 pgoff_t offset, pgoff_t last_index)
6683 pgoff_t req_size = last_index - offset + 1;
6685 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6686 return offset + req_size;
6689 struct inode *btrfs_alloc_inode(struct super_block *sb)
6691 struct btrfs_inode *ei;
6692 struct inode *inode;
6694 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6699 ei->space_info = NULL;
6703 ei->last_sub_trans = 0;
6704 ei->logged_trans = 0;
6705 ei->delalloc_bytes = 0;
6706 ei->reserved_bytes = 0;
6707 ei->disk_i_size = 0;
6709 ei->index_cnt = (u64)-1;
6710 ei->last_unlink_trans = 0;
6712 atomic_set(&ei->outstanding_extents, 0);
6713 atomic_set(&ei->reserved_extents, 0);
6715 ei->ordered_data_close = 0;
6716 ei->orphan_meta_reserved = 0;
6717 ei->dummy_inode = 0;
6718 ei->force_compress = BTRFS_COMPRESS_NONE;
6720 inode = &ei->vfs_inode;
6721 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6722 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6723 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6724 mutex_init(&ei->log_mutex);
6725 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6726 INIT_LIST_HEAD(&ei->i_orphan);
6727 INIT_LIST_HEAD(&ei->delalloc_inodes);
6728 INIT_LIST_HEAD(&ei->ordered_operations);
6729 RB_CLEAR_NODE(&ei->rb_node);
6734 static void btrfs_i_callback(struct rcu_head *head)
6736 struct inode *inode = container_of(head, struct inode, i_rcu);
6737 INIT_LIST_HEAD(&inode->i_dentry);
6738 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6741 void btrfs_destroy_inode(struct inode *inode)
6743 struct btrfs_ordered_extent *ordered;
6744 struct btrfs_root *root = BTRFS_I(inode)->root;
6746 WARN_ON(!list_empty(&inode->i_dentry));
6747 WARN_ON(inode->i_data.nrpages);
6748 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6749 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6752 * This can happen where we create an inode, but somebody else also
6753 * created the same inode and we need to destroy the one we already
6760 * Make sure we're properly removed from the ordered operation
6764 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6765 spin_lock(&root->fs_info->ordered_extent_lock);
6766 list_del_init(&BTRFS_I(inode)->ordered_operations);
6767 spin_unlock(&root->fs_info->ordered_extent_lock);
6770 if (root == root->fs_info->tree_root) {
6771 struct btrfs_block_group_cache *block_group;
6773 block_group = btrfs_lookup_block_group(root->fs_info,
6774 BTRFS_I(inode)->block_group);
6775 if (block_group && block_group->inode == inode) {
6776 spin_lock(&block_group->lock);
6777 block_group->inode = NULL;
6778 spin_unlock(&block_group->lock);
6779 btrfs_put_block_group(block_group);
6780 } else if (block_group) {
6781 btrfs_put_block_group(block_group);
6785 spin_lock(&root->orphan_lock);
6786 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6787 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6789 list_del_init(&BTRFS_I(inode)->i_orphan);
6791 spin_unlock(&root->orphan_lock);
6794 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6798 printk(KERN_ERR "btrfs found ordered "
6799 "extent %llu %llu on inode cleanup\n",
6800 (unsigned long long)ordered->file_offset,
6801 (unsigned long long)ordered->len);
6802 btrfs_remove_ordered_extent(inode, ordered);
6803 btrfs_put_ordered_extent(ordered);
6804 btrfs_put_ordered_extent(ordered);
6807 inode_tree_del(inode);
6808 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6810 call_rcu(&inode->i_rcu, btrfs_i_callback);
6813 int btrfs_drop_inode(struct inode *inode)
6815 struct btrfs_root *root = BTRFS_I(inode)->root;
6817 if (btrfs_root_refs(&root->root_item) == 0 &&
6818 root != root->fs_info->tree_root)
6821 return generic_drop_inode(inode);
6824 static void init_once(void *foo)
6826 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6828 inode_init_once(&ei->vfs_inode);
6831 void btrfs_destroy_cachep(void)
6833 if (btrfs_inode_cachep)
6834 kmem_cache_destroy(btrfs_inode_cachep);
6835 if (btrfs_trans_handle_cachep)
6836 kmem_cache_destroy(btrfs_trans_handle_cachep);
6837 if (btrfs_transaction_cachep)
6838 kmem_cache_destroy(btrfs_transaction_cachep);
6839 if (btrfs_path_cachep)
6840 kmem_cache_destroy(btrfs_path_cachep);
6841 if (btrfs_free_space_cachep)
6842 kmem_cache_destroy(btrfs_free_space_cachep);
6845 int btrfs_init_cachep(void)
6847 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6848 sizeof(struct btrfs_inode), 0,
6849 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6850 if (!btrfs_inode_cachep)
6853 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6854 sizeof(struct btrfs_trans_handle), 0,
6855 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6856 if (!btrfs_trans_handle_cachep)
6859 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6860 sizeof(struct btrfs_transaction), 0,
6861 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6862 if (!btrfs_transaction_cachep)
6865 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6866 sizeof(struct btrfs_path), 0,
6867 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6868 if (!btrfs_path_cachep)
6871 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6872 sizeof(struct btrfs_free_space), 0,
6873 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6874 if (!btrfs_free_space_cachep)
6879 btrfs_destroy_cachep();
6883 static int btrfs_getattr(struct vfsmount *mnt,
6884 struct dentry *dentry, struct kstat *stat)
6886 struct inode *inode = dentry->d_inode;
6887 generic_fillattr(inode, stat);
6888 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6889 stat->blksize = PAGE_CACHE_SIZE;
6890 stat->blocks = (inode_get_bytes(inode) +
6891 BTRFS_I(inode)->delalloc_bytes) >> 9;
6896 * If a file is moved, it will inherit the cow and compression flags of the new
6899 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6901 struct btrfs_inode *b_dir = BTRFS_I(dir);
6902 struct btrfs_inode *b_inode = BTRFS_I(inode);
6904 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6905 b_inode->flags |= BTRFS_INODE_NODATACOW;
6907 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6909 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6910 b_inode->flags |= BTRFS_INODE_COMPRESS;
6912 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6915 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6916 struct inode *new_dir, struct dentry *new_dentry)
6918 struct btrfs_trans_handle *trans;
6919 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6920 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6921 struct inode *new_inode = new_dentry->d_inode;
6922 struct inode *old_inode = old_dentry->d_inode;
6923 struct timespec ctime = CURRENT_TIME;
6928 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6931 /* we only allow rename subvolume link between subvolumes */
6932 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6935 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6936 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6939 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6940 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6943 * we're using rename to replace one file with another.
6944 * and the replacement file is large. Start IO on it now so
6945 * we don't add too much work to the end of the transaction
6947 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6948 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6949 filemap_flush(old_inode->i_mapping);
6951 /* close the racy window with snapshot create/destroy ioctl */
6952 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6953 down_read(&root->fs_info->subvol_sem);
6955 * We want to reserve the absolute worst case amount of items. So if
6956 * both inodes are subvols and we need to unlink them then that would
6957 * require 4 item modifications, but if they are both normal inodes it
6958 * would require 5 item modifications, so we'll assume their normal
6959 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6960 * should cover the worst case number of items we'll modify.
6962 trans = btrfs_start_transaction(root, 20);
6964 return PTR_ERR(trans);
6966 btrfs_set_trans_block_group(trans, new_dir);
6969 btrfs_record_root_in_trans(trans, dest);
6971 ret = btrfs_set_inode_index(new_dir, &index);
6975 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6976 /* force full log commit if subvolume involved. */
6977 root->fs_info->last_trans_log_full_commit = trans->transid;
6979 ret = btrfs_insert_inode_ref(trans, dest,
6980 new_dentry->d_name.name,
6981 new_dentry->d_name.len,
6983 new_dir->i_ino, index);
6987 * this is an ugly little race, but the rename is required
6988 * to make sure that if we crash, the inode is either at the
6989 * old name or the new one. pinning the log transaction lets
6990 * us make sure we don't allow a log commit to come in after
6991 * we unlink the name but before we add the new name back in.
6993 btrfs_pin_log_trans(root);
6996 * make sure the inode gets flushed if it is replacing
6999 if (new_inode && new_inode->i_size &&
7000 old_inode && S_ISREG(old_inode->i_mode)) {
7001 btrfs_add_ordered_operation(trans, root, old_inode);
7004 old_dir->i_ctime = old_dir->i_mtime = ctime;
7005 new_dir->i_ctime = new_dir->i_mtime = ctime;
7006 old_inode->i_ctime = ctime;
7008 if (old_dentry->d_parent != new_dentry->d_parent)
7009 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7011 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7012 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7013 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7014 old_dentry->d_name.name,
7015 old_dentry->d_name.len);
7017 ret = __btrfs_unlink_inode(trans, root, old_dir,
7018 old_dentry->d_inode,
7019 old_dentry->d_name.name,
7020 old_dentry->d_name.len);
7022 ret = btrfs_update_inode(trans, root, old_inode);
7027 new_inode->i_ctime = CURRENT_TIME;
7028 if (unlikely(new_inode->i_ino ==
7029 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7030 root_objectid = BTRFS_I(new_inode)->location.objectid;
7031 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7033 new_dentry->d_name.name,
7034 new_dentry->d_name.len);
7035 BUG_ON(new_inode->i_nlink == 0);
7037 ret = btrfs_unlink_inode(trans, dest, new_dir,
7038 new_dentry->d_inode,
7039 new_dentry->d_name.name,
7040 new_dentry->d_name.len);
7043 if (new_inode->i_nlink == 0) {
7044 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7049 fixup_inode_flags(new_dir, old_inode);
7051 ret = btrfs_add_link(trans, new_dir, old_inode,
7052 new_dentry->d_name.name,
7053 new_dentry->d_name.len, 0, index);
7056 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7057 struct dentry *parent = dget_parent(new_dentry);
7058 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7060 btrfs_end_log_trans(root);
7063 btrfs_end_transaction_throttle(trans, root);
7065 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7066 up_read(&root->fs_info->subvol_sem);
7072 * some fairly slow code that needs optimization. This walks the list
7073 * of all the inodes with pending delalloc and forces them to disk.
7075 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7077 struct list_head *head = &root->fs_info->delalloc_inodes;
7078 struct btrfs_inode *binode;
7079 struct inode *inode;
7081 if (root->fs_info->sb->s_flags & MS_RDONLY)
7084 spin_lock(&root->fs_info->delalloc_lock);
7085 while (!list_empty(head)) {
7086 binode = list_entry(head->next, struct btrfs_inode,
7088 inode = igrab(&binode->vfs_inode);
7090 list_del_init(&binode->delalloc_inodes);
7091 spin_unlock(&root->fs_info->delalloc_lock);
7093 filemap_flush(inode->i_mapping);
7095 btrfs_add_delayed_iput(inode);
7100 spin_lock(&root->fs_info->delalloc_lock);
7102 spin_unlock(&root->fs_info->delalloc_lock);
7104 /* the filemap_flush will queue IO into the worker threads, but
7105 * we have to make sure the IO is actually started and that
7106 * ordered extents get created before we return
7108 atomic_inc(&root->fs_info->async_submit_draining);
7109 while (atomic_read(&root->fs_info->nr_async_submits) ||
7110 atomic_read(&root->fs_info->async_delalloc_pages)) {
7111 wait_event(root->fs_info->async_submit_wait,
7112 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7113 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7115 atomic_dec(&root->fs_info->async_submit_draining);
7119 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7122 struct btrfs_inode *binode;
7123 struct inode *inode = NULL;
7125 spin_lock(&root->fs_info->delalloc_lock);
7126 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7127 binode = list_entry(root->fs_info->delalloc_inodes.next,
7128 struct btrfs_inode, delalloc_inodes);
7129 inode = igrab(&binode->vfs_inode);
7131 list_move_tail(&binode->delalloc_inodes,
7132 &root->fs_info->delalloc_inodes);
7136 list_del_init(&binode->delalloc_inodes);
7137 cond_resched_lock(&root->fs_info->delalloc_lock);
7139 spin_unlock(&root->fs_info->delalloc_lock);
7143 filemap_write_and_wait(inode->i_mapping);
7145 * We have to do this because compression doesn't
7146 * actually set PG_writeback until it submits the pages
7147 * for IO, which happens in an async thread, so we could
7148 * race and not actually wait for any writeback pages
7149 * because they've not been submitted yet. Technically
7150 * this could still be the case for the ordered stuff
7151 * since the async thread may not have started to do its
7152 * work yet. If this becomes the case then we need to
7153 * figure out a way to make sure that in writepage we
7154 * wait for any async pages to be submitted before
7155 * returning so that fdatawait does what its supposed to
7158 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7160 filemap_flush(inode->i_mapping);
7163 btrfs_add_delayed_iput(inode);
7171 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7172 const char *symname)
7174 struct btrfs_trans_handle *trans;
7175 struct btrfs_root *root = BTRFS_I(dir)->root;
7176 struct btrfs_path *path;
7177 struct btrfs_key key;
7178 struct inode *inode = NULL;
7186 struct btrfs_file_extent_item *ei;
7187 struct extent_buffer *leaf;
7188 unsigned long nr = 0;
7190 name_len = strlen(symname) + 1;
7191 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7192 return -ENAMETOOLONG;
7194 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7198 * 2 items for inode item and ref
7199 * 2 items for dir items
7200 * 1 item for xattr if selinux is on
7202 trans = btrfs_start_transaction(root, 5);
7204 return PTR_ERR(trans);
7206 btrfs_set_trans_block_group(trans, dir);
7208 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7209 dentry->d_name.len, dir->i_ino, objectid,
7210 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7212 err = PTR_ERR(inode);
7216 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7222 btrfs_set_trans_block_group(trans, inode);
7223 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7227 inode->i_mapping->a_ops = &btrfs_aops;
7228 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7229 inode->i_fop = &btrfs_file_operations;
7230 inode->i_op = &btrfs_file_inode_operations;
7231 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7233 btrfs_update_inode_block_group(trans, inode);
7234 btrfs_update_inode_block_group(trans, dir);
7238 path = btrfs_alloc_path();
7240 key.objectid = inode->i_ino;
7242 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7243 datasize = btrfs_file_extent_calc_inline_size(name_len);
7244 err = btrfs_insert_empty_item(trans, root, path, &key,
7250 leaf = path->nodes[0];
7251 ei = btrfs_item_ptr(leaf, path->slots[0],
7252 struct btrfs_file_extent_item);
7253 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7254 btrfs_set_file_extent_type(leaf, ei,
7255 BTRFS_FILE_EXTENT_INLINE);
7256 btrfs_set_file_extent_encryption(leaf, ei, 0);
7257 btrfs_set_file_extent_compression(leaf, ei, 0);
7258 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7259 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7261 ptr = btrfs_file_extent_inline_start(ei);
7262 write_extent_buffer(leaf, symname, ptr, name_len);
7263 btrfs_mark_buffer_dirty(leaf);
7264 btrfs_free_path(path);
7266 inode->i_op = &btrfs_symlink_inode_operations;
7267 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7268 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7269 inode_set_bytes(inode, name_len);
7270 btrfs_i_size_write(inode, name_len - 1);
7271 err = btrfs_update_inode(trans, root, inode);
7276 nr = trans->blocks_used;
7277 btrfs_end_transaction_throttle(trans, root);
7279 inode_dec_link_count(inode);
7282 btrfs_btree_balance_dirty(root, nr);
7286 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7287 u64 start, u64 num_bytes, u64 min_size,
7288 loff_t actual_len, u64 *alloc_hint,
7289 struct btrfs_trans_handle *trans)
7291 struct btrfs_root *root = BTRFS_I(inode)->root;
7292 struct btrfs_key ins;
7293 u64 cur_offset = start;
7296 bool own_trans = true;
7300 while (num_bytes > 0) {
7302 trans = btrfs_start_transaction(root, 3);
7303 if (IS_ERR(trans)) {
7304 ret = PTR_ERR(trans);
7309 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7310 0, *alloc_hint, (u64)-1, &ins, 1);
7313 btrfs_end_transaction(trans, root);
7317 ret = insert_reserved_file_extent(trans, inode,
7318 cur_offset, ins.objectid,
7319 ins.offset, ins.offset,
7320 ins.offset, 0, 0, 0,
7321 BTRFS_FILE_EXTENT_PREALLOC);
7323 btrfs_drop_extent_cache(inode, cur_offset,
7324 cur_offset + ins.offset -1, 0);
7326 num_bytes -= ins.offset;
7327 cur_offset += ins.offset;
7328 *alloc_hint = ins.objectid + ins.offset;
7330 inode->i_ctime = CURRENT_TIME;
7331 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7332 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7333 (actual_len > inode->i_size) &&
7334 (cur_offset > inode->i_size)) {
7335 if (cur_offset > actual_len)
7336 i_size = actual_len;
7338 i_size = cur_offset;
7339 i_size_write(inode, i_size);
7340 btrfs_ordered_update_i_size(inode, i_size, NULL);
7343 ret = btrfs_update_inode(trans, root, inode);
7347 btrfs_end_transaction(trans, root);
7352 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7353 u64 start, u64 num_bytes, u64 min_size,
7354 loff_t actual_len, u64 *alloc_hint)
7356 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7357 min_size, actual_len, alloc_hint,
7361 int btrfs_prealloc_file_range_trans(struct inode *inode,
7362 struct btrfs_trans_handle *trans, int mode,
7363 u64 start, u64 num_bytes, u64 min_size,
7364 loff_t actual_len, u64 *alloc_hint)
7366 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7367 min_size, actual_len, alloc_hint, trans);
7370 static int btrfs_set_page_dirty(struct page *page)
7372 return __set_page_dirty_nobuffers(page);
7375 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7377 struct btrfs_root *root = BTRFS_I(inode)->root;
7379 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7381 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7383 return generic_permission(inode, mask, flags, btrfs_check_acl);
7386 static const struct inode_operations btrfs_dir_inode_operations = {
7387 .getattr = btrfs_getattr,
7388 .lookup = btrfs_lookup,
7389 .create = btrfs_create,
7390 .unlink = btrfs_unlink,
7392 .mkdir = btrfs_mkdir,
7393 .rmdir = btrfs_rmdir,
7394 .rename = btrfs_rename,
7395 .symlink = btrfs_symlink,
7396 .setattr = btrfs_setattr,
7397 .mknod = btrfs_mknod,
7398 .setxattr = btrfs_setxattr,
7399 .getxattr = btrfs_getxattr,
7400 .listxattr = btrfs_listxattr,
7401 .removexattr = btrfs_removexattr,
7402 .permission = btrfs_permission,
7404 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7405 .lookup = btrfs_lookup,
7406 .permission = btrfs_permission,
7409 static const struct file_operations btrfs_dir_file_operations = {
7410 .llseek = generic_file_llseek,
7411 .read = generic_read_dir,
7412 .readdir = btrfs_real_readdir,
7413 .unlocked_ioctl = btrfs_ioctl,
7414 #ifdef CONFIG_COMPAT
7415 .compat_ioctl = btrfs_ioctl,
7417 .release = btrfs_release_file,
7418 .fsync = btrfs_sync_file,
7421 static struct extent_io_ops btrfs_extent_io_ops = {
7422 .fill_delalloc = run_delalloc_range,
7423 .submit_bio_hook = btrfs_submit_bio_hook,
7424 .merge_bio_hook = btrfs_merge_bio_hook,
7425 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7426 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7427 .writepage_start_hook = btrfs_writepage_start_hook,
7428 .readpage_io_failed_hook = btrfs_io_failed_hook,
7429 .set_bit_hook = btrfs_set_bit_hook,
7430 .clear_bit_hook = btrfs_clear_bit_hook,
7431 .merge_extent_hook = btrfs_merge_extent_hook,
7432 .split_extent_hook = btrfs_split_extent_hook,
7436 * btrfs doesn't support the bmap operation because swapfiles
7437 * use bmap to make a mapping of extents in the file. They assume
7438 * these extents won't change over the life of the file and they
7439 * use the bmap result to do IO directly to the drive.
7441 * the btrfs bmap call would return logical addresses that aren't
7442 * suitable for IO and they also will change frequently as COW
7443 * operations happen. So, swapfile + btrfs == corruption.
7445 * For now we're avoiding this by dropping bmap.
7447 static const struct address_space_operations btrfs_aops = {
7448 .readpage = btrfs_readpage,
7449 .writepage = btrfs_writepage,
7450 .writepages = btrfs_writepages,
7451 .readpages = btrfs_readpages,
7452 .direct_IO = btrfs_direct_IO,
7453 .invalidatepage = btrfs_invalidatepage,
7454 .releasepage = btrfs_releasepage,
7455 .set_page_dirty = btrfs_set_page_dirty,
7456 .error_remove_page = generic_error_remove_page,
7459 static const struct address_space_operations btrfs_symlink_aops = {
7460 .readpage = btrfs_readpage,
7461 .writepage = btrfs_writepage,
7462 .invalidatepage = btrfs_invalidatepage,
7463 .releasepage = btrfs_releasepage,
7466 static const struct inode_operations btrfs_file_inode_operations = {
7467 .getattr = btrfs_getattr,
7468 .setattr = btrfs_setattr,
7469 .setxattr = btrfs_setxattr,
7470 .getxattr = btrfs_getxattr,
7471 .listxattr = btrfs_listxattr,
7472 .removexattr = btrfs_removexattr,
7473 .permission = btrfs_permission,
7474 .fiemap = btrfs_fiemap,
7476 static const struct inode_operations btrfs_special_inode_operations = {
7477 .getattr = btrfs_getattr,
7478 .setattr = btrfs_setattr,
7479 .permission = btrfs_permission,
7480 .setxattr = btrfs_setxattr,
7481 .getxattr = btrfs_getxattr,
7482 .listxattr = btrfs_listxattr,
7483 .removexattr = btrfs_removexattr,
7485 static const struct inode_operations btrfs_symlink_inode_operations = {
7486 .readlink = generic_readlink,
7487 .follow_link = page_follow_link_light,
7488 .put_link = page_put_link,
7489 .getattr = btrfs_getattr,
7490 .permission = btrfs_permission,
7491 .setxattr = btrfs_setxattr,
7492 .getxattr = btrfs_getxattr,
7493 .listxattr = btrfs_listxattr,
7494 .removexattr = btrfs_removexattr,
7497 const struct dentry_operations btrfs_dentry_operations = {
7498 .d_delete = btrfs_dentry_delete,