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"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96 struct inode *inode, struct inode *dir)
100 err = btrfs_init_acl(trans, inode, dir);
102 err = btrfs_xattr_security_init(trans, inode, dir);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
112 struct btrfs_root *root, struct inode *inode,
113 u64 start, size_t size, size_t compressed_size,
114 struct page **compressed_pages)
116 struct btrfs_key key;
117 struct btrfs_path *path;
118 struct extent_buffer *leaf;
119 struct page *page = NULL;
122 struct btrfs_file_extent_item *ei;
125 size_t cur_size = size;
127 unsigned long offset;
128 int compress_type = BTRFS_COMPRESS_NONE;
130 if (compressed_size && compressed_pages) {
131 compress_type = root->fs_info->compress_type;
132 cur_size = compressed_size;
135 path = btrfs_alloc_path();
139 path->leave_spinning = 1;
140 btrfs_set_trans_block_group(trans, inode);
142 key.objectid = inode->i_ino;
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
165 if (compress_type != BTRFS_COMPRESS_NONE) {
168 while (compressed_size > 0) {
169 cpage = compressed_pages[i];
170 cur_size = min_t(unsigned long, compressed_size,
173 kaddr = kmap_atomic(cpage, KM_USER0);
174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
175 kunmap_atomic(kaddr, KM_USER0);
179 compressed_size -= cur_size;
181 btrfs_set_file_extent_compression(leaf, ei,
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
210 btrfs_free_path(path);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
223 size_t compressed_size,
224 struct page **compressed_pages)
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((u64)root->sectorsize - 1);
232 u64 data_len = inline_len;
236 data_len = compressed_size;
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
242 (actual_end & (root->sectorsize - 1)) == 0) ||
244 data_len > root->fs_info->max_inline) {
248 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
263 struct async_extent {
268 unsigned long nr_pages;
270 struct list_head list;
275 struct btrfs_root *root;
276 struct page *locked_page;
279 struct list_head extents;
280 struct btrfs_work work;
283 static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
287 unsigned long nr_pages,
290 struct async_extent *async_extent;
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
293 BUG_ON(!async_extent);
294 async_extent->start = start;
295 async_extent->ram_size = ram_size;
296 async_extent->compressed_size = compressed_size;
297 async_extent->pages = pages;
298 async_extent->nr_pages = nr_pages;
299 async_extent->compress_type = compress_type;
300 list_add_tail(&async_extent->list, &cow->extents);
305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
320 static noinline int compress_file_range(struct inode *inode,
321 struct page *locked_page,
323 struct async_cow *async_cow,
326 struct btrfs_root *root = BTRFS_I(inode)->root;
327 struct btrfs_trans_handle *trans;
329 u64 blocksize = root->sectorsize;
331 u64 isize = i_size_read(inode);
333 struct page **pages = NULL;
334 unsigned long nr_pages;
335 unsigned long nr_pages_ret = 0;
336 unsigned long total_compressed = 0;
337 unsigned long total_in = 0;
338 unsigned long max_compressed = 128 * 1024;
339 unsigned long max_uncompressed = 128 * 1024;
342 int compress_type = root->fs_info->compress_type;
344 actual_end = min_t(u64, isize, end + 1);
347 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
348 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
351 * we don't want to send crud past the end of i_size through
352 * compression, that's just a waste of CPU time. So, if the
353 * end of the file is before the start of our current
354 * requested range of bytes, we bail out to the uncompressed
355 * cleanup code that can deal with all of this.
357 * It isn't really the fastest way to fix things, but this is a
358 * very uncommon corner.
360 if (actual_end <= start)
361 goto cleanup_and_bail_uncompressed;
363 total_compressed = actual_end - start;
365 /* we want to make sure that amount of ram required to uncompress
366 * an extent is reasonable, so we limit the total size in ram
367 * of a compressed extent to 128k. This is a crucial number
368 * because it also controls how easily we can spread reads across
369 * cpus for decompression.
371 * We also want to make sure the amount of IO required to do
372 * a random read is reasonably small, so we limit the size of
373 * a compressed extent to 128k.
375 total_compressed = min(total_compressed, max_uncompressed);
376 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
377 num_bytes = max(blocksize, num_bytes);
382 * we do compression for mount -o compress and when the
383 * inode has not been flagged as nocompress. This flag can
384 * change at any time if we discover bad compression ratios.
386 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
387 (btrfs_test_opt(root, COMPRESS) ||
388 (BTRFS_I(inode)->force_compress) ||
389 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
391 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
394 if (BTRFS_I(inode)->force_compress)
395 compress_type = BTRFS_I(inode)->force_compress;
397 ret = btrfs_compress_pages(compress_type,
398 inode->i_mapping, start,
399 total_compressed, pages,
400 nr_pages, &nr_pages_ret,
406 unsigned long offset = total_compressed &
407 (PAGE_CACHE_SIZE - 1);
408 struct page *page = pages[nr_pages_ret - 1];
411 /* zero the tail end of the last page, we might be
412 * sending it down to disk
415 kaddr = kmap_atomic(page, KM_USER0);
416 memset(kaddr + offset, 0,
417 PAGE_CACHE_SIZE - offset);
418 kunmap_atomic(kaddr, KM_USER0);
424 trans = btrfs_join_transaction(root, 1);
425 BUG_ON(IS_ERR(trans));
426 btrfs_set_trans_block_group(trans, inode);
427 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
429 /* lets try to make an inline extent */
430 if (ret || total_in < (actual_end - start)) {
431 /* we didn't compress the entire range, try
432 * to make an uncompressed inline extent.
434 ret = cow_file_range_inline(trans, root, inode,
435 start, end, 0, NULL);
437 /* try making a compressed inline extent */
438 ret = cow_file_range_inline(trans, root, inode,
440 total_compressed, pages);
444 * inline extent creation worked, we don't need
445 * to create any more async work items. Unlock
446 * and free up our temp pages.
448 extent_clear_unlock_delalloc(inode,
449 &BTRFS_I(inode)->io_tree,
451 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
452 EXTENT_CLEAR_DELALLOC |
453 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
455 btrfs_end_transaction(trans, root);
458 btrfs_end_transaction(trans, root);
463 * we aren't doing an inline extent round the compressed size
464 * up to a block size boundary so the allocator does sane
467 total_compressed = (total_compressed + blocksize - 1) &
471 * one last check to make sure the compression is really a
472 * win, compare the page count read with the blocks on disk
474 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
475 ~(PAGE_CACHE_SIZE - 1);
476 if (total_compressed >= total_in) {
479 num_bytes = total_in;
482 if (!will_compress && pages) {
484 * the compression code ran but failed to make things smaller,
485 * free any pages it allocated and our page pointer array
487 for (i = 0; i < nr_pages_ret; i++) {
488 WARN_ON(pages[i]->mapping);
489 page_cache_release(pages[i]);
493 total_compressed = 0;
496 /* flag the file so we don't compress in the future */
497 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
498 !(BTRFS_I(inode)->force_compress)) {
499 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
505 /* the async work queues will take care of doing actual
506 * allocation on disk for these compressed pages,
507 * and will submit them to the elevator.
509 add_async_extent(async_cow, start, num_bytes,
510 total_compressed, pages, nr_pages_ret,
513 if (start + num_bytes < end) {
520 cleanup_and_bail_uncompressed:
522 * No compression, but we still need to write the pages in
523 * the file we've been given so far. redirty the locked
524 * page if it corresponds to our extent and set things up
525 * for the async work queue to run cow_file_range to do
526 * the normal delalloc dance
528 if (page_offset(locked_page) >= start &&
529 page_offset(locked_page) <= end) {
530 __set_page_dirty_nobuffers(locked_page);
531 /* unlocked later on in the async handlers */
533 add_async_extent(async_cow, start, end - start + 1,
534 0, NULL, 0, BTRFS_COMPRESS_NONE);
542 for (i = 0; i < nr_pages_ret; i++) {
543 WARN_ON(pages[i]->mapping);
544 page_cache_release(pages[i]);
552 * phase two of compressed writeback. This is the ordered portion
553 * of the code, which only gets called in the order the work was
554 * queued. We walk all the async extents created by compress_file_range
555 * and send them down to the disk.
557 static noinline int submit_compressed_extents(struct inode *inode,
558 struct async_cow *async_cow)
560 struct async_extent *async_extent;
562 struct btrfs_trans_handle *trans;
563 struct btrfs_key ins;
564 struct extent_map *em;
565 struct btrfs_root *root = BTRFS_I(inode)->root;
566 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
567 struct extent_io_tree *io_tree;
570 if (list_empty(&async_cow->extents))
574 while (!list_empty(&async_cow->extents)) {
575 async_extent = list_entry(async_cow->extents.next,
576 struct async_extent, list);
577 list_del(&async_extent->list);
579 io_tree = &BTRFS_I(inode)->io_tree;
582 /* did the compression code fall back to uncompressed IO? */
583 if (!async_extent->pages) {
584 int page_started = 0;
585 unsigned long nr_written = 0;
587 lock_extent(io_tree, async_extent->start,
588 async_extent->start +
589 async_extent->ram_size - 1, GFP_NOFS);
591 /* allocate blocks */
592 ret = cow_file_range(inode, async_cow->locked_page,
594 async_extent->start +
595 async_extent->ram_size - 1,
596 &page_started, &nr_written, 0);
599 * if page_started, cow_file_range inserted an
600 * inline extent and took care of all the unlocking
601 * and IO for us. Otherwise, we need to submit
602 * all those pages down to the drive.
604 if (!page_started && !ret)
605 extent_write_locked_range(io_tree,
606 inode, async_extent->start,
607 async_extent->start +
608 async_extent->ram_size - 1,
616 lock_extent(io_tree, async_extent->start,
617 async_extent->start + async_extent->ram_size - 1,
620 trans = btrfs_join_transaction(root, 1);
621 BUG_ON(IS_ERR(trans));
622 ret = btrfs_reserve_extent(trans, root,
623 async_extent->compressed_size,
624 async_extent->compressed_size,
627 btrfs_end_transaction(trans, root);
631 for (i = 0; i < async_extent->nr_pages; i++) {
632 WARN_ON(async_extent->pages[i]->mapping);
633 page_cache_release(async_extent->pages[i]);
635 kfree(async_extent->pages);
636 async_extent->nr_pages = 0;
637 async_extent->pages = NULL;
638 unlock_extent(io_tree, async_extent->start,
639 async_extent->start +
640 async_extent->ram_size - 1, GFP_NOFS);
645 * here we're doing allocation and writeback of the
648 btrfs_drop_extent_cache(inode, async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1, 0);
652 em = alloc_extent_map(GFP_NOFS);
654 em->start = async_extent->start;
655 em->len = async_extent->ram_size;
656 em->orig_start = em->start;
658 em->block_start = ins.objectid;
659 em->block_len = ins.offset;
660 em->bdev = root->fs_info->fs_devices->latest_bdev;
661 em->compress_type = async_extent->compress_type;
662 set_bit(EXTENT_FLAG_PINNED, &em->flags);
663 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
666 write_lock(&em_tree->lock);
667 ret = add_extent_mapping(em_tree, em);
668 write_unlock(&em_tree->lock);
669 if (ret != -EEXIST) {
673 btrfs_drop_extent_cache(inode, async_extent->start,
674 async_extent->start +
675 async_extent->ram_size - 1, 0);
678 ret = btrfs_add_ordered_extent_compress(inode,
681 async_extent->ram_size,
683 BTRFS_ORDERED_COMPRESSED,
684 async_extent->compress_type);
688 * clear dirty, set writeback and unlock the pages.
690 extent_clear_unlock_delalloc(inode,
691 &BTRFS_I(inode)->io_tree,
693 async_extent->start +
694 async_extent->ram_size - 1,
695 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
696 EXTENT_CLEAR_UNLOCK |
697 EXTENT_CLEAR_DELALLOC |
698 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
700 ret = btrfs_submit_compressed_write(inode,
702 async_extent->ram_size,
704 ins.offset, async_extent->pages,
705 async_extent->nr_pages);
708 alloc_hint = ins.objectid + ins.offset;
716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
719 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
720 struct extent_map *em;
723 read_lock(&em_tree->lock);
724 em = search_extent_mapping(em_tree, start, num_bytes);
727 * if block start isn't an actual block number then find the
728 * first block in this inode and use that as a hint. If that
729 * block is also bogus then just don't worry about it.
731 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
733 em = search_extent_mapping(em_tree, 0, 0);
734 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
735 alloc_hint = em->block_start;
739 alloc_hint = em->block_start;
743 read_unlock(&em_tree->lock);
749 * when extent_io.c finds a delayed allocation range in the file,
750 * the call backs end up in this code. The basic idea is to
751 * allocate extents on disk for the range, and create ordered data structs
752 * in ram to track those extents.
754 * locked_page is the page that writepage had locked already. We use
755 * it to make sure we don't do extra locks or unlocks.
757 * *page_started is set to one if we unlock locked_page and do everything
758 * required to start IO on it. It may be clean and already done with
761 static noinline int cow_file_range(struct inode *inode,
762 struct page *locked_page,
763 u64 start, u64 end, int *page_started,
764 unsigned long *nr_written,
767 struct btrfs_root *root = BTRFS_I(inode)->root;
768 struct btrfs_trans_handle *trans;
771 unsigned long ram_size;
774 u64 blocksize = root->sectorsize;
775 struct btrfs_key ins;
776 struct extent_map *em;
777 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
780 BUG_ON(root == root->fs_info->tree_root);
781 trans = btrfs_join_transaction(root, 1);
782 BUG_ON(IS_ERR(trans));
783 btrfs_set_trans_block_group(trans, inode);
784 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
786 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
787 num_bytes = max(blocksize, num_bytes);
788 disk_num_bytes = num_bytes;
792 /* lets try to make an inline extent */
793 ret = cow_file_range_inline(trans, root, inode,
794 start, end, 0, NULL);
796 extent_clear_unlock_delalloc(inode,
797 &BTRFS_I(inode)->io_tree,
799 EXTENT_CLEAR_UNLOCK_PAGE |
800 EXTENT_CLEAR_UNLOCK |
801 EXTENT_CLEAR_DELALLOC |
803 EXTENT_SET_WRITEBACK |
804 EXTENT_END_WRITEBACK);
806 *nr_written = *nr_written +
807 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
814 BUG_ON(disk_num_bytes >
815 btrfs_super_total_bytes(&root->fs_info->super_copy));
817 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
818 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
820 while (disk_num_bytes > 0) {
823 cur_alloc_size = disk_num_bytes;
824 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
825 root->sectorsize, 0, alloc_hint,
829 em = alloc_extent_map(GFP_NOFS);
832 em->orig_start = em->start;
833 ram_size = ins.offset;
834 em->len = ins.offset;
836 em->block_start = ins.objectid;
837 em->block_len = ins.offset;
838 em->bdev = root->fs_info->fs_devices->latest_bdev;
839 set_bit(EXTENT_FLAG_PINNED, &em->flags);
842 write_lock(&em_tree->lock);
843 ret = add_extent_mapping(em_tree, em);
844 write_unlock(&em_tree->lock);
845 if (ret != -EEXIST) {
849 btrfs_drop_extent_cache(inode, start,
850 start + ram_size - 1, 0);
853 cur_alloc_size = ins.offset;
854 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
855 ram_size, cur_alloc_size, 0);
858 if (root->root_key.objectid ==
859 BTRFS_DATA_RELOC_TREE_OBJECTID) {
860 ret = btrfs_reloc_clone_csums(inode, start,
865 if (disk_num_bytes < cur_alloc_size)
868 /* we're not doing compressed IO, don't unlock the first
869 * page (which the caller expects to stay locked), don't
870 * clear any dirty bits and don't set any writeback bits
872 * Do set the Private2 bit so we know this page was properly
873 * setup for writepage
875 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
876 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
879 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
880 start, start + ram_size - 1,
882 disk_num_bytes -= cur_alloc_size;
883 num_bytes -= cur_alloc_size;
884 alloc_hint = ins.objectid + ins.offset;
885 start += cur_alloc_size;
889 btrfs_end_transaction(trans, root);
895 * work queue call back to started compression on a file and pages
897 static noinline void async_cow_start(struct btrfs_work *work)
899 struct async_cow *async_cow;
901 async_cow = container_of(work, struct async_cow, work);
903 compress_file_range(async_cow->inode, async_cow->locked_page,
904 async_cow->start, async_cow->end, async_cow,
907 async_cow->inode = NULL;
911 * work queue call back to submit previously compressed pages
913 static noinline void async_cow_submit(struct btrfs_work *work)
915 struct async_cow *async_cow;
916 struct btrfs_root *root;
917 unsigned long nr_pages;
919 async_cow = container_of(work, struct async_cow, work);
921 root = async_cow->root;
922 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
925 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
927 if (atomic_read(&root->fs_info->async_delalloc_pages) <
929 waitqueue_active(&root->fs_info->async_submit_wait))
930 wake_up(&root->fs_info->async_submit_wait);
932 if (async_cow->inode)
933 submit_compressed_extents(async_cow->inode, async_cow);
936 static noinline void async_cow_free(struct btrfs_work *work)
938 struct async_cow *async_cow;
939 async_cow = container_of(work, struct async_cow, work);
943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
944 u64 start, u64 end, int *page_started,
945 unsigned long *nr_written)
947 struct async_cow *async_cow;
948 struct btrfs_root *root = BTRFS_I(inode)->root;
949 unsigned long nr_pages;
951 int limit = 10 * 1024 * 1042;
953 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
954 1, 0, NULL, GFP_NOFS);
955 while (start < end) {
956 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
957 async_cow->inode = inode;
958 async_cow->root = root;
959 async_cow->locked_page = locked_page;
960 async_cow->start = start;
962 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
965 cur_end = min(end, start + 512 * 1024 - 1);
967 async_cow->end = cur_end;
968 INIT_LIST_HEAD(&async_cow->extents);
970 async_cow->work.func = async_cow_start;
971 async_cow->work.ordered_func = async_cow_submit;
972 async_cow->work.ordered_free = async_cow_free;
973 async_cow->work.flags = 0;
975 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
977 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
979 btrfs_queue_worker(&root->fs_info->delalloc_workers,
982 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
983 wait_event(root->fs_info->async_submit_wait,
984 (atomic_read(&root->fs_info->async_delalloc_pages) <
988 while (atomic_read(&root->fs_info->async_submit_draining) &&
989 atomic_read(&root->fs_info->async_delalloc_pages)) {
990 wait_event(root->fs_info->async_submit_wait,
991 (atomic_read(&root->fs_info->async_delalloc_pages) ==
995 *nr_written += nr_pages;
1002 static noinline int csum_exist_in_range(struct btrfs_root *root,
1003 u64 bytenr, u64 num_bytes)
1006 struct btrfs_ordered_sum *sums;
1009 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1010 bytenr + num_bytes - 1, &list);
1011 if (ret == 0 && list_empty(&list))
1014 while (!list_empty(&list)) {
1015 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1016 list_del(&sums->list);
1023 * when nowcow writeback call back. This checks for snapshots or COW copies
1024 * of the extents that exist in the file, and COWs the file as required.
1026 * If no cow copies or snapshots exist, we write directly to the existing
1029 static noinline int run_delalloc_nocow(struct inode *inode,
1030 struct page *locked_page,
1031 u64 start, u64 end, int *page_started, int force,
1032 unsigned long *nr_written)
1034 struct btrfs_root *root = BTRFS_I(inode)->root;
1035 struct btrfs_trans_handle *trans;
1036 struct extent_buffer *leaf;
1037 struct btrfs_path *path;
1038 struct btrfs_file_extent_item *fi;
1039 struct btrfs_key found_key;
1051 bool nolock = false;
1053 path = btrfs_alloc_path();
1055 if (root == root->fs_info->tree_root) {
1057 trans = btrfs_join_transaction_nolock(root, 1);
1059 trans = btrfs_join_transaction(root, 1);
1061 BUG_ON(IS_ERR(trans));
1063 cow_start = (u64)-1;
1066 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1069 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1070 leaf = path->nodes[0];
1071 btrfs_item_key_to_cpu(leaf, &found_key,
1072 path->slots[0] - 1);
1073 if (found_key.objectid == inode->i_ino &&
1074 found_key.type == BTRFS_EXTENT_DATA_KEY)
1079 leaf = path->nodes[0];
1080 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081 ret = btrfs_next_leaf(root, path);
1086 leaf = path->nodes[0];
1092 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094 if (found_key.objectid > inode->i_ino ||
1095 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1096 found_key.offset > end)
1099 if (found_key.offset > cur_offset) {
1100 extent_end = found_key.offset;
1105 fi = btrfs_item_ptr(leaf, path->slots[0],
1106 struct btrfs_file_extent_item);
1107 extent_type = btrfs_file_extent_type(leaf, fi);
1109 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1110 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1111 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1112 extent_offset = btrfs_file_extent_offset(leaf, fi);
1113 extent_end = found_key.offset +
1114 btrfs_file_extent_num_bytes(leaf, fi);
1115 if (extent_end <= start) {
1119 if (disk_bytenr == 0)
1121 if (btrfs_file_extent_compression(leaf, fi) ||
1122 btrfs_file_extent_encryption(leaf, fi) ||
1123 btrfs_file_extent_other_encoding(leaf, fi))
1125 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127 if (btrfs_extent_readonly(root, disk_bytenr))
1129 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1131 extent_offset, disk_bytenr))
1133 disk_bytenr += extent_offset;
1134 disk_bytenr += cur_offset - found_key.offset;
1135 num_bytes = min(end + 1, extent_end) - cur_offset;
1137 * force cow if csum exists in the range.
1138 * this ensure that csum for a given extent are
1139 * either valid or do not exist.
1141 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1144 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1145 extent_end = found_key.offset +
1146 btrfs_file_extent_inline_len(leaf, fi);
1147 extent_end = ALIGN(extent_end, root->sectorsize);
1152 if (extent_end <= start) {
1157 if (cow_start == (u64)-1)
1158 cow_start = cur_offset;
1159 cur_offset = extent_end;
1160 if (cur_offset > end)
1166 btrfs_release_path(root, path);
1167 if (cow_start != (u64)-1) {
1168 ret = cow_file_range(inode, locked_page, cow_start,
1169 found_key.offset - 1, page_started,
1172 cow_start = (u64)-1;
1175 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1176 struct extent_map *em;
1177 struct extent_map_tree *em_tree;
1178 em_tree = &BTRFS_I(inode)->extent_tree;
1179 em = alloc_extent_map(GFP_NOFS);
1181 em->start = cur_offset;
1182 em->orig_start = em->start;
1183 em->len = num_bytes;
1184 em->block_len = num_bytes;
1185 em->block_start = disk_bytenr;
1186 em->bdev = root->fs_info->fs_devices->latest_bdev;
1187 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189 write_lock(&em_tree->lock);
1190 ret = add_extent_mapping(em_tree, em);
1191 write_unlock(&em_tree->lock);
1192 if (ret != -EEXIST) {
1193 free_extent_map(em);
1196 btrfs_drop_extent_cache(inode, em->start,
1197 em->start + em->len - 1, 0);
1199 type = BTRFS_ORDERED_PREALLOC;
1201 type = BTRFS_ORDERED_NOCOW;
1204 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1205 num_bytes, num_bytes, type);
1208 if (root->root_key.objectid ==
1209 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1210 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1215 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1216 cur_offset, cur_offset + num_bytes - 1,
1217 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1218 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1219 EXTENT_SET_PRIVATE2);
1220 cur_offset = extent_end;
1221 if (cur_offset > end)
1224 btrfs_release_path(root, path);
1226 if (cur_offset <= end && cow_start == (u64)-1)
1227 cow_start = cur_offset;
1228 if (cow_start != (u64)-1) {
1229 ret = cow_file_range(inode, locked_page, cow_start, end,
1230 page_started, nr_written, 1);
1235 ret = btrfs_end_transaction_nolock(trans, root);
1238 ret = btrfs_end_transaction(trans, root);
1241 btrfs_free_path(path);
1246 * extent_io.c call back to do delayed allocation processing
1248 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1249 u64 start, u64 end, int *page_started,
1250 unsigned long *nr_written)
1253 struct btrfs_root *root = BTRFS_I(inode)->root;
1255 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1256 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257 page_started, 1, nr_written);
1258 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1259 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260 page_started, 0, nr_written);
1261 else if (!btrfs_test_opt(root, COMPRESS) &&
1262 !(BTRFS_I(inode)->force_compress) &&
1263 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1264 ret = cow_file_range(inode, locked_page, start, end,
1265 page_started, nr_written, 1);
1267 ret = cow_file_range_async(inode, locked_page, start, end,
1268 page_started, nr_written);
1272 static int btrfs_split_extent_hook(struct inode *inode,
1273 struct extent_state *orig, u64 split)
1275 /* not delalloc, ignore it */
1276 if (!(orig->state & EXTENT_DELALLOC))
1279 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1284 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285 * extents so we can keep track of new extents that are just merged onto old
1286 * extents, such as when we are doing sequential writes, so we can properly
1287 * account for the metadata space we'll need.
1289 static int btrfs_merge_extent_hook(struct inode *inode,
1290 struct extent_state *new,
1291 struct extent_state *other)
1293 /* not delalloc, ignore it */
1294 if (!(other->state & EXTENT_DELALLOC))
1297 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1302 * extent_io.c set_bit_hook, used to track delayed allocation
1303 * bytes in this file, and to maintain the list of inodes that
1304 * have pending delalloc work to be done.
1306 static int btrfs_set_bit_hook(struct inode *inode,
1307 struct extent_state *state, int *bits)
1311 * set_bit and clear bit hooks normally require _irqsave/restore
1312 * but in this case, we are only testeing for the DELALLOC
1313 * bit, which is only set or cleared with irqs on
1315 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1316 struct btrfs_root *root = BTRFS_I(inode)->root;
1317 u64 len = state->end + 1 - state->start;
1318 int do_list = (root->root_key.objectid !=
1319 BTRFS_ROOT_TREE_OBJECTID);
1321 if (*bits & EXTENT_FIRST_DELALLOC)
1322 *bits &= ~EXTENT_FIRST_DELALLOC;
1324 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326 spin_lock(&root->fs_info->delalloc_lock);
1327 BTRFS_I(inode)->delalloc_bytes += len;
1328 root->fs_info->delalloc_bytes += len;
1329 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1330 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1331 &root->fs_info->delalloc_inodes);
1333 spin_unlock(&root->fs_info->delalloc_lock);
1339 * extent_io.c clear_bit_hook, see set_bit_hook for why
1341 static int btrfs_clear_bit_hook(struct inode *inode,
1342 struct extent_state *state, int *bits)
1345 * set_bit and clear bit hooks normally require _irqsave/restore
1346 * but in this case, we are only testeing for the DELALLOC
1347 * bit, which is only set or cleared with irqs on
1349 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1350 struct btrfs_root *root = BTRFS_I(inode)->root;
1351 u64 len = state->end + 1 - state->start;
1352 int do_list = (root->root_key.objectid !=
1353 BTRFS_ROOT_TREE_OBJECTID);
1355 if (*bits & EXTENT_FIRST_DELALLOC)
1356 *bits &= ~EXTENT_FIRST_DELALLOC;
1357 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1358 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360 if (*bits & EXTENT_DO_ACCOUNTING)
1361 btrfs_delalloc_release_metadata(inode, len);
1363 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365 btrfs_free_reserved_data_space(inode, len);
1367 spin_lock(&root->fs_info->delalloc_lock);
1368 root->fs_info->delalloc_bytes -= len;
1369 BTRFS_I(inode)->delalloc_bytes -= len;
1371 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1372 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1373 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375 spin_unlock(&root->fs_info->delalloc_lock);
1381 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382 * we don't create bios that span stripes or chunks
1384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1385 size_t size, struct bio *bio,
1386 unsigned long bio_flags)
1388 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1389 struct btrfs_mapping_tree *map_tree;
1390 u64 logical = (u64)bio->bi_sector << 9;
1395 if (bio_flags & EXTENT_BIO_COMPRESSED)
1398 length = bio->bi_size;
1399 map_tree = &root->fs_info->mapping_tree;
1400 map_length = length;
1401 ret = btrfs_map_block(map_tree, READ, logical,
1402 &map_length, NULL, 0);
1404 if (map_length < length + size)
1410 * in order to insert checksums into the metadata in large chunks,
1411 * we wait until bio submission time. All the pages in the bio are
1412 * checksummed and sums are attached onto the ordered extent record.
1414 * At IO completion time the cums attached on the ordered extent record
1415 * are inserted into the btree
1417 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1418 struct bio *bio, int mirror_num,
1419 unsigned long bio_flags,
1422 struct btrfs_root *root = BTRFS_I(inode)->root;
1425 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1431 * in order to insert checksums into the metadata in large chunks,
1432 * we wait until bio submission time. All the pages in the bio are
1433 * checksummed and sums are attached onto the ordered extent record.
1435 * At IO completion time the cums attached on the ordered extent record
1436 * are inserted into the btree
1438 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1439 int mirror_num, unsigned long bio_flags,
1442 struct btrfs_root *root = BTRFS_I(inode)->root;
1443 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1447 * extent_io.c submission hook. This does the right thing for csum calculation
1448 * on write, or reading the csums from the tree before a read
1450 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1451 int mirror_num, unsigned long bio_flags,
1454 struct btrfs_root *root = BTRFS_I(inode)->root;
1458 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460 if (root == root->fs_info->tree_root)
1461 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1466 if (!(rw & REQ_WRITE)) {
1467 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1468 return btrfs_submit_compressed_read(inode, bio,
1469 mirror_num, bio_flags);
1470 } else if (!skip_sum)
1471 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1473 } else if (!skip_sum) {
1474 /* csum items have already been cloned */
1475 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1477 /* we're doing a write, do the async checksumming */
1478 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1479 inode, rw, bio, mirror_num,
1480 bio_flags, bio_offset,
1481 __btrfs_submit_bio_start,
1482 __btrfs_submit_bio_done);
1486 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1490 * given a list of ordered sums record them in the inode. This happens
1491 * at IO completion time based on sums calculated at bio submission time.
1493 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1494 struct inode *inode, u64 file_offset,
1495 struct list_head *list)
1497 struct btrfs_ordered_sum *sum;
1499 btrfs_set_trans_block_group(trans, inode);
1501 list_for_each_entry(sum, list, list) {
1502 btrfs_csum_file_blocks(trans,
1503 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1508 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1509 struct extent_state **cached_state)
1511 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1513 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1514 cached_state, GFP_NOFS);
1517 /* see btrfs_writepage_start_hook for details on why this is required */
1518 struct btrfs_writepage_fixup {
1520 struct btrfs_work work;
1523 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1525 struct btrfs_writepage_fixup *fixup;
1526 struct btrfs_ordered_extent *ordered;
1527 struct extent_state *cached_state = NULL;
1529 struct inode *inode;
1533 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1537 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1538 ClearPageChecked(page);
1542 inode = page->mapping->host;
1543 page_start = page_offset(page);
1544 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1546 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1547 &cached_state, GFP_NOFS);
1549 /* already ordered? We're done */
1550 if (PagePrivate2(page))
1553 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1555 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1556 page_end, &cached_state, GFP_NOFS);
1558 btrfs_start_ordered_extent(inode, ordered, 1);
1563 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1564 ClearPageChecked(page);
1566 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1567 &cached_state, GFP_NOFS);
1570 page_cache_release(page);
1575 * There are a few paths in the higher layers of the kernel that directly
1576 * set the page dirty bit without asking the filesystem if it is a
1577 * good idea. This causes problems because we want to make sure COW
1578 * properly happens and the data=ordered rules are followed.
1580 * In our case any range that doesn't have the ORDERED bit set
1581 * hasn't been properly setup for IO. We kick off an async process
1582 * to fix it up. The async helper will wait for ordered extents, set
1583 * the delalloc bit and make it safe to write the page.
1585 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1587 struct inode *inode = page->mapping->host;
1588 struct btrfs_writepage_fixup *fixup;
1589 struct btrfs_root *root = BTRFS_I(inode)->root;
1591 /* this page is properly in the ordered list */
1592 if (TestClearPagePrivate2(page))
1595 if (PageChecked(page))
1598 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1602 SetPageChecked(page);
1603 page_cache_get(page);
1604 fixup->work.func = btrfs_writepage_fixup_worker;
1606 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1610 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1611 struct inode *inode, u64 file_pos,
1612 u64 disk_bytenr, u64 disk_num_bytes,
1613 u64 num_bytes, u64 ram_bytes,
1614 u8 compression, u8 encryption,
1615 u16 other_encoding, int extent_type)
1617 struct btrfs_root *root = BTRFS_I(inode)->root;
1618 struct btrfs_file_extent_item *fi;
1619 struct btrfs_path *path;
1620 struct extent_buffer *leaf;
1621 struct btrfs_key ins;
1625 path = btrfs_alloc_path();
1628 path->leave_spinning = 1;
1631 * we may be replacing one extent in the tree with another.
1632 * The new extent is pinned in the extent map, and we don't want
1633 * to drop it from the cache until it is completely in the btree.
1635 * So, tell btrfs_drop_extents to leave this extent in the cache.
1636 * the caller is expected to unpin it and allow it to be merged
1639 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1643 ins.objectid = inode->i_ino;
1644 ins.offset = file_pos;
1645 ins.type = BTRFS_EXTENT_DATA_KEY;
1646 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1648 leaf = path->nodes[0];
1649 fi = btrfs_item_ptr(leaf, path->slots[0],
1650 struct btrfs_file_extent_item);
1651 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1652 btrfs_set_file_extent_type(leaf, fi, extent_type);
1653 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1654 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1655 btrfs_set_file_extent_offset(leaf, fi, 0);
1656 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1657 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1658 btrfs_set_file_extent_compression(leaf, fi, compression);
1659 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1660 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1662 btrfs_unlock_up_safe(path, 1);
1663 btrfs_set_lock_blocking(leaf);
1665 btrfs_mark_buffer_dirty(leaf);
1667 inode_add_bytes(inode, num_bytes);
1669 ins.objectid = disk_bytenr;
1670 ins.offset = disk_num_bytes;
1671 ins.type = BTRFS_EXTENT_ITEM_KEY;
1672 ret = btrfs_alloc_reserved_file_extent(trans, root,
1673 root->root_key.objectid,
1674 inode->i_ino, file_pos, &ins);
1676 btrfs_free_path(path);
1682 * helper function for btrfs_finish_ordered_io, this
1683 * just reads in some of the csum leaves to prime them into ram
1684 * before we start the transaction. It limits the amount of btree
1685 * reads required while inside the transaction.
1687 /* as ordered data IO finishes, this gets called so we can finish
1688 * an ordered extent if the range of bytes in the file it covers are
1691 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1693 struct btrfs_root *root = BTRFS_I(inode)->root;
1694 struct btrfs_trans_handle *trans = NULL;
1695 struct btrfs_ordered_extent *ordered_extent = NULL;
1696 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1697 struct extent_state *cached_state = NULL;
1698 int compress_type = 0;
1700 bool nolock = false;
1702 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1706 BUG_ON(!ordered_extent);
1708 nolock = (root == root->fs_info->tree_root);
1710 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1711 BUG_ON(!list_empty(&ordered_extent->list));
1712 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1715 trans = btrfs_join_transaction_nolock(root, 1);
1717 trans = btrfs_join_transaction(root, 1);
1718 BUG_ON(IS_ERR(trans));
1719 btrfs_set_trans_block_group(trans, inode);
1720 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1721 ret = btrfs_update_inode(trans, root, inode);
1727 lock_extent_bits(io_tree, ordered_extent->file_offset,
1728 ordered_extent->file_offset + ordered_extent->len - 1,
1729 0, &cached_state, GFP_NOFS);
1732 trans = btrfs_join_transaction_nolock(root, 1);
1734 trans = btrfs_join_transaction(root, 1);
1735 BUG_ON(IS_ERR(trans));
1736 btrfs_set_trans_block_group(trans, inode);
1737 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1739 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1740 compress_type = ordered_extent->compress_type;
1741 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1742 BUG_ON(compress_type);
1743 ret = btrfs_mark_extent_written(trans, inode,
1744 ordered_extent->file_offset,
1745 ordered_extent->file_offset +
1746 ordered_extent->len);
1749 BUG_ON(root == root->fs_info->tree_root);
1750 ret = insert_reserved_file_extent(trans, inode,
1751 ordered_extent->file_offset,
1752 ordered_extent->start,
1753 ordered_extent->disk_len,
1754 ordered_extent->len,
1755 ordered_extent->len,
1756 compress_type, 0, 0,
1757 BTRFS_FILE_EXTENT_REG);
1758 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1759 ordered_extent->file_offset,
1760 ordered_extent->len);
1763 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1764 ordered_extent->file_offset +
1765 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1767 add_pending_csums(trans, inode, ordered_extent->file_offset,
1768 &ordered_extent->list);
1770 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1771 ret = btrfs_update_inode(trans, root, inode);
1776 btrfs_end_transaction_nolock(trans, root);
1778 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1780 btrfs_end_transaction(trans, root);
1784 btrfs_put_ordered_extent(ordered_extent);
1785 /* once for the tree */
1786 btrfs_put_ordered_extent(ordered_extent);
1791 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1792 struct extent_state *state, int uptodate)
1794 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1796 ClearPagePrivate2(page);
1797 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1801 * When IO fails, either with EIO or csum verification fails, we
1802 * try other mirrors that might have a good copy of the data. This
1803 * io_failure_record is used to record state as we go through all the
1804 * mirrors. If another mirror has good data, the page is set up to date
1805 * and things continue. If a good mirror can't be found, the original
1806 * bio end_io callback is called to indicate things have failed.
1808 struct io_failure_record {
1813 unsigned long bio_flags;
1817 static int btrfs_io_failed_hook(struct bio *failed_bio,
1818 struct page *page, u64 start, u64 end,
1819 struct extent_state *state)
1821 struct io_failure_record *failrec = NULL;
1823 struct extent_map *em;
1824 struct inode *inode = page->mapping->host;
1825 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1826 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1833 ret = get_state_private(failure_tree, start, &private);
1835 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1838 failrec->start = start;
1839 failrec->len = end - start + 1;
1840 failrec->last_mirror = 0;
1841 failrec->bio_flags = 0;
1843 read_lock(&em_tree->lock);
1844 em = lookup_extent_mapping(em_tree, start, failrec->len);
1845 if (em->start > start || em->start + em->len < start) {
1846 free_extent_map(em);
1849 read_unlock(&em_tree->lock);
1851 if (!em || IS_ERR(em)) {
1855 logical = start - em->start;
1856 logical = em->block_start + logical;
1857 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1858 logical = em->block_start;
1859 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1860 extent_set_compress_type(&failrec->bio_flags,
1863 failrec->logical = logical;
1864 free_extent_map(em);
1865 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1866 EXTENT_DIRTY, GFP_NOFS);
1867 set_state_private(failure_tree, start,
1868 (u64)(unsigned long)failrec);
1870 failrec = (struct io_failure_record *)(unsigned long)private;
1872 num_copies = btrfs_num_copies(
1873 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1874 failrec->logical, failrec->len);
1875 failrec->last_mirror++;
1877 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1878 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1881 if (state && state->start != failrec->start)
1883 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1885 if (!state || failrec->last_mirror > num_copies) {
1886 set_state_private(failure_tree, failrec->start, 0);
1887 clear_extent_bits(failure_tree, failrec->start,
1888 failrec->start + failrec->len - 1,
1889 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1893 bio = bio_alloc(GFP_NOFS, 1);
1894 bio->bi_private = state;
1895 bio->bi_end_io = failed_bio->bi_end_io;
1896 bio->bi_sector = failrec->logical >> 9;
1897 bio->bi_bdev = failed_bio->bi_bdev;
1900 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1901 if (failed_bio->bi_rw & REQ_WRITE)
1906 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1907 failrec->last_mirror,
1908 failrec->bio_flags, 0);
1913 * each time an IO finishes, we do a fast check in the IO failure tree
1914 * to see if we need to process or clean up an io_failure_record
1916 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1919 u64 private_failure;
1920 struct io_failure_record *failure;
1924 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1925 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1926 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1927 start, &private_failure);
1929 failure = (struct io_failure_record *)(unsigned long)
1931 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1933 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1935 failure->start + failure->len - 1,
1936 EXTENT_DIRTY | EXTENT_LOCKED,
1945 * when reads are done, we need to check csums to verify the data is correct
1946 * if there's a match, we allow the bio to finish. If not, we go through
1947 * the io_failure_record routines to find good copies
1949 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1950 struct extent_state *state)
1952 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1953 struct inode *inode = page->mapping->host;
1954 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1956 u64 private = ~(u32)0;
1958 struct btrfs_root *root = BTRFS_I(inode)->root;
1961 if (PageChecked(page)) {
1962 ClearPageChecked(page);
1966 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1969 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1970 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1971 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1976 if (state && state->start == start) {
1977 private = state->private;
1980 ret = get_state_private(io_tree, start, &private);
1982 kaddr = kmap_atomic(page, KM_USER0);
1986 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1987 btrfs_csum_final(csum, (char *)&csum);
1988 if (csum != private)
1991 kunmap_atomic(kaddr, KM_USER0);
1993 /* if the io failure tree for this inode is non-empty,
1994 * check to see if we've recovered from a failed IO
1996 btrfs_clean_io_failures(inode, start);
2000 if (printk_ratelimit()) {
2001 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2002 "private %llu\n", page->mapping->host->i_ino,
2003 (unsigned long long)start, csum,
2004 (unsigned long long)private);
2006 memset(kaddr + offset, 1, end - start + 1);
2007 flush_dcache_page(page);
2008 kunmap_atomic(kaddr, KM_USER0);
2014 struct delayed_iput {
2015 struct list_head list;
2016 struct inode *inode;
2019 void btrfs_add_delayed_iput(struct inode *inode)
2021 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2022 struct delayed_iput *delayed;
2024 if (atomic_add_unless(&inode->i_count, -1, 1))
2027 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2028 delayed->inode = inode;
2030 spin_lock(&fs_info->delayed_iput_lock);
2031 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2032 spin_unlock(&fs_info->delayed_iput_lock);
2035 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2038 struct btrfs_fs_info *fs_info = root->fs_info;
2039 struct delayed_iput *delayed;
2042 spin_lock(&fs_info->delayed_iput_lock);
2043 empty = list_empty(&fs_info->delayed_iputs);
2044 spin_unlock(&fs_info->delayed_iput_lock);
2048 down_read(&root->fs_info->cleanup_work_sem);
2049 spin_lock(&fs_info->delayed_iput_lock);
2050 list_splice_init(&fs_info->delayed_iputs, &list);
2051 spin_unlock(&fs_info->delayed_iput_lock);
2053 while (!list_empty(&list)) {
2054 delayed = list_entry(list.next, struct delayed_iput, list);
2055 list_del(&delayed->list);
2056 iput(delayed->inode);
2059 up_read(&root->fs_info->cleanup_work_sem);
2063 * calculate extra metadata reservation when snapshotting a subvolume
2064 * contains orphan files.
2066 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2067 struct btrfs_pending_snapshot *pending,
2068 u64 *bytes_to_reserve)
2070 struct btrfs_root *root;
2071 struct btrfs_block_rsv *block_rsv;
2075 root = pending->root;
2076 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2079 block_rsv = root->orphan_block_rsv;
2081 /* orphan block reservation for the snapshot */
2082 num_bytes = block_rsv->size;
2085 * after the snapshot is created, COWing tree blocks may use more
2086 * space than it frees. So we should make sure there is enough
2089 index = trans->transid & 0x1;
2090 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2091 num_bytes += block_rsv->size -
2092 (block_rsv->reserved + block_rsv->freed[index]);
2095 *bytes_to_reserve += num_bytes;
2098 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2099 struct btrfs_pending_snapshot *pending)
2101 struct btrfs_root *root = pending->root;
2102 struct btrfs_root *snap = pending->snap;
2103 struct btrfs_block_rsv *block_rsv;
2108 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2111 /* refill source subvolume's orphan block reservation */
2112 block_rsv = root->orphan_block_rsv;
2113 index = trans->transid & 0x1;
2114 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2115 num_bytes = block_rsv->size -
2116 (block_rsv->reserved + block_rsv->freed[index]);
2117 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2118 root->orphan_block_rsv,
2123 /* setup orphan block reservation for the snapshot */
2124 block_rsv = btrfs_alloc_block_rsv(snap);
2127 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2128 snap->orphan_block_rsv = block_rsv;
2130 num_bytes = root->orphan_block_rsv->size;
2131 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2132 block_rsv, num_bytes);
2136 /* insert orphan item for the snapshot */
2137 WARN_ON(!root->orphan_item_inserted);
2138 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2139 snap->root_key.objectid);
2141 snap->orphan_item_inserted = 1;
2145 enum btrfs_orphan_cleanup_state {
2146 ORPHAN_CLEANUP_STARTED = 1,
2147 ORPHAN_CLEANUP_DONE = 2,
2151 * This is called in transaction commmit time. If there are no orphan
2152 * files in the subvolume, it removes orphan item and frees block_rsv
2155 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2156 struct btrfs_root *root)
2160 if (!list_empty(&root->orphan_list) ||
2161 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2164 if (root->orphan_item_inserted &&
2165 btrfs_root_refs(&root->root_item) > 0) {
2166 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2167 root->root_key.objectid);
2169 root->orphan_item_inserted = 0;
2172 if (root->orphan_block_rsv) {
2173 WARN_ON(root->orphan_block_rsv->size > 0);
2174 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2175 root->orphan_block_rsv = NULL;
2180 * This creates an orphan entry for the given inode in case something goes
2181 * wrong in the middle of an unlink/truncate.
2183 * NOTE: caller of this function should reserve 5 units of metadata for
2186 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2188 struct btrfs_root *root = BTRFS_I(inode)->root;
2189 struct btrfs_block_rsv *block_rsv = NULL;
2194 if (!root->orphan_block_rsv) {
2195 block_rsv = btrfs_alloc_block_rsv(root);
2199 spin_lock(&root->orphan_lock);
2200 if (!root->orphan_block_rsv) {
2201 root->orphan_block_rsv = block_rsv;
2202 } else if (block_rsv) {
2203 btrfs_free_block_rsv(root, block_rsv);
2207 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2208 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2211 * For proper ENOSPC handling, we should do orphan
2212 * cleanup when mounting. But this introduces backward
2213 * compatibility issue.
2215 if (!xchg(&root->orphan_item_inserted, 1))
2222 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2225 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2226 BTRFS_I(inode)->orphan_meta_reserved = 1;
2229 spin_unlock(&root->orphan_lock);
2232 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2234 /* grab metadata reservation from transaction handle */
2236 ret = btrfs_orphan_reserve_metadata(trans, inode);
2240 /* insert an orphan item to track this unlinked/truncated file */
2242 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2246 /* insert an orphan item to track subvolume contains orphan files */
2248 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2249 root->root_key.objectid);
2256 * We have done the truncate/delete so we can go ahead and remove the orphan
2257 * item for this particular inode.
2259 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2261 struct btrfs_root *root = BTRFS_I(inode)->root;
2262 int delete_item = 0;
2263 int release_rsv = 0;
2266 spin_lock(&root->orphan_lock);
2267 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2268 list_del_init(&BTRFS_I(inode)->i_orphan);
2272 if (BTRFS_I(inode)->orphan_meta_reserved) {
2273 BTRFS_I(inode)->orphan_meta_reserved = 0;
2276 spin_unlock(&root->orphan_lock);
2278 if (trans && delete_item) {
2279 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2284 btrfs_orphan_release_metadata(inode);
2290 * this cleans up any orphans that may be left on the list from the last use
2293 int btrfs_orphan_cleanup(struct btrfs_root *root)
2295 struct btrfs_path *path;
2296 struct extent_buffer *leaf;
2297 struct btrfs_key key, found_key;
2298 struct btrfs_trans_handle *trans;
2299 struct inode *inode;
2300 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2302 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2305 path = btrfs_alloc_path();
2312 key.objectid = BTRFS_ORPHAN_OBJECTID;
2313 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2314 key.offset = (u64)-1;
2317 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2322 * if ret == 0 means we found what we were searching for, which
2323 * is weird, but possible, so only screw with path if we didnt
2324 * find the key and see if we have stuff that matches
2328 if (path->slots[0] == 0)
2333 /* pull out the item */
2334 leaf = path->nodes[0];
2335 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2337 /* make sure the item matches what we want */
2338 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2340 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2343 /* release the path since we're done with it */
2344 btrfs_release_path(root, path);
2347 * this is where we are basically btrfs_lookup, without the
2348 * crossing root thing. we store the inode number in the
2349 * offset of the orphan item.
2351 found_key.objectid = found_key.offset;
2352 found_key.type = BTRFS_INODE_ITEM_KEY;
2353 found_key.offset = 0;
2354 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2355 if (IS_ERR(inode)) {
2356 ret = PTR_ERR(inode);
2361 * add this inode to the orphan list so btrfs_orphan_del does
2362 * the proper thing when we hit it
2364 spin_lock(&root->orphan_lock);
2365 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2366 spin_unlock(&root->orphan_lock);
2369 * if this is a bad inode, means we actually succeeded in
2370 * removing the inode, but not the orphan record, which means
2371 * we need to manually delete the orphan since iput will just
2372 * do a destroy_inode
2374 if (is_bad_inode(inode)) {
2375 trans = btrfs_start_transaction(root, 0);
2376 if (IS_ERR(trans)) {
2377 ret = PTR_ERR(trans);
2380 btrfs_orphan_del(trans, inode);
2381 btrfs_end_transaction(trans, root);
2386 /* if we have links, this was a truncate, lets do that */
2387 if (inode->i_nlink) {
2388 if (!S_ISREG(inode->i_mode)) {
2394 ret = btrfs_truncate(inode);
2399 /* this will do delete_inode and everything for us */
2404 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2406 if (root->orphan_block_rsv)
2407 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2410 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2411 trans = btrfs_join_transaction(root, 1);
2413 btrfs_end_transaction(trans, root);
2417 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2419 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2423 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2424 btrfs_free_path(path);
2429 * very simple check to peek ahead in the leaf looking for xattrs. If we
2430 * don't find any xattrs, we know there can't be any acls.
2432 * slot is the slot the inode is in, objectid is the objectid of the inode
2434 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2435 int slot, u64 objectid)
2437 u32 nritems = btrfs_header_nritems(leaf);
2438 struct btrfs_key found_key;
2442 while (slot < nritems) {
2443 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2445 /* we found a different objectid, there must not be acls */
2446 if (found_key.objectid != objectid)
2449 /* we found an xattr, assume we've got an acl */
2450 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2454 * we found a key greater than an xattr key, there can't
2455 * be any acls later on
2457 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2464 * it goes inode, inode backrefs, xattrs, extents,
2465 * so if there are a ton of hard links to an inode there can
2466 * be a lot of backrefs. Don't waste time searching too hard,
2467 * this is just an optimization
2472 /* we hit the end of the leaf before we found an xattr or
2473 * something larger than an xattr. We have to assume the inode
2480 * read an inode from the btree into the in-memory inode
2482 static void btrfs_read_locked_inode(struct inode *inode)
2484 struct btrfs_path *path;
2485 struct extent_buffer *leaf;
2486 struct btrfs_inode_item *inode_item;
2487 struct btrfs_timespec *tspec;
2488 struct btrfs_root *root = BTRFS_I(inode)->root;
2489 struct btrfs_key location;
2491 u64 alloc_group_block;
2495 path = btrfs_alloc_path();
2497 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2499 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2503 leaf = path->nodes[0];
2504 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2505 struct btrfs_inode_item);
2507 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2508 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2509 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2510 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2511 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2513 tspec = btrfs_inode_atime(inode_item);
2514 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2515 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2517 tspec = btrfs_inode_mtime(inode_item);
2518 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2519 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2521 tspec = btrfs_inode_ctime(inode_item);
2522 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2523 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2525 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2526 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2527 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2528 inode->i_generation = BTRFS_I(inode)->generation;
2530 rdev = btrfs_inode_rdev(leaf, inode_item);
2532 BTRFS_I(inode)->index_cnt = (u64)-1;
2533 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2535 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2538 * try to precache a NULL acl entry for files that don't have
2539 * any xattrs or acls
2541 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2543 cache_no_acl(inode);
2545 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2546 alloc_group_block, 0);
2547 btrfs_free_path(path);
2550 switch (inode->i_mode & S_IFMT) {
2552 inode->i_mapping->a_ops = &btrfs_aops;
2553 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2554 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2555 inode->i_fop = &btrfs_file_operations;
2556 inode->i_op = &btrfs_file_inode_operations;
2559 inode->i_fop = &btrfs_dir_file_operations;
2560 if (root == root->fs_info->tree_root)
2561 inode->i_op = &btrfs_dir_ro_inode_operations;
2563 inode->i_op = &btrfs_dir_inode_operations;
2566 inode->i_op = &btrfs_symlink_inode_operations;
2567 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2568 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2571 inode->i_op = &btrfs_special_inode_operations;
2572 init_special_inode(inode, inode->i_mode, rdev);
2576 btrfs_update_iflags(inode);
2580 btrfs_free_path(path);
2581 make_bad_inode(inode);
2585 * given a leaf and an inode, copy the inode fields into the leaf
2587 static void fill_inode_item(struct btrfs_trans_handle *trans,
2588 struct extent_buffer *leaf,
2589 struct btrfs_inode_item *item,
2590 struct inode *inode)
2592 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2593 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2594 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2595 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2596 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2598 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2599 inode->i_atime.tv_sec);
2600 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2601 inode->i_atime.tv_nsec);
2603 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2604 inode->i_mtime.tv_sec);
2605 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2606 inode->i_mtime.tv_nsec);
2608 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2609 inode->i_ctime.tv_sec);
2610 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2611 inode->i_ctime.tv_nsec);
2613 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2614 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2615 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2616 btrfs_set_inode_transid(leaf, item, trans->transid);
2617 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2618 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2619 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2623 * copy everything in the in-memory inode into the btree.
2625 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2626 struct btrfs_root *root, struct inode *inode)
2628 struct btrfs_inode_item *inode_item;
2629 struct btrfs_path *path;
2630 struct extent_buffer *leaf;
2633 path = btrfs_alloc_path();
2635 path->leave_spinning = 1;
2636 ret = btrfs_lookup_inode(trans, root, path,
2637 &BTRFS_I(inode)->location, 1);
2644 btrfs_unlock_up_safe(path, 1);
2645 leaf = path->nodes[0];
2646 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2647 struct btrfs_inode_item);
2649 fill_inode_item(trans, leaf, inode_item, inode);
2650 btrfs_mark_buffer_dirty(leaf);
2651 btrfs_set_inode_last_trans(trans, inode);
2654 btrfs_free_path(path);
2660 * unlink helper that gets used here in inode.c and in the tree logging
2661 * recovery code. It remove a link in a directory with a given name, and
2662 * also drops the back refs in the inode to the directory
2664 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2665 struct btrfs_root *root,
2666 struct inode *dir, struct inode *inode,
2667 const char *name, int name_len)
2669 struct btrfs_path *path;
2671 struct extent_buffer *leaf;
2672 struct btrfs_dir_item *di;
2673 struct btrfs_key key;
2676 path = btrfs_alloc_path();
2682 path->leave_spinning = 1;
2683 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2684 name, name_len, -1);
2693 leaf = path->nodes[0];
2694 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2695 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2698 btrfs_release_path(root, path);
2700 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2702 dir->i_ino, &index);
2704 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2705 "inode %lu parent %lu\n", name_len, name,
2706 inode->i_ino, dir->i_ino);
2710 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2711 index, name, name_len, -1);
2720 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2721 btrfs_release_path(root, path);
2723 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2725 BUG_ON(ret != 0 && ret != -ENOENT);
2727 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2732 btrfs_free_path(path);
2736 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2737 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2738 btrfs_update_inode(trans, root, dir);
2739 btrfs_drop_nlink(inode);
2740 ret = btrfs_update_inode(trans, root, inode);
2745 /* helper to check if there is any shared block in the path */
2746 static int check_path_shared(struct btrfs_root *root,
2747 struct btrfs_path *path)
2749 struct extent_buffer *eb;
2753 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2756 if (!path->nodes[level])
2758 eb = path->nodes[level];
2759 if (!btrfs_block_can_be_shared(root, eb))
2761 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2770 * helper to start transaction for unlink and rmdir.
2772 * unlink and rmdir are special in btrfs, they do not always free space.
2773 * so in enospc case, we should make sure they will free space before
2774 * allowing them to use the global metadata reservation.
2776 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2777 struct dentry *dentry)
2779 struct btrfs_trans_handle *trans;
2780 struct btrfs_root *root = BTRFS_I(dir)->root;
2781 struct btrfs_path *path;
2782 struct btrfs_inode_ref *ref;
2783 struct btrfs_dir_item *di;
2784 struct inode *inode = dentry->d_inode;
2790 trans = btrfs_start_transaction(root, 10);
2791 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2794 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2795 return ERR_PTR(-ENOSPC);
2797 /* check if there is someone else holds reference */
2798 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2799 return ERR_PTR(-ENOSPC);
2801 if (atomic_read(&inode->i_count) > 2)
2802 return ERR_PTR(-ENOSPC);
2804 if (xchg(&root->fs_info->enospc_unlink, 1))
2805 return ERR_PTR(-ENOSPC);
2807 path = btrfs_alloc_path();
2809 root->fs_info->enospc_unlink = 0;
2810 return ERR_PTR(-ENOMEM);
2813 trans = btrfs_start_transaction(root, 0);
2814 if (IS_ERR(trans)) {
2815 btrfs_free_path(path);
2816 root->fs_info->enospc_unlink = 0;
2820 path->skip_locking = 1;
2821 path->search_commit_root = 1;
2823 ret = btrfs_lookup_inode(trans, root, path,
2824 &BTRFS_I(dir)->location, 0);
2830 if (check_path_shared(root, path))
2835 btrfs_release_path(root, path);
2837 ret = btrfs_lookup_inode(trans, root, path,
2838 &BTRFS_I(inode)->location, 0);
2844 if (check_path_shared(root, path))
2849 btrfs_release_path(root, path);
2851 if (ret == 0 && S_ISREG(inode->i_mode)) {
2852 ret = btrfs_lookup_file_extent(trans, root, path,
2853 inode->i_ino, (u64)-1, 0);
2859 if (check_path_shared(root, path))
2861 btrfs_release_path(root, path);
2869 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2870 dentry->d_name.name, dentry->d_name.len, 0);
2876 if (check_path_shared(root, path))
2882 btrfs_release_path(root, path);
2884 ref = btrfs_lookup_inode_ref(trans, root, path,
2885 dentry->d_name.name, dentry->d_name.len,
2886 inode->i_ino, dir->i_ino, 0);
2892 if (check_path_shared(root, path))
2894 index = btrfs_inode_ref_index(path->nodes[0], ref);
2895 btrfs_release_path(root, path);
2897 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2898 dentry->d_name.name, dentry->d_name.len, 0);
2903 BUG_ON(ret == -ENOENT);
2904 if (check_path_shared(root, path))
2909 btrfs_free_path(path);
2911 btrfs_end_transaction(trans, root);
2912 root->fs_info->enospc_unlink = 0;
2913 return ERR_PTR(err);
2916 trans->block_rsv = &root->fs_info->global_block_rsv;
2920 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2921 struct btrfs_root *root)
2923 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2924 BUG_ON(!root->fs_info->enospc_unlink);
2925 root->fs_info->enospc_unlink = 0;
2927 btrfs_end_transaction_throttle(trans, root);
2930 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2932 struct btrfs_root *root = BTRFS_I(dir)->root;
2933 struct btrfs_trans_handle *trans;
2934 struct inode *inode = dentry->d_inode;
2936 unsigned long nr = 0;
2938 trans = __unlink_start_trans(dir, dentry);
2940 return PTR_ERR(trans);
2942 btrfs_set_trans_block_group(trans, dir);
2944 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2946 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2947 dentry->d_name.name, dentry->d_name.len);
2950 if (inode->i_nlink == 0) {
2951 ret = btrfs_orphan_add(trans, inode);
2955 nr = trans->blocks_used;
2956 __unlink_end_trans(trans, root);
2957 btrfs_btree_balance_dirty(root, nr);
2961 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2962 struct btrfs_root *root,
2963 struct inode *dir, u64 objectid,
2964 const char *name, int name_len)
2966 struct btrfs_path *path;
2967 struct extent_buffer *leaf;
2968 struct btrfs_dir_item *di;
2969 struct btrfs_key key;
2973 path = btrfs_alloc_path();
2977 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2978 name, name_len, -1);
2979 BUG_ON(!di || IS_ERR(di));
2981 leaf = path->nodes[0];
2982 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2983 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2984 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2986 btrfs_release_path(root, path);
2988 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2989 objectid, root->root_key.objectid,
2990 dir->i_ino, &index, name, name_len);
2992 BUG_ON(ret != -ENOENT);
2993 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2995 BUG_ON(!di || IS_ERR(di));
2997 leaf = path->nodes[0];
2998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2999 btrfs_release_path(root, path);
3003 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3004 index, name, name_len, -1);
3005 BUG_ON(!di || IS_ERR(di));
3007 leaf = path->nodes[0];
3008 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3009 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3010 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3012 btrfs_release_path(root, path);
3014 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3015 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3016 ret = btrfs_update_inode(trans, root, dir);
3019 btrfs_free_path(path);
3023 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3025 struct inode *inode = dentry->d_inode;
3027 struct btrfs_root *root = BTRFS_I(dir)->root;
3028 struct btrfs_trans_handle *trans;
3029 unsigned long nr = 0;
3031 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3032 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3035 trans = __unlink_start_trans(dir, dentry);
3037 return PTR_ERR(trans);
3039 btrfs_set_trans_block_group(trans, dir);
3041 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3042 err = btrfs_unlink_subvol(trans, root, dir,
3043 BTRFS_I(inode)->location.objectid,
3044 dentry->d_name.name,
3045 dentry->d_name.len);
3049 err = btrfs_orphan_add(trans, inode);
3053 /* now the directory is empty */
3054 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3055 dentry->d_name.name, dentry->d_name.len);
3057 btrfs_i_size_write(inode, 0);
3059 nr = trans->blocks_used;
3060 __unlink_end_trans(trans, root);
3061 btrfs_btree_balance_dirty(root, nr);
3068 * when truncating bytes in a file, it is possible to avoid reading
3069 * the leaves that contain only checksum items. This can be the
3070 * majority of the IO required to delete a large file, but it must
3071 * be done carefully.
3073 * The keys in the level just above the leaves are checked to make sure
3074 * the lowest key in a given leaf is a csum key, and starts at an offset
3075 * after the new size.
3077 * Then the key for the next leaf is checked to make sure it also has
3078 * a checksum item for the same file. If it does, we know our target leaf
3079 * contains only checksum items, and it can be safely freed without reading
3082 * This is just an optimization targeted at large files. It may do
3083 * nothing. It will return 0 unless things went badly.
3085 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3086 struct btrfs_root *root,
3087 struct btrfs_path *path,
3088 struct inode *inode, u64 new_size)
3090 struct btrfs_key key;
3093 struct btrfs_key found_key;
3094 struct btrfs_key other_key;
3095 struct btrfs_leaf_ref *ref;
3099 path->lowest_level = 1;
3100 key.objectid = inode->i_ino;
3101 key.type = BTRFS_CSUM_ITEM_KEY;
3102 key.offset = new_size;
3104 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3108 if (path->nodes[1] == NULL) {
3113 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3114 nritems = btrfs_header_nritems(path->nodes[1]);
3119 if (path->slots[1] >= nritems)
3122 /* did we find a key greater than anything we want to delete? */
3123 if (found_key.objectid > inode->i_ino ||
3124 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3127 /* we check the next key in the node to make sure the leave contains
3128 * only checksum items. This comparison doesn't work if our
3129 * leaf is the last one in the node
3131 if (path->slots[1] + 1 >= nritems) {
3133 /* search forward from the last key in the node, this
3134 * will bring us into the next node in the tree
3136 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3138 /* unlikely, but we inc below, so check to be safe */
3139 if (found_key.offset == (u64)-1)
3142 /* search_forward needs a path with locks held, do the
3143 * search again for the original key. It is possible
3144 * this will race with a balance and return a path that
3145 * we could modify, but this drop is just an optimization
3146 * and is allowed to miss some leaves.
3148 btrfs_release_path(root, path);
3151 /* setup a max key for search_forward */
3152 other_key.offset = (u64)-1;
3153 other_key.type = key.type;
3154 other_key.objectid = key.objectid;
3156 path->keep_locks = 1;
3157 ret = btrfs_search_forward(root, &found_key, &other_key,
3159 path->keep_locks = 0;
3160 if (ret || found_key.objectid != key.objectid ||
3161 found_key.type != key.type) {
3166 key.offset = found_key.offset;
3167 btrfs_release_path(root, path);
3172 /* we know there's one more slot after us in the tree,
3173 * read that key so we can verify it is also a checksum item
3175 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3177 if (found_key.objectid < inode->i_ino)
3180 if (found_key.type != key.type || found_key.offset < new_size)
3184 * if the key for the next leaf isn't a csum key from this objectid,
3185 * we can't be sure there aren't good items inside this leaf.
3188 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3191 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3192 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3194 * it is safe to delete this leaf, it contains only
3195 * csum items from this inode at an offset >= new_size
3197 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3200 if (root->ref_cows && leaf_gen < trans->transid) {
3201 ref = btrfs_alloc_leaf_ref(root, 0);
3203 ref->root_gen = root->root_key.offset;
3204 ref->bytenr = leaf_start;
3206 ref->generation = leaf_gen;
3209 btrfs_sort_leaf_ref(ref);
3211 ret = btrfs_add_leaf_ref(root, ref, 0);
3213 btrfs_free_leaf_ref(root, ref);
3219 btrfs_release_path(root, path);
3221 if (other_key.objectid == inode->i_ino &&
3222 other_key.type == key.type && other_key.offset > key.offset) {
3223 key.offset = other_key.offset;
3229 /* fixup any changes we've made to the path */
3230 path->lowest_level = 0;
3231 path->keep_locks = 0;
3232 btrfs_release_path(root, path);
3239 * this can truncate away extent items, csum items and directory items.
3240 * It starts at a high offset and removes keys until it can't find
3241 * any higher than new_size
3243 * csum items that cross the new i_size are truncated to the new size
3246 * min_type is the minimum key type to truncate down to. If set to 0, this
3247 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3249 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3250 struct btrfs_root *root,
3251 struct inode *inode,
3252 u64 new_size, u32 min_type)
3254 struct btrfs_path *path;
3255 struct extent_buffer *leaf;
3256 struct btrfs_file_extent_item *fi;
3257 struct btrfs_key key;
3258 struct btrfs_key found_key;
3259 u64 extent_start = 0;
3260 u64 extent_num_bytes = 0;
3261 u64 extent_offset = 0;
3263 u64 mask = root->sectorsize - 1;
3264 u32 found_type = (u8)-1;
3267 int pending_del_nr = 0;
3268 int pending_del_slot = 0;
3269 int extent_type = -1;
3274 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3276 if (root->ref_cows || root == root->fs_info->tree_root)
3277 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3279 path = btrfs_alloc_path();
3283 key.objectid = inode->i_ino;
3284 key.offset = (u64)-1;
3288 path->leave_spinning = 1;
3289 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3296 /* there are no items in the tree for us to truncate, we're
3299 if (path->slots[0] == 0)
3306 leaf = path->nodes[0];
3307 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3308 found_type = btrfs_key_type(&found_key);
3311 if (found_key.objectid != inode->i_ino)
3314 if (found_type < min_type)
3317 item_end = found_key.offset;
3318 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3319 fi = btrfs_item_ptr(leaf, path->slots[0],
3320 struct btrfs_file_extent_item);
3321 extent_type = btrfs_file_extent_type(leaf, fi);
3322 encoding = btrfs_file_extent_compression(leaf, fi);
3323 encoding |= btrfs_file_extent_encryption(leaf, fi);
3324 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3326 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3328 btrfs_file_extent_num_bytes(leaf, fi);
3329 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3330 item_end += btrfs_file_extent_inline_len(leaf,
3335 if (found_type > min_type) {
3338 if (item_end < new_size)
3340 if (found_key.offset >= new_size)
3346 /* FIXME, shrink the extent if the ref count is only 1 */
3347 if (found_type != BTRFS_EXTENT_DATA_KEY)
3350 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3352 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3353 if (!del_item && !encoding) {
3354 u64 orig_num_bytes =
3355 btrfs_file_extent_num_bytes(leaf, fi);
3356 extent_num_bytes = new_size -
3357 found_key.offset + root->sectorsize - 1;
3358 extent_num_bytes = extent_num_bytes &
3359 ~((u64)root->sectorsize - 1);
3360 btrfs_set_file_extent_num_bytes(leaf, fi,
3362 num_dec = (orig_num_bytes -
3364 if (root->ref_cows && extent_start != 0)
3365 inode_sub_bytes(inode, num_dec);
3366 btrfs_mark_buffer_dirty(leaf);
3369 btrfs_file_extent_disk_num_bytes(leaf,
3371 extent_offset = found_key.offset -
3372 btrfs_file_extent_offset(leaf, fi);
3374 /* FIXME blocksize != 4096 */
3375 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3376 if (extent_start != 0) {
3379 inode_sub_bytes(inode, num_dec);
3382 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3384 * we can't truncate inline items that have had
3388 btrfs_file_extent_compression(leaf, fi) == 0 &&
3389 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3390 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3391 u32 size = new_size - found_key.offset;
3393 if (root->ref_cows) {
3394 inode_sub_bytes(inode, item_end + 1 -
3398 btrfs_file_extent_calc_inline_size(size);
3399 ret = btrfs_truncate_item(trans, root, path,
3402 } else if (root->ref_cows) {
3403 inode_sub_bytes(inode, item_end + 1 -
3409 if (!pending_del_nr) {
3410 /* no pending yet, add ourselves */
3411 pending_del_slot = path->slots[0];
3413 } else if (pending_del_nr &&
3414 path->slots[0] + 1 == pending_del_slot) {
3415 /* hop on the pending chunk */
3417 pending_del_slot = path->slots[0];
3424 if (found_extent && (root->ref_cows ||
3425 root == root->fs_info->tree_root)) {
3426 btrfs_set_path_blocking(path);
3427 ret = btrfs_free_extent(trans, root, extent_start,
3428 extent_num_bytes, 0,
3429 btrfs_header_owner(leaf),
3430 inode->i_ino, extent_offset);
3434 if (found_type == BTRFS_INODE_ITEM_KEY)
3437 if (path->slots[0] == 0 ||
3438 path->slots[0] != pending_del_slot) {
3439 if (root->ref_cows) {
3443 if (pending_del_nr) {
3444 ret = btrfs_del_items(trans, root, path,
3450 btrfs_release_path(root, path);
3457 if (pending_del_nr) {
3458 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3462 btrfs_free_path(path);
3467 * taken from block_truncate_page, but does cow as it zeros out
3468 * any bytes left in the last page in the file.
3470 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3472 struct inode *inode = mapping->host;
3473 struct btrfs_root *root = BTRFS_I(inode)->root;
3474 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3475 struct btrfs_ordered_extent *ordered;
3476 struct extent_state *cached_state = NULL;
3478 u32 blocksize = root->sectorsize;
3479 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3480 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3486 if ((offset & (blocksize - 1)) == 0)
3488 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3494 page = grab_cache_page(mapping, index);
3496 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3500 page_start = page_offset(page);
3501 page_end = page_start + PAGE_CACHE_SIZE - 1;
3503 if (!PageUptodate(page)) {
3504 ret = btrfs_readpage(NULL, page);
3506 if (page->mapping != mapping) {
3508 page_cache_release(page);
3511 if (!PageUptodate(page)) {
3516 wait_on_page_writeback(page);
3518 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3520 set_page_extent_mapped(page);
3522 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3524 unlock_extent_cached(io_tree, page_start, page_end,
3525 &cached_state, GFP_NOFS);
3527 page_cache_release(page);
3528 btrfs_start_ordered_extent(inode, ordered, 1);
3529 btrfs_put_ordered_extent(ordered);
3533 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3534 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3535 0, 0, &cached_state, GFP_NOFS);
3537 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3540 unlock_extent_cached(io_tree, page_start, page_end,
3541 &cached_state, GFP_NOFS);
3546 if (offset != PAGE_CACHE_SIZE) {
3548 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3549 flush_dcache_page(page);
3552 ClearPageChecked(page);
3553 set_page_dirty(page);
3554 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3559 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3561 page_cache_release(page);
3567 * This function puts in dummy file extents for the area we're creating a hole
3568 * for. So if we are truncating this file to a larger size we need to insert
3569 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3570 * the range between oldsize and size
3572 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3574 struct btrfs_trans_handle *trans;
3575 struct btrfs_root *root = BTRFS_I(inode)->root;
3576 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3577 struct extent_map *em = NULL;
3578 struct extent_state *cached_state = NULL;
3579 u64 mask = root->sectorsize - 1;
3580 u64 hole_start = (oldsize + mask) & ~mask;
3581 u64 block_end = (size + mask) & ~mask;
3587 if (size <= hole_start)
3591 struct btrfs_ordered_extent *ordered;
3592 btrfs_wait_ordered_range(inode, hole_start,
3593 block_end - hole_start);
3594 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3595 &cached_state, GFP_NOFS);
3596 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3599 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3600 &cached_state, GFP_NOFS);
3601 btrfs_put_ordered_extent(ordered);
3604 cur_offset = hole_start;
3606 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3607 block_end - cur_offset, 0);
3608 BUG_ON(IS_ERR(em) || !em);
3609 last_byte = min(extent_map_end(em), block_end);
3610 last_byte = (last_byte + mask) & ~mask;
3611 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3613 hole_size = last_byte - cur_offset;
3615 trans = btrfs_start_transaction(root, 2);
3616 if (IS_ERR(trans)) {
3617 err = PTR_ERR(trans);
3620 btrfs_set_trans_block_group(trans, inode);
3622 err = btrfs_drop_extents(trans, inode, cur_offset,
3623 cur_offset + hole_size,
3628 err = btrfs_insert_file_extent(trans, root,
3629 inode->i_ino, cur_offset, 0,
3630 0, hole_size, 0, hole_size,
3635 btrfs_drop_extent_cache(inode, hole_start,
3638 btrfs_end_transaction(trans, root);
3640 free_extent_map(em);
3642 cur_offset = last_byte;
3643 if (cur_offset >= block_end)
3647 free_extent_map(em);
3648 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3653 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3655 loff_t oldsize = i_size_read(inode);
3658 if (newsize == oldsize)
3661 if (newsize > oldsize) {
3662 i_size_write(inode, newsize);
3663 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3664 truncate_pagecache(inode, oldsize, newsize);
3665 ret = btrfs_cont_expand(inode, oldsize, newsize);
3667 btrfs_setsize(inode, oldsize);
3671 mark_inode_dirty(inode);
3675 * We're truncating a file that used to have good data down to
3676 * zero. Make sure it gets into the ordered flush list so that
3677 * any new writes get down to disk quickly.
3680 BTRFS_I(inode)->ordered_data_close = 1;
3682 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3683 truncate_setsize(inode, newsize);
3684 ret = btrfs_truncate(inode);
3690 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3692 struct inode *inode = dentry->d_inode;
3693 struct btrfs_root *root = BTRFS_I(inode)->root;
3696 if (btrfs_root_readonly(root))
3699 err = inode_change_ok(inode, attr);
3703 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3704 err = btrfs_setsize(inode, attr->ia_size);
3709 if (attr->ia_valid) {
3710 setattr_copy(inode, attr);
3711 mark_inode_dirty(inode);
3713 if (attr->ia_valid & ATTR_MODE)
3714 err = btrfs_acl_chmod(inode);
3720 void btrfs_evict_inode(struct inode *inode)
3722 struct btrfs_trans_handle *trans;
3723 struct btrfs_root *root = BTRFS_I(inode)->root;
3727 trace_btrfs_inode_evict(inode);
3729 truncate_inode_pages(&inode->i_data, 0);
3730 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3731 root == root->fs_info->tree_root))
3734 if (is_bad_inode(inode)) {
3735 btrfs_orphan_del(NULL, inode);
3738 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3739 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3741 if (root->fs_info->log_root_recovering) {
3742 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3746 if (inode->i_nlink > 0) {
3747 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3751 btrfs_i_size_write(inode, 0);
3754 trans = btrfs_start_transaction(root, 0);
3755 BUG_ON(IS_ERR(trans));
3756 btrfs_set_trans_block_group(trans, inode);
3757 trans->block_rsv = root->orphan_block_rsv;
3759 ret = btrfs_block_rsv_check(trans, root,
3760 root->orphan_block_rsv, 0, 5);
3762 BUG_ON(ret != -EAGAIN);
3763 ret = btrfs_commit_transaction(trans, root);
3768 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3772 nr = trans->blocks_used;
3773 btrfs_end_transaction(trans, root);
3775 btrfs_btree_balance_dirty(root, nr);
3780 ret = btrfs_orphan_del(trans, inode);
3784 nr = trans->blocks_used;
3785 btrfs_end_transaction(trans, root);
3786 btrfs_btree_balance_dirty(root, nr);
3788 end_writeback(inode);
3793 * this returns the key found in the dir entry in the location pointer.
3794 * If no dir entries were found, location->objectid is 0.
3796 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3797 struct btrfs_key *location)
3799 const char *name = dentry->d_name.name;
3800 int namelen = dentry->d_name.len;
3801 struct btrfs_dir_item *di;
3802 struct btrfs_path *path;
3803 struct btrfs_root *root = BTRFS_I(dir)->root;
3806 path = btrfs_alloc_path();
3809 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3814 if (!di || IS_ERR(di))
3817 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3819 btrfs_free_path(path);
3822 location->objectid = 0;
3827 * when we hit a tree root in a directory, the btrfs part of the inode
3828 * needs to be changed to reflect the root directory of the tree root. This
3829 * is kind of like crossing a mount point.
3831 static int fixup_tree_root_location(struct btrfs_root *root,
3833 struct dentry *dentry,
3834 struct btrfs_key *location,
3835 struct btrfs_root **sub_root)
3837 struct btrfs_path *path;
3838 struct btrfs_root *new_root;
3839 struct btrfs_root_ref *ref;
3840 struct extent_buffer *leaf;
3844 path = btrfs_alloc_path();
3851 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3852 BTRFS_I(dir)->root->root_key.objectid,
3853 location->objectid);
3860 leaf = path->nodes[0];
3861 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3862 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3863 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3866 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3867 (unsigned long)(ref + 1),
3868 dentry->d_name.len);
3872 btrfs_release_path(root->fs_info->tree_root, path);
3874 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3875 if (IS_ERR(new_root)) {
3876 err = PTR_ERR(new_root);
3880 if (btrfs_root_refs(&new_root->root_item) == 0) {
3885 *sub_root = new_root;
3886 location->objectid = btrfs_root_dirid(&new_root->root_item);
3887 location->type = BTRFS_INODE_ITEM_KEY;
3888 location->offset = 0;
3891 btrfs_free_path(path);
3895 static void inode_tree_add(struct inode *inode)
3897 struct btrfs_root *root = BTRFS_I(inode)->root;
3898 struct btrfs_inode *entry;
3900 struct rb_node *parent;
3902 p = &root->inode_tree.rb_node;
3905 if (inode_unhashed(inode))
3908 spin_lock(&root->inode_lock);
3911 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3913 if (inode->i_ino < entry->vfs_inode.i_ino)
3914 p = &parent->rb_left;
3915 else if (inode->i_ino > entry->vfs_inode.i_ino)
3916 p = &parent->rb_right;
3918 WARN_ON(!(entry->vfs_inode.i_state &
3919 (I_WILL_FREE | I_FREEING)));
3920 rb_erase(parent, &root->inode_tree);
3921 RB_CLEAR_NODE(parent);
3922 spin_unlock(&root->inode_lock);
3926 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3927 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3928 spin_unlock(&root->inode_lock);
3931 static void inode_tree_del(struct inode *inode)
3933 struct btrfs_root *root = BTRFS_I(inode)->root;
3936 spin_lock(&root->inode_lock);
3937 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3938 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3939 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3940 empty = RB_EMPTY_ROOT(&root->inode_tree);
3942 spin_unlock(&root->inode_lock);
3945 * Free space cache has inodes in the tree root, but the tree root has a
3946 * root_refs of 0, so this could end up dropping the tree root as a
3947 * snapshot, so we need the extra !root->fs_info->tree_root check to
3948 * make sure we don't drop it.
3950 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3951 root != root->fs_info->tree_root) {
3952 synchronize_srcu(&root->fs_info->subvol_srcu);
3953 spin_lock(&root->inode_lock);
3954 empty = RB_EMPTY_ROOT(&root->inode_tree);
3955 spin_unlock(&root->inode_lock);
3957 btrfs_add_dead_root(root);
3961 int btrfs_invalidate_inodes(struct btrfs_root *root)
3963 struct rb_node *node;
3964 struct rb_node *prev;
3965 struct btrfs_inode *entry;
3966 struct inode *inode;
3969 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3971 spin_lock(&root->inode_lock);
3973 node = root->inode_tree.rb_node;
3977 entry = rb_entry(node, struct btrfs_inode, rb_node);
3979 if (objectid < entry->vfs_inode.i_ino)
3980 node = node->rb_left;
3981 else if (objectid > entry->vfs_inode.i_ino)
3982 node = node->rb_right;
3988 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3989 if (objectid <= entry->vfs_inode.i_ino) {
3993 prev = rb_next(prev);
3997 entry = rb_entry(node, struct btrfs_inode, rb_node);
3998 objectid = entry->vfs_inode.i_ino + 1;
3999 inode = igrab(&entry->vfs_inode);
4001 spin_unlock(&root->inode_lock);
4002 if (atomic_read(&inode->i_count) > 1)
4003 d_prune_aliases(inode);
4005 * btrfs_drop_inode will have it removed from
4006 * the inode cache when its usage count
4011 spin_lock(&root->inode_lock);
4015 if (cond_resched_lock(&root->inode_lock))
4018 node = rb_next(node);
4020 spin_unlock(&root->inode_lock);
4024 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4026 struct btrfs_iget_args *args = p;
4027 inode->i_ino = args->ino;
4028 BTRFS_I(inode)->root = args->root;
4029 btrfs_set_inode_space_info(args->root, inode);
4033 static int btrfs_find_actor(struct inode *inode, void *opaque)
4035 struct btrfs_iget_args *args = opaque;
4036 return args->ino == inode->i_ino &&
4037 args->root == BTRFS_I(inode)->root;
4040 static struct inode *btrfs_iget_locked(struct super_block *s,
4042 struct btrfs_root *root)
4044 struct inode *inode;
4045 struct btrfs_iget_args args;
4046 args.ino = objectid;
4049 inode = iget5_locked(s, objectid, btrfs_find_actor,
4050 btrfs_init_locked_inode,
4055 /* Get an inode object given its location and corresponding root.
4056 * Returns in *is_new if the inode was read from disk
4058 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4059 struct btrfs_root *root, int *new)
4061 struct inode *inode;
4063 inode = btrfs_iget_locked(s, location->objectid, root);
4065 return ERR_PTR(-ENOMEM);
4067 if (inode->i_state & I_NEW) {
4068 BTRFS_I(inode)->root = root;
4069 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4070 btrfs_read_locked_inode(inode);
4072 inode_tree_add(inode);
4073 unlock_new_inode(inode);
4081 static struct inode *new_simple_dir(struct super_block *s,
4082 struct btrfs_key *key,
4083 struct btrfs_root *root)
4085 struct inode *inode = new_inode(s);
4088 return ERR_PTR(-ENOMEM);
4090 BTRFS_I(inode)->root = root;
4091 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4092 BTRFS_I(inode)->dummy_inode = 1;
4094 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4095 inode->i_op = &simple_dir_inode_operations;
4096 inode->i_fop = &simple_dir_operations;
4097 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4098 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4103 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4105 struct inode *inode;
4106 struct btrfs_root *root = BTRFS_I(dir)->root;
4107 struct btrfs_root *sub_root = root;
4108 struct btrfs_key location;
4112 if (dentry->d_name.len > BTRFS_NAME_LEN)
4113 return ERR_PTR(-ENAMETOOLONG);
4115 ret = btrfs_inode_by_name(dir, dentry, &location);
4118 return ERR_PTR(ret);
4120 if (location.objectid == 0)
4123 if (location.type == BTRFS_INODE_ITEM_KEY) {
4124 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4128 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4130 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4131 ret = fixup_tree_root_location(root, dir, dentry,
4132 &location, &sub_root);
4135 inode = ERR_PTR(ret);
4137 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4139 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4141 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4143 if (!IS_ERR(inode) && root != sub_root) {
4144 down_read(&root->fs_info->cleanup_work_sem);
4145 if (!(inode->i_sb->s_flags & MS_RDONLY))
4146 ret = btrfs_orphan_cleanup(sub_root);
4147 up_read(&root->fs_info->cleanup_work_sem);
4149 inode = ERR_PTR(ret);
4155 static int btrfs_dentry_delete(const struct dentry *dentry)
4157 struct btrfs_root *root;
4159 if (!dentry->d_inode && !IS_ROOT(dentry))
4160 dentry = dentry->d_parent;
4162 if (dentry->d_inode) {
4163 root = BTRFS_I(dentry->d_inode)->root;
4164 if (btrfs_root_refs(&root->root_item) == 0)
4170 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4171 struct nameidata *nd)
4173 struct inode *inode;
4175 inode = btrfs_lookup_dentry(dir, dentry);
4177 return ERR_CAST(inode);
4179 return d_splice_alias(inode, dentry);
4182 static unsigned char btrfs_filetype_table[] = {
4183 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4186 static int btrfs_real_readdir(struct file *filp, void *dirent,
4189 struct inode *inode = filp->f_dentry->d_inode;
4190 struct btrfs_root *root = BTRFS_I(inode)->root;
4191 struct btrfs_item *item;
4192 struct btrfs_dir_item *di;
4193 struct btrfs_key key;
4194 struct btrfs_key found_key;
4195 struct btrfs_path *path;
4198 struct extent_buffer *leaf;
4201 unsigned char d_type;
4206 int key_type = BTRFS_DIR_INDEX_KEY;
4211 /* FIXME, use a real flag for deciding about the key type */
4212 if (root->fs_info->tree_root == root)
4213 key_type = BTRFS_DIR_ITEM_KEY;
4215 /* special case for "." */
4216 if (filp->f_pos == 0) {
4217 over = filldir(dirent, ".", 1,
4224 /* special case for .., just use the back ref */
4225 if (filp->f_pos == 1) {
4226 u64 pino = parent_ino(filp->f_path.dentry);
4227 over = filldir(dirent, "..", 2,
4233 path = btrfs_alloc_path();
4236 btrfs_set_key_type(&key, key_type);
4237 key.offset = filp->f_pos;
4238 key.objectid = inode->i_ino;
4240 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4246 leaf = path->nodes[0];
4247 nritems = btrfs_header_nritems(leaf);
4248 slot = path->slots[0];
4249 if (advance || slot >= nritems) {
4250 if (slot >= nritems - 1) {
4251 ret = btrfs_next_leaf(root, path);
4254 leaf = path->nodes[0];
4255 nritems = btrfs_header_nritems(leaf);
4256 slot = path->slots[0];
4264 item = btrfs_item_nr(leaf, slot);
4265 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4267 if (found_key.objectid != key.objectid)
4269 if (btrfs_key_type(&found_key) != key_type)
4271 if (found_key.offset < filp->f_pos)
4274 filp->f_pos = found_key.offset;
4276 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4278 di_total = btrfs_item_size(leaf, item);
4280 while (di_cur < di_total) {
4281 struct btrfs_key location;
4283 if (verify_dir_item(root, leaf, di))
4286 name_len = btrfs_dir_name_len(leaf, di);
4287 if (name_len <= sizeof(tmp_name)) {
4288 name_ptr = tmp_name;
4290 name_ptr = kmalloc(name_len, GFP_NOFS);
4296 read_extent_buffer(leaf, name_ptr,
4297 (unsigned long)(di + 1), name_len);
4299 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4300 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4302 /* is this a reference to our own snapshot? If so
4305 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4306 location.objectid == root->root_key.objectid) {
4310 over = filldir(dirent, name_ptr, name_len,
4311 found_key.offset, location.objectid,
4315 if (name_ptr != tmp_name)
4320 di_len = btrfs_dir_name_len(leaf, di) +
4321 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4323 di = (struct btrfs_dir_item *)((char *)di + di_len);
4327 /* Reached end of directory/root. Bump pos past the last item. */
4328 if (key_type == BTRFS_DIR_INDEX_KEY)
4330 * 32-bit glibc will use getdents64, but then strtol -
4331 * so the last number we can serve is this.
4333 filp->f_pos = 0x7fffffff;
4339 btrfs_free_path(path);
4343 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4345 struct btrfs_root *root = BTRFS_I(inode)->root;
4346 struct btrfs_trans_handle *trans;
4348 bool nolock = false;
4350 if (BTRFS_I(inode)->dummy_inode)
4354 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4356 if (wbc->sync_mode == WB_SYNC_ALL) {
4358 trans = btrfs_join_transaction_nolock(root, 1);
4360 trans = btrfs_join_transaction(root, 1);
4362 return PTR_ERR(trans);
4363 btrfs_set_trans_block_group(trans, inode);
4365 ret = btrfs_end_transaction_nolock(trans, root);
4367 ret = btrfs_commit_transaction(trans, root);
4373 * This is somewhat expensive, updating the tree every time the
4374 * inode changes. But, it is most likely to find the inode in cache.
4375 * FIXME, needs more benchmarking...there are no reasons other than performance
4376 * to keep or drop this code.
4378 void btrfs_dirty_inode(struct inode *inode)
4380 struct btrfs_root *root = BTRFS_I(inode)->root;
4381 struct btrfs_trans_handle *trans;
4384 if (BTRFS_I(inode)->dummy_inode)
4387 trans = btrfs_join_transaction(root, 1);
4388 BUG_ON(IS_ERR(trans));
4389 btrfs_set_trans_block_group(trans, inode);
4391 ret = btrfs_update_inode(trans, root, inode);
4392 if (ret && ret == -ENOSPC) {
4393 /* whoops, lets try again with the full transaction */
4394 btrfs_end_transaction(trans, root);
4395 trans = btrfs_start_transaction(root, 1);
4396 if (IS_ERR(trans)) {
4397 if (printk_ratelimit()) {
4398 printk(KERN_ERR "btrfs: fail to "
4399 "dirty inode %lu error %ld\n",
4400 inode->i_ino, PTR_ERR(trans));
4404 btrfs_set_trans_block_group(trans, inode);
4406 ret = btrfs_update_inode(trans, root, inode);
4408 if (printk_ratelimit()) {
4409 printk(KERN_ERR "btrfs: fail to "
4410 "dirty inode %lu error %d\n",
4415 btrfs_end_transaction(trans, root);
4419 * find the highest existing sequence number in a directory
4420 * and then set the in-memory index_cnt variable to reflect
4421 * free sequence numbers
4423 static int btrfs_set_inode_index_count(struct inode *inode)
4425 struct btrfs_root *root = BTRFS_I(inode)->root;
4426 struct btrfs_key key, found_key;
4427 struct btrfs_path *path;
4428 struct extent_buffer *leaf;
4431 key.objectid = inode->i_ino;
4432 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4433 key.offset = (u64)-1;
4435 path = btrfs_alloc_path();
4439 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4442 /* FIXME: we should be able to handle this */
4448 * MAGIC NUMBER EXPLANATION:
4449 * since we search a directory based on f_pos we have to start at 2
4450 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4451 * else has to start at 2
4453 if (path->slots[0] == 0) {
4454 BTRFS_I(inode)->index_cnt = 2;
4460 leaf = path->nodes[0];
4461 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4463 if (found_key.objectid != inode->i_ino ||
4464 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4465 BTRFS_I(inode)->index_cnt = 2;
4469 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4471 btrfs_free_path(path);
4476 * helper to find a free sequence number in a given directory. This current
4477 * code is very simple, later versions will do smarter things in the btree
4479 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4483 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4484 ret = btrfs_set_inode_index_count(dir);
4489 *index = BTRFS_I(dir)->index_cnt;
4490 BTRFS_I(dir)->index_cnt++;
4495 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4496 struct btrfs_root *root,
4498 const char *name, int name_len,
4499 u64 ref_objectid, u64 objectid,
4500 u64 alloc_hint, int mode, u64 *index)
4502 struct inode *inode;
4503 struct btrfs_inode_item *inode_item;
4504 struct btrfs_key *location;
4505 struct btrfs_path *path;
4506 struct btrfs_inode_ref *ref;
4507 struct btrfs_key key[2];
4513 path = btrfs_alloc_path();
4516 inode = new_inode(root->fs_info->sb);
4518 return ERR_PTR(-ENOMEM);
4521 trace_btrfs_inode_request(dir);
4523 ret = btrfs_set_inode_index(dir, index);
4526 return ERR_PTR(ret);
4530 * index_cnt is ignored for everything but a dir,
4531 * btrfs_get_inode_index_count has an explanation for the magic
4534 BTRFS_I(inode)->index_cnt = 2;
4535 BTRFS_I(inode)->root = root;
4536 BTRFS_I(inode)->generation = trans->transid;
4537 inode->i_generation = BTRFS_I(inode)->generation;
4538 btrfs_set_inode_space_info(root, inode);
4544 BTRFS_I(inode)->block_group =
4545 btrfs_find_block_group(root, 0, alloc_hint, owner);
4547 key[0].objectid = objectid;
4548 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4551 key[1].objectid = objectid;
4552 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4553 key[1].offset = ref_objectid;
4555 sizes[0] = sizeof(struct btrfs_inode_item);
4556 sizes[1] = name_len + sizeof(*ref);
4558 path->leave_spinning = 1;
4559 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4563 inode_init_owner(inode, dir, mode);
4564 inode->i_ino = objectid;
4565 inode_set_bytes(inode, 0);
4566 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4567 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4568 struct btrfs_inode_item);
4569 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4571 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4572 struct btrfs_inode_ref);
4573 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4574 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4575 ptr = (unsigned long)(ref + 1);
4576 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4578 btrfs_mark_buffer_dirty(path->nodes[0]);
4579 btrfs_free_path(path);
4581 location = &BTRFS_I(inode)->location;
4582 location->objectid = objectid;
4583 location->offset = 0;
4584 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4586 btrfs_inherit_iflags(inode, dir);
4588 if ((mode & S_IFREG)) {
4589 if (btrfs_test_opt(root, NODATASUM))
4590 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4591 if (btrfs_test_opt(root, NODATACOW) ||
4592 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4593 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4596 insert_inode_hash(inode);
4597 inode_tree_add(inode);
4599 trace_btrfs_inode_new(inode);
4604 BTRFS_I(dir)->index_cnt--;
4605 btrfs_free_path(path);
4607 return ERR_PTR(ret);
4610 static inline u8 btrfs_inode_type(struct inode *inode)
4612 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4616 * utility function to add 'inode' into 'parent_inode' with
4617 * a give name and a given sequence number.
4618 * if 'add_backref' is true, also insert a backref from the
4619 * inode to the parent directory.
4621 int btrfs_add_link(struct btrfs_trans_handle *trans,
4622 struct inode *parent_inode, struct inode *inode,
4623 const char *name, int name_len, int add_backref, u64 index)
4626 struct btrfs_key key;
4627 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4629 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4630 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4632 key.objectid = inode->i_ino;
4633 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4637 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4638 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4639 key.objectid, root->root_key.objectid,
4640 parent_inode->i_ino,
4641 index, name, name_len);
4642 } else if (add_backref) {
4643 ret = btrfs_insert_inode_ref(trans, root,
4644 name, name_len, inode->i_ino,
4645 parent_inode->i_ino, index);
4649 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4650 parent_inode->i_ino, &key,
4651 btrfs_inode_type(inode), index);
4654 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4656 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4657 ret = btrfs_update_inode(trans, root, parent_inode);
4662 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4663 struct inode *dir, struct dentry *dentry,
4664 struct inode *inode, int backref, u64 index)
4666 int err = btrfs_add_link(trans, dir, inode,
4667 dentry->d_name.name, dentry->d_name.len,
4670 d_instantiate(dentry, inode);
4678 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4679 int mode, dev_t rdev)
4681 struct btrfs_trans_handle *trans;
4682 struct btrfs_root *root = BTRFS_I(dir)->root;
4683 struct inode *inode = NULL;
4687 unsigned long nr = 0;
4690 if (!new_valid_dev(rdev))
4693 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4698 * 2 for inode item and ref
4700 * 1 for xattr if selinux is on
4702 trans = btrfs_start_transaction(root, 5);
4704 return PTR_ERR(trans);
4706 btrfs_set_trans_block_group(trans, dir);
4708 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4709 dentry->d_name.len, dir->i_ino, objectid,
4710 BTRFS_I(dir)->block_group, mode, &index);
4711 err = PTR_ERR(inode);
4715 err = btrfs_init_inode_security(trans, inode, dir);
4721 btrfs_set_trans_block_group(trans, inode);
4722 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4726 inode->i_op = &btrfs_special_inode_operations;
4727 init_special_inode(inode, inode->i_mode, rdev);
4728 btrfs_update_inode(trans, root, inode);
4730 btrfs_update_inode_block_group(trans, inode);
4731 btrfs_update_inode_block_group(trans, dir);
4733 nr = trans->blocks_used;
4734 btrfs_end_transaction_throttle(trans, root);
4735 btrfs_btree_balance_dirty(root, nr);
4737 inode_dec_link_count(inode);
4743 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4744 int mode, struct nameidata *nd)
4746 struct btrfs_trans_handle *trans;
4747 struct btrfs_root *root = BTRFS_I(dir)->root;
4748 struct inode *inode = NULL;
4751 unsigned long nr = 0;
4755 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4759 * 2 for inode item and ref
4761 * 1 for xattr if selinux is on
4763 trans = btrfs_start_transaction(root, 5);
4765 return PTR_ERR(trans);
4767 btrfs_set_trans_block_group(trans, dir);
4769 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4770 dentry->d_name.len, dir->i_ino, objectid,
4771 BTRFS_I(dir)->block_group, mode, &index);
4772 err = PTR_ERR(inode);
4776 err = btrfs_init_inode_security(trans, inode, dir);
4782 btrfs_set_trans_block_group(trans, inode);
4783 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4787 inode->i_mapping->a_ops = &btrfs_aops;
4788 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4789 inode->i_fop = &btrfs_file_operations;
4790 inode->i_op = &btrfs_file_inode_operations;
4791 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4793 btrfs_update_inode_block_group(trans, inode);
4794 btrfs_update_inode_block_group(trans, dir);
4796 nr = trans->blocks_used;
4797 btrfs_end_transaction_throttle(trans, root);
4799 inode_dec_link_count(inode);
4802 btrfs_btree_balance_dirty(root, nr);
4806 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4807 struct dentry *dentry)
4809 struct btrfs_trans_handle *trans;
4810 struct btrfs_root *root = BTRFS_I(dir)->root;
4811 struct inode *inode = old_dentry->d_inode;
4813 unsigned long nr = 0;
4817 if (inode->i_nlink == 0)
4820 /* do not allow sys_link's with other subvols of the same device */
4821 if (root->objectid != BTRFS_I(inode)->root->objectid)
4824 btrfs_inc_nlink(inode);
4825 inode->i_ctime = CURRENT_TIME;
4827 err = btrfs_set_inode_index(dir, &index);
4832 * 2 items for inode and inode ref
4833 * 2 items for dir items
4834 * 1 item for parent inode
4836 trans = btrfs_start_transaction(root, 5);
4837 if (IS_ERR(trans)) {
4838 err = PTR_ERR(trans);
4842 btrfs_set_trans_block_group(trans, dir);
4845 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4850 struct dentry *parent = dget_parent(dentry);
4851 btrfs_update_inode_block_group(trans, dir);
4852 err = btrfs_update_inode(trans, root, inode);
4854 btrfs_log_new_name(trans, inode, NULL, parent);
4858 nr = trans->blocks_used;
4859 btrfs_end_transaction_throttle(trans, root);
4862 inode_dec_link_count(inode);
4865 btrfs_btree_balance_dirty(root, nr);
4869 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4871 struct inode *inode = NULL;
4872 struct btrfs_trans_handle *trans;
4873 struct btrfs_root *root = BTRFS_I(dir)->root;
4875 int drop_on_err = 0;
4878 unsigned long nr = 1;
4880 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4885 * 2 items for inode and ref
4886 * 2 items for dir items
4887 * 1 for xattr if selinux is on
4889 trans = btrfs_start_transaction(root, 5);
4891 return PTR_ERR(trans);
4892 btrfs_set_trans_block_group(trans, dir);
4894 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4895 dentry->d_name.len, dir->i_ino, objectid,
4896 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4898 if (IS_ERR(inode)) {
4899 err = PTR_ERR(inode);
4905 err = btrfs_init_inode_security(trans, inode, dir);
4909 inode->i_op = &btrfs_dir_inode_operations;
4910 inode->i_fop = &btrfs_dir_file_operations;
4911 btrfs_set_trans_block_group(trans, inode);
4913 btrfs_i_size_write(inode, 0);
4914 err = btrfs_update_inode(trans, root, inode);
4918 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4919 dentry->d_name.len, 0, index);
4923 d_instantiate(dentry, inode);
4925 btrfs_update_inode_block_group(trans, inode);
4926 btrfs_update_inode_block_group(trans, dir);
4929 nr = trans->blocks_used;
4930 btrfs_end_transaction_throttle(trans, root);
4933 btrfs_btree_balance_dirty(root, nr);
4937 /* helper for btfs_get_extent. Given an existing extent in the tree,
4938 * and an extent that you want to insert, deal with overlap and insert
4939 * the new extent into the tree.
4941 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4942 struct extent_map *existing,
4943 struct extent_map *em,
4944 u64 map_start, u64 map_len)
4948 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4949 start_diff = map_start - em->start;
4950 em->start = map_start;
4952 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4953 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4954 em->block_start += start_diff;
4955 em->block_len -= start_diff;
4957 return add_extent_mapping(em_tree, em);
4960 static noinline int uncompress_inline(struct btrfs_path *path,
4961 struct inode *inode, struct page *page,
4962 size_t pg_offset, u64 extent_offset,
4963 struct btrfs_file_extent_item *item)
4966 struct extent_buffer *leaf = path->nodes[0];
4969 unsigned long inline_size;
4973 WARN_ON(pg_offset != 0);
4974 compress_type = btrfs_file_extent_compression(leaf, item);
4975 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4976 inline_size = btrfs_file_extent_inline_item_len(leaf,
4977 btrfs_item_nr(leaf, path->slots[0]));
4978 tmp = kmalloc(inline_size, GFP_NOFS);
4979 ptr = btrfs_file_extent_inline_start(item);
4981 read_extent_buffer(leaf, tmp, ptr, inline_size);
4983 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4984 ret = btrfs_decompress(compress_type, tmp, page,
4985 extent_offset, inline_size, max_size);
4987 char *kaddr = kmap_atomic(page, KM_USER0);
4988 unsigned long copy_size = min_t(u64,
4989 PAGE_CACHE_SIZE - pg_offset,
4990 max_size - extent_offset);
4991 memset(kaddr + pg_offset, 0, copy_size);
4992 kunmap_atomic(kaddr, KM_USER0);
4999 * a bit scary, this does extent mapping from logical file offset to the disk.
5000 * the ugly parts come from merging extents from the disk with the in-ram
5001 * representation. This gets more complex because of the data=ordered code,
5002 * where the in-ram extents might be locked pending data=ordered completion.
5004 * This also copies inline extents directly into the page.
5007 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5008 size_t pg_offset, u64 start, u64 len,
5014 u64 extent_start = 0;
5016 u64 objectid = inode->i_ino;
5018 struct btrfs_path *path = NULL;
5019 struct btrfs_root *root = BTRFS_I(inode)->root;
5020 struct btrfs_file_extent_item *item;
5021 struct extent_buffer *leaf;
5022 struct btrfs_key found_key;
5023 struct extent_map *em = NULL;
5024 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5025 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5026 struct btrfs_trans_handle *trans = NULL;
5030 read_lock(&em_tree->lock);
5031 em = lookup_extent_mapping(em_tree, start, len);
5033 em->bdev = root->fs_info->fs_devices->latest_bdev;
5034 read_unlock(&em_tree->lock);
5037 if (em->start > start || em->start + em->len <= start)
5038 free_extent_map(em);
5039 else if (em->block_start == EXTENT_MAP_INLINE && page)
5040 free_extent_map(em);
5044 em = alloc_extent_map(GFP_NOFS);
5049 em->bdev = root->fs_info->fs_devices->latest_bdev;
5050 em->start = EXTENT_MAP_HOLE;
5051 em->orig_start = EXTENT_MAP_HOLE;
5053 em->block_len = (u64)-1;
5056 path = btrfs_alloc_path();
5060 ret = btrfs_lookup_file_extent(trans, root, path,
5061 objectid, start, trans != NULL);
5068 if (path->slots[0] == 0)
5073 leaf = path->nodes[0];
5074 item = btrfs_item_ptr(leaf, path->slots[0],
5075 struct btrfs_file_extent_item);
5076 /* are we inside the extent that was found? */
5077 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5078 found_type = btrfs_key_type(&found_key);
5079 if (found_key.objectid != objectid ||
5080 found_type != BTRFS_EXTENT_DATA_KEY) {
5084 found_type = btrfs_file_extent_type(leaf, item);
5085 extent_start = found_key.offset;
5086 compress_type = btrfs_file_extent_compression(leaf, item);
5087 if (found_type == BTRFS_FILE_EXTENT_REG ||
5088 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5089 extent_end = extent_start +
5090 btrfs_file_extent_num_bytes(leaf, item);
5091 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5093 size = btrfs_file_extent_inline_len(leaf, item);
5094 extent_end = (extent_start + size + root->sectorsize - 1) &
5095 ~((u64)root->sectorsize - 1);
5098 if (start >= extent_end) {
5100 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5101 ret = btrfs_next_leaf(root, path);
5108 leaf = path->nodes[0];
5110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5111 if (found_key.objectid != objectid ||
5112 found_key.type != BTRFS_EXTENT_DATA_KEY)
5114 if (start + len <= found_key.offset)
5117 em->len = found_key.offset - start;
5121 if (found_type == BTRFS_FILE_EXTENT_REG ||
5122 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5123 em->start = extent_start;
5124 em->len = extent_end - extent_start;
5125 em->orig_start = extent_start -
5126 btrfs_file_extent_offset(leaf, item);
5127 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5129 em->block_start = EXTENT_MAP_HOLE;
5132 if (compress_type != BTRFS_COMPRESS_NONE) {
5133 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5134 em->compress_type = compress_type;
5135 em->block_start = bytenr;
5136 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5139 bytenr += btrfs_file_extent_offset(leaf, item);
5140 em->block_start = bytenr;
5141 em->block_len = em->len;
5142 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5143 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5146 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5150 size_t extent_offset;
5153 em->block_start = EXTENT_MAP_INLINE;
5154 if (!page || create) {
5155 em->start = extent_start;
5156 em->len = extent_end - extent_start;
5160 size = btrfs_file_extent_inline_len(leaf, item);
5161 extent_offset = page_offset(page) + pg_offset - extent_start;
5162 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5163 size - extent_offset);
5164 em->start = extent_start + extent_offset;
5165 em->len = (copy_size + root->sectorsize - 1) &
5166 ~((u64)root->sectorsize - 1);
5167 em->orig_start = EXTENT_MAP_INLINE;
5168 if (compress_type) {
5169 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5170 em->compress_type = compress_type;
5172 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5173 if (create == 0 && !PageUptodate(page)) {
5174 if (btrfs_file_extent_compression(leaf, item) !=
5175 BTRFS_COMPRESS_NONE) {
5176 ret = uncompress_inline(path, inode, page,
5178 extent_offset, item);
5182 read_extent_buffer(leaf, map + pg_offset, ptr,
5184 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5185 memset(map + pg_offset + copy_size, 0,
5186 PAGE_CACHE_SIZE - pg_offset -
5191 flush_dcache_page(page);
5192 } else if (create && PageUptodate(page)) {
5196 free_extent_map(em);
5198 btrfs_release_path(root, path);
5199 trans = btrfs_join_transaction(root, 1);
5201 return ERR_CAST(trans);
5205 write_extent_buffer(leaf, map + pg_offset, ptr,
5208 btrfs_mark_buffer_dirty(leaf);
5210 set_extent_uptodate(io_tree, em->start,
5211 extent_map_end(em) - 1, GFP_NOFS);
5214 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5221 em->block_start = EXTENT_MAP_HOLE;
5222 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5224 btrfs_release_path(root, path);
5225 if (em->start > start || extent_map_end(em) <= start) {
5226 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5227 "[%llu %llu]\n", (unsigned long long)em->start,
5228 (unsigned long long)em->len,
5229 (unsigned long long)start,
5230 (unsigned long long)len);
5236 write_lock(&em_tree->lock);
5237 ret = add_extent_mapping(em_tree, em);
5238 /* it is possible that someone inserted the extent into the tree
5239 * while we had the lock dropped. It is also possible that
5240 * an overlapping map exists in the tree
5242 if (ret == -EEXIST) {
5243 struct extent_map *existing;
5247 existing = lookup_extent_mapping(em_tree, start, len);
5248 if (existing && (existing->start > start ||
5249 existing->start + existing->len <= start)) {
5250 free_extent_map(existing);
5254 existing = lookup_extent_mapping(em_tree, em->start,
5257 err = merge_extent_mapping(em_tree, existing,
5260 free_extent_map(existing);
5262 free_extent_map(em);
5267 free_extent_map(em);
5271 free_extent_map(em);
5276 write_unlock(&em_tree->lock);
5279 trace_btrfs_get_extent(root, em);
5282 btrfs_free_path(path);
5284 ret = btrfs_end_transaction(trans, root);
5289 free_extent_map(em);
5290 return ERR_PTR(err);
5295 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5296 size_t pg_offset, u64 start, u64 len,
5299 struct extent_map *em;
5300 struct extent_map *hole_em = NULL;
5301 u64 range_start = start;
5307 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5312 * if our em maps to a hole, there might
5313 * actually be delalloc bytes behind it
5315 if (em->block_start != EXTENT_MAP_HOLE)
5321 /* check to see if we've wrapped (len == -1 or similar) */
5330 /* ok, we didn't find anything, lets look for delalloc */
5331 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5332 end, len, EXTENT_DELALLOC, 1);
5333 found_end = range_start + found;
5334 if (found_end < range_start)
5335 found_end = (u64)-1;
5338 * we didn't find anything useful, return
5339 * the original results from get_extent()
5341 if (range_start > end || found_end <= start) {
5347 /* adjust the range_start to make sure it doesn't
5348 * go backwards from the start they passed in
5350 range_start = max(start,range_start);
5351 found = found_end - range_start;
5354 u64 hole_start = start;
5357 em = alloc_extent_map(GFP_NOFS);
5363 * when btrfs_get_extent can't find anything it
5364 * returns one huge hole
5366 * make sure what it found really fits our range, and
5367 * adjust to make sure it is based on the start from
5371 u64 calc_end = extent_map_end(hole_em);
5373 if (calc_end <= start || (hole_em->start > end)) {
5374 free_extent_map(hole_em);
5377 hole_start = max(hole_em->start, start);
5378 hole_len = calc_end - hole_start;
5382 if (hole_em && range_start > hole_start) {
5383 /* our hole starts before our delalloc, so we
5384 * have to return just the parts of the hole
5385 * that go until the delalloc starts
5387 em->len = min(hole_len,
5388 range_start - hole_start);
5389 em->start = hole_start;
5390 em->orig_start = hole_start;
5392 * don't adjust block start at all,
5393 * it is fixed at EXTENT_MAP_HOLE
5395 em->block_start = hole_em->block_start;
5396 em->block_len = hole_len;
5398 em->start = range_start;
5400 em->orig_start = range_start;
5401 em->block_start = EXTENT_MAP_DELALLOC;
5402 em->block_len = found;
5404 } else if (hole_em) {
5409 free_extent_map(hole_em);
5411 free_extent_map(em);
5412 return ERR_PTR(err);
5417 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5420 struct btrfs_root *root = BTRFS_I(inode)->root;
5421 struct btrfs_trans_handle *trans;
5422 struct extent_map *em;
5423 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5424 struct btrfs_key ins;
5428 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5430 trans = btrfs_join_transaction(root, 0);
5432 return ERR_CAST(trans);
5434 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5436 alloc_hint = get_extent_allocation_hint(inode, start, len);
5437 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5438 alloc_hint, (u64)-1, &ins, 1);
5444 em = alloc_extent_map(GFP_NOFS);
5446 em = ERR_PTR(-ENOMEM);
5451 em->orig_start = em->start;
5452 em->len = ins.offset;
5454 em->block_start = ins.objectid;
5455 em->block_len = ins.offset;
5456 em->bdev = root->fs_info->fs_devices->latest_bdev;
5457 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5460 write_lock(&em_tree->lock);
5461 ret = add_extent_mapping(em_tree, em);
5462 write_unlock(&em_tree->lock);
5465 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5468 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5469 ins.offset, ins.offset, 0);
5471 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5475 btrfs_end_transaction(trans, root);
5480 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5481 * block must be cow'd
5483 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5484 struct inode *inode, u64 offset, u64 len)
5486 struct btrfs_path *path;
5488 struct extent_buffer *leaf;
5489 struct btrfs_root *root = BTRFS_I(inode)->root;
5490 struct btrfs_file_extent_item *fi;
5491 struct btrfs_key key;
5499 path = btrfs_alloc_path();
5503 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5508 slot = path->slots[0];
5511 /* can't find the item, must cow */
5518 leaf = path->nodes[0];
5519 btrfs_item_key_to_cpu(leaf, &key, slot);
5520 if (key.objectid != inode->i_ino ||
5521 key.type != BTRFS_EXTENT_DATA_KEY) {
5522 /* not our file or wrong item type, must cow */
5526 if (key.offset > offset) {
5527 /* Wrong offset, must cow */
5531 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5532 found_type = btrfs_file_extent_type(leaf, fi);
5533 if (found_type != BTRFS_FILE_EXTENT_REG &&
5534 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5535 /* not a regular extent, must cow */
5538 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5539 backref_offset = btrfs_file_extent_offset(leaf, fi);
5541 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5542 if (extent_end < offset + len) {
5543 /* extent doesn't include our full range, must cow */
5547 if (btrfs_extent_readonly(root, disk_bytenr))
5551 * look for other files referencing this extent, if we
5552 * find any we must cow
5554 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5555 key.offset - backref_offset, disk_bytenr))
5559 * adjust disk_bytenr and num_bytes to cover just the bytes
5560 * in this extent we are about to write. If there
5561 * are any csums in that range we have to cow in order
5562 * to keep the csums correct
5564 disk_bytenr += backref_offset;
5565 disk_bytenr += offset - key.offset;
5566 num_bytes = min(offset + len, extent_end) - offset;
5567 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5570 * all of the above have passed, it is safe to overwrite this extent
5575 btrfs_free_path(path);
5579 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5580 struct buffer_head *bh_result, int create)
5582 struct extent_map *em;
5583 struct btrfs_root *root = BTRFS_I(inode)->root;
5584 u64 start = iblock << inode->i_blkbits;
5585 u64 len = bh_result->b_size;
5586 struct btrfs_trans_handle *trans;
5588 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5593 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5594 * io. INLINE is special, and we could probably kludge it in here, but
5595 * it's still buffered so for safety lets just fall back to the generic
5598 * For COMPRESSED we _have_ to read the entire extent in so we can
5599 * decompress it, so there will be buffering required no matter what we
5600 * do, so go ahead and fallback to buffered.
5602 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5603 * to buffered IO. Don't blame me, this is the price we pay for using
5606 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5607 em->block_start == EXTENT_MAP_INLINE) {
5608 free_extent_map(em);
5612 /* Just a good old fashioned hole, return */
5613 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5614 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5615 free_extent_map(em);
5616 /* DIO will do one hole at a time, so just unlock a sector */
5617 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5618 start + root->sectorsize - 1, GFP_NOFS);
5623 * We don't allocate a new extent in the following cases
5625 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5627 * 2) The extent is marked as PREALLOC. We're good to go here and can
5628 * just use the extent.
5632 len = em->len - (start - em->start);
5636 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5637 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5638 em->block_start != EXTENT_MAP_HOLE)) {
5643 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5644 type = BTRFS_ORDERED_PREALLOC;
5646 type = BTRFS_ORDERED_NOCOW;
5647 len = min(len, em->len - (start - em->start));
5648 block_start = em->block_start + (start - em->start);
5651 * we're not going to log anything, but we do need
5652 * to make sure the current transaction stays open
5653 * while we look for nocow cross refs
5655 trans = btrfs_join_transaction(root, 0);
5659 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5660 ret = btrfs_add_ordered_extent_dio(inode, start,
5661 block_start, len, len, type);
5662 btrfs_end_transaction(trans, root);
5664 free_extent_map(em);
5669 btrfs_end_transaction(trans, root);
5673 * this will cow the extent, reset the len in case we changed
5676 len = bh_result->b_size;
5677 free_extent_map(em);
5678 em = btrfs_new_extent_direct(inode, start, len);
5681 len = min(len, em->len - (start - em->start));
5683 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5684 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5687 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5689 bh_result->b_size = len;
5690 bh_result->b_bdev = em->bdev;
5691 set_buffer_mapped(bh_result);
5692 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5693 set_buffer_new(bh_result);
5695 free_extent_map(em);
5700 struct btrfs_dio_private {
5701 struct inode *inode;
5708 /* number of bios pending for this dio */
5709 atomic_t pending_bios;
5714 struct bio *orig_bio;
5717 static void btrfs_endio_direct_read(struct bio *bio, int err)
5719 struct btrfs_dio_private *dip = bio->bi_private;
5720 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5721 struct bio_vec *bvec = bio->bi_io_vec;
5722 struct inode *inode = dip->inode;
5723 struct btrfs_root *root = BTRFS_I(inode)->root;
5725 u32 *private = dip->csums;
5727 start = dip->logical_offset;
5729 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5730 struct page *page = bvec->bv_page;
5733 unsigned long flags;
5735 local_irq_save(flags);
5736 kaddr = kmap_atomic(page, KM_IRQ0);
5737 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5738 csum, bvec->bv_len);
5739 btrfs_csum_final(csum, (char *)&csum);
5740 kunmap_atomic(kaddr, KM_IRQ0);
5741 local_irq_restore(flags);
5743 flush_dcache_page(bvec->bv_page);
5744 if (csum != *private) {
5745 printk(KERN_ERR "btrfs csum failed ino %lu off"
5746 " %llu csum %u private %u\n",
5747 inode->i_ino, (unsigned long long)start,
5753 start += bvec->bv_len;
5756 } while (bvec <= bvec_end);
5758 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5759 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5760 bio->bi_private = dip->private;
5765 /* If we had a csum failure make sure to clear the uptodate flag */
5767 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5768 dio_end_io(bio, err);
5771 static void btrfs_endio_direct_write(struct bio *bio, int err)
5773 struct btrfs_dio_private *dip = bio->bi_private;
5774 struct inode *inode = dip->inode;
5775 struct btrfs_root *root = BTRFS_I(inode)->root;
5776 struct btrfs_trans_handle *trans;
5777 struct btrfs_ordered_extent *ordered = NULL;
5778 struct extent_state *cached_state = NULL;
5779 u64 ordered_offset = dip->logical_offset;
5780 u64 ordered_bytes = dip->bytes;
5786 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5794 trans = btrfs_join_transaction(root, 1);
5795 if (IS_ERR(trans)) {
5799 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5801 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5802 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5804 ret = btrfs_update_inode(trans, root, inode);
5809 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5810 ordered->file_offset + ordered->len - 1, 0,
5811 &cached_state, GFP_NOFS);
5813 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5814 ret = btrfs_mark_extent_written(trans, inode,
5815 ordered->file_offset,
5816 ordered->file_offset +
5823 ret = insert_reserved_file_extent(trans, inode,
5824 ordered->file_offset,
5830 BTRFS_FILE_EXTENT_REG);
5831 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5832 ordered->file_offset, ordered->len);
5840 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5841 btrfs_ordered_update_i_size(inode, 0, ordered);
5842 btrfs_update_inode(trans, root, inode);
5844 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5845 ordered->file_offset + ordered->len - 1,
5846 &cached_state, GFP_NOFS);
5848 btrfs_delalloc_release_metadata(inode, ordered->len);
5849 btrfs_end_transaction(trans, root);
5850 ordered_offset = ordered->file_offset + ordered->len;
5851 btrfs_put_ordered_extent(ordered);
5852 btrfs_put_ordered_extent(ordered);
5856 * our bio might span multiple ordered extents. If we haven't
5857 * completed the accounting for the whole dio, go back and try again
5859 if (ordered_offset < dip->logical_offset + dip->bytes) {
5860 ordered_bytes = dip->logical_offset + dip->bytes -
5865 bio->bi_private = dip->private;
5870 /* If we had an error make sure to clear the uptodate flag */
5872 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5873 dio_end_io(bio, err);
5876 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5877 struct bio *bio, int mirror_num,
5878 unsigned long bio_flags, u64 offset)
5881 struct btrfs_root *root = BTRFS_I(inode)->root;
5882 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5887 static void btrfs_end_dio_bio(struct bio *bio, int err)
5889 struct btrfs_dio_private *dip = bio->bi_private;
5892 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5893 "sector %#Lx len %u err no %d\n",
5894 dip->inode->i_ino, bio->bi_rw,
5895 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5899 * before atomic variable goto zero, we must make sure
5900 * dip->errors is perceived to be set.
5902 smp_mb__before_atomic_dec();
5905 /* if there are more bios still pending for this dio, just exit */
5906 if (!atomic_dec_and_test(&dip->pending_bios))
5910 bio_io_error(dip->orig_bio);
5912 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5913 bio_endio(dip->orig_bio, 0);
5919 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5920 u64 first_sector, gfp_t gfp_flags)
5922 int nr_vecs = bio_get_nr_vecs(bdev);
5923 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5926 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5927 int rw, u64 file_offset, int skip_sum,
5930 int write = rw & REQ_WRITE;
5931 struct btrfs_root *root = BTRFS_I(inode)->root;
5935 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5939 if (write && !skip_sum) {
5940 ret = btrfs_wq_submit_bio(root->fs_info,
5941 inode, rw, bio, 0, 0,
5943 __btrfs_submit_bio_start_direct_io,
5944 __btrfs_submit_bio_done);
5946 } else if (!skip_sum)
5947 btrfs_lookup_bio_sums_dio(root, inode, bio,
5948 file_offset, csums);
5950 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5956 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5959 struct inode *inode = dip->inode;
5960 struct btrfs_root *root = BTRFS_I(inode)->root;
5961 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5963 struct bio *orig_bio = dip->orig_bio;
5964 struct bio_vec *bvec = orig_bio->bi_io_vec;
5965 u64 start_sector = orig_bio->bi_sector;
5966 u64 file_offset = dip->logical_offset;
5970 u32 *csums = dip->csums;
5972 int write = rw & REQ_WRITE;
5974 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5977 bio->bi_private = dip;
5978 bio->bi_end_io = btrfs_end_dio_bio;
5979 atomic_inc(&dip->pending_bios);
5981 map_length = orig_bio->bi_size;
5982 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5983 &map_length, NULL, 0);
5989 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5990 if (unlikely(map_length < submit_len + bvec->bv_len ||
5991 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5992 bvec->bv_offset) < bvec->bv_len)) {
5994 * inc the count before we submit the bio so
5995 * we know the end IO handler won't happen before
5996 * we inc the count. Otherwise, the dip might get freed
5997 * before we're done setting it up
5999 atomic_inc(&dip->pending_bios);
6000 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6001 file_offset, skip_sum,
6005 atomic_dec(&dip->pending_bios);
6009 /* Write's use the ordered csums */
6010 if (!write && !skip_sum)
6011 csums = csums + nr_pages;
6012 start_sector += submit_len >> 9;
6013 file_offset += submit_len;
6018 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6019 start_sector, GFP_NOFS);
6022 bio->bi_private = dip;
6023 bio->bi_end_io = btrfs_end_dio_bio;
6025 map_length = orig_bio->bi_size;
6026 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6027 &map_length, NULL, 0);
6033 submit_len += bvec->bv_len;
6039 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6048 * before atomic variable goto zero, we must
6049 * make sure dip->errors is perceived to be set.
6051 smp_mb__before_atomic_dec();
6052 if (atomic_dec_and_test(&dip->pending_bios))
6053 bio_io_error(dip->orig_bio);
6055 /* bio_end_io() will handle error, so we needn't return it */
6059 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6062 struct btrfs_root *root = BTRFS_I(inode)->root;
6063 struct btrfs_dio_private *dip;
6064 struct bio_vec *bvec = bio->bi_io_vec;
6066 int write = rw & REQ_WRITE;
6069 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6071 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6078 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6079 if (!write && !skip_sum) {
6080 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6088 dip->private = bio->bi_private;
6090 dip->logical_offset = file_offset;
6094 dip->bytes += bvec->bv_len;
6096 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6098 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6099 bio->bi_private = dip;
6101 dip->orig_bio = bio;
6102 atomic_set(&dip->pending_bios, 0);
6105 bio->bi_end_io = btrfs_endio_direct_write;
6107 bio->bi_end_io = btrfs_endio_direct_read;
6109 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6114 * If this is a write, we need to clean up the reserved space and kill
6115 * the ordered extent.
6118 struct btrfs_ordered_extent *ordered;
6119 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6120 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6121 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6122 btrfs_free_reserved_extent(root, ordered->start,
6124 btrfs_put_ordered_extent(ordered);
6125 btrfs_put_ordered_extent(ordered);
6127 bio_endio(bio, ret);
6130 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6131 const struct iovec *iov, loff_t offset,
6132 unsigned long nr_segs)
6137 unsigned blocksize_mask = root->sectorsize - 1;
6138 ssize_t retval = -EINVAL;
6139 loff_t end = offset;
6141 if (offset & blocksize_mask)
6144 /* Check the memory alignment. Blocks cannot straddle pages */
6145 for (seg = 0; seg < nr_segs; seg++) {
6146 addr = (unsigned long)iov[seg].iov_base;
6147 size = iov[seg].iov_len;
6149 if ((addr & blocksize_mask) || (size & blocksize_mask))
6156 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6157 const struct iovec *iov, loff_t offset,
6158 unsigned long nr_segs)
6160 struct file *file = iocb->ki_filp;
6161 struct inode *inode = file->f_mapping->host;
6162 struct btrfs_ordered_extent *ordered;
6163 struct extent_state *cached_state = NULL;
6164 u64 lockstart, lockend;
6166 int writing = rw & WRITE;
6168 size_t count = iov_length(iov, nr_segs);
6170 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6176 lockend = offset + count - 1;
6179 ret = btrfs_delalloc_reserve_space(inode, count);
6185 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6186 0, &cached_state, GFP_NOFS);
6188 * We're concerned with the entire range that we're going to be
6189 * doing DIO to, so we need to make sure theres no ordered
6190 * extents in this range.
6192 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6193 lockend - lockstart + 1);
6196 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6197 &cached_state, GFP_NOFS);
6198 btrfs_start_ordered_extent(inode, ordered, 1);
6199 btrfs_put_ordered_extent(ordered);
6204 * we don't use btrfs_set_extent_delalloc because we don't want
6205 * the dirty or uptodate bits
6208 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6209 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6210 EXTENT_DELALLOC, 0, NULL, &cached_state,
6213 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6214 lockend, EXTENT_LOCKED | write_bits,
6215 1, 0, &cached_state, GFP_NOFS);
6220 free_extent_state(cached_state);
6221 cached_state = NULL;
6223 ret = __blockdev_direct_IO(rw, iocb, inode,
6224 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6225 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6226 btrfs_submit_direct, 0);
6228 if (ret < 0 && ret != -EIOCBQUEUED) {
6229 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6230 offset + iov_length(iov, nr_segs) - 1,
6231 EXTENT_LOCKED | write_bits, 1, 0,
6232 &cached_state, GFP_NOFS);
6233 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6235 * We're falling back to buffered, unlock the section we didn't
6238 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6239 offset + iov_length(iov, nr_segs) - 1,
6240 EXTENT_LOCKED | write_bits, 1, 0,
6241 &cached_state, GFP_NOFS);
6244 free_extent_state(cached_state);
6248 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6249 __u64 start, __u64 len)
6251 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6254 int btrfs_readpage(struct file *file, struct page *page)
6256 struct extent_io_tree *tree;
6257 tree = &BTRFS_I(page->mapping->host)->io_tree;
6258 return extent_read_full_page(tree, page, btrfs_get_extent);
6261 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6263 struct extent_io_tree *tree;
6266 if (current->flags & PF_MEMALLOC) {
6267 redirty_page_for_writepage(wbc, page);
6271 tree = &BTRFS_I(page->mapping->host)->io_tree;
6272 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6275 int btrfs_writepages(struct address_space *mapping,
6276 struct writeback_control *wbc)
6278 struct extent_io_tree *tree;
6280 tree = &BTRFS_I(mapping->host)->io_tree;
6281 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6285 btrfs_readpages(struct file *file, struct address_space *mapping,
6286 struct list_head *pages, unsigned nr_pages)
6288 struct extent_io_tree *tree;
6289 tree = &BTRFS_I(mapping->host)->io_tree;
6290 return extent_readpages(tree, mapping, pages, nr_pages,
6293 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6295 struct extent_io_tree *tree;
6296 struct extent_map_tree *map;
6299 tree = &BTRFS_I(page->mapping->host)->io_tree;
6300 map = &BTRFS_I(page->mapping->host)->extent_tree;
6301 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6303 ClearPagePrivate(page);
6304 set_page_private(page, 0);
6305 page_cache_release(page);
6310 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6312 if (PageWriteback(page) || PageDirty(page))
6314 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6317 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6319 struct extent_io_tree *tree;
6320 struct btrfs_ordered_extent *ordered;
6321 struct extent_state *cached_state = NULL;
6322 u64 page_start = page_offset(page);
6323 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6327 * we have the page locked, so new writeback can't start,
6328 * and the dirty bit won't be cleared while we are here.
6330 * Wait for IO on this page so that we can safely clear
6331 * the PagePrivate2 bit and do ordered accounting
6333 wait_on_page_writeback(page);
6335 tree = &BTRFS_I(page->mapping->host)->io_tree;
6337 btrfs_releasepage(page, GFP_NOFS);
6340 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6342 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6346 * IO on this page will never be started, so we need
6347 * to account for any ordered extents now
6349 clear_extent_bit(tree, page_start, page_end,
6350 EXTENT_DIRTY | EXTENT_DELALLOC |
6351 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6352 &cached_state, GFP_NOFS);
6354 * whoever cleared the private bit is responsible
6355 * for the finish_ordered_io
6357 if (TestClearPagePrivate2(page)) {
6358 btrfs_finish_ordered_io(page->mapping->host,
6359 page_start, page_end);
6361 btrfs_put_ordered_extent(ordered);
6362 cached_state = NULL;
6363 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6366 clear_extent_bit(tree, page_start, page_end,
6367 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6368 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6369 __btrfs_releasepage(page, GFP_NOFS);
6371 ClearPageChecked(page);
6372 if (PagePrivate(page)) {
6373 ClearPagePrivate(page);
6374 set_page_private(page, 0);
6375 page_cache_release(page);
6380 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6381 * called from a page fault handler when a page is first dirtied. Hence we must
6382 * be careful to check for EOF conditions here. We set the page up correctly
6383 * for a written page which means we get ENOSPC checking when writing into
6384 * holes and correct delalloc and unwritten extent mapping on filesystems that
6385 * support these features.
6387 * We are not allowed to take the i_mutex here so we have to play games to
6388 * protect against truncate races as the page could now be beyond EOF. Because
6389 * vmtruncate() writes the inode size before removing pages, once we have the
6390 * page lock we can determine safely if the page is beyond EOF. If it is not
6391 * beyond EOF, then the page is guaranteed safe against truncation until we
6394 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6396 struct page *page = vmf->page;
6397 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6398 struct btrfs_root *root = BTRFS_I(inode)->root;
6399 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6400 struct btrfs_ordered_extent *ordered;
6401 struct extent_state *cached_state = NULL;
6403 unsigned long zero_start;
6409 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6413 else /* -ENOSPC, -EIO, etc */
6414 ret = VM_FAULT_SIGBUS;
6418 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6421 size = i_size_read(inode);
6422 page_start = page_offset(page);
6423 page_end = page_start + PAGE_CACHE_SIZE - 1;
6425 if ((page->mapping != inode->i_mapping) ||
6426 (page_start >= size)) {
6427 /* page got truncated out from underneath us */
6430 wait_on_page_writeback(page);
6432 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6434 set_page_extent_mapped(page);
6437 * we can't set the delalloc bits if there are pending ordered
6438 * extents. Drop our locks and wait for them to finish
6440 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6442 unlock_extent_cached(io_tree, page_start, page_end,
6443 &cached_state, GFP_NOFS);
6445 btrfs_start_ordered_extent(inode, ordered, 1);
6446 btrfs_put_ordered_extent(ordered);
6451 * XXX - page_mkwrite gets called every time the page is dirtied, even
6452 * if it was already dirty, so for space accounting reasons we need to
6453 * clear any delalloc bits for the range we are fixing to save. There
6454 * is probably a better way to do this, but for now keep consistent with
6455 * prepare_pages in the normal write path.
6457 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6458 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6459 0, 0, &cached_state, GFP_NOFS);
6461 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6464 unlock_extent_cached(io_tree, page_start, page_end,
6465 &cached_state, GFP_NOFS);
6466 ret = VM_FAULT_SIGBUS;
6471 /* page is wholly or partially inside EOF */
6472 if (page_start + PAGE_CACHE_SIZE > size)
6473 zero_start = size & ~PAGE_CACHE_MASK;
6475 zero_start = PAGE_CACHE_SIZE;
6477 if (zero_start != PAGE_CACHE_SIZE) {
6479 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6480 flush_dcache_page(page);
6483 ClearPageChecked(page);
6484 set_page_dirty(page);
6485 SetPageUptodate(page);
6487 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6488 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6490 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6494 return VM_FAULT_LOCKED;
6496 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6501 static int btrfs_truncate(struct inode *inode)
6503 struct btrfs_root *root = BTRFS_I(inode)->root;
6506 struct btrfs_trans_handle *trans;
6508 u64 mask = root->sectorsize - 1;
6510 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6514 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6515 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6517 trans = btrfs_start_transaction(root, 5);
6519 return PTR_ERR(trans);
6521 btrfs_set_trans_block_group(trans, inode);
6523 ret = btrfs_orphan_add(trans, inode);
6525 btrfs_end_transaction(trans, root);
6529 nr = trans->blocks_used;
6530 btrfs_end_transaction(trans, root);
6531 btrfs_btree_balance_dirty(root, nr);
6533 /* Now start a transaction for the truncate */
6534 trans = btrfs_start_transaction(root, 0);
6536 return PTR_ERR(trans);
6537 btrfs_set_trans_block_group(trans, inode);
6538 trans->block_rsv = root->orphan_block_rsv;
6541 * setattr is responsible for setting the ordered_data_close flag,
6542 * but that is only tested during the last file release. That
6543 * could happen well after the next commit, leaving a great big
6544 * window where new writes may get lost if someone chooses to write
6545 * to this file after truncating to zero
6547 * The inode doesn't have any dirty data here, and so if we commit
6548 * this is a noop. If someone immediately starts writing to the inode
6549 * it is very likely we'll catch some of their writes in this
6550 * transaction, and the commit will find this file on the ordered
6551 * data list with good things to send down.
6553 * This is a best effort solution, there is still a window where
6554 * using truncate to replace the contents of the file will
6555 * end up with a zero length file after a crash.
6557 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6558 btrfs_add_ordered_operation(trans, root, inode);
6562 trans = btrfs_start_transaction(root, 0);
6564 return PTR_ERR(trans);
6565 btrfs_set_trans_block_group(trans, inode);
6566 trans->block_rsv = root->orphan_block_rsv;
6569 ret = btrfs_block_rsv_check(trans, root,
6570 root->orphan_block_rsv, 0, 5);
6571 if (ret == -EAGAIN) {
6572 ret = btrfs_commit_transaction(trans, root);
6582 ret = btrfs_truncate_inode_items(trans, root, inode,
6584 BTRFS_EXTENT_DATA_KEY);
6585 if (ret != -EAGAIN) {
6590 ret = btrfs_update_inode(trans, root, inode);
6596 nr = trans->blocks_used;
6597 btrfs_end_transaction(trans, root);
6599 btrfs_btree_balance_dirty(root, nr);
6602 if (ret == 0 && inode->i_nlink > 0) {
6603 ret = btrfs_orphan_del(trans, inode);
6606 } else if (ret && inode->i_nlink > 0) {
6608 * Failed to do the truncate, remove us from the in memory
6611 ret = btrfs_orphan_del(NULL, inode);
6614 ret = btrfs_update_inode(trans, root, inode);
6618 nr = trans->blocks_used;
6619 ret = btrfs_end_transaction_throttle(trans, root);
6622 btrfs_btree_balance_dirty(root, nr);
6628 * create a new subvolume directory/inode (helper for the ioctl).
6630 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6631 struct btrfs_root *new_root,
6632 u64 new_dirid, u64 alloc_hint)
6634 struct inode *inode;
6638 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6639 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6641 return PTR_ERR(inode);
6642 inode->i_op = &btrfs_dir_inode_operations;
6643 inode->i_fop = &btrfs_dir_file_operations;
6646 btrfs_i_size_write(inode, 0);
6648 err = btrfs_update_inode(trans, new_root, inode);
6655 /* helper function for file defrag and space balancing. This
6656 * forces readahead on a given range of bytes in an inode
6658 unsigned long btrfs_force_ra(struct address_space *mapping,
6659 struct file_ra_state *ra, struct file *file,
6660 pgoff_t offset, pgoff_t last_index)
6662 pgoff_t req_size = last_index - offset + 1;
6664 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6665 return offset + req_size;
6668 struct inode *btrfs_alloc_inode(struct super_block *sb)
6670 struct btrfs_inode *ei;
6671 struct inode *inode;
6673 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6678 ei->space_info = NULL;
6682 ei->last_sub_trans = 0;
6683 ei->logged_trans = 0;
6684 ei->delalloc_bytes = 0;
6685 ei->reserved_bytes = 0;
6686 ei->disk_i_size = 0;
6688 ei->index_cnt = (u64)-1;
6689 ei->last_unlink_trans = 0;
6691 atomic_set(&ei->outstanding_extents, 0);
6692 atomic_set(&ei->reserved_extents, 0);
6694 ei->ordered_data_close = 0;
6695 ei->orphan_meta_reserved = 0;
6696 ei->dummy_inode = 0;
6697 ei->force_compress = BTRFS_COMPRESS_NONE;
6699 inode = &ei->vfs_inode;
6700 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6701 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6702 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6703 mutex_init(&ei->log_mutex);
6704 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6705 INIT_LIST_HEAD(&ei->i_orphan);
6706 INIT_LIST_HEAD(&ei->delalloc_inodes);
6707 INIT_LIST_HEAD(&ei->ordered_operations);
6708 RB_CLEAR_NODE(&ei->rb_node);
6713 static void btrfs_i_callback(struct rcu_head *head)
6715 struct inode *inode = container_of(head, struct inode, i_rcu);
6716 INIT_LIST_HEAD(&inode->i_dentry);
6717 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6720 void btrfs_destroy_inode(struct inode *inode)
6722 struct btrfs_ordered_extent *ordered;
6723 struct btrfs_root *root = BTRFS_I(inode)->root;
6725 WARN_ON(!list_empty(&inode->i_dentry));
6726 WARN_ON(inode->i_data.nrpages);
6727 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6728 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6731 * This can happen where we create an inode, but somebody else also
6732 * created the same inode and we need to destroy the one we already
6739 * Make sure we're properly removed from the ordered operation
6743 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6744 spin_lock(&root->fs_info->ordered_extent_lock);
6745 list_del_init(&BTRFS_I(inode)->ordered_operations);
6746 spin_unlock(&root->fs_info->ordered_extent_lock);
6749 if (root == root->fs_info->tree_root) {
6750 struct btrfs_block_group_cache *block_group;
6752 block_group = btrfs_lookup_block_group(root->fs_info,
6753 BTRFS_I(inode)->block_group);
6754 if (block_group && block_group->inode == inode) {
6755 spin_lock(&block_group->lock);
6756 block_group->inode = NULL;
6757 spin_unlock(&block_group->lock);
6758 btrfs_put_block_group(block_group);
6759 } else if (block_group) {
6760 btrfs_put_block_group(block_group);
6764 spin_lock(&root->orphan_lock);
6765 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6766 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6768 list_del_init(&BTRFS_I(inode)->i_orphan);
6770 spin_unlock(&root->orphan_lock);
6773 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6777 printk(KERN_ERR "btrfs found ordered "
6778 "extent %llu %llu on inode cleanup\n",
6779 (unsigned long long)ordered->file_offset,
6780 (unsigned long long)ordered->len);
6781 btrfs_remove_ordered_extent(inode, ordered);
6782 btrfs_put_ordered_extent(ordered);
6783 btrfs_put_ordered_extent(ordered);
6786 inode_tree_del(inode);
6787 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6789 call_rcu(&inode->i_rcu, btrfs_i_callback);
6792 int btrfs_drop_inode(struct inode *inode)
6794 struct btrfs_root *root = BTRFS_I(inode)->root;
6796 if (btrfs_root_refs(&root->root_item) == 0 &&
6797 root != root->fs_info->tree_root)
6800 return generic_drop_inode(inode);
6803 static void init_once(void *foo)
6805 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6807 inode_init_once(&ei->vfs_inode);
6810 void btrfs_destroy_cachep(void)
6812 if (btrfs_inode_cachep)
6813 kmem_cache_destroy(btrfs_inode_cachep);
6814 if (btrfs_trans_handle_cachep)
6815 kmem_cache_destroy(btrfs_trans_handle_cachep);
6816 if (btrfs_transaction_cachep)
6817 kmem_cache_destroy(btrfs_transaction_cachep);
6818 if (btrfs_path_cachep)
6819 kmem_cache_destroy(btrfs_path_cachep);
6820 if (btrfs_free_space_cachep)
6821 kmem_cache_destroy(btrfs_free_space_cachep);
6824 int btrfs_init_cachep(void)
6826 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6827 sizeof(struct btrfs_inode), 0,
6828 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6829 if (!btrfs_inode_cachep)
6832 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6833 sizeof(struct btrfs_trans_handle), 0,
6834 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6835 if (!btrfs_trans_handle_cachep)
6838 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6839 sizeof(struct btrfs_transaction), 0,
6840 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6841 if (!btrfs_transaction_cachep)
6844 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6845 sizeof(struct btrfs_path), 0,
6846 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6847 if (!btrfs_path_cachep)
6850 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6851 sizeof(struct btrfs_free_space), 0,
6852 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6853 if (!btrfs_free_space_cachep)
6858 btrfs_destroy_cachep();
6862 static int btrfs_getattr(struct vfsmount *mnt,
6863 struct dentry *dentry, struct kstat *stat)
6865 struct inode *inode = dentry->d_inode;
6866 generic_fillattr(inode, stat);
6867 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6868 stat->blksize = PAGE_CACHE_SIZE;
6869 stat->blocks = (inode_get_bytes(inode) +
6870 BTRFS_I(inode)->delalloc_bytes) >> 9;
6875 * If a file is moved, it will inherit the cow and compression flags of the new
6878 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6880 struct btrfs_inode *b_dir = BTRFS_I(dir);
6881 struct btrfs_inode *b_inode = BTRFS_I(inode);
6883 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6884 b_inode->flags |= BTRFS_INODE_NODATACOW;
6886 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6888 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6889 b_inode->flags |= BTRFS_INODE_COMPRESS;
6891 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6894 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6895 struct inode *new_dir, struct dentry *new_dentry)
6897 struct btrfs_trans_handle *trans;
6898 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6899 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6900 struct inode *new_inode = new_dentry->d_inode;
6901 struct inode *old_inode = old_dentry->d_inode;
6902 struct timespec ctime = CURRENT_TIME;
6907 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6910 /* we only allow rename subvolume link between subvolumes */
6911 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6914 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6915 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6918 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6919 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6922 * we're using rename to replace one file with another.
6923 * and the replacement file is large. Start IO on it now so
6924 * we don't add too much work to the end of the transaction
6926 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6927 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6928 filemap_flush(old_inode->i_mapping);
6930 /* close the racy window with snapshot create/destroy ioctl */
6931 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6932 down_read(&root->fs_info->subvol_sem);
6934 * We want to reserve the absolute worst case amount of items. So if
6935 * both inodes are subvols and we need to unlink them then that would
6936 * require 4 item modifications, but if they are both normal inodes it
6937 * would require 5 item modifications, so we'll assume their normal
6938 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6939 * should cover the worst case number of items we'll modify.
6941 trans = btrfs_start_transaction(root, 20);
6943 return PTR_ERR(trans);
6945 btrfs_set_trans_block_group(trans, new_dir);
6948 btrfs_record_root_in_trans(trans, dest);
6950 ret = btrfs_set_inode_index(new_dir, &index);
6954 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6955 /* force full log commit if subvolume involved. */
6956 root->fs_info->last_trans_log_full_commit = trans->transid;
6958 ret = btrfs_insert_inode_ref(trans, dest,
6959 new_dentry->d_name.name,
6960 new_dentry->d_name.len,
6962 new_dir->i_ino, index);
6966 * this is an ugly little race, but the rename is required
6967 * to make sure that if we crash, the inode is either at the
6968 * old name or the new one. pinning the log transaction lets
6969 * us make sure we don't allow a log commit to come in after
6970 * we unlink the name but before we add the new name back in.
6972 btrfs_pin_log_trans(root);
6975 * make sure the inode gets flushed if it is replacing
6978 if (new_inode && new_inode->i_size &&
6979 old_inode && S_ISREG(old_inode->i_mode)) {
6980 btrfs_add_ordered_operation(trans, root, old_inode);
6983 old_dir->i_ctime = old_dir->i_mtime = ctime;
6984 new_dir->i_ctime = new_dir->i_mtime = ctime;
6985 old_inode->i_ctime = ctime;
6987 if (old_dentry->d_parent != new_dentry->d_parent)
6988 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6990 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6991 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6992 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6993 old_dentry->d_name.name,
6994 old_dentry->d_name.len);
6996 btrfs_inc_nlink(old_dentry->d_inode);
6997 ret = btrfs_unlink_inode(trans, root, old_dir,
6998 old_dentry->d_inode,
6999 old_dentry->d_name.name,
7000 old_dentry->d_name.len);
7005 new_inode->i_ctime = CURRENT_TIME;
7006 if (unlikely(new_inode->i_ino ==
7007 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7008 root_objectid = BTRFS_I(new_inode)->location.objectid;
7009 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7011 new_dentry->d_name.name,
7012 new_dentry->d_name.len);
7013 BUG_ON(new_inode->i_nlink == 0);
7015 ret = btrfs_unlink_inode(trans, dest, new_dir,
7016 new_dentry->d_inode,
7017 new_dentry->d_name.name,
7018 new_dentry->d_name.len);
7021 if (new_inode->i_nlink == 0) {
7022 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7027 fixup_inode_flags(new_dir, old_inode);
7029 ret = btrfs_add_link(trans, new_dir, old_inode,
7030 new_dentry->d_name.name,
7031 new_dentry->d_name.len, 0, index);
7034 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7035 struct dentry *parent = dget_parent(new_dentry);
7036 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7038 btrfs_end_log_trans(root);
7041 btrfs_end_transaction_throttle(trans, root);
7043 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7044 up_read(&root->fs_info->subvol_sem);
7050 * some fairly slow code that needs optimization. This walks the list
7051 * of all the inodes with pending delalloc and forces them to disk.
7053 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7055 struct list_head *head = &root->fs_info->delalloc_inodes;
7056 struct btrfs_inode *binode;
7057 struct inode *inode;
7059 if (root->fs_info->sb->s_flags & MS_RDONLY)
7062 spin_lock(&root->fs_info->delalloc_lock);
7063 while (!list_empty(head)) {
7064 binode = list_entry(head->next, struct btrfs_inode,
7066 inode = igrab(&binode->vfs_inode);
7068 list_del_init(&binode->delalloc_inodes);
7069 spin_unlock(&root->fs_info->delalloc_lock);
7071 filemap_flush(inode->i_mapping);
7073 btrfs_add_delayed_iput(inode);
7078 spin_lock(&root->fs_info->delalloc_lock);
7080 spin_unlock(&root->fs_info->delalloc_lock);
7082 /* the filemap_flush will queue IO into the worker threads, but
7083 * we have to make sure the IO is actually started and that
7084 * ordered extents get created before we return
7086 atomic_inc(&root->fs_info->async_submit_draining);
7087 while (atomic_read(&root->fs_info->nr_async_submits) ||
7088 atomic_read(&root->fs_info->async_delalloc_pages)) {
7089 wait_event(root->fs_info->async_submit_wait,
7090 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7091 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7093 atomic_dec(&root->fs_info->async_submit_draining);
7097 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7100 struct btrfs_inode *binode;
7101 struct inode *inode = NULL;
7103 spin_lock(&root->fs_info->delalloc_lock);
7104 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7105 binode = list_entry(root->fs_info->delalloc_inodes.next,
7106 struct btrfs_inode, delalloc_inodes);
7107 inode = igrab(&binode->vfs_inode);
7109 list_move_tail(&binode->delalloc_inodes,
7110 &root->fs_info->delalloc_inodes);
7114 list_del_init(&binode->delalloc_inodes);
7115 cond_resched_lock(&root->fs_info->delalloc_lock);
7117 spin_unlock(&root->fs_info->delalloc_lock);
7121 filemap_write_and_wait(inode->i_mapping);
7123 * We have to do this because compression doesn't
7124 * actually set PG_writeback until it submits the pages
7125 * for IO, which happens in an async thread, so we could
7126 * race and not actually wait for any writeback pages
7127 * because they've not been submitted yet. Technically
7128 * this could still be the case for the ordered stuff
7129 * since the async thread may not have started to do its
7130 * work yet. If this becomes the case then we need to
7131 * figure out a way to make sure that in writepage we
7132 * wait for any async pages to be submitted before
7133 * returning so that fdatawait does what its supposed to
7136 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7138 filemap_flush(inode->i_mapping);
7141 btrfs_add_delayed_iput(inode);
7149 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7150 const char *symname)
7152 struct btrfs_trans_handle *trans;
7153 struct btrfs_root *root = BTRFS_I(dir)->root;
7154 struct btrfs_path *path;
7155 struct btrfs_key key;
7156 struct inode *inode = NULL;
7164 struct btrfs_file_extent_item *ei;
7165 struct extent_buffer *leaf;
7166 unsigned long nr = 0;
7168 name_len = strlen(symname) + 1;
7169 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7170 return -ENAMETOOLONG;
7172 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7176 * 2 items for inode item and ref
7177 * 2 items for dir items
7178 * 1 item for xattr if selinux is on
7180 trans = btrfs_start_transaction(root, 5);
7182 return PTR_ERR(trans);
7184 btrfs_set_trans_block_group(trans, dir);
7186 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7187 dentry->d_name.len, dir->i_ino, objectid,
7188 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7190 err = PTR_ERR(inode);
7194 err = btrfs_init_inode_security(trans, inode, dir);
7200 btrfs_set_trans_block_group(trans, inode);
7201 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7205 inode->i_mapping->a_ops = &btrfs_aops;
7206 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7207 inode->i_fop = &btrfs_file_operations;
7208 inode->i_op = &btrfs_file_inode_operations;
7209 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7211 btrfs_update_inode_block_group(trans, inode);
7212 btrfs_update_inode_block_group(trans, dir);
7216 path = btrfs_alloc_path();
7218 key.objectid = inode->i_ino;
7220 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7221 datasize = btrfs_file_extent_calc_inline_size(name_len);
7222 err = btrfs_insert_empty_item(trans, root, path, &key,
7228 leaf = path->nodes[0];
7229 ei = btrfs_item_ptr(leaf, path->slots[0],
7230 struct btrfs_file_extent_item);
7231 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7232 btrfs_set_file_extent_type(leaf, ei,
7233 BTRFS_FILE_EXTENT_INLINE);
7234 btrfs_set_file_extent_encryption(leaf, ei, 0);
7235 btrfs_set_file_extent_compression(leaf, ei, 0);
7236 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7237 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7239 ptr = btrfs_file_extent_inline_start(ei);
7240 write_extent_buffer(leaf, symname, ptr, name_len);
7241 btrfs_mark_buffer_dirty(leaf);
7242 btrfs_free_path(path);
7244 inode->i_op = &btrfs_symlink_inode_operations;
7245 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7246 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7247 inode_set_bytes(inode, name_len);
7248 btrfs_i_size_write(inode, name_len - 1);
7249 err = btrfs_update_inode(trans, root, inode);
7254 nr = trans->blocks_used;
7255 btrfs_end_transaction_throttle(trans, root);
7257 inode_dec_link_count(inode);
7260 btrfs_btree_balance_dirty(root, nr);
7264 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7265 u64 start, u64 num_bytes, u64 min_size,
7266 loff_t actual_len, u64 *alloc_hint,
7267 struct btrfs_trans_handle *trans)
7269 struct btrfs_root *root = BTRFS_I(inode)->root;
7270 struct btrfs_key ins;
7271 u64 cur_offset = start;
7274 bool own_trans = true;
7278 while (num_bytes > 0) {
7280 trans = btrfs_start_transaction(root, 3);
7281 if (IS_ERR(trans)) {
7282 ret = PTR_ERR(trans);
7287 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7288 0, *alloc_hint, (u64)-1, &ins, 1);
7291 btrfs_end_transaction(trans, root);
7295 ret = insert_reserved_file_extent(trans, inode,
7296 cur_offset, ins.objectid,
7297 ins.offset, ins.offset,
7298 ins.offset, 0, 0, 0,
7299 BTRFS_FILE_EXTENT_PREALLOC);
7301 btrfs_drop_extent_cache(inode, cur_offset,
7302 cur_offset + ins.offset -1, 0);
7304 num_bytes -= ins.offset;
7305 cur_offset += ins.offset;
7306 *alloc_hint = ins.objectid + ins.offset;
7308 inode->i_ctime = CURRENT_TIME;
7309 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7310 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7311 (actual_len > inode->i_size) &&
7312 (cur_offset > inode->i_size)) {
7313 if (cur_offset > actual_len)
7314 i_size = actual_len;
7316 i_size = cur_offset;
7317 i_size_write(inode, i_size);
7318 btrfs_ordered_update_i_size(inode, i_size, NULL);
7321 ret = btrfs_update_inode(trans, root, inode);
7325 btrfs_end_transaction(trans, root);
7330 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7331 u64 start, u64 num_bytes, u64 min_size,
7332 loff_t actual_len, u64 *alloc_hint)
7334 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7335 min_size, actual_len, alloc_hint,
7339 int btrfs_prealloc_file_range_trans(struct inode *inode,
7340 struct btrfs_trans_handle *trans, int mode,
7341 u64 start, u64 num_bytes, u64 min_size,
7342 loff_t actual_len, u64 *alloc_hint)
7344 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7345 min_size, actual_len, alloc_hint, trans);
7348 static int btrfs_set_page_dirty(struct page *page)
7350 return __set_page_dirty_nobuffers(page);
7353 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7355 struct btrfs_root *root = BTRFS_I(inode)->root;
7357 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7359 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7361 return generic_permission(inode, mask, flags, btrfs_check_acl);
7364 static const struct inode_operations btrfs_dir_inode_operations = {
7365 .getattr = btrfs_getattr,
7366 .lookup = btrfs_lookup,
7367 .create = btrfs_create,
7368 .unlink = btrfs_unlink,
7370 .mkdir = btrfs_mkdir,
7371 .rmdir = btrfs_rmdir,
7372 .rename = btrfs_rename,
7373 .symlink = btrfs_symlink,
7374 .setattr = btrfs_setattr,
7375 .mknod = btrfs_mknod,
7376 .setxattr = btrfs_setxattr,
7377 .getxattr = btrfs_getxattr,
7378 .listxattr = btrfs_listxattr,
7379 .removexattr = btrfs_removexattr,
7380 .permission = btrfs_permission,
7382 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7383 .lookup = btrfs_lookup,
7384 .permission = btrfs_permission,
7387 static const struct file_operations btrfs_dir_file_operations = {
7388 .llseek = generic_file_llseek,
7389 .read = generic_read_dir,
7390 .readdir = btrfs_real_readdir,
7391 .unlocked_ioctl = btrfs_ioctl,
7392 #ifdef CONFIG_COMPAT
7393 .compat_ioctl = btrfs_ioctl,
7395 .release = btrfs_release_file,
7396 .fsync = btrfs_sync_file,
7399 static struct extent_io_ops btrfs_extent_io_ops = {
7400 .fill_delalloc = run_delalloc_range,
7401 .submit_bio_hook = btrfs_submit_bio_hook,
7402 .merge_bio_hook = btrfs_merge_bio_hook,
7403 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7404 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7405 .writepage_start_hook = btrfs_writepage_start_hook,
7406 .readpage_io_failed_hook = btrfs_io_failed_hook,
7407 .set_bit_hook = btrfs_set_bit_hook,
7408 .clear_bit_hook = btrfs_clear_bit_hook,
7409 .merge_extent_hook = btrfs_merge_extent_hook,
7410 .split_extent_hook = btrfs_split_extent_hook,
7414 * btrfs doesn't support the bmap operation because swapfiles
7415 * use bmap to make a mapping of extents in the file. They assume
7416 * these extents won't change over the life of the file and they
7417 * use the bmap result to do IO directly to the drive.
7419 * the btrfs bmap call would return logical addresses that aren't
7420 * suitable for IO and they also will change frequently as COW
7421 * operations happen. So, swapfile + btrfs == corruption.
7423 * For now we're avoiding this by dropping bmap.
7425 static const struct address_space_operations btrfs_aops = {
7426 .readpage = btrfs_readpage,
7427 .writepage = btrfs_writepage,
7428 .writepages = btrfs_writepages,
7429 .readpages = btrfs_readpages,
7430 .sync_page = block_sync_page,
7431 .direct_IO = btrfs_direct_IO,
7432 .invalidatepage = btrfs_invalidatepage,
7433 .releasepage = btrfs_releasepage,
7434 .set_page_dirty = btrfs_set_page_dirty,
7435 .error_remove_page = generic_error_remove_page,
7438 static const struct address_space_operations btrfs_symlink_aops = {
7439 .readpage = btrfs_readpage,
7440 .writepage = btrfs_writepage,
7441 .invalidatepage = btrfs_invalidatepage,
7442 .releasepage = btrfs_releasepage,
7445 static const struct inode_operations btrfs_file_inode_operations = {
7446 .getattr = btrfs_getattr,
7447 .setattr = btrfs_setattr,
7448 .setxattr = btrfs_setxattr,
7449 .getxattr = btrfs_getxattr,
7450 .listxattr = btrfs_listxattr,
7451 .removexattr = btrfs_removexattr,
7452 .permission = btrfs_permission,
7453 .fiemap = btrfs_fiemap,
7455 static const struct inode_operations btrfs_special_inode_operations = {
7456 .getattr = btrfs_getattr,
7457 .setattr = btrfs_setattr,
7458 .permission = btrfs_permission,
7459 .setxattr = btrfs_setxattr,
7460 .getxattr = btrfs_getxattr,
7461 .listxattr = btrfs_listxattr,
7462 .removexattr = btrfs_removexattr,
7464 static const struct inode_operations btrfs_symlink_inode_operations = {
7465 .readlink = generic_readlink,
7466 .follow_link = page_follow_link_light,
7467 .put_link = page_put_link,
7468 .getattr = btrfs_getattr,
7469 .permission = btrfs_permission,
7470 .setxattr = btrfs_setxattr,
7471 .getxattr = btrfs_getxattr,
7472 .listxattr = btrfs_listxattr,
7473 .removexattr = btrfs_removexattr,
7476 const struct dentry_operations btrfs_dentry_operations = {
7477 .d_delete = btrfs_dentry_delete,