2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
40 #include <linux/falloc.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
52 #include "ref-cache.h"
53 #include "compression.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
95 * a very lame attempt at stopping writes when the FS is 85% full. There
96 * are countless ways this is incorrect, but it is better than nothing.
98 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
107 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
108 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
109 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
117 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
119 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
131 struct page **compressed_pages)
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
144 unsigned long offset;
145 int use_compress = 0;
147 if (compressed_size && compressed_pages) {
149 cur_size = compressed_size;
152 path = btrfs_alloc_path(); if (!path)
155 btrfs_set_trans_block_group(trans, inode);
157 key.objectid = inode->i_ino;
159 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
160 inode_add_bytes(inode, size);
161 datasize = btrfs_file_extent_calc_inline_size(cur_size);
163 inode_add_bytes(inode, size);
164 ret = btrfs_insert_empty_item(trans, root, path, &key,
169 printk("got bad ret %d\n", ret);
172 leaf = path->nodes[0];
173 ei = btrfs_item_ptr(leaf, path->slots[0],
174 struct btrfs_file_extent_item);
175 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
176 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
177 btrfs_set_file_extent_encryption(leaf, ei, 0);
178 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
179 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
180 ptr = btrfs_file_extent_inline_start(ei);
185 while(compressed_size > 0) {
186 cpage = compressed_pages[i];
187 cur_size = min(compressed_size,
191 write_extent_buffer(leaf, kaddr, ptr, cur_size);
196 compressed_size -= cur_size;
198 btrfs_set_file_extent_compression(leaf, ei,
199 BTRFS_COMPRESS_ZLIB);
201 page = find_get_page(inode->i_mapping,
202 start >> PAGE_CACHE_SHIFT);
203 btrfs_set_file_extent_compression(leaf, ei, 0);
204 kaddr = kmap_atomic(page, KM_USER0);
205 offset = start & (PAGE_CACHE_SIZE - 1);
206 write_extent_buffer(leaf, kaddr + offset, ptr, size);
207 kunmap_atomic(kaddr, KM_USER0);
208 page_cache_release(page);
210 btrfs_mark_buffer_dirty(leaf);
211 btrfs_free_path(path);
213 BTRFS_I(inode)->disk_i_size = inode->i_size;
214 btrfs_update_inode(trans, root, inode);
217 btrfs_free_path(path);
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
227 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
228 struct btrfs_root *root,
229 struct inode *inode, u64 start, u64 end,
230 size_t compressed_size,
231 struct page **compressed_pages)
233 u64 isize = i_size_read(inode);
234 u64 actual_end = min(end + 1, isize);
235 u64 inline_len = actual_end - start;
236 u64 aligned_end = (end + root->sectorsize - 1) &
237 ~((u64)root->sectorsize - 1);
239 u64 data_len = inline_len;
243 data_len = compressed_size;
246 actual_end >= PAGE_CACHE_SIZE ||
247 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
249 (actual_end & (root->sectorsize - 1)) == 0) ||
251 data_len > root->fs_info->max_inline) {
255 ret = btrfs_drop_extents(trans, root, inode, start,
256 aligned_end, start, &hint_byte);
259 if (isize > actual_end)
260 inline_len = min_t(u64, isize, actual_end);
261 ret = insert_inline_extent(trans, root, inode, start,
262 inline_len, compressed_size,
265 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
269 struct async_extent {
274 unsigned long nr_pages;
275 struct list_head list;
280 struct btrfs_root *root;
281 struct page *locked_page;
284 struct list_head extents;
285 struct btrfs_work work;
288 static noinline int add_async_extent(struct async_cow *cow,
289 u64 start, u64 ram_size,
292 unsigned long nr_pages)
294 struct async_extent *async_extent;
296 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
302 list_add_tail(&async_extent->list, &cow->extents);
307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
322 static noinline int compress_file_range(struct inode *inode,
323 struct page *locked_page,
325 struct async_cow *async_cow,
328 struct btrfs_root *root = BTRFS_I(inode)->root;
329 struct btrfs_trans_handle *trans;
333 u64 blocksize = root->sectorsize;
336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
342 unsigned long max_uncompressed = 128 * 1024;
350 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
351 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
353 actual_end = min_t(u64, i_size_read(inode), end + 1);
354 total_compressed = actual_end - start;
356 /* we want to make sure that amount of ram required to uncompress
357 * an extent is reasonable, so we limit the total size in ram
358 * of a compressed extent to 128k. This is a crucial number
359 * because it also controls how easily we can spread reads across
360 * cpus for decompression.
362 * We also want to make sure the amount of IO required to do
363 * a random read is reasonably small, so we limit the size of
364 * a compressed extent to 128k.
366 total_compressed = min(total_compressed, max_uncompressed);
367 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
368 num_bytes = max(blocksize, num_bytes);
369 disk_num_bytes = num_bytes;
374 * we do compression for mount -o compress and when the
375 * inode has not been flagged as nocompress. This flag can
376 * change at any time if we discover bad compression ratios.
378 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
379 btrfs_test_opt(root, COMPRESS)) {
381 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
383 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
384 total_compressed, pages,
385 nr_pages, &nr_pages_ret,
391 unsigned long offset = total_compressed &
392 (PAGE_CACHE_SIZE - 1);
393 struct page *page = pages[nr_pages_ret - 1];
396 /* zero the tail end of the last page, we might be
397 * sending it down to disk
400 kaddr = kmap_atomic(page, KM_USER0);
401 memset(kaddr + offset, 0,
402 PAGE_CACHE_SIZE - offset);
403 kunmap_atomic(kaddr, KM_USER0);
409 trans = btrfs_join_transaction(root, 1);
411 btrfs_set_trans_block_group(trans, inode);
413 /* lets try to make an inline extent */
414 if (ret || total_in < (actual_end - start)) {
415 /* we didn't compress the entire range, try
416 * to make an uncompressed inline extent.
418 ret = cow_file_range_inline(trans, root, inode,
419 start, end, 0, NULL);
421 /* try making a compressed inline extent */
422 ret = cow_file_range_inline(trans, root, inode,
424 total_compressed, pages);
426 btrfs_end_transaction(trans, root);
429 * inline extent creation worked, we don't need
430 * to create any more async work items. Unlock
431 * and free up our temp pages.
433 extent_clear_unlock_delalloc(inode,
434 &BTRFS_I(inode)->io_tree,
435 start, end, NULL, 1, 0,
444 * we aren't doing an inline extent round the compressed size
445 * up to a block size boundary so the allocator does sane
448 total_compressed = (total_compressed + blocksize - 1) &
452 * one last check to make sure the compression is really a
453 * win, compare the page count read with the blocks on disk
455 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
456 ~(PAGE_CACHE_SIZE - 1);
457 if (total_compressed >= total_in) {
460 disk_num_bytes = total_compressed;
461 num_bytes = total_in;
464 if (!will_compress && pages) {
466 * the compression code ran but failed to make things smaller,
467 * free any pages it allocated and our page pointer array
469 for (i = 0; i < nr_pages_ret; i++) {
470 WARN_ON(pages[i]->mapping);
471 page_cache_release(pages[i]);
475 total_compressed = 0;
478 /* flag the file so we don't compress in the future */
479 btrfs_set_flag(inode, NOCOMPRESS);
484 /* the async work queues will take care of doing actual
485 * allocation on disk for these compressed pages,
486 * and will submit them to the elevator.
488 add_async_extent(async_cow, start, num_bytes,
489 total_compressed, pages, nr_pages_ret);
491 if (start + num_bytes < end) {
499 * No compression, but we still need to write the pages in
500 * the file we've been given so far. redirty the locked
501 * page if it corresponds to our extent and set things up
502 * for the async work queue to run cow_file_range to do
503 * the normal delalloc dance
505 if (page_offset(locked_page) >= start &&
506 page_offset(locked_page) <= end) {
507 __set_page_dirty_nobuffers(locked_page);
508 /* unlocked later on in the async handlers */
510 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
518 for (i = 0; i < nr_pages_ret; i++) {
519 WARN_ON(pages[i]->mapping);
520 page_cache_release(pages[i]);
529 * phase two of compressed writeback. This is the ordered portion
530 * of the code, which only gets called in the order the work was
531 * queued. We walk all the async extents created by compress_file_range
532 * and send them down to the disk.
534 static noinline int submit_compressed_extents(struct inode *inode,
535 struct async_cow *async_cow)
537 struct async_extent *async_extent;
539 struct btrfs_trans_handle *trans;
540 struct btrfs_key ins;
541 struct extent_map *em;
542 struct btrfs_root *root = BTRFS_I(inode)->root;
543 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
544 struct extent_io_tree *io_tree;
547 if (list_empty(&async_cow->extents))
550 trans = btrfs_join_transaction(root, 1);
552 while(!list_empty(&async_cow->extents)) {
553 async_extent = list_entry(async_cow->extents.next,
554 struct async_extent, list);
555 list_del(&async_extent->list);
557 io_tree = &BTRFS_I(inode)->io_tree;
559 /* did the compression code fall back to uncompressed IO? */
560 if (!async_extent->pages) {
561 int page_started = 0;
562 unsigned long nr_written = 0;
564 lock_extent(io_tree, async_extent->start,
565 async_extent->start + async_extent->ram_size - 1,
568 /* allocate blocks */
569 cow_file_range(inode, async_cow->locked_page,
571 async_extent->start +
572 async_extent->ram_size - 1,
573 &page_started, &nr_written, 0);
576 * if page_started, cow_file_range inserted an
577 * inline extent and took care of all the unlocking
578 * and IO for us. Otherwise, we need to submit
579 * all those pages down to the drive.
582 extent_write_locked_range(io_tree,
583 inode, async_extent->start,
584 async_extent->start +
585 async_extent->ram_size - 1,
593 lock_extent(io_tree, async_extent->start,
594 async_extent->start + async_extent->ram_size - 1,
597 * here we're doing allocation and writeback of the
600 btrfs_drop_extent_cache(inode, async_extent->start,
601 async_extent->start +
602 async_extent->ram_size - 1, 0);
604 ret = btrfs_reserve_extent(trans, root,
605 async_extent->compressed_size,
606 async_extent->compressed_size,
610 em = alloc_extent_map(GFP_NOFS);
611 em->start = async_extent->start;
612 em->len = async_extent->ram_size;
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 /* don't do orphan cleanup if the fs is readonly. */
1812 if (root->fs_info->sb->s_flags & MS_RDONLY)
1815 path = btrfs_alloc_path();
1820 key.objectid = BTRFS_ORPHAN_OBJECTID;
1821 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1822 key.offset = (u64)-1;
1826 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1828 printk(KERN_ERR "Error searching slot for orphan: %d"
1834 * if ret == 0 means we found what we were searching for, which
1835 * is weird, but possible, so only screw with path if we didnt
1836 * find the key and see if we have stuff that matches
1839 if (path->slots[0] == 0)
1844 /* pull out the item */
1845 leaf = path->nodes[0];
1846 item = btrfs_item_nr(leaf, path->slots[0]);
1847 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1849 /* make sure the item matches what we want */
1850 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1852 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1855 /* release the path since we're done with it */
1856 btrfs_release_path(root, path);
1859 * this is where we are basically btrfs_lookup, without the
1860 * crossing root thing. we store the inode number in the
1861 * offset of the orphan item.
1863 inode = btrfs_iget_locked(root->fs_info->sb,
1864 found_key.offset, root);
1868 if (inode->i_state & I_NEW) {
1869 BTRFS_I(inode)->root = root;
1871 /* have to set the location manually */
1872 BTRFS_I(inode)->location.objectid = inode->i_ino;
1873 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1874 BTRFS_I(inode)->location.offset = 0;
1876 btrfs_read_locked_inode(inode);
1877 unlock_new_inode(inode);
1881 * add this inode to the orphan list so btrfs_orphan_del does
1882 * the proper thing when we hit it
1884 spin_lock(&root->list_lock);
1885 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1886 spin_unlock(&root->list_lock);
1889 * if this is a bad inode, means we actually succeeded in
1890 * removing the inode, but not the orphan record, which means
1891 * we need to manually delete the orphan since iput will just
1892 * do a destroy_inode
1894 if (is_bad_inode(inode)) {
1895 trans = btrfs_start_transaction(root, 1);
1896 btrfs_orphan_del(trans, inode);
1897 btrfs_end_transaction(trans, root);
1902 /* if we have links, this was a truncate, lets do that */
1903 if (inode->i_nlink) {
1905 btrfs_truncate(inode);
1910 /* this will do delete_inode and everything for us */
1915 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1917 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1919 btrfs_free_path(path);
1923 * read an inode from the btree into the in-memory inode
1925 void btrfs_read_locked_inode(struct inode *inode)
1927 struct btrfs_path *path;
1928 struct extent_buffer *leaf;
1929 struct btrfs_inode_item *inode_item;
1930 struct btrfs_timespec *tspec;
1931 struct btrfs_root *root = BTRFS_I(inode)->root;
1932 struct btrfs_key location;
1933 u64 alloc_group_block;
1937 path = btrfs_alloc_path();
1939 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1941 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1945 leaf = path->nodes[0];
1946 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1947 struct btrfs_inode_item);
1949 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1950 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1951 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1952 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1953 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1955 tspec = btrfs_inode_atime(inode_item);
1956 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1957 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1959 tspec = btrfs_inode_mtime(inode_item);
1960 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1961 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1963 tspec = btrfs_inode_ctime(inode_item);
1964 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1965 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1967 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1968 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1969 inode->i_generation = BTRFS_I(inode)->generation;
1971 rdev = btrfs_inode_rdev(leaf, inode_item);
1973 BTRFS_I(inode)->index_cnt = (u64)-1;
1975 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1976 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1978 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1979 if (!BTRFS_I(inode)->block_group) {
1980 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1982 BTRFS_BLOCK_GROUP_METADATA, 0);
1984 btrfs_free_path(path);
1987 switch (inode->i_mode & S_IFMT) {
1989 inode->i_mapping->a_ops = &btrfs_aops;
1990 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1991 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1992 inode->i_fop = &btrfs_file_operations;
1993 inode->i_op = &btrfs_file_inode_operations;
1996 inode->i_fop = &btrfs_dir_file_operations;
1997 if (root == root->fs_info->tree_root)
1998 inode->i_op = &btrfs_dir_ro_inode_operations;
2000 inode->i_op = &btrfs_dir_inode_operations;
2003 inode->i_op = &btrfs_symlink_inode_operations;
2004 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2005 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2008 init_special_inode(inode, inode->i_mode, rdev);
2014 btrfs_free_path(path);
2015 make_bad_inode(inode);
2019 * given a leaf and an inode, copy the inode fields into the leaf
2021 static void fill_inode_item(struct btrfs_trans_handle *trans,
2022 struct extent_buffer *leaf,
2023 struct btrfs_inode_item *item,
2024 struct inode *inode)
2026 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2027 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2028 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2029 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2030 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2032 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2033 inode->i_atime.tv_sec);
2034 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2035 inode->i_atime.tv_nsec);
2037 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2038 inode->i_mtime.tv_sec);
2039 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2040 inode->i_mtime.tv_nsec);
2042 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2043 inode->i_ctime.tv_sec);
2044 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2045 inode->i_ctime.tv_nsec);
2047 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2048 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2049 btrfs_set_inode_transid(leaf, item, trans->transid);
2050 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2051 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2052 btrfs_set_inode_block_group(leaf, item,
2053 BTRFS_I(inode)->block_group->key.objectid);
2057 * copy everything in the in-memory inode into the btree.
2059 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
2060 struct btrfs_root *root,
2061 struct inode *inode)
2063 struct btrfs_inode_item *inode_item;
2064 struct btrfs_path *path;
2065 struct extent_buffer *leaf;
2068 path = btrfs_alloc_path();
2070 ret = btrfs_lookup_inode(trans, root, path,
2071 &BTRFS_I(inode)->location, 1);
2078 leaf = path->nodes[0];
2079 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2080 struct btrfs_inode_item);
2082 fill_inode_item(trans, leaf, inode_item, inode);
2083 btrfs_mark_buffer_dirty(leaf);
2084 btrfs_set_inode_last_trans(trans, inode);
2087 btrfs_free_path(path);
2093 * unlink helper that gets used here in inode.c and in the tree logging
2094 * recovery code. It remove a link in a directory with a given name, and
2095 * also drops the back refs in the inode to the directory
2097 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2098 struct btrfs_root *root,
2099 struct inode *dir, struct inode *inode,
2100 const char *name, int name_len)
2102 struct btrfs_path *path;
2104 struct extent_buffer *leaf;
2105 struct btrfs_dir_item *di;
2106 struct btrfs_key key;
2109 path = btrfs_alloc_path();
2115 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2116 name, name_len, -1);
2125 leaf = path->nodes[0];
2126 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2127 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2130 btrfs_release_path(root, path);
2132 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2134 dir->i_ino, &index);
2136 printk("failed to delete reference to %.*s, "
2137 "inode %lu parent %lu\n", name_len, name,
2138 inode->i_ino, dir->i_ino);
2142 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2143 index, name, name_len, -1);
2152 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2153 btrfs_release_path(root, path);
2155 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2157 BUG_ON(ret != 0 && ret != -ENOENT);
2159 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2161 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2165 btrfs_free_path(path);
2169 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2170 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2171 btrfs_update_inode(trans, root, dir);
2172 btrfs_drop_nlink(inode);
2173 ret = btrfs_update_inode(trans, root, inode);
2174 dir->i_sb->s_dirt = 1;
2179 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2181 struct btrfs_root *root;
2182 struct btrfs_trans_handle *trans;
2183 struct inode *inode = dentry->d_inode;
2185 unsigned long nr = 0;
2187 root = BTRFS_I(dir)->root;
2189 ret = btrfs_check_free_space(root, 1, 1);
2193 trans = btrfs_start_transaction(root, 1);
2195 btrfs_set_trans_block_group(trans, dir);
2196 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2197 dentry->d_name.name, dentry->d_name.len);
2199 if (inode->i_nlink == 0)
2200 ret = btrfs_orphan_add(trans, inode);
2202 nr = trans->blocks_used;
2204 btrfs_end_transaction_throttle(trans, root);
2206 btrfs_btree_balance_dirty(root, nr);
2210 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2212 struct inode *inode = dentry->d_inode;
2215 struct btrfs_root *root = BTRFS_I(dir)->root;
2216 struct btrfs_trans_handle *trans;
2217 unsigned long nr = 0;
2219 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2223 ret = btrfs_check_free_space(root, 1, 1);
2227 trans = btrfs_start_transaction(root, 1);
2228 btrfs_set_trans_block_group(trans, dir);
2230 err = btrfs_orphan_add(trans, inode);
2234 /* now the directory is empty */
2235 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2236 dentry->d_name.name, dentry->d_name.len);
2238 btrfs_i_size_write(inode, 0);
2242 nr = trans->blocks_used;
2243 ret = btrfs_end_transaction_throttle(trans, root);
2245 btrfs_btree_balance_dirty(root, nr);
2253 * when truncating bytes in a file, it is possible to avoid reading
2254 * the leaves that contain only checksum items. This can be the
2255 * majority of the IO required to delete a large file, but it must
2256 * be done carefully.
2258 * The keys in the level just above the leaves are checked to make sure
2259 * the lowest key in a given leaf is a csum key, and starts at an offset
2260 * after the new size.
2262 * Then the key for the next leaf is checked to make sure it also has
2263 * a checksum item for the same file. If it does, we know our target leaf
2264 * contains only checksum items, and it can be safely freed without reading
2267 * This is just an optimization targeted at large files. It may do
2268 * nothing. It will return 0 unless things went badly.
2270 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2271 struct btrfs_root *root,
2272 struct btrfs_path *path,
2273 struct inode *inode, u64 new_size)
2275 struct btrfs_key key;
2278 struct btrfs_key found_key;
2279 struct btrfs_key other_key;
2280 struct btrfs_leaf_ref *ref;
2284 path->lowest_level = 1;
2285 key.objectid = inode->i_ino;
2286 key.type = BTRFS_CSUM_ITEM_KEY;
2287 key.offset = new_size;
2289 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2293 if (path->nodes[1] == NULL) {
2298 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2299 nritems = btrfs_header_nritems(path->nodes[1]);
2304 if (path->slots[1] >= nritems)
2307 /* did we find a key greater than anything we want to delete? */
2308 if (found_key.objectid > inode->i_ino ||
2309 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2312 /* we check the next key in the node to make sure the leave contains
2313 * only checksum items. This comparison doesn't work if our
2314 * leaf is the last one in the node
2316 if (path->slots[1] + 1 >= nritems) {
2318 /* search forward from the last key in the node, this
2319 * will bring us into the next node in the tree
2321 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2323 /* unlikely, but we inc below, so check to be safe */
2324 if (found_key.offset == (u64)-1)
2327 /* search_forward needs a path with locks held, do the
2328 * search again for the original key. It is possible
2329 * this will race with a balance and return a path that
2330 * we could modify, but this drop is just an optimization
2331 * and is allowed to miss some leaves.
2333 btrfs_release_path(root, path);
2336 /* setup a max key for search_forward */
2337 other_key.offset = (u64)-1;
2338 other_key.type = key.type;
2339 other_key.objectid = key.objectid;
2341 path->keep_locks = 1;
2342 ret = btrfs_search_forward(root, &found_key, &other_key,
2344 path->keep_locks = 0;
2345 if (ret || found_key.objectid != key.objectid ||
2346 found_key.type != key.type) {
2351 key.offset = found_key.offset;
2352 btrfs_release_path(root, path);
2357 /* we know there's one more slot after us in the tree,
2358 * read that key so we can verify it is also a checksum item
2360 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2362 if (found_key.objectid < inode->i_ino)
2365 if (found_key.type != key.type || found_key.offset < new_size)
2369 * if the key for the next leaf isn't a csum key from this objectid,
2370 * we can't be sure there aren't good items inside this leaf.
2373 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2376 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2377 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2379 * it is safe to delete this leaf, it contains only
2380 * csum items from this inode at an offset >= new_size
2382 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2385 if (root->ref_cows && leaf_gen < trans->transid) {
2386 ref = btrfs_alloc_leaf_ref(root, 0);
2388 ref->root_gen = root->root_key.offset;
2389 ref->bytenr = leaf_start;
2391 ref->generation = leaf_gen;
2394 ret = btrfs_add_leaf_ref(root, ref, 0);
2396 btrfs_free_leaf_ref(root, ref);
2402 btrfs_release_path(root, path);
2404 if (other_key.objectid == inode->i_ino &&
2405 other_key.type == key.type && other_key.offset > key.offset) {
2406 key.offset = other_key.offset;
2412 /* fixup any changes we've made to the path */
2413 path->lowest_level = 0;
2414 path->keep_locks = 0;
2415 btrfs_release_path(root, path);
2420 * this can truncate away extent items, csum items and directory items.
2421 * It starts at a high offset and removes keys until it can't find
2422 * any higher than new_size
2424 * csum items that cross the new i_size are truncated to the new size
2427 * min_type is the minimum key type to truncate down to. If set to 0, this
2428 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2430 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2431 struct btrfs_root *root,
2432 struct inode *inode,
2433 u64 new_size, u32 min_type)
2436 struct btrfs_path *path;
2437 struct btrfs_key key;
2438 struct btrfs_key found_key;
2440 struct extent_buffer *leaf;
2441 struct btrfs_file_extent_item *fi;
2442 u64 extent_start = 0;
2443 u64 extent_num_bytes = 0;
2449 int pending_del_nr = 0;
2450 int pending_del_slot = 0;
2451 int extent_type = -1;
2453 u64 mask = root->sectorsize - 1;
2456 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2457 path = btrfs_alloc_path();
2461 /* FIXME, add redo link to tree so we don't leak on crash */
2462 key.objectid = inode->i_ino;
2463 key.offset = (u64)-1;
2466 btrfs_init_path(path);
2468 ret = drop_csum_leaves(trans, root, path, inode, new_size);
2472 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2477 /* there are no items in the tree for us to truncate, we're
2480 if (path->slots[0] == 0) {
2489 leaf = path->nodes[0];
2490 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2491 found_type = btrfs_key_type(&found_key);
2494 if (found_key.objectid != inode->i_ino)
2497 if (found_type < min_type)
2500 item_end = found_key.offset;
2501 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2502 fi = btrfs_item_ptr(leaf, path->slots[0],
2503 struct btrfs_file_extent_item);
2504 extent_type = btrfs_file_extent_type(leaf, fi);
2505 encoding = btrfs_file_extent_compression(leaf, fi);
2506 encoding |= btrfs_file_extent_encryption(leaf, fi);
2507 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2509 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2511 btrfs_file_extent_num_bytes(leaf, fi);
2512 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2513 item_end += btrfs_file_extent_inline_len(leaf,
2518 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2519 ret = btrfs_csum_truncate(trans, root, path,
2523 if (item_end < new_size) {
2524 if (found_type == BTRFS_DIR_ITEM_KEY) {
2525 found_type = BTRFS_INODE_ITEM_KEY;
2526 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2527 found_type = BTRFS_CSUM_ITEM_KEY;
2528 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2529 found_type = BTRFS_XATTR_ITEM_KEY;
2530 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2531 found_type = BTRFS_INODE_REF_KEY;
2532 } else if (found_type) {
2537 btrfs_set_key_type(&key, found_type);
2540 if (found_key.offset >= new_size)
2546 /* FIXME, shrink the extent if the ref count is only 1 */
2547 if (found_type != BTRFS_EXTENT_DATA_KEY)
2550 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2552 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2553 if (!del_item && !encoding) {
2554 u64 orig_num_bytes =
2555 btrfs_file_extent_num_bytes(leaf, fi);
2556 extent_num_bytes = new_size -
2557 found_key.offset + root->sectorsize - 1;
2558 extent_num_bytes = extent_num_bytes &
2559 ~((u64)root->sectorsize - 1);
2560 btrfs_set_file_extent_num_bytes(leaf, fi,
2562 num_dec = (orig_num_bytes -
2564 if (root->ref_cows && extent_start != 0)
2565 inode_sub_bytes(inode, num_dec);
2566 btrfs_mark_buffer_dirty(leaf);
2569 btrfs_file_extent_disk_num_bytes(leaf,
2571 /* FIXME blocksize != 4096 */
2572 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2573 if (extent_start != 0) {
2576 inode_sub_bytes(inode, num_dec);
2578 root_gen = btrfs_header_generation(leaf);
2579 root_owner = btrfs_header_owner(leaf);
2581 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2583 * we can't truncate inline items that have had
2587 btrfs_file_extent_compression(leaf, fi) == 0 &&
2588 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2589 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2590 u32 size = new_size - found_key.offset;
2592 if (root->ref_cows) {
2593 inode_sub_bytes(inode, item_end + 1 -
2597 btrfs_file_extent_calc_inline_size(size);
2598 ret = btrfs_truncate_item(trans, root, path,
2601 } else if (root->ref_cows) {
2602 inode_sub_bytes(inode, item_end + 1 -
2608 if (!pending_del_nr) {
2609 /* no pending yet, add ourselves */
2610 pending_del_slot = path->slots[0];
2612 } else if (pending_del_nr &&
2613 path->slots[0] + 1 == pending_del_slot) {
2614 /* hop on the pending chunk */
2616 pending_del_slot = path->slots[0];
2618 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2624 ret = btrfs_free_extent(trans, root, extent_start,
2626 leaf->start, root_owner,
2627 root_gen, inode->i_ino, 0);
2631 if (path->slots[0] == 0) {
2634 btrfs_release_path(root, path);
2639 if (pending_del_nr &&
2640 path->slots[0] + 1 != pending_del_slot) {
2641 struct btrfs_key debug;
2643 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2645 ret = btrfs_del_items(trans, root, path,
2650 btrfs_release_path(root, path);
2656 if (pending_del_nr) {
2657 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2660 btrfs_free_path(path);
2661 inode->i_sb->s_dirt = 1;
2666 * taken from block_truncate_page, but does cow as it zeros out
2667 * any bytes left in the last page in the file.
2669 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2671 struct inode *inode = mapping->host;
2672 struct btrfs_root *root = BTRFS_I(inode)->root;
2673 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2674 struct btrfs_ordered_extent *ordered;
2676 u32 blocksize = root->sectorsize;
2677 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2678 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2684 if ((offset & (blocksize - 1)) == 0)
2689 page = grab_cache_page(mapping, index);
2693 page_start = page_offset(page);
2694 page_end = page_start + PAGE_CACHE_SIZE - 1;
2696 if (!PageUptodate(page)) {
2697 ret = btrfs_readpage(NULL, page);
2699 if (page->mapping != mapping) {
2701 page_cache_release(page);
2704 if (!PageUptodate(page)) {
2709 wait_on_page_writeback(page);
2711 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2712 set_page_extent_mapped(page);
2714 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2716 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2718 page_cache_release(page);
2719 btrfs_start_ordered_extent(inode, ordered, 1);
2720 btrfs_put_ordered_extent(ordered);
2724 btrfs_set_extent_delalloc(inode, page_start, page_end);
2726 if (offset != PAGE_CACHE_SIZE) {
2728 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2729 flush_dcache_page(page);
2732 ClearPageChecked(page);
2733 set_page_dirty(page);
2734 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2738 page_cache_release(page);
2743 int btrfs_cont_expand(struct inode *inode, loff_t size)
2745 struct btrfs_trans_handle *trans;
2746 struct btrfs_root *root = BTRFS_I(inode)->root;
2747 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2748 struct extent_map *em;
2749 u64 mask = root->sectorsize - 1;
2750 u64 hole_start = (inode->i_size + mask) & ~mask;
2751 u64 block_end = (size + mask) & ~mask;
2757 if (size <= hole_start)
2760 err = btrfs_check_free_space(root, 1, 0);
2764 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2767 struct btrfs_ordered_extent *ordered;
2768 btrfs_wait_ordered_range(inode, hole_start,
2769 block_end - hole_start);
2770 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2771 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2774 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2775 btrfs_put_ordered_extent(ordered);
2778 trans = btrfs_start_transaction(root, 1);
2779 btrfs_set_trans_block_group(trans, inode);
2781 cur_offset = hole_start;
2783 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2784 block_end - cur_offset, 0);
2785 BUG_ON(IS_ERR(em) || !em);
2786 last_byte = min(extent_map_end(em), block_end);
2787 last_byte = (last_byte + mask) & ~mask;
2788 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2790 hole_size = last_byte - cur_offset;
2791 err = btrfs_drop_extents(trans, root, inode,
2793 cur_offset + hole_size,
2794 cur_offset, &hint_byte);
2797 err = btrfs_insert_file_extent(trans, root,
2798 inode->i_ino, cur_offset, 0,
2799 0, hole_size, 0, hole_size,
2801 btrfs_drop_extent_cache(inode, hole_start,
2804 free_extent_map(em);
2805 cur_offset = last_byte;
2806 if (err || cur_offset >= block_end)
2810 btrfs_end_transaction(trans, root);
2811 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2815 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2817 struct inode *inode = dentry->d_inode;
2820 err = inode_change_ok(inode, attr);
2824 if (S_ISREG(inode->i_mode) &&
2825 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2826 err = btrfs_cont_expand(inode, attr->ia_size);
2831 err = inode_setattr(inode, attr);
2833 if (!err && ((attr->ia_valid & ATTR_MODE)))
2834 err = btrfs_acl_chmod(inode);
2838 void btrfs_delete_inode(struct inode *inode)
2840 struct btrfs_trans_handle *trans;
2841 struct btrfs_root *root = BTRFS_I(inode)->root;
2845 truncate_inode_pages(&inode->i_data, 0);
2846 if (is_bad_inode(inode)) {
2847 btrfs_orphan_del(NULL, inode);
2850 btrfs_wait_ordered_range(inode, 0, (u64)-1);
2852 btrfs_i_size_write(inode, 0);
2853 trans = btrfs_start_transaction(root, 1);
2855 btrfs_set_trans_block_group(trans, inode);
2856 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2858 btrfs_orphan_del(NULL, inode);
2859 goto no_delete_lock;
2862 btrfs_orphan_del(trans, inode);
2864 nr = trans->blocks_used;
2867 btrfs_end_transaction(trans, root);
2868 btrfs_btree_balance_dirty(root, nr);
2872 nr = trans->blocks_used;
2873 btrfs_end_transaction(trans, root);
2874 btrfs_btree_balance_dirty(root, nr);
2880 * this returns the key found in the dir entry in the location pointer.
2881 * If no dir entries were found, location->objectid is 0.
2883 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2884 struct btrfs_key *location)
2886 const char *name = dentry->d_name.name;
2887 int namelen = dentry->d_name.len;
2888 struct btrfs_dir_item *di;
2889 struct btrfs_path *path;
2890 struct btrfs_root *root = BTRFS_I(dir)->root;
2893 path = btrfs_alloc_path();
2896 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2900 if (!di || IS_ERR(di)) {
2903 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2905 btrfs_free_path(path);
2908 location->objectid = 0;
2913 * when we hit a tree root in a directory, the btrfs part of the inode
2914 * needs to be changed to reflect the root directory of the tree root. This
2915 * is kind of like crossing a mount point.
2917 static int fixup_tree_root_location(struct btrfs_root *root,
2918 struct btrfs_key *location,
2919 struct btrfs_root **sub_root,
2920 struct dentry *dentry)
2922 struct btrfs_root_item *ri;
2924 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2926 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2929 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2930 dentry->d_name.name,
2931 dentry->d_name.len);
2932 if (IS_ERR(*sub_root))
2933 return PTR_ERR(*sub_root);
2935 ri = &(*sub_root)->root_item;
2936 location->objectid = btrfs_root_dirid(ri);
2937 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2938 location->offset = 0;
2943 static noinline void init_btrfs_i(struct inode *inode)
2945 struct btrfs_inode *bi = BTRFS_I(inode);
2948 bi->i_default_acl = NULL;
2952 bi->logged_trans = 0;
2953 bi->delalloc_bytes = 0;
2954 bi->disk_i_size = 0;
2956 bi->index_cnt = (u64)-1;
2957 bi->log_dirty_trans = 0;
2958 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2959 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2960 inode->i_mapping, GFP_NOFS);
2961 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2962 inode->i_mapping, GFP_NOFS);
2963 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2964 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2965 mutex_init(&BTRFS_I(inode)->csum_mutex);
2966 mutex_init(&BTRFS_I(inode)->extent_mutex);
2967 mutex_init(&BTRFS_I(inode)->log_mutex);
2970 static int btrfs_init_locked_inode(struct inode *inode, void *p)
2972 struct btrfs_iget_args *args = p;
2973 inode->i_ino = args->ino;
2974 init_btrfs_i(inode);
2975 BTRFS_I(inode)->root = args->root;
2979 static int btrfs_find_actor(struct inode *inode, void *opaque)
2981 struct btrfs_iget_args *args = opaque;
2982 return (args->ino == inode->i_ino &&
2983 args->root == BTRFS_I(inode)->root);
2986 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2987 struct btrfs_root *root, int wait)
2989 struct inode *inode;
2990 struct btrfs_iget_args args;
2991 args.ino = objectid;
2995 inode = ilookup5(s, objectid, btrfs_find_actor,
2998 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3004 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3005 struct btrfs_root *root)
3007 struct inode *inode;
3008 struct btrfs_iget_args args;
3009 args.ino = objectid;
3012 inode = iget5_locked(s, objectid, btrfs_find_actor,
3013 btrfs_init_locked_inode,
3018 /* Get an inode object given its location and corresponding root.
3019 * Returns in *is_new if the inode was read from disk
3021 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3022 struct btrfs_root *root, int *is_new)
3024 struct inode *inode;
3026 inode = btrfs_iget_locked(s, location->objectid, root);
3028 return ERR_PTR(-EACCES);
3030 if (inode->i_state & I_NEW) {
3031 BTRFS_I(inode)->root = root;
3032 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3033 btrfs_read_locked_inode(inode);
3034 unlock_new_inode(inode);
3045 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3046 struct nameidata *nd)
3048 struct inode * inode;
3049 struct btrfs_inode *bi = BTRFS_I(dir);
3050 struct btrfs_root *root = bi->root;
3051 struct btrfs_root *sub_root = root;
3052 struct btrfs_key location;
3053 int ret, new, do_orphan = 0;
3055 if (dentry->d_name.len > BTRFS_NAME_LEN)
3056 return ERR_PTR(-ENAMETOOLONG);
3058 ret = btrfs_inode_by_name(dir, dentry, &location);
3061 return ERR_PTR(ret);
3064 if (location.objectid) {
3065 ret = fixup_tree_root_location(root, &location, &sub_root,
3068 return ERR_PTR(ret);
3070 return ERR_PTR(-ENOENT);
3071 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3073 return ERR_CAST(inode);
3075 /* the inode and parent dir are two different roots */
3076 if (new && root != sub_root) {
3078 sub_root->inode = inode;
3083 if (unlikely(do_orphan))
3084 btrfs_orphan_cleanup(sub_root);
3086 return d_splice_alias(inode, dentry);
3089 static unsigned char btrfs_filetype_table[] = {
3090 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3093 static int btrfs_real_readdir(struct file *filp, void *dirent,
3096 struct inode *inode = filp->f_dentry->d_inode;
3097 struct btrfs_root *root = BTRFS_I(inode)->root;
3098 struct btrfs_item *item;
3099 struct btrfs_dir_item *di;
3100 struct btrfs_key key;
3101 struct btrfs_key found_key;
3102 struct btrfs_path *path;
3105 struct extent_buffer *leaf;
3108 unsigned char d_type;
3113 int key_type = BTRFS_DIR_INDEX_KEY;
3118 /* FIXME, use a real flag for deciding about the key type */
3119 if (root->fs_info->tree_root == root)
3120 key_type = BTRFS_DIR_ITEM_KEY;
3122 /* special case for "." */
3123 if (filp->f_pos == 0) {
3124 over = filldir(dirent, ".", 1,
3131 /* special case for .., just use the back ref */
3132 if (filp->f_pos == 1) {
3133 u64 pino = parent_ino(filp->f_path.dentry);
3134 over = filldir(dirent, "..", 2,
3141 path = btrfs_alloc_path();
3144 btrfs_set_key_type(&key, key_type);
3145 key.offset = filp->f_pos;
3146 key.objectid = inode->i_ino;
3148 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3154 leaf = path->nodes[0];
3155 nritems = btrfs_header_nritems(leaf);
3156 slot = path->slots[0];
3157 if (advance || slot >= nritems) {
3158 if (slot >= nritems - 1) {
3159 ret = btrfs_next_leaf(root, path);
3162 leaf = path->nodes[0];
3163 nritems = btrfs_header_nritems(leaf);
3164 slot = path->slots[0];
3171 item = btrfs_item_nr(leaf, slot);
3172 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3174 if (found_key.objectid != key.objectid)
3176 if (btrfs_key_type(&found_key) != key_type)
3178 if (found_key.offset < filp->f_pos)
3181 filp->f_pos = found_key.offset;
3183 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3185 di_total = btrfs_item_size(leaf, item);
3187 while (di_cur < di_total) {
3188 struct btrfs_key location;
3190 name_len = btrfs_dir_name_len(leaf, di);
3191 if (name_len <= sizeof(tmp_name)) {
3192 name_ptr = tmp_name;
3194 name_ptr = kmalloc(name_len, GFP_NOFS);
3200 read_extent_buffer(leaf, name_ptr,
3201 (unsigned long)(di + 1), name_len);
3203 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3204 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3205 over = filldir(dirent, name_ptr, name_len,
3206 found_key.offset, location.objectid,
3209 if (name_ptr != tmp_name)
3215 di_len = btrfs_dir_name_len(leaf, di) +
3216 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3218 di = (struct btrfs_dir_item *)((char *)di + di_len);
3222 /* Reached end of directory/root. Bump pos past the last item. */
3223 if (key_type == BTRFS_DIR_INDEX_KEY)
3224 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3230 btrfs_free_path(path);
3234 int btrfs_write_inode(struct inode *inode, int wait)
3236 struct btrfs_root *root = BTRFS_I(inode)->root;
3237 struct btrfs_trans_handle *trans;
3240 if (root->fs_info->closing > 1)
3244 trans = btrfs_join_transaction(root, 1);
3245 btrfs_set_trans_block_group(trans, inode);
3246 ret = btrfs_commit_transaction(trans, root);
3252 * This is somewhat expensive, updating the tree every time the
3253 * inode changes. But, it is most likely to find the inode in cache.
3254 * FIXME, needs more benchmarking...there are no reasons other than performance
3255 * to keep or drop this code.
3257 void btrfs_dirty_inode(struct inode *inode)
3259 struct btrfs_root *root = BTRFS_I(inode)->root;
3260 struct btrfs_trans_handle *trans;
3262 trans = btrfs_join_transaction(root, 1);
3263 btrfs_set_trans_block_group(trans, inode);
3264 btrfs_update_inode(trans, root, inode);
3265 btrfs_end_transaction(trans, root);
3269 * find the highest existing sequence number in a directory
3270 * and then set the in-memory index_cnt variable to reflect
3271 * free sequence numbers
3273 static int btrfs_set_inode_index_count(struct inode *inode)
3275 struct btrfs_root *root = BTRFS_I(inode)->root;
3276 struct btrfs_key key, found_key;
3277 struct btrfs_path *path;
3278 struct extent_buffer *leaf;
3281 key.objectid = inode->i_ino;
3282 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3283 key.offset = (u64)-1;
3285 path = btrfs_alloc_path();
3289 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3292 /* FIXME: we should be able to handle this */
3298 * MAGIC NUMBER EXPLANATION:
3299 * since we search a directory based on f_pos we have to start at 2
3300 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3301 * else has to start at 2
3303 if (path->slots[0] == 0) {
3304 BTRFS_I(inode)->index_cnt = 2;
3310 leaf = path->nodes[0];
3311 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3313 if (found_key.objectid != inode->i_ino ||
3314 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3315 BTRFS_I(inode)->index_cnt = 2;
3319 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3321 btrfs_free_path(path);
3326 * helper to find a free sequence number in a given directory. This current
3327 * code is very simple, later versions will do smarter things in the btree
3329 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
3334 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3335 ret = btrfs_set_inode_index_count(dir);
3341 *index = BTRFS_I(dir)->index_cnt;
3342 BTRFS_I(dir)->index_cnt++;
3347 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3348 struct btrfs_root *root,
3350 const char *name, int name_len,
3353 struct btrfs_block_group_cache *group,
3354 int mode, u64 *index)
3356 struct inode *inode;
3357 struct btrfs_inode_item *inode_item;
3358 struct btrfs_block_group_cache *new_inode_group;
3359 struct btrfs_key *location;
3360 struct btrfs_path *path;
3361 struct btrfs_inode_ref *ref;
3362 struct btrfs_key key[2];
3368 path = btrfs_alloc_path();
3371 inode = new_inode(root->fs_info->sb);
3373 return ERR_PTR(-ENOMEM);
3376 ret = btrfs_set_inode_index(dir, inode, index);
3378 return ERR_PTR(ret);
3381 * index_cnt is ignored for everything but a dir,
3382 * btrfs_get_inode_index_count has an explanation for the magic
3385 init_btrfs_i(inode);
3386 BTRFS_I(inode)->index_cnt = 2;
3387 BTRFS_I(inode)->root = root;
3388 BTRFS_I(inode)->generation = trans->transid;
3394 new_inode_group = btrfs_find_block_group(root, group, 0,
3395 BTRFS_BLOCK_GROUP_METADATA, owner);
3396 if (!new_inode_group) {
3397 printk("find_block group failed\n");
3398 new_inode_group = group;
3400 BTRFS_I(inode)->block_group = new_inode_group;
3402 key[0].objectid = objectid;
3403 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3406 key[1].objectid = objectid;
3407 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3408 key[1].offset = ref_objectid;
3410 sizes[0] = sizeof(struct btrfs_inode_item);
3411 sizes[1] = name_len + sizeof(*ref);
3413 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3417 if (objectid > root->highest_inode)
3418 root->highest_inode = objectid;
3420 inode->i_uid = current->fsuid;
3421 inode->i_gid = current->fsgid;
3422 inode->i_mode = mode;
3423 inode->i_ino = objectid;
3424 inode_set_bytes(inode, 0);
3425 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3426 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3427 struct btrfs_inode_item);
3428 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3430 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3431 struct btrfs_inode_ref);
3432 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3433 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3434 ptr = (unsigned long)(ref + 1);
3435 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3437 btrfs_mark_buffer_dirty(path->nodes[0]);
3438 btrfs_free_path(path);
3440 location = &BTRFS_I(inode)->location;
3441 location->objectid = objectid;
3442 location->offset = 0;
3443 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3445 insert_inode_hash(inode);
3449 BTRFS_I(dir)->index_cnt--;
3450 btrfs_free_path(path);
3451 return ERR_PTR(ret);
3454 static inline u8 btrfs_inode_type(struct inode *inode)
3456 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3460 * utility function to add 'inode' into 'parent_inode' with
3461 * a give name and a given sequence number.
3462 * if 'add_backref' is true, also insert a backref from the
3463 * inode to the parent directory.
3465 int btrfs_add_link(struct btrfs_trans_handle *trans,
3466 struct inode *parent_inode, struct inode *inode,
3467 const char *name, int name_len, int add_backref, u64 index)
3470 struct btrfs_key key;
3471 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3473 key.objectid = inode->i_ino;
3474 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3477 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3478 parent_inode->i_ino,
3479 &key, btrfs_inode_type(inode),
3483 ret = btrfs_insert_inode_ref(trans, root,
3486 parent_inode->i_ino,
3489 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3491 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3492 ret = btrfs_update_inode(trans, root, parent_inode);
3497 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3498 struct dentry *dentry, struct inode *inode,
3499 int backref, u64 index)
3501 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3502 inode, dentry->d_name.name,
3503 dentry->d_name.len, backref, index);
3505 d_instantiate(dentry, inode);
3513 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3514 int mode, dev_t rdev)
3516 struct btrfs_trans_handle *trans;
3517 struct btrfs_root *root = BTRFS_I(dir)->root;
3518 struct inode *inode = NULL;
3522 unsigned long nr = 0;
3525 if (!new_valid_dev(rdev))
3528 err = btrfs_check_free_space(root, 1, 0);
3532 trans = btrfs_start_transaction(root, 1);
3533 btrfs_set_trans_block_group(trans, dir);
3535 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3541 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3543 dentry->d_parent->d_inode->i_ino, objectid,
3544 BTRFS_I(dir)->block_group, mode, &index);
3545 err = PTR_ERR(inode);
3549 err = btrfs_init_acl(inode, dir);
3555 btrfs_set_trans_block_group(trans, inode);
3556 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3560 inode->i_op = &btrfs_special_inode_operations;
3561 init_special_inode(inode, inode->i_mode, rdev);
3562 btrfs_update_inode(trans, root, inode);
3564 dir->i_sb->s_dirt = 1;
3565 btrfs_update_inode_block_group(trans, inode);
3566 btrfs_update_inode_block_group(trans, dir);
3568 nr = trans->blocks_used;
3569 btrfs_end_transaction_throttle(trans, root);
3572 inode_dec_link_count(inode);
3575 btrfs_btree_balance_dirty(root, nr);
3579 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3580 int mode, struct nameidata *nd)
3582 struct btrfs_trans_handle *trans;
3583 struct btrfs_root *root = BTRFS_I(dir)->root;
3584 struct inode *inode = NULL;
3587 unsigned long nr = 0;
3591 err = btrfs_check_free_space(root, 1, 0);
3594 trans = btrfs_start_transaction(root, 1);
3595 btrfs_set_trans_block_group(trans, dir);
3597 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3603 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3605 dentry->d_parent->d_inode->i_ino,
3606 objectid, BTRFS_I(dir)->block_group, mode,
3608 err = PTR_ERR(inode);
3612 err = btrfs_init_acl(inode, dir);
3618 btrfs_set_trans_block_group(trans, inode);
3619 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3623 inode->i_mapping->a_ops = &btrfs_aops;
3624 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3625 inode->i_fop = &btrfs_file_operations;
3626 inode->i_op = &btrfs_file_inode_operations;
3627 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3629 dir->i_sb->s_dirt = 1;
3630 btrfs_update_inode_block_group(trans, inode);
3631 btrfs_update_inode_block_group(trans, dir);
3633 nr = trans->blocks_used;
3634 btrfs_end_transaction_throttle(trans, root);
3637 inode_dec_link_count(inode);
3640 btrfs_btree_balance_dirty(root, nr);
3644 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3645 struct dentry *dentry)
3647 struct btrfs_trans_handle *trans;
3648 struct btrfs_root *root = BTRFS_I(dir)->root;
3649 struct inode *inode = old_dentry->d_inode;
3651 unsigned long nr = 0;
3655 if (inode->i_nlink == 0)
3658 btrfs_inc_nlink(inode);
3659 err = btrfs_check_free_space(root, 1, 0);
3662 err = btrfs_set_inode_index(dir, inode, &index);
3666 trans = btrfs_start_transaction(root, 1);
3668 btrfs_set_trans_block_group(trans, dir);
3669 atomic_inc(&inode->i_count);
3671 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3676 dir->i_sb->s_dirt = 1;
3677 btrfs_update_inode_block_group(trans, dir);
3678 err = btrfs_update_inode(trans, root, inode);
3683 nr = trans->blocks_used;
3684 btrfs_end_transaction_throttle(trans, root);
3687 inode_dec_link_count(inode);
3690 btrfs_btree_balance_dirty(root, nr);
3694 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3696 struct inode *inode = NULL;
3697 struct btrfs_trans_handle *trans;
3698 struct btrfs_root *root = BTRFS_I(dir)->root;
3700 int drop_on_err = 0;
3703 unsigned long nr = 1;
3705 err = btrfs_check_free_space(root, 1, 0);
3709 trans = btrfs_start_transaction(root, 1);
3710 btrfs_set_trans_block_group(trans, dir);
3712 if (IS_ERR(trans)) {
3713 err = PTR_ERR(trans);
3717 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3723 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3725 dentry->d_parent->d_inode->i_ino, objectid,
3726 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3728 if (IS_ERR(inode)) {
3729 err = PTR_ERR(inode);
3735 err = btrfs_init_acl(inode, dir);
3739 inode->i_op = &btrfs_dir_inode_operations;
3740 inode->i_fop = &btrfs_dir_file_operations;
3741 btrfs_set_trans_block_group(trans, inode);
3743 btrfs_i_size_write(inode, 0);
3744 err = btrfs_update_inode(trans, root, inode);
3748 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3749 inode, dentry->d_name.name,
3750 dentry->d_name.len, 0, index);
3754 d_instantiate(dentry, inode);
3756 dir->i_sb->s_dirt = 1;
3757 btrfs_update_inode_block_group(trans, inode);
3758 btrfs_update_inode_block_group(trans, dir);
3761 nr = trans->blocks_used;
3762 btrfs_end_transaction_throttle(trans, root);
3767 btrfs_btree_balance_dirty(root, nr);
3771 /* helper for btfs_get_extent. Given an existing extent in the tree,
3772 * and an extent that you want to insert, deal with overlap and insert
3773 * the new extent into the tree.
3775 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3776 struct extent_map *existing,
3777 struct extent_map *em,
3778 u64 map_start, u64 map_len)
3782 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3783 start_diff = map_start - em->start;
3784 em->start = map_start;
3786 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3787 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3788 em->block_start += start_diff;
3789 em->block_len -= start_diff;
3791 return add_extent_mapping(em_tree, em);
3794 static noinline int uncompress_inline(struct btrfs_path *path,
3795 struct inode *inode, struct page *page,
3796 size_t pg_offset, u64 extent_offset,
3797 struct btrfs_file_extent_item *item)
3800 struct extent_buffer *leaf = path->nodes[0];
3803 unsigned long inline_size;
3806 WARN_ON(pg_offset != 0);
3807 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3808 inline_size = btrfs_file_extent_inline_item_len(leaf,
3809 btrfs_item_nr(leaf, path->slots[0]));
3810 tmp = kmalloc(inline_size, GFP_NOFS);
3811 ptr = btrfs_file_extent_inline_start(item);
3813 read_extent_buffer(leaf, tmp, ptr, inline_size);
3815 max_size = min(PAGE_CACHE_SIZE, max_size);
3816 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3817 inline_size, max_size);
3819 char *kaddr = kmap_atomic(page, KM_USER0);
3820 unsigned long copy_size = min_t(u64,
3821 PAGE_CACHE_SIZE - pg_offset,
3822 max_size - extent_offset);
3823 memset(kaddr + pg_offset, 0, copy_size);
3824 kunmap_atomic(kaddr, KM_USER0);
3831 * a bit scary, this does extent mapping from logical file offset to the disk.
3832 * the ugly parts come from merging extents from the disk with the
3833 * in-ram representation. This gets more complex because of the data=ordered code,
3834 * where the in-ram extents might be locked pending data=ordered completion.
3836 * This also copies inline extents directly into the page.
3838 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3839 size_t pg_offset, u64 start, u64 len,
3845 u64 extent_start = 0;
3847 u64 objectid = inode->i_ino;
3849 struct btrfs_path *path = NULL;
3850 struct btrfs_root *root = BTRFS_I(inode)->root;
3851 struct btrfs_file_extent_item *item;
3852 struct extent_buffer *leaf;
3853 struct btrfs_key found_key;
3854 struct extent_map *em = NULL;
3855 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3856 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3857 struct btrfs_trans_handle *trans = NULL;
3861 spin_lock(&em_tree->lock);
3862 em = lookup_extent_mapping(em_tree, start, len);
3864 em->bdev = root->fs_info->fs_devices->latest_bdev;
3865 spin_unlock(&em_tree->lock);
3868 if (em->start > start || em->start + em->len <= start)
3869 free_extent_map(em);
3870 else if (em->block_start == EXTENT_MAP_INLINE && page)
3871 free_extent_map(em);
3875 em = alloc_extent_map(GFP_NOFS);
3880 em->bdev = root->fs_info->fs_devices->latest_bdev;
3881 em->start = EXTENT_MAP_HOLE;
3882 em->orig_start = EXTENT_MAP_HOLE;
3884 em->block_len = (u64)-1;
3887 path = btrfs_alloc_path();
3891 ret = btrfs_lookup_file_extent(trans, root, path,
3892 objectid, start, trans != NULL);
3899 if (path->slots[0] == 0)
3904 leaf = path->nodes[0];
3905 item = btrfs_item_ptr(leaf, path->slots[0],
3906 struct btrfs_file_extent_item);
3907 /* are we inside the extent that was found? */
3908 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3909 found_type = btrfs_key_type(&found_key);
3910 if (found_key.objectid != objectid ||
3911 found_type != BTRFS_EXTENT_DATA_KEY) {
3915 found_type = btrfs_file_extent_type(leaf, item);
3916 extent_start = found_key.offset;
3917 compressed = btrfs_file_extent_compression(leaf, item);
3918 if (found_type == BTRFS_FILE_EXTENT_REG ||
3919 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3920 extent_end = extent_start +
3921 btrfs_file_extent_num_bytes(leaf, item);
3922 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3924 size = btrfs_file_extent_inline_len(leaf, item);
3925 extent_end = (extent_start + size + root->sectorsize - 1) &
3926 ~((u64)root->sectorsize - 1);
3929 if (start >= extent_end) {
3931 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3932 ret = btrfs_next_leaf(root, path);
3939 leaf = path->nodes[0];
3941 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3942 if (found_key.objectid != objectid ||
3943 found_key.type != BTRFS_EXTENT_DATA_KEY)
3945 if (start + len <= found_key.offset)
3948 em->len = found_key.offset - start;
3952 if (found_type == BTRFS_FILE_EXTENT_REG ||
3953 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3954 em->start = extent_start;
3955 em->len = extent_end - extent_start;
3956 em->orig_start = extent_start -
3957 btrfs_file_extent_offset(leaf, item);
3958 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3960 em->block_start = EXTENT_MAP_HOLE;
3964 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3965 em->block_start = bytenr;
3966 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3969 bytenr += btrfs_file_extent_offset(leaf, item);
3970 em->block_start = bytenr;
3971 em->block_len = em->len;
3972 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3973 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3976 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3980 size_t extent_offset;
3983 em->block_start = EXTENT_MAP_INLINE;
3984 if (!page || create) {
3985 em->start = extent_start;
3986 em->len = extent_end - extent_start;
3990 size = btrfs_file_extent_inline_len(leaf, item);
3991 extent_offset = page_offset(page) + pg_offset - extent_start;
3992 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3993 size - extent_offset);
3994 em->start = extent_start + extent_offset;
3995 em->len = (copy_size + root->sectorsize - 1) &
3996 ~((u64)root->sectorsize - 1);
3997 em->orig_start = EXTENT_MAP_INLINE;
3999 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4000 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4001 if (create == 0 && !PageUptodate(page)) {
4002 if (btrfs_file_extent_compression(leaf, item) ==
4003 BTRFS_COMPRESS_ZLIB) {
4004 ret = uncompress_inline(path, inode, page,
4006 extent_offset, item);
4010 read_extent_buffer(leaf, map + pg_offset, ptr,
4014 flush_dcache_page(page);
4015 } else if (create && PageUptodate(page)) {
4018 free_extent_map(em);
4020 btrfs_release_path(root, path);
4021 trans = btrfs_join_transaction(root, 1);
4025 write_extent_buffer(leaf, map + pg_offset, ptr,
4028 btrfs_mark_buffer_dirty(leaf);
4030 set_extent_uptodate(io_tree, em->start,
4031 extent_map_end(em) - 1, GFP_NOFS);
4034 printk("unkknown found_type %d\n", found_type);
4041 em->block_start = EXTENT_MAP_HOLE;
4042 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4044 btrfs_release_path(root, path);
4045 if (em->start > start || extent_map_end(em) <= start) {
4046 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
4052 spin_lock(&em_tree->lock);
4053 ret = add_extent_mapping(em_tree, em);
4054 /* it is possible that someone inserted the extent into the tree
4055 * while we had the lock dropped. It is also possible that
4056 * an overlapping map exists in the tree
4058 if (ret == -EEXIST) {
4059 struct extent_map *existing;
4063 existing = lookup_extent_mapping(em_tree, start, len);
4064 if (existing && (existing->start > start ||
4065 existing->start + existing->len <= start)) {
4066 free_extent_map(existing);
4070 existing = lookup_extent_mapping(em_tree, em->start,
4073 err = merge_extent_mapping(em_tree, existing,
4076 free_extent_map(existing);
4078 free_extent_map(em);
4083 printk("failing to insert %Lu %Lu\n",
4085 free_extent_map(em);
4089 free_extent_map(em);
4094 spin_unlock(&em_tree->lock);
4097 btrfs_free_path(path);
4099 ret = btrfs_end_transaction(trans, root);
4105 free_extent_map(em);
4107 return ERR_PTR(err);
4112 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4113 const struct iovec *iov, loff_t offset,
4114 unsigned long nr_segs)
4119 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4121 return extent_bmap(mapping, iblock, btrfs_get_extent);
4124 int btrfs_readpage(struct file *file, struct page *page)
4126 struct extent_io_tree *tree;
4127 tree = &BTRFS_I(page->mapping->host)->io_tree;
4128 return extent_read_full_page(tree, page, btrfs_get_extent);
4131 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4133 struct extent_io_tree *tree;
4136 if (current->flags & PF_MEMALLOC) {
4137 redirty_page_for_writepage(wbc, page);
4141 tree = &BTRFS_I(page->mapping->host)->io_tree;
4142 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4145 int btrfs_writepages(struct address_space *mapping,
4146 struct writeback_control *wbc)
4148 struct extent_io_tree *tree;
4150 tree = &BTRFS_I(mapping->host)->io_tree;
4151 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4155 btrfs_readpages(struct file *file, struct address_space *mapping,
4156 struct list_head *pages, unsigned nr_pages)
4158 struct extent_io_tree *tree;
4159 tree = &BTRFS_I(mapping->host)->io_tree;
4160 return extent_readpages(tree, mapping, pages, nr_pages,
4163 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4165 struct extent_io_tree *tree;
4166 struct extent_map_tree *map;
4169 tree = &BTRFS_I(page->mapping->host)->io_tree;
4170 map = &BTRFS_I(page->mapping->host)->extent_tree;
4171 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4173 ClearPagePrivate(page);
4174 set_page_private(page, 0);
4175 page_cache_release(page);
4180 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4182 if (PageWriteback(page) || PageDirty(page))
4184 return __btrfs_releasepage(page, gfp_flags);
4187 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4189 struct extent_io_tree *tree;
4190 struct btrfs_ordered_extent *ordered;
4191 u64 page_start = page_offset(page);
4192 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4194 wait_on_page_writeback(page);
4195 tree = &BTRFS_I(page->mapping->host)->io_tree;
4197 btrfs_releasepage(page, GFP_NOFS);
4201 lock_extent(tree, page_start, page_end, GFP_NOFS);
4202 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4206 * IO on this page will never be started, so we need
4207 * to account for any ordered extents now
4209 clear_extent_bit(tree, page_start, page_end,
4210 EXTENT_DIRTY | EXTENT_DELALLOC |
4211 EXTENT_LOCKED, 1, 0, GFP_NOFS);
4212 btrfs_finish_ordered_io(page->mapping->host,
4213 page_start, page_end);
4214 btrfs_put_ordered_extent(ordered);
4215 lock_extent(tree, page_start, page_end, GFP_NOFS);
4217 clear_extent_bit(tree, page_start, page_end,
4218 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4221 __btrfs_releasepage(page, GFP_NOFS);
4223 ClearPageChecked(page);
4224 if (PagePrivate(page)) {
4225 ClearPagePrivate(page);
4226 set_page_private(page, 0);
4227 page_cache_release(page);
4232 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4233 * called from a page fault handler when a page is first dirtied. Hence we must
4234 * be careful to check for EOF conditions here. We set the page up correctly
4235 * for a written page which means we get ENOSPC checking when writing into
4236 * holes and correct delalloc and unwritten extent mapping on filesystems that
4237 * support these features.
4239 * We are not allowed to take the i_mutex here so we have to play games to
4240 * protect against truncate races as the page could now be beyond EOF. Because
4241 * vmtruncate() writes the inode size before removing pages, once we have the
4242 * page lock we can determine safely if the page is beyond EOF. If it is not
4243 * beyond EOF, then the page is guaranteed safe against truncation until we
4246 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4248 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4249 struct btrfs_root *root = BTRFS_I(inode)->root;
4250 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4251 struct btrfs_ordered_extent *ordered;
4253 unsigned long zero_start;
4259 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4266 size = i_size_read(inode);
4267 page_start = page_offset(page);
4268 page_end = page_start + PAGE_CACHE_SIZE - 1;
4270 if ((page->mapping != inode->i_mapping) ||
4271 (page_start >= size)) {
4272 /* page got truncated out from underneath us */
4275 wait_on_page_writeback(page);
4277 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4278 set_page_extent_mapped(page);
4281 * we can't set the delalloc bits if there are pending ordered
4282 * extents. Drop our locks and wait for them to finish
4284 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4286 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4288 btrfs_start_ordered_extent(inode, ordered, 1);
4289 btrfs_put_ordered_extent(ordered);
4293 btrfs_set_extent_delalloc(inode, page_start, page_end);
4296 /* page is wholly or partially inside EOF */
4297 if (page_start + PAGE_CACHE_SIZE > size)
4298 zero_start = size & ~PAGE_CACHE_MASK;
4300 zero_start = PAGE_CACHE_SIZE;
4302 if (zero_start != PAGE_CACHE_SIZE) {
4304 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4305 flush_dcache_page(page);
4308 ClearPageChecked(page);
4309 set_page_dirty(page);
4310 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4318 static void btrfs_truncate(struct inode *inode)
4320 struct btrfs_root *root = BTRFS_I(inode)->root;
4322 struct btrfs_trans_handle *trans;
4324 u64 mask = root->sectorsize - 1;
4326 if (!S_ISREG(inode->i_mode))
4328 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4331 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4332 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4334 trans = btrfs_start_transaction(root, 1);
4335 btrfs_set_trans_block_group(trans, inode);
4336 btrfs_i_size_write(inode, inode->i_size);
4338 ret = btrfs_orphan_add(trans, inode);
4341 /* FIXME, add redo link to tree so we don't leak on crash */
4342 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4343 BTRFS_EXTENT_DATA_KEY);
4344 btrfs_update_inode(trans, root, inode);
4346 ret = btrfs_orphan_del(trans, inode);
4350 nr = trans->blocks_used;
4351 ret = btrfs_end_transaction_throttle(trans, root);
4353 btrfs_btree_balance_dirty(root, nr);
4357 * Invalidate a single dcache entry at the root of the filesystem.
4358 * Needed after creation of snapshot or subvolume.
4360 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
4363 struct dentry *alias, *entry;
4366 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
4370 /* change me if btrfs ever gets a d_hash operation */
4371 qstr.hash = full_name_hash(qstr.name, qstr.len);
4372 entry = d_lookup(alias, &qstr);
4375 d_invalidate(entry);
4382 * create a new subvolume directory/inode (helper for the ioctl).
4384 int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4385 struct btrfs_trans_handle *trans, u64 new_dirid,
4386 struct btrfs_block_group_cache *block_group)
4388 struct inode *inode;
4392 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4393 new_dirid, block_group, S_IFDIR | 0700, &index);
4395 return PTR_ERR(inode);
4396 inode->i_op = &btrfs_dir_inode_operations;
4397 inode->i_fop = &btrfs_dir_file_operations;
4398 new_root->inode = inode;
4401 btrfs_i_size_write(inode, 0);
4403 error = btrfs_update_inode(trans, new_root, inode);
4407 atomic_inc(&inode->i_count);
4408 d_instantiate(dentry, inode);
4412 /* helper function for file defrag and space balancing. This
4413 * forces readahead on a given range of bytes in an inode
4415 unsigned long btrfs_force_ra(struct address_space *mapping,
4416 struct file_ra_state *ra, struct file *file,
4417 pgoff_t offset, pgoff_t last_index)
4419 pgoff_t req_size = last_index - offset + 1;
4421 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4422 return offset + req_size;
4425 struct inode *btrfs_alloc_inode(struct super_block *sb)
4427 struct btrfs_inode *ei;
4429 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4433 ei->logged_trans = 0;
4434 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4435 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4436 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4437 INIT_LIST_HEAD(&ei->i_orphan);
4438 return &ei->vfs_inode;
4441 void btrfs_destroy_inode(struct inode *inode)
4443 struct btrfs_ordered_extent *ordered;
4444 WARN_ON(!list_empty(&inode->i_dentry));
4445 WARN_ON(inode->i_data.nrpages);
4447 if (BTRFS_I(inode)->i_acl &&
4448 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4449 posix_acl_release(BTRFS_I(inode)->i_acl);
4450 if (BTRFS_I(inode)->i_default_acl &&
4451 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4452 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4454 spin_lock(&BTRFS_I(inode)->root->list_lock);
4455 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4456 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4457 " list\n", inode->i_ino);
4460 spin_unlock(&BTRFS_I(inode)->root->list_lock);
4463 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4467 printk("found ordered extent %Lu %Lu\n",
4468 ordered->file_offset, ordered->len);
4469 btrfs_remove_ordered_extent(inode, ordered);
4470 btrfs_put_ordered_extent(ordered);
4471 btrfs_put_ordered_extent(ordered);
4474 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4475 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4478 static void init_once(void *foo)
4480 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4482 inode_init_once(&ei->vfs_inode);
4485 void btrfs_destroy_cachep(void)
4487 if (btrfs_inode_cachep)
4488 kmem_cache_destroy(btrfs_inode_cachep);
4489 if (btrfs_trans_handle_cachep)
4490 kmem_cache_destroy(btrfs_trans_handle_cachep);
4491 if (btrfs_transaction_cachep)
4492 kmem_cache_destroy(btrfs_transaction_cachep);
4493 if (btrfs_bit_radix_cachep)
4494 kmem_cache_destroy(btrfs_bit_radix_cachep);
4495 if (btrfs_path_cachep)
4496 kmem_cache_destroy(btrfs_path_cachep);
4499 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4500 unsigned long extra_flags,
4501 void (*ctor)(void *))
4503 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4504 SLAB_MEM_SPREAD | extra_flags), ctor);
4507 int btrfs_init_cachep(void)
4509 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4510 sizeof(struct btrfs_inode),
4512 if (!btrfs_inode_cachep)
4514 btrfs_trans_handle_cachep =
4515 btrfs_cache_create("btrfs_trans_handle_cache",
4516 sizeof(struct btrfs_trans_handle),
4518 if (!btrfs_trans_handle_cachep)
4520 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4521 sizeof(struct btrfs_transaction),
4523 if (!btrfs_transaction_cachep)
4525 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4526 sizeof(struct btrfs_path),
4528 if (!btrfs_path_cachep)
4530 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4531 SLAB_DESTROY_BY_RCU, NULL);
4532 if (!btrfs_bit_radix_cachep)
4536 btrfs_destroy_cachep();
4540 static int btrfs_getattr(struct vfsmount *mnt,
4541 struct dentry *dentry, struct kstat *stat)
4543 struct inode *inode = dentry->d_inode;
4544 generic_fillattr(inode, stat);
4545 stat->blksize = PAGE_CACHE_SIZE;
4546 stat->blocks = (inode_get_bytes(inode) +
4547 BTRFS_I(inode)->delalloc_bytes) >> 9;
4551 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4552 struct inode * new_dir,struct dentry *new_dentry)
4554 struct btrfs_trans_handle *trans;
4555 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4556 struct inode *new_inode = new_dentry->d_inode;
4557 struct inode *old_inode = old_dentry->d_inode;
4558 struct timespec ctime = CURRENT_TIME;
4562 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4563 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4567 ret = btrfs_check_free_space(root, 1, 0);
4571 trans = btrfs_start_transaction(root, 1);
4573 btrfs_set_trans_block_group(trans, new_dir);
4575 btrfs_inc_nlink(old_dentry->d_inode);
4576 old_dir->i_ctime = old_dir->i_mtime = ctime;
4577 new_dir->i_ctime = new_dir->i_mtime = ctime;
4578 old_inode->i_ctime = ctime;
4580 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4581 old_dentry->d_name.name,
4582 old_dentry->d_name.len);
4587 new_inode->i_ctime = CURRENT_TIME;
4588 ret = btrfs_unlink_inode(trans, root, new_dir,
4589 new_dentry->d_inode,
4590 new_dentry->d_name.name,
4591 new_dentry->d_name.len);
4594 if (new_inode->i_nlink == 0) {
4595 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4601 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
4605 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4606 old_inode, new_dentry->d_name.name,
4607 new_dentry->d_name.len, 1, index);
4612 btrfs_end_transaction_throttle(trans, root);
4618 * some fairly slow code that needs optimization. This walks the list
4619 * of all the inodes with pending delalloc and forces them to disk.
4621 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4623 struct list_head *head = &root->fs_info->delalloc_inodes;
4624 struct btrfs_inode *binode;
4625 struct inode *inode;
4626 unsigned long flags;
4628 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4629 while(!list_empty(head)) {
4630 binode = list_entry(head->next, struct btrfs_inode,
4632 inode = igrab(&binode->vfs_inode);
4634 list_del_init(&binode->delalloc_inodes);
4635 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4637 filemap_flush(inode->i_mapping);
4641 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4643 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4645 /* the filemap_flush will queue IO into the worker threads, but
4646 * we have to make sure the IO is actually started and that
4647 * ordered extents get created before we return
4649 atomic_inc(&root->fs_info->async_submit_draining);
4650 while(atomic_read(&root->fs_info->nr_async_submits) ||
4651 atomic_read(&root->fs_info->async_delalloc_pages)) {
4652 wait_event(root->fs_info->async_submit_wait,
4653 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4654 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4656 atomic_dec(&root->fs_info->async_submit_draining);
4660 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4661 const char *symname)
4663 struct btrfs_trans_handle *trans;
4664 struct btrfs_root *root = BTRFS_I(dir)->root;
4665 struct btrfs_path *path;
4666 struct btrfs_key key;
4667 struct inode *inode = NULL;
4675 struct btrfs_file_extent_item *ei;
4676 struct extent_buffer *leaf;
4677 unsigned long nr = 0;
4679 name_len = strlen(symname) + 1;
4680 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4681 return -ENAMETOOLONG;
4683 err = btrfs_check_free_space(root, 1, 0);
4687 trans = btrfs_start_transaction(root, 1);
4688 btrfs_set_trans_block_group(trans, dir);
4690 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4696 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4698 dentry->d_parent->d_inode->i_ino, objectid,
4699 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4701 err = PTR_ERR(inode);
4705 err = btrfs_init_acl(inode, dir);
4711 btrfs_set_trans_block_group(trans, inode);
4712 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4716 inode->i_mapping->a_ops = &btrfs_aops;
4717 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4718 inode->i_fop = &btrfs_file_operations;
4719 inode->i_op = &btrfs_file_inode_operations;
4720 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4722 dir->i_sb->s_dirt = 1;
4723 btrfs_update_inode_block_group(trans, inode);
4724 btrfs_update_inode_block_group(trans, dir);
4728 path = btrfs_alloc_path();
4730 key.objectid = inode->i_ino;
4732 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4733 datasize = btrfs_file_extent_calc_inline_size(name_len);
4734 err = btrfs_insert_empty_item(trans, root, path, &key,
4740 leaf = path->nodes[0];
4741 ei = btrfs_item_ptr(leaf, path->slots[0],
4742 struct btrfs_file_extent_item);
4743 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4744 btrfs_set_file_extent_type(leaf, ei,
4745 BTRFS_FILE_EXTENT_INLINE);
4746 btrfs_set_file_extent_encryption(leaf, ei, 0);
4747 btrfs_set_file_extent_compression(leaf, ei, 0);
4748 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4749 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4751 ptr = btrfs_file_extent_inline_start(ei);
4752 write_extent_buffer(leaf, symname, ptr, name_len);
4753 btrfs_mark_buffer_dirty(leaf);
4754 btrfs_free_path(path);
4756 inode->i_op = &btrfs_symlink_inode_operations;
4757 inode->i_mapping->a_ops = &btrfs_symlink_aops;
4758 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4759 inode_set_bytes(inode, name_len);
4760 btrfs_i_size_write(inode, name_len - 1);
4761 err = btrfs_update_inode(trans, root, inode);
4766 nr = trans->blocks_used;
4767 btrfs_end_transaction_throttle(trans, root);
4770 inode_dec_link_count(inode);
4773 btrfs_btree_balance_dirty(root, nr);
4777 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4778 u64 alloc_hint, int mode)
4780 struct btrfs_trans_handle *trans;
4781 struct btrfs_root *root = BTRFS_I(inode)->root;
4782 struct btrfs_key ins;
4784 u64 cur_offset = start;
4785 u64 num_bytes = end - start;
4788 trans = btrfs_join_transaction(root, 1);
4790 btrfs_set_trans_block_group(trans, inode);
4792 while (num_bytes > 0) {
4793 alloc_size = min(num_bytes, root->fs_info->max_extent);
4794 ret = btrfs_reserve_extent(trans, root, alloc_size,
4795 root->sectorsize, 0, alloc_hint,
4801 ret = insert_reserved_file_extent(trans, inode,
4802 cur_offset, ins.objectid,
4803 ins.offset, ins.offset,
4804 ins.offset, 0, 0, 0,
4805 BTRFS_FILE_EXTENT_PREALLOC);
4807 num_bytes -= ins.offset;
4808 cur_offset += ins.offset;
4809 alloc_hint = ins.objectid + ins.offset;
4812 if (cur_offset > start) {
4813 inode->i_ctime = CURRENT_TIME;
4814 btrfs_set_flag(inode, PREALLOC);
4815 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4816 cur_offset > i_size_read(inode))
4817 btrfs_i_size_write(inode, cur_offset);
4818 ret = btrfs_update_inode(trans, root, inode);
4822 btrfs_end_transaction(trans, root);
4826 static long btrfs_fallocate(struct inode *inode, int mode,
4827 loff_t offset, loff_t len)
4834 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4835 struct extent_map *em;
4838 alloc_start = offset & ~mask;
4839 alloc_end = (offset + len + mask) & ~mask;
4841 mutex_lock(&inode->i_mutex);
4842 if (alloc_start > inode->i_size) {
4843 ret = btrfs_cont_expand(inode, alloc_start);
4849 struct btrfs_ordered_extent *ordered;
4850 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4851 alloc_end - 1, GFP_NOFS);
4852 ordered = btrfs_lookup_first_ordered_extent(inode,
4855 ordered->file_offset + ordered->len > alloc_start &&
4856 ordered->file_offset < alloc_end) {
4857 btrfs_put_ordered_extent(ordered);
4858 unlock_extent(&BTRFS_I(inode)->io_tree,
4859 alloc_start, alloc_end - 1, GFP_NOFS);
4860 btrfs_wait_ordered_range(inode, alloc_start,
4861 alloc_end - alloc_start);
4864 btrfs_put_ordered_extent(ordered);
4869 cur_offset = alloc_start;
4871 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4872 alloc_end - cur_offset, 0);
4873 BUG_ON(IS_ERR(em) || !em);
4874 last_byte = min(extent_map_end(em), alloc_end);
4875 last_byte = (last_byte + mask) & ~mask;
4876 if (em->block_start == EXTENT_MAP_HOLE) {
4877 ret = prealloc_file_range(inode, cur_offset,
4878 last_byte, alloc_hint, mode);
4880 free_extent_map(em);
4884 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4885 alloc_hint = em->block_start;
4886 free_extent_map(em);
4888 cur_offset = last_byte;
4889 if (cur_offset >= alloc_end) {
4894 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4897 mutex_unlock(&inode->i_mutex);
4901 static int btrfs_set_page_dirty(struct page *page)
4903 return __set_page_dirty_nobuffers(page);
4906 static int btrfs_permission(struct inode *inode, int mask)
4908 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4910 return generic_permission(inode, mask, btrfs_check_acl);
4913 static struct inode_operations btrfs_dir_inode_operations = {
4914 .lookup = btrfs_lookup,
4915 .create = btrfs_create,
4916 .unlink = btrfs_unlink,
4918 .mkdir = btrfs_mkdir,
4919 .rmdir = btrfs_rmdir,
4920 .rename = btrfs_rename,
4921 .symlink = btrfs_symlink,
4922 .setattr = btrfs_setattr,
4923 .mknod = btrfs_mknod,
4924 .setxattr = btrfs_setxattr,
4925 .getxattr = btrfs_getxattr,
4926 .listxattr = btrfs_listxattr,
4927 .removexattr = btrfs_removexattr,
4928 .permission = btrfs_permission,
4930 static struct inode_operations btrfs_dir_ro_inode_operations = {
4931 .lookup = btrfs_lookup,
4932 .permission = btrfs_permission,
4934 static struct file_operations btrfs_dir_file_operations = {
4935 .llseek = generic_file_llseek,
4936 .read = generic_read_dir,
4937 .readdir = btrfs_real_readdir,
4938 .unlocked_ioctl = btrfs_ioctl,
4939 #ifdef CONFIG_COMPAT
4940 .compat_ioctl = btrfs_ioctl,
4942 .release = btrfs_release_file,
4943 .fsync = btrfs_sync_file,
4946 static struct extent_io_ops btrfs_extent_io_ops = {
4947 .fill_delalloc = run_delalloc_range,
4948 .submit_bio_hook = btrfs_submit_bio_hook,
4949 .merge_bio_hook = btrfs_merge_bio_hook,
4950 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4951 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4952 .writepage_start_hook = btrfs_writepage_start_hook,
4953 .readpage_io_failed_hook = btrfs_io_failed_hook,
4954 .set_bit_hook = btrfs_set_bit_hook,
4955 .clear_bit_hook = btrfs_clear_bit_hook,
4958 static struct address_space_operations btrfs_aops = {
4959 .readpage = btrfs_readpage,
4960 .writepage = btrfs_writepage,
4961 .writepages = btrfs_writepages,
4962 .readpages = btrfs_readpages,
4963 .sync_page = block_sync_page,
4965 .direct_IO = btrfs_direct_IO,
4966 .invalidatepage = btrfs_invalidatepage,
4967 .releasepage = btrfs_releasepage,
4968 .set_page_dirty = btrfs_set_page_dirty,
4971 static struct address_space_operations btrfs_symlink_aops = {
4972 .readpage = btrfs_readpage,
4973 .writepage = btrfs_writepage,
4974 .invalidatepage = btrfs_invalidatepage,
4975 .releasepage = btrfs_releasepage,
4978 static struct inode_operations btrfs_file_inode_operations = {
4979 .truncate = btrfs_truncate,
4980 .getattr = btrfs_getattr,
4981 .setattr = btrfs_setattr,
4982 .setxattr = btrfs_setxattr,
4983 .getxattr = btrfs_getxattr,
4984 .listxattr = btrfs_listxattr,
4985 .removexattr = btrfs_removexattr,
4986 .permission = btrfs_permission,
4987 .fallocate = btrfs_fallocate,
4989 static struct inode_operations btrfs_special_inode_operations = {
4990 .getattr = btrfs_getattr,
4991 .setattr = btrfs_setattr,
4992 .permission = btrfs_permission,
4993 .setxattr = btrfs_setxattr,
4994 .getxattr = btrfs_getxattr,
4995 .listxattr = btrfs_listxattr,
4996 .removexattr = btrfs_removexattr,
4998 static struct inode_operations btrfs_symlink_inode_operations = {
4999 .readlink = generic_readlink,
5000 .follow_link = page_follow_link_light,
5001 .put_link = page_put_link,
5002 .permission = btrfs_permission,