2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93 struct inode *inode, struct inode *dir)
97 err = btrfs_init_acl(trans, inode, dir);
99 err = btrfs_xattr_security_init(trans, inode, dir);
104 * this does all the hard work for inserting an inline extent into
105 * the btree. The caller should have done a btrfs_drop_extents so that
106 * no overlapping inline items exist in the btree
108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root, struct inode *inode,
110 u64 start, size_t size, size_t compressed_size,
111 struct page **compressed_pages)
113 struct btrfs_key key;
114 struct btrfs_path *path;
115 struct extent_buffer *leaf;
116 struct page *page = NULL;
119 struct btrfs_file_extent_item *ei;
122 size_t cur_size = size;
124 unsigned long offset;
125 int compress_type = BTRFS_COMPRESS_NONE;
127 if (compressed_size && compressed_pages) {
128 compress_type = root->fs_info->compress_type;
129 cur_size = compressed_size;
132 path = btrfs_alloc_path();
136 path->leave_spinning = 1;
137 btrfs_set_trans_block_group(trans, inode);
139 key.objectid = inode->i_ino;
141 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142 datasize = btrfs_file_extent_calc_inline_size(cur_size);
144 inode_add_bytes(inode, size);
145 ret = btrfs_insert_empty_item(trans, root, path, &key,
152 leaf = path->nodes[0];
153 ei = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_file_extent_item);
155 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157 btrfs_set_file_extent_encryption(leaf, ei, 0);
158 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160 ptr = btrfs_file_extent_inline_start(ei);
162 if (compress_type != BTRFS_COMPRESS_NONE) {
165 while (compressed_size > 0) {
166 cpage = compressed_pages[i];
167 cur_size = min_t(unsigned long, compressed_size,
170 kaddr = kmap_atomic(cpage, KM_USER0);
171 write_extent_buffer(leaf, kaddr, ptr, cur_size);
172 kunmap_atomic(kaddr, KM_USER0);
176 compressed_size -= cur_size;
178 btrfs_set_file_extent_compression(leaf, ei,
181 page = find_get_page(inode->i_mapping,
182 start >> PAGE_CACHE_SHIFT);
183 btrfs_set_file_extent_compression(leaf, ei, 0);
184 kaddr = kmap_atomic(page, KM_USER0);
185 offset = start & (PAGE_CACHE_SIZE - 1);
186 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187 kunmap_atomic(kaddr, KM_USER0);
188 page_cache_release(page);
190 btrfs_mark_buffer_dirty(leaf);
191 btrfs_free_path(path);
194 * we're an inline extent, so nobody can
195 * extend the file past i_size without locking
196 * a page we already have locked.
198 * We must do any isize and inode updates
199 * before we unlock the pages. Otherwise we
200 * could end up racing with unlink.
202 BTRFS_I(inode)->disk_i_size = inode->i_size;
203 btrfs_update_inode(trans, root, inode);
207 btrfs_free_path(path);
213 * conditionally insert an inline extent into the file. This
214 * does the checks required to make sure the data is small enough
215 * to fit as an inline extent.
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218 struct btrfs_root *root,
219 struct inode *inode, u64 start, u64 end,
220 size_t compressed_size,
221 struct page **compressed_pages)
223 u64 isize = i_size_read(inode);
224 u64 actual_end = min(end + 1, isize);
225 u64 inline_len = actual_end - start;
226 u64 aligned_end = (end + root->sectorsize - 1) &
227 ~((u64)root->sectorsize - 1);
229 u64 data_len = inline_len;
233 data_len = compressed_size;
236 actual_end >= PAGE_CACHE_SIZE ||
237 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
239 (actual_end & (root->sectorsize - 1)) == 0) ||
241 data_len > root->fs_info->max_inline) {
245 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
249 if (isize > actual_end)
250 inline_len = min_t(u64, isize, actual_end);
251 ret = insert_inline_extent(trans, root, inode, start,
252 inline_len, compressed_size,
255 btrfs_delalloc_release_metadata(inode, end + 1 - start);
256 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
260 struct async_extent {
265 unsigned long nr_pages;
267 struct list_head list;
272 struct btrfs_root *root;
273 struct page *locked_page;
276 struct list_head extents;
277 struct btrfs_work work;
280 static noinline int add_async_extent(struct async_cow *cow,
281 u64 start, u64 ram_size,
284 unsigned long nr_pages,
287 struct async_extent *async_extent;
289 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
290 async_extent->start = start;
291 async_extent->ram_size = ram_size;
292 async_extent->compressed_size = compressed_size;
293 async_extent->pages = pages;
294 async_extent->nr_pages = nr_pages;
295 async_extent->compress_type = compress_type;
296 list_add_tail(&async_extent->list, &cow->extents);
301 * we create compressed extents in two phases. The first
302 * phase compresses a range of pages that have already been
303 * locked (both pages and state bits are locked).
305 * This is done inside an ordered work queue, and the compression
306 * is spread across many cpus. The actual IO submission is step
307 * two, and the ordered work queue takes care of making sure that
308 * happens in the same order things were put onto the queue by
309 * writepages and friends.
311 * If this code finds it can't get good compression, it puts an
312 * entry onto the work queue to write the uncompressed bytes. This
313 * makes sure that both compressed inodes and uncompressed inodes
314 * are written in the same order that pdflush sent them down.
316 static noinline int compress_file_range(struct inode *inode,
317 struct page *locked_page,
319 struct async_cow *async_cow,
322 struct btrfs_root *root = BTRFS_I(inode)->root;
323 struct btrfs_trans_handle *trans;
325 u64 blocksize = root->sectorsize;
327 u64 isize = i_size_read(inode);
329 struct page **pages = NULL;
330 unsigned long nr_pages;
331 unsigned long nr_pages_ret = 0;
332 unsigned long total_compressed = 0;
333 unsigned long total_in = 0;
334 unsigned long max_compressed = 128 * 1024;
335 unsigned long max_uncompressed = 128 * 1024;
338 int compress_type = root->fs_info->compress_type;
340 actual_end = min_t(u64, isize, end + 1);
343 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
344 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
347 * we don't want to send crud past the end of i_size through
348 * compression, that's just a waste of CPU time. So, if the
349 * end of the file is before the start of our current
350 * requested range of bytes, we bail out to the uncompressed
351 * cleanup code that can deal with all of this.
353 * It isn't really the fastest way to fix things, but this is a
354 * very uncommon corner.
356 if (actual_end <= start)
357 goto cleanup_and_bail_uncompressed;
359 total_compressed = actual_end - start;
361 /* we want to make sure that amount of ram required to uncompress
362 * an extent is reasonable, so we limit the total size in ram
363 * of a compressed extent to 128k. This is a crucial number
364 * because it also controls how easily we can spread reads across
365 * cpus for decompression.
367 * We also want to make sure the amount of IO required to do
368 * a random read is reasonably small, so we limit the size of
369 * a compressed extent to 128k.
371 total_compressed = min(total_compressed, max_uncompressed);
372 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
373 num_bytes = max(blocksize, num_bytes);
378 * we do compression for mount -o compress and when the
379 * inode has not been flagged as nocompress. This flag can
380 * change at any time if we discover bad compression ratios.
382 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
383 (btrfs_test_opt(root, COMPRESS) ||
384 (BTRFS_I(inode)->force_compress))) {
386 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
388 if (BTRFS_I(inode)->force_compress)
389 compress_type = BTRFS_I(inode)->force_compress;
391 ret = btrfs_compress_pages(compress_type,
392 inode->i_mapping, start,
393 total_compressed, pages,
394 nr_pages, &nr_pages_ret,
400 unsigned long offset = total_compressed &
401 (PAGE_CACHE_SIZE - 1);
402 struct page *page = pages[nr_pages_ret - 1];
405 /* zero the tail end of the last page, we might be
406 * sending it down to disk
409 kaddr = kmap_atomic(page, KM_USER0);
410 memset(kaddr + offset, 0,
411 PAGE_CACHE_SIZE - offset);
412 kunmap_atomic(kaddr, KM_USER0);
418 trans = btrfs_join_transaction(root, 1);
419 BUG_ON(IS_ERR(trans));
420 btrfs_set_trans_block_group(trans, inode);
421 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
423 /* lets try to make an inline extent */
424 if (ret || total_in < (actual_end - start)) {
425 /* we didn't compress the entire range, try
426 * to make an uncompressed inline extent.
428 ret = cow_file_range_inline(trans, root, inode,
429 start, end, 0, NULL);
431 /* try making a compressed inline extent */
432 ret = cow_file_range_inline(trans, root, inode,
434 total_compressed, pages);
438 * inline extent creation worked, we don't need
439 * to create any more async work items. Unlock
440 * and free up our temp pages.
442 extent_clear_unlock_delalloc(inode,
443 &BTRFS_I(inode)->io_tree,
445 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
446 EXTENT_CLEAR_DELALLOC |
447 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
449 btrfs_end_transaction(trans, root);
452 btrfs_end_transaction(trans, root);
457 * we aren't doing an inline extent round the compressed size
458 * up to a block size boundary so the allocator does sane
461 total_compressed = (total_compressed + blocksize - 1) &
465 * one last check to make sure the compression is really a
466 * win, compare the page count read with the blocks on disk
468 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
469 ~(PAGE_CACHE_SIZE - 1);
470 if (total_compressed >= total_in) {
473 num_bytes = total_in;
476 if (!will_compress && pages) {
478 * the compression code ran but failed to make things smaller,
479 * free any pages it allocated and our page pointer array
481 for (i = 0; i < nr_pages_ret; i++) {
482 WARN_ON(pages[i]->mapping);
483 page_cache_release(pages[i]);
487 total_compressed = 0;
490 /* flag the file so we don't compress in the future */
491 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
492 !(BTRFS_I(inode)->force_compress)) {
493 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
499 /* the async work queues will take care of doing actual
500 * allocation on disk for these compressed pages,
501 * and will submit them to the elevator.
503 add_async_extent(async_cow, start, num_bytes,
504 total_compressed, pages, nr_pages_ret,
507 if (start + num_bytes < end) {
514 cleanup_and_bail_uncompressed:
516 * No compression, but we still need to write the pages in
517 * the file we've been given so far. redirty the locked
518 * page if it corresponds to our extent and set things up
519 * for the async work queue to run cow_file_range to do
520 * the normal delalloc dance
522 if (page_offset(locked_page) >= start &&
523 page_offset(locked_page) <= end) {
524 __set_page_dirty_nobuffers(locked_page);
525 /* unlocked later on in the async handlers */
527 add_async_extent(async_cow, start, end - start + 1,
528 0, NULL, 0, BTRFS_COMPRESS_NONE);
536 for (i = 0; i < nr_pages_ret; i++) {
537 WARN_ON(pages[i]->mapping);
538 page_cache_release(pages[i]);
546 * phase two of compressed writeback. This is the ordered portion
547 * of the code, which only gets called in the order the work was
548 * queued. We walk all the async extents created by compress_file_range
549 * and send them down to the disk.
551 static noinline int submit_compressed_extents(struct inode *inode,
552 struct async_cow *async_cow)
554 struct async_extent *async_extent;
556 struct btrfs_trans_handle *trans;
557 struct btrfs_key ins;
558 struct extent_map *em;
559 struct btrfs_root *root = BTRFS_I(inode)->root;
560 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
561 struct extent_io_tree *io_tree;
564 if (list_empty(&async_cow->extents))
568 while (!list_empty(&async_cow->extents)) {
569 async_extent = list_entry(async_cow->extents.next,
570 struct async_extent, list);
571 list_del(&async_extent->list);
573 io_tree = &BTRFS_I(inode)->io_tree;
576 /* did the compression code fall back to uncompressed IO? */
577 if (!async_extent->pages) {
578 int page_started = 0;
579 unsigned long nr_written = 0;
581 lock_extent(io_tree, async_extent->start,
582 async_extent->start +
583 async_extent->ram_size - 1, GFP_NOFS);
585 /* allocate blocks */
586 ret = cow_file_range(inode, async_cow->locked_page,
588 async_extent->start +
589 async_extent->ram_size - 1,
590 &page_started, &nr_written, 0);
593 * if page_started, cow_file_range inserted an
594 * inline extent and took care of all the unlocking
595 * and IO for us. Otherwise, we need to submit
596 * all those pages down to the drive.
598 if (!page_started && !ret)
599 extent_write_locked_range(io_tree,
600 inode, async_extent->start,
601 async_extent->start +
602 async_extent->ram_size - 1,
610 lock_extent(io_tree, async_extent->start,
611 async_extent->start + async_extent->ram_size - 1,
614 trans = btrfs_join_transaction(root, 1);
615 BUG_ON(IS_ERR(trans));
616 ret = btrfs_reserve_extent(trans, root,
617 async_extent->compressed_size,
618 async_extent->compressed_size,
621 btrfs_end_transaction(trans, root);
625 for (i = 0; i < async_extent->nr_pages; i++) {
626 WARN_ON(async_extent->pages[i]->mapping);
627 page_cache_release(async_extent->pages[i]);
629 kfree(async_extent->pages);
630 async_extent->nr_pages = 0;
631 async_extent->pages = NULL;
632 unlock_extent(io_tree, async_extent->start,
633 async_extent->start +
634 async_extent->ram_size - 1, GFP_NOFS);
639 * here we're doing allocation and writeback of the
642 btrfs_drop_extent_cache(inode, async_extent->start,
643 async_extent->start +
644 async_extent->ram_size - 1, 0);
646 em = alloc_extent_map(GFP_NOFS);
647 em->start = async_extent->start;
648 em->len = async_extent->ram_size;
649 em->orig_start = em->start;
651 em->block_start = ins.objectid;
652 em->block_len = ins.offset;
653 em->bdev = root->fs_info->fs_devices->latest_bdev;
654 em->compress_type = async_extent->compress_type;
655 set_bit(EXTENT_FLAG_PINNED, &em->flags);
656 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
659 write_lock(&em_tree->lock);
660 ret = add_extent_mapping(em_tree, em);
661 write_unlock(&em_tree->lock);
662 if (ret != -EEXIST) {
666 btrfs_drop_extent_cache(inode, async_extent->start,
667 async_extent->start +
668 async_extent->ram_size - 1, 0);
671 ret = btrfs_add_ordered_extent_compress(inode,
674 async_extent->ram_size,
676 BTRFS_ORDERED_COMPRESSED,
677 async_extent->compress_type);
681 * clear dirty, set writeback and unlock the pages.
683 extent_clear_unlock_delalloc(inode,
684 &BTRFS_I(inode)->io_tree,
686 async_extent->start +
687 async_extent->ram_size - 1,
688 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
689 EXTENT_CLEAR_UNLOCK |
690 EXTENT_CLEAR_DELALLOC |
691 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
693 ret = btrfs_submit_compressed_write(inode,
695 async_extent->ram_size,
697 ins.offset, async_extent->pages,
698 async_extent->nr_pages);
701 alloc_hint = ins.objectid + ins.offset;
709 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
712 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
713 struct extent_map *em;
716 read_lock(&em_tree->lock);
717 em = search_extent_mapping(em_tree, start, num_bytes);
720 * if block start isn't an actual block number then find the
721 * first block in this inode and use that as a hint. If that
722 * block is also bogus then just don't worry about it.
724 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
726 em = search_extent_mapping(em_tree, 0, 0);
727 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
728 alloc_hint = em->block_start;
732 alloc_hint = em->block_start;
736 read_unlock(&em_tree->lock);
742 * when extent_io.c finds a delayed allocation range in the file,
743 * the call backs end up in this code. The basic idea is to
744 * allocate extents on disk for the range, and create ordered data structs
745 * in ram to track those extents.
747 * locked_page is the page that writepage had locked already. We use
748 * it to make sure we don't do extra locks or unlocks.
750 * *page_started is set to one if we unlock locked_page and do everything
751 * required to start IO on it. It may be clean and already done with
754 static noinline int cow_file_range(struct inode *inode,
755 struct page *locked_page,
756 u64 start, u64 end, int *page_started,
757 unsigned long *nr_written,
760 struct btrfs_root *root = BTRFS_I(inode)->root;
761 struct btrfs_trans_handle *trans;
764 unsigned long ram_size;
767 u64 blocksize = root->sectorsize;
768 struct btrfs_key ins;
769 struct extent_map *em;
770 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
773 BUG_ON(root == root->fs_info->tree_root);
774 trans = btrfs_join_transaction(root, 1);
775 BUG_ON(IS_ERR(trans));
776 btrfs_set_trans_block_group(trans, inode);
777 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
779 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
780 num_bytes = max(blocksize, num_bytes);
781 disk_num_bytes = num_bytes;
785 /* lets try to make an inline extent */
786 ret = cow_file_range_inline(trans, root, inode,
787 start, end, 0, NULL);
789 extent_clear_unlock_delalloc(inode,
790 &BTRFS_I(inode)->io_tree,
792 EXTENT_CLEAR_UNLOCK_PAGE |
793 EXTENT_CLEAR_UNLOCK |
794 EXTENT_CLEAR_DELALLOC |
796 EXTENT_SET_WRITEBACK |
797 EXTENT_END_WRITEBACK);
799 *nr_written = *nr_written +
800 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
807 BUG_ON(disk_num_bytes >
808 btrfs_super_total_bytes(&root->fs_info->super_copy));
810 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
811 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
813 while (disk_num_bytes > 0) {
816 cur_alloc_size = disk_num_bytes;
817 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
818 root->sectorsize, 0, alloc_hint,
822 em = alloc_extent_map(GFP_NOFS);
824 em->orig_start = em->start;
825 ram_size = ins.offset;
826 em->len = ins.offset;
828 em->block_start = ins.objectid;
829 em->block_len = ins.offset;
830 em->bdev = root->fs_info->fs_devices->latest_bdev;
831 set_bit(EXTENT_FLAG_PINNED, &em->flags);
834 write_lock(&em_tree->lock);
835 ret = add_extent_mapping(em_tree, em);
836 write_unlock(&em_tree->lock);
837 if (ret != -EEXIST) {
841 btrfs_drop_extent_cache(inode, start,
842 start + ram_size - 1, 0);
845 cur_alloc_size = ins.offset;
846 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
847 ram_size, cur_alloc_size, 0);
850 if (root->root_key.objectid ==
851 BTRFS_DATA_RELOC_TREE_OBJECTID) {
852 ret = btrfs_reloc_clone_csums(inode, start,
857 if (disk_num_bytes < cur_alloc_size)
860 /* we're not doing compressed IO, don't unlock the first
861 * page (which the caller expects to stay locked), don't
862 * clear any dirty bits and don't set any writeback bits
864 * Do set the Private2 bit so we know this page was properly
865 * setup for writepage
867 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
868 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
871 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
872 start, start + ram_size - 1,
874 disk_num_bytes -= cur_alloc_size;
875 num_bytes -= cur_alloc_size;
876 alloc_hint = ins.objectid + ins.offset;
877 start += cur_alloc_size;
881 btrfs_end_transaction(trans, root);
887 * work queue call back to started compression on a file and pages
889 static noinline void async_cow_start(struct btrfs_work *work)
891 struct async_cow *async_cow;
893 async_cow = container_of(work, struct async_cow, work);
895 compress_file_range(async_cow->inode, async_cow->locked_page,
896 async_cow->start, async_cow->end, async_cow,
899 async_cow->inode = NULL;
903 * work queue call back to submit previously compressed pages
905 static noinline void async_cow_submit(struct btrfs_work *work)
907 struct async_cow *async_cow;
908 struct btrfs_root *root;
909 unsigned long nr_pages;
911 async_cow = container_of(work, struct async_cow, work);
913 root = async_cow->root;
914 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
917 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
919 if (atomic_read(&root->fs_info->async_delalloc_pages) <
921 waitqueue_active(&root->fs_info->async_submit_wait))
922 wake_up(&root->fs_info->async_submit_wait);
924 if (async_cow->inode)
925 submit_compressed_extents(async_cow->inode, async_cow);
928 static noinline void async_cow_free(struct btrfs_work *work)
930 struct async_cow *async_cow;
931 async_cow = container_of(work, struct async_cow, work);
935 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
936 u64 start, u64 end, int *page_started,
937 unsigned long *nr_written)
939 struct async_cow *async_cow;
940 struct btrfs_root *root = BTRFS_I(inode)->root;
941 unsigned long nr_pages;
943 int limit = 10 * 1024 * 1042;
945 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
946 1, 0, NULL, GFP_NOFS);
947 while (start < end) {
948 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
949 async_cow->inode = inode;
950 async_cow->root = root;
951 async_cow->locked_page = locked_page;
952 async_cow->start = start;
954 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
957 cur_end = min(end, start + 512 * 1024 - 1);
959 async_cow->end = cur_end;
960 INIT_LIST_HEAD(&async_cow->extents);
962 async_cow->work.func = async_cow_start;
963 async_cow->work.ordered_func = async_cow_submit;
964 async_cow->work.ordered_free = async_cow_free;
965 async_cow->work.flags = 0;
967 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
969 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
971 btrfs_queue_worker(&root->fs_info->delalloc_workers,
974 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
975 wait_event(root->fs_info->async_submit_wait,
976 (atomic_read(&root->fs_info->async_delalloc_pages) <
980 while (atomic_read(&root->fs_info->async_submit_draining) &&
981 atomic_read(&root->fs_info->async_delalloc_pages)) {
982 wait_event(root->fs_info->async_submit_wait,
983 (atomic_read(&root->fs_info->async_delalloc_pages) ==
987 *nr_written += nr_pages;
994 static noinline int csum_exist_in_range(struct btrfs_root *root,
995 u64 bytenr, u64 num_bytes)
998 struct btrfs_ordered_sum *sums;
1001 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1002 bytenr + num_bytes - 1, &list);
1003 if (ret == 0 && list_empty(&list))
1006 while (!list_empty(&list)) {
1007 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1008 list_del(&sums->list);
1015 * when nowcow writeback call back. This checks for snapshots or COW copies
1016 * of the extents that exist in the file, and COWs the file as required.
1018 * If no cow copies or snapshots exist, we write directly to the existing
1021 static noinline int run_delalloc_nocow(struct inode *inode,
1022 struct page *locked_page,
1023 u64 start, u64 end, int *page_started, int force,
1024 unsigned long *nr_written)
1026 struct btrfs_root *root = BTRFS_I(inode)->root;
1027 struct btrfs_trans_handle *trans;
1028 struct extent_buffer *leaf;
1029 struct btrfs_path *path;
1030 struct btrfs_file_extent_item *fi;
1031 struct btrfs_key found_key;
1043 bool nolock = false;
1045 path = btrfs_alloc_path();
1047 if (root == root->fs_info->tree_root) {
1049 trans = btrfs_join_transaction_nolock(root, 1);
1051 trans = btrfs_join_transaction(root, 1);
1053 BUG_ON(IS_ERR(trans));
1055 cow_start = (u64)-1;
1058 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1061 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1062 leaf = path->nodes[0];
1063 btrfs_item_key_to_cpu(leaf, &found_key,
1064 path->slots[0] - 1);
1065 if (found_key.objectid == inode->i_ino &&
1066 found_key.type == BTRFS_EXTENT_DATA_KEY)
1071 leaf = path->nodes[0];
1072 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1073 ret = btrfs_next_leaf(root, path);
1078 leaf = path->nodes[0];
1084 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1086 if (found_key.objectid > inode->i_ino ||
1087 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1088 found_key.offset > end)
1091 if (found_key.offset > cur_offset) {
1092 extent_end = found_key.offset;
1097 fi = btrfs_item_ptr(leaf, path->slots[0],
1098 struct btrfs_file_extent_item);
1099 extent_type = btrfs_file_extent_type(leaf, fi);
1101 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1102 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1103 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1104 extent_offset = btrfs_file_extent_offset(leaf, fi);
1105 extent_end = found_key.offset +
1106 btrfs_file_extent_num_bytes(leaf, fi);
1107 if (extent_end <= start) {
1111 if (disk_bytenr == 0)
1113 if (btrfs_file_extent_compression(leaf, fi) ||
1114 btrfs_file_extent_encryption(leaf, fi) ||
1115 btrfs_file_extent_other_encoding(leaf, fi))
1117 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1119 if (btrfs_extent_readonly(root, disk_bytenr))
1121 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1123 extent_offset, disk_bytenr))
1125 disk_bytenr += extent_offset;
1126 disk_bytenr += cur_offset - found_key.offset;
1127 num_bytes = min(end + 1, extent_end) - cur_offset;
1129 * force cow if csum exists in the range.
1130 * this ensure that csum for a given extent are
1131 * either valid or do not exist.
1133 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1136 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1137 extent_end = found_key.offset +
1138 btrfs_file_extent_inline_len(leaf, fi);
1139 extent_end = ALIGN(extent_end, root->sectorsize);
1144 if (extent_end <= start) {
1149 if (cow_start == (u64)-1)
1150 cow_start = cur_offset;
1151 cur_offset = extent_end;
1152 if (cur_offset > end)
1158 btrfs_release_path(root, path);
1159 if (cow_start != (u64)-1) {
1160 ret = cow_file_range(inode, locked_page, cow_start,
1161 found_key.offset - 1, page_started,
1164 cow_start = (u64)-1;
1167 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1168 struct extent_map *em;
1169 struct extent_map_tree *em_tree;
1170 em_tree = &BTRFS_I(inode)->extent_tree;
1171 em = alloc_extent_map(GFP_NOFS);
1172 em->start = cur_offset;
1173 em->orig_start = em->start;
1174 em->len = num_bytes;
1175 em->block_len = num_bytes;
1176 em->block_start = disk_bytenr;
1177 em->bdev = root->fs_info->fs_devices->latest_bdev;
1178 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1180 write_lock(&em_tree->lock);
1181 ret = add_extent_mapping(em_tree, em);
1182 write_unlock(&em_tree->lock);
1183 if (ret != -EEXIST) {
1184 free_extent_map(em);
1187 btrfs_drop_extent_cache(inode, em->start,
1188 em->start + em->len - 1, 0);
1190 type = BTRFS_ORDERED_PREALLOC;
1192 type = BTRFS_ORDERED_NOCOW;
1195 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1196 num_bytes, num_bytes, type);
1199 if (root->root_key.objectid ==
1200 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1201 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1206 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1207 cur_offset, cur_offset + num_bytes - 1,
1208 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1209 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1210 EXTENT_SET_PRIVATE2);
1211 cur_offset = extent_end;
1212 if (cur_offset > end)
1215 btrfs_release_path(root, path);
1217 if (cur_offset <= end && cow_start == (u64)-1)
1218 cow_start = cur_offset;
1219 if (cow_start != (u64)-1) {
1220 ret = cow_file_range(inode, locked_page, cow_start, end,
1221 page_started, nr_written, 1);
1226 ret = btrfs_end_transaction_nolock(trans, root);
1229 ret = btrfs_end_transaction(trans, root);
1232 btrfs_free_path(path);
1237 * extent_io.c call back to do delayed allocation processing
1239 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1240 u64 start, u64 end, int *page_started,
1241 unsigned long *nr_written)
1244 struct btrfs_root *root = BTRFS_I(inode)->root;
1246 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1247 ret = run_delalloc_nocow(inode, locked_page, start, end,
1248 page_started, 1, nr_written);
1249 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1250 ret = run_delalloc_nocow(inode, locked_page, start, end,
1251 page_started, 0, nr_written);
1252 else if (!btrfs_test_opt(root, COMPRESS) &&
1253 !(BTRFS_I(inode)->force_compress))
1254 ret = cow_file_range(inode, locked_page, start, end,
1255 page_started, nr_written, 1);
1257 ret = cow_file_range_async(inode, locked_page, start, end,
1258 page_started, nr_written);
1262 static int btrfs_split_extent_hook(struct inode *inode,
1263 struct extent_state *orig, u64 split)
1265 /* not delalloc, ignore it */
1266 if (!(orig->state & EXTENT_DELALLOC))
1269 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1274 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1275 * extents so we can keep track of new extents that are just merged onto old
1276 * extents, such as when we are doing sequential writes, so we can properly
1277 * account for the metadata space we'll need.
1279 static int btrfs_merge_extent_hook(struct inode *inode,
1280 struct extent_state *new,
1281 struct extent_state *other)
1283 /* not delalloc, ignore it */
1284 if (!(other->state & EXTENT_DELALLOC))
1287 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1292 * extent_io.c set_bit_hook, used to track delayed allocation
1293 * bytes in this file, and to maintain the list of inodes that
1294 * have pending delalloc work to be done.
1296 static int btrfs_set_bit_hook(struct inode *inode,
1297 struct extent_state *state, int *bits)
1301 * set_bit and clear bit hooks normally require _irqsave/restore
1302 * but in this case, we are only testeing for the DELALLOC
1303 * bit, which is only set or cleared with irqs on
1305 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1306 struct btrfs_root *root = BTRFS_I(inode)->root;
1307 u64 len = state->end + 1 - state->start;
1308 int do_list = (root->root_key.objectid !=
1309 BTRFS_ROOT_TREE_OBJECTID);
1311 if (*bits & EXTENT_FIRST_DELALLOC)
1312 *bits &= ~EXTENT_FIRST_DELALLOC;
1314 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1316 spin_lock(&root->fs_info->delalloc_lock);
1317 BTRFS_I(inode)->delalloc_bytes += len;
1318 root->fs_info->delalloc_bytes += len;
1319 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1320 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1321 &root->fs_info->delalloc_inodes);
1323 spin_unlock(&root->fs_info->delalloc_lock);
1329 * extent_io.c clear_bit_hook, see set_bit_hook for why
1331 static int btrfs_clear_bit_hook(struct inode *inode,
1332 struct extent_state *state, int *bits)
1335 * set_bit and clear bit hooks normally require _irqsave/restore
1336 * but in this case, we are only testeing for the DELALLOC
1337 * bit, which is only set or cleared with irqs on
1339 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1340 struct btrfs_root *root = BTRFS_I(inode)->root;
1341 u64 len = state->end + 1 - state->start;
1342 int do_list = (root->root_key.objectid !=
1343 BTRFS_ROOT_TREE_OBJECTID);
1345 if (*bits & EXTENT_FIRST_DELALLOC)
1346 *bits &= ~EXTENT_FIRST_DELALLOC;
1347 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1348 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1350 if (*bits & EXTENT_DO_ACCOUNTING)
1351 btrfs_delalloc_release_metadata(inode, len);
1353 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1355 btrfs_free_reserved_data_space(inode, len);
1357 spin_lock(&root->fs_info->delalloc_lock);
1358 root->fs_info->delalloc_bytes -= len;
1359 BTRFS_I(inode)->delalloc_bytes -= len;
1361 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1362 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1363 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1365 spin_unlock(&root->fs_info->delalloc_lock);
1371 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1372 * we don't create bios that span stripes or chunks
1374 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1375 size_t size, struct bio *bio,
1376 unsigned long bio_flags)
1378 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1379 struct btrfs_mapping_tree *map_tree;
1380 u64 logical = (u64)bio->bi_sector << 9;
1385 if (bio_flags & EXTENT_BIO_COMPRESSED)
1388 length = bio->bi_size;
1389 map_tree = &root->fs_info->mapping_tree;
1390 map_length = length;
1391 ret = btrfs_map_block(map_tree, READ, logical,
1392 &map_length, NULL, 0);
1394 if (map_length < length + size)
1400 * in order to insert checksums into the metadata in large chunks,
1401 * we wait until bio submission time. All the pages in the bio are
1402 * checksummed and sums are attached onto the ordered extent record.
1404 * At IO completion time the cums attached on the ordered extent record
1405 * are inserted into the btree
1407 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1408 struct bio *bio, int mirror_num,
1409 unsigned long bio_flags,
1412 struct btrfs_root *root = BTRFS_I(inode)->root;
1415 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1421 * in order to insert checksums into the metadata in large chunks,
1422 * we wait until bio submission time. All the pages in the bio are
1423 * checksummed and sums are attached onto the ordered extent record.
1425 * At IO completion time the cums attached on the ordered extent record
1426 * are inserted into the btree
1428 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1429 int mirror_num, unsigned long bio_flags,
1432 struct btrfs_root *root = BTRFS_I(inode)->root;
1433 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1437 * extent_io.c submission hook. This does the right thing for csum calculation
1438 * on write, or reading the csums from the tree before a read
1440 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1441 int mirror_num, unsigned long bio_flags,
1444 struct btrfs_root *root = BTRFS_I(inode)->root;
1448 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1450 if (root == root->fs_info->tree_root)
1451 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1453 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1456 if (!(rw & REQ_WRITE)) {
1457 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1458 return btrfs_submit_compressed_read(inode, bio,
1459 mirror_num, bio_flags);
1460 } else if (!skip_sum)
1461 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1463 } else if (!skip_sum) {
1464 /* csum items have already been cloned */
1465 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1467 /* we're doing a write, do the async checksumming */
1468 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1469 inode, rw, bio, mirror_num,
1470 bio_flags, bio_offset,
1471 __btrfs_submit_bio_start,
1472 __btrfs_submit_bio_done);
1476 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1480 * given a list of ordered sums record them in the inode. This happens
1481 * at IO completion time based on sums calculated at bio submission time.
1483 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1484 struct inode *inode, u64 file_offset,
1485 struct list_head *list)
1487 struct btrfs_ordered_sum *sum;
1489 btrfs_set_trans_block_group(trans, inode);
1491 list_for_each_entry(sum, list, list) {
1492 btrfs_csum_file_blocks(trans,
1493 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1498 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1499 struct extent_state **cached_state)
1501 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1503 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1504 cached_state, GFP_NOFS);
1507 /* see btrfs_writepage_start_hook for details on why this is required */
1508 struct btrfs_writepage_fixup {
1510 struct btrfs_work work;
1513 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1515 struct btrfs_writepage_fixup *fixup;
1516 struct btrfs_ordered_extent *ordered;
1517 struct extent_state *cached_state = NULL;
1519 struct inode *inode;
1523 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1527 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1528 ClearPageChecked(page);
1532 inode = page->mapping->host;
1533 page_start = page_offset(page);
1534 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1536 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1537 &cached_state, GFP_NOFS);
1539 /* already ordered? We're done */
1540 if (PagePrivate2(page))
1543 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1545 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1546 page_end, &cached_state, GFP_NOFS);
1548 btrfs_start_ordered_extent(inode, ordered, 1);
1553 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1554 ClearPageChecked(page);
1556 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1557 &cached_state, GFP_NOFS);
1560 page_cache_release(page);
1565 * There are a few paths in the higher layers of the kernel that directly
1566 * set the page dirty bit without asking the filesystem if it is a
1567 * good idea. This causes problems because we want to make sure COW
1568 * properly happens and the data=ordered rules are followed.
1570 * In our case any range that doesn't have the ORDERED bit set
1571 * hasn't been properly setup for IO. We kick off an async process
1572 * to fix it up. The async helper will wait for ordered extents, set
1573 * the delalloc bit and make it safe to write the page.
1575 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1577 struct inode *inode = page->mapping->host;
1578 struct btrfs_writepage_fixup *fixup;
1579 struct btrfs_root *root = BTRFS_I(inode)->root;
1581 /* this page is properly in the ordered list */
1582 if (TestClearPagePrivate2(page))
1585 if (PageChecked(page))
1588 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1592 SetPageChecked(page);
1593 page_cache_get(page);
1594 fixup->work.func = btrfs_writepage_fixup_worker;
1596 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1600 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1601 struct inode *inode, u64 file_pos,
1602 u64 disk_bytenr, u64 disk_num_bytes,
1603 u64 num_bytes, u64 ram_bytes,
1604 u8 compression, u8 encryption,
1605 u16 other_encoding, int extent_type)
1607 struct btrfs_root *root = BTRFS_I(inode)->root;
1608 struct btrfs_file_extent_item *fi;
1609 struct btrfs_path *path;
1610 struct extent_buffer *leaf;
1611 struct btrfs_key ins;
1615 path = btrfs_alloc_path();
1618 path->leave_spinning = 1;
1621 * we may be replacing one extent in the tree with another.
1622 * The new extent is pinned in the extent map, and we don't want
1623 * to drop it from the cache until it is completely in the btree.
1625 * So, tell btrfs_drop_extents to leave this extent in the cache.
1626 * the caller is expected to unpin it and allow it to be merged
1629 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1633 ins.objectid = inode->i_ino;
1634 ins.offset = file_pos;
1635 ins.type = BTRFS_EXTENT_DATA_KEY;
1636 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1638 leaf = path->nodes[0];
1639 fi = btrfs_item_ptr(leaf, path->slots[0],
1640 struct btrfs_file_extent_item);
1641 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1642 btrfs_set_file_extent_type(leaf, fi, extent_type);
1643 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1644 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1645 btrfs_set_file_extent_offset(leaf, fi, 0);
1646 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1647 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1648 btrfs_set_file_extent_compression(leaf, fi, compression);
1649 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1650 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1652 btrfs_unlock_up_safe(path, 1);
1653 btrfs_set_lock_blocking(leaf);
1655 btrfs_mark_buffer_dirty(leaf);
1657 inode_add_bytes(inode, num_bytes);
1659 ins.objectid = disk_bytenr;
1660 ins.offset = disk_num_bytes;
1661 ins.type = BTRFS_EXTENT_ITEM_KEY;
1662 ret = btrfs_alloc_reserved_file_extent(trans, root,
1663 root->root_key.objectid,
1664 inode->i_ino, file_pos, &ins);
1666 btrfs_free_path(path);
1672 * helper function for btrfs_finish_ordered_io, this
1673 * just reads in some of the csum leaves to prime them into ram
1674 * before we start the transaction. It limits the amount of btree
1675 * reads required while inside the transaction.
1677 /* as ordered data IO finishes, this gets called so we can finish
1678 * an ordered extent if the range of bytes in the file it covers are
1681 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1683 struct btrfs_root *root = BTRFS_I(inode)->root;
1684 struct btrfs_trans_handle *trans = NULL;
1685 struct btrfs_ordered_extent *ordered_extent = NULL;
1686 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1687 struct extent_state *cached_state = NULL;
1688 int compress_type = 0;
1690 bool nolock = false;
1692 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1696 BUG_ON(!ordered_extent);
1698 nolock = (root == root->fs_info->tree_root);
1700 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1701 BUG_ON(!list_empty(&ordered_extent->list));
1702 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1705 trans = btrfs_join_transaction_nolock(root, 1);
1707 trans = btrfs_join_transaction(root, 1);
1708 BUG_ON(IS_ERR(trans));
1709 btrfs_set_trans_block_group(trans, inode);
1710 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1711 ret = btrfs_update_inode(trans, root, inode);
1717 lock_extent_bits(io_tree, ordered_extent->file_offset,
1718 ordered_extent->file_offset + ordered_extent->len - 1,
1719 0, &cached_state, GFP_NOFS);
1722 trans = btrfs_join_transaction_nolock(root, 1);
1724 trans = btrfs_join_transaction(root, 1);
1725 BUG_ON(IS_ERR(trans));
1726 btrfs_set_trans_block_group(trans, inode);
1727 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1729 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1730 compress_type = ordered_extent->compress_type;
1731 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1732 BUG_ON(compress_type);
1733 ret = btrfs_mark_extent_written(trans, inode,
1734 ordered_extent->file_offset,
1735 ordered_extent->file_offset +
1736 ordered_extent->len);
1739 BUG_ON(root == root->fs_info->tree_root);
1740 ret = insert_reserved_file_extent(trans, inode,
1741 ordered_extent->file_offset,
1742 ordered_extent->start,
1743 ordered_extent->disk_len,
1744 ordered_extent->len,
1745 ordered_extent->len,
1746 compress_type, 0, 0,
1747 BTRFS_FILE_EXTENT_REG);
1748 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1749 ordered_extent->file_offset,
1750 ordered_extent->len);
1753 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1754 ordered_extent->file_offset +
1755 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1757 add_pending_csums(trans, inode, ordered_extent->file_offset,
1758 &ordered_extent->list);
1760 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1761 ret = btrfs_update_inode(trans, root, inode);
1766 btrfs_end_transaction_nolock(trans, root);
1768 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1770 btrfs_end_transaction(trans, root);
1774 btrfs_put_ordered_extent(ordered_extent);
1775 /* once for the tree */
1776 btrfs_put_ordered_extent(ordered_extent);
1781 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1782 struct extent_state *state, int uptodate)
1784 ClearPagePrivate2(page);
1785 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1789 * When IO fails, either with EIO or csum verification fails, we
1790 * try other mirrors that might have a good copy of the data. This
1791 * io_failure_record is used to record state as we go through all the
1792 * mirrors. If another mirror has good data, the page is set up to date
1793 * and things continue. If a good mirror can't be found, the original
1794 * bio end_io callback is called to indicate things have failed.
1796 struct io_failure_record {
1801 unsigned long bio_flags;
1805 static int btrfs_io_failed_hook(struct bio *failed_bio,
1806 struct page *page, u64 start, u64 end,
1807 struct extent_state *state)
1809 struct io_failure_record *failrec = NULL;
1811 struct extent_map *em;
1812 struct inode *inode = page->mapping->host;
1813 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1814 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1821 ret = get_state_private(failure_tree, start, &private);
1823 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1826 failrec->start = start;
1827 failrec->len = end - start + 1;
1828 failrec->last_mirror = 0;
1829 failrec->bio_flags = 0;
1831 read_lock(&em_tree->lock);
1832 em = lookup_extent_mapping(em_tree, start, failrec->len);
1833 if (em->start > start || em->start + em->len < start) {
1834 free_extent_map(em);
1837 read_unlock(&em_tree->lock);
1839 if (!em || IS_ERR(em)) {
1843 logical = start - em->start;
1844 logical = em->block_start + logical;
1845 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1846 logical = em->block_start;
1847 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1848 extent_set_compress_type(&failrec->bio_flags,
1851 failrec->logical = logical;
1852 free_extent_map(em);
1853 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1854 EXTENT_DIRTY, GFP_NOFS);
1855 set_state_private(failure_tree, start,
1856 (u64)(unsigned long)failrec);
1858 failrec = (struct io_failure_record *)(unsigned long)private;
1860 num_copies = btrfs_num_copies(
1861 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1862 failrec->logical, failrec->len);
1863 failrec->last_mirror++;
1865 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1866 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1869 if (state && state->start != failrec->start)
1871 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1873 if (!state || failrec->last_mirror > num_copies) {
1874 set_state_private(failure_tree, failrec->start, 0);
1875 clear_extent_bits(failure_tree, failrec->start,
1876 failrec->start + failrec->len - 1,
1877 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1881 bio = bio_alloc(GFP_NOFS, 1);
1882 bio->bi_private = state;
1883 bio->bi_end_io = failed_bio->bi_end_io;
1884 bio->bi_sector = failrec->logical >> 9;
1885 bio->bi_bdev = failed_bio->bi_bdev;
1888 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1889 if (failed_bio->bi_rw & REQ_WRITE)
1894 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1895 failrec->last_mirror,
1896 failrec->bio_flags, 0);
1901 * each time an IO finishes, we do a fast check in the IO failure tree
1902 * to see if we need to process or clean up an io_failure_record
1904 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1907 u64 private_failure;
1908 struct io_failure_record *failure;
1912 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1913 (u64)-1, 1, EXTENT_DIRTY)) {
1914 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1915 start, &private_failure);
1917 failure = (struct io_failure_record *)(unsigned long)
1919 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1921 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1923 failure->start + failure->len - 1,
1924 EXTENT_DIRTY | EXTENT_LOCKED,
1933 * when reads are done, we need to check csums to verify the data is correct
1934 * if there's a match, we allow the bio to finish. If not, we go through
1935 * the io_failure_record routines to find good copies
1937 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1938 struct extent_state *state)
1940 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1941 struct inode *inode = page->mapping->host;
1942 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1944 u64 private = ~(u32)0;
1946 struct btrfs_root *root = BTRFS_I(inode)->root;
1949 if (PageChecked(page)) {
1950 ClearPageChecked(page);
1954 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1957 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1958 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1959 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1964 if (state && state->start == start) {
1965 private = state->private;
1968 ret = get_state_private(io_tree, start, &private);
1970 kaddr = kmap_atomic(page, KM_USER0);
1974 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1975 btrfs_csum_final(csum, (char *)&csum);
1976 if (csum != private)
1979 kunmap_atomic(kaddr, KM_USER0);
1981 /* if the io failure tree for this inode is non-empty,
1982 * check to see if we've recovered from a failed IO
1984 btrfs_clean_io_failures(inode, start);
1988 if (printk_ratelimit()) {
1989 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1990 "private %llu\n", page->mapping->host->i_ino,
1991 (unsigned long long)start, csum,
1992 (unsigned long long)private);
1994 memset(kaddr + offset, 1, end - start + 1);
1995 flush_dcache_page(page);
1996 kunmap_atomic(kaddr, KM_USER0);
2002 struct delayed_iput {
2003 struct list_head list;
2004 struct inode *inode;
2007 void btrfs_add_delayed_iput(struct inode *inode)
2009 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2010 struct delayed_iput *delayed;
2012 if (atomic_add_unless(&inode->i_count, -1, 1))
2015 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2016 delayed->inode = inode;
2018 spin_lock(&fs_info->delayed_iput_lock);
2019 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2020 spin_unlock(&fs_info->delayed_iput_lock);
2023 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2026 struct btrfs_fs_info *fs_info = root->fs_info;
2027 struct delayed_iput *delayed;
2030 spin_lock(&fs_info->delayed_iput_lock);
2031 empty = list_empty(&fs_info->delayed_iputs);
2032 spin_unlock(&fs_info->delayed_iput_lock);
2036 down_read(&root->fs_info->cleanup_work_sem);
2037 spin_lock(&fs_info->delayed_iput_lock);
2038 list_splice_init(&fs_info->delayed_iputs, &list);
2039 spin_unlock(&fs_info->delayed_iput_lock);
2041 while (!list_empty(&list)) {
2042 delayed = list_entry(list.next, struct delayed_iput, list);
2043 list_del(&delayed->list);
2044 iput(delayed->inode);
2047 up_read(&root->fs_info->cleanup_work_sem);
2051 * calculate extra metadata reservation when snapshotting a subvolume
2052 * contains orphan files.
2054 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2055 struct btrfs_pending_snapshot *pending,
2056 u64 *bytes_to_reserve)
2058 struct btrfs_root *root;
2059 struct btrfs_block_rsv *block_rsv;
2063 root = pending->root;
2064 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2067 block_rsv = root->orphan_block_rsv;
2069 /* orphan block reservation for the snapshot */
2070 num_bytes = block_rsv->size;
2073 * after the snapshot is created, COWing tree blocks may use more
2074 * space than it frees. So we should make sure there is enough
2077 index = trans->transid & 0x1;
2078 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2079 num_bytes += block_rsv->size -
2080 (block_rsv->reserved + block_rsv->freed[index]);
2083 *bytes_to_reserve += num_bytes;
2086 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2087 struct btrfs_pending_snapshot *pending)
2089 struct btrfs_root *root = pending->root;
2090 struct btrfs_root *snap = pending->snap;
2091 struct btrfs_block_rsv *block_rsv;
2096 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2099 /* refill source subvolume's orphan block reservation */
2100 block_rsv = root->orphan_block_rsv;
2101 index = trans->transid & 0x1;
2102 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2103 num_bytes = block_rsv->size -
2104 (block_rsv->reserved + block_rsv->freed[index]);
2105 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2106 root->orphan_block_rsv,
2111 /* setup orphan block reservation for the snapshot */
2112 block_rsv = btrfs_alloc_block_rsv(snap);
2115 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2116 snap->orphan_block_rsv = block_rsv;
2118 num_bytes = root->orphan_block_rsv->size;
2119 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2120 block_rsv, num_bytes);
2124 /* insert orphan item for the snapshot */
2125 WARN_ON(!root->orphan_item_inserted);
2126 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2127 snap->root_key.objectid);
2129 snap->orphan_item_inserted = 1;
2133 enum btrfs_orphan_cleanup_state {
2134 ORPHAN_CLEANUP_STARTED = 1,
2135 ORPHAN_CLEANUP_DONE = 2,
2139 * This is called in transaction commmit time. If there are no orphan
2140 * files in the subvolume, it removes orphan item and frees block_rsv
2143 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2144 struct btrfs_root *root)
2148 if (!list_empty(&root->orphan_list) ||
2149 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2152 if (root->orphan_item_inserted &&
2153 btrfs_root_refs(&root->root_item) > 0) {
2154 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2155 root->root_key.objectid);
2157 root->orphan_item_inserted = 0;
2160 if (root->orphan_block_rsv) {
2161 WARN_ON(root->orphan_block_rsv->size > 0);
2162 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2163 root->orphan_block_rsv = NULL;
2168 * This creates an orphan entry for the given inode in case something goes
2169 * wrong in the middle of an unlink/truncate.
2171 * NOTE: caller of this function should reserve 5 units of metadata for
2174 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2176 struct btrfs_root *root = BTRFS_I(inode)->root;
2177 struct btrfs_block_rsv *block_rsv = NULL;
2182 if (!root->orphan_block_rsv) {
2183 block_rsv = btrfs_alloc_block_rsv(root);
2187 spin_lock(&root->orphan_lock);
2188 if (!root->orphan_block_rsv) {
2189 root->orphan_block_rsv = block_rsv;
2190 } else if (block_rsv) {
2191 btrfs_free_block_rsv(root, block_rsv);
2195 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2196 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2199 * For proper ENOSPC handling, we should do orphan
2200 * cleanup when mounting. But this introduces backward
2201 * compatibility issue.
2203 if (!xchg(&root->orphan_item_inserted, 1))
2210 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2213 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2214 BTRFS_I(inode)->orphan_meta_reserved = 1;
2217 spin_unlock(&root->orphan_lock);
2220 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2222 /* grab metadata reservation from transaction handle */
2224 ret = btrfs_orphan_reserve_metadata(trans, inode);
2228 /* insert an orphan item to track this unlinked/truncated file */
2230 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2234 /* insert an orphan item to track subvolume contains orphan files */
2236 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2237 root->root_key.objectid);
2244 * We have done the truncate/delete so we can go ahead and remove the orphan
2245 * item for this particular inode.
2247 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2249 struct btrfs_root *root = BTRFS_I(inode)->root;
2250 int delete_item = 0;
2251 int release_rsv = 0;
2254 spin_lock(&root->orphan_lock);
2255 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2256 list_del_init(&BTRFS_I(inode)->i_orphan);
2260 if (BTRFS_I(inode)->orphan_meta_reserved) {
2261 BTRFS_I(inode)->orphan_meta_reserved = 0;
2264 spin_unlock(&root->orphan_lock);
2266 if (trans && delete_item) {
2267 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2272 btrfs_orphan_release_metadata(inode);
2278 * this cleans up any orphans that may be left on the list from the last use
2281 void btrfs_orphan_cleanup(struct btrfs_root *root)
2283 struct btrfs_path *path;
2284 struct extent_buffer *leaf;
2285 struct btrfs_key key, found_key;
2286 struct btrfs_trans_handle *trans;
2287 struct inode *inode;
2288 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2290 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2293 path = btrfs_alloc_path();
2297 key.objectid = BTRFS_ORPHAN_OBJECTID;
2298 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2299 key.offset = (u64)-1;
2302 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2304 printk(KERN_ERR "Error searching slot for orphan: %d"
2310 * if ret == 0 means we found what we were searching for, which
2311 * is weird, but possible, so only screw with path if we didnt
2312 * find the key and see if we have stuff that matches
2315 if (path->slots[0] == 0)
2320 /* pull out the item */
2321 leaf = path->nodes[0];
2322 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2324 /* make sure the item matches what we want */
2325 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2327 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2330 /* release the path since we're done with it */
2331 btrfs_release_path(root, path);
2334 * this is where we are basically btrfs_lookup, without the
2335 * crossing root thing. we store the inode number in the
2336 * offset of the orphan item.
2338 found_key.objectid = found_key.offset;
2339 found_key.type = BTRFS_INODE_ITEM_KEY;
2340 found_key.offset = 0;
2341 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2342 BUG_ON(IS_ERR(inode));
2345 * add this inode to the orphan list so btrfs_orphan_del does
2346 * the proper thing when we hit it
2348 spin_lock(&root->orphan_lock);
2349 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2350 spin_unlock(&root->orphan_lock);
2353 * if this is a bad inode, means we actually succeeded in
2354 * removing the inode, but not the orphan record, which means
2355 * we need to manually delete the orphan since iput will just
2356 * do a destroy_inode
2358 if (is_bad_inode(inode)) {
2359 trans = btrfs_start_transaction(root, 0);
2360 btrfs_orphan_del(trans, inode);
2361 btrfs_end_transaction(trans, root);
2366 /* if we have links, this was a truncate, lets do that */
2367 if (inode->i_nlink) {
2369 btrfs_truncate(inode);
2374 /* this will do delete_inode and everything for us */
2377 btrfs_free_path(path);
2379 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2381 if (root->orphan_block_rsv)
2382 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2385 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2386 trans = btrfs_join_transaction(root, 1);
2387 BUG_ON(IS_ERR(trans));
2388 btrfs_end_transaction(trans, root);
2392 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2394 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2398 * very simple check to peek ahead in the leaf looking for xattrs. If we
2399 * don't find any xattrs, we know there can't be any acls.
2401 * slot is the slot the inode is in, objectid is the objectid of the inode
2403 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2404 int slot, u64 objectid)
2406 u32 nritems = btrfs_header_nritems(leaf);
2407 struct btrfs_key found_key;
2411 while (slot < nritems) {
2412 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2414 /* we found a different objectid, there must not be acls */
2415 if (found_key.objectid != objectid)
2418 /* we found an xattr, assume we've got an acl */
2419 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2423 * we found a key greater than an xattr key, there can't
2424 * be any acls later on
2426 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2433 * it goes inode, inode backrefs, xattrs, extents,
2434 * so if there are a ton of hard links to an inode there can
2435 * be a lot of backrefs. Don't waste time searching too hard,
2436 * this is just an optimization
2441 /* we hit the end of the leaf before we found an xattr or
2442 * something larger than an xattr. We have to assume the inode
2449 * read an inode from the btree into the in-memory inode
2451 static void btrfs_read_locked_inode(struct inode *inode)
2453 struct btrfs_path *path;
2454 struct extent_buffer *leaf;
2455 struct btrfs_inode_item *inode_item;
2456 struct btrfs_timespec *tspec;
2457 struct btrfs_root *root = BTRFS_I(inode)->root;
2458 struct btrfs_key location;
2460 u64 alloc_group_block;
2464 path = btrfs_alloc_path();
2466 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2468 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2472 leaf = path->nodes[0];
2473 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2474 struct btrfs_inode_item);
2476 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2477 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2478 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2479 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2480 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2482 tspec = btrfs_inode_atime(inode_item);
2483 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2484 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2486 tspec = btrfs_inode_mtime(inode_item);
2487 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2488 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2490 tspec = btrfs_inode_ctime(inode_item);
2491 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2492 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2494 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2495 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2496 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2497 inode->i_generation = BTRFS_I(inode)->generation;
2499 rdev = btrfs_inode_rdev(leaf, inode_item);
2501 BTRFS_I(inode)->index_cnt = (u64)-1;
2502 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2504 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2507 * try to precache a NULL acl entry for files that don't have
2508 * any xattrs or acls
2510 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2512 cache_no_acl(inode);
2514 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2515 alloc_group_block, 0);
2516 btrfs_free_path(path);
2519 switch (inode->i_mode & S_IFMT) {
2521 inode->i_mapping->a_ops = &btrfs_aops;
2522 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2523 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2524 inode->i_fop = &btrfs_file_operations;
2525 inode->i_op = &btrfs_file_inode_operations;
2528 inode->i_fop = &btrfs_dir_file_operations;
2529 if (root == root->fs_info->tree_root)
2530 inode->i_op = &btrfs_dir_ro_inode_operations;
2532 inode->i_op = &btrfs_dir_inode_operations;
2535 inode->i_op = &btrfs_symlink_inode_operations;
2536 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2537 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2540 inode->i_op = &btrfs_special_inode_operations;
2541 init_special_inode(inode, inode->i_mode, rdev);
2545 btrfs_update_iflags(inode);
2549 btrfs_free_path(path);
2550 make_bad_inode(inode);
2554 * given a leaf and an inode, copy the inode fields into the leaf
2556 static void fill_inode_item(struct btrfs_trans_handle *trans,
2557 struct extent_buffer *leaf,
2558 struct btrfs_inode_item *item,
2559 struct inode *inode)
2561 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2562 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2563 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2564 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2565 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2567 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2568 inode->i_atime.tv_sec);
2569 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2570 inode->i_atime.tv_nsec);
2572 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2573 inode->i_mtime.tv_sec);
2574 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2575 inode->i_mtime.tv_nsec);
2577 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2578 inode->i_ctime.tv_sec);
2579 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2580 inode->i_ctime.tv_nsec);
2582 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2583 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2584 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2585 btrfs_set_inode_transid(leaf, item, trans->transid);
2586 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2587 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2588 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2592 * copy everything in the in-memory inode into the btree.
2594 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2595 struct btrfs_root *root, struct inode *inode)
2597 struct btrfs_inode_item *inode_item;
2598 struct btrfs_path *path;
2599 struct extent_buffer *leaf;
2602 path = btrfs_alloc_path();
2604 path->leave_spinning = 1;
2605 ret = btrfs_lookup_inode(trans, root, path,
2606 &BTRFS_I(inode)->location, 1);
2613 btrfs_unlock_up_safe(path, 1);
2614 leaf = path->nodes[0];
2615 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2616 struct btrfs_inode_item);
2618 fill_inode_item(trans, leaf, inode_item, inode);
2619 btrfs_mark_buffer_dirty(leaf);
2620 btrfs_set_inode_last_trans(trans, inode);
2623 btrfs_free_path(path);
2629 * unlink helper that gets used here in inode.c and in the tree logging
2630 * recovery code. It remove a link in a directory with a given name, and
2631 * also drops the back refs in the inode to the directory
2633 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2634 struct btrfs_root *root,
2635 struct inode *dir, struct inode *inode,
2636 const char *name, int name_len)
2638 struct btrfs_path *path;
2640 struct extent_buffer *leaf;
2641 struct btrfs_dir_item *di;
2642 struct btrfs_key key;
2645 path = btrfs_alloc_path();
2651 path->leave_spinning = 1;
2652 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2653 name, name_len, -1);
2662 leaf = path->nodes[0];
2663 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2664 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2667 btrfs_release_path(root, path);
2669 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2671 dir->i_ino, &index);
2673 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2674 "inode %lu parent %lu\n", name_len, name,
2675 inode->i_ino, dir->i_ino);
2679 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2680 index, name, name_len, -1);
2689 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2690 btrfs_release_path(root, path);
2692 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2694 BUG_ON(ret != 0 && ret != -ENOENT);
2696 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2701 btrfs_free_path(path);
2705 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2706 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2707 btrfs_update_inode(trans, root, dir);
2708 btrfs_drop_nlink(inode);
2709 ret = btrfs_update_inode(trans, root, inode);
2714 /* helper to check if there is any shared block in the path */
2715 static int check_path_shared(struct btrfs_root *root,
2716 struct btrfs_path *path)
2718 struct extent_buffer *eb;
2722 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2725 if (!path->nodes[level])
2727 eb = path->nodes[level];
2728 if (!btrfs_block_can_be_shared(root, eb))
2730 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2739 * helper to start transaction for unlink and rmdir.
2741 * unlink and rmdir are special in btrfs, they do not always free space.
2742 * so in enospc case, we should make sure they will free space before
2743 * allowing them to use the global metadata reservation.
2745 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2746 struct dentry *dentry)
2748 struct btrfs_trans_handle *trans;
2749 struct btrfs_root *root = BTRFS_I(dir)->root;
2750 struct btrfs_path *path;
2751 struct btrfs_inode_ref *ref;
2752 struct btrfs_dir_item *di;
2753 struct inode *inode = dentry->d_inode;
2759 trans = btrfs_start_transaction(root, 10);
2760 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2763 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2764 return ERR_PTR(-ENOSPC);
2766 /* check if there is someone else holds reference */
2767 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2768 return ERR_PTR(-ENOSPC);
2770 if (atomic_read(&inode->i_count) > 2)
2771 return ERR_PTR(-ENOSPC);
2773 if (xchg(&root->fs_info->enospc_unlink, 1))
2774 return ERR_PTR(-ENOSPC);
2776 path = btrfs_alloc_path();
2778 root->fs_info->enospc_unlink = 0;
2779 return ERR_PTR(-ENOMEM);
2782 trans = btrfs_start_transaction(root, 0);
2783 if (IS_ERR(trans)) {
2784 btrfs_free_path(path);
2785 root->fs_info->enospc_unlink = 0;
2789 path->skip_locking = 1;
2790 path->search_commit_root = 1;
2792 ret = btrfs_lookup_inode(trans, root, path,
2793 &BTRFS_I(dir)->location, 0);
2799 if (check_path_shared(root, path))
2804 btrfs_release_path(root, path);
2806 ret = btrfs_lookup_inode(trans, root, path,
2807 &BTRFS_I(inode)->location, 0);
2813 if (check_path_shared(root, path))
2818 btrfs_release_path(root, path);
2820 if (ret == 0 && S_ISREG(inode->i_mode)) {
2821 ret = btrfs_lookup_file_extent(trans, root, path,
2822 inode->i_ino, (u64)-1, 0);
2828 if (check_path_shared(root, path))
2830 btrfs_release_path(root, path);
2838 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2839 dentry->d_name.name, dentry->d_name.len, 0);
2845 if (check_path_shared(root, path))
2851 btrfs_release_path(root, path);
2853 ref = btrfs_lookup_inode_ref(trans, root, path,
2854 dentry->d_name.name, dentry->d_name.len,
2855 inode->i_ino, dir->i_ino, 0);
2861 if (check_path_shared(root, path))
2863 index = btrfs_inode_ref_index(path->nodes[0], ref);
2864 btrfs_release_path(root, path);
2866 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2867 dentry->d_name.name, dentry->d_name.len, 0);
2872 BUG_ON(ret == -ENOENT);
2873 if (check_path_shared(root, path))
2878 btrfs_free_path(path);
2880 btrfs_end_transaction(trans, root);
2881 root->fs_info->enospc_unlink = 0;
2882 return ERR_PTR(err);
2885 trans->block_rsv = &root->fs_info->global_block_rsv;
2889 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2890 struct btrfs_root *root)
2892 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2893 BUG_ON(!root->fs_info->enospc_unlink);
2894 root->fs_info->enospc_unlink = 0;
2896 btrfs_end_transaction_throttle(trans, root);
2899 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2901 struct btrfs_root *root = BTRFS_I(dir)->root;
2902 struct btrfs_trans_handle *trans;
2903 struct inode *inode = dentry->d_inode;
2905 unsigned long nr = 0;
2907 trans = __unlink_start_trans(dir, dentry);
2909 return PTR_ERR(trans);
2911 btrfs_set_trans_block_group(trans, dir);
2913 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2915 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2916 dentry->d_name.name, dentry->d_name.len);
2919 if (inode->i_nlink == 0) {
2920 ret = btrfs_orphan_add(trans, inode);
2924 nr = trans->blocks_used;
2925 __unlink_end_trans(trans, root);
2926 btrfs_btree_balance_dirty(root, nr);
2930 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2931 struct btrfs_root *root,
2932 struct inode *dir, u64 objectid,
2933 const char *name, int name_len)
2935 struct btrfs_path *path;
2936 struct extent_buffer *leaf;
2937 struct btrfs_dir_item *di;
2938 struct btrfs_key key;
2942 path = btrfs_alloc_path();
2946 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2947 name, name_len, -1);
2948 BUG_ON(!di || IS_ERR(di));
2950 leaf = path->nodes[0];
2951 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2952 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2953 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2955 btrfs_release_path(root, path);
2957 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2958 objectid, root->root_key.objectid,
2959 dir->i_ino, &index, name, name_len);
2961 BUG_ON(ret != -ENOENT);
2962 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2964 BUG_ON(!di || IS_ERR(di));
2966 leaf = path->nodes[0];
2967 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2968 btrfs_release_path(root, path);
2972 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2973 index, name, name_len, -1);
2974 BUG_ON(!di || IS_ERR(di));
2976 leaf = path->nodes[0];
2977 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2978 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2979 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2981 btrfs_release_path(root, path);
2983 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2984 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2985 ret = btrfs_update_inode(trans, root, dir);
2988 btrfs_free_path(path);
2992 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2994 struct inode *inode = dentry->d_inode;
2996 struct btrfs_root *root = BTRFS_I(dir)->root;
2997 struct btrfs_trans_handle *trans;
2998 unsigned long nr = 0;
3000 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3001 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3004 trans = __unlink_start_trans(dir, dentry);
3006 return PTR_ERR(trans);
3008 btrfs_set_trans_block_group(trans, dir);
3010 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3011 err = btrfs_unlink_subvol(trans, root, dir,
3012 BTRFS_I(inode)->location.objectid,
3013 dentry->d_name.name,
3014 dentry->d_name.len);
3018 err = btrfs_orphan_add(trans, inode);
3022 /* now the directory is empty */
3023 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3024 dentry->d_name.name, dentry->d_name.len);
3026 btrfs_i_size_write(inode, 0);
3028 nr = trans->blocks_used;
3029 __unlink_end_trans(trans, root);
3030 btrfs_btree_balance_dirty(root, nr);
3037 * when truncating bytes in a file, it is possible to avoid reading
3038 * the leaves that contain only checksum items. This can be the
3039 * majority of the IO required to delete a large file, but it must
3040 * be done carefully.
3042 * The keys in the level just above the leaves are checked to make sure
3043 * the lowest key in a given leaf is a csum key, and starts at an offset
3044 * after the new size.
3046 * Then the key for the next leaf is checked to make sure it also has
3047 * a checksum item for the same file. If it does, we know our target leaf
3048 * contains only checksum items, and it can be safely freed without reading
3051 * This is just an optimization targeted at large files. It may do
3052 * nothing. It will return 0 unless things went badly.
3054 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3055 struct btrfs_root *root,
3056 struct btrfs_path *path,
3057 struct inode *inode, u64 new_size)
3059 struct btrfs_key key;
3062 struct btrfs_key found_key;
3063 struct btrfs_key other_key;
3064 struct btrfs_leaf_ref *ref;
3068 path->lowest_level = 1;
3069 key.objectid = inode->i_ino;
3070 key.type = BTRFS_CSUM_ITEM_KEY;
3071 key.offset = new_size;
3073 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3077 if (path->nodes[1] == NULL) {
3082 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3083 nritems = btrfs_header_nritems(path->nodes[1]);
3088 if (path->slots[1] >= nritems)
3091 /* did we find a key greater than anything we want to delete? */
3092 if (found_key.objectid > inode->i_ino ||
3093 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3096 /* we check the next key in the node to make sure the leave contains
3097 * only checksum items. This comparison doesn't work if our
3098 * leaf is the last one in the node
3100 if (path->slots[1] + 1 >= nritems) {
3102 /* search forward from the last key in the node, this
3103 * will bring us into the next node in the tree
3105 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3107 /* unlikely, but we inc below, so check to be safe */
3108 if (found_key.offset == (u64)-1)
3111 /* search_forward needs a path with locks held, do the
3112 * search again for the original key. It is possible
3113 * this will race with a balance and return a path that
3114 * we could modify, but this drop is just an optimization
3115 * and is allowed to miss some leaves.
3117 btrfs_release_path(root, path);
3120 /* setup a max key for search_forward */
3121 other_key.offset = (u64)-1;
3122 other_key.type = key.type;
3123 other_key.objectid = key.objectid;
3125 path->keep_locks = 1;
3126 ret = btrfs_search_forward(root, &found_key, &other_key,
3128 path->keep_locks = 0;
3129 if (ret || found_key.objectid != key.objectid ||
3130 found_key.type != key.type) {
3135 key.offset = found_key.offset;
3136 btrfs_release_path(root, path);
3141 /* we know there's one more slot after us in the tree,
3142 * read that key so we can verify it is also a checksum item
3144 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3146 if (found_key.objectid < inode->i_ino)
3149 if (found_key.type != key.type || found_key.offset < new_size)
3153 * if the key for the next leaf isn't a csum key from this objectid,
3154 * we can't be sure there aren't good items inside this leaf.
3157 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3160 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3161 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3163 * it is safe to delete this leaf, it contains only
3164 * csum items from this inode at an offset >= new_size
3166 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3169 if (root->ref_cows && leaf_gen < trans->transid) {
3170 ref = btrfs_alloc_leaf_ref(root, 0);
3172 ref->root_gen = root->root_key.offset;
3173 ref->bytenr = leaf_start;
3175 ref->generation = leaf_gen;
3178 btrfs_sort_leaf_ref(ref);
3180 ret = btrfs_add_leaf_ref(root, ref, 0);
3182 btrfs_free_leaf_ref(root, ref);
3188 btrfs_release_path(root, path);
3190 if (other_key.objectid == inode->i_ino &&
3191 other_key.type == key.type && other_key.offset > key.offset) {
3192 key.offset = other_key.offset;
3198 /* fixup any changes we've made to the path */
3199 path->lowest_level = 0;
3200 path->keep_locks = 0;
3201 btrfs_release_path(root, path);
3208 * this can truncate away extent items, csum items and directory items.
3209 * It starts at a high offset and removes keys until it can't find
3210 * any higher than new_size
3212 * csum items that cross the new i_size are truncated to the new size
3215 * min_type is the minimum key type to truncate down to. If set to 0, this
3216 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3218 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3219 struct btrfs_root *root,
3220 struct inode *inode,
3221 u64 new_size, u32 min_type)
3223 struct btrfs_path *path;
3224 struct extent_buffer *leaf;
3225 struct btrfs_file_extent_item *fi;
3226 struct btrfs_key key;
3227 struct btrfs_key found_key;
3228 u64 extent_start = 0;
3229 u64 extent_num_bytes = 0;
3230 u64 extent_offset = 0;
3232 u64 mask = root->sectorsize - 1;
3233 u32 found_type = (u8)-1;
3236 int pending_del_nr = 0;
3237 int pending_del_slot = 0;
3238 int extent_type = -1;
3243 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3245 if (root->ref_cows || root == root->fs_info->tree_root)
3246 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3248 path = btrfs_alloc_path();
3252 key.objectid = inode->i_ino;
3253 key.offset = (u64)-1;
3257 path->leave_spinning = 1;
3258 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3265 /* there are no items in the tree for us to truncate, we're
3268 if (path->slots[0] == 0)
3275 leaf = path->nodes[0];
3276 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3277 found_type = btrfs_key_type(&found_key);
3280 if (found_key.objectid != inode->i_ino)
3283 if (found_type < min_type)
3286 item_end = found_key.offset;
3287 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3288 fi = btrfs_item_ptr(leaf, path->slots[0],
3289 struct btrfs_file_extent_item);
3290 extent_type = btrfs_file_extent_type(leaf, fi);
3291 encoding = btrfs_file_extent_compression(leaf, fi);
3292 encoding |= btrfs_file_extent_encryption(leaf, fi);
3293 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3295 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3297 btrfs_file_extent_num_bytes(leaf, fi);
3298 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3299 item_end += btrfs_file_extent_inline_len(leaf,
3304 if (found_type > min_type) {
3307 if (item_end < new_size)
3309 if (found_key.offset >= new_size)
3315 /* FIXME, shrink the extent if the ref count is only 1 */
3316 if (found_type != BTRFS_EXTENT_DATA_KEY)
3319 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3321 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3322 if (!del_item && !encoding) {
3323 u64 orig_num_bytes =
3324 btrfs_file_extent_num_bytes(leaf, fi);
3325 extent_num_bytes = new_size -
3326 found_key.offset + root->sectorsize - 1;
3327 extent_num_bytes = extent_num_bytes &
3328 ~((u64)root->sectorsize - 1);
3329 btrfs_set_file_extent_num_bytes(leaf, fi,
3331 num_dec = (orig_num_bytes -
3333 if (root->ref_cows && extent_start != 0)
3334 inode_sub_bytes(inode, num_dec);
3335 btrfs_mark_buffer_dirty(leaf);
3338 btrfs_file_extent_disk_num_bytes(leaf,
3340 extent_offset = found_key.offset -
3341 btrfs_file_extent_offset(leaf, fi);
3343 /* FIXME blocksize != 4096 */
3344 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3345 if (extent_start != 0) {
3348 inode_sub_bytes(inode, num_dec);
3351 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3353 * we can't truncate inline items that have had
3357 btrfs_file_extent_compression(leaf, fi) == 0 &&
3358 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3359 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3360 u32 size = new_size - found_key.offset;
3362 if (root->ref_cows) {
3363 inode_sub_bytes(inode, item_end + 1 -
3367 btrfs_file_extent_calc_inline_size(size);
3368 ret = btrfs_truncate_item(trans, root, path,
3371 } else if (root->ref_cows) {
3372 inode_sub_bytes(inode, item_end + 1 -
3378 if (!pending_del_nr) {
3379 /* no pending yet, add ourselves */
3380 pending_del_slot = path->slots[0];
3382 } else if (pending_del_nr &&
3383 path->slots[0] + 1 == pending_del_slot) {
3384 /* hop on the pending chunk */
3386 pending_del_slot = path->slots[0];
3393 if (found_extent && (root->ref_cows ||
3394 root == root->fs_info->tree_root)) {
3395 btrfs_set_path_blocking(path);
3396 ret = btrfs_free_extent(trans, root, extent_start,
3397 extent_num_bytes, 0,
3398 btrfs_header_owner(leaf),
3399 inode->i_ino, extent_offset);
3403 if (found_type == BTRFS_INODE_ITEM_KEY)
3406 if (path->slots[0] == 0 ||
3407 path->slots[0] != pending_del_slot) {
3408 if (root->ref_cows) {
3412 if (pending_del_nr) {
3413 ret = btrfs_del_items(trans, root, path,
3419 btrfs_release_path(root, path);
3426 if (pending_del_nr) {
3427 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3431 btrfs_free_path(path);
3436 * taken from block_truncate_page, but does cow as it zeros out
3437 * any bytes left in the last page in the file.
3439 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3441 struct inode *inode = mapping->host;
3442 struct btrfs_root *root = BTRFS_I(inode)->root;
3443 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3444 struct btrfs_ordered_extent *ordered;
3445 struct extent_state *cached_state = NULL;
3447 u32 blocksize = root->sectorsize;
3448 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3449 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3455 if ((offset & (blocksize - 1)) == 0)
3457 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3463 page = grab_cache_page(mapping, index);
3465 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3469 page_start = page_offset(page);
3470 page_end = page_start + PAGE_CACHE_SIZE - 1;
3472 if (!PageUptodate(page)) {
3473 ret = btrfs_readpage(NULL, page);
3475 if (page->mapping != mapping) {
3477 page_cache_release(page);
3480 if (!PageUptodate(page)) {
3485 wait_on_page_writeback(page);
3487 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3489 set_page_extent_mapped(page);
3491 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3493 unlock_extent_cached(io_tree, page_start, page_end,
3494 &cached_state, GFP_NOFS);
3496 page_cache_release(page);
3497 btrfs_start_ordered_extent(inode, ordered, 1);
3498 btrfs_put_ordered_extent(ordered);
3502 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3503 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3504 0, 0, &cached_state, GFP_NOFS);
3506 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3509 unlock_extent_cached(io_tree, page_start, page_end,
3510 &cached_state, GFP_NOFS);
3515 if (offset != PAGE_CACHE_SIZE) {
3517 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3518 flush_dcache_page(page);
3521 ClearPageChecked(page);
3522 set_page_dirty(page);
3523 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3528 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3530 page_cache_release(page);
3535 int btrfs_cont_expand(struct inode *inode, loff_t size)
3537 struct btrfs_trans_handle *trans;
3538 struct btrfs_root *root = BTRFS_I(inode)->root;
3539 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3540 struct extent_map *em = NULL;
3541 struct extent_state *cached_state = NULL;
3542 u64 mask = root->sectorsize - 1;
3543 u64 hole_start = (inode->i_size + mask) & ~mask;
3544 u64 block_end = (size + mask) & ~mask;
3550 if (size <= hole_start)
3554 struct btrfs_ordered_extent *ordered;
3555 btrfs_wait_ordered_range(inode, hole_start,
3556 block_end - hole_start);
3557 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3558 &cached_state, GFP_NOFS);
3559 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3562 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3563 &cached_state, GFP_NOFS);
3564 btrfs_put_ordered_extent(ordered);
3567 cur_offset = hole_start;
3569 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3570 block_end - cur_offset, 0);
3571 BUG_ON(IS_ERR(em) || !em);
3572 last_byte = min(extent_map_end(em), block_end);
3573 last_byte = (last_byte + mask) & ~mask;
3574 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3576 hole_size = last_byte - cur_offset;
3578 trans = btrfs_start_transaction(root, 2);
3579 if (IS_ERR(trans)) {
3580 err = PTR_ERR(trans);
3583 btrfs_set_trans_block_group(trans, inode);
3585 err = btrfs_drop_extents(trans, inode, cur_offset,
3586 cur_offset + hole_size,
3590 err = btrfs_insert_file_extent(trans, root,
3591 inode->i_ino, cur_offset, 0,
3592 0, hole_size, 0, hole_size,
3596 btrfs_drop_extent_cache(inode, hole_start,
3599 btrfs_end_transaction(trans, root);
3601 free_extent_map(em);
3603 cur_offset = last_byte;
3604 if (cur_offset >= block_end)
3608 free_extent_map(em);
3609 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3614 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3616 struct btrfs_root *root = BTRFS_I(inode)->root;
3617 struct btrfs_trans_handle *trans;
3621 if (attr->ia_size == inode->i_size)
3624 if (attr->ia_size > inode->i_size) {
3625 unsigned long limit;
3626 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3627 if (attr->ia_size > inode->i_sb->s_maxbytes)
3629 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3630 send_sig(SIGXFSZ, current, 0);
3635 trans = btrfs_start_transaction(root, 5);
3637 return PTR_ERR(trans);
3639 btrfs_set_trans_block_group(trans, inode);
3641 ret = btrfs_orphan_add(trans, inode);
3644 nr = trans->blocks_used;
3645 btrfs_end_transaction(trans, root);
3646 btrfs_btree_balance_dirty(root, nr);
3648 if (attr->ia_size > inode->i_size) {
3649 ret = btrfs_cont_expand(inode, attr->ia_size);
3651 btrfs_truncate(inode);
3655 i_size_write(inode, attr->ia_size);
3656 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3658 trans = btrfs_start_transaction(root, 0);
3659 BUG_ON(IS_ERR(trans));
3660 btrfs_set_trans_block_group(trans, inode);
3661 trans->block_rsv = root->orphan_block_rsv;
3662 BUG_ON(!trans->block_rsv);
3664 ret = btrfs_update_inode(trans, root, inode);
3666 if (inode->i_nlink > 0) {
3667 ret = btrfs_orphan_del(trans, inode);
3670 nr = trans->blocks_used;
3671 btrfs_end_transaction(trans, root);
3672 btrfs_btree_balance_dirty(root, nr);
3677 * We're truncating a file that used to have good data down to
3678 * zero. Make sure it gets into the ordered flush list so that
3679 * any new writes get down to disk quickly.
3681 if (attr->ia_size == 0)
3682 BTRFS_I(inode)->ordered_data_close = 1;
3684 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3685 ret = vmtruncate(inode, attr->ia_size);
3691 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3693 struct inode *inode = dentry->d_inode;
3694 struct btrfs_root *root = BTRFS_I(inode)->root;
3697 if (btrfs_root_readonly(root))
3700 err = inode_change_ok(inode, attr);
3704 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3705 err = btrfs_setattr_size(inode, attr);
3710 if (attr->ia_valid) {
3711 setattr_copy(inode, attr);
3712 mark_inode_dirty(inode);
3714 if (attr->ia_valid & ATTR_MODE)
3715 err = btrfs_acl_chmod(inode);
3721 void btrfs_evict_inode(struct inode *inode)
3723 struct btrfs_trans_handle *trans;
3724 struct btrfs_root *root = BTRFS_I(inode)->root;
3728 truncate_inode_pages(&inode->i_data, 0);
3729 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3730 root == root->fs_info->tree_root))
3733 if (is_bad_inode(inode)) {
3734 btrfs_orphan_del(NULL, inode);
3737 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3738 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3740 if (root->fs_info->log_root_recovering) {
3741 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3745 if (inode->i_nlink > 0) {
3746 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3750 btrfs_i_size_write(inode, 0);
3753 trans = btrfs_start_transaction(root, 0);
3754 BUG_ON(IS_ERR(trans));
3755 btrfs_set_trans_block_group(trans, inode);
3756 trans->block_rsv = root->orphan_block_rsv;
3758 ret = btrfs_block_rsv_check(trans, root,
3759 root->orphan_block_rsv, 0, 5);
3761 BUG_ON(ret != -EAGAIN);
3762 ret = btrfs_commit_transaction(trans, root);
3767 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3771 nr = trans->blocks_used;
3772 btrfs_end_transaction(trans, root);
3774 btrfs_btree_balance_dirty(root, nr);
3779 ret = btrfs_orphan_del(trans, inode);
3783 nr = trans->blocks_used;
3784 btrfs_end_transaction(trans, root);
3785 btrfs_btree_balance_dirty(root, nr);
3787 end_writeback(inode);
3792 * this returns the key found in the dir entry in the location pointer.
3793 * If no dir entries were found, location->objectid is 0.
3795 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3796 struct btrfs_key *location)
3798 const char *name = dentry->d_name.name;
3799 int namelen = dentry->d_name.len;
3800 struct btrfs_dir_item *di;
3801 struct btrfs_path *path;
3802 struct btrfs_root *root = BTRFS_I(dir)->root;
3805 path = btrfs_alloc_path();
3808 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3813 if (!di || IS_ERR(di))
3816 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3818 btrfs_free_path(path);
3821 location->objectid = 0;
3826 * when we hit a tree root in a directory, the btrfs part of the inode
3827 * needs to be changed to reflect the root directory of the tree root. This
3828 * is kind of like crossing a mount point.
3830 static int fixup_tree_root_location(struct btrfs_root *root,
3832 struct dentry *dentry,
3833 struct btrfs_key *location,
3834 struct btrfs_root **sub_root)
3836 struct btrfs_path *path;
3837 struct btrfs_root *new_root;
3838 struct btrfs_root_ref *ref;
3839 struct extent_buffer *leaf;
3843 path = btrfs_alloc_path();
3850 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3851 BTRFS_I(dir)->root->root_key.objectid,
3852 location->objectid);
3859 leaf = path->nodes[0];
3860 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3861 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3862 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3865 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3866 (unsigned long)(ref + 1),
3867 dentry->d_name.len);
3871 btrfs_release_path(root->fs_info->tree_root, path);
3873 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3874 if (IS_ERR(new_root)) {
3875 err = PTR_ERR(new_root);
3879 if (btrfs_root_refs(&new_root->root_item) == 0) {
3884 *sub_root = new_root;
3885 location->objectid = btrfs_root_dirid(&new_root->root_item);
3886 location->type = BTRFS_INODE_ITEM_KEY;
3887 location->offset = 0;
3890 btrfs_free_path(path);
3894 static void inode_tree_add(struct inode *inode)
3896 struct btrfs_root *root = BTRFS_I(inode)->root;
3897 struct btrfs_inode *entry;
3899 struct rb_node *parent;
3901 p = &root->inode_tree.rb_node;
3904 if (hlist_unhashed(&inode->i_hash))
3907 spin_lock(&root->inode_lock);
3910 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3912 if (inode->i_ino < entry->vfs_inode.i_ino)
3913 p = &parent->rb_left;
3914 else if (inode->i_ino > entry->vfs_inode.i_ino)
3915 p = &parent->rb_right;
3917 WARN_ON(!(entry->vfs_inode.i_state &
3918 (I_WILL_FREE | I_FREEING)));
3919 rb_erase(parent, &root->inode_tree);
3920 RB_CLEAR_NODE(parent);
3921 spin_unlock(&root->inode_lock);
3925 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3926 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3927 spin_unlock(&root->inode_lock);
3930 static void inode_tree_del(struct inode *inode)
3932 struct btrfs_root *root = BTRFS_I(inode)->root;
3935 spin_lock(&root->inode_lock);
3936 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3937 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3938 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3939 empty = RB_EMPTY_ROOT(&root->inode_tree);
3941 spin_unlock(&root->inode_lock);
3944 * Free space cache has inodes in the tree root, but the tree root has a
3945 * root_refs of 0, so this could end up dropping the tree root as a
3946 * snapshot, so we need the extra !root->fs_info->tree_root check to
3947 * make sure we don't drop it.
3949 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3950 root != root->fs_info->tree_root) {
3951 synchronize_srcu(&root->fs_info->subvol_srcu);
3952 spin_lock(&root->inode_lock);
3953 empty = RB_EMPTY_ROOT(&root->inode_tree);
3954 spin_unlock(&root->inode_lock);
3956 btrfs_add_dead_root(root);
3960 int btrfs_invalidate_inodes(struct btrfs_root *root)
3962 struct rb_node *node;
3963 struct rb_node *prev;
3964 struct btrfs_inode *entry;
3965 struct inode *inode;
3968 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3970 spin_lock(&root->inode_lock);
3972 node = root->inode_tree.rb_node;
3976 entry = rb_entry(node, struct btrfs_inode, rb_node);
3978 if (objectid < entry->vfs_inode.i_ino)
3979 node = node->rb_left;
3980 else if (objectid > entry->vfs_inode.i_ino)
3981 node = node->rb_right;
3987 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3988 if (objectid <= entry->vfs_inode.i_ino) {
3992 prev = rb_next(prev);
3996 entry = rb_entry(node, struct btrfs_inode, rb_node);
3997 objectid = entry->vfs_inode.i_ino + 1;
3998 inode = igrab(&entry->vfs_inode);
4000 spin_unlock(&root->inode_lock);
4001 if (atomic_read(&inode->i_count) > 1)
4002 d_prune_aliases(inode);
4004 * btrfs_drop_inode will have it removed from
4005 * the inode cache when its usage count
4010 spin_lock(&root->inode_lock);
4014 if (cond_resched_lock(&root->inode_lock))
4017 node = rb_next(node);
4019 spin_unlock(&root->inode_lock);
4023 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4025 struct btrfs_iget_args *args = p;
4026 inode->i_ino = args->ino;
4027 BTRFS_I(inode)->root = args->root;
4028 btrfs_set_inode_space_info(args->root, inode);
4032 static int btrfs_find_actor(struct inode *inode, void *opaque)
4034 struct btrfs_iget_args *args = opaque;
4035 return args->ino == inode->i_ino &&
4036 args->root == BTRFS_I(inode)->root;
4039 static struct inode *btrfs_iget_locked(struct super_block *s,
4041 struct btrfs_root *root)
4043 struct inode *inode;
4044 struct btrfs_iget_args args;
4045 args.ino = objectid;
4048 inode = iget5_locked(s, objectid, btrfs_find_actor,
4049 btrfs_init_locked_inode,
4054 /* Get an inode object given its location and corresponding root.
4055 * Returns in *is_new if the inode was read from disk
4057 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4058 struct btrfs_root *root, int *new)
4060 struct inode *inode;
4062 inode = btrfs_iget_locked(s, location->objectid, root);
4064 return ERR_PTR(-ENOMEM);
4066 if (inode->i_state & I_NEW) {
4067 BTRFS_I(inode)->root = root;
4068 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4069 btrfs_read_locked_inode(inode);
4071 inode_tree_add(inode);
4072 unlock_new_inode(inode);
4080 static struct inode *new_simple_dir(struct super_block *s,
4081 struct btrfs_key *key,
4082 struct btrfs_root *root)
4084 struct inode *inode = new_inode(s);
4087 return ERR_PTR(-ENOMEM);
4089 BTRFS_I(inode)->root = root;
4090 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4091 BTRFS_I(inode)->dummy_inode = 1;
4093 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4094 inode->i_op = &simple_dir_inode_operations;
4095 inode->i_fop = &simple_dir_operations;
4096 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4097 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4102 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4104 struct inode *inode;
4105 struct btrfs_root *root = BTRFS_I(dir)->root;
4106 struct btrfs_root *sub_root = root;
4107 struct btrfs_key location;
4111 dentry->d_op = &btrfs_dentry_operations;
4113 if (dentry->d_name.len > BTRFS_NAME_LEN)
4114 return ERR_PTR(-ENAMETOOLONG);
4116 ret = btrfs_inode_by_name(dir, dentry, &location);
4119 return ERR_PTR(ret);
4121 if (location.objectid == 0)
4124 if (location.type == BTRFS_INODE_ITEM_KEY) {
4125 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4129 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4131 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4132 ret = fixup_tree_root_location(root, dir, dentry,
4133 &location, &sub_root);
4136 inode = ERR_PTR(ret);
4138 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4140 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4142 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4144 if (!IS_ERR(inode) && root != sub_root) {
4145 down_read(&root->fs_info->cleanup_work_sem);
4146 if (!(inode->i_sb->s_flags & MS_RDONLY))
4147 btrfs_orphan_cleanup(sub_root);
4148 up_read(&root->fs_info->cleanup_work_sem);
4154 static int btrfs_dentry_delete(struct dentry *dentry)
4156 struct btrfs_root *root;
4158 if (!dentry->d_inode && !IS_ROOT(dentry))
4159 dentry = dentry->d_parent;
4161 if (dentry->d_inode) {
4162 root = BTRFS_I(dentry->d_inode)->root;
4163 if (btrfs_root_refs(&root->root_item) == 0)
4169 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4170 struct nameidata *nd)
4172 struct inode *inode;
4174 inode = btrfs_lookup_dentry(dir, dentry);
4176 return ERR_CAST(inode);
4178 return d_splice_alias(inode, dentry);
4181 static unsigned char btrfs_filetype_table[] = {
4182 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4185 static int btrfs_real_readdir(struct file *filp, void *dirent,
4188 struct inode *inode = filp->f_dentry->d_inode;
4189 struct btrfs_root *root = BTRFS_I(inode)->root;
4190 struct btrfs_item *item;
4191 struct btrfs_dir_item *di;
4192 struct btrfs_key key;
4193 struct btrfs_key found_key;
4194 struct btrfs_path *path;
4197 struct extent_buffer *leaf;
4200 unsigned char d_type;
4205 int key_type = BTRFS_DIR_INDEX_KEY;
4210 /* FIXME, use a real flag for deciding about the key type */
4211 if (root->fs_info->tree_root == root)
4212 key_type = BTRFS_DIR_ITEM_KEY;
4214 /* special case for "." */
4215 if (filp->f_pos == 0) {
4216 over = filldir(dirent, ".", 1,
4223 /* special case for .., just use the back ref */
4224 if (filp->f_pos == 1) {
4225 u64 pino = parent_ino(filp->f_path.dentry);
4226 over = filldir(dirent, "..", 2,
4232 path = btrfs_alloc_path();
4235 btrfs_set_key_type(&key, key_type);
4236 key.offset = filp->f_pos;
4237 key.objectid = inode->i_ino;
4239 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4245 leaf = path->nodes[0];
4246 nritems = btrfs_header_nritems(leaf);
4247 slot = path->slots[0];
4248 if (advance || slot >= nritems) {
4249 if (slot >= nritems - 1) {
4250 ret = btrfs_next_leaf(root, path);
4253 leaf = path->nodes[0];
4254 nritems = btrfs_header_nritems(leaf);
4255 slot = path->slots[0];
4263 item = btrfs_item_nr(leaf, slot);
4264 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4266 if (found_key.objectid != key.objectid)
4268 if (btrfs_key_type(&found_key) != key_type)
4270 if (found_key.offset < filp->f_pos)
4273 filp->f_pos = found_key.offset;
4275 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4277 di_total = btrfs_item_size(leaf, item);
4279 while (di_cur < di_total) {
4280 struct btrfs_key location;
4282 name_len = btrfs_dir_name_len(leaf, di);
4283 if (name_len <= sizeof(tmp_name)) {
4284 name_ptr = tmp_name;
4286 name_ptr = kmalloc(name_len, GFP_NOFS);
4292 read_extent_buffer(leaf, name_ptr,
4293 (unsigned long)(di + 1), name_len);
4295 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4296 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4298 /* is this a reference to our own snapshot? If so
4301 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4302 location.objectid == root->root_key.objectid) {
4306 over = filldir(dirent, name_ptr, name_len,
4307 found_key.offset, location.objectid,
4311 if (name_ptr != tmp_name)
4316 di_len = btrfs_dir_name_len(leaf, di) +
4317 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4319 di = (struct btrfs_dir_item *)((char *)di + di_len);
4323 /* Reached end of directory/root. Bump pos past the last item. */
4324 if (key_type == BTRFS_DIR_INDEX_KEY)
4326 * 32-bit glibc will use getdents64, but then strtol -
4327 * so the last number we can serve is this.
4329 filp->f_pos = 0x7fffffff;
4335 btrfs_free_path(path);
4339 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4341 struct btrfs_root *root = BTRFS_I(inode)->root;
4342 struct btrfs_trans_handle *trans;
4344 bool nolock = false;
4346 if (BTRFS_I(inode)->dummy_inode)
4350 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4352 if (wbc->sync_mode == WB_SYNC_ALL) {
4354 trans = btrfs_join_transaction_nolock(root, 1);
4356 trans = btrfs_join_transaction(root, 1);
4358 return PTR_ERR(trans);
4359 btrfs_set_trans_block_group(trans, inode);
4361 ret = btrfs_end_transaction_nolock(trans, root);
4363 ret = btrfs_commit_transaction(trans, root);
4369 * This is somewhat expensive, updating the tree every time the
4370 * inode changes. But, it is most likely to find the inode in cache.
4371 * FIXME, needs more benchmarking...there are no reasons other than performance
4372 * to keep or drop this code.
4374 void btrfs_dirty_inode(struct inode *inode)
4376 struct btrfs_root *root = BTRFS_I(inode)->root;
4377 struct btrfs_trans_handle *trans;
4380 if (BTRFS_I(inode)->dummy_inode)
4383 trans = btrfs_join_transaction(root, 1);
4384 BUG_ON(IS_ERR(trans));
4385 btrfs_set_trans_block_group(trans, inode);
4387 ret = btrfs_update_inode(trans, root, inode);
4388 if (ret && ret == -ENOSPC) {
4389 /* whoops, lets try again with the full transaction */
4390 btrfs_end_transaction(trans, root);
4391 trans = btrfs_start_transaction(root, 1);
4392 if (IS_ERR(trans)) {
4393 if (printk_ratelimit()) {
4394 printk(KERN_ERR "btrfs: fail to "
4395 "dirty inode %lu error %ld\n",
4396 inode->i_ino, PTR_ERR(trans));
4400 btrfs_set_trans_block_group(trans, inode);
4402 ret = btrfs_update_inode(trans, root, inode);
4404 if (printk_ratelimit()) {
4405 printk(KERN_ERR "btrfs: fail to "
4406 "dirty inode %lu error %d\n",
4411 btrfs_end_transaction(trans, root);
4415 * find the highest existing sequence number in a directory
4416 * and then set the in-memory index_cnt variable to reflect
4417 * free sequence numbers
4419 static int btrfs_set_inode_index_count(struct inode *inode)
4421 struct btrfs_root *root = BTRFS_I(inode)->root;
4422 struct btrfs_key key, found_key;
4423 struct btrfs_path *path;
4424 struct extent_buffer *leaf;
4427 key.objectid = inode->i_ino;
4428 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4429 key.offset = (u64)-1;
4431 path = btrfs_alloc_path();
4435 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4438 /* FIXME: we should be able to handle this */
4444 * MAGIC NUMBER EXPLANATION:
4445 * since we search a directory based on f_pos we have to start at 2
4446 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4447 * else has to start at 2
4449 if (path->slots[0] == 0) {
4450 BTRFS_I(inode)->index_cnt = 2;
4456 leaf = path->nodes[0];
4457 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4459 if (found_key.objectid != inode->i_ino ||
4460 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4461 BTRFS_I(inode)->index_cnt = 2;
4465 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4467 btrfs_free_path(path);
4472 * helper to find a free sequence number in a given directory. This current
4473 * code is very simple, later versions will do smarter things in the btree
4475 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4479 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4480 ret = btrfs_set_inode_index_count(dir);
4485 *index = BTRFS_I(dir)->index_cnt;
4486 BTRFS_I(dir)->index_cnt++;
4491 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4492 struct btrfs_root *root,
4494 const char *name, int name_len,
4495 u64 ref_objectid, u64 objectid,
4496 u64 alloc_hint, int mode, u64 *index)
4498 struct inode *inode;
4499 struct btrfs_inode_item *inode_item;
4500 struct btrfs_key *location;
4501 struct btrfs_path *path;
4502 struct btrfs_inode_ref *ref;
4503 struct btrfs_key key[2];
4509 path = btrfs_alloc_path();
4512 inode = new_inode(root->fs_info->sb);
4514 return ERR_PTR(-ENOMEM);
4517 ret = btrfs_set_inode_index(dir, index);
4520 return ERR_PTR(ret);
4524 * index_cnt is ignored for everything but a dir,
4525 * btrfs_get_inode_index_count has an explanation for the magic
4528 BTRFS_I(inode)->index_cnt = 2;
4529 BTRFS_I(inode)->root = root;
4530 BTRFS_I(inode)->generation = trans->transid;
4531 inode->i_generation = BTRFS_I(inode)->generation;
4532 btrfs_set_inode_space_info(root, inode);
4538 BTRFS_I(inode)->block_group =
4539 btrfs_find_block_group(root, 0, alloc_hint, owner);
4541 key[0].objectid = objectid;
4542 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4545 key[1].objectid = objectid;
4546 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4547 key[1].offset = ref_objectid;
4549 sizes[0] = sizeof(struct btrfs_inode_item);
4550 sizes[1] = name_len + sizeof(*ref);
4552 path->leave_spinning = 1;
4553 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4557 inode_init_owner(inode, dir, mode);
4558 inode->i_ino = objectid;
4559 inode_set_bytes(inode, 0);
4560 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4561 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4562 struct btrfs_inode_item);
4563 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4565 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4566 struct btrfs_inode_ref);
4567 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4568 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4569 ptr = (unsigned long)(ref + 1);
4570 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4572 btrfs_mark_buffer_dirty(path->nodes[0]);
4573 btrfs_free_path(path);
4575 location = &BTRFS_I(inode)->location;
4576 location->objectid = objectid;
4577 location->offset = 0;
4578 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4580 btrfs_inherit_iflags(inode, dir);
4582 if ((mode & S_IFREG)) {
4583 if (btrfs_test_opt(root, NODATASUM))
4584 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4585 if (btrfs_test_opt(root, NODATACOW))
4586 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4589 insert_inode_hash(inode);
4590 inode_tree_add(inode);
4594 BTRFS_I(dir)->index_cnt--;
4595 btrfs_free_path(path);
4597 return ERR_PTR(ret);
4600 static inline u8 btrfs_inode_type(struct inode *inode)
4602 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4606 * utility function to add 'inode' into 'parent_inode' with
4607 * a give name and a given sequence number.
4608 * if 'add_backref' is true, also insert a backref from the
4609 * inode to the parent directory.
4611 int btrfs_add_link(struct btrfs_trans_handle *trans,
4612 struct inode *parent_inode, struct inode *inode,
4613 const char *name, int name_len, int add_backref, u64 index)
4616 struct btrfs_key key;
4617 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4619 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4620 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4622 key.objectid = inode->i_ino;
4623 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4627 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4628 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4629 key.objectid, root->root_key.objectid,
4630 parent_inode->i_ino,
4631 index, name, name_len);
4632 } else if (add_backref) {
4633 ret = btrfs_insert_inode_ref(trans, root,
4634 name, name_len, inode->i_ino,
4635 parent_inode->i_ino, index);
4639 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4640 parent_inode->i_ino, &key,
4641 btrfs_inode_type(inode), index);
4644 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4646 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4647 ret = btrfs_update_inode(trans, root, parent_inode);
4652 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4653 struct inode *dir, struct dentry *dentry,
4654 struct inode *inode, int backref, u64 index)
4656 int err = btrfs_add_link(trans, dir, inode,
4657 dentry->d_name.name, dentry->d_name.len,
4660 d_instantiate(dentry, inode);
4668 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4669 int mode, dev_t rdev)
4671 struct btrfs_trans_handle *trans;
4672 struct btrfs_root *root = BTRFS_I(dir)->root;
4673 struct inode *inode = NULL;
4677 unsigned long nr = 0;
4680 if (!new_valid_dev(rdev))
4683 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4688 * 2 for inode item and ref
4690 * 1 for xattr if selinux is on
4692 trans = btrfs_start_transaction(root, 5);
4694 return PTR_ERR(trans);
4696 btrfs_set_trans_block_group(trans, dir);
4698 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4699 dentry->d_name.len, dir->i_ino, objectid,
4700 BTRFS_I(dir)->block_group, mode, &index);
4701 err = PTR_ERR(inode);
4705 err = btrfs_init_inode_security(trans, inode, dir);
4711 btrfs_set_trans_block_group(trans, inode);
4712 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4716 inode->i_op = &btrfs_special_inode_operations;
4717 init_special_inode(inode, inode->i_mode, rdev);
4718 btrfs_update_inode(trans, root, inode);
4720 btrfs_update_inode_block_group(trans, inode);
4721 btrfs_update_inode_block_group(trans, dir);
4723 nr = trans->blocks_used;
4724 btrfs_end_transaction_throttle(trans, root);
4725 btrfs_btree_balance_dirty(root, nr);
4727 inode_dec_link_count(inode);
4733 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4734 int mode, struct nameidata *nd)
4736 struct btrfs_trans_handle *trans;
4737 struct btrfs_root *root = BTRFS_I(dir)->root;
4738 struct inode *inode = NULL;
4741 unsigned long nr = 0;
4745 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4749 * 2 for inode item and ref
4751 * 1 for xattr if selinux is on
4753 trans = btrfs_start_transaction(root, 5);
4755 return PTR_ERR(trans);
4757 btrfs_set_trans_block_group(trans, dir);
4759 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4760 dentry->d_name.len, dir->i_ino, objectid,
4761 BTRFS_I(dir)->block_group, mode, &index);
4762 err = PTR_ERR(inode);
4766 err = btrfs_init_inode_security(trans, inode, dir);
4772 btrfs_set_trans_block_group(trans, inode);
4773 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4777 inode->i_mapping->a_ops = &btrfs_aops;
4778 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4779 inode->i_fop = &btrfs_file_operations;
4780 inode->i_op = &btrfs_file_inode_operations;
4781 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4783 btrfs_update_inode_block_group(trans, inode);
4784 btrfs_update_inode_block_group(trans, dir);
4786 nr = trans->blocks_used;
4787 btrfs_end_transaction_throttle(trans, root);
4789 inode_dec_link_count(inode);
4792 btrfs_btree_balance_dirty(root, nr);
4796 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4797 struct dentry *dentry)
4799 struct btrfs_trans_handle *trans;
4800 struct btrfs_root *root = BTRFS_I(dir)->root;
4801 struct inode *inode = old_dentry->d_inode;
4803 unsigned long nr = 0;
4807 if (inode->i_nlink == 0)
4810 /* do not allow sys_link's with other subvols of the same device */
4811 if (root->objectid != BTRFS_I(inode)->root->objectid)
4814 btrfs_inc_nlink(inode);
4815 inode->i_ctime = CURRENT_TIME;
4817 err = btrfs_set_inode_index(dir, &index);
4822 * 1 item for inode ref
4823 * 2 items for dir items
4825 trans = btrfs_start_transaction(root, 3);
4826 if (IS_ERR(trans)) {
4827 err = PTR_ERR(trans);
4831 btrfs_set_trans_block_group(trans, dir);
4832 atomic_inc(&inode->i_count);
4834 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4839 struct dentry *parent = dget_parent(dentry);
4840 btrfs_update_inode_block_group(trans, dir);
4841 err = btrfs_update_inode(trans, root, inode);
4843 btrfs_log_new_name(trans, inode, NULL, parent);
4847 nr = trans->blocks_used;
4848 btrfs_end_transaction_throttle(trans, root);
4851 inode_dec_link_count(inode);
4854 btrfs_btree_balance_dirty(root, nr);
4858 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4860 struct inode *inode = NULL;
4861 struct btrfs_trans_handle *trans;
4862 struct btrfs_root *root = BTRFS_I(dir)->root;
4864 int drop_on_err = 0;
4867 unsigned long nr = 1;
4869 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4874 * 2 items for inode and ref
4875 * 2 items for dir items
4876 * 1 for xattr if selinux is on
4878 trans = btrfs_start_transaction(root, 5);
4880 return PTR_ERR(trans);
4881 btrfs_set_trans_block_group(trans, dir);
4883 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4884 dentry->d_name.len, dir->i_ino, objectid,
4885 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4887 if (IS_ERR(inode)) {
4888 err = PTR_ERR(inode);
4894 err = btrfs_init_inode_security(trans, inode, dir);
4898 inode->i_op = &btrfs_dir_inode_operations;
4899 inode->i_fop = &btrfs_dir_file_operations;
4900 btrfs_set_trans_block_group(trans, inode);
4902 btrfs_i_size_write(inode, 0);
4903 err = btrfs_update_inode(trans, root, inode);
4907 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4908 dentry->d_name.len, 0, index);
4912 d_instantiate(dentry, inode);
4914 btrfs_update_inode_block_group(trans, inode);
4915 btrfs_update_inode_block_group(trans, dir);
4918 nr = trans->blocks_used;
4919 btrfs_end_transaction_throttle(trans, root);
4922 btrfs_btree_balance_dirty(root, nr);
4926 /* helper for btfs_get_extent. Given an existing extent in the tree,
4927 * and an extent that you want to insert, deal with overlap and insert
4928 * the new extent into the tree.
4930 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4931 struct extent_map *existing,
4932 struct extent_map *em,
4933 u64 map_start, u64 map_len)
4937 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4938 start_diff = map_start - em->start;
4939 em->start = map_start;
4941 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4942 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4943 em->block_start += start_diff;
4944 em->block_len -= start_diff;
4946 return add_extent_mapping(em_tree, em);
4949 static noinline int uncompress_inline(struct btrfs_path *path,
4950 struct inode *inode, struct page *page,
4951 size_t pg_offset, u64 extent_offset,
4952 struct btrfs_file_extent_item *item)
4955 struct extent_buffer *leaf = path->nodes[0];
4958 unsigned long inline_size;
4962 WARN_ON(pg_offset != 0);
4963 compress_type = btrfs_file_extent_compression(leaf, item);
4964 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4965 inline_size = btrfs_file_extent_inline_item_len(leaf,
4966 btrfs_item_nr(leaf, path->slots[0]));
4967 tmp = kmalloc(inline_size, GFP_NOFS);
4968 ptr = btrfs_file_extent_inline_start(item);
4970 read_extent_buffer(leaf, tmp, ptr, inline_size);
4972 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4973 ret = btrfs_decompress(compress_type, tmp, page,
4974 extent_offset, inline_size, max_size);
4976 char *kaddr = kmap_atomic(page, KM_USER0);
4977 unsigned long copy_size = min_t(u64,
4978 PAGE_CACHE_SIZE - pg_offset,
4979 max_size - extent_offset);
4980 memset(kaddr + pg_offset, 0, copy_size);
4981 kunmap_atomic(kaddr, KM_USER0);
4988 * a bit scary, this does extent mapping from logical file offset to the disk.
4989 * the ugly parts come from merging extents from the disk with the in-ram
4990 * representation. This gets more complex because of the data=ordered code,
4991 * where the in-ram extents might be locked pending data=ordered completion.
4993 * This also copies inline extents directly into the page.
4996 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4997 size_t pg_offset, u64 start, u64 len,
5003 u64 extent_start = 0;
5005 u64 objectid = inode->i_ino;
5007 struct btrfs_path *path = NULL;
5008 struct btrfs_root *root = BTRFS_I(inode)->root;
5009 struct btrfs_file_extent_item *item;
5010 struct extent_buffer *leaf;
5011 struct btrfs_key found_key;
5012 struct extent_map *em = NULL;
5013 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5014 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5015 struct btrfs_trans_handle *trans = NULL;
5019 read_lock(&em_tree->lock);
5020 em = lookup_extent_mapping(em_tree, start, len);
5022 em->bdev = root->fs_info->fs_devices->latest_bdev;
5023 read_unlock(&em_tree->lock);
5026 if (em->start > start || em->start + em->len <= start)
5027 free_extent_map(em);
5028 else if (em->block_start == EXTENT_MAP_INLINE && page)
5029 free_extent_map(em);
5033 em = alloc_extent_map(GFP_NOFS);
5038 em->bdev = root->fs_info->fs_devices->latest_bdev;
5039 em->start = EXTENT_MAP_HOLE;
5040 em->orig_start = EXTENT_MAP_HOLE;
5042 em->block_len = (u64)-1;
5045 path = btrfs_alloc_path();
5049 ret = btrfs_lookup_file_extent(trans, root, path,
5050 objectid, start, trans != NULL);
5057 if (path->slots[0] == 0)
5062 leaf = path->nodes[0];
5063 item = btrfs_item_ptr(leaf, path->slots[0],
5064 struct btrfs_file_extent_item);
5065 /* are we inside the extent that was found? */
5066 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5067 found_type = btrfs_key_type(&found_key);
5068 if (found_key.objectid != objectid ||
5069 found_type != BTRFS_EXTENT_DATA_KEY) {
5073 found_type = btrfs_file_extent_type(leaf, item);
5074 extent_start = found_key.offset;
5075 compress_type = btrfs_file_extent_compression(leaf, item);
5076 if (found_type == BTRFS_FILE_EXTENT_REG ||
5077 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5078 extent_end = extent_start +
5079 btrfs_file_extent_num_bytes(leaf, item);
5080 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5082 size = btrfs_file_extent_inline_len(leaf, item);
5083 extent_end = (extent_start + size + root->sectorsize - 1) &
5084 ~((u64)root->sectorsize - 1);
5087 if (start >= extent_end) {
5089 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5090 ret = btrfs_next_leaf(root, path);
5097 leaf = path->nodes[0];
5099 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5100 if (found_key.objectid != objectid ||
5101 found_key.type != BTRFS_EXTENT_DATA_KEY)
5103 if (start + len <= found_key.offset)
5106 em->len = found_key.offset - start;
5110 if (found_type == BTRFS_FILE_EXTENT_REG ||
5111 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5112 em->start = extent_start;
5113 em->len = extent_end - extent_start;
5114 em->orig_start = extent_start -
5115 btrfs_file_extent_offset(leaf, item);
5116 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5118 em->block_start = EXTENT_MAP_HOLE;
5121 if (compress_type != BTRFS_COMPRESS_NONE) {
5122 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5123 em->compress_type = compress_type;
5124 em->block_start = bytenr;
5125 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5128 bytenr += btrfs_file_extent_offset(leaf, item);
5129 em->block_start = bytenr;
5130 em->block_len = em->len;
5131 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5132 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5135 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5139 size_t extent_offset;
5142 em->block_start = EXTENT_MAP_INLINE;
5143 if (!page || create) {
5144 em->start = extent_start;
5145 em->len = extent_end - extent_start;
5149 size = btrfs_file_extent_inline_len(leaf, item);
5150 extent_offset = page_offset(page) + pg_offset - extent_start;
5151 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5152 size - extent_offset);
5153 em->start = extent_start + extent_offset;
5154 em->len = (copy_size + root->sectorsize - 1) &
5155 ~((u64)root->sectorsize - 1);
5156 em->orig_start = EXTENT_MAP_INLINE;
5157 if (compress_type) {
5158 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5159 em->compress_type = compress_type;
5161 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5162 if (create == 0 && !PageUptodate(page)) {
5163 if (btrfs_file_extent_compression(leaf, item) !=
5164 BTRFS_COMPRESS_NONE) {
5165 ret = uncompress_inline(path, inode, page,
5167 extent_offset, item);
5171 read_extent_buffer(leaf, map + pg_offset, ptr,
5173 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5174 memset(map + pg_offset + copy_size, 0,
5175 PAGE_CACHE_SIZE - pg_offset -
5180 flush_dcache_page(page);
5181 } else if (create && PageUptodate(page)) {
5185 free_extent_map(em);
5187 btrfs_release_path(root, path);
5188 trans = btrfs_join_transaction(root, 1);
5190 return ERR_CAST(trans);
5194 write_extent_buffer(leaf, map + pg_offset, ptr,
5197 btrfs_mark_buffer_dirty(leaf);
5199 set_extent_uptodate(io_tree, em->start,
5200 extent_map_end(em) - 1, GFP_NOFS);
5203 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5210 em->block_start = EXTENT_MAP_HOLE;
5211 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5213 btrfs_release_path(root, path);
5214 if (em->start > start || extent_map_end(em) <= start) {
5215 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5216 "[%llu %llu]\n", (unsigned long long)em->start,
5217 (unsigned long long)em->len,
5218 (unsigned long long)start,
5219 (unsigned long long)len);
5225 write_lock(&em_tree->lock);
5226 ret = add_extent_mapping(em_tree, em);
5227 /* it is possible that someone inserted the extent into the tree
5228 * while we had the lock dropped. It is also possible that
5229 * an overlapping map exists in the tree
5231 if (ret == -EEXIST) {
5232 struct extent_map *existing;
5236 existing = lookup_extent_mapping(em_tree, start, len);
5237 if (existing && (existing->start > start ||
5238 existing->start + existing->len <= start)) {
5239 free_extent_map(existing);
5243 existing = lookup_extent_mapping(em_tree, em->start,
5246 err = merge_extent_mapping(em_tree, existing,
5249 free_extent_map(existing);
5251 free_extent_map(em);
5256 free_extent_map(em);
5260 free_extent_map(em);
5265 write_unlock(&em_tree->lock);
5268 btrfs_free_path(path);
5270 ret = btrfs_end_transaction(trans, root);
5275 free_extent_map(em);
5276 return ERR_PTR(err);
5281 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5284 struct btrfs_root *root = BTRFS_I(inode)->root;
5285 struct btrfs_trans_handle *trans;
5286 struct extent_map *em;
5287 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5288 struct btrfs_key ins;
5292 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5294 trans = btrfs_join_transaction(root, 0);
5296 return ERR_CAST(trans);
5298 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5300 alloc_hint = get_extent_allocation_hint(inode, start, len);
5301 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5302 alloc_hint, (u64)-1, &ins, 1);
5308 em = alloc_extent_map(GFP_NOFS);
5310 em = ERR_PTR(-ENOMEM);
5315 em->orig_start = em->start;
5316 em->len = ins.offset;
5318 em->block_start = ins.objectid;
5319 em->block_len = ins.offset;
5320 em->bdev = root->fs_info->fs_devices->latest_bdev;
5321 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5324 write_lock(&em_tree->lock);
5325 ret = add_extent_mapping(em_tree, em);
5326 write_unlock(&em_tree->lock);
5329 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5332 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5333 ins.offset, ins.offset, 0);
5335 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5339 btrfs_end_transaction(trans, root);
5344 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5345 * block must be cow'd
5347 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5348 struct inode *inode, u64 offset, u64 len)
5350 struct btrfs_path *path;
5352 struct extent_buffer *leaf;
5353 struct btrfs_root *root = BTRFS_I(inode)->root;
5354 struct btrfs_file_extent_item *fi;
5355 struct btrfs_key key;
5363 path = btrfs_alloc_path();
5367 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5372 slot = path->slots[0];
5375 /* can't find the item, must cow */
5382 leaf = path->nodes[0];
5383 btrfs_item_key_to_cpu(leaf, &key, slot);
5384 if (key.objectid != inode->i_ino ||
5385 key.type != BTRFS_EXTENT_DATA_KEY) {
5386 /* not our file or wrong item type, must cow */
5390 if (key.offset > offset) {
5391 /* Wrong offset, must cow */
5395 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5396 found_type = btrfs_file_extent_type(leaf, fi);
5397 if (found_type != BTRFS_FILE_EXTENT_REG &&
5398 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5399 /* not a regular extent, must cow */
5402 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5403 backref_offset = btrfs_file_extent_offset(leaf, fi);
5405 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5406 if (extent_end < offset + len) {
5407 /* extent doesn't include our full range, must cow */
5411 if (btrfs_extent_readonly(root, disk_bytenr))
5415 * look for other files referencing this extent, if we
5416 * find any we must cow
5418 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5419 key.offset - backref_offset, disk_bytenr))
5423 * adjust disk_bytenr and num_bytes to cover just the bytes
5424 * in this extent we are about to write. If there
5425 * are any csums in that range we have to cow in order
5426 * to keep the csums correct
5428 disk_bytenr += backref_offset;
5429 disk_bytenr += offset - key.offset;
5430 num_bytes = min(offset + len, extent_end) - offset;
5431 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5434 * all of the above have passed, it is safe to overwrite this extent
5439 btrfs_free_path(path);
5443 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5444 struct buffer_head *bh_result, int create)
5446 struct extent_map *em;
5447 struct btrfs_root *root = BTRFS_I(inode)->root;
5448 u64 start = iblock << inode->i_blkbits;
5449 u64 len = bh_result->b_size;
5450 struct btrfs_trans_handle *trans;
5452 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5457 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5458 * io. INLINE is special, and we could probably kludge it in here, but
5459 * it's still buffered so for safety lets just fall back to the generic
5462 * For COMPRESSED we _have_ to read the entire extent in so we can
5463 * decompress it, so there will be buffering required no matter what we
5464 * do, so go ahead and fallback to buffered.
5466 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5467 * to buffered IO. Don't blame me, this is the price we pay for using
5470 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5471 em->block_start == EXTENT_MAP_INLINE) {
5472 free_extent_map(em);
5476 /* Just a good old fashioned hole, return */
5477 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5478 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5479 free_extent_map(em);
5480 /* DIO will do one hole at a time, so just unlock a sector */
5481 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5482 start + root->sectorsize - 1, GFP_NOFS);
5487 * We don't allocate a new extent in the following cases
5489 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5491 * 2) The extent is marked as PREALLOC. We're good to go here and can
5492 * just use the extent.
5496 len = em->len - (start - em->start);
5500 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5501 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5502 em->block_start != EXTENT_MAP_HOLE)) {
5507 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5508 type = BTRFS_ORDERED_PREALLOC;
5510 type = BTRFS_ORDERED_NOCOW;
5511 len = min(len, em->len - (start - em->start));
5512 block_start = em->block_start + (start - em->start);
5515 * we're not going to log anything, but we do need
5516 * to make sure the current transaction stays open
5517 * while we look for nocow cross refs
5519 trans = btrfs_join_transaction(root, 0);
5523 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5524 ret = btrfs_add_ordered_extent_dio(inode, start,
5525 block_start, len, len, type);
5526 btrfs_end_transaction(trans, root);
5528 free_extent_map(em);
5533 btrfs_end_transaction(trans, root);
5537 * this will cow the extent, reset the len in case we changed
5540 len = bh_result->b_size;
5541 free_extent_map(em);
5542 em = btrfs_new_extent_direct(inode, start, len);
5545 len = min(len, em->len - (start - em->start));
5547 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5548 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5551 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5553 bh_result->b_size = len;
5554 bh_result->b_bdev = em->bdev;
5555 set_buffer_mapped(bh_result);
5556 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5557 set_buffer_new(bh_result);
5559 free_extent_map(em);
5564 struct btrfs_dio_private {
5565 struct inode *inode;
5572 /* number of bios pending for this dio */
5573 atomic_t pending_bios;
5578 struct bio *orig_bio;
5581 static void btrfs_endio_direct_read(struct bio *bio, int err)
5583 struct btrfs_dio_private *dip = bio->bi_private;
5584 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5585 struct bio_vec *bvec = bio->bi_io_vec;
5586 struct inode *inode = dip->inode;
5587 struct btrfs_root *root = BTRFS_I(inode)->root;
5589 u32 *private = dip->csums;
5591 start = dip->logical_offset;
5593 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5594 struct page *page = bvec->bv_page;
5597 unsigned long flags;
5599 local_irq_save(flags);
5600 kaddr = kmap_atomic(page, KM_IRQ0);
5601 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5602 csum, bvec->bv_len);
5603 btrfs_csum_final(csum, (char *)&csum);
5604 kunmap_atomic(kaddr, KM_IRQ0);
5605 local_irq_restore(flags);
5607 flush_dcache_page(bvec->bv_page);
5608 if (csum != *private) {
5609 printk(KERN_ERR "btrfs csum failed ino %lu off"
5610 " %llu csum %u private %u\n",
5611 inode->i_ino, (unsigned long long)start,
5617 start += bvec->bv_len;
5620 } while (bvec <= bvec_end);
5622 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5623 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5624 bio->bi_private = dip->private;
5628 dio_end_io(bio, err);
5631 static void btrfs_endio_direct_write(struct bio *bio, int err)
5633 struct btrfs_dio_private *dip = bio->bi_private;
5634 struct inode *inode = dip->inode;
5635 struct btrfs_root *root = BTRFS_I(inode)->root;
5636 struct btrfs_trans_handle *trans;
5637 struct btrfs_ordered_extent *ordered = NULL;
5638 struct extent_state *cached_state = NULL;
5639 u64 ordered_offset = dip->logical_offset;
5640 u64 ordered_bytes = dip->bytes;
5646 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5654 trans = btrfs_join_transaction(root, 1);
5655 if (IS_ERR(trans)) {
5659 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5661 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5662 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5664 ret = btrfs_update_inode(trans, root, inode);
5669 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5670 ordered->file_offset + ordered->len - 1, 0,
5671 &cached_state, GFP_NOFS);
5673 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5674 ret = btrfs_mark_extent_written(trans, inode,
5675 ordered->file_offset,
5676 ordered->file_offset +
5683 ret = insert_reserved_file_extent(trans, inode,
5684 ordered->file_offset,
5690 BTRFS_FILE_EXTENT_REG);
5691 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5692 ordered->file_offset, ordered->len);
5700 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5701 btrfs_ordered_update_i_size(inode, 0, ordered);
5702 btrfs_update_inode(trans, root, inode);
5704 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5705 ordered->file_offset + ordered->len - 1,
5706 &cached_state, GFP_NOFS);
5708 btrfs_delalloc_release_metadata(inode, ordered->len);
5709 btrfs_end_transaction(trans, root);
5710 ordered_offset = ordered->file_offset + ordered->len;
5711 btrfs_put_ordered_extent(ordered);
5712 btrfs_put_ordered_extent(ordered);
5716 * our bio might span multiple ordered extents. If we haven't
5717 * completed the accounting for the whole dio, go back and try again
5719 if (ordered_offset < dip->logical_offset + dip->bytes) {
5720 ordered_bytes = dip->logical_offset + dip->bytes -
5725 bio->bi_private = dip->private;
5729 dio_end_io(bio, err);
5732 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5733 struct bio *bio, int mirror_num,
5734 unsigned long bio_flags, u64 offset)
5737 struct btrfs_root *root = BTRFS_I(inode)->root;
5738 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5743 static void btrfs_end_dio_bio(struct bio *bio, int err)
5745 struct btrfs_dio_private *dip = bio->bi_private;
5748 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5749 "sector %#Lx len %u err no %d\n",
5750 dip->inode->i_ino, bio->bi_rw,
5751 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5755 * before atomic variable goto zero, we must make sure
5756 * dip->errors is perceived to be set.
5758 smp_mb__before_atomic_dec();
5761 /* if there are more bios still pending for this dio, just exit */
5762 if (!atomic_dec_and_test(&dip->pending_bios))
5766 bio_io_error(dip->orig_bio);
5768 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5769 bio_endio(dip->orig_bio, 0);
5775 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5776 u64 first_sector, gfp_t gfp_flags)
5778 int nr_vecs = bio_get_nr_vecs(bdev);
5779 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5782 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5783 int rw, u64 file_offset, int skip_sum,
5786 int write = rw & REQ_WRITE;
5787 struct btrfs_root *root = BTRFS_I(inode)->root;
5791 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5795 if (write && !skip_sum) {
5796 ret = btrfs_wq_submit_bio(root->fs_info,
5797 inode, rw, bio, 0, 0,
5799 __btrfs_submit_bio_start_direct_io,
5800 __btrfs_submit_bio_done);
5802 } else if (!skip_sum)
5803 btrfs_lookup_bio_sums_dio(root, inode, bio,
5804 file_offset, csums);
5806 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5812 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5815 struct inode *inode = dip->inode;
5816 struct btrfs_root *root = BTRFS_I(inode)->root;
5817 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5819 struct bio *orig_bio = dip->orig_bio;
5820 struct bio_vec *bvec = orig_bio->bi_io_vec;
5821 u64 start_sector = orig_bio->bi_sector;
5822 u64 file_offset = dip->logical_offset;
5826 u32 *csums = dip->csums;
5829 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5832 bio->bi_private = dip;
5833 bio->bi_end_io = btrfs_end_dio_bio;
5834 atomic_inc(&dip->pending_bios);
5836 map_length = orig_bio->bi_size;
5837 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5838 &map_length, NULL, 0);
5844 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5845 if (unlikely(map_length < submit_len + bvec->bv_len ||
5846 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5847 bvec->bv_offset) < bvec->bv_len)) {
5849 * inc the count before we submit the bio so
5850 * we know the end IO handler won't happen before
5851 * we inc the count. Otherwise, the dip might get freed
5852 * before we're done setting it up
5854 atomic_inc(&dip->pending_bios);
5855 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5856 file_offset, skip_sum,
5860 atomic_dec(&dip->pending_bios);
5865 csums = csums + nr_pages;
5866 start_sector += submit_len >> 9;
5867 file_offset += submit_len;
5872 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5873 start_sector, GFP_NOFS);
5876 bio->bi_private = dip;
5877 bio->bi_end_io = btrfs_end_dio_bio;
5879 map_length = orig_bio->bi_size;
5880 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5881 &map_length, NULL, 0);
5887 submit_len += bvec->bv_len;
5893 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5902 * before atomic variable goto zero, we must
5903 * make sure dip->errors is perceived to be set.
5905 smp_mb__before_atomic_dec();
5906 if (atomic_dec_and_test(&dip->pending_bios))
5907 bio_io_error(dip->orig_bio);
5909 /* bio_end_io() will handle error, so we needn't return it */
5913 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5916 struct btrfs_root *root = BTRFS_I(inode)->root;
5917 struct btrfs_dio_private *dip;
5918 struct bio_vec *bvec = bio->bi_io_vec;
5920 int write = rw & REQ_WRITE;
5923 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5925 dip = kmalloc(sizeof(*dip), GFP_NOFS);
5933 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5940 dip->private = bio->bi_private;
5942 dip->logical_offset = file_offset;
5946 dip->bytes += bvec->bv_len;
5948 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5950 dip->disk_bytenr = (u64)bio->bi_sector << 9;
5951 bio->bi_private = dip;
5953 dip->orig_bio = bio;
5954 atomic_set(&dip->pending_bios, 0);
5957 bio->bi_end_io = btrfs_endio_direct_write;
5959 bio->bi_end_io = btrfs_endio_direct_read;
5961 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
5966 * If this is a write, we need to clean up the reserved space and kill
5967 * the ordered extent.
5970 struct btrfs_ordered_extent *ordered;
5971 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
5972 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
5973 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
5974 btrfs_free_reserved_extent(root, ordered->start,
5976 btrfs_put_ordered_extent(ordered);
5977 btrfs_put_ordered_extent(ordered);
5979 bio_endio(bio, ret);
5982 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
5983 const struct iovec *iov, loff_t offset,
5984 unsigned long nr_segs)
5989 unsigned blocksize_mask = root->sectorsize - 1;
5990 ssize_t retval = -EINVAL;
5991 loff_t end = offset;
5993 if (offset & blocksize_mask)
5996 /* Check the memory alignment. Blocks cannot straddle pages */
5997 for (seg = 0; seg < nr_segs; seg++) {
5998 addr = (unsigned long)iov[seg].iov_base;
5999 size = iov[seg].iov_len;
6001 if ((addr & blocksize_mask) || (size & blocksize_mask))
6008 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6009 const struct iovec *iov, loff_t offset,
6010 unsigned long nr_segs)
6012 struct file *file = iocb->ki_filp;
6013 struct inode *inode = file->f_mapping->host;
6014 struct btrfs_ordered_extent *ordered;
6015 struct extent_state *cached_state = NULL;
6016 u64 lockstart, lockend;
6018 int writing = rw & WRITE;
6020 size_t count = iov_length(iov, nr_segs);
6022 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6028 lockend = offset + count - 1;
6031 ret = btrfs_delalloc_reserve_space(inode, count);
6037 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6038 0, &cached_state, GFP_NOFS);
6040 * We're concerned with the entire range that we're going to be
6041 * doing DIO to, so we need to make sure theres no ordered
6042 * extents in this range.
6044 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6045 lockend - lockstart + 1);
6048 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6049 &cached_state, GFP_NOFS);
6050 btrfs_start_ordered_extent(inode, ordered, 1);
6051 btrfs_put_ordered_extent(ordered);
6056 * we don't use btrfs_set_extent_delalloc because we don't want
6057 * the dirty or uptodate bits
6060 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6061 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6062 EXTENT_DELALLOC, 0, NULL, &cached_state,
6065 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6066 lockend, EXTENT_LOCKED | write_bits,
6067 1, 0, &cached_state, GFP_NOFS);
6072 free_extent_state(cached_state);
6073 cached_state = NULL;
6075 ret = __blockdev_direct_IO(rw, iocb, inode,
6076 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6077 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6078 btrfs_submit_direct, 0);
6080 if (ret < 0 && ret != -EIOCBQUEUED) {
6081 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6082 offset + iov_length(iov, nr_segs) - 1,
6083 EXTENT_LOCKED | write_bits, 1, 0,
6084 &cached_state, GFP_NOFS);
6085 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6087 * We're falling back to buffered, unlock the section we didn't
6090 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6091 offset + iov_length(iov, nr_segs) - 1,
6092 EXTENT_LOCKED | write_bits, 1, 0,
6093 &cached_state, GFP_NOFS);
6096 free_extent_state(cached_state);
6100 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6101 __u64 start, __u64 len)
6103 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
6106 int btrfs_readpage(struct file *file, struct page *page)
6108 struct extent_io_tree *tree;
6109 tree = &BTRFS_I(page->mapping->host)->io_tree;
6110 return extent_read_full_page(tree, page, btrfs_get_extent);
6113 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6115 struct extent_io_tree *tree;
6118 if (current->flags & PF_MEMALLOC) {
6119 redirty_page_for_writepage(wbc, page);
6123 tree = &BTRFS_I(page->mapping->host)->io_tree;
6124 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6127 int btrfs_writepages(struct address_space *mapping,
6128 struct writeback_control *wbc)
6130 struct extent_io_tree *tree;
6132 tree = &BTRFS_I(mapping->host)->io_tree;
6133 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6137 btrfs_readpages(struct file *file, struct address_space *mapping,
6138 struct list_head *pages, unsigned nr_pages)
6140 struct extent_io_tree *tree;
6141 tree = &BTRFS_I(mapping->host)->io_tree;
6142 return extent_readpages(tree, mapping, pages, nr_pages,
6145 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6147 struct extent_io_tree *tree;
6148 struct extent_map_tree *map;
6151 tree = &BTRFS_I(page->mapping->host)->io_tree;
6152 map = &BTRFS_I(page->mapping->host)->extent_tree;
6153 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6155 ClearPagePrivate(page);
6156 set_page_private(page, 0);
6157 page_cache_release(page);
6162 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6164 if (PageWriteback(page) || PageDirty(page))
6166 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6169 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6171 struct extent_io_tree *tree;
6172 struct btrfs_ordered_extent *ordered;
6173 struct extent_state *cached_state = NULL;
6174 u64 page_start = page_offset(page);
6175 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6179 * we have the page locked, so new writeback can't start,
6180 * and the dirty bit won't be cleared while we are here.
6182 * Wait for IO on this page so that we can safely clear
6183 * the PagePrivate2 bit and do ordered accounting
6185 wait_on_page_writeback(page);
6187 tree = &BTRFS_I(page->mapping->host)->io_tree;
6189 btrfs_releasepage(page, GFP_NOFS);
6192 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6194 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6198 * IO on this page will never be started, so we need
6199 * to account for any ordered extents now
6201 clear_extent_bit(tree, page_start, page_end,
6202 EXTENT_DIRTY | EXTENT_DELALLOC |
6203 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6204 &cached_state, GFP_NOFS);
6206 * whoever cleared the private bit is responsible
6207 * for the finish_ordered_io
6209 if (TestClearPagePrivate2(page)) {
6210 btrfs_finish_ordered_io(page->mapping->host,
6211 page_start, page_end);
6213 btrfs_put_ordered_extent(ordered);
6214 cached_state = NULL;
6215 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6218 clear_extent_bit(tree, page_start, page_end,
6219 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6220 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6221 __btrfs_releasepage(page, GFP_NOFS);
6223 ClearPageChecked(page);
6224 if (PagePrivate(page)) {
6225 ClearPagePrivate(page);
6226 set_page_private(page, 0);
6227 page_cache_release(page);
6232 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6233 * called from a page fault handler when a page is first dirtied. Hence we must
6234 * be careful to check for EOF conditions here. We set the page up correctly
6235 * for a written page which means we get ENOSPC checking when writing into
6236 * holes and correct delalloc and unwritten extent mapping on filesystems that
6237 * support these features.
6239 * We are not allowed to take the i_mutex here so we have to play games to
6240 * protect against truncate races as the page could now be beyond EOF. Because
6241 * vmtruncate() writes the inode size before removing pages, once we have the
6242 * page lock we can determine safely if the page is beyond EOF. If it is not
6243 * beyond EOF, then the page is guaranteed safe against truncation until we
6246 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6248 struct page *page = vmf->page;
6249 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6250 struct btrfs_root *root = BTRFS_I(inode)->root;
6251 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6252 struct btrfs_ordered_extent *ordered;
6253 struct extent_state *cached_state = NULL;
6255 unsigned long zero_start;
6261 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6265 else /* -ENOSPC, -EIO, etc */
6266 ret = VM_FAULT_SIGBUS;
6270 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6273 size = i_size_read(inode);
6274 page_start = page_offset(page);
6275 page_end = page_start + PAGE_CACHE_SIZE - 1;
6277 if ((page->mapping != inode->i_mapping) ||
6278 (page_start >= size)) {
6279 /* page got truncated out from underneath us */
6282 wait_on_page_writeback(page);
6284 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6286 set_page_extent_mapped(page);
6289 * we can't set the delalloc bits if there are pending ordered
6290 * extents. Drop our locks and wait for them to finish
6292 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6294 unlock_extent_cached(io_tree, page_start, page_end,
6295 &cached_state, GFP_NOFS);
6297 btrfs_start_ordered_extent(inode, ordered, 1);
6298 btrfs_put_ordered_extent(ordered);
6303 * XXX - page_mkwrite gets called every time the page is dirtied, even
6304 * if it was already dirty, so for space accounting reasons we need to
6305 * clear any delalloc bits for the range we are fixing to save. There
6306 * is probably a better way to do this, but for now keep consistent with
6307 * prepare_pages in the normal write path.
6309 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6310 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6311 0, 0, &cached_state, GFP_NOFS);
6313 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6316 unlock_extent_cached(io_tree, page_start, page_end,
6317 &cached_state, GFP_NOFS);
6318 ret = VM_FAULT_SIGBUS;
6323 /* page is wholly or partially inside EOF */
6324 if (page_start + PAGE_CACHE_SIZE > size)
6325 zero_start = size & ~PAGE_CACHE_MASK;
6327 zero_start = PAGE_CACHE_SIZE;
6329 if (zero_start != PAGE_CACHE_SIZE) {
6331 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6332 flush_dcache_page(page);
6335 ClearPageChecked(page);
6336 set_page_dirty(page);
6337 SetPageUptodate(page);
6339 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6340 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6342 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6346 return VM_FAULT_LOCKED;
6348 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6353 static void btrfs_truncate(struct inode *inode)
6355 struct btrfs_root *root = BTRFS_I(inode)->root;
6357 struct btrfs_trans_handle *trans;
6359 u64 mask = root->sectorsize - 1;
6361 if (!S_ISREG(inode->i_mode)) {
6366 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6370 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6371 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6373 trans = btrfs_start_transaction(root, 0);
6374 BUG_ON(IS_ERR(trans));
6375 btrfs_set_trans_block_group(trans, inode);
6376 trans->block_rsv = root->orphan_block_rsv;
6379 * setattr is responsible for setting the ordered_data_close flag,
6380 * but that is only tested during the last file release. That
6381 * could happen well after the next commit, leaving a great big
6382 * window where new writes may get lost if someone chooses to write
6383 * to this file after truncating to zero
6385 * The inode doesn't have any dirty data here, and so if we commit
6386 * this is a noop. If someone immediately starts writing to the inode
6387 * it is very likely we'll catch some of their writes in this
6388 * transaction, and the commit will find this file on the ordered
6389 * data list with good things to send down.
6391 * This is a best effort solution, there is still a window where
6392 * using truncate to replace the contents of the file will
6393 * end up with a zero length file after a crash.
6395 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6396 btrfs_add_ordered_operation(trans, root, inode);
6400 trans = btrfs_start_transaction(root, 0);
6401 BUG_ON(IS_ERR(trans));
6402 btrfs_set_trans_block_group(trans, inode);
6403 trans->block_rsv = root->orphan_block_rsv;
6406 ret = btrfs_block_rsv_check(trans, root,
6407 root->orphan_block_rsv, 0, 5);
6409 BUG_ON(ret != -EAGAIN);
6410 ret = btrfs_commit_transaction(trans, root);
6416 ret = btrfs_truncate_inode_items(trans, root, inode,
6418 BTRFS_EXTENT_DATA_KEY);
6422 ret = btrfs_update_inode(trans, root, inode);
6425 nr = trans->blocks_used;
6426 btrfs_end_transaction(trans, root);
6428 btrfs_btree_balance_dirty(root, nr);
6431 if (ret == 0 && inode->i_nlink > 0) {
6432 ret = btrfs_orphan_del(trans, inode);
6436 ret = btrfs_update_inode(trans, root, inode);
6439 nr = trans->blocks_used;
6440 ret = btrfs_end_transaction_throttle(trans, root);
6442 btrfs_btree_balance_dirty(root, nr);
6446 * create a new subvolume directory/inode (helper for the ioctl).
6448 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6449 struct btrfs_root *new_root,
6450 u64 new_dirid, u64 alloc_hint)
6452 struct inode *inode;
6456 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6457 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6459 return PTR_ERR(inode);
6460 inode->i_op = &btrfs_dir_inode_operations;
6461 inode->i_fop = &btrfs_dir_file_operations;
6464 btrfs_i_size_write(inode, 0);
6466 err = btrfs_update_inode(trans, new_root, inode);
6473 /* helper function for file defrag and space balancing. This
6474 * forces readahead on a given range of bytes in an inode
6476 unsigned long btrfs_force_ra(struct address_space *mapping,
6477 struct file_ra_state *ra, struct file *file,
6478 pgoff_t offset, pgoff_t last_index)
6480 pgoff_t req_size = last_index - offset + 1;
6482 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6483 return offset + req_size;
6486 struct inode *btrfs_alloc_inode(struct super_block *sb)
6488 struct btrfs_inode *ei;
6489 struct inode *inode;
6491 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6496 ei->space_info = NULL;
6500 ei->last_sub_trans = 0;
6501 ei->logged_trans = 0;
6502 ei->delalloc_bytes = 0;
6503 ei->reserved_bytes = 0;
6504 ei->disk_i_size = 0;
6506 ei->index_cnt = (u64)-1;
6507 ei->last_unlink_trans = 0;
6509 spin_lock_init(&ei->accounting_lock);
6510 atomic_set(&ei->outstanding_extents, 0);
6511 ei->reserved_extents = 0;
6513 ei->ordered_data_close = 0;
6514 ei->orphan_meta_reserved = 0;
6515 ei->dummy_inode = 0;
6516 ei->force_compress = BTRFS_COMPRESS_NONE;
6518 inode = &ei->vfs_inode;
6519 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6520 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6521 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6522 mutex_init(&ei->log_mutex);
6523 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6524 INIT_LIST_HEAD(&ei->i_orphan);
6525 INIT_LIST_HEAD(&ei->delalloc_inodes);
6526 INIT_LIST_HEAD(&ei->ordered_operations);
6527 RB_CLEAR_NODE(&ei->rb_node);
6532 void btrfs_destroy_inode(struct inode *inode)
6534 struct btrfs_ordered_extent *ordered;
6535 struct btrfs_root *root = BTRFS_I(inode)->root;
6537 WARN_ON(!list_empty(&inode->i_dentry));
6538 WARN_ON(inode->i_data.nrpages);
6539 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6540 WARN_ON(BTRFS_I(inode)->reserved_extents);
6543 * This can happen where we create an inode, but somebody else also
6544 * created the same inode and we need to destroy the one we already
6551 * Make sure we're properly removed from the ordered operation
6555 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6556 spin_lock(&root->fs_info->ordered_extent_lock);
6557 list_del_init(&BTRFS_I(inode)->ordered_operations);
6558 spin_unlock(&root->fs_info->ordered_extent_lock);
6561 if (root == root->fs_info->tree_root) {
6562 struct btrfs_block_group_cache *block_group;
6564 block_group = btrfs_lookup_block_group(root->fs_info,
6565 BTRFS_I(inode)->block_group);
6566 if (block_group && block_group->inode == inode) {
6567 spin_lock(&block_group->lock);
6568 block_group->inode = NULL;
6569 spin_unlock(&block_group->lock);
6570 btrfs_put_block_group(block_group);
6571 } else if (block_group) {
6572 btrfs_put_block_group(block_group);
6576 spin_lock(&root->orphan_lock);
6577 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6578 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6580 list_del_init(&BTRFS_I(inode)->i_orphan);
6582 spin_unlock(&root->orphan_lock);
6585 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6589 printk(KERN_ERR "btrfs found ordered "
6590 "extent %llu %llu on inode cleanup\n",
6591 (unsigned long long)ordered->file_offset,
6592 (unsigned long long)ordered->len);
6593 btrfs_remove_ordered_extent(inode, ordered);
6594 btrfs_put_ordered_extent(ordered);
6595 btrfs_put_ordered_extent(ordered);
6598 inode_tree_del(inode);
6599 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6601 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6604 int btrfs_drop_inode(struct inode *inode)
6606 struct btrfs_root *root = BTRFS_I(inode)->root;
6608 if (btrfs_root_refs(&root->root_item) == 0 &&
6609 root != root->fs_info->tree_root)
6612 return generic_drop_inode(inode);
6615 static void init_once(void *foo)
6617 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6619 inode_init_once(&ei->vfs_inode);
6622 void btrfs_destroy_cachep(void)
6624 if (btrfs_inode_cachep)
6625 kmem_cache_destroy(btrfs_inode_cachep);
6626 if (btrfs_trans_handle_cachep)
6627 kmem_cache_destroy(btrfs_trans_handle_cachep);
6628 if (btrfs_transaction_cachep)
6629 kmem_cache_destroy(btrfs_transaction_cachep);
6630 if (btrfs_path_cachep)
6631 kmem_cache_destroy(btrfs_path_cachep);
6634 int btrfs_init_cachep(void)
6636 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6637 sizeof(struct btrfs_inode), 0,
6638 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6639 if (!btrfs_inode_cachep)
6642 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6643 sizeof(struct btrfs_trans_handle), 0,
6644 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6645 if (!btrfs_trans_handle_cachep)
6648 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6649 sizeof(struct btrfs_transaction), 0,
6650 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6651 if (!btrfs_transaction_cachep)
6654 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6655 sizeof(struct btrfs_path), 0,
6656 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6657 if (!btrfs_path_cachep)
6662 btrfs_destroy_cachep();
6666 static int btrfs_getattr(struct vfsmount *mnt,
6667 struct dentry *dentry, struct kstat *stat)
6669 struct inode *inode = dentry->d_inode;
6670 generic_fillattr(inode, stat);
6671 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6672 stat->blksize = PAGE_CACHE_SIZE;
6673 stat->blocks = (inode_get_bytes(inode) +
6674 BTRFS_I(inode)->delalloc_bytes) >> 9;
6678 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6679 struct inode *new_dir, struct dentry *new_dentry)
6681 struct btrfs_trans_handle *trans;
6682 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6683 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6684 struct inode *new_inode = new_dentry->d_inode;
6685 struct inode *old_inode = old_dentry->d_inode;
6686 struct timespec ctime = CURRENT_TIME;
6691 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6694 /* we only allow rename subvolume link between subvolumes */
6695 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6698 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6699 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6702 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6703 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6706 * we're using rename to replace one file with another.
6707 * and the replacement file is large. Start IO on it now so
6708 * we don't add too much work to the end of the transaction
6710 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6711 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6712 filemap_flush(old_inode->i_mapping);
6714 /* close the racy window with snapshot create/destroy ioctl */
6715 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6716 down_read(&root->fs_info->subvol_sem);
6718 * We want to reserve the absolute worst case amount of items. So if
6719 * both inodes are subvols and we need to unlink them then that would
6720 * require 4 item modifications, but if they are both normal inodes it
6721 * would require 5 item modifications, so we'll assume their normal
6722 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6723 * should cover the worst case number of items we'll modify.
6725 trans = btrfs_start_transaction(root, 20);
6727 return PTR_ERR(trans);
6729 btrfs_set_trans_block_group(trans, new_dir);
6732 btrfs_record_root_in_trans(trans, dest);
6734 ret = btrfs_set_inode_index(new_dir, &index);
6738 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6739 /* force full log commit if subvolume involved. */
6740 root->fs_info->last_trans_log_full_commit = trans->transid;
6742 ret = btrfs_insert_inode_ref(trans, dest,
6743 new_dentry->d_name.name,
6744 new_dentry->d_name.len,
6746 new_dir->i_ino, index);
6750 * this is an ugly little race, but the rename is required
6751 * to make sure that if we crash, the inode is either at the
6752 * old name or the new one. pinning the log transaction lets
6753 * us make sure we don't allow a log commit to come in after
6754 * we unlink the name but before we add the new name back in.
6756 btrfs_pin_log_trans(root);
6759 * make sure the inode gets flushed if it is replacing
6762 if (new_inode && new_inode->i_size &&
6763 old_inode && S_ISREG(old_inode->i_mode)) {
6764 btrfs_add_ordered_operation(trans, root, old_inode);
6767 old_dir->i_ctime = old_dir->i_mtime = ctime;
6768 new_dir->i_ctime = new_dir->i_mtime = ctime;
6769 old_inode->i_ctime = ctime;
6771 if (old_dentry->d_parent != new_dentry->d_parent)
6772 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6774 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6775 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6776 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6777 old_dentry->d_name.name,
6778 old_dentry->d_name.len);
6780 btrfs_inc_nlink(old_dentry->d_inode);
6781 ret = btrfs_unlink_inode(trans, root, old_dir,
6782 old_dentry->d_inode,
6783 old_dentry->d_name.name,
6784 old_dentry->d_name.len);
6789 new_inode->i_ctime = CURRENT_TIME;
6790 if (unlikely(new_inode->i_ino ==
6791 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6792 root_objectid = BTRFS_I(new_inode)->location.objectid;
6793 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6795 new_dentry->d_name.name,
6796 new_dentry->d_name.len);
6797 BUG_ON(new_inode->i_nlink == 0);
6799 ret = btrfs_unlink_inode(trans, dest, new_dir,
6800 new_dentry->d_inode,
6801 new_dentry->d_name.name,
6802 new_dentry->d_name.len);
6805 if (new_inode->i_nlink == 0) {
6806 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6811 ret = btrfs_add_link(trans, new_dir, old_inode,
6812 new_dentry->d_name.name,
6813 new_dentry->d_name.len, 0, index);
6816 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6817 struct dentry *parent = dget_parent(new_dentry);
6818 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6820 btrfs_end_log_trans(root);
6823 btrfs_end_transaction_throttle(trans, root);
6825 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6826 up_read(&root->fs_info->subvol_sem);
6832 * some fairly slow code that needs optimization. This walks the list
6833 * of all the inodes with pending delalloc and forces them to disk.
6835 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6837 struct list_head *head = &root->fs_info->delalloc_inodes;
6838 struct btrfs_inode *binode;
6839 struct inode *inode;
6841 if (root->fs_info->sb->s_flags & MS_RDONLY)
6844 spin_lock(&root->fs_info->delalloc_lock);
6845 while (!list_empty(head)) {
6846 binode = list_entry(head->next, struct btrfs_inode,
6848 inode = igrab(&binode->vfs_inode);
6850 list_del_init(&binode->delalloc_inodes);
6851 spin_unlock(&root->fs_info->delalloc_lock);
6853 filemap_flush(inode->i_mapping);
6855 btrfs_add_delayed_iput(inode);
6860 spin_lock(&root->fs_info->delalloc_lock);
6862 spin_unlock(&root->fs_info->delalloc_lock);
6864 /* the filemap_flush will queue IO into the worker threads, but
6865 * we have to make sure the IO is actually started and that
6866 * ordered extents get created before we return
6868 atomic_inc(&root->fs_info->async_submit_draining);
6869 while (atomic_read(&root->fs_info->nr_async_submits) ||
6870 atomic_read(&root->fs_info->async_delalloc_pages)) {
6871 wait_event(root->fs_info->async_submit_wait,
6872 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
6873 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
6875 atomic_dec(&root->fs_info->async_submit_draining);
6879 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
6882 struct btrfs_inode *binode;
6883 struct inode *inode = NULL;
6885 spin_lock(&root->fs_info->delalloc_lock);
6886 while (!list_empty(&root->fs_info->delalloc_inodes)) {
6887 binode = list_entry(root->fs_info->delalloc_inodes.next,
6888 struct btrfs_inode, delalloc_inodes);
6889 inode = igrab(&binode->vfs_inode);
6891 list_move_tail(&binode->delalloc_inodes,
6892 &root->fs_info->delalloc_inodes);
6896 list_del_init(&binode->delalloc_inodes);
6897 cond_resched_lock(&root->fs_info->delalloc_lock);
6899 spin_unlock(&root->fs_info->delalloc_lock);
6903 filemap_write_and_wait(inode->i_mapping);
6905 * We have to do this because compression doesn't
6906 * actually set PG_writeback until it submits the pages
6907 * for IO, which happens in an async thread, so we could
6908 * race and not actually wait for any writeback pages
6909 * because they've not been submitted yet. Technically
6910 * this could still be the case for the ordered stuff
6911 * since the async thread may not have started to do its
6912 * work yet. If this becomes the case then we need to
6913 * figure out a way to make sure that in writepage we
6914 * wait for any async pages to be submitted before
6915 * returning so that fdatawait does what its supposed to
6918 btrfs_wait_ordered_range(inode, 0, (u64)-1);
6920 filemap_flush(inode->i_mapping);
6923 btrfs_add_delayed_iput(inode);
6931 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
6932 const char *symname)
6934 struct btrfs_trans_handle *trans;
6935 struct btrfs_root *root = BTRFS_I(dir)->root;
6936 struct btrfs_path *path;
6937 struct btrfs_key key;
6938 struct inode *inode = NULL;
6946 struct btrfs_file_extent_item *ei;
6947 struct extent_buffer *leaf;
6948 unsigned long nr = 0;
6950 name_len = strlen(symname) + 1;
6951 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
6952 return -ENAMETOOLONG;
6954 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
6958 * 2 items for inode item and ref
6959 * 2 items for dir items
6960 * 1 item for xattr if selinux is on
6962 trans = btrfs_start_transaction(root, 5);
6964 return PTR_ERR(trans);
6966 btrfs_set_trans_block_group(trans, dir);
6968 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6969 dentry->d_name.len, dir->i_ino, objectid,
6970 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
6972 err = PTR_ERR(inode);
6976 err = btrfs_init_inode_security(trans, inode, dir);
6982 btrfs_set_trans_block_group(trans, inode);
6983 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6987 inode->i_mapping->a_ops = &btrfs_aops;
6988 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6989 inode->i_fop = &btrfs_file_operations;
6990 inode->i_op = &btrfs_file_inode_operations;
6991 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6993 btrfs_update_inode_block_group(trans, inode);
6994 btrfs_update_inode_block_group(trans, dir);
6998 path = btrfs_alloc_path();
7000 key.objectid = inode->i_ino;
7002 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7003 datasize = btrfs_file_extent_calc_inline_size(name_len);
7004 err = btrfs_insert_empty_item(trans, root, path, &key,
7010 leaf = path->nodes[0];
7011 ei = btrfs_item_ptr(leaf, path->slots[0],
7012 struct btrfs_file_extent_item);
7013 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7014 btrfs_set_file_extent_type(leaf, ei,
7015 BTRFS_FILE_EXTENT_INLINE);
7016 btrfs_set_file_extent_encryption(leaf, ei, 0);
7017 btrfs_set_file_extent_compression(leaf, ei, 0);
7018 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7019 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7021 ptr = btrfs_file_extent_inline_start(ei);
7022 write_extent_buffer(leaf, symname, ptr, name_len);
7023 btrfs_mark_buffer_dirty(leaf);
7024 btrfs_free_path(path);
7026 inode->i_op = &btrfs_symlink_inode_operations;
7027 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7028 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7029 inode_set_bytes(inode, name_len);
7030 btrfs_i_size_write(inode, name_len - 1);
7031 err = btrfs_update_inode(trans, root, inode);
7036 nr = trans->blocks_used;
7037 btrfs_end_transaction_throttle(trans, root);
7039 inode_dec_link_count(inode);
7042 btrfs_btree_balance_dirty(root, nr);
7046 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7047 u64 start, u64 num_bytes, u64 min_size,
7048 loff_t actual_len, u64 *alloc_hint,
7049 struct btrfs_trans_handle *trans)
7051 struct btrfs_root *root = BTRFS_I(inode)->root;
7052 struct btrfs_key ins;
7053 u64 cur_offset = start;
7056 bool own_trans = true;
7060 while (num_bytes > 0) {
7062 trans = btrfs_start_transaction(root, 3);
7063 if (IS_ERR(trans)) {
7064 ret = PTR_ERR(trans);
7069 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7070 0, *alloc_hint, (u64)-1, &ins, 1);
7073 btrfs_end_transaction(trans, root);
7077 ret = insert_reserved_file_extent(trans, inode,
7078 cur_offset, ins.objectid,
7079 ins.offset, ins.offset,
7080 ins.offset, 0, 0, 0,
7081 BTRFS_FILE_EXTENT_PREALLOC);
7083 btrfs_drop_extent_cache(inode, cur_offset,
7084 cur_offset + ins.offset -1, 0);
7086 num_bytes -= ins.offset;
7087 cur_offset += ins.offset;
7088 *alloc_hint = ins.objectid + ins.offset;
7090 inode->i_ctime = CURRENT_TIME;
7091 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7092 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7093 (actual_len > inode->i_size) &&
7094 (cur_offset > inode->i_size)) {
7095 if (cur_offset > actual_len)
7096 i_size = actual_len;
7098 i_size = cur_offset;
7099 i_size_write(inode, i_size);
7100 btrfs_ordered_update_i_size(inode, i_size, NULL);
7103 ret = btrfs_update_inode(trans, root, inode);
7107 btrfs_end_transaction(trans, root);
7112 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7113 u64 start, u64 num_bytes, u64 min_size,
7114 loff_t actual_len, u64 *alloc_hint)
7116 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7117 min_size, actual_len, alloc_hint,
7121 int btrfs_prealloc_file_range_trans(struct inode *inode,
7122 struct btrfs_trans_handle *trans, int mode,
7123 u64 start, u64 num_bytes, u64 min_size,
7124 loff_t actual_len, u64 *alloc_hint)
7126 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7127 min_size, actual_len, alloc_hint, trans);
7130 static long btrfs_fallocate(struct inode *inode, int mode,
7131 loff_t offset, loff_t len)
7133 struct extent_state *cached_state = NULL;
7140 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
7141 struct extent_map *em;
7144 alloc_start = offset & ~mask;
7145 alloc_end = (offset + len + mask) & ~mask;
7148 * wait for ordered IO before we have any locks. We'll loop again
7149 * below with the locks held.
7151 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
7153 mutex_lock(&inode->i_mutex);
7154 ret = inode_newsize_ok(inode, alloc_end);
7158 if (alloc_start > inode->i_size) {
7159 ret = btrfs_cont_expand(inode, alloc_start);
7164 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
7168 locked_end = alloc_end - 1;
7170 struct btrfs_ordered_extent *ordered;
7172 /* the extent lock is ordered inside the running
7175 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
7176 locked_end, 0, &cached_state, GFP_NOFS);
7177 ordered = btrfs_lookup_first_ordered_extent(inode,
7180 ordered->file_offset + ordered->len > alloc_start &&
7181 ordered->file_offset < alloc_end) {
7182 btrfs_put_ordered_extent(ordered);
7183 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7184 alloc_start, locked_end,
7185 &cached_state, GFP_NOFS);
7187 * we can't wait on the range with the transaction
7188 * running or with the extent lock held
7190 btrfs_wait_ordered_range(inode, alloc_start,
7191 alloc_end - alloc_start);
7194 btrfs_put_ordered_extent(ordered);
7199 cur_offset = alloc_start;
7201 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
7202 alloc_end - cur_offset, 0);
7203 BUG_ON(IS_ERR(em) || !em);
7204 last_byte = min(extent_map_end(em), alloc_end);
7205 last_byte = (last_byte + mask) & ~mask;
7206 if (em->block_start == EXTENT_MAP_HOLE ||
7207 (cur_offset >= inode->i_size &&
7208 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7209 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
7210 last_byte - cur_offset,
7211 1 << inode->i_blkbits,
7215 free_extent_map(em);
7219 free_extent_map(em);
7221 cur_offset = last_byte;
7222 if (cur_offset >= alloc_end) {
7227 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
7228 &cached_state, GFP_NOFS);
7230 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
7232 mutex_unlock(&inode->i_mutex);
7236 static int btrfs_set_page_dirty(struct page *page)
7238 return __set_page_dirty_nobuffers(page);
7241 static int btrfs_permission(struct inode *inode, int mask)
7243 struct btrfs_root *root = BTRFS_I(inode)->root;
7245 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7247 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7249 return generic_permission(inode, mask, btrfs_check_acl);
7252 static const struct inode_operations btrfs_dir_inode_operations = {
7253 .getattr = btrfs_getattr,
7254 .lookup = btrfs_lookup,
7255 .create = btrfs_create,
7256 .unlink = btrfs_unlink,
7258 .mkdir = btrfs_mkdir,
7259 .rmdir = btrfs_rmdir,
7260 .rename = btrfs_rename,
7261 .symlink = btrfs_symlink,
7262 .setattr = btrfs_setattr,
7263 .mknod = btrfs_mknod,
7264 .setxattr = btrfs_setxattr,
7265 .getxattr = btrfs_getxattr,
7266 .listxattr = btrfs_listxattr,
7267 .removexattr = btrfs_removexattr,
7268 .permission = btrfs_permission,
7270 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7271 .lookup = btrfs_lookup,
7272 .permission = btrfs_permission,
7275 static const struct file_operations btrfs_dir_file_operations = {
7276 .llseek = generic_file_llseek,
7277 .read = generic_read_dir,
7278 .readdir = btrfs_real_readdir,
7279 .unlocked_ioctl = btrfs_ioctl,
7280 #ifdef CONFIG_COMPAT
7281 .compat_ioctl = btrfs_ioctl,
7283 .release = btrfs_release_file,
7284 .fsync = btrfs_sync_file,
7287 static struct extent_io_ops btrfs_extent_io_ops = {
7288 .fill_delalloc = run_delalloc_range,
7289 .submit_bio_hook = btrfs_submit_bio_hook,
7290 .merge_bio_hook = btrfs_merge_bio_hook,
7291 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7292 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7293 .writepage_start_hook = btrfs_writepage_start_hook,
7294 .readpage_io_failed_hook = btrfs_io_failed_hook,
7295 .set_bit_hook = btrfs_set_bit_hook,
7296 .clear_bit_hook = btrfs_clear_bit_hook,
7297 .merge_extent_hook = btrfs_merge_extent_hook,
7298 .split_extent_hook = btrfs_split_extent_hook,
7302 * btrfs doesn't support the bmap operation because swapfiles
7303 * use bmap to make a mapping of extents in the file. They assume
7304 * these extents won't change over the life of the file and they
7305 * use the bmap result to do IO directly to the drive.
7307 * the btrfs bmap call would return logical addresses that aren't
7308 * suitable for IO and they also will change frequently as COW
7309 * operations happen. So, swapfile + btrfs == corruption.
7311 * For now we're avoiding this by dropping bmap.
7313 static const struct address_space_operations btrfs_aops = {
7314 .readpage = btrfs_readpage,
7315 .writepage = btrfs_writepage,
7316 .writepages = btrfs_writepages,
7317 .readpages = btrfs_readpages,
7318 .sync_page = block_sync_page,
7319 .direct_IO = btrfs_direct_IO,
7320 .invalidatepage = btrfs_invalidatepage,
7321 .releasepage = btrfs_releasepage,
7322 .set_page_dirty = btrfs_set_page_dirty,
7323 .error_remove_page = generic_error_remove_page,
7326 static const struct address_space_operations btrfs_symlink_aops = {
7327 .readpage = btrfs_readpage,
7328 .writepage = btrfs_writepage,
7329 .invalidatepage = btrfs_invalidatepage,
7330 .releasepage = btrfs_releasepage,
7333 static const struct inode_operations btrfs_file_inode_operations = {
7334 .truncate = btrfs_truncate,
7335 .getattr = btrfs_getattr,
7336 .setattr = btrfs_setattr,
7337 .setxattr = btrfs_setxattr,
7338 .getxattr = btrfs_getxattr,
7339 .listxattr = btrfs_listxattr,
7340 .removexattr = btrfs_removexattr,
7341 .permission = btrfs_permission,
7342 .fallocate = btrfs_fallocate,
7343 .fiemap = btrfs_fiemap,
7345 static const struct inode_operations btrfs_special_inode_operations = {
7346 .getattr = btrfs_getattr,
7347 .setattr = btrfs_setattr,
7348 .permission = btrfs_permission,
7349 .setxattr = btrfs_setxattr,
7350 .getxattr = btrfs_getxattr,
7351 .listxattr = btrfs_listxattr,
7352 .removexattr = btrfs_removexattr,
7354 static const struct inode_operations btrfs_symlink_inode_operations = {
7355 .readlink = generic_readlink,
7356 .follow_link = page_follow_link_light,
7357 .put_link = page_put_link,
7358 .getattr = btrfs_getattr,
7359 .permission = btrfs_permission,
7360 .setxattr = btrfs_setxattr,
7361 .getxattr = btrfs_getxattr,
7362 .listxattr = btrfs_listxattr,
7363 .removexattr = btrfs_removexattr,
7366 const struct dentry_operations btrfs_dentry_operations = {
7367 .d_delete = btrfs_dentry_delete,