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)
100 err = btrfs_init_acl(trans, inode, dir);
102 err = btrfs_xattr_security_init(trans, inode, dir);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
112 struct btrfs_root *root, struct inode *inode,
113 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;
130 if (compressed_size && compressed_pages)
131 cur_size = compressed_size;
133 path = btrfs_alloc_path();
137 path->leave_spinning = 1;
138 btrfs_set_trans_block_group(trans, inode);
140 key.objectid = inode->i_ino;
142 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
143 datasize = btrfs_file_extent_calc_inline_size(cur_size);
145 inode_add_bytes(inode, size);
146 ret = btrfs_insert_empty_item(trans, root, path, &key,
153 leaf = path->nodes[0];
154 ei = btrfs_item_ptr(leaf, path->slots[0],
155 struct btrfs_file_extent_item);
156 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158 btrfs_set_file_extent_encryption(leaf, ei, 0);
159 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161 ptr = btrfs_file_extent_inline_start(ei);
163 if (compress_type != BTRFS_COMPRESS_NONE) {
166 while (compressed_size > 0) {
167 cpage = compressed_pages[i];
168 cur_size = min_t(unsigned long, compressed_size,
171 kaddr = kmap_atomic(cpage, KM_USER0);
172 write_extent_buffer(leaf, kaddr, ptr, cur_size);
173 kunmap_atomic(kaddr, KM_USER0);
177 compressed_size -= cur_size;
179 btrfs_set_file_extent_compression(leaf, ei,
182 page = find_get_page(inode->i_mapping,
183 start >> PAGE_CACHE_SHIFT);
184 btrfs_set_file_extent_compression(leaf, ei, 0);
185 kaddr = kmap_atomic(page, KM_USER0);
186 offset = start & (PAGE_CACHE_SIZE - 1);
187 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188 kunmap_atomic(kaddr, KM_USER0);
189 page_cache_release(page);
191 btrfs_mark_buffer_dirty(leaf);
192 btrfs_free_path(path);
195 * we're an inline extent, so nobody can
196 * extend the file past i_size without locking
197 * a page we already have locked.
199 * We must do any isize and inode updates
200 * before we unlock the pages. Otherwise we
201 * could end up racing with unlink.
203 BTRFS_I(inode)->disk_i_size = inode->i_size;
204 btrfs_update_inode(trans, root, inode);
208 btrfs_free_path(path);
214 * conditionally insert an inline extent into the file. This
215 * does the checks required to make sure the data is small enough
216 * to fit as an inline extent.
218 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
219 struct btrfs_root *root,
220 struct inode *inode, u64 start, u64 end,
221 size_t compressed_size, int compress_type,
222 struct page **compressed_pages)
224 u64 isize = i_size_read(inode);
225 u64 actual_end = min(end + 1, isize);
226 u64 inline_len = actual_end - start;
227 u64 aligned_end = (end + root->sectorsize - 1) &
228 ~((u64)root->sectorsize - 1);
230 u64 data_len = inline_len;
234 data_len = compressed_size;
237 actual_end >= PAGE_CACHE_SIZE ||
238 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
240 (actual_end & (root->sectorsize - 1)) == 0) ||
242 data_len > root->fs_info->max_inline) {
246 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
250 if (isize > actual_end)
251 inline_len = min_t(u64, isize, actual_end);
252 ret = insert_inline_extent(trans, root, inode, start,
253 inline_len, compressed_size,
254 compress_type, compressed_pages);
256 btrfs_delalloc_release_metadata(inode, end + 1 - start);
257 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
261 struct async_extent {
266 unsigned long nr_pages;
268 struct list_head list;
273 struct btrfs_root *root;
274 struct page *locked_page;
277 struct list_head extents;
278 struct btrfs_work work;
281 static noinline int add_async_extent(struct async_cow *cow,
282 u64 start, u64 ram_size,
285 unsigned long nr_pages,
288 struct async_extent *async_extent;
290 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
291 BUG_ON(!async_extent);
292 async_extent->start = start;
293 async_extent->ram_size = ram_size;
294 async_extent->compressed_size = compressed_size;
295 async_extent->pages = pages;
296 async_extent->nr_pages = nr_pages;
297 async_extent->compress_type = compress_type;
298 list_add_tail(&async_extent->list, &cow->extents);
303 * we create compressed extents in two phases. The first
304 * phase compresses a range of pages that have already been
305 * locked (both pages and state bits are locked).
307 * This is done inside an ordered work queue, and the compression
308 * is spread across many cpus. The actual IO submission is step
309 * two, and the ordered work queue takes care of making sure that
310 * happens in the same order things were put onto the queue by
311 * writepages and friends.
313 * If this code finds it can't get good compression, it puts an
314 * entry onto the work queue to write the uncompressed bytes. This
315 * makes sure that both compressed inodes and uncompressed inodes
316 * are written in the same order that pdflush sent them down.
318 static noinline int compress_file_range(struct inode *inode,
319 struct page *locked_page,
321 struct async_cow *async_cow,
324 struct btrfs_root *root = BTRFS_I(inode)->root;
325 struct btrfs_trans_handle *trans;
327 u64 blocksize = root->sectorsize;
329 u64 isize = i_size_read(inode);
331 struct page **pages = NULL;
332 unsigned long nr_pages;
333 unsigned long nr_pages_ret = 0;
334 unsigned long total_compressed = 0;
335 unsigned long total_in = 0;
336 unsigned long max_compressed = 128 * 1024;
337 unsigned long max_uncompressed = 128 * 1024;
340 int compress_type = root->fs_info->compress_type;
342 actual_end = min_t(u64, isize, end + 1);
345 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
346 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
349 * we don't want to send crud past the end of i_size through
350 * compression, that's just a waste of CPU time. So, if the
351 * end of the file is before the start of our current
352 * requested range of bytes, we bail out to the uncompressed
353 * cleanup code that can deal with all of this.
355 * It isn't really the fastest way to fix things, but this is a
356 * very uncommon corner.
358 if (actual_end <= start)
359 goto cleanup_and_bail_uncompressed;
361 total_compressed = actual_end - start;
363 /* we want to make sure that amount of ram required to uncompress
364 * an extent is reasonable, so we limit the total size in ram
365 * of a compressed extent to 128k. This is a crucial number
366 * because it also controls how easily we can spread reads across
367 * cpus for decompression.
369 * We also want to make sure the amount of IO required to do
370 * a random read is reasonably small, so we limit the size of
371 * a compressed extent to 128k.
373 total_compressed = min(total_compressed, max_uncompressed);
374 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
375 num_bytes = max(blocksize, num_bytes);
380 * we do compression for mount -o compress and when the
381 * inode has not been flagged as nocompress. This flag can
382 * change at any time if we discover bad compression ratios.
384 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
385 (btrfs_test_opt(root, COMPRESS) ||
386 (BTRFS_I(inode)->force_compress) ||
387 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
389 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
392 if (BTRFS_I(inode)->force_compress)
393 compress_type = BTRFS_I(inode)->force_compress;
395 ret = btrfs_compress_pages(compress_type,
396 inode->i_mapping, start,
397 total_compressed, pages,
398 nr_pages, &nr_pages_ret,
404 unsigned long offset = total_compressed &
405 (PAGE_CACHE_SIZE - 1);
406 struct page *page = pages[nr_pages_ret - 1];
409 /* zero the tail end of the last page, we might be
410 * sending it down to disk
413 kaddr = kmap_atomic(page, KM_USER0);
414 memset(kaddr + offset, 0,
415 PAGE_CACHE_SIZE - offset);
416 kunmap_atomic(kaddr, KM_USER0);
422 trans = btrfs_join_transaction(root, 1);
423 BUG_ON(IS_ERR(trans));
424 btrfs_set_trans_block_group(trans, inode);
425 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
427 /* lets try to make an inline extent */
428 if (ret || total_in < (actual_end - start)) {
429 /* we didn't compress the entire range, try
430 * to make an uncompressed inline extent.
432 ret = cow_file_range_inline(trans, root, inode,
433 start, end, 0, 0, NULL);
435 /* try making a compressed inline extent */
436 ret = cow_file_range_inline(trans, root, inode,
439 compress_type, pages);
443 * inline extent creation worked, we don't need
444 * to create any more async work items. Unlock
445 * and free up our temp pages.
447 extent_clear_unlock_delalloc(inode,
448 &BTRFS_I(inode)->io_tree,
450 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
451 EXTENT_CLEAR_DELALLOC |
452 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
454 btrfs_end_transaction(trans, root);
457 btrfs_end_transaction(trans, root);
462 * we aren't doing an inline extent round the compressed size
463 * up to a block size boundary so the allocator does sane
466 total_compressed = (total_compressed + blocksize - 1) &
470 * one last check to make sure the compression is really a
471 * win, compare the page count read with the blocks on disk
473 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
474 ~(PAGE_CACHE_SIZE - 1);
475 if (total_compressed >= total_in) {
478 num_bytes = total_in;
481 if (!will_compress && pages) {
483 * the compression code ran but failed to make things smaller,
484 * free any pages it allocated and our page pointer array
486 for (i = 0; i < nr_pages_ret; i++) {
487 WARN_ON(pages[i]->mapping);
488 page_cache_release(pages[i]);
492 total_compressed = 0;
495 /* flag the file so we don't compress in the future */
496 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
497 !(BTRFS_I(inode)->force_compress)) {
498 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
504 /* the async work queues will take care of doing actual
505 * allocation on disk for these compressed pages,
506 * and will submit them to the elevator.
508 add_async_extent(async_cow, start, num_bytes,
509 total_compressed, pages, nr_pages_ret,
512 if (start + num_bytes < end) {
519 cleanup_and_bail_uncompressed:
521 * No compression, but we still need to write the pages in
522 * the file we've been given so far. redirty the locked
523 * page if it corresponds to our extent and set things up
524 * for the async work queue to run cow_file_range to do
525 * the normal delalloc dance
527 if (page_offset(locked_page) >= start &&
528 page_offset(locked_page) <= end) {
529 __set_page_dirty_nobuffers(locked_page);
530 /* unlocked later on in the async handlers */
532 add_async_extent(async_cow, start, end - start + 1,
533 0, NULL, 0, BTRFS_COMPRESS_NONE);
541 for (i = 0; i < nr_pages_ret; i++) {
542 WARN_ON(pages[i]->mapping);
543 page_cache_release(pages[i]);
551 * phase two of compressed writeback. This is the ordered portion
552 * of the code, which only gets called in the order the work was
553 * queued. We walk all the async extents created by compress_file_range
554 * and send them down to the disk.
556 static noinline int submit_compressed_extents(struct inode *inode,
557 struct async_cow *async_cow)
559 struct async_extent *async_extent;
561 struct btrfs_trans_handle *trans;
562 struct btrfs_key ins;
563 struct extent_map *em;
564 struct btrfs_root *root = BTRFS_I(inode)->root;
565 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
566 struct extent_io_tree *io_tree;
569 if (list_empty(&async_cow->extents))
573 while (!list_empty(&async_cow->extents)) {
574 async_extent = list_entry(async_cow->extents.next,
575 struct async_extent, list);
576 list_del(&async_extent->list);
578 io_tree = &BTRFS_I(inode)->io_tree;
581 /* did the compression code fall back to uncompressed IO? */
582 if (!async_extent->pages) {
583 int page_started = 0;
584 unsigned long nr_written = 0;
586 lock_extent(io_tree, async_extent->start,
587 async_extent->start +
588 async_extent->ram_size - 1, GFP_NOFS);
590 /* allocate blocks */
591 ret = cow_file_range(inode, async_cow->locked_page,
593 async_extent->start +
594 async_extent->ram_size - 1,
595 &page_started, &nr_written, 0);
598 * if page_started, cow_file_range inserted an
599 * inline extent and took care of all the unlocking
600 * and IO for us. Otherwise, we need to submit
601 * all those pages down to the drive.
603 if (!page_started && !ret)
604 extent_write_locked_range(io_tree,
605 inode, async_extent->start,
606 async_extent->start +
607 async_extent->ram_size - 1,
615 lock_extent(io_tree, async_extent->start,
616 async_extent->start + async_extent->ram_size - 1,
619 trans = btrfs_join_transaction(root, 1);
620 BUG_ON(IS_ERR(trans));
621 ret = btrfs_reserve_extent(trans, root,
622 async_extent->compressed_size,
623 async_extent->compressed_size,
626 btrfs_end_transaction(trans, root);
630 for (i = 0; i < async_extent->nr_pages; i++) {
631 WARN_ON(async_extent->pages[i]->mapping);
632 page_cache_release(async_extent->pages[i]);
634 kfree(async_extent->pages);
635 async_extent->nr_pages = 0;
636 async_extent->pages = NULL;
637 unlock_extent(io_tree, async_extent->start,
638 async_extent->start +
639 async_extent->ram_size - 1, GFP_NOFS);
644 * here we're doing allocation and writeback of the
647 btrfs_drop_extent_cache(inode, async_extent->start,
648 async_extent->start +
649 async_extent->ram_size - 1, 0);
651 em = alloc_extent_map(GFP_NOFS);
653 em->start = async_extent->start;
654 em->len = async_extent->ram_size;
655 em->orig_start = em->start;
657 em->block_start = ins.objectid;
658 em->block_len = ins.offset;
659 em->bdev = root->fs_info->fs_devices->latest_bdev;
660 em->compress_type = async_extent->compress_type;
661 set_bit(EXTENT_FLAG_PINNED, &em->flags);
662 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
665 write_lock(&em_tree->lock);
666 ret = add_extent_mapping(em_tree, em);
667 write_unlock(&em_tree->lock);
668 if (ret != -EEXIST) {
672 btrfs_drop_extent_cache(inode, async_extent->start,
673 async_extent->start +
674 async_extent->ram_size - 1, 0);
677 ret = btrfs_add_ordered_extent_compress(inode,
680 async_extent->ram_size,
682 BTRFS_ORDERED_COMPRESSED,
683 async_extent->compress_type);
687 * clear dirty, set writeback and unlock the pages.
689 extent_clear_unlock_delalloc(inode,
690 &BTRFS_I(inode)->io_tree,
692 async_extent->start +
693 async_extent->ram_size - 1,
694 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
695 EXTENT_CLEAR_UNLOCK |
696 EXTENT_CLEAR_DELALLOC |
697 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
699 ret = btrfs_submit_compressed_write(inode,
701 async_extent->ram_size,
703 ins.offset, async_extent->pages,
704 async_extent->nr_pages);
707 alloc_hint = ins.objectid + ins.offset;
715 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
718 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
719 struct extent_map *em;
722 read_lock(&em_tree->lock);
723 em = search_extent_mapping(em_tree, start, num_bytes);
726 * if block start isn't an actual block number then find the
727 * first block in this inode and use that as a hint. If that
728 * block is also bogus then just don't worry about it.
730 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
732 em = search_extent_mapping(em_tree, 0, 0);
733 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
734 alloc_hint = em->block_start;
738 alloc_hint = em->block_start;
742 read_unlock(&em_tree->lock);
748 * when extent_io.c finds a delayed allocation range in the file,
749 * the call backs end up in this code. The basic idea is to
750 * allocate extents on disk for the range, and create ordered data structs
751 * in ram to track those extents.
753 * locked_page is the page that writepage had locked already. We use
754 * it to make sure we don't do extra locks or unlocks.
756 * *page_started is set to one if we unlock locked_page and do everything
757 * required to start IO on it. It may be clean and already done with
760 static noinline int cow_file_range(struct inode *inode,
761 struct page *locked_page,
762 u64 start, u64 end, int *page_started,
763 unsigned long *nr_written,
766 struct btrfs_root *root = BTRFS_I(inode)->root;
767 struct btrfs_trans_handle *trans;
770 unsigned long ram_size;
773 u64 blocksize = root->sectorsize;
774 struct btrfs_key ins;
775 struct extent_map *em;
776 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
779 BUG_ON(root == root->fs_info->tree_root);
780 trans = btrfs_join_transaction(root, 1);
781 BUG_ON(IS_ERR(trans));
782 btrfs_set_trans_block_group(trans, inode);
783 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
785 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
786 num_bytes = max(blocksize, num_bytes);
787 disk_num_bytes = num_bytes;
791 /* lets try to make an inline extent */
792 ret = cow_file_range_inline(trans, root, inode,
793 start, end, 0, 0, NULL);
795 extent_clear_unlock_delalloc(inode,
796 &BTRFS_I(inode)->io_tree,
798 EXTENT_CLEAR_UNLOCK_PAGE |
799 EXTENT_CLEAR_UNLOCK |
800 EXTENT_CLEAR_DELALLOC |
802 EXTENT_SET_WRITEBACK |
803 EXTENT_END_WRITEBACK);
805 *nr_written = *nr_written +
806 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
813 BUG_ON(disk_num_bytes >
814 btrfs_super_total_bytes(&root->fs_info->super_copy));
816 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
817 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
819 while (disk_num_bytes > 0) {
822 cur_alloc_size = disk_num_bytes;
823 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
824 root->sectorsize, 0, alloc_hint,
828 em = alloc_extent_map(GFP_NOFS);
831 em->orig_start = em->start;
832 ram_size = ins.offset;
833 em->len = ins.offset;
835 em->block_start = ins.objectid;
836 em->block_len = ins.offset;
837 em->bdev = root->fs_info->fs_devices->latest_bdev;
838 set_bit(EXTENT_FLAG_PINNED, &em->flags);
841 write_lock(&em_tree->lock);
842 ret = add_extent_mapping(em_tree, em);
843 write_unlock(&em_tree->lock);
844 if (ret != -EEXIST) {
848 btrfs_drop_extent_cache(inode, start,
849 start + ram_size - 1, 0);
852 cur_alloc_size = ins.offset;
853 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
854 ram_size, cur_alloc_size, 0);
857 if (root->root_key.objectid ==
858 BTRFS_DATA_RELOC_TREE_OBJECTID) {
859 ret = btrfs_reloc_clone_csums(inode, start,
864 if (disk_num_bytes < cur_alloc_size)
867 /* we're not doing compressed IO, don't unlock the first
868 * page (which the caller expects to stay locked), don't
869 * clear any dirty bits and don't set any writeback bits
871 * Do set the Private2 bit so we know this page was properly
872 * setup for writepage
874 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
875 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
878 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
879 start, start + ram_size - 1,
881 disk_num_bytes -= cur_alloc_size;
882 num_bytes -= cur_alloc_size;
883 alloc_hint = ins.objectid + ins.offset;
884 start += cur_alloc_size;
888 btrfs_end_transaction(trans, root);
894 * work queue call back to started compression on a file and pages
896 static noinline void async_cow_start(struct btrfs_work *work)
898 struct async_cow *async_cow;
900 async_cow = container_of(work, struct async_cow, work);
902 compress_file_range(async_cow->inode, async_cow->locked_page,
903 async_cow->start, async_cow->end, async_cow,
906 async_cow->inode = NULL;
910 * work queue call back to submit previously compressed pages
912 static noinline void async_cow_submit(struct btrfs_work *work)
914 struct async_cow *async_cow;
915 struct btrfs_root *root;
916 unsigned long nr_pages;
918 async_cow = container_of(work, struct async_cow, work);
920 root = async_cow->root;
921 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
924 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
926 if (atomic_read(&root->fs_info->async_delalloc_pages) <
928 waitqueue_active(&root->fs_info->async_submit_wait))
929 wake_up(&root->fs_info->async_submit_wait);
931 if (async_cow->inode)
932 submit_compressed_extents(async_cow->inode, async_cow);
935 static noinline void async_cow_free(struct btrfs_work *work)
937 struct async_cow *async_cow;
938 async_cow = container_of(work, struct async_cow, work);
942 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
943 u64 start, u64 end, int *page_started,
944 unsigned long *nr_written)
946 struct async_cow *async_cow;
947 struct btrfs_root *root = BTRFS_I(inode)->root;
948 unsigned long nr_pages;
950 int limit = 10 * 1024 * 1042;
952 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
953 1, 0, NULL, GFP_NOFS);
954 while (start < end) {
955 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
956 async_cow->inode = inode;
957 async_cow->root = root;
958 async_cow->locked_page = locked_page;
959 async_cow->start = start;
961 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
964 cur_end = min(end, start + 512 * 1024 - 1);
966 async_cow->end = cur_end;
967 INIT_LIST_HEAD(&async_cow->extents);
969 async_cow->work.func = async_cow_start;
970 async_cow->work.ordered_func = async_cow_submit;
971 async_cow->work.ordered_free = async_cow_free;
972 async_cow->work.flags = 0;
974 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
976 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
978 btrfs_queue_worker(&root->fs_info->delalloc_workers,
981 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
982 wait_event(root->fs_info->async_submit_wait,
983 (atomic_read(&root->fs_info->async_delalloc_pages) <
987 while (atomic_read(&root->fs_info->async_submit_draining) &&
988 atomic_read(&root->fs_info->async_delalloc_pages)) {
989 wait_event(root->fs_info->async_submit_wait,
990 (atomic_read(&root->fs_info->async_delalloc_pages) ==
994 *nr_written += nr_pages;
1001 static noinline int csum_exist_in_range(struct btrfs_root *root,
1002 u64 bytenr, u64 num_bytes)
1005 struct btrfs_ordered_sum *sums;
1008 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1009 bytenr + num_bytes - 1, &list);
1010 if (ret == 0 && list_empty(&list))
1013 while (!list_empty(&list)) {
1014 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1015 list_del(&sums->list);
1022 * when nowcow writeback call back. This checks for snapshots or COW copies
1023 * of the extents that exist in the file, and COWs the file as required.
1025 * If no cow copies or snapshots exist, we write directly to the existing
1028 static noinline int run_delalloc_nocow(struct inode *inode,
1029 struct page *locked_page,
1030 u64 start, u64 end, int *page_started, int force,
1031 unsigned long *nr_written)
1033 struct btrfs_root *root = BTRFS_I(inode)->root;
1034 struct btrfs_trans_handle *trans;
1035 struct extent_buffer *leaf;
1036 struct btrfs_path *path;
1037 struct btrfs_file_extent_item *fi;
1038 struct btrfs_key found_key;
1050 bool nolock = false;
1052 path = btrfs_alloc_path();
1054 if (root == root->fs_info->tree_root) {
1056 trans = btrfs_join_transaction_nolock(root, 1);
1058 trans = btrfs_join_transaction(root, 1);
1060 BUG_ON(IS_ERR(trans));
1062 cow_start = (u64)-1;
1065 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1068 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1069 leaf = path->nodes[0];
1070 btrfs_item_key_to_cpu(leaf, &found_key,
1071 path->slots[0] - 1);
1072 if (found_key.objectid == inode->i_ino &&
1073 found_key.type == BTRFS_EXTENT_DATA_KEY)
1078 leaf = path->nodes[0];
1079 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1080 ret = btrfs_next_leaf(root, path);
1085 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1093 if (found_key.objectid > inode->i_ino ||
1094 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1095 found_key.offset > end)
1098 if (found_key.offset > cur_offset) {
1099 extent_end = found_key.offset;
1104 fi = btrfs_item_ptr(leaf, path->slots[0],
1105 struct btrfs_file_extent_item);
1106 extent_type = btrfs_file_extent_type(leaf, fi);
1108 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1109 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1110 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1111 extent_offset = btrfs_file_extent_offset(leaf, fi);
1112 extent_end = found_key.offset +
1113 btrfs_file_extent_num_bytes(leaf, fi);
1114 if (extent_end <= start) {
1118 if (disk_bytenr == 0)
1120 if (btrfs_file_extent_compression(leaf, fi) ||
1121 btrfs_file_extent_encryption(leaf, fi) ||
1122 btrfs_file_extent_other_encoding(leaf, fi))
1124 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1126 if (btrfs_extent_readonly(root, disk_bytenr))
1128 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1130 extent_offset, disk_bytenr))
1132 disk_bytenr += extent_offset;
1133 disk_bytenr += cur_offset - found_key.offset;
1134 num_bytes = min(end + 1, extent_end) - cur_offset;
1136 * force cow if csum exists in the range.
1137 * this ensure that csum for a given extent are
1138 * either valid or do not exist.
1140 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1144 extent_end = found_key.offset +
1145 btrfs_file_extent_inline_len(leaf, fi);
1146 extent_end = ALIGN(extent_end, root->sectorsize);
1151 if (extent_end <= start) {
1156 if (cow_start == (u64)-1)
1157 cow_start = cur_offset;
1158 cur_offset = extent_end;
1159 if (cur_offset > end)
1165 btrfs_release_path(root, path);
1166 if (cow_start != (u64)-1) {
1167 ret = cow_file_range(inode, locked_page, cow_start,
1168 found_key.offset - 1, page_started,
1171 cow_start = (u64)-1;
1174 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1175 struct extent_map *em;
1176 struct extent_map_tree *em_tree;
1177 em_tree = &BTRFS_I(inode)->extent_tree;
1178 em = alloc_extent_map(GFP_NOFS);
1180 em->start = cur_offset;
1181 em->orig_start = em->start;
1182 em->len = num_bytes;
1183 em->block_len = num_bytes;
1184 em->block_start = disk_bytenr;
1185 em->bdev = root->fs_info->fs_devices->latest_bdev;
1186 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1188 write_lock(&em_tree->lock);
1189 ret = add_extent_mapping(em_tree, em);
1190 write_unlock(&em_tree->lock);
1191 if (ret != -EEXIST) {
1192 free_extent_map(em);
1195 btrfs_drop_extent_cache(inode, em->start,
1196 em->start + em->len - 1, 0);
1198 type = BTRFS_ORDERED_PREALLOC;
1200 type = BTRFS_ORDERED_NOCOW;
1203 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1204 num_bytes, num_bytes, type);
1207 if (root->root_key.objectid ==
1208 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1209 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1214 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1215 cur_offset, cur_offset + num_bytes - 1,
1216 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1217 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1218 EXTENT_SET_PRIVATE2);
1219 cur_offset = extent_end;
1220 if (cur_offset > end)
1223 btrfs_release_path(root, path);
1225 if (cur_offset <= end && cow_start == (u64)-1)
1226 cow_start = cur_offset;
1227 if (cow_start != (u64)-1) {
1228 ret = cow_file_range(inode, locked_page, cow_start, end,
1229 page_started, nr_written, 1);
1234 ret = btrfs_end_transaction_nolock(trans, root);
1237 ret = btrfs_end_transaction(trans, root);
1240 btrfs_free_path(path);
1245 * extent_io.c call back to do delayed allocation processing
1247 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1248 u64 start, u64 end, int *page_started,
1249 unsigned long *nr_written)
1252 struct btrfs_root *root = BTRFS_I(inode)->root;
1254 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1255 ret = run_delalloc_nocow(inode, locked_page, start, end,
1256 page_started, 1, nr_written);
1257 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1258 ret = run_delalloc_nocow(inode, locked_page, start, end,
1259 page_started, 0, nr_written);
1260 else if (!btrfs_test_opt(root, COMPRESS) &&
1261 !(BTRFS_I(inode)->force_compress) &&
1262 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1263 ret = cow_file_range(inode, locked_page, start, end,
1264 page_started, nr_written, 1);
1266 ret = cow_file_range_async(inode, locked_page, start, end,
1267 page_started, nr_written);
1271 static int btrfs_split_extent_hook(struct inode *inode,
1272 struct extent_state *orig, u64 split)
1274 /* not delalloc, ignore it */
1275 if (!(orig->state & EXTENT_DELALLOC))
1278 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1283 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1284 * extents so we can keep track of new extents that are just merged onto old
1285 * extents, such as when we are doing sequential writes, so we can properly
1286 * account for the metadata space we'll need.
1288 static int btrfs_merge_extent_hook(struct inode *inode,
1289 struct extent_state *new,
1290 struct extent_state *other)
1292 /* not delalloc, ignore it */
1293 if (!(other->state & EXTENT_DELALLOC))
1296 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1301 * extent_io.c set_bit_hook, used to track delayed allocation
1302 * bytes in this file, and to maintain the list of inodes that
1303 * have pending delalloc work to be done.
1305 static int btrfs_set_bit_hook(struct inode *inode,
1306 struct extent_state *state, int *bits)
1310 * set_bit and clear bit hooks normally require _irqsave/restore
1311 * but in this case, we are only testeing for the DELALLOC
1312 * bit, which is only set or cleared with irqs on
1314 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1315 struct btrfs_root *root = BTRFS_I(inode)->root;
1316 u64 len = state->end + 1 - state->start;
1317 int do_list = (root->root_key.objectid !=
1318 BTRFS_ROOT_TREE_OBJECTID);
1320 if (*bits & EXTENT_FIRST_DELALLOC)
1321 *bits &= ~EXTENT_FIRST_DELALLOC;
1323 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1325 spin_lock(&root->fs_info->delalloc_lock);
1326 BTRFS_I(inode)->delalloc_bytes += len;
1327 root->fs_info->delalloc_bytes += len;
1328 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1329 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1330 &root->fs_info->delalloc_inodes);
1332 spin_unlock(&root->fs_info->delalloc_lock);
1338 * extent_io.c clear_bit_hook, see set_bit_hook for why
1340 static int btrfs_clear_bit_hook(struct inode *inode,
1341 struct extent_state *state, int *bits)
1344 * set_bit and clear bit hooks normally require _irqsave/restore
1345 * but in this case, we are only testeing for the DELALLOC
1346 * bit, which is only set or cleared with irqs on
1348 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1349 struct btrfs_root *root = BTRFS_I(inode)->root;
1350 u64 len = state->end + 1 - state->start;
1351 int do_list = (root->root_key.objectid !=
1352 BTRFS_ROOT_TREE_OBJECTID);
1354 if (*bits & EXTENT_FIRST_DELALLOC)
1355 *bits &= ~EXTENT_FIRST_DELALLOC;
1356 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1357 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1359 if (*bits & EXTENT_DO_ACCOUNTING)
1360 btrfs_delalloc_release_metadata(inode, len);
1362 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1364 btrfs_free_reserved_data_space(inode, len);
1366 spin_lock(&root->fs_info->delalloc_lock);
1367 root->fs_info->delalloc_bytes -= len;
1368 BTRFS_I(inode)->delalloc_bytes -= len;
1370 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1371 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1372 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1374 spin_unlock(&root->fs_info->delalloc_lock);
1380 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1381 * we don't create bios that span stripes or chunks
1383 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1384 size_t size, struct bio *bio,
1385 unsigned long bio_flags)
1387 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1388 struct btrfs_mapping_tree *map_tree;
1389 u64 logical = (u64)bio->bi_sector << 9;
1394 if (bio_flags & EXTENT_BIO_COMPRESSED)
1397 length = bio->bi_size;
1398 map_tree = &root->fs_info->mapping_tree;
1399 map_length = length;
1400 ret = btrfs_map_block(map_tree, READ, logical,
1401 &map_length, NULL, 0);
1403 if (map_length < length + size)
1409 * in order to insert checksums into the metadata in large chunks,
1410 * we wait until bio submission time. All the pages in the bio are
1411 * checksummed and sums are attached onto the ordered extent record.
1413 * At IO completion time the cums attached on the ordered extent record
1414 * are inserted into the btree
1416 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1417 struct bio *bio, int mirror_num,
1418 unsigned long bio_flags,
1421 struct btrfs_root *root = BTRFS_I(inode)->root;
1424 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1430 * in order to insert checksums into the metadata in large chunks,
1431 * we wait until bio submission time. All the pages in the bio are
1432 * checksummed and sums are attached onto the ordered extent record.
1434 * At IO completion time the cums attached on the ordered extent record
1435 * are inserted into the btree
1437 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1438 int mirror_num, unsigned long bio_flags,
1441 struct btrfs_root *root = BTRFS_I(inode)->root;
1442 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1446 * extent_io.c submission hook. This does the right thing for csum calculation
1447 * on write, or reading the csums from the tree before a read
1449 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1450 int mirror_num, unsigned long bio_flags,
1453 struct btrfs_root *root = BTRFS_I(inode)->root;
1457 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1459 if (root == root->fs_info->tree_root)
1460 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1462 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465 if (!(rw & REQ_WRITE)) {
1466 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1467 return btrfs_submit_compressed_read(inode, bio,
1468 mirror_num, bio_flags);
1469 } else if (!skip_sum) {
1470 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1475 } else if (!skip_sum) {
1476 /* csum items have already been cloned */
1477 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1479 /* we're doing a write, do the async checksumming */
1480 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1481 inode, rw, bio, mirror_num,
1482 bio_flags, bio_offset,
1483 __btrfs_submit_bio_start,
1484 __btrfs_submit_bio_done);
1488 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1492 * given a list of ordered sums record them in the inode. This happens
1493 * at IO completion time based on sums calculated at bio submission time.
1495 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1496 struct inode *inode, u64 file_offset,
1497 struct list_head *list)
1499 struct btrfs_ordered_sum *sum;
1501 btrfs_set_trans_block_group(trans, inode);
1503 list_for_each_entry(sum, list, list) {
1504 btrfs_csum_file_blocks(trans,
1505 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1510 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1511 struct extent_state **cached_state)
1513 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1515 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1516 cached_state, GFP_NOFS);
1519 /* see btrfs_writepage_start_hook for details on why this is required */
1520 struct btrfs_writepage_fixup {
1522 struct btrfs_work work;
1525 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1527 struct btrfs_writepage_fixup *fixup;
1528 struct btrfs_ordered_extent *ordered;
1529 struct extent_state *cached_state = NULL;
1531 struct inode *inode;
1535 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1539 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1540 ClearPageChecked(page);
1544 inode = page->mapping->host;
1545 page_start = page_offset(page);
1546 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1548 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1549 &cached_state, GFP_NOFS);
1551 /* already ordered? We're done */
1552 if (PagePrivate2(page))
1555 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1557 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1558 page_end, &cached_state, GFP_NOFS);
1560 btrfs_start_ordered_extent(inode, ordered, 1);
1565 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1566 ClearPageChecked(page);
1568 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1569 &cached_state, GFP_NOFS);
1572 page_cache_release(page);
1577 * There are a few paths in the higher layers of the kernel that directly
1578 * set the page dirty bit without asking the filesystem if it is a
1579 * good idea. This causes problems because we want to make sure COW
1580 * properly happens and the data=ordered rules are followed.
1582 * In our case any range that doesn't have the ORDERED bit set
1583 * hasn't been properly setup for IO. We kick off an async process
1584 * to fix it up. The async helper will wait for ordered extents, set
1585 * the delalloc bit and make it safe to write the page.
1587 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1589 struct inode *inode = page->mapping->host;
1590 struct btrfs_writepage_fixup *fixup;
1591 struct btrfs_root *root = BTRFS_I(inode)->root;
1593 /* this page is properly in the ordered list */
1594 if (TestClearPagePrivate2(page))
1597 if (PageChecked(page))
1600 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1604 SetPageChecked(page);
1605 page_cache_get(page);
1606 fixup->work.func = btrfs_writepage_fixup_worker;
1608 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1612 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1613 struct inode *inode, u64 file_pos,
1614 u64 disk_bytenr, u64 disk_num_bytes,
1615 u64 num_bytes, u64 ram_bytes,
1616 u8 compression, u8 encryption,
1617 u16 other_encoding, int extent_type)
1619 struct btrfs_root *root = BTRFS_I(inode)->root;
1620 struct btrfs_file_extent_item *fi;
1621 struct btrfs_path *path;
1622 struct extent_buffer *leaf;
1623 struct btrfs_key ins;
1627 path = btrfs_alloc_path();
1630 path->leave_spinning = 1;
1633 * we may be replacing one extent in the tree with another.
1634 * The new extent is pinned in the extent map, and we don't want
1635 * to drop it from the cache until it is completely in the btree.
1637 * So, tell btrfs_drop_extents to leave this extent in the cache.
1638 * the caller is expected to unpin it and allow it to be merged
1641 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1645 ins.objectid = inode->i_ino;
1646 ins.offset = file_pos;
1647 ins.type = BTRFS_EXTENT_DATA_KEY;
1648 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1650 leaf = path->nodes[0];
1651 fi = btrfs_item_ptr(leaf, path->slots[0],
1652 struct btrfs_file_extent_item);
1653 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1654 btrfs_set_file_extent_type(leaf, fi, extent_type);
1655 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1656 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1657 btrfs_set_file_extent_offset(leaf, fi, 0);
1658 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1659 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1660 btrfs_set_file_extent_compression(leaf, fi, compression);
1661 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1662 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1664 btrfs_unlock_up_safe(path, 1);
1665 btrfs_set_lock_blocking(leaf);
1667 btrfs_mark_buffer_dirty(leaf);
1669 inode_add_bytes(inode, num_bytes);
1671 ins.objectid = disk_bytenr;
1672 ins.offset = disk_num_bytes;
1673 ins.type = BTRFS_EXTENT_ITEM_KEY;
1674 ret = btrfs_alloc_reserved_file_extent(trans, root,
1675 root->root_key.objectid,
1676 inode->i_ino, file_pos, &ins);
1678 btrfs_free_path(path);
1684 * helper function for btrfs_finish_ordered_io, this
1685 * just reads in some of the csum leaves to prime them into ram
1686 * before we start the transaction. It limits the amount of btree
1687 * reads required while inside the transaction.
1689 /* as ordered data IO finishes, this gets called so we can finish
1690 * an ordered extent if the range of bytes in the file it covers are
1693 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1695 struct btrfs_root *root = BTRFS_I(inode)->root;
1696 struct btrfs_trans_handle *trans = NULL;
1697 struct btrfs_ordered_extent *ordered_extent = NULL;
1698 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1699 struct extent_state *cached_state = NULL;
1700 int compress_type = 0;
1702 bool nolock = false;
1704 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1708 BUG_ON(!ordered_extent);
1710 nolock = (root == root->fs_info->tree_root);
1712 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1713 BUG_ON(!list_empty(&ordered_extent->list));
1714 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717 trans = btrfs_join_transaction_nolock(root, 1);
1719 trans = btrfs_join_transaction(root, 1);
1720 BUG_ON(IS_ERR(trans));
1721 btrfs_set_trans_block_group(trans, inode);
1722 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1723 ret = btrfs_update_inode(trans, root, inode);
1729 lock_extent_bits(io_tree, ordered_extent->file_offset,
1730 ordered_extent->file_offset + ordered_extent->len - 1,
1731 0, &cached_state, GFP_NOFS);
1734 trans = btrfs_join_transaction_nolock(root, 1);
1736 trans = btrfs_join_transaction(root, 1);
1737 BUG_ON(IS_ERR(trans));
1738 btrfs_set_trans_block_group(trans, inode);
1739 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1742 compress_type = ordered_extent->compress_type;
1743 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1744 BUG_ON(compress_type);
1745 ret = btrfs_mark_extent_written(trans, inode,
1746 ordered_extent->file_offset,
1747 ordered_extent->file_offset +
1748 ordered_extent->len);
1751 BUG_ON(root == root->fs_info->tree_root);
1752 ret = insert_reserved_file_extent(trans, inode,
1753 ordered_extent->file_offset,
1754 ordered_extent->start,
1755 ordered_extent->disk_len,
1756 ordered_extent->len,
1757 ordered_extent->len,
1758 compress_type, 0, 0,
1759 BTRFS_FILE_EXTENT_REG);
1760 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1761 ordered_extent->file_offset,
1762 ordered_extent->len);
1765 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1766 ordered_extent->file_offset +
1767 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1769 add_pending_csums(trans, inode, ordered_extent->file_offset,
1770 &ordered_extent->list);
1772 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1774 ret = btrfs_update_inode(trans, root, inode);
1781 btrfs_end_transaction_nolock(trans, root);
1783 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1785 btrfs_end_transaction(trans, root);
1789 btrfs_put_ordered_extent(ordered_extent);
1790 /* once for the tree */
1791 btrfs_put_ordered_extent(ordered_extent);
1796 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1797 struct extent_state *state, int uptodate)
1799 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1801 ClearPagePrivate2(page);
1802 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1806 * When IO fails, either with EIO or csum verification fails, we
1807 * try other mirrors that might have a good copy of the data. This
1808 * io_failure_record is used to record state as we go through all the
1809 * mirrors. If another mirror has good data, the page is set up to date
1810 * and things continue. If a good mirror can't be found, the original
1811 * bio end_io callback is called to indicate things have failed.
1813 struct io_failure_record {
1818 unsigned long bio_flags;
1822 static int btrfs_io_failed_hook(struct bio *failed_bio,
1823 struct page *page, u64 start, u64 end,
1824 struct extent_state *state)
1826 struct io_failure_record *failrec = NULL;
1828 struct extent_map *em;
1829 struct inode *inode = page->mapping->host;
1830 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1831 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1838 ret = get_state_private(failure_tree, start, &private);
1840 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1843 failrec->start = start;
1844 failrec->len = end - start + 1;
1845 failrec->last_mirror = 0;
1846 failrec->bio_flags = 0;
1848 read_lock(&em_tree->lock);
1849 em = lookup_extent_mapping(em_tree, start, failrec->len);
1850 if (em->start > start || em->start + em->len < start) {
1851 free_extent_map(em);
1854 read_unlock(&em_tree->lock);
1856 if (!em || IS_ERR(em)) {
1860 logical = start - em->start;
1861 logical = em->block_start + logical;
1862 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1863 logical = em->block_start;
1864 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1865 extent_set_compress_type(&failrec->bio_flags,
1868 failrec->logical = logical;
1869 free_extent_map(em);
1870 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1871 EXTENT_DIRTY, GFP_NOFS);
1872 set_state_private(failure_tree, start,
1873 (u64)(unsigned long)failrec);
1875 failrec = (struct io_failure_record *)(unsigned long)private;
1877 num_copies = btrfs_num_copies(
1878 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1879 failrec->logical, failrec->len);
1880 failrec->last_mirror++;
1882 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1883 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1886 if (state && state->start != failrec->start)
1888 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1890 if (!state || failrec->last_mirror > num_copies) {
1891 set_state_private(failure_tree, failrec->start, 0);
1892 clear_extent_bits(failure_tree, failrec->start,
1893 failrec->start + failrec->len - 1,
1894 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1898 bio = bio_alloc(GFP_NOFS, 1);
1899 bio->bi_private = state;
1900 bio->bi_end_io = failed_bio->bi_end_io;
1901 bio->bi_sector = failrec->logical >> 9;
1902 bio->bi_bdev = failed_bio->bi_bdev;
1905 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1906 if (failed_bio->bi_rw & REQ_WRITE)
1911 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1912 failrec->last_mirror,
1913 failrec->bio_flags, 0);
1918 * each time an IO finishes, we do a fast check in the IO failure tree
1919 * to see if we need to process or clean up an io_failure_record
1921 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1924 u64 private_failure;
1925 struct io_failure_record *failure;
1929 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1930 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1931 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1932 start, &private_failure);
1934 failure = (struct io_failure_record *)(unsigned long)
1936 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1938 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1940 failure->start + failure->len - 1,
1941 EXTENT_DIRTY | EXTENT_LOCKED,
1950 * when reads are done, we need to check csums to verify the data is correct
1951 * if there's a match, we allow the bio to finish. If not, we go through
1952 * the io_failure_record routines to find good copies
1954 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1955 struct extent_state *state)
1957 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1958 struct inode *inode = page->mapping->host;
1959 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1961 u64 private = ~(u32)0;
1963 struct btrfs_root *root = BTRFS_I(inode)->root;
1966 if (PageChecked(page)) {
1967 ClearPageChecked(page);
1971 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1974 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1975 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1976 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1981 if (state && state->start == start) {
1982 private = state->private;
1985 ret = get_state_private(io_tree, start, &private);
1987 kaddr = kmap_atomic(page, KM_USER0);
1991 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1992 btrfs_csum_final(csum, (char *)&csum);
1993 if (csum != private)
1996 kunmap_atomic(kaddr, KM_USER0);
1998 /* if the io failure tree for this inode is non-empty,
1999 * check to see if we've recovered from a failed IO
2001 btrfs_clean_io_failures(inode, start);
2005 if (printk_ratelimit()) {
2006 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2007 "private %llu\n", page->mapping->host->i_ino,
2008 (unsigned long long)start, csum,
2009 (unsigned long long)private);
2011 memset(kaddr + offset, 1, end - start + 1);
2012 flush_dcache_page(page);
2013 kunmap_atomic(kaddr, KM_USER0);
2019 struct delayed_iput {
2020 struct list_head list;
2021 struct inode *inode;
2024 void btrfs_add_delayed_iput(struct inode *inode)
2026 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2027 struct delayed_iput *delayed;
2029 if (atomic_add_unless(&inode->i_count, -1, 1))
2032 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2033 delayed->inode = inode;
2035 spin_lock(&fs_info->delayed_iput_lock);
2036 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2037 spin_unlock(&fs_info->delayed_iput_lock);
2040 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2043 struct btrfs_fs_info *fs_info = root->fs_info;
2044 struct delayed_iput *delayed;
2047 spin_lock(&fs_info->delayed_iput_lock);
2048 empty = list_empty(&fs_info->delayed_iputs);
2049 spin_unlock(&fs_info->delayed_iput_lock);
2053 down_read(&root->fs_info->cleanup_work_sem);
2054 spin_lock(&fs_info->delayed_iput_lock);
2055 list_splice_init(&fs_info->delayed_iputs, &list);
2056 spin_unlock(&fs_info->delayed_iput_lock);
2058 while (!list_empty(&list)) {
2059 delayed = list_entry(list.next, struct delayed_iput, list);
2060 list_del(&delayed->list);
2061 iput(delayed->inode);
2064 up_read(&root->fs_info->cleanup_work_sem);
2068 * calculate extra metadata reservation when snapshotting a subvolume
2069 * contains orphan files.
2071 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2072 struct btrfs_pending_snapshot *pending,
2073 u64 *bytes_to_reserve)
2075 struct btrfs_root *root;
2076 struct btrfs_block_rsv *block_rsv;
2080 root = pending->root;
2081 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2084 block_rsv = root->orphan_block_rsv;
2086 /* orphan block reservation for the snapshot */
2087 num_bytes = block_rsv->size;
2090 * after the snapshot is created, COWing tree blocks may use more
2091 * space than it frees. So we should make sure there is enough
2094 index = trans->transid & 0x1;
2095 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2096 num_bytes += block_rsv->size -
2097 (block_rsv->reserved + block_rsv->freed[index]);
2100 *bytes_to_reserve += num_bytes;
2103 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2104 struct btrfs_pending_snapshot *pending)
2106 struct btrfs_root *root = pending->root;
2107 struct btrfs_root *snap = pending->snap;
2108 struct btrfs_block_rsv *block_rsv;
2113 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2116 /* refill source subvolume's orphan block reservation */
2117 block_rsv = root->orphan_block_rsv;
2118 index = trans->transid & 0x1;
2119 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2120 num_bytes = block_rsv->size -
2121 (block_rsv->reserved + block_rsv->freed[index]);
2122 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2123 root->orphan_block_rsv,
2128 /* setup orphan block reservation for the snapshot */
2129 block_rsv = btrfs_alloc_block_rsv(snap);
2132 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2133 snap->orphan_block_rsv = block_rsv;
2135 num_bytes = root->orphan_block_rsv->size;
2136 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2137 block_rsv, num_bytes);
2141 /* insert orphan item for the snapshot */
2142 WARN_ON(!root->orphan_item_inserted);
2143 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2144 snap->root_key.objectid);
2146 snap->orphan_item_inserted = 1;
2150 enum btrfs_orphan_cleanup_state {
2151 ORPHAN_CLEANUP_STARTED = 1,
2152 ORPHAN_CLEANUP_DONE = 2,
2156 * This is called in transaction commmit time. If there are no orphan
2157 * files in the subvolume, it removes orphan item and frees block_rsv
2160 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2161 struct btrfs_root *root)
2165 if (!list_empty(&root->orphan_list) ||
2166 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2169 if (root->orphan_item_inserted &&
2170 btrfs_root_refs(&root->root_item) > 0) {
2171 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2172 root->root_key.objectid);
2174 root->orphan_item_inserted = 0;
2177 if (root->orphan_block_rsv) {
2178 WARN_ON(root->orphan_block_rsv->size > 0);
2179 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2180 root->orphan_block_rsv = NULL;
2185 * This creates an orphan entry for the given inode in case something goes
2186 * wrong in the middle of an unlink/truncate.
2188 * NOTE: caller of this function should reserve 5 units of metadata for
2191 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2193 struct btrfs_root *root = BTRFS_I(inode)->root;
2194 struct btrfs_block_rsv *block_rsv = NULL;
2199 if (!root->orphan_block_rsv) {
2200 block_rsv = btrfs_alloc_block_rsv(root);
2204 spin_lock(&root->orphan_lock);
2205 if (!root->orphan_block_rsv) {
2206 root->orphan_block_rsv = block_rsv;
2207 } else if (block_rsv) {
2208 btrfs_free_block_rsv(root, block_rsv);
2212 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2213 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2216 * For proper ENOSPC handling, we should do orphan
2217 * cleanup when mounting. But this introduces backward
2218 * compatibility issue.
2220 if (!xchg(&root->orphan_item_inserted, 1))
2228 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2229 BTRFS_I(inode)->orphan_meta_reserved = 1;
2232 spin_unlock(&root->orphan_lock);
2235 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2237 /* grab metadata reservation from transaction handle */
2239 ret = btrfs_orphan_reserve_metadata(trans, inode);
2243 /* insert an orphan item to track this unlinked/truncated file */
2245 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2249 /* insert an orphan item to track subvolume contains orphan files */
2251 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2252 root->root_key.objectid);
2259 * We have done the truncate/delete so we can go ahead and remove the orphan
2260 * item for this particular inode.
2262 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2264 struct btrfs_root *root = BTRFS_I(inode)->root;
2265 int delete_item = 0;
2266 int release_rsv = 0;
2269 spin_lock(&root->orphan_lock);
2270 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2271 list_del_init(&BTRFS_I(inode)->i_orphan);
2275 if (BTRFS_I(inode)->orphan_meta_reserved) {
2276 BTRFS_I(inode)->orphan_meta_reserved = 0;
2279 spin_unlock(&root->orphan_lock);
2281 if (trans && delete_item) {
2282 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2287 btrfs_orphan_release_metadata(inode);
2293 * this cleans up any orphans that may be left on the list from the last use
2296 int btrfs_orphan_cleanup(struct btrfs_root *root)
2298 struct btrfs_path *path;
2299 struct extent_buffer *leaf;
2300 struct btrfs_key key, found_key;
2301 struct btrfs_trans_handle *trans;
2302 struct inode *inode;
2303 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2305 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2308 path = btrfs_alloc_path();
2315 key.objectid = BTRFS_ORPHAN_OBJECTID;
2316 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2317 key.offset = (u64)-1;
2320 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2325 * if ret == 0 means we found what we were searching for, which
2326 * is weird, but possible, so only screw with path if we didnt
2327 * find the key and see if we have stuff that matches
2331 if (path->slots[0] == 0)
2336 /* pull out the item */
2337 leaf = path->nodes[0];
2338 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2340 /* make sure the item matches what we want */
2341 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2343 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2346 /* release the path since we're done with it */
2347 btrfs_release_path(root, path);
2350 * this is where we are basically btrfs_lookup, without the
2351 * crossing root thing. we store the inode number in the
2352 * offset of the orphan item.
2354 found_key.objectid = found_key.offset;
2355 found_key.type = BTRFS_INODE_ITEM_KEY;
2356 found_key.offset = 0;
2357 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2358 if (IS_ERR(inode)) {
2359 ret = PTR_ERR(inode);
2364 * add this inode to the orphan list so btrfs_orphan_del does
2365 * the proper thing when we hit it
2367 spin_lock(&root->orphan_lock);
2368 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2369 spin_unlock(&root->orphan_lock);
2372 * if this is a bad inode, means we actually succeeded in
2373 * removing the inode, but not the orphan record, which means
2374 * we need to manually delete the orphan since iput will just
2375 * do a destroy_inode
2377 if (is_bad_inode(inode)) {
2378 trans = btrfs_start_transaction(root, 0);
2379 if (IS_ERR(trans)) {
2380 ret = PTR_ERR(trans);
2383 btrfs_orphan_del(trans, inode);
2384 btrfs_end_transaction(trans, root);
2389 /* if we have links, this was a truncate, lets do that */
2390 if (inode->i_nlink) {
2391 if (!S_ISREG(inode->i_mode)) {
2397 ret = btrfs_truncate(inode);
2402 /* this will do delete_inode and everything for us */
2407 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2409 if (root->orphan_block_rsv)
2410 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2413 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2414 trans = btrfs_join_transaction(root, 1);
2416 btrfs_end_transaction(trans, root);
2420 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2422 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2426 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2427 btrfs_free_path(path);
2432 * very simple check to peek ahead in the leaf looking for xattrs. If we
2433 * don't find any xattrs, we know there can't be any acls.
2435 * slot is the slot the inode is in, objectid is the objectid of the inode
2437 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2438 int slot, u64 objectid)
2440 u32 nritems = btrfs_header_nritems(leaf);
2441 struct btrfs_key found_key;
2445 while (slot < nritems) {
2446 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2448 /* we found a different objectid, there must not be acls */
2449 if (found_key.objectid != objectid)
2452 /* we found an xattr, assume we've got an acl */
2453 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2457 * we found a key greater than an xattr key, there can't
2458 * be any acls later on
2460 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2467 * it goes inode, inode backrefs, xattrs, extents,
2468 * so if there are a ton of hard links to an inode there can
2469 * be a lot of backrefs. Don't waste time searching too hard,
2470 * this is just an optimization
2475 /* we hit the end of the leaf before we found an xattr or
2476 * something larger than an xattr. We have to assume the inode
2483 * read an inode from the btree into the in-memory inode
2485 static void btrfs_read_locked_inode(struct inode *inode)
2487 struct btrfs_path *path;
2488 struct extent_buffer *leaf;
2489 struct btrfs_inode_item *inode_item;
2490 struct btrfs_timespec *tspec;
2491 struct btrfs_root *root = BTRFS_I(inode)->root;
2492 struct btrfs_key location;
2494 u64 alloc_group_block;
2498 path = btrfs_alloc_path();
2500 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2502 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2506 leaf = path->nodes[0];
2507 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2508 struct btrfs_inode_item);
2510 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2511 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2512 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2513 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2514 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2516 tspec = btrfs_inode_atime(inode_item);
2517 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2518 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2520 tspec = btrfs_inode_mtime(inode_item);
2521 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2522 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2524 tspec = btrfs_inode_ctime(inode_item);
2525 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2526 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2528 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2529 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2530 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2531 inode->i_generation = BTRFS_I(inode)->generation;
2533 rdev = btrfs_inode_rdev(leaf, inode_item);
2535 BTRFS_I(inode)->index_cnt = (u64)-1;
2536 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2538 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2541 * try to precache a NULL acl entry for files that don't have
2542 * any xattrs or acls
2544 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2546 cache_no_acl(inode);
2548 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2549 alloc_group_block, 0);
2550 btrfs_free_path(path);
2553 switch (inode->i_mode & S_IFMT) {
2555 inode->i_mapping->a_ops = &btrfs_aops;
2556 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2557 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2558 inode->i_fop = &btrfs_file_operations;
2559 inode->i_op = &btrfs_file_inode_operations;
2562 inode->i_fop = &btrfs_dir_file_operations;
2563 if (root == root->fs_info->tree_root)
2564 inode->i_op = &btrfs_dir_ro_inode_operations;
2566 inode->i_op = &btrfs_dir_inode_operations;
2569 inode->i_op = &btrfs_symlink_inode_operations;
2570 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2571 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2574 inode->i_op = &btrfs_special_inode_operations;
2575 init_special_inode(inode, inode->i_mode, rdev);
2579 btrfs_update_iflags(inode);
2583 btrfs_free_path(path);
2584 make_bad_inode(inode);
2588 * given a leaf and an inode, copy the inode fields into the leaf
2590 static void fill_inode_item(struct btrfs_trans_handle *trans,
2591 struct extent_buffer *leaf,
2592 struct btrfs_inode_item *item,
2593 struct inode *inode)
2595 if (!leaf->map_token)
2596 map_private_extent_buffer(leaf, (unsigned long)item,
2597 sizeof(struct btrfs_inode_item),
2598 &leaf->map_token, &leaf->kaddr,
2599 &leaf->map_start, &leaf->map_len,
2602 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2603 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2604 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2605 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2606 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2608 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2609 inode->i_atime.tv_sec);
2610 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2611 inode->i_atime.tv_nsec);
2613 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2614 inode->i_mtime.tv_sec);
2615 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2616 inode->i_mtime.tv_nsec);
2618 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2619 inode->i_ctime.tv_sec);
2620 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2621 inode->i_ctime.tv_nsec);
2623 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2624 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2625 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2626 btrfs_set_inode_transid(leaf, item, trans->transid);
2627 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2628 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2629 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2631 if (leaf->map_token) {
2632 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2633 leaf->map_token = NULL;
2638 * copy everything in the in-memory inode into the btree.
2640 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2641 struct btrfs_root *root, struct inode *inode)
2643 struct btrfs_inode_item *inode_item;
2644 struct btrfs_path *path;
2645 struct extent_buffer *leaf;
2648 path = btrfs_alloc_path();
2650 path->leave_spinning = 1;
2651 ret = btrfs_lookup_inode(trans, root, path,
2652 &BTRFS_I(inode)->location, 1);
2659 btrfs_unlock_up_safe(path, 1);
2660 leaf = path->nodes[0];
2661 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2662 struct btrfs_inode_item);
2664 fill_inode_item(trans, leaf, inode_item, inode);
2665 btrfs_mark_buffer_dirty(leaf);
2666 btrfs_set_inode_last_trans(trans, inode);
2669 btrfs_free_path(path);
2675 * unlink helper that gets used here in inode.c and in the tree logging
2676 * recovery code. It remove a link in a directory with a given name, and
2677 * also drops the back refs in the inode to the directory
2679 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2680 struct btrfs_root *root,
2681 struct inode *dir, struct inode *inode,
2682 const char *name, int name_len)
2684 struct btrfs_path *path;
2686 struct extent_buffer *leaf;
2687 struct btrfs_dir_item *di;
2688 struct btrfs_key key;
2691 path = btrfs_alloc_path();
2697 path->leave_spinning = 1;
2698 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2699 name, name_len, -1);
2708 leaf = path->nodes[0];
2709 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2710 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2713 btrfs_release_path(root, path);
2715 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2717 dir->i_ino, &index);
2719 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2720 "inode %lu parent %lu\n", name_len, name,
2721 inode->i_ino, dir->i_ino);
2725 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2726 index, name, name_len, -1);
2735 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2736 btrfs_release_path(root, path);
2738 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2740 BUG_ON(ret != 0 && ret != -ENOENT);
2742 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2747 btrfs_free_path(path);
2751 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2752 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2753 btrfs_update_inode(trans, root, dir);
2758 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2759 struct btrfs_root *root,
2760 struct inode *dir, struct inode *inode,
2761 const char *name, int name_len)
2764 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2766 btrfs_drop_nlink(inode);
2767 ret = btrfs_update_inode(trans, root, inode);
2773 /* helper to check if there is any shared block in the path */
2774 static int check_path_shared(struct btrfs_root *root,
2775 struct btrfs_path *path)
2777 struct extent_buffer *eb;
2781 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2784 if (!path->nodes[level])
2786 eb = path->nodes[level];
2787 if (!btrfs_block_can_be_shared(root, eb))
2789 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2798 * helper to start transaction for unlink and rmdir.
2800 * unlink and rmdir are special in btrfs, they do not always free space.
2801 * so in enospc case, we should make sure they will free space before
2802 * allowing them to use the global metadata reservation.
2804 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2805 struct dentry *dentry)
2807 struct btrfs_trans_handle *trans;
2808 struct btrfs_root *root = BTRFS_I(dir)->root;
2809 struct btrfs_path *path;
2810 struct btrfs_inode_ref *ref;
2811 struct btrfs_dir_item *di;
2812 struct inode *inode = dentry->d_inode;
2818 trans = btrfs_start_transaction(root, 10);
2819 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2822 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2823 return ERR_PTR(-ENOSPC);
2825 /* check if there is someone else holds reference */
2826 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2827 return ERR_PTR(-ENOSPC);
2829 if (atomic_read(&inode->i_count) > 2)
2830 return ERR_PTR(-ENOSPC);
2832 if (xchg(&root->fs_info->enospc_unlink, 1))
2833 return ERR_PTR(-ENOSPC);
2835 path = btrfs_alloc_path();
2837 root->fs_info->enospc_unlink = 0;
2838 return ERR_PTR(-ENOMEM);
2841 trans = btrfs_start_transaction(root, 0);
2842 if (IS_ERR(trans)) {
2843 btrfs_free_path(path);
2844 root->fs_info->enospc_unlink = 0;
2848 path->skip_locking = 1;
2849 path->search_commit_root = 1;
2851 ret = btrfs_lookup_inode(trans, root, path,
2852 &BTRFS_I(dir)->location, 0);
2858 if (check_path_shared(root, path))
2863 btrfs_release_path(root, path);
2865 ret = btrfs_lookup_inode(trans, root, path,
2866 &BTRFS_I(inode)->location, 0);
2872 if (check_path_shared(root, path))
2877 btrfs_release_path(root, path);
2879 if (ret == 0 && S_ISREG(inode->i_mode)) {
2880 ret = btrfs_lookup_file_extent(trans, root, path,
2881 inode->i_ino, (u64)-1, 0);
2887 if (check_path_shared(root, path))
2889 btrfs_release_path(root, path);
2897 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2898 dentry->d_name.name, dentry->d_name.len, 0);
2904 if (check_path_shared(root, path))
2910 btrfs_release_path(root, path);
2912 ref = btrfs_lookup_inode_ref(trans, root, path,
2913 dentry->d_name.name, dentry->d_name.len,
2914 inode->i_ino, dir->i_ino, 0);
2920 if (check_path_shared(root, path))
2922 index = btrfs_inode_ref_index(path->nodes[0], ref);
2923 btrfs_release_path(root, path);
2925 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2926 dentry->d_name.name, dentry->d_name.len, 0);
2931 BUG_ON(ret == -ENOENT);
2932 if (check_path_shared(root, path))
2937 btrfs_free_path(path);
2939 btrfs_end_transaction(trans, root);
2940 root->fs_info->enospc_unlink = 0;
2941 return ERR_PTR(err);
2944 trans->block_rsv = &root->fs_info->global_block_rsv;
2948 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2949 struct btrfs_root *root)
2951 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2952 BUG_ON(!root->fs_info->enospc_unlink);
2953 root->fs_info->enospc_unlink = 0;
2955 btrfs_end_transaction_throttle(trans, root);
2958 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2960 struct btrfs_root *root = BTRFS_I(dir)->root;
2961 struct btrfs_trans_handle *trans;
2962 struct inode *inode = dentry->d_inode;
2964 unsigned long nr = 0;
2966 trans = __unlink_start_trans(dir, dentry);
2968 return PTR_ERR(trans);
2970 btrfs_set_trans_block_group(trans, dir);
2972 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2974 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2975 dentry->d_name.name, dentry->d_name.len);
2978 if (inode->i_nlink == 0) {
2979 ret = btrfs_orphan_add(trans, inode);
2983 nr = trans->blocks_used;
2984 __unlink_end_trans(trans, root);
2985 btrfs_btree_balance_dirty(root, nr);
2989 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2990 struct btrfs_root *root,
2991 struct inode *dir, u64 objectid,
2992 const char *name, int name_len)
2994 struct btrfs_path *path;
2995 struct extent_buffer *leaf;
2996 struct btrfs_dir_item *di;
2997 struct btrfs_key key;
3001 path = btrfs_alloc_path();
3005 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3006 name, name_len, -1);
3007 BUG_ON(!di || IS_ERR(di));
3009 leaf = path->nodes[0];
3010 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3011 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3012 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3014 btrfs_release_path(root, path);
3016 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3017 objectid, root->root_key.objectid,
3018 dir->i_ino, &index, name, name_len);
3020 BUG_ON(ret != -ENOENT);
3021 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3023 BUG_ON(!di || IS_ERR(di));
3025 leaf = path->nodes[0];
3026 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3027 btrfs_release_path(root, path);
3031 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3032 index, name, name_len, -1);
3033 BUG_ON(!di || IS_ERR(di));
3035 leaf = path->nodes[0];
3036 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3037 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3038 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3040 btrfs_release_path(root, path);
3042 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3043 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3044 ret = btrfs_update_inode(trans, root, dir);
3047 btrfs_free_path(path);
3051 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3053 struct inode *inode = dentry->d_inode;
3055 struct btrfs_root *root = BTRFS_I(dir)->root;
3056 struct btrfs_trans_handle *trans;
3057 unsigned long nr = 0;
3059 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3060 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3063 trans = __unlink_start_trans(dir, dentry);
3065 return PTR_ERR(trans);
3067 btrfs_set_trans_block_group(trans, dir);
3069 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3070 err = btrfs_unlink_subvol(trans, root, dir,
3071 BTRFS_I(inode)->location.objectid,
3072 dentry->d_name.name,
3073 dentry->d_name.len);
3077 err = btrfs_orphan_add(trans, inode);
3081 /* now the directory is empty */
3082 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3083 dentry->d_name.name, dentry->d_name.len);
3085 btrfs_i_size_write(inode, 0);
3087 nr = trans->blocks_used;
3088 __unlink_end_trans(trans, root);
3089 btrfs_btree_balance_dirty(root, nr);
3096 * when truncating bytes in a file, it is possible to avoid reading
3097 * the leaves that contain only checksum items. This can be the
3098 * majority of the IO required to delete a large file, but it must
3099 * be done carefully.
3101 * The keys in the level just above the leaves are checked to make sure
3102 * the lowest key in a given leaf is a csum key, and starts at an offset
3103 * after the new size.
3105 * Then the key for the next leaf is checked to make sure it also has
3106 * a checksum item for the same file. If it does, we know our target leaf
3107 * contains only checksum items, and it can be safely freed without reading
3110 * This is just an optimization targeted at large files. It may do
3111 * nothing. It will return 0 unless things went badly.
3113 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3114 struct btrfs_root *root,
3115 struct btrfs_path *path,
3116 struct inode *inode, u64 new_size)
3118 struct btrfs_key key;
3121 struct btrfs_key found_key;
3122 struct btrfs_key other_key;
3123 struct btrfs_leaf_ref *ref;
3127 path->lowest_level = 1;
3128 key.objectid = inode->i_ino;
3129 key.type = BTRFS_CSUM_ITEM_KEY;
3130 key.offset = new_size;
3132 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3136 if (path->nodes[1] == NULL) {
3141 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3142 nritems = btrfs_header_nritems(path->nodes[1]);
3147 if (path->slots[1] >= nritems)
3150 /* did we find a key greater than anything we want to delete? */
3151 if (found_key.objectid > inode->i_ino ||
3152 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3155 /* we check the next key in the node to make sure the leave contains
3156 * only checksum items. This comparison doesn't work if our
3157 * leaf is the last one in the node
3159 if (path->slots[1] + 1 >= nritems) {
3161 /* search forward from the last key in the node, this
3162 * will bring us into the next node in the tree
3164 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3166 /* unlikely, but we inc below, so check to be safe */
3167 if (found_key.offset == (u64)-1)
3170 /* search_forward needs a path with locks held, do the
3171 * search again for the original key. It is possible
3172 * this will race with a balance and return a path that
3173 * we could modify, but this drop is just an optimization
3174 * and is allowed to miss some leaves.
3176 btrfs_release_path(root, path);
3179 /* setup a max key for search_forward */
3180 other_key.offset = (u64)-1;
3181 other_key.type = key.type;
3182 other_key.objectid = key.objectid;
3184 path->keep_locks = 1;
3185 ret = btrfs_search_forward(root, &found_key, &other_key,
3187 path->keep_locks = 0;
3188 if (ret || found_key.objectid != key.objectid ||
3189 found_key.type != key.type) {
3194 key.offset = found_key.offset;
3195 btrfs_release_path(root, path);
3200 /* we know there's one more slot after us in the tree,
3201 * read that key so we can verify it is also a checksum item
3203 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3205 if (found_key.objectid < inode->i_ino)
3208 if (found_key.type != key.type || found_key.offset < new_size)
3212 * if the key for the next leaf isn't a csum key from this objectid,
3213 * we can't be sure there aren't good items inside this leaf.
3216 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3219 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3220 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3222 * it is safe to delete this leaf, it contains only
3223 * csum items from this inode at an offset >= new_size
3225 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3228 if (root->ref_cows && leaf_gen < trans->transid) {
3229 ref = btrfs_alloc_leaf_ref(root, 0);
3231 ref->root_gen = root->root_key.offset;
3232 ref->bytenr = leaf_start;
3234 ref->generation = leaf_gen;
3237 btrfs_sort_leaf_ref(ref);
3239 ret = btrfs_add_leaf_ref(root, ref, 0);
3241 btrfs_free_leaf_ref(root, ref);
3247 btrfs_release_path(root, path);
3249 if (other_key.objectid == inode->i_ino &&
3250 other_key.type == key.type && other_key.offset > key.offset) {
3251 key.offset = other_key.offset;
3257 /* fixup any changes we've made to the path */
3258 path->lowest_level = 0;
3259 path->keep_locks = 0;
3260 btrfs_release_path(root, path);
3267 * this can truncate away extent items, csum items and directory items.
3268 * It starts at a high offset and removes keys until it can't find
3269 * any higher than new_size
3271 * csum items that cross the new i_size are truncated to the new size
3274 * min_type is the minimum key type to truncate down to. If set to 0, this
3275 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3277 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3278 struct btrfs_root *root,
3279 struct inode *inode,
3280 u64 new_size, u32 min_type)
3282 struct btrfs_path *path;
3283 struct extent_buffer *leaf;
3284 struct btrfs_file_extent_item *fi;
3285 struct btrfs_key key;
3286 struct btrfs_key found_key;
3287 u64 extent_start = 0;
3288 u64 extent_num_bytes = 0;
3289 u64 extent_offset = 0;
3291 u64 mask = root->sectorsize - 1;
3292 u32 found_type = (u8)-1;
3295 int pending_del_nr = 0;
3296 int pending_del_slot = 0;
3297 int extent_type = -1;
3302 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3304 if (root->ref_cows || root == root->fs_info->tree_root)
3305 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3307 path = btrfs_alloc_path();
3311 key.objectid = inode->i_ino;
3312 key.offset = (u64)-1;
3316 path->leave_spinning = 1;
3317 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3324 /* there are no items in the tree for us to truncate, we're
3327 if (path->slots[0] == 0)
3334 leaf = path->nodes[0];
3335 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3336 found_type = btrfs_key_type(&found_key);
3339 if (found_key.objectid != inode->i_ino)
3342 if (found_type < min_type)
3345 item_end = found_key.offset;
3346 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3347 fi = btrfs_item_ptr(leaf, path->slots[0],
3348 struct btrfs_file_extent_item);
3349 extent_type = btrfs_file_extent_type(leaf, fi);
3350 encoding = btrfs_file_extent_compression(leaf, fi);
3351 encoding |= btrfs_file_extent_encryption(leaf, fi);
3352 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3354 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3356 btrfs_file_extent_num_bytes(leaf, fi);
3357 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3358 item_end += btrfs_file_extent_inline_len(leaf,
3363 if (found_type > min_type) {
3366 if (item_end < new_size)
3368 if (found_key.offset >= new_size)
3374 /* FIXME, shrink the extent if the ref count is only 1 */
3375 if (found_type != BTRFS_EXTENT_DATA_KEY)
3378 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3380 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3381 if (!del_item && !encoding) {
3382 u64 orig_num_bytes =
3383 btrfs_file_extent_num_bytes(leaf, fi);
3384 extent_num_bytes = new_size -
3385 found_key.offset + root->sectorsize - 1;
3386 extent_num_bytes = extent_num_bytes &
3387 ~((u64)root->sectorsize - 1);
3388 btrfs_set_file_extent_num_bytes(leaf, fi,
3390 num_dec = (orig_num_bytes -
3392 if (root->ref_cows && extent_start != 0)
3393 inode_sub_bytes(inode, num_dec);
3394 btrfs_mark_buffer_dirty(leaf);
3397 btrfs_file_extent_disk_num_bytes(leaf,
3399 extent_offset = found_key.offset -
3400 btrfs_file_extent_offset(leaf, fi);
3402 /* FIXME blocksize != 4096 */
3403 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3404 if (extent_start != 0) {
3407 inode_sub_bytes(inode, num_dec);
3410 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3412 * we can't truncate inline items that have had
3416 btrfs_file_extent_compression(leaf, fi) == 0 &&
3417 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3418 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3419 u32 size = new_size - found_key.offset;
3421 if (root->ref_cows) {
3422 inode_sub_bytes(inode, item_end + 1 -
3426 btrfs_file_extent_calc_inline_size(size);
3427 ret = btrfs_truncate_item(trans, root, path,
3430 } else if (root->ref_cows) {
3431 inode_sub_bytes(inode, item_end + 1 -
3437 if (!pending_del_nr) {
3438 /* no pending yet, add ourselves */
3439 pending_del_slot = path->slots[0];
3441 } else if (pending_del_nr &&
3442 path->slots[0] + 1 == pending_del_slot) {
3443 /* hop on the pending chunk */
3445 pending_del_slot = path->slots[0];
3452 if (found_extent && (root->ref_cows ||
3453 root == root->fs_info->tree_root)) {
3454 btrfs_set_path_blocking(path);
3455 ret = btrfs_free_extent(trans, root, extent_start,
3456 extent_num_bytes, 0,
3457 btrfs_header_owner(leaf),
3458 inode->i_ino, extent_offset);
3462 if (found_type == BTRFS_INODE_ITEM_KEY)
3465 if (path->slots[0] == 0 ||
3466 path->slots[0] != pending_del_slot) {
3467 if (root->ref_cows) {
3471 if (pending_del_nr) {
3472 ret = btrfs_del_items(trans, root, path,
3478 btrfs_release_path(root, path);
3485 if (pending_del_nr) {
3486 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3490 btrfs_free_path(path);
3495 * taken from block_truncate_page, but does cow as it zeros out
3496 * any bytes left in the last page in the file.
3498 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3500 struct inode *inode = mapping->host;
3501 struct btrfs_root *root = BTRFS_I(inode)->root;
3502 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3503 struct btrfs_ordered_extent *ordered;
3504 struct extent_state *cached_state = NULL;
3506 u32 blocksize = root->sectorsize;
3507 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3508 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3514 if ((offset & (blocksize - 1)) == 0)
3516 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3522 page = grab_cache_page(mapping, index);
3524 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3528 page_start = page_offset(page);
3529 page_end = page_start + PAGE_CACHE_SIZE - 1;
3531 if (!PageUptodate(page)) {
3532 ret = btrfs_readpage(NULL, page);
3534 if (page->mapping != mapping) {
3536 page_cache_release(page);
3539 if (!PageUptodate(page)) {
3544 wait_on_page_writeback(page);
3546 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3548 set_page_extent_mapped(page);
3550 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3552 unlock_extent_cached(io_tree, page_start, page_end,
3553 &cached_state, GFP_NOFS);
3555 page_cache_release(page);
3556 btrfs_start_ordered_extent(inode, ordered, 1);
3557 btrfs_put_ordered_extent(ordered);
3561 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3562 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3563 0, 0, &cached_state, GFP_NOFS);
3565 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3568 unlock_extent_cached(io_tree, page_start, page_end,
3569 &cached_state, GFP_NOFS);
3574 if (offset != PAGE_CACHE_SIZE) {
3576 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3577 flush_dcache_page(page);
3580 ClearPageChecked(page);
3581 set_page_dirty(page);
3582 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3587 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3589 page_cache_release(page);
3595 * This function puts in dummy file extents for the area we're creating a hole
3596 * for. So if we are truncating this file to a larger size we need to insert
3597 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3598 * the range between oldsize and size
3600 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3602 struct btrfs_trans_handle *trans;
3603 struct btrfs_root *root = BTRFS_I(inode)->root;
3604 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3605 struct extent_map *em = NULL;
3606 struct extent_state *cached_state = NULL;
3607 u64 mask = root->sectorsize - 1;
3608 u64 hole_start = (oldsize + mask) & ~mask;
3609 u64 block_end = (size + mask) & ~mask;
3615 if (size <= hole_start)
3619 struct btrfs_ordered_extent *ordered;
3620 btrfs_wait_ordered_range(inode, hole_start,
3621 block_end - hole_start);
3622 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3623 &cached_state, GFP_NOFS);
3624 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3627 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3628 &cached_state, GFP_NOFS);
3629 btrfs_put_ordered_extent(ordered);
3632 cur_offset = hole_start;
3634 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3635 block_end - cur_offset, 0);
3636 BUG_ON(IS_ERR(em) || !em);
3637 last_byte = min(extent_map_end(em), block_end);
3638 last_byte = (last_byte + mask) & ~mask;
3639 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3641 hole_size = last_byte - cur_offset;
3643 trans = btrfs_start_transaction(root, 2);
3644 if (IS_ERR(trans)) {
3645 err = PTR_ERR(trans);
3648 btrfs_set_trans_block_group(trans, inode);
3650 err = btrfs_drop_extents(trans, inode, cur_offset,
3651 cur_offset + hole_size,
3656 err = btrfs_insert_file_extent(trans, root,
3657 inode->i_ino, cur_offset, 0,
3658 0, hole_size, 0, hole_size,
3663 btrfs_drop_extent_cache(inode, hole_start,
3666 btrfs_end_transaction(trans, root);
3668 free_extent_map(em);
3670 cur_offset = last_byte;
3671 if (cur_offset >= block_end)
3675 free_extent_map(em);
3676 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3681 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3683 loff_t oldsize = i_size_read(inode);
3686 if (newsize == oldsize)
3689 if (newsize > oldsize) {
3690 i_size_write(inode, newsize);
3691 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3692 truncate_pagecache(inode, oldsize, newsize);
3693 ret = btrfs_cont_expand(inode, oldsize, newsize);
3695 btrfs_setsize(inode, oldsize);
3699 mark_inode_dirty(inode);
3703 * We're truncating a file that used to have good data down to
3704 * zero. Make sure it gets into the ordered flush list so that
3705 * any new writes get down to disk quickly.
3708 BTRFS_I(inode)->ordered_data_close = 1;
3710 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3711 truncate_setsize(inode, newsize);
3712 ret = btrfs_truncate(inode);
3718 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3720 struct inode *inode = dentry->d_inode;
3721 struct btrfs_root *root = BTRFS_I(inode)->root;
3724 if (btrfs_root_readonly(root))
3727 err = inode_change_ok(inode, attr);
3731 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3732 err = btrfs_setsize(inode, attr->ia_size);
3737 if (attr->ia_valid) {
3738 setattr_copy(inode, attr);
3739 mark_inode_dirty(inode);
3741 if (attr->ia_valid & ATTR_MODE)
3742 err = btrfs_acl_chmod(inode);
3748 void btrfs_evict_inode(struct inode *inode)
3750 struct btrfs_trans_handle *trans;
3751 struct btrfs_root *root = BTRFS_I(inode)->root;
3755 trace_btrfs_inode_evict(inode);
3757 truncate_inode_pages(&inode->i_data, 0);
3758 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3759 root == root->fs_info->tree_root))
3762 if (is_bad_inode(inode)) {
3763 btrfs_orphan_del(NULL, inode);
3766 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3767 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3769 if (root->fs_info->log_root_recovering) {
3770 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3774 if (inode->i_nlink > 0) {
3775 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3779 btrfs_i_size_write(inode, 0);
3782 trans = btrfs_start_transaction(root, 0);
3783 BUG_ON(IS_ERR(trans));
3784 btrfs_set_trans_block_group(trans, inode);
3785 trans->block_rsv = root->orphan_block_rsv;
3787 ret = btrfs_block_rsv_check(trans, root,
3788 root->orphan_block_rsv, 0, 5);
3790 BUG_ON(ret != -EAGAIN);
3791 ret = btrfs_commit_transaction(trans, root);
3796 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3800 nr = trans->blocks_used;
3801 btrfs_end_transaction(trans, root);
3803 btrfs_btree_balance_dirty(root, nr);
3808 ret = btrfs_orphan_del(trans, inode);
3812 nr = trans->blocks_used;
3813 btrfs_end_transaction(trans, root);
3814 btrfs_btree_balance_dirty(root, nr);
3816 end_writeback(inode);
3821 * this returns the key found in the dir entry in the location pointer.
3822 * If no dir entries were found, location->objectid is 0.
3824 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3825 struct btrfs_key *location)
3827 const char *name = dentry->d_name.name;
3828 int namelen = dentry->d_name.len;
3829 struct btrfs_dir_item *di;
3830 struct btrfs_path *path;
3831 struct btrfs_root *root = BTRFS_I(dir)->root;
3834 path = btrfs_alloc_path();
3837 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3842 if (!di || IS_ERR(di))
3845 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3847 btrfs_free_path(path);
3850 location->objectid = 0;
3855 * when we hit a tree root in a directory, the btrfs part of the inode
3856 * needs to be changed to reflect the root directory of the tree root. This
3857 * is kind of like crossing a mount point.
3859 static int fixup_tree_root_location(struct btrfs_root *root,
3861 struct dentry *dentry,
3862 struct btrfs_key *location,
3863 struct btrfs_root **sub_root)
3865 struct btrfs_path *path;
3866 struct btrfs_root *new_root;
3867 struct btrfs_root_ref *ref;
3868 struct extent_buffer *leaf;
3872 path = btrfs_alloc_path();
3879 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3880 BTRFS_I(dir)->root->root_key.objectid,
3881 location->objectid);
3888 leaf = path->nodes[0];
3889 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3890 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3891 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3894 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3895 (unsigned long)(ref + 1),
3896 dentry->d_name.len);
3900 btrfs_release_path(root->fs_info->tree_root, path);
3902 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3903 if (IS_ERR(new_root)) {
3904 err = PTR_ERR(new_root);
3908 if (btrfs_root_refs(&new_root->root_item) == 0) {
3913 *sub_root = new_root;
3914 location->objectid = btrfs_root_dirid(&new_root->root_item);
3915 location->type = BTRFS_INODE_ITEM_KEY;
3916 location->offset = 0;
3919 btrfs_free_path(path);
3923 static void inode_tree_add(struct inode *inode)
3925 struct btrfs_root *root = BTRFS_I(inode)->root;
3926 struct btrfs_inode *entry;
3928 struct rb_node *parent;
3930 p = &root->inode_tree.rb_node;
3933 if (inode_unhashed(inode))
3936 spin_lock(&root->inode_lock);
3939 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3941 if (inode->i_ino < entry->vfs_inode.i_ino)
3942 p = &parent->rb_left;
3943 else if (inode->i_ino > entry->vfs_inode.i_ino)
3944 p = &parent->rb_right;
3946 WARN_ON(!(entry->vfs_inode.i_state &
3947 (I_WILL_FREE | I_FREEING)));
3948 rb_erase(parent, &root->inode_tree);
3949 RB_CLEAR_NODE(parent);
3950 spin_unlock(&root->inode_lock);
3954 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3955 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3956 spin_unlock(&root->inode_lock);
3959 static void inode_tree_del(struct inode *inode)
3961 struct btrfs_root *root = BTRFS_I(inode)->root;
3964 spin_lock(&root->inode_lock);
3965 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3966 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3967 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3968 empty = RB_EMPTY_ROOT(&root->inode_tree);
3970 spin_unlock(&root->inode_lock);
3973 * Free space cache has inodes in the tree root, but the tree root has a
3974 * root_refs of 0, so this could end up dropping the tree root as a
3975 * snapshot, so we need the extra !root->fs_info->tree_root check to
3976 * make sure we don't drop it.
3978 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3979 root != root->fs_info->tree_root) {
3980 synchronize_srcu(&root->fs_info->subvol_srcu);
3981 spin_lock(&root->inode_lock);
3982 empty = RB_EMPTY_ROOT(&root->inode_tree);
3983 spin_unlock(&root->inode_lock);
3985 btrfs_add_dead_root(root);
3989 int btrfs_invalidate_inodes(struct btrfs_root *root)
3991 struct rb_node *node;
3992 struct rb_node *prev;
3993 struct btrfs_inode *entry;
3994 struct inode *inode;
3997 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3999 spin_lock(&root->inode_lock);
4001 node = root->inode_tree.rb_node;
4005 entry = rb_entry(node, struct btrfs_inode, rb_node);
4007 if (objectid < entry->vfs_inode.i_ino)
4008 node = node->rb_left;
4009 else if (objectid > entry->vfs_inode.i_ino)
4010 node = node->rb_right;
4016 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4017 if (objectid <= entry->vfs_inode.i_ino) {
4021 prev = rb_next(prev);
4025 entry = rb_entry(node, struct btrfs_inode, rb_node);
4026 objectid = entry->vfs_inode.i_ino + 1;
4027 inode = igrab(&entry->vfs_inode);
4029 spin_unlock(&root->inode_lock);
4030 if (atomic_read(&inode->i_count) > 1)
4031 d_prune_aliases(inode);
4033 * btrfs_drop_inode will have it removed from
4034 * the inode cache when its usage count
4039 spin_lock(&root->inode_lock);
4043 if (cond_resched_lock(&root->inode_lock))
4046 node = rb_next(node);
4048 spin_unlock(&root->inode_lock);
4052 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4054 struct btrfs_iget_args *args = p;
4055 inode->i_ino = args->ino;
4056 BTRFS_I(inode)->root = args->root;
4057 btrfs_set_inode_space_info(args->root, inode);
4061 static int btrfs_find_actor(struct inode *inode, void *opaque)
4063 struct btrfs_iget_args *args = opaque;
4064 return args->ino == inode->i_ino &&
4065 args->root == BTRFS_I(inode)->root;
4068 static struct inode *btrfs_iget_locked(struct super_block *s,
4070 struct btrfs_root *root)
4072 struct inode *inode;
4073 struct btrfs_iget_args args;
4074 args.ino = objectid;
4077 inode = iget5_locked(s, objectid, btrfs_find_actor,
4078 btrfs_init_locked_inode,
4083 /* Get an inode object given its location and corresponding root.
4084 * Returns in *is_new if the inode was read from disk
4086 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4087 struct btrfs_root *root, int *new)
4089 struct inode *inode;
4091 inode = btrfs_iget_locked(s, location->objectid, root);
4093 return ERR_PTR(-ENOMEM);
4095 if (inode->i_state & I_NEW) {
4096 BTRFS_I(inode)->root = root;
4097 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4098 btrfs_read_locked_inode(inode);
4099 inode_tree_add(inode);
4100 unlock_new_inode(inode);
4108 static struct inode *new_simple_dir(struct super_block *s,
4109 struct btrfs_key *key,
4110 struct btrfs_root *root)
4112 struct inode *inode = new_inode(s);
4115 return ERR_PTR(-ENOMEM);
4117 BTRFS_I(inode)->root = root;
4118 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4119 BTRFS_I(inode)->dummy_inode = 1;
4121 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4122 inode->i_op = &simple_dir_inode_operations;
4123 inode->i_fop = &simple_dir_operations;
4124 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4125 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4130 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4132 struct inode *inode;
4133 struct btrfs_root *root = BTRFS_I(dir)->root;
4134 struct btrfs_root *sub_root = root;
4135 struct btrfs_key location;
4139 if (dentry->d_name.len > BTRFS_NAME_LEN)
4140 return ERR_PTR(-ENAMETOOLONG);
4142 ret = btrfs_inode_by_name(dir, dentry, &location);
4145 return ERR_PTR(ret);
4147 if (location.objectid == 0)
4150 if (location.type == BTRFS_INODE_ITEM_KEY) {
4151 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4155 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4157 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4158 ret = fixup_tree_root_location(root, dir, dentry,
4159 &location, &sub_root);
4162 inode = ERR_PTR(ret);
4164 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4166 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4168 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4170 if (!IS_ERR(inode) && root != sub_root) {
4171 down_read(&root->fs_info->cleanup_work_sem);
4172 if (!(inode->i_sb->s_flags & MS_RDONLY))
4173 ret = btrfs_orphan_cleanup(sub_root);
4174 up_read(&root->fs_info->cleanup_work_sem);
4176 inode = ERR_PTR(ret);
4182 static int btrfs_dentry_delete(const struct dentry *dentry)
4184 struct btrfs_root *root;
4186 if (!dentry->d_inode && !IS_ROOT(dentry))
4187 dentry = dentry->d_parent;
4189 if (dentry->d_inode) {
4190 root = BTRFS_I(dentry->d_inode)->root;
4191 if (btrfs_root_refs(&root->root_item) == 0)
4197 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4198 struct nameidata *nd)
4200 struct inode *inode;
4202 inode = btrfs_lookup_dentry(dir, dentry);
4204 return ERR_CAST(inode);
4206 return d_splice_alias(inode, dentry);
4209 static unsigned char btrfs_filetype_table[] = {
4210 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4213 static int btrfs_real_readdir(struct file *filp, void *dirent,
4216 struct inode *inode = filp->f_dentry->d_inode;
4217 struct btrfs_root *root = BTRFS_I(inode)->root;
4218 struct btrfs_item *item;
4219 struct btrfs_dir_item *di;
4220 struct btrfs_key key;
4221 struct btrfs_key found_key;
4222 struct btrfs_path *path;
4225 struct extent_buffer *leaf;
4228 unsigned char d_type;
4233 int key_type = BTRFS_DIR_INDEX_KEY;
4238 /* FIXME, use a real flag for deciding about the key type */
4239 if (root->fs_info->tree_root == root)
4240 key_type = BTRFS_DIR_ITEM_KEY;
4242 /* special case for "." */
4243 if (filp->f_pos == 0) {
4244 over = filldir(dirent, ".", 1,
4251 /* special case for .., just use the back ref */
4252 if (filp->f_pos == 1) {
4253 u64 pino = parent_ino(filp->f_path.dentry);
4254 over = filldir(dirent, "..", 2,
4260 path = btrfs_alloc_path();
4263 btrfs_set_key_type(&key, key_type);
4264 key.offset = filp->f_pos;
4265 key.objectid = inode->i_ino;
4267 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4273 leaf = path->nodes[0];
4274 nritems = btrfs_header_nritems(leaf);
4275 slot = path->slots[0];
4276 if (advance || slot >= nritems) {
4277 if (slot >= nritems - 1) {
4278 ret = btrfs_next_leaf(root, path);
4281 leaf = path->nodes[0];
4282 nritems = btrfs_header_nritems(leaf);
4283 slot = path->slots[0];
4291 item = btrfs_item_nr(leaf, slot);
4292 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4294 if (found_key.objectid != key.objectid)
4296 if (btrfs_key_type(&found_key) != key_type)
4298 if (found_key.offset < filp->f_pos)
4301 filp->f_pos = found_key.offset;
4303 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4305 di_total = btrfs_item_size(leaf, item);
4307 while (di_cur < di_total) {
4308 struct btrfs_key location;
4310 if (verify_dir_item(root, leaf, di))
4313 name_len = btrfs_dir_name_len(leaf, di);
4314 if (name_len <= sizeof(tmp_name)) {
4315 name_ptr = tmp_name;
4317 name_ptr = kmalloc(name_len, GFP_NOFS);
4323 read_extent_buffer(leaf, name_ptr,
4324 (unsigned long)(di + 1), name_len);
4326 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4327 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4329 /* is this a reference to our own snapshot? If so
4332 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4333 location.objectid == root->root_key.objectid) {
4337 over = filldir(dirent, name_ptr, name_len,
4338 found_key.offset, location.objectid,
4342 if (name_ptr != tmp_name)
4347 di_len = btrfs_dir_name_len(leaf, di) +
4348 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4350 di = (struct btrfs_dir_item *)((char *)di + di_len);
4354 /* Reached end of directory/root. Bump pos past the last item. */
4355 if (key_type == BTRFS_DIR_INDEX_KEY)
4357 * 32-bit glibc will use getdents64, but then strtol -
4358 * so the last number we can serve is this.
4360 filp->f_pos = 0x7fffffff;
4366 btrfs_free_path(path);
4370 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4372 struct btrfs_root *root = BTRFS_I(inode)->root;
4373 struct btrfs_trans_handle *trans;
4375 bool nolock = false;
4377 if (BTRFS_I(inode)->dummy_inode)
4381 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4383 if (wbc->sync_mode == WB_SYNC_ALL) {
4385 trans = btrfs_join_transaction_nolock(root, 1);
4387 trans = btrfs_join_transaction(root, 1);
4389 return PTR_ERR(trans);
4390 btrfs_set_trans_block_group(trans, inode);
4392 ret = btrfs_end_transaction_nolock(trans, root);
4394 ret = btrfs_commit_transaction(trans, root);
4400 * This is somewhat expensive, updating the tree every time the
4401 * inode changes. But, it is most likely to find the inode in cache.
4402 * FIXME, needs more benchmarking...there are no reasons other than performance
4403 * to keep or drop this code.
4405 void btrfs_dirty_inode(struct inode *inode)
4407 struct btrfs_root *root = BTRFS_I(inode)->root;
4408 struct btrfs_trans_handle *trans;
4411 if (BTRFS_I(inode)->dummy_inode)
4414 trans = btrfs_join_transaction(root, 1);
4415 BUG_ON(IS_ERR(trans));
4416 btrfs_set_trans_block_group(trans, inode);
4418 ret = btrfs_update_inode(trans, root, inode);
4419 if (ret && ret == -ENOSPC) {
4420 /* whoops, lets try again with the full transaction */
4421 btrfs_end_transaction(trans, root);
4422 trans = btrfs_start_transaction(root, 1);
4423 if (IS_ERR(trans)) {
4424 if (printk_ratelimit()) {
4425 printk(KERN_ERR "btrfs: fail to "
4426 "dirty inode %lu error %ld\n",
4427 inode->i_ino, PTR_ERR(trans));
4431 btrfs_set_trans_block_group(trans, inode);
4433 ret = btrfs_update_inode(trans, root, inode);
4435 if (printk_ratelimit()) {
4436 printk(KERN_ERR "btrfs: fail to "
4437 "dirty inode %lu error %d\n",
4442 btrfs_end_transaction(trans, root);
4446 * find the highest existing sequence number in a directory
4447 * and then set the in-memory index_cnt variable to reflect
4448 * free sequence numbers
4450 static int btrfs_set_inode_index_count(struct inode *inode)
4452 struct btrfs_root *root = BTRFS_I(inode)->root;
4453 struct btrfs_key key, found_key;
4454 struct btrfs_path *path;
4455 struct extent_buffer *leaf;
4458 key.objectid = inode->i_ino;
4459 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4460 key.offset = (u64)-1;
4462 path = btrfs_alloc_path();
4466 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4469 /* FIXME: we should be able to handle this */
4475 * MAGIC NUMBER EXPLANATION:
4476 * since we search a directory based on f_pos we have to start at 2
4477 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4478 * else has to start at 2
4480 if (path->slots[0] == 0) {
4481 BTRFS_I(inode)->index_cnt = 2;
4487 leaf = path->nodes[0];
4488 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4490 if (found_key.objectid != inode->i_ino ||
4491 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4492 BTRFS_I(inode)->index_cnt = 2;
4496 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4498 btrfs_free_path(path);
4503 * helper to find a free sequence number in a given directory. This current
4504 * code is very simple, later versions will do smarter things in the btree
4506 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4510 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4511 ret = btrfs_set_inode_index_count(dir);
4516 *index = BTRFS_I(dir)->index_cnt;
4517 BTRFS_I(dir)->index_cnt++;
4522 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4523 struct btrfs_root *root,
4525 const char *name, int name_len,
4526 u64 ref_objectid, u64 objectid,
4527 u64 alloc_hint, int mode, u64 *index)
4529 struct inode *inode;
4530 struct btrfs_inode_item *inode_item;
4531 struct btrfs_key *location;
4532 struct btrfs_path *path;
4533 struct btrfs_inode_ref *ref;
4534 struct btrfs_key key[2];
4540 path = btrfs_alloc_path();
4543 inode = new_inode(root->fs_info->sb);
4545 return ERR_PTR(-ENOMEM);
4548 trace_btrfs_inode_request(dir);
4550 ret = btrfs_set_inode_index(dir, index);
4553 return ERR_PTR(ret);
4557 * index_cnt is ignored for everything but a dir,
4558 * btrfs_get_inode_index_count has an explanation for the magic
4561 BTRFS_I(inode)->index_cnt = 2;
4562 BTRFS_I(inode)->root = root;
4563 BTRFS_I(inode)->generation = trans->transid;
4564 inode->i_generation = BTRFS_I(inode)->generation;
4565 btrfs_set_inode_space_info(root, inode);
4571 BTRFS_I(inode)->block_group =
4572 btrfs_find_block_group(root, 0, alloc_hint, owner);
4574 key[0].objectid = objectid;
4575 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4578 key[1].objectid = objectid;
4579 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4580 key[1].offset = ref_objectid;
4582 sizes[0] = sizeof(struct btrfs_inode_item);
4583 sizes[1] = name_len + sizeof(*ref);
4585 path->leave_spinning = 1;
4586 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4590 inode_init_owner(inode, dir, mode);
4591 inode->i_ino = objectid;
4592 inode_set_bytes(inode, 0);
4593 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4594 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4595 struct btrfs_inode_item);
4596 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4598 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4599 struct btrfs_inode_ref);
4600 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4601 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4602 ptr = (unsigned long)(ref + 1);
4603 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4605 btrfs_mark_buffer_dirty(path->nodes[0]);
4606 btrfs_free_path(path);
4608 location = &BTRFS_I(inode)->location;
4609 location->objectid = objectid;
4610 location->offset = 0;
4611 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4613 btrfs_inherit_iflags(inode, dir);
4615 if ((mode & S_IFREG)) {
4616 if (btrfs_test_opt(root, NODATASUM))
4617 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4618 if (btrfs_test_opt(root, NODATACOW) ||
4619 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4620 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4623 insert_inode_hash(inode);
4624 inode_tree_add(inode);
4626 trace_btrfs_inode_new(inode);
4631 BTRFS_I(dir)->index_cnt--;
4632 btrfs_free_path(path);
4634 return ERR_PTR(ret);
4637 static inline u8 btrfs_inode_type(struct inode *inode)
4639 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4643 * utility function to add 'inode' into 'parent_inode' with
4644 * a give name and a given sequence number.
4645 * if 'add_backref' is true, also insert a backref from the
4646 * inode to the parent directory.
4648 int btrfs_add_link(struct btrfs_trans_handle *trans,
4649 struct inode *parent_inode, struct inode *inode,
4650 const char *name, int name_len, int add_backref, u64 index)
4653 struct btrfs_key key;
4654 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4656 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4657 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4659 key.objectid = inode->i_ino;
4660 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4664 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4665 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4666 key.objectid, root->root_key.objectid,
4667 parent_inode->i_ino,
4668 index, name, name_len);
4669 } else if (add_backref) {
4670 ret = btrfs_insert_inode_ref(trans, root,
4671 name, name_len, inode->i_ino,
4672 parent_inode->i_ino, index);
4676 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4677 parent_inode->i_ino, &key,
4678 btrfs_inode_type(inode), index);
4681 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4683 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4684 ret = btrfs_update_inode(trans, root, parent_inode);
4689 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4690 struct inode *dir, struct dentry *dentry,
4691 struct inode *inode, int backref, u64 index)
4693 int err = btrfs_add_link(trans, dir, inode,
4694 dentry->d_name.name, dentry->d_name.len,
4697 d_instantiate(dentry, inode);
4705 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4706 int mode, dev_t rdev)
4708 struct btrfs_trans_handle *trans;
4709 struct btrfs_root *root = BTRFS_I(dir)->root;
4710 struct inode *inode = NULL;
4714 unsigned long nr = 0;
4717 if (!new_valid_dev(rdev))
4720 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4725 * 2 for inode item and ref
4727 * 1 for xattr if selinux is on
4729 trans = btrfs_start_transaction(root, 5);
4731 return PTR_ERR(trans);
4733 btrfs_set_trans_block_group(trans, dir);
4735 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4736 dentry->d_name.len, dir->i_ino, objectid,
4737 BTRFS_I(dir)->block_group, mode, &index);
4738 err = PTR_ERR(inode);
4742 err = btrfs_init_inode_security(trans, inode, dir);
4748 btrfs_set_trans_block_group(trans, inode);
4749 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4753 inode->i_op = &btrfs_special_inode_operations;
4754 init_special_inode(inode, inode->i_mode, rdev);
4755 btrfs_update_inode(trans, root, inode);
4757 btrfs_update_inode_block_group(trans, inode);
4758 btrfs_update_inode_block_group(trans, dir);
4760 nr = trans->blocks_used;
4761 btrfs_end_transaction_throttle(trans, root);
4762 btrfs_btree_balance_dirty(root, nr);
4764 inode_dec_link_count(inode);
4770 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4771 int mode, struct nameidata *nd)
4773 struct btrfs_trans_handle *trans;
4774 struct btrfs_root *root = BTRFS_I(dir)->root;
4775 struct inode *inode = NULL;
4778 unsigned long nr = 0;
4782 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4786 * 2 for inode item and ref
4788 * 1 for xattr if selinux is on
4790 trans = btrfs_start_transaction(root, 5);
4792 return PTR_ERR(trans);
4794 btrfs_set_trans_block_group(trans, dir);
4796 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4797 dentry->d_name.len, dir->i_ino, objectid,
4798 BTRFS_I(dir)->block_group, mode, &index);
4799 err = PTR_ERR(inode);
4803 err = btrfs_init_inode_security(trans, inode, dir);
4809 btrfs_set_trans_block_group(trans, inode);
4810 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4814 inode->i_mapping->a_ops = &btrfs_aops;
4815 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4816 inode->i_fop = &btrfs_file_operations;
4817 inode->i_op = &btrfs_file_inode_operations;
4818 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4820 btrfs_update_inode_block_group(trans, inode);
4821 btrfs_update_inode_block_group(trans, dir);
4823 nr = trans->blocks_used;
4824 btrfs_end_transaction_throttle(trans, root);
4826 inode_dec_link_count(inode);
4829 btrfs_btree_balance_dirty(root, nr);
4833 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4834 struct dentry *dentry)
4836 struct btrfs_trans_handle *trans;
4837 struct btrfs_root *root = BTRFS_I(dir)->root;
4838 struct inode *inode = old_dentry->d_inode;
4840 unsigned long nr = 0;
4844 if (inode->i_nlink == 0)
4847 /* do not allow sys_link's with other subvols of the same device */
4848 if (root->objectid != BTRFS_I(inode)->root->objectid)
4851 if (inode->i_nlink == ~0U)
4854 btrfs_inc_nlink(inode);
4855 inode->i_ctime = CURRENT_TIME;
4857 err = btrfs_set_inode_index(dir, &index);
4862 * 2 items for inode and inode ref
4863 * 2 items for dir items
4864 * 1 item for parent inode
4866 trans = btrfs_start_transaction(root, 5);
4867 if (IS_ERR(trans)) {
4868 err = PTR_ERR(trans);
4872 btrfs_set_trans_block_group(trans, dir);
4875 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4880 struct dentry *parent = dget_parent(dentry);
4881 btrfs_update_inode_block_group(trans, dir);
4882 err = btrfs_update_inode(trans, root, inode);
4884 btrfs_log_new_name(trans, inode, NULL, parent);
4888 nr = trans->blocks_used;
4889 btrfs_end_transaction_throttle(trans, root);
4892 inode_dec_link_count(inode);
4895 btrfs_btree_balance_dirty(root, nr);
4899 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4901 struct inode *inode = NULL;
4902 struct btrfs_trans_handle *trans;
4903 struct btrfs_root *root = BTRFS_I(dir)->root;
4905 int drop_on_err = 0;
4908 unsigned long nr = 1;
4910 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4915 * 2 items for inode and ref
4916 * 2 items for dir items
4917 * 1 for xattr if selinux is on
4919 trans = btrfs_start_transaction(root, 5);
4921 return PTR_ERR(trans);
4922 btrfs_set_trans_block_group(trans, dir);
4924 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4925 dentry->d_name.len, dir->i_ino, objectid,
4926 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4928 if (IS_ERR(inode)) {
4929 err = PTR_ERR(inode);
4935 err = btrfs_init_inode_security(trans, inode, dir);
4939 inode->i_op = &btrfs_dir_inode_operations;
4940 inode->i_fop = &btrfs_dir_file_operations;
4941 btrfs_set_trans_block_group(trans, inode);
4943 btrfs_i_size_write(inode, 0);
4944 err = btrfs_update_inode(trans, root, inode);
4948 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4949 dentry->d_name.len, 0, index);
4953 d_instantiate(dentry, inode);
4955 btrfs_update_inode_block_group(trans, inode);
4956 btrfs_update_inode_block_group(trans, dir);
4959 nr = trans->blocks_used;
4960 btrfs_end_transaction_throttle(trans, root);
4963 btrfs_btree_balance_dirty(root, nr);
4967 /* helper for btfs_get_extent. Given an existing extent in the tree,
4968 * and an extent that you want to insert, deal with overlap and insert
4969 * the new extent into the tree.
4971 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4972 struct extent_map *existing,
4973 struct extent_map *em,
4974 u64 map_start, u64 map_len)
4978 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4979 start_diff = map_start - em->start;
4980 em->start = map_start;
4982 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4983 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4984 em->block_start += start_diff;
4985 em->block_len -= start_diff;
4987 return add_extent_mapping(em_tree, em);
4990 static noinline int uncompress_inline(struct btrfs_path *path,
4991 struct inode *inode, struct page *page,
4992 size_t pg_offset, u64 extent_offset,
4993 struct btrfs_file_extent_item *item)
4996 struct extent_buffer *leaf = path->nodes[0];
4999 unsigned long inline_size;
5003 WARN_ON(pg_offset != 0);
5004 compress_type = btrfs_file_extent_compression(leaf, item);
5005 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5006 inline_size = btrfs_file_extent_inline_item_len(leaf,
5007 btrfs_item_nr(leaf, path->slots[0]));
5008 tmp = kmalloc(inline_size, GFP_NOFS);
5009 ptr = btrfs_file_extent_inline_start(item);
5011 read_extent_buffer(leaf, tmp, ptr, inline_size);
5013 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5014 ret = btrfs_decompress(compress_type, tmp, page,
5015 extent_offset, inline_size, max_size);
5017 char *kaddr = kmap_atomic(page, KM_USER0);
5018 unsigned long copy_size = min_t(u64,
5019 PAGE_CACHE_SIZE - pg_offset,
5020 max_size - extent_offset);
5021 memset(kaddr + pg_offset, 0, copy_size);
5022 kunmap_atomic(kaddr, KM_USER0);
5029 * a bit scary, this does extent mapping from logical file offset to the disk.
5030 * the ugly parts come from merging extents from the disk with the in-ram
5031 * representation. This gets more complex because of the data=ordered code,
5032 * where the in-ram extents might be locked pending data=ordered completion.
5034 * This also copies inline extents directly into the page.
5037 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5038 size_t pg_offset, u64 start, u64 len,
5044 u64 extent_start = 0;
5046 u64 objectid = inode->i_ino;
5048 struct btrfs_path *path = NULL;
5049 struct btrfs_root *root = BTRFS_I(inode)->root;
5050 struct btrfs_file_extent_item *item;
5051 struct extent_buffer *leaf;
5052 struct btrfs_key found_key;
5053 struct extent_map *em = NULL;
5054 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5055 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5056 struct btrfs_trans_handle *trans = NULL;
5060 read_lock(&em_tree->lock);
5061 em = lookup_extent_mapping(em_tree, start, len);
5063 em->bdev = root->fs_info->fs_devices->latest_bdev;
5064 read_unlock(&em_tree->lock);
5067 if (em->start > start || em->start + em->len <= start)
5068 free_extent_map(em);
5069 else if (em->block_start == EXTENT_MAP_INLINE && page)
5070 free_extent_map(em);
5074 em = alloc_extent_map(GFP_NOFS);
5079 em->bdev = root->fs_info->fs_devices->latest_bdev;
5080 em->start = EXTENT_MAP_HOLE;
5081 em->orig_start = EXTENT_MAP_HOLE;
5083 em->block_len = (u64)-1;
5086 path = btrfs_alloc_path();
5090 ret = btrfs_lookup_file_extent(trans, root, path,
5091 objectid, start, trans != NULL);
5098 if (path->slots[0] == 0)
5103 leaf = path->nodes[0];
5104 item = btrfs_item_ptr(leaf, path->slots[0],
5105 struct btrfs_file_extent_item);
5106 /* are we inside the extent that was found? */
5107 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5108 found_type = btrfs_key_type(&found_key);
5109 if (found_key.objectid != objectid ||
5110 found_type != BTRFS_EXTENT_DATA_KEY) {
5114 found_type = btrfs_file_extent_type(leaf, item);
5115 extent_start = found_key.offset;
5116 compress_type = btrfs_file_extent_compression(leaf, item);
5117 if (found_type == BTRFS_FILE_EXTENT_REG ||
5118 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5119 extent_end = extent_start +
5120 btrfs_file_extent_num_bytes(leaf, item);
5121 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5123 size = btrfs_file_extent_inline_len(leaf, item);
5124 extent_end = (extent_start + size + root->sectorsize - 1) &
5125 ~((u64)root->sectorsize - 1);
5128 if (start >= extent_end) {
5130 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5131 ret = btrfs_next_leaf(root, path);
5138 leaf = path->nodes[0];
5140 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5141 if (found_key.objectid != objectid ||
5142 found_key.type != BTRFS_EXTENT_DATA_KEY)
5144 if (start + len <= found_key.offset)
5147 em->len = found_key.offset - start;
5151 if (found_type == BTRFS_FILE_EXTENT_REG ||
5152 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5153 em->start = extent_start;
5154 em->len = extent_end - extent_start;
5155 em->orig_start = extent_start -
5156 btrfs_file_extent_offset(leaf, item);
5157 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5159 em->block_start = EXTENT_MAP_HOLE;
5162 if (compress_type != BTRFS_COMPRESS_NONE) {
5163 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5164 em->compress_type = compress_type;
5165 em->block_start = bytenr;
5166 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5169 bytenr += btrfs_file_extent_offset(leaf, item);
5170 em->block_start = bytenr;
5171 em->block_len = em->len;
5172 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5173 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5176 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5180 size_t extent_offset;
5183 em->block_start = EXTENT_MAP_INLINE;
5184 if (!page || create) {
5185 em->start = extent_start;
5186 em->len = extent_end - extent_start;
5190 size = btrfs_file_extent_inline_len(leaf, item);
5191 extent_offset = page_offset(page) + pg_offset - extent_start;
5192 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5193 size - extent_offset);
5194 em->start = extent_start + extent_offset;
5195 em->len = (copy_size + root->sectorsize - 1) &
5196 ~((u64)root->sectorsize - 1);
5197 em->orig_start = EXTENT_MAP_INLINE;
5198 if (compress_type) {
5199 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5200 em->compress_type = compress_type;
5202 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5203 if (create == 0 && !PageUptodate(page)) {
5204 if (btrfs_file_extent_compression(leaf, item) !=
5205 BTRFS_COMPRESS_NONE) {
5206 ret = uncompress_inline(path, inode, page,
5208 extent_offset, item);
5212 read_extent_buffer(leaf, map + pg_offset, ptr,
5214 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5215 memset(map + pg_offset + copy_size, 0,
5216 PAGE_CACHE_SIZE - pg_offset -
5221 flush_dcache_page(page);
5222 } else if (create && PageUptodate(page)) {
5226 free_extent_map(em);
5228 btrfs_release_path(root, path);
5229 trans = btrfs_join_transaction(root, 1);
5231 return ERR_CAST(trans);
5235 write_extent_buffer(leaf, map + pg_offset, ptr,
5238 btrfs_mark_buffer_dirty(leaf);
5240 set_extent_uptodate(io_tree, em->start,
5241 extent_map_end(em) - 1, GFP_NOFS);
5244 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5251 em->block_start = EXTENT_MAP_HOLE;
5252 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5254 btrfs_release_path(root, path);
5255 if (em->start > start || extent_map_end(em) <= start) {
5256 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5257 "[%llu %llu]\n", (unsigned long long)em->start,
5258 (unsigned long long)em->len,
5259 (unsigned long long)start,
5260 (unsigned long long)len);
5266 write_lock(&em_tree->lock);
5267 ret = add_extent_mapping(em_tree, em);
5268 /* it is possible that someone inserted the extent into the tree
5269 * while we had the lock dropped. It is also possible that
5270 * an overlapping map exists in the tree
5272 if (ret == -EEXIST) {
5273 struct extent_map *existing;
5277 existing = lookup_extent_mapping(em_tree, start, len);
5278 if (existing && (existing->start > start ||
5279 existing->start + existing->len <= start)) {
5280 free_extent_map(existing);
5284 existing = lookup_extent_mapping(em_tree, em->start,
5287 err = merge_extent_mapping(em_tree, existing,
5290 free_extent_map(existing);
5292 free_extent_map(em);
5297 free_extent_map(em);
5301 free_extent_map(em);
5306 write_unlock(&em_tree->lock);
5309 trace_btrfs_get_extent(root, em);
5312 btrfs_free_path(path);
5314 ret = btrfs_end_transaction(trans, root);
5319 free_extent_map(em);
5320 return ERR_PTR(err);
5325 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5326 size_t pg_offset, u64 start, u64 len,
5329 struct extent_map *em;
5330 struct extent_map *hole_em = NULL;
5331 u64 range_start = start;
5337 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5342 * if our em maps to a hole, there might
5343 * actually be delalloc bytes behind it
5345 if (em->block_start != EXTENT_MAP_HOLE)
5351 /* check to see if we've wrapped (len == -1 or similar) */
5360 /* ok, we didn't find anything, lets look for delalloc */
5361 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5362 end, len, EXTENT_DELALLOC, 1);
5363 found_end = range_start + found;
5364 if (found_end < range_start)
5365 found_end = (u64)-1;
5368 * we didn't find anything useful, return
5369 * the original results from get_extent()
5371 if (range_start > end || found_end <= start) {
5377 /* adjust the range_start to make sure it doesn't
5378 * go backwards from the start they passed in
5380 range_start = max(start,range_start);
5381 found = found_end - range_start;
5384 u64 hole_start = start;
5387 em = alloc_extent_map(GFP_NOFS);
5393 * when btrfs_get_extent can't find anything it
5394 * returns one huge hole
5396 * make sure what it found really fits our range, and
5397 * adjust to make sure it is based on the start from
5401 u64 calc_end = extent_map_end(hole_em);
5403 if (calc_end <= start || (hole_em->start > end)) {
5404 free_extent_map(hole_em);
5407 hole_start = max(hole_em->start, start);
5408 hole_len = calc_end - hole_start;
5412 if (hole_em && range_start > hole_start) {
5413 /* our hole starts before our delalloc, so we
5414 * have to return just the parts of the hole
5415 * that go until the delalloc starts
5417 em->len = min(hole_len,
5418 range_start - hole_start);
5419 em->start = hole_start;
5420 em->orig_start = hole_start;
5422 * don't adjust block start at all,
5423 * it is fixed at EXTENT_MAP_HOLE
5425 em->block_start = hole_em->block_start;
5426 em->block_len = hole_len;
5428 em->start = range_start;
5430 em->orig_start = range_start;
5431 em->block_start = EXTENT_MAP_DELALLOC;
5432 em->block_len = found;
5434 } else if (hole_em) {
5439 free_extent_map(hole_em);
5441 free_extent_map(em);
5442 return ERR_PTR(err);
5447 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5450 struct btrfs_root *root = BTRFS_I(inode)->root;
5451 struct btrfs_trans_handle *trans;
5452 struct extent_map *em;
5453 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5454 struct btrfs_key ins;
5458 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5460 trans = btrfs_join_transaction(root, 0);
5462 return ERR_CAST(trans);
5464 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5466 alloc_hint = get_extent_allocation_hint(inode, start, len);
5467 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5468 alloc_hint, (u64)-1, &ins, 1);
5474 em = alloc_extent_map(GFP_NOFS);
5476 em = ERR_PTR(-ENOMEM);
5481 em->orig_start = em->start;
5482 em->len = ins.offset;
5484 em->block_start = ins.objectid;
5485 em->block_len = ins.offset;
5486 em->bdev = root->fs_info->fs_devices->latest_bdev;
5487 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5490 write_lock(&em_tree->lock);
5491 ret = add_extent_mapping(em_tree, em);
5492 write_unlock(&em_tree->lock);
5495 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5498 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5499 ins.offset, ins.offset, 0);
5501 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5505 btrfs_end_transaction(trans, root);
5510 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5511 * block must be cow'd
5513 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5514 struct inode *inode, u64 offset, u64 len)
5516 struct btrfs_path *path;
5518 struct extent_buffer *leaf;
5519 struct btrfs_root *root = BTRFS_I(inode)->root;
5520 struct btrfs_file_extent_item *fi;
5521 struct btrfs_key key;
5529 path = btrfs_alloc_path();
5533 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5538 slot = path->slots[0];
5541 /* can't find the item, must cow */
5548 leaf = path->nodes[0];
5549 btrfs_item_key_to_cpu(leaf, &key, slot);
5550 if (key.objectid != inode->i_ino ||
5551 key.type != BTRFS_EXTENT_DATA_KEY) {
5552 /* not our file or wrong item type, must cow */
5556 if (key.offset > offset) {
5557 /* Wrong offset, must cow */
5561 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5562 found_type = btrfs_file_extent_type(leaf, fi);
5563 if (found_type != BTRFS_FILE_EXTENT_REG &&
5564 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5565 /* not a regular extent, must cow */
5568 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5569 backref_offset = btrfs_file_extent_offset(leaf, fi);
5571 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5572 if (extent_end < offset + len) {
5573 /* extent doesn't include our full range, must cow */
5577 if (btrfs_extent_readonly(root, disk_bytenr))
5581 * look for other files referencing this extent, if we
5582 * find any we must cow
5584 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5585 key.offset - backref_offset, disk_bytenr))
5589 * adjust disk_bytenr and num_bytes to cover just the bytes
5590 * in this extent we are about to write. If there
5591 * are any csums in that range we have to cow in order
5592 * to keep the csums correct
5594 disk_bytenr += backref_offset;
5595 disk_bytenr += offset - key.offset;
5596 num_bytes = min(offset + len, extent_end) - offset;
5597 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5600 * all of the above have passed, it is safe to overwrite this extent
5605 btrfs_free_path(path);
5609 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5610 struct buffer_head *bh_result, int create)
5612 struct extent_map *em;
5613 struct btrfs_root *root = BTRFS_I(inode)->root;
5614 u64 start = iblock << inode->i_blkbits;
5615 u64 len = bh_result->b_size;
5616 struct btrfs_trans_handle *trans;
5618 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5623 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5624 * io. INLINE is special, and we could probably kludge it in here, but
5625 * it's still buffered so for safety lets just fall back to the generic
5628 * For COMPRESSED we _have_ to read the entire extent in so we can
5629 * decompress it, so there will be buffering required no matter what we
5630 * do, so go ahead and fallback to buffered.
5632 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5633 * to buffered IO. Don't blame me, this is the price we pay for using
5636 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5637 em->block_start == EXTENT_MAP_INLINE) {
5638 free_extent_map(em);
5642 /* Just a good old fashioned hole, return */
5643 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5644 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5645 free_extent_map(em);
5646 /* DIO will do one hole at a time, so just unlock a sector */
5647 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5648 start + root->sectorsize - 1, GFP_NOFS);
5653 * We don't allocate a new extent in the following cases
5655 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5657 * 2) The extent is marked as PREALLOC. We're good to go here and can
5658 * just use the extent.
5662 len = em->len - (start - em->start);
5666 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5667 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5668 em->block_start != EXTENT_MAP_HOLE)) {
5673 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5674 type = BTRFS_ORDERED_PREALLOC;
5676 type = BTRFS_ORDERED_NOCOW;
5677 len = min(len, em->len - (start - em->start));
5678 block_start = em->block_start + (start - em->start);
5681 * we're not going to log anything, but we do need
5682 * to make sure the current transaction stays open
5683 * while we look for nocow cross refs
5685 trans = btrfs_join_transaction(root, 0);
5689 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5690 ret = btrfs_add_ordered_extent_dio(inode, start,
5691 block_start, len, len, type);
5692 btrfs_end_transaction(trans, root);
5694 free_extent_map(em);
5699 btrfs_end_transaction(trans, root);
5703 * this will cow the extent, reset the len in case we changed
5706 len = bh_result->b_size;
5707 free_extent_map(em);
5708 em = btrfs_new_extent_direct(inode, start, len);
5711 len = min(len, em->len - (start - em->start));
5713 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5714 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5717 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5719 bh_result->b_size = len;
5720 bh_result->b_bdev = em->bdev;
5721 set_buffer_mapped(bh_result);
5722 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5723 set_buffer_new(bh_result);
5725 free_extent_map(em);
5730 struct btrfs_dio_private {
5731 struct inode *inode;
5738 /* number of bios pending for this dio */
5739 atomic_t pending_bios;
5744 struct bio *orig_bio;
5747 static void btrfs_endio_direct_read(struct bio *bio, int err)
5749 struct btrfs_dio_private *dip = bio->bi_private;
5750 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5751 struct bio_vec *bvec = bio->bi_io_vec;
5752 struct inode *inode = dip->inode;
5753 struct btrfs_root *root = BTRFS_I(inode)->root;
5755 u32 *private = dip->csums;
5757 start = dip->logical_offset;
5759 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5760 struct page *page = bvec->bv_page;
5763 unsigned long flags;
5765 local_irq_save(flags);
5766 kaddr = kmap_atomic(page, KM_IRQ0);
5767 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5768 csum, bvec->bv_len);
5769 btrfs_csum_final(csum, (char *)&csum);
5770 kunmap_atomic(kaddr, KM_IRQ0);
5771 local_irq_restore(flags);
5773 flush_dcache_page(bvec->bv_page);
5774 if (csum != *private) {
5775 printk(KERN_ERR "btrfs csum failed ino %lu off"
5776 " %llu csum %u private %u\n",
5777 inode->i_ino, (unsigned long long)start,
5783 start += bvec->bv_len;
5786 } while (bvec <= bvec_end);
5788 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5789 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5790 bio->bi_private = dip->private;
5795 /* If we had a csum failure make sure to clear the uptodate flag */
5797 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5798 dio_end_io(bio, err);
5801 static void btrfs_endio_direct_write(struct bio *bio, int err)
5803 struct btrfs_dio_private *dip = bio->bi_private;
5804 struct inode *inode = dip->inode;
5805 struct btrfs_root *root = BTRFS_I(inode)->root;
5806 struct btrfs_trans_handle *trans;
5807 struct btrfs_ordered_extent *ordered = NULL;
5808 struct extent_state *cached_state = NULL;
5809 u64 ordered_offset = dip->logical_offset;
5810 u64 ordered_bytes = dip->bytes;
5816 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5824 trans = btrfs_join_transaction(root, 1);
5825 if (IS_ERR(trans)) {
5829 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5831 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5832 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5834 ret = btrfs_update_inode(trans, root, inode);
5839 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5840 ordered->file_offset + ordered->len - 1, 0,
5841 &cached_state, GFP_NOFS);
5843 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5844 ret = btrfs_mark_extent_written(trans, inode,
5845 ordered->file_offset,
5846 ordered->file_offset +
5853 ret = insert_reserved_file_extent(trans, inode,
5854 ordered->file_offset,
5860 BTRFS_FILE_EXTENT_REG);
5861 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5862 ordered->file_offset, ordered->len);
5870 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5871 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5873 btrfs_update_inode(trans, root, inode);
5876 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5877 ordered->file_offset + ordered->len - 1,
5878 &cached_state, GFP_NOFS);
5880 btrfs_delalloc_release_metadata(inode, ordered->len);
5881 btrfs_end_transaction(trans, root);
5882 ordered_offset = ordered->file_offset + ordered->len;
5883 btrfs_put_ordered_extent(ordered);
5884 btrfs_put_ordered_extent(ordered);
5888 * our bio might span multiple ordered extents. If we haven't
5889 * completed the accounting for the whole dio, go back and try again
5891 if (ordered_offset < dip->logical_offset + dip->bytes) {
5892 ordered_bytes = dip->logical_offset + dip->bytes -
5897 bio->bi_private = dip->private;
5902 /* If we had an error make sure to clear the uptodate flag */
5904 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5905 dio_end_io(bio, err);
5908 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5909 struct bio *bio, int mirror_num,
5910 unsigned long bio_flags, u64 offset)
5913 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5919 static void btrfs_end_dio_bio(struct bio *bio, int err)
5921 struct btrfs_dio_private *dip = bio->bi_private;
5924 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5925 "sector %#Lx len %u err no %d\n",
5926 dip->inode->i_ino, bio->bi_rw,
5927 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5931 * before atomic variable goto zero, we must make sure
5932 * dip->errors is perceived to be set.
5934 smp_mb__before_atomic_dec();
5937 /* if there are more bios still pending for this dio, just exit */
5938 if (!atomic_dec_and_test(&dip->pending_bios))
5942 bio_io_error(dip->orig_bio);
5944 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5945 bio_endio(dip->orig_bio, 0);
5951 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5952 u64 first_sector, gfp_t gfp_flags)
5954 int nr_vecs = bio_get_nr_vecs(bdev);
5955 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5958 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5959 int rw, u64 file_offset, int skip_sum,
5962 int write = rw & REQ_WRITE;
5963 struct btrfs_root *root = BTRFS_I(inode)->root;
5967 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5971 if (write && !skip_sum) {
5972 ret = btrfs_wq_submit_bio(root->fs_info,
5973 inode, rw, bio, 0, 0,
5975 __btrfs_submit_bio_start_direct_io,
5976 __btrfs_submit_bio_done);
5978 } else if (!skip_sum) {
5979 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5980 file_offset, csums);
5985 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5991 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5994 struct inode *inode = dip->inode;
5995 struct btrfs_root *root = BTRFS_I(inode)->root;
5996 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5998 struct bio *orig_bio = dip->orig_bio;
5999 struct bio_vec *bvec = orig_bio->bi_io_vec;
6000 u64 start_sector = orig_bio->bi_sector;
6001 u64 file_offset = dip->logical_offset;
6005 u32 *csums = dip->csums;
6007 int write = rw & REQ_WRITE;
6009 map_length = orig_bio->bi_size;
6010 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6011 &map_length, NULL, 0);
6017 if (map_length >= orig_bio->bi_size) {
6022 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6025 bio->bi_private = dip;
6026 bio->bi_end_io = btrfs_end_dio_bio;
6027 atomic_inc(&dip->pending_bios);
6029 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6030 if (unlikely(map_length < submit_len + bvec->bv_len ||
6031 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6032 bvec->bv_offset) < bvec->bv_len)) {
6034 * inc the count before we submit the bio so
6035 * we know the end IO handler won't happen before
6036 * we inc the count. Otherwise, the dip might get freed
6037 * before we're done setting it up
6039 atomic_inc(&dip->pending_bios);
6040 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6041 file_offset, skip_sum,
6045 atomic_dec(&dip->pending_bios);
6049 /* Write's use the ordered csums */
6050 if (!write && !skip_sum)
6051 csums = csums + nr_pages;
6052 start_sector += submit_len >> 9;
6053 file_offset += submit_len;
6058 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6059 start_sector, GFP_NOFS);
6062 bio->bi_private = dip;
6063 bio->bi_end_io = btrfs_end_dio_bio;
6065 map_length = orig_bio->bi_size;
6066 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6067 &map_length, NULL, 0);
6073 submit_len += bvec->bv_len;
6080 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6089 * before atomic variable goto zero, we must
6090 * make sure dip->errors is perceived to be set.
6092 smp_mb__before_atomic_dec();
6093 if (atomic_dec_and_test(&dip->pending_bios))
6094 bio_io_error(dip->orig_bio);
6096 /* bio_end_io() will handle error, so we needn't return it */
6100 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6103 struct btrfs_root *root = BTRFS_I(inode)->root;
6104 struct btrfs_dio_private *dip;
6105 struct bio_vec *bvec = bio->bi_io_vec;
6107 int write = rw & REQ_WRITE;
6110 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6112 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6119 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6120 if (!write && !skip_sum) {
6121 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6129 dip->private = bio->bi_private;
6131 dip->logical_offset = file_offset;
6135 dip->bytes += bvec->bv_len;
6137 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6139 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6140 bio->bi_private = dip;
6142 dip->orig_bio = bio;
6143 atomic_set(&dip->pending_bios, 0);
6146 bio->bi_end_io = btrfs_endio_direct_write;
6148 bio->bi_end_io = btrfs_endio_direct_read;
6150 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6155 * If this is a write, we need to clean up the reserved space and kill
6156 * the ordered extent.
6159 struct btrfs_ordered_extent *ordered;
6160 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6161 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6162 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6163 btrfs_free_reserved_extent(root, ordered->start,
6165 btrfs_put_ordered_extent(ordered);
6166 btrfs_put_ordered_extent(ordered);
6168 bio_endio(bio, ret);
6171 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6172 const struct iovec *iov, loff_t offset,
6173 unsigned long nr_segs)
6178 unsigned blocksize_mask = root->sectorsize - 1;
6179 ssize_t retval = -EINVAL;
6180 loff_t end = offset;
6182 if (offset & blocksize_mask)
6185 /* Check the memory alignment. Blocks cannot straddle pages */
6186 for (seg = 0; seg < nr_segs; seg++) {
6187 addr = (unsigned long)iov[seg].iov_base;
6188 size = iov[seg].iov_len;
6190 if ((addr & blocksize_mask) || (size & blocksize_mask))
6197 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6198 const struct iovec *iov, loff_t offset,
6199 unsigned long nr_segs)
6201 struct file *file = iocb->ki_filp;
6202 struct inode *inode = file->f_mapping->host;
6203 struct btrfs_ordered_extent *ordered;
6204 struct extent_state *cached_state = NULL;
6205 u64 lockstart, lockend;
6207 int writing = rw & WRITE;
6209 size_t count = iov_length(iov, nr_segs);
6211 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6217 lockend = offset + count - 1;
6220 ret = btrfs_delalloc_reserve_space(inode, count);
6226 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6227 0, &cached_state, GFP_NOFS);
6229 * We're concerned with the entire range that we're going to be
6230 * doing DIO to, so we need to make sure theres no ordered
6231 * extents in this range.
6233 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6234 lockend - lockstart + 1);
6237 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6238 &cached_state, GFP_NOFS);
6239 btrfs_start_ordered_extent(inode, ordered, 1);
6240 btrfs_put_ordered_extent(ordered);
6245 * we don't use btrfs_set_extent_delalloc because we don't want
6246 * the dirty or uptodate bits
6249 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6250 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6251 EXTENT_DELALLOC, 0, NULL, &cached_state,
6254 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6255 lockend, EXTENT_LOCKED | write_bits,
6256 1, 0, &cached_state, GFP_NOFS);
6261 free_extent_state(cached_state);
6262 cached_state = NULL;
6264 ret = __blockdev_direct_IO(rw, iocb, inode,
6265 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6266 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6267 btrfs_submit_direct, 0);
6269 if (ret < 0 && ret != -EIOCBQUEUED) {
6270 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6271 offset + iov_length(iov, nr_segs) - 1,
6272 EXTENT_LOCKED | write_bits, 1, 0,
6273 &cached_state, GFP_NOFS);
6274 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6276 * We're falling back to buffered, unlock the section we didn't
6279 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6280 offset + iov_length(iov, nr_segs) - 1,
6281 EXTENT_LOCKED | write_bits, 1, 0,
6282 &cached_state, GFP_NOFS);
6285 free_extent_state(cached_state);
6289 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6290 __u64 start, __u64 len)
6292 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6295 int btrfs_readpage(struct file *file, struct page *page)
6297 struct extent_io_tree *tree;
6298 tree = &BTRFS_I(page->mapping->host)->io_tree;
6299 return extent_read_full_page(tree, page, btrfs_get_extent);
6302 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6304 struct extent_io_tree *tree;
6307 if (current->flags & PF_MEMALLOC) {
6308 redirty_page_for_writepage(wbc, page);
6312 tree = &BTRFS_I(page->mapping->host)->io_tree;
6313 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6316 int btrfs_writepages(struct address_space *mapping,
6317 struct writeback_control *wbc)
6319 struct extent_io_tree *tree;
6321 tree = &BTRFS_I(mapping->host)->io_tree;
6322 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6326 btrfs_readpages(struct file *file, struct address_space *mapping,
6327 struct list_head *pages, unsigned nr_pages)
6329 struct extent_io_tree *tree;
6330 tree = &BTRFS_I(mapping->host)->io_tree;
6331 return extent_readpages(tree, mapping, pages, nr_pages,
6334 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6336 struct extent_io_tree *tree;
6337 struct extent_map_tree *map;
6340 tree = &BTRFS_I(page->mapping->host)->io_tree;
6341 map = &BTRFS_I(page->mapping->host)->extent_tree;
6342 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6344 ClearPagePrivate(page);
6345 set_page_private(page, 0);
6346 page_cache_release(page);
6351 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6353 if (PageWriteback(page) || PageDirty(page))
6355 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6358 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6360 struct extent_io_tree *tree;
6361 struct btrfs_ordered_extent *ordered;
6362 struct extent_state *cached_state = NULL;
6363 u64 page_start = page_offset(page);
6364 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6368 * we have the page locked, so new writeback can't start,
6369 * and the dirty bit won't be cleared while we are here.
6371 * Wait for IO on this page so that we can safely clear
6372 * the PagePrivate2 bit and do ordered accounting
6374 wait_on_page_writeback(page);
6376 tree = &BTRFS_I(page->mapping->host)->io_tree;
6378 btrfs_releasepage(page, GFP_NOFS);
6381 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6383 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6387 * IO on this page will never be started, so we need
6388 * to account for any ordered extents now
6390 clear_extent_bit(tree, page_start, page_end,
6391 EXTENT_DIRTY | EXTENT_DELALLOC |
6392 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6393 &cached_state, GFP_NOFS);
6395 * whoever cleared the private bit is responsible
6396 * for the finish_ordered_io
6398 if (TestClearPagePrivate2(page)) {
6399 btrfs_finish_ordered_io(page->mapping->host,
6400 page_start, page_end);
6402 btrfs_put_ordered_extent(ordered);
6403 cached_state = NULL;
6404 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6407 clear_extent_bit(tree, page_start, page_end,
6408 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6409 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6410 __btrfs_releasepage(page, GFP_NOFS);
6412 ClearPageChecked(page);
6413 if (PagePrivate(page)) {
6414 ClearPagePrivate(page);
6415 set_page_private(page, 0);
6416 page_cache_release(page);
6421 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6422 * called from a page fault handler when a page is first dirtied. Hence we must
6423 * be careful to check for EOF conditions here. We set the page up correctly
6424 * for a written page which means we get ENOSPC checking when writing into
6425 * holes and correct delalloc and unwritten extent mapping on filesystems that
6426 * support these features.
6428 * We are not allowed to take the i_mutex here so we have to play games to
6429 * protect against truncate races as the page could now be beyond EOF. Because
6430 * vmtruncate() writes the inode size before removing pages, once we have the
6431 * page lock we can determine safely if the page is beyond EOF. If it is not
6432 * beyond EOF, then the page is guaranteed safe against truncation until we
6435 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6437 struct page *page = vmf->page;
6438 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6439 struct btrfs_root *root = BTRFS_I(inode)->root;
6440 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6441 struct btrfs_ordered_extent *ordered;
6442 struct extent_state *cached_state = NULL;
6444 unsigned long zero_start;
6450 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6454 else /* -ENOSPC, -EIO, etc */
6455 ret = VM_FAULT_SIGBUS;
6459 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6462 size = i_size_read(inode);
6463 page_start = page_offset(page);
6464 page_end = page_start + PAGE_CACHE_SIZE - 1;
6466 if ((page->mapping != inode->i_mapping) ||
6467 (page_start >= size)) {
6468 /* page got truncated out from underneath us */
6471 wait_on_page_writeback(page);
6473 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6475 set_page_extent_mapped(page);
6478 * we can't set the delalloc bits if there are pending ordered
6479 * extents. Drop our locks and wait for them to finish
6481 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6483 unlock_extent_cached(io_tree, page_start, page_end,
6484 &cached_state, GFP_NOFS);
6486 btrfs_start_ordered_extent(inode, ordered, 1);
6487 btrfs_put_ordered_extent(ordered);
6492 * XXX - page_mkwrite gets called every time the page is dirtied, even
6493 * if it was already dirty, so for space accounting reasons we need to
6494 * clear any delalloc bits for the range we are fixing to save. There
6495 * is probably a better way to do this, but for now keep consistent with
6496 * prepare_pages in the normal write path.
6498 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6499 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6500 0, 0, &cached_state, GFP_NOFS);
6502 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6505 unlock_extent_cached(io_tree, page_start, page_end,
6506 &cached_state, GFP_NOFS);
6507 ret = VM_FAULT_SIGBUS;
6512 /* page is wholly or partially inside EOF */
6513 if (page_start + PAGE_CACHE_SIZE > size)
6514 zero_start = size & ~PAGE_CACHE_MASK;
6516 zero_start = PAGE_CACHE_SIZE;
6518 if (zero_start != PAGE_CACHE_SIZE) {
6520 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6521 flush_dcache_page(page);
6524 ClearPageChecked(page);
6525 set_page_dirty(page);
6526 SetPageUptodate(page);
6528 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6529 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6531 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6535 return VM_FAULT_LOCKED;
6537 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6542 static int btrfs_truncate(struct inode *inode)
6544 struct btrfs_root *root = BTRFS_I(inode)->root;
6547 struct btrfs_trans_handle *trans;
6549 u64 mask = root->sectorsize - 1;
6551 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6555 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6556 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6558 trans = btrfs_start_transaction(root, 5);
6560 return PTR_ERR(trans);
6562 btrfs_set_trans_block_group(trans, inode);
6564 ret = btrfs_orphan_add(trans, inode);
6566 btrfs_end_transaction(trans, root);
6570 nr = trans->blocks_used;
6571 btrfs_end_transaction(trans, root);
6572 btrfs_btree_balance_dirty(root, nr);
6574 /* Now start a transaction for the truncate */
6575 trans = btrfs_start_transaction(root, 0);
6577 return PTR_ERR(trans);
6578 btrfs_set_trans_block_group(trans, inode);
6579 trans->block_rsv = root->orphan_block_rsv;
6582 * setattr is responsible for setting the ordered_data_close flag,
6583 * but that is only tested during the last file release. That
6584 * could happen well after the next commit, leaving a great big
6585 * window where new writes may get lost if someone chooses to write
6586 * to this file after truncating to zero
6588 * The inode doesn't have any dirty data here, and so if we commit
6589 * this is a noop. If someone immediately starts writing to the inode
6590 * it is very likely we'll catch some of their writes in this
6591 * transaction, and the commit will find this file on the ordered
6592 * data list with good things to send down.
6594 * This is a best effort solution, there is still a window where
6595 * using truncate to replace the contents of the file will
6596 * end up with a zero length file after a crash.
6598 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6599 btrfs_add_ordered_operation(trans, root, inode);
6603 trans = btrfs_start_transaction(root, 0);
6605 return PTR_ERR(trans);
6606 btrfs_set_trans_block_group(trans, inode);
6607 trans->block_rsv = root->orphan_block_rsv;
6610 ret = btrfs_block_rsv_check(trans, root,
6611 root->orphan_block_rsv, 0, 5);
6612 if (ret == -EAGAIN) {
6613 ret = btrfs_commit_transaction(trans, root);
6623 ret = btrfs_truncate_inode_items(trans, root, inode,
6625 BTRFS_EXTENT_DATA_KEY);
6626 if (ret != -EAGAIN) {
6631 ret = btrfs_update_inode(trans, root, inode);
6637 nr = trans->blocks_used;
6638 btrfs_end_transaction(trans, root);
6640 btrfs_btree_balance_dirty(root, nr);
6643 if (ret == 0 && inode->i_nlink > 0) {
6644 ret = btrfs_orphan_del(trans, inode);
6647 } else if (ret && inode->i_nlink > 0) {
6649 * Failed to do the truncate, remove us from the in memory
6652 ret = btrfs_orphan_del(NULL, inode);
6655 ret = btrfs_update_inode(trans, root, inode);
6659 nr = trans->blocks_used;
6660 ret = btrfs_end_transaction_throttle(trans, root);
6663 btrfs_btree_balance_dirty(root, nr);
6669 * create a new subvolume directory/inode (helper for the ioctl).
6671 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6672 struct btrfs_root *new_root,
6673 u64 new_dirid, u64 alloc_hint)
6675 struct inode *inode;
6679 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6680 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6682 return PTR_ERR(inode);
6683 inode->i_op = &btrfs_dir_inode_operations;
6684 inode->i_fop = &btrfs_dir_file_operations;
6687 btrfs_i_size_write(inode, 0);
6689 err = btrfs_update_inode(trans, new_root, inode);
6696 /* helper function for file defrag and space balancing. This
6697 * forces readahead on a given range of bytes in an inode
6699 unsigned long btrfs_force_ra(struct address_space *mapping,
6700 struct file_ra_state *ra, struct file *file,
6701 pgoff_t offset, pgoff_t last_index)
6703 pgoff_t req_size = last_index - offset + 1;
6705 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6706 return offset + req_size;
6709 struct inode *btrfs_alloc_inode(struct super_block *sb)
6711 struct btrfs_inode *ei;
6712 struct inode *inode;
6714 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6719 ei->space_info = NULL;
6723 ei->last_sub_trans = 0;
6724 ei->logged_trans = 0;
6725 ei->delalloc_bytes = 0;
6726 ei->reserved_bytes = 0;
6727 ei->disk_i_size = 0;
6729 ei->index_cnt = (u64)-1;
6730 ei->last_unlink_trans = 0;
6732 atomic_set(&ei->outstanding_extents, 0);
6733 atomic_set(&ei->reserved_extents, 0);
6735 ei->ordered_data_close = 0;
6736 ei->orphan_meta_reserved = 0;
6737 ei->dummy_inode = 0;
6738 ei->force_compress = BTRFS_COMPRESS_NONE;
6740 inode = &ei->vfs_inode;
6741 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6742 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6743 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6744 mutex_init(&ei->log_mutex);
6745 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6746 INIT_LIST_HEAD(&ei->i_orphan);
6747 INIT_LIST_HEAD(&ei->delalloc_inodes);
6748 INIT_LIST_HEAD(&ei->ordered_operations);
6749 RB_CLEAR_NODE(&ei->rb_node);
6754 static void btrfs_i_callback(struct rcu_head *head)
6756 struct inode *inode = container_of(head, struct inode, i_rcu);
6757 INIT_LIST_HEAD(&inode->i_dentry);
6758 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6761 void btrfs_destroy_inode(struct inode *inode)
6763 struct btrfs_ordered_extent *ordered;
6764 struct btrfs_root *root = BTRFS_I(inode)->root;
6766 WARN_ON(!list_empty(&inode->i_dentry));
6767 WARN_ON(inode->i_data.nrpages);
6768 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6769 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6772 * This can happen where we create an inode, but somebody else also
6773 * created the same inode and we need to destroy the one we already
6780 * Make sure we're properly removed from the ordered operation
6784 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6785 spin_lock(&root->fs_info->ordered_extent_lock);
6786 list_del_init(&BTRFS_I(inode)->ordered_operations);
6787 spin_unlock(&root->fs_info->ordered_extent_lock);
6790 if (root == root->fs_info->tree_root) {
6791 struct btrfs_block_group_cache *block_group;
6793 block_group = btrfs_lookup_block_group(root->fs_info,
6794 BTRFS_I(inode)->block_group);
6795 if (block_group && block_group->inode == inode) {
6796 spin_lock(&block_group->lock);
6797 block_group->inode = NULL;
6798 spin_unlock(&block_group->lock);
6799 btrfs_put_block_group(block_group);
6800 } else if (block_group) {
6801 btrfs_put_block_group(block_group);
6805 spin_lock(&root->orphan_lock);
6806 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6807 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6809 list_del_init(&BTRFS_I(inode)->i_orphan);
6811 spin_unlock(&root->orphan_lock);
6814 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6818 printk(KERN_ERR "btrfs found ordered "
6819 "extent %llu %llu on inode cleanup\n",
6820 (unsigned long long)ordered->file_offset,
6821 (unsigned long long)ordered->len);
6822 btrfs_remove_ordered_extent(inode, ordered);
6823 btrfs_put_ordered_extent(ordered);
6824 btrfs_put_ordered_extent(ordered);
6827 inode_tree_del(inode);
6828 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6830 call_rcu(&inode->i_rcu, btrfs_i_callback);
6833 int btrfs_drop_inode(struct inode *inode)
6835 struct btrfs_root *root = BTRFS_I(inode)->root;
6837 if (btrfs_root_refs(&root->root_item) == 0 &&
6838 root != root->fs_info->tree_root)
6841 return generic_drop_inode(inode);
6844 static void init_once(void *foo)
6846 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6848 inode_init_once(&ei->vfs_inode);
6851 void btrfs_destroy_cachep(void)
6853 if (btrfs_inode_cachep)
6854 kmem_cache_destroy(btrfs_inode_cachep);
6855 if (btrfs_trans_handle_cachep)
6856 kmem_cache_destroy(btrfs_trans_handle_cachep);
6857 if (btrfs_transaction_cachep)
6858 kmem_cache_destroy(btrfs_transaction_cachep);
6859 if (btrfs_path_cachep)
6860 kmem_cache_destroy(btrfs_path_cachep);
6861 if (btrfs_free_space_cachep)
6862 kmem_cache_destroy(btrfs_free_space_cachep);
6865 int btrfs_init_cachep(void)
6867 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6868 sizeof(struct btrfs_inode), 0,
6869 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6870 if (!btrfs_inode_cachep)
6873 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6874 sizeof(struct btrfs_trans_handle), 0,
6875 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6876 if (!btrfs_trans_handle_cachep)
6879 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6880 sizeof(struct btrfs_transaction), 0,
6881 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6882 if (!btrfs_transaction_cachep)
6885 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6886 sizeof(struct btrfs_path), 0,
6887 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6888 if (!btrfs_path_cachep)
6891 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6892 sizeof(struct btrfs_free_space), 0,
6893 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6894 if (!btrfs_free_space_cachep)
6899 btrfs_destroy_cachep();
6903 static int btrfs_getattr(struct vfsmount *mnt,
6904 struct dentry *dentry, struct kstat *stat)
6906 struct inode *inode = dentry->d_inode;
6907 generic_fillattr(inode, stat);
6908 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6909 stat->blksize = PAGE_CACHE_SIZE;
6910 stat->blocks = (inode_get_bytes(inode) +
6911 BTRFS_I(inode)->delalloc_bytes) >> 9;
6916 * If a file is moved, it will inherit the cow and compression flags of the new
6919 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6921 struct btrfs_inode *b_dir = BTRFS_I(dir);
6922 struct btrfs_inode *b_inode = BTRFS_I(inode);
6924 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6925 b_inode->flags |= BTRFS_INODE_NODATACOW;
6927 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6929 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6930 b_inode->flags |= BTRFS_INODE_COMPRESS;
6932 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6935 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6936 struct inode *new_dir, struct dentry *new_dentry)
6938 struct btrfs_trans_handle *trans;
6939 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6940 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6941 struct inode *new_inode = new_dentry->d_inode;
6942 struct inode *old_inode = old_dentry->d_inode;
6943 struct timespec ctime = CURRENT_TIME;
6948 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6951 /* we only allow rename subvolume link between subvolumes */
6952 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6955 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6956 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6959 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6960 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6963 * we're using rename to replace one file with another.
6964 * and the replacement file is large. Start IO on it now so
6965 * we don't add too much work to the end of the transaction
6967 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6968 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6969 filemap_flush(old_inode->i_mapping);
6971 /* close the racy window with snapshot create/destroy ioctl */
6972 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6973 down_read(&root->fs_info->subvol_sem);
6975 * We want to reserve the absolute worst case amount of items. So if
6976 * both inodes are subvols and we need to unlink them then that would
6977 * require 4 item modifications, but if they are both normal inodes it
6978 * would require 5 item modifications, so we'll assume their normal
6979 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6980 * should cover the worst case number of items we'll modify.
6982 trans = btrfs_start_transaction(root, 20);
6983 if (IS_ERR(trans)) {
6984 ret = PTR_ERR(trans);
6988 btrfs_set_trans_block_group(trans, new_dir);
6991 btrfs_record_root_in_trans(trans, dest);
6993 ret = btrfs_set_inode_index(new_dir, &index);
6997 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6998 /* force full log commit if subvolume involved. */
6999 root->fs_info->last_trans_log_full_commit = trans->transid;
7001 ret = btrfs_insert_inode_ref(trans, dest,
7002 new_dentry->d_name.name,
7003 new_dentry->d_name.len,
7005 new_dir->i_ino, index);
7009 * this is an ugly little race, but the rename is required
7010 * to make sure that if we crash, the inode is either at the
7011 * old name or the new one. pinning the log transaction lets
7012 * us make sure we don't allow a log commit to come in after
7013 * we unlink the name but before we add the new name back in.
7015 btrfs_pin_log_trans(root);
7018 * make sure the inode gets flushed if it is replacing
7021 if (new_inode && new_inode->i_size &&
7022 old_inode && S_ISREG(old_inode->i_mode)) {
7023 btrfs_add_ordered_operation(trans, root, old_inode);
7026 old_dir->i_ctime = old_dir->i_mtime = ctime;
7027 new_dir->i_ctime = new_dir->i_mtime = ctime;
7028 old_inode->i_ctime = ctime;
7030 if (old_dentry->d_parent != new_dentry->d_parent)
7031 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7033 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7034 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7035 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7036 old_dentry->d_name.name,
7037 old_dentry->d_name.len);
7039 ret = __btrfs_unlink_inode(trans, root, old_dir,
7040 old_dentry->d_inode,
7041 old_dentry->d_name.name,
7042 old_dentry->d_name.len);
7044 ret = btrfs_update_inode(trans, root, old_inode);
7049 new_inode->i_ctime = CURRENT_TIME;
7050 if (unlikely(new_inode->i_ino ==
7051 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7052 root_objectid = BTRFS_I(new_inode)->location.objectid;
7053 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7055 new_dentry->d_name.name,
7056 new_dentry->d_name.len);
7057 BUG_ON(new_inode->i_nlink == 0);
7059 ret = btrfs_unlink_inode(trans, dest, new_dir,
7060 new_dentry->d_inode,
7061 new_dentry->d_name.name,
7062 new_dentry->d_name.len);
7065 if (new_inode->i_nlink == 0) {
7066 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7071 fixup_inode_flags(new_dir, old_inode);
7073 ret = btrfs_add_link(trans, new_dir, old_inode,
7074 new_dentry->d_name.name,
7075 new_dentry->d_name.len, 0, index);
7078 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7079 struct dentry *parent = dget_parent(new_dentry);
7080 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7082 btrfs_end_log_trans(root);
7085 btrfs_end_transaction_throttle(trans, root);
7087 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7088 up_read(&root->fs_info->subvol_sem);
7094 * some fairly slow code that needs optimization. This walks the list
7095 * of all the inodes with pending delalloc and forces them to disk.
7097 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7099 struct list_head *head = &root->fs_info->delalloc_inodes;
7100 struct btrfs_inode *binode;
7101 struct inode *inode;
7103 if (root->fs_info->sb->s_flags & MS_RDONLY)
7106 spin_lock(&root->fs_info->delalloc_lock);
7107 while (!list_empty(head)) {
7108 binode = list_entry(head->next, struct btrfs_inode,
7110 inode = igrab(&binode->vfs_inode);
7112 list_del_init(&binode->delalloc_inodes);
7113 spin_unlock(&root->fs_info->delalloc_lock);
7115 filemap_flush(inode->i_mapping);
7117 btrfs_add_delayed_iput(inode);
7122 spin_lock(&root->fs_info->delalloc_lock);
7124 spin_unlock(&root->fs_info->delalloc_lock);
7126 /* the filemap_flush will queue IO into the worker threads, but
7127 * we have to make sure the IO is actually started and that
7128 * ordered extents get created before we return
7130 atomic_inc(&root->fs_info->async_submit_draining);
7131 while (atomic_read(&root->fs_info->nr_async_submits) ||
7132 atomic_read(&root->fs_info->async_delalloc_pages)) {
7133 wait_event(root->fs_info->async_submit_wait,
7134 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7135 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7137 atomic_dec(&root->fs_info->async_submit_draining);
7141 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7144 struct btrfs_inode *binode;
7145 struct inode *inode = NULL;
7147 spin_lock(&root->fs_info->delalloc_lock);
7148 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7149 binode = list_entry(root->fs_info->delalloc_inodes.next,
7150 struct btrfs_inode, delalloc_inodes);
7151 inode = igrab(&binode->vfs_inode);
7153 list_move_tail(&binode->delalloc_inodes,
7154 &root->fs_info->delalloc_inodes);
7158 list_del_init(&binode->delalloc_inodes);
7159 cond_resched_lock(&root->fs_info->delalloc_lock);
7161 spin_unlock(&root->fs_info->delalloc_lock);
7165 filemap_write_and_wait(inode->i_mapping);
7167 * We have to do this because compression doesn't
7168 * actually set PG_writeback until it submits the pages
7169 * for IO, which happens in an async thread, so we could
7170 * race and not actually wait for any writeback pages
7171 * because they've not been submitted yet. Technically
7172 * this could still be the case for the ordered stuff
7173 * since the async thread may not have started to do its
7174 * work yet. If this becomes the case then we need to
7175 * figure out a way to make sure that in writepage we
7176 * wait for any async pages to be submitted before
7177 * returning so that fdatawait does what its supposed to
7180 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7182 filemap_flush(inode->i_mapping);
7185 btrfs_add_delayed_iput(inode);
7193 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7194 const char *symname)
7196 struct btrfs_trans_handle *trans;
7197 struct btrfs_root *root = BTRFS_I(dir)->root;
7198 struct btrfs_path *path;
7199 struct btrfs_key key;
7200 struct inode *inode = NULL;
7208 struct btrfs_file_extent_item *ei;
7209 struct extent_buffer *leaf;
7210 unsigned long nr = 0;
7212 name_len = strlen(symname) + 1;
7213 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7214 return -ENAMETOOLONG;
7216 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7220 * 2 items for inode item and ref
7221 * 2 items for dir items
7222 * 1 item for xattr if selinux is on
7224 trans = btrfs_start_transaction(root, 5);
7226 return PTR_ERR(trans);
7228 btrfs_set_trans_block_group(trans, dir);
7230 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7231 dentry->d_name.len, dir->i_ino, objectid,
7232 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7234 err = PTR_ERR(inode);
7238 err = btrfs_init_inode_security(trans, inode, dir);
7244 btrfs_set_trans_block_group(trans, inode);
7245 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7249 inode->i_mapping->a_ops = &btrfs_aops;
7250 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7251 inode->i_fop = &btrfs_file_operations;
7252 inode->i_op = &btrfs_file_inode_operations;
7253 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7255 btrfs_update_inode_block_group(trans, inode);
7256 btrfs_update_inode_block_group(trans, dir);
7260 path = btrfs_alloc_path();
7262 key.objectid = inode->i_ino;
7264 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7265 datasize = btrfs_file_extent_calc_inline_size(name_len);
7266 err = btrfs_insert_empty_item(trans, root, path, &key,
7272 leaf = path->nodes[0];
7273 ei = btrfs_item_ptr(leaf, path->slots[0],
7274 struct btrfs_file_extent_item);
7275 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7276 btrfs_set_file_extent_type(leaf, ei,
7277 BTRFS_FILE_EXTENT_INLINE);
7278 btrfs_set_file_extent_encryption(leaf, ei, 0);
7279 btrfs_set_file_extent_compression(leaf, ei, 0);
7280 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7281 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7283 ptr = btrfs_file_extent_inline_start(ei);
7284 write_extent_buffer(leaf, symname, ptr, name_len);
7285 btrfs_mark_buffer_dirty(leaf);
7286 btrfs_free_path(path);
7288 inode->i_op = &btrfs_symlink_inode_operations;
7289 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7290 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7291 inode_set_bytes(inode, name_len);
7292 btrfs_i_size_write(inode, name_len - 1);
7293 err = btrfs_update_inode(trans, root, inode);
7298 nr = trans->blocks_used;
7299 btrfs_end_transaction_throttle(trans, root);
7301 inode_dec_link_count(inode);
7304 btrfs_btree_balance_dirty(root, nr);
7308 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7309 u64 start, u64 num_bytes, u64 min_size,
7310 loff_t actual_len, u64 *alloc_hint,
7311 struct btrfs_trans_handle *trans)
7313 struct btrfs_root *root = BTRFS_I(inode)->root;
7314 struct btrfs_key ins;
7315 u64 cur_offset = start;
7318 bool own_trans = true;
7322 while (num_bytes > 0) {
7324 trans = btrfs_start_transaction(root, 3);
7325 if (IS_ERR(trans)) {
7326 ret = PTR_ERR(trans);
7331 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7332 0, *alloc_hint, (u64)-1, &ins, 1);
7335 btrfs_end_transaction(trans, root);
7339 ret = insert_reserved_file_extent(trans, inode,
7340 cur_offset, ins.objectid,
7341 ins.offset, ins.offset,
7342 ins.offset, 0, 0, 0,
7343 BTRFS_FILE_EXTENT_PREALLOC);
7345 btrfs_drop_extent_cache(inode, cur_offset,
7346 cur_offset + ins.offset -1, 0);
7348 num_bytes -= ins.offset;
7349 cur_offset += ins.offset;
7350 *alloc_hint = ins.objectid + ins.offset;
7352 inode->i_ctime = CURRENT_TIME;
7353 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7354 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7355 (actual_len > inode->i_size) &&
7356 (cur_offset > inode->i_size)) {
7357 if (cur_offset > actual_len)
7358 i_size = actual_len;
7360 i_size = cur_offset;
7361 i_size_write(inode, i_size);
7362 btrfs_ordered_update_i_size(inode, i_size, NULL);
7365 ret = btrfs_update_inode(trans, root, inode);
7369 btrfs_end_transaction(trans, root);
7374 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7375 u64 start, u64 num_bytes, u64 min_size,
7376 loff_t actual_len, u64 *alloc_hint)
7378 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7379 min_size, actual_len, alloc_hint,
7383 int btrfs_prealloc_file_range_trans(struct inode *inode,
7384 struct btrfs_trans_handle *trans, int mode,
7385 u64 start, u64 num_bytes, u64 min_size,
7386 loff_t actual_len, u64 *alloc_hint)
7388 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7389 min_size, actual_len, alloc_hint, trans);
7392 static int btrfs_set_page_dirty(struct page *page)
7394 return __set_page_dirty_nobuffers(page);
7397 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7399 struct btrfs_root *root = BTRFS_I(inode)->root;
7401 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7403 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7405 return generic_permission(inode, mask, flags, btrfs_check_acl);
7408 static const struct inode_operations btrfs_dir_inode_operations = {
7409 .getattr = btrfs_getattr,
7410 .lookup = btrfs_lookup,
7411 .create = btrfs_create,
7412 .unlink = btrfs_unlink,
7414 .mkdir = btrfs_mkdir,
7415 .rmdir = btrfs_rmdir,
7416 .rename = btrfs_rename,
7417 .symlink = btrfs_symlink,
7418 .setattr = btrfs_setattr,
7419 .mknod = btrfs_mknod,
7420 .setxattr = btrfs_setxattr,
7421 .getxattr = btrfs_getxattr,
7422 .listxattr = btrfs_listxattr,
7423 .removexattr = btrfs_removexattr,
7424 .permission = btrfs_permission,
7426 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7427 .lookup = btrfs_lookup,
7428 .permission = btrfs_permission,
7431 static const struct file_operations btrfs_dir_file_operations = {
7432 .llseek = generic_file_llseek,
7433 .read = generic_read_dir,
7434 .readdir = btrfs_real_readdir,
7435 .unlocked_ioctl = btrfs_ioctl,
7436 #ifdef CONFIG_COMPAT
7437 .compat_ioctl = btrfs_ioctl,
7439 .release = btrfs_release_file,
7440 .fsync = btrfs_sync_file,
7443 static struct extent_io_ops btrfs_extent_io_ops = {
7444 .fill_delalloc = run_delalloc_range,
7445 .submit_bio_hook = btrfs_submit_bio_hook,
7446 .merge_bio_hook = btrfs_merge_bio_hook,
7447 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7448 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7449 .writepage_start_hook = btrfs_writepage_start_hook,
7450 .readpage_io_failed_hook = btrfs_io_failed_hook,
7451 .set_bit_hook = btrfs_set_bit_hook,
7452 .clear_bit_hook = btrfs_clear_bit_hook,
7453 .merge_extent_hook = btrfs_merge_extent_hook,
7454 .split_extent_hook = btrfs_split_extent_hook,
7458 * btrfs doesn't support the bmap operation because swapfiles
7459 * use bmap to make a mapping of extents in the file. They assume
7460 * these extents won't change over the life of the file and they
7461 * use the bmap result to do IO directly to the drive.
7463 * the btrfs bmap call would return logical addresses that aren't
7464 * suitable for IO and they also will change frequently as COW
7465 * operations happen. So, swapfile + btrfs == corruption.
7467 * For now we're avoiding this by dropping bmap.
7469 static const struct address_space_operations btrfs_aops = {
7470 .readpage = btrfs_readpage,
7471 .writepage = btrfs_writepage,
7472 .writepages = btrfs_writepages,
7473 .readpages = btrfs_readpages,
7474 .sync_page = block_sync_page,
7475 .direct_IO = btrfs_direct_IO,
7476 .invalidatepage = btrfs_invalidatepage,
7477 .releasepage = btrfs_releasepage,
7478 .set_page_dirty = btrfs_set_page_dirty,
7479 .error_remove_page = generic_error_remove_page,
7482 static const struct address_space_operations btrfs_symlink_aops = {
7483 .readpage = btrfs_readpage,
7484 .writepage = btrfs_writepage,
7485 .invalidatepage = btrfs_invalidatepage,
7486 .releasepage = btrfs_releasepage,
7489 static const struct inode_operations btrfs_file_inode_operations = {
7490 .getattr = btrfs_getattr,
7491 .setattr = btrfs_setattr,
7492 .setxattr = btrfs_setxattr,
7493 .getxattr = btrfs_getxattr,
7494 .listxattr = btrfs_listxattr,
7495 .removexattr = btrfs_removexattr,
7496 .permission = btrfs_permission,
7497 .fiemap = btrfs_fiemap,
7499 static const struct inode_operations btrfs_special_inode_operations = {
7500 .getattr = btrfs_getattr,
7501 .setattr = btrfs_setattr,
7502 .permission = btrfs_permission,
7503 .setxattr = btrfs_setxattr,
7504 .getxattr = btrfs_getxattr,
7505 .listxattr = btrfs_listxattr,
7506 .removexattr = btrfs_removexattr,
7508 static const struct inode_operations btrfs_symlink_inode_operations = {
7509 .readlink = generic_readlink,
7510 .follow_link = page_follow_link_light,
7511 .put_link = page_put_link,
7512 .getattr = btrfs_getattr,
7513 .permission = btrfs_permission,
7514 .setxattr = btrfs_setxattr,
7515 .getxattr = btrfs_getxattr,
7516 .listxattr = btrfs_listxattr,
7517 .removexattr = btrfs_removexattr,
7520 const struct dentry_operations btrfs_dentry_operations = {
7521 .d_delete = btrfs_dentry_delete,