2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
50 #include "compression.h"
53 struct btrfs_iget_args {
55 struct btrfs_root *root;
58 static struct inode_operations btrfs_dir_inode_operations;
59 static struct inode_operations btrfs_symlink_inode_operations;
60 static struct inode_operations btrfs_dir_ro_inode_operations;
61 static struct inode_operations btrfs_special_inode_operations;
62 static struct inode_operations btrfs_file_inode_operations;
63 static struct address_space_operations btrfs_aops;
64 static struct address_space_operations btrfs_symlink_aops;
65 static struct file_operations btrfs_dir_file_operations;
66 static struct extent_io_ops btrfs_extent_io_ops;
68 static struct kmem_cache *btrfs_inode_cachep;
69 struct kmem_cache *btrfs_trans_handle_cachep;
70 struct kmem_cache *btrfs_transaction_cachep;
71 struct kmem_cache *btrfs_path_cachep;
74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
75 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
76 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
77 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
78 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
79 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
80 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
81 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
84 static void btrfs_truncate(struct inode *inode);
85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
86 static noinline int cow_file_range(struct inode *inode,
87 struct page *locked_page,
88 u64 start, u64 end, int *page_started,
89 unsigned long *nr_written, int unlock);
91 static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
95 err = btrfs_init_acl(inode, dir);
97 err = btrfs_xattr_security_init(inode, dir);
102 * this does all the hard work for inserting an inline extent into
103 * the btree. The caller should have done a btrfs_drop_extents so that
104 * no overlapping inline items exist in the btree
106 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root, struct inode *inode,
108 u64 start, size_t size, size_t compressed_size,
109 struct page **compressed_pages)
111 struct btrfs_key key;
112 struct btrfs_path *path;
113 struct extent_buffer *leaf;
114 struct page *page = NULL;
117 struct btrfs_file_extent_item *ei;
120 size_t cur_size = size;
122 unsigned long offset;
123 int use_compress = 0;
125 if (compressed_size && compressed_pages) {
127 cur_size = compressed_size;
130 path = btrfs_alloc_path();
134 path->leave_spinning = 1;
135 btrfs_set_trans_block_group(trans, inode);
137 key.objectid = inode->i_ino;
139 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
140 datasize = btrfs_file_extent_calc_inline_size(cur_size);
142 inode_add_bytes(inode, size);
143 ret = btrfs_insert_empty_item(trans, root, path, &key,
150 leaf = path->nodes[0];
151 ei = btrfs_item_ptr(leaf, path->slots[0],
152 struct btrfs_file_extent_item);
153 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
154 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
155 btrfs_set_file_extent_encryption(leaf, ei, 0);
156 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
157 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
158 ptr = btrfs_file_extent_inline_start(ei);
163 while (compressed_size > 0) {
164 cpage = compressed_pages[i];
165 cur_size = min_t(unsigned long, compressed_size,
168 kaddr = kmap_atomic(cpage, KM_USER0);
169 write_extent_buffer(leaf, kaddr, ptr, cur_size);
170 kunmap_atomic(kaddr, KM_USER0);
174 compressed_size -= cur_size;
176 btrfs_set_file_extent_compression(leaf, ei,
177 BTRFS_COMPRESS_ZLIB);
179 page = find_get_page(inode->i_mapping,
180 start >> PAGE_CACHE_SHIFT);
181 btrfs_set_file_extent_compression(leaf, ei, 0);
182 kaddr = kmap_atomic(page, KM_USER0);
183 offset = start & (PAGE_CACHE_SIZE - 1);
184 write_extent_buffer(leaf, kaddr + offset, ptr, size);
185 kunmap_atomic(kaddr, KM_USER0);
186 page_cache_release(page);
188 btrfs_mark_buffer_dirty(leaf);
189 btrfs_free_path(path);
191 BTRFS_I(inode)->disk_i_size = inode->i_size;
192 btrfs_update_inode(trans, root, inode);
195 btrfs_free_path(path);
201 * conditionally insert an inline extent into the file. This
202 * does the checks required to make sure the data is small enough
203 * to fit as an inline extent.
205 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
206 struct btrfs_root *root,
207 struct inode *inode, u64 start, u64 end,
208 size_t compressed_size,
209 struct page **compressed_pages)
211 u64 isize = i_size_read(inode);
212 u64 actual_end = min(end + 1, isize);
213 u64 inline_len = actual_end - start;
214 u64 aligned_end = (end + root->sectorsize - 1) &
215 ~((u64)root->sectorsize - 1);
217 u64 data_len = inline_len;
221 data_len = compressed_size;
224 actual_end >= PAGE_CACHE_SIZE ||
225 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227 (actual_end & (root->sectorsize - 1)) == 0) ||
229 data_len > root->fs_info->max_inline) {
233 ret = btrfs_drop_extents(trans, root, inode, start,
234 aligned_end, aligned_end, start,
238 if (isize > actual_end)
239 inline_len = min_t(u64, isize, actual_end);
240 ret = insert_inline_extent(trans, root, inode, start,
241 inline_len, compressed_size,
244 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
248 struct async_extent {
253 unsigned long nr_pages;
254 struct list_head list;
259 struct btrfs_root *root;
260 struct page *locked_page;
263 struct list_head extents;
264 struct btrfs_work work;
267 static noinline int add_async_extent(struct async_cow *cow,
268 u64 start, u64 ram_size,
271 unsigned long nr_pages)
273 struct async_extent *async_extent;
275 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
276 async_extent->start = start;
277 async_extent->ram_size = ram_size;
278 async_extent->compressed_size = compressed_size;
279 async_extent->pages = pages;
280 async_extent->nr_pages = nr_pages;
281 list_add_tail(&async_extent->list, &cow->extents);
286 * we create compressed extents in two phases. The first
287 * phase compresses a range of pages that have already been
288 * locked (both pages and state bits are locked).
290 * This is done inside an ordered work queue, and the compression
291 * is spread across many cpus. The actual IO submission is step
292 * two, and the ordered work queue takes care of making sure that
293 * happens in the same order things were put onto the queue by
294 * writepages and friends.
296 * If this code finds it can't get good compression, it puts an
297 * entry onto the work queue to write the uncompressed bytes. This
298 * makes sure that both compressed inodes and uncompressed inodes
299 * are written in the same order that pdflush sent them down.
301 static noinline int compress_file_range(struct inode *inode,
302 struct page *locked_page,
304 struct async_cow *async_cow,
307 struct btrfs_root *root = BTRFS_I(inode)->root;
308 struct btrfs_trans_handle *trans;
312 u64 blocksize = root->sectorsize;
314 u64 isize = i_size_read(inode);
316 struct page **pages = NULL;
317 unsigned long nr_pages;
318 unsigned long nr_pages_ret = 0;
319 unsigned long total_compressed = 0;
320 unsigned long total_in = 0;
321 unsigned long max_compressed = 128 * 1024;
322 unsigned long max_uncompressed = 128 * 1024;
328 actual_end = min_t(u64, isize, end + 1);
331 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
332 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
335 * we don't want to send crud past the end of i_size through
336 * compression, that's just a waste of CPU time. So, if the
337 * end of the file is before the start of our current
338 * requested range of bytes, we bail out to the uncompressed
339 * cleanup code that can deal with all of this.
341 * It isn't really the fastest way to fix things, but this is a
342 * very uncommon corner.
344 if (actual_end <= start)
345 goto cleanup_and_bail_uncompressed;
347 total_compressed = actual_end - start;
349 /* we want to make sure that amount of ram required to uncompress
350 * an extent is reasonable, so we limit the total size in ram
351 * of a compressed extent to 128k. This is a crucial number
352 * because it also controls how easily we can spread reads across
353 * cpus for decompression.
355 * We also want to make sure the amount of IO required to do
356 * a random read is reasonably small, so we limit the size of
357 * a compressed extent to 128k.
359 total_compressed = min(total_compressed, max_uncompressed);
360 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
361 num_bytes = max(blocksize, num_bytes);
362 disk_num_bytes = num_bytes;
367 * we do compression for mount -o compress and when the
368 * inode has not been flagged as nocompress. This flag can
369 * change at any time if we discover bad compression ratios.
371 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
372 btrfs_test_opt(root, COMPRESS)) {
374 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
376 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
377 total_compressed, pages,
378 nr_pages, &nr_pages_ret,
384 unsigned long offset = total_compressed &
385 (PAGE_CACHE_SIZE - 1);
386 struct page *page = pages[nr_pages_ret - 1];
389 /* zero the tail end of the last page, we might be
390 * sending it down to disk
393 kaddr = kmap_atomic(page, KM_USER0);
394 memset(kaddr + offset, 0,
395 PAGE_CACHE_SIZE - offset);
396 kunmap_atomic(kaddr, KM_USER0);
402 trans = btrfs_join_transaction(root, 1);
404 btrfs_set_trans_block_group(trans, inode);
406 /* lets try to make an inline extent */
407 if (ret || total_in < (actual_end - start)) {
408 /* we didn't compress the entire range, try
409 * to make an uncompressed inline extent.
411 ret = cow_file_range_inline(trans, root, inode,
412 start, end, 0, NULL);
414 /* try making a compressed inline extent */
415 ret = cow_file_range_inline(trans, root, inode,
417 total_compressed, pages);
419 btrfs_end_transaction(trans, root);
422 * inline extent creation worked, we don't need
423 * to create any more async work items. Unlock
424 * and free up our temp pages.
426 extent_clear_unlock_delalloc(inode,
427 &BTRFS_I(inode)->io_tree,
428 start, end, NULL, 1, 0,
437 * we aren't doing an inline extent round the compressed size
438 * up to a block size boundary so the allocator does sane
441 total_compressed = (total_compressed + blocksize - 1) &
445 * one last check to make sure the compression is really a
446 * win, compare the page count read with the blocks on disk
448 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
449 ~(PAGE_CACHE_SIZE - 1);
450 if (total_compressed >= total_in) {
453 disk_num_bytes = total_compressed;
454 num_bytes = total_in;
457 if (!will_compress && pages) {
459 * the compression code ran but failed to make things smaller,
460 * free any pages it allocated and our page pointer array
462 for (i = 0; i < nr_pages_ret; i++) {
463 WARN_ON(pages[i]->mapping);
464 page_cache_release(pages[i]);
468 total_compressed = 0;
471 /* flag the file so we don't compress in the future */
472 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
477 /* the async work queues will take care of doing actual
478 * allocation on disk for these compressed pages,
479 * and will submit them to the elevator.
481 add_async_extent(async_cow, start, num_bytes,
482 total_compressed, pages, nr_pages_ret);
484 if (start + num_bytes < end && start + num_bytes < actual_end) {
491 cleanup_and_bail_uncompressed:
493 * No compression, but we still need to write the pages in
494 * the file we've been given so far. redirty the locked
495 * page if it corresponds to our extent and set things up
496 * for the async work queue to run cow_file_range to do
497 * the normal delalloc dance
499 if (page_offset(locked_page) >= start &&
500 page_offset(locked_page) <= end) {
501 __set_page_dirty_nobuffers(locked_page);
502 /* unlocked later on in the async handlers */
504 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
512 for (i = 0; i < nr_pages_ret; i++) {
513 WARN_ON(pages[i]->mapping);
514 page_cache_release(pages[i]);
522 * phase two of compressed writeback. This is the ordered portion
523 * of the code, which only gets called in the order the work was
524 * queued. We walk all the async extents created by compress_file_range
525 * and send them down to the disk.
527 static noinline int submit_compressed_extents(struct inode *inode,
528 struct async_cow *async_cow)
530 struct async_extent *async_extent;
532 struct btrfs_trans_handle *trans;
533 struct btrfs_key ins;
534 struct extent_map *em;
535 struct btrfs_root *root = BTRFS_I(inode)->root;
536 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
537 struct extent_io_tree *io_tree;
540 if (list_empty(&async_cow->extents))
543 trans = btrfs_join_transaction(root, 1);
545 while (!list_empty(&async_cow->extents)) {
546 async_extent = list_entry(async_cow->extents.next,
547 struct async_extent, list);
548 list_del(&async_extent->list);
550 io_tree = &BTRFS_I(inode)->io_tree;
552 /* did the compression code fall back to uncompressed IO? */
553 if (!async_extent->pages) {
554 int page_started = 0;
555 unsigned long nr_written = 0;
557 lock_extent(io_tree, async_extent->start,
558 async_extent->start +
559 async_extent->ram_size - 1, GFP_NOFS);
561 /* allocate blocks */
562 cow_file_range(inode, async_cow->locked_page,
564 async_extent->start +
565 async_extent->ram_size - 1,
566 &page_started, &nr_written, 0);
569 * if page_started, cow_file_range inserted an
570 * inline extent and took care of all the unlocking
571 * and IO for us. Otherwise, we need to submit
572 * all those pages down to the drive.
575 extent_write_locked_range(io_tree,
576 inode, async_extent->start,
577 async_extent->start +
578 async_extent->ram_size - 1,
586 lock_extent(io_tree, async_extent->start,
587 async_extent->start + async_extent->ram_size - 1,
590 * here we're doing allocation and writeback of the
593 btrfs_drop_extent_cache(inode, async_extent->start,
594 async_extent->start +
595 async_extent->ram_size - 1, 0);
597 ret = btrfs_reserve_extent(trans, root,
598 async_extent->compressed_size,
599 async_extent->compressed_size,
603 em = alloc_extent_map(GFP_NOFS);
604 em->start = async_extent->start;
605 em->len = async_extent->ram_size;
606 em->orig_start = em->start;
608 em->block_start = ins.objectid;
609 em->block_len = ins.offset;
610 em->bdev = root->fs_info->fs_devices->latest_bdev;
611 set_bit(EXTENT_FLAG_PINNED, &em->flags);
612 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
615 write_lock(&em_tree->lock);
616 ret = add_extent_mapping(em_tree, em);
617 write_unlock(&em_tree->lock);
618 if (ret != -EEXIST) {
622 btrfs_drop_extent_cache(inode, async_extent->start,
623 async_extent->start +
624 async_extent->ram_size - 1, 0);
627 ret = btrfs_add_ordered_extent(inode, async_extent->start,
629 async_extent->ram_size,
631 BTRFS_ORDERED_COMPRESSED);
634 btrfs_end_transaction(trans, root);
637 * clear dirty, set writeback and unlock the pages.
639 extent_clear_unlock_delalloc(inode,
640 &BTRFS_I(inode)->io_tree,
642 async_extent->start +
643 async_extent->ram_size - 1,
644 NULL, 1, 1, 0, 1, 1, 0, 0);
646 ret = btrfs_submit_compressed_write(inode,
648 async_extent->ram_size,
650 ins.offset, async_extent->pages,
651 async_extent->nr_pages);
654 trans = btrfs_join_transaction(root, 1);
655 alloc_hint = ins.objectid + ins.offset;
660 btrfs_end_transaction(trans, root);
665 * when extent_io.c finds a delayed allocation range in the file,
666 * the call backs end up in this code. The basic idea is to
667 * allocate extents on disk for the range, and create ordered data structs
668 * in ram to track those extents.
670 * locked_page is the page that writepage had locked already. We use
671 * it to make sure we don't do extra locks or unlocks.
673 * *page_started is set to one if we unlock locked_page and do everything
674 * required to start IO on it. It may be clean and already done with
677 static noinline int cow_file_range(struct inode *inode,
678 struct page *locked_page,
679 u64 start, u64 end, int *page_started,
680 unsigned long *nr_written,
683 struct btrfs_root *root = BTRFS_I(inode)->root;
684 struct btrfs_trans_handle *trans;
687 unsigned long ram_size;
690 u64 blocksize = root->sectorsize;
692 u64 isize = i_size_read(inode);
693 struct btrfs_key ins;
694 struct extent_map *em;
695 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
698 trans = btrfs_join_transaction(root, 1);
700 btrfs_set_trans_block_group(trans, inode);
702 actual_end = min_t(u64, isize, end + 1);
704 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
705 num_bytes = max(blocksize, num_bytes);
706 disk_num_bytes = num_bytes;
710 /* lets try to make an inline extent */
711 ret = cow_file_range_inline(trans, root, inode,
712 start, end, 0, NULL);
714 extent_clear_unlock_delalloc(inode,
715 &BTRFS_I(inode)->io_tree,
716 start, end, NULL, 1, 1,
718 *nr_written = *nr_written +
719 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
726 BUG_ON(disk_num_bytes >
727 btrfs_super_total_bytes(&root->fs_info->super_copy));
729 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
731 while (disk_num_bytes > 0) {
732 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
733 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
734 root->sectorsize, 0, alloc_hint,
738 em = alloc_extent_map(GFP_NOFS);
740 em->orig_start = em->start;
742 ram_size = ins.offset;
743 em->len = ins.offset;
745 em->block_start = ins.objectid;
746 em->block_len = ins.offset;
747 em->bdev = root->fs_info->fs_devices->latest_bdev;
748 set_bit(EXTENT_FLAG_PINNED, &em->flags);
751 write_lock(&em_tree->lock);
752 ret = add_extent_mapping(em_tree, em);
753 write_unlock(&em_tree->lock);
754 if (ret != -EEXIST) {
758 btrfs_drop_extent_cache(inode, start,
759 start + ram_size - 1, 0);
762 cur_alloc_size = ins.offset;
763 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
764 ram_size, cur_alloc_size, 0);
767 if (root->root_key.objectid ==
768 BTRFS_DATA_RELOC_TREE_OBJECTID) {
769 ret = btrfs_reloc_clone_csums(inode, start,
774 if (disk_num_bytes < cur_alloc_size)
777 /* we're not doing compressed IO, don't unlock the first
778 * page (which the caller expects to stay locked), don't
779 * clear any dirty bits and don't set any writeback bits
781 * Do set the Private2 bit so we know this page was properly
782 * setup for writepage
784 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
785 start, start + ram_size - 1,
786 locked_page, unlock, 1,
788 disk_num_bytes -= cur_alloc_size;
789 num_bytes -= cur_alloc_size;
790 alloc_hint = ins.objectid + ins.offset;
791 start += cur_alloc_size;
795 btrfs_end_transaction(trans, root);
801 * work queue call back to started compression on a file and pages
803 static noinline void async_cow_start(struct btrfs_work *work)
805 struct async_cow *async_cow;
807 async_cow = container_of(work, struct async_cow, work);
809 compress_file_range(async_cow->inode, async_cow->locked_page,
810 async_cow->start, async_cow->end, async_cow,
813 async_cow->inode = NULL;
817 * work queue call back to submit previously compressed pages
819 static noinline void async_cow_submit(struct btrfs_work *work)
821 struct async_cow *async_cow;
822 struct btrfs_root *root;
823 unsigned long nr_pages;
825 async_cow = container_of(work, struct async_cow, work);
827 root = async_cow->root;
828 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
831 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
833 if (atomic_read(&root->fs_info->async_delalloc_pages) <
835 waitqueue_active(&root->fs_info->async_submit_wait))
836 wake_up(&root->fs_info->async_submit_wait);
838 if (async_cow->inode)
839 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 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
860 EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS);
861 while (start < end) {
862 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
863 async_cow->inode = inode;
864 async_cow->root = root;
865 async_cow->locked_page = locked_page;
866 async_cow->start = start;
868 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
871 cur_end = min(end, start + 512 * 1024 - 1);
873 async_cow->end = cur_end;
874 INIT_LIST_HEAD(&async_cow->extents);
876 async_cow->work.func = async_cow_start;
877 async_cow->work.ordered_func = async_cow_submit;
878 async_cow->work.ordered_free = async_cow_free;
879 async_cow->work.flags = 0;
881 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
883 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
885 btrfs_queue_worker(&root->fs_info->delalloc_workers,
888 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
889 wait_event(root->fs_info->async_submit_wait,
890 (atomic_read(&root->fs_info->async_delalloc_pages) <
894 while (atomic_read(&root->fs_info->async_submit_draining) &&
895 atomic_read(&root->fs_info->async_delalloc_pages)) {
896 wait_event(root->fs_info->async_submit_wait,
897 (atomic_read(&root->fs_info->async_delalloc_pages) ==
901 *nr_written += nr_pages;
908 static noinline int csum_exist_in_range(struct btrfs_root *root,
909 u64 bytenr, u64 num_bytes)
912 struct btrfs_ordered_sum *sums;
915 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
916 bytenr + num_bytes - 1, &list);
917 if (ret == 0 && list_empty(&list))
920 while (!list_empty(&list)) {
921 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
922 list_del(&sums->list);
929 * when nowcow writeback call back. This checks for snapshots or COW copies
930 * of the extents that exist in the file, and COWs the file as required.
932 * If no cow copies or snapshots exist, we write directly to the existing
935 static noinline int run_delalloc_nocow(struct inode *inode,
936 struct page *locked_page,
937 u64 start, u64 end, int *page_started, int force,
938 unsigned long *nr_written)
940 struct btrfs_root *root = BTRFS_I(inode)->root;
941 struct btrfs_trans_handle *trans;
942 struct extent_buffer *leaf;
943 struct btrfs_path *path;
944 struct btrfs_file_extent_item *fi;
945 struct btrfs_key found_key;
958 path = btrfs_alloc_path();
960 trans = btrfs_join_transaction(root, 1);
966 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
969 if (ret > 0 && path->slots[0] > 0 && check_prev) {
970 leaf = path->nodes[0];
971 btrfs_item_key_to_cpu(leaf, &found_key,
973 if (found_key.objectid == inode->i_ino &&
974 found_key.type == BTRFS_EXTENT_DATA_KEY)
979 leaf = path->nodes[0];
980 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
981 ret = btrfs_next_leaf(root, path);
986 leaf = path->nodes[0];
992 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
994 if (found_key.objectid > inode->i_ino ||
995 found_key.type > BTRFS_EXTENT_DATA_KEY ||
996 found_key.offset > end)
999 if (found_key.offset > cur_offset) {
1000 extent_end = found_key.offset;
1004 fi = btrfs_item_ptr(leaf, path->slots[0],
1005 struct btrfs_file_extent_item);
1006 extent_type = btrfs_file_extent_type(leaf, fi);
1008 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1009 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1010 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1011 extent_offset = btrfs_file_extent_offset(leaf, fi);
1012 extent_end = found_key.offset +
1013 btrfs_file_extent_num_bytes(leaf, fi);
1014 if (extent_end <= start) {
1018 if (disk_bytenr == 0)
1020 if (btrfs_file_extent_compression(leaf, fi) ||
1021 btrfs_file_extent_encryption(leaf, fi) ||
1022 btrfs_file_extent_other_encoding(leaf, fi))
1024 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1026 if (btrfs_extent_readonly(root, disk_bytenr))
1028 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1030 extent_offset, disk_bytenr))
1032 disk_bytenr += extent_offset;
1033 disk_bytenr += cur_offset - found_key.offset;
1034 num_bytes = min(end + 1, extent_end) - cur_offset;
1036 * force cow if csum exists in the range.
1037 * this ensure that csum for a given extent are
1038 * either valid or do not exist.
1040 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1043 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1044 extent_end = found_key.offset +
1045 btrfs_file_extent_inline_len(leaf, fi);
1046 extent_end = ALIGN(extent_end, root->sectorsize);
1051 if (extent_end <= start) {
1056 if (cow_start == (u64)-1)
1057 cow_start = cur_offset;
1058 cur_offset = extent_end;
1059 if (cur_offset > end)
1065 btrfs_release_path(root, path);
1066 if (cow_start != (u64)-1) {
1067 ret = cow_file_range(inode, locked_page, cow_start,
1068 found_key.offset - 1, page_started,
1071 cow_start = (u64)-1;
1074 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1075 struct extent_map *em;
1076 struct extent_map_tree *em_tree;
1077 em_tree = &BTRFS_I(inode)->extent_tree;
1078 em = alloc_extent_map(GFP_NOFS);
1079 em->start = cur_offset;
1080 em->orig_start = em->start;
1081 em->len = num_bytes;
1082 em->block_len = num_bytes;
1083 em->block_start = disk_bytenr;
1084 em->bdev = root->fs_info->fs_devices->latest_bdev;
1085 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1087 write_lock(&em_tree->lock);
1088 ret = add_extent_mapping(em_tree, em);
1089 write_unlock(&em_tree->lock);
1090 if (ret != -EEXIST) {
1091 free_extent_map(em);
1094 btrfs_drop_extent_cache(inode, em->start,
1095 em->start + em->len - 1, 0);
1097 type = BTRFS_ORDERED_PREALLOC;
1099 type = BTRFS_ORDERED_NOCOW;
1102 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1103 num_bytes, num_bytes, type);
1106 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1107 cur_offset, cur_offset + num_bytes - 1,
1108 locked_page, 1, 1, 1, 0, 0, 0, 1);
1109 cur_offset = extent_end;
1110 if (cur_offset > end)
1113 btrfs_release_path(root, path);
1115 if (cur_offset <= end && cow_start == (u64)-1)
1116 cow_start = cur_offset;
1117 if (cow_start != (u64)-1) {
1118 ret = cow_file_range(inode, locked_page, cow_start, end,
1119 page_started, nr_written, 1);
1123 ret = btrfs_end_transaction(trans, root);
1125 btrfs_free_path(path);
1130 * extent_io.c call back to do delayed allocation processing
1132 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1133 u64 start, u64 end, int *page_started,
1134 unsigned long *nr_written)
1137 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1140 ret = run_delalloc_nocow(inode, locked_page, start, end,
1141 page_started, 1, nr_written);
1142 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1143 ret = run_delalloc_nocow(inode, locked_page, start, end,
1144 page_started, 0, nr_written);
1145 else if (!btrfs_test_opt(root, COMPRESS))
1146 ret = cow_file_range(inode, locked_page, start, end,
1147 page_started, nr_written, 1);
1149 ret = cow_file_range_async(inode, locked_page, start, end,
1150 page_started, nr_written);
1155 * extent_io.c set_bit_hook, used to track delayed allocation
1156 * bytes in this file, and to maintain the list of inodes that
1157 * have pending delalloc work to be done.
1159 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1160 unsigned long old, unsigned long bits)
1163 * set_bit and clear bit hooks normally require _irqsave/restore
1164 * but in this case, we are only testeing for the DELALLOC
1165 * bit, which is only set or cleared with irqs on
1167 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1168 struct btrfs_root *root = BTRFS_I(inode)->root;
1169 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1170 spin_lock(&root->fs_info->delalloc_lock);
1171 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1172 root->fs_info->delalloc_bytes += end - start + 1;
1173 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1174 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1175 &root->fs_info->delalloc_inodes);
1177 spin_unlock(&root->fs_info->delalloc_lock);
1183 * extent_io.c clear_bit_hook, see set_bit_hook for why
1185 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1186 unsigned long old, unsigned long bits)
1189 * set_bit and clear bit hooks normally require _irqsave/restore
1190 * but in this case, we are only testeing for the DELALLOC
1191 * bit, which is only set or cleared with irqs on
1193 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1194 struct btrfs_root *root = BTRFS_I(inode)->root;
1196 spin_lock(&root->fs_info->delalloc_lock);
1197 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1198 printk(KERN_INFO "btrfs warning: delalloc account "
1200 (unsigned long long)end - start + 1,
1201 (unsigned long long)
1202 root->fs_info->delalloc_bytes);
1203 btrfs_delalloc_free_space(root, inode, (u64)-1);
1204 root->fs_info->delalloc_bytes = 0;
1205 BTRFS_I(inode)->delalloc_bytes = 0;
1207 btrfs_delalloc_free_space(root, inode,
1209 root->fs_info->delalloc_bytes -= end - start + 1;
1210 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1212 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1213 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1214 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1216 spin_unlock(&root->fs_info->delalloc_lock);
1222 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1223 * we don't create bios that span stripes or chunks
1225 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1226 size_t size, struct bio *bio,
1227 unsigned long bio_flags)
1229 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1230 struct btrfs_mapping_tree *map_tree;
1231 u64 logical = (u64)bio->bi_sector << 9;
1236 if (bio_flags & EXTENT_BIO_COMPRESSED)
1239 length = bio->bi_size;
1240 map_tree = &root->fs_info->mapping_tree;
1241 map_length = length;
1242 ret = btrfs_map_block(map_tree, READ, logical,
1243 &map_length, NULL, 0);
1245 if (map_length < length + size)
1251 * in order to insert checksums into the metadata in large chunks,
1252 * we wait until bio submission time. All the pages in the bio are
1253 * checksummed and sums are attached onto the ordered extent record.
1255 * At IO completion time the cums attached on the ordered extent record
1256 * are inserted into the btree
1258 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1259 struct bio *bio, int mirror_num,
1260 unsigned long bio_flags)
1262 struct btrfs_root *root = BTRFS_I(inode)->root;
1265 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1271 * in order to insert checksums into the metadata in large chunks,
1272 * we wait until bio submission time. All the pages in the bio are
1273 * checksummed and sums are attached onto the ordered extent record.
1275 * At IO completion time the cums attached on the ordered extent record
1276 * are inserted into the btree
1278 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1279 int mirror_num, unsigned long bio_flags)
1281 struct btrfs_root *root = BTRFS_I(inode)->root;
1282 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1286 * extent_io.c submission hook. This does the right thing for csum calculation
1287 * on write, or reading the csums from the tree before a read
1289 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1290 int mirror_num, unsigned long bio_flags)
1292 struct btrfs_root *root = BTRFS_I(inode)->root;
1296 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1298 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1301 if (!(rw & (1 << BIO_RW))) {
1302 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1303 return btrfs_submit_compressed_read(inode, bio,
1304 mirror_num, bio_flags);
1305 } else if (!skip_sum)
1306 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1308 } else if (!skip_sum) {
1309 /* csum items have already been cloned */
1310 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1312 /* we're doing a write, do the async checksumming */
1313 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1314 inode, rw, bio, mirror_num,
1315 bio_flags, __btrfs_submit_bio_start,
1316 __btrfs_submit_bio_done);
1320 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1324 * given a list of ordered sums record them in the inode. This happens
1325 * at IO completion time based on sums calculated at bio submission time.
1327 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1328 struct inode *inode, u64 file_offset,
1329 struct list_head *list)
1331 struct btrfs_ordered_sum *sum;
1333 btrfs_set_trans_block_group(trans, inode);
1335 list_for_each_entry(sum, list, list) {
1336 btrfs_csum_file_blocks(trans,
1337 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1342 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1344 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1346 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1350 /* see btrfs_writepage_start_hook for details on why this is required */
1351 struct btrfs_writepage_fixup {
1353 struct btrfs_work work;
1356 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1358 struct btrfs_writepage_fixup *fixup;
1359 struct btrfs_ordered_extent *ordered;
1361 struct inode *inode;
1365 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1369 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1370 ClearPageChecked(page);
1374 inode = page->mapping->host;
1375 page_start = page_offset(page);
1376 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1378 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1380 /* already ordered? We're done */
1381 if (PagePrivate2(page))
1384 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1386 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1387 page_end, GFP_NOFS);
1389 btrfs_start_ordered_extent(inode, ordered, 1);
1393 btrfs_set_extent_delalloc(inode, page_start, page_end);
1394 ClearPageChecked(page);
1396 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1399 page_cache_release(page);
1403 * There are a few paths in the higher layers of the kernel that directly
1404 * set the page dirty bit without asking the filesystem if it is a
1405 * good idea. This causes problems because we want to make sure COW
1406 * properly happens and the data=ordered rules are followed.
1408 * In our case any range that doesn't have the ORDERED bit set
1409 * hasn't been properly setup for IO. We kick off an async process
1410 * to fix it up. The async helper will wait for ordered extents, set
1411 * the delalloc bit and make it safe to write the page.
1413 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1415 struct inode *inode = page->mapping->host;
1416 struct btrfs_writepage_fixup *fixup;
1417 struct btrfs_root *root = BTRFS_I(inode)->root;
1419 /* this page is properly in the ordered list */
1420 if (TestClearPagePrivate2(page))
1423 if (PageChecked(page))
1426 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1430 SetPageChecked(page);
1431 page_cache_get(page);
1432 fixup->work.func = btrfs_writepage_fixup_worker;
1434 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439 struct inode *inode, u64 file_pos,
1440 u64 disk_bytenr, u64 disk_num_bytes,
1441 u64 num_bytes, u64 ram_bytes,
1443 u8 compression, u8 encryption,
1444 u16 other_encoding, int extent_type)
1446 struct btrfs_root *root = BTRFS_I(inode)->root;
1447 struct btrfs_file_extent_item *fi;
1448 struct btrfs_path *path;
1449 struct extent_buffer *leaf;
1450 struct btrfs_key ins;
1454 path = btrfs_alloc_path();
1457 path->leave_spinning = 1;
1460 * we may be replacing one extent in the tree with another.
1461 * The new extent is pinned in the extent map, and we don't want
1462 * to drop it from the cache until it is completely in the btree.
1464 * So, tell btrfs_drop_extents to leave this extent in the cache.
1465 * the caller is expected to unpin it and allow it to be merged
1468 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1469 file_pos + num_bytes, locked_end,
1470 file_pos, &hint, 0);
1473 ins.objectid = inode->i_ino;
1474 ins.offset = file_pos;
1475 ins.type = BTRFS_EXTENT_DATA_KEY;
1476 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1478 leaf = path->nodes[0];
1479 fi = btrfs_item_ptr(leaf, path->slots[0],
1480 struct btrfs_file_extent_item);
1481 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1482 btrfs_set_file_extent_type(leaf, fi, extent_type);
1483 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1484 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1485 btrfs_set_file_extent_offset(leaf, fi, 0);
1486 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1487 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1488 btrfs_set_file_extent_compression(leaf, fi, compression);
1489 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1490 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1492 btrfs_unlock_up_safe(path, 1);
1493 btrfs_set_lock_blocking(leaf);
1495 btrfs_mark_buffer_dirty(leaf);
1497 inode_add_bytes(inode, num_bytes);
1499 ins.objectid = disk_bytenr;
1500 ins.offset = disk_num_bytes;
1501 ins.type = BTRFS_EXTENT_ITEM_KEY;
1502 ret = btrfs_alloc_reserved_file_extent(trans, root,
1503 root->root_key.objectid,
1504 inode->i_ino, file_pos, &ins);
1506 btrfs_free_path(path);
1512 * helper function for btrfs_finish_ordered_io, this
1513 * just reads in some of the csum leaves to prime them into ram
1514 * before we start the transaction. It limits the amount of btree
1515 * reads required while inside the transaction.
1517 static noinline void reada_csum(struct btrfs_root *root,
1518 struct btrfs_path *path,
1519 struct btrfs_ordered_extent *ordered_extent)
1521 struct btrfs_ordered_sum *sum;
1524 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1526 bytenr = sum->sums[0].bytenr;
1529 * we don't care about the results, the point of this search is
1530 * just to get the btree leaves into ram
1532 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1535 /* as ordered data IO finishes, this gets called so we can finish
1536 * an ordered extent if the range of bytes in the file it covers are
1539 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1541 struct btrfs_root *root = BTRFS_I(inode)->root;
1542 struct btrfs_trans_handle *trans;
1543 struct btrfs_ordered_extent *ordered_extent = NULL;
1544 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1545 struct btrfs_path *path;
1549 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1554 * before we join the transaction, try to do some of our IO.
1555 * This will limit the amount of IO that we have to do with
1556 * the transaction running. We're unlikely to need to do any
1557 * IO if the file extents are new, the disk_i_size checks
1558 * covers the most common case.
1560 if (start < BTRFS_I(inode)->disk_i_size) {
1561 path = btrfs_alloc_path();
1563 ret = btrfs_lookup_file_extent(NULL, root, path,
1566 ordered_extent = btrfs_lookup_ordered_extent(inode,
1568 if (!list_empty(&ordered_extent->list)) {
1569 btrfs_release_path(root, path);
1570 reada_csum(root, path, ordered_extent);
1572 btrfs_free_path(path);
1576 trans = btrfs_join_transaction(root, 1);
1578 if (!ordered_extent)
1579 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1580 BUG_ON(!ordered_extent);
1581 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1584 lock_extent(io_tree, ordered_extent->file_offset,
1585 ordered_extent->file_offset + ordered_extent->len - 1,
1588 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1590 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1592 ret = btrfs_mark_extent_written(trans, root, inode,
1593 ordered_extent->file_offset,
1594 ordered_extent->file_offset +
1595 ordered_extent->len);
1598 ret = insert_reserved_file_extent(trans, inode,
1599 ordered_extent->file_offset,
1600 ordered_extent->start,
1601 ordered_extent->disk_len,
1602 ordered_extent->len,
1603 ordered_extent->len,
1604 ordered_extent->file_offset +
1605 ordered_extent->len,
1607 BTRFS_FILE_EXTENT_REG);
1608 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1609 ordered_extent->file_offset,
1610 ordered_extent->len);
1613 unlock_extent(io_tree, ordered_extent->file_offset,
1614 ordered_extent->file_offset + ordered_extent->len - 1,
1617 add_pending_csums(trans, inode, ordered_extent->file_offset,
1618 &ordered_extent->list);
1620 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1621 btrfs_ordered_update_i_size(inode, ordered_extent);
1622 btrfs_update_inode(trans, root, inode);
1623 btrfs_remove_ordered_extent(inode, ordered_extent);
1624 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1627 btrfs_put_ordered_extent(ordered_extent);
1628 /* once for the tree */
1629 btrfs_put_ordered_extent(ordered_extent);
1631 btrfs_end_transaction(trans, root);
1635 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1636 struct extent_state *state, int uptodate)
1638 ClearPagePrivate2(page);
1639 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1643 * When IO fails, either with EIO or csum verification fails, we
1644 * try other mirrors that might have a good copy of the data. This
1645 * io_failure_record is used to record state as we go through all the
1646 * mirrors. If another mirror has good data, the page is set up to date
1647 * and things continue. If a good mirror can't be found, the original
1648 * bio end_io callback is called to indicate things have failed.
1650 struct io_failure_record {
1655 unsigned long bio_flags;
1659 static int btrfs_io_failed_hook(struct bio *failed_bio,
1660 struct page *page, u64 start, u64 end,
1661 struct extent_state *state)
1663 struct io_failure_record *failrec = NULL;
1665 struct extent_map *em;
1666 struct inode *inode = page->mapping->host;
1667 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1668 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1675 ret = get_state_private(failure_tree, start, &private);
1677 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1680 failrec->start = start;
1681 failrec->len = end - start + 1;
1682 failrec->last_mirror = 0;
1683 failrec->bio_flags = 0;
1685 read_lock(&em_tree->lock);
1686 em = lookup_extent_mapping(em_tree, start, failrec->len);
1687 if (em->start > start || em->start + em->len < start) {
1688 free_extent_map(em);
1691 read_unlock(&em_tree->lock);
1693 if (!em || IS_ERR(em)) {
1697 logical = start - em->start;
1698 logical = em->block_start + logical;
1699 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1700 logical = em->block_start;
1701 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1703 failrec->logical = logical;
1704 free_extent_map(em);
1705 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1706 EXTENT_DIRTY, GFP_NOFS);
1707 set_state_private(failure_tree, start,
1708 (u64)(unsigned long)failrec);
1710 failrec = (struct io_failure_record *)(unsigned long)private;
1712 num_copies = btrfs_num_copies(
1713 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1714 failrec->logical, failrec->len);
1715 failrec->last_mirror++;
1717 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1718 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1721 if (state && state->start != failrec->start)
1723 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1725 if (!state || failrec->last_mirror > num_copies) {
1726 set_state_private(failure_tree, failrec->start, 0);
1727 clear_extent_bits(failure_tree, failrec->start,
1728 failrec->start + failrec->len - 1,
1729 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1733 bio = bio_alloc(GFP_NOFS, 1);
1734 bio->bi_private = state;
1735 bio->bi_end_io = failed_bio->bi_end_io;
1736 bio->bi_sector = failrec->logical >> 9;
1737 bio->bi_bdev = failed_bio->bi_bdev;
1740 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1741 if (failed_bio->bi_rw & (1 << BIO_RW))
1746 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1747 failrec->last_mirror,
1748 failrec->bio_flags);
1753 * each time an IO finishes, we do a fast check in the IO failure tree
1754 * to see if we need to process or clean up an io_failure_record
1756 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1759 u64 private_failure;
1760 struct io_failure_record *failure;
1764 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1765 (u64)-1, 1, EXTENT_DIRTY)) {
1766 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1767 start, &private_failure);
1769 failure = (struct io_failure_record *)(unsigned long)
1771 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1773 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1775 failure->start + failure->len - 1,
1776 EXTENT_DIRTY | EXTENT_LOCKED,
1785 * when reads are done, we need to check csums to verify the data is correct
1786 * if there's a match, we allow the bio to finish. If not, we go through
1787 * the io_failure_record routines to find good copies
1789 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1790 struct extent_state *state)
1792 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1793 struct inode *inode = page->mapping->host;
1794 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1796 u64 private = ~(u32)0;
1798 struct btrfs_root *root = BTRFS_I(inode)->root;
1801 if (PageChecked(page)) {
1802 ClearPageChecked(page);
1806 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1809 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1810 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1811 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1816 if (state && state->start == start) {
1817 private = state->private;
1820 ret = get_state_private(io_tree, start, &private);
1822 kaddr = kmap_atomic(page, KM_USER0);
1826 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1827 btrfs_csum_final(csum, (char *)&csum);
1828 if (csum != private)
1831 kunmap_atomic(kaddr, KM_USER0);
1833 /* if the io failure tree for this inode is non-empty,
1834 * check to see if we've recovered from a failed IO
1836 btrfs_clean_io_failures(inode, start);
1840 if (printk_ratelimit()) {
1841 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1842 "private %llu\n", page->mapping->host->i_ino,
1843 (unsigned long long)start, csum,
1844 (unsigned long long)private);
1846 memset(kaddr + offset, 1, end - start + 1);
1847 flush_dcache_page(page);
1848 kunmap_atomic(kaddr, KM_USER0);
1855 * This creates an orphan entry for the given inode in case something goes
1856 * wrong in the middle of an unlink/truncate.
1858 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1860 struct btrfs_root *root = BTRFS_I(inode)->root;
1863 spin_lock(&root->list_lock);
1865 /* already on the orphan list, we're good */
1866 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1867 spin_unlock(&root->list_lock);
1871 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1873 spin_unlock(&root->list_lock);
1876 * insert an orphan item to track this unlinked/truncated file
1878 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1884 * We have done the truncate/delete so we can go ahead and remove the orphan
1885 * item for this particular inode.
1887 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1889 struct btrfs_root *root = BTRFS_I(inode)->root;
1892 spin_lock(&root->list_lock);
1894 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1895 spin_unlock(&root->list_lock);
1899 list_del_init(&BTRFS_I(inode)->i_orphan);
1901 spin_unlock(&root->list_lock);
1905 spin_unlock(&root->list_lock);
1907 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1913 * this cleans up any orphans that may be left on the list from the last use
1916 void btrfs_orphan_cleanup(struct btrfs_root *root)
1918 struct btrfs_path *path;
1919 struct extent_buffer *leaf;
1920 struct btrfs_item *item;
1921 struct btrfs_key key, found_key;
1922 struct btrfs_trans_handle *trans;
1923 struct inode *inode;
1924 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1926 path = btrfs_alloc_path();
1931 key.objectid = BTRFS_ORPHAN_OBJECTID;
1932 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1933 key.offset = (u64)-1;
1937 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1939 printk(KERN_ERR "Error searching slot for orphan: %d"
1945 * if ret == 0 means we found what we were searching for, which
1946 * is weird, but possible, so only screw with path if we didnt
1947 * find the key and see if we have stuff that matches
1950 if (path->slots[0] == 0)
1955 /* pull out the item */
1956 leaf = path->nodes[0];
1957 item = btrfs_item_nr(leaf, path->slots[0]);
1958 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1960 /* make sure the item matches what we want */
1961 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1963 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1966 /* release the path since we're done with it */
1967 btrfs_release_path(root, path);
1970 * this is where we are basically btrfs_lookup, without the
1971 * crossing root thing. we store the inode number in the
1972 * offset of the orphan item.
1974 found_key.objectid = found_key.offset;
1975 found_key.type = BTRFS_INODE_ITEM_KEY;
1976 found_key.offset = 0;
1977 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1982 * add this inode to the orphan list so btrfs_orphan_del does
1983 * the proper thing when we hit it
1985 spin_lock(&root->list_lock);
1986 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1987 spin_unlock(&root->list_lock);
1990 * if this is a bad inode, means we actually succeeded in
1991 * removing the inode, but not the orphan record, which means
1992 * we need to manually delete the orphan since iput will just
1993 * do a destroy_inode
1995 if (is_bad_inode(inode)) {
1996 trans = btrfs_start_transaction(root, 1);
1997 btrfs_orphan_del(trans, inode);
1998 btrfs_end_transaction(trans, root);
2003 /* if we have links, this was a truncate, lets do that */
2004 if (inode->i_nlink) {
2006 btrfs_truncate(inode);
2011 /* this will do delete_inode and everything for us */
2016 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2018 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2020 btrfs_free_path(path);
2024 * very simple check to peek ahead in the leaf looking for xattrs. If we
2025 * don't find any xattrs, we know there can't be any acls.
2027 * slot is the slot the inode is in, objectid is the objectid of the inode
2029 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2030 int slot, u64 objectid)
2032 u32 nritems = btrfs_header_nritems(leaf);
2033 struct btrfs_key found_key;
2037 while (slot < nritems) {
2038 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2040 /* we found a different objectid, there must not be acls */
2041 if (found_key.objectid != objectid)
2044 /* we found an xattr, assume we've got an acl */
2045 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2049 * we found a key greater than an xattr key, there can't
2050 * be any acls later on
2052 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2059 * it goes inode, inode backrefs, xattrs, extents,
2060 * so if there are a ton of hard links to an inode there can
2061 * be a lot of backrefs. Don't waste time searching too hard,
2062 * this is just an optimization
2067 /* we hit the end of the leaf before we found an xattr or
2068 * something larger than an xattr. We have to assume the inode
2075 * read an inode from the btree into the in-memory inode
2077 static void btrfs_read_locked_inode(struct inode *inode)
2079 struct btrfs_path *path;
2080 struct extent_buffer *leaf;
2081 struct btrfs_inode_item *inode_item;
2082 struct btrfs_timespec *tspec;
2083 struct btrfs_root *root = BTRFS_I(inode)->root;
2084 struct btrfs_key location;
2086 u64 alloc_group_block;
2090 path = btrfs_alloc_path();
2092 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2094 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2098 leaf = path->nodes[0];
2099 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2100 struct btrfs_inode_item);
2102 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2103 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2104 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2105 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2106 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2108 tspec = btrfs_inode_atime(inode_item);
2109 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2110 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2112 tspec = btrfs_inode_mtime(inode_item);
2113 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2114 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2116 tspec = btrfs_inode_ctime(inode_item);
2117 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2118 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2120 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2121 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2122 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2123 inode->i_generation = BTRFS_I(inode)->generation;
2125 rdev = btrfs_inode_rdev(leaf, inode_item);
2127 BTRFS_I(inode)->index_cnt = (u64)-1;
2128 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2130 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2133 * try to precache a NULL acl entry for files that don't have
2134 * any xattrs or acls
2136 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2138 cache_no_acl(inode);
2140 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2141 alloc_group_block, 0);
2142 btrfs_free_path(path);
2145 switch (inode->i_mode & S_IFMT) {
2147 inode->i_mapping->a_ops = &btrfs_aops;
2148 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2149 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2150 inode->i_fop = &btrfs_file_operations;
2151 inode->i_op = &btrfs_file_inode_operations;
2154 inode->i_fop = &btrfs_dir_file_operations;
2155 if (root == root->fs_info->tree_root)
2156 inode->i_op = &btrfs_dir_ro_inode_operations;
2158 inode->i_op = &btrfs_dir_inode_operations;
2161 inode->i_op = &btrfs_symlink_inode_operations;
2162 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2163 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2166 inode->i_op = &btrfs_special_inode_operations;
2167 init_special_inode(inode, inode->i_mode, rdev);
2171 btrfs_update_iflags(inode);
2175 btrfs_free_path(path);
2176 make_bad_inode(inode);
2180 * given a leaf and an inode, copy the inode fields into the leaf
2182 static void fill_inode_item(struct btrfs_trans_handle *trans,
2183 struct extent_buffer *leaf,
2184 struct btrfs_inode_item *item,
2185 struct inode *inode)
2187 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2188 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2189 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2190 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2191 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2193 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2194 inode->i_atime.tv_sec);
2195 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2196 inode->i_atime.tv_nsec);
2198 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2199 inode->i_mtime.tv_sec);
2200 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2201 inode->i_mtime.tv_nsec);
2203 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2204 inode->i_ctime.tv_sec);
2205 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2206 inode->i_ctime.tv_nsec);
2208 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2209 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2210 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2211 btrfs_set_inode_transid(leaf, item, trans->transid);
2212 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2213 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2214 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2218 * copy everything in the in-memory inode into the btree.
2220 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2221 struct btrfs_root *root, struct inode *inode)
2223 struct btrfs_inode_item *inode_item;
2224 struct btrfs_path *path;
2225 struct extent_buffer *leaf;
2228 path = btrfs_alloc_path();
2230 path->leave_spinning = 1;
2231 ret = btrfs_lookup_inode(trans, root, path,
2232 &BTRFS_I(inode)->location, 1);
2239 btrfs_unlock_up_safe(path, 1);
2240 leaf = path->nodes[0];
2241 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2242 struct btrfs_inode_item);
2244 fill_inode_item(trans, leaf, inode_item, inode);
2245 btrfs_mark_buffer_dirty(leaf);
2246 btrfs_set_inode_last_trans(trans, inode);
2249 btrfs_free_path(path);
2255 * unlink helper that gets used here in inode.c and in the tree logging
2256 * recovery code. It remove a link in a directory with a given name, and
2257 * also drops the back refs in the inode to the directory
2259 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2260 struct btrfs_root *root,
2261 struct inode *dir, struct inode *inode,
2262 const char *name, int name_len)
2264 struct btrfs_path *path;
2266 struct extent_buffer *leaf;
2267 struct btrfs_dir_item *di;
2268 struct btrfs_key key;
2271 path = btrfs_alloc_path();
2277 path->leave_spinning = 1;
2278 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2279 name, name_len, -1);
2288 leaf = path->nodes[0];
2289 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2290 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2293 btrfs_release_path(root, path);
2295 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2297 dir->i_ino, &index);
2299 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2300 "inode %lu parent %lu\n", name_len, name,
2301 inode->i_ino, dir->i_ino);
2305 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2306 index, name, name_len, -1);
2315 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2316 btrfs_release_path(root, path);
2318 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2320 BUG_ON(ret != 0 && ret != -ENOENT);
2322 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2326 btrfs_free_path(path);
2330 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2331 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2332 btrfs_update_inode(trans, root, dir);
2333 btrfs_drop_nlink(inode);
2334 ret = btrfs_update_inode(trans, root, inode);
2339 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2341 struct btrfs_root *root;
2342 struct btrfs_trans_handle *trans;
2343 struct inode *inode = dentry->d_inode;
2345 unsigned long nr = 0;
2347 root = BTRFS_I(dir)->root;
2349 trans = btrfs_start_transaction(root, 1);
2351 btrfs_set_trans_block_group(trans, dir);
2353 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2355 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2356 dentry->d_name.name, dentry->d_name.len);
2358 if (inode->i_nlink == 0)
2359 ret = btrfs_orphan_add(trans, inode);
2361 nr = trans->blocks_used;
2363 btrfs_end_transaction_throttle(trans, root);
2364 btrfs_btree_balance_dirty(root, nr);
2368 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2370 struct inode *inode = dentry->d_inode;
2373 struct btrfs_root *root = BTRFS_I(dir)->root;
2374 struct btrfs_trans_handle *trans;
2375 unsigned long nr = 0;
2378 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2379 * the root of a subvolume or snapshot
2381 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2382 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2386 trans = btrfs_start_transaction(root, 1);
2387 btrfs_set_trans_block_group(trans, dir);
2389 err = btrfs_orphan_add(trans, inode);
2393 /* now the directory is empty */
2394 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2395 dentry->d_name.name, dentry->d_name.len);
2397 btrfs_i_size_write(inode, 0);
2400 nr = trans->blocks_used;
2401 ret = btrfs_end_transaction_throttle(trans, root);
2402 btrfs_btree_balance_dirty(root, nr);
2411 * when truncating bytes in a file, it is possible to avoid reading
2412 * the leaves that contain only checksum items. This can be the
2413 * majority of the IO required to delete a large file, but it must
2414 * be done carefully.
2416 * The keys in the level just above the leaves are checked to make sure
2417 * the lowest key in a given leaf is a csum key, and starts at an offset
2418 * after the new size.
2420 * Then the key for the next leaf is checked to make sure it also has
2421 * a checksum item for the same file. If it does, we know our target leaf
2422 * contains only checksum items, and it can be safely freed without reading
2425 * This is just an optimization targeted at large files. It may do
2426 * nothing. It will return 0 unless things went badly.
2428 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2429 struct btrfs_root *root,
2430 struct btrfs_path *path,
2431 struct inode *inode, u64 new_size)
2433 struct btrfs_key key;
2436 struct btrfs_key found_key;
2437 struct btrfs_key other_key;
2438 struct btrfs_leaf_ref *ref;
2442 path->lowest_level = 1;
2443 key.objectid = inode->i_ino;
2444 key.type = BTRFS_CSUM_ITEM_KEY;
2445 key.offset = new_size;
2447 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2451 if (path->nodes[1] == NULL) {
2456 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2457 nritems = btrfs_header_nritems(path->nodes[1]);
2462 if (path->slots[1] >= nritems)
2465 /* did we find a key greater than anything we want to delete? */
2466 if (found_key.objectid > inode->i_ino ||
2467 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2470 /* we check the next key in the node to make sure the leave contains
2471 * only checksum items. This comparison doesn't work if our
2472 * leaf is the last one in the node
2474 if (path->slots[1] + 1 >= nritems) {
2476 /* search forward from the last key in the node, this
2477 * will bring us into the next node in the tree
2479 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2481 /* unlikely, but we inc below, so check to be safe */
2482 if (found_key.offset == (u64)-1)
2485 /* search_forward needs a path with locks held, do the
2486 * search again for the original key. It is possible
2487 * this will race with a balance and return a path that
2488 * we could modify, but this drop is just an optimization
2489 * and is allowed to miss some leaves.
2491 btrfs_release_path(root, path);
2494 /* setup a max key for search_forward */
2495 other_key.offset = (u64)-1;
2496 other_key.type = key.type;
2497 other_key.objectid = key.objectid;
2499 path->keep_locks = 1;
2500 ret = btrfs_search_forward(root, &found_key, &other_key,
2502 path->keep_locks = 0;
2503 if (ret || found_key.objectid != key.objectid ||
2504 found_key.type != key.type) {
2509 key.offset = found_key.offset;
2510 btrfs_release_path(root, path);
2515 /* we know there's one more slot after us in the tree,
2516 * read that key so we can verify it is also a checksum item
2518 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2520 if (found_key.objectid < inode->i_ino)
2523 if (found_key.type != key.type || found_key.offset < new_size)
2527 * if the key for the next leaf isn't a csum key from this objectid,
2528 * we can't be sure there aren't good items inside this leaf.
2531 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2534 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2535 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2537 * it is safe to delete this leaf, it contains only
2538 * csum items from this inode at an offset >= new_size
2540 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2543 if (root->ref_cows && leaf_gen < trans->transid) {
2544 ref = btrfs_alloc_leaf_ref(root, 0);
2546 ref->root_gen = root->root_key.offset;
2547 ref->bytenr = leaf_start;
2549 ref->generation = leaf_gen;
2552 btrfs_sort_leaf_ref(ref);
2554 ret = btrfs_add_leaf_ref(root, ref, 0);
2556 btrfs_free_leaf_ref(root, ref);
2562 btrfs_release_path(root, path);
2564 if (other_key.objectid == inode->i_ino &&
2565 other_key.type == key.type && other_key.offset > key.offset) {
2566 key.offset = other_key.offset;
2572 /* fixup any changes we've made to the path */
2573 path->lowest_level = 0;
2574 path->keep_locks = 0;
2575 btrfs_release_path(root, path);
2582 * this can truncate away extent items, csum items and directory items.
2583 * It starts at a high offset and removes keys until it can't find
2584 * any higher than new_size
2586 * csum items that cross the new i_size are truncated to the new size
2589 * min_type is the minimum key type to truncate down to. If set to 0, this
2590 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2592 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2593 struct btrfs_root *root,
2594 struct inode *inode,
2595 u64 new_size, u32 min_type)
2598 struct btrfs_path *path;
2599 struct btrfs_key key;
2600 struct btrfs_key found_key;
2601 u32 found_type = (u8)-1;
2602 struct extent_buffer *leaf;
2603 struct btrfs_file_extent_item *fi;
2604 u64 extent_start = 0;
2605 u64 extent_num_bytes = 0;
2606 u64 extent_offset = 0;
2610 int pending_del_nr = 0;
2611 int pending_del_slot = 0;
2612 int extent_type = -1;
2614 u64 mask = root->sectorsize - 1;
2617 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2618 path = btrfs_alloc_path();
2622 /* FIXME, add redo link to tree so we don't leak on crash */
2623 key.objectid = inode->i_ino;
2624 key.offset = (u64)-1;
2628 path->leave_spinning = 1;
2629 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2634 /* there are no items in the tree for us to truncate, we're
2637 if (path->slots[0] == 0) {
2646 leaf = path->nodes[0];
2647 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2648 found_type = btrfs_key_type(&found_key);
2651 if (found_key.objectid != inode->i_ino)
2654 if (found_type < min_type)
2657 item_end = found_key.offset;
2658 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2659 fi = btrfs_item_ptr(leaf, path->slots[0],
2660 struct btrfs_file_extent_item);
2661 extent_type = btrfs_file_extent_type(leaf, fi);
2662 encoding = btrfs_file_extent_compression(leaf, fi);
2663 encoding |= btrfs_file_extent_encryption(leaf, fi);
2664 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2666 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2668 btrfs_file_extent_num_bytes(leaf, fi);
2669 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2670 item_end += btrfs_file_extent_inline_len(leaf,
2675 if (item_end < new_size) {
2676 if (found_type == BTRFS_DIR_ITEM_KEY)
2677 found_type = BTRFS_INODE_ITEM_KEY;
2678 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2679 found_type = BTRFS_EXTENT_DATA_KEY;
2680 else if (found_type == BTRFS_EXTENT_DATA_KEY)
2681 found_type = BTRFS_XATTR_ITEM_KEY;
2682 else if (found_type == BTRFS_XATTR_ITEM_KEY)
2683 found_type = BTRFS_INODE_REF_KEY;
2684 else if (found_type)
2688 btrfs_set_key_type(&key, found_type);
2691 if (found_key.offset >= new_size)
2697 /* FIXME, shrink the extent if the ref count is only 1 */
2698 if (found_type != BTRFS_EXTENT_DATA_KEY)
2701 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2703 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2704 if (!del_item && !encoding) {
2705 u64 orig_num_bytes =
2706 btrfs_file_extent_num_bytes(leaf, fi);
2707 extent_num_bytes = new_size -
2708 found_key.offset + root->sectorsize - 1;
2709 extent_num_bytes = extent_num_bytes &
2710 ~((u64)root->sectorsize - 1);
2711 btrfs_set_file_extent_num_bytes(leaf, fi,
2713 num_dec = (orig_num_bytes -
2715 if (root->ref_cows && extent_start != 0)
2716 inode_sub_bytes(inode, num_dec);
2717 btrfs_mark_buffer_dirty(leaf);
2720 btrfs_file_extent_disk_num_bytes(leaf,
2722 extent_offset = found_key.offset -
2723 btrfs_file_extent_offset(leaf, fi);
2725 /* FIXME blocksize != 4096 */
2726 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2727 if (extent_start != 0) {
2730 inode_sub_bytes(inode, num_dec);
2733 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2735 * we can't truncate inline items that have had
2739 btrfs_file_extent_compression(leaf, fi) == 0 &&
2740 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2741 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2742 u32 size = new_size - found_key.offset;
2744 if (root->ref_cows) {
2745 inode_sub_bytes(inode, item_end + 1 -
2749 btrfs_file_extent_calc_inline_size(size);
2750 ret = btrfs_truncate_item(trans, root, path,
2753 } else if (root->ref_cows) {
2754 inode_sub_bytes(inode, item_end + 1 -
2760 if (!pending_del_nr) {
2761 /* no pending yet, add ourselves */
2762 pending_del_slot = path->slots[0];
2764 } else if (pending_del_nr &&
2765 path->slots[0] + 1 == pending_del_slot) {
2766 /* hop on the pending chunk */
2768 pending_del_slot = path->slots[0];
2775 if (found_extent && root->ref_cows) {
2776 btrfs_set_path_blocking(path);
2777 ret = btrfs_free_extent(trans, root, extent_start,
2778 extent_num_bytes, 0,
2779 btrfs_header_owner(leaf),
2780 inode->i_ino, extent_offset);
2784 if (path->slots[0] == 0) {
2787 btrfs_release_path(root, path);
2788 if (found_type == BTRFS_INODE_ITEM_KEY)
2794 if (pending_del_nr &&
2795 path->slots[0] + 1 != pending_del_slot) {
2796 struct btrfs_key debug;
2798 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2800 ret = btrfs_del_items(trans, root, path,
2805 btrfs_release_path(root, path);
2806 if (found_type == BTRFS_INODE_ITEM_KEY)
2813 if (pending_del_nr) {
2814 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2817 btrfs_free_path(path);
2822 * taken from block_truncate_page, but does cow as it zeros out
2823 * any bytes left in the last page in the file.
2825 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2827 struct inode *inode = mapping->host;
2828 struct btrfs_root *root = BTRFS_I(inode)->root;
2829 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2830 struct btrfs_ordered_extent *ordered;
2832 u32 blocksize = root->sectorsize;
2833 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2834 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2840 if ((offset & (blocksize - 1)) == 0)
2845 page = grab_cache_page(mapping, index);
2849 page_start = page_offset(page);
2850 page_end = page_start + PAGE_CACHE_SIZE - 1;
2852 if (!PageUptodate(page)) {
2853 ret = btrfs_readpage(NULL, page);
2855 if (page->mapping != mapping) {
2857 page_cache_release(page);
2860 if (!PageUptodate(page)) {
2865 wait_on_page_writeback(page);
2867 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2868 set_page_extent_mapped(page);
2870 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2872 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2874 page_cache_release(page);
2875 btrfs_start_ordered_extent(inode, ordered, 1);
2876 btrfs_put_ordered_extent(ordered);
2880 btrfs_set_extent_delalloc(inode, page_start, page_end);
2882 if (offset != PAGE_CACHE_SIZE) {
2884 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2885 flush_dcache_page(page);
2888 ClearPageChecked(page);
2889 set_page_dirty(page);
2890 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2894 page_cache_release(page);
2899 int btrfs_cont_expand(struct inode *inode, loff_t size)
2901 struct btrfs_trans_handle *trans;
2902 struct btrfs_root *root = BTRFS_I(inode)->root;
2903 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2904 struct extent_map *em;
2905 u64 mask = root->sectorsize - 1;
2906 u64 hole_start = (inode->i_size + mask) & ~mask;
2907 u64 block_end = (size + mask) & ~mask;
2913 if (size <= hole_start)
2916 err = btrfs_check_metadata_free_space(root);
2920 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2923 struct btrfs_ordered_extent *ordered;
2924 btrfs_wait_ordered_range(inode, hole_start,
2925 block_end - hole_start);
2926 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2927 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2930 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2931 btrfs_put_ordered_extent(ordered);
2934 trans = btrfs_start_transaction(root, 1);
2935 btrfs_set_trans_block_group(trans, inode);
2937 cur_offset = hole_start;
2939 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2940 block_end - cur_offset, 0);
2941 BUG_ON(IS_ERR(em) || !em);
2942 last_byte = min(extent_map_end(em), block_end);
2943 last_byte = (last_byte + mask) & ~mask;
2944 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2946 hole_size = last_byte - cur_offset;
2947 err = btrfs_drop_extents(trans, root, inode,
2949 cur_offset + hole_size,
2951 cur_offset, &hint_byte, 1);
2954 err = btrfs_insert_file_extent(trans, root,
2955 inode->i_ino, cur_offset, 0,
2956 0, hole_size, 0, hole_size,
2958 btrfs_drop_extent_cache(inode, hole_start,
2961 free_extent_map(em);
2962 cur_offset = last_byte;
2963 if (err || cur_offset >= block_end)
2967 btrfs_end_transaction(trans, root);
2968 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2972 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2974 struct inode *inode = dentry->d_inode;
2977 err = inode_change_ok(inode, attr);
2981 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2982 if (attr->ia_size > inode->i_size) {
2983 err = btrfs_cont_expand(inode, attr->ia_size);
2986 } else if (inode->i_size > 0 &&
2987 attr->ia_size == 0) {
2989 /* we're truncating a file that used to have good
2990 * data down to zero. Make sure it gets into
2991 * the ordered flush list so that any new writes
2992 * get down to disk quickly.
2994 BTRFS_I(inode)->ordered_data_close = 1;
2998 err = inode_setattr(inode, attr);
3000 if (!err && ((attr->ia_valid & ATTR_MODE)))
3001 err = btrfs_acl_chmod(inode);
3005 void btrfs_delete_inode(struct inode *inode)
3007 struct btrfs_trans_handle *trans;
3008 struct btrfs_root *root = BTRFS_I(inode)->root;
3012 truncate_inode_pages(&inode->i_data, 0);
3013 if (is_bad_inode(inode)) {
3014 btrfs_orphan_del(NULL, inode);
3017 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3019 btrfs_i_size_write(inode, 0);
3020 trans = btrfs_join_transaction(root, 1);
3022 btrfs_set_trans_block_group(trans, inode);
3023 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3025 btrfs_orphan_del(NULL, inode);
3026 goto no_delete_lock;
3029 btrfs_orphan_del(trans, inode);
3031 nr = trans->blocks_used;
3034 btrfs_end_transaction(trans, root);
3035 btrfs_btree_balance_dirty(root, nr);
3039 nr = trans->blocks_used;
3040 btrfs_end_transaction(trans, root);
3041 btrfs_btree_balance_dirty(root, nr);
3047 * this returns the key found in the dir entry in the location pointer.
3048 * If no dir entries were found, location->objectid is 0.
3050 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3051 struct btrfs_key *location)
3053 const char *name = dentry->d_name.name;
3054 int namelen = dentry->d_name.len;
3055 struct btrfs_dir_item *di;
3056 struct btrfs_path *path;
3057 struct btrfs_root *root = BTRFS_I(dir)->root;
3060 path = btrfs_alloc_path();
3063 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3068 if (!di || IS_ERR(di))
3071 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3073 btrfs_free_path(path);
3076 location->objectid = 0;
3081 * when we hit a tree root in a directory, the btrfs part of the inode
3082 * needs to be changed to reflect the root directory of the tree root. This
3083 * is kind of like crossing a mount point.
3085 static int fixup_tree_root_location(struct btrfs_root *root,
3086 struct btrfs_key *location,
3087 struct btrfs_root **sub_root,
3088 struct dentry *dentry)
3090 struct btrfs_root_item *ri;
3092 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3094 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3097 *sub_root = btrfs_read_fs_root(root->fs_info, location,
3098 dentry->d_name.name,
3099 dentry->d_name.len);
3100 if (IS_ERR(*sub_root))
3101 return PTR_ERR(*sub_root);
3103 ri = &(*sub_root)->root_item;
3104 location->objectid = btrfs_root_dirid(ri);
3105 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3106 location->offset = 0;
3111 static void inode_tree_add(struct inode *inode)
3113 struct btrfs_root *root = BTRFS_I(inode)->root;
3114 struct btrfs_inode *entry;
3116 struct rb_node *parent;
3119 p = &root->inode_tree.rb_node;
3122 spin_lock(&root->inode_lock);
3125 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3127 if (inode->i_ino < entry->vfs_inode.i_ino)
3128 p = &parent->rb_left;
3129 else if (inode->i_ino > entry->vfs_inode.i_ino)
3130 p = &parent->rb_right;
3132 WARN_ON(!(entry->vfs_inode.i_state &
3133 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3134 rb_erase(parent, &root->inode_tree);
3135 RB_CLEAR_NODE(parent);
3136 spin_unlock(&root->inode_lock);
3140 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3141 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3142 spin_unlock(&root->inode_lock);
3145 static void inode_tree_del(struct inode *inode)
3147 struct btrfs_root *root = BTRFS_I(inode)->root;
3149 spin_lock(&root->inode_lock);
3150 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3151 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3152 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3154 spin_unlock(&root->inode_lock);
3157 static noinline void init_btrfs_i(struct inode *inode)
3159 struct btrfs_inode *bi = BTRFS_I(inode);
3164 bi->logged_trans = 0;
3165 bi->delalloc_bytes = 0;
3166 bi->reserved_bytes = 0;
3167 bi->disk_i_size = 0;
3169 bi->index_cnt = (u64)-1;
3170 bi->last_unlink_trans = 0;
3171 bi->ordered_data_close = 0;
3172 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3173 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3174 inode->i_mapping, GFP_NOFS);
3175 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3176 inode->i_mapping, GFP_NOFS);
3177 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3178 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3179 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3180 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3181 mutex_init(&BTRFS_I(inode)->extent_mutex);
3182 mutex_init(&BTRFS_I(inode)->log_mutex);
3185 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3187 struct btrfs_iget_args *args = p;
3188 inode->i_ino = args->ino;
3189 init_btrfs_i(inode);
3190 BTRFS_I(inode)->root = args->root;
3191 btrfs_set_inode_space_info(args->root, inode);
3195 static int btrfs_find_actor(struct inode *inode, void *opaque)
3197 struct btrfs_iget_args *args = opaque;
3198 return args->ino == inode->i_ino &&
3199 args->root == BTRFS_I(inode)->root;
3202 static struct inode *btrfs_iget_locked(struct super_block *s,
3204 struct btrfs_root *root)
3206 struct inode *inode;
3207 struct btrfs_iget_args args;
3208 args.ino = objectid;
3211 inode = iget5_locked(s, objectid, btrfs_find_actor,
3212 btrfs_init_locked_inode,
3217 /* Get an inode object given its location and corresponding root.
3218 * Returns in *is_new if the inode was read from disk
3220 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3221 struct btrfs_root *root)
3223 struct inode *inode;
3225 inode = btrfs_iget_locked(s, location->objectid, root);
3227 return ERR_PTR(-ENOMEM);
3229 if (inode->i_state & I_NEW) {
3230 BTRFS_I(inode)->root = root;
3231 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3232 btrfs_read_locked_inode(inode);
3234 inode_tree_add(inode);
3235 unlock_new_inode(inode);
3241 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3243 struct inode *inode;
3244 struct btrfs_inode *bi = BTRFS_I(dir);
3245 struct btrfs_root *root = bi->root;
3246 struct btrfs_root *sub_root = root;
3247 struct btrfs_key location;
3250 if (dentry->d_name.len > BTRFS_NAME_LEN)
3251 return ERR_PTR(-ENAMETOOLONG);
3253 ret = btrfs_inode_by_name(dir, dentry, &location);
3256 return ERR_PTR(ret);
3259 if (location.objectid) {
3260 ret = fixup_tree_root_location(root, &location, &sub_root,
3263 return ERR_PTR(ret);
3265 return ERR_PTR(-ENOENT);
3266 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3268 return ERR_CAST(inode);
3273 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3274 struct nameidata *nd)
3276 struct inode *inode;
3278 if (dentry->d_name.len > BTRFS_NAME_LEN)
3279 return ERR_PTR(-ENAMETOOLONG);
3281 inode = btrfs_lookup_dentry(dir, dentry);
3283 return ERR_CAST(inode);
3285 return d_splice_alias(inode, dentry);
3288 static unsigned char btrfs_filetype_table[] = {
3289 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3292 static int btrfs_real_readdir(struct file *filp, void *dirent,
3295 struct inode *inode = filp->f_dentry->d_inode;
3296 struct btrfs_root *root = BTRFS_I(inode)->root;
3297 struct btrfs_item *item;
3298 struct btrfs_dir_item *di;
3299 struct btrfs_key key;
3300 struct btrfs_key found_key;
3301 struct btrfs_path *path;
3304 struct extent_buffer *leaf;
3307 unsigned char d_type;
3312 int key_type = BTRFS_DIR_INDEX_KEY;
3317 /* FIXME, use a real flag for deciding about the key type */
3318 if (root->fs_info->tree_root == root)
3319 key_type = BTRFS_DIR_ITEM_KEY;
3321 /* special case for "." */
3322 if (filp->f_pos == 0) {
3323 over = filldir(dirent, ".", 1,
3330 /* special case for .., just use the back ref */
3331 if (filp->f_pos == 1) {
3332 u64 pino = parent_ino(filp->f_path.dentry);
3333 over = filldir(dirent, "..", 2,
3339 path = btrfs_alloc_path();
3342 btrfs_set_key_type(&key, key_type);
3343 key.offset = filp->f_pos;
3344 key.objectid = inode->i_ino;
3346 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3352 leaf = path->nodes[0];
3353 nritems = btrfs_header_nritems(leaf);
3354 slot = path->slots[0];
3355 if (advance || slot >= nritems) {
3356 if (slot >= nritems - 1) {
3357 ret = btrfs_next_leaf(root, path);
3360 leaf = path->nodes[0];
3361 nritems = btrfs_header_nritems(leaf);
3362 slot = path->slots[0];
3370 item = btrfs_item_nr(leaf, slot);
3371 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3373 if (found_key.objectid != key.objectid)
3375 if (btrfs_key_type(&found_key) != key_type)
3377 if (found_key.offset < filp->f_pos)
3380 filp->f_pos = found_key.offset;
3382 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3384 di_total = btrfs_item_size(leaf, item);
3386 while (di_cur < di_total) {
3387 struct btrfs_key location;
3389 name_len = btrfs_dir_name_len(leaf, di);
3390 if (name_len <= sizeof(tmp_name)) {
3391 name_ptr = tmp_name;
3393 name_ptr = kmalloc(name_len, GFP_NOFS);
3399 read_extent_buffer(leaf, name_ptr,
3400 (unsigned long)(di + 1), name_len);
3402 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3403 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3405 /* is this a reference to our own snapshot? If so
3408 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3409 location.objectid == root->root_key.objectid) {
3413 over = filldir(dirent, name_ptr, name_len,
3414 found_key.offset, location.objectid,
3418 if (name_ptr != tmp_name)
3423 di_len = btrfs_dir_name_len(leaf, di) +
3424 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3426 di = (struct btrfs_dir_item *)((char *)di + di_len);
3430 /* Reached end of directory/root. Bump pos past the last item. */
3431 if (key_type == BTRFS_DIR_INDEX_KEY)
3432 filp->f_pos = INT_LIMIT(off_t);
3438 btrfs_free_path(path);
3442 int btrfs_write_inode(struct inode *inode, int wait)
3444 struct btrfs_root *root = BTRFS_I(inode)->root;
3445 struct btrfs_trans_handle *trans;
3448 if (root->fs_info->btree_inode == inode)
3452 trans = btrfs_join_transaction(root, 1);
3453 btrfs_set_trans_block_group(trans, inode);
3454 ret = btrfs_commit_transaction(trans, root);
3460 * This is somewhat expensive, updating the tree every time the
3461 * inode changes. But, it is most likely to find the inode in cache.
3462 * FIXME, needs more benchmarking...there are no reasons other than performance
3463 * to keep or drop this code.
3465 void btrfs_dirty_inode(struct inode *inode)
3467 struct btrfs_root *root = BTRFS_I(inode)->root;
3468 struct btrfs_trans_handle *trans;
3470 trans = btrfs_join_transaction(root, 1);
3471 btrfs_set_trans_block_group(trans, inode);
3472 btrfs_update_inode(trans, root, inode);
3473 btrfs_end_transaction(trans, root);
3477 * find the highest existing sequence number in a directory
3478 * and then set the in-memory index_cnt variable to reflect
3479 * free sequence numbers
3481 static int btrfs_set_inode_index_count(struct inode *inode)
3483 struct btrfs_root *root = BTRFS_I(inode)->root;
3484 struct btrfs_key key, found_key;
3485 struct btrfs_path *path;
3486 struct extent_buffer *leaf;
3489 key.objectid = inode->i_ino;
3490 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3491 key.offset = (u64)-1;
3493 path = btrfs_alloc_path();
3497 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3500 /* FIXME: we should be able to handle this */
3506 * MAGIC NUMBER EXPLANATION:
3507 * since we search a directory based on f_pos we have to start at 2
3508 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3509 * else has to start at 2
3511 if (path->slots[0] == 0) {
3512 BTRFS_I(inode)->index_cnt = 2;
3518 leaf = path->nodes[0];
3519 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3521 if (found_key.objectid != inode->i_ino ||
3522 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3523 BTRFS_I(inode)->index_cnt = 2;
3527 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3529 btrfs_free_path(path);
3534 * helper to find a free sequence number in a given directory. This current
3535 * code is very simple, later versions will do smarter things in the btree
3537 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3541 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3542 ret = btrfs_set_inode_index_count(dir);
3547 *index = BTRFS_I(dir)->index_cnt;
3548 BTRFS_I(dir)->index_cnt++;
3553 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3554 struct btrfs_root *root,
3556 const char *name, int name_len,
3557 u64 ref_objectid, u64 objectid,
3558 u64 alloc_hint, int mode, u64 *index)
3560 struct inode *inode;
3561 struct btrfs_inode_item *inode_item;
3562 struct btrfs_key *location;
3563 struct btrfs_path *path;
3564 struct btrfs_inode_ref *ref;
3565 struct btrfs_key key[2];
3571 path = btrfs_alloc_path();
3574 inode = new_inode(root->fs_info->sb);
3576 return ERR_PTR(-ENOMEM);
3579 ret = btrfs_set_inode_index(dir, index);
3582 return ERR_PTR(ret);
3586 * index_cnt is ignored for everything but a dir,
3587 * btrfs_get_inode_index_count has an explanation for the magic
3590 init_btrfs_i(inode);
3591 BTRFS_I(inode)->index_cnt = 2;
3592 BTRFS_I(inode)->root = root;
3593 BTRFS_I(inode)->generation = trans->transid;
3594 btrfs_set_inode_space_info(root, inode);
3600 BTRFS_I(inode)->block_group =
3601 btrfs_find_block_group(root, 0, alloc_hint, owner);
3603 key[0].objectid = objectid;
3604 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3607 key[1].objectid = objectid;
3608 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3609 key[1].offset = ref_objectid;
3611 sizes[0] = sizeof(struct btrfs_inode_item);
3612 sizes[1] = name_len + sizeof(*ref);
3614 path->leave_spinning = 1;
3615 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3619 if (objectid > root->highest_inode)
3620 root->highest_inode = objectid;
3622 inode->i_uid = current_fsuid();
3624 if (dir && (dir->i_mode & S_ISGID)) {
3625 inode->i_gid = dir->i_gid;
3629 inode->i_gid = current_fsgid();
3631 inode->i_mode = mode;
3632 inode->i_ino = objectid;
3633 inode_set_bytes(inode, 0);
3634 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3635 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3636 struct btrfs_inode_item);
3637 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3639 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3640 struct btrfs_inode_ref);
3641 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3642 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3643 ptr = (unsigned long)(ref + 1);
3644 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3646 btrfs_mark_buffer_dirty(path->nodes[0]);
3647 btrfs_free_path(path);
3649 location = &BTRFS_I(inode)->location;
3650 location->objectid = objectid;
3651 location->offset = 0;
3652 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3654 btrfs_inherit_iflags(inode, dir);
3656 if ((mode & S_IFREG)) {
3657 if (btrfs_test_opt(root, NODATASUM))
3658 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
3659 if (btrfs_test_opt(root, NODATACOW))
3660 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
3663 insert_inode_hash(inode);
3664 inode_tree_add(inode);
3668 BTRFS_I(dir)->index_cnt--;
3669 btrfs_free_path(path);
3671 return ERR_PTR(ret);
3674 static inline u8 btrfs_inode_type(struct inode *inode)
3676 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3680 * utility function to add 'inode' into 'parent_inode' with
3681 * a give name and a given sequence number.
3682 * if 'add_backref' is true, also insert a backref from the
3683 * inode to the parent directory.
3685 int btrfs_add_link(struct btrfs_trans_handle *trans,
3686 struct inode *parent_inode, struct inode *inode,
3687 const char *name, int name_len, int add_backref, u64 index)
3690 struct btrfs_key key;
3691 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3693 key.objectid = inode->i_ino;
3694 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3697 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3698 parent_inode->i_ino,
3699 &key, btrfs_inode_type(inode),
3703 ret = btrfs_insert_inode_ref(trans, root,
3706 parent_inode->i_ino,
3709 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3711 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3712 ret = btrfs_update_inode(trans, root, parent_inode);
3717 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3718 struct dentry *dentry, struct inode *inode,
3719 int backref, u64 index)
3721 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3722 inode, dentry->d_name.name,
3723 dentry->d_name.len, backref, index);
3725 d_instantiate(dentry, inode);
3733 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3734 int mode, dev_t rdev)
3736 struct btrfs_trans_handle *trans;
3737 struct btrfs_root *root = BTRFS_I(dir)->root;
3738 struct inode *inode = NULL;
3742 unsigned long nr = 0;
3745 if (!new_valid_dev(rdev))
3748 err = btrfs_check_metadata_free_space(root);
3752 trans = btrfs_start_transaction(root, 1);
3753 btrfs_set_trans_block_group(trans, dir);
3755 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3761 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3763 dentry->d_parent->d_inode->i_ino, objectid,
3764 BTRFS_I(dir)->block_group, mode, &index);
3765 err = PTR_ERR(inode);
3769 err = btrfs_init_inode_security(inode, dir);
3775 btrfs_set_trans_block_group(trans, inode);
3776 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3780 inode->i_op = &btrfs_special_inode_operations;
3781 init_special_inode(inode, inode->i_mode, rdev);
3782 btrfs_update_inode(trans, root, inode);
3784 btrfs_update_inode_block_group(trans, inode);
3785 btrfs_update_inode_block_group(trans, dir);
3787 nr = trans->blocks_used;
3788 btrfs_end_transaction_throttle(trans, root);
3791 inode_dec_link_count(inode);
3794 btrfs_btree_balance_dirty(root, nr);
3798 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3799 int mode, struct nameidata *nd)
3801 struct btrfs_trans_handle *trans;
3802 struct btrfs_root *root = BTRFS_I(dir)->root;
3803 struct inode *inode = NULL;
3806 unsigned long nr = 0;
3810 err = btrfs_check_metadata_free_space(root);
3813 trans = btrfs_start_transaction(root, 1);
3814 btrfs_set_trans_block_group(trans, dir);
3816 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3822 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3824 dentry->d_parent->d_inode->i_ino,
3825 objectid, BTRFS_I(dir)->block_group, mode,
3827 err = PTR_ERR(inode);
3831 err = btrfs_init_inode_security(inode, dir);
3837 btrfs_set_trans_block_group(trans, inode);
3838 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3842 inode->i_mapping->a_ops = &btrfs_aops;
3843 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3844 inode->i_fop = &btrfs_file_operations;
3845 inode->i_op = &btrfs_file_inode_operations;
3846 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3848 btrfs_update_inode_block_group(trans, inode);
3849 btrfs_update_inode_block_group(trans, dir);
3851 nr = trans->blocks_used;
3852 btrfs_end_transaction_throttle(trans, root);
3855 inode_dec_link_count(inode);
3858 btrfs_btree_balance_dirty(root, nr);
3862 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3863 struct dentry *dentry)
3865 struct btrfs_trans_handle *trans;
3866 struct btrfs_root *root = BTRFS_I(dir)->root;
3867 struct inode *inode = old_dentry->d_inode;
3869 unsigned long nr = 0;
3873 if (inode->i_nlink == 0)
3876 btrfs_inc_nlink(inode);
3877 err = btrfs_check_metadata_free_space(root);
3880 err = btrfs_set_inode_index(dir, &index);
3884 trans = btrfs_start_transaction(root, 1);
3886 btrfs_set_trans_block_group(trans, dir);
3887 atomic_inc(&inode->i_count);
3889 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3894 btrfs_update_inode_block_group(trans, dir);
3895 err = btrfs_update_inode(trans, root, inode);
3900 nr = trans->blocks_used;
3902 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3903 btrfs_end_transaction_throttle(trans, root);
3906 inode_dec_link_count(inode);
3909 btrfs_btree_balance_dirty(root, nr);
3913 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3915 struct inode *inode = NULL;
3916 struct btrfs_trans_handle *trans;
3917 struct btrfs_root *root = BTRFS_I(dir)->root;
3919 int drop_on_err = 0;
3922 unsigned long nr = 1;
3924 err = btrfs_check_metadata_free_space(root);
3928 trans = btrfs_start_transaction(root, 1);
3929 btrfs_set_trans_block_group(trans, dir);
3931 if (IS_ERR(trans)) {
3932 err = PTR_ERR(trans);
3936 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3942 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3944 dentry->d_parent->d_inode->i_ino, objectid,
3945 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3947 if (IS_ERR(inode)) {
3948 err = PTR_ERR(inode);
3954 err = btrfs_init_inode_security(inode, dir);
3958 inode->i_op = &btrfs_dir_inode_operations;
3959 inode->i_fop = &btrfs_dir_file_operations;
3960 btrfs_set_trans_block_group(trans, inode);
3962 btrfs_i_size_write(inode, 0);
3963 err = btrfs_update_inode(trans, root, inode);
3967 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3968 inode, dentry->d_name.name,
3969 dentry->d_name.len, 0, index);
3973 d_instantiate(dentry, inode);
3975 btrfs_update_inode_block_group(trans, inode);
3976 btrfs_update_inode_block_group(trans, dir);
3979 nr = trans->blocks_used;
3980 btrfs_end_transaction_throttle(trans, root);
3985 btrfs_btree_balance_dirty(root, nr);
3989 /* helper for btfs_get_extent. Given an existing extent in the tree,
3990 * and an extent that you want to insert, deal with overlap and insert
3991 * the new extent into the tree.
3993 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3994 struct extent_map *existing,
3995 struct extent_map *em,
3996 u64 map_start, u64 map_len)
4000 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4001 start_diff = map_start - em->start;
4002 em->start = map_start;
4004 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4005 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4006 em->block_start += start_diff;
4007 em->block_len -= start_diff;
4009 return add_extent_mapping(em_tree, em);
4012 static noinline int uncompress_inline(struct btrfs_path *path,
4013 struct inode *inode, struct page *page,
4014 size_t pg_offset, u64 extent_offset,
4015 struct btrfs_file_extent_item *item)
4018 struct extent_buffer *leaf = path->nodes[0];
4021 unsigned long inline_size;
4024 WARN_ON(pg_offset != 0);
4025 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4026 inline_size = btrfs_file_extent_inline_item_len(leaf,
4027 btrfs_item_nr(leaf, path->slots[0]));
4028 tmp = kmalloc(inline_size, GFP_NOFS);
4029 ptr = btrfs_file_extent_inline_start(item);
4031 read_extent_buffer(leaf, tmp, ptr, inline_size);
4033 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4034 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4035 inline_size, max_size);
4037 char *kaddr = kmap_atomic(page, KM_USER0);
4038 unsigned long copy_size = min_t(u64,
4039 PAGE_CACHE_SIZE - pg_offset,
4040 max_size - extent_offset);
4041 memset(kaddr + pg_offset, 0, copy_size);
4042 kunmap_atomic(kaddr, KM_USER0);
4049 * a bit scary, this does extent mapping from logical file offset to the disk.
4050 * the ugly parts come from merging extents from the disk with the in-ram
4051 * representation. This gets more complex because of the data=ordered code,
4052 * where the in-ram extents might be locked pending data=ordered completion.
4054 * This also copies inline extents directly into the page.
4057 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4058 size_t pg_offset, u64 start, u64 len,
4064 u64 extent_start = 0;
4066 u64 objectid = inode->i_ino;
4068 struct btrfs_path *path = NULL;
4069 struct btrfs_root *root = BTRFS_I(inode)->root;
4070 struct btrfs_file_extent_item *item;
4071 struct extent_buffer *leaf;
4072 struct btrfs_key found_key;
4073 struct extent_map *em = NULL;
4074 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4075 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4076 struct btrfs_trans_handle *trans = NULL;
4080 read_lock(&em_tree->lock);
4081 em = lookup_extent_mapping(em_tree, start, len);
4083 em->bdev = root->fs_info->fs_devices->latest_bdev;
4084 read_unlock(&em_tree->lock);
4087 if (em->start > start || em->start + em->len <= start)
4088 free_extent_map(em);
4089 else if (em->block_start == EXTENT_MAP_INLINE && page)
4090 free_extent_map(em);
4094 em = alloc_extent_map(GFP_NOFS);
4099 em->bdev = root->fs_info->fs_devices->latest_bdev;
4100 em->start = EXTENT_MAP_HOLE;
4101 em->orig_start = EXTENT_MAP_HOLE;
4103 em->block_len = (u64)-1;
4106 path = btrfs_alloc_path();
4110 ret = btrfs_lookup_file_extent(trans, root, path,
4111 objectid, start, trans != NULL);
4118 if (path->slots[0] == 0)
4123 leaf = path->nodes[0];
4124 item = btrfs_item_ptr(leaf, path->slots[0],
4125 struct btrfs_file_extent_item);
4126 /* are we inside the extent that was found? */
4127 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4128 found_type = btrfs_key_type(&found_key);
4129 if (found_key.objectid != objectid ||
4130 found_type != BTRFS_EXTENT_DATA_KEY) {
4134 found_type = btrfs_file_extent_type(leaf, item);
4135 extent_start = found_key.offset;
4136 compressed = btrfs_file_extent_compression(leaf, item);
4137 if (found_type == BTRFS_FILE_EXTENT_REG ||
4138 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4139 extent_end = extent_start +
4140 btrfs_file_extent_num_bytes(leaf, item);
4141 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4143 size = btrfs_file_extent_inline_len(leaf, item);
4144 extent_end = (extent_start + size + root->sectorsize - 1) &
4145 ~((u64)root->sectorsize - 1);
4148 if (start >= extent_end) {
4150 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4151 ret = btrfs_next_leaf(root, path);
4158 leaf = path->nodes[0];
4160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4161 if (found_key.objectid != objectid ||
4162 found_key.type != BTRFS_EXTENT_DATA_KEY)
4164 if (start + len <= found_key.offset)
4167 em->len = found_key.offset - start;
4171 if (found_type == BTRFS_FILE_EXTENT_REG ||
4172 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4173 em->start = extent_start;
4174 em->len = extent_end - extent_start;
4175 em->orig_start = extent_start -
4176 btrfs_file_extent_offset(leaf, item);
4177 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4179 em->block_start = EXTENT_MAP_HOLE;
4183 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4184 em->block_start = bytenr;
4185 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4188 bytenr += btrfs_file_extent_offset(leaf, item);
4189 em->block_start = bytenr;
4190 em->block_len = em->len;
4191 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4192 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4195 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4199 size_t extent_offset;
4202 em->block_start = EXTENT_MAP_INLINE;
4203 if (!page || create) {
4204 em->start = extent_start;
4205 em->len = extent_end - extent_start;
4209 size = btrfs_file_extent_inline_len(leaf, item);
4210 extent_offset = page_offset(page) + pg_offset - extent_start;
4211 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4212 size - extent_offset);
4213 em->start = extent_start + extent_offset;
4214 em->len = (copy_size + root->sectorsize - 1) &
4215 ~((u64)root->sectorsize - 1);
4216 em->orig_start = EXTENT_MAP_INLINE;
4218 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4219 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4220 if (create == 0 && !PageUptodate(page)) {
4221 if (btrfs_file_extent_compression(leaf, item) ==
4222 BTRFS_COMPRESS_ZLIB) {
4223 ret = uncompress_inline(path, inode, page,
4225 extent_offset, item);
4229 read_extent_buffer(leaf, map + pg_offset, ptr,
4231 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4232 memset(map + pg_offset + copy_size, 0,
4233 PAGE_CACHE_SIZE - pg_offset -
4238 flush_dcache_page(page);
4239 } else if (create && PageUptodate(page)) {
4242 free_extent_map(em);
4244 btrfs_release_path(root, path);
4245 trans = btrfs_join_transaction(root, 1);
4249 write_extent_buffer(leaf, map + pg_offset, ptr,
4252 btrfs_mark_buffer_dirty(leaf);
4254 set_extent_uptodate(io_tree, em->start,
4255 extent_map_end(em) - 1, GFP_NOFS);
4258 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4265 em->block_start = EXTENT_MAP_HOLE;
4266 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4268 btrfs_release_path(root, path);
4269 if (em->start > start || extent_map_end(em) <= start) {
4270 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4271 "[%llu %llu]\n", (unsigned long long)em->start,
4272 (unsigned long long)em->len,
4273 (unsigned long long)start,
4274 (unsigned long long)len);
4280 write_lock(&em_tree->lock);
4281 ret = add_extent_mapping(em_tree, em);
4282 /* it is possible that someone inserted the extent into the tree
4283 * while we had the lock dropped. It is also possible that
4284 * an overlapping map exists in the tree
4286 if (ret == -EEXIST) {
4287 struct extent_map *existing;
4291 existing = lookup_extent_mapping(em_tree, start, len);
4292 if (existing && (existing->start > start ||
4293 existing->start + existing->len <= start)) {
4294 free_extent_map(existing);
4298 existing = lookup_extent_mapping(em_tree, em->start,
4301 err = merge_extent_mapping(em_tree, existing,
4304 free_extent_map(existing);
4306 free_extent_map(em);
4311 free_extent_map(em);
4315 free_extent_map(em);
4320 write_unlock(&em_tree->lock);
4323 btrfs_free_path(path);
4325 ret = btrfs_end_transaction(trans, root);
4330 free_extent_map(em);
4331 return ERR_PTR(err);
4336 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4337 const struct iovec *iov, loff_t offset,
4338 unsigned long nr_segs)
4343 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4344 __u64 start, __u64 len)
4346 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4349 int btrfs_readpage(struct file *file, struct page *page)
4351 struct extent_io_tree *tree;
4352 tree = &BTRFS_I(page->mapping->host)->io_tree;
4353 return extent_read_full_page(tree, page, btrfs_get_extent);
4356 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4358 struct extent_io_tree *tree;
4361 if (current->flags & PF_MEMALLOC) {
4362 redirty_page_for_writepage(wbc, page);
4366 tree = &BTRFS_I(page->mapping->host)->io_tree;
4367 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4370 int btrfs_writepages(struct address_space *mapping,
4371 struct writeback_control *wbc)
4373 struct extent_io_tree *tree;
4375 tree = &BTRFS_I(mapping->host)->io_tree;
4376 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4380 btrfs_readpages(struct file *file, struct address_space *mapping,
4381 struct list_head *pages, unsigned nr_pages)
4383 struct extent_io_tree *tree;
4384 tree = &BTRFS_I(mapping->host)->io_tree;
4385 return extent_readpages(tree, mapping, pages, nr_pages,
4388 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4390 struct extent_io_tree *tree;
4391 struct extent_map_tree *map;
4394 tree = &BTRFS_I(page->mapping->host)->io_tree;
4395 map = &BTRFS_I(page->mapping->host)->extent_tree;
4396 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4398 ClearPagePrivate(page);
4399 set_page_private(page, 0);
4400 page_cache_release(page);
4405 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4407 if (PageWriteback(page) || PageDirty(page))
4409 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4412 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4414 struct extent_io_tree *tree;
4415 struct btrfs_ordered_extent *ordered;
4416 u64 page_start = page_offset(page);
4417 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4421 * we have the page locked, so new writeback can't start,
4422 * and the dirty bit won't be cleared while we are here.
4424 * Wait for IO on this page so that we can safely clear
4425 * the PagePrivate2 bit and do ordered accounting
4427 wait_on_page_writeback(page);
4429 tree = &BTRFS_I(page->mapping->host)->io_tree;
4431 btrfs_releasepage(page, GFP_NOFS);
4434 lock_extent(tree, page_start, page_end, GFP_NOFS);
4435 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4439 * IO on this page will never be started, so we need
4440 * to account for any ordered extents now
4442 clear_extent_bit(tree, page_start, page_end,
4443 EXTENT_DIRTY | EXTENT_DELALLOC |
4444 EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
4446 * whoever cleared the private bit is responsible
4447 * for the finish_ordered_io
4449 if (TestClearPagePrivate2(page)) {
4450 btrfs_finish_ordered_io(page->mapping->host,
4451 page_start, page_end);
4453 btrfs_put_ordered_extent(ordered);
4454 lock_extent(tree, page_start, page_end, GFP_NOFS);
4456 clear_extent_bit(tree, page_start, page_end,
4457 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC,
4458 1, 1, NULL, GFP_NOFS);
4459 __btrfs_releasepage(page, GFP_NOFS);
4461 ClearPageChecked(page);
4462 if (PagePrivate(page)) {
4463 ClearPagePrivate(page);
4464 set_page_private(page, 0);
4465 page_cache_release(page);
4470 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4471 * called from a page fault handler when a page is first dirtied. Hence we must
4472 * be careful to check for EOF conditions here. We set the page up correctly
4473 * for a written page which means we get ENOSPC checking when writing into
4474 * holes and correct delalloc and unwritten extent mapping on filesystems that
4475 * support these features.
4477 * We are not allowed to take the i_mutex here so we have to play games to
4478 * protect against truncate races as the page could now be beyond EOF. Because
4479 * vmtruncate() writes the inode size before removing pages, once we have the
4480 * page lock we can determine safely if the page is beyond EOF. If it is not
4481 * beyond EOF, then the page is guaranteed safe against truncation until we
4484 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4486 struct page *page = vmf->page;
4487 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4488 struct btrfs_root *root = BTRFS_I(inode)->root;
4489 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4490 struct btrfs_ordered_extent *ordered;
4492 unsigned long zero_start;
4498 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4502 else /* -ENOSPC, -EIO, etc */
4503 ret = VM_FAULT_SIGBUS;
4507 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4510 size = i_size_read(inode);
4511 page_start = page_offset(page);
4512 page_end = page_start + PAGE_CACHE_SIZE - 1;
4514 if ((page->mapping != inode->i_mapping) ||
4515 (page_start >= size)) {
4516 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4517 /* page got truncated out from underneath us */
4520 wait_on_page_writeback(page);
4522 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4523 set_page_extent_mapped(page);
4526 * we can't set the delalloc bits if there are pending ordered
4527 * extents. Drop our locks and wait for them to finish
4529 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4531 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4533 btrfs_start_ordered_extent(inode, ordered, 1);
4534 btrfs_put_ordered_extent(ordered);
4538 btrfs_set_extent_delalloc(inode, page_start, page_end);
4541 /* page is wholly or partially inside EOF */
4542 if (page_start + PAGE_CACHE_SIZE > size)
4543 zero_start = size & ~PAGE_CACHE_MASK;
4545 zero_start = PAGE_CACHE_SIZE;
4547 if (zero_start != PAGE_CACHE_SIZE) {
4549 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4550 flush_dcache_page(page);
4553 ClearPageChecked(page);
4554 set_page_dirty(page);
4555 SetPageUptodate(page);
4557 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4558 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4562 return VM_FAULT_LOCKED;
4568 static void btrfs_truncate(struct inode *inode)
4570 struct btrfs_root *root = BTRFS_I(inode)->root;
4572 struct btrfs_trans_handle *trans;
4574 u64 mask = root->sectorsize - 1;
4576 if (!S_ISREG(inode->i_mode))
4578 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4581 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4582 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4584 trans = btrfs_start_transaction(root, 1);
4587 * setattr is responsible for setting the ordered_data_close flag,
4588 * but that is only tested during the last file release. That
4589 * could happen well after the next commit, leaving a great big
4590 * window where new writes may get lost if someone chooses to write
4591 * to this file after truncating to zero
4593 * The inode doesn't have any dirty data here, and so if we commit
4594 * this is a noop. If someone immediately starts writing to the inode
4595 * it is very likely we'll catch some of their writes in this
4596 * transaction, and the commit will find this file on the ordered
4597 * data list with good things to send down.
4599 * This is a best effort solution, there is still a window where
4600 * using truncate to replace the contents of the file will
4601 * end up with a zero length file after a crash.
4603 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4604 btrfs_add_ordered_operation(trans, root, inode);
4606 btrfs_set_trans_block_group(trans, inode);
4607 btrfs_i_size_write(inode, inode->i_size);
4609 ret = btrfs_orphan_add(trans, inode);
4612 /* FIXME, add redo link to tree so we don't leak on crash */
4613 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4614 BTRFS_EXTENT_DATA_KEY);
4615 btrfs_update_inode(trans, root, inode);
4617 ret = btrfs_orphan_del(trans, inode);
4621 nr = trans->blocks_used;
4622 ret = btrfs_end_transaction_throttle(trans, root);
4624 btrfs_btree_balance_dirty(root, nr);
4628 * create a new subvolume directory/inode (helper for the ioctl).
4630 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4631 struct btrfs_root *new_root, struct dentry *dentry,
4632 u64 new_dirid, u64 alloc_hint)
4634 struct inode *inode;
4638 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4639 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4641 return PTR_ERR(inode);
4642 inode->i_op = &btrfs_dir_inode_operations;
4643 inode->i_fop = &btrfs_dir_file_operations;
4646 btrfs_i_size_write(inode, 0);
4648 error = btrfs_update_inode(trans, new_root, inode);
4652 d_instantiate(dentry, inode);
4656 /* helper function for file defrag and space balancing. This
4657 * forces readahead on a given range of bytes in an inode
4659 unsigned long btrfs_force_ra(struct address_space *mapping,
4660 struct file_ra_state *ra, struct file *file,
4661 pgoff_t offset, pgoff_t last_index)
4663 pgoff_t req_size = last_index - offset + 1;
4665 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4666 return offset + req_size;
4669 struct inode *btrfs_alloc_inode(struct super_block *sb)
4671 struct btrfs_inode *ei;
4673 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4677 ei->logged_trans = 0;
4678 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4679 INIT_LIST_HEAD(&ei->i_orphan);
4680 INIT_LIST_HEAD(&ei->ordered_operations);
4681 return &ei->vfs_inode;
4684 void btrfs_destroy_inode(struct inode *inode)
4686 struct btrfs_ordered_extent *ordered;
4687 struct btrfs_root *root = BTRFS_I(inode)->root;
4689 WARN_ON(!list_empty(&inode->i_dentry));
4690 WARN_ON(inode->i_data.nrpages);
4693 * Make sure we're properly removed from the ordered operation
4697 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4698 spin_lock(&root->fs_info->ordered_extent_lock);
4699 list_del_init(&BTRFS_I(inode)->ordered_operations);
4700 spin_unlock(&root->fs_info->ordered_extent_lock);
4703 spin_lock(&root->list_lock);
4704 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4705 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4706 " list\n", inode->i_ino);
4709 spin_unlock(&root->list_lock);
4712 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4716 printk(KERN_ERR "btrfs found ordered "
4717 "extent %llu %llu on inode cleanup\n",
4718 (unsigned long long)ordered->file_offset,
4719 (unsigned long long)ordered->len);
4720 btrfs_remove_ordered_extent(inode, ordered);
4721 btrfs_put_ordered_extent(ordered);
4722 btrfs_put_ordered_extent(ordered);
4725 inode_tree_del(inode);
4726 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4727 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4730 static void init_once(void *foo)
4732 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4734 inode_init_once(&ei->vfs_inode);
4737 void btrfs_destroy_cachep(void)
4739 if (btrfs_inode_cachep)
4740 kmem_cache_destroy(btrfs_inode_cachep);
4741 if (btrfs_trans_handle_cachep)
4742 kmem_cache_destroy(btrfs_trans_handle_cachep);
4743 if (btrfs_transaction_cachep)
4744 kmem_cache_destroy(btrfs_transaction_cachep);
4745 if (btrfs_path_cachep)
4746 kmem_cache_destroy(btrfs_path_cachep);
4749 int btrfs_init_cachep(void)
4751 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4752 sizeof(struct btrfs_inode), 0,
4753 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4754 if (!btrfs_inode_cachep)
4757 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4758 sizeof(struct btrfs_trans_handle), 0,
4759 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4760 if (!btrfs_trans_handle_cachep)
4763 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4764 sizeof(struct btrfs_transaction), 0,
4765 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4766 if (!btrfs_transaction_cachep)
4769 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4770 sizeof(struct btrfs_path), 0,
4771 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4772 if (!btrfs_path_cachep)
4777 btrfs_destroy_cachep();
4781 static int btrfs_getattr(struct vfsmount *mnt,
4782 struct dentry *dentry, struct kstat *stat)
4784 struct inode *inode = dentry->d_inode;
4785 generic_fillattr(inode, stat);
4786 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4787 stat->blksize = PAGE_CACHE_SIZE;
4788 stat->blocks = (inode_get_bytes(inode) +
4789 BTRFS_I(inode)->delalloc_bytes) >> 9;
4793 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4794 struct inode *new_dir, struct dentry *new_dentry)
4796 struct btrfs_trans_handle *trans;
4797 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4798 struct inode *new_inode = new_dentry->d_inode;
4799 struct inode *old_inode = old_dentry->d_inode;
4800 struct timespec ctime = CURRENT_TIME;
4804 /* we're not allowed to rename between subvolumes */
4805 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4806 BTRFS_I(new_dir)->root->root_key.objectid)
4809 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4810 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4814 /* to rename a snapshot or subvolume, we need to juggle the
4815 * backrefs. This isn't coded yet
4817 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4820 ret = btrfs_check_metadata_free_space(root);
4825 * we're using rename to replace one file with another.
4826 * and the replacement file is large. Start IO on it now so
4827 * we don't add too much work to the end of the transaction
4829 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
4830 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4831 filemap_flush(old_inode->i_mapping);
4833 trans = btrfs_start_transaction(root, 1);
4836 * make sure the inode gets flushed if it is replacing
4839 if (new_inode && new_inode->i_size &&
4840 old_inode && S_ISREG(old_inode->i_mode)) {
4841 btrfs_add_ordered_operation(trans, root, old_inode);
4845 * this is an ugly little race, but the rename is required to make
4846 * sure that if we crash, the inode is either at the old name
4847 * or the new one. pinning the log transaction lets us make sure
4848 * we don't allow a log commit to come in after we unlink the
4849 * name but before we add the new name back in.
4851 btrfs_pin_log_trans(root);
4853 btrfs_set_trans_block_group(trans, new_dir);
4855 btrfs_inc_nlink(old_dentry->d_inode);
4856 old_dir->i_ctime = old_dir->i_mtime = ctime;
4857 new_dir->i_ctime = new_dir->i_mtime = ctime;
4858 old_inode->i_ctime = ctime;
4860 if (old_dentry->d_parent != new_dentry->d_parent)
4861 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4863 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4864 old_dentry->d_name.name,
4865 old_dentry->d_name.len);
4870 new_inode->i_ctime = CURRENT_TIME;
4871 ret = btrfs_unlink_inode(trans, root, new_dir,
4872 new_dentry->d_inode,
4873 new_dentry->d_name.name,
4874 new_dentry->d_name.len);
4877 if (new_inode->i_nlink == 0) {
4878 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4884 ret = btrfs_set_inode_index(new_dir, &index);
4888 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4889 old_inode, new_dentry->d_name.name,
4890 new_dentry->d_name.len, 1, index);
4894 btrfs_log_new_name(trans, old_inode, old_dir,
4895 new_dentry->d_parent);
4898 /* this btrfs_end_log_trans just allows the current
4899 * log-sub transaction to complete
4901 btrfs_end_log_trans(root);
4902 btrfs_end_transaction_throttle(trans, root);
4908 * some fairly slow code that needs optimization. This walks the list
4909 * of all the inodes with pending delalloc and forces them to disk.
4911 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4913 struct list_head *head = &root->fs_info->delalloc_inodes;
4914 struct btrfs_inode *binode;
4915 struct inode *inode;
4917 if (root->fs_info->sb->s_flags & MS_RDONLY)
4920 spin_lock(&root->fs_info->delalloc_lock);
4921 while (!list_empty(head)) {
4922 binode = list_entry(head->next, struct btrfs_inode,
4924 inode = igrab(&binode->vfs_inode);
4926 list_del_init(&binode->delalloc_inodes);
4927 spin_unlock(&root->fs_info->delalloc_lock);
4929 filemap_flush(inode->i_mapping);
4933 spin_lock(&root->fs_info->delalloc_lock);
4935 spin_unlock(&root->fs_info->delalloc_lock);
4937 /* the filemap_flush will queue IO into the worker threads, but
4938 * we have to make sure the IO is actually started and that
4939 * ordered extents get created before we return
4941 atomic_inc(&root->fs_info->async_submit_draining);
4942 while (atomic_read(&root->fs_info->nr_async_submits) ||
4943 atomic_read(&root->fs_info->async_delalloc_pages)) {
4944 wait_event(root->fs_info->async_submit_wait,
4945 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4946 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4948 atomic_dec(&root->fs_info->async_submit_draining);
4952 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4953 const char *symname)
4955 struct btrfs_trans_handle *trans;
4956 struct btrfs_root *root = BTRFS_I(dir)->root;
4957 struct btrfs_path *path;
4958 struct btrfs_key key;
4959 struct inode *inode = NULL;
4967 struct btrfs_file_extent_item *ei;
4968 struct extent_buffer *leaf;
4969 unsigned long nr = 0;
4971 name_len = strlen(symname) + 1;
4972 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4973 return -ENAMETOOLONG;
4975 err = btrfs_check_metadata_free_space(root);
4979 trans = btrfs_start_transaction(root, 1);
4980 btrfs_set_trans_block_group(trans, dir);
4982 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4988 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4990 dentry->d_parent->d_inode->i_ino, objectid,
4991 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4993 err = PTR_ERR(inode);
4997 err = btrfs_init_inode_security(inode, dir);
5003 btrfs_set_trans_block_group(trans, inode);
5004 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5008 inode->i_mapping->a_ops = &btrfs_aops;
5009 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5010 inode->i_fop = &btrfs_file_operations;
5011 inode->i_op = &btrfs_file_inode_operations;
5012 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5014 btrfs_update_inode_block_group(trans, inode);
5015 btrfs_update_inode_block_group(trans, dir);
5019 path = btrfs_alloc_path();
5021 key.objectid = inode->i_ino;
5023 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5024 datasize = btrfs_file_extent_calc_inline_size(name_len);
5025 err = btrfs_insert_empty_item(trans, root, path, &key,
5031 leaf = path->nodes[0];
5032 ei = btrfs_item_ptr(leaf, path->slots[0],
5033 struct btrfs_file_extent_item);
5034 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5035 btrfs_set_file_extent_type(leaf, ei,
5036 BTRFS_FILE_EXTENT_INLINE);
5037 btrfs_set_file_extent_encryption(leaf, ei, 0);
5038 btrfs_set_file_extent_compression(leaf, ei, 0);
5039 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5040 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5042 ptr = btrfs_file_extent_inline_start(ei);
5043 write_extent_buffer(leaf, symname, ptr, name_len);
5044 btrfs_mark_buffer_dirty(leaf);
5045 btrfs_free_path(path);
5047 inode->i_op = &btrfs_symlink_inode_operations;
5048 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5049 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5050 inode_set_bytes(inode, name_len);
5051 btrfs_i_size_write(inode, name_len - 1);
5052 err = btrfs_update_inode(trans, root, inode);
5057 nr = trans->blocks_used;
5058 btrfs_end_transaction_throttle(trans, root);
5061 inode_dec_link_count(inode);
5064 btrfs_btree_balance_dirty(root, nr);
5068 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5069 struct inode *inode, u64 start, u64 end,
5070 u64 locked_end, u64 alloc_hint, int mode)
5072 struct btrfs_root *root = BTRFS_I(inode)->root;
5073 struct btrfs_key ins;
5075 u64 cur_offset = start;
5076 u64 num_bytes = end - start;
5079 while (num_bytes > 0) {
5080 alloc_size = min(num_bytes, root->fs_info->max_extent);
5081 ret = btrfs_reserve_extent(trans, root, alloc_size,
5082 root->sectorsize, 0, alloc_hint,
5088 ret = insert_reserved_file_extent(trans, inode,
5089 cur_offset, ins.objectid,
5090 ins.offset, ins.offset,
5091 ins.offset, locked_end,
5093 BTRFS_FILE_EXTENT_PREALLOC);
5095 btrfs_drop_extent_cache(inode, cur_offset,
5096 cur_offset + ins.offset -1, 0);
5097 num_bytes -= ins.offset;
5098 cur_offset += ins.offset;
5099 alloc_hint = ins.objectid + ins.offset;
5102 if (cur_offset > start) {
5103 inode->i_ctime = CURRENT_TIME;
5104 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5105 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5106 cur_offset > i_size_read(inode))
5107 btrfs_i_size_write(inode, cur_offset);
5108 ret = btrfs_update_inode(trans, root, inode);
5115 static long btrfs_fallocate(struct inode *inode, int mode,
5116 loff_t offset, loff_t len)
5124 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5125 struct extent_map *em;
5126 struct btrfs_trans_handle *trans;
5127 struct btrfs_root *root;
5130 alloc_start = offset & ~mask;
5131 alloc_end = (offset + len + mask) & ~mask;
5134 * wait for ordered IO before we have any locks. We'll loop again
5135 * below with the locks held.
5137 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5139 mutex_lock(&inode->i_mutex);
5140 if (alloc_start > inode->i_size) {
5141 ret = btrfs_cont_expand(inode, alloc_start);
5146 root = BTRFS_I(inode)->root;
5148 ret = btrfs_check_data_free_space(root, inode,
5149 alloc_end - alloc_start);
5153 locked_end = alloc_end - 1;
5155 struct btrfs_ordered_extent *ordered;
5157 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5163 /* the extent lock is ordered inside the running
5166 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5168 ordered = btrfs_lookup_first_ordered_extent(inode,
5171 ordered->file_offset + ordered->len > alloc_start &&
5172 ordered->file_offset < alloc_end) {
5173 btrfs_put_ordered_extent(ordered);
5174 unlock_extent(&BTRFS_I(inode)->io_tree,
5175 alloc_start, locked_end, GFP_NOFS);
5176 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5179 * we can't wait on the range with the transaction
5180 * running or with the extent lock held
5182 btrfs_wait_ordered_range(inode, alloc_start,
5183 alloc_end - alloc_start);
5186 btrfs_put_ordered_extent(ordered);
5191 cur_offset = alloc_start;
5193 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5194 alloc_end - cur_offset, 0);
5195 BUG_ON(IS_ERR(em) || !em);
5196 last_byte = min(extent_map_end(em), alloc_end);
5197 last_byte = (last_byte + mask) & ~mask;
5198 if (em->block_start == EXTENT_MAP_HOLE) {
5199 ret = prealloc_file_range(trans, inode, cur_offset,
5200 last_byte, locked_end + 1,
5203 free_extent_map(em);
5207 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5208 alloc_hint = em->block_start;
5209 free_extent_map(em);
5211 cur_offset = last_byte;
5212 if (cur_offset >= alloc_end) {
5217 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5220 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5222 btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
5224 mutex_unlock(&inode->i_mutex);
5228 static int btrfs_set_page_dirty(struct page *page)
5230 return __set_page_dirty_nobuffers(page);
5233 static int btrfs_permission(struct inode *inode, int mask)
5235 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5237 return generic_permission(inode, mask, btrfs_check_acl);
5240 static struct inode_operations btrfs_dir_inode_operations = {
5241 .getattr = btrfs_getattr,
5242 .lookup = btrfs_lookup,
5243 .create = btrfs_create,
5244 .unlink = btrfs_unlink,
5246 .mkdir = btrfs_mkdir,
5247 .rmdir = btrfs_rmdir,
5248 .rename = btrfs_rename,
5249 .symlink = btrfs_symlink,
5250 .setattr = btrfs_setattr,
5251 .mknod = btrfs_mknod,
5252 .setxattr = btrfs_setxattr,
5253 .getxattr = btrfs_getxattr,
5254 .listxattr = btrfs_listxattr,
5255 .removexattr = btrfs_removexattr,
5256 .permission = btrfs_permission,
5258 static struct inode_operations btrfs_dir_ro_inode_operations = {
5259 .lookup = btrfs_lookup,
5260 .permission = btrfs_permission,
5262 static struct file_operations btrfs_dir_file_operations = {
5263 .llseek = generic_file_llseek,
5264 .read = generic_read_dir,
5265 .readdir = btrfs_real_readdir,
5266 .unlocked_ioctl = btrfs_ioctl,
5267 #ifdef CONFIG_COMPAT
5268 .compat_ioctl = btrfs_ioctl,
5270 .release = btrfs_release_file,
5271 .fsync = btrfs_sync_file,
5274 static struct extent_io_ops btrfs_extent_io_ops = {
5275 .fill_delalloc = run_delalloc_range,
5276 .submit_bio_hook = btrfs_submit_bio_hook,
5277 .merge_bio_hook = btrfs_merge_bio_hook,
5278 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5279 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5280 .writepage_start_hook = btrfs_writepage_start_hook,
5281 .readpage_io_failed_hook = btrfs_io_failed_hook,
5282 .set_bit_hook = btrfs_set_bit_hook,
5283 .clear_bit_hook = btrfs_clear_bit_hook,
5287 * btrfs doesn't support the bmap operation because swapfiles
5288 * use bmap to make a mapping of extents in the file. They assume
5289 * these extents won't change over the life of the file and they
5290 * use the bmap result to do IO directly to the drive.
5292 * the btrfs bmap call would return logical addresses that aren't
5293 * suitable for IO and they also will change frequently as COW
5294 * operations happen. So, swapfile + btrfs == corruption.
5296 * For now we're avoiding this by dropping bmap.
5298 static struct address_space_operations btrfs_aops = {
5299 .readpage = btrfs_readpage,
5300 .writepage = btrfs_writepage,
5301 .writepages = btrfs_writepages,
5302 .readpages = btrfs_readpages,
5303 .sync_page = block_sync_page,
5304 .direct_IO = btrfs_direct_IO,
5305 .invalidatepage = btrfs_invalidatepage,
5306 .releasepage = btrfs_releasepage,
5307 .set_page_dirty = btrfs_set_page_dirty,
5310 static struct address_space_operations btrfs_symlink_aops = {
5311 .readpage = btrfs_readpage,
5312 .writepage = btrfs_writepage,
5313 .invalidatepage = btrfs_invalidatepage,
5314 .releasepage = btrfs_releasepage,
5317 static struct inode_operations btrfs_file_inode_operations = {
5318 .truncate = btrfs_truncate,
5319 .getattr = btrfs_getattr,
5320 .setattr = btrfs_setattr,
5321 .setxattr = btrfs_setxattr,
5322 .getxattr = btrfs_getxattr,
5323 .listxattr = btrfs_listxattr,
5324 .removexattr = btrfs_removexattr,
5325 .permission = btrfs_permission,
5326 .fallocate = btrfs_fallocate,
5327 .fiemap = btrfs_fiemap,
5329 static struct inode_operations btrfs_special_inode_operations = {
5330 .getattr = btrfs_getattr,
5331 .setattr = btrfs_setattr,
5332 .permission = btrfs_permission,
5333 .setxattr = btrfs_setxattr,
5334 .getxattr = btrfs_getxattr,
5335 .listxattr = btrfs_listxattr,
5336 .removexattr = btrfs_removexattr,
5338 static struct inode_operations btrfs_symlink_inode_operations = {
5339 .readlink = generic_readlink,
5340 .follow_link = page_follow_link_light,
5341 .put_link = page_put_link,
5342 .permission = btrfs_permission,
5343 .setxattr = btrfs_setxattr,
5344 .getxattr = btrfs_getxattr,
5345 .listxattr = btrfs_listxattr,
5346 .removexattr = btrfs_removexattr,