2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
50 #include "compression.h"
53 struct btrfs_iget_args {
55 struct btrfs_root *root;
58 static const struct inode_operations btrfs_dir_inode_operations;
59 static const struct inode_operations btrfs_symlink_inode_operations;
60 static const struct inode_operations btrfs_dir_ro_inode_operations;
61 static const struct inode_operations btrfs_special_inode_operations;
62 static const struct inode_operations btrfs_file_inode_operations;
63 static const struct address_space_operations btrfs_aops;
64 static const struct address_space_operations btrfs_symlink_aops;
65 static const struct file_operations btrfs_dir_file_operations;
66 static struct extent_io_ops btrfs_extent_io_ops;
68 static struct kmem_cache *btrfs_inode_cachep;
69 struct kmem_cache *btrfs_trans_handle_cachep;
70 struct kmem_cache *btrfs_transaction_cachep;
71 struct kmem_cache *btrfs_path_cachep;
74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
75 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
76 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
77 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
78 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
79 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
80 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
81 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
84 static void btrfs_truncate(struct inode *inode);
85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
86 static noinline int cow_file_range(struct inode *inode,
87 struct page *locked_page,
88 u64 start, u64 end, int *page_started,
89 unsigned long *nr_written, int unlock);
91 static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
95 err = btrfs_init_acl(inode, dir);
97 err = btrfs_xattr_security_init(inode, dir);
102 * this does all the hard work for inserting an inline extent into
103 * the btree. The caller should have done a btrfs_drop_extents so that
104 * no overlapping inline items exist in the btree
106 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root, struct inode *inode,
108 u64 start, size_t size, size_t compressed_size,
109 struct page **compressed_pages)
111 struct btrfs_key key;
112 struct btrfs_path *path;
113 struct extent_buffer *leaf;
114 struct page *page = NULL;
117 struct btrfs_file_extent_item *ei;
120 size_t cur_size = size;
122 unsigned long offset;
123 int use_compress = 0;
125 if (compressed_size && compressed_pages) {
127 cur_size = compressed_size;
130 path = btrfs_alloc_path();
134 path->leave_spinning = 1;
135 btrfs_set_trans_block_group(trans, inode);
137 key.objectid = inode->i_ino;
139 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
140 datasize = btrfs_file_extent_calc_inline_size(cur_size);
142 inode_add_bytes(inode, size);
143 ret = btrfs_insert_empty_item(trans, root, path, &key,
150 leaf = path->nodes[0];
151 ei = btrfs_item_ptr(leaf, path->slots[0],
152 struct btrfs_file_extent_item);
153 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
154 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
155 btrfs_set_file_extent_encryption(leaf, ei, 0);
156 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
157 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
158 ptr = btrfs_file_extent_inline_start(ei);
163 while (compressed_size > 0) {
164 cpage = compressed_pages[i];
165 cur_size = min_t(unsigned long, compressed_size,
168 kaddr = kmap_atomic(cpage, KM_USER0);
169 write_extent_buffer(leaf, kaddr, ptr, cur_size);
170 kunmap_atomic(kaddr, KM_USER0);
174 compressed_size -= cur_size;
176 btrfs_set_file_extent_compression(leaf, ei,
177 BTRFS_COMPRESS_ZLIB);
179 page = find_get_page(inode->i_mapping,
180 start >> PAGE_CACHE_SHIFT);
181 btrfs_set_file_extent_compression(leaf, ei, 0);
182 kaddr = kmap_atomic(page, KM_USER0);
183 offset = start & (PAGE_CACHE_SIZE - 1);
184 write_extent_buffer(leaf, kaddr + offset, ptr, size);
185 kunmap_atomic(kaddr, KM_USER0);
186 page_cache_release(page);
188 btrfs_mark_buffer_dirty(leaf);
189 btrfs_free_path(path);
191 BTRFS_I(inode)->disk_i_size = inode->i_size;
192 btrfs_update_inode(trans, root, inode);
195 btrfs_free_path(path);
201 * conditionally insert an inline extent into the file. This
202 * does the checks required to make sure the data is small enough
203 * to fit as an inline extent.
205 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
206 struct btrfs_root *root,
207 struct inode *inode, u64 start, u64 end,
208 size_t compressed_size,
209 struct page **compressed_pages)
211 u64 isize = i_size_read(inode);
212 u64 actual_end = min(end + 1, isize);
213 u64 inline_len = actual_end - start;
214 u64 aligned_end = (end + root->sectorsize - 1) &
215 ~((u64)root->sectorsize - 1);
217 u64 data_len = inline_len;
221 data_len = compressed_size;
224 actual_end >= PAGE_CACHE_SIZE ||
225 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227 (actual_end & (root->sectorsize - 1)) == 0) ||
229 data_len > root->fs_info->max_inline) {
233 ret = btrfs_drop_extents(trans, root, inode, start,
234 aligned_end, aligned_end, start,
238 if (isize > actual_end)
239 inline_len = min_t(u64, isize, actual_end);
240 ret = insert_inline_extent(trans, root, inode, start,
241 inline_len, compressed_size,
244 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
248 struct async_extent {
253 unsigned long nr_pages;
254 struct list_head list;
259 struct btrfs_root *root;
260 struct page *locked_page;
263 struct list_head extents;
264 struct btrfs_work work;
267 static noinline int add_async_extent(struct async_cow *cow,
268 u64 start, u64 ram_size,
271 unsigned long nr_pages)
273 struct async_extent *async_extent;
275 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
276 async_extent->start = start;
277 async_extent->ram_size = ram_size;
278 async_extent->compressed_size = compressed_size;
279 async_extent->pages = pages;
280 async_extent->nr_pages = nr_pages;
281 list_add_tail(&async_extent->list, &cow->extents);
286 * we create compressed extents in two phases. The first
287 * phase compresses a range of pages that have already been
288 * locked (both pages and state bits are locked).
290 * This is done inside an ordered work queue, and the compression
291 * is spread across many cpus. The actual IO submission is step
292 * two, and the ordered work queue takes care of making sure that
293 * happens in the same order things were put onto the queue by
294 * writepages and friends.
296 * If this code finds it can't get good compression, it puts an
297 * entry onto the work queue to write the uncompressed bytes. This
298 * makes sure that both compressed inodes and uncompressed inodes
299 * are written in the same order that pdflush sent them down.
301 static noinline int compress_file_range(struct inode *inode,
302 struct page *locked_page,
304 struct async_cow *async_cow,
307 struct btrfs_root *root = BTRFS_I(inode)->root;
308 struct btrfs_trans_handle *trans;
312 u64 blocksize = root->sectorsize;
314 u64 isize = i_size_read(inode);
316 struct page **pages = NULL;
317 unsigned long nr_pages;
318 unsigned long nr_pages_ret = 0;
319 unsigned long total_compressed = 0;
320 unsigned long total_in = 0;
321 unsigned long max_compressed = 128 * 1024;
322 unsigned long max_uncompressed = 128 * 1024;
328 actual_end = min_t(u64, isize, end + 1);
331 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
332 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
335 * we don't want to send crud past the end of i_size through
336 * compression, that's just a waste of CPU time. So, if the
337 * end of the file is before the start of our current
338 * requested range of bytes, we bail out to the uncompressed
339 * cleanup code that can deal with all of this.
341 * It isn't really the fastest way to fix things, but this is a
342 * very uncommon corner.
344 if (actual_end <= start)
345 goto cleanup_and_bail_uncompressed;
347 total_compressed = actual_end - start;
349 /* we want to make sure that amount of ram required to uncompress
350 * an extent is reasonable, so we limit the total size in ram
351 * of a compressed extent to 128k. This is a crucial number
352 * because it also controls how easily we can spread reads across
353 * cpus for decompression.
355 * We also want to make sure the amount of IO required to do
356 * a random read is reasonably small, so we limit the size of
357 * a compressed extent to 128k.
359 total_compressed = min(total_compressed, max_uncompressed);
360 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
361 num_bytes = max(blocksize, num_bytes);
362 disk_num_bytes = num_bytes;
367 * we do compression for mount -o compress and when the
368 * inode has not been flagged as nocompress. This flag can
369 * change at any time if we discover bad compression ratios.
371 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
372 btrfs_test_opt(root, COMPRESS)) {
374 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
376 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
377 total_compressed, pages,
378 nr_pages, &nr_pages_ret,
384 unsigned long offset = total_compressed &
385 (PAGE_CACHE_SIZE - 1);
386 struct page *page = pages[nr_pages_ret - 1];
389 /* zero the tail end of the last page, we might be
390 * sending it down to disk
393 kaddr = kmap_atomic(page, KM_USER0);
394 memset(kaddr + offset, 0,
395 PAGE_CACHE_SIZE - offset);
396 kunmap_atomic(kaddr, KM_USER0);
402 trans = btrfs_join_transaction(root, 1);
404 btrfs_set_trans_block_group(trans, inode);
406 /* lets try to make an inline extent */
407 if (ret || total_in < (actual_end - start)) {
408 /* we didn't compress the entire range, try
409 * to make an uncompressed inline extent.
411 ret = cow_file_range_inline(trans, root, inode,
412 start, end, 0, NULL);
414 /* try making a compressed inline extent */
415 ret = cow_file_range_inline(trans, root, inode,
417 total_compressed, pages);
419 btrfs_end_transaction(trans, root);
422 * inline extent creation worked, we don't need
423 * to create any more async work items. Unlock
424 * and free up our temp pages.
426 extent_clear_unlock_delalloc(inode,
427 &BTRFS_I(inode)->io_tree,
429 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
430 EXTENT_CLEAR_DELALLOC |
431 EXTENT_CLEAR_ACCOUNTING |
432 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
440 * we aren't doing an inline extent round the compressed size
441 * up to a block size boundary so the allocator does sane
444 total_compressed = (total_compressed + blocksize - 1) &
448 * one last check to make sure the compression is really a
449 * win, compare the page count read with the blocks on disk
451 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
452 ~(PAGE_CACHE_SIZE - 1);
453 if (total_compressed >= total_in) {
456 disk_num_bytes = total_compressed;
457 num_bytes = total_in;
460 if (!will_compress && pages) {
462 * the compression code ran but failed to make things smaller,
463 * free any pages it allocated and our page pointer array
465 for (i = 0; i < nr_pages_ret; i++) {
466 WARN_ON(pages[i]->mapping);
467 page_cache_release(pages[i]);
471 total_compressed = 0;
474 /* flag the file so we don't compress in the future */
475 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
480 /* the async work queues will take care of doing actual
481 * allocation on disk for these compressed pages,
482 * and will submit them to the elevator.
484 add_async_extent(async_cow, start, num_bytes,
485 total_compressed, pages, nr_pages_ret);
487 if (start + num_bytes < end && start + num_bytes < actual_end) {
494 cleanup_and_bail_uncompressed:
496 * No compression, but we still need to write the pages in
497 * the file we've been given so far. redirty the locked
498 * page if it corresponds to our extent and set things up
499 * for the async work queue to run cow_file_range to do
500 * the normal delalloc dance
502 if (page_offset(locked_page) >= start &&
503 page_offset(locked_page) <= end) {
504 __set_page_dirty_nobuffers(locked_page);
505 /* unlocked later on in the async handlers */
507 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
515 for (i = 0; i < nr_pages_ret; i++) {
516 WARN_ON(pages[i]->mapping);
517 page_cache_release(pages[i]);
525 * phase two of compressed writeback. This is the ordered portion
526 * of the code, which only gets called in the order the work was
527 * queued. We walk all the async extents created by compress_file_range
528 * and send them down to the disk.
530 static noinline int submit_compressed_extents(struct inode *inode,
531 struct async_cow *async_cow)
533 struct async_extent *async_extent;
535 struct btrfs_trans_handle *trans;
536 struct btrfs_key ins;
537 struct extent_map *em;
538 struct btrfs_root *root = BTRFS_I(inode)->root;
539 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
540 struct extent_io_tree *io_tree;
543 if (list_empty(&async_cow->extents))
546 trans = btrfs_join_transaction(root, 1);
548 while (!list_empty(&async_cow->extents)) {
549 async_extent = list_entry(async_cow->extents.next,
550 struct async_extent, list);
551 list_del(&async_extent->list);
553 io_tree = &BTRFS_I(inode)->io_tree;
556 /* did the compression code fall back to uncompressed IO? */
557 if (!async_extent->pages) {
558 int page_started = 0;
559 unsigned long nr_written = 0;
561 lock_extent(io_tree, async_extent->start,
562 async_extent->start +
563 async_extent->ram_size - 1, GFP_NOFS);
565 /* allocate blocks */
566 ret = cow_file_range(inode, async_cow->locked_page,
568 async_extent->start +
569 async_extent->ram_size - 1,
570 &page_started, &nr_written, 0);
573 * if page_started, cow_file_range inserted an
574 * inline extent and took care of all the unlocking
575 * and IO for us. Otherwise, we need to submit
576 * all those pages down to the drive.
578 if (!page_started && !ret)
579 extent_write_locked_range(io_tree,
580 inode, async_extent->start,
581 async_extent->start +
582 async_extent->ram_size - 1,
590 lock_extent(io_tree, async_extent->start,
591 async_extent->start + async_extent->ram_size - 1,
594 * here we're doing allocation and writeback of the
597 btrfs_drop_extent_cache(inode, async_extent->start,
598 async_extent->start +
599 async_extent->ram_size - 1, 0);
601 ret = btrfs_reserve_extent(trans, root,
602 async_extent->compressed_size,
603 async_extent->compressed_size,
608 for (i = 0; i < async_extent->nr_pages; i++) {
609 WARN_ON(async_extent->pages[i]->mapping);
610 page_cache_release(async_extent->pages[i]);
612 kfree(async_extent->pages);
613 async_extent->nr_pages = 0;
614 async_extent->pages = NULL;
615 unlock_extent(io_tree, async_extent->start,
616 async_extent->start +
617 async_extent->ram_size - 1, GFP_NOFS);
621 em = alloc_extent_map(GFP_NOFS);
622 em->start = async_extent->start;
623 em->len = async_extent->ram_size;
624 em->orig_start = em->start;
626 em->block_start = ins.objectid;
627 em->block_len = ins.offset;
628 em->bdev = root->fs_info->fs_devices->latest_bdev;
629 set_bit(EXTENT_FLAG_PINNED, &em->flags);
630 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
633 write_lock(&em_tree->lock);
634 ret = add_extent_mapping(em_tree, em);
635 write_unlock(&em_tree->lock);
636 if (ret != -EEXIST) {
640 btrfs_drop_extent_cache(inode, async_extent->start,
641 async_extent->start +
642 async_extent->ram_size - 1, 0);
645 ret = btrfs_add_ordered_extent(inode, async_extent->start,
647 async_extent->ram_size,
649 BTRFS_ORDERED_COMPRESSED);
652 btrfs_end_transaction(trans, root);
655 * clear dirty, set writeback and unlock the pages.
657 extent_clear_unlock_delalloc(inode,
658 &BTRFS_I(inode)->io_tree,
660 async_extent->start +
661 async_extent->ram_size - 1,
662 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
663 EXTENT_CLEAR_UNLOCK |
664 EXTENT_CLEAR_DELALLOC |
665 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
667 ret = btrfs_submit_compressed_write(inode,
669 async_extent->ram_size,
671 ins.offset, async_extent->pages,
672 async_extent->nr_pages);
675 trans = btrfs_join_transaction(root, 1);
676 alloc_hint = ins.objectid + ins.offset;
681 btrfs_end_transaction(trans, root);
686 * when extent_io.c finds a delayed allocation range in the file,
687 * the call backs end up in this code. The basic idea is to
688 * allocate extents on disk for the range, and create ordered data structs
689 * in ram to track those extents.
691 * locked_page is the page that writepage had locked already. We use
692 * it to make sure we don't do extra locks or unlocks.
694 * *page_started is set to one if we unlock locked_page and do everything
695 * required to start IO on it. It may be clean and already done with
698 static noinline int cow_file_range(struct inode *inode,
699 struct page *locked_page,
700 u64 start, u64 end, int *page_started,
701 unsigned long *nr_written,
704 struct btrfs_root *root = BTRFS_I(inode)->root;
705 struct btrfs_trans_handle *trans;
708 unsigned long ram_size;
711 u64 blocksize = root->sectorsize;
713 u64 isize = i_size_read(inode);
714 struct btrfs_key ins;
715 struct extent_map *em;
716 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
719 trans = btrfs_join_transaction(root, 1);
721 btrfs_set_trans_block_group(trans, inode);
723 actual_end = min_t(u64, isize, end + 1);
725 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
726 num_bytes = max(blocksize, num_bytes);
727 disk_num_bytes = num_bytes;
731 /* lets try to make an inline extent */
732 ret = cow_file_range_inline(trans, root, inode,
733 start, end, 0, NULL);
735 extent_clear_unlock_delalloc(inode,
736 &BTRFS_I(inode)->io_tree,
738 EXTENT_CLEAR_UNLOCK_PAGE |
739 EXTENT_CLEAR_UNLOCK |
740 EXTENT_CLEAR_DELALLOC |
741 EXTENT_CLEAR_ACCOUNTING |
743 EXTENT_SET_WRITEBACK |
744 EXTENT_END_WRITEBACK);
745 *nr_written = *nr_written +
746 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
753 BUG_ON(disk_num_bytes >
754 btrfs_super_total_bytes(&root->fs_info->super_copy));
757 read_lock(&BTRFS_I(inode)->extent_tree.lock);
758 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
762 * if block start isn't an actual block number then find the
763 * first block in this inode and use that as a hint. If that
764 * block is also bogus then just don't worry about it.
766 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
768 em = search_extent_mapping(em_tree, 0, 0);
769 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
770 alloc_hint = em->block_start;
774 alloc_hint = em->block_start;
778 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
779 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
781 while (disk_num_bytes > 0) {
784 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
785 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
786 root->sectorsize, 0, alloc_hint,
790 em = alloc_extent_map(GFP_NOFS);
792 em->orig_start = em->start;
793 ram_size = ins.offset;
794 em->len = ins.offset;
796 em->block_start = ins.objectid;
797 em->block_len = ins.offset;
798 em->bdev = root->fs_info->fs_devices->latest_bdev;
799 set_bit(EXTENT_FLAG_PINNED, &em->flags);
802 write_lock(&em_tree->lock);
803 ret = add_extent_mapping(em_tree, em);
804 write_unlock(&em_tree->lock);
805 if (ret != -EEXIST) {
809 btrfs_drop_extent_cache(inode, start,
810 start + ram_size - 1, 0);
813 cur_alloc_size = ins.offset;
814 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
815 ram_size, cur_alloc_size, 0);
818 if (root->root_key.objectid ==
819 BTRFS_DATA_RELOC_TREE_OBJECTID) {
820 ret = btrfs_reloc_clone_csums(inode, start,
825 if (disk_num_bytes < cur_alloc_size)
828 /* we're not doing compressed IO, don't unlock the first
829 * page (which the caller expects to stay locked), don't
830 * clear any dirty bits and don't set any writeback bits
832 * Do set the Private2 bit so we know this page was properly
833 * setup for writepage
835 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
836 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
839 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
840 start, start + ram_size - 1,
842 disk_num_bytes -= cur_alloc_size;
843 num_bytes -= cur_alloc_size;
844 alloc_hint = ins.objectid + ins.offset;
845 start += cur_alloc_size;
849 btrfs_end_transaction(trans, root);
855 * work queue call back to started compression on a file and pages
857 static noinline void async_cow_start(struct btrfs_work *work)
859 struct async_cow *async_cow;
861 async_cow = container_of(work, struct async_cow, work);
863 compress_file_range(async_cow->inode, async_cow->locked_page,
864 async_cow->start, async_cow->end, async_cow,
867 async_cow->inode = NULL;
871 * work queue call back to submit previously compressed pages
873 static noinline void async_cow_submit(struct btrfs_work *work)
875 struct async_cow *async_cow;
876 struct btrfs_root *root;
877 unsigned long nr_pages;
879 async_cow = container_of(work, struct async_cow, work);
881 root = async_cow->root;
882 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
885 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
887 if (atomic_read(&root->fs_info->async_delalloc_pages) <
889 waitqueue_active(&root->fs_info->async_submit_wait))
890 wake_up(&root->fs_info->async_submit_wait);
892 if (async_cow->inode)
893 submit_compressed_extents(async_cow->inode, async_cow);
896 static noinline void async_cow_free(struct btrfs_work *work)
898 struct async_cow *async_cow;
899 async_cow = container_of(work, struct async_cow, work);
903 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
904 u64 start, u64 end, int *page_started,
905 unsigned long *nr_written)
907 struct async_cow *async_cow;
908 struct btrfs_root *root = BTRFS_I(inode)->root;
909 unsigned long nr_pages;
911 int limit = 10 * 1024 * 1042;
913 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
914 1, 0, NULL, GFP_NOFS);
915 while (start < end) {
916 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
917 async_cow->inode = inode;
918 async_cow->root = root;
919 async_cow->locked_page = locked_page;
920 async_cow->start = start;
922 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
925 cur_end = min(end, start + 512 * 1024 - 1);
927 async_cow->end = cur_end;
928 INIT_LIST_HEAD(&async_cow->extents);
930 async_cow->work.func = async_cow_start;
931 async_cow->work.ordered_func = async_cow_submit;
932 async_cow->work.ordered_free = async_cow_free;
933 async_cow->work.flags = 0;
935 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
937 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
939 btrfs_queue_worker(&root->fs_info->delalloc_workers,
942 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
943 wait_event(root->fs_info->async_submit_wait,
944 (atomic_read(&root->fs_info->async_delalloc_pages) <
948 while (atomic_read(&root->fs_info->async_submit_draining) &&
949 atomic_read(&root->fs_info->async_delalloc_pages)) {
950 wait_event(root->fs_info->async_submit_wait,
951 (atomic_read(&root->fs_info->async_delalloc_pages) ==
955 *nr_written += nr_pages;
962 static noinline int csum_exist_in_range(struct btrfs_root *root,
963 u64 bytenr, u64 num_bytes)
966 struct btrfs_ordered_sum *sums;
969 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
970 bytenr + num_bytes - 1, &list);
971 if (ret == 0 && list_empty(&list))
974 while (!list_empty(&list)) {
975 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
976 list_del(&sums->list);
983 * when nowcow writeback call back. This checks for snapshots or COW copies
984 * of the extents that exist in the file, and COWs the file as required.
986 * If no cow copies or snapshots exist, we write directly to the existing
989 static noinline int run_delalloc_nocow(struct inode *inode,
990 struct page *locked_page,
991 u64 start, u64 end, int *page_started, int force,
992 unsigned long *nr_written)
994 struct btrfs_root *root = BTRFS_I(inode)->root;
995 struct btrfs_trans_handle *trans;
996 struct extent_buffer *leaf;
997 struct btrfs_path *path;
998 struct btrfs_file_extent_item *fi;
999 struct btrfs_key found_key;
1012 path = btrfs_alloc_path();
1014 trans = btrfs_join_transaction(root, 1);
1017 cow_start = (u64)-1;
1020 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1023 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1024 leaf = path->nodes[0];
1025 btrfs_item_key_to_cpu(leaf, &found_key,
1026 path->slots[0] - 1);
1027 if (found_key.objectid == inode->i_ino &&
1028 found_key.type == BTRFS_EXTENT_DATA_KEY)
1033 leaf = path->nodes[0];
1034 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1035 ret = btrfs_next_leaf(root, path);
1040 leaf = path->nodes[0];
1046 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1048 if (found_key.objectid > inode->i_ino ||
1049 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1050 found_key.offset > end)
1053 if (found_key.offset > cur_offset) {
1054 extent_end = found_key.offset;
1059 fi = btrfs_item_ptr(leaf, path->slots[0],
1060 struct btrfs_file_extent_item);
1061 extent_type = btrfs_file_extent_type(leaf, fi);
1063 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1064 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1065 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1066 extent_offset = btrfs_file_extent_offset(leaf, fi);
1067 extent_end = found_key.offset +
1068 btrfs_file_extent_num_bytes(leaf, fi);
1069 if (extent_end <= start) {
1073 if (disk_bytenr == 0)
1075 if (btrfs_file_extent_compression(leaf, fi) ||
1076 btrfs_file_extent_encryption(leaf, fi) ||
1077 btrfs_file_extent_other_encoding(leaf, fi))
1079 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1081 if (btrfs_extent_readonly(root, disk_bytenr))
1083 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1085 extent_offset, disk_bytenr))
1087 disk_bytenr += extent_offset;
1088 disk_bytenr += cur_offset - found_key.offset;
1089 num_bytes = min(end + 1, extent_end) - cur_offset;
1091 * force cow if csum exists in the range.
1092 * this ensure that csum for a given extent are
1093 * either valid or do not exist.
1095 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1098 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1099 extent_end = found_key.offset +
1100 btrfs_file_extent_inline_len(leaf, fi);
1101 extent_end = ALIGN(extent_end, root->sectorsize);
1106 if (extent_end <= start) {
1111 if (cow_start == (u64)-1)
1112 cow_start = cur_offset;
1113 cur_offset = extent_end;
1114 if (cur_offset > end)
1120 btrfs_release_path(root, path);
1121 if (cow_start != (u64)-1) {
1122 ret = cow_file_range(inode, locked_page, cow_start,
1123 found_key.offset - 1, page_started,
1126 cow_start = (u64)-1;
1129 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1130 struct extent_map *em;
1131 struct extent_map_tree *em_tree;
1132 em_tree = &BTRFS_I(inode)->extent_tree;
1133 em = alloc_extent_map(GFP_NOFS);
1134 em->start = cur_offset;
1135 em->orig_start = em->start;
1136 em->len = num_bytes;
1137 em->block_len = num_bytes;
1138 em->block_start = disk_bytenr;
1139 em->bdev = root->fs_info->fs_devices->latest_bdev;
1140 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1142 write_lock(&em_tree->lock);
1143 ret = add_extent_mapping(em_tree, em);
1144 write_unlock(&em_tree->lock);
1145 if (ret != -EEXIST) {
1146 free_extent_map(em);
1149 btrfs_drop_extent_cache(inode, em->start,
1150 em->start + em->len - 1, 0);
1152 type = BTRFS_ORDERED_PREALLOC;
1154 type = BTRFS_ORDERED_NOCOW;
1157 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1158 num_bytes, num_bytes, type);
1161 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1162 cur_offset, cur_offset + num_bytes - 1,
1163 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1164 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1165 EXTENT_SET_PRIVATE2);
1166 cur_offset = extent_end;
1167 if (cur_offset > end)
1170 btrfs_release_path(root, path);
1172 if (cur_offset <= end && cow_start == (u64)-1)
1173 cow_start = cur_offset;
1174 if (cow_start != (u64)-1) {
1175 ret = cow_file_range(inode, locked_page, cow_start, end,
1176 page_started, nr_written, 1);
1180 ret = btrfs_end_transaction(trans, root);
1182 btrfs_free_path(path);
1187 * extent_io.c call back to do delayed allocation processing
1189 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1190 u64 start, u64 end, int *page_started,
1191 unsigned long *nr_written)
1194 struct btrfs_root *root = BTRFS_I(inode)->root;
1196 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1197 ret = run_delalloc_nocow(inode, locked_page, start, end,
1198 page_started, 1, nr_written);
1199 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1200 ret = run_delalloc_nocow(inode, locked_page, start, end,
1201 page_started, 0, nr_written);
1202 else if (!btrfs_test_opt(root, COMPRESS))
1203 ret = cow_file_range(inode, locked_page, start, end,
1204 page_started, nr_written, 1);
1206 ret = cow_file_range_async(inode, locked_page, start, end,
1207 page_started, nr_written);
1211 static int btrfs_split_extent_hook(struct inode *inode,
1212 struct extent_state *orig, u64 split)
1214 struct btrfs_root *root = BTRFS_I(inode)->root;
1217 if (!(orig->state & EXTENT_DELALLOC))
1220 size = orig->end - orig->start + 1;
1221 if (size > root->fs_info->max_extent) {
1225 new_size = orig->end - split + 1;
1226 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1227 root->fs_info->max_extent);
1230 * if we break a large extent up then leave oustanding_extents
1231 * be, since we've already accounted for the large extent.
1233 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1234 root->fs_info->max_extent) < num_extents)
1238 spin_lock(&BTRFS_I(inode)->accounting_lock);
1239 BTRFS_I(inode)->outstanding_extents++;
1240 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1246 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1247 * extents so we can keep track of new extents that are just merged onto old
1248 * extents, such as when we are doing sequential writes, so we can properly
1249 * account for the metadata space we'll need.
1251 static int btrfs_merge_extent_hook(struct inode *inode,
1252 struct extent_state *new,
1253 struct extent_state *other)
1255 struct btrfs_root *root = BTRFS_I(inode)->root;
1256 u64 new_size, old_size;
1259 /* not delalloc, ignore it */
1260 if (!(other->state & EXTENT_DELALLOC))
1263 old_size = other->end - other->start + 1;
1264 if (new->start < other->start)
1265 new_size = other->end - new->start + 1;
1267 new_size = new->end - other->start + 1;
1269 /* we're not bigger than the max, unreserve the space and go */
1270 if (new_size <= root->fs_info->max_extent) {
1271 spin_lock(&BTRFS_I(inode)->accounting_lock);
1272 BTRFS_I(inode)->outstanding_extents--;
1273 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1278 * If we grew by another max_extent, just return, we want to keep that
1281 num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1282 root->fs_info->max_extent);
1283 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1284 root->fs_info->max_extent) > num_extents)
1287 spin_lock(&BTRFS_I(inode)->accounting_lock);
1288 BTRFS_I(inode)->outstanding_extents--;
1289 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1295 * extent_io.c set_bit_hook, used to track delayed allocation
1296 * bytes in this file, and to maintain the list of inodes that
1297 * have pending delalloc work to be done.
1299 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1300 unsigned long old, unsigned long bits)
1304 * set_bit and clear bit hooks normally require _irqsave/restore
1305 * but in this case, we are only testeing for the DELALLOC
1306 * bit, which is only set or cleared with irqs on
1308 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1309 struct btrfs_root *root = BTRFS_I(inode)->root;
1311 spin_lock(&BTRFS_I(inode)->accounting_lock);
1312 BTRFS_I(inode)->outstanding_extents++;
1313 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1314 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1315 spin_lock(&root->fs_info->delalloc_lock);
1316 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1317 root->fs_info->delalloc_bytes += end - start + 1;
1318 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1319 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1320 &root->fs_info->delalloc_inodes);
1322 spin_unlock(&root->fs_info->delalloc_lock);
1328 * extent_io.c clear_bit_hook, see set_bit_hook for why
1330 static int btrfs_clear_bit_hook(struct inode *inode,
1331 struct extent_state *state, unsigned long bits)
1334 * set_bit and clear bit hooks normally require _irqsave/restore
1335 * but in this case, we are only testeing for the DELALLOC
1336 * bit, which is only set or cleared with irqs on
1338 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1339 struct btrfs_root *root = BTRFS_I(inode)->root;
1341 if (bits & EXTENT_DO_ACCOUNTING) {
1342 spin_lock(&BTRFS_I(inode)->accounting_lock);
1343 BTRFS_I(inode)->outstanding_extents--;
1344 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1345 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1348 spin_lock(&root->fs_info->delalloc_lock);
1349 if (state->end - state->start + 1 >
1350 root->fs_info->delalloc_bytes) {
1351 printk(KERN_INFO "btrfs warning: delalloc account "
1353 (unsigned long long)
1354 state->end - state->start + 1,
1355 (unsigned long long)
1356 root->fs_info->delalloc_bytes);
1357 btrfs_delalloc_free_space(root, inode, (u64)-1);
1358 root->fs_info->delalloc_bytes = 0;
1359 BTRFS_I(inode)->delalloc_bytes = 0;
1361 btrfs_delalloc_free_space(root, inode,
1364 root->fs_info->delalloc_bytes -= state->end -
1366 BTRFS_I(inode)->delalloc_bytes -= state->end -
1369 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1370 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1371 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1373 spin_unlock(&root->fs_info->delalloc_lock);
1379 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1380 * we don't create bios that span stripes or chunks
1382 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1383 size_t size, struct bio *bio,
1384 unsigned long bio_flags)
1386 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1387 struct btrfs_mapping_tree *map_tree;
1388 u64 logical = (u64)bio->bi_sector << 9;
1393 if (bio_flags & EXTENT_BIO_COMPRESSED)
1396 length = bio->bi_size;
1397 map_tree = &root->fs_info->mapping_tree;
1398 map_length = length;
1399 ret = btrfs_map_block(map_tree, READ, logical,
1400 &map_length, NULL, 0);
1402 if (map_length < length + size)
1408 * in order to insert checksums into the metadata in large chunks,
1409 * we wait until bio submission time. All the pages in the bio are
1410 * checksummed and sums are attached onto the ordered extent record.
1412 * At IO completion time the cums attached on the ordered extent record
1413 * are inserted into the btree
1415 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1416 struct bio *bio, int mirror_num,
1417 unsigned long bio_flags)
1419 struct btrfs_root *root = BTRFS_I(inode)->root;
1422 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1428 * in order to insert checksums into the metadata in large chunks,
1429 * we wait until bio submission time. All the pages in the bio are
1430 * checksummed and sums are attached onto the ordered extent record.
1432 * At IO completion time the cums attached on the ordered extent record
1433 * are inserted into the btree
1435 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1436 int mirror_num, unsigned long bio_flags)
1438 struct btrfs_root *root = BTRFS_I(inode)->root;
1439 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1443 * extent_io.c submission hook. This does the right thing for csum calculation
1444 * on write, or reading the csums from the tree before a read
1446 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1447 int mirror_num, unsigned long bio_flags)
1449 struct btrfs_root *root = BTRFS_I(inode)->root;
1453 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1455 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1458 if (!(rw & (1 << BIO_RW))) {
1459 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1460 return btrfs_submit_compressed_read(inode, bio,
1461 mirror_num, bio_flags);
1462 } else if (!skip_sum)
1463 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1465 } else if (!skip_sum) {
1466 /* csum items have already been cloned */
1467 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1469 /* we're doing a write, do the async checksumming */
1470 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1471 inode, rw, bio, mirror_num,
1472 bio_flags, __btrfs_submit_bio_start,
1473 __btrfs_submit_bio_done);
1477 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1481 * given a list of ordered sums record them in the inode. This happens
1482 * at IO completion time based on sums calculated at bio submission time.
1484 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1485 struct inode *inode, u64 file_offset,
1486 struct list_head *list)
1488 struct btrfs_ordered_sum *sum;
1490 btrfs_set_trans_block_group(trans, inode);
1492 list_for_each_entry(sum, list, list) {
1493 btrfs_csum_file_blocks(trans,
1494 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1499 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1501 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1503 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1507 /* see btrfs_writepage_start_hook for details on why this is required */
1508 struct btrfs_writepage_fixup {
1510 struct btrfs_work work;
1513 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1515 struct btrfs_writepage_fixup *fixup;
1516 struct btrfs_ordered_extent *ordered;
1518 struct inode *inode;
1522 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1526 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1527 ClearPageChecked(page);
1531 inode = page->mapping->host;
1532 page_start = page_offset(page);
1533 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1535 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1537 /* already ordered? We're done */
1538 if (PagePrivate2(page))
1541 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1543 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1544 page_end, GFP_NOFS);
1546 btrfs_start_ordered_extent(inode, ordered, 1);
1550 btrfs_set_extent_delalloc(inode, page_start, page_end);
1551 ClearPageChecked(page);
1553 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1556 page_cache_release(page);
1560 * There are a few paths in the higher layers of the kernel that directly
1561 * set the page dirty bit without asking the filesystem if it is a
1562 * good idea. This causes problems because we want to make sure COW
1563 * properly happens and the data=ordered rules are followed.
1565 * In our case any range that doesn't have the ORDERED bit set
1566 * hasn't been properly setup for IO. We kick off an async process
1567 * to fix it up. The async helper will wait for ordered extents, set
1568 * the delalloc bit and make it safe to write the page.
1570 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1572 struct inode *inode = page->mapping->host;
1573 struct btrfs_writepage_fixup *fixup;
1574 struct btrfs_root *root = BTRFS_I(inode)->root;
1576 /* this page is properly in the ordered list */
1577 if (TestClearPagePrivate2(page))
1580 if (PageChecked(page))
1583 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1587 SetPageChecked(page);
1588 page_cache_get(page);
1589 fixup->work.func = btrfs_writepage_fixup_worker;
1591 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1595 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1596 struct inode *inode, u64 file_pos,
1597 u64 disk_bytenr, u64 disk_num_bytes,
1598 u64 num_bytes, u64 ram_bytes,
1600 u8 compression, u8 encryption,
1601 u16 other_encoding, int extent_type)
1603 struct btrfs_root *root = BTRFS_I(inode)->root;
1604 struct btrfs_file_extent_item *fi;
1605 struct btrfs_path *path;
1606 struct extent_buffer *leaf;
1607 struct btrfs_key ins;
1611 path = btrfs_alloc_path();
1614 path->leave_spinning = 1;
1617 * we may be replacing one extent in the tree with another.
1618 * The new extent is pinned in the extent map, and we don't want
1619 * to drop it from the cache until it is completely in the btree.
1621 * So, tell btrfs_drop_extents to leave this extent in the cache.
1622 * the caller is expected to unpin it and allow it to be merged
1625 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1626 file_pos + num_bytes, locked_end,
1627 file_pos, &hint, 0);
1630 ins.objectid = inode->i_ino;
1631 ins.offset = file_pos;
1632 ins.type = BTRFS_EXTENT_DATA_KEY;
1633 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1635 leaf = path->nodes[0];
1636 fi = btrfs_item_ptr(leaf, path->slots[0],
1637 struct btrfs_file_extent_item);
1638 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1639 btrfs_set_file_extent_type(leaf, fi, extent_type);
1640 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1641 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1642 btrfs_set_file_extent_offset(leaf, fi, 0);
1643 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1644 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1645 btrfs_set_file_extent_compression(leaf, fi, compression);
1646 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1647 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1649 btrfs_unlock_up_safe(path, 1);
1650 btrfs_set_lock_blocking(leaf);
1652 btrfs_mark_buffer_dirty(leaf);
1654 inode_add_bytes(inode, num_bytes);
1656 ins.objectid = disk_bytenr;
1657 ins.offset = disk_num_bytes;
1658 ins.type = BTRFS_EXTENT_ITEM_KEY;
1659 ret = btrfs_alloc_reserved_file_extent(trans, root,
1660 root->root_key.objectid,
1661 inode->i_ino, file_pos, &ins);
1663 btrfs_free_path(path);
1669 * helper function for btrfs_finish_ordered_io, this
1670 * just reads in some of the csum leaves to prime them into ram
1671 * before we start the transaction. It limits the amount of btree
1672 * reads required while inside the transaction.
1674 static noinline void reada_csum(struct btrfs_root *root,
1675 struct btrfs_path *path,
1676 struct btrfs_ordered_extent *ordered_extent)
1678 struct btrfs_ordered_sum *sum;
1681 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1683 bytenr = sum->sums[0].bytenr;
1686 * we don't care about the results, the point of this search is
1687 * just to get the btree leaves into ram
1689 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1692 /* as ordered data IO finishes, this gets called so we can finish
1693 * an ordered extent if the range of bytes in the file it covers are
1696 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1698 struct btrfs_root *root = BTRFS_I(inode)->root;
1699 struct btrfs_trans_handle *trans;
1700 struct btrfs_ordered_extent *ordered_extent = NULL;
1701 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1702 struct btrfs_path *path;
1706 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1711 * before we join the transaction, try to do some of our IO.
1712 * This will limit the amount of IO that we have to do with
1713 * the transaction running. We're unlikely to need to do any
1714 * IO if the file extents are new, the disk_i_size checks
1715 * covers the most common case.
1717 if (start < BTRFS_I(inode)->disk_i_size) {
1718 path = btrfs_alloc_path();
1720 ret = btrfs_lookup_file_extent(NULL, root, path,
1723 ordered_extent = btrfs_lookup_ordered_extent(inode,
1725 if (!list_empty(&ordered_extent->list)) {
1726 btrfs_release_path(root, path);
1727 reada_csum(root, path, ordered_extent);
1729 btrfs_free_path(path);
1733 trans = btrfs_join_transaction(root, 1);
1735 if (!ordered_extent)
1736 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1737 BUG_ON(!ordered_extent);
1738 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1741 lock_extent(io_tree, ordered_extent->file_offset,
1742 ordered_extent->file_offset + ordered_extent->len - 1,
1745 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1747 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1749 ret = btrfs_mark_extent_written(trans, root, inode,
1750 ordered_extent->file_offset,
1751 ordered_extent->file_offset +
1752 ordered_extent->len);
1755 ret = insert_reserved_file_extent(trans, inode,
1756 ordered_extent->file_offset,
1757 ordered_extent->start,
1758 ordered_extent->disk_len,
1759 ordered_extent->len,
1760 ordered_extent->len,
1761 ordered_extent->file_offset +
1762 ordered_extent->len,
1764 BTRFS_FILE_EXTENT_REG);
1765 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1766 ordered_extent->file_offset,
1767 ordered_extent->len);
1770 unlock_extent(io_tree, ordered_extent->file_offset,
1771 ordered_extent->file_offset + ordered_extent->len - 1,
1774 add_pending_csums(trans, inode, ordered_extent->file_offset,
1775 &ordered_extent->list);
1777 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1778 btrfs_ordered_update_i_size(inode, ordered_extent);
1779 btrfs_update_inode(trans, root, inode);
1780 btrfs_remove_ordered_extent(inode, ordered_extent);
1781 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1784 btrfs_put_ordered_extent(ordered_extent);
1785 /* once for the tree */
1786 btrfs_put_ordered_extent(ordered_extent);
1788 btrfs_end_transaction(trans, root);
1792 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1793 struct extent_state *state, int uptodate)
1795 ClearPagePrivate2(page);
1796 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1800 * When IO fails, either with EIO or csum verification fails, we
1801 * try other mirrors that might have a good copy of the data. This
1802 * io_failure_record is used to record state as we go through all the
1803 * mirrors. If another mirror has good data, the page is set up to date
1804 * and things continue. If a good mirror can't be found, the original
1805 * bio end_io callback is called to indicate things have failed.
1807 struct io_failure_record {
1812 unsigned long bio_flags;
1816 static int btrfs_io_failed_hook(struct bio *failed_bio,
1817 struct page *page, u64 start, u64 end,
1818 struct extent_state *state)
1820 struct io_failure_record *failrec = NULL;
1822 struct extent_map *em;
1823 struct inode *inode = page->mapping->host;
1824 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1825 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1832 ret = get_state_private(failure_tree, start, &private);
1834 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1837 failrec->start = start;
1838 failrec->len = end - start + 1;
1839 failrec->last_mirror = 0;
1840 failrec->bio_flags = 0;
1842 read_lock(&em_tree->lock);
1843 em = lookup_extent_mapping(em_tree, start, failrec->len);
1844 if (em->start > start || em->start + em->len < start) {
1845 free_extent_map(em);
1848 read_unlock(&em_tree->lock);
1850 if (!em || IS_ERR(em)) {
1854 logical = start - em->start;
1855 logical = em->block_start + logical;
1856 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1857 logical = em->block_start;
1858 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1860 failrec->logical = logical;
1861 free_extent_map(em);
1862 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1863 EXTENT_DIRTY, GFP_NOFS);
1864 set_state_private(failure_tree, start,
1865 (u64)(unsigned long)failrec);
1867 failrec = (struct io_failure_record *)(unsigned long)private;
1869 num_copies = btrfs_num_copies(
1870 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1871 failrec->logical, failrec->len);
1872 failrec->last_mirror++;
1874 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1875 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1878 if (state && state->start != failrec->start)
1880 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1882 if (!state || failrec->last_mirror > num_copies) {
1883 set_state_private(failure_tree, failrec->start, 0);
1884 clear_extent_bits(failure_tree, failrec->start,
1885 failrec->start + failrec->len - 1,
1886 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1890 bio = bio_alloc(GFP_NOFS, 1);
1891 bio->bi_private = state;
1892 bio->bi_end_io = failed_bio->bi_end_io;
1893 bio->bi_sector = failrec->logical >> 9;
1894 bio->bi_bdev = failed_bio->bi_bdev;
1897 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1898 if (failed_bio->bi_rw & (1 << BIO_RW))
1903 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1904 failrec->last_mirror,
1905 failrec->bio_flags);
1910 * each time an IO finishes, we do a fast check in the IO failure tree
1911 * to see if we need to process or clean up an io_failure_record
1913 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1916 u64 private_failure;
1917 struct io_failure_record *failure;
1921 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1922 (u64)-1, 1, EXTENT_DIRTY)) {
1923 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1924 start, &private_failure);
1926 failure = (struct io_failure_record *)(unsigned long)
1928 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1930 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1932 failure->start + failure->len - 1,
1933 EXTENT_DIRTY | EXTENT_LOCKED,
1942 * when reads are done, we need to check csums to verify the data is correct
1943 * if there's a match, we allow the bio to finish. If not, we go through
1944 * the io_failure_record routines to find good copies
1946 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1947 struct extent_state *state)
1949 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1950 struct inode *inode = page->mapping->host;
1951 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1953 u64 private = ~(u32)0;
1955 struct btrfs_root *root = BTRFS_I(inode)->root;
1958 if (PageChecked(page)) {
1959 ClearPageChecked(page);
1963 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1966 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1967 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1968 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1973 if (state && state->start == start) {
1974 private = state->private;
1977 ret = get_state_private(io_tree, start, &private);
1979 kaddr = kmap_atomic(page, KM_USER0);
1983 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1984 btrfs_csum_final(csum, (char *)&csum);
1985 if (csum != private)
1988 kunmap_atomic(kaddr, KM_USER0);
1990 /* if the io failure tree for this inode is non-empty,
1991 * check to see if we've recovered from a failed IO
1993 btrfs_clean_io_failures(inode, start);
1997 if (printk_ratelimit()) {
1998 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1999 "private %llu\n", page->mapping->host->i_ino,
2000 (unsigned long long)start, csum,
2001 (unsigned long long)private);
2003 memset(kaddr + offset, 1, end - start + 1);
2004 flush_dcache_page(page);
2005 kunmap_atomic(kaddr, KM_USER0);
2012 * This creates an orphan entry for the given inode in case something goes
2013 * wrong in the middle of an unlink/truncate.
2015 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2017 struct btrfs_root *root = BTRFS_I(inode)->root;
2020 spin_lock(&root->list_lock);
2022 /* already on the orphan list, we're good */
2023 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2024 spin_unlock(&root->list_lock);
2028 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2030 spin_unlock(&root->list_lock);
2033 * insert an orphan item to track this unlinked/truncated file
2035 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2041 * We have done the truncate/delete so we can go ahead and remove the orphan
2042 * item for this particular inode.
2044 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2046 struct btrfs_root *root = BTRFS_I(inode)->root;
2049 spin_lock(&root->list_lock);
2051 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2052 spin_unlock(&root->list_lock);
2056 list_del_init(&BTRFS_I(inode)->i_orphan);
2058 spin_unlock(&root->list_lock);
2062 spin_unlock(&root->list_lock);
2064 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2070 * this cleans up any orphans that may be left on the list from the last use
2073 void btrfs_orphan_cleanup(struct btrfs_root *root)
2075 struct btrfs_path *path;
2076 struct extent_buffer *leaf;
2077 struct btrfs_item *item;
2078 struct btrfs_key key, found_key;
2079 struct btrfs_trans_handle *trans;
2080 struct inode *inode;
2081 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2083 path = btrfs_alloc_path();
2088 key.objectid = BTRFS_ORPHAN_OBJECTID;
2089 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2090 key.offset = (u64)-1;
2094 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2096 printk(KERN_ERR "Error searching slot for orphan: %d"
2102 * if ret == 0 means we found what we were searching for, which
2103 * is weird, but possible, so only screw with path if we didnt
2104 * find the key and see if we have stuff that matches
2107 if (path->slots[0] == 0)
2112 /* pull out the item */
2113 leaf = path->nodes[0];
2114 item = btrfs_item_nr(leaf, path->slots[0]);
2115 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2117 /* make sure the item matches what we want */
2118 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2120 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2123 /* release the path since we're done with it */
2124 btrfs_release_path(root, path);
2127 * this is where we are basically btrfs_lookup, without the
2128 * crossing root thing. we store the inode number in the
2129 * offset of the orphan item.
2131 found_key.objectid = found_key.offset;
2132 found_key.type = BTRFS_INODE_ITEM_KEY;
2133 found_key.offset = 0;
2134 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
2139 * add this inode to the orphan list so btrfs_orphan_del does
2140 * the proper thing when we hit it
2142 spin_lock(&root->list_lock);
2143 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2144 spin_unlock(&root->list_lock);
2147 * if this is a bad inode, means we actually succeeded in
2148 * removing the inode, but not the orphan record, which means
2149 * we need to manually delete the orphan since iput will just
2150 * do a destroy_inode
2152 if (is_bad_inode(inode)) {
2153 trans = btrfs_start_transaction(root, 1);
2154 btrfs_orphan_del(trans, inode);
2155 btrfs_end_transaction(trans, root);
2160 /* if we have links, this was a truncate, lets do that */
2161 if (inode->i_nlink) {
2163 btrfs_truncate(inode);
2168 /* this will do delete_inode and everything for us */
2173 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2175 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2177 btrfs_free_path(path);
2181 * very simple check to peek ahead in the leaf looking for xattrs. If we
2182 * don't find any xattrs, we know there can't be any acls.
2184 * slot is the slot the inode is in, objectid is the objectid of the inode
2186 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2187 int slot, u64 objectid)
2189 u32 nritems = btrfs_header_nritems(leaf);
2190 struct btrfs_key found_key;
2194 while (slot < nritems) {
2195 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2197 /* we found a different objectid, there must not be acls */
2198 if (found_key.objectid != objectid)
2201 /* we found an xattr, assume we've got an acl */
2202 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2206 * we found a key greater than an xattr key, there can't
2207 * be any acls later on
2209 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2216 * it goes inode, inode backrefs, xattrs, extents,
2217 * so if there are a ton of hard links to an inode there can
2218 * be a lot of backrefs. Don't waste time searching too hard,
2219 * this is just an optimization
2224 /* we hit the end of the leaf before we found an xattr or
2225 * something larger than an xattr. We have to assume the inode
2232 * read an inode from the btree into the in-memory inode
2234 static void btrfs_read_locked_inode(struct inode *inode)
2236 struct btrfs_path *path;
2237 struct extent_buffer *leaf;
2238 struct btrfs_inode_item *inode_item;
2239 struct btrfs_timespec *tspec;
2240 struct btrfs_root *root = BTRFS_I(inode)->root;
2241 struct btrfs_key location;
2243 u64 alloc_group_block;
2247 path = btrfs_alloc_path();
2249 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2251 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2255 leaf = path->nodes[0];
2256 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2257 struct btrfs_inode_item);
2259 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2260 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2261 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2262 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2263 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2265 tspec = btrfs_inode_atime(inode_item);
2266 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2267 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2269 tspec = btrfs_inode_mtime(inode_item);
2270 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2271 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2273 tspec = btrfs_inode_ctime(inode_item);
2274 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2275 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2277 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2278 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2279 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2280 inode->i_generation = BTRFS_I(inode)->generation;
2282 rdev = btrfs_inode_rdev(leaf, inode_item);
2284 BTRFS_I(inode)->index_cnt = (u64)-1;
2285 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2287 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2290 * try to precache a NULL acl entry for files that don't have
2291 * any xattrs or acls
2293 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2295 cache_no_acl(inode);
2297 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2298 alloc_group_block, 0);
2299 btrfs_free_path(path);
2302 switch (inode->i_mode & S_IFMT) {
2304 inode->i_mapping->a_ops = &btrfs_aops;
2305 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2306 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2307 inode->i_fop = &btrfs_file_operations;
2308 inode->i_op = &btrfs_file_inode_operations;
2311 inode->i_fop = &btrfs_dir_file_operations;
2312 if (root == root->fs_info->tree_root)
2313 inode->i_op = &btrfs_dir_ro_inode_operations;
2315 inode->i_op = &btrfs_dir_inode_operations;
2318 inode->i_op = &btrfs_symlink_inode_operations;
2319 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2320 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2323 inode->i_op = &btrfs_special_inode_operations;
2324 init_special_inode(inode, inode->i_mode, rdev);
2328 btrfs_update_iflags(inode);
2332 btrfs_free_path(path);
2333 make_bad_inode(inode);
2337 * given a leaf and an inode, copy the inode fields into the leaf
2339 static void fill_inode_item(struct btrfs_trans_handle *trans,
2340 struct extent_buffer *leaf,
2341 struct btrfs_inode_item *item,
2342 struct inode *inode)
2344 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2345 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2346 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2347 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2348 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2350 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2351 inode->i_atime.tv_sec);
2352 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2353 inode->i_atime.tv_nsec);
2355 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2356 inode->i_mtime.tv_sec);
2357 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2358 inode->i_mtime.tv_nsec);
2360 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2361 inode->i_ctime.tv_sec);
2362 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2363 inode->i_ctime.tv_nsec);
2365 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2366 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2367 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2368 btrfs_set_inode_transid(leaf, item, trans->transid);
2369 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2370 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2371 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2375 * copy everything in the in-memory inode into the btree.
2377 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2378 struct btrfs_root *root, struct inode *inode)
2380 struct btrfs_inode_item *inode_item;
2381 struct btrfs_path *path;
2382 struct extent_buffer *leaf;
2385 path = btrfs_alloc_path();
2387 path->leave_spinning = 1;
2388 ret = btrfs_lookup_inode(trans, root, path,
2389 &BTRFS_I(inode)->location, 1);
2396 btrfs_unlock_up_safe(path, 1);
2397 leaf = path->nodes[0];
2398 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2399 struct btrfs_inode_item);
2401 fill_inode_item(trans, leaf, inode_item, inode);
2402 btrfs_mark_buffer_dirty(leaf);
2403 btrfs_set_inode_last_trans(trans, inode);
2406 btrfs_free_path(path);
2412 * unlink helper that gets used here in inode.c and in the tree logging
2413 * recovery code. It remove a link in a directory with a given name, and
2414 * also drops the back refs in the inode to the directory
2416 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2417 struct btrfs_root *root,
2418 struct inode *dir, struct inode *inode,
2419 const char *name, int name_len)
2421 struct btrfs_path *path;
2423 struct extent_buffer *leaf;
2424 struct btrfs_dir_item *di;
2425 struct btrfs_key key;
2428 path = btrfs_alloc_path();
2434 path->leave_spinning = 1;
2435 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2436 name, name_len, -1);
2445 leaf = path->nodes[0];
2446 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2447 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2450 btrfs_release_path(root, path);
2452 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2454 dir->i_ino, &index);
2456 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2457 "inode %lu parent %lu\n", name_len, name,
2458 inode->i_ino, dir->i_ino);
2462 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2463 index, name, name_len, -1);
2472 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2473 btrfs_release_path(root, path);
2475 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2477 BUG_ON(ret != 0 && ret != -ENOENT);
2479 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2483 btrfs_free_path(path);
2487 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2488 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2489 btrfs_update_inode(trans, root, dir);
2490 btrfs_drop_nlink(inode);
2491 ret = btrfs_update_inode(trans, root, inode);
2496 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2498 struct btrfs_root *root;
2499 struct btrfs_trans_handle *trans;
2500 struct inode *inode = dentry->d_inode;
2502 unsigned long nr = 0;
2504 root = BTRFS_I(dir)->root;
2507 * 5 items for unlink inode
2510 ret = btrfs_reserve_metadata_space(root, 6);
2514 trans = btrfs_start_transaction(root, 1);
2515 if (IS_ERR(trans)) {
2516 btrfs_unreserve_metadata_space(root, 6);
2517 return PTR_ERR(trans);
2520 btrfs_set_trans_block_group(trans, dir);
2522 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2524 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2525 dentry->d_name.name, dentry->d_name.len);
2527 if (inode->i_nlink == 0)
2528 ret = btrfs_orphan_add(trans, inode);
2530 nr = trans->blocks_used;
2532 btrfs_end_transaction_throttle(trans, root);
2533 btrfs_unreserve_metadata_space(root, 6);
2534 btrfs_btree_balance_dirty(root, nr);
2538 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2539 struct btrfs_root *root,
2540 struct inode *dir, u64 objectid,
2541 const char *name, int name_len)
2543 struct btrfs_path *path;
2544 struct extent_buffer *leaf;
2545 struct btrfs_dir_item *di;
2546 struct btrfs_key key;
2550 path = btrfs_alloc_path();
2554 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2555 name, name_len, -1);
2556 BUG_ON(!di || IS_ERR(di));
2558 leaf = path->nodes[0];
2559 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2560 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2561 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2563 btrfs_release_path(root, path);
2565 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2566 objectid, root->root_key.objectid,
2567 dir->i_ino, &index, name, name_len);
2569 BUG_ON(ret != -ENOENT);
2570 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2572 BUG_ON(!di || IS_ERR(di));
2574 leaf = path->nodes[0];
2575 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2576 btrfs_release_path(root, path);
2580 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2581 index, name, name_len, -1);
2582 BUG_ON(!di || IS_ERR(di));
2584 leaf = path->nodes[0];
2585 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2586 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2587 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2589 btrfs_release_path(root, path);
2591 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2592 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2593 ret = btrfs_update_inode(trans, root, dir);
2595 dir->i_sb->s_dirt = 1;
2597 btrfs_free_path(path);
2601 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2603 struct inode *inode = dentry->d_inode;
2606 struct btrfs_root *root = BTRFS_I(dir)->root;
2607 struct btrfs_trans_handle *trans;
2608 unsigned long nr = 0;
2610 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2611 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2614 ret = btrfs_reserve_metadata_space(root, 5);
2618 trans = btrfs_start_transaction(root, 1);
2619 if (IS_ERR(trans)) {
2620 btrfs_unreserve_metadata_space(root, 5);
2621 return PTR_ERR(trans);
2624 btrfs_set_trans_block_group(trans, dir);
2626 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2627 err = btrfs_unlink_subvol(trans, root, dir,
2628 BTRFS_I(inode)->location.objectid,
2629 dentry->d_name.name,
2630 dentry->d_name.len);
2634 err = btrfs_orphan_add(trans, inode);
2638 /* now the directory is empty */
2639 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2640 dentry->d_name.name, dentry->d_name.len);
2642 btrfs_i_size_write(inode, 0);
2644 nr = trans->blocks_used;
2645 ret = btrfs_end_transaction_throttle(trans, root);
2646 btrfs_unreserve_metadata_space(root, 5);
2647 btrfs_btree_balance_dirty(root, nr);
2656 * when truncating bytes in a file, it is possible to avoid reading
2657 * the leaves that contain only checksum items. This can be the
2658 * majority of the IO required to delete a large file, but it must
2659 * be done carefully.
2661 * The keys in the level just above the leaves are checked to make sure
2662 * the lowest key in a given leaf is a csum key, and starts at an offset
2663 * after the new size.
2665 * Then the key for the next leaf is checked to make sure it also has
2666 * a checksum item for the same file. If it does, we know our target leaf
2667 * contains only checksum items, and it can be safely freed without reading
2670 * This is just an optimization targeted at large files. It may do
2671 * nothing. It will return 0 unless things went badly.
2673 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2674 struct btrfs_root *root,
2675 struct btrfs_path *path,
2676 struct inode *inode, u64 new_size)
2678 struct btrfs_key key;
2681 struct btrfs_key found_key;
2682 struct btrfs_key other_key;
2683 struct btrfs_leaf_ref *ref;
2687 path->lowest_level = 1;
2688 key.objectid = inode->i_ino;
2689 key.type = BTRFS_CSUM_ITEM_KEY;
2690 key.offset = new_size;
2692 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2696 if (path->nodes[1] == NULL) {
2701 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2702 nritems = btrfs_header_nritems(path->nodes[1]);
2707 if (path->slots[1] >= nritems)
2710 /* did we find a key greater than anything we want to delete? */
2711 if (found_key.objectid > inode->i_ino ||
2712 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2715 /* we check the next key in the node to make sure the leave contains
2716 * only checksum items. This comparison doesn't work if our
2717 * leaf is the last one in the node
2719 if (path->slots[1] + 1 >= nritems) {
2721 /* search forward from the last key in the node, this
2722 * will bring us into the next node in the tree
2724 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2726 /* unlikely, but we inc below, so check to be safe */
2727 if (found_key.offset == (u64)-1)
2730 /* search_forward needs a path with locks held, do the
2731 * search again for the original key. It is possible
2732 * this will race with a balance and return a path that
2733 * we could modify, but this drop is just an optimization
2734 * and is allowed to miss some leaves.
2736 btrfs_release_path(root, path);
2739 /* setup a max key for search_forward */
2740 other_key.offset = (u64)-1;
2741 other_key.type = key.type;
2742 other_key.objectid = key.objectid;
2744 path->keep_locks = 1;
2745 ret = btrfs_search_forward(root, &found_key, &other_key,
2747 path->keep_locks = 0;
2748 if (ret || found_key.objectid != key.objectid ||
2749 found_key.type != key.type) {
2754 key.offset = found_key.offset;
2755 btrfs_release_path(root, path);
2760 /* we know there's one more slot after us in the tree,
2761 * read that key so we can verify it is also a checksum item
2763 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2765 if (found_key.objectid < inode->i_ino)
2768 if (found_key.type != key.type || found_key.offset < new_size)
2772 * if the key for the next leaf isn't a csum key from this objectid,
2773 * we can't be sure there aren't good items inside this leaf.
2776 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2779 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2780 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2782 * it is safe to delete this leaf, it contains only
2783 * csum items from this inode at an offset >= new_size
2785 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2788 if (root->ref_cows && leaf_gen < trans->transid) {
2789 ref = btrfs_alloc_leaf_ref(root, 0);
2791 ref->root_gen = root->root_key.offset;
2792 ref->bytenr = leaf_start;
2794 ref->generation = leaf_gen;
2797 btrfs_sort_leaf_ref(ref);
2799 ret = btrfs_add_leaf_ref(root, ref, 0);
2801 btrfs_free_leaf_ref(root, ref);
2807 btrfs_release_path(root, path);
2809 if (other_key.objectid == inode->i_ino &&
2810 other_key.type == key.type && other_key.offset > key.offset) {
2811 key.offset = other_key.offset;
2817 /* fixup any changes we've made to the path */
2818 path->lowest_level = 0;
2819 path->keep_locks = 0;
2820 btrfs_release_path(root, path);
2827 * this can truncate away extent items, csum items and directory items.
2828 * It starts at a high offset and removes keys until it can't find
2829 * any higher than new_size
2831 * csum items that cross the new i_size are truncated to the new size
2834 * min_type is the minimum key type to truncate down to. If set to 0, this
2835 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2837 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2838 struct btrfs_root *root,
2839 struct inode *inode,
2840 u64 new_size, u32 min_type)
2843 struct btrfs_path *path;
2844 struct btrfs_key key;
2845 struct btrfs_key found_key;
2846 u32 found_type = (u8)-1;
2847 struct extent_buffer *leaf;
2848 struct btrfs_file_extent_item *fi;
2849 u64 extent_start = 0;
2850 u64 extent_num_bytes = 0;
2851 u64 extent_offset = 0;
2855 int pending_del_nr = 0;
2856 int pending_del_slot = 0;
2857 int extent_type = -1;
2859 u64 mask = root->sectorsize - 1;
2862 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2863 path = btrfs_alloc_path();
2867 /* FIXME, add redo link to tree so we don't leak on crash */
2868 key.objectid = inode->i_ino;
2869 key.offset = (u64)-1;
2873 path->leave_spinning = 1;
2874 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2879 /* there are no items in the tree for us to truncate, we're
2882 if (path->slots[0] == 0) {
2891 leaf = path->nodes[0];
2892 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2893 found_type = btrfs_key_type(&found_key);
2896 if (found_key.objectid != inode->i_ino)
2899 if (found_type < min_type)
2902 item_end = found_key.offset;
2903 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2904 fi = btrfs_item_ptr(leaf, path->slots[0],
2905 struct btrfs_file_extent_item);
2906 extent_type = btrfs_file_extent_type(leaf, fi);
2907 encoding = btrfs_file_extent_compression(leaf, fi);
2908 encoding |= btrfs_file_extent_encryption(leaf, fi);
2909 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2911 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2913 btrfs_file_extent_num_bytes(leaf, fi);
2914 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2915 item_end += btrfs_file_extent_inline_len(leaf,
2920 if (item_end < new_size) {
2921 if (found_type == BTRFS_DIR_ITEM_KEY)
2922 found_type = BTRFS_INODE_ITEM_KEY;
2923 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2924 found_type = BTRFS_EXTENT_DATA_KEY;
2925 else if (found_type == BTRFS_EXTENT_DATA_KEY)
2926 found_type = BTRFS_XATTR_ITEM_KEY;
2927 else if (found_type == BTRFS_XATTR_ITEM_KEY)
2928 found_type = BTRFS_INODE_REF_KEY;
2929 else if (found_type)
2933 btrfs_set_key_type(&key, found_type);
2936 if (found_key.offset >= new_size)
2942 /* FIXME, shrink the extent if the ref count is only 1 */
2943 if (found_type != BTRFS_EXTENT_DATA_KEY)
2946 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2948 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2949 if (!del_item && !encoding) {
2950 u64 orig_num_bytes =
2951 btrfs_file_extent_num_bytes(leaf, fi);
2952 extent_num_bytes = new_size -
2953 found_key.offset + root->sectorsize - 1;
2954 extent_num_bytes = extent_num_bytes &
2955 ~((u64)root->sectorsize - 1);
2956 btrfs_set_file_extent_num_bytes(leaf, fi,
2958 num_dec = (orig_num_bytes -
2960 if (root->ref_cows && extent_start != 0)
2961 inode_sub_bytes(inode, num_dec);
2962 btrfs_mark_buffer_dirty(leaf);
2965 btrfs_file_extent_disk_num_bytes(leaf,
2967 extent_offset = found_key.offset -
2968 btrfs_file_extent_offset(leaf, fi);
2970 /* FIXME blocksize != 4096 */
2971 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2972 if (extent_start != 0) {
2975 inode_sub_bytes(inode, num_dec);
2978 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2980 * we can't truncate inline items that have had
2984 btrfs_file_extent_compression(leaf, fi) == 0 &&
2985 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2986 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2987 u32 size = new_size - found_key.offset;
2989 if (root->ref_cows) {
2990 inode_sub_bytes(inode, item_end + 1 -
2994 btrfs_file_extent_calc_inline_size(size);
2995 ret = btrfs_truncate_item(trans, root, path,
2998 } else if (root->ref_cows) {
2999 inode_sub_bytes(inode, item_end + 1 -
3005 if (!pending_del_nr) {
3006 /* no pending yet, add ourselves */
3007 pending_del_slot = path->slots[0];
3009 } else if (pending_del_nr &&
3010 path->slots[0] + 1 == pending_del_slot) {
3011 /* hop on the pending chunk */
3013 pending_del_slot = path->slots[0];
3020 if (found_extent && root->ref_cows) {
3021 btrfs_set_path_blocking(path);
3022 ret = btrfs_free_extent(trans, root, extent_start,
3023 extent_num_bytes, 0,
3024 btrfs_header_owner(leaf),
3025 inode->i_ino, extent_offset);
3029 if (path->slots[0] == 0) {
3032 btrfs_release_path(root, path);
3033 if (found_type == BTRFS_INODE_ITEM_KEY)
3039 if (pending_del_nr &&
3040 path->slots[0] + 1 != pending_del_slot) {
3041 struct btrfs_key debug;
3043 btrfs_item_key_to_cpu(path->nodes[0], &debug,
3045 ret = btrfs_del_items(trans, root, path,
3050 btrfs_release_path(root, path);
3051 if (found_type == BTRFS_INODE_ITEM_KEY)
3058 if (pending_del_nr) {
3059 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3062 btrfs_free_path(path);
3067 * taken from block_truncate_page, but does cow as it zeros out
3068 * any bytes left in the last page in the file.
3070 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3072 struct inode *inode = mapping->host;
3073 struct btrfs_root *root = BTRFS_I(inode)->root;
3074 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3075 struct btrfs_ordered_extent *ordered;
3077 u32 blocksize = root->sectorsize;
3078 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3079 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3085 if ((offset & (blocksize - 1)) == 0)
3087 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3091 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3097 page = grab_cache_page(mapping, index);
3099 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3100 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3104 page_start = page_offset(page);
3105 page_end = page_start + PAGE_CACHE_SIZE - 1;
3107 if (!PageUptodate(page)) {
3108 ret = btrfs_readpage(NULL, page);
3110 if (page->mapping != mapping) {
3112 page_cache_release(page);
3115 if (!PageUptodate(page)) {
3120 wait_on_page_writeback(page);
3122 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3123 set_page_extent_mapped(page);
3125 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3127 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3129 page_cache_release(page);
3130 btrfs_start_ordered_extent(inode, ordered, 1);
3131 btrfs_put_ordered_extent(ordered);
3135 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
3136 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3139 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3141 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3146 if (offset != PAGE_CACHE_SIZE) {
3148 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3149 flush_dcache_page(page);
3152 ClearPageChecked(page);
3153 set_page_dirty(page);
3154 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3158 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3159 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3161 page_cache_release(page);
3166 int btrfs_cont_expand(struct inode *inode, loff_t size)
3168 struct btrfs_trans_handle *trans;
3169 struct btrfs_root *root = BTRFS_I(inode)->root;
3170 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3171 struct extent_map *em;
3172 u64 mask = root->sectorsize - 1;
3173 u64 hole_start = (inode->i_size + mask) & ~mask;
3174 u64 block_end = (size + mask) & ~mask;
3180 if (size <= hole_start)
3183 err = btrfs_truncate_page(inode->i_mapping, inode->i_size);
3188 struct btrfs_ordered_extent *ordered;
3189 btrfs_wait_ordered_range(inode, hole_start,
3190 block_end - hole_start);
3191 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3192 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3195 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3196 btrfs_put_ordered_extent(ordered);
3199 trans = btrfs_start_transaction(root, 1);
3200 btrfs_set_trans_block_group(trans, inode);
3202 cur_offset = hole_start;
3204 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3205 block_end - cur_offset, 0);
3206 BUG_ON(IS_ERR(em) || !em);
3207 last_byte = min(extent_map_end(em), block_end);
3208 last_byte = (last_byte + mask) & ~mask;
3209 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
3211 hole_size = last_byte - cur_offset;
3212 err = btrfs_drop_extents(trans, root, inode,
3214 cur_offset + hole_size,
3216 cur_offset, &hint_byte, 1);
3220 err = btrfs_reserve_metadata_space(root, 1);
3224 err = btrfs_insert_file_extent(trans, root,
3225 inode->i_ino, cur_offset, 0,
3226 0, hole_size, 0, hole_size,
3228 btrfs_drop_extent_cache(inode, hole_start,
3230 btrfs_unreserve_metadata_space(root, 1);
3232 free_extent_map(em);
3233 cur_offset = last_byte;
3234 if (err || cur_offset >= block_end)
3238 btrfs_end_transaction(trans, root);
3239 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3243 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3245 struct inode *inode = dentry->d_inode;
3248 err = inode_change_ok(inode, attr);
3252 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3253 if (attr->ia_size > inode->i_size) {
3254 err = btrfs_cont_expand(inode, attr->ia_size);
3257 } else if (inode->i_size > 0 &&
3258 attr->ia_size == 0) {
3260 /* we're truncating a file that used to have good
3261 * data down to zero. Make sure it gets into
3262 * the ordered flush list so that any new writes
3263 * get down to disk quickly.
3265 BTRFS_I(inode)->ordered_data_close = 1;
3269 err = inode_setattr(inode, attr);
3271 if (!err && ((attr->ia_valid & ATTR_MODE)))
3272 err = btrfs_acl_chmod(inode);
3276 void btrfs_delete_inode(struct inode *inode)
3278 struct btrfs_trans_handle *trans;
3279 struct btrfs_root *root = BTRFS_I(inode)->root;
3283 truncate_inode_pages(&inode->i_data, 0);
3284 if (is_bad_inode(inode)) {
3285 btrfs_orphan_del(NULL, inode);
3288 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3290 if (inode->i_nlink > 0) {
3291 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3295 btrfs_i_size_write(inode, 0);
3296 trans = btrfs_join_transaction(root, 1);
3298 btrfs_set_trans_block_group(trans, inode);
3299 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3301 btrfs_orphan_del(NULL, inode);
3302 goto no_delete_lock;
3305 btrfs_orphan_del(trans, inode);
3307 nr = trans->blocks_used;
3310 btrfs_end_transaction(trans, root);
3311 btrfs_btree_balance_dirty(root, nr);
3315 nr = trans->blocks_used;
3316 btrfs_end_transaction(trans, root);
3317 btrfs_btree_balance_dirty(root, nr);
3323 * this returns the key found in the dir entry in the location pointer.
3324 * If no dir entries were found, location->objectid is 0.
3326 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3327 struct btrfs_key *location)
3329 const char *name = dentry->d_name.name;
3330 int namelen = dentry->d_name.len;
3331 struct btrfs_dir_item *di;
3332 struct btrfs_path *path;
3333 struct btrfs_root *root = BTRFS_I(dir)->root;
3336 path = btrfs_alloc_path();
3339 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3344 if (!di || IS_ERR(di))
3347 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3349 btrfs_free_path(path);
3352 location->objectid = 0;
3357 * when we hit a tree root in a directory, the btrfs part of the inode
3358 * needs to be changed to reflect the root directory of the tree root. This
3359 * is kind of like crossing a mount point.
3361 static int fixup_tree_root_location(struct btrfs_root *root,
3363 struct dentry *dentry,
3364 struct btrfs_key *location,
3365 struct btrfs_root **sub_root)
3367 struct btrfs_path *path;
3368 struct btrfs_root *new_root;
3369 struct btrfs_root_ref *ref;
3370 struct extent_buffer *leaf;
3374 path = btrfs_alloc_path();
3381 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3382 BTRFS_I(dir)->root->root_key.objectid,
3383 location->objectid);
3390 leaf = path->nodes[0];
3391 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3392 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3393 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3396 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3397 (unsigned long)(ref + 1),
3398 dentry->d_name.len);
3402 btrfs_release_path(root->fs_info->tree_root, path);
3404 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3405 if (IS_ERR(new_root)) {
3406 err = PTR_ERR(new_root);
3410 if (btrfs_root_refs(&new_root->root_item) == 0) {
3415 *sub_root = new_root;
3416 location->objectid = btrfs_root_dirid(&new_root->root_item);
3417 location->type = BTRFS_INODE_ITEM_KEY;
3418 location->offset = 0;
3421 btrfs_free_path(path);
3425 static void inode_tree_add(struct inode *inode)
3427 struct btrfs_root *root = BTRFS_I(inode)->root;
3428 struct btrfs_inode *entry;
3430 struct rb_node *parent;
3432 p = &root->inode_tree.rb_node;
3435 if (hlist_unhashed(&inode->i_hash))
3438 spin_lock(&root->inode_lock);
3441 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3443 if (inode->i_ino < entry->vfs_inode.i_ino)
3444 p = &parent->rb_left;
3445 else if (inode->i_ino > entry->vfs_inode.i_ino)
3446 p = &parent->rb_right;
3448 WARN_ON(!(entry->vfs_inode.i_state &
3449 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3450 rb_erase(parent, &root->inode_tree);
3451 RB_CLEAR_NODE(parent);
3452 spin_unlock(&root->inode_lock);
3456 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3457 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3458 spin_unlock(&root->inode_lock);
3461 static void inode_tree_del(struct inode *inode)
3463 struct btrfs_root *root = BTRFS_I(inode)->root;
3466 spin_lock(&root->inode_lock);
3467 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3468 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3469 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3470 empty = RB_EMPTY_ROOT(&root->inode_tree);
3472 spin_unlock(&root->inode_lock);
3474 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3475 synchronize_srcu(&root->fs_info->subvol_srcu);
3476 spin_lock(&root->inode_lock);
3477 empty = RB_EMPTY_ROOT(&root->inode_tree);
3478 spin_unlock(&root->inode_lock);
3480 btrfs_add_dead_root(root);
3484 int btrfs_invalidate_inodes(struct btrfs_root *root)
3486 struct rb_node *node;
3487 struct rb_node *prev;
3488 struct btrfs_inode *entry;
3489 struct inode *inode;
3492 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3494 spin_lock(&root->inode_lock);
3496 node = root->inode_tree.rb_node;
3500 entry = rb_entry(node, struct btrfs_inode, rb_node);
3502 if (objectid < entry->vfs_inode.i_ino)
3503 node = node->rb_left;
3504 else if (objectid > entry->vfs_inode.i_ino)
3505 node = node->rb_right;
3511 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3512 if (objectid <= entry->vfs_inode.i_ino) {
3516 prev = rb_next(prev);
3520 entry = rb_entry(node, struct btrfs_inode, rb_node);
3521 objectid = entry->vfs_inode.i_ino + 1;
3522 inode = igrab(&entry->vfs_inode);
3524 spin_unlock(&root->inode_lock);
3525 if (atomic_read(&inode->i_count) > 1)
3526 d_prune_aliases(inode);
3528 * btrfs_drop_inode will remove it from
3529 * the inode cache when its usage count
3534 spin_lock(&root->inode_lock);
3538 if (cond_resched_lock(&root->inode_lock))
3541 node = rb_next(node);
3543 spin_unlock(&root->inode_lock);
3547 static noinline void init_btrfs_i(struct inode *inode)
3549 struct btrfs_inode *bi = BTRFS_I(inode);
3554 bi->last_sub_trans = 0;
3555 bi->logged_trans = 0;
3556 bi->delalloc_bytes = 0;
3557 bi->reserved_bytes = 0;
3558 bi->disk_i_size = 0;
3560 bi->index_cnt = (u64)-1;
3561 bi->last_unlink_trans = 0;
3562 bi->ordered_data_close = 0;
3563 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3564 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3565 inode->i_mapping, GFP_NOFS);
3566 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3567 inode->i_mapping, GFP_NOFS);
3568 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3569 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3570 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3571 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3572 mutex_init(&BTRFS_I(inode)->extent_mutex);
3573 mutex_init(&BTRFS_I(inode)->log_mutex);
3576 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3578 struct btrfs_iget_args *args = p;
3579 inode->i_ino = args->ino;
3580 init_btrfs_i(inode);
3581 BTRFS_I(inode)->root = args->root;
3582 btrfs_set_inode_space_info(args->root, inode);
3586 static int btrfs_find_actor(struct inode *inode, void *opaque)
3588 struct btrfs_iget_args *args = opaque;
3589 return args->ino == inode->i_ino &&
3590 args->root == BTRFS_I(inode)->root;
3593 static struct inode *btrfs_iget_locked(struct super_block *s,
3595 struct btrfs_root *root)
3597 struct inode *inode;
3598 struct btrfs_iget_args args;
3599 args.ino = objectid;
3602 inode = iget5_locked(s, objectid, btrfs_find_actor,
3603 btrfs_init_locked_inode,
3608 /* Get an inode object given its location and corresponding root.
3609 * Returns in *is_new if the inode was read from disk
3611 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3612 struct btrfs_root *root)
3614 struct inode *inode;
3616 inode = btrfs_iget_locked(s, location->objectid, root);
3618 return ERR_PTR(-ENOMEM);
3620 if (inode->i_state & I_NEW) {
3621 BTRFS_I(inode)->root = root;
3622 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3623 btrfs_read_locked_inode(inode);
3625 inode_tree_add(inode);
3626 unlock_new_inode(inode);
3632 static struct inode *new_simple_dir(struct super_block *s,
3633 struct btrfs_key *key,
3634 struct btrfs_root *root)
3636 struct inode *inode = new_inode(s);
3639 return ERR_PTR(-ENOMEM);
3641 init_btrfs_i(inode);
3643 BTRFS_I(inode)->root = root;
3644 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3645 BTRFS_I(inode)->dummy_inode = 1;
3647 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3648 inode->i_op = &simple_dir_inode_operations;
3649 inode->i_fop = &simple_dir_operations;
3650 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3651 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3656 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3658 struct inode *inode;
3659 struct btrfs_root *root = BTRFS_I(dir)->root;
3660 struct btrfs_root *sub_root = root;
3661 struct btrfs_key location;
3665 dentry->d_op = &btrfs_dentry_operations;
3667 if (dentry->d_name.len > BTRFS_NAME_LEN)
3668 return ERR_PTR(-ENAMETOOLONG);
3670 ret = btrfs_inode_by_name(dir, dentry, &location);
3673 return ERR_PTR(ret);
3675 if (location.objectid == 0)
3678 if (location.type == BTRFS_INODE_ITEM_KEY) {
3679 inode = btrfs_iget(dir->i_sb, &location, root);
3683 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3685 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3686 ret = fixup_tree_root_location(root, dir, dentry,
3687 &location, &sub_root);
3690 inode = ERR_PTR(ret);
3692 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3694 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3696 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3701 static int btrfs_dentry_delete(struct dentry *dentry)
3703 struct btrfs_root *root;
3705 if (!dentry->d_inode && !IS_ROOT(dentry))
3706 dentry = dentry->d_parent;
3708 if (dentry->d_inode) {
3709 root = BTRFS_I(dentry->d_inode)->root;
3710 if (btrfs_root_refs(&root->root_item) == 0)
3716 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3717 struct nameidata *nd)
3719 struct inode *inode;
3721 inode = btrfs_lookup_dentry(dir, dentry);
3723 return ERR_CAST(inode);
3725 return d_splice_alias(inode, dentry);
3728 static unsigned char btrfs_filetype_table[] = {
3729 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3732 static int btrfs_real_readdir(struct file *filp, void *dirent,
3735 struct inode *inode = filp->f_dentry->d_inode;
3736 struct btrfs_root *root = BTRFS_I(inode)->root;
3737 struct btrfs_item *item;
3738 struct btrfs_dir_item *di;
3739 struct btrfs_key key;
3740 struct btrfs_key found_key;
3741 struct btrfs_path *path;
3744 struct extent_buffer *leaf;
3747 unsigned char d_type;
3752 int key_type = BTRFS_DIR_INDEX_KEY;
3757 /* FIXME, use a real flag for deciding about the key type */
3758 if (root->fs_info->tree_root == root)
3759 key_type = BTRFS_DIR_ITEM_KEY;
3761 /* special case for "." */
3762 if (filp->f_pos == 0) {
3763 over = filldir(dirent, ".", 1,
3770 /* special case for .., just use the back ref */
3771 if (filp->f_pos == 1) {
3772 u64 pino = parent_ino(filp->f_path.dentry);
3773 over = filldir(dirent, "..", 2,
3779 path = btrfs_alloc_path();
3782 btrfs_set_key_type(&key, key_type);
3783 key.offset = filp->f_pos;
3784 key.objectid = inode->i_ino;
3786 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3792 leaf = path->nodes[0];
3793 nritems = btrfs_header_nritems(leaf);
3794 slot = path->slots[0];
3795 if (advance || slot >= nritems) {
3796 if (slot >= nritems - 1) {
3797 ret = btrfs_next_leaf(root, path);
3800 leaf = path->nodes[0];
3801 nritems = btrfs_header_nritems(leaf);
3802 slot = path->slots[0];
3810 item = btrfs_item_nr(leaf, slot);
3811 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3813 if (found_key.objectid != key.objectid)
3815 if (btrfs_key_type(&found_key) != key_type)
3817 if (found_key.offset < filp->f_pos)
3820 filp->f_pos = found_key.offset;
3822 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3824 di_total = btrfs_item_size(leaf, item);
3826 while (di_cur < di_total) {
3827 struct btrfs_key location;
3829 name_len = btrfs_dir_name_len(leaf, di);
3830 if (name_len <= sizeof(tmp_name)) {
3831 name_ptr = tmp_name;
3833 name_ptr = kmalloc(name_len, GFP_NOFS);
3839 read_extent_buffer(leaf, name_ptr,
3840 (unsigned long)(di + 1), name_len);
3842 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3843 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3845 /* is this a reference to our own snapshot? If so
3848 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3849 location.objectid == root->root_key.objectid) {
3853 over = filldir(dirent, name_ptr, name_len,
3854 found_key.offset, location.objectid,
3858 if (name_ptr != tmp_name)
3863 di_len = btrfs_dir_name_len(leaf, di) +
3864 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3866 di = (struct btrfs_dir_item *)((char *)di + di_len);
3870 /* Reached end of directory/root. Bump pos past the last item. */
3871 if (key_type == BTRFS_DIR_INDEX_KEY)
3872 filp->f_pos = INT_LIMIT(off_t);
3878 btrfs_free_path(path);
3882 int btrfs_write_inode(struct inode *inode, int wait)
3884 struct btrfs_root *root = BTRFS_I(inode)->root;
3885 struct btrfs_trans_handle *trans;
3888 if (root->fs_info->btree_inode == inode)
3892 trans = btrfs_join_transaction(root, 1);
3893 btrfs_set_trans_block_group(trans, inode);
3894 ret = btrfs_commit_transaction(trans, root);
3900 * This is somewhat expensive, updating the tree every time the
3901 * inode changes. But, it is most likely to find the inode in cache.
3902 * FIXME, needs more benchmarking...there are no reasons other than performance
3903 * to keep or drop this code.
3905 void btrfs_dirty_inode(struct inode *inode)
3907 struct btrfs_root *root = BTRFS_I(inode)->root;
3908 struct btrfs_trans_handle *trans;
3910 trans = btrfs_join_transaction(root, 1);
3911 btrfs_set_trans_block_group(trans, inode);
3912 btrfs_update_inode(trans, root, inode);
3913 btrfs_end_transaction(trans, root);
3917 * find the highest existing sequence number in a directory
3918 * and then set the in-memory index_cnt variable to reflect
3919 * free sequence numbers
3921 static int btrfs_set_inode_index_count(struct inode *inode)
3923 struct btrfs_root *root = BTRFS_I(inode)->root;
3924 struct btrfs_key key, found_key;
3925 struct btrfs_path *path;
3926 struct extent_buffer *leaf;
3929 key.objectid = inode->i_ino;
3930 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3931 key.offset = (u64)-1;
3933 path = btrfs_alloc_path();
3937 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3940 /* FIXME: we should be able to handle this */
3946 * MAGIC NUMBER EXPLANATION:
3947 * since we search a directory based on f_pos we have to start at 2
3948 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3949 * else has to start at 2
3951 if (path->slots[0] == 0) {
3952 BTRFS_I(inode)->index_cnt = 2;
3958 leaf = path->nodes[0];
3959 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3961 if (found_key.objectid != inode->i_ino ||
3962 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3963 BTRFS_I(inode)->index_cnt = 2;
3967 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3969 btrfs_free_path(path);
3974 * helper to find a free sequence number in a given directory. This current
3975 * code is very simple, later versions will do smarter things in the btree
3977 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3981 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3982 ret = btrfs_set_inode_index_count(dir);
3987 *index = BTRFS_I(dir)->index_cnt;
3988 BTRFS_I(dir)->index_cnt++;
3993 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3994 struct btrfs_root *root,
3996 const char *name, int name_len,
3997 u64 ref_objectid, u64 objectid,
3998 u64 alloc_hint, int mode, u64 *index)
4000 struct inode *inode;
4001 struct btrfs_inode_item *inode_item;
4002 struct btrfs_key *location;
4003 struct btrfs_path *path;
4004 struct btrfs_inode_ref *ref;
4005 struct btrfs_key key[2];
4011 path = btrfs_alloc_path();
4014 inode = new_inode(root->fs_info->sb);
4016 return ERR_PTR(-ENOMEM);
4019 ret = btrfs_set_inode_index(dir, index);
4022 return ERR_PTR(ret);
4026 * index_cnt is ignored for everything but a dir,
4027 * btrfs_get_inode_index_count has an explanation for the magic
4030 init_btrfs_i(inode);
4031 BTRFS_I(inode)->index_cnt = 2;
4032 BTRFS_I(inode)->root = root;
4033 BTRFS_I(inode)->generation = trans->transid;
4034 btrfs_set_inode_space_info(root, inode);
4040 BTRFS_I(inode)->block_group =
4041 btrfs_find_block_group(root, 0, alloc_hint, owner);
4043 key[0].objectid = objectid;
4044 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4047 key[1].objectid = objectid;
4048 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4049 key[1].offset = ref_objectid;
4051 sizes[0] = sizeof(struct btrfs_inode_item);
4052 sizes[1] = name_len + sizeof(*ref);
4054 path->leave_spinning = 1;
4055 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4059 inode->i_uid = current_fsuid();
4061 if (dir && (dir->i_mode & S_ISGID)) {
4062 inode->i_gid = dir->i_gid;
4066 inode->i_gid = current_fsgid();
4068 inode->i_mode = mode;
4069 inode->i_ino = objectid;
4070 inode_set_bytes(inode, 0);
4071 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4072 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4073 struct btrfs_inode_item);
4074 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4076 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4077 struct btrfs_inode_ref);
4078 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4079 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4080 ptr = (unsigned long)(ref + 1);
4081 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4083 btrfs_mark_buffer_dirty(path->nodes[0]);
4084 btrfs_free_path(path);
4086 location = &BTRFS_I(inode)->location;
4087 location->objectid = objectid;
4088 location->offset = 0;
4089 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4091 btrfs_inherit_iflags(inode, dir);
4093 if ((mode & S_IFREG)) {
4094 if (btrfs_test_opt(root, NODATASUM))
4095 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4096 if (btrfs_test_opt(root, NODATACOW))
4097 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4100 insert_inode_hash(inode);
4101 inode_tree_add(inode);
4105 BTRFS_I(dir)->index_cnt--;
4106 btrfs_free_path(path);
4108 return ERR_PTR(ret);
4111 static inline u8 btrfs_inode_type(struct inode *inode)
4113 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4117 * utility function to add 'inode' into 'parent_inode' with
4118 * a give name and a given sequence number.
4119 * if 'add_backref' is true, also insert a backref from the
4120 * inode to the parent directory.
4122 int btrfs_add_link(struct btrfs_trans_handle *trans,
4123 struct inode *parent_inode, struct inode *inode,
4124 const char *name, int name_len, int add_backref, u64 index)
4127 struct btrfs_key key;
4128 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4130 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4131 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4133 key.objectid = inode->i_ino;
4134 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4138 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4139 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4140 key.objectid, root->root_key.objectid,
4141 parent_inode->i_ino,
4142 index, name, name_len);
4143 } else if (add_backref) {
4144 ret = btrfs_insert_inode_ref(trans, root,
4145 name, name_len, inode->i_ino,
4146 parent_inode->i_ino, index);
4150 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4151 parent_inode->i_ino, &key,
4152 btrfs_inode_type(inode), index);
4155 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4157 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4158 ret = btrfs_update_inode(trans, root, parent_inode);
4163 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4164 struct dentry *dentry, struct inode *inode,
4165 int backref, u64 index)
4167 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4168 inode, dentry->d_name.name,
4169 dentry->d_name.len, backref, index);
4171 d_instantiate(dentry, inode);
4179 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4180 int mode, dev_t rdev)
4182 struct btrfs_trans_handle *trans;
4183 struct btrfs_root *root = BTRFS_I(dir)->root;
4184 struct inode *inode = NULL;
4188 unsigned long nr = 0;
4191 if (!new_valid_dev(rdev))
4195 * 2 for inode item and ref
4197 * 1 for xattr if selinux is on
4199 err = btrfs_reserve_metadata_space(root, 5);
4203 trans = btrfs_start_transaction(root, 1);
4206 btrfs_set_trans_block_group(trans, dir);
4208 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4214 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4216 dentry->d_parent->d_inode->i_ino, objectid,
4217 BTRFS_I(dir)->block_group, mode, &index);
4218 err = PTR_ERR(inode);
4222 err = btrfs_init_inode_security(inode, dir);
4228 btrfs_set_trans_block_group(trans, inode);
4229 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4233 inode->i_op = &btrfs_special_inode_operations;
4234 init_special_inode(inode, inode->i_mode, rdev);
4235 btrfs_update_inode(trans, root, inode);
4237 btrfs_update_inode_block_group(trans, inode);
4238 btrfs_update_inode_block_group(trans, dir);
4240 nr = trans->blocks_used;
4241 btrfs_end_transaction_throttle(trans, root);
4243 btrfs_unreserve_metadata_space(root, 5);
4245 inode_dec_link_count(inode);
4248 btrfs_btree_balance_dirty(root, nr);
4252 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4253 int mode, struct nameidata *nd)
4255 struct btrfs_trans_handle *trans;
4256 struct btrfs_root *root = BTRFS_I(dir)->root;
4257 struct inode *inode = NULL;
4260 unsigned long nr = 0;
4265 * 2 for inode item and ref
4267 * 1 for xattr if selinux is on
4269 err = btrfs_reserve_metadata_space(root, 5);
4273 trans = btrfs_start_transaction(root, 1);
4276 btrfs_set_trans_block_group(trans, dir);
4278 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4284 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4286 dentry->d_parent->d_inode->i_ino,
4287 objectid, BTRFS_I(dir)->block_group, mode,
4289 err = PTR_ERR(inode);
4293 err = btrfs_init_inode_security(inode, dir);
4299 btrfs_set_trans_block_group(trans, inode);
4300 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4304 inode->i_mapping->a_ops = &btrfs_aops;
4305 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4306 inode->i_fop = &btrfs_file_operations;
4307 inode->i_op = &btrfs_file_inode_operations;
4308 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4310 btrfs_update_inode_block_group(trans, inode);
4311 btrfs_update_inode_block_group(trans, dir);
4313 nr = trans->blocks_used;
4314 btrfs_end_transaction_throttle(trans, root);
4316 btrfs_unreserve_metadata_space(root, 5);
4318 inode_dec_link_count(inode);
4321 btrfs_btree_balance_dirty(root, nr);
4325 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4326 struct dentry *dentry)
4328 struct btrfs_trans_handle *trans;
4329 struct btrfs_root *root = BTRFS_I(dir)->root;
4330 struct inode *inode = old_dentry->d_inode;
4332 unsigned long nr = 0;
4336 if (inode->i_nlink == 0)
4340 * 1 item for inode ref
4341 * 2 items for dir items
4343 err = btrfs_reserve_metadata_space(root, 3);
4347 btrfs_inc_nlink(inode);
4349 err = btrfs_set_inode_index(dir, &index);
4353 trans = btrfs_start_transaction(root, 1);
4355 btrfs_set_trans_block_group(trans, dir);
4356 atomic_inc(&inode->i_count);
4358 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4363 btrfs_update_inode_block_group(trans, dir);
4364 err = btrfs_update_inode(trans, root, inode);
4366 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4369 nr = trans->blocks_used;
4370 btrfs_end_transaction_throttle(trans, root);
4372 btrfs_unreserve_metadata_space(root, 3);
4374 inode_dec_link_count(inode);
4377 btrfs_btree_balance_dirty(root, nr);
4381 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4383 struct inode *inode = NULL;
4384 struct btrfs_trans_handle *trans;
4385 struct btrfs_root *root = BTRFS_I(dir)->root;
4387 int drop_on_err = 0;
4390 unsigned long nr = 1;
4393 * 2 items for inode and ref
4394 * 2 items for dir items
4395 * 1 for xattr if selinux is on
4397 err = btrfs_reserve_metadata_space(root, 5);
4401 trans = btrfs_start_transaction(root, 1);
4406 btrfs_set_trans_block_group(trans, dir);
4408 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4414 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4416 dentry->d_parent->d_inode->i_ino, objectid,
4417 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4419 if (IS_ERR(inode)) {
4420 err = PTR_ERR(inode);
4426 err = btrfs_init_inode_security(inode, dir);
4430 inode->i_op = &btrfs_dir_inode_operations;
4431 inode->i_fop = &btrfs_dir_file_operations;
4432 btrfs_set_trans_block_group(trans, inode);
4434 btrfs_i_size_write(inode, 0);
4435 err = btrfs_update_inode(trans, root, inode);
4439 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4440 inode, dentry->d_name.name,
4441 dentry->d_name.len, 0, index);
4445 d_instantiate(dentry, inode);
4447 btrfs_update_inode_block_group(trans, inode);
4448 btrfs_update_inode_block_group(trans, dir);
4451 nr = trans->blocks_used;
4452 btrfs_end_transaction_throttle(trans, root);
4455 btrfs_unreserve_metadata_space(root, 5);
4458 btrfs_btree_balance_dirty(root, nr);
4462 /* helper for btfs_get_extent. Given an existing extent in the tree,
4463 * and an extent that you want to insert, deal with overlap and insert
4464 * the new extent into the tree.
4466 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4467 struct extent_map *existing,
4468 struct extent_map *em,
4469 u64 map_start, u64 map_len)
4473 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4474 start_diff = map_start - em->start;
4475 em->start = map_start;
4477 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4478 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4479 em->block_start += start_diff;
4480 em->block_len -= start_diff;
4482 return add_extent_mapping(em_tree, em);
4485 static noinline int uncompress_inline(struct btrfs_path *path,
4486 struct inode *inode, struct page *page,
4487 size_t pg_offset, u64 extent_offset,
4488 struct btrfs_file_extent_item *item)
4491 struct extent_buffer *leaf = path->nodes[0];
4494 unsigned long inline_size;
4497 WARN_ON(pg_offset != 0);
4498 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4499 inline_size = btrfs_file_extent_inline_item_len(leaf,
4500 btrfs_item_nr(leaf, path->slots[0]));
4501 tmp = kmalloc(inline_size, GFP_NOFS);
4502 ptr = btrfs_file_extent_inline_start(item);
4504 read_extent_buffer(leaf, tmp, ptr, inline_size);
4506 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4507 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4508 inline_size, max_size);
4510 char *kaddr = kmap_atomic(page, KM_USER0);
4511 unsigned long copy_size = min_t(u64,
4512 PAGE_CACHE_SIZE - pg_offset,
4513 max_size - extent_offset);
4514 memset(kaddr + pg_offset, 0, copy_size);
4515 kunmap_atomic(kaddr, KM_USER0);
4522 * a bit scary, this does extent mapping from logical file offset to the disk.
4523 * the ugly parts come from merging extents from the disk with the in-ram
4524 * representation. This gets more complex because of the data=ordered code,
4525 * where the in-ram extents might be locked pending data=ordered completion.
4527 * This also copies inline extents directly into the page.
4530 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4531 size_t pg_offset, u64 start, u64 len,
4537 u64 extent_start = 0;
4539 u64 objectid = inode->i_ino;
4541 struct btrfs_path *path = NULL;
4542 struct btrfs_root *root = BTRFS_I(inode)->root;
4543 struct btrfs_file_extent_item *item;
4544 struct extent_buffer *leaf;
4545 struct btrfs_key found_key;
4546 struct extent_map *em = NULL;
4547 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4548 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4549 struct btrfs_trans_handle *trans = NULL;
4553 read_lock(&em_tree->lock);
4554 em = lookup_extent_mapping(em_tree, start, len);
4556 em->bdev = root->fs_info->fs_devices->latest_bdev;
4557 read_unlock(&em_tree->lock);
4560 if (em->start > start || em->start + em->len <= start)
4561 free_extent_map(em);
4562 else if (em->block_start == EXTENT_MAP_INLINE && page)
4563 free_extent_map(em);
4567 em = alloc_extent_map(GFP_NOFS);
4572 em->bdev = root->fs_info->fs_devices->latest_bdev;
4573 em->start = EXTENT_MAP_HOLE;
4574 em->orig_start = EXTENT_MAP_HOLE;
4576 em->block_len = (u64)-1;
4579 path = btrfs_alloc_path();
4583 ret = btrfs_lookup_file_extent(trans, root, path,
4584 objectid, start, trans != NULL);
4591 if (path->slots[0] == 0)
4596 leaf = path->nodes[0];
4597 item = btrfs_item_ptr(leaf, path->slots[0],
4598 struct btrfs_file_extent_item);
4599 /* are we inside the extent that was found? */
4600 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4601 found_type = btrfs_key_type(&found_key);
4602 if (found_key.objectid != objectid ||
4603 found_type != BTRFS_EXTENT_DATA_KEY) {
4607 found_type = btrfs_file_extent_type(leaf, item);
4608 extent_start = found_key.offset;
4609 compressed = btrfs_file_extent_compression(leaf, item);
4610 if (found_type == BTRFS_FILE_EXTENT_REG ||
4611 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4612 extent_end = extent_start +
4613 btrfs_file_extent_num_bytes(leaf, item);
4614 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4616 size = btrfs_file_extent_inline_len(leaf, item);
4617 extent_end = (extent_start + size + root->sectorsize - 1) &
4618 ~((u64)root->sectorsize - 1);
4621 if (start >= extent_end) {
4623 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4624 ret = btrfs_next_leaf(root, path);
4631 leaf = path->nodes[0];
4633 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4634 if (found_key.objectid != objectid ||
4635 found_key.type != BTRFS_EXTENT_DATA_KEY)
4637 if (start + len <= found_key.offset)
4640 em->len = found_key.offset - start;
4644 if (found_type == BTRFS_FILE_EXTENT_REG ||
4645 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4646 em->start = extent_start;
4647 em->len = extent_end - extent_start;
4648 em->orig_start = extent_start -
4649 btrfs_file_extent_offset(leaf, item);
4650 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4652 em->block_start = EXTENT_MAP_HOLE;
4656 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4657 em->block_start = bytenr;
4658 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4661 bytenr += btrfs_file_extent_offset(leaf, item);
4662 em->block_start = bytenr;
4663 em->block_len = em->len;
4664 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4665 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4668 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4672 size_t extent_offset;
4675 em->block_start = EXTENT_MAP_INLINE;
4676 if (!page || create) {
4677 em->start = extent_start;
4678 em->len = extent_end - extent_start;
4682 size = btrfs_file_extent_inline_len(leaf, item);
4683 extent_offset = page_offset(page) + pg_offset - extent_start;
4684 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4685 size - extent_offset);
4686 em->start = extent_start + extent_offset;
4687 em->len = (copy_size + root->sectorsize - 1) &
4688 ~((u64)root->sectorsize - 1);
4689 em->orig_start = EXTENT_MAP_INLINE;
4691 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4692 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4693 if (create == 0 && !PageUptodate(page)) {
4694 if (btrfs_file_extent_compression(leaf, item) ==
4695 BTRFS_COMPRESS_ZLIB) {
4696 ret = uncompress_inline(path, inode, page,
4698 extent_offset, item);
4702 read_extent_buffer(leaf, map + pg_offset, ptr,
4704 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4705 memset(map + pg_offset + copy_size, 0,
4706 PAGE_CACHE_SIZE - pg_offset -
4711 flush_dcache_page(page);
4712 } else if (create && PageUptodate(page)) {
4715 free_extent_map(em);
4717 btrfs_release_path(root, path);
4718 trans = btrfs_join_transaction(root, 1);
4722 write_extent_buffer(leaf, map + pg_offset, ptr,
4725 btrfs_mark_buffer_dirty(leaf);
4727 set_extent_uptodate(io_tree, em->start,
4728 extent_map_end(em) - 1, GFP_NOFS);
4731 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4738 em->block_start = EXTENT_MAP_HOLE;
4739 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4741 btrfs_release_path(root, path);
4742 if (em->start > start || extent_map_end(em) <= start) {
4743 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4744 "[%llu %llu]\n", (unsigned long long)em->start,
4745 (unsigned long long)em->len,
4746 (unsigned long long)start,
4747 (unsigned long long)len);
4753 write_lock(&em_tree->lock);
4754 ret = add_extent_mapping(em_tree, em);
4755 /* it is possible that someone inserted the extent into the tree
4756 * while we had the lock dropped. It is also possible that
4757 * an overlapping map exists in the tree
4759 if (ret == -EEXIST) {
4760 struct extent_map *existing;
4764 existing = lookup_extent_mapping(em_tree, start, len);
4765 if (existing && (existing->start > start ||
4766 existing->start + existing->len <= start)) {
4767 free_extent_map(existing);
4771 existing = lookup_extent_mapping(em_tree, em->start,
4774 err = merge_extent_mapping(em_tree, existing,
4777 free_extent_map(existing);
4779 free_extent_map(em);
4784 free_extent_map(em);
4788 free_extent_map(em);
4793 write_unlock(&em_tree->lock);
4796 btrfs_free_path(path);
4798 ret = btrfs_end_transaction(trans, root);
4803 free_extent_map(em);
4804 return ERR_PTR(err);
4809 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4810 const struct iovec *iov, loff_t offset,
4811 unsigned long nr_segs)
4816 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4817 __u64 start, __u64 len)
4819 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4822 int btrfs_readpage(struct file *file, struct page *page)
4824 struct extent_io_tree *tree;
4825 tree = &BTRFS_I(page->mapping->host)->io_tree;
4826 return extent_read_full_page(tree, page, btrfs_get_extent);
4829 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4831 struct extent_io_tree *tree;
4834 if (current->flags & PF_MEMALLOC) {
4835 redirty_page_for_writepage(wbc, page);
4839 tree = &BTRFS_I(page->mapping->host)->io_tree;
4840 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4843 int btrfs_writepages(struct address_space *mapping,
4844 struct writeback_control *wbc)
4846 struct extent_io_tree *tree;
4848 tree = &BTRFS_I(mapping->host)->io_tree;
4849 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4853 btrfs_readpages(struct file *file, struct address_space *mapping,
4854 struct list_head *pages, unsigned nr_pages)
4856 struct extent_io_tree *tree;
4857 tree = &BTRFS_I(mapping->host)->io_tree;
4858 return extent_readpages(tree, mapping, pages, nr_pages,
4861 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4863 struct extent_io_tree *tree;
4864 struct extent_map_tree *map;
4867 tree = &BTRFS_I(page->mapping->host)->io_tree;
4868 map = &BTRFS_I(page->mapping->host)->extent_tree;
4869 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4871 ClearPagePrivate(page);
4872 set_page_private(page, 0);
4873 page_cache_release(page);
4878 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4880 if (PageWriteback(page) || PageDirty(page))
4882 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4885 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4887 struct extent_io_tree *tree;
4888 struct btrfs_ordered_extent *ordered;
4889 u64 page_start = page_offset(page);
4890 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4894 * we have the page locked, so new writeback can't start,
4895 * and the dirty bit won't be cleared while we are here.
4897 * Wait for IO on this page so that we can safely clear
4898 * the PagePrivate2 bit and do ordered accounting
4900 wait_on_page_writeback(page);
4902 tree = &BTRFS_I(page->mapping->host)->io_tree;
4904 btrfs_releasepage(page, GFP_NOFS);
4907 lock_extent(tree, page_start, page_end, GFP_NOFS);
4908 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4912 * IO on this page will never be started, so we need
4913 * to account for any ordered extents now
4915 clear_extent_bit(tree, page_start, page_end,
4916 EXTENT_DIRTY | EXTENT_DELALLOC |
4917 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
4920 * whoever cleared the private bit is responsible
4921 * for the finish_ordered_io
4923 if (TestClearPagePrivate2(page)) {
4924 btrfs_finish_ordered_io(page->mapping->host,
4925 page_start, page_end);
4927 btrfs_put_ordered_extent(ordered);
4928 lock_extent(tree, page_start, page_end, GFP_NOFS);
4930 clear_extent_bit(tree, page_start, page_end,
4931 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4932 EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
4933 __btrfs_releasepage(page, GFP_NOFS);
4935 ClearPageChecked(page);
4936 if (PagePrivate(page)) {
4937 ClearPagePrivate(page);
4938 set_page_private(page, 0);
4939 page_cache_release(page);
4944 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4945 * called from a page fault handler when a page is first dirtied. Hence we must
4946 * be careful to check for EOF conditions here. We set the page up correctly
4947 * for a written page which means we get ENOSPC checking when writing into
4948 * holes and correct delalloc and unwritten extent mapping on filesystems that
4949 * support these features.
4951 * We are not allowed to take the i_mutex here so we have to play games to
4952 * protect against truncate races as the page could now be beyond EOF. Because
4953 * vmtruncate() writes the inode size before removing pages, once we have the
4954 * page lock we can determine safely if the page is beyond EOF. If it is not
4955 * beyond EOF, then the page is guaranteed safe against truncation until we
4958 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4960 struct page *page = vmf->page;
4961 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4962 struct btrfs_root *root = BTRFS_I(inode)->root;
4963 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4964 struct btrfs_ordered_extent *ordered;
4966 unsigned long zero_start;
4972 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4976 else /* -ENOSPC, -EIO, etc */
4977 ret = VM_FAULT_SIGBUS;
4981 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
4983 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4984 ret = VM_FAULT_SIGBUS;
4988 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4991 size = i_size_read(inode);
4992 page_start = page_offset(page);
4993 page_end = page_start + PAGE_CACHE_SIZE - 1;
4995 if ((page->mapping != inode->i_mapping) ||
4996 (page_start >= size)) {
4997 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4998 /* page got truncated out from underneath us */
5001 wait_on_page_writeback(page);
5003 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5004 set_page_extent_mapped(page);
5007 * we can't set the delalloc bits if there are pending ordered
5008 * extents. Drop our locks and wait for them to finish
5010 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5012 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5014 btrfs_start_ordered_extent(inode, ordered, 1);
5015 btrfs_put_ordered_extent(ordered);
5020 * XXX - page_mkwrite gets called every time the page is dirtied, even
5021 * if it was already dirty, so for space accounting reasons we need to
5022 * clear any delalloc bits for the range we are fixing to save. There
5023 * is probably a better way to do this, but for now keep consistent with
5024 * prepare_pages in the normal write path.
5026 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
5027 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5030 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
5032 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5033 ret = VM_FAULT_SIGBUS;
5034 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5039 /* page is wholly or partially inside EOF */
5040 if (page_start + PAGE_CACHE_SIZE > size)
5041 zero_start = size & ~PAGE_CACHE_MASK;
5043 zero_start = PAGE_CACHE_SIZE;
5045 if (zero_start != PAGE_CACHE_SIZE) {
5047 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5048 flush_dcache_page(page);
5051 ClearPageChecked(page);
5052 set_page_dirty(page);
5053 SetPageUptodate(page);
5055 BTRFS_I(inode)->last_trans = root->fs_info->generation;
5056 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5058 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5061 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
5063 return VM_FAULT_LOCKED;
5069 static void btrfs_truncate(struct inode *inode)
5071 struct btrfs_root *root = BTRFS_I(inode)->root;
5073 struct btrfs_trans_handle *trans;
5075 u64 mask = root->sectorsize - 1;
5077 if (!S_ISREG(inode->i_mode))
5079 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
5082 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5085 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5087 trans = btrfs_start_transaction(root, 1);
5090 * setattr is responsible for setting the ordered_data_close flag,
5091 * but that is only tested during the last file release. That
5092 * could happen well after the next commit, leaving a great big
5093 * window where new writes may get lost if someone chooses to write
5094 * to this file after truncating to zero
5096 * The inode doesn't have any dirty data here, and so if we commit
5097 * this is a noop. If someone immediately starts writing to the inode
5098 * it is very likely we'll catch some of their writes in this
5099 * transaction, and the commit will find this file on the ordered
5100 * data list with good things to send down.
5102 * This is a best effort solution, there is still a window where
5103 * using truncate to replace the contents of the file will
5104 * end up with a zero length file after a crash.
5106 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5107 btrfs_add_ordered_operation(trans, root, inode);
5109 btrfs_set_trans_block_group(trans, inode);
5110 btrfs_i_size_write(inode, inode->i_size);
5112 ret = btrfs_orphan_add(trans, inode);
5115 /* FIXME, add redo link to tree so we don't leak on crash */
5116 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
5117 BTRFS_EXTENT_DATA_KEY);
5118 btrfs_update_inode(trans, root, inode);
5120 ret = btrfs_orphan_del(trans, inode);
5124 nr = trans->blocks_used;
5125 ret = btrfs_end_transaction_throttle(trans, root);
5127 btrfs_btree_balance_dirty(root, nr);
5131 * create a new subvolume directory/inode (helper for the ioctl).
5133 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5134 struct btrfs_root *new_root,
5135 u64 new_dirid, u64 alloc_hint)
5137 struct inode *inode;
5141 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5142 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5144 return PTR_ERR(inode);
5145 inode->i_op = &btrfs_dir_inode_operations;
5146 inode->i_fop = &btrfs_dir_file_operations;
5149 btrfs_i_size_write(inode, 0);
5151 err = btrfs_update_inode(trans, new_root, inode);
5158 /* helper function for file defrag and space balancing. This
5159 * forces readahead on a given range of bytes in an inode
5161 unsigned long btrfs_force_ra(struct address_space *mapping,
5162 struct file_ra_state *ra, struct file *file,
5163 pgoff_t offset, pgoff_t last_index)
5165 pgoff_t req_size = last_index - offset + 1;
5167 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5168 return offset + req_size;
5171 struct inode *btrfs_alloc_inode(struct super_block *sb)
5173 struct btrfs_inode *ei;
5175 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5179 ei->last_sub_trans = 0;
5180 ei->logged_trans = 0;
5181 ei->outstanding_extents = 0;
5182 ei->reserved_extents = 0;
5184 spin_lock_init(&ei->accounting_lock);
5185 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5186 INIT_LIST_HEAD(&ei->i_orphan);
5187 INIT_LIST_HEAD(&ei->ordered_operations);
5188 return &ei->vfs_inode;
5191 void btrfs_destroy_inode(struct inode *inode)
5193 struct btrfs_ordered_extent *ordered;
5194 struct btrfs_root *root = BTRFS_I(inode)->root;
5196 WARN_ON(!list_empty(&inode->i_dentry));
5197 WARN_ON(inode->i_data.nrpages);
5200 * This can happen where we create an inode, but somebody else also
5201 * created the same inode and we need to destroy the one we already
5208 * Make sure we're properly removed from the ordered operation
5212 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5213 spin_lock(&root->fs_info->ordered_extent_lock);
5214 list_del_init(&BTRFS_I(inode)->ordered_operations);
5215 spin_unlock(&root->fs_info->ordered_extent_lock);
5218 spin_lock(&root->list_lock);
5219 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5220 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
5221 " list\n", inode->i_ino);
5224 spin_unlock(&root->list_lock);
5227 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5231 printk(KERN_ERR "btrfs found ordered "
5232 "extent %llu %llu on inode cleanup\n",
5233 (unsigned long long)ordered->file_offset,
5234 (unsigned long long)ordered->len);
5235 btrfs_remove_ordered_extent(inode, ordered);
5236 btrfs_put_ordered_extent(ordered);
5237 btrfs_put_ordered_extent(ordered);
5240 inode_tree_del(inode);
5241 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5243 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5246 void btrfs_drop_inode(struct inode *inode)
5248 struct btrfs_root *root = BTRFS_I(inode)->root;
5250 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5251 generic_delete_inode(inode);
5253 generic_drop_inode(inode);
5256 static void init_once(void *foo)
5258 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5260 inode_init_once(&ei->vfs_inode);
5263 void btrfs_destroy_cachep(void)
5265 if (btrfs_inode_cachep)
5266 kmem_cache_destroy(btrfs_inode_cachep);
5267 if (btrfs_trans_handle_cachep)
5268 kmem_cache_destroy(btrfs_trans_handle_cachep);
5269 if (btrfs_transaction_cachep)
5270 kmem_cache_destroy(btrfs_transaction_cachep);
5271 if (btrfs_path_cachep)
5272 kmem_cache_destroy(btrfs_path_cachep);
5275 int btrfs_init_cachep(void)
5277 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5278 sizeof(struct btrfs_inode), 0,
5279 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5280 if (!btrfs_inode_cachep)
5283 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5284 sizeof(struct btrfs_trans_handle), 0,
5285 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5286 if (!btrfs_trans_handle_cachep)
5289 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5290 sizeof(struct btrfs_transaction), 0,
5291 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5292 if (!btrfs_transaction_cachep)
5295 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5296 sizeof(struct btrfs_path), 0,
5297 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5298 if (!btrfs_path_cachep)
5303 btrfs_destroy_cachep();
5307 static int btrfs_getattr(struct vfsmount *mnt,
5308 struct dentry *dentry, struct kstat *stat)
5310 struct inode *inode = dentry->d_inode;
5311 generic_fillattr(inode, stat);
5312 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5313 stat->blksize = PAGE_CACHE_SIZE;
5314 stat->blocks = (inode_get_bytes(inode) +
5315 BTRFS_I(inode)->delalloc_bytes) >> 9;
5319 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5320 struct inode *new_dir, struct dentry *new_dentry)
5322 struct btrfs_trans_handle *trans;
5323 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5324 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5325 struct inode *new_inode = new_dentry->d_inode;
5326 struct inode *old_inode = old_dentry->d_inode;
5327 struct timespec ctime = CURRENT_TIME;
5332 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5335 /* we only allow rename subvolume link between subvolumes */
5336 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5339 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5340 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5343 if (S_ISDIR(old_inode->i_mode) && new_inode &&
5344 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5348 * We want to reserve the absolute worst case amount of items. So if
5349 * both inodes are subvols and we need to unlink them then that would
5350 * require 4 item modifications, but if they are both normal inodes it
5351 * would require 5 item modifications, so we'll assume their normal
5352 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5353 * should cover the worst case number of items we'll modify.
5355 ret = btrfs_reserve_metadata_space(root, 11);
5360 * we're using rename to replace one file with another.
5361 * and the replacement file is large. Start IO on it now so
5362 * we don't add too much work to the end of the transaction
5364 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5365 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5366 filemap_flush(old_inode->i_mapping);
5368 /* close the racy window with snapshot create/destroy ioctl */
5369 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5370 down_read(&root->fs_info->subvol_sem);
5372 trans = btrfs_start_transaction(root, 1);
5373 btrfs_set_trans_block_group(trans, new_dir);
5376 btrfs_record_root_in_trans(trans, dest);
5378 ret = btrfs_set_inode_index(new_dir, &index);
5382 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5383 /* force full log commit if subvolume involved. */
5384 root->fs_info->last_trans_log_full_commit = trans->transid;
5386 ret = btrfs_insert_inode_ref(trans, dest,
5387 new_dentry->d_name.name,
5388 new_dentry->d_name.len,
5390 new_dir->i_ino, index);
5394 * this is an ugly little race, but the rename is required
5395 * to make sure that if we crash, the inode is either at the
5396 * old name or the new one. pinning the log transaction lets
5397 * us make sure we don't allow a log commit to come in after
5398 * we unlink the name but before we add the new name back in.
5400 btrfs_pin_log_trans(root);
5403 * make sure the inode gets flushed if it is replacing
5406 if (new_inode && new_inode->i_size &&
5407 old_inode && S_ISREG(old_inode->i_mode)) {
5408 btrfs_add_ordered_operation(trans, root, old_inode);
5411 old_dir->i_ctime = old_dir->i_mtime = ctime;
5412 new_dir->i_ctime = new_dir->i_mtime = ctime;
5413 old_inode->i_ctime = ctime;
5415 if (old_dentry->d_parent != new_dentry->d_parent)
5416 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5418 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5419 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5420 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5421 old_dentry->d_name.name,
5422 old_dentry->d_name.len);
5424 btrfs_inc_nlink(old_dentry->d_inode);
5425 ret = btrfs_unlink_inode(trans, root, old_dir,
5426 old_dentry->d_inode,
5427 old_dentry->d_name.name,
5428 old_dentry->d_name.len);
5433 new_inode->i_ctime = CURRENT_TIME;
5434 if (unlikely(new_inode->i_ino ==
5435 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5436 root_objectid = BTRFS_I(new_inode)->location.objectid;
5437 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5439 new_dentry->d_name.name,
5440 new_dentry->d_name.len);
5441 BUG_ON(new_inode->i_nlink == 0);
5443 ret = btrfs_unlink_inode(trans, dest, new_dir,
5444 new_dentry->d_inode,
5445 new_dentry->d_name.name,
5446 new_dentry->d_name.len);
5449 if (new_inode->i_nlink == 0) {
5450 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5455 ret = btrfs_add_link(trans, new_dir, old_inode,
5456 new_dentry->d_name.name,
5457 new_dentry->d_name.len, 0, index);
5460 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5461 btrfs_log_new_name(trans, old_inode, old_dir,
5462 new_dentry->d_parent);
5463 btrfs_end_log_trans(root);
5466 btrfs_end_transaction_throttle(trans, root);
5468 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5469 up_read(&root->fs_info->subvol_sem);
5471 btrfs_unreserve_metadata_space(root, 11);
5476 * some fairly slow code that needs optimization. This walks the list
5477 * of all the inodes with pending delalloc and forces them to disk.
5479 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
5481 struct list_head *head = &root->fs_info->delalloc_inodes;
5482 struct btrfs_inode *binode;
5483 struct inode *inode;
5485 if (root->fs_info->sb->s_flags & MS_RDONLY)
5488 spin_lock(&root->fs_info->delalloc_lock);
5489 while (!list_empty(head)) {
5490 binode = list_entry(head->next, struct btrfs_inode,
5492 inode = igrab(&binode->vfs_inode);
5494 list_del_init(&binode->delalloc_inodes);
5495 spin_unlock(&root->fs_info->delalloc_lock);
5497 filemap_flush(inode->i_mapping);
5501 spin_lock(&root->fs_info->delalloc_lock);
5503 spin_unlock(&root->fs_info->delalloc_lock);
5505 /* the filemap_flush will queue IO into the worker threads, but
5506 * we have to make sure the IO is actually started and that
5507 * ordered extents get created before we return
5509 atomic_inc(&root->fs_info->async_submit_draining);
5510 while (atomic_read(&root->fs_info->nr_async_submits) ||
5511 atomic_read(&root->fs_info->async_delalloc_pages)) {
5512 wait_event(root->fs_info->async_submit_wait,
5513 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5514 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5516 atomic_dec(&root->fs_info->async_submit_draining);
5520 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5521 const char *symname)
5523 struct btrfs_trans_handle *trans;
5524 struct btrfs_root *root = BTRFS_I(dir)->root;
5525 struct btrfs_path *path;
5526 struct btrfs_key key;
5527 struct inode *inode = NULL;
5535 struct btrfs_file_extent_item *ei;
5536 struct extent_buffer *leaf;
5537 unsigned long nr = 0;
5539 name_len = strlen(symname) + 1;
5540 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5541 return -ENAMETOOLONG;
5544 * 2 items for inode item and ref
5545 * 2 items for dir items
5546 * 1 item for xattr if selinux is on
5548 err = btrfs_reserve_metadata_space(root, 5);
5552 trans = btrfs_start_transaction(root, 1);
5555 btrfs_set_trans_block_group(trans, dir);
5557 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5563 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5565 dentry->d_parent->d_inode->i_ino, objectid,
5566 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5568 err = PTR_ERR(inode);
5572 err = btrfs_init_inode_security(inode, dir);
5578 btrfs_set_trans_block_group(trans, inode);
5579 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5583 inode->i_mapping->a_ops = &btrfs_aops;
5584 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5585 inode->i_fop = &btrfs_file_operations;
5586 inode->i_op = &btrfs_file_inode_operations;
5587 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5589 btrfs_update_inode_block_group(trans, inode);
5590 btrfs_update_inode_block_group(trans, dir);
5594 path = btrfs_alloc_path();
5596 key.objectid = inode->i_ino;
5598 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5599 datasize = btrfs_file_extent_calc_inline_size(name_len);
5600 err = btrfs_insert_empty_item(trans, root, path, &key,
5606 leaf = path->nodes[0];
5607 ei = btrfs_item_ptr(leaf, path->slots[0],
5608 struct btrfs_file_extent_item);
5609 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5610 btrfs_set_file_extent_type(leaf, ei,
5611 BTRFS_FILE_EXTENT_INLINE);
5612 btrfs_set_file_extent_encryption(leaf, ei, 0);
5613 btrfs_set_file_extent_compression(leaf, ei, 0);
5614 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5615 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5617 ptr = btrfs_file_extent_inline_start(ei);
5618 write_extent_buffer(leaf, symname, ptr, name_len);
5619 btrfs_mark_buffer_dirty(leaf);
5620 btrfs_free_path(path);
5622 inode->i_op = &btrfs_symlink_inode_operations;
5623 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5624 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5625 inode_set_bytes(inode, name_len);
5626 btrfs_i_size_write(inode, name_len - 1);
5627 err = btrfs_update_inode(trans, root, inode);
5632 nr = trans->blocks_used;
5633 btrfs_end_transaction_throttle(trans, root);
5635 btrfs_unreserve_metadata_space(root, 5);
5637 inode_dec_link_count(inode);
5640 btrfs_btree_balance_dirty(root, nr);
5644 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5645 struct inode *inode, u64 start, u64 end,
5646 u64 locked_end, u64 alloc_hint, int mode)
5648 struct btrfs_root *root = BTRFS_I(inode)->root;
5649 struct btrfs_key ins;
5651 u64 cur_offset = start;
5652 u64 num_bytes = end - start;
5655 while (num_bytes > 0) {
5656 alloc_size = min(num_bytes, root->fs_info->max_extent);
5658 ret = btrfs_reserve_metadata_space(root, 1);
5662 ret = btrfs_reserve_extent(trans, root, alloc_size,
5663 root->sectorsize, 0, alloc_hint,
5669 ret = insert_reserved_file_extent(trans, inode,
5670 cur_offset, ins.objectid,
5671 ins.offset, ins.offset,
5672 ins.offset, locked_end,
5674 BTRFS_FILE_EXTENT_PREALLOC);
5676 btrfs_drop_extent_cache(inode, cur_offset,
5677 cur_offset + ins.offset -1, 0);
5678 num_bytes -= ins.offset;
5679 cur_offset += ins.offset;
5680 alloc_hint = ins.objectid + ins.offset;
5681 btrfs_unreserve_metadata_space(root, 1);
5684 if (cur_offset > start) {
5685 inode->i_ctime = CURRENT_TIME;
5686 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5687 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5688 cur_offset > i_size_read(inode))
5689 btrfs_i_size_write(inode, cur_offset);
5690 ret = btrfs_update_inode(trans, root, inode);
5697 static long btrfs_fallocate(struct inode *inode, int mode,
5698 loff_t offset, loff_t len)
5706 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5707 struct extent_map *em;
5708 struct btrfs_trans_handle *trans;
5709 struct btrfs_root *root;
5712 alloc_start = offset & ~mask;
5713 alloc_end = (offset + len + mask) & ~mask;
5716 * wait for ordered IO before we have any locks. We'll loop again
5717 * below with the locks held.
5719 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5721 mutex_lock(&inode->i_mutex);
5722 if (alloc_start > inode->i_size) {
5723 ret = btrfs_cont_expand(inode, alloc_start);
5728 root = BTRFS_I(inode)->root;
5730 ret = btrfs_check_data_free_space(root, inode,
5731 alloc_end - alloc_start);
5735 locked_end = alloc_end - 1;
5737 struct btrfs_ordered_extent *ordered;
5739 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5745 /* the extent lock is ordered inside the running
5748 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5750 ordered = btrfs_lookup_first_ordered_extent(inode,
5753 ordered->file_offset + ordered->len > alloc_start &&
5754 ordered->file_offset < alloc_end) {
5755 btrfs_put_ordered_extent(ordered);
5756 unlock_extent(&BTRFS_I(inode)->io_tree,
5757 alloc_start, locked_end, GFP_NOFS);
5758 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5761 * we can't wait on the range with the transaction
5762 * running or with the extent lock held
5764 btrfs_wait_ordered_range(inode, alloc_start,
5765 alloc_end - alloc_start);
5768 btrfs_put_ordered_extent(ordered);
5773 cur_offset = alloc_start;
5775 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5776 alloc_end - cur_offset, 0);
5777 BUG_ON(IS_ERR(em) || !em);
5778 last_byte = min(extent_map_end(em), alloc_end);
5779 last_byte = (last_byte + mask) & ~mask;
5780 if (em->block_start == EXTENT_MAP_HOLE) {
5781 ret = prealloc_file_range(trans, inode, cur_offset,
5782 last_byte, locked_end + 1,
5785 free_extent_map(em);
5789 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5790 alloc_hint = em->block_start;
5791 free_extent_map(em);
5793 cur_offset = last_byte;
5794 if (cur_offset >= alloc_end) {
5799 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5802 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5804 btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
5806 mutex_unlock(&inode->i_mutex);
5810 static int btrfs_set_page_dirty(struct page *page)
5812 return __set_page_dirty_nobuffers(page);
5815 static int btrfs_permission(struct inode *inode, int mask)
5817 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5819 return generic_permission(inode, mask, btrfs_check_acl);
5822 static const struct inode_operations btrfs_dir_inode_operations = {
5823 .getattr = btrfs_getattr,
5824 .lookup = btrfs_lookup,
5825 .create = btrfs_create,
5826 .unlink = btrfs_unlink,
5828 .mkdir = btrfs_mkdir,
5829 .rmdir = btrfs_rmdir,
5830 .rename = btrfs_rename,
5831 .symlink = btrfs_symlink,
5832 .setattr = btrfs_setattr,
5833 .mknod = btrfs_mknod,
5834 .setxattr = btrfs_setxattr,
5835 .getxattr = btrfs_getxattr,
5836 .listxattr = btrfs_listxattr,
5837 .removexattr = btrfs_removexattr,
5838 .permission = btrfs_permission,
5840 static const struct inode_operations btrfs_dir_ro_inode_operations = {
5841 .lookup = btrfs_lookup,
5842 .permission = btrfs_permission,
5845 static const struct file_operations btrfs_dir_file_operations = {
5846 .llseek = generic_file_llseek,
5847 .read = generic_read_dir,
5848 .readdir = btrfs_real_readdir,
5849 .unlocked_ioctl = btrfs_ioctl,
5850 #ifdef CONFIG_COMPAT
5851 .compat_ioctl = btrfs_ioctl,
5853 .release = btrfs_release_file,
5854 .fsync = btrfs_sync_file,
5857 static struct extent_io_ops btrfs_extent_io_ops = {
5858 .fill_delalloc = run_delalloc_range,
5859 .submit_bio_hook = btrfs_submit_bio_hook,
5860 .merge_bio_hook = btrfs_merge_bio_hook,
5861 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5862 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5863 .writepage_start_hook = btrfs_writepage_start_hook,
5864 .readpage_io_failed_hook = btrfs_io_failed_hook,
5865 .set_bit_hook = btrfs_set_bit_hook,
5866 .clear_bit_hook = btrfs_clear_bit_hook,
5867 .merge_extent_hook = btrfs_merge_extent_hook,
5868 .split_extent_hook = btrfs_split_extent_hook,
5872 * btrfs doesn't support the bmap operation because swapfiles
5873 * use bmap to make a mapping of extents in the file. They assume
5874 * these extents won't change over the life of the file and they
5875 * use the bmap result to do IO directly to the drive.
5877 * the btrfs bmap call would return logical addresses that aren't
5878 * suitable for IO and they also will change frequently as COW
5879 * operations happen. So, swapfile + btrfs == corruption.
5881 * For now we're avoiding this by dropping bmap.
5883 static const struct address_space_operations btrfs_aops = {
5884 .readpage = btrfs_readpage,
5885 .writepage = btrfs_writepage,
5886 .writepages = btrfs_writepages,
5887 .readpages = btrfs_readpages,
5888 .sync_page = block_sync_page,
5889 .direct_IO = btrfs_direct_IO,
5890 .invalidatepage = btrfs_invalidatepage,
5891 .releasepage = btrfs_releasepage,
5892 .set_page_dirty = btrfs_set_page_dirty,
5893 .error_remove_page = generic_error_remove_page,
5896 static const struct address_space_operations btrfs_symlink_aops = {
5897 .readpage = btrfs_readpage,
5898 .writepage = btrfs_writepage,
5899 .invalidatepage = btrfs_invalidatepage,
5900 .releasepage = btrfs_releasepage,
5903 static const struct inode_operations btrfs_file_inode_operations = {
5904 .truncate = btrfs_truncate,
5905 .getattr = btrfs_getattr,
5906 .setattr = btrfs_setattr,
5907 .setxattr = btrfs_setxattr,
5908 .getxattr = btrfs_getxattr,
5909 .listxattr = btrfs_listxattr,
5910 .removexattr = btrfs_removexattr,
5911 .permission = btrfs_permission,
5912 .fallocate = btrfs_fallocate,
5913 .fiemap = btrfs_fiemap,
5915 static const struct inode_operations btrfs_special_inode_operations = {
5916 .getattr = btrfs_getattr,
5917 .setattr = btrfs_setattr,
5918 .permission = btrfs_permission,
5919 .setxattr = btrfs_setxattr,
5920 .getxattr = btrfs_getxattr,
5921 .listxattr = btrfs_listxattr,
5922 .removexattr = btrfs_removexattr,
5924 static const struct inode_operations btrfs_symlink_inode_operations = {
5925 .readlink = generic_readlink,
5926 .follow_link = page_follow_link_light,
5927 .put_link = page_put_link,
5928 .permission = btrfs_permission,
5929 .setxattr = btrfs_setxattr,
5930 .getxattr = btrfs_getxattr,
5931 .listxattr = btrfs_listxattr,
5932 .removexattr = btrfs_removexattr,
5935 const struct dentry_operations btrfs_dentry_operations = {
5936 .d_delete = btrfs_dentry_delete,