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>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93 struct inode *inode, struct inode *dir)
97 err = btrfs_init_acl(trans, inode, dir);
99 err = btrfs_xattr_security_init(trans, inode, dir);
104 * this does all the hard work for inserting an inline extent into
105 * the btree. The caller should have done a btrfs_drop_extents so that
106 * no overlapping inline items exist in the btree
108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root, struct inode *inode,
110 u64 start, size_t size, size_t compressed_size,
111 struct page **compressed_pages)
113 struct btrfs_key key;
114 struct btrfs_path *path;
115 struct extent_buffer *leaf;
116 struct page *page = NULL;
119 struct btrfs_file_extent_item *ei;
122 size_t cur_size = size;
124 unsigned long offset;
125 int use_compress = 0;
127 if (compressed_size && compressed_pages) {
129 cur_size = compressed_size;
132 path = btrfs_alloc_path();
136 path->leave_spinning = 1;
137 btrfs_set_trans_block_group(trans, inode);
139 key.objectid = inode->i_ino;
141 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142 datasize = btrfs_file_extent_calc_inline_size(cur_size);
144 inode_add_bytes(inode, size);
145 ret = btrfs_insert_empty_item(trans, root, path, &key,
152 leaf = path->nodes[0];
153 ei = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_file_extent_item);
155 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157 btrfs_set_file_extent_encryption(leaf, ei, 0);
158 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160 ptr = btrfs_file_extent_inline_start(ei);
165 while (compressed_size > 0) {
166 cpage = compressed_pages[i];
167 cur_size = min_t(unsigned long, compressed_size,
170 kaddr = kmap_atomic(cpage, KM_USER0);
171 write_extent_buffer(leaf, kaddr, ptr, cur_size);
172 kunmap_atomic(kaddr, KM_USER0);
176 compressed_size -= cur_size;
178 btrfs_set_file_extent_compression(leaf, ei,
179 BTRFS_COMPRESS_ZLIB);
181 page = find_get_page(inode->i_mapping,
182 start >> PAGE_CACHE_SHIFT);
183 btrfs_set_file_extent_compression(leaf, ei, 0);
184 kaddr = kmap_atomic(page, KM_USER0);
185 offset = start & (PAGE_CACHE_SIZE - 1);
186 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187 kunmap_atomic(kaddr, KM_USER0);
188 page_cache_release(page);
190 btrfs_mark_buffer_dirty(leaf);
191 btrfs_free_path(path);
194 * we're an inline extent, so nobody can
195 * extend the file past i_size without locking
196 * a page we already have locked.
198 * We must do any isize and inode updates
199 * before we unlock the pages. Otherwise we
200 * could end up racing with unlink.
202 BTRFS_I(inode)->disk_i_size = inode->i_size;
203 btrfs_update_inode(trans, root, inode);
207 btrfs_free_path(path);
213 * conditionally insert an inline extent into the file. This
214 * does the checks required to make sure the data is small enough
215 * to fit as an inline extent.
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218 struct btrfs_root *root,
219 struct inode *inode, u64 start, u64 end,
220 size_t compressed_size,
221 struct page **compressed_pages)
223 u64 isize = i_size_read(inode);
224 u64 actual_end = min(end + 1, isize);
225 u64 inline_len = actual_end - start;
226 u64 aligned_end = (end + root->sectorsize - 1) &
227 ~((u64)root->sectorsize - 1);
229 u64 data_len = inline_len;
233 data_len = compressed_size;
236 actual_end >= PAGE_CACHE_SIZE ||
237 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
239 (actual_end & (root->sectorsize - 1)) == 0) ||
241 data_len > root->fs_info->max_inline) {
245 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
249 if (isize > actual_end)
250 inline_len = min_t(u64, isize, actual_end);
251 ret = insert_inline_extent(trans, root, inode, start,
252 inline_len, compressed_size,
255 btrfs_delalloc_release_metadata(inode, end + 1 - start);
256 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
260 struct async_extent {
265 unsigned long nr_pages;
266 struct list_head list;
271 struct btrfs_root *root;
272 struct page *locked_page;
275 struct list_head extents;
276 struct btrfs_work work;
279 static noinline int add_async_extent(struct async_cow *cow,
280 u64 start, u64 ram_size,
283 unsigned long nr_pages)
285 struct async_extent *async_extent;
287 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
288 async_extent->start = start;
289 async_extent->ram_size = ram_size;
290 async_extent->compressed_size = compressed_size;
291 async_extent->pages = pages;
292 async_extent->nr_pages = nr_pages;
293 list_add_tail(&async_extent->list, &cow->extents);
298 * we create compressed extents in two phases. The first
299 * phase compresses a range of pages that have already been
300 * locked (both pages and state bits are locked).
302 * This is done inside an ordered work queue, and the compression
303 * is spread across many cpus. The actual IO submission is step
304 * two, and the ordered work queue takes care of making sure that
305 * happens in the same order things were put onto the queue by
306 * writepages and friends.
308 * If this code finds it can't get good compression, it puts an
309 * entry onto the work queue to write the uncompressed bytes. This
310 * makes sure that both compressed inodes and uncompressed inodes
311 * are written in the same order that pdflush sent them down.
313 static noinline int compress_file_range(struct inode *inode,
314 struct page *locked_page,
316 struct async_cow *async_cow,
319 struct btrfs_root *root = BTRFS_I(inode)->root;
320 struct btrfs_trans_handle *trans;
324 u64 blocksize = root->sectorsize;
326 u64 isize = i_size_read(inode);
328 struct page **pages = NULL;
329 unsigned long nr_pages;
330 unsigned long nr_pages_ret = 0;
331 unsigned long total_compressed = 0;
332 unsigned long total_in = 0;
333 unsigned long max_compressed = 128 * 1024;
334 unsigned long max_uncompressed = 128 * 1024;
340 actual_end = min_t(u64, isize, end + 1);
343 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
344 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
347 * we don't want to send crud past the end of i_size through
348 * compression, that's just a waste of CPU time. So, if the
349 * end of the file is before the start of our current
350 * requested range of bytes, we bail out to the uncompressed
351 * cleanup code that can deal with all of this.
353 * It isn't really the fastest way to fix things, but this is a
354 * very uncommon corner.
356 if (actual_end <= start)
357 goto cleanup_and_bail_uncompressed;
359 total_compressed = actual_end - start;
361 /* we want to make sure that amount of ram required to uncompress
362 * an extent is reasonable, so we limit the total size in ram
363 * of a compressed extent to 128k. This is a crucial number
364 * because it also controls how easily we can spread reads across
365 * cpus for decompression.
367 * We also want to make sure the amount of IO required to do
368 * a random read is reasonably small, so we limit the size of
369 * a compressed extent to 128k.
371 total_compressed = min(total_compressed, max_uncompressed);
372 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
373 num_bytes = max(blocksize, num_bytes);
374 disk_num_bytes = num_bytes;
379 * we do compression for mount -o compress and when the
380 * inode has not been flagged as nocompress. This flag can
381 * change at any time if we discover bad compression ratios.
383 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
384 (btrfs_test_opt(root, COMPRESS) ||
385 (BTRFS_I(inode)->force_compress))) {
387 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
389 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
390 total_compressed, pages,
391 nr_pages, &nr_pages_ret,
397 unsigned long offset = total_compressed &
398 (PAGE_CACHE_SIZE - 1);
399 struct page *page = pages[nr_pages_ret - 1];
402 /* zero the tail end of the last page, we might be
403 * sending it down to disk
406 kaddr = kmap_atomic(page, KM_USER0);
407 memset(kaddr + offset, 0,
408 PAGE_CACHE_SIZE - offset);
409 kunmap_atomic(kaddr, KM_USER0);
415 trans = btrfs_join_transaction(root, 1);
417 btrfs_set_trans_block_group(trans, inode);
418 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
420 /* lets try to make an inline extent */
421 if (ret || total_in < (actual_end - start)) {
422 /* we didn't compress the entire range, try
423 * to make an uncompressed inline extent.
425 ret = cow_file_range_inline(trans, root, inode,
426 start, end, 0, NULL);
428 /* try making a compressed inline extent */
429 ret = cow_file_range_inline(trans, root, inode,
431 total_compressed, pages);
435 * inline extent creation worked, we don't need
436 * to create any more async work items. Unlock
437 * and free up our temp pages.
439 extent_clear_unlock_delalloc(inode,
440 &BTRFS_I(inode)->io_tree,
442 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
443 EXTENT_CLEAR_DELALLOC |
444 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
446 btrfs_end_transaction(trans, root);
449 btrfs_end_transaction(trans, root);
454 * we aren't doing an inline extent round the compressed size
455 * up to a block size boundary so the allocator does sane
458 total_compressed = (total_compressed + blocksize - 1) &
462 * one last check to make sure the compression is really a
463 * win, compare the page count read with the blocks on disk
465 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
466 ~(PAGE_CACHE_SIZE - 1);
467 if (total_compressed >= total_in) {
470 disk_num_bytes = total_compressed;
471 num_bytes = total_in;
474 if (!will_compress && pages) {
476 * the compression code ran but failed to make things smaller,
477 * free any pages it allocated and our page pointer array
479 for (i = 0; i < nr_pages_ret; i++) {
480 WARN_ON(pages[i]->mapping);
481 page_cache_release(pages[i]);
485 total_compressed = 0;
488 /* flag the file so we don't compress in the future */
489 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
490 !(BTRFS_I(inode)->force_compress)) {
491 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
497 /* the async work queues will take care of doing actual
498 * allocation on disk for these compressed pages,
499 * and will submit them to the elevator.
501 add_async_extent(async_cow, start, num_bytes,
502 total_compressed, pages, nr_pages_ret);
504 if (start + num_bytes < end && start + num_bytes < actual_end) {
511 cleanup_and_bail_uncompressed:
513 * No compression, but we still need to write the pages in
514 * the file we've been given so far. redirty the locked
515 * page if it corresponds to our extent and set things up
516 * for the async work queue to run cow_file_range to do
517 * the normal delalloc dance
519 if (page_offset(locked_page) >= start &&
520 page_offset(locked_page) <= end) {
521 __set_page_dirty_nobuffers(locked_page);
522 /* unlocked later on in the async handlers */
524 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
532 for (i = 0; i < nr_pages_ret; i++) {
533 WARN_ON(pages[i]->mapping);
534 page_cache_release(pages[i]);
542 * phase two of compressed writeback. This is the ordered portion
543 * of the code, which only gets called in the order the work was
544 * queued. We walk all the async extents created by compress_file_range
545 * and send them down to the disk.
547 static noinline int submit_compressed_extents(struct inode *inode,
548 struct async_cow *async_cow)
550 struct async_extent *async_extent;
552 struct btrfs_trans_handle *trans;
553 struct btrfs_key ins;
554 struct extent_map *em;
555 struct btrfs_root *root = BTRFS_I(inode)->root;
556 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
557 struct extent_io_tree *io_tree;
560 if (list_empty(&async_cow->extents))
564 while (!list_empty(&async_cow->extents)) {
565 async_extent = list_entry(async_cow->extents.next,
566 struct async_extent, list);
567 list_del(&async_extent->list);
569 io_tree = &BTRFS_I(inode)->io_tree;
572 /* did the compression code fall back to uncompressed IO? */
573 if (!async_extent->pages) {
574 int page_started = 0;
575 unsigned long nr_written = 0;
577 lock_extent(io_tree, async_extent->start,
578 async_extent->start +
579 async_extent->ram_size - 1, GFP_NOFS);
581 /* allocate blocks */
582 ret = cow_file_range(inode, async_cow->locked_page,
584 async_extent->start +
585 async_extent->ram_size - 1,
586 &page_started, &nr_written, 0);
589 * if page_started, cow_file_range inserted an
590 * inline extent and took care of all the unlocking
591 * and IO for us. Otherwise, we need to submit
592 * all those pages down to the drive.
594 if (!page_started && !ret)
595 extent_write_locked_range(io_tree,
596 inode, async_extent->start,
597 async_extent->start +
598 async_extent->ram_size - 1,
606 lock_extent(io_tree, async_extent->start,
607 async_extent->start + async_extent->ram_size - 1,
610 trans = btrfs_join_transaction(root, 1);
611 ret = btrfs_reserve_extent(trans, root,
612 async_extent->compressed_size,
613 async_extent->compressed_size,
616 btrfs_end_transaction(trans, root);
620 for (i = 0; i < async_extent->nr_pages; i++) {
621 WARN_ON(async_extent->pages[i]->mapping);
622 page_cache_release(async_extent->pages[i]);
624 kfree(async_extent->pages);
625 async_extent->nr_pages = 0;
626 async_extent->pages = NULL;
627 unlock_extent(io_tree, async_extent->start,
628 async_extent->start +
629 async_extent->ram_size - 1, GFP_NOFS);
634 * here we're doing allocation and writeback of the
637 btrfs_drop_extent_cache(inode, async_extent->start,
638 async_extent->start +
639 async_extent->ram_size - 1, 0);
641 em = alloc_extent_map(GFP_NOFS);
642 em->start = async_extent->start;
643 em->len = async_extent->ram_size;
644 em->orig_start = em->start;
646 em->block_start = ins.objectid;
647 em->block_len = ins.offset;
648 em->bdev = root->fs_info->fs_devices->latest_bdev;
649 set_bit(EXTENT_FLAG_PINNED, &em->flags);
650 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
653 write_lock(&em_tree->lock);
654 ret = add_extent_mapping(em_tree, em);
655 write_unlock(&em_tree->lock);
656 if (ret != -EEXIST) {
660 btrfs_drop_extent_cache(inode, async_extent->start,
661 async_extent->start +
662 async_extent->ram_size - 1, 0);
665 ret = btrfs_add_ordered_extent(inode, async_extent->start,
667 async_extent->ram_size,
669 BTRFS_ORDERED_COMPRESSED);
673 * clear dirty, set writeback and unlock the pages.
675 extent_clear_unlock_delalloc(inode,
676 &BTRFS_I(inode)->io_tree,
678 async_extent->start +
679 async_extent->ram_size - 1,
680 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
681 EXTENT_CLEAR_UNLOCK |
682 EXTENT_CLEAR_DELALLOC |
683 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
685 ret = btrfs_submit_compressed_write(inode,
687 async_extent->ram_size,
689 ins.offset, async_extent->pages,
690 async_extent->nr_pages);
693 alloc_hint = ins.objectid + ins.offset;
701 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
704 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
705 struct extent_map *em;
708 read_lock(&em_tree->lock);
709 em = search_extent_mapping(em_tree, start, num_bytes);
712 * if block start isn't an actual block number then find the
713 * first block in this inode and use that as a hint. If that
714 * block is also bogus then just don't worry about it.
716 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
718 em = search_extent_mapping(em_tree, 0, 0);
719 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
720 alloc_hint = em->block_start;
724 alloc_hint = em->block_start;
728 read_unlock(&em_tree->lock);
734 * when extent_io.c finds a delayed allocation range in the file,
735 * the call backs end up in this code. The basic idea is to
736 * allocate extents on disk for the range, and create ordered data structs
737 * in ram to track those extents.
739 * locked_page is the page that writepage had locked already. We use
740 * it to make sure we don't do extra locks or unlocks.
742 * *page_started is set to one if we unlock locked_page and do everything
743 * required to start IO on it. It may be clean and already done with
746 static noinline int cow_file_range(struct inode *inode,
747 struct page *locked_page,
748 u64 start, u64 end, int *page_started,
749 unsigned long *nr_written,
752 struct btrfs_root *root = BTRFS_I(inode)->root;
753 struct btrfs_trans_handle *trans;
756 unsigned long ram_size;
759 u64 blocksize = root->sectorsize;
761 u64 isize = i_size_read(inode);
762 struct btrfs_key ins;
763 struct extent_map *em;
764 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
767 trans = btrfs_join_transaction(root, 1);
769 btrfs_set_trans_block_group(trans, inode);
770 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
772 actual_end = min_t(u64, isize, end + 1);
774 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
775 num_bytes = max(blocksize, num_bytes);
776 disk_num_bytes = num_bytes;
780 /* lets try to make an inline extent */
781 ret = cow_file_range_inline(trans, root, inode,
782 start, end, 0, NULL);
784 extent_clear_unlock_delalloc(inode,
785 &BTRFS_I(inode)->io_tree,
787 EXTENT_CLEAR_UNLOCK_PAGE |
788 EXTENT_CLEAR_UNLOCK |
789 EXTENT_CLEAR_DELALLOC |
791 EXTENT_SET_WRITEBACK |
792 EXTENT_END_WRITEBACK);
794 *nr_written = *nr_written +
795 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
802 BUG_ON(disk_num_bytes >
803 btrfs_super_total_bytes(&root->fs_info->super_copy));
805 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
806 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
808 while (disk_num_bytes > 0) {
811 cur_alloc_size = disk_num_bytes;
812 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
813 root->sectorsize, 0, alloc_hint,
817 em = alloc_extent_map(GFP_NOFS);
819 em->orig_start = em->start;
820 ram_size = ins.offset;
821 em->len = ins.offset;
823 em->block_start = ins.objectid;
824 em->block_len = ins.offset;
825 em->bdev = root->fs_info->fs_devices->latest_bdev;
826 set_bit(EXTENT_FLAG_PINNED, &em->flags);
829 write_lock(&em_tree->lock);
830 ret = add_extent_mapping(em_tree, em);
831 write_unlock(&em_tree->lock);
832 if (ret != -EEXIST) {
836 btrfs_drop_extent_cache(inode, start,
837 start + ram_size - 1, 0);
840 cur_alloc_size = ins.offset;
841 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
842 ram_size, cur_alloc_size, 0);
845 if (root->root_key.objectid ==
846 BTRFS_DATA_RELOC_TREE_OBJECTID) {
847 ret = btrfs_reloc_clone_csums(inode, start,
852 if (disk_num_bytes < cur_alloc_size)
855 /* we're not doing compressed IO, don't unlock the first
856 * page (which the caller expects to stay locked), don't
857 * clear any dirty bits and don't set any writeback bits
859 * Do set the Private2 bit so we know this page was properly
860 * setup for writepage
862 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
863 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
866 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
867 start, start + ram_size - 1,
869 disk_num_bytes -= cur_alloc_size;
870 num_bytes -= cur_alloc_size;
871 alloc_hint = ins.objectid + ins.offset;
872 start += cur_alloc_size;
876 btrfs_end_transaction(trans, root);
882 * work queue call back to started compression on a file and pages
884 static noinline void async_cow_start(struct btrfs_work *work)
886 struct async_cow *async_cow;
888 async_cow = container_of(work, struct async_cow, work);
890 compress_file_range(async_cow->inode, async_cow->locked_page,
891 async_cow->start, async_cow->end, async_cow,
894 async_cow->inode = NULL;
898 * work queue call back to submit previously compressed pages
900 static noinline void async_cow_submit(struct btrfs_work *work)
902 struct async_cow *async_cow;
903 struct btrfs_root *root;
904 unsigned long nr_pages;
906 async_cow = container_of(work, struct async_cow, work);
908 root = async_cow->root;
909 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
912 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
914 if (atomic_read(&root->fs_info->async_delalloc_pages) <
916 waitqueue_active(&root->fs_info->async_submit_wait))
917 wake_up(&root->fs_info->async_submit_wait);
919 if (async_cow->inode)
920 submit_compressed_extents(async_cow->inode, async_cow);
923 static noinline void async_cow_free(struct btrfs_work *work)
925 struct async_cow *async_cow;
926 async_cow = container_of(work, struct async_cow, work);
930 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
931 u64 start, u64 end, int *page_started,
932 unsigned long *nr_written)
934 struct async_cow *async_cow;
935 struct btrfs_root *root = BTRFS_I(inode)->root;
936 unsigned long nr_pages;
938 int limit = 10 * 1024 * 1042;
940 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
941 1, 0, NULL, GFP_NOFS);
942 while (start < end) {
943 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
944 async_cow->inode = inode;
945 async_cow->root = root;
946 async_cow->locked_page = locked_page;
947 async_cow->start = start;
949 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
952 cur_end = min(end, start + 512 * 1024 - 1);
954 async_cow->end = cur_end;
955 INIT_LIST_HEAD(&async_cow->extents);
957 async_cow->work.func = async_cow_start;
958 async_cow->work.ordered_func = async_cow_submit;
959 async_cow->work.ordered_free = async_cow_free;
960 async_cow->work.flags = 0;
962 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
964 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
966 btrfs_queue_worker(&root->fs_info->delalloc_workers,
969 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
970 wait_event(root->fs_info->async_submit_wait,
971 (atomic_read(&root->fs_info->async_delalloc_pages) <
975 while (atomic_read(&root->fs_info->async_submit_draining) &&
976 atomic_read(&root->fs_info->async_delalloc_pages)) {
977 wait_event(root->fs_info->async_submit_wait,
978 (atomic_read(&root->fs_info->async_delalloc_pages) ==
982 *nr_written += nr_pages;
989 static noinline int csum_exist_in_range(struct btrfs_root *root,
990 u64 bytenr, u64 num_bytes)
993 struct btrfs_ordered_sum *sums;
996 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
997 bytenr + num_bytes - 1, &list);
998 if (ret == 0 && list_empty(&list))
1001 while (!list_empty(&list)) {
1002 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1003 list_del(&sums->list);
1010 * when nowcow writeback call back. This checks for snapshots or COW copies
1011 * of the extents that exist in the file, and COWs the file as required.
1013 * If no cow copies or snapshots exist, we write directly to the existing
1016 static noinline int run_delalloc_nocow(struct inode *inode,
1017 struct page *locked_page,
1018 u64 start, u64 end, int *page_started, int force,
1019 unsigned long *nr_written)
1021 struct btrfs_root *root = BTRFS_I(inode)->root;
1022 struct btrfs_trans_handle *trans;
1023 struct extent_buffer *leaf;
1024 struct btrfs_path *path;
1025 struct btrfs_file_extent_item *fi;
1026 struct btrfs_key found_key;
1039 path = btrfs_alloc_path();
1041 trans = btrfs_join_transaction(root, 1);
1044 cow_start = (u64)-1;
1047 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1050 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1051 leaf = path->nodes[0];
1052 btrfs_item_key_to_cpu(leaf, &found_key,
1053 path->slots[0] - 1);
1054 if (found_key.objectid == inode->i_ino &&
1055 found_key.type == BTRFS_EXTENT_DATA_KEY)
1060 leaf = path->nodes[0];
1061 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1062 ret = btrfs_next_leaf(root, path);
1067 leaf = path->nodes[0];
1073 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1075 if (found_key.objectid > inode->i_ino ||
1076 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1077 found_key.offset > end)
1080 if (found_key.offset > cur_offset) {
1081 extent_end = found_key.offset;
1086 fi = btrfs_item_ptr(leaf, path->slots[0],
1087 struct btrfs_file_extent_item);
1088 extent_type = btrfs_file_extent_type(leaf, fi);
1090 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1091 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1092 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1093 extent_offset = btrfs_file_extent_offset(leaf, fi);
1094 extent_end = found_key.offset +
1095 btrfs_file_extent_num_bytes(leaf, fi);
1096 if (extent_end <= start) {
1100 if (disk_bytenr == 0)
1102 if (btrfs_file_extent_compression(leaf, fi) ||
1103 btrfs_file_extent_encryption(leaf, fi) ||
1104 btrfs_file_extent_other_encoding(leaf, fi))
1106 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1108 if (btrfs_extent_readonly(root, disk_bytenr))
1110 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1112 extent_offset, disk_bytenr))
1114 disk_bytenr += extent_offset;
1115 disk_bytenr += cur_offset - found_key.offset;
1116 num_bytes = min(end + 1, extent_end) - cur_offset;
1118 * force cow if csum exists in the range.
1119 * this ensure that csum for a given extent are
1120 * either valid or do not exist.
1122 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1125 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1126 extent_end = found_key.offset +
1127 btrfs_file_extent_inline_len(leaf, fi);
1128 extent_end = ALIGN(extent_end, root->sectorsize);
1133 if (extent_end <= start) {
1138 if (cow_start == (u64)-1)
1139 cow_start = cur_offset;
1140 cur_offset = extent_end;
1141 if (cur_offset > end)
1147 btrfs_release_path(root, path);
1148 if (cow_start != (u64)-1) {
1149 ret = cow_file_range(inode, locked_page, cow_start,
1150 found_key.offset - 1, page_started,
1153 cow_start = (u64)-1;
1156 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1157 struct extent_map *em;
1158 struct extent_map_tree *em_tree;
1159 em_tree = &BTRFS_I(inode)->extent_tree;
1160 em = alloc_extent_map(GFP_NOFS);
1161 em->start = cur_offset;
1162 em->orig_start = em->start;
1163 em->len = num_bytes;
1164 em->block_len = num_bytes;
1165 em->block_start = disk_bytenr;
1166 em->bdev = root->fs_info->fs_devices->latest_bdev;
1167 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1169 write_lock(&em_tree->lock);
1170 ret = add_extent_mapping(em_tree, em);
1171 write_unlock(&em_tree->lock);
1172 if (ret != -EEXIST) {
1173 free_extent_map(em);
1176 btrfs_drop_extent_cache(inode, em->start,
1177 em->start + em->len - 1, 0);
1179 type = BTRFS_ORDERED_PREALLOC;
1181 type = BTRFS_ORDERED_NOCOW;
1184 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1185 num_bytes, num_bytes, type);
1188 if (root->root_key.objectid ==
1189 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1190 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1195 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1196 cur_offset, cur_offset + num_bytes - 1,
1197 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1198 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1199 EXTENT_SET_PRIVATE2);
1200 cur_offset = extent_end;
1201 if (cur_offset > end)
1204 btrfs_release_path(root, path);
1206 if (cur_offset <= end && cow_start == (u64)-1)
1207 cow_start = cur_offset;
1208 if (cow_start != (u64)-1) {
1209 ret = cow_file_range(inode, locked_page, cow_start, end,
1210 page_started, nr_written, 1);
1214 ret = btrfs_end_transaction(trans, root);
1216 btrfs_free_path(path);
1221 * extent_io.c call back to do delayed allocation processing
1223 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1224 u64 start, u64 end, int *page_started,
1225 unsigned long *nr_written)
1228 struct btrfs_root *root = BTRFS_I(inode)->root;
1230 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1231 ret = run_delalloc_nocow(inode, locked_page, start, end,
1232 page_started, 1, nr_written);
1233 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1234 ret = run_delalloc_nocow(inode, locked_page, start, end,
1235 page_started, 0, nr_written);
1236 else if (!btrfs_test_opt(root, COMPRESS) &&
1237 !(BTRFS_I(inode)->force_compress))
1238 ret = cow_file_range(inode, locked_page, start, end,
1239 page_started, nr_written, 1);
1241 ret = cow_file_range_async(inode, locked_page, start, end,
1242 page_started, nr_written);
1246 static int btrfs_split_extent_hook(struct inode *inode,
1247 struct extent_state *orig, u64 split)
1249 /* not delalloc, ignore it */
1250 if (!(orig->state & EXTENT_DELALLOC))
1253 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1258 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1259 * extents so we can keep track of new extents that are just merged onto old
1260 * extents, such as when we are doing sequential writes, so we can properly
1261 * account for the metadata space we'll need.
1263 static int btrfs_merge_extent_hook(struct inode *inode,
1264 struct extent_state *new,
1265 struct extent_state *other)
1267 /* not delalloc, ignore it */
1268 if (!(other->state & EXTENT_DELALLOC))
1271 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1276 * extent_io.c set_bit_hook, used to track delayed allocation
1277 * bytes in this file, and to maintain the list of inodes that
1278 * have pending delalloc work to be done.
1280 static int btrfs_set_bit_hook(struct inode *inode,
1281 struct extent_state *state, int *bits)
1285 * set_bit and clear bit hooks normally require _irqsave/restore
1286 * but in this case, we are only testeing for the DELALLOC
1287 * bit, which is only set or cleared with irqs on
1289 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1290 struct btrfs_root *root = BTRFS_I(inode)->root;
1291 u64 len = state->end + 1 - state->start;
1293 if (*bits & EXTENT_FIRST_DELALLOC)
1294 *bits &= ~EXTENT_FIRST_DELALLOC;
1296 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1298 spin_lock(&root->fs_info->delalloc_lock);
1299 BTRFS_I(inode)->delalloc_bytes += len;
1300 root->fs_info->delalloc_bytes += len;
1301 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1302 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1303 &root->fs_info->delalloc_inodes);
1305 spin_unlock(&root->fs_info->delalloc_lock);
1311 * extent_io.c clear_bit_hook, see set_bit_hook for why
1313 static int btrfs_clear_bit_hook(struct inode *inode,
1314 struct extent_state *state, int *bits)
1317 * set_bit and clear bit hooks normally require _irqsave/restore
1318 * but in this case, we are only testeing for the DELALLOC
1319 * bit, which is only set or cleared with irqs on
1321 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1322 struct btrfs_root *root = BTRFS_I(inode)->root;
1323 u64 len = state->end + 1 - state->start;
1325 if (*bits & EXTENT_FIRST_DELALLOC)
1326 *bits &= ~EXTENT_FIRST_DELALLOC;
1327 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1328 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1330 if (*bits & EXTENT_DO_ACCOUNTING)
1331 btrfs_delalloc_release_metadata(inode, len);
1333 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1334 btrfs_free_reserved_data_space(inode, len);
1336 spin_lock(&root->fs_info->delalloc_lock);
1337 root->fs_info->delalloc_bytes -= len;
1338 BTRFS_I(inode)->delalloc_bytes -= len;
1340 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1341 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1342 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1344 spin_unlock(&root->fs_info->delalloc_lock);
1350 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1351 * we don't create bios that span stripes or chunks
1353 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1354 size_t size, struct bio *bio,
1355 unsigned long bio_flags)
1357 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1358 struct btrfs_mapping_tree *map_tree;
1359 u64 logical = (u64)bio->bi_sector << 9;
1364 if (bio_flags & EXTENT_BIO_COMPRESSED)
1367 length = bio->bi_size;
1368 map_tree = &root->fs_info->mapping_tree;
1369 map_length = length;
1370 ret = btrfs_map_block(map_tree, READ, logical,
1371 &map_length, NULL, 0);
1373 if (map_length < length + size)
1379 * in order to insert checksums into the metadata in large chunks,
1380 * we wait until bio submission time. All the pages in the bio are
1381 * checksummed and sums are attached onto the ordered extent record.
1383 * At IO completion time the cums attached on the ordered extent record
1384 * are inserted into the btree
1386 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1387 struct bio *bio, int mirror_num,
1388 unsigned long bio_flags,
1391 struct btrfs_root *root = BTRFS_I(inode)->root;
1394 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1400 * in order to insert checksums into the metadata in large chunks,
1401 * we wait until bio submission time. All the pages in the bio are
1402 * checksummed and sums are attached onto the ordered extent record.
1404 * At IO completion time the cums attached on the ordered extent record
1405 * are inserted into the btree
1407 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1408 int mirror_num, unsigned long bio_flags,
1411 struct btrfs_root *root = BTRFS_I(inode)->root;
1412 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1416 * extent_io.c submission hook. This does the right thing for csum calculation
1417 * on write, or reading the csums from the tree before a read
1419 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1420 int mirror_num, unsigned long bio_flags,
1423 struct btrfs_root *root = BTRFS_I(inode)->root;
1427 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1429 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1432 if (!(rw & (1 << BIO_RW))) {
1433 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1434 return btrfs_submit_compressed_read(inode, bio,
1435 mirror_num, bio_flags);
1436 } else if (!skip_sum)
1437 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1439 } else if (!skip_sum) {
1440 /* csum items have already been cloned */
1441 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1443 /* we're doing a write, do the async checksumming */
1444 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1445 inode, rw, bio, mirror_num,
1446 bio_flags, bio_offset,
1447 __btrfs_submit_bio_start,
1448 __btrfs_submit_bio_done);
1452 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1456 * given a list of ordered sums record them in the inode. This happens
1457 * at IO completion time based on sums calculated at bio submission time.
1459 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1460 struct inode *inode, u64 file_offset,
1461 struct list_head *list)
1463 struct btrfs_ordered_sum *sum;
1465 btrfs_set_trans_block_group(trans, inode);
1467 list_for_each_entry(sum, list, list) {
1468 btrfs_csum_file_blocks(trans,
1469 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1474 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1475 struct extent_state **cached_state)
1477 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1479 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1480 cached_state, GFP_NOFS);
1483 /* see btrfs_writepage_start_hook for details on why this is required */
1484 struct btrfs_writepage_fixup {
1486 struct btrfs_work work;
1489 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1491 struct btrfs_writepage_fixup *fixup;
1492 struct btrfs_ordered_extent *ordered;
1493 struct extent_state *cached_state = NULL;
1495 struct inode *inode;
1499 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1503 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1504 ClearPageChecked(page);
1508 inode = page->mapping->host;
1509 page_start = page_offset(page);
1510 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1512 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1513 &cached_state, GFP_NOFS);
1515 /* already ordered? We're done */
1516 if (PagePrivate2(page))
1519 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1521 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1522 page_end, &cached_state, GFP_NOFS);
1524 btrfs_start_ordered_extent(inode, ordered, 1);
1529 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1530 ClearPageChecked(page);
1532 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1533 &cached_state, GFP_NOFS);
1536 page_cache_release(page);
1540 * There are a few paths in the higher layers of the kernel that directly
1541 * set the page dirty bit without asking the filesystem if it is a
1542 * good idea. This causes problems because we want to make sure COW
1543 * properly happens and the data=ordered rules are followed.
1545 * In our case any range that doesn't have the ORDERED bit set
1546 * hasn't been properly setup for IO. We kick off an async process
1547 * to fix it up. The async helper will wait for ordered extents, set
1548 * the delalloc bit and make it safe to write the page.
1550 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1552 struct inode *inode = page->mapping->host;
1553 struct btrfs_writepage_fixup *fixup;
1554 struct btrfs_root *root = BTRFS_I(inode)->root;
1556 /* this page is properly in the ordered list */
1557 if (TestClearPagePrivate2(page))
1560 if (PageChecked(page))
1563 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1567 SetPageChecked(page);
1568 page_cache_get(page);
1569 fixup->work.func = btrfs_writepage_fixup_worker;
1571 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1575 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1576 struct inode *inode, u64 file_pos,
1577 u64 disk_bytenr, u64 disk_num_bytes,
1578 u64 num_bytes, u64 ram_bytes,
1579 u8 compression, u8 encryption,
1580 u16 other_encoding, int extent_type)
1582 struct btrfs_root *root = BTRFS_I(inode)->root;
1583 struct btrfs_file_extent_item *fi;
1584 struct btrfs_path *path;
1585 struct extent_buffer *leaf;
1586 struct btrfs_key ins;
1590 path = btrfs_alloc_path();
1593 path->leave_spinning = 1;
1596 * we may be replacing one extent in the tree with another.
1597 * The new extent is pinned in the extent map, and we don't want
1598 * to drop it from the cache until it is completely in the btree.
1600 * So, tell btrfs_drop_extents to leave this extent in the cache.
1601 * the caller is expected to unpin it and allow it to be merged
1604 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1608 ins.objectid = inode->i_ino;
1609 ins.offset = file_pos;
1610 ins.type = BTRFS_EXTENT_DATA_KEY;
1611 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1613 leaf = path->nodes[0];
1614 fi = btrfs_item_ptr(leaf, path->slots[0],
1615 struct btrfs_file_extent_item);
1616 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1617 btrfs_set_file_extent_type(leaf, fi, extent_type);
1618 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1619 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1620 btrfs_set_file_extent_offset(leaf, fi, 0);
1621 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1622 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1623 btrfs_set_file_extent_compression(leaf, fi, compression);
1624 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1625 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1627 btrfs_unlock_up_safe(path, 1);
1628 btrfs_set_lock_blocking(leaf);
1630 btrfs_mark_buffer_dirty(leaf);
1632 inode_add_bytes(inode, num_bytes);
1634 ins.objectid = disk_bytenr;
1635 ins.offset = disk_num_bytes;
1636 ins.type = BTRFS_EXTENT_ITEM_KEY;
1637 ret = btrfs_alloc_reserved_file_extent(trans, root,
1638 root->root_key.objectid,
1639 inode->i_ino, file_pos, &ins);
1641 btrfs_free_path(path);
1647 * helper function for btrfs_finish_ordered_io, this
1648 * just reads in some of the csum leaves to prime them into ram
1649 * before we start the transaction. It limits the amount of btree
1650 * reads required while inside the transaction.
1652 /* as ordered data IO finishes, this gets called so we can finish
1653 * an ordered extent if the range of bytes in the file it covers are
1656 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1658 struct btrfs_root *root = BTRFS_I(inode)->root;
1659 struct btrfs_trans_handle *trans = NULL;
1660 struct btrfs_ordered_extent *ordered_extent = NULL;
1661 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1662 struct extent_state *cached_state = NULL;
1666 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1670 BUG_ON(!ordered_extent);
1672 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1673 BUG_ON(!list_empty(&ordered_extent->list));
1674 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1676 trans = btrfs_join_transaction(root, 1);
1677 btrfs_set_trans_block_group(trans, inode);
1678 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1679 ret = btrfs_update_inode(trans, root, inode);
1685 lock_extent_bits(io_tree, ordered_extent->file_offset,
1686 ordered_extent->file_offset + ordered_extent->len - 1,
1687 0, &cached_state, GFP_NOFS);
1689 trans = btrfs_join_transaction(root, 1);
1690 btrfs_set_trans_block_group(trans, inode);
1691 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1693 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1695 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1697 ret = btrfs_mark_extent_written(trans, inode,
1698 ordered_extent->file_offset,
1699 ordered_extent->file_offset +
1700 ordered_extent->len);
1703 ret = insert_reserved_file_extent(trans, inode,
1704 ordered_extent->file_offset,
1705 ordered_extent->start,
1706 ordered_extent->disk_len,
1707 ordered_extent->len,
1708 ordered_extent->len,
1710 BTRFS_FILE_EXTENT_REG);
1711 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1712 ordered_extent->file_offset,
1713 ordered_extent->len);
1716 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1717 ordered_extent->file_offset +
1718 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1720 add_pending_csums(trans, inode, ordered_extent->file_offset,
1721 &ordered_extent->list);
1723 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1724 ret = btrfs_update_inode(trans, root, inode);
1727 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1729 btrfs_end_transaction(trans, root);
1731 btrfs_put_ordered_extent(ordered_extent);
1732 /* once for the tree */
1733 btrfs_put_ordered_extent(ordered_extent);
1738 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1739 struct extent_state *state, int uptodate)
1741 ClearPagePrivate2(page);
1742 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1746 * When IO fails, either with EIO or csum verification fails, we
1747 * try other mirrors that might have a good copy of the data. This
1748 * io_failure_record is used to record state as we go through all the
1749 * mirrors. If another mirror has good data, the page is set up to date
1750 * and things continue. If a good mirror can't be found, the original
1751 * bio end_io callback is called to indicate things have failed.
1753 struct io_failure_record {
1758 unsigned long bio_flags;
1762 static int btrfs_io_failed_hook(struct bio *failed_bio,
1763 struct page *page, u64 start, u64 end,
1764 struct extent_state *state)
1766 struct io_failure_record *failrec = NULL;
1768 struct extent_map *em;
1769 struct inode *inode = page->mapping->host;
1770 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1771 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1778 ret = get_state_private(failure_tree, start, &private);
1780 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1783 failrec->start = start;
1784 failrec->len = end - start + 1;
1785 failrec->last_mirror = 0;
1786 failrec->bio_flags = 0;
1788 read_lock(&em_tree->lock);
1789 em = lookup_extent_mapping(em_tree, start, failrec->len);
1790 if (em->start > start || em->start + em->len < start) {
1791 free_extent_map(em);
1794 read_unlock(&em_tree->lock);
1796 if (!em || IS_ERR(em)) {
1800 logical = start - em->start;
1801 logical = em->block_start + logical;
1802 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1803 logical = em->block_start;
1804 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1806 failrec->logical = logical;
1807 free_extent_map(em);
1808 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1809 EXTENT_DIRTY, GFP_NOFS);
1810 set_state_private(failure_tree, start,
1811 (u64)(unsigned long)failrec);
1813 failrec = (struct io_failure_record *)(unsigned long)private;
1815 num_copies = btrfs_num_copies(
1816 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1817 failrec->logical, failrec->len);
1818 failrec->last_mirror++;
1820 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1821 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1824 if (state && state->start != failrec->start)
1826 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1828 if (!state || failrec->last_mirror > num_copies) {
1829 set_state_private(failure_tree, failrec->start, 0);
1830 clear_extent_bits(failure_tree, failrec->start,
1831 failrec->start + failrec->len - 1,
1832 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1836 bio = bio_alloc(GFP_NOFS, 1);
1837 bio->bi_private = state;
1838 bio->bi_end_io = failed_bio->bi_end_io;
1839 bio->bi_sector = failrec->logical >> 9;
1840 bio->bi_bdev = failed_bio->bi_bdev;
1843 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1844 if (failed_bio->bi_rw & (1 << BIO_RW))
1849 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1850 failrec->last_mirror,
1851 failrec->bio_flags, 0);
1856 * each time an IO finishes, we do a fast check in the IO failure tree
1857 * to see if we need to process or clean up an io_failure_record
1859 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1862 u64 private_failure;
1863 struct io_failure_record *failure;
1867 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1868 (u64)-1, 1, EXTENT_DIRTY)) {
1869 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1870 start, &private_failure);
1872 failure = (struct io_failure_record *)(unsigned long)
1874 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1876 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1878 failure->start + failure->len - 1,
1879 EXTENT_DIRTY | EXTENT_LOCKED,
1888 * when reads are done, we need to check csums to verify the data is correct
1889 * if there's a match, we allow the bio to finish. If not, we go through
1890 * the io_failure_record routines to find good copies
1892 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1893 struct extent_state *state)
1895 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1896 struct inode *inode = page->mapping->host;
1897 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1899 u64 private = ~(u32)0;
1901 struct btrfs_root *root = BTRFS_I(inode)->root;
1904 if (PageChecked(page)) {
1905 ClearPageChecked(page);
1909 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1912 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1913 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1914 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1919 if (state && state->start == start) {
1920 private = state->private;
1923 ret = get_state_private(io_tree, start, &private);
1925 kaddr = kmap_atomic(page, KM_USER0);
1929 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1930 btrfs_csum_final(csum, (char *)&csum);
1931 if (csum != private)
1934 kunmap_atomic(kaddr, KM_USER0);
1936 /* if the io failure tree for this inode is non-empty,
1937 * check to see if we've recovered from a failed IO
1939 btrfs_clean_io_failures(inode, start);
1943 if (printk_ratelimit()) {
1944 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1945 "private %llu\n", page->mapping->host->i_ino,
1946 (unsigned long long)start, csum,
1947 (unsigned long long)private);
1949 memset(kaddr + offset, 1, end - start + 1);
1950 flush_dcache_page(page);
1951 kunmap_atomic(kaddr, KM_USER0);
1957 struct delayed_iput {
1958 struct list_head list;
1959 struct inode *inode;
1962 void btrfs_add_delayed_iput(struct inode *inode)
1964 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1965 struct delayed_iput *delayed;
1967 if (atomic_add_unless(&inode->i_count, -1, 1))
1970 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1971 delayed->inode = inode;
1973 spin_lock(&fs_info->delayed_iput_lock);
1974 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1975 spin_unlock(&fs_info->delayed_iput_lock);
1978 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1981 struct btrfs_fs_info *fs_info = root->fs_info;
1982 struct delayed_iput *delayed;
1985 spin_lock(&fs_info->delayed_iput_lock);
1986 empty = list_empty(&fs_info->delayed_iputs);
1987 spin_unlock(&fs_info->delayed_iput_lock);
1991 down_read(&root->fs_info->cleanup_work_sem);
1992 spin_lock(&fs_info->delayed_iput_lock);
1993 list_splice_init(&fs_info->delayed_iputs, &list);
1994 spin_unlock(&fs_info->delayed_iput_lock);
1996 while (!list_empty(&list)) {
1997 delayed = list_entry(list.next, struct delayed_iput, list);
1998 list_del(&delayed->list);
1999 iput(delayed->inode);
2002 up_read(&root->fs_info->cleanup_work_sem);
2006 * calculate extra metadata reservation when snapshotting a subvolume
2007 * contains orphan files.
2009 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2010 struct btrfs_pending_snapshot *pending,
2011 u64 *bytes_to_reserve)
2013 struct btrfs_root *root;
2014 struct btrfs_block_rsv *block_rsv;
2018 root = pending->root;
2019 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2022 block_rsv = root->orphan_block_rsv;
2024 /* orphan block reservation for the snapshot */
2025 num_bytes = block_rsv->size;
2028 * after the snapshot is created, COWing tree blocks may use more
2029 * space than it frees. So we should make sure there is enough
2032 index = trans->transid & 0x1;
2033 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2034 num_bytes += block_rsv->size -
2035 (block_rsv->reserved + block_rsv->freed[index]);
2038 *bytes_to_reserve += num_bytes;
2041 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2042 struct btrfs_pending_snapshot *pending)
2044 struct btrfs_root *root = pending->root;
2045 struct btrfs_root *snap = pending->snap;
2046 struct btrfs_block_rsv *block_rsv;
2051 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2054 /* refill source subvolume's orphan block reservation */
2055 block_rsv = root->orphan_block_rsv;
2056 index = trans->transid & 0x1;
2057 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2058 num_bytes = block_rsv->size -
2059 (block_rsv->reserved + block_rsv->freed[index]);
2060 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2061 root->orphan_block_rsv,
2066 /* setup orphan block reservation for the snapshot */
2067 block_rsv = btrfs_alloc_block_rsv(snap);
2070 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2071 snap->orphan_block_rsv = block_rsv;
2073 num_bytes = root->orphan_block_rsv->size;
2074 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2075 block_rsv, num_bytes);
2079 /* insert orphan item for the snapshot */
2080 WARN_ON(!root->orphan_item_inserted);
2081 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2082 snap->root_key.objectid);
2084 snap->orphan_item_inserted = 1;
2088 enum btrfs_orphan_cleanup_state {
2089 ORPHAN_CLEANUP_STARTED = 1,
2090 ORPHAN_CLEANUP_DONE = 2,
2094 * This is called in transaction commmit time. If there are no orphan
2095 * files in the subvolume, it removes orphan item and frees block_rsv
2098 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2099 struct btrfs_root *root)
2103 if (!list_empty(&root->orphan_list) ||
2104 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2107 if (root->orphan_item_inserted &&
2108 btrfs_root_refs(&root->root_item) > 0) {
2109 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2110 root->root_key.objectid);
2112 root->orphan_item_inserted = 0;
2115 if (root->orphan_block_rsv) {
2116 WARN_ON(root->orphan_block_rsv->size > 0);
2117 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2118 root->orphan_block_rsv = NULL;
2123 * This creates an orphan entry for the given inode in case something goes
2124 * wrong in the middle of an unlink/truncate.
2126 * NOTE: caller of this function should reserve 5 units of metadata for
2129 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2131 struct btrfs_root *root = BTRFS_I(inode)->root;
2132 struct btrfs_block_rsv *block_rsv = NULL;
2137 if (!root->orphan_block_rsv) {
2138 block_rsv = btrfs_alloc_block_rsv(root);
2142 spin_lock(&root->orphan_lock);
2143 if (!root->orphan_block_rsv) {
2144 root->orphan_block_rsv = block_rsv;
2145 } else if (block_rsv) {
2146 btrfs_free_block_rsv(root, block_rsv);
2150 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2151 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2154 * For proper ENOSPC handling, we should do orphan
2155 * cleanup when mounting. But this introduces backward
2156 * compatibility issue.
2158 if (!xchg(&root->orphan_item_inserted, 1))
2165 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2168 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2169 BTRFS_I(inode)->orphan_meta_reserved = 1;
2172 spin_unlock(&root->orphan_lock);
2175 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2177 /* grab metadata reservation from transaction handle */
2179 ret = btrfs_orphan_reserve_metadata(trans, inode);
2183 /* insert an orphan item to track this unlinked/truncated file */
2185 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2189 /* insert an orphan item to track subvolume contains orphan files */
2191 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2192 root->root_key.objectid);
2199 * We have done the truncate/delete so we can go ahead and remove the orphan
2200 * item for this particular inode.
2202 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2204 struct btrfs_root *root = BTRFS_I(inode)->root;
2205 int delete_item = 0;
2206 int release_rsv = 0;
2209 spin_lock(&root->orphan_lock);
2210 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2211 list_del_init(&BTRFS_I(inode)->i_orphan);
2215 if (BTRFS_I(inode)->orphan_meta_reserved) {
2216 BTRFS_I(inode)->orphan_meta_reserved = 0;
2219 spin_unlock(&root->orphan_lock);
2221 if (trans && delete_item) {
2222 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2227 btrfs_orphan_release_metadata(inode);
2233 * this cleans up any orphans that may be left on the list from the last use
2236 void btrfs_orphan_cleanup(struct btrfs_root *root)
2238 struct btrfs_path *path;
2239 struct extent_buffer *leaf;
2240 struct btrfs_item *item;
2241 struct btrfs_key key, found_key;
2242 struct btrfs_trans_handle *trans;
2243 struct inode *inode;
2244 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2246 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2249 path = btrfs_alloc_path();
2253 key.objectid = BTRFS_ORPHAN_OBJECTID;
2254 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2255 key.offset = (u64)-1;
2258 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2260 printk(KERN_ERR "Error searching slot for orphan: %d"
2266 * if ret == 0 means we found what we were searching for, which
2267 * is weird, but possible, so only screw with path if we didnt
2268 * find the key and see if we have stuff that matches
2271 if (path->slots[0] == 0)
2276 /* pull out the item */
2277 leaf = path->nodes[0];
2278 item = btrfs_item_nr(leaf, path->slots[0]);
2279 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2281 /* make sure the item matches what we want */
2282 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2284 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2287 /* release the path since we're done with it */
2288 btrfs_release_path(root, path);
2291 * this is where we are basically btrfs_lookup, without the
2292 * crossing root thing. we store the inode number in the
2293 * offset of the orphan item.
2295 found_key.objectid = found_key.offset;
2296 found_key.type = BTRFS_INODE_ITEM_KEY;
2297 found_key.offset = 0;
2298 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2299 BUG_ON(IS_ERR(inode));
2302 * add this inode to the orphan list so btrfs_orphan_del does
2303 * the proper thing when we hit it
2305 spin_lock(&root->orphan_lock);
2306 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2307 spin_unlock(&root->orphan_lock);
2310 * if this is a bad inode, means we actually succeeded in
2311 * removing the inode, but not the orphan record, which means
2312 * we need to manually delete the orphan since iput will just
2313 * do a destroy_inode
2315 if (is_bad_inode(inode)) {
2316 trans = btrfs_start_transaction(root, 0);
2317 btrfs_orphan_del(trans, inode);
2318 btrfs_end_transaction(trans, root);
2323 /* if we have links, this was a truncate, lets do that */
2324 if (inode->i_nlink) {
2326 btrfs_truncate(inode);
2331 /* this will do delete_inode and everything for us */
2334 btrfs_free_path(path);
2336 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2338 if (root->orphan_block_rsv)
2339 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2342 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2343 trans = btrfs_join_transaction(root, 1);
2344 btrfs_end_transaction(trans, root);
2348 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2350 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2354 * very simple check to peek ahead in the leaf looking for xattrs. If we
2355 * don't find any xattrs, we know there can't be any acls.
2357 * slot is the slot the inode is in, objectid is the objectid of the inode
2359 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2360 int slot, u64 objectid)
2362 u32 nritems = btrfs_header_nritems(leaf);
2363 struct btrfs_key found_key;
2367 while (slot < nritems) {
2368 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2370 /* we found a different objectid, there must not be acls */
2371 if (found_key.objectid != objectid)
2374 /* we found an xattr, assume we've got an acl */
2375 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2379 * we found a key greater than an xattr key, there can't
2380 * be any acls later on
2382 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2389 * it goes inode, inode backrefs, xattrs, extents,
2390 * so if there are a ton of hard links to an inode there can
2391 * be a lot of backrefs. Don't waste time searching too hard,
2392 * this is just an optimization
2397 /* we hit the end of the leaf before we found an xattr or
2398 * something larger than an xattr. We have to assume the inode
2405 * read an inode from the btree into the in-memory inode
2407 static void btrfs_read_locked_inode(struct inode *inode)
2409 struct btrfs_path *path;
2410 struct extent_buffer *leaf;
2411 struct btrfs_inode_item *inode_item;
2412 struct btrfs_timespec *tspec;
2413 struct btrfs_root *root = BTRFS_I(inode)->root;
2414 struct btrfs_key location;
2416 u64 alloc_group_block;
2420 path = btrfs_alloc_path();
2422 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2424 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2428 leaf = path->nodes[0];
2429 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2430 struct btrfs_inode_item);
2432 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2433 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2434 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2435 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2436 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2438 tspec = btrfs_inode_atime(inode_item);
2439 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2440 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2442 tspec = btrfs_inode_mtime(inode_item);
2443 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2444 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2446 tspec = btrfs_inode_ctime(inode_item);
2447 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2448 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2450 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2451 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2452 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2453 inode->i_generation = BTRFS_I(inode)->generation;
2455 rdev = btrfs_inode_rdev(leaf, inode_item);
2457 BTRFS_I(inode)->index_cnt = (u64)-1;
2458 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2460 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2463 * try to precache a NULL acl entry for files that don't have
2464 * any xattrs or acls
2466 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2468 cache_no_acl(inode);
2470 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2471 alloc_group_block, 0);
2472 btrfs_free_path(path);
2475 switch (inode->i_mode & S_IFMT) {
2477 inode->i_mapping->a_ops = &btrfs_aops;
2478 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2479 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2480 inode->i_fop = &btrfs_file_operations;
2481 inode->i_op = &btrfs_file_inode_operations;
2484 inode->i_fop = &btrfs_dir_file_operations;
2485 if (root == root->fs_info->tree_root)
2486 inode->i_op = &btrfs_dir_ro_inode_operations;
2488 inode->i_op = &btrfs_dir_inode_operations;
2491 inode->i_op = &btrfs_symlink_inode_operations;
2492 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2493 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2496 inode->i_op = &btrfs_special_inode_operations;
2497 init_special_inode(inode, inode->i_mode, rdev);
2501 btrfs_update_iflags(inode);
2505 btrfs_free_path(path);
2506 make_bad_inode(inode);
2510 * given a leaf and an inode, copy the inode fields into the leaf
2512 static void fill_inode_item(struct btrfs_trans_handle *trans,
2513 struct extent_buffer *leaf,
2514 struct btrfs_inode_item *item,
2515 struct inode *inode)
2517 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2518 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2519 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2520 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2521 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2523 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2524 inode->i_atime.tv_sec);
2525 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2526 inode->i_atime.tv_nsec);
2528 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2529 inode->i_mtime.tv_sec);
2530 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2531 inode->i_mtime.tv_nsec);
2533 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2534 inode->i_ctime.tv_sec);
2535 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2536 inode->i_ctime.tv_nsec);
2538 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2539 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2540 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2541 btrfs_set_inode_transid(leaf, item, trans->transid);
2542 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2543 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2544 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2548 * copy everything in the in-memory inode into the btree.
2550 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2551 struct btrfs_root *root, struct inode *inode)
2553 struct btrfs_inode_item *inode_item;
2554 struct btrfs_path *path;
2555 struct extent_buffer *leaf;
2558 path = btrfs_alloc_path();
2560 path->leave_spinning = 1;
2561 ret = btrfs_lookup_inode(trans, root, path,
2562 &BTRFS_I(inode)->location, 1);
2569 btrfs_unlock_up_safe(path, 1);
2570 leaf = path->nodes[0];
2571 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2572 struct btrfs_inode_item);
2574 fill_inode_item(trans, leaf, inode_item, inode);
2575 btrfs_mark_buffer_dirty(leaf);
2576 btrfs_set_inode_last_trans(trans, inode);
2579 btrfs_free_path(path);
2585 * unlink helper that gets used here in inode.c and in the tree logging
2586 * recovery code. It remove a link in a directory with a given name, and
2587 * also drops the back refs in the inode to the directory
2589 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2590 struct btrfs_root *root,
2591 struct inode *dir, struct inode *inode,
2592 const char *name, int name_len)
2594 struct btrfs_path *path;
2596 struct extent_buffer *leaf;
2597 struct btrfs_dir_item *di;
2598 struct btrfs_key key;
2601 path = btrfs_alloc_path();
2607 path->leave_spinning = 1;
2608 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2609 name, name_len, -1);
2618 leaf = path->nodes[0];
2619 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2620 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2623 btrfs_release_path(root, path);
2625 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2627 dir->i_ino, &index);
2629 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2630 "inode %lu parent %lu\n", name_len, name,
2631 inode->i_ino, dir->i_ino);
2635 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2636 index, name, name_len, -1);
2645 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2646 btrfs_release_path(root, path);
2648 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2650 BUG_ON(ret != 0 && ret != -ENOENT);
2652 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2656 btrfs_free_path(path);
2660 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2661 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2662 btrfs_update_inode(trans, root, dir);
2663 btrfs_drop_nlink(inode);
2664 ret = btrfs_update_inode(trans, root, inode);
2669 /* helper to check if there is any shared block in the path */
2670 static int check_path_shared(struct btrfs_root *root,
2671 struct btrfs_path *path)
2673 struct extent_buffer *eb;
2678 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2679 if (!path->nodes[level])
2681 eb = path->nodes[level];
2682 if (!btrfs_block_can_be_shared(root, eb))
2684 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2693 * helper to start transaction for unlink and rmdir.
2695 * unlink and rmdir are special in btrfs, they do not always free space.
2696 * so in enospc case, we should make sure they will free space before
2697 * allowing them to use the global metadata reservation.
2699 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2700 struct dentry *dentry)
2702 struct btrfs_trans_handle *trans;
2703 struct btrfs_root *root = BTRFS_I(dir)->root;
2704 struct btrfs_path *path;
2705 struct btrfs_inode_ref *ref;
2706 struct btrfs_dir_item *di;
2707 struct inode *inode = dentry->d_inode;
2713 trans = btrfs_start_transaction(root, 10);
2714 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2717 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2718 return ERR_PTR(-ENOSPC);
2720 /* check if there is someone else holds reference */
2721 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2722 return ERR_PTR(-ENOSPC);
2724 if (atomic_read(&inode->i_count) > 2)
2725 return ERR_PTR(-ENOSPC);
2727 if (xchg(&root->fs_info->enospc_unlink, 1))
2728 return ERR_PTR(-ENOSPC);
2730 path = btrfs_alloc_path();
2732 root->fs_info->enospc_unlink = 0;
2733 return ERR_PTR(-ENOMEM);
2736 trans = btrfs_start_transaction(root, 0);
2737 if (IS_ERR(trans)) {
2738 btrfs_free_path(path);
2739 root->fs_info->enospc_unlink = 0;
2743 path->skip_locking = 1;
2744 path->search_commit_root = 1;
2746 ret = btrfs_lookup_inode(trans, root, path,
2747 &BTRFS_I(dir)->location, 0);
2753 if (check_path_shared(root, path))
2758 btrfs_release_path(root, path);
2760 ret = btrfs_lookup_inode(trans, root, path,
2761 &BTRFS_I(inode)->location, 0);
2767 if (check_path_shared(root, path))
2772 btrfs_release_path(root, path);
2774 if (ret == 0 && S_ISREG(inode->i_mode)) {
2775 ret = btrfs_lookup_file_extent(trans, root, path,
2776 inode->i_ino, (u64)-1, 0);
2782 if (check_path_shared(root, path))
2784 btrfs_release_path(root, path);
2792 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2793 dentry->d_name.name, dentry->d_name.len, 0);
2799 if (check_path_shared(root, path))
2805 btrfs_release_path(root, path);
2807 ref = btrfs_lookup_inode_ref(trans, root, path,
2808 dentry->d_name.name, dentry->d_name.len,
2809 inode->i_ino, dir->i_ino, 0);
2815 if (check_path_shared(root, path))
2817 index = btrfs_inode_ref_index(path->nodes[0], ref);
2818 btrfs_release_path(root, path);
2820 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2821 dentry->d_name.name, dentry->d_name.len, 0);
2826 BUG_ON(ret == -ENOENT);
2827 if (check_path_shared(root, path))
2832 btrfs_free_path(path);
2834 btrfs_end_transaction(trans, root);
2835 root->fs_info->enospc_unlink = 0;
2836 return ERR_PTR(err);
2839 trans->block_rsv = &root->fs_info->global_block_rsv;
2843 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *root)
2846 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2847 BUG_ON(!root->fs_info->enospc_unlink);
2848 root->fs_info->enospc_unlink = 0;
2850 btrfs_end_transaction_throttle(trans, root);
2853 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2855 struct btrfs_root *root = BTRFS_I(dir)->root;
2856 struct btrfs_trans_handle *trans;
2857 struct inode *inode = dentry->d_inode;
2859 unsigned long nr = 0;
2861 trans = __unlink_start_trans(dir, dentry);
2863 return PTR_ERR(trans);
2865 btrfs_set_trans_block_group(trans, dir);
2867 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2869 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2870 dentry->d_name.name, dentry->d_name.len);
2873 if (inode->i_nlink == 0) {
2874 ret = btrfs_orphan_add(trans, inode);
2878 nr = trans->blocks_used;
2879 __unlink_end_trans(trans, root);
2880 btrfs_btree_balance_dirty(root, nr);
2884 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2885 struct btrfs_root *root,
2886 struct inode *dir, u64 objectid,
2887 const char *name, int name_len)
2889 struct btrfs_path *path;
2890 struct extent_buffer *leaf;
2891 struct btrfs_dir_item *di;
2892 struct btrfs_key key;
2896 path = btrfs_alloc_path();
2900 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2901 name, name_len, -1);
2902 BUG_ON(!di || IS_ERR(di));
2904 leaf = path->nodes[0];
2905 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2906 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2907 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2909 btrfs_release_path(root, path);
2911 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2912 objectid, root->root_key.objectid,
2913 dir->i_ino, &index, name, name_len);
2915 BUG_ON(ret != -ENOENT);
2916 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2918 BUG_ON(!di || IS_ERR(di));
2920 leaf = path->nodes[0];
2921 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2922 btrfs_release_path(root, path);
2926 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2927 index, name, name_len, -1);
2928 BUG_ON(!di || IS_ERR(di));
2930 leaf = path->nodes[0];
2931 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2932 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2933 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2935 btrfs_release_path(root, path);
2937 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2938 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2939 ret = btrfs_update_inode(trans, root, dir);
2941 dir->i_sb->s_dirt = 1;
2943 btrfs_free_path(path);
2947 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2949 struct inode *inode = dentry->d_inode;
2951 struct btrfs_root *root = BTRFS_I(dir)->root;
2952 struct btrfs_trans_handle *trans;
2953 unsigned long nr = 0;
2955 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2956 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2959 trans = __unlink_start_trans(dir, dentry);
2961 return PTR_ERR(trans);
2963 btrfs_set_trans_block_group(trans, dir);
2965 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2966 err = btrfs_unlink_subvol(trans, root, dir,
2967 BTRFS_I(inode)->location.objectid,
2968 dentry->d_name.name,
2969 dentry->d_name.len);
2973 err = btrfs_orphan_add(trans, inode);
2977 /* now the directory is empty */
2978 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2979 dentry->d_name.name, dentry->d_name.len);
2981 btrfs_i_size_write(inode, 0);
2983 nr = trans->blocks_used;
2984 __unlink_end_trans(trans, root);
2985 btrfs_btree_balance_dirty(root, nr);
2992 * when truncating bytes in a file, it is possible to avoid reading
2993 * the leaves that contain only checksum items. This can be the
2994 * majority of the IO required to delete a large file, but it must
2995 * be done carefully.
2997 * The keys in the level just above the leaves are checked to make sure
2998 * the lowest key in a given leaf is a csum key, and starts at an offset
2999 * after the new size.
3001 * Then the key for the next leaf is checked to make sure it also has
3002 * a checksum item for the same file. If it does, we know our target leaf
3003 * contains only checksum items, and it can be safely freed without reading
3006 * This is just an optimization targeted at large files. It may do
3007 * nothing. It will return 0 unless things went badly.
3009 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3010 struct btrfs_root *root,
3011 struct btrfs_path *path,
3012 struct inode *inode, u64 new_size)
3014 struct btrfs_key key;
3017 struct btrfs_key found_key;
3018 struct btrfs_key other_key;
3019 struct btrfs_leaf_ref *ref;
3023 path->lowest_level = 1;
3024 key.objectid = inode->i_ino;
3025 key.type = BTRFS_CSUM_ITEM_KEY;
3026 key.offset = new_size;
3028 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3032 if (path->nodes[1] == NULL) {
3037 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3038 nritems = btrfs_header_nritems(path->nodes[1]);
3043 if (path->slots[1] >= nritems)
3046 /* did we find a key greater than anything we want to delete? */
3047 if (found_key.objectid > inode->i_ino ||
3048 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3051 /* we check the next key in the node to make sure the leave contains
3052 * only checksum items. This comparison doesn't work if our
3053 * leaf is the last one in the node
3055 if (path->slots[1] + 1 >= nritems) {
3057 /* search forward from the last key in the node, this
3058 * will bring us into the next node in the tree
3060 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3062 /* unlikely, but we inc below, so check to be safe */
3063 if (found_key.offset == (u64)-1)
3066 /* search_forward needs a path with locks held, do the
3067 * search again for the original key. It is possible
3068 * this will race with a balance and return a path that
3069 * we could modify, but this drop is just an optimization
3070 * and is allowed to miss some leaves.
3072 btrfs_release_path(root, path);
3075 /* setup a max key for search_forward */
3076 other_key.offset = (u64)-1;
3077 other_key.type = key.type;
3078 other_key.objectid = key.objectid;
3080 path->keep_locks = 1;
3081 ret = btrfs_search_forward(root, &found_key, &other_key,
3083 path->keep_locks = 0;
3084 if (ret || found_key.objectid != key.objectid ||
3085 found_key.type != key.type) {
3090 key.offset = found_key.offset;
3091 btrfs_release_path(root, path);
3096 /* we know there's one more slot after us in the tree,
3097 * read that key so we can verify it is also a checksum item
3099 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3101 if (found_key.objectid < inode->i_ino)
3104 if (found_key.type != key.type || found_key.offset < new_size)
3108 * if the key for the next leaf isn't a csum key from this objectid,
3109 * we can't be sure there aren't good items inside this leaf.
3112 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3115 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3116 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3118 * it is safe to delete this leaf, it contains only
3119 * csum items from this inode at an offset >= new_size
3121 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3124 if (root->ref_cows && leaf_gen < trans->transid) {
3125 ref = btrfs_alloc_leaf_ref(root, 0);
3127 ref->root_gen = root->root_key.offset;
3128 ref->bytenr = leaf_start;
3130 ref->generation = leaf_gen;
3133 btrfs_sort_leaf_ref(ref);
3135 ret = btrfs_add_leaf_ref(root, ref, 0);
3137 btrfs_free_leaf_ref(root, ref);
3143 btrfs_release_path(root, path);
3145 if (other_key.objectid == inode->i_ino &&
3146 other_key.type == key.type && other_key.offset > key.offset) {
3147 key.offset = other_key.offset;
3153 /* fixup any changes we've made to the path */
3154 path->lowest_level = 0;
3155 path->keep_locks = 0;
3156 btrfs_release_path(root, path);
3163 * this can truncate away extent items, csum items and directory items.
3164 * It starts at a high offset and removes keys until it can't find
3165 * any higher than new_size
3167 * csum items that cross the new i_size are truncated to the new size
3170 * min_type is the minimum key type to truncate down to. If set to 0, this
3171 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3173 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3174 struct btrfs_root *root,
3175 struct inode *inode,
3176 u64 new_size, u32 min_type)
3178 struct btrfs_path *path;
3179 struct extent_buffer *leaf;
3180 struct btrfs_file_extent_item *fi;
3181 struct btrfs_key key;
3182 struct btrfs_key found_key;
3183 u64 extent_start = 0;
3184 u64 extent_num_bytes = 0;
3185 u64 extent_offset = 0;
3187 u64 mask = root->sectorsize - 1;
3188 u32 found_type = (u8)-1;
3191 int pending_del_nr = 0;
3192 int pending_del_slot = 0;
3193 int extent_type = -1;
3198 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3201 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3203 path = btrfs_alloc_path();
3207 key.objectid = inode->i_ino;
3208 key.offset = (u64)-1;
3212 path->leave_spinning = 1;
3213 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3220 /* there are no items in the tree for us to truncate, we're
3223 if (path->slots[0] == 0)
3230 leaf = path->nodes[0];
3231 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3232 found_type = btrfs_key_type(&found_key);
3235 if (found_key.objectid != inode->i_ino)
3238 if (found_type < min_type)
3241 item_end = found_key.offset;
3242 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3243 fi = btrfs_item_ptr(leaf, path->slots[0],
3244 struct btrfs_file_extent_item);
3245 extent_type = btrfs_file_extent_type(leaf, fi);
3246 encoding = btrfs_file_extent_compression(leaf, fi);
3247 encoding |= btrfs_file_extent_encryption(leaf, fi);
3248 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3250 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3252 btrfs_file_extent_num_bytes(leaf, fi);
3253 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3254 item_end += btrfs_file_extent_inline_len(leaf,
3259 if (found_type > min_type) {
3262 if (item_end < new_size)
3264 if (found_key.offset >= new_size)
3270 /* FIXME, shrink the extent if the ref count is only 1 */
3271 if (found_type != BTRFS_EXTENT_DATA_KEY)
3274 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3276 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3277 if (!del_item && !encoding) {
3278 u64 orig_num_bytes =
3279 btrfs_file_extent_num_bytes(leaf, fi);
3280 extent_num_bytes = new_size -
3281 found_key.offset + root->sectorsize - 1;
3282 extent_num_bytes = extent_num_bytes &
3283 ~((u64)root->sectorsize - 1);
3284 btrfs_set_file_extent_num_bytes(leaf, fi,
3286 num_dec = (orig_num_bytes -
3288 if (root->ref_cows && extent_start != 0)
3289 inode_sub_bytes(inode, num_dec);
3290 btrfs_mark_buffer_dirty(leaf);
3293 btrfs_file_extent_disk_num_bytes(leaf,
3295 extent_offset = found_key.offset -
3296 btrfs_file_extent_offset(leaf, fi);
3298 /* FIXME blocksize != 4096 */
3299 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3300 if (extent_start != 0) {
3303 inode_sub_bytes(inode, num_dec);
3306 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3308 * we can't truncate inline items that have had
3312 btrfs_file_extent_compression(leaf, fi) == 0 &&
3313 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3314 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3315 u32 size = new_size - found_key.offset;
3317 if (root->ref_cows) {
3318 inode_sub_bytes(inode, item_end + 1 -
3322 btrfs_file_extent_calc_inline_size(size);
3323 ret = btrfs_truncate_item(trans, root, path,
3326 } else if (root->ref_cows) {
3327 inode_sub_bytes(inode, item_end + 1 -
3333 if (!pending_del_nr) {
3334 /* no pending yet, add ourselves */
3335 pending_del_slot = path->slots[0];
3337 } else if (pending_del_nr &&
3338 path->slots[0] + 1 == pending_del_slot) {
3339 /* hop on the pending chunk */
3341 pending_del_slot = path->slots[0];
3348 if (found_extent && root->ref_cows) {
3349 btrfs_set_path_blocking(path);
3350 ret = btrfs_free_extent(trans, root, extent_start,
3351 extent_num_bytes, 0,
3352 btrfs_header_owner(leaf),
3353 inode->i_ino, extent_offset);
3357 if (found_type == BTRFS_INODE_ITEM_KEY)
3360 if (path->slots[0] == 0 ||
3361 path->slots[0] != pending_del_slot) {
3362 if (root->ref_cows) {
3366 if (pending_del_nr) {
3367 ret = btrfs_del_items(trans, root, path,
3373 btrfs_release_path(root, path);
3380 if (pending_del_nr) {
3381 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3385 btrfs_free_path(path);
3390 * taken from block_truncate_page, but does cow as it zeros out
3391 * any bytes left in the last page in the file.
3393 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3395 struct inode *inode = mapping->host;
3396 struct btrfs_root *root = BTRFS_I(inode)->root;
3397 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3398 struct btrfs_ordered_extent *ordered;
3399 struct extent_state *cached_state = NULL;
3401 u32 blocksize = root->sectorsize;
3402 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3403 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3409 if ((offset & (blocksize - 1)) == 0)
3411 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3417 page = grab_cache_page(mapping, index);
3419 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3423 page_start = page_offset(page);
3424 page_end = page_start + PAGE_CACHE_SIZE - 1;
3426 if (!PageUptodate(page)) {
3427 ret = btrfs_readpage(NULL, page);
3429 if (page->mapping != mapping) {
3431 page_cache_release(page);
3434 if (!PageUptodate(page)) {
3439 wait_on_page_writeback(page);
3441 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3443 set_page_extent_mapped(page);
3445 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3447 unlock_extent_cached(io_tree, page_start, page_end,
3448 &cached_state, GFP_NOFS);
3450 page_cache_release(page);
3451 btrfs_start_ordered_extent(inode, ordered, 1);
3452 btrfs_put_ordered_extent(ordered);
3456 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3457 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3458 0, 0, &cached_state, GFP_NOFS);
3460 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3463 unlock_extent_cached(io_tree, page_start, page_end,
3464 &cached_state, GFP_NOFS);
3469 if (offset != PAGE_CACHE_SIZE) {
3471 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3472 flush_dcache_page(page);
3475 ClearPageChecked(page);
3476 set_page_dirty(page);
3477 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3482 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3484 page_cache_release(page);
3489 int btrfs_cont_expand(struct inode *inode, loff_t size)
3491 struct btrfs_trans_handle *trans;
3492 struct btrfs_root *root = BTRFS_I(inode)->root;
3493 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3494 struct extent_map *em = NULL;
3495 struct extent_state *cached_state = NULL;
3496 u64 mask = root->sectorsize - 1;
3497 u64 hole_start = (inode->i_size + mask) & ~mask;
3498 u64 block_end = (size + mask) & ~mask;
3504 if (size <= hole_start)
3508 struct btrfs_ordered_extent *ordered;
3509 btrfs_wait_ordered_range(inode, hole_start,
3510 block_end - hole_start);
3511 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3512 &cached_state, GFP_NOFS);
3513 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3516 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3517 &cached_state, GFP_NOFS);
3518 btrfs_put_ordered_extent(ordered);
3521 cur_offset = hole_start;
3523 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3524 block_end - cur_offset, 0);
3525 BUG_ON(IS_ERR(em) || !em);
3526 last_byte = min(extent_map_end(em), block_end);
3527 last_byte = (last_byte + mask) & ~mask;
3528 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3530 hole_size = last_byte - cur_offset;
3532 trans = btrfs_start_transaction(root, 2);
3533 if (IS_ERR(trans)) {
3534 err = PTR_ERR(trans);
3537 btrfs_set_trans_block_group(trans, inode);
3539 err = btrfs_drop_extents(trans, inode, cur_offset,
3540 cur_offset + hole_size,
3544 err = btrfs_insert_file_extent(trans, root,
3545 inode->i_ino, cur_offset, 0,
3546 0, hole_size, 0, hole_size,
3550 btrfs_drop_extent_cache(inode, hole_start,
3553 btrfs_end_transaction(trans, root);
3555 free_extent_map(em);
3557 cur_offset = last_byte;
3558 if (cur_offset >= block_end)
3562 free_extent_map(em);
3563 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3568 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3570 struct btrfs_root *root = BTRFS_I(inode)->root;
3571 struct btrfs_trans_handle *trans;
3575 if (attr->ia_size == inode->i_size)
3578 if (attr->ia_size > inode->i_size) {
3579 unsigned long limit;
3580 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3581 if (attr->ia_size > inode->i_sb->s_maxbytes)
3583 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3584 send_sig(SIGXFSZ, current, 0);
3589 trans = btrfs_start_transaction(root, 5);
3591 return PTR_ERR(trans);
3593 btrfs_set_trans_block_group(trans, inode);
3595 ret = btrfs_orphan_add(trans, inode);
3598 nr = trans->blocks_used;
3599 btrfs_end_transaction(trans, root);
3600 btrfs_btree_balance_dirty(root, nr);
3602 if (attr->ia_size > inode->i_size) {
3603 ret = btrfs_cont_expand(inode, attr->ia_size);
3605 btrfs_truncate(inode);
3609 i_size_write(inode, attr->ia_size);
3610 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3612 trans = btrfs_start_transaction(root, 0);
3613 BUG_ON(IS_ERR(trans));
3614 btrfs_set_trans_block_group(trans, inode);
3615 trans->block_rsv = root->orphan_block_rsv;
3616 BUG_ON(!trans->block_rsv);
3618 ret = btrfs_update_inode(trans, root, inode);
3620 if (inode->i_nlink > 0) {
3621 ret = btrfs_orphan_del(trans, inode);
3624 nr = trans->blocks_used;
3625 btrfs_end_transaction(trans, root);
3626 btrfs_btree_balance_dirty(root, nr);
3631 * We're truncating a file that used to have good data down to
3632 * zero. Make sure it gets into the ordered flush list so that
3633 * any new writes get down to disk quickly.
3635 if (attr->ia_size == 0)
3636 BTRFS_I(inode)->ordered_data_close = 1;
3638 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3639 ret = vmtruncate(inode, attr->ia_size);
3645 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3647 struct inode *inode = dentry->d_inode;
3650 err = inode_change_ok(inode, attr);
3654 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3655 err = btrfs_setattr_size(inode, attr);
3659 attr->ia_valid &= ~ATTR_SIZE;
3662 err = inode_setattr(inode, attr);
3664 if (!err && ((attr->ia_valid & ATTR_MODE)))
3665 err = btrfs_acl_chmod(inode);
3669 void btrfs_delete_inode(struct inode *inode)
3671 struct btrfs_trans_handle *trans;
3672 struct btrfs_root *root = BTRFS_I(inode)->root;
3676 truncate_inode_pages(&inode->i_data, 0);
3677 if (is_bad_inode(inode)) {
3678 btrfs_orphan_del(NULL, inode);
3681 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3683 if (root->fs_info->log_root_recovering) {
3684 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3688 if (inode->i_nlink > 0) {
3689 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3693 btrfs_i_size_write(inode, 0);
3696 trans = btrfs_start_transaction(root, 0);
3697 BUG_ON(IS_ERR(trans));
3698 btrfs_set_trans_block_group(trans, inode);
3699 trans->block_rsv = root->orphan_block_rsv;
3701 ret = btrfs_block_rsv_check(trans, root,
3702 root->orphan_block_rsv, 0, 5);
3704 BUG_ON(ret != -EAGAIN);
3705 ret = btrfs_commit_transaction(trans, root);
3710 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3714 nr = trans->blocks_used;
3715 btrfs_end_transaction(trans, root);
3717 btrfs_btree_balance_dirty(root, nr);
3722 ret = btrfs_orphan_del(trans, inode);
3726 nr = trans->blocks_used;
3727 btrfs_end_transaction(trans, root);
3728 btrfs_btree_balance_dirty(root, nr);
3735 * this returns the key found in the dir entry in the location pointer.
3736 * If no dir entries were found, location->objectid is 0.
3738 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3739 struct btrfs_key *location)
3741 const char *name = dentry->d_name.name;
3742 int namelen = dentry->d_name.len;
3743 struct btrfs_dir_item *di;
3744 struct btrfs_path *path;
3745 struct btrfs_root *root = BTRFS_I(dir)->root;
3748 path = btrfs_alloc_path();
3751 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3756 if (!di || IS_ERR(di))
3759 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3761 btrfs_free_path(path);
3764 location->objectid = 0;
3769 * when we hit a tree root in a directory, the btrfs part of the inode
3770 * needs to be changed to reflect the root directory of the tree root. This
3771 * is kind of like crossing a mount point.
3773 static int fixup_tree_root_location(struct btrfs_root *root,
3775 struct dentry *dentry,
3776 struct btrfs_key *location,
3777 struct btrfs_root **sub_root)
3779 struct btrfs_path *path;
3780 struct btrfs_root *new_root;
3781 struct btrfs_root_ref *ref;
3782 struct extent_buffer *leaf;
3786 path = btrfs_alloc_path();
3793 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3794 BTRFS_I(dir)->root->root_key.objectid,
3795 location->objectid);
3802 leaf = path->nodes[0];
3803 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3804 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3805 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3808 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3809 (unsigned long)(ref + 1),
3810 dentry->d_name.len);
3814 btrfs_release_path(root->fs_info->tree_root, path);
3816 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3817 if (IS_ERR(new_root)) {
3818 err = PTR_ERR(new_root);
3822 if (btrfs_root_refs(&new_root->root_item) == 0) {
3827 *sub_root = new_root;
3828 location->objectid = btrfs_root_dirid(&new_root->root_item);
3829 location->type = BTRFS_INODE_ITEM_KEY;
3830 location->offset = 0;
3833 btrfs_free_path(path);
3837 static void inode_tree_add(struct inode *inode)
3839 struct btrfs_root *root = BTRFS_I(inode)->root;
3840 struct btrfs_inode *entry;
3842 struct rb_node *parent;
3844 p = &root->inode_tree.rb_node;
3847 if (hlist_unhashed(&inode->i_hash))
3850 spin_lock(&root->inode_lock);
3853 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3855 if (inode->i_ino < entry->vfs_inode.i_ino)
3856 p = &parent->rb_left;
3857 else if (inode->i_ino > entry->vfs_inode.i_ino)
3858 p = &parent->rb_right;
3860 WARN_ON(!(entry->vfs_inode.i_state &
3861 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3862 rb_erase(parent, &root->inode_tree);
3863 RB_CLEAR_NODE(parent);
3864 spin_unlock(&root->inode_lock);
3868 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3869 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3870 spin_unlock(&root->inode_lock);
3873 static void inode_tree_del(struct inode *inode)
3875 struct btrfs_root *root = BTRFS_I(inode)->root;
3878 spin_lock(&root->inode_lock);
3879 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3880 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3881 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3882 empty = RB_EMPTY_ROOT(&root->inode_tree);
3884 spin_unlock(&root->inode_lock);
3886 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3887 synchronize_srcu(&root->fs_info->subvol_srcu);
3888 spin_lock(&root->inode_lock);
3889 empty = RB_EMPTY_ROOT(&root->inode_tree);
3890 spin_unlock(&root->inode_lock);
3892 btrfs_add_dead_root(root);
3896 int btrfs_invalidate_inodes(struct btrfs_root *root)
3898 struct rb_node *node;
3899 struct rb_node *prev;
3900 struct btrfs_inode *entry;
3901 struct inode *inode;
3904 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3906 spin_lock(&root->inode_lock);
3908 node = root->inode_tree.rb_node;
3912 entry = rb_entry(node, struct btrfs_inode, rb_node);
3914 if (objectid < entry->vfs_inode.i_ino)
3915 node = node->rb_left;
3916 else if (objectid > entry->vfs_inode.i_ino)
3917 node = node->rb_right;
3923 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3924 if (objectid <= entry->vfs_inode.i_ino) {
3928 prev = rb_next(prev);
3932 entry = rb_entry(node, struct btrfs_inode, rb_node);
3933 objectid = entry->vfs_inode.i_ino + 1;
3934 inode = igrab(&entry->vfs_inode);
3936 spin_unlock(&root->inode_lock);
3937 if (atomic_read(&inode->i_count) > 1)
3938 d_prune_aliases(inode);
3940 * btrfs_drop_inode will remove it from
3941 * the inode cache when its usage count
3946 spin_lock(&root->inode_lock);
3950 if (cond_resched_lock(&root->inode_lock))
3953 node = rb_next(node);
3955 spin_unlock(&root->inode_lock);
3959 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3961 struct btrfs_iget_args *args = p;
3962 inode->i_ino = args->ino;
3963 BTRFS_I(inode)->root = args->root;
3964 btrfs_set_inode_space_info(args->root, inode);
3968 static int btrfs_find_actor(struct inode *inode, void *opaque)
3970 struct btrfs_iget_args *args = opaque;
3971 return args->ino == inode->i_ino &&
3972 args->root == BTRFS_I(inode)->root;
3975 static struct inode *btrfs_iget_locked(struct super_block *s,
3977 struct btrfs_root *root)
3979 struct inode *inode;
3980 struct btrfs_iget_args args;
3981 args.ino = objectid;
3984 inode = iget5_locked(s, objectid, btrfs_find_actor,
3985 btrfs_init_locked_inode,
3990 /* Get an inode object given its location and corresponding root.
3991 * Returns in *is_new if the inode was read from disk
3993 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3994 struct btrfs_root *root, int *new)
3996 struct inode *inode;
3998 inode = btrfs_iget_locked(s, location->objectid, root);
4000 return ERR_PTR(-ENOMEM);
4002 if (inode->i_state & I_NEW) {
4003 BTRFS_I(inode)->root = root;
4004 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4005 btrfs_read_locked_inode(inode);
4007 inode_tree_add(inode);
4008 unlock_new_inode(inode);
4016 static struct inode *new_simple_dir(struct super_block *s,
4017 struct btrfs_key *key,
4018 struct btrfs_root *root)
4020 struct inode *inode = new_inode(s);
4023 return ERR_PTR(-ENOMEM);
4025 BTRFS_I(inode)->root = root;
4026 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4027 BTRFS_I(inode)->dummy_inode = 1;
4029 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4030 inode->i_op = &simple_dir_inode_operations;
4031 inode->i_fop = &simple_dir_operations;
4032 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4033 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4038 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4040 struct inode *inode;
4041 struct btrfs_root *root = BTRFS_I(dir)->root;
4042 struct btrfs_root *sub_root = root;
4043 struct btrfs_key location;
4047 dentry->d_op = &btrfs_dentry_operations;
4049 if (dentry->d_name.len > BTRFS_NAME_LEN)
4050 return ERR_PTR(-ENAMETOOLONG);
4052 ret = btrfs_inode_by_name(dir, dentry, &location);
4055 return ERR_PTR(ret);
4057 if (location.objectid == 0)
4060 if (location.type == BTRFS_INODE_ITEM_KEY) {
4061 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4065 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4067 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4068 ret = fixup_tree_root_location(root, dir, dentry,
4069 &location, &sub_root);
4072 inode = ERR_PTR(ret);
4074 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4076 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4078 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4080 if (root != sub_root) {
4081 down_read(&root->fs_info->cleanup_work_sem);
4082 if (!(inode->i_sb->s_flags & MS_RDONLY))
4083 btrfs_orphan_cleanup(sub_root);
4084 up_read(&root->fs_info->cleanup_work_sem);
4090 static int btrfs_dentry_delete(struct dentry *dentry)
4092 struct btrfs_root *root;
4094 if (!dentry->d_inode && !IS_ROOT(dentry))
4095 dentry = dentry->d_parent;
4097 if (dentry->d_inode) {
4098 root = BTRFS_I(dentry->d_inode)->root;
4099 if (btrfs_root_refs(&root->root_item) == 0)
4105 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4106 struct nameidata *nd)
4108 struct inode *inode;
4110 inode = btrfs_lookup_dentry(dir, dentry);
4112 return ERR_CAST(inode);
4114 return d_splice_alias(inode, dentry);
4117 static unsigned char btrfs_filetype_table[] = {
4118 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4121 static int btrfs_real_readdir(struct file *filp, void *dirent,
4124 struct inode *inode = filp->f_dentry->d_inode;
4125 struct btrfs_root *root = BTRFS_I(inode)->root;
4126 struct btrfs_item *item;
4127 struct btrfs_dir_item *di;
4128 struct btrfs_key key;
4129 struct btrfs_key found_key;
4130 struct btrfs_path *path;
4133 struct extent_buffer *leaf;
4136 unsigned char d_type;
4141 int key_type = BTRFS_DIR_INDEX_KEY;
4146 /* FIXME, use a real flag for deciding about the key type */
4147 if (root->fs_info->tree_root == root)
4148 key_type = BTRFS_DIR_ITEM_KEY;
4150 /* special case for "." */
4151 if (filp->f_pos == 0) {
4152 over = filldir(dirent, ".", 1,
4159 /* special case for .., just use the back ref */
4160 if (filp->f_pos == 1) {
4161 u64 pino = parent_ino(filp->f_path.dentry);
4162 over = filldir(dirent, "..", 2,
4168 path = btrfs_alloc_path();
4171 btrfs_set_key_type(&key, key_type);
4172 key.offset = filp->f_pos;
4173 key.objectid = inode->i_ino;
4175 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4181 leaf = path->nodes[0];
4182 nritems = btrfs_header_nritems(leaf);
4183 slot = path->slots[0];
4184 if (advance || slot >= nritems) {
4185 if (slot >= nritems - 1) {
4186 ret = btrfs_next_leaf(root, path);
4189 leaf = path->nodes[0];
4190 nritems = btrfs_header_nritems(leaf);
4191 slot = path->slots[0];
4199 item = btrfs_item_nr(leaf, slot);
4200 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4202 if (found_key.objectid != key.objectid)
4204 if (btrfs_key_type(&found_key) != key_type)
4206 if (found_key.offset < filp->f_pos)
4209 filp->f_pos = found_key.offset;
4211 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4213 di_total = btrfs_item_size(leaf, item);
4215 while (di_cur < di_total) {
4216 struct btrfs_key location;
4218 name_len = btrfs_dir_name_len(leaf, di);
4219 if (name_len <= sizeof(tmp_name)) {
4220 name_ptr = tmp_name;
4222 name_ptr = kmalloc(name_len, GFP_NOFS);
4228 read_extent_buffer(leaf, name_ptr,
4229 (unsigned long)(di + 1), name_len);
4231 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4232 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4234 /* is this a reference to our own snapshot? If so
4237 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4238 location.objectid == root->root_key.objectid) {
4242 over = filldir(dirent, name_ptr, name_len,
4243 found_key.offset, location.objectid,
4247 if (name_ptr != tmp_name)
4252 di_len = btrfs_dir_name_len(leaf, di) +
4253 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4255 di = (struct btrfs_dir_item *)((char *)di + di_len);
4259 /* Reached end of directory/root. Bump pos past the last item. */
4260 if (key_type == BTRFS_DIR_INDEX_KEY)
4262 * 32-bit glibc will use getdents64, but then strtol -
4263 * so the last number we can serve is this.
4265 filp->f_pos = 0x7fffffff;
4271 btrfs_free_path(path);
4275 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4277 struct btrfs_root *root = BTRFS_I(inode)->root;
4278 struct btrfs_trans_handle *trans;
4281 if (BTRFS_I(inode)->dummy_inode)
4284 if (wbc->sync_mode == WB_SYNC_ALL) {
4285 trans = btrfs_join_transaction(root, 1);
4286 btrfs_set_trans_block_group(trans, inode);
4287 ret = btrfs_commit_transaction(trans, root);
4293 * This is somewhat expensive, updating the tree every time the
4294 * inode changes. But, it is most likely to find the inode in cache.
4295 * FIXME, needs more benchmarking...there are no reasons other than performance
4296 * to keep or drop this code.
4298 void btrfs_dirty_inode(struct inode *inode)
4300 struct btrfs_root *root = BTRFS_I(inode)->root;
4301 struct btrfs_trans_handle *trans;
4304 if (BTRFS_I(inode)->dummy_inode)
4307 trans = btrfs_join_transaction(root, 1);
4308 btrfs_set_trans_block_group(trans, inode);
4310 ret = btrfs_update_inode(trans, root, inode);
4311 if (ret && ret == -ENOSPC) {
4312 /* whoops, lets try again with the full transaction */
4313 btrfs_end_transaction(trans, root);
4314 trans = btrfs_start_transaction(root, 1);
4315 if (IS_ERR(trans)) {
4316 if (printk_ratelimit()) {
4317 printk(KERN_ERR "btrfs: fail to "
4318 "dirty inode %lu error %ld\n",
4319 inode->i_ino, PTR_ERR(trans));
4323 btrfs_set_trans_block_group(trans, inode);
4325 ret = btrfs_update_inode(trans, root, inode);
4327 if (printk_ratelimit()) {
4328 printk(KERN_ERR "btrfs: fail to "
4329 "dirty inode %lu error %d\n",
4334 btrfs_end_transaction(trans, root);
4338 * find the highest existing sequence number in a directory
4339 * and then set the in-memory index_cnt variable to reflect
4340 * free sequence numbers
4342 static int btrfs_set_inode_index_count(struct inode *inode)
4344 struct btrfs_root *root = BTRFS_I(inode)->root;
4345 struct btrfs_key key, found_key;
4346 struct btrfs_path *path;
4347 struct extent_buffer *leaf;
4350 key.objectid = inode->i_ino;
4351 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4352 key.offset = (u64)-1;
4354 path = btrfs_alloc_path();
4358 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4361 /* FIXME: we should be able to handle this */
4367 * MAGIC NUMBER EXPLANATION:
4368 * since we search a directory based on f_pos we have to start at 2
4369 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4370 * else has to start at 2
4372 if (path->slots[0] == 0) {
4373 BTRFS_I(inode)->index_cnt = 2;
4379 leaf = path->nodes[0];
4380 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4382 if (found_key.objectid != inode->i_ino ||
4383 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4384 BTRFS_I(inode)->index_cnt = 2;
4388 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4390 btrfs_free_path(path);
4395 * helper to find a free sequence number in a given directory. This current
4396 * code is very simple, later versions will do smarter things in the btree
4398 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4402 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4403 ret = btrfs_set_inode_index_count(dir);
4408 *index = BTRFS_I(dir)->index_cnt;
4409 BTRFS_I(dir)->index_cnt++;
4414 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4415 struct btrfs_root *root,
4417 const char *name, int name_len,
4418 u64 ref_objectid, u64 objectid,
4419 u64 alloc_hint, int mode, u64 *index)
4421 struct inode *inode;
4422 struct btrfs_inode_item *inode_item;
4423 struct btrfs_key *location;
4424 struct btrfs_path *path;
4425 struct btrfs_inode_ref *ref;
4426 struct btrfs_key key[2];
4432 path = btrfs_alloc_path();
4435 inode = new_inode(root->fs_info->sb);
4437 return ERR_PTR(-ENOMEM);
4440 ret = btrfs_set_inode_index(dir, index);
4443 return ERR_PTR(ret);
4447 * index_cnt is ignored for everything but a dir,
4448 * btrfs_get_inode_index_count has an explanation for the magic
4451 BTRFS_I(inode)->index_cnt = 2;
4452 BTRFS_I(inode)->root = root;
4453 BTRFS_I(inode)->generation = trans->transid;
4454 btrfs_set_inode_space_info(root, inode);
4460 BTRFS_I(inode)->block_group =
4461 btrfs_find_block_group(root, 0, alloc_hint, owner);
4463 key[0].objectid = objectid;
4464 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4467 key[1].objectid = objectid;
4468 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4469 key[1].offset = ref_objectid;
4471 sizes[0] = sizeof(struct btrfs_inode_item);
4472 sizes[1] = name_len + sizeof(*ref);
4474 path->leave_spinning = 1;
4475 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4479 inode_init_owner(inode, dir, mode);
4480 inode->i_ino = objectid;
4481 inode_set_bytes(inode, 0);
4482 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4483 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4484 struct btrfs_inode_item);
4485 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4487 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4488 struct btrfs_inode_ref);
4489 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4490 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4491 ptr = (unsigned long)(ref + 1);
4492 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4494 btrfs_mark_buffer_dirty(path->nodes[0]);
4495 btrfs_free_path(path);
4497 location = &BTRFS_I(inode)->location;
4498 location->objectid = objectid;
4499 location->offset = 0;
4500 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4502 btrfs_inherit_iflags(inode, dir);
4504 if ((mode & S_IFREG)) {
4505 if (btrfs_test_opt(root, NODATASUM))
4506 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4507 if (btrfs_test_opt(root, NODATACOW))
4508 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4511 insert_inode_hash(inode);
4512 inode_tree_add(inode);
4516 BTRFS_I(dir)->index_cnt--;
4517 btrfs_free_path(path);
4519 return ERR_PTR(ret);
4522 static inline u8 btrfs_inode_type(struct inode *inode)
4524 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4528 * utility function to add 'inode' into 'parent_inode' with
4529 * a give name and a given sequence number.
4530 * if 'add_backref' is true, also insert a backref from the
4531 * inode to the parent directory.
4533 int btrfs_add_link(struct btrfs_trans_handle *trans,
4534 struct inode *parent_inode, struct inode *inode,
4535 const char *name, int name_len, int add_backref, u64 index)
4538 struct btrfs_key key;
4539 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4541 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4542 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4544 key.objectid = inode->i_ino;
4545 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4549 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4550 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4551 key.objectid, root->root_key.objectid,
4552 parent_inode->i_ino,
4553 index, name, name_len);
4554 } else if (add_backref) {
4555 ret = btrfs_insert_inode_ref(trans, root,
4556 name, name_len, inode->i_ino,
4557 parent_inode->i_ino, index);
4561 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4562 parent_inode->i_ino, &key,
4563 btrfs_inode_type(inode), index);
4566 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4568 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4569 ret = btrfs_update_inode(trans, root, parent_inode);
4574 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4575 struct dentry *dentry, struct inode *inode,
4576 int backref, u64 index)
4578 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4579 inode, dentry->d_name.name,
4580 dentry->d_name.len, backref, index);
4582 d_instantiate(dentry, inode);
4590 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4591 int mode, dev_t rdev)
4593 struct btrfs_trans_handle *trans;
4594 struct btrfs_root *root = BTRFS_I(dir)->root;
4595 struct inode *inode = NULL;
4599 unsigned long nr = 0;
4602 if (!new_valid_dev(rdev))
4605 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4610 * 2 for inode item and ref
4612 * 1 for xattr if selinux is on
4614 trans = btrfs_start_transaction(root, 5);
4616 return PTR_ERR(trans);
4618 btrfs_set_trans_block_group(trans, dir);
4620 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4622 dentry->d_parent->d_inode->i_ino, objectid,
4623 BTRFS_I(dir)->block_group, mode, &index);
4624 err = PTR_ERR(inode);
4628 err = btrfs_init_inode_security(trans, inode, dir);
4634 btrfs_set_trans_block_group(trans, inode);
4635 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4639 inode->i_op = &btrfs_special_inode_operations;
4640 init_special_inode(inode, inode->i_mode, rdev);
4641 btrfs_update_inode(trans, root, inode);
4643 btrfs_update_inode_block_group(trans, inode);
4644 btrfs_update_inode_block_group(trans, dir);
4646 nr = trans->blocks_used;
4647 btrfs_end_transaction_throttle(trans, root);
4648 btrfs_btree_balance_dirty(root, nr);
4650 inode_dec_link_count(inode);
4656 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4657 int mode, struct nameidata *nd)
4659 struct btrfs_trans_handle *trans;
4660 struct btrfs_root *root = BTRFS_I(dir)->root;
4661 struct inode *inode = NULL;
4664 unsigned long nr = 0;
4668 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4672 * 2 for inode item and ref
4674 * 1 for xattr if selinux is on
4676 trans = btrfs_start_transaction(root, 5);
4678 return PTR_ERR(trans);
4680 btrfs_set_trans_block_group(trans, dir);
4682 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4684 dentry->d_parent->d_inode->i_ino,
4685 objectid, BTRFS_I(dir)->block_group, mode,
4687 err = PTR_ERR(inode);
4691 err = btrfs_init_inode_security(trans, inode, dir);
4697 btrfs_set_trans_block_group(trans, inode);
4698 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4702 inode->i_mapping->a_ops = &btrfs_aops;
4703 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4704 inode->i_fop = &btrfs_file_operations;
4705 inode->i_op = &btrfs_file_inode_operations;
4706 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4708 btrfs_update_inode_block_group(trans, inode);
4709 btrfs_update_inode_block_group(trans, dir);
4711 nr = trans->blocks_used;
4712 btrfs_end_transaction_throttle(trans, root);
4714 inode_dec_link_count(inode);
4717 btrfs_btree_balance_dirty(root, nr);
4721 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4722 struct dentry *dentry)
4724 struct btrfs_trans_handle *trans;
4725 struct btrfs_root *root = BTRFS_I(dir)->root;
4726 struct inode *inode = old_dentry->d_inode;
4728 unsigned long nr = 0;
4732 if (inode->i_nlink == 0)
4735 /* do not allow sys_link's with other subvols of the same device */
4736 if (root->objectid != BTRFS_I(inode)->root->objectid)
4739 btrfs_inc_nlink(inode);
4741 err = btrfs_set_inode_index(dir, &index);
4746 * 1 item for inode ref
4747 * 2 items for dir items
4749 trans = btrfs_start_transaction(root, 3);
4750 if (IS_ERR(trans)) {
4751 err = PTR_ERR(trans);
4755 btrfs_set_trans_block_group(trans, dir);
4756 atomic_inc(&inode->i_count);
4758 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4763 btrfs_update_inode_block_group(trans, dir);
4764 err = btrfs_update_inode(trans, root, inode);
4766 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4769 nr = trans->blocks_used;
4770 btrfs_end_transaction_throttle(trans, root);
4773 inode_dec_link_count(inode);
4776 btrfs_btree_balance_dirty(root, nr);
4780 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4782 struct inode *inode = NULL;
4783 struct btrfs_trans_handle *trans;
4784 struct btrfs_root *root = BTRFS_I(dir)->root;
4786 int drop_on_err = 0;
4789 unsigned long nr = 1;
4791 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4796 * 2 items for inode and ref
4797 * 2 items for dir items
4798 * 1 for xattr if selinux is on
4800 trans = btrfs_start_transaction(root, 5);
4802 return PTR_ERR(trans);
4803 btrfs_set_trans_block_group(trans, dir);
4805 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4807 dentry->d_parent->d_inode->i_ino, objectid,
4808 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4810 if (IS_ERR(inode)) {
4811 err = PTR_ERR(inode);
4817 err = btrfs_init_inode_security(trans, inode, dir);
4821 inode->i_op = &btrfs_dir_inode_operations;
4822 inode->i_fop = &btrfs_dir_file_operations;
4823 btrfs_set_trans_block_group(trans, inode);
4825 btrfs_i_size_write(inode, 0);
4826 err = btrfs_update_inode(trans, root, inode);
4830 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4831 inode, dentry->d_name.name,
4832 dentry->d_name.len, 0, index);
4836 d_instantiate(dentry, inode);
4838 btrfs_update_inode_block_group(trans, inode);
4839 btrfs_update_inode_block_group(trans, dir);
4842 nr = trans->blocks_used;
4843 btrfs_end_transaction_throttle(trans, root);
4846 btrfs_btree_balance_dirty(root, nr);
4850 /* helper for btfs_get_extent. Given an existing extent in the tree,
4851 * and an extent that you want to insert, deal with overlap and insert
4852 * the new extent into the tree.
4854 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4855 struct extent_map *existing,
4856 struct extent_map *em,
4857 u64 map_start, u64 map_len)
4861 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4862 start_diff = map_start - em->start;
4863 em->start = map_start;
4865 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4866 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4867 em->block_start += start_diff;
4868 em->block_len -= start_diff;
4870 return add_extent_mapping(em_tree, em);
4873 static noinline int uncompress_inline(struct btrfs_path *path,
4874 struct inode *inode, struct page *page,
4875 size_t pg_offset, u64 extent_offset,
4876 struct btrfs_file_extent_item *item)
4879 struct extent_buffer *leaf = path->nodes[0];
4882 unsigned long inline_size;
4885 WARN_ON(pg_offset != 0);
4886 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4887 inline_size = btrfs_file_extent_inline_item_len(leaf,
4888 btrfs_item_nr(leaf, path->slots[0]));
4889 tmp = kmalloc(inline_size, GFP_NOFS);
4890 ptr = btrfs_file_extent_inline_start(item);
4892 read_extent_buffer(leaf, tmp, ptr, inline_size);
4894 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4895 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4896 inline_size, max_size);
4898 char *kaddr = kmap_atomic(page, KM_USER0);
4899 unsigned long copy_size = min_t(u64,
4900 PAGE_CACHE_SIZE - pg_offset,
4901 max_size - extent_offset);
4902 memset(kaddr + pg_offset, 0, copy_size);
4903 kunmap_atomic(kaddr, KM_USER0);
4910 * a bit scary, this does extent mapping from logical file offset to the disk.
4911 * the ugly parts come from merging extents from the disk with the in-ram
4912 * representation. This gets more complex because of the data=ordered code,
4913 * where the in-ram extents might be locked pending data=ordered completion.
4915 * This also copies inline extents directly into the page.
4918 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4919 size_t pg_offset, u64 start, u64 len,
4925 u64 extent_start = 0;
4927 u64 objectid = inode->i_ino;
4929 struct btrfs_path *path = NULL;
4930 struct btrfs_root *root = BTRFS_I(inode)->root;
4931 struct btrfs_file_extent_item *item;
4932 struct extent_buffer *leaf;
4933 struct btrfs_key found_key;
4934 struct extent_map *em = NULL;
4935 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4936 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4937 struct btrfs_trans_handle *trans = NULL;
4941 read_lock(&em_tree->lock);
4942 em = lookup_extent_mapping(em_tree, start, len);
4944 em->bdev = root->fs_info->fs_devices->latest_bdev;
4945 read_unlock(&em_tree->lock);
4948 if (em->start > start || em->start + em->len <= start)
4949 free_extent_map(em);
4950 else if (em->block_start == EXTENT_MAP_INLINE && page)
4951 free_extent_map(em);
4955 em = alloc_extent_map(GFP_NOFS);
4960 em->bdev = root->fs_info->fs_devices->latest_bdev;
4961 em->start = EXTENT_MAP_HOLE;
4962 em->orig_start = EXTENT_MAP_HOLE;
4964 em->block_len = (u64)-1;
4967 path = btrfs_alloc_path();
4971 ret = btrfs_lookup_file_extent(trans, root, path,
4972 objectid, start, trans != NULL);
4979 if (path->slots[0] == 0)
4984 leaf = path->nodes[0];
4985 item = btrfs_item_ptr(leaf, path->slots[0],
4986 struct btrfs_file_extent_item);
4987 /* are we inside the extent that was found? */
4988 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4989 found_type = btrfs_key_type(&found_key);
4990 if (found_key.objectid != objectid ||
4991 found_type != BTRFS_EXTENT_DATA_KEY) {
4995 found_type = btrfs_file_extent_type(leaf, item);
4996 extent_start = found_key.offset;
4997 compressed = btrfs_file_extent_compression(leaf, item);
4998 if (found_type == BTRFS_FILE_EXTENT_REG ||
4999 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5000 extent_end = extent_start +
5001 btrfs_file_extent_num_bytes(leaf, item);
5002 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5004 size = btrfs_file_extent_inline_len(leaf, item);
5005 extent_end = (extent_start + size + root->sectorsize - 1) &
5006 ~((u64)root->sectorsize - 1);
5009 if (start >= extent_end) {
5011 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5012 ret = btrfs_next_leaf(root, path);
5019 leaf = path->nodes[0];
5021 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5022 if (found_key.objectid != objectid ||
5023 found_key.type != BTRFS_EXTENT_DATA_KEY)
5025 if (start + len <= found_key.offset)
5028 em->len = found_key.offset - start;
5032 if (found_type == BTRFS_FILE_EXTENT_REG ||
5033 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5034 em->start = extent_start;
5035 em->len = extent_end - extent_start;
5036 em->orig_start = extent_start -
5037 btrfs_file_extent_offset(leaf, item);
5038 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5040 em->block_start = EXTENT_MAP_HOLE;
5044 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5045 em->block_start = bytenr;
5046 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5049 bytenr += btrfs_file_extent_offset(leaf, item);
5050 em->block_start = bytenr;
5051 em->block_len = em->len;
5052 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5053 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5056 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5060 size_t extent_offset;
5063 em->block_start = EXTENT_MAP_INLINE;
5064 if (!page || create) {
5065 em->start = extent_start;
5066 em->len = extent_end - extent_start;
5070 size = btrfs_file_extent_inline_len(leaf, item);
5071 extent_offset = page_offset(page) + pg_offset - extent_start;
5072 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5073 size - extent_offset);
5074 em->start = extent_start + extent_offset;
5075 em->len = (copy_size + root->sectorsize - 1) &
5076 ~((u64)root->sectorsize - 1);
5077 em->orig_start = EXTENT_MAP_INLINE;
5079 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5080 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5081 if (create == 0 && !PageUptodate(page)) {
5082 if (btrfs_file_extent_compression(leaf, item) ==
5083 BTRFS_COMPRESS_ZLIB) {
5084 ret = uncompress_inline(path, inode, page,
5086 extent_offset, item);
5090 read_extent_buffer(leaf, map + pg_offset, ptr,
5092 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5093 memset(map + pg_offset + copy_size, 0,
5094 PAGE_CACHE_SIZE - pg_offset -
5099 flush_dcache_page(page);
5100 } else if (create && PageUptodate(page)) {
5104 free_extent_map(em);
5106 btrfs_release_path(root, path);
5107 trans = btrfs_join_transaction(root, 1);
5111 write_extent_buffer(leaf, map + pg_offset, ptr,
5114 btrfs_mark_buffer_dirty(leaf);
5116 set_extent_uptodate(io_tree, em->start,
5117 extent_map_end(em) - 1, GFP_NOFS);
5120 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5127 em->block_start = EXTENT_MAP_HOLE;
5128 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5130 btrfs_release_path(root, path);
5131 if (em->start > start || extent_map_end(em) <= start) {
5132 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5133 "[%llu %llu]\n", (unsigned long long)em->start,
5134 (unsigned long long)em->len,
5135 (unsigned long long)start,
5136 (unsigned long long)len);
5142 write_lock(&em_tree->lock);
5143 ret = add_extent_mapping(em_tree, em);
5144 /* it is possible that someone inserted the extent into the tree
5145 * while we had the lock dropped. It is also possible that
5146 * an overlapping map exists in the tree
5148 if (ret == -EEXIST) {
5149 struct extent_map *existing;
5153 existing = lookup_extent_mapping(em_tree, start, len);
5154 if (existing && (existing->start > start ||
5155 existing->start + existing->len <= start)) {
5156 free_extent_map(existing);
5160 existing = lookup_extent_mapping(em_tree, em->start,
5163 err = merge_extent_mapping(em_tree, existing,
5166 free_extent_map(existing);
5168 free_extent_map(em);
5173 free_extent_map(em);
5177 free_extent_map(em);
5182 write_unlock(&em_tree->lock);
5185 btrfs_free_path(path);
5187 ret = btrfs_end_transaction(trans, root);
5192 free_extent_map(em);
5193 return ERR_PTR(err);
5198 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5201 struct btrfs_root *root = BTRFS_I(inode)->root;
5202 struct btrfs_trans_handle *trans;
5203 struct extent_map *em;
5204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5205 struct btrfs_key ins;
5209 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5211 trans = btrfs_join_transaction(root, 0);
5213 return ERR_PTR(-ENOMEM);
5215 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5217 alloc_hint = get_extent_allocation_hint(inode, start, len);
5218 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5219 alloc_hint, (u64)-1, &ins, 1);
5225 em = alloc_extent_map(GFP_NOFS);
5227 em = ERR_PTR(-ENOMEM);
5232 em->orig_start = em->start;
5233 em->len = ins.offset;
5235 em->block_start = ins.objectid;
5236 em->block_len = ins.offset;
5237 em->bdev = root->fs_info->fs_devices->latest_bdev;
5238 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5241 write_lock(&em_tree->lock);
5242 ret = add_extent_mapping(em_tree, em);
5243 write_unlock(&em_tree->lock);
5246 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5249 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5250 ins.offset, ins.offset, 0);
5252 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5256 btrfs_end_transaction(trans, root);
5261 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5262 * block must be cow'd
5264 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5265 struct inode *inode, u64 offset, u64 len)
5267 struct btrfs_path *path;
5269 struct extent_buffer *leaf;
5270 struct btrfs_root *root = BTRFS_I(inode)->root;
5271 struct btrfs_file_extent_item *fi;
5272 struct btrfs_key key;
5280 path = btrfs_alloc_path();
5284 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5289 slot = path->slots[0];
5292 /* can't find the item, must cow */
5299 leaf = path->nodes[0];
5300 btrfs_item_key_to_cpu(leaf, &key, slot);
5301 if (key.objectid != inode->i_ino ||
5302 key.type != BTRFS_EXTENT_DATA_KEY) {
5303 /* not our file or wrong item type, must cow */
5307 if (key.offset > offset) {
5308 /* Wrong offset, must cow */
5312 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5313 found_type = btrfs_file_extent_type(leaf, fi);
5314 if (found_type != BTRFS_FILE_EXTENT_REG &&
5315 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5316 /* not a regular extent, must cow */
5319 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5320 backref_offset = btrfs_file_extent_offset(leaf, fi);
5322 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5323 if (extent_end < offset + len) {
5324 /* extent doesn't include our full range, must cow */
5328 if (btrfs_extent_readonly(root, disk_bytenr))
5332 * look for other files referencing this extent, if we
5333 * find any we must cow
5335 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5336 key.offset - backref_offset, disk_bytenr))
5340 * adjust disk_bytenr and num_bytes to cover just the bytes
5341 * in this extent we are about to write. If there
5342 * are any csums in that range we have to cow in order
5343 * to keep the csums correct
5345 disk_bytenr += backref_offset;
5346 disk_bytenr += offset - key.offset;
5347 num_bytes = min(offset + len, extent_end) - offset;
5348 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5351 * all of the above have passed, it is safe to overwrite this extent
5356 btrfs_free_path(path);
5360 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5361 struct buffer_head *bh_result, int create)
5363 struct extent_map *em;
5364 struct btrfs_root *root = BTRFS_I(inode)->root;
5365 u64 start = iblock << inode->i_blkbits;
5366 u64 len = bh_result->b_size;
5367 struct btrfs_trans_handle *trans;
5369 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5374 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5375 * io. INLINE is special, and we could probably kludge it in here, but
5376 * it's still buffered so for safety lets just fall back to the generic
5379 * For COMPRESSED we _have_ to read the entire extent in so we can
5380 * decompress it, so there will be buffering required no matter what we
5381 * do, so go ahead and fallback to buffered.
5383 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5384 * to buffered IO. Don't blame me, this is the price we pay for using
5387 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5388 em->block_start == EXTENT_MAP_INLINE) {
5389 free_extent_map(em);
5393 /* Just a good old fashioned hole, return */
5394 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5395 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5396 free_extent_map(em);
5397 /* DIO will do one hole at a time, so just unlock a sector */
5398 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5399 start + root->sectorsize - 1, GFP_NOFS);
5404 * We don't allocate a new extent in the following cases
5406 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5408 * 2) The extent is marked as PREALLOC. We're good to go here and can
5409 * just use the extent.
5413 len = em->len - (start - em->start);
5417 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5418 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5419 em->block_start != EXTENT_MAP_HOLE)) {
5424 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5425 type = BTRFS_ORDERED_PREALLOC;
5427 type = BTRFS_ORDERED_NOCOW;
5428 len = min(len, em->len - (start - em->start));
5429 block_start = em->block_start + (start - em->start);
5432 * we're not going to log anything, but we do need
5433 * to make sure the current transaction stays open
5434 * while we look for nocow cross refs
5436 trans = btrfs_join_transaction(root, 0);
5440 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5441 ret = btrfs_add_ordered_extent_dio(inode, start,
5442 block_start, len, len, type);
5443 btrfs_end_transaction(trans, root);
5445 free_extent_map(em);
5450 btrfs_end_transaction(trans, root);
5454 * this will cow the extent, reset the len in case we changed
5457 len = bh_result->b_size;
5458 free_extent_map(em);
5459 em = btrfs_new_extent_direct(inode, start, len);
5462 len = min(len, em->len - (start - em->start));
5464 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5465 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5468 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5470 bh_result->b_size = len;
5471 bh_result->b_bdev = em->bdev;
5472 set_buffer_mapped(bh_result);
5473 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5474 set_buffer_new(bh_result);
5476 free_extent_map(em);
5481 struct btrfs_dio_private {
5482 struct inode *inode;
5490 static void btrfs_endio_direct_read(struct bio *bio, int err)
5492 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5493 struct bio_vec *bvec = bio->bi_io_vec;
5494 struct btrfs_dio_private *dip = bio->bi_private;
5495 struct inode *inode = dip->inode;
5496 struct btrfs_root *root = BTRFS_I(inode)->root;
5498 u32 *private = dip->csums;
5500 start = dip->logical_offset;
5502 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5503 struct page *page = bvec->bv_page;
5506 unsigned long flags;
5508 local_irq_save(flags);
5509 kaddr = kmap_atomic(page, KM_IRQ0);
5510 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5511 csum, bvec->bv_len);
5512 btrfs_csum_final(csum, (char *)&csum);
5513 kunmap_atomic(kaddr, KM_IRQ0);
5514 local_irq_restore(flags);
5516 flush_dcache_page(bvec->bv_page);
5517 if (csum != *private) {
5518 printk(KERN_ERR "btrfs csum failed ino %lu off"
5519 " %llu csum %u private %u\n",
5520 inode->i_ino, (unsigned long long)start,
5526 start += bvec->bv_len;
5529 } while (bvec <= bvec_end);
5531 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5532 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5533 bio->bi_private = dip->private;
5537 dio_end_io(bio, err);
5540 static void btrfs_endio_direct_write(struct bio *bio, int err)
5542 struct btrfs_dio_private *dip = bio->bi_private;
5543 struct inode *inode = dip->inode;
5544 struct btrfs_root *root = BTRFS_I(inode)->root;
5545 struct btrfs_trans_handle *trans;
5546 struct btrfs_ordered_extent *ordered = NULL;
5547 struct extent_state *cached_state = NULL;
5553 ret = btrfs_dec_test_ordered_pending(inode, &ordered,
5554 dip->logical_offset, dip->bytes);
5560 trans = btrfs_join_transaction(root, 1);
5565 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5567 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5568 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5570 ret = btrfs_update_inode(trans, root, inode);
5575 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5576 ordered->file_offset + ordered->len - 1, 0,
5577 &cached_state, GFP_NOFS);
5579 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5580 ret = btrfs_mark_extent_written(trans, inode,
5581 ordered->file_offset,
5582 ordered->file_offset +
5589 ret = insert_reserved_file_extent(trans, inode,
5590 ordered->file_offset,
5596 BTRFS_FILE_EXTENT_REG);
5597 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5598 ordered->file_offset, ordered->len);
5606 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5607 btrfs_ordered_update_i_size(inode, 0, ordered);
5608 btrfs_update_inode(trans, root, inode);
5610 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5611 ordered->file_offset + ordered->len - 1,
5612 &cached_state, GFP_NOFS);
5614 btrfs_delalloc_release_metadata(inode, ordered->len);
5615 btrfs_end_transaction(trans, root);
5616 btrfs_put_ordered_extent(ordered);
5617 btrfs_put_ordered_extent(ordered);
5619 bio->bi_private = dip->private;
5623 dio_end_io(bio, err);
5626 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5627 struct bio *bio, int mirror_num,
5628 unsigned long bio_flags, u64 offset)
5631 struct btrfs_root *root = BTRFS_I(inode)->root;
5632 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5637 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5640 struct btrfs_root *root = BTRFS_I(inode)->root;
5641 struct btrfs_dio_private *dip;
5642 struct bio_vec *bvec = bio->bi_io_vec;
5645 int write = rw & (1 << BIO_RW);
5648 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5650 dip = kmalloc(sizeof(*dip), GFP_NOFS);
5658 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5665 dip->private = bio->bi_private;
5667 dip->logical_offset = file_offset;
5669 start = dip->logical_offset;
5672 dip->bytes += bvec->bv_len;
5674 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5676 dip->disk_bytenr = (u64)bio->bi_sector << 9;
5677 bio->bi_private = dip;
5680 bio->bi_end_io = btrfs_endio_direct_write;
5682 bio->bi_end_io = btrfs_endio_direct_read;
5684 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5688 if (write && !skip_sum) {
5689 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
5690 inode, rw, bio, 0, 0,
5691 dip->logical_offset,
5692 __btrfs_submit_bio_start_direct_io,
5693 __btrfs_submit_bio_done);
5697 } else if (!skip_sum)
5698 btrfs_lookup_bio_sums_dio(root, inode, bio,
5699 dip->logical_offset, dip->csums);
5701 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5710 * If this is a write, we need to clean up the reserved space and kill
5711 * the ordered extent.
5714 struct btrfs_ordered_extent *ordered;
5715 ordered = btrfs_lookup_ordered_extent(inode,
5716 dip->logical_offset);
5717 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
5718 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
5719 btrfs_free_reserved_extent(root, ordered->start,
5721 btrfs_put_ordered_extent(ordered);
5722 btrfs_put_ordered_extent(ordered);
5724 bio_endio(bio, ret);
5727 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
5728 const struct iovec *iov, loff_t offset,
5729 unsigned long nr_segs)
5734 unsigned blocksize_mask = root->sectorsize - 1;
5735 ssize_t retval = -EINVAL;
5736 loff_t end = offset;
5738 if (offset & blocksize_mask)
5741 /* Check the memory alignment. Blocks cannot straddle pages */
5742 for (seg = 0; seg < nr_segs; seg++) {
5743 addr = (unsigned long)iov[seg].iov_base;
5744 size = iov[seg].iov_len;
5746 if ((addr & blocksize_mask) || (size & blocksize_mask))
5753 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
5754 const struct iovec *iov, loff_t offset,
5755 unsigned long nr_segs)
5757 struct file *file = iocb->ki_filp;
5758 struct inode *inode = file->f_mapping->host;
5759 struct btrfs_ordered_extent *ordered;
5760 struct extent_state *cached_state = NULL;
5761 u64 lockstart, lockend;
5763 int writing = rw & WRITE;
5765 size_t count = iov_length(iov, nr_segs);
5767 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
5773 lockend = offset + count - 1;
5776 ret = btrfs_delalloc_reserve_space(inode, count);
5782 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5783 0, &cached_state, GFP_NOFS);
5785 * We're concerned with the entire range that we're going to be
5786 * doing DIO to, so we need to make sure theres no ordered
5787 * extents in this range.
5789 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5790 lockend - lockstart + 1);
5793 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5794 &cached_state, GFP_NOFS);
5795 btrfs_start_ordered_extent(inode, ordered, 1);
5796 btrfs_put_ordered_extent(ordered);
5801 * we don't use btrfs_set_extent_delalloc because we don't want
5802 * the dirty or uptodate bits
5805 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
5806 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5807 EXTENT_DELALLOC, 0, NULL, &cached_state,
5810 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5811 lockend, EXTENT_LOCKED | write_bits,
5812 1, 0, &cached_state, GFP_NOFS);
5817 free_extent_state(cached_state);
5818 cached_state = NULL;
5820 ret = __blockdev_direct_IO(rw, iocb, inode,
5821 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
5822 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
5823 btrfs_submit_direct, 0);
5825 if (ret < 0 && ret != -EIOCBQUEUED) {
5826 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
5827 offset + iov_length(iov, nr_segs) - 1,
5828 EXTENT_LOCKED | write_bits, 1, 0,
5829 &cached_state, GFP_NOFS);
5830 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
5832 * We're falling back to buffered, unlock the section we didn't
5835 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
5836 offset + iov_length(iov, nr_segs) - 1,
5837 EXTENT_LOCKED | write_bits, 1, 0,
5838 &cached_state, GFP_NOFS);
5841 free_extent_state(cached_state);
5845 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5846 __u64 start, __u64 len)
5848 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
5851 int btrfs_readpage(struct file *file, struct page *page)
5853 struct extent_io_tree *tree;
5854 tree = &BTRFS_I(page->mapping->host)->io_tree;
5855 return extent_read_full_page(tree, page, btrfs_get_extent);
5858 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
5860 struct extent_io_tree *tree;
5863 if (current->flags & PF_MEMALLOC) {
5864 redirty_page_for_writepage(wbc, page);
5868 tree = &BTRFS_I(page->mapping->host)->io_tree;
5869 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
5872 int btrfs_writepages(struct address_space *mapping,
5873 struct writeback_control *wbc)
5875 struct extent_io_tree *tree;
5877 tree = &BTRFS_I(mapping->host)->io_tree;
5878 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
5882 btrfs_readpages(struct file *file, struct address_space *mapping,
5883 struct list_head *pages, unsigned nr_pages)
5885 struct extent_io_tree *tree;
5886 tree = &BTRFS_I(mapping->host)->io_tree;
5887 return extent_readpages(tree, mapping, pages, nr_pages,
5890 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5892 struct extent_io_tree *tree;
5893 struct extent_map_tree *map;
5896 tree = &BTRFS_I(page->mapping->host)->io_tree;
5897 map = &BTRFS_I(page->mapping->host)->extent_tree;
5898 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
5900 ClearPagePrivate(page);
5901 set_page_private(page, 0);
5902 page_cache_release(page);
5907 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5909 if (PageWriteback(page) || PageDirty(page))
5911 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
5914 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
5916 struct extent_io_tree *tree;
5917 struct btrfs_ordered_extent *ordered;
5918 struct extent_state *cached_state = NULL;
5919 u64 page_start = page_offset(page);
5920 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
5924 * we have the page locked, so new writeback can't start,
5925 * and the dirty bit won't be cleared while we are here.
5927 * Wait for IO on this page so that we can safely clear
5928 * the PagePrivate2 bit and do ordered accounting
5930 wait_on_page_writeback(page);
5932 tree = &BTRFS_I(page->mapping->host)->io_tree;
5934 btrfs_releasepage(page, GFP_NOFS);
5937 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5939 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5943 * IO on this page will never be started, so we need
5944 * to account for any ordered extents now
5946 clear_extent_bit(tree, page_start, page_end,
5947 EXTENT_DIRTY | EXTENT_DELALLOC |
5948 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5949 &cached_state, GFP_NOFS);
5951 * whoever cleared the private bit is responsible
5952 * for the finish_ordered_io
5954 if (TestClearPagePrivate2(page)) {
5955 btrfs_finish_ordered_io(page->mapping->host,
5956 page_start, page_end);
5958 btrfs_put_ordered_extent(ordered);
5959 cached_state = NULL;
5960 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5963 clear_extent_bit(tree, page_start, page_end,
5964 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5965 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
5966 __btrfs_releasepage(page, GFP_NOFS);
5968 ClearPageChecked(page);
5969 if (PagePrivate(page)) {
5970 ClearPagePrivate(page);
5971 set_page_private(page, 0);
5972 page_cache_release(page);
5977 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5978 * called from a page fault handler when a page is first dirtied. Hence we must
5979 * be careful to check for EOF conditions here. We set the page up correctly
5980 * for a written page which means we get ENOSPC checking when writing into
5981 * holes and correct delalloc and unwritten extent mapping on filesystems that
5982 * support these features.
5984 * We are not allowed to take the i_mutex here so we have to play games to
5985 * protect against truncate races as the page could now be beyond EOF. Because
5986 * vmtruncate() writes the inode size before removing pages, once we have the
5987 * page lock we can determine safely if the page is beyond EOF. If it is not
5988 * beyond EOF, then the page is guaranteed safe against truncation until we
5991 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5993 struct page *page = vmf->page;
5994 struct inode *inode = fdentry(vma->vm_file)->d_inode;
5995 struct btrfs_root *root = BTRFS_I(inode)->root;
5996 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5997 struct btrfs_ordered_extent *ordered;
5998 struct extent_state *cached_state = NULL;
6000 unsigned long zero_start;
6006 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6010 else /* -ENOSPC, -EIO, etc */
6011 ret = VM_FAULT_SIGBUS;
6015 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6018 size = i_size_read(inode);
6019 page_start = page_offset(page);
6020 page_end = page_start + PAGE_CACHE_SIZE - 1;
6022 if ((page->mapping != inode->i_mapping) ||
6023 (page_start >= size)) {
6024 /* page got truncated out from underneath us */
6027 wait_on_page_writeback(page);
6029 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6031 set_page_extent_mapped(page);
6034 * we can't set the delalloc bits if there are pending ordered
6035 * extents. Drop our locks and wait for them to finish
6037 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6039 unlock_extent_cached(io_tree, page_start, page_end,
6040 &cached_state, GFP_NOFS);
6042 btrfs_start_ordered_extent(inode, ordered, 1);
6043 btrfs_put_ordered_extent(ordered);
6048 * XXX - page_mkwrite gets called every time the page is dirtied, even
6049 * if it was already dirty, so for space accounting reasons we need to
6050 * clear any delalloc bits for the range we are fixing to save. There
6051 * is probably a better way to do this, but for now keep consistent with
6052 * prepare_pages in the normal write path.
6054 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6055 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6056 0, 0, &cached_state, GFP_NOFS);
6058 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6061 unlock_extent_cached(io_tree, page_start, page_end,
6062 &cached_state, GFP_NOFS);
6063 ret = VM_FAULT_SIGBUS;
6068 /* page is wholly or partially inside EOF */
6069 if (page_start + PAGE_CACHE_SIZE > size)
6070 zero_start = size & ~PAGE_CACHE_MASK;
6072 zero_start = PAGE_CACHE_SIZE;
6074 if (zero_start != PAGE_CACHE_SIZE) {
6076 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6077 flush_dcache_page(page);
6080 ClearPageChecked(page);
6081 set_page_dirty(page);
6082 SetPageUptodate(page);
6084 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6085 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6087 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6091 return VM_FAULT_LOCKED;
6093 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6098 static void btrfs_truncate(struct inode *inode)
6100 struct btrfs_root *root = BTRFS_I(inode)->root;
6102 struct btrfs_trans_handle *trans;
6104 u64 mask = root->sectorsize - 1;
6106 if (!S_ISREG(inode->i_mode)) {
6111 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6115 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6116 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6118 trans = btrfs_start_transaction(root, 0);
6119 BUG_ON(IS_ERR(trans));
6120 btrfs_set_trans_block_group(trans, inode);
6121 trans->block_rsv = root->orphan_block_rsv;
6124 * setattr is responsible for setting the ordered_data_close flag,
6125 * but that is only tested during the last file release. That
6126 * could happen well after the next commit, leaving a great big
6127 * window where new writes may get lost if someone chooses to write
6128 * to this file after truncating to zero
6130 * The inode doesn't have any dirty data here, and so if we commit
6131 * this is a noop. If someone immediately starts writing to the inode
6132 * it is very likely we'll catch some of their writes in this
6133 * transaction, and the commit will find this file on the ordered
6134 * data list with good things to send down.
6136 * This is a best effort solution, there is still a window where
6137 * using truncate to replace the contents of the file will
6138 * end up with a zero length file after a crash.
6140 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6141 btrfs_add_ordered_operation(trans, root, inode);
6145 trans = btrfs_start_transaction(root, 0);
6146 BUG_ON(IS_ERR(trans));
6147 btrfs_set_trans_block_group(trans, inode);
6148 trans->block_rsv = root->orphan_block_rsv;
6151 ret = btrfs_block_rsv_check(trans, root,
6152 root->orphan_block_rsv, 0, 5);
6154 BUG_ON(ret != -EAGAIN);
6155 ret = btrfs_commit_transaction(trans, root);
6161 ret = btrfs_truncate_inode_items(trans, root, inode,
6163 BTRFS_EXTENT_DATA_KEY);
6167 ret = btrfs_update_inode(trans, root, inode);
6170 nr = trans->blocks_used;
6171 btrfs_end_transaction(trans, root);
6173 btrfs_btree_balance_dirty(root, nr);
6176 if (ret == 0 && inode->i_nlink > 0) {
6177 ret = btrfs_orphan_del(trans, inode);
6181 ret = btrfs_update_inode(trans, root, inode);
6184 nr = trans->blocks_used;
6185 ret = btrfs_end_transaction_throttle(trans, root);
6187 btrfs_btree_balance_dirty(root, nr);
6191 * create a new subvolume directory/inode (helper for the ioctl).
6193 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6194 struct btrfs_root *new_root,
6195 u64 new_dirid, u64 alloc_hint)
6197 struct inode *inode;
6201 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6202 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6204 return PTR_ERR(inode);
6205 inode->i_op = &btrfs_dir_inode_operations;
6206 inode->i_fop = &btrfs_dir_file_operations;
6209 btrfs_i_size_write(inode, 0);
6211 err = btrfs_update_inode(trans, new_root, inode);
6218 /* helper function for file defrag and space balancing. This
6219 * forces readahead on a given range of bytes in an inode
6221 unsigned long btrfs_force_ra(struct address_space *mapping,
6222 struct file_ra_state *ra, struct file *file,
6223 pgoff_t offset, pgoff_t last_index)
6225 pgoff_t req_size = last_index - offset + 1;
6227 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6228 return offset + req_size;
6231 struct inode *btrfs_alloc_inode(struct super_block *sb)
6233 struct btrfs_inode *ei;
6234 struct inode *inode;
6236 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6241 ei->space_info = NULL;
6245 ei->last_sub_trans = 0;
6246 ei->logged_trans = 0;
6247 ei->delalloc_bytes = 0;
6248 ei->reserved_bytes = 0;
6249 ei->disk_i_size = 0;
6251 ei->index_cnt = (u64)-1;
6252 ei->last_unlink_trans = 0;
6254 spin_lock_init(&ei->accounting_lock);
6255 atomic_set(&ei->outstanding_extents, 0);
6256 ei->reserved_extents = 0;
6258 ei->ordered_data_close = 0;
6259 ei->orphan_meta_reserved = 0;
6260 ei->dummy_inode = 0;
6261 ei->force_compress = 0;
6263 inode = &ei->vfs_inode;
6264 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6265 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6266 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6267 mutex_init(&ei->log_mutex);
6268 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6269 INIT_LIST_HEAD(&ei->i_orphan);
6270 INIT_LIST_HEAD(&ei->delalloc_inodes);
6271 INIT_LIST_HEAD(&ei->ordered_operations);
6272 RB_CLEAR_NODE(&ei->rb_node);
6277 void btrfs_destroy_inode(struct inode *inode)
6279 struct btrfs_ordered_extent *ordered;
6280 struct btrfs_root *root = BTRFS_I(inode)->root;
6282 WARN_ON(!list_empty(&inode->i_dentry));
6283 WARN_ON(inode->i_data.nrpages);
6284 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6285 WARN_ON(BTRFS_I(inode)->reserved_extents);
6288 * This can happen where we create an inode, but somebody else also
6289 * created the same inode and we need to destroy the one we already
6296 * Make sure we're properly removed from the ordered operation
6300 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6301 spin_lock(&root->fs_info->ordered_extent_lock);
6302 list_del_init(&BTRFS_I(inode)->ordered_operations);
6303 spin_unlock(&root->fs_info->ordered_extent_lock);
6306 spin_lock(&root->orphan_lock);
6307 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6308 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6310 list_del_init(&BTRFS_I(inode)->i_orphan);
6312 spin_unlock(&root->orphan_lock);
6315 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6319 printk(KERN_ERR "btrfs found ordered "
6320 "extent %llu %llu on inode cleanup\n",
6321 (unsigned long long)ordered->file_offset,
6322 (unsigned long long)ordered->len);
6323 btrfs_remove_ordered_extent(inode, ordered);
6324 btrfs_put_ordered_extent(ordered);
6325 btrfs_put_ordered_extent(ordered);
6328 inode_tree_del(inode);
6329 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6331 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6334 void btrfs_drop_inode(struct inode *inode)
6336 struct btrfs_root *root = BTRFS_I(inode)->root;
6337 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
6338 generic_delete_inode(inode);
6340 generic_drop_inode(inode);
6343 static void init_once(void *foo)
6345 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6347 inode_init_once(&ei->vfs_inode);
6350 void btrfs_destroy_cachep(void)
6352 if (btrfs_inode_cachep)
6353 kmem_cache_destroy(btrfs_inode_cachep);
6354 if (btrfs_trans_handle_cachep)
6355 kmem_cache_destroy(btrfs_trans_handle_cachep);
6356 if (btrfs_transaction_cachep)
6357 kmem_cache_destroy(btrfs_transaction_cachep);
6358 if (btrfs_path_cachep)
6359 kmem_cache_destroy(btrfs_path_cachep);
6362 int btrfs_init_cachep(void)
6364 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6365 sizeof(struct btrfs_inode), 0,
6366 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6367 if (!btrfs_inode_cachep)
6370 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6371 sizeof(struct btrfs_trans_handle), 0,
6372 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6373 if (!btrfs_trans_handle_cachep)
6376 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6377 sizeof(struct btrfs_transaction), 0,
6378 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6379 if (!btrfs_transaction_cachep)
6382 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6383 sizeof(struct btrfs_path), 0,
6384 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6385 if (!btrfs_path_cachep)
6390 btrfs_destroy_cachep();
6394 static int btrfs_getattr(struct vfsmount *mnt,
6395 struct dentry *dentry, struct kstat *stat)
6397 struct inode *inode = dentry->d_inode;
6398 generic_fillattr(inode, stat);
6399 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6400 stat->blksize = PAGE_CACHE_SIZE;
6401 stat->blocks = (inode_get_bytes(inode) +
6402 BTRFS_I(inode)->delalloc_bytes) >> 9;
6406 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6407 struct inode *new_dir, struct dentry *new_dentry)
6409 struct btrfs_trans_handle *trans;
6410 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6411 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6412 struct inode *new_inode = new_dentry->d_inode;
6413 struct inode *old_inode = old_dentry->d_inode;
6414 struct timespec ctime = CURRENT_TIME;
6419 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6422 /* we only allow rename subvolume link between subvolumes */
6423 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6426 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6427 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6430 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6431 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6434 * we're using rename to replace one file with another.
6435 * and the replacement file is large. Start IO on it now so
6436 * we don't add too much work to the end of the transaction
6438 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6439 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6440 filemap_flush(old_inode->i_mapping);
6442 /* close the racy window with snapshot create/destroy ioctl */
6443 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6444 down_read(&root->fs_info->subvol_sem);
6446 * We want to reserve the absolute worst case amount of items. So if
6447 * both inodes are subvols and we need to unlink them then that would
6448 * require 4 item modifications, but if they are both normal inodes it
6449 * would require 5 item modifications, so we'll assume their normal
6450 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6451 * should cover the worst case number of items we'll modify.
6453 trans = btrfs_start_transaction(root, 20);
6455 return PTR_ERR(trans);
6457 btrfs_set_trans_block_group(trans, new_dir);
6460 btrfs_record_root_in_trans(trans, dest);
6462 ret = btrfs_set_inode_index(new_dir, &index);
6466 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6467 /* force full log commit if subvolume involved. */
6468 root->fs_info->last_trans_log_full_commit = trans->transid;
6470 ret = btrfs_insert_inode_ref(trans, dest,
6471 new_dentry->d_name.name,
6472 new_dentry->d_name.len,
6474 new_dir->i_ino, index);
6478 * this is an ugly little race, but the rename is required
6479 * to make sure that if we crash, the inode is either at the
6480 * old name or the new one. pinning the log transaction lets
6481 * us make sure we don't allow a log commit to come in after
6482 * we unlink the name but before we add the new name back in.
6484 btrfs_pin_log_trans(root);
6487 * make sure the inode gets flushed if it is replacing
6490 if (new_inode && new_inode->i_size &&
6491 old_inode && S_ISREG(old_inode->i_mode)) {
6492 btrfs_add_ordered_operation(trans, root, old_inode);
6495 old_dir->i_ctime = old_dir->i_mtime = ctime;
6496 new_dir->i_ctime = new_dir->i_mtime = ctime;
6497 old_inode->i_ctime = ctime;
6499 if (old_dentry->d_parent != new_dentry->d_parent)
6500 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6502 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6503 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6504 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6505 old_dentry->d_name.name,
6506 old_dentry->d_name.len);
6508 btrfs_inc_nlink(old_dentry->d_inode);
6509 ret = btrfs_unlink_inode(trans, root, old_dir,
6510 old_dentry->d_inode,
6511 old_dentry->d_name.name,
6512 old_dentry->d_name.len);
6517 new_inode->i_ctime = CURRENT_TIME;
6518 if (unlikely(new_inode->i_ino ==
6519 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6520 root_objectid = BTRFS_I(new_inode)->location.objectid;
6521 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6523 new_dentry->d_name.name,
6524 new_dentry->d_name.len);
6525 BUG_ON(new_inode->i_nlink == 0);
6527 ret = btrfs_unlink_inode(trans, dest, new_dir,
6528 new_dentry->d_inode,
6529 new_dentry->d_name.name,
6530 new_dentry->d_name.len);
6533 if (new_inode->i_nlink == 0) {
6534 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6539 ret = btrfs_add_link(trans, new_dir, old_inode,
6540 new_dentry->d_name.name,
6541 new_dentry->d_name.len, 0, index);
6544 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6545 btrfs_log_new_name(trans, old_inode, old_dir,
6546 new_dentry->d_parent);
6547 btrfs_end_log_trans(root);
6550 btrfs_end_transaction_throttle(trans, root);
6552 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6553 up_read(&root->fs_info->subvol_sem);
6559 * some fairly slow code that needs optimization. This walks the list
6560 * of all the inodes with pending delalloc and forces them to disk.
6562 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6564 struct list_head *head = &root->fs_info->delalloc_inodes;
6565 struct btrfs_inode *binode;
6566 struct inode *inode;
6568 if (root->fs_info->sb->s_flags & MS_RDONLY)
6571 spin_lock(&root->fs_info->delalloc_lock);
6572 while (!list_empty(head)) {
6573 binode = list_entry(head->next, struct btrfs_inode,
6575 inode = igrab(&binode->vfs_inode);
6577 list_del_init(&binode->delalloc_inodes);
6578 spin_unlock(&root->fs_info->delalloc_lock);
6580 filemap_flush(inode->i_mapping);
6582 btrfs_add_delayed_iput(inode);
6587 spin_lock(&root->fs_info->delalloc_lock);
6589 spin_unlock(&root->fs_info->delalloc_lock);
6591 /* the filemap_flush will queue IO into the worker threads, but
6592 * we have to make sure the IO is actually started and that
6593 * ordered extents get created before we return
6595 atomic_inc(&root->fs_info->async_submit_draining);
6596 while (atomic_read(&root->fs_info->nr_async_submits) ||
6597 atomic_read(&root->fs_info->async_delalloc_pages)) {
6598 wait_event(root->fs_info->async_submit_wait,
6599 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
6600 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
6602 atomic_dec(&root->fs_info->async_submit_draining);
6606 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput)
6608 struct btrfs_inode *binode;
6609 struct inode *inode = NULL;
6611 spin_lock(&root->fs_info->delalloc_lock);
6612 while (!list_empty(&root->fs_info->delalloc_inodes)) {
6613 binode = list_entry(root->fs_info->delalloc_inodes.next,
6614 struct btrfs_inode, delalloc_inodes);
6615 inode = igrab(&binode->vfs_inode);
6617 list_move_tail(&binode->delalloc_inodes,
6618 &root->fs_info->delalloc_inodes);
6622 list_del_init(&binode->delalloc_inodes);
6623 cond_resched_lock(&root->fs_info->delalloc_lock);
6625 spin_unlock(&root->fs_info->delalloc_lock);
6628 write_inode_now(inode, 0);
6630 btrfs_add_delayed_iput(inode);
6638 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
6639 const char *symname)
6641 struct btrfs_trans_handle *trans;
6642 struct btrfs_root *root = BTRFS_I(dir)->root;
6643 struct btrfs_path *path;
6644 struct btrfs_key key;
6645 struct inode *inode = NULL;
6653 struct btrfs_file_extent_item *ei;
6654 struct extent_buffer *leaf;
6655 unsigned long nr = 0;
6657 name_len = strlen(symname) + 1;
6658 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
6659 return -ENAMETOOLONG;
6661 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
6665 * 2 items for inode item and ref
6666 * 2 items for dir items
6667 * 1 item for xattr if selinux is on
6669 trans = btrfs_start_transaction(root, 5);
6671 return PTR_ERR(trans);
6673 btrfs_set_trans_block_group(trans, dir);
6675 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6677 dentry->d_parent->d_inode->i_ino, objectid,
6678 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
6680 err = PTR_ERR(inode);
6684 err = btrfs_init_inode_security(trans, inode, dir);
6690 btrfs_set_trans_block_group(trans, inode);
6691 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
6695 inode->i_mapping->a_ops = &btrfs_aops;
6696 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6697 inode->i_fop = &btrfs_file_operations;
6698 inode->i_op = &btrfs_file_inode_operations;
6699 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6701 btrfs_update_inode_block_group(trans, inode);
6702 btrfs_update_inode_block_group(trans, dir);
6706 path = btrfs_alloc_path();
6708 key.objectid = inode->i_ino;
6710 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
6711 datasize = btrfs_file_extent_calc_inline_size(name_len);
6712 err = btrfs_insert_empty_item(trans, root, path, &key,
6718 leaf = path->nodes[0];
6719 ei = btrfs_item_ptr(leaf, path->slots[0],
6720 struct btrfs_file_extent_item);
6721 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
6722 btrfs_set_file_extent_type(leaf, ei,
6723 BTRFS_FILE_EXTENT_INLINE);
6724 btrfs_set_file_extent_encryption(leaf, ei, 0);
6725 btrfs_set_file_extent_compression(leaf, ei, 0);
6726 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
6727 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
6729 ptr = btrfs_file_extent_inline_start(ei);
6730 write_extent_buffer(leaf, symname, ptr, name_len);
6731 btrfs_mark_buffer_dirty(leaf);
6732 btrfs_free_path(path);
6734 inode->i_op = &btrfs_symlink_inode_operations;
6735 inode->i_mapping->a_ops = &btrfs_symlink_aops;
6736 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6737 inode_set_bytes(inode, name_len);
6738 btrfs_i_size_write(inode, name_len - 1);
6739 err = btrfs_update_inode(trans, root, inode);
6744 nr = trans->blocks_used;
6745 btrfs_end_transaction_throttle(trans, root);
6747 inode_dec_link_count(inode);
6750 btrfs_btree_balance_dirty(root, nr);
6754 int btrfs_prealloc_file_range(struct inode *inode, int mode,
6755 u64 start, u64 num_bytes, u64 min_size,
6756 loff_t actual_len, u64 *alloc_hint)
6758 struct btrfs_trans_handle *trans;
6759 struct btrfs_root *root = BTRFS_I(inode)->root;
6760 struct btrfs_key ins;
6761 u64 cur_offset = start;
6764 while (num_bytes > 0) {
6765 trans = btrfs_start_transaction(root, 3);
6766 if (IS_ERR(trans)) {
6767 ret = PTR_ERR(trans);
6771 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
6772 0, *alloc_hint, (u64)-1, &ins, 1);
6774 btrfs_end_transaction(trans, root);
6778 ret = insert_reserved_file_extent(trans, inode,
6779 cur_offset, ins.objectid,
6780 ins.offset, ins.offset,
6781 ins.offset, 0, 0, 0,
6782 BTRFS_FILE_EXTENT_PREALLOC);
6784 btrfs_drop_extent_cache(inode, cur_offset,
6785 cur_offset + ins.offset -1, 0);
6787 num_bytes -= ins.offset;
6788 cur_offset += ins.offset;
6789 *alloc_hint = ins.objectid + ins.offset;
6791 inode->i_ctime = CURRENT_TIME;
6792 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
6793 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
6794 (actual_len > inode->i_size) &&
6795 (cur_offset > inode->i_size)) {
6796 if (cur_offset > actual_len)
6797 i_size_write(inode, actual_len);
6799 i_size_write(inode, cur_offset);
6800 i_size_write(inode, cur_offset);
6801 btrfs_ordered_update_i_size(inode, cur_offset, NULL);
6804 ret = btrfs_update_inode(trans, root, inode);
6807 btrfs_end_transaction(trans, root);
6812 static long btrfs_fallocate(struct inode *inode, int mode,
6813 loff_t offset, loff_t len)
6815 struct extent_state *cached_state = NULL;
6822 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
6823 struct extent_map *em;
6826 alloc_start = offset & ~mask;
6827 alloc_end = (offset + len + mask) & ~mask;
6830 * wait for ordered IO before we have any locks. We'll loop again
6831 * below with the locks held.
6833 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
6835 mutex_lock(&inode->i_mutex);
6836 if (alloc_start > inode->i_size) {
6837 ret = btrfs_cont_expand(inode, alloc_start);
6842 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
6846 locked_end = alloc_end - 1;
6848 struct btrfs_ordered_extent *ordered;
6850 /* the extent lock is ordered inside the running
6853 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
6854 locked_end, 0, &cached_state, GFP_NOFS);
6855 ordered = btrfs_lookup_first_ordered_extent(inode,
6858 ordered->file_offset + ordered->len > alloc_start &&
6859 ordered->file_offset < alloc_end) {
6860 btrfs_put_ordered_extent(ordered);
6861 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
6862 alloc_start, locked_end,
6863 &cached_state, GFP_NOFS);
6865 * we can't wait on the range with the transaction
6866 * running or with the extent lock held
6868 btrfs_wait_ordered_range(inode, alloc_start,
6869 alloc_end - alloc_start);
6872 btrfs_put_ordered_extent(ordered);
6877 cur_offset = alloc_start;
6879 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
6880 alloc_end - cur_offset, 0);
6881 BUG_ON(IS_ERR(em) || !em);
6882 last_byte = min(extent_map_end(em), alloc_end);
6883 last_byte = (last_byte + mask) & ~mask;
6884 if (em->block_start == EXTENT_MAP_HOLE ||
6885 (cur_offset >= inode->i_size &&
6886 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6887 ret = btrfs_prealloc_file_range(inode, 0, cur_offset,
6888 last_byte - cur_offset,
6889 1 << inode->i_blkbits,
6893 free_extent_map(em);
6897 free_extent_map(em);
6899 cur_offset = last_byte;
6900 if (cur_offset >= alloc_end) {
6905 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
6906 &cached_state, GFP_NOFS);
6908 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
6910 mutex_unlock(&inode->i_mutex);
6914 static int btrfs_set_page_dirty(struct page *page)
6916 return __set_page_dirty_nobuffers(page);
6919 static int btrfs_permission(struct inode *inode, int mask)
6921 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
6923 return generic_permission(inode, mask, btrfs_check_acl);
6926 static const struct inode_operations btrfs_dir_inode_operations = {
6927 .getattr = btrfs_getattr,
6928 .lookup = btrfs_lookup,
6929 .create = btrfs_create,
6930 .unlink = btrfs_unlink,
6932 .mkdir = btrfs_mkdir,
6933 .rmdir = btrfs_rmdir,
6934 .rename = btrfs_rename,
6935 .symlink = btrfs_symlink,
6936 .setattr = btrfs_setattr,
6937 .mknod = btrfs_mknod,
6938 .setxattr = btrfs_setxattr,
6939 .getxattr = btrfs_getxattr,
6940 .listxattr = btrfs_listxattr,
6941 .removexattr = btrfs_removexattr,
6942 .permission = btrfs_permission,
6944 static const struct inode_operations btrfs_dir_ro_inode_operations = {
6945 .lookup = btrfs_lookup,
6946 .permission = btrfs_permission,
6949 static const struct file_operations btrfs_dir_file_operations = {
6950 .llseek = generic_file_llseek,
6951 .read = generic_read_dir,
6952 .readdir = btrfs_real_readdir,
6953 .unlocked_ioctl = btrfs_ioctl,
6954 #ifdef CONFIG_COMPAT
6955 .compat_ioctl = btrfs_ioctl,
6957 .release = btrfs_release_file,
6958 .fsync = btrfs_sync_file,
6961 static struct extent_io_ops btrfs_extent_io_ops = {
6962 .fill_delalloc = run_delalloc_range,
6963 .submit_bio_hook = btrfs_submit_bio_hook,
6964 .merge_bio_hook = btrfs_merge_bio_hook,
6965 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
6966 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
6967 .writepage_start_hook = btrfs_writepage_start_hook,
6968 .readpage_io_failed_hook = btrfs_io_failed_hook,
6969 .set_bit_hook = btrfs_set_bit_hook,
6970 .clear_bit_hook = btrfs_clear_bit_hook,
6971 .merge_extent_hook = btrfs_merge_extent_hook,
6972 .split_extent_hook = btrfs_split_extent_hook,
6976 * btrfs doesn't support the bmap operation because swapfiles
6977 * use bmap to make a mapping of extents in the file. They assume
6978 * these extents won't change over the life of the file and they
6979 * use the bmap result to do IO directly to the drive.
6981 * the btrfs bmap call would return logical addresses that aren't
6982 * suitable for IO and they also will change frequently as COW
6983 * operations happen. So, swapfile + btrfs == corruption.
6985 * For now we're avoiding this by dropping bmap.
6987 static const struct address_space_operations btrfs_aops = {
6988 .readpage = btrfs_readpage,
6989 .writepage = btrfs_writepage,
6990 .writepages = btrfs_writepages,
6991 .readpages = btrfs_readpages,
6992 .sync_page = block_sync_page,
6993 .direct_IO = btrfs_direct_IO,
6994 .invalidatepage = btrfs_invalidatepage,
6995 .releasepage = btrfs_releasepage,
6996 .set_page_dirty = btrfs_set_page_dirty,
6997 .error_remove_page = generic_error_remove_page,
7000 static const struct address_space_operations btrfs_symlink_aops = {
7001 .readpage = btrfs_readpage,
7002 .writepage = btrfs_writepage,
7003 .invalidatepage = btrfs_invalidatepage,
7004 .releasepage = btrfs_releasepage,
7007 static const struct inode_operations btrfs_file_inode_operations = {
7008 .truncate = btrfs_truncate,
7009 .getattr = btrfs_getattr,
7010 .setattr = btrfs_setattr,
7011 .setxattr = btrfs_setxattr,
7012 .getxattr = btrfs_getxattr,
7013 .listxattr = btrfs_listxattr,
7014 .removexattr = btrfs_removexattr,
7015 .permission = btrfs_permission,
7016 .fallocate = btrfs_fallocate,
7017 .fiemap = btrfs_fiemap,
7019 static const struct inode_operations btrfs_special_inode_operations = {
7020 .getattr = btrfs_getattr,
7021 .setattr = btrfs_setattr,
7022 .permission = btrfs_permission,
7023 .setxattr = btrfs_setxattr,
7024 .getxattr = btrfs_getxattr,
7025 .listxattr = btrfs_listxattr,
7026 .removexattr = btrfs_removexattr,
7028 static const struct inode_operations btrfs_symlink_inode_operations = {
7029 .readlink = generic_readlink,
7030 .follow_link = page_follow_link_light,
7031 .put_link = page_put_link,
7032 .permission = btrfs_permission,
7033 .setxattr = btrfs_setxattr,
7034 .getxattr = btrfs_getxattr,
7035 .listxattr = btrfs_listxattr,
7036 .removexattr = btrfs_removexattr,
7039 const struct dentry_operations btrfs_dentry_operations = {
7040 .d_delete = btrfs_dentry_delete,