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_t(unsigned long, 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;
613 em->orig_start = em->start;
615 em->block_start = ins.objectid;
616 em->block_len = ins.offset;
617 em->bdev = root->fs_info->fs_devices->latest_bdev;
618 set_bit(EXTENT_FLAG_PINNED, &em->flags);
619 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
622 spin_lock(&em_tree->lock);
623 ret = add_extent_mapping(em_tree, em);
624 spin_unlock(&em_tree->lock);
625 if (ret != -EEXIST) {
629 btrfs_drop_extent_cache(inode, async_extent->start,
630 async_extent->start +
631 async_extent->ram_size - 1, 0);
634 ret = btrfs_add_ordered_extent(inode, async_extent->start,
636 async_extent->ram_size,
638 BTRFS_ORDERED_COMPRESSED);
641 btrfs_end_transaction(trans, root);
644 * clear dirty, set writeback and unlock the pages.
646 extent_clear_unlock_delalloc(inode,
647 &BTRFS_I(inode)->io_tree,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 NULL, 1, 1, 0, 1, 1, 0);
653 ret = btrfs_submit_compressed_write(inode,
655 async_extent->ram_size,
657 ins.offset, async_extent->pages,
658 async_extent->nr_pages);
661 trans = btrfs_join_transaction(root, 1);
662 alloc_hint = ins.objectid + ins.offset;
667 btrfs_end_transaction(trans, root);
672 * when extent_io.c finds a delayed allocation range in the file,
673 * the call backs end up in this code. The basic idea is to
674 * allocate extents on disk for the range, and create ordered data structs
675 * in ram to track those extents.
677 * locked_page is the page that writepage had locked already. We use
678 * it to make sure we don't do extra locks or unlocks.
680 * *page_started is set to one if we unlock locked_page and do everything
681 * required to start IO on it. It may be clean and already done with
684 static noinline int cow_file_range(struct inode *inode,
685 struct page *locked_page,
686 u64 start, u64 end, int *page_started,
687 unsigned long *nr_written,
690 struct btrfs_root *root = BTRFS_I(inode)->root;
691 struct btrfs_trans_handle *trans;
694 unsigned long ram_size;
697 u64 blocksize = root->sectorsize;
699 struct btrfs_key ins;
700 struct extent_map *em;
701 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
704 trans = btrfs_join_transaction(root, 1);
706 btrfs_set_trans_block_group(trans, inode);
708 actual_end = min_t(u64, i_size_read(inode), end + 1);
710 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
711 num_bytes = max(blocksize, num_bytes);
712 disk_num_bytes = num_bytes;
716 /* lets try to make an inline extent */
717 ret = cow_file_range_inline(trans, root, inode,
718 start, end, 0, NULL);
720 extent_clear_unlock_delalloc(inode,
721 &BTRFS_I(inode)->io_tree,
722 start, end, NULL, 1, 1,
724 *nr_written = *nr_written +
725 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
732 BUG_ON(disk_num_bytes >
733 btrfs_super_total_bytes(&root->fs_info->super_copy));
735 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
737 while(disk_num_bytes > 0) {
738 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
739 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
740 root->sectorsize, 0, alloc_hint,
745 em = alloc_extent_map(GFP_NOFS);
747 em->orig_start = em->start;
749 ram_size = ins.offset;
750 em->len = ins.offset;
752 em->block_start = ins.objectid;
753 em->block_len = ins.offset;
754 em->bdev = root->fs_info->fs_devices->latest_bdev;
755 set_bit(EXTENT_FLAG_PINNED, &em->flags);
758 spin_lock(&em_tree->lock);
759 ret = add_extent_mapping(em_tree, em);
760 spin_unlock(&em_tree->lock);
761 if (ret != -EEXIST) {
765 btrfs_drop_extent_cache(inode, start,
766 start + ram_size - 1, 0);
769 cur_alloc_size = ins.offset;
770 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771 ram_size, cur_alloc_size, 0);
774 if (disk_num_bytes < cur_alloc_size) {
775 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
779 /* we're not doing compressed IO, don't unlock the first
780 * page (which the caller expects to stay locked), don't
781 * clear any dirty bits and don't set any writeback bits
783 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
784 start, start + ram_size - 1,
785 locked_page, unlock, 1,
787 disk_num_bytes -= cur_alloc_size;
788 num_bytes -= cur_alloc_size;
789 alloc_hint = ins.objectid + ins.offset;
790 start += cur_alloc_size;
794 btrfs_end_transaction(trans, root);
800 * work queue call back to started compression on a file and pages
802 static noinline void async_cow_start(struct btrfs_work *work)
804 struct async_cow *async_cow;
806 async_cow = container_of(work, struct async_cow, work);
808 compress_file_range(async_cow->inode, async_cow->locked_page,
809 async_cow->start, async_cow->end, async_cow,
812 async_cow->inode = NULL;
816 * work queue call back to submit previously compressed pages
818 static noinline void async_cow_submit(struct btrfs_work *work)
820 struct async_cow *async_cow;
821 struct btrfs_root *root;
822 unsigned long nr_pages;
824 async_cow = container_of(work, struct async_cow, work);
826 root = async_cow->root;
827 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
830 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
832 if (atomic_read(&root->fs_info->async_delalloc_pages) <
834 waitqueue_active(&root->fs_info->async_submit_wait))
835 wake_up(&root->fs_info->async_submit_wait);
837 if (async_cow->inode) {
838 submit_compressed_extents(async_cow->inode, async_cow);
842 static noinline void async_cow_free(struct btrfs_work *work)
844 struct async_cow *async_cow;
845 async_cow = container_of(work, struct async_cow, work);
849 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
850 u64 start, u64 end, int *page_started,
851 unsigned long *nr_written)
853 struct async_cow *async_cow;
854 struct btrfs_root *root = BTRFS_I(inode)->root;
855 unsigned long nr_pages;
857 int limit = 10 * 1024 * 1042;
859 if (!btrfs_test_opt(root, COMPRESS)) {
860 return cow_file_range(inode, locked_page, start, end,
861 page_started, nr_written, 1);
864 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
865 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
867 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
868 async_cow->inode = inode;
869 async_cow->root = root;
870 async_cow->locked_page = locked_page;
871 async_cow->start = start;
873 if (btrfs_test_flag(inode, NOCOMPRESS))
876 cur_end = min(end, start + 512 * 1024 - 1);
878 async_cow->end = cur_end;
879 INIT_LIST_HEAD(&async_cow->extents);
881 async_cow->work.func = async_cow_start;
882 async_cow->work.ordered_func = async_cow_submit;
883 async_cow->work.ordered_free = async_cow_free;
884 async_cow->work.flags = 0;
886 while(atomic_read(&root->fs_info->async_submit_draining) &&
887 atomic_read(&root->fs_info->async_delalloc_pages)) {
888 wait_event(root->fs_info->async_submit_wait,
889 (atomic_read(&root->fs_info->async_delalloc_pages)
893 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
895 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
897 btrfs_queue_worker(&root->fs_info->delalloc_workers,
900 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
901 wait_event(root->fs_info->async_submit_wait,
902 (atomic_read(&root->fs_info->async_delalloc_pages) <
906 while(atomic_read(&root->fs_info->async_submit_draining) &&
907 atomic_read(&root->fs_info->async_delalloc_pages)) {
908 wait_event(root->fs_info->async_submit_wait,
909 (atomic_read(&root->fs_info->async_delalloc_pages) ==
913 *nr_written += nr_pages;
921 * when nowcow writeback call back. This checks for snapshots or COW copies
922 * of the extents that exist in the file, and COWs the file as required.
924 * If no cow copies or snapshots exist, we write directly to the existing
927 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
928 u64 start, u64 end, int *page_started, int force,
929 unsigned long *nr_written)
931 struct btrfs_root *root = BTRFS_I(inode)->root;
932 struct btrfs_trans_handle *trans;
933 struct extent_buffer *leaf;
934 struct btrfs_path *path;
935 struct btrfs_file_extent_item *fi;
936 struct btrfs_key found_key;
948 path = btrfs_alloc_path();
950 trans = btrfs_join_transaction(root, 1);
956 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
959 if (ret > 0 && path->slots[0] > 0 && check_prev) {
960 leaf = path->nodes[0];
961 btrfs_item_key_to_cpu(leaf, &found_key,
963 if (found_key.objectid == inode->i_ino &&
964 found_key.type == BTRFS_EXTENT_DATA_KEY)
969 leaf = path->nodes[0];
970 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
971 ret = btrfs_next_leaf(root, path);
976 leaf = path->nodes[0];
981 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
983 if (found_key.objectid > inode->i_ino ||
984 found_key.type > BTRFS_EXTENT_DATA_KEY ||
985 found_key.offset > end)
988 if (found_key.offset > cur_offset) {
989 extent_end = found_key.offset;
993 fi = btrfs_item_ptr(leaf, path->slots[0],
994 struct btrfs_file_extent_item);
995 extent_type = btrfs_file_extent_type(leaf, fi);
997 if (extent_type == BTRFS_FILE_EXTENT_REG ||
998 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
999 struct btrfs_block_group_cache *block_group;
1000 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1001 extent_end = found_key.offset +
1002 btrfs_file_extent_num_bytes(leaf, fi);
1003 if (extent_end <= start) {
1007 if (btrfs_file_extent_compression(leaf, fi) ||
1008 btrfs_file_extent_encryption(leaf, fi) ||
1009 btrfs_file_extent_other_encoding(leaf, fi))
1011 if (disk_bytenr == 0)
1013 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1015 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1017 block_group = btrfs_lookup_block_group(root->fs_info,
1019 if (!block_group || block_group->ro)
1021 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1023 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1024 extent_end = found_key.offset +
1025 btrfs_file_extent_inline_len(leaf, fi);
1026 extent_end = ALIGN(extent_end, root->sectorsize);
1031 if (extent_end <= start) {
1036 if (cow_start == (u64)-1)
1037 cow_start = cur_offset;
1038 cur_offset = extent_end;
1039 if (cur_offset > end)
1045 btrfs_release_path(root, path);
1046 if (cow_start != (u64)-1) {
1047 ret = cow_file_range(inode, locked_page, cow_start,
1048 found_key.offset - 1, page_started,
1051 cow_start = (u64)-1;
1054 disk_bytenr += cur_offset - found_key.offset;
1055 num_bytes = min(end + 1, extent_end) - cur_offset;
1056 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1057 struct extent_map *em;
1058 struct extent_map_tree *em_tree;
1059 em_tree = &BTRFS_I(inode)->extent_tree;
1060 em = alloc_extent_map(GFP_NOFS);
1061 em->start = cur_offset;
1062 em->orig_start = em->start;
1063 em->len = num_bytes;
1064 em->block_len = num_bytes;
1065 em->block_start = disk_bytenr;
1066 em->bdev = root->fs_info->fs_devices->latest_bdev;
1067 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1069 spin_lock(&em_tree->lock);
1070 ret = add_extent_mapping(em_tree, em);
1071 spin_unlock(&em_tree->lock);
1072 if (ret != -EEXIST) {
1073 free_extent_map(em);
1076 btrfs_drop_extent_cache(inode, em->start,
1077 em->start + em->len - 1, 0);
1079 type = BTRFS_ORDERED_PREALLOC;
1081 type = BTRFS_ORDERED_NOCOW;
1084 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1085 num_bytes, num_bytes, type);
1088 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1089 cur_offset, cur_offset + num_bytes - 1,
1090 locked_page, 1, 1, 1, 0, 0, 0);
1091 cur_offset = extent_end;
1092 if (cur_offset > end)
1095 btrfs_release_path(root, path);
1097 if (cur_offset <= end && cow_start == (u64)-1)
1098 cow_start = cur_offset;
1099 if (cow_start != (u64)-1) {
1100 ret = cow_file_range(inode, locked_page, cow_start, end,
1101 page_started, nr_written, 1);
1105 ret = btrfs_end_transaction(trans, root);
1107 btrfs_free_path(path);
1112 * extent_io.c call back to do delayed allocation processing
1114 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1115 u64 start, u64 end, int *page_started,
1116 unsigned long *nr_written)
1118 struct btrfs_root *root = BTRFS_I(inode)->root;
1121 if (btrfs_test_opt(root, NODATACOW) ||
1122 btrfs_test_flag(inode, NODATACOW))
1123 ret = run_delalloc_nocow(inode, locked_page, start, end,
1124 page_started, 0, nr_written);
1125 else if (btrfs_test_flag(inode, PREALLOC))
1126 ret = run_delalloc_nocow(inode, locked_page, start, end,
1127 page_started, 1, nr_written);
1129 ret = cow_file_range_async(inode, locked_page, start, end,
1130 page_started, nr_written);
1136 * extent_io.c set_bit_hook, used to track delayed allocation
1137 * bytes in this file, and to maintain the list of inodes that
1138 * have pending delalloc work to be done.
1140 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1141 unsigned long old, unsigned long bits)
1143 unsigned long flags;
1144 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1145 struct btrfs_root *root = BTRFS_I(inode)->root;
1146 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1147 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1148 root->fs_info->delalloc_bytes += end - start + 1;
1149 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1150 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1151 &root->fs_info->delalloc_inodes);
1153 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1159 * extent_io.c clear_bit_hook, see set_bit_hook for why
1161 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1162 unsigned long old, unsigned long bits)
1164 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1165 struct btrfs_root *root = BTRFS_I(inode)->root;
1166 unsigned long flags;
1168 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1169 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1170 printk("warning: delalloc account %Lu %Lu\n",
1171 end - start + 1, root->fs_info->delalloc_bytes);
1172 root->fs_info->delalloc_bytes = 0;
1173 BTRFS_I(inode)->delalloc_bytes = 0;
1175 root->fs_info->delalloc_bytes -= end - start + 1;
1176 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1178 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1179 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1180 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1182 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1188 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1189 * we don't create bios that span stripes or chunks
1191 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1192 size_t size, struct bio *bio,
1193 unsigned long bio_flags)
1195 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1196 struct btrfs_mapping_tree *map_tree;
1197 u64 logical = (u64)bio->bi_sector << 9;
1202 if (bio_flags & EXTENT_BIO_COMPRESSED)
1205 length = bio->bi_size;
1206 map_tree = &root->fs_info->mapping_tree;
1207 map_length = length;
1208 ret = btrfs_map_block(map_tree, READ, logical,
1209 &map_length, NULL, 0);
1211 if (map_length < length + size) {
1218 * in order to insert checksums into the metadata in large chunks,
1219 * we wait until bio submission time. All the pages in the bio are
1220 * checksummed and sums are attached onto the ordered extent record.
1222 * At IO completion time the cums attached on the ordered extent record
1223 * are inserted into the btree
1225 int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
1226 int mirror_num, unsigned long bio_flags)
1228 struct btrfs_root *root = BTRFS_I(inode)->root;
1231 ret = btrfs_csum_one_bio(root, inode, bio);
1237 * in order to insert checksums into the metadata in large chunks,
1238 * we wait until bio submission time. All the pages in the bio are
1239 * checksummed and sums are attached onto the ordered extent record.
1241 * At IO completion time the cums attached on the ordered extent record
1242 * are inserted into the btree
1244 int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1245 int mirror_num, unsigned long bio_flags)
1247 struct btrfs_root *root = BTRFS_I(inode)->root;
1248 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1252 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1253 * or reading the csums from the tree before a read
1255 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1256 int mirror_num, unsigned long bio_flags)
1258 struct btrfs_root *root = BTRFS_I(inode)->root;
1262 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1265 skip_sum = btrfs_test_opt(root, NODATASUM) ||
1266 btrfs_test_flag(inode, NODATASUM);
1268 if (!(rw & (1 << BIO_RW))) {
1270 if (bio_flags & EXTENT_BIO_COMPRESSED)
1271 return btrfs_submit_compressed_read(inode, bio,
1272 mirror_num, bio_flags);
1274 btrfs_lookup_bio_sums(root, inode, bio);
1276 } else if (!skip_sum) {
1277 /* we're doing a write, do the async checksumming */
1278 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1279 inode, rw, bio, mirror_num,
1280 bio_flags, __btrfs_submit_bio_start,
1281 __btrfs_submit_bio_done);
1285 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1289 * given a list of ordered sums record them in the inode. This happens
1290 * at IO completion time based on sums calculated at bio submission time.
1292 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1293 struct inode *inode, u64 file_offset,
1294 struct list_head *list)
1296 struct list_head *cur;
1297 struct btrfs_ordered_sum *sum;
1299 btrfs_set_trans_block_group(trans, inode);
1300 list_for_each(cur, list) {
1301 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1302 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
1308 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1310 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1313 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1317 /* see btrfs_writepage_start_hook for details on why this is required */
1318 struct btrfs_writepage_fixup {
1320 struct btrfs_work work;
1323 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1325 struct btrfs_writepage_fixup *fixup;
1326 struct btrfs_ordered_extent *ordered;
1328 struct inode *inode;
1332 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1336 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1337 ClearPageChecked(page);
1341 inode = page->mapping->host;
1342 page_start = page_offset(page);
1343 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1345 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1347 /* already ordered? We're done */
1348 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1349 EXTENT_ORDERED, 0)) {
1353 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1355 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1356 page_end, GFP_NOFS);
1358 btrfs_start_ordered_extent(inode, ordered, 1);
1362 btrfs_set_extent_delalloc(inode, page_start, page_end);
1363 ClearPageChecked(page);
1365 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1368 page_cache_release(page);
1372 * There are a few paths in the higher layers of the kernel that directly
1373 * set the page dirty bit without asking the filesystem if it is a
1374 * good idea. This causes problems because we want to make sure COW
1375 * properly happens and the data=ordered rules are followed.
1377 * In our case any range that doesn't have the ORDERED bit set
1378 * hasn't been properly setup for IO. We kick off an async process
1379 * to fix it up. The async helper will wait for ordered extents, set
1380 * the delalloc bit and make it safe to write the page.
1382 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1384 struct inode *inode = page->mapping->host;
1385 struct btrfs_writepage_fixup *fixup;
1386 struct btrfs_root *root = BTRFS_I(inode)->root;
1389 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1394 if (PageChecked(page))
1397 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1401 SetPageChecked(page);
1402 page_cache_get(page);
1403 fixup->work.func = btrfs_writepage_fixup_worker;
1405 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1409 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1410 struct inode *inode, u64 file_pos,
1411 u64 disk_bytenr, u64 disk_num_bytes,
1412 u64 num_bytes, u64 ram_bytes,
1413 u8 compression, u8 encryption,
1414 u16 other_encoding, int extent_type)
1416 struct btrfs_root *root = BTRFS_I(inode)->root;
1417 struct btrfs_file_extent_item *fi;
1418 struct btrfs_path *path;
1419 struct extent_buffer *leaf;
1420 struct btrfs_key ins;
1424 path = btrfs_alloc_path();
1427 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1428 file_pos + num_bytes, file_pos, &hint);
1431 ins.objectid = inode->i_ino;
1432 ins.offset = file_pos;
1433 ins.type = BTRFS_EXTENT_DATA_KEY;
1434 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1436 leaf = path->nodes[0];
1437 fi = btrfs_item_ptr(leaf, path->slots[0],
1438 struct btrfs_file_extent_item);
1439 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1440 btrfs_set_file_extent_type(leaf, fi, extent_type);
1441 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1442 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1443 btrfs_set_file_extent_offset(leaf, fi, 0);
1444 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1445 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1446 btrfs_set_file_extent_compression(leaf, fi, compression);
1447 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1448 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1449 btrfs_mark_buffer_dirty(leaf);
1451 inode_add_bytes(inode, num_bytes);
1452 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1454 ins.objectid = disk_bytenr;
1455 ins.offset = disk_num_bytes;
1456 ins.type = BTRFS_EXTENT_ITEM_KEY;
1457 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1458 root->root_key.objectid,
1459 trans->transid, inode->i_ino, &ins);
1462 btrfs_free_path(path);
1466 /* as ordered data IO finishes, this gets called so we can finish
1467 * an ordered extent if the range of bytes in the file it covers are
1470 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1472 struct btrfs_root *root = BTRFS_I(inode)->root;
1473 struct btrfs_trans_handle *trans;
1474 struct btrfs_ordered_extent *ordered_extent;
1475 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1479 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1483 trans = btrfs_join_transaction(root, 1);
1485 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1486 BUG_ON(!ordered_extent);
1487 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1490 lock_extent(io_tree, ordered_extent->file_offset,
1491 ordered_extent->file_offset + ordered_extent->len - 1,
1494 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1496 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1498 ret = btrfs_mark_extent_written(trans, root, inode,
1499 ordered_extent->file_offset,
1500 ordered_extent->file_offset +
1501 ordered_extent->len);
1504 ret = insert_reserved_file_extent(trans, inode,
1505 ordered_extent->file_offset,
1506 ordered_extent->start,
1507 ordered_extent->disk_len,
1508 ordered_extent->len,
1509 ordered_extent->len,
1511 BTRFS_FILE_EXTENT_REG);
1514 unlock_extent(io_tree, ordered_extent->file_offset,
1515 ordered_extent->file_offset + ordered_extent->len - 1,
1518 add_pending_csums(trans, inode, ordered_extent->file_offset,
1519 &ordered_extent->list);
1521 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1522 btrfs_ordered_update_i_size(inode, ordered_extent);
1523 btrfs_update_inode(trans, root, inode);
1524 btrfs_remove_ordered_extent(inode, ordered_extent);
1525 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1528 btrfs_put_ordered_extent(ordered_extent);
1529 /* once for the tree */
1530 btrfs_put_ordered_extent(ordered_extent);
1532 btrfs_end_transaction(trans, root);
1536 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1537 struct extent_state *state, int uptodate)
1539 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1543 * When IO fails, either with EIO or csum verification fails, we
1544 * try other mirrors that might have a good copy of the data. This
1545 * io_failure_record is used to record state as we go through all the
1546 * mirrors. If another mirror has good data, the page is set up to date
1547 * and things continue. If a good mirror can't be found, the original
1548 * bio end_io callback is called to indicate things have failed.
1550 struct io_failure_record {
1558 int btrfs_io_failed_hook(struct bio *failed_bio,
1559 struct page *page, u64 start, u64 end,
1560 struct extent_state *state)
1562 struct io_failure_record *failrec = NULL;
1564 struct extent_map *em;
1565 struct inode *inode = page->mapping->host;
1566 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1567 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1573 unsigned long bio_flags = 0;
1575 ret = get_state_private(failure_tree, start, &private);
1577 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1580 failrec->start = start;
1581 failrec->len = end - start + 1;
1582 failrec->last_mirror = 0;
1584 spin_lock(&em_tree->lock);
1585 em = lookup_extent_mapping(em_tree, start, failrec->len);
1586 if (em->start > start || em->start + em->len < start) {
1587 free_extent_map(em);
1590 spin_unlock(&em_tree->lock);
1592 if (!em || IS_ERR(em)) {
1596 logical = start - em->start;
1597 logical = em->block_start + logical;
1598 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1599 bio_flags = EXTENT_BIO_COMPRESSED;
1600 failrec->logical = logical;
1601 free_extent_map(em);
1602 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1603 EXTENT_DIRTY, GFP_NOFS);
1604 set_state_private(failure_tree, start,
1605 (u64)(unsigned long)failrec);
1607 failrec = (struct io_failure_record *)(unsigned long)private;
1609 num_copies = btrfs_num_copies(
1610 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1611 failrec->logical, failrec->len);
1612 failrec->last_mirror++;
1614 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1615 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1618 if (state && state->start != failrec->start)
1620 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1622 if (!state || failrec->last_mirror > num_copies) {
1623 set_state_private(failure_tree, failrec->start, 0);
1624 clear_extent_bits(failure_tree, failrec->start,
1625 failrec->start + failrec->len - 1,
1626 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1630 bio = bio_alloc(GFP_NOFS, 1);
1631 bio->bi_private = state;
1632 bio->bi_end_io = failed_bio->bi_end_io;
1633 bio->bi_sector = failrec->logical >> 9;
1634 bio->bi_bdev = failed_bio->bi_bdev;
1636 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1637 if (failed_bio->bi_rw & (1 << BIO_RW))
1642 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1643 failrec->last_mirror,
1649 * each time an IO finishes, we do a fast check in the IO failure tree
1650 * to see if we need to process or clean up an io_failure_record
1652 int btrfs_clean_io_failures(struct inode *inode, u64 start)
1655 u64 private_failure;
1656 struct io_failure_record *failure;
1660 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1661 (u64)-1, 1, EXTENT_DIRTY)) {
1662 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1663 start, &private_failure);
1665 failure = (struct io_failure_record *)(unsigned long)
1667 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1669 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1671 failure->start + failure->len - 1,
1672 EXTENT_DIRTY | EXTENT_LOCKED,
1681 * when reads are done, we need to check csums to verify the data is correct
1682 * if there's a match, we allow the bio to finish. If not, we go through
1683 * the io_failure_record routines to find good copies
1685 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1686 struct extent_state *state)
1688 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1689 struct inode *inode = page->mapping->host;
1690 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1692 u64 private = ~(u32)0;
1694 struct btrfs_root *root = BTRFS_I(inode)->root;
1696 unsigned long flags;
1698 if (btrfs_test_opt(root, NODATASUM) ||
1699 btrfs_test_flag(inode, NODATASUM))
1701 if (state && state->start == start) {
1702 private = state->private;
1705 ret = get_state_private(io_tree, start, &private);
1707 local_irq_save(flags);
1708 kaddr = kmap_atomic(page, KM_IRQ0);
1712 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1713 btrfs_csum_final(csum, (char *)&csum);
1714 if (csum != private) {
1717 kunmap_atomic(kaddr, KM_IRQ0);
1718 local_irq_restore(flags);
1720 /* if the io failure tree for this inode is non-empty,
1721 * check to see if we've recovered from a failed IO
1723 btrfs_clean_io_failures(inode, start);
1727 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1728 page->mapping->host->i_ino, (unsigned long long)start, csum,
1730 memset(kaddr + offset, 1, end - start + 1);
1731 flush_dcache_page(page);
1732 kunmap_atomic(kaddr, KM_IRQ0);
1733 local_irq_restore(flags);
1740 * This creates an orphan entry for the given inode in case something goes
1741 * wrong in the middle of an unlink/truncate.
1743 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1745 struct btrfs_root *root = BTRFS_I(inode)->root;
1748 spin_lock(&root->list_lock);
1750 /* already on the orphan list, we're good */
1751 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1752 spin_unlock(&root->list_lock);
1756 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1758 spin_unlock(&root->list_lock);
1761 * insert an orphan item to track this unlinked/truncated file
1763 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1769 * We have done the truncate/delete so we can go ahead and remove the orphan
1770 * item for this particular inode.
1772 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1774 struct btrfs_root *root = BTRFS_I(inode)->root;
1777 spin_lock(&root->list_lock);
1779 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1780 spin_unlock(&root->list_lock);
1784 list_del_init(&BTRFS_I(inode)->i_orphan);
1786 spin_unlock(&root->list_lock);
1790 spin_unlock(&root->list_lock);
1792 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1798 * this cleans up any orphans that may be left on the list from the last use
1801 void btrfs_orphan_cleanup(struct btrfs_root *root)
1803 struct btrfs_path *path;
1804 struct extent_buffer *leaf;
1805 struct btrfs_item *item;
1806 struct btrfs_key key, found_key;
1807 struct btrfs_trans_handle *trans;
1808 struct inode *inode;
1809 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1811 path = btrfs_alloc_path();
1816 key.objectid = BTRFS_ORPHAN_OBJECTID;
1817 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1818 key.offset = (u64)-1;
1822 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1824 printk(KERN_ERR "Error searching slot for orphan: %d"
1830 * if ret == 0 means we found what we were searching for, which
1831 * is weird, but possible, so only screw with path if we didnt
1832 * find the key and see if we have stuff that matches
1835 if (path->slots[0] == 0)
1840 /* pull out the item */
1841 leaf = path->nodes[0];
1842 item = btrfs_item_nr(leaf, path->slots[0]);
1843 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1845 /* make sure the item matches what we want */
1846 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1848 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1851 /* release the path since we're done with it */
1852 btrfs_release_path(root, path);
1855 * this is where we are basically btrfs_lookup, without the
1856 * crossing root thing. we store the inode number in the
1857 * offset of the orphan item.
1859 inode = btrfs_iget_locked(root->fs_info->sb,
1860 found_key.offset, root);
1864 if (inode->i_state & I_NEW) {
1865 BTRFS_I(inode)->root = root;
1867 /* have to set the location manually */
1868 BTRFS_I(inode)->location.objectid = inode->i_ino;
1869 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1870 BTRFS_I(inode)->location.offset = 0;
1872 btrfs_read_locked_inode(inode);
1873 unlock_new_inode(inode);
1877 * add this inode to the orphan list so btrfs_orphan_del does
1878 * the proper thing when we hit it
1880 spin_lock(&root->list_lock);
1881 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1882 spin_unlock(&root->list_lock);
1885 * if this is a bad inode, means we actually succeeded in
1886 * removing the inode, but not the orphan record, which means
1887 * we need to manually delete the orphan since iput will just
1888 * do a destroy_inode
1890 if (is_bad_inode(inode)) {
1891 trans = btrfs_start_transaction(root, 1);
1892 btrfs_orphan_del(trans, inode);
1893 btrfs_end_transaction(trans, root);
1898 /* if we have links, this was a truncate, lets do that */
1899 if (inode->i_nlink) {
1901 btrfs_truncate(inode);
1906 /* this will do delete_inode and everything for us */
1911 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1913 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1915 btrfs_free_path(path);
1919 * read an inode from the btree into the in-memory inode
1921 void btrfs_read_locked_inode(struct inode *inode)
1923 struct btrfs_path *path;
1924 struct extent_buffer *leaf;
1925 struct btrfs_inode_item *inode_item;
1926 struct btrfs_timespec *tspec;
1927 struct btrfs_root *root = BTRFS_I(inode)->root;
1928 struct btrfs_key location;
1929 u64 alloc_group_block;
1933 path = btrfs_alloc_path();
1935 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1937 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1941 leaf = path->nodes[0];
1942 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1943 struct btrfs_inode_item);
1945 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1946 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1947 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1948 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1949 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1951 tspec = btrfs_inode_atime(inode_item);
1952 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1953 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1955 tspec = btrfs_inode_mtime(inode_item);
1956 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1957 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1959 tspec = btrfs_inode_ctime(inode_item);
1960 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1961 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1963 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1964 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1965 inode->i_generation = BTRFS_I(inode)->generation;
1967 rdev = btrfs_inode_rdev(leaf, inode_item);
1969 BTRFS_I(inode)->index_cnt = (u64)-1;
1971 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1972 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1974 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1975 if (!BTRFS_I(inode)->block_group) {
1976 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1978 BTRFS_BLOCK_GROUP_METADATA, 0);
1980 btrfs_free_path(path);
1983 switch (inode->i_mode & S_IFMT) {
1985 inode->i_mapping->a_ops = &btrfs_aops;
1986 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1987 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1988 inode->i_fop = &btrfs_file_operations;
1989 inode->i_op = &btrfs_file_inode_operations;
1992 inode->i_fop = &btrfs_dir_file_operations;
1993 if (root == root->fs_info->tree_root)
1994 inode->i_op = &btrfs_dir_ro_inode_operations;
1996 inode->i_op = &btrfs_dir_inode_operations;
1999 inode->i_op = &btrfs_symlink_inode_operations;
2000 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2001 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2004 init_special_inode(inode, inode->i_mode, rdev);
2010 btrfs_free_path(path);
2011 make_bad_inode(inode);
2015 * given a leaf and an inode, copy the inode fields into the leaf
2017 static void fill_inode_item(struct btrfs_trans_handle *trans,
2018 struct extent_buffer *leaf,
2019 struct btrfs_inode_item *item,
2020 struct inode *inode)
2022 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2023 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2024 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2025 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2026 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2028 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2029 inode->i_atime.tv_sec);
2030 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2031 inode->i_atime.tv_nsec);
2033 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2034 inode->i_mtime.tv_sec);
2035 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2036 inode->i_mtime.tv_nsec);
2038 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2039 inode->i_ctime.tv_sec);
2040 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2041 inode->i_ctime.tv_nsec);
2043 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2044 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2045 btrfs_set_inode_transid(leaf, item, trans->transid);
2046 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2047 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2048 btrfs_set_inode_block_group(leaf, item,
2049 BTRFS_I(inode)->block_group->key.objectid);
2053 * copy everything in the in-memory inode into the btree.
2055 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
2056 struct btrfs_root *root,
2057 struct inode *inode)
2059 struct btrfs_inode_item *inode_item;
2060 struct btrfs_path *path;
2061 struct extent_buffer *leaf;
2064 path = btrfs_alloc_path();
2066 ret = btrfs_lookup_inode(trans, root, path,
2067 &BTRFS_I(inode)->location, 1);
2074 leaf = path->nodes[0];
2075 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2076 struct btrfs_inode_item);
2078 fill_inode_item(trans, leaf, inode_item, inode);
2079 btrfs_mark_buffer_dirty(leaf);
2080 btrfs_set_inode_last_trans(trans, inode);
2083 btrfs_free_path(path);
2089 * unlink helper that gets used here in inode.c and in the tree logging
2090 * recovery code. It remove a link in a directory with a given name, and
2091 * also drops the back refs in the inode to the directory
2093 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2094 struct btrfs_root *root,
2095 struct inode *dir, struct inode *inode,
2096 const char *name, int name_len)
2098 struct btrfs_path *path;
2100 struct extent_buffer *leaf;
2101 struct btrfs_dir_item *di;
2102 struct btrfs_key key;
2105 path = btrfs_alloc_path();
2111 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2112 name, name_len, -1);
2121 leaf = path->nodes[0];
2122 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2123 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2126 btrfs_release_path(root, path);
2128 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2130 dir->i_ino, &index);
2132 printk("failed to delete reference to %.*s, "
2133 "inode %lu parent %lu\n", name_len, name,
2134 inode->i_ino, dir->i_ino);
2138 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2139 index, name, name_len, -1);
2148 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2149 btrfs_release_path(root, path);
2151 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2153 BUG_ON(ret != 0 && ret != -ENOENT);
2155 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2157 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2161 btrfs_free_path(path);
2165 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2166 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2167 btrfs_update_inode(trans, root, dir);
2168 btrfs_drop_nlink(inode);
2169 ret = btrfs_update_inode(trans, root, inode);
2170 dir->i_sb->s_dirt = 1;
2175 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2177 struct btrfs_root *root;
2178 struct btrfs_trans_handle *trans;
2179 struct inode *inode = dentry->d_inode;
2181 unsigned long nr = 0;
2183 root = BTRFS_I(dir)->root;
2185 ret = btrfs_check_free_space(root, 1, 1);
2189 trans = btrfs_start_transaction(root, 1);
2191 btrfs_set_trans_block_group(trans, dir);
2192 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2193 dentry->d_name.name, dentry->d_name.len);
2195 if (inode->i_nlink == 0)
2196 ret = btrfs_orphan_add(trans, inode);
2198 nr = trans->blocks_used;
2200 btrfs_end_transaction_throttle(trans, root);
2202 btrfs_btree_balance_dirty(root, nr);
2206 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2208 struct inode *inode = dentry->d_inode;
2211 struct btrfs_root *root = BTRFS_I(dir)->root;
2212 struct btrfs_trans_handle *trans;
2213 unsigned long nr = 0;
2215 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2219 ret = btrfs_check_free_space(root, 1, 1);
2223 trans = btrfs_start_transaction(root, 1);
2224 btrfs_set_trans_block_group(trans, dir);
2226 err = btrfs_orphan_add(trans, inode);
2230 /* now the directory is empty */
2231 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2232 dentry->d_name.name, dentry->d_name.len);
2234 btrfs_i_size_write(inode, 0);
2238 nr = trans->blocks_used;
2239 ret = btrfs_end_transaction_throttle(trans, root);
2241 btrfs_btree_balance_dirty(root, nr);
2249 * when truncating bytes in a file, it is possible to avoid reading
2250 * the leaves that contain only checksum items. This can be the
2251 * majority of the IO required to delete a large file, but it must
2252 * be done carefully.
2254 * The keys in the level just above the leaves are checked to make sure
2255 * the lowest key in a given leaf is a csum key, and starts at an offset
2256 * after the new size.
2258 * Then the key for the next leaf is checked to make sure it also has
2259 * a checksum item for the same file. If it does, we know our target leaf
2260 * contains only checksum items, and it can be safely freed without reading
2263 * This is just an optimization targeted at large files. It may do
2264 * nothing. It will return 0 unless things went badly.
2266 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2267 struct btrfs_root *root,
2268 struct btrfs_path *path,
2269 struct inode *inode, u64 new_size)
2271 struct btrfs_key key;
2274 struct btrfs_key found_key;
2275 struct btrfs_key other_key;
2276 struct btrfs_leaf_ref *ref;
2280 path->lowest_level = 1;
2281 key.objectid = inode->i_ino;
2282 key.type = BTRFS_CSUM_ITEM_KEY;
2283 key.offset = new_size;
2285 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2289 if (path->nodes[1] == NULL) {
2294 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2295 nritems = btrfs_header_nritems(path->nodes[1]);
2300 if (path->slots[1] >= nritems)
2303 /* did we find a key greater than anything we want to delete? */
2304 if (found_key.objectid > inode->i_ino ||
2305 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2308 /* we check the next key in the node to make sure the leave contains
2309 * only checksum items. This comparison doesn't work if our
2310 * leaf is the last one in the node
2312 if (path->slots[1] + 1 >= nritems) {
2314 /* search forward from the last key in the node, this
2315 * will bring us into the next node in the tree
2317 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2319 /* unlikely, but we inc below, so check to be safe */
2320 if (found_key.offset == (u64)-1)
2323 /* search_forward needs a path with locks held, do the
2324 * search again for the original key. It is possible
2325 * this will race with a balance and return a path that
2326 * we could modify, but this drop is just an optimization
2327 * and is allowed to miss some leaves.
2329 btrfs_release_path(root, path);
2332 /* setup a max key for search_forward */
2333 other_key.offset = (u64)-1;
2334 other_key.type = key.type;
2335 other_key.objectid = key.objectid;
2337 path->keep_locks = 1;
2338 ret = btrfs_search_forward(root, &found_key, &other_key,
2340 path->keep_locks = 0;
2341 if (ret || found_key.objectid != key.objectid ||
2342 found_key.type != key.type) {
2347 key.offset = found_key.offset;
2348 btrfs_release_path(root, path);
2353 /* we know there's one more slot after us in the tree,
2354 * read that key so we can verify it is also a checksum item
2356 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2358 if (found_key.objectid < inode->i_ino)
2361 if (found_key.type != key.type || found_key.offset < new_size)
2365 * if the key for the next leaf isn't a csum key from this objectid,
2366 * we can't be sure there aren't good items inside this leaf.
2369 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2372 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2373 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2375 * it is safe to delete this leaf, it contains only
2376 * csum items from this inode at an offset >= new_size
2378 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2381 if (root->ref_cows && leaf_gen < trans->transid) {
2382 ref = btrfs_alloc_leaf_ref(root, 0);
2384 ref->root_gen = root->root_key.offset;
2385 ref->bytenr = leaf_start;
2387 ref->generation = leaf_gen;
2390 ret = btrfs_add_leaf_ref(root, ref, 0);
2392 btrfs_free_leaf_ref(root, ref);
2398 btrfs_release_path(root, path);
2400 if (other_key.objectid == inode->i_ino &&
2401 other_key.type == key.type && other_key.offset > key.offset) {
2402 key.offset = other_key.offset;
2408 /* fixup any changes we've made to the path */
2409 path->lowest_level = 0;
2410 path->keep_locks = 0;
2411 btrfs_release_path(root, path);
2416 * this can truncate away extent items, csum items and directory items.
2417 * It starts at a high offset and removes keys until it can't find
2418 * any higher than new_size
2420 * csum items that cross the new i_size are truncated to the new size
2423 * min_type is the minimum key type to truncate down to. If set to 0, this
2424 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2426 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2427 struct btrfs_root *root,
2428 struct inode *inode,
2429 u64 new_size, u32 min_type)
2432 struct btrfs_path *path;
2433 struct btrfs_key key;
2434 struct btrfs_key found_key;
2436 struct extent_buffer *leaf;
2437 struct btrfs_file_extent_item *fi;
2438 u64 extent_start = 0;
2439 u64 extent_num_bytes = 0;
2445 int pending_del_nr = 0;
2446 int pending_del_slot = 0;
2447 int extent_type = -1;
2449 u64 mask = root->sectorsize - 1;
2452 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2453 path = btrfs_alloc_path();
2457 /* FIXME, add redo link to tree so we don't leak on crash */
2458 key.objectid = inode->i_ino;
2459 key.offset = (u64)-1;
2462 btrfs_init_path(path);
2464 ret = drop_csum_leaves(trans, root, path, inode, new_size);
2468 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2473 /* there are no items in the tree for us to truncate, we're
2476 if (path->slots[0] == 0) {
2485 leaf = path->nodes[0];
2486 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2487 found_type = btrfs_key_type(&found_key);
2490 if (found_key.objectid != inode->i_ino)
2493 if (found_type < min_type)
2496 item_end = found_key.offset;
2497 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2498 fi = btrfs_item_ptr(leaf, path->slots[0],
2499 struct btrfs_file_extent_item);
2500 extent_type = btrfs_file_extent_type(leaf, fi);
2501 encoding = btrfs_file_extent_compression(leaf, fi);
2502 encoding |= btrfs_file_extent_encryption(leaf, fi);
2503 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2505 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2507 btrfs_file_extent_num_bytes(leaf, fi);
2508 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2509 item_end += btrfs_file_extent_inline_len(leaf,
2514 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2515 ret = btrfs_csum_truncate(trans, root, path,
2519 if (item_end < new_size) {
2520 if (found_type == BTRFS_DIR_ITEM_KEY) {
2521 found_type = BTRFS_INODE_ITEM_KEY;
2522 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2523 found_type = BTRFS_CSUM_ITEM_KEY;
2524 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2525 found_type = BTRFS_XATTR_ITEM_KEY;
2526 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2527 found_type = BTRFS_INODE_REF_KEY;
2528 } else if (found_type) {
2533 btrfs_set_key_type(&key, found_type);
2536 if (found_key.offset >= new_size)
2542 /* FIXME, shrink the extent if the ref count is only 1 */
2543 if (found_type != BTRFS_EXTENT_DATA_KEY)
2546 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2548 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2549 if (!del_item && !encoding) {
2550 u64 orig_num_bytes =
2551 btrfs_file_extent_num_bytes(leaf, fi);
2552 extent_num_bytes = new_size -
2553 found_key.offset + root->sectorsize - 1;
2554 extent_num_bytes = extent_num_bytes &
2555 ~((u64)root->sectorsize - 1);
2556 btrfs_set_file_extent_num_bytes(leaf, fi,
2558 num_dec = (orig_num_bytes -
2560 if (root->ref_cows && extent_start != 0)
2561 inode_sub_bytes(inode, num_dec);
2562 btrfs_mark_buffer_dirty(leaf);
2565 btrfs_file_extent_disk_num_bytes(leaf,
2567 /* FIXME blocksize != 4096 */
2568 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2569 if (extent_start != 0) {
2572 inode_sub_bytes(inode, num_dec);
2574 root_gen = btrfs_header_generation(leaf);
2575 root_owner = btrfs_header_owner(leaf);
2577 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2579 * we can't truncate inline items that have had
2583 btrfs_file_extent_compression(leaf, fi) == 0 &&
2584 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2585 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2586 u32 size = new_size - found_key.offset;
2588 if (root->ref_cows) {
2589 inode_sub_bytes(inode, item_end + 1 -
2593 btrfs_file_extent_calc_inline_size(size);
2594 ret = btrfs_truncate_item(trans, root, path,
2597 } else if (root->ref_cows) {
2598 inode_sub_bytes(inode, item_end + 1 -
2604 if (!pending_del_nr) {
2605 /* no pending yet, add ourselves */
2606 pending_del_slot = path->slots[0];
2608 } else if (pending_del_nr &&
2609 path->slots[0] + 1 == pending_del_slot) {
2610 /* hop on the pending chunk */
2612 pending_del_slot = path->slots[0];
2614 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2620 ret = btrfs_free_extent(trans, root, extent_start,
2622 leaf->start, root_owner,
2623 root_gen, inode->i_ino, 0);
2627 if (path->slots[0] == 0) {
2630 btrfs_release_path(root, path);
2635 if (pending_del_nr &&
2636 path->slots[0] + 1 != pending_del_slot) {
2637 struct btrfs_key debug;
2639 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2641 ret = btrfs_del_items(trans, root, path,
2646 btrfs_release_path(root, path);
2652 if (pending_del_nr) {
2653 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2656 btrfs_free_path(path);
2657 inode->i_sb->s_dirt = 1;
2662 * taken from block_truncate_page, but does cow as it zeros out
2663 * any bytes left in the last page in the file.
2665 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2667 struct inode *inode = mapping->host;
2668 struct btrfs_root *root = BTRFS_I(inode)->root;
2669 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2670 struct btrfs_ordered_extent *ordered;
2672 u32 blocksize = root->sectorsize;
2673 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2674 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2680 if ((offset & (blocksize - 1)) == 0)
2685 page = grab_cache_page(mapping, index);
2689 page_start = page_offset(page);
2690 page_end = page_start + PAGE_CACHE_SIZE - 1;
2692 if (!PageUptodate(page)) {
2693 ret = btrfs_readpage(NULL, page);
2695 if (page->mapping != mapping) {
2697 page_cache_release(page);
2700 if (!PageUptodate(page)) {
2705 wait_on_page_writeback(page);
2707 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2708 set_page_extent_mapped(page);
2710 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2712 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2714 page_cache_release(page);
2715 btrfs_start_ordered_extent(inode, ordered, 1);
2716 btrfs_put_ordered_extent(ordered);
2720 btrfs_set_extent_delalloc(inode, page_start, page_end);
2722 if (offset != PAGE_CACHE_SIZE) {
2724 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2725 flush_dcache_page(page);
2728 ClearPageChecked(page);
2729 set_page_dirty(page);
2730 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2734 page_cache_release(page);
2739 int btrfs_cont_expand(struct inode *inode, loff_t size)
2741 struct btrfs_trans_handle *trans;
2742 struct btrfs_root *root = BTRFS_I(inode)->root;
2743 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2744 struct extent_map *em;
2745 u64 mask = root->sectorsize - 1;
2746 u64 hole_start = (inode->i_size + mask) & ~mask;
2747 u64 block_end = (size + mask) & ~mask;
2753 if (size <= hole_start)
2756 err = btrfs_check_free_space(root, 1, 0);
2760 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2763 struct btrfs_ordered_extent *ordered;
2764 btrfs_wait_ordered_range(inode, hole_start,
2765 block_end - hole_start);
2766 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2767 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2770 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2771 btrfs_put_ordered_extent(ordered);
2774 trans = btrfs_start_transaction(root, 1);
2775 btrfs_set_trans_block_group(trans, inode);
2777 cur_offset = hole_start;
2779 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2780 block_end - cur_offset, 0);
2781 BUG_ON(IS_ERR(em) || !em);
2782 last_byte = min(extent_map_end(em), block_end);
2783 last_byte = (last_byte + mask) & ~mask;
2784 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2786 hole_size = last_byte - cur_offset;
2787 err = btrfs_drop_extents(trans, root, inode,
2789 cur_offset + hole_size,
2790 cur_offset, &hint_byte);
2793 err = btrfs_insert_file_extent(trans, root,
2794 inode->i_ino, cur_offset, 0,
2795 0, hole_size, 0, hole_size,
2797 btrfs_drop_extent_cache(inode, hole_start,
2800 free_extent_map(em);
2801 cur_offset = last_byte;
2802 if (err || cur_offset >= block_end)
2806 btrfs_end_transaction(trans, root);
2807 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2811 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2813 struct inode *inode = dentry->d_inode;
2816 err = inode_change_ok(inode, attr);
2820 if (S_ISREG(inode->i_mode) &&
2821 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2822 err = btrfs_cont_expand(inode, attr->ia_size);
2827 err = inode_setattr(inode, attr);
2829 if (!err && ((attr->ia_valid & ATTR_MODE)))
2830 err = btrfs_acl_chmod(inode);
2834 void btrfs_delete_inode(struct inode *inode)
2836 struct btrfs_trans_handle *trans;
2837 struct btrfs_root *root = BTRFS_I(inode)->root;
2841 truncate_inode_pages(&inode->i_data, 0);
2842 if (is_bad_inode(inode)) {
2843 btrfs_orphan_del(NULL, inode);
2846 btrfs_wait_ordered_range(inode, 0, (u64)-1);
2848 btrfs_i_size_write(inode, 0);
2849 trans = btrfs_start_transaction(root, 1);
2851 btrfs_set_trans_block_group(trans, inode);
2852 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2854 btrfs_orphan_del(NULL, inode);
2855 goto no_delete_lock;
2858 btrfs_orphan_del(trans, inode);
2860 nr = trans->blocks_used;
2863 btrfs_end_transaction(trans, root);
2864 btrfs_btree_balance_dirty(root, nr);
2868 nr = trans->blocks_used;
2869 btrfs_end_transaction(trans, root);
2870 btrfs_btree_balance_dirty(root, nr);
2876 * this returns the key found in the dir entry in the location pointer.
2877 * If no dir entries were found, location->objectid is 0.
2879 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2880 struct btrfs_key *location)
2882 const char *name = dentry->d_name.name;
2883 int namelen = dentry->d_name.len;
2884 struct btrfs_dir_item *di;
2885 struct btrfs_path *path;
2886 struct btrfs_root *root = BTRFS_I(dir)->root;
2889 path = btrfs_alloc_path();
2892 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2896 if (!di || IS_ERR(di)) {
2899 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2901 btrfs_free_path(path);
2904 location->objectid = 0;
2909 * when we hit a tree root in a directory, the btrfs part of the inode
2910 * needs to be changed to reflect the root directory of the tree root. This
2911 * is kind of like crossing a mount point.
2913 static int fixup_tree_root_location(struct btrfs_root *root,
2914 struct btrfs_key *location,
2915 struct btrfs_root **sub_root,
2916 struct dentry *dentry)
2918 struct btrfs_root_item *ri;
2920 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2922 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2925 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2926 dentry->d_name.name,
2927 dentry->d_name.len);
2928 if (IS_ERR(*sub_root))
2929 return PTR_ERR(*sub_root);
2931 ri = &(*sub_root)->root_item;
2932 location->objectid = btrfs_root_dirid(ri);
2933 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2934 location->offset = 0;
2939 static noinline void init_btrfs_i(struct inode *inode)
2941 struct btrfs_inode *bi = BTRFS_I(inode);
2944 bi->i_default_acl = NULL;
2948 bi->logged_trans = 0;
2949 bi->delalloc_bytes = 0;
2950 bi->disk_i_size = 0;
2952 bi->index_cnt = (u64)-1;
2953 bi->log_dirty_trans = 0;
2954 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2955 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2956 inode->i_mapping, GFP_NOFS);
2957 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2958 inode->i_mapping, GFP_NOFS);
2959 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2960 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2961 mutex_init(&BTRFS_I(inode)->csum_mutex);
2962 mutex_init(&BTRFS_I(inode)->extent_mutex);
2963 mutex_init(&BTRFS_I(inode)->log_mutex);
2966 static int btrfs_init_locked_inode(struct inode *inode, void *p)
2968 struct btrfs_iget_args *args = p;
2969 inode->i_ino = args->ino;
2970 init_btrfs_i(inode);
2971 BTRFS_I(inode)->root = args->root;
2975 static int btrfs_find_actor(struct inode *inode, void *opaque)
2977 struct btrfs_iget_args *args = opaque;
2978 return (args->ino == inode->i_ino &&
2979 args->root == BTRFS_I(inode)->root);
2982 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2983 struct btrfs_root *root, int wait)
2985 struct inode *inode;
2986 struct btrfs_iget_args args;
2987 args.ino = objectid;
2991 inode = ilookup5(s, objectid, btrfs_find_actor,
2994 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3000 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3001 struct btrfs_root *root)
3003 struct inode *inode;
3004 struct btrfs_iget_args args;
3005 args.ino = objectid;
3008 inode = iget5_locked(s, objectid, btrfs_find_actor,
3009 btrfs_init_locked_inode,
3014 /* Get an inode object given its location and corresponding root.
3015 * Returns in *is_new if the inode was read from disk
3017 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3018 struct btrfs_root *root, int *is_new)
3020 struct inode *inode;
3022 inode = btrfs_iget_locked(s, location->objectid, root);
3024 return ERR_PTR(-EACCES);
3026 if (inode->i_state & I_NEW) {
3027 BTRFS_I(inode)->root = root;
3028 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3029 btrfs_read_locked_inode(inode);
3030 unlock_new_inode(inode);
3041 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3042 struct nameidata *nd)
3044 struct inode * inode;
3045 struct btrfs_inode *bi = BTRFS_I(dir);
3046 struct btrfs_root *root = bi->root;
3047 struct btrfs_root *sub_root = root;
3048 struct btrfs_key location;
3051 if (dentry->d_name.len > BTRFS_NAME_LEN)
3052 return ERR_PTR(-ENAMETOOLONG);
3054 ret = btrfs_inode_by_name(dir, dentry, &location);
3057 return ERR_PTR(ret);
3060 if (location.objectid) {
3061 ret = fixup_tree_root_location(root, &location, &sub_root,
3064 return ERR_PTR(ret);
3066 return ERR_PTR(-ENOENT);
3067 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3069 return ERR_CAST(inode);
3071 /* the inode and parent dir are two different roots */
3072 if (new && root != sub_root) {
3074 sub_root->inode = inode;
3078 return d_splice_alias(inode, dentry);
3081 static unsigned char btrfs_filetype_table[] = {
3082 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3085 static int btrfs_real_readdir(struct file *filp, void *dirent,
3088 struct inode *inode = filp->f_dentry->d_inode;
3089 struct btrfs_root *root = BTRFS_I(inode)->root;
3090 struct btrfs_item *item;
3091 struct btrfs_dir_item *di;
3092 struct btrfs_key key;
3093 struct btrfs_key found_key;
3094 struct btrfs_path *path;
3097 struct extent_buffer *leaf;
3100 unsigned char d_type;
3105 int key_type = BTRFS_DIR_INDEX_KEY;
3110 /* FIXME, use a real flag for deciding about the key type */
3111 if (root->fs_info->tree_root == root)
3112 key_type = BTRFS_DIR_ITEM_KEY;
3114 /* special case for "." */
3115 if (filp->f_pos == 0) {
3116 over = filldir(dirent, ".", 1,
3123 /* special case for .., just use the back ref */
3124 if (filp->f_pos == 1) {
3125 u64 pino = parent_ino(filp->f_path.dentry);
3126 over = filldir(dirent, "..", 2,
3133 path = btrfs_alloc_path();
3136 btrfs_set_key_type(&key, key_type);
3137 key.offset = filp->f_pos;
3138 key.objectid = inode->i_ino;
3140 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3146 leaf = path->nodes[0];
3147 nritems = btrfs_header_nritems(leaf);
3148 slot = path->slots[0];
3149 if (advance || slot >= nritems) {
3150 if (slot >= nritems - 1) {
3151 ret = btrfs_next_leaf(root, path);
3154 leaf = path->nodes[0];
3155 nritems = btrfs_header_nritems(leaf);
3156 slot = path->slots[0];
3163 item = btrfs_item_nr(leaf, slot);
3164 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3166 if (found_key.objectid != key.objectid)
3168 if (btrfs_key_type(&found_key) != key_type)
3170 if (found_key.offset < filp->f_pos)
3173 filp->f_pos = found_key.offset;
3175 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3177 di_total = btrfs_item_size(leaf, item);
3179 while (di_cur < di_total) {
3180 struct btrfs_key location;
3182 name_len = btrfs_dir_name_len(leaf, di);
3183 if (name_len <= sizeof(tmp_name)) {
3184 name_ptr = tmp_name;
3186 name_ptr = kmalloc(name_len, GFP_NOFS);
3192 read_extent_buffer(leaf, name_ptr,
3193 (unsigned long)(di + 1), name_len);
3195 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3196 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3197 over = filldir(dirent, name_ptr, name_len,
3198 found_key.offset, location.objectid,
3201 if (name_ptr != tmp_name)
3207 di_len = btrfs_dir_name_len(leaf, di) +
3208 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3210 di = (struct btrfs_dir_item *)((char *)di + di_len);
3214 /* Reached end of directory/root. Bump pos past the last item. */
3215 if (key_type == BTRFS_DIR_INDEX_KEY)
3216 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3222 btrfs_free_path(path);
3226 int btrfs_write_inode(struct inode *inode, int wait)
3228 struct btrfs_root *root = BTRFS_I(inode)->root;
3229 struct btrfs_trans_handle *trans;
3232 if (root->fs_info->btree_inode == inode)
3236 trans = btrfs_join_transaction(root, 1);
3237 btrfs_set_trans_block_group(trans, inode);
3238 ret = btrfs_commit_transaction(trans, root);
3244 * This is somewhat expensive, updating the tree every time the
3245 * inode changes. But, it is most likely to find the inode in cache.
3246 * FIXME, needs more benchmarking...there are no reasons other than performance
3247 * to keep or drop this code.
3249 void btrfs_dirty_inode(struct inode *inode)
3251 struct btrfs_root *root = BTRFS_I(inode)->root;
3252 struct btrfs_trans_handle *trans;
3254 trans = btrfs_join_transaction(root, 1);
3255 btrfs_set_trans_block_group(trans, inode);
3256 btrfs_update_inode(trans, root, inode);
3257 btrfs_end_transaction(trans, root);
3261 * find the highest existing sequence number in a directory
3262 * and then set the in-memory index_cnt variable to reflect
3263 * free sequence numbers
3265 static int btrfs_set_inode_index_count(struct inode *inode)
3267 struct btrfs_root *root = BTRFS_I(inode)->root;
3268 struct btrfs_key key, found_key;
3269 struct btrfs_path *path;
3270 struct extent_buffer *leaf;
3273 key.objectid = inode->i_ino;
3274 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3275 key.offset = (u64)-1;
3277 path = btrfs_alloc_path();
3281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3284 /* FIXME: we should be able to handle this */
3290 * MAGIC NUMBER EXPLANATION:
3291 * since we search a directory based on f_pos we have to start at 2
3292 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3293 * else has to start at 2
3295 if (path->slots[0] == 0) {
3296 BTRFS_I(inode)->index_cnt = 2;
3302 leaf = path->nodes[0];
3303 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3305 if (found_key.objectid != inode->i_ino ||
3306 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3307 BTRFS_I(inode)->index_cnt = 2;
3311 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3313 btrfs_free_path(path);
3318 * helper to find a free sequence number in a given directory. This current
3319 * code is very simple, later versions will do smarter things in the btree
3321 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
3326 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3327 ret = btrfs_set_inode_index_count(dir);
3333 *index = BTRFS_I(dir)->index_cnt;
3334 BTRFS_I(dir)->index_cnt++;
3339 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3340 struct btrfs_root *root,
3342 const char *name, int name_len,
3345 struct btrfs_block_group_cache *group,
3346 int mode, u64 *index)
3348 struct inode *inode;
3349 struct btrfs_inode_item *inode_item;
3350 struct btrfs_block_group_cache *new_inode_group;
3351 struct btrfs_key *location;
3352 struct btrfs_path *path;
3353 struct btrfs_inode_ref *ref;
3354 struct btrfs_key key[2];
3360 path = btrfs_alloc_path();
3363 inode = new_inode(root->fs_info->sb);
3365 return ERR_PTR(-ENOMEM);
3368 ret = btrfs_set_inode_index(dir, inode, index);
3370 return ERR_PTR(ret);
3373 * index_cnt is ignored for everything but a dir,
3374 * btrfs_get_inode_index_count has an explanation for the magic
3377 init_btrfs_i(inode);
3378 BTRFS_I(inode)->index_cnt = 2;
3379 BTRFS_I(inode)->root = root;
3380 BTRFS_I(inode)->generation = trans->transid;
3386 new_inode_group = btrfs_find_block_group(root, group, 0,
3387 BTRFS_BLOCK_GROUP_METADATA, owner);
3388 if (!new_inode_group) {
3389 printk("find_block group failed\n");
3390 new_inode_group = group;
3392 BTRFS_I(inode)->block_group = new_inode_group;
3394 key[0].objectid = objectid;
3395 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3398 key[1].objectid = objectid;
3399 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3400 key[1].offset = ref_objectid;
3402 sizes[0] = sizeof(struct btrfs_inode_item);
3403 sizes[1] = name_len + sizeof(*ref);
3405 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3409 if (objectid > root->highest_inode)
3410 root->highest_inode = objectid;
3412 inode->i_uid = current->fsuid;
3413 inode->i_gid = current->fsgid;
3414 inode->i_mode = mode;
3415 inode->i_ino = objectid;
3416 inode_set_bytes(inode, 0);
3417 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3418 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3419 struct btrfs_inode_item);
3420 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3422 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3423 struct btrfs_inode_ref);
3424 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3425 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3426 ptr = (unsigned long)(ref + 1);
3427 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3429 btrfs_mark_buffer_dirty(path->nodes[0]);
3430 btrfs_free_path(path);
3432 location = &BTRFS_I(inode)->location;
3433 location->objectid = objectid;
3434 location->offset = 0;
3435 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3437 insert_inode_hash(inode);
3441 BTRFS_I(dir)->index_cnt--;
3442 btrfs_free_path(path);
3443 return ERR_PTR(ret);
3446 static inline u8 btrfs_inode_type(struct inode *inode)
3448 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3452 * utility function to add 'inode' into 'parent_inode' with
3453 * a give name and a given sequence number.
3454 * if 'add_backref' is true, also insert a backref from the
3455 * inode to the parent directory.
3457 int btrfs_add_link(struct btrfs_trans_handle *trans,
3458 struct inode *parent_inode, struct inode *inode,
3459 const char *name, int name_len, int add_backref, u64 index)
3462 struct btrfs_key key;
3463 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3465 key.objectid = inode->i_ino;
3466 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3469 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3470 parent_inode->i_ino,
3471 &key, btrfs_inode_type(inode),
3475 ret = btrfs_insert_inode_ref(trans, root,
3478 parent_inode->i_ino,
3481 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3483 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3484 ret = btrfs_update_inode(trans, root, parent_inode);
3489 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3490 struct dentry *dentry, struct inode *inode,
3491 int backref, u64 index)
3493 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3494 inode, dentry->d_name.name,
3495 dentry->d_name.len, backref, index);
3497 d_instantiate(dentry, inode);
3505 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3506 int mode, dev_t rdev)
3508 struct btrfs_trans_handle *trans;
3509 struct btrfs_root *root = BTRFS_I(dir)->root;
3510 struct inode *inode = NULL;
3514 unsigned long nr = 0;
3517 if (!new_valid_dev(rdev))
3520 err = btrfs_check_free_space(root, 1, 0);
3524 trans = btrfs_start_transaction(root, 1);
3525 btrfs_set_trans_block_group(trans, dir);
3527 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3533 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3535 dentry->d_parent->d_inode->i_ino, objectid,
3536 BTRFS_I(dir)->block_group, mode, &index);
3537 err = PTR_ERR(inode);
3541 err = btrfs_init_acl(inode, dir);
3547 btrfs_set_trans_block_group(trans, inode);
3548 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3552 inode->i_op = &btrfs_special_inode_operations;
3553 init_special_inode(inode, inode->i_mode, rdev);
3554 btrfs_update_inode(trans, root, inode);
3556 dir->i_sb->s_dirt = 1;
3557 btrfs_update_inode_block_group(trans, inode);
3558 btrfs_update_inode_block_group(trans, dir);
3560 nr = trans->blocks_used;
3561 btrfs_end_transaction_throttle(trans, root);
3564 inode_dec_link_count(inode);
3567 btrfs_btree_balance_dirty(root, nr);
3571 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3572 int mode, struct nameidata *nd)
3574 struct btrfs_trans_handle *trans;
3575 struct btrfs_root *root = BTRFS_I(dir)->root;
3576 struct inode *inode = NULL;
3579 unsigned long nr = 0;
3583 err = btrfs_check_free_space(root, 1, 0);
3586 trans = btrfs_start_transaction(root, 1);
3587 btrfs_set_trans_block_group(trans, dir);
3589 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3595 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3597 dentry->d_parent->d_inode->i_ino,
3598 objectid, BTRFS_I(dir)->block_group, mode,
3600 err = PTR_ERR(inode);
3604 err = btrfs_init_acl(inode, dir);
3610 btrfs_set_trans_block_group(trans, inode);
3611 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3615 inode->i_mapping->a_ops = &btrfs_aops;
3616 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3617 inode->i_fop = &btrfs_file_operations;
3618 inode->i_op = &btrfs_file_inode_operations;
3619 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3621 dir->i_sb->s_dirt = 1;
3622 btrfs_update_inode_block_group(trans, inode);
3623 btrfs_update_inode_block_group(trans, dir);
3625 nr = trans->blocks_used;
3626 btrfs_end_transaction_throttle(trans, root);
3629 inode_dec_link_count(inode);
3632 btrfs_btree_balance_dirty(root, nr);
3636 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3637 struct dentry *dentry)
3639 struct btrfs_trans_handle *trans;
3640 struct btrfs_root *root = BTRFS_I(dir)->root;
3641 struct inode *inode = old_dentry->d_inode;
3643 unsigned long nr = 0;
3647 if (inode->i_nlink == 0)
3650 btrfs_inc_nlink(inode);
3651 err = btrfs_check_free_space(root, 1, 0);
3654 err = btrfs_set_inode_index(dir, inode, &index);
3658 trans = btrfs_start_transaction(root, 1);
3660 btrfs_set_trans_block_group(trans, dir);
3661 atomic_inc(&inode->i_count);
3663 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3668 dir->i_sb->s_dirt = 1;
3669 btrfs_update_inode_block_group(trans, dir);
3670 err = btrfs_update_inode(trans, root, inode);
3675 nr = trans->blocks_used;
3676 btrfs_end_transaction_throttle(trans, root);
3679 inode_dec_link_count(inode);
3682 btrfs_btree_balance_dirty(root, nr);
3686 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3688 struct inode *inode = NULL;
3689 struct btrfs_trans_handle *trans;
3690 struct btrfs_root *root = BTRFS_I(dir)->root;
3692 int drop_on_err = 0;
3695 unsigned long nr = 1;
3697 err = btrfs_check_free_space(root, 1, 0);
3701 trans = btrfs_start_transaction(root, 1);
3702 btrfs_set_trans_block_group(trans, dir);
3704 if (IS_ERR(trans)) {
3705 err = PTR_ERR(trans);
3709 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3715 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3717 dentry->d_parent->d_inode->i_ino, objectid,
3718 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3720 if (IS_ERR(inode)) {
3721 err = PTR_ERR(inode);
3727 err = btrfs_init_acl(inode, dir);
3731 inode->i_op = &btrfs_dir_inode_operations;
3732 inode->i_fop = &btrfs_dir_file_operations;
3733 btrfs_set_trans_block_group(trans, inode);
3735 btrfs_i_size_write(inode, 0);
3736 err = btrfs_update_inode(trans, root, inode);
3740 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3741 inode, dentry->d_name.name,
3742 dentry->d_name.len, 0, index);
3746 d_instantiate(dentry, inode);
3748 dir->i_sb->s_dirt = 1;
3749 btrfs_update_inode_block_group(trans, inode);
3750 btrfs_update_inode_block_group(trans, dir);
3753 nr = trans->blocks_used;
3754 btrfs_end_transaction_throttle(trans, root);
3759 btrfs_btree_balance_dirty(root, nr);
3763 /* helper for btfs_get_extent. Given an existing extent in the tree,
3764 * and an extent that you want to insert, deal with overlap and insert
3765 * the new extent into the tree.
3767 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3768 struct extent_map *existing,
3769 struct extent_map *em,
3770 u64 map_start, u64 map_len)
3774 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3775 start_diff = map_start - em->start;
3776 em->start = map_start;
3778 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3779 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3780 em->block_start += start_diff;
3781 em->block_len -= start_diff;
3783 return add_extent_mapping(em_tree, em);
3786 static noinline int uncompress_inline(struct btrfs_path *path,
3787 struct inode *inode, struct page *page,
3788 size_t pg_offset, u64 extent_offset,
3789 struct btrfs_file_extent_item *item)
3792 struct extent_buffer *leaf = path->nodes[0];
3795 unsigned long inline_size;
3798 WARN_ON(pg_offset != 0);
3799 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3800 inline_size = btrfs_file_extent_inline_item_len(leaf,
3801 btrfs_item_nr(leaf, path->slots[0]));
3802 tmp = kmalloc(inline_size, GFP_NOFS);
3803 ptr = btrfs_file_extent_inline_start(item);
3805 read_extent_buffer(leaf, tmp, ptr, inline_size);
3807 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3808 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3809 inline_size, max_size);
3811 char *kaddr = kmap_atomic(page, KM_USER0);
3812 unsigned long copy_size = min_t(u64,
3813 PAGE_CACHE_SIZE - pg_offset,
3814 max_size - extent_offset);
3815 memset(kaddr + pg_offset, 0, copy_size);
3816 kunmap_atomic(kaddr, KM_USER0);
3823 * a bit scary, this does extent mapping from logical file offset to the disk.
3824 * the ugly parts come from merging extents from the disk with the
3825 * in-ram representation. This gets more complex because of the data=ordered code,
3826 * where the in-ram extents might be locked pending data=ordered completion.
3828 * This also copies inline extents directly into the page.
3830 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3831 size_t pg_offset, u64 start, u64 len,
3837 u64 extent_start = 0;
3839 u64 objectid = inode->i_ino;
3841 struct btrfs_path *path = NULL;
3842 struct btrfs_root *root = BTRFS_I(inode)->root;
3843 struct btrfs_file_extent_item *item;
3844 struct extent_buffer *leaf;
3845 struct btrfs_key found_key;
3846 struct extent_map *em = NULL;
3847 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3848 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3849 struct btrfs_trans_handle *trans = NULL;
3853 spin_lock(&em_tree->lock);
3854 em = lookup_extent_mapping(em_tree, start, len);
3856 em->bdev = root->fs_info->fs_devices->latest_bdev;
3857 spin_unlock(&em_tree->lock);
3860 if (em->start > start || em->start + em->len <= start)
3861 free_extent_map(em);
3862 else if (em->block_start == EXTENT_MAP_INLINE && page)
3863 free_extent_map(em);
3867 em = alloc_extent_map(GFP_NOFS);
3872 em->bdev = root->fs_info->fs_devices->latest_bdev;
3873 em->start = EXTENT_MAP_HOLE;
3874 em->orig_start = EXTENT_MAP_HOLE;
3876 em->block_len = (u64)-1;
3879 path = btrfs_alloc_path();
3883 ret = btrfs_lookup_file_extent(trans, root, path,
3884 objectid, start, trans != NULL);
3891 if (path->slots[0] == 0)
3896 leaf = path->nodes[0];
3897 item = btrfs_item_ptr(leaf, path->slots[0],
3898 struct btrfs_file_extent_item);
3899 /* are we inside the extent that was found? */
3900 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3901 found_type = btrfs_key_type(&found_key);
3902 if (found_key.objectid != objectid ||
3903 found_type != BTRFS_EXTENT_DATA_KEY) {
3907 found_type = btrfs_file_extent_type(leaf, item);
3908 extent_start = found_key.offset;
3909 compressed = btrfs_file_extent_compression(leaf, item);
3910 if (found_type == BTRFS_FILE_EXTENT_REG ||
3911 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3912 extent_end = extent_start +
3913 btrfs_file_extent_num_bytes(leaf, item);
3914 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3916 size = btrfs_file_extent_inline_len(leaf, item);
3917 extent_end = (extent_start + size + root->sectorsize - 1) &
3918 ~((u64)root->sectorsize - 1);
3921 if (start >= extent_end) {
3923 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3924 ret = btrfs_next_leaf(root, path);
3931 leaf = path->nodes[0];
3933 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3934 if (found_key.objectid != objectid ||
3935 found_key.type != BTRFS_EXTENT_DATA_KEY)
3937 if (start + len <= found_key.offset)
3940 em->len = found_key.offset - start;
3944 if (found_type == BTRFS_FILE_EXTENT_REG ||
3945 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3946 em->start = extent_start;
3947 em->len = extent_end - extent_start;
3948 em->orig_start = extent_start -
3949 btrfs_file_extent_offset(leaf, item);
3950 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3952 em->block_start = EXTENT_MAP_HOLE;
3956 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3957 em->block_start = bytenr;
3958 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3961 bytenr += btrfs_file_extent_offset(leaf, item);
3962 em->block_start = bytenr;
3963 em->block_len = em->len;
3964 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3965 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3968 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3972 size_t extent_offset;
3975 em->block_start = EXTENT_MAP_INLINE;
3976 if (!page || create) {
3977 em->start = extent_start;
3978 em->len = extent_end - extent_start;
3982 size = btrfs_file_extent_inline_len(leaf, item);
3983 extent_offset = page_offset(page) + pg_offset - extent_start;
3984 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3985 size - extent_offset);
3986 em->start = extent_start + extent_offset;
3987 em->len = (copy_size + root->sectorsize - 1) &
3988 ~((u64)root->sectorsize - 1);
3989 em->orig_start = EXTENT_MAP_INLINE;
3991 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3992 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
3993 if (create == 0 && !PageUptodate(page)) {
3994 if (btrfs_file_extent_compression(leaf, item) ==
3995 BTRFS_COMPRESS_ZLIB) {
3996 ret = uncompress_inline(path, inode, page,
3998 extent_offset, item);
4002 read_extent_buffer(leaf, map + pg_offset, ptr,
4006 flush_dcache_page(page);
4007 } else if (create && PageUptodate(page)) {
4010 free_extent_map(em);
4012 btrfs_release_path(root, path);
4013 trans = btrfs_join_transaction(root, 1);
4017 write_extent_buffer(leaf, map + pg_offset, ptr,
4020 btrfs_mark_buffer_dirty(leaf);
4022 set_extent_uptodate(io_tree, em->start,
4023 extent_map_end(em) - 1, GFP_NOFS);
4026 printk("unkknown found_type %d\n", found_type);
4033 em->block_start = EXTENT_MAP_HOLE;
4034 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4036 btrfs_release_path(root, path);
4037 if (em->start > start || extent_map_end(em) <= start) {
4038 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
4044 spin_lock(&em_tree->lock);
4045 ret = add_extent_mapping(em_tree, em);
4046 /* it is possible that someone inserted the extent into the tree
4047 * while we had the lock dropped. It is also possible that
4048 * an overlapping map exists in the tree
4050 if (ret == -EEXIST) {
4051 struct extent_map *existing;
4055 existing = lookup_extent_mapping(em_tree, start, len);
4056 if (existing && (existing->start > start ||
4057 existing->start + existing->len <= start)) {
4058 free_extent_map(existing);
4062 existing = lookup_extent_mapping(em_tree, em->start,
4065 err = merge_extent_mapping(em_tree, existing,
4068 free_extent_map(existing);
4070 free_extent_map(em);
4075 printk("failing to insert %Lu %Lu\n",
4077 free_extent_map(em);
4081 free_extent_map(em);
4086 spin_unlock(&em_tree->lock);
4089 btrfs_free_path(path);
4091 ret = btrfs_end_transaction(trans, root);
4097 free_extent_map(em);
4099 return ERR_PTR(err);
4104 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4105 const struct iovec *iov, loff_t offset,
4106 unsigned long nr_segs)
4111 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4113 return extent_bmap(mapping, iblock, btrfs_get_extent);
4116 int btrfs_readpage(struct file *file, struct page *page)
4118 struct extent_io_tree *tree;
4119 tree = &BTRFS_I(page->mapping->host)->io_tree;
4120 return extent_read_full_page(tree, page, btrfs_get_extent);
4123 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4125 struct extent_io_tree *tree;
4128 if (current->flags & PF_MEMALLOC) {
4129 redirty_page_for_writepage(wbc, page);
4133 tree = &BTRFS_I(page->mapping->host)->io_tree;
4134 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4137 int btrfs_writepages(struct address_space *mapping,
4138 struct writeback_control *wbc)
4140 struct extent_io_tree *tree;
4142 tree = &BTRFS_I(mapping->host)->io_tree;
4143 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4147 btrfs_readpages(struct file *file, struct address_space *mapping,
4148 struct list_head *pages, unsigned nr_pages)
4150 struct extent_io_tree *tree;
4151 tree = &BTRFS_I(mapping->host)->io_tree;
4152 return extent_readpages(tree, mapping, pages, nr_pages,
4155 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4157 struct extent_io_tree *tree;
4158 struct extent_map_tree *map;
4161 tree = &BTRFS_I(page->mapping->host)->io_tree;
4162 map = &BTRFS_I(page->mapping->host)->extent_tree;
4163 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4165 ClearPagePrivate(page);
4166 set_page_private(page, 0);
4167 page_cache_release(page);
4172 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4174 if (PageWriteback(page) || PageDirty(page))
4176 return __btrfs_releasepage(page, gfp_flags);
4179 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4181 struct extent_io_tree *tree;
4182 struct btrfs_ordered_extent *ordered;
4183 u64 page_start = page_offset(page);
4184 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4186 wait_on_page_writeback(page);
4187 tree = &BTRFS_I(page->mapping->host)->io_tree;
4189 btrfs_releasepage(page, GFP_NOFS);
4193 lock_extent(tree, page_start, page_end, GFP_NOFS);
4194 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4198 * IO on this page will never be started, so we need
4199 * to account for any ordered extents now
4201 clear_extent_bit(tree, page_start, page_end,
4202 EXTENT_DIRTY | EXTENT_DELALLOC |
4203 EXTENT_LOCKED, 1, 0, GFP_NOFS);
4204 btrfs_finish_ordered_io(page->mapping->host,
4205 page_start, page_end);
4206 btrfs_put_ordered_extent(ordered);
4207 lock_extent(tree, page_start, page_end, GFP_NOFS);
4209 clear_extent_bit(tree, page_start, page_end,
4210 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4213 __btrfs_releasepage(page, GFP_NOFS);
4215 ClearPageChecked(page);
4216 if (PagePrivate(page)) {
4217 ClearPagePrivate(page);
4218 set_page_private(page, 0);
4219 page_cache_release(page);
4224 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4225 * called from a page fault handler when a page is first dirtied. Hence we must
4226 * be careful to check for EOF conditions here. We set the page up correctly
4227 * for a written page which means we get ENOSPC checking when writing into
4228 * holes and correct delalloc and unwritten extent mapping on filesystems that
4229 * support these features.
4231 * We are not allowed to take the i_mutex here so we have to play games to
4232 * protect against truncate races as the page could now be beyond EOF. Because
4233 * vmtruncate() writes the inode size before removing pages, once we have the
4234 * page lock we can determine safely if the page is beyond EOF. If it is not
4235 * beyond EOF, then the page is guaranteed safe against truncation until we
4238 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4240 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4241 struct btrfs_root *root = BTRFS_I(inode)->root;
4242 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4243 struct btrfs_ordered_extent *ordered;
4245 unsigned long zero_start;
4251 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4258 size = i_size_read(inode);
4259 page_start = page_offset(page);
4260 page_end = page_start + PAGE_CACHE_SIZE - 1;
4262 if ((page->mapping != inode->i_mapping) ||
4263 (page_start >= size)) {
4264 /* page got truncated out from underneath us */
4267 wait_on_page_writeback(page);
4269 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4270 set_page_extent_mapped(page);
4273 * we can't set the delalloc bits if there are pending ordered
4274 * extents. Drop our locks and wait for them to finish
4276 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4278 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4280 btrfs_start_ordered_extent(inode, ordered, 1);
4281 btrfs_put_ordered_extent(ordered);
4285 btrfs_set_extent_delalloc(inode, page_start, page_end);
4288 /* page is wholly or partially inside EOF */
4289 if (page_start + PAGE_CACHE_SIZE > size)
4290 zero_start = size & ~PAGE_CACHE_MASK;
4292 zero_start = PAGE_CACHE_SIZE;
4294 if (zero_start != PAGE_CACHE_SIZE) {
4296 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4297 flush_dcache_page(page);
4300 ClearPageChecked(page);
4301 set_page_dirty(page);
4302 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4310 static void btrfs_truncate(struct inode *inode)
4312 struct btrfs_root *root = BTRFS_I(inode)->root;
4314 struct btrfs_trans_handle *trans;
4316 u64 mask = root->sectorsize - 1;
4318 if (!S_ISREG(inode->i_mode))
4320 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4323 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4324 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4326 trans = btrfs_start_transaction(root, 1);
4327 btrfs_set_trans_block_group(trans, inode);
4328 btrfs_i_size_write(inode, inode->i_size);
4330 ret = btrfs_orphan_add(trans, inode);
4333 /* FIXME, add redo link to tree so we don't leak on crash */
4334 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4335 BTRFS_EXTENT_DATA_KEY);
4336 btrfs_update_inode(trans, root, inode);
4338 ret = btrfs_orphan_del(trans, inode);
4342 nr = trans->blocks_used;
4343 ret = btrfs_end_transaction_throttle(trans, root);
4345 btrfs_btree_balance_dirty(root, nr);
4349 * Invalidate a single dcache entry at the root of the filesystem.
4350 * Needed after creation of snapshot or subvolume.
4352 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
4355 struct dentry *alias, *entry;
4358 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
4362 /* change me if btrfs ever gets a d_hash operation */
4363 qstr.hash = full_name_hash(qstr.name, qstr.len);
4364 entry = d_lookup(alias, &qstr);
4367 d_invalidate(entry);
4374 * create a new subvolume directory/inode (helper for the ioctl).
4376 int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4377 struct btrfs_trans_handle *trans, u64 new_dirid,
4378 struct btrfs_block_group_cache *block_group)
4380 struct inode *inode;
4384 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4385 new_dirid, block_group, S_IFDIR | 0700, &index);
4387 return PTR_ERR(inode);
4388 inode->i_op = &btrfs_dir_inode_operations;
4389 inode->i_fop = &btrfs_dir_file_operations;
4390 new_root->inode = inode;
4393 btrfs_i_size_write(inode, 0);
4395 error = btrfs_update_inode(trans, new_root, inode);
4399 atomic_inc(&inode->i_count);
4400 d_instantiate(dentry, inode);
4404 /* helper function for file defrag and space balancing. This
4405 * forces readahead on a given range of bytes in an inode
4407 unsigned long btrfs_force_ra(struct address_space *mapping,
4408 struct file_ra_state *ra, struct file *file,
4409 pgoff_t offset, pgoff_t last_index)
4411 pgoff_t req_size = last_index - offset + 1;
4413 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4414 return offset + req_size;
4417 struct inode *btrfs_alloc_inode(struct super_block *sb)
4419 struct btrfs_inode *ei;
4421 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4425 ei->logged_trans = 0;
4426 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4427 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4428 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4429 INIT_LIST_HEAD(&ei->i_orphan);
4430 return &ei->vfs_inode;
4433 void btrfs_destroy_inode(struct inode *inode)
4435 struct btrfs_ordered_extent *ordered;
4436 WARN_ON(!list_empty(&inode->i_dentry));
4437 WARN_ON(inode->i_data.nrpages);
4439 if (BTRFS_I(inode)->i_acl &&
4440 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4441 posix_acl_release(BTRFS_I(inode)->i_acl);
4442 if (BTRFS_I(inode)->i_default_acl &&
4443 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4444 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4446 spin_lock(&BTRFS_I(inode)->root->list_lock);
4447 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4448 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4449 " list\n", inode->i_ino);
4452 spin_unlock(&BTRFS_I(inode)->root->list_lock);
4455 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4459 printk("found ordered extent %Lu %Lu\n",
4460 ordered->file_offset, ordered->len);
4461 btrfs_remove_ordered_extent(inode, ordered);
4462 btrfs_put_ordered_extent(ordered);
4463 btrfs_put_ordered_extent(ordered);
4466 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4467 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4470 static void init_once(void *foo)
4472 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4474 inode_init_once(&ei->vfs_inode);
4477 void btrfs_destroy_cachep(void)
4479 if (btrfs_inode_cachep)
4480 kmem_cache_destroy(btrfs_inode_cachep);
4481 if (btrfs_trans_handle_cachep)
4482 kmem_cache_destroy(btrfs_trans_handle_cachep);
4483 if (btrfs_transaction_cachep)
4484 kmem_cache_destroy(btrfs_transaction_cachep);
4485 if (btrfs_bit_radix_cachep)
4486 kmem_cache_destroy(btrfs_bit_radix_cachep);
4487 if (btrfs_path_cachep)
4488 kmem_cache_destroy(btrfs_path_cachep);
4491 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4492 unsigned long extra_flags,
4493 void (*ctor)(void *))
4495 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4496 SLAB_MEM_SPREAD | extra_flags), ctor);
4499 int btrfs_init_cachep(void)
4501 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4502 sizeof(struct btrfs_inode),
4504 if (!btrfs_inode_cachep)
4506 btrfs_trans_handle_cachep =
4507 btrfs_cache_create("btrfs_trans_handle_cache",
4508 sizeof(struct btrfs_trans_handle),
4510 if (!btrfs_trans_handle_cachep)
4512 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4513 sizeof(struct btrfs_transaction),
4515 if (!btrfs_transaction_cachep)
4517 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4518 sizeof(struct btrfs_path),
4520 if (!btrfs_path_cachep)
4522 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4523 SLAB_DESTROY_BY_RCU, NULL);
4524 if (!btrfs_bit_radix_cachep)
4528 btrfs_destroy_cachep();
4532 static int btrfs_getattr(struct vfsmount *mnt,
4533 struct dentry *dentry, struct kstat *stat)
4535 struct inode *inode = dentry->d_inode;
4536 generic_fillattr(inode, stat);
4537 stat->blksize = PAGE_CACHE_SIZE;
4538 stat->blocks = (inode_get_bytes(inode) +
4539 BTRFS_I(inode)->delalloc_bytes) >> 9;
4543 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4544 struct inode * new_dir,struct dentry *new_dentry)
4546 struct btrfs_trans_handle *trans;
4547 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4548 struct inode *new_inode = new_dentry->d_inode;
4549 struct inode *old_inode = old_dentry->d_inode;
4550 struct timespec ctime = CURRENT_TIME;
4554 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4555 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4559 ret = btrfs_check_free_space(root, 1, 0);
4563 trans = btrfs_start_transaction(root, 1);
4565 btrfs_set_trans_block_group(trans, new_dir);
4567 btrfs_inc_nlink(old_dentry->d_inode);
4568 old_dir->i_ctime = old_dir->i_mtime = ctime;
4569 new_dir->i_ctime = new_dir->i_mtime = ctime;
4570 old_inode->i_ctime = ctime;
4572 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4573 old_dentry->d_name.name,
4574 old_dentry->d_name.len);
4579 new_inode->i_ctime = CURRENT_TIME;
4580 ret = btrfs_unlink_inode(trans, root, new_dir,
4581 new_dentry->d_inode,
4582 new_dentry->d_name.name,
4583 new_dentry->d_name.len);
4586 if (new_inode->i_nlink == 0) {
4587 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4593 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
4597 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4598 old_inode, new_dentry->d_name.name,
4599 new_dentry->d_name.len, 1, index);
4604 btrfs_end_transaction_throttle(trans, root);
4610 * some fairly slow code that needs optimization. This walks the list
4611 * of all the inodes with pending delalloc and forces them to disk.
4613 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4615 struct list_head *head = &root->fs_info->delalloc_inodes;
4616 struct btrfs_inode *binode;
4617 struct inode *inode;
4618 unsigned long flags;
4620 if (root->fs_info->sb->s_flags & MS_RDONLY)
4623 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4624 while(!list_empty(head)) {
4625 binode = list_entry(head->next, struct btrfs_inode,
4627 inode = igrab(&binode->vfs_inode);
4629 list_del_init(&binode->delalloc_inodes);
4630 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4632 filemap_flush(inode->i_mapping);
4636 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4638 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4640 /* the filemap_flush will queue IO into the worker threads, but
4641 * we have to make sure the IO is actually started and that
4642 * ordered extents get created before we return
4644 atomic_inc(&root->fs_info->async_submit_draining);
4645 while(atomic_read(&root->fs_info->nr_async_submits) ||
4646 atomic_read(&root->fs_info->async_delalloc_pages)) {
4647 wait_event(root->fs_info->async_submit_wait,
4648 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4649 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4651 atomic_dec(&root->fs_info->async_submit_draining);
4655 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4656 const char *symname)
4658 struct btrfs_trans_handle *trans;
4659 struct btrfs_root *root = BTRFS_I(dir)->root;
4660 struct btrfs_path *path;
4661 struct btrfs_key key;
4662 struct inode *inode = NULL;
4670 struct btrfs_file_extent_item *ei;
4671 struct extent_buffer *leaf;
4672 unsigned long nr = 0;
4674 name_len = strlen(symname) + 1;
4675 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4676 return -ENAMETOOLONG;
4678 err = btrfs_check_free_space(root, 1, 0);
4682 trans = btrfs_start_transaction(root, 1);
4683 btrfs_set_trans_block_group(trans, dir);
4685 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4691 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4693 dentry->d_parent->d_inode->i_ino, objectid,
4694 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4696 err = PTR_ERR(inode);
4700 err = btrfs_init_acl(inode, dir);
4706 btrfs_set_trans_block_group(trans, inode);
4707 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4711 inode->i_mapping->a_ops = &btrfs_aops;
4712 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4713 inode->i_fop = &btrfs_file_operations;
4714 inode->i_op = &btrfs_file_inode_operations;
4715 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4717 dir->i_sb->s_dirt = 1;
4718 btrfs_update_inode_block_group(trans, inode);
4719 btrfs_update_inode_block_group(trans, dir);
4723 path = btrfs_alloc_path();
4725 key.objectid = inode->i_ino;
4727 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4728 datasize = btrfs_file_extent_calc_inline_size(name_len);
4729 err = btrfs_insert_empty_item(trans, root, path, &key,
4735 leaf = path->nodes[0];
4736 ei = btrfs_item_ptr(leaf, path->slots[0],
4737 struct btrfs_file_extent_item);
4738 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4739 btrfs_set_file_extent_type(leaf, ei,
4740 BTRFS_FILE_EXTENT_INLINE);
4741 btrfs_set_file_extent_encryption(leaf, ei, 0);
4742 btrfs_set_file_extent_compression(leaf, ei, 0);
4743 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4744 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4746 ptr = btrfs_file_extent_inline_start(ei);
4747 write_extent_buffer(leaf, symname, ptr, name_len);
4748 btrfs_mark_buffer_dirty(leaf);
4749 btrfs_free_path(path);
4751 inode->i_op = &btrfs_symlink_inode_operations;
4752 inode->i_mapping->a_ops = &btrfs_symlink_aops;
4753 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4754 inode_set_bytes(inode, name_len);
4755 btrfs_i_size_write(inode, name_len - 1);
4756 err = btrfs_update_inode(trans, root, inode);
4761 nr = trans->blocks_used;
4762 btrfs_end_transaction_throttle(trans, root);
4765 inode_dec_link_count(inode);
4768 btrfs_btree_balance_dirty(root, nr);
4772 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4773 u64 alloc_hint, int mode)
4775 struct btrfs_trans_handle *trans;
4776 struct btrfs_root *root = BTRFS_I(inode)->root;
4777 struct btrfs_key ins;
4779 u64 cur_offset = start;
4780 u64 num_bytes = end - start;
4783 trans = btrfs_join_transaction(root, 1);
4785 btrfs_set_trans_block_group(trans, inode);
4787 while (num_bytes > 0) {
4788 alloc_size = min(num_bytes, root->fs_info->max_extent);
4789 ret = btrfs_reserve_extent(trans, root, alloc_size,
4790 root->sectorsize, 0, alloc_hint,
4796 ret = insert_reserved_file_extent(trans, inode,
4797 cur_offset, ins.objectid,
4798 ins.offset, ins.offset,
4799 ins.offset, 0, 0, 0,
4800 BTRFS_FILE_EXTENT_PREALLOC);
4802 num_bytes -= ins.offset;
4803 cur_offset += ins.offset;
4804 alloc_hint = ins.objectid + ins.offset;
4807 if (cur_offset > start) {
4808 inode->i_ctime = CURRENT_TIME;
4809 btrfs_set_flag(inode, PREALLOC);
4810 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4811 cur_offset > i_size_read(inode))
4812 btrfs_i_size_write(inode, cur_offset);
4813 ret = btrfs_update_inode(trans, root, inode);
4817 btrfs_end_transaction(trans, root);
4821 static long btrfs_fallocate(struct inode *inode, int mode,
4822 loff_t offset, loff_t len)
4829 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4830 struct extent_map *em;
4833 alloc_start = offset & ~mask;
4834 alloc_end = (offset + len + mask) & ~mask;
4836 mutex_lock(&inode->i_mutex);
4837 if (alloc_start > inode->i_size) {
4838 ret = btrfs_cont_expand(inode, alloc_start);
4844 struct btrfs_ordered_extent *ordered;
4845 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4846 alloc_end - 1, GFP_NOFS);
4847 ordered = btrfs_lookup_first_ordered_extent(inode,
4850 ordered->file_offset + ordered->len > alloc_start &&
4851 ordered->file_offset < alloc_end) {
4852 btrfs_put_ordered_extent(ordered);
4853 unlock_extent(&BTRFS_I(inode)->io_tree,
4854 alloc_start, alloc_end - 1, GFP_NOFS);
4855 btrfs_wait_ordered_range(inode, alloc_start,
4856 alloc_end - alloc_start);
4859 btrfs_put_ordered_extent(ordered);
4864 cur_offset = alloc_start;
4866 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4867 alloc_end - cur_offset, 0);
4868 BUG_ON(IS_ERR(em) || !em);
4869 last_byte = min(extent_map_end(em), alloc_end);
4870 last_byte = (last_byte + mask) & ~mask;
4871 if (em->block_start == EXTENT_MAP_HOLE) {
4872 ret = prealloc_file_range(inode, cur_offset,
4873 last_byte, alloc_hint, mode);
4875 free_extent_map(em);
4879 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4880 alloc_hint = em->block_start;
4881 free_extent_map(em);
4883 cur_offset = last_byte;
4884 if (cur_offset >= alloc_end) {
4889 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4892 mutex_unlock(&inode->i_mutex);
4896 static int btrfs_set_page_dirty(struct page *page)
4898 return __set_page_dirty_nobuffers(page);
4901 static int btrfs_permission(struct inode *inode, int mask)
4903 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4905 return generic_permission(inode, mask, btrfs_check_acl);
4908 static struct inode_operations btrfs_dir_inode_operations = {
4909 .lookup = btrfs_lookup,
4910 .create = btrfs_create,
4911 .unlink = btrfs_unlink,
4913 .mkdir = btrfs_mkdir,
4914 .rmdir = btrfs_rmdir,
4915 .rename = btrfs_rename,
4916 .symlink = btrfs_symlink,
4917 .setattr = btrfs_setattr,
4918 .mknod = btrfs_mknod,
4919 .setxattr = btrfs_setxattr,
4920 .getxattr = btrfs_getxattr,
4921 .listxattr = btrfs_listxattr,
4922 .removexattr = btrfs_removexattr,
4923 .permission = btrfs_permission,
4925 static struct inode_operations btrfs_dir_ro_inode_operations = {
4926 .lookup = btrfs_lookup,
4927 .permission = btrfs_permission,
4929 static struct file_operations btrfs_dir_file_operations = {
4930 .llseek = generic_file_llseek,
4931 .read = generic_read_dir,
4932 .readdir = btrfs_real_readdir,
4933 .unlocked_ioctl = btrfs_ioctl,
4934 #ifdef CONFIG_COMPAT
4935 .compat_ioctl = btrfs_ioctl,
4937 .release = btrfs_release_file,
4938 .fsync = btrfs_sync_file,
4941 static struct extent_io_ops btrfs_extent_io_ops = {
4942 .fill_delalloc = run_delalloc_range,
4943 .submit_bio_hook = btrfs_submit_bio_hook,
4944 .merge_bio_hook = btrfs_merge_bio_hook,
4945 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4946 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4947 .writepage_start_hook = btrfs_writepage_start_hook,
4948 .readpage_io_failed_hook = btrfs_io_failed_hook,
4949 .set_bit_hook = btrfs_set_bit_hook,
4950 .clear_bit_hook = btrfs_clear_bit_hook,
4953 static struct address_space_operations btrfs_aops = {
4954 .readpage = btrfs_readpage,
4955 .writepage = btrfs_writepage,
4956 .writepages = btrfs_writepages,
4957 .readpages = btrfs_readpages,
4958 .sync_page = block_sync_page,
4960 .direct_IO = btrfs_direct_IO,
4961 .invalidatepage = btrfs_invalidatepage,
4962 .releasepage = btrfs_releasepage,
4963 .set_page_dirty = btrfs_set_page_dirty,
4966 static struct address_space_operations btrfs_symlink_aops = {
4967 .readpage = btrfs_readpage,
4968 .writepage = btrfs_writepage,
4969 .invalidatepage = btrfs_invalidatepage,
4970 .releasepage = btrfs_releasepage,
4973 static struct inode_operations btrfs_file_inode_operations = {
4974 .truncate = btrfs_truncate,
4975 .getattr = btrfs_getattr,
4976 .setattr = btrfs_setattr,
4977 .setxattr = btrfs_setxattr,
4978 .getxattr = btrfs_getxattr,
4979 .listxattr = btrfs_listxattr,
4980 .removexattr = btrfs_removexattr,
4981 .permission = btrfs_permission,
4982 .fallocate = btrfs_fallocate,
4984 static struct inode_operations btrfs_special_inode_operations = {
4985 .getattr = btrfs_getattr,
4986 .setattr = btrfs_setattr,
4987 .permission = btrfs_permission,
4988 .setxattr = btrfs_setxattr,
4989 .getxattr = btrfs_getxattr,
4990 .listxattr = btrfs_listxattr,
4991 .removexattr = btrfs_removexattr,
4993 static struct inode_operations btrfs_symlink_inode_operations = {
4994 .readlink = generic_readlink,
4995 .follow_link = page_follow_link_light,
4996 .put_link = page_put_link,
4997 .permission = btrfs_permission,