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/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
40 #include <linux/falloc.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
52 #include "ref-cache.h"
53 #include "compression.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
95 * a very lame attempt at stopping writes when the FS is 85% full. There
96 * are countless ways this is incorrect, but it is better than nothing.
98 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
107 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
108 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
109 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
117 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
119 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
131 struct page **compressed_pages)
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
144 unsigned long offset;
145 int use_compress = 0;
147 if (compressed_size && compressed_pages) {
149 cur_size = compressed_size;
152 path = btrfs_alloc_path(); if (!path)
155 btrfs_set_trans_block_group(trans, inode);
157 key.objectid = inode->i_ino;
159 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
160 inode_add_bytes(inode, size);
161 datasize = btrfs_file_extent_calc_inline_size(cur_size);
163 inode_add_bytes(inode, size);
164 ret = btrfs_insert_empty_item(trans, root, path, &key,
169 printk("got bad ret %d\n", ret);
172 leaf = path->nodes[0];
173 ei = btrfs_item_ptr(leaf, path->slots[0],
174 struct btrfs_file_extent_item);
175 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
176 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
177 btrfs_set_file_extent_encryption(leaf, ei, 0);
178 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
179 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
180 ptr = btrfs_file_extent_inline_start(ei);
185 while(compressed_size > 0) {
186 cpage = compressed_pages[i];
187 cur_size = min(compressed_size,
191 write_extent_buffer(leaf, kaddr, ptr, cur_size);
196 compressed_size -= cur_size;
198 btrfs_set_file_extent_compression(leaf, ei,
199 BTRFS_COMPRESS_ZLIB);
201 page = find_get_page(inode->i_mapping,
202 start >> PAGE_CACHE_SHIFT);
203 btrfs_set_file_extent_compression(leaf, ei, 0);
204 kaddr = kmap_atomic(page, KM_USER0);
205 offset = start & (PAGE_CACHE_SIZE - 1);
206 write_extent_buffer(leaf, kaddr + offset, ptr, size);
207 kunmap_atomic(kaddr, KM_USER0);
208 page_cache_release(page);
210 btrfs_mark_buffer_dirty(leaf);
211 btrfs_free_path(path);
213 BTRFS_I(inode)->disk_i_size = inode->i_size;
214 btrfs_update_inode(trans, root, inode);
217 btrfs_free_path(path);
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
227 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
228 struct btrfs_root *root,
229 struct inode *inode, u64 start, u64 end,
230 size_t compressed_size,
231 struct page **compressed_pages)
233 u64 isize = i_size_read(inode);
234 u64 actual_end = min(end + 1, isize);
235 u64 inline_len = actual_end - start;
236 u64 aligned_end = (end + root->sectorsize - 1) &
237 ~((u64)root->sectorsize - 1);
239 u64 data_len = inline_len;
243 data_len = compressed_size;
246 actual_end >= PAGE_CACHE_SIZE ||
247 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
249 (actual_end & (root->sectorsize - 1)) == 0) ||
251 data_len > root->fs_info->max_inline) {
255 ret = btrfs_drop_extents(trans, root, inode, start,
256 aligned_end, start, &hint_byte);
259 if (isize > actual_end)
260 inline_len = min_t(u64, isize, actual_end);
261 ret = insert_inline_extent(trans, root, inode, start,
262 inline_len, compressed_size,
265 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
269 struct async_extent {
274 unsigned long nr_pages;
275 struct list_head list;
280 struct btrfs_root *root;
281 struct page *locked_page;
284 struct list_head extents;
285 struct btrfs_work work;
288 static noinline int add_async_extent(struct async_cow *cow,
289 u64 start, u64 ram_size,
292 unsigned long nr_pages)
294 struct async_extent *async_extent;
296 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
302 list_add_tail(&async_extent->list, &cow->extents);
307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
322 static noinline int compress_file_range(struct inode *inode,
323 struct page *locked_page,
325 struct async_cow *async_cow,
328 struct btrfs_root *root = BTRFS_I(inode)->root;
329 struct btrfs_trans_handle *trans;
333 u64 blocksize = root->sectorsize;
336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
342 unsigned long max_uncompressed = 128 * 1024;
350 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
351 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
353 actual_end = min_t(u64, i_size_read(inode), end + 1);
354 total_compressed = actual_end - start;
356 /* we want to make sure that amount of ram required to uncompress
357 * an extent is reasonable, so we limit the total size in ram
358 * of a compressed extent to 128k. This is a crucial number
359 * because it also controls how easily we can spread reads across
360 * cpus for decompression.
362 * We also want to make sure the amount of IO required to do
363 * a random read is reasonably small, so we limit the size of
364 * a compressed extent to 128k.
366 total_compressed = min(total_compressed, max_uncompressed);
367 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
368 num_bytes = max(blocksize, num_bytes);
369 disk_num_bytes = num_bytes;
374 * we do compression for mount -o compress and when the
375 * inode has not been flagged as nocompress. This flag can
376 * change at any time if we discover bad compression ratios.
378 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
379 btrfs_test_opt(root, COMPRESS)) {
381 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
383 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
384 total_compressed, pages,
385 nr_pages, &nr_pages_ret,
391 unsigned long offset = total_compressed &
392 (PAGE_CACHE_SIZE - 1);
393 struct page *page = pages[nr_pages_ret - 1];
396 /* zero the tail end of the last page, we might be
397 * sending it down to disk
400 kaddr = kmap_atomic(page, KM_USER0);
401 memset(kaddr + offset, 0,
402 PAGE_CACHE_SIZE - offset);
403 kunmap_atomic(kaddr, KM_USER0);
409 trans = btrfs_join_transaction(root, 1);
411 btrfs_set_trans_block_group(trans, inode);
413 /* lets try to make an inline extent */
414 if (ret || total_in < (actual_end - start)) {
415 /* we didn't compress the entire range, try
416 * to make an uncompressed inline extent.
418 ret = cow_file_range_inline(trans, root, inode,
419 start, end, 0, NULL);
421 /* try making a compressed inline extent */
422 ret = cow_file_range_inline(trans, root, inode,
424 total_compressed, pages);
426 btrfs_end_transaction(trans, root);
429 * inline extent creation worked, we don't need
430 * to create any more async work items. Unlock
431 * and free up our temp pages.
433 extent_clear_unlock_delalloc(inode,
434 &BTRFS_I(inode)->io_tree,
435 start, end, NULL, 1, 0,
444 * we aren't doing an inline extent round the compressed size
445 * up to a block size boundary so the allocator does sane
448 total_compressed = (total_compressed + blocksize - 1) &
452 * one last check to make sure the compression is really a
453 * win, compare the page count read with the blocks on disk
455 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
456 ~(PAGE_CACHE_SIZE - 1);
457 if (total_compressed >= total_in) {
460 disk_num_bytes = total_compressed;
461 num_bytes = total_in;
464 if (!will_compress && pages) {
466 * the compression code ran but failed to make things smaller,
467 * free any pages it allocated and our page pointer array
469 for (i = 0; i < nr_pages_ret; i++) {
470 WARN_ON(pages[i]->mapping);
471 page_cache_release(pages[i]);
475 total_compressed = 0;
478 /* flag the file so we don't compress in the future */
479 btrfs_set_flag(inode, NOCOMPRESS);
484 /* the async work queues will take care of doing actual
485 * allocation on disk for these compressed pages,
486 * and will submit them to the elevator.
488 add_async_extent(async_cow, start, num_bytes,
489 total_compressed, pages, nr_pages_ret);
491 if (start + num_bytes < end) {
499 * No compression, but we still need to write the pages in
500 * the file we've been given so far. redirty the locked
501 * page if it corresponds to our extent and set things up
502 * for the async work queue to run cow_file_range to do
503 * the normal delalloc dance
505 if (page_offset(locked_page) >= start &&
506 page_offset(locked_page) <= end) {
507 __set_page_dirty_nobuffers(locked_page);
508 /* unlocked later on in the async handlers */
510 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
518 for (i = 0; i < nr_pages_ret; i++) {
519 WARN_ON(pages[i]->mapping);
520 page_cache_release(pages[i]);
529 * phase two of compressed writeback. This is the ordered portion
530 * of the code, which only gets called in the order the work was
531 * queued. We walk all the async extents created by compress_file_range
532 * and send them down to the disk.
534 static noinline int submit_compressed_extents(struct inode *inode,
535 struct async_cow *async_cow)
537 struct async_extent *async_extent;
539 struct btrfs_trans_handle *trans;
540 struct btrfs_key ins;
541 struct extent_map *em;
542 struct btrfs_root *root = BTRFS_I(inode)->root;
543 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
544 struct extent_io_tree *io_tree;
547 if (list_empty(&async_cow->extents))
550 trans = btrfs_join_transaction(root, 1);
552 while(!list_empty(&async_cow->extents)) {
553 async_extent = list_entry(async_cow->extents.next,
554 struct async_extent, list);
555 list_del(&async_extent->list);
557 io_tree = &BTRFS_I(inode)->io_tree;
559 /* did the compression code fall back to uncompressed IO? */
560 if (!async_extent->pages) {
561 int page_started = 0;
562 unsigned long nr_written = 0;
564 lock_extent(io_tree, async_extent->start,
565 async_extent->start + async_extent->ram_size - 1,
568 /* allocate blocks */
569 cow_file_range(inode, async_cow->locked_page,
571 async_extent->start +
572 async_extent->ram_size - 1,
573 &page_started, &nr_written, 0);
576 * if page_started, cow_file_range inserted an
577 * inline extent and took care of all the unlocking
578 * and IO for us. Otherwise, we need to submit
579 * all those pages down to the drive.
582 extent_write_locked_range(io_tree,
583 inode, async_extent->start,
584 async_extent->start +
585 async_extent->ram_size - 1,
593 lock_extent(io_tree, async_extent->start,
594 async_extent->start + async_extent->ram_size - 1,
597 * here we're doing allocation and writeback of the
600 btrfs_drop_extent_cache(inode, async_extent->start,
601 async_extent->start +
602 async_extent->ram_size - 1, 0);
604 ret = btrfs_reserve_extent(trans, root,
605 async_extent->compressed_size,
606 async_extent->compressed_size,
610 em = alloc_extent_map(GFP_NOFS);
611 em->start = async_extent->start;
612 em->len = async_extent->ram_size;
614 em->block_start = ins.objectid;
615 em->block_len = ins.offset;
616 em->bdev = root->fs_info->fs_devices->latest_bdev;
617 set_bit(EXTENT_FLAG_PINNED, &em->flags);
618 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
621 spin_lock(&em_tree->lock);
622 ret = add_extent_mapping(em_tree, em);
623 spin_unlock(&em_tree->lock);
624 if (ret != -EEXIST) {
628 btrfs_drop_extent_cache(inode, async_extent->start,
629 async_extent->start +
630 async_extent->ram_size - 1, 0);
633 ret = btrfs_add_ordered_extent(inode, async_extent->start,
635 async_extent->ram_size,
637 BTRFS_ORDERED_COMPRESSED);
640 btrfs_end_transaction(trans, root);
643 * clear dirty, set writeback and unlock the pages.
645 extent_clear_unlock_delalloc(inode,
646 &BTRFS_I(inode)->io_tree,
648 async_extent->start +
649 async_extent->ram_size - 1,
650 NULL, 1, 1, 0, 1, 1, 0);
652 ret = btrfs_submit_compressed_write(inode,
654 async_extent->ram_size,
656 ins.offset, async_extent->pages,
657 async_extent->nr_pages);
660 trans = btrfs_join_transaction(root, 1);
661 alloc_hint = ins.objectid + ins.offset;
666 btrfs_end_transaction(trans, root);
671 * when extent_io.c finds a delayed allocation range in the file,
672 * the call backs end up in this code. The basic idea is to
673 * allocate extents on disk for the range, and create ordered data structs
674 * in ram to track those extents.
676 * locked_page is the page that writepage had locked already. We use
677 * it to make sure we don't do extra locks or unlocks.
679 * *page_started is set to one if we unlock locked_page and do everything
680 * required to start IO on it. It may be clean and already done with
683 static noinline int cow_file_range(struct inode *inode,
684 struct page *locked_page,
685 u64 start, u64 end, int *page_started,
686 unsigned long *nr_written,
689 struct btrfs_root *root = BTRFS_I(inode)->root;
690 struct btrfs_trans_handle *trans;
693 unsigned long ram_size;
696 u64 blocksize = root->sectorsize;
698 struct btrfs_key ins;
699 struct extent_map *em;
700 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
703 trans = btrfs_join_transaction(root, 1);
705 btrfs_set_trans_block_group(trans, inode);
707 actual_end = min_t(u64, i_size_read(inode), end + 1);
709 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
710 num_bytes = max(blocksize, num_bytes);
711 disk_num_bytes = num_bytes;
715 /* lets try to make an inline extent */
716 ret = cow_file_range_inline(trans, root, inode,
717 start, end, 0, NULL);
719 extent_clear_unlock_delalloc(inode,
720 &BTRFS_I(inode)->io_tree,
721 start, end, NULL, 1, 1,
723 *nr_written = *nr_written +
724 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
731 BUG_ON(disk_num_bytes >
732 btrfs_super_total_bytes(&root->fs_info->super_copy));
734 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
736 while(disk_num_bytes > 0) {
737 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
738 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
739 root->sectorsize, 0, alloc_hint,
744 em = alloc_extent_map(GFP_NOFS);
747 ram_size = ins.offset;
748 em->len = ins.offset;
750 em->block_start = ins.objectid;
751 em->block_len = ins.offset;
752 em->bdev = root->fs_info->fs_devices->latest_bdev;
753 set_bit(EXTENT_FLAG_PINNED, &em->flags);
756 spin_lock(&em_tree->lock);
757 ret = add_extent_mapping(em_tree, em);
758 spin_unlock(&em_tree->lock);
759 if (ret != -EEXIST) {
763 btrfs_drop_extent_cache(inode, start,
764 start + ram_size - 1, 0);
767 cur_alloc_size = ins.offset;
768 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
769 ram_size, cur_alloc_size, 0);
772 if (disk_num_bytes < cur_alloc_size) {
773 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
777 /* we're not doing compressed IO, don't unlock the first
778 * page (which the caller expects to stay locked), don't
779 * clear any dirty bits and don't set any writeback bits
781 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
782 start, start + ram_size - 1,
783 locked_page, unlock, 1,
785 disk_num_bytes -= cur_alloc_size;
786 num_bytes -= cur_alloc_size;
787 alloc_hint = ins.objectid + ins.offset;
788 start += cur_alloc_size;
792 btrfs_end_transaction(trans, root);
798 * work queue call back to started compression on a file and pages
800 static noinline void async_cow_start(struct btrfs_work *work)
802 struct async_cow *async_cow;
804 async_cow = container_of(work, struct async_cow, work);
806 compress_file_range(async_cow->inode, async_cow->locked_page,
807 async_cow->start, async_cow->end, async_cow,
810 async_cow->inode = NULL;
814 * work queue call back to submit previously compressed pages
816 static noinline void async_cow_submit(struct btrfs_work *work)
818 struct async_cow *async_cow;
819 struct btrfs_root *root;
820 unsigned long nr_pages;
822 async_cow = container_of(work, struct async_cow, work);
824 root = async_cow->root;
825 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
828 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
830 if (atomic_read(&root->fs_info->async_delalloc_pages) <
832 waitqueue_active(&root->fs_info->async_submit_wait))
833 wake_up(&root->fs_info->async_submit_wait);
835 if (async_cow->inode) {
836 submit_compressed_extents(async_cow->inode, async_cow);
840 static noinline void async_cow_free(struct btrfs_work *work)
842 struct async_cow *async_cow;
843 async_cow = container_of(work, struct async_cow, work);
847 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
848 u64 start, u64 end, int *page_started,
849 unsigned long *nr_written)
851 struct async_cow *async_cow;
852 struct btrfs_root *root = BTRFS_I(inode)->root;
853 unsigned long nr_pages;
855 int limit = 10 * 1024 * 1042;
857 if (!btrfs_test_opt(root, COMPRESS)) {
858 return cow_file_range(inode, locked_page, start, end,
859 page_started, nr_written, 1);
862 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
863 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
865 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
866 async_cow->inode = inode;
867 async_cow->root = root;
868 async_cow->locked_page = locked_page;
869 async_cow->start = start;
871 if (btrfs_test_flag(inode, NOCOMPRESS))
874 cur_end = min(end, start + 512 * 1024 - 1);
876 async_cow->end = cur_end;
877 INIT_LIST_HEAD(&async_cow->extents);
879 async_cow->work.func = async_cow_start;
880 async_cow->work.ordered_func = async_cow_submit;
881 async_cow->work.ordered_free = async_cow_free;
882 async_cow->work.flags = 0;
884 while(atomic_read(&root->fs_info->async_submit_draining) &&
885 atomic_read(&root->fs_info->async_delalloc_pages)) {
886 wait_event(root->fs_info->async_submit_wait,
887 (atomic_read(&root->fs_info->async_delalloc_pages)
891 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
893 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
895 btrfs_queue_worker(&root->fs_info->delalloc_workers,
898 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
899 wait_event(root->fs_info->async_submit_wait,
900 (atomic_read(&root->fs_info->async_delalloc_pages) <
904 while(atomic_read(&root->fs_info->async_submit_draining) &&
905 atomic_read(&root->fs_info->async_delalloc_pages)) {
906 wait_event(root->fs_info->async_submit_wait,
907 (atomic_read(&root->fs_info->async_delalloc_pages) ==
911 *nr_written += nr_pages;
919 * when nowcow writeback call back. This checks for snapshots or COW copies
920 * of the extents that exist in the file, and COWs the file as required.
922 * If no cow copies or snapshots exist, we write directly to the existing
925 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
926 u64 start, u64 end, int *page_started, int force,
927 unsigned long *nr_written)
929 struct btrfs_root *root = BTRFS_I(inode)->root;
930 struct btrfs_trans_handle *trans;
931 struct extent_buffer *leaf;
932 struct btrfs_path *path;
933 struct btrfs_file_extent_item *fi;
934 struct btrfs_key found_key;
946 path = btrfs_alloc_path();
948 trans = btrfs_join_transaction(root, 1);
954 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
957 if (ret > 0 && path->slots[0] > 0 && check_prev) {
958 leaf = path->nodes[0];
959 btrfs_item_key_to_cpu(leaf, &found_key,
961 if (found_key.objectid == inode->i_ino &&
962 found_key.type == BTRFS_EXTENT_DATA_KEY)
967 leaf = path->nodes[0];
968 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
969 ret = btrfs_next_leaf(root, path);
974 leaf = path->nodes[0];
979 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
981 if (found_key.objectid > inode->i_ino ||
982 found_key.type > BTRFS_EXTENT_DATA_KEY ||
983 found_key.offset > end)
986 if (found_key.offset > cur_offset) {
987 extent_end = found_key.offset;
991 fi = btrfs_item_ptr(leaf, path->slots[0],
992 struct btrfs_file_extent_item);
993 extent_type = btrfs_file_extent_type(leaf, fi);
995 if (extent_type == BTRFS_FILE_EXTENT_REG ||
996 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
997 struct btrfs_block_group_cache *block_group;
998 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
999 extent_end = found_key.offset +
1000 btrfs_file_extent_num_bytes(leaf, fi);
1001 if (extent_end <= start) {
1005 if (btrfs_file_extent_compression(leaf, fi) ||
1006 btrfs_file_extent_encryption(leaf, fi) ||
1007 btrfs_file_extent_other_encoding(leaf, fi))
1009 if (disk_bytenr == 0)
1011 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1013 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1015 block_group = btrfs_lookup_block_group(root->fs_info,
1017 if (!block_group || block_group->ro)
1019 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1021 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1022 extent_end = found_key.offset +
1023 btrfs_file_extent_inline_len(leaf, fi);
1024 extent_end = ALIGN(extent_end, root->sectorsize);
1029 if (extent_end <= start) {
1034 if (cow_start == (u64)-1)
1035 cow_start = cur_offset;
1036 cur_offset = extent_end;
1037 if (cur_offset > end)
1043 btrfs_release_path(root, path);
1044 if (cow_start != (u64)-1) {
1045 ret = cow_file_range(inode, locked_page, cow_start,
1046 found_key.offset - 1, page_started,
1049 cow_start = (u64)-1;
1052 disk_bytenr += cur_offset - found_key.offset;
1053 num_bytes = min(end + 1, extent_end) - cur_offset;
1054 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1055 struct extent_map *em;
1056 struct extent_map_tree *em_tree;
1057 em_tree = &BTRFS_I(inode)->extent_tree;
1058 em = alloc_extent_map(GFP_NOFS);
1059 em->start = cur_offset;
1060 em->len = num_bytes;
1061 em->block_len = num_bytes;
1062 em->block_start = disk_bytenr;
1063 em->bdev = root->fs_info->fs_devices->latest_bdev;
1064 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1066 spin_lock(&em_tree->lock);
1067 ret = add_extent_mapping(em_tree, em);
1068 spin_unlock(&em_tree->lock);
1069 if (ret != -EEXIST) {
1070 free_extent_map(em);
1073 btrfs_drop_extent_cache(inode, em->start,
1074 em->start + em->len - 1, 0);
1076 type = BTRFS_ORDERED_PREALLOC;
1078 type = BTRFS_ORDERED_NOCOW;
1081 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1082 num_bytes, num_bytes, type);
1085 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1086 cur_offset, cur_offset + num_bytes - 1,
1087 locked_page, 1, 1, 1, 0, 0, 0);
1088 cur_offset = extent_end;
1089 if (cur_offset > end)
1092 btrfs_release_path(root, path);
1094 if (cur_offset <= end && cow_start == (u64)-1)
1095 cow_start = cur_offset;
1096 if (cow_start != (u64)-1) {
1097 ret = cow_file_range(inode, locked_page, cow_start, end,
1098 page_started, nr_written, 1);
1102 ret = btrfs_end_transaction(trans, root);
1104 btrfs_free_path(path);
1109 * extent_io.c call back to do delayed allocation processing
1111 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1112 u64 start, u64 end, int *page_started,
1113 unsigned long *nr_written)
1115 struct btrfs_root *root = BTRFS_I(inode)->root;
1118 if (btrfs_test_opt(root, NODATACOW) ||
1119 btrfs_test_flag(inode, NODATACOW))
1120 ret = run_delalloc_nocow(inode, locked_page, start, end,
1121 page_started, 0, nr_written);
1122 else if (btrfs_test_flag(inode, PREALLOC))
1123 ret = run_delalloc_nocow(inode, locked_page, start, end,
1124 page_started, 1, nr_written);
1126 ret = cow_file_range_async(inode, locked_page, start, end,
1127 page_started, nr_written);
1133 * extent_io.c set_bit_hook, used to track delayed allocation
1134 * bytes in this file, and to maintain the list of inodes that
1135 * have pending delalloc work to be done.
1137 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1138 unsigned long old, unsigned long bits)
1140 unsigned long flags;
1141 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1142 struct btrfs_root *root = BTRFS_I(inode)->root;
1143 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1144 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1145 root->fs_info->delalloc_bytes += end - start + 1;
1146 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1147 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1148 &root->fs_info->delalloc_inodes);
1150 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1156 * extent_io.c clear_bit_hook, see set_bit_hook for why
1158 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1159 unsigned long old, unsigned long bits)
1161 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1162 struct btrfs_root *root = BTRFS_I(inode)->root;
1163 unsigned long flags;
1165 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1166 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1167 printk("warning: delalloc account %Lu %Lu\n",
1168 end - start + 1, root->fs_info->delalloc_bytes);
1169 root->fs_info->delalloc_bytes = 0;
1170 BTRFS_I(inode)->delalloc_bytes = 0;
1172 root->fs_info->delalloc_bytes -= end - start + 1;
1173 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1175 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1176 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1177 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1179 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1185 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1186 * we don't create bios that span stripes or chunks
1188 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1189 size_t size, struct bio *bio,
1190 unsigned long bio_flags)
1192 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1193 struct btrfs_mapping_tree *map_tree;
1194 u64 logical = (u64)bio->bi_sector << 9;
1199 if (bio_flags & EXTENT_BIO_COMPRESSED)
1202 length = bio->bi_size;
1203 map_tree = &root->fs_info->mapping_tree;
1204 map_length = length;
1205 ret = btrfs_map_block(map_tree, READ, logical,
1206 &map_length, NULL, 0);
1208 if (map_length < length + size) {
1215 * in order to insert checksums into the metadata in large chunks,
1216 * we wait until bio submission time. All the pages in the bio are
1217 * checksummed and sums are attached onto the ordered extent record.
1219 * At IO completion time the cums attached on the ordered extent record
1220 * are inserted into the btree
1222 int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
1223 int mirror_num, unsigned long bio_flags)
1225 struct btrfs_root *root = BTRFS_I(inode)->root;
1228 ret = btrfs_csum_one_bio(root, inode, bio);
1234 * in order to insert checksums into the metadata in large chunks,
1235 * we wait until bio submission time. All the pages in the bio are
1236 * checksummed and sums are attached onto the ordered extent record.
1238 * At IO completion time the cums attached on the ordered extent record
1239 * are inserted into the btree
1241 int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1242 int mirror_num, unsigned long bio_flags)
1244 struct btrfs_root *root = BTRFS_I(inode)->root;
1245 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1249 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1250 * or reading the csums from the tree before a read
1252 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1253 int mirror_num, unsigned long bio_flags)
1255 struct btrfs_root *root = BTRFS_I(inode)->root;
1259 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1262 skip_sum = btrfs_test_opt(root, NODATASUM) ||
1263 btrfs_test_flag(inode, NODATASUM);
1265 if (!(rw & (1 << BIO_RW))) {
1267 if (bio_flags & EXTENT_BIO_COMPRESSED)
1268 return btrfs_submit_compressed_read(inode, bio,
1269 mirror_num, bio_flags);
1271 btrfs_lookup_bio_sums(root, inode, bio);
1273 } else if (!skip_sum) {
1274 /* we're doing a write, do the async checksumming */
1275 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1276 inode, rw, bio, mirror_num,
1277 bio_flags, __btrfs_submit_bio_start,
1278 __btrfs_submit_bio_done);
1282 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1286 * given a list of ordered sums record them in the inode. This happens
1287 * at IO completion time based on sums calculated at bio submission time.
1289 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1290 struct inode *inode, u64 file_offset,
1291 struct list_head *list)
1293 struct list_head *cur;
1294 struct btrfs_ordered_sum *sum;
1296 btrfs_set_trans_block_group(trans, inode);
1297 list_for_each(cur, list) {
1298 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1299 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
1305 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1307 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1310 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1314 /* see btrfs_writepage_start_hook for details on why this is required */
1315 struct btrfs_writepage_fixup {
1317 struct btrfs_work work;
1320 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1322 struct btrfs_writepage_fixup *fixup;
1323 struct btrfs_ordered_extent *ordered;
1325 struct inode *inode;
1329 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1333 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1334 ClearPageChecked(page);
1338 inode = page->mapping->host;
1339 page_start = page_offset(page);
1340 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1342 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1344 /* already ordered? We're done */
1345 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1346 EXTENT_ORDERED, 0)) {
1350 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1352 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1353 page_end, GFP_NOFS);
1355 btrfs_start_ordered_extent(inode, ordered, 1);
1359 btrfs_set_extent_delalloc(inode, page_start, page_end);
1360 ClearPageChecked(page);
1362 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1365 page_cache_release(page);
1369 * There are a few paths in the higher layers of the kernel that directly
1370 * set the page dirty bit without asking the filesystem if it is a
1371 * good idea. This causes problems because we want to make sure COW
1372 * properly happens and the data=ordered rules are followed.
1374 * In our case any range that doesn't have the ORDERED bit set
1375 * hasn't been properly setup for IO. We kick off an async process
1376 * to fix it up. The async helper will wait for ordered extents, set
1377 * the delalloc bit and make it safe to write the page.
1379 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1381 struct inode *inode = page->mapping->host;
1382 struct btrfs_writepage_fixup *fixup;
1383 struct btrfs_root *root = BTRFS_I(inode)->root;
1386 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1391 if (PageChecked(page))
1394 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1398 SetPageChecked(page);
1399 page_cache_get(page);
1400 fixup->work.func = btrfs_writepage_fixup_worker;
1402 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1406 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1407 struct inode *inode, u64 file_pos,
1408 u64 disk_bytenr, u64 disk_num_bytes,
1409 u64 num_bytes, u64 ram_bytes,
1410 u8 compression, u8 encryption,
1411 u16 other_encoding, int extent_type)
1413 struct btrfs_root *root = BTRFS_I(inode)->root;
1414 struct btrfs_file_extent_item *fi;
1415 struct btrfs_path *path;
1416 struct extent_buffer *leaf;
1417 struct btrfs_key ins;
1421 path = btrfs_alloc_path();
1424 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1425 file_pos + num_bytes, file_pos, &hint);
1428 ins.objectid = inode->i_ino;
1429 ins.offset = file_pos;
1430 ins.type = BTRFS_EXTENT_DATA_KEY;
1431 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1433 leaf = path->nodes[0];
1434 fi = btrfs_item_ptr(leaf, path->slots[0],
1435 struct btrfs_file_extent_item);
1436 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1437 btrfs_set_file_extent_type(leaf, fi, extent_type);
1438 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1439 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1440 btrfs_set_file_extent_offset(leaf, fi, 0);
1441 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1442 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1443 btrfs_set_file_extent_compression(leaf, fi, compression);
1444 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1445 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1446 btrfs_mark_buffer_dirty(leaf);
1448 inode_add_bytes(inode, num_bytes);
1449 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1451 ins.objectid = disk_bytenr;
1452 ins.offset = disk_num_bytes;
1453 ins.type = BTRFS_EXTENT_ITEM_KEY;
1454 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1455 root->root_key.objectid,
1456 trans->transid, inode->i_ino, &ins);
1459 btrfs_free_path(path);
1463 /* as ordered data IO finishes, this gets called so we can finish
1464 * an ordered extent if the range of bytes in the file it covers are
1467 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1469 struct btrfs_root *root = BTRFS_I(inode)->root;
1470 struct btrfs_trans_handle *trans;
1471 struct btrfs_ordered_extent *ordered_extent;
1472 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1476 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1480 trans = btrfs_join_transaction(root, 1);
1482 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1483 BUG_ON(!ordered_extent);
1484 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1487 lock_extent(io_tree, ordered_extent->file_offset,
1488 ordered_extent->file_offset + ordered_extent->len - 1,
1491 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1493 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1495 ret = btrfs_mark_extent_written(trans, root, inode,
1496 ordered_extent->file_offset,
1497 ordered_extent->file_offset +
1498 ordered_extent->len);
1501 ret = insert_reserved_file_extent(trans, inode,
1502 ordered_extent->file_offset,
1503 ordered_extent->start,
1504 ordered_extent->disk_len,
1505 ordered_extent->len,
1506 ordered_extent->len,
1508 BTRFS_FILE_EXTENT_REG);
1511 unlock_extent(io_tree, ordered_extent->file_offset,
1512 ordered_extent->file_offset + ordered_extent->len - 1,
1515 add_pending_csums(trans, inode, ordered_extent->file_offset,
1516 &ordered_extent->list);
1518 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1519 btrfs_ordered_update_i_size(inode, ordered_extent);
1520 btrfs_update_inode(trans, root, inode);
1521 btrfs_remove_ordered_extent(inode, ordered_extent);
1522 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1525 btrfs_put_ordered_extent(ordered_extent);
1526 /* once for the tree */
1527 btrfs_put_ordered_extent(ordered_extent);
1529 btrfs_end_transaction(trans, root);
1533 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1534 struct extent_state *state, int uptodate)
1536 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1540 * When IO fails, either with EIO or csum verification fails, we
1541 * try other mirrors that might have a good copy of the data. This
1542 * io_failure_record is used to record state as we go through all the
1543 * mirrors. If another mirror has good data, the page is set up to date
1544 * and things continue. If a good mirror can't be found, the original
1545 * bio end_io callback is called to indicate things have failed.
1547 struct io_failure_record {
1555 int btrfs_io_failed_hook(struct bio *failed_bio,
1556 struct page *page, u64 start, u64 end,
1557 struct extent_state *state)
1559 struct io_failure_record *failrec = NULL;
1561 struct extent_map *em;
1562 struct inode *inode = page->mapping->host;
1563 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1564 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1570 unsigned long bio_flags = 0;
1572 ret = get_state_private(failure_tree, start, &private);
1574 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1577 failrec->start = start;
1578 failrec->len = end - start + 1;
1579 failrec->last_mirror = 0;
1581 spin_lock(&em_tree->lock);
1582 em = lookup_extent_mapping(em_tree, start, failrec->len);
1583 if (em->start > start || em->start + em->len < start) {
1584 free_extent_map(em);
1587 spin_unlock(&em_tree->lock);
1589 if (!em || IS_ERR(em)) {
1593 logical = start - em->start;
1594 logical = em->block_start + logical;
1595 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1596 bio_flags = EXTENT_BIO_COMPRESSED;
1597 failrec->logical = logical;
1598 free_extent_map(em);
1599 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1600 EXTENT_DIRTY, GFP_NOFS);
1601 set_state_private(failure_tree, start,
1602 (u64)(unsigned long)failrec);
1604 failrec = (struct io_failure_record *)(unsigned long)private;
1606 num_copies = btrfs_num_copies(
1607 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1608 failrec->logical, failrec->len);
1609 failrec->last_mirror++;
1611 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1612 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1615 if (state && state->start != failrec->start)
1617 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1619 if (!state || failrec->last_mirror > num_copies) {
1620 set_state_private(failure_tree, failrec->start, 0);
1621 clear_extent_bits(failure_tree, failrec->start,
1622 failrec->start + failrec->len - 1,
1623 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1627 bio = bio_alloc(GFP_NOFS, 1);
1628 bio->bi_private = state;
1629 bio->bi_end_io = failed_bio->bi_end_io;
1630 bio->bi_sector = failrec->logical >> 9;
1631 bio->bi_bdev = failed_bio->bi_bdev;
1633 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1634 if (failed_bio->bi_rw & (1 << BIO_RW))
1639 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1640 failrec->last_mirror,
1646 * each time an IO finishes, we do a fast check in the IO failure tree
1647 * to see if we need to process or clean up an io_failure_record
1649 int btrfs_clean_io_failures(struct inode *inode, u64 start)
1652 u64 private_failure;
1653 struct io_failure_record *failure;
1657 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1658 (u64)-1, 1, EXTENT_DIRTY)) {
1659 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1660 start, &private_failure);
1662 failure = (struct io_failure_record *)(unsigned long)
1664 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1666 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1668 failure->start + failure->len - 1,
1669 EXTENT_DIRTY | EXTENT_LOCKED,
1678 * when reads are done, we need to check csums to verify the data is correct
1679 * if there's a match, we allow the bio to finish. If not, we go through
1680 * the io_failure_record routines to find good copies
1682 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1683 struct extent_state *state)
1685 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1686 struct inode *inode = page->mapping->host;
1687 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1689 u64 private = ~(u32)0;
1691 struct btrfs_root *root = BTRFS_I(inode)->root;
1693 unsigned long flags;
1695 if (btrfs_test_opt(root, NODATASUM) ||
1696 btrfs_test_flag(inode, NODATASUM))
1698 if (state && state->start == start) {
1699 private = state->private;
1702 ret = get_state_private(io_tree, start, &private);
1704 local_irq_save(flags);
1705 kaddr = kmap_atomic(page, KM_IRQ0);
1709 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1710 btrfs_csum_final(csum, (char *)&csum);
1711 if (csum != private) {
1714 kunmap_atomic(kaddr, KM_IRQ0);
1715 local_irq_restore(flags);
1717 /* if the io failure tree for this inode is non-empty,
1718 * check to see if we've recovered from a failed IO
1720 btrfs_clean_io_failures(inode, start);
1724 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1725 page->mapping->host->i_ino, (unsigned long long)start, csum,
1727 memset(kaddr + offset, 1, end - start + 1);
1728 flush_dcache_page(page);
1729 kunmap_atomic(kaddr, KM_IRQ0);
1730 local_irq_restore(flags);
1737 * This creates an orphan entry for the given inode in case something goes
1738 * wrong in the middle of an unlink/truncate.
1740 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1742 struct btrfs_root *root = BTRFS_I(inode)->root;
1745 spin_lock(&root->list_lock);
1747 /* already on the orphan list, we're good */
1748 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1749 spin_unlock(&root->list_lock);
1753 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1755 spin_unlock(&root->list_lock);
1758 * insert an orphan item to track this unlinked/truncated file
1760 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1766 * We have done the truncate/delete so we can go ahead and remove the orphan
1767 * item for this particular inode.
1769 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1771 struct btrfs_root *root = BTRFS_I(inode)->root;
1774 spin_lock(&root->list_lock);
1776 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1777 spin_unlock(&root->list_lock);
1781 list_del_init(&BTRFS_I(inode)->i_orphan);
1783 spin_unlock(&root->list_lock);
1787 spin_unlock(&root->list_lock);
1789 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1795 * this cleans up any orphans that may be left on the list from the last use
1798 void btrfs_orphan_cleanup(struct btrfs_root *root)
1800 struct btrfs_path *path;
1801 struct extent_buffer *leaf;
1802 struct btrfs_item *item;
1803 struct btrfs_key key, found_key;
1804 struct btrfs_trans_handle *trans;
1805 struct inode *inode;
1806 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1808 /* don't do orphan cleanup if the fs is readonly. */
1809 if (root->fs_info->sb->s_flags & MS_RDONLY)
1812 path = btrfs_alloc_path();
1817 key.objectid = BTRFS_ORPHAN_OBJECTID;
1818 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1819 key.offset = (u64)-1;
1823 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1825 printk(KERN_ERR "Error searching slot for orphan: %d"
1831 * if ret == 0 means we found what we were searching for, which
1832 * is weird, but possible, so only screw with path if we didnt
1833 * find the key and see if we have stuff that matches
1836 if (path->slots[0] == 0)
1841 /* pull out the item */
1842 leaf = path->nodes[0];
1843 item = btrfs_item_nr(leaf, path->slots[0]);
1844 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1846 /* make sure the item matches what we want */
1847 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1849 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1852 /* release the path since we're done with it */
1853 btrfs_release_path(root, path);
1856 * this is where we are basically btrfs_lookup, without the
1857 * crossing root thing. we store the inode number in the
1858 * offset of the orphan item.
1860 inode = btrfs_iget_locked(root->fs_info->sb,
1861 found_key.offset, root);
1865 if (inode->i_state & I_NEW) {
1866 BTRFS_I(inode)->root = root;
1868 /* have to set the location manually */
1869 BTRFS_I(inode)->location.objectid = inode->i_ino;
1870 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1871 BTRFS_I(inode)->location.offset = 0;
1873 btrfs_read_locked_inode(inode);
1874 unlock_new_inode(inode);
1878 * add this inode to the orphan list so btrfs_orphan_del does
1879 * the proper thing when we hit it
1881 spin_lock(&root->list_lock);
1882 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1883 spin_unlock(&root->list_lock);
1886 * if this is a bad inode, means we actually succeeded in
1887 * removing the inode, but not the orphan record, which means
1888 * we need to manually delete the orphan since iput will just
1889 * do a destroy_inode
1891 if (is_bad_inode(inode)) {
1892 trans = btrfs_start_transaction(root, 1);
1893 btrfs_orphan_del(trans, inode);
1894 btrfs_end_transaction(trans, root);
1899 /* if we have links, this was a truncate, lets do that */
1900 if (inode->i_nlink) {
1902 btrfs_truncate(inode);
1907 /* this will do delete_inode and everything for us */
1912 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1914 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1916 btrfs_free_path(path);
1920 * read an inode from the btree into the in-memory inode
1922 void btrfs_read_locked_inode(struct inode *inode)
1924 struct btrfs_path *path;
1925 struct extent_buffer *leaf;
1926 struct btrfs_inode_item *inode_item;
1927 struct btrfs_timespec *tspec;
1928 struct btrfs_root *root = BTRFS_I(inode)->root;
1929 struct btrfs_key location;
1930 u64 alloc_group_block;
1934 path = btrfs_alloc_path();
1936 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1938 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1942 leaf = path->nodes[0];
1943 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1944 struct btrfs_inode_item);
1946 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1947 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1948 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1949 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1950 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1952 tspec = btrfs_inode_atime(inode_item);
1953 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1954 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1956 tspec = btrfs_inode_mtime(inode_item);
1957 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1958 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1960 tspec = btrfs_inode_ctime(inode_item);
1961 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1962 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1964 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1965 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1966 inode->i_generation = BTRFS_I(inode)->generation;
1968 rdev = btrfs_inode_rdev(leaf, inode_item);
1970 BTRFS_I(inode)->index_cnt = (u64)-1;
1972 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1973 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1975 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1976 if (!BTRFS_I(inode)->block_group) {
1977 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1979 BTRFS_BLOCK_GROUP_METADATA, 0);
1981 btrfs_free_path(path);
1984 switch (inode->i_mode & S_IFMT) {
1986 inode->i_mapping->a_ops = &btrfs_aops;
1987 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1988 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1989 inode->i_fop = &btrfs_file_operations;
1990 inode->i_op = &btrfs_file_inode_operations;
1993 inode->i_fop = &btrfs_dir_file_operations;
1994 if (root == root->fs_info->tree_root)
1995 inode->i_op = &btrfs_dir_ro_inode_operations;
1997 inode->i_op = &btrfs_dir_inode_operations;
2000 inode->i_op = &btrfs_symlink_inode_operations;
2001 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2002 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2005 init_special_inode(inode, inode->i_mode, rdev);
2011 btrfs_free_path(path);
2012 make_bad_inode(inode);
2016 * given a leaf and an inode, copy the inode fields into the leaf
2018 static void fill_inode_item(struct btrfs_trans_handle *trans,
2019 struct extent_buffer *leaf,
2020 struct btrfs_inode_item *item,
2021 struct inode *inode)
2023 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2024 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2025 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2026 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2027 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2029 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2030 inode->i_atime.tv_sec);
2031 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2032 inode->i_atime.tv_nsec);
2034 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2035 inode->i_mtime.tv_sec);
2036 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2037 inode->i_mtime.tv_nsec);
2039 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2040 inode->i_ctime.tv_sec);
2041 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2042 inode->i_ctime.tv_nsec);
2044 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2045 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2046 btrfs_set_inode_transid(leaf, item, trans->transid);
2047 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2048 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2049 btrfs_set_inode_block_group(leaf, item,
2050 BTRFS_I(inode)->block_group->key.objectid);
2054 * copy everything in the in-memory inode into the btree.
2056 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
2057 struct btrfs_root *root,
2058 struct inode *inode)
2060 struct btrfs_inode_item *inode_item;
2061 struct btrfs_path *path;
2062 struct extent_buffer *leaf;
2065 path = btrfs_alloc_path();
2067 ret = btrfs_lookup_inode(trans, root, path,
2068 &BTRFS_I(inode)->location, 1);
2075 leaf = path->nodes[0];
2076 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2077 struct btrfs_inode_item);
2079 fill_inode_item(trans, leaf, inode_item, inode);
2080 btrfs_mark_buffer_dirty(leaf);
2081 btrfs_set_inode_last_trans(trans, inode);
2084 btrfs_free_path(path);
2090 * unlink helper that gets used here in inode.c and in the tree logging
2091 * recovery code. It remove a link in a directory with a given name, and
2092 * also drops the back refs in the inode to the directory
2094 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2095 struct btrfs_root *root,
2096 struct inode *dir, struct inode *inode,
2097 const char *name, int name_len)
2099 struct btrfs_path *path;
2101 struct extent_buffer *leaf;
2102 struct btrfs_dir_item *di;
2103 struct btrfs_key key;
2106 path = btrfs_alloc_path();
2112 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2113 name, name_len, -1);
2122 leaf = path->nodes[0];
2123 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2124 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2127 btrfs_release_path(root, path);
2129 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2131 dir->i_ino, &index);
2133 printk("failed to delete reference to %.*s, "
2134 "inode %lu parent %lu\n", name_len, name,
2135 inode->i_ino, dir->i_ino);
2139 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2140 index, name, name_len, -1);
2149 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2150 btrfs_release_path(root, path);
2152 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2154 BUG_ON(ret != 0 && ret != -ENOENT);
2156 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2158 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2162 btrfs_free_path(path);
2166 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2167 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2168 btrfs_update_inode(trans, root, dir);
2169 btrfs_drop_nlink(inode);
2170 ret = btrfs_update_inode(trans, root, inode);
2171 dir->i_sb->s_dirt = 1;
2176 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2178 struct btrfs_root *root;
2179 struct btrfs_trans_handle *trans;
2180 struct inode *inode = dentry->d_inode;
2182 unsigned long nr = 0;
2184 root = BTRFS_I(dir)->root;
2186 ret = btrfs_check_free_space(root, 1, 1);
2190 trans = btrfs_start_transaction(root, 1);
2192 btrfs_set_trans_block_group(trans, dir);
2193 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2194 dentry->d_name.name, dentry->d_name.len);
2196 if (inode->i_nlink == 0)
2197 ret = btrfs_orphan_add(trans, inode);
2199 nr = trans->blocks_used;
2201 btrfs_end_transaction_throttle(trans, root);
2203 btrfs_btree_balance_dirty(root, nr);
2207 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2209 struct inode *inode = dentry->d_inode;
2212 struct btrfs_root *root = BTRFS_I(dir)->root;
2213 struct btrfs_trans_handle *trans;
2214 unsigned long nr = 0;
2216 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2220 ret = btrfs_check_free_space(root, 1, 1);
2224 trans = btrfs_start_transaction(root, 1);
2225 btrfs_set_trans_block_group(trans, dir);
2227 err = btrfs_orphan_add(trans, inode);
2231 /* now the directory is empty */
2232 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2233 dentry->d_name.name, dentry->d_name.len);
2235 btrfs_i_size_write(inode, 0);
2239 nr = trans->blocks_used;
2240 ret = btrfs_end_transaction_throttle(trans, root);
2242 btrfs_btree_balance_dirty(root, nr);
2250 * when truncating bytes in a file, it is possible to avoid reading
2251 * the leaves that contain only checksum items. This can be the
2252 * majority of the IO required to delete a large file, but it must
2253 * be done carefully.
2255 * The keys in the level just above the leaves are checked to make sure
2256 * the lowest key in a given leaf is a csum key, and starts at an offset
2257 * after the new size.
2259 * Then the key for the next leaf is checked to make sure it also has
2260 * a checksum item for the same file. If it does, we know our target leaf
2261 * contains only checksum items, and it can be safely freed without reading
2264 * This is just an optimization targeted at large files. It may do
2265 * nothing. It will return 0 unless things went badly.
2267 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2268 struct btrfs_root *root,
2269 struct btrfs_path *path,
2270 struct inode *inode, u64 new_size)
2272 struct btrfs_key key;
2275 struct btrfs_key found_key;
2276 struct btrfs_key other_key;
2277 struct btrfs_leaf_ref *ref;
2281 path->lowest_level = 1;
2282 key.objectid = inode->i_ino;
2283 key.type = BTRFS_CSUM_ITEM_KEY;
2284 key.offset = new_size;
2286 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2290 if (path->nodes[1] == NULL) {
2295 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2296 nritems = btrfs_header_nritems(path->nodes[1]);
2301 if (path->slots[1] >= nritems)
2304 /* did we find a key greater than anything we want to delete? */
2305 if (found_key.objectid > inode->i_ino ||
2306 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2309 /* we check the next key in the node to make sure the leave contains
2310 * only checksum items. This comparison doesn't work if our
2311 * leaf is the last one in the node
2313 if (path->slots[1] + 1 >= nritems) {
2315 /* search forward from the last key in the node, this
2316 * will bring us into the next node in the tree
2318 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2320 /* unlikely, but we inc below, so check to be safe */
2321 if (found_key.offset == (u64)-1)
2324 /* search_forward needs a path with locks held, do the
2325 * search again for the original key. It is possible
2326 * this will race with a balance and return a path that
2327 * we could modify, but this drop is just an optimization
2328 * and is allowed to miss some leaves.
2330 btrfs_release_path(root, path);
2333 /* setup a max key for search_forward */
2334 other_key.offset = (u64)-1;
2335 other_key.type = key.type;
2336 other_key.objectid = key.objectid;
2338 path->keep_locks = 1;
2339 ret = btrfs_search_forward(root, &found_key, &other_key,
2341 path->keep_locks = 0;
2342 if (ret || found_key.objectid != key.objectid ||
2343 found_key.type != key.type) {
2348 key.offset = found_key.offset;
2349 btrfs_release_path(root, path);
2354 /* we know there's one more slot after us in the tree,
2355 * read that key so we can verify it is also a checksum item
2357 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2359 if (found_key.objectid < inode->i_ino)
2362 if (found_key.type != key.type || found_key.offset < new_size)
2366 * if the key for the next leaf isn't a csum key from this objectid,
2367 * we can't be sure there aren't good items inside this leaf.
2370 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2373 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2374 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2376 * it is safe to delete this leaf, it contains only
2377 * csum items from this inode at an offset >= new_size
2379 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2382 if (root->ref_cows && leaf_gen < trans->transid) {
2383 ref = btrfs_alloc_leaf_ref(root, 0);
2385 ref->root_gen = root->root_key.offset;
2386 ref->bytenr = leaf_start;
2388 ref->generation = leaf_gen;
2391 ret = btrfs_add_leaf_ref(root, ref, 0);
2393 btrfs_free_leaf_ref(root, ref);
2399 btrfs_release_path(root, path);
2401 if (other_key.objectid == inode->i_ino &&
2402 other_key.type == key.type && other_key.offset > key.offset) {
2403 key.offset = other_key.offset;
2409 /* fixup any changes we've made to the path */
2410 path->lowest_level = 0;
2411 path->keep_locks = 0;
2412 btrfs_release_path(root, path);
2417 * this can truncate away extent items, csum items and directory items.
2418 * It starts at a high offset and removes keys until it can't find
2419 * any higher than new_size
2421 * csum items that cross the new i_size are truncated to the new size
2424 * min_type is the minimum key type to truncate down to. If set to 0, this
2425 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2427 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2428 struct btrfs_root *root,
2429 struct inode *inode,
2430 u64 new_size, u32 min_type)
2433 struct btrfs_path *path;
2434 struct btrfs_key key;
2435 struct btrfs_key found_key;
2437 struct extent_buffer *leaf;
2438 struct btrfs_file_extent_item *fi;
2439 u64 extent_start = 0;
2440 u64 extent_num_bytes = 0;
2446 int pending_del_nr = 0;
2447 int pending_del_slot = 0;
2448 int extent_type = -1;
2450 u64 mask = root->sectorsize - 1;
2453 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2454 path = btrfs_alloc_path();
2458 /* FIXME, add redo link to tree so we don't leak on crash */
2459 key.objectid = inode->i_ino;
2460 key.offset = (u64)-1;
2463 btrfs_init_path(path);
2465 ret = drop_csum_leaves(trans, root, path, inode, new_size);
2469 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2474 /* there are no items in the tree for us to truncate, we're
2477 if (path->slots[0] == 0) {
2486 leaf = path->nodes[0];
2487 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2488 found_type = btrfs_key_type(&found_key);
2491 if (found_key.objectid != inode->i_ino)
2494 if (found_type < min_type)
2497 item_end = found_key.offset;
2498 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2499 fi = btrfs_item_ptr(leaf, path->slots[0],
2500 struct btrfs_file_extent_item);
2501 extent_type = btrfs_file_extent_type(leaf, fi);
2502 encoding = btrfs_file_extent_compression(leaf, fi);
2503 encoding |= btrfs_file_extent_encryption(leaf, fi);
2504 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2506 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2508 btrfs_file_extent_num_bytes(leaf, fi);
2509 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2510 item_end += btrfs_file_extent_inline_len(leaf,
2515 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2516 ret = btrfs_csum_truncate(trans, root, path,
2520 if (item_end < new_size) {
2521 if (found_type == BTRFS_DIR_ITEM_KEY) {
2522 found_type = BTRFS_INODE_ITEM_KEY;
2523 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2524 found_type = BTRFS_CSUM_ITEM_KEY;
2525 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2526 found_type = BTRFS_XATTR_ITEM_KEY;
2527 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2528 found_type = BTRFS_INODE_REF_KEY;
2529 } else if (found_type) {
2534 btrfs_set_key_type(&key, found_type);
2537 if (found_key.offset >= new_size)
2543 /* FIXME, shrink the extent if the ref count is only 1 */
2544 if (found_type != BTRFS_EXTENT_DATA_KEY)
2547 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2549 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2550 if (!del_item && !encoding) {
2551 u64 orig_num_bytes =
2552 btrfs_file_extent_num_bytes(leaf, fi);
2553 extent_num_bytes = new_size -
2554 found_key.offset + root->sectorsize - 1;
2555 extent_num_bytes = extent_num_bytes &
2556 ~((u64)root->sectorsize - 1);
2557 btrfs_set_file_extent_num_bytes(leaf, fi,
2559 num_dec = (orig_num_bytes -
2561 if (root->ref_cows && extent_start != 0)
2562 inode_sub_bytes(inode, num_dec);
2563 btrfs_mark_buffer_dirty(leaf);
2566 btrfs_file_extent_disk_num_bytes(leaf,
2568 /* FIXME blocksize != 4096 */
2569 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2570 if (extent_start != 0) {
2573 inode_sub_bytes(inode, num_dec);
2575 root_gen = btrfs_header_generation(leaf);
2576 root_owner = btrfs_header_owner(leaf);
2578 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2580 * we can't truncate inline items that have had
2584 btrfs_file_extent_compression(leaf, fi) == 0 &&
2585 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2586 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2587 u32 size = new_size - found_key.offset;
2589 if (root->ref_cows) {
2590 inode_sub_bytes(inode, item_end + 1 -
2594 btrfs_file_extent_calc_inline_size(size);
2595 ret = btrfs_truncate_item(trans, root, path,
2598 } else if (root->ref_cows) {
2599 inode_sub_bytes(inode, item_end + 1 -
2605 if (!pending_del_nr) {
2606 /* no pending yet, add ourselves */
2607 pending_del_slot = path->slots[0];
2609 } else if (pending_del_nr &&
2610 path->slots[0] + 1 == pending_del_slot) {
2611 /* hop on the pending chunk */
2613 pending_del_slot = path->slots[0];
2615 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2621 ret = btrfs_free_extent(trans, root, extent_start,
2623 leaf->start, root_owner,
2624 root_gen, inode->i_ino, 0);
2628 if (path->slots[0] == 0) {
2631 btrfs_release_path(root, path);
2636 if (pending_del_nr &&
2637 path->slots[0] + 1 != pending_del_slot) {
2638 struct btrfs_key debug;
2640 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2642 ret = btrfs_del_items(trans, root, path,
2647 btrfs_release_path(root, path);
2653 if (pending_del_nr) {
2654 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2657 btrfs_free_path(path);
2658 inode->i_sb->s_dirt = 1;
2663 * taken from block_truncate_page, but does cow as it zeros out
2664 * any bytes left in the last page in the file.
2666 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2668 struct inode *inode = mapping->host;
2669 struct btrfs_root *root = BTRFS_I(inode)->root;
2670 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2671 struct btrfs_ordered_extent *ordered;
2673 u32 blocksize = root->sectorsize;
2674 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2675 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2681 if ((offset & (blocksize - 1)) == 0)
2686 page = grab_cache_page(mapping, index);
2690 page_start = page_offset(page);
2691 page_end = page_start + PAGE_CACHE_SIZE - 1;
2693 if (!PageUptodate(page)) {
2694 ret = btrfs_readpage(NULL, page);
2696 if (page->mapping != mapping) {
2698 page_cache_release(page);
2701 if (!PageUptodate(page)) {
2706 wait_on_page_writeback(page);
2708 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2709 set_page_extent_mapped(page);
2711 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2713 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2715 page_cache_release(page);
2716 btrfs_start_ordered_extent(inode, ordered, 1);
2717 btrfs_put_ordered_extent(ordered);
2721 btrfs_set_extent_delalloc(inode, page_start, page_end);
2723 if (offset != PAGE_CACHE_SIZE) {
2725 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2726 flush_dcache_page(page);
2729 ClearPageChecked(page);
2730 set_page_dirty(page);
2731 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2735 page_cache_release(page);
2740 int btrfs_cont_expand(struct inode *inode, loff_t size)
2742 struct btrfs_trans_handle *trans;
2743 struct btrfs_root *root = BTRFS_I(inode)->root;
2744 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2745 struct extent_map *em;
2746 u64 mask = root->sectorsize - 1;
2747 u64 hole_start = (inode->i_size + mask) & ~mask;
2748 u64 block_end = (size + mask) & ~mask;
2754 if (size <= hole_start)
2757 err = btrfs_check_free_space(root, 1, 0);
2761 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2764 struct btrfs_ordered_extent *ordered;
2765 btrfs_wait_ordered_range(inode, hole_start,
2766 block_end - hole_start);
2767 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2768 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2771 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2772 btrfs_put_ordered_extent(ordered);
2775 trans = btrfs_start_transaction(root, 1);
2776 btrfs_set_trans_block_group(trans, inode);
2778 cur_offset = hole_start;
2780 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2781 block_end - cur_offset, 0);
2782 BUG_ON(IS_ERR(em) || !em);
2783 last_byte = min(extent_map_end(em), block_end);
2784 last_byte = (last_byte + mask) & ~mask;
2785 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2787 hole_size = last_byte - cur_offset;
2788 err = btrfs_drop_extents(trans, root, inode,
2790 cur_offset + hole_size,
2791 cur_offset, &hint_byte);
2794 err = btrfs_insert_file_extent(trans, root,
2795 inode->i_ino, cur_offset, 0,
2796 0, hole_size, 0, hole_size,
2798 btrfs_drop_extent_cache(inode, hole_start,
2801 free_extent_map(em);
2802 cur_offset = last_byte;
2803 if (err || cur_offset >= block_end)
2807 btrfs_end_transaction(trans, root);
2808 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2812 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2814 struct inode *inode = dentry->d_inode;
2817 err = inode_change_ok(inode, attr);
2821 if (S_ISREG(inode->i_mode) &&
2822 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2823 err = btrfs_cont_expand(inode, attr->ia_size);
2828 err = inode_setattr(inode, attr);
2830 if (!err && ((attr->ia_valid & ATTR_MODE)))
2831 err = btrfs_acl_chmod(inode);
2835 void btrfs_delete_inode(struct inode *inode)
2837 struct btrfs_trans_handle *trans;
2838 struct btrfs_root *root = BTRFS_I(inode)->root;
2842 truncate_inode_pages(&inode->i_data, 0);
2843 if (is_bad_inode(inode)) {
2844 btrfs_orphan_del(NULL, inode);
2847 btrfs_wait_ordered_range(inode, 0, (u64)-1);
2849 btrfs_i_size_write(inode, 0);
2850 trans = btrfs_start_transaction(root, 1);
2852 btrfs_set_trans_block_group(trans, inode);
2853 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2855 btrfs_orphan_del(NULL, inode);
2856 goto no_delete_lock;
2859 btrfs_orphan_del(trans, inode);
2861 nr = trans->blocks_used;
2864 btrfs_end_transaction(trans, root);
2865 btrfs_btree_balance_dirty(root, nr);
2869 nr = trans->blocks_used;
2870 btrfs_end_transaction(trans, root);
2871 btrfs_btree_balance_dirty(root, nr);
2877 * this returns the key found in the dir entry in the location pointer.
2878 * If no dir entries were found, location->objectid is 0.
2880 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2881 struct btrfs_key *location)
2883 const char *name = dentry->d_name.name;
2884 int namelen = dentry->d_name.len;
2885 struct btrfs_dir_item *di;
2886 struct btrfs_path *path;
2887 struct btrfs_root *root = BTRFS_I(dir)->root;
2890 path = btrfs_alloc_path();
2893 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2897 if (!di || IS_ERR(di)) {
2900 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2902 btrfs_free_path(path);
2905 location->objectid = 0;
2910 * when we hit a tree root in a directory, the btrfs part of the inode
2911 * needs to be changed to reflect the root directory of the tree root. This
2912 * is kind of like crossing a mount point.
2914 static int fixup_tree_root_location(struct btrfs_root *root,
2915 struct btrfs_key *location,
2916 struct btrfs_root **sub_root,
2917 struct dentry *dentry)
2919 struct btrfs_root_item *ri;
2921 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2923 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2926 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2927 dentry->d_name.name,
2928 dentry->d_name.len);
2929 if (IS_ERR(*sub_root))
2930 return PTR_ERR(*sub_root);
2932 ri = &(*sub_root)->root_item;
2933 location->objectid = btrfs_root_dirid(ri);
2934 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2935 location->offset = 0;
2940 static noinline void init_btrfs_i(struct inode *inode)
2942 struct btrfs_inode *bi = BTRFS_I(inode);
2945 bi->i_default_acl = NULL;
2949 bi->logged_trans = 0;
2950 bi->delalloc_bytes = 0;
2951 bi->disk_i_size = 0;
2953 bi->index_cnt = (u64)-1;
2954 bi->log_dirty_trans = 0;
2955 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2956 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2957 inode->i_mapping, GFP_NOFS);
2958 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2959 inode->i_mapping, GFP_NOFS);
2960 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2961 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2962 mutex_init(&BTRFS_I(inode)->csum_mutex);
2963 mutex_init(&BTRFS_I(inode)->extent_mutex);
2964 mutex_init(&BTRFS_I(inode)->log_mutex);
2967 static int btrfs_init_locked_inode(struct inode *inode, void *p)
2969 struct btrfs_iget_args *args = p;
2970 inode->i_ino = args->ino;
2971 init_btrfs_i(inode);
2972 BTRFS_I(inode)->root = args->root;
2976 static int btrfs_find_actor(struct inode *inode, void *opaque)
2978 struct btrfs_iget_args *args = opaque;
2979 return (args->ino == inode->i_ino &&
2980 args->root == BTRFS_I(inode)->root);
2983 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2984 struct btrfs_root *root, int wait)
2986 struct inode *inode;
2987 struct btrfs_iget_args args;
2988 args.ino = objectid;
2992 inode = ilookup5(s, objectid, btrfs_find_actor,
2995 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3001 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3002 struct btrfs_root *root)
3004 struct inode *inode;
3005 struct btrfs_iget_args args;
3006 args.ino = objectid;
3009 inode = iget5_locked(s, objectid, btrfs_find_actor,
3010 btrfs_init_locked_inode,
3015 /* Get an inode object given its location and corresponding root.
3016 * Returns in *is_new if the inode was read from disk
3018 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3019 struct btrfs_root *root, int *is_new)
3021 struct inode *inode;
3023 inode = btrfs_iget_locked(s, location->objectid, root);
3025 return ERR_PTR(-EACCES);
3027 if (inode->i_state & I_NEW) {
3028 BTRFS_I(inode)->root = root;
3029 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3030 btrfs_read_locked_inode(inode);
3031 unlock_new_inode(inode);
3042 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3043 struct nameidata *nd)
3045 struct inode * inode;
3046 struct btrfs_inode *bi = BTRFS_I(dir);
3047 struct btrfs_root *root = bi->root;
3048 struct btrfs_root *sub_root = root;
3049 struct btrfs_key location;
3050 int ret, new, do_orphan = 0;
3052 if (dentry->d_name.len > BTRFS_NAME_LEN)
3053 return ERR_PTR(-ENAMETOOLONG);
3055 ret = btrfs_inode_by_name(dir, dentry, &location);
3058 return ERR_PTR(ret);
3061 if (location.objectid) {
3062 ret = fixup_tree_root_location(root, &location, &sub_root,
3065 return ERR_PTR(ret);
3067 return ERR_PTR(-ENOENT);
3068 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3070 return ERR_CAST(inode);
3072 /* the inode and parent dir are two different roots */
3073 if (new && root != sub_root) {
3075 sub_root->inode = inode;
3080 if (unlikely(do_orphan))
3081 btrfs_orphan_cleanup(sub_root);
3083 return d_splice_alias(inode, dentry);
3086 static unsigned char btrfs_filetype_table[] = {
3087 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3090 static int btrfs_real_readdir(struct file *filp, void *dirent,
3093 struct inode *inode = filp->f_dentry->d_inode;
3094 struct btrfs_root *root = BTRFS_I(inode)->root;
3095 struct btrfs_item *item;
3096 struct btrfs_dir_item *di;
3097 struct btrfs_key key;
3098 struct btrfs_key found_key;
3099 struct btrfs_path *path;
3102 struct extent_buffer *leaf;
3105 unsigned char d_type;
3110 int key_type = BTRFS_DIR_INDEX_KEY;
3115 /* FIXME, use a real flag for deciding about the key type */
3116 if (root->fs_info->tree_root == root)
3117 key_type = BTRFS_DIR_ITEM_KEY;
3119 /* special case for "." */
3120 if (filp->f_pos == 0) {
3121 over = filldir(dirent, ".", 1,
3128 /* special case for .., just use the back ref */
3129 if (filp->f_pos == 1) {
3130 u64 pino = parent_ino(filp->f_path.dentry);
3131 over = filldir(dirent, "..", 2,
3138 path = btrfs_alloc_path();
3141 btrfs_set_key_type(&key, key_type);
3142 key.offset = filp->f_pos;
3143 key.objectid = inode->i_ino;
3145 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3151 leaf = path->nodes[0];
3152 nritems = btrfs_header_nritems(leaf);
3153 slot = path->slots[0];
3154 if (advance || slot >= nritems) {
3155 if (slot >= nritems - 1) {
3156 ret = btrfs_next_leaf(root, path);
3159 leaf = path->nodes[0];
3160 nritems = btrfs_header_nritems(leaf);
3161 slot = path->slots[0];
3168 item = btrfs_item_nr(leaf, slot);
3169 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3171 if (found_key.objectid != key.objectid)
3173 if (btrfs_key_type(&found_key) != key_type)
3175 if (found_key.offset < filp->f_pos)
3178 filp->f_pos = found_key.offset;
3180 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3182 di_total = btrfs_item_size(leaf, item);
3184 while (di_cur < di_total) {
3185 struct btrfs_key location;
3187 name_len = btrfs_dir_name_len(leaf, di);
3188 if (name_len <= sizeof(tmp_name)) {
3189 name_ptr = tmp_name;
3191 name_ptr = kmalloc(name_len, GFP_NOFS);
3197 read_extent_buffer(leaf, name_ptr,
3198 (unsigned long)(di + 1), name_len);
3200 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3201 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3202 over = filldir(dirent, name_ptr, name_len,
3203 found_key.offset, location.objectid,
3206 if (name_ptr != tmp_name)
3212 di_len = btrfs_dir_name_len(leaf, di) +
3213 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3215 di = (struct btrfs_dir_item *)((char *)di + di_len);
3219 /* Reached end of directory/root. Bump pos past the last item. */
3220 if (key_type == BTRFS_DIR_INDEX_KEY)
3221 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3227 btrfs_free_path(path);
3231 int btrfs_write_inode(struct inode *inode, int wait)
3233 struct btrfs_root *root = BTRFS_I(inode)->root;
3234 struct btrfs_trans_handle *trans;
3237 if (root->fs_info->closing > 1)
3241 trans = btrfs_join_transaction(root, 1);
3242 btrfs_set_trans_block_group(trans, inode);
3243 ret = btrfs_commit_transaction(trans, root);
3249 * This is somewhat expensive, updating the tree every time the
3250 * inode changes. But, it is most likely to find the inode in cache.
3251 * FIXME, needs more benchmarking...there are no reasons other than performance
3252 * to keep or drop this code.
3254 void btrfs_dirty_inode(struct inode *inode)
3256 struct btrfs_root *root = BTRFS_I(inode)->root;
3257 struct btrfs_trans_handle *trans;
3259 trans = btrfs_join_transaction(root, 1);
3260 btrfs_set_trans_block_group(trans, inode);
3261 btrfs_update_inode(trans, root, inode);
3262 btrfs_end_transaction(trans, root);
3266 * find the highest existing sequence number in a directory
3267 * and then set the in-memory index_cnt variable to reflect
3268 * free sequence numbers
3270 static int btrfs_set_inode_index_count(struct inode *inode)
3272 struct btrfs_root *root = BTRFS_I(inode)->root;
3273 struct btrfs_key key, found_key;
3274 struct btrfs_path *path;
3275 struct extent_buffer *leaf;
3278 key.objectid = inode->i_ino;
3279 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3280 key.offset = (u64)-1;
3282 path = btrfs_alloc_path();
3286 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3289 /* FIXME: we should be able to handle this */
3295 * MAGIC NUMBER EXPLANATION:
3296 * since we search a directory based on f_pos we have to start at 2
3297 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3298 * else has to start at 2
3300 if (path->slots[0] == 0) {
3301 BTRFS_I(inode)->index_cnt = 2;
3307 leaf = path->nodes[0];
3308 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3310 if (found_key.objectid != inode->i_ino ||
3311 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3312 BTRFS_I(inode)->index_cnt = 2;
3316 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3318 btrfs_free_path(path);
3323 * helper to find a free sequence number in a given directory. This current
3324 * code is very simple, later versions will do smarter things in the btree
3326 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
3331 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3332 ret = btrfs_set_inode_index_count(dir);
3338 *index = BTRFS_I(dir)->index_cnt;
3339 BTRFS_I(dir)->index_cnt++;
3344 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3345 struct btrfs_root *root,
3347 const char *name, int name_len,
3350 struct btrfs_block_group_cache *group,
3351 int mode, u64 *index)
3353 struct inode *inode;
3354 struct btrfs_inode_item *inode_item;
3355 struct btrfs_block_group_cache *new_inode_group;
3356 struct btrfs_key *location;
3357 struct btrfs_path *path;
3358 struct btrfs_inode_ref *ref;
3359 struct btrfs_key key[2];
3365 path = btrfs_alloc_path();
3368 inode = new_inode(root->fs_info->sb);
3370 return ERR_PTR(-ENOMEM);
3373 ret = btrfs_set_inode_index(dir, inode, index);
3375 return ERR_PTR(ret);
3378 * index_cnt is ignored for everything but a dir,
3379 * btrfs_get_inode_index_count has an explanation for the magic
3382 init_btrfs_i(inode);
3383 BTRFS_I(inode)->index_cnt = 2;
3384 BTRFS_I(inode)->root = root;
3385 BTRFS_I(inode)->generation = trans->transid;
3391 new_inode_group = btrfs_find_block_group(root, group, 0,
3392 BTRFS_BLOCK_GROUP_METADATA, owner);
3393 if (!new_inode_group) {
3394 printk("find_block group failed\n");
3395 new_inode_group = group;
3397 BTRFS_I(inode)->block_group = new_inode_group;
3399 key[0].objectid = objectid;
3400 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3403 key[1].objectid = objectid;
3404 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3405 key[1].offset = ref_objectid;
3407 sizes[0] = sizeof(struct btrfs_inode_item);
3408 sizes[1] = name_len + sizeof(*ref);
3410 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3414 if (objectid > root->highest_inode)
3415 root->highest_inode = objectid;
3417 inode->i_uid = current->fsuid;
3418 inode->i_gid = current->fsgid;
3419 inode->i_mode = mode;
3420 inode->i_ino = objectid;
3421 inode_set_bytes(inode, 0);
3422 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3423 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3424 struct btrfs_inode_item);
3425 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3427 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3428 struct btrfs_inode_ref);
3429 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3430 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3431 ptr = (unsigned long)(ref + 1);
3432 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3434 btrfs_mark_buffer_dirty(path->nodes[0]);
3435 btrfs_free_path(path);
3437 location = &BTRFS_I(inode)->location;
3438 location->objectid = objectid;
3439 location->offset = 0;
3440 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3442 insert_inode_hash(inode);
3446 BTRFS_I(dir)->index_cnt--;
3447 btrfs_free_path(path);
3448 return ERR_PTR(ret);
3451 static inline u8 btrfs_inode_type(struct inode *inode)
3453 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3457 * utility function to add 'inode' into 'parent_inode' with
3458 * a give name and a given sequence number.
3459 * if 'add_backref' is true, also insert a backref from the
3460 * inode to the parent directory.
3462 int btrfs_add_link(struct btrfs_trans_handle *trans,
3463 struct inode *parent_inode, struct inode *inode,
3464 const char *name, int name_len, int add_backref, u64 index)
3467 struct btrfs_key key;
3468 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3470 key.objectid = inode->i_ino;
3471 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3474 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3475 parent_inode->i_ino,
3476 &key, btrfs_inode_type(inode),
3480 ret = btrfs_insert_inode_ref(trans, root,
3483 parent_inode->i_ino,
3486 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3488 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3489 ret = btrfs_update_inode(trans, root, parent_inode);
3494 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3495 struct dentry *dentry, struct inode *inode,
3496 int backref, u64 index)
3498 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3499 inode, dentry->d_name.name,
3500 dentry->d_name.len, backref, index);
3502 d_instantiate(dentry, inode);
3510 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3511 int mode, dev_t rdev)
3513 struct btrfs_trans_handle *trans;
3514 struct btrfs_root *root = BTRFS_I(dir)->root;
3515 struct inode *inode = NULL;
3519 unsigned long nr = 0;
3522 if (!new_valid_dev(rdev))
3525 err = btrfs_check_free_space(root, 1, 0);
3529 trans = btrfs_start_transaction(root, 1);
3530 btrfs_set_trans_block_group(trans, dir);
3532 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3538 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3540 dentry->d_parent->d_inode->i_ino, objectid,
3541 BTRFS_I(dir)->block_group, mode, &index);
3542 err = PTR_ERR(inode);
3546 err = btrfs_init_acl(inode, dir);
3552 btrfs_set_trans_block_group(trans, inode);
3553 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3557 inode->i_op = &btrfs_special_inode_operations;
3558 init_special_inode(inode, inode->i_mode, rdev);
3559 btrfs_update_inode(trans, root, inode);
3561 dir->i_sb->s_dirt = 1;
3562 btrfs_update_inode_block_group(trans, inode);
3563 btrfs_update_inode_block_group(trans, dir);
3565 nr = trans->blocks_used;
3566 btrfs_end_transaction_throttle(trans, root);
3569 inode_dec_link_count(inode);
3572 btrfs_btree_balance_dirty(root, nr);
3576 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3577 int mode, struct nameidata *nd)
3579 struct btrfs_trans_handle *trans;
3580 struct btrfs_root *root = BTRFS_I(dir)->root;
3581 struct inode *inode = NULL;
3584 unsigned long nr = 0;
3588 err = btrfs_check_free_space(root, 1, 0);
3591 trans = btrfs_start_transaction(root, 1);
3592 btrfs_set_trans_block_group(trans, dir);
3594 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3600 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3602 dentry->d_parent->d_inode->i_ino,
3603 objectid, BTRFS_I(dir)->block_group, mode,
3605 err = PTR_ERR(inode);
3609 err = btrfs_init_acl(inode, dir);
3615 btrfs_set_trans_block_group(trans, inode);
3616 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3620 inode->i_mapping->a_ops = &btrfs_aops;
3621 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3622 inode->i_fop = &btrfs_file_operations;
3623 inode->i_op = &btrfs_file_inode_operations;
3624 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3626 dir->i_sb->s_dirt = 1;
3627 btrfs_update_inode_block_group(trans, inode);
3628 btrfs_update_inode_block_group(trans, dir);
3630 nr = trans->blocks_used;
3631 btrfs_end_transaction_throttle(trans, root);
3634 inode_dec_link_count(inode);
3637 btrfs_btree_balance_dirty(root, nr);
3641 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3642 struct dentry *dentry)
3644 struct btrfs_trans_handle *trans;
3645 struct btrfs_root *root = BTRFS_I(dir)->root;
3646 struct inode *inode = old_dentry->d_inode;
3648 unsigned long nr = 0;
3652 if (inode->i_nlink == 0)
3655 btrfs_inc_nlink(inode);
3656 err = btrfs_check_free_space(root, 1, 0);
3659 err = btrfs_set_inode_index(dir, inode, &index);
3663 trans = btrfs_start_transaction(root, 1);
3665 btrfs_set_trans_block_group(trans, dir);
3666 atomic_inc(&inode->i_count);
3668 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3673 dir->i_sb->s_dirt = 1;
3674 btrfs_update_inode_block_group(trans, dir);
3675 err = btrfs_update_inode(trans, root, inode);
3680 nr = trans->blocks_used;
3681 btrfs_end_transaction_throttle(trans, root);
3684 inode_dec_link_count(inode);
3687 btrfs_btree_balance_dirty(root, nr);
3691 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3693 struct inode *inode = NULL;
3694 struct btrfs_trans_handle *trans;
3695 struct btrfs_root *root = BTRFS_I(dir)->root;
3697 int drop_on_err = 0;
3700 unsigned long nr = 1;
3702 err = btrfs_check_free_space(root, 1, 0);
3706 trans = btrfs_start_transaction(root, 1);
3707 btrfs_set_trans_block_group(trans, dir);
3709 if (IS_ERR(trans)) {
3710 err = PTR_ERR(trans);
3714 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3720 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3722 dentry->d_parent->d_inode->i_ino, objectid,
3723 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3725 if (IS_ERR(inode)) {
3726 err = PTR_ERR(inode);
3732 err = btrfs_init_acl(inode, dir);
3736 inode->i_op = &btrfs_dir_inode_operations;
3737 inode->i_fop = &btrfs_dir_file_operations;
3738 btrfs_set_trans_block_group(trans, inode);
3740 btrfs_i_size_write(inode, 0);
3741 err = btrfs_update_inode(trans, root, inode);
3745 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3746 inode, dentry->d_name.name,
3747 dentry->d_name.len, 0, index);
3751 d_instantiate(dentry, inode);
3753 dir->i_sb->s_dirt = 1;
3754 btrfs_update_inode_block_group(trans, inode);
3755 btrfs_update_inode_block_group(trans, dir);
3758 nr = trans->blocks_used;
3759 btrfs_end_transaction_throttle(trans, root);
3764 btrfs_btree_balance_dirty(root, nr);
3768 /* helper for btfs_get_extent. Given an existing extent in the tree,
3769 * and an extent that you want to insert, deal with overlap and insert
3770 * the new extent into the tree.
3772 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3773 struct extent_map *existing,
3774 struct extent_map *em,
3775 u64 map_start, u64 map_len)
3779 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3780 start_diff = map_start - em->start;
3781 em->start = map_start;
3783 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3784 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3785 em->block_start += start_diff;
3786 em->block_len -= start_diff;
3788 return add_extent_mapping(em_tree, em);
3791 static noinline int uncompress_inline(struct btrfs_path *path,
3792 struct inode *inode, struct page *page,
3793 size_t pg_offset, u64 extent_offset,
3794 struct btrfs_file_extent_item *item)
3797 struct extent_buffer *leaf = path->nodes[0];
3800 unsigned long inline_size;
3803 WARN_ON(pg_offset != 0);
3804 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3805 inline_size = btrfs_file_extent_inline_item_len(leaf,
3806 btrfs_item_nr(leaf, path->slots[0]));
3807 tmp = kmalloc(inline_size, GFP_NOFS);
3808 ptr = btrfs_file_extent_inline_start(item);
3810 read_extent_buffer(leaf, tmp, ptr, inline_size);
3812 max_size = min(PAGE_CACHE_SIZE, max_size);
3813 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3814 inline_size, max_size);
3816 char *kaddr = kmap_atomic(page, KM_USER0);
3817 unsigned long copy_size = min_t(u64,
3818 PAGE_CACHE_SIZE - pg_offset,
3819 max_size - extent_offset);
3820 memset(kaddr + pg_offset, 0, copy_size);
3821 kunmap_atomic(kaddr, KM_USER0);
3828 * a bit scary, this does extent mapping from logical file offset to the disk.
3829 * the ugly parts come from merging extents from the disk with the
3830 * in-ram representation. This gets more complex because of the data=ordered code,
3831 * where the in-ram extents might be locked pending data=ordered completion.
3833 * This also copies inline extents directly into the page.
3835 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3836 size_t pg_offset, u64 start, u64 len,
3842 u64 extent_start = 0;
3844 u64 objectid = inode->i_ino;
3846 struct btrfs_path *path = NULL;
3847 struct btrfs_root *root = BTRFS_I(inode)->root;
3848 struct btrfs_file_extent_item *item;
3849 struct extent_buffer *leaf;
3850 struct btrfs_key found_key;
3851 struct extent_map *em = NULL;
3852 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3853 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3854 struct btrfs_trans_handle *trans = NULL;
3858 spin_lock(&em_tree->lock);
3859 em = lookup_extent_mapping(em_tree, start, len);
3861 em->bdev = root->fs_info->fs_devices->latest_bdev;
3862 spin_unlock(&em_tree->lock);
3865 if (em->start > start || em->start + em->len <= start)
3866 free_extent_map(em);
3867 else if (em->block_start == EXTENT_MAP_INLINE && page)
3868 free_extent_map(em);
3872 em = alloc_extent_map(GFP_NOFS);
3877 em->bdev = root->fs_info->fs_devices->latest_bdev;
3878 em->start = EXTENT_MAP_HOLE;
3880 em->block_len = (u64)-1;
3883 path = btrfs_alloc_path();
3887 ret = btrfs_lookup_file_extent(trans, root, path,
3888 objectid, start, trans != NULL);
3895 if (path->slots[0] == 0)
3900 leaf = path->nodes[0];
3901 item = btrfs_item_ptr(leaf, path->slots[0],
3902 struct btrfs_file_extent_item);
3903 /* are we inside the extent that was found? */
3904 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3905 found_type = btrfs_key_type(&found_key);
3906 if (found_key.objectid != objectid ||
3907 found_type != BTRFS_EXTENT_DATA_KEY) {
3911 found_type = btrfs_file_extent_type(leaf, item);
3912 extent_start = found_key.offset;
3913 compressed = btrfs_file_extent_compression(leaf, item);
3914 if (found_type == BTRFS_FILE_EXTENT_REG ||
3915 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3916 extent_end = extent_start +
3917 btrfs_file_extent_num_bytes(leaf, item);
3918 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3920 size = btrfs_file_extent_inline_len(leaf, item);
3921 extent_end = (extent_start + size + root->sectorsize - 1) &
3922 ~((u64)root->sectorsize - 1);
3925 if (start >= extent_end) {
3927 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3928 ret = btrfs_next_leaf(root, path);
3935 leaf = path->nodes[0];
3937 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3938 if (found_key.objectid != objectid ||
3939 found_key.type != BTRFS_EXTENT_DATA_KEY)
3941 if (start + len <= found_key.offset)
3944 em->len = found_key.offset - start;
3948 if (found_type == BTRFS_FILE_EXTENT_REG ||
3949 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3950 em->start = extent_start;
3951 em->len = extent_end - extent_start;
3952 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3954 em->block_start = EXTENT_MAP_HOLE;
3958 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3959 em->block_start = bytenr;
3960 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3963 bytenr += btrfs_file_extent_offset(leaf, item);
3964 em->block_start = bytenr;
3965 em->block_len = em->len;
3966 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3967 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3970 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3974 size_t extent_offset;
3977 em->block_start = EXTENT_MAP_INLINE;
3978 if (!page || create) {
3979 em->start = extent_start;
3980 em->len = extent_end - extent_start;
3984 size = btrfs_file_extent_inline_len(leaf, item);
3985 extent_offset = page_offset(page) + pg_offset - extent_start;
3986 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3987 size - extent_offset);
3988 em->start = extent_start + extent_offset;
3989 em->len = (copy_size + root->sectorsize - 1) &
3990 ~((u64)root->sectorsize - 1);
3992 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3993 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
3994 if (create == 0 && !PageUptodate(page)) {
3995 if (btrfs_file_extent_compression(leaf, item) ==
3996 BTRFS_COMPRESS_ZLIB) {
3997 ret = uncompress_inline(path, inode, page,
3999 extent_offset, item);
4003 read_extent_buffer(leaf, map + pg_offset, ptr,
4007 flush_dcache_page(page);
4008 } else if (create && PageUptodate(page)) {
4011 free_extent_map(em);
4013 btrfs_release_path(root, path);
4014 trans = btrfs_join_transaction(root, 1);
4018 write_extent_buffer(leaf, map + pg_offset, ptr,
4021 btrfs_mark_buffer_dirty(leaf);
4023 set_extent_uptodate(io_tree, em->start,
4024 extent_map_end(em) - 1, GFP_NOFS);
4027 printk("unkknown found_type %d\n", found_type);
4034 em->block_start = EXTENT_MAP_HOLE;
4035 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4037 btrfs_release_path(root, path);
4038 if (em->start > start || extent_map_end(em) <= start) {
4039 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
4045 spin_lock(&em_tree->lock);
4046 ret = add_extent_mapping(em_tree, em);
4047 /* it is possible that someone inserted the extent into the tree
4048 * while we had the lock dropped. It is also possible that
4049 * an overlapping map exists in the tree
4051 if (ret == -EEXIST) {
4052 struct extent_map *existing;
4056 existing = lookup_extent_mapping(em_tree, start, len);
4057 if (existing && (existing->start > start ||
4058 existing->start + existing->len <= start)) {
4059 free_extent_map(existing);
4063 existing = lookup_extent_mapping(em_tree, em->start,
4066 err = merge_extent_mapping(em_tree, existing,
4069 free_extent_map(existing);
4071 free_extent_map(em);
4076 printk("failing to insert %Lu %Lu\n",
4078 free_extent_map(em);
4082 free_extent_map(em);
4087 spin_unlock(&em_tree->lock);
4090 btrfs_free_path(path);
4092 ret = btrfs_end_transaction(trans, root);
4098 free_extent_map(em);
4100 return ERR_PTR(err);
4105 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4106 const struct iovec *iov, loff_t offset,
4107 unsigned long nr_segs)
4112 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4114 return extent_bmap(mapping, iblock, btrfs_get_extent);
4117 int btrfs_readpage(struct file *file, struct page *page)
4119 struct extent_io_tree *tree;
4120 tree = &BTRFS_I(page->mapping->host)->io_tree;
4121 return extent_read_full_page(tree, page, btrfs_get_extent);
4124 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4126 struct extent_io_tree *tree;
4129 if (current->flags & PF_MEMALLOC) {
4130 redirty_page_for_writepage(wbc, page);
4134 tree = &BTRFS_I(page->mapping->host)->io_tree;
4135 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4138 int btrfs_writepages(struct address_space *mapping,
4139 struct writeback_control *wbc)
4141 struct extent_io_tree *tree;
4143 tree = &BTRFS_I(mapping->host)->io_tree;
4144 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4148 btrfs_readpages(struct file *file, struct address_space *mapping,
4149 struct list_head *pages, unsigned nr_pages)
4151 struct extent_io_tree *tree;
4152 tree = &BTRFS_I(mapping->host)->io_tree;
4153 return extent_readpages(tree, mapping, pages, nr_pages,
4156 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4158 struct extent_io_tree *tree;
4159 struct extent_map_tree *map;
4162 tree = &BTRFS_I(page->mapping->host)->io_tree;
4163 map = &BTRFS_I(page->mapping->host)->extent_tree;
4164 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4166 ClearPagePrivate(page);
4167 set_page_private(page, 0);
4168 page_cache_release(page);
4173 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4175 if (PageWriteback(page) || PageDirty(page))
4177 return __btrfs_releasepage(page, gfp_flags);
4180 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4182 struct extent_io_tree *tree;
4183 struct btrfs_ordered_extent *ordered;
4184 u64 page_start = page_offset(page);
4185 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4187 wait_on_page_writeback(page);
4188 tree = &BTRFS_I(page->mapping->host)->io_tree;
4190 btrfs_releasepage(page, GFP_NOFS);
4194 lock_extent(tree, page_start, page_end, GFP_NOFS);
4195 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4199 * IO on this page will never be started, so we need
4200 * to account for any ordered extents now
4202 clear_extent_bit(tree, page_start, page_end,
4203 EXTENT_DIRTY | EXTENT_DELALLOC |
4204 EXTENT_LOCKED, 1, 0, GFP_NOFS);
4205 btrfs_finish_ordered_io(page->mapping->host,
4206 page_start, page_end);
4207 btrfs_put_ordered_extent(ordered);
4208 lock_extent(tree, page_start, page_end, GFP_NOFS);
4210 clear_extent_bit(tree, page_start, page_end,
4211 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4214 __btrfs_releasepage(page, GFP_NOFS);
4216 ClearPageChecked(page);
4217 if (PagePrivate(page)) {
4218 ClearPagePrivate(page);
4219 set_page_private(page, 0);
4220 page_cache_release(page);
4225 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4226 * called from a page fault handler when a page is first dirtied. Hence we must
4227 * be careful to check for EOF conditions here. We set the page up correctly
4228 * for a written page which means we get ENOSPC checking when writing into
4229 * holes and correct delalloc and unwritten extent mapping on filesystems that
4230 * support these features.
4232 * We are not allowed to take the i_mutex here so we have to play games to
4233 * protect against truncate races as the page could now be beyond EOF. Because
4234 * vmtruncate() writes the inode size before removing pages, once we have the
4235 * page lock we can determine safely if the page is beyond EOF. If it is not
4236 * beyond EOF, then the page is guaranteed safe against truncation until we
4239 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4241 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4242 struct btrfs_root *root = BTRFS_I(inode)->root;
4243 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4244 struct btrfs_ordered_extent *ordered;
4246 unsigned long zero_start;
4252 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4259 size = i_size_read(inode);
4260 page_start = page_offset(page);
4261 page_end = page_start + PAGE_CACHE_SIZE - 1;
4263 if ((page->mapping != inode->i_mapping) ||
4264 (page_start >= size)) {
4265 /* page got truncated out from underneath us */
4268 wait_on_page_writeback(page);
4270 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4271 set_page_extent_mapped(page);
4274 * we can't set the delalloc bits if there are pending ordered
4275 * extents. Drop our locks and wait for them to finish
4277 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4279 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4281 btrfs_start_ordered_extent(inode, ordered, 1);
4282 btrfs_put_ordered_extent(ordered);
4286 btrfs_set_extent_delalloc(inode, page_start, page_end);
4289 /* page is wholly or partially inside EOF */
4290 if (page_start + PAGE_CACHE_SIZE > size)
4291 zero_start = size & ~PAGE_CACHE_MASK;
4293 zero_start = PAGE_CACHE_SIZE;
4295 if (zero_start != PAGE_CACHE_SIZE) {
4297 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4298 flush_dcache_page(page);
4301 ClearPageChecked(page);
4302 set_page_dirty(page);
4303 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4311 static void btrfs_truncate(struct inode *inode)
4313 struct btrfs_root *root = BTRFS_I(inode)->root;
4315 struct btrfs_trans_handle *trans;
4317 u64 mask = root->sectorsize - 1;
4319 if (!S_ISREG(inode->i_mode))
4321 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4324 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4325 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4327 trans = btrfs_start_transaction(root, 1);
4328 btrfs_set_trans_block_group(trans, inode);
4329 btrfs_i_size_write(inode, inode->i_size);
4331 ret = btrfs_orphan_add(trans, inode);
4334 /* FIXME, add redo link to tree so we don't leak on crash */
4335 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4336 BTRFS_EXTENT_DATA_KEY);
4337 btrfs_update_inode(trans, root, inode);
4339 ret = btrfs_orphan_del(trans, inode);
4343 nr = trans->blocks_used;
4344 ret = btrfs_end_transaction_throttle(trans, root);
4346 btrfs_btree_balance_dirty(root, nr);
4350 * Invalidate a single dcache entry at the root of the filesystem.
4351 * Needed after creation of snapshot or subvolume.
4353 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
4356 struct dentry *alias, *entry;
4359 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
4363 /* change me if btrfs ever gets a d_hash operation */
4364 qstr.hash = full_name_hash(qstr.name, qstr.len);
4365 entry = d_lookup(alias, &qstr);
4368 d_invalidate(entry);
4375 * create a new subvolume directory/inode (helper for the ioctl).
4377 int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4378 struct btrfs_trans_handle *trans, u64 new_dirid,
4379 struct btrfs_block_group_cache *block_group)
4381 struct inode *inode;
4385 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4386 new_dirid, block_group, S_IFDIR | 0700, &index);
4388 return PTR_ERR(inode);
4389 inode->i_op = &btrfs_dir_inode_operations;
4390 inode->i_fop = &btrfs_dir_file_operations;
4391 new_root->inode = inode;
4394 btrfs_i_size_write(inode, 0);
4396 error = btrfs_update_inode(trans, new_root, inode);
4400 atomic_inc(&inode->i_count);
4401 d_instantiate(dentry, inode);
4405 /* helper function for file defrag and space balancing. This
4406 * forces readahead on a given range of bytes in an inode
4408 unsigned long btrfs_force_ra(struct address_space *mapping,
4409 struct file_ra_state *ra, struct file *file,
4410 pgoff_t offset, pgoff_t last_index)
4412 pgoff_t req_size = last_index - offset + 1;
4414 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4415 return offset + req_size;
4418 struct inode *btrfs_alloc_inode(struct super_block *sb)
4420 struct btrfs_inode *ei;
4422 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4426 ei->logged_trans = 0;
4427 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4428 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4429 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4430 INIT_LIST_HEAD(&ei->i_orphan);
4431 return &ei->vfs_inode;
4434 void btrfs_destroy_inode(struct inode *inode)
4436 struct btrfs_ordered_extent *ordered;
4437 WARN_ON(!list_empty(&inode->i_dentry));
4438 WARN_ON(inode->i_data.nrpages);
4440 if (BTRFS_I(inode)->i_acl &&
4441 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4442 posix_acl_release(BTRFS_I(inode)->i_acl);
4443 if (BTRFS_I(inode)->i_default_acl &&
4444 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4445 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4447 spin_lock(&BTRFS_I(inode)->root->list_lock);
4448 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4449 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4450 " list\n", inode->i_ino);
4453 spin_unlock(&BTRFS_I(inode)->root->list_lock);
4456 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4460 printk("found ordered extent %Lu %Lu\n",
4461 ordered->file_offset, ordered->len);
4462 btrfs_remove_ordered_extent(inode, ordered);
4463 btrfs_put_ordered_extent(ordered);
4464 btrfs_put_ordered_extent(ordered);
4467 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4468 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4471 static void init_once(void *foo)
4473 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4475 inode_init_once(&ei->vfs_inode);
4478 void btrfs_destroy_cachep(void)
4480 if (btrfs_inode_cachep)
4481 kmem_cache_destroy(btrfs_inode_cachep);
4482 if (btrfs_trans_handle_cachep)
4483 kmem_cache_destroy(btrfs_trans_handle_cachep);
4484 if (btrfs_transaction_cachep)
4485 kmem_cache_destroy(btrfs_transaction_cachep);
4486 if (btrfs_bit_radix_cachep)
4487 kmem_cache_destroy(btrfs_bit_radix_cachep);
4488 if (btrfs_path_cachep)
4489 kmem_cache_destroy(btrfs_path_cachep);
4492 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4493 unsigned long extra_flags,
4494 void (*ctor)(void *))
4496 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4497 SLAB_MEM_SPREAD | extra_flags), ctor);
4500 int btrfs_init_cachep(void)
4502 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4503 sizeof(struct btrfs_inode),
4505 if (!btrfs_inode_cachep)
4507 btrfs_trans_handle_cachep =
4508 btrfs_cache_create("btrfs_trans_handle_cache",
4509 sizeof(struct btrfs_trans_handle),
4511 if (!btrfs_trans_handle_cachep)
4513 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4514 sizeof(struct btrfs_transaction),
4516 if (!btrfs_transaction_cachep)
4518 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4519 sizeof(struct btrfs_path),
4521 if (!btrfs_path_cachep)
4523 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4524 SLAB_DESTROY_BY_RCU, NULL);
4525 if (!btrfs_bit_radix_cachep)
4529 btrfs_destroy_cachep();
4533 static int btrfs_getattr(struct vfsmount *mnt,
4534 struct dentry *dentry, struct kstat *stat)
4536 struct inode *inode = dentry->d_inode;
4537 generic_fillattr(inode, stat);
4538 stat->blksize = PAGE_CACHE_SIZE;
4539 stat->blocks = (inode_get_bytes(inode) +
4540 BTRFS_I(inode)->delalloc_bytes) >> 9;
4544 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4545 struct inode * new_dir,struct dentry *new_dentry)
4547 struct btrfs_trans_handle *trans;
4548 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4549 struct inode *new_inode = new_dentry->d_inode;
4550 struct inode *old_inode = old_dentry->d_inode;
4551 struct timespec ctime = CURRENT_TIME;
4555 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4556 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4560 ret = btrfs_check_free_space(root, 1, 0);
4564 trans = btrfs_start_transaction(root, 1);
4566 btrfs_set_trans_block_group(trans, new_dir);
4568 btrfs_inc_nlink(old_dentry->d_inode);
4569 old_dir->i_ctime = old_dir->i_mtime = ctime;
4570 new_dir->i_ctime = new_dir->i_mtime = ctime;
4571 old_inode->i_ctime = ctime;
4573 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4574 old_dentry->d_name.name,
4575 old_dentry->d_name.len);
4580 new_inode->i_ctime = CURRENT_TIME;
4581 ret = btrfs_unlink_inode(trans, root, new_dir,
4582 new_dentry->d_inode,
4583 new_dentry->d_name.name,
4584 new_dentry->d_name.len);
4587 if (new_inode->i_nlink == 0) {
4588 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4594 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
4598 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4599 old_inode, new_dentry->d_name.name,
4600 new_dentry->d_name.len, 1, index);
4605 btrfs_end_transaction_throttle(trans, root);
4611 * some fairly slow code that needs optimization. This walks the list
4612 * of all the inodes with pending delalloc and forces them to disk.
4614 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4616 struct list_head *head = &root->fs_info->delalloc_inodes;
4617 struct btrfs_inode *binode;
4618 struct inode *inode;
4619 unsigned long flags;
4621 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4622 while(!list_empty(head)) {
4623 binode = list_entry(head->next, struct btrfs_inode,
4625 inode = igrab(&binode->vfs_inode);
4627 list_del_init(&binode->delalloc_inodes);
4628 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4630 filemap_flush(inode->i_mapping);
4634 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4636 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4638 /* the filemap_flush will queue IO into the worker threads, but
4639 * we have to make sure the IO is actually started and that
4640 * ordered extents get created before we return
4642 atomic_inc(&root->fs_info->async_submit_draining);
4643 while(atomic_read(&root->fs_info->nr_async_submits) ||
4644 atomic_read(&root->fs_info->async_delalloc_pages)) {
4645 wait_event(root->fs_info->async_submit_wait,
4646 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4647 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4649 atomic_dec(&root->fs_info->async_submit_draining);
4653 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4654 const char *symname)
4656 struct btrfs_trans_handle *trans;
4657 struct btrfs_root *root = BTRFS_I(dir)->root;
4658 struct btrfs_path *path;
4659 struct btrfs_key key;
4660 struct inode *inode = NULL;
4668 struct btrfs_file_extent_item *ei;
4669 struct extent_buffer *leaf;
4670 unsigned long nr = 0;
4672 name_len = strlen(symname) + 1;
4673 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4674 return -ENAMETOOLONG;
4676 err = btrfs_check_free_space(root, 1, 0);
4680 trans = btrfs_start_transaction(root, 1);
4681 btrfs_set_trans_block_group(trans, dir);
4683 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4689 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4691 dentry->d_parent->d_inode->i_ino, objectid,
4692 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4694 err = PTR_ERR(inode);
4698 err = btrfs_init_acl(inode, dir);
4704 btrfs_set_trans_block_group(trans, inode);
4705 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4709 inode->i_mapping->a_ops = &btrfs_aops;
4710 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4711 inode->i_fop = &btrfs_file_operations;
4712 inode->i_op = &btrfs_file_inode_operations;
4713 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4715 dir->i_sb->s_dirt = 1;
4716 btrfs_update_inode_block_group(trans, inode);
4717 btrfs_update_inode_block_group(trans, dir);
4721 path = btrfs_alloc_path();
4723 key.objectid = inode->i_ino;
4725 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4726 datasize = btrfs_file_extent_calc_inline_size(name_len);
4727 err = btrfs_insert_empty_item(trans, root, path, &key,
4733 leaf = path->nodes[0];
4734 ei = btrfs_item_ptr(leaf, path->slots[0],
4735 struct btrfs_file_extent_item);
4736 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4737 btrfs_set_file_extent_type(leaf, ei,
4738 BTRFS_FILE_EXTENT_INLINE);
4739 btrfs_set_file_extent_encryption(leaf, ei, 0);
4740 btrfs_set_file_extent_compression(leaf, ei, 0);
4741 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4742 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4744 ptr = btrfs_file_extent_inline_start(ei);
4745 write_extent_buffer(leaf, symname, ptr, name_len);
4746 btrfs_mark_buffer_dirty(leaf);
4747 btrfs_free_path(path);
4749 inode->i_op = &btrfs_symlink_inode_operations;
4750 inode->i_mapping->a_ops = &btrfs_symlink_aops;
4751 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4752 inode_set_bytes(inode, name_len);
4753 btrfs_i_size_write(inode, name_len - 1);
4754 err = btrfs_update_inode(trans, root, inode);
4759 nr = trans->blocks_used;
4760 btrfs_end_transaction_throttle(trans, root);
4763 inode_dec_link_count(inode);
4766 btrfs_btree_balance_dirty(root, nr);
4770 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4771 u64 alloc_hint, int mode)
4773 struct btrfs_trans_handle *trans;
4774 struct btrfs_root *root = BTRFS_I(inode)->root;
4775 struct btrfs_key ins;
4777 u64 cur_offset = start;
4778 u64 num_bytes = end - start;
4781 trans = btrfs_join_transaction(root, 1);
4783 btrfs_set_trans_block_group(trans, inode);
4785 while (num_bytes > 0) {
4786 alloc_size = min(num_bytes, root->fs_info->max_extent);
4787 ret = btrfs_reserve_extent(trans, root, alloc_size,
4788 root->sectorsize, 0, alloc_hint,
4794 ret = insert_reserved_file_extent(trans, inode,
4795 cur_offset, ins.objectid,
4796 ins.offset, ins.offset,
4797 ins.offset, 0, 0, 0,
4798 BTRFS_FILE_EXTENT_PREALLOC);
4800 num_bytes -= ins.offset;
4801 cur_offset += ins.offset;
4802 alloc_hint = ins.objectid + ins.offset;
4805 if (cur_offset > start) {
4806 inode->i_ctime = CURRENT_TIME;
4807 btrfs_set_flag(inode, PREALLOC);
4808 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4809 cur_offset > i_size_read(inode))
4810 btrfs_i_size_write(inode, cur_offset);
4811 ret = btrfs_update_inode(trans, root, inode);
4815 btrfs_end_transaction(trans, root);
4819 static long btrfs_fallocate(struct inode *inode, int mode,
4820 loff_t offset, loff_t len)
4827 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4828 struct extent_map *em;
4831 alloc_start = offset & ~mask;
4832 alloc_end = (offset + len + mask) & ~mask;
4834 mutex_lock(&inode->i_mutex);
4835 if (alloc_start > inode->i_size) {
4836 ret = btrfs_cont_expand(inode, alloc_start);
4842 struct btrfs_ordered_extent *ordered;
4843 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4844 alloc_end - 1, GFP_NOFS);
4845 ordered = btrfs_lookup_first_ordered_extent(inode,
4848 ordered->file_offset + ordered->len > alloc_start &&
4849 ordered->file_offset < alloc_end) {
4850 btrfs_put_ordered_extent(ordered);
4851 unlock_extent(&BTRFS_I(inode)->io_tree,
4852 alloc_start, alloc_end - 1, GFP_NOFS);
4853 btrfs_wait_ordered_range(inode, alloc_start,
4854 alloc_end - alloc_start);
4857 btrfs_put_ordered_extent(ordered);
4862 cur_offset = alloc_start;
4864 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4865 alloc_end - cur_offset, 0);
4866 BUG_ON(IS_ERR(em) || !em);
4867 last_byte = min(extent_map_end(em), alloc_end);
4868 last_byte = (last_byte + mask) & ~mask;
4869 if (em->block_start == EXTENT_MAP_HOLE) {
4870 ret = prealloc_file_range(inode, cur_offset,
4871 last_byte, alloc_hint, mode);
4873 free_extent_map(em);
4877 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4878 alloc_hint = em->block_start;
4879 free_extent_map(em);
4881 cur_offset = last_byte;
4882 if (cur_offset >= alloc_end) {
4887 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4890 mutex_unlock(&inode->i_mutex);
4894 static int btrfs_set_page_dirty(struct page *page)
4896 return __set_page_dirty_nobuffers(page);
4899 static int btrfs_permission(struct inode *inode, int mask)
4901 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4903 return generic_permission(inode, mask, btrfs_check_acl);
4906 static struct inode_operations btrfs_dir_inode_operations = {
4907 .lookup = btrfs_lookup,
4908 .create = btrfs_create,
4909 .unlink = btrfs_unlink,
4911 .mkdir = btrfs_mkdir,
4912 .rmdir = btrfs_rmdir,
4913 .rename = btrfs_rename,
4914 .symlink = btrfs_symlink,
4915 .setattr = btrfs_setattr,
4916 .mknod = btrfs_mknod,
4917 .setxattr = btrfs_setxattr,
4918 .getxattr = btrfs_getxattr,
4919 .listxattr = btrfs_listxattr,
4920 .removexattr = btrfs_removexattr,
4921 .permission = btrfs_permission,
4923 static struct inode_operations btrfs_dir_ro_inode_operations = {
4924 .lookup = btrfs_lookup,
4925 .permission = btrfs_permission,
4927 static struct file_operations btrfs_dir_file_operations = {
4928 .llseek = generic_file_llseek,
4929 .read = generic_read_dir,
4930 .readdir = btrfs_real_readdir,
4931 .unlocked_ioctl = btrfs_ioctl,
4932 #ifdef CONFIG_COMPAT
4933 .compat_ioctl = btrfs_ioctl,
4935 .release = btrfs_release_file,
4936 .fsync = btrfs_sync_file,
4939 static struct extent_io_ops btrfs_extent_io_ops = {
4940 .fill_delalloc = run_delalloc_range,
4941 .submit_bio_hook = btrfs_submit_bio_hook,
4942 .merge_bio_hook = btrfs_merge_bio_hook,
4943 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4944 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4945 .writepage_start_hook = btrfs_writepage_start_hook,
4946 .readpage_io_failed_hook = btrfs_io_failed_hook,
4947 .set_bit_hook = btrfs_set_bit_hook,
4948 .clear_bit_hook = btrfs_clear_bit_hook,
4951 static struct address_space_operations btrfs_aops = {
4952 .readpage = btrfs_readpage,
4953 .writepage = btrfs_writepage,
4954 .writepages = btrfs_writepages,
4955 .readpages = btrfs_readpages,
4956 .sync_page = block_sync_page,
4958 .direct_IO = btrfs_direct_IO,
4959 .invalidatepage = btrfs_invalidatepage,
4960 .releasepage = btrfs_releasepage,
4961 .set_page_dirty = btrfs_set_page_dirty,
4964 static struct address_space_operations btrfs_symlink_aops = {
4965 .readpage = btrfs_readpage,
4966 .writepage = btrfs_writepage,
4967 .invalidatepage = btrfs_invalidatepage,
4968 .releasepage = btrfs_releasepage,
4971 static struct inode_operations btrfs_file_inode_operations = {
4972 .truncate = btrfs_truncate,
4973 .getattr = btrfs_getattr,
4974 .setattr = btrfs_setattr,
4975 .setxattr = btrfs_setxattr,
4976 .getxattr = btrfs_getxattr,
4977 .listxattr = btrfs_listxattr,
4978 .removexattr = btrfs_removexattr,
4979 .permission = btrfs_permission,
4980 .fallocate = btrfs_fallocate,
4982 static struct inode_operations btrfs_special_inode_operations = {
4983 .getattr = btrfs_getattr,
4984 .setattr = btrfs_setattr,
4985 .permission = btrfs_permission,
4986 .setxattr = btrfs_setxattr,
4987 .getxattr = btrfs_getxattr,
4988 .listxattr = btrfs_listxattr,
4989 .removexattr = btrfs_removexattr,
4991 static struct inode_operations btrfs_symlink_inode_operations = {
4992 .readlink = generic_readlink,
4993 .follow_link = page_follow_link_light,
4994 .put_link = page_put_link,
4995 .permission = btrfs_permission,