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>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
93 static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
106 err = btrfs_init_acl(trans, inode, dir);
108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
121 struct page **compressed_pages)
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
129 struct btrfs_file_extent_item *ei;
132 size_t cur_size = size;
134 unsigned long offset;
136 if (compressed_size && compressed_pages)
137 cur_size = compressed_size;
139 path = btrfs_alloc_path();
143 path->leave_spinning = 1;
145 key.objectid = btrfs_ino(inode);
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
157 leaf = path->nodes[0];
158 ei = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162 btrfs_set_file_extent_encryption(leaf, ei, 0);
163 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165 ptr = btrfs_file_extent_inline_start(ei);
167 if (compress_type != BTRFS_COMPRESS_NONE) {
170 while (compressed_size > 0) {
171 cpage = compressed_pages[i];
172 cur_size = min_t(unsigned long, compressed_size,
175 kaddr = kmap_atomic(cpage);
176 write_extent_buffer(leaf, kaddr, ptr, cur_size);
177 kunmap_atomic(kaddr);
181 compressed_size -= cur_size;
183 btrfs_set_file_extent_compression(leaf, ei,
186 page = find_get_page(inode->i_mapping,
187 start >> PAGE_CACHE_SHIFT);
188 btrfs_set_file_extent_compression(leaf, ei, 0);
189 kaddr = kmap_atomic(page);
190 offset = start & (PAGE_CACHE_SIZE - 1);
191 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192 kunmap_atomic(kaddr);
193 page_cache_release(page);
195 btrfs_mark_buffer_dirty(leaf);
196 btrfs_free_path(path);
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
207 BTRFS_I(inode)->disk_i_size = inode->i_size;
208 ret = btrfs_update_inode(trans, root, inode);
212 btrfs_free_path(path);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
223 struct btrfs_root *root,
224 struct inode *inode, u64 start, u64 end,
225 size_t compressed_size, int compress_type,
226 struct page **compressed_pages)
228 u64 isize = i_size_read(inode);
229 u64 actual_end = min(end + 1, isize);
230 u64 inline_len = actual_end - start;
231 u64 aligned_end = (end + root->sectorsize - 1) &
232 ~((u64)root->sectorsize - 1);
234 u64 data_len = inline_len;
238 data_len = compressed_size;
241 actual_end >= PAGE_CACHE_SIZE ||
242 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
244 (actual_end & (root->sectorsize - 1)) == 0) ||
246 data_len > root->fs_info->max_inline) {
250 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
259 compress_type, compressed_pages);
260 if (ret && ret != -ENOSPC) {
261 btrfs_abort_transaction(trans, root, ret);
263 } else if (ret == -ENOSPC) {
267 btrfs_delalloc_release_metadata(inode, end + 1 - start);
268 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
272 struct async_extent {
277 unsigned long nr_pages;
279 struct list_head list;
284 struct btrfs_root *root;
285 struct page *locked_page;
288 struct list_head extents;
289 struct btrfs_work work;
292 static noinline int add_async_extent(struct async_cow *cow,
293 u64 start, u64 ram_size,
296 unsigned long nr_pages,
299 struct async_extent *async_extent;
301 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
302 BUG_ON(!async_extent); /* -ENOMEM */
303 async_extent->start = start;
304 async_extent->ram_size = ram_size;
305 async_extent->compressed_size = compressed_size;
306 async_extent->pages = pages;
307 async_extent->nr_pages = nr_pages;
308 async_extent->compress_type = compress_type;
309 list_add_tail(&async_extent->list, &cow->extents);
314 * we create compressed extents in two phases. The first
315 * phase compresses a range of pages that have already been
316 * locked (both pages and state bits are locked).
318 * This is done inside an ordered work queue, and the compression
319 * is spread across many cpus. The actual IO submission is step
320 * two, and the ordered work queue takes care of making sure that
321 * happens in the same order things were put onto the queue by
322 * writepages and friends.
324 * If this code finds it can't get good compression, it puts an
325 * entry onto the work queue to write the uncompressed bytes. This
326 * makes sure that both compressed inodes and uncompressed inodes
327 * are written in the same order that pdflush sent them down.
329 static noinline int compress_file_range(struct inode *inode,
330 struct page *locked_page,
332 struct async_cow *async_cow,
335 struct btrfs_root *root = BTRFS_I(inode)->root;
336 struct btrfs_trans_handle *trans;
338 u64 blocksize = root->sectorsize;
340 u64 isize = i_size_read(inode);
342 struct page **pages = NULL;
343 unsigned long nr_pages;
344 unsigned long nr_pages_ret = 0;
345 unsigned long total_compressed = 0;
346 unsigned long total_in = 0;
347 unsigned long max_compressed = 128 * 1024;
348 unsigned long max_uncompressed = 128 * 1024;
351 int compress_type = root->fs_info->compress_type;
353 /* if this is a small write inside eof, kick off a defrag */
354 if ((end - start + 1) < 16 * 1024 &&
355 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
356 btrfs_add_inode_defrag(NULL, inode);
358 actual_end = min_t(u64, isize, end + 1);
361 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
362 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
365 * we don't want to send crud past the end of i_size through
366 * compression, that's just a waste of CPU time. So, if the
367 * end of the file is before the start of our current
368 * requested range of bytes, we bail out to the uncompressed
369 * cleanup code that can deal with all of this.
371 * It isn't really the fastest way to fix things, but this is a
372 * very uncommon corner.
374 if (actual_end <= start)
375 goto cleanup_and_bail_uncompressed;
377 total_compressed = actual_end - start;
379 /* we want to make sure that amount of ram required to uncompress
380 * an extent is reasonable, so we limit the total size in ram
381 * of a compressed extent to 128k. This is a crucial number
382 * because it also controls how easily we can spread reads across
383 * cpus for decompression.
385 * We also want to make sure the amount of IO required to do
386 * a random read is reasonably small, so we limit the size of
387 * a compressed extent to 128k.
389 total_compressed = min(total_compressed, max_uncompressed);
390 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
391 num_bytes = max(blocksize, num_bytes);
396 * we do compression for mount -o compress and when the
397 * inode has not been flagged as nocompress. This flag can
398 * change at any time if we discover bad compression ratios.
400 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
401 (btrfs_test_opt(root, COMPRESS) ||
402 (BTRFS_I(inode)->force_compress) ||
403 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
405 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
407 /* just bail out to the uncompressed code */
411 if (BTRFS_I(inode)->force_compress)
412 compress_type = BTRFS_I(inode)->force_compress;
414 ret = btrfs_compress_pages(compress_type,
415 inode->i_mapping, start,
416 total_compressed, pages,
417 nr_pages, &nr_pages_ret,
423 unsigned long offset = total_compressed &
424 (PAGE_CACHE_SIZE - 1);
425 struct page *page = pages[nr_pages_ret - 1];
428 /* zero the tail end of the last page, we might be
429 * sending it down to disk
432 kaddr = kmap_atomic(page);
433 memset(kaddr + offset, 0,
434 PAGE_CACHE_SIZE - offset);
435 kunmap_atomic(kaddr);
442 trans = btrfs_join_transaction(root);
444 ret = PTR_ERR(trans);
446 goto cleanup_and_out;
448 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
450 /* lets try to make an inline extent */
451 if (ret || total_in < (actual_end - start)) {
452 /* we didn't compress the entire range, try
453 * to make an uncompressed inline extent.
455 ret = cow_file_range_inline(trans, root, inode,
456 start, end, 0, 0, NULL);
458 /* try making a compressed inline extent */
459 ret = cow_file_range_inline(trans, root, inode,
462 compress_type, pages);
466 * inline extent creation worked or returned error,
467 * we don't need to create any more async work items.
468 * Unlock and free up our temp pages.
470 extent_clear_unlock_delalloc(inode,
471 &BTRFS_I(inode)->io_tree,
473 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
474 EXTENT_CLEAR_DELALLOC |
475 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
477 btrfs_end_transaction(trans, root);
480 btrfs_end_transaction(trans, root);
485 * we aren't doing an inline extent round the compressed size
486 * up to a block size boundary so the allocator does sane
489 total_compressed = (total_compressed + blocksize - 1) &
493 * one last check to make sure the compression is really a
494 * win, compare the page count read with the blocks on disk
496 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
497 ~(PAGE_CACHE_SIZE - 1);
498 if (total_compressed >= total_in) {
501 num_bytes = total_in;
504 if (!will_compress && pages) {
506 * the compression code ran but failed to make things smaller,
507 * free any pages it allocated and our page pointer array
509 for (i = 0; i < nr_pages_ret; i++) {
510 WARN_ON(pages[i]->mapping);
511 page_cache_release(pages[i]);
515 total_compressed = 0;
518 /* flag the file so we don't compress in the future */
519 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
520 !(BTRFS_I(inode)->force_compress)) {
521 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
527 /* the async work queues will take care of doing actual
528 * allocation on disk for these compressed pages,
529 * and will submit them to the elevator.
531 add_async_extent(async_cow, start, num_bytes,
532 total_compressed, pages, nr_pages_ret,
535 if (start + num_bytes < end) {
542 cleanup_and_bail_uncompressed:
544 * No compression, but we still need to write the pages in
545 * the file we've been given so far. redirty the locked
546 * page if it corresponds to our extent and set things up
547 * for the async work queue to run cow_file_range to do
548 * the normal delalloc dance
550 if (page_offset(locked_page) >= start &&
551 page_offset(locked_page) <= end) {
552 __set_page_dirty_nobuffers(locked_page);
553 /* unlocked later on in the async handlers */
555 add_async_extent(async_cow, start, end - start + 1,
556 0, NULL, 0, BTRFS_COMPRESS_NONE);
564 for (i = 0; i < nr_pages_ret; i++) {
565 WARN_ON(pages[i]->mapping);
566 page_cache_release(pages[i]);
573 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
575 EXTENT_CLEAR_UNLOCK_PAGE |
577 EXTENT_CLEAR_DELALLOC |
578 EXTENT_SET_WRITEBACK |
579 EXTENT_END_WRITEBACK);
580 if (!trans || IS_ERR(trans))
581 btrfs_error(root->fs_info, ret, "Failed to join transaction");
583 btrfs_abort_transaction(trans, root, ret);
588 * phase two of compressed writeback. This is the ordered portion
589 * of the code, which only gets called in the order the work was
590 * queued. We walk all the async extents created by compress_file_range
591 * and send them down to the disk.
593 static noinline int submit_compressed_extents(struct inode *inode,
594 struct async_cow *async_cow)
596 struct async_extent *async_extent;
598 struct btrfs_trans_handle *trans;
599 struct btrfs_key ins;
600 struct extent_map *em;
601 struct btrfs_root *root = BTRFS_I(inode)->root;
602 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
603 struct extent_io_tree *io_tree;
606 if (list_empty(&async_cow->extents))
610 while (!list_empty(&async_cow->extents)) {
611 async_extent = list_entry(async_cow->extents.next,
612 struct async_extent, list);
613 list_del(&async_extent->list);
615 io_tree = &BTRFS_I(inode)->io_tree;
618 /* did the compression code fall back to uncompressed IO? */
619 if (!async_extent->pages) {
620 int page_started = 0;
621 unsigned long nr_written = 0;
623 lock_extent(io_tree, async_extent->start,
624 async_extent->start +
625 async_extent->ram_size - 1);
627 /* allocate blocks */
628 ret = cow_file_range(inode, async_cow->locked_page,
630 async_extent->start +
631 async_extent->ram_size - 1,
632 &page_started, &nr_written, 0);
637 * if page_started, cow_file_range inserted an
638 * inline extent and took care of all the unlocking
639 * and IO for us. Otherwise, we need to submit
640 * all those pages down to the drive.
642 if (!page_started && !ret)
643 extent_write_locked_range(io_tree,
644 inode, async_extent->start,
645 async_extent->start +
646 async_extent->ram_size - 1,
654 lock_extent(io_tree, async_extent->start,
655 async_extent->start + async_extent->ram_size - 1);
657 trans = btrfs_join_transaction(root);
659 ret = PTR_ERR(trans);
661 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
662 ret = btrfs_reserve_extent(trans, root,
663 async_extent->compressed_size,
664 async_extent->compressed_size,
665 0, alloc_hint, &ins, 1);
667 btrfs_abort_transaction(trans, root, ret);
668 btrfs_end_transaction(trans, root);
673 for (i = 0; i < async_extent->nr_pages; i++) {
674 WARN_ON(async_extent->pages[i]->mapping);
675 page_cache_release(async_extent->pages[i]);
677 kfree(async_extent->pages);
678 async_extent->nr_pages = 0;
679 async_extent->pages = NULL;
680 unlock_extent(io_tree, async_extent->start,
681 async_extent->start +
682 async_extent->ram_size - 1);
685 goto out_free; /* JDM: Requeue? */
689 * here we're doing allocation and writeback of the
692 btrfs_drop_extent_cache(inode, async_extent->start,
693 async_extent->start +
694 async_extent->ram_size - 1, 0);
696 em = alloc_extent_map();
697 BUG_ON(!em); /* -ENOMEM */
698 em->start = async_extent->start;
699 em->len = async_extent->ram_size;
700 em->orig_start = em->start;
702 em->block_start = ins.objectid;
703 em->block_len = ins.offset;
704 em->bdev = root->fs_info->fs_devices->latest_bdev;
705 em->compress_type = async_extent->compress_type;
706 set_bit(EXTENT_FLAG_PINNED, &em->flags);
707 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
710 write_lock(&em_tree->lock);
711 ret = add_extent_mapping(em_tree, em);
712 write_unlock(&em_tree->lock);
713 if (ret != -EEXIST) {
717 btrfs_drop_extent_cache(inode, async_extent->start,
718 async_extent->start +
719 async_extent->ram_size - 1, 0);
722 ret = btrfs_add_ordered_extent_compress(inode,
725 async_extent->ram_size,
727 BTRFS_ORDERED_COMPRESSED,
728 async_extent->compress_type);
729 BUG_ON(ret); /* -ENOMEM */
732 * clear dirty, set writeback and unlock the pages.
734 extent_clear_unlock_delalloc(inode,
735 &BTRFS_I(inode)->io_tree,
737 async_extent->start +
738 async_extent->ram_size - 1,
739 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
740 EXTENT_CLEAR_UNLOCK |
741 EXTENT_CLEAR_DELALLOC |
742 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
744 ret = btrfs_submit_compressed_write(inode,
746 async_extent->ram_size,
748 ins.offset, async_extent->pages,
749 async_extent->nr_pages);
751 BUG_ON(ret); /* -ENOMEM */
752 alloc_hint = ins.objectid + ins.offset;
764 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
767 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
768 struct extent_map *em;
771 read_lock(&em_tree->lock);
772 em = search_extent_mapping(em_tree, start, num_bytes);
775 * if block start isn't an actual block number then find the
776 * first block in this inode and use that as a hint. If that
777 * block is also bogus then just don't worry about it.
779 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
781 em = search_extent_mapping(em_tree, 0, 0);
782 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
783 alloc_hint = em->block_start;
787 alloc_hint = em->block_start;
791 read_unlock(&em_tree->lock);
797 * when extent_io.c finds a delayed allocation range in the file,
798 * the call backs end up in this code. The basic idea is to
799 * allocate extents on disk for the range, and create ordered data structs
800 * in ram to track those extents.
802 * locked_page is the page that writepage had locked already. We use
803 * it to make sure we don't do extra locks or unlocks.
805 * *page_started is set to one if we unlock locked_page and do everything
806 * required to start IO on it. It may be clean and already done with
809 static noinline int cow_file_range(struct inode *inode,
810 struct page *locked_page,
811 u64 start, u64 end, int *page_started,
812 unsigned long *nr_written,
815 struct btrfs_root *root = BTRFS_I(inode)->root;
816 struct btrfs_trans_handle *trans;
819 unsigned long ram_size;
822 u64 blocksize = root->sectorsize;
823 struct btrfs_key ins;
824 struct extent_map *em;
825 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
828 BUG_ON(btrfs_is_free_space_inode(inode));
829 trans = btrfs_join_transaction(root);
831 extent_clear_unlock_delalloc(inode,
832 &BTRFS_I(inode)->io_tree,
833 start, end, locked_page,
834 EXTENT_CLEAR_UNLOCK_PAGE |
835 EXTENT_CLEAR_UNLOCK |
836 EXTENT_CLEAR_DELALLOC |
838 EXTENT_SET_WRITEBACK |
839 EXTENT_END_WRITEBACK);
840 return PTR_ERR(trans);
842 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
844 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
845 num_bytes = max(blocksize, num_bytes);
846 disk_num_bytes = num_bytes;
849 /* if this is a small write inside eof, kick off defrag */
850 if (num_bytes < 64 * 1024 &&
851 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
852 btrfs_add_inode_defrag(trans, inode);
855 /* lets try to make an inline extent */
856 ret = cow_file_range_inline(trans, root, inode,
857 start, end, 0, 0, NULL);
859 extent_clear_unlock_delalloc(inode,
860 &BTRFS_I(inode)->io_tree,
862 EXTENT_CLEAR_UNLOCK_PAGE |
863 EXTENT_CLEAR_UNLOCK |
864 EXTENT_CLEAR_DELALLOC |
866 EXTENT_SET_WRITEBACK |
867 EXTENT_END_WRITEBACK);
869 *nr_written = *nr_written +
870 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
873 } else if (ret < 0) {
874 btrfs_abort_transaction(trans, root, ret);
879 BUG_ON(disk_num_bytes >
880 btrfs_super_total_bytes(root->fs_info->super_copy));
882 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
883 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
885 while (disk_num_bytes > 0) {
888 cur_alloc_size = disk_num_bytes;
889 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
890 root->sectorsize, 0, alloc_hint,
893 btrfs_abort_transaction(trans, root, ret);
897 em = alloc_extent_map();
898 BUG_ON(!em); /* -ENOMEM */
900 em->orig_start = em->start;
901 ram_size = ins.offset;
902 em->len = ins.offset;
904 em->block_start = ins.objectid;
905 em->block_len = ins.offset;
906 em->bdev = root->fs_info->fs_devices->latest_bdev;
907 set_bit(EXTENT_FLAG_PINNED, &em->flags);
910 write_lock(&em_tree->lock);
911 ret = add_extent_mapping(em_tree, em);
912 write_unlock(&em_tree->lock);
913 if (ret != -EEXIST) {
917 btrfs_drop_extent_cache(inode, start,
918 start + ram_size - 1, 0);
921 cur_alloc_size = ins.offset;
922 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
923 ram_size, cur_alloc_size, 0);
924 BUG_ON(ret); /* -ENOMEM */
926 if (root->root_key.objectid ==
927 BTRFS_DATA_RELOC_TREE_OBJECTID) {
928 ret = btrfs_reloc_clone_csums(inode, start,
931 btrfs_abort_transaction(trans, root, ret);
936 if (disk_num_bytes < cur_alloc_size)
939 /* we're not doing compressed IO, don't unlock the first
940 * page (which the caller expects to stay locked), don't
941 * clear any dirty bits and don't set any writeback bits
943 * Do set the Private2 bit so we know this page was properly
944 * setup for writepage
946 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
947 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
950 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
951 start, start + ram_size - 1,
953 disk_num_bytes -= cur_alloc_size;
954 num_bytes -= cur_alloc_size;
955 alloc_hint = ins.objectid + ins.offset;
956 start += cur_alloc_size;
960 btrfs_end_transaction(trans, root);
964 extent_clear_unlock_delalloc(inode,
965 &BTRFS_I(inode)->io_tree,
966 start, end, locked_page,
967 EXTENT_CLEAR_UNLOCK_PAGE |
968 EXTENT_CLEAR_UNLOCK |
969 EXTENT_CLEAR_DELALLOC |
971 EXTENT_SET_WRITEBACK |
972 EXTENT_END_WRITEBACK);
978 * work queue call back to started compression on a file and pages
980 static noinline void async_cow_start(struct btrfs_work *work)
982 struct async_cow *async_cow;
984 async_cow = container_of(work, struct async_cow, work);
986 compress_file_range(async_cow->inode, async_cow->locked_page,
987 async_cow->start, async_cow->end, async_cow,
989 if (num_added == 0) {
990 btrfs_add_delayed_iput(async_cow->inode);
991 async_cow->inode = NULL;
996 * work queue call back to submit previously compressed pages
998 static noinline void async_cow_submit(struct btrfs_work *work)
1000 struct async_cow *async_cow;
1001 struct btrfs_root *root;
1002 unsigned long nr_pages;
1004 async_cow = container_of(work, struct async_cow, work);
1006 root = async_cow->root;
1007 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1010 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1012 if (atomic_read(&root->fs_info->async_delalloc_pages) <
1014 waitqueue_active(&root->fs_info->async_submit_wait))
1015 wake_up(&root->fs_info->async_submit_wait);
1017 if (async_cow->inode)
1018 submit_compressed_extents(async_cow->inode, async_cow);
1021 static noinline void async_cow_free(struct btrfs_work *work)
1023 struct async_cow *async_cow;
1024 async_cow = container_of(work, struct async_cow, work);
1025 if (async_cow->inode)
1026 btrfs_add_delayed_iput(async_cow->inode);
1030 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1031 u64 start, u64 end, int *page_started,
1032 unsigned long *nr_written)
1034 struct async_cow *async_cow;
1035 struct btrfs_root *root = BTRFS_I(inode)->root;
1036 unsigned long nr_pages;
1038 int limit = 10 * 1024 * 1024;
1040 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1041 1, 0, NULL, GFP_NOFS);
1042 while (start < end) {
1043 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1044 BUG_ON(!async_cow); /* -ENOMEM */
1045 async_cow->inode = igrab(inode);
1046 async_cow->root = root;
1047 async_cow->locked_page = locked_page;
1048 async_cow->start = start;
1050 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1053 cur_end = min(end, start + 512 * 1024 - 1);
1055 async_cow->end = cur_end;
1056 INIT_LIST_HEAD(&async_cow->extents);
1058 async_cow->work.func = async_cow_start;
1059 async_cow->work.ordered_func = async_cow_submit;
1060 async_cow->work.ordered_free = async_cow_free;
1061 async_cow->work.flags = 0;
1063 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1065 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1067 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1070 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1071 wait_event(root->fs_info->async_submit_wait,
1072 (atomic_read(&root->fs_info->async_delalloc_pages) <
1076 while (atomic_read(&root->fs_info->async_submit_draining) &&
1077 atomic_read(&root->fs_info->async_delalloc_pages)) {
1078 wait_event(root->fs_info->async_submit_wait,
1079 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1083 *nr_written += nr_pages;
1084 start = cur_end + 1;
1090 static noinline int csum_exist_in_range(struct btrfs_root *root,
1091 u64 bytenr, u64 num_bytes)
1094 struct btrfs_ordered_sum *sums;
1097 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1098 bytenr + num_bytes - 1, &list, 0);
1099 if (ret == 0 && list_empty(&list))
1102 while (!list_empty(&list)) {
1103 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1104 list_del(&sums->list);
1111 * when nowcow writeback call back. This checks for snapshots or COW copies
1112 * of the extents that exist in the file, and COWs the file as required.
1114 * If no cow copies or snapshots exist, we write directly to the existing
1117 static noinline int run_delalloc_nocow(struct inode *inode,
1118 struct page *locked_page,
1119 u64 start, u64 end, int *page_started, int force,
1120 unsigned long *nr_written)
1122 struct btrfs_root *root = BTRFS_I(inode)->root;
1123 struct btrfs_trans_handle *trans;
1124 struct extent_buffer *leaf;
1125 struct btrfs_path *path;
1126 struct btrfs_file_extent_item *fi;
1127 struct btrfs_key found_key;
1140 u64 ino = btrfs_ino(inode);
1142 path = btrfs_alloc_path();
1144 extent_clear_unlock_delalloc(inode,
1145 &BTRFS_I(inode)->io_tree,
1146 start, end, locked_page,
1147 EXTENT_CLEAR_UNLOCK_PAGE |
1148 EXTENT_CLEAR_UNLOCK |
1149 EXTENT_CLEAR_DELALLOC |
1150 EXTENT_CLEAR_DIRTY |
1151 EXTENT_SET_WRITEBACK |
1152 EXTENT_END_WRITEBACK);
1156 nolock = btrfs_is_free_space_inode(inode);
1159 trans = btrfs_join_transaction_nolock(root);
1161 trans = btrfs_join_transaction(root);
1163 if (IS_ERR(trans)) {
1164 extent_clear_unlock_delalloc(inode,
1165 &BTRFS_I(inode)->io_tree,
1166 start, end, locked_page,
1167 EXTENT_CLEAR_UNLOCK_PAGE |
1168 EXTENT_CLEAR_UNLOCK |
1169 EXTENT_CLEAR_DELALLOC |
1170 EXTENT_CLEAR_DIRTY |
1171 EXTENT_SET_WRITEBACK |
1172 EXTENT_END_WRITEBACK);
1173 btrfs_free_path(path);
1174 return PTR_ERR(trans);
1177 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1179 cow_start = (u64)-1;
1182 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1185 btrfs_abort_transaction(trans, root, ret);
1188 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1189 leaf = path->nodes[0];
1190 btrfs_item_key_to_cpu(leaf, &found_key,
1191 path->slots[0] - 1);
1192 if (found_key.objectid == ino &&
1193 found_key.type == BTRFS_EXTENT_DATA_KEY)
1198 leaf = path->nodes[0];
1199 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1200 ret = btrfs_next_leaf(root, path);
1202 btrfs_abort_transaction(trans, root, ret);
1207 leaf = path->nodes[0];
1213 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1215 if (found_key.objectid > ino ||
1216 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1217 found_key.offset > end)
1220 if (found_key.offset > cur_offset) {
1221 extent_end = found_key.offset;
1226 fi = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_file_extent_item);
1228 extent_type = btrfs_file_extent_type(leaf, fi);
1230 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1231 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1232 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1233 extent_offset = btrfs_file_extent_offset(leaf, fi);
1234 extent_end = found_key.offset +
1235 btrfs_file_extent_num_bytes(leaf, fi);
1236 if (extent_end <= start) {
1240 if (disk_bytenr == 0)
1242 if (btrfs_file_extent_compression(leaf, fi) ||
1243 btrfs_file_extent_encryption(leaf, fi) ||
1244 btrfs_file_extent_other_encoding(leaf, fi))
1246 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1248 if (btrfs_extent_readonly(root, disk_bytenr))
1250 if (btrfs_cross_ref_exist(trans, root, ino,
1252 extent_offset, disk_bytenr))
1254 disk_bytenr += extent_offset;
1255 disk_bytenr += cur_offset - found_key.offset;
1256 num_bytes = min(end + 1, extent_end) - cur_offset;
1258 * force cow if csum exists in the range.
1259 * this ensure that csum for a given extent are
1260 * either valid or do not exist.
1262 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1265 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266 extent_end = found_key.offset +
1267 btrfs_file_extent_inline_len(leaf, fi);
1268 extent_end = ALIGN(extent_end, root->sectorsize);
1273 if (extent_end <= start) {
1278 if (cow_start == (u64)-1)
1279 cow_start = cur_offset;
1280 cur_offset = extent_end;
1281 if (cur_offset > end)
1287 btrfs_release_path(path);
1288 if (cow_start != (u64)-1) {
1289 ret = cow_file_range(inode, locked_page, cow_start,
1290 found_key.offset - 1, page_started,
1293 btrfs_abort_transaction(trans, root, ret);
1296 cow_start = (u64)-1;
1299 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300 struct extent_map *em;
1301 struct extent_map_tree *em_tree;
1302 em_tree = &BTRFS_I(inode)->extent_tree;
1303 em = alloc_extent_map();
1304 BUG_ON(!em); /* -ENOMEM */
1305 em->start = cur_offset;
1306 em->orig_start = em->start;
1307 em->len = num_bytes;
1308 em->block_len = num_bytes;
1309 em->block_start = disk_bytenr;
1310 em->bdev = root->fs_info->fs_devices->latest_bdev;
1311 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1313 write_lock(&em_tree->lock);
1314 ret = add_extent_mapping(em_tree, em);
1315 write_unlock(&em_tree->lock);
1316 if (ret != -EEXIST) {
1317 free_extent_map(em);
1320 btrfs_drop_extent_cache(inode, em->start,
1321 em->start + em->len - 1, 0);
1323 type = BTRFS_ORDERED_PREALLOC;
1325 type = BTRFS_ORDERED_NOCOW;
1328 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1329 num_bytes, num_bytes, type);
1330 BUG_ON(ret); /* -ENOMEM */
1332 if (root->root_key.objectid ==
1333 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1334 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1337 btrfs_abort_transaction(trans, root, ret);
1342 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1343 cur_offset, cur_offset + num_bytes - 1,
1344 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1345 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1346 EXTENT_SET_PRIVATE2);
1347 cur_offset = extent_end;
1348 if (cur_offset > end)
1351 btrfs_release_path(path);
1353 if (cur_offset <= end && cow_start == (u64)-1) {
1354 cow_start = cur_offset;
1358 if (cow_start != (u64)-1) {
1359 ret = cow_file_range(inode, locked_page, cow_start, end,
1360 page_started, nr_written, 1);
1362 btrfs_abort_transaction(trans, root, ret);
1369 err = btrfs_end_transaction_nolock(trans, root);
1371 err = btrfs_end_transaction(trans, root);
1376 if (ret && cur_offset < end)
1377 extent_clear_unlock_delalloc(inode,
1378 &BTRFS_I(inode)->io_tree,
1379 cur_offset, end, locked_page,
1380 EXTENT_CLEAR_UNLOCK_PAGE |
1381 EXTENT_CLEAR_UNLOCK |
1382 EXTENT_CLEAR_DELALLOC |
1383 EXTENT_CLEAR_DIRTY |
1384 EXTENT_SET_WRITEBACK |
1385 EXTENT_END_WRITEBACK);
1387 btrfs_free_path(path);
1392 * extent_io.c call back to do delayed allocation processing
1394 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1395 u64 start, u64 end, int *page_started,
1396 unsigned long *nr_written)
1399 struct btrfs_root *root = BTRFS_I(inode)->root;
1401 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1402 ret = run_delalloc_nocow(inode, locked_page, start, end,
1403 page_started, 1, nr_written);
1404 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1405 ret = run_delalloc_nocow(inode, locked_page, start, end,
1406 page_started, 0, nr_written);
1407 } else if (!btrfs_test_opt(root, COMPRESS) &&
1408 !(BTRFS_I(inode)->force_compress) &&
1409 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1410 ret = cow_file_range(inode, locked_page, start, end,
1411 page_started, nr_written, 1);
1413 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1414 &BTRFS_I(inode)->runtime_flags);
1415 ret = cow_file_range_async(inode, locked_page, start, end,
1416 page_started, nr_written);
1421 static void btrfs_split_extent_hook(struct inode *inode,
1422 struct extent_state *orig, u64 split)
1424 /* not delalloc, ignore it */
1425 if (!(orig->state & EXTENT_DELALLOC))
1428 spin_lock(&BTRFS_I(inode)->lock);
1429 BTRFS_I(inode)->outstanding_extents++;
1430 spin_unlock(&BTRFS_I(inode)->lock);
1434 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1435 * extents so we can keep track of new extents that are just merged onto old
1436 * extents, such as when we are doing sequential writes, so we can properly
1437 * account for the metadata space we'll need.
1439 static void btrfs_merge_extent_hook(struct inode *inode,
1440 struct extent_state *new,
1441 struct extent_state *other)
1443 /* not delalloc, ignore it */
1444 if (!(other->state & EXTENT_DELALLOC))
1447 spin_lock(&BTRFS_I(inode)->lock);
1448 BTRFS_I(inode)->outstanding_extents--;
1449 spin_unlock(&BTRFS_I(inode)->lock);
1453 * extent_io.c set_bit_hook, used to track delayed allocation
1454 * bytes in this file, and to maintain the list of inodes that
1455 * have pending delalloc work to be done.
1457 static void btrfs_set_bit_hook(struct inode *inode,
1458 struct extent_state *state, int *bits)
1462 * set_bit and clear bit hooks normally require _irqsave/restore
1463 * but in this case, we are only testing for the DELALLOC
1464 * bit, which is only set or cleared with irqs on
1466 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1467 struct btrfs_root *root = BTRFS_I(inode)->root;
1468 u64 len = state->end + 1 - state->start;
1469 bool do_list = !btrfs_is_free_space_inode(inode);
1471 if (*bits & EXTENT_FIRST_DELALLOC) {
1472 *bits &= ~EXTENT_FIRST_DELALLOC;
1474 spin_lock(&BTRFS_I(inode)->lock);
1475 BTRFS_I(inode)->outstanding_extents++;
1476 spin_unlock(&BTRFS_I(inode)->lock);
1479 spin_lock(&root->fs_info->delalloc_lock);
1480 BTRFS_I(inode)->delalloc_bytes += len;
1481 root->fs_info->delalloc_bytes += len;
1482 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1483 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1484 &root->fs_info->delalloc_inodes);
1486 spin_unlock(&root->fs_info->delalloc_lock);
1491 * extent_io.c clear_bit_hook, see set_bit_hook for why
1493 static void btrfs_clear_bit_hook(struct inode *inode,
1494 struct extent_state *state, int *bits)
1497 * set_bit and clear bit hooks normally require _irqsave/restore
1498 * but in this case, we are only testing for the DELALLOC
1499 * bit, which is only set or cleared with irqs on
1501 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1502 struct btrfs_root *root = BTRFS_I(inode)->root;
1503 u64 len = state->end + 1 - state->start;
1504 bool do_list = !btrfs_is_free_space_inode(inode);
1506 if (*bits & EXTENT_FIRST_DELALLOC) {
1507 *bits &= ~EXTENT_FIRST_DELALLOC;
1508 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1509 spin_lock(&BTRFS_I(inode)->lock);
1510 BTRFS_I(inode)->outstanding_extents--;
1511 spin_unlock(&BTRFS_I(inode)->lock);
1514 if (*bits & EXTENT_DO_ACCOUNTING)
1515 btrfs_delalloc_release_metadata(inode, len);
1517 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1519 btrfs_free_reserved_data_space(inode, len);
1521 spin_lock(&root->fs_info->delalloc_lock);
1522 root->fs_info->delalloc_bytes -= len;
1523 BTRFS_I(inode)->delalloc_bytes -= len;
1525 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1526 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1527 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1529 spin_unlock(&root->fs_info->delalloc_lock);
1534 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1535 * we don't create bios that span stripes or chunks
1537 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1538 size_t size, struct bio *bio,
1539 unsigned long bio_flags)
1541 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1542 struct btrfs_mapping_tree *map_tree;
1543 u64 logical = (u64)bio->bi_sector << 9;
1548 if (bio_flags & EXTENT_BIO_COMPRESSED)
1551 length = bio->bi_size;
1552 map_tree = &root->fs_info->mapping_tree;
1553 map_length = length;
1554 ret = btrfs_map_block(map_tree, READ, logical,
1555 &map_length, NULL, 0);
1556 /* Will always return 0 or 1 with map_multi == NULL */
1558 if (map_length < length + size)
1564 * in order to insert checksums into the metadata in large chunks,
1565 * we wait until bio submission time. All the pages in the bio are
1566 * checksummed and sums are attached onto the ordered extent record.
1568 * At IO completion time the cums attached on the ordered extent record
1569 * are inserted into the btree
1571 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1572 struct bio *bio, int mirror_num,
1573 unsigned long bio_flags,
1576 struct btrfs_root *root = BTRFS_I(inode)->root;
1579 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1580 BUG_ON(ret); /* -ENOMEM */
1585 * in order to insert checksums into the metadata in large chunks,
1586 * we wait until bio submission time. All the pages in the bio are
1587 * checksummed and sums are attached onto the ordered extent record.
1589 * At IO completion time the cums attached on the ordered extent record
1590 * are inserted into the btree
1592 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1593 int mirror_num, unsigned long bio_flags,
1596 struct btrfs_root *root = BTRFS_I(inode)->root;
1597 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1601 * extent_io.c submission hook. This does the right thing for csum calculation
1602 * on write, or reading the csums from the tree before a read
1604 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1605 int mirror_num, unsigned long bio_flags,
1608 struct btrfs_root *root = BTRFS_I(inode)->root;
1613 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1615 if (btrfs_is_free_space_inode(inode))
1618 if (!(rw & REQ_WRITE)) {
1619 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1623 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1624 return btrfs_submit_compressed_read(inode, bio,
1625 mirror_num, bio_flags);
1626 } else if (!skip_sum) {
1627 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1632 } else if (!skip_sum) {
1633 /* csum items have already been cloned */
1634 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1636 /* we're doing a write, do the async checksumming */
1637 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1638 inode, rw, bio, mirror_num,
1639 bio_flags, bio_offset,
1640 __btrfs_submit_bio_start,
1641 __btrfs_submit_bio_done);
1645 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1649 * given a list of ordered sums record them in the inode. This happens
1650 * at IO completion time based on sums calculated at bio submission time.
1652 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1653 struct inode *inode, u64 file_offset,
1654 struct list_head *list)
1656 struct btrfs_ordered_sum *sum;
1658 list_for_each_entry(sum, list, list) {
1659 btrfs_csum_file_blocks(trans,
1660 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1665 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1666 struct extent_state **cached_state)
1668 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1670 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1671 cached_state, GFP_NOFS);
1674 /* see btrfs_writepage_start_hook for details on why this is required */
1675 struct btrfs_writepage_fixup {
1677 struct btrfs_work work;
1680 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1682 struct btrfs_writepage_fixup *fixup;
1683 struct btrfs_ordered_extent *ordered;
1684 struct extent_state *cached_state = NULL;
1686 struct inode *inode;
1691 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1695 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1696 ClearPageChecked(page);
1700 inode = page->mapping->host;
1701 page_start = page_offset(page);
1702 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1704 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1707 /* already ordered? We're done */
1708 if (PagePrivate2(page))
1711 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1713 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1714 page_end, &cached_state, GFP_NOFS);
1716 btrfs_start_ordered_extent(inode, ordered, 1);
1717 btrfs_put_ordered_extent(ordered);
1721 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1723 mapping_set_error(page->mapping, ret);
1724 end_extent_writepage(page, ret, page_start, page_end);
1725 ClearPageChecked(page);
1729 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1730 ClearPageChecked(page);
1731 set_page_dirty(page);
1733 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1734 &cached_state, GFP_NOFS);
1737 page_cache_release(page);
1742 * There are a few paths in the higher layers of the kernel that directly
1743 * set the page dirty bit without asking the filesystem if it is a
1744 * good idea. This causes problems because we want to make sure COW
1745 * properly happens and the data=ordered rules are followed.
1747 * In our case any range that doesn't have the ORDERED bit set
1748 * hasn't been properly setup for IO. We kick off an async process
1749 * to fix it up. The async helper will wait for ordered extents, set
1750 * the delalloc bit and make it safe to write the page.
1752 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1754 struct inode *inode = page->mapping->host;
1755 struct btrfs_writepage_fixup *fixup;
1756 struct btrfs_root *root = BTRFS_I(inode)->root;
1758 /* this page is properly in the ordered list */
1759 if (TestClearPagePrivate2(page))
1762 if (PageChecked(page))
1765 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1769 SetPageChecked(page);
1770 page_cache_get(page);
1771 fixup->work.func = btrfs_writepage_fixup_worker;
1773 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1777 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1778 struct inode *inode, u64 file_pos,
1779 u64 disk_bytenr, u64 disk_num_bytes,
1780 u64 num_bytes, u64 ram_bytes,
1781 u8 compression, u8 encryption,
1782 u16 other_encoding, int extent_type)
1784 struct btrfs_root *root = BTRFS_I(inode)->root;
1785 struct btrfs_file_extent_item *fi;
1786 struct btrfs_path *path;
1787 struct extent_buffer *leaf;
1788 struct btrfs_key ins;
1792 path = btrfs_alloc_path();
1796 path->leave_spinning = 1;
1799 * we may be replacing one extent in the tree with another.
1800 * The new extent is pinned in the extent map, and we don't want
1801 * to drop it from the cache until it is completely in the btree.
1803 * So, tell btrfs_drop_extents to leave this extent in the cache.
1804 * the caller is expected to unpin it and allow it to be merged
1807 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1812 ins.objectid = btrfs_ino(inode);
1813 ins.offset = file_pos;
1814 ins.type = BTRFS_EXTENT_DATA_KEY;
1815 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1818 leaf = path->nodes[0];
1819 fi = btrfs_item_ptr(leaf, path->slots[0],
1820 struct btrfs_file_extent_item);
1821 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1822 btrfs_set_file_extent_type(leaf, fi, extent_type);
1823 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1824 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1825 btrfs_set_file_extent_offset(leaf, fi, 0);
1826 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1827 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1828 btrfs_set_file_extent_compression(leaf, fi, compression);
1829 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1830 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1832 btrfs_unlock_up_safe(path, 1);
1833 btrfs_set_lock_blocking(leaf);
1835 btrfs_mark_buffer_dirty(leaf);
1837 inode_add_bytes(inode, num_bytes);
1839 ins.objectid = disk_bytenr;
1840 ins.offset = disk_num_bytes;
1841 ins.type = BTRFS_EXTENT_ITEM_KEY;
1842 ret = btrfs_alloc_reserved_file_extent(trans, root,
1843 root->root_key.objectid,
1844 btrfs_ino(inode), file_pos, &ins);
1846 btrfs_free_path(path);
1852 * helper function for btrfs_finish_ordered_io, this
1853 * just reads in some of the csum leaves to prime them into ram
1854 * before we start the transaction. It limits the amount of btree
1855 * reads required while inside the transaction.
1857 /* as ordered data IO finishes, this gets called so we can finish
1858 * an ordered extent if the range of bytes in the file it covers are
1861 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1863 struct inode *inode = ordered_extent->inode;
1864 struct btrfs_root *root = BTRFS_I(inode)->root;
1865 struct btrfs_trans_handle *trans = NULL;
1866 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1867 struct extent_state *cached_state = NULL;
1868 int compress_type = 0;
1872 nolock = btrfs_is_free_space_inode(inode);
1874 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1879 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1880 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1881 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1884 trans = btrfs_join_transaction_nolock(root);
1886 trans = btrfs_join_transaction(root);
1888 return PTR_ERR(trans);
1889 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1890 ret = btrfs_update_inode_fallback(trans, root, inode);
1891 if (ret) /* -ENOMEM or corruption */
1892 btrfs_abort_transaction(trans, root, ret);
1897 lock_extent_bits(io_tree, ordered_extent->file_offset,
1898 ordered_extent->file_offset + ordered_extent->len - 1,
1902 trans = btrfs_join_transaction_nolock(root);
1904 trans = btrfs_join_transaction(root);
1905 if (IS_ERR(trans)) {
1906 ret = PTR_ERR(trans);
1910 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1912 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1913 compress_type = ordered_extent->compress_type;
1914 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1915 BUG_ON(compress_type);
1916 ret = btrfs_mark_extent_written(trans, inode,
1917 ordered_extent->file_offset,
1918 ordered_extent->file_offset +
1919 ordered_extent->len);
1921 BUG_ON(root == root->fs_info->tree_root);
1922 ret = insert_reserved_file_extent(trans, inode,
1923 ordered_extent->file_offset,
1924 ordered_extent->start,
1925 ordered_extent->disk_len,
1926 ordered_extent->len,
1927 ordered_extent->len,
1928 compress_type, 0, 0,
1929 BTRFS_FILE_EXTENT_REG);
1930 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1931 ordered_extent->file_offset,
1932 ordered_extent->len);
1936 btrfs_abort_transaction(trans, root, ret);
1940 add_pending_csums(trans, inode, ordered_extent->file_offset,
1941 &ordered_extent->list);
1943 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1944 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1945 ret = btrfs_update_inode_fallback(trans, root, inode);
1946 if (ret) { /* -ENOMEM or corruption */
1947 btrfs_abort_transaction(trans, root, ret);
1953 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1954 ordered_extent->file_offset +
1955 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1957 if (root != root->fs_info->tree_root)
1958 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1961 btrfs_end_transaction_nolock(trans, root);
1963 btrfs_end_transaction(trans, root);
1967 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1968 ordered_extent->file_offset +
1969 ordered_extent->len - 1, NULL, GFP_NOFS);
1972 * This needs to be dont to make sure anybody waiting knows we are done
1973 * upating everything for this ordered extent.
1975 btrfs_remove_ordered_extent(inode, ordered_extent);
1978 btrfs_put_ordered_extent(ordered_extent);
1979 /* once for the tree */
1980 btrfs_put_ordered_extent(ordered_extent);
1985 static void finish_ordered_fn(struct btrfs_work *work)
1987 struct btrfs_ordered_extent *ordered_extent;
1988 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1989 btrfs_finish_ordered_io(ordered_extent);
1992 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1993 struct extent_state *state, int uptodate)
1995 struct inode *inode = page->mapping->host;
1996 struct btrfs_root *root = BTRFS_I(inode)->root;
1997 struct btrfs_ordered_extent *ordered_extent = NULL;
1998 struct btrfs_workers *workers;
2000 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2002 ClearPagePrivate2(page);
2003 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2004 end - start + 1, uptodate))
2007 ordered_extent->work.func = finish_ordered_fn;
2008 ordered_extent->work.flags = 0;
2010 if (btrfs_is_free_space_inode(inode))
2011 workers = &root->fs_info->endio_freespace_worker;
2013 workers = &root->fs_info->endio_write_workers;
2014 btrfs_queue_worker(workers, &ordered_extent->work);
2020 * when reads are done, we need to check csums to verify the data is correct
2021 * if there's a match, we allow the bio to finish. If not, the code in
2022 * extent_io.c will try to find good copies for us.
2024 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2025 struct extent_state *state, int mirror)
2027 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2028 struct inode *inode = page->mapping->host;
2029 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2031 u64 private = ~(u32)0;
2033 struct btrfs_root *root = BTRFS_I(inode)->root;
2036 if (PageChecked(page)) {
2037 ClearPageChecked(page);
2041 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2044 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2045 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2046 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2051 if (state && state->start == start) {
2052 private = state->private;
2055 ret = get_state_private(io_tree, start, &private);
2057 kaddr = kmap_atomic(page);
2061 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2062 btrfs_csum_final(csum, (char *)&csum);
2063 if (csum != private)
2066 kunmap_atomic(kaddr);
2071 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2073 (unsigned long long)btrfs_ino(page->mapping->host),
2074 (unsigned long long)start, csum,
2075 (unsigned long long)private);
2076 memset(kaddr + offset, 1, end - start + 1);
2077 flush_dcache_page(page);
2078 kunmap_atomic(kaddr);
2084 struct delayed_iput {
2085 struct list_head list;
2086 struct inode *inode;
2089 /* JDM: If this is fs-wide, why can't we add a pointer to
2090 * btrfs_inode instead and avoid the allocation? */
2091 void btrfs_add_delayed_iput(struct inode *inode)
2093 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2094 struct delayed_iput *delayed;
2096 if (atomic_add_unless(&inode->i_count, -1, 1))
2099 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2100 delayed->inode = inode;
2102 spin_lock(&fs_info->delayed_iput_lock);
2103 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2104 spin_unlock(&fs_info->delayed_iput_lock);
2107 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2110 struct btrfs_fs_info *fs_info = root->fs_info;
2111 struct delayed_iput *delayed;
2114 spin_lock(&fs_info->delayed_iput_lock);
2115 empty = list_empty(&fs_info->delayed_iputs);
2116 spin_unlock(&fs_info->delayed_iput_lock);
2120 down_read(&root->fs_info->cleanup_work_sem);
2121 spin_lock(&fs_info->delayed_iput_lock);
2122 list_splice_init(&fs_info->delayed_iputs, &list);
2123 spin_unlock(&fs_info->delayed_iput_lock);
2125 while (!list_empty(&list)) {
2126 delayed = list_entry(list.next, struct delayed_iput, list);
2127 list_del(&delayed->list);
2128 iput(delayed->inode);
2131 up_read(&root->fs_info->cleanup_work_sem);
2134 enum btrfs_orphan_cleanup_state {
2135 ORPHAN_CLEANUP_STARTED = 1,
2136 ORPHAN_CLEANUP_DONE = 2,
2140 * This is called in transaction commit time. If there are no orphan
2141 * files in the subvolume, it removes orphan item and frees block_rsv
2144 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2145 struct btrfs_root *root)
2147 struct btrfs_block_rsv *block_rsv;
2150 if (atomic_read(&root->orphan_inodes) ||
2151 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2154 spin_lock(&root->orphan_lock);
2155 if (atomic_read(&root->orphan_inodes)) {
2156 spin_unlock(&root->orphan_lock);
2160 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2161 spin_unlock(&root->orphan_lock);
2165 block_rsv = root->orphan_block_rsv;
2166 root->orphan_block_rsv = NULL;
2167 spin_unlock(&root->orphan_lock);
2169 if (root->orphan_item_inserted &&
2170 btrfs_root_refs(&root->root_item) > 0) {
2171 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2172 root->root_key.objectid);
2174 root->orphan_item_inserted = 0;
2178 WARN_ON(block_rsv->size > 0);
2179 btrfs_free_block_rsv(root, block_rsv);
2184 * This creates an orphan entry for the given inode in case something goes
2185 * wrong in the middle of an unlink/truncate.
2187 * NOTE: caller of this function should reserve 5 units of metadata for
2190 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2192 struct btrfs_root *root = BTRFS_I(inode)->root;
2193 struct btrfs_block_rsv *block_rsv = NULL;
2198 if (!root->orphan_block_rsv) {
2199 block_rsv = btrfs_alloc_block_rsv(root);
2204 spin_lock(&root->orphan_lock);
2205 if (!root->orphan_block_rsv) {
2206 root->orphan_block_rsv = block_rsv;
2207 } else if (block_rsv) {
2208 btrfs_free_block_rsv(root, block_rsv);
2212 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2213 &BTRFS_I(inode)->runtime_flags)) {
2216 * For proper ENOSPC handling, we should do orphan
2217 * cleanup when mounting. But this introduces backward
2218 * compatibility issue.
2220 if (!xchg(&root->orphan_item_inserted, 1))
2226 atomic_dec(&root->orphan_inodes);
2229 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2230 &BTRFS_I(inode)->runtime_flags))
2232 spin_unlock(&root->orphan_lock);
2234 /* grab metadata reservation from transaction handle */
2236 ret = btrfs_orphan_reserve_metadata(trans, inode);
2237 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2240 /* insert an orphan item to track this unlinked/truncated file */
2242 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2243 if (ret && ret != -EEXIST) {
2244 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2245 &BTRFS_I(inode)->runtime_flags);
2246 btrfs_abort_transaction(trans, root, ret);
2252 /* insert an orphan item to track subvolume contains orphan files */
2254 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2255 root->root_key.objectid);
2256 if (ret && ret != -EEXIST) {
2257 btrfs_abort_transaction(trans, root, ret);
2265 * We have done the truncate/delete so we can go ahead and remove the orphan
2266 * item for this particular inode.
2268 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2270 struct btrfs_root *root = BTRFS_I(inode)->root;
2271 int delete_item = 0;
2272 int release_rsv = 0;
2275 spin_lock(&root->orphan_lock);
2276 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2277 &BTRFS_I(inode)->runtime_flags))
2280 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2281 &BTRFS_I(inode)->runtime_flags))
2283 spin_unlock(&root->orphan_lock);
2285 if (trans && delete_item) {
2286 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2287 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2291 btrfs_orphan_release_metadata(inode);
2292 atomic_dec(&root->orphan_inodes);
2299 * this cleans up any orphans that may be left on the list from the last use
2302 int btrfs_orphan_cleanup(struct btrfs_root *root)
2304 struct btrfs_path *path;
2305 struct extent_buffer *leaf;
2306 struct btrfs_key key, found_key;
2307 struct btrfs_trans_handle *trans;
2308 struct inode *inode;
2309 u64 last_objectid = 0;
2310 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2312 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2315 path = btrfs_alloc_path();
2322 key.objectid = BTRFS_ORPHAN_OBJECTID;
2323 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2324 key.offset = (u64)-1;
2327 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2332 * if ret == 0 means we found what we were searching for, which
2333 * is weird, but possible, so only screw with path if we didn't
2334 * find the key and see if we have stuff that matches
2338 if (path->slots[0] == 0)
2343 /* pull out the item */
2344 leaf = path->nodes[0];
2345 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2347 /* make sure the item matches what we want */
2348 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2350 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2353 /* release the path since we're done with it */
2354 btrfs_release_path(path);
2357 * this is where we are basically btrfs_lookup, without the
2358 * crossing root thing. we store the inode number in the
2359 * offset of the orphan item.
2362 if (found_key.offset == last_objectid) {
2363 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2364 "stopping orphan cleanup\n");
2369 last_objectid = found_key.offset;
2371 found_key.objectid = found_key.offset;
2372 found_key.type = BTRFS_INODE_ITEM_KEY;
2373 found_key.offset = 0;
2374 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2375 ret = PTR_RET(inode);
2376 if (ret && ret != -ESTALE)
2379 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2380 struct btrfs_root *dead_root;
2381 struct btrfs_fs_info *fs_info = root->fs_info;
2382 int is_dead_root = 0;
2385 * this is an orphan in the tree root. Currently these
2386 * could come from 2 sources:
2387 * a) a snapshot deletion in progress
2388 * b) a free space cache inode
2389 * We need to distinguish those two, as the snapshot
2390 * orphan must not get deleted.
2391 * find_dead_roots already ran before us, so if this
2392 * is a snapshot deletion, we should find the root
2393 * in the dead_roots list
2395 spin_lock(&fs_info->trans_lock);
2396 list_for_each_entry(dead_root, &fs_info->dead_roots,
2398 if (dead_root->root_key.objectid ==
2399 found_key.objectid) {
2404 spin_unlock(&fs_info->trans_lock);
2406 /* prevent this orphan from being found again */
2407 key.offset = found_key.objectid - 1;
2412 * Inode is already gone but the orphan item is still there,
2413 * kill the orphan item.
2415 if (ret == -ESTALE) {
2416 trans = btrfs_start_transaction(root, 1);
2417 if (IS_ERR(trans)) {
2418 ret = PTR_ERR(trans);
2421 printk(KERN_ERR "auto deleting %Lu\n",
2422 found_key.objectid);
2423 ret = btrfs_del_orphan_item(trans, root,
2424 found_key.objectid);
2425 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2426 btrfs_end_transaction(trans, root);
2431 * add this inode to the orphan list so btrfs_orphan_del does
2432 * the proper thing when we hit it
2434 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2435 &BTRFS_I(inode)->runtime_flags);
2437 /* if we have links, this was a truncate, lets do that */
2438 if (inode->i_nlink) {
2439 if (!S_ISREG(inode->i_mode)) {
2445 ret = btrfs_truncate(inode);
2450 /* this will do delete_inode and everything for us */
2455 /* release the path since we're done with it */
2456 btrfs_release_path(path);
2458 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2460 if (root->orphan_block_rsv)
2461 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2464 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2465 trans = btrfs_join_transaction(root);
2467 btrfs_end_transaction(trans, root);
2471 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2473 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2477 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2478 btrfs_free_path(path);
2483 * very simple check to peek ahead in the leaf looking for xattrs. If we
2484 * don't find any xattrs, we know there can't be any acls.
2486 * slot is the slot the inode is in, objectid is the objectid of the inode
2488 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2489 int slot, u64 objectid)
2491 u32 nritems = btrfs_header_nritems(leaf);
2492 struct btrfs_key found_key;
2496 while (slot < nritems) {
2497 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2499 /* we found a different objectid, there must not be acls */
2500 if (found_key.objectid != objectid)
2503 /* we found an xattr, assume we've got an acl */
2504 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2508 * we found a key greater than an xattr key, there can't
2509 * be any acls later on
2511 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2518 * it goes inode, inode backrefs, xattrs, extents,
2519 * so if there are a ton of hard links to an inode there can
2520 * be a lot of backrefs. Don't waste time searching too hard,
2521 * this is just an optimization
2526 /* we hit the end of the leaf before we found an xattr or
2527 * something larger than an xattr. We have to assume the inode
2534 * read an inode from the btree into the in-memory inode
2536 static void btrfs_read_locked_inode(struct inode *inode)
2538 struct btrfs_path *path;
2539 struct extent_buffer *leaf;
2540 struct btrfs_inode_item *inode_item;
2541 struct btrfs_timespec *tspec;
2542 struct btrfs_root *root = BTRFS_I(inode)->root;
2543 struct btrfs_key location;
2547 bool filled = false;
2549 ret = btrfs_fill_inode(inode, &rdev);
2553 path = btrfs_alloc_path();
2557 path->leave_spinning = 1;
2558 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2560 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2564 leaf = path->nodes[0];
2569 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2570 struct btrfs_inode_item);
2571 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2572 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2573 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2574 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2575 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2577 tspec = btrfs_inode_atime(inode_item);
2578 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2579 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2581 tspec = btrfs_inode_mtime(inode_item);
2582 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2583 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2585 tspec = btrfs_inode_ctime(inode_item);
2586 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2587 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2589 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2590 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2591 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2592 inode->i_generation = BTRFS_I(inode)->generation;
2594 rdev = btrfs_inode_rdev(leaf, inode_item);
2596 BTRFS_I(inode)->index_cnt = (u64)-1;
2597 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2600 * try to precache a NULL acl entry for files that don't have
2601 * any xattrs or acls
2603 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2606 cache_no_acl(inode);
2608 btrfs_free_path(path);
2610 switch (inode->i_mode & S_IFMT) {
2612 inode->i_mapping->a_ops = &btrfs_aops;
2613 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2614 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2615 inode->i_fop = &btrfs_file_operations;
2616 inode->i_op = &btrfs_file_inode_operations;
2619 inode->i_fop = &btrfs_dir_file_operations;
2620 if (root == root->fs_info->tree_root)
2621 inode->i_op = &btrfs_dir_ro_inode_operations;
2623 inode->i_op = &btrfs_dir_inode_operations;
2626 inode->i_op = &btrfs_symlink_inode_operations;
2627 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2628 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2631 inode->i_op = &btrfs_special_inode_operations;
2632 init_special_inode(inode, inode->i_mode, rdev);
2636 btrfs_update_iflags(inode);
2640 btrfs_free_path(path);
2641 make_bad_inode(inode);
2645 * given a leaf and an inode, copy the inode fields into the leaf
2647 static void fill_inode_item(struct btrfs_trans_handle *trans,
2648 struct extent_buffer *leaf,
2649 struct btrfs_inode_item *item,
2650 struct inode *inode)
2652 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2653 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2654 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2655 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2656 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2658 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2659 inode->i_atime.tv_sec);
2660 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2661 inode->i_atime.tv_nsec);
2663 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2664 inode->i_mtime.tv_sec);
2665 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2666 inode->i_mtime.tv_nsec);
2668 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2669 inode->i_ctime.tv_sec);
2670 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2671 inode->i_ctime.tv_nsec);
2673 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2674 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2675 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2676 btrfs_set_inode_transid(leaf, item, trans->transid);
2677 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2678 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2679 btrfs_set_inode_block_group(leaf, item, 0);
2683 * copy everything in the in-memory inode into the btree.
2685 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2686 struct btrfs_root *root, struct inode *inode)
2688 struct btrfs_inode_item *inode_item;
2689 struct btrfs_path *path;
2690 struct extent_buffer *leaf;
2693 path = btrfs_alloc_path();
2697 path->leave_spinning = 1;
2698 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2706 btrfs_unlock_up_safe(path, 1);
2707 leaf = path->nodes[0];
2708 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2709 struct btrfs_inode_item);
2711 fill_inode_item(trans, leaf, inode_item, inode);
2712 btrfs_mark_buffer_dirty(leaf);
2713 btrfs_set_inode_last_trans(trans, inode);
2716 btrfs_free_path(path);
2721 * copy everything in the in-memory inode into the btree.
2723 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *root, struct inode *inode)
2729 * If the inode is a free space inode, we can deadlock during commit
2730 * if we put it into the delayed code.
2732 * The data relocation inode should also be directly updated
2735 if (!btrfs_is_free_space_inode(inode)
2736 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2737 btrfs_update_root_times(trans, root);
2739 ret = btrfs_delayed_update_inode(trans, root, inode);
2741 btrfs_set_inode_last_trans(trans, inode);
2745 return btrfs_update_inode_item(trans, root, inode);
2748 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2749 struct btrfs_root *root, struct inode *inode)
2753 ret = btrfs_update_inode(trans, root, inode);
2755 return btrfs_update_inode_item(trans, root, inode);
2760 * unlink helper that gets used here in inode.c and in the tree logging
2761 * recovery code. It remove a link in a directory with a given name, and
2762 * also drops the back refs in the inode to the directory
2764 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root,
2766 struct inode *dir, struct inode *inode,
2767 const char *name, int name_len)
2769 struct btrfs_path *path;
2771 struct extent_buffer *leaf;
2772 struct btrfs_dir_item *di;
2773 struct btrfs_key key;
2775 u64 ino = btrfs_ino(inode);
2776 u64 dir_ino = btrfs_ino(dir);
2778 path = btrfs_alloc_path();
2784 path->leave_spinning = 1;
2785 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2786 name, name_len, -1);
2795 leaf = path->nodes[0];
2796 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2797 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2800 btrfs_release_path(path);
2802 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2805 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2806 "inode %llu parent %llu\n", name_len, name,
2807 (unsigned long long)ino, (unsigned long long)dir_ino);
2808 btrfs_abort_transaction(trans, root, ret);
2812 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2814 btrfs_abort_transaction(trans, root, ret);
2818 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2820 if (ret != 0 && ret != -ENOENT) {
2821 btrfs_abort_transaction(trans, root, ret);
2825 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2830 btrfs_free_path(path);
2834 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2835 inode_inc_iversion(inode);
2836 inode_inc_iversion(dir);
2837 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2838 ret = btrfs_update_inode(trans, root, dir);
2843 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *root,
2845 struct inode *dir, struct inode *inode,
2846 const char *name, int name_len)
2849 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2851 btrfs_drop_nlink(inode);
2852 ret = btrfs_update_inode(trans, root, inode);
2858 /* helper to check if there is any shared block in the path */
2859 static int check_path_shared(struct btrfs_root *root,
2860 struct btrfs_path *path)
2862 struct extent_buffer *eb;
2866 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2869 if (!path->nodes[level])
2871 eb = path->nodes[level];
2872 if (!btrfs_block_can_be_shared(root, eb))
2874 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2883 * helper to start transaction for unlink and rmdir.
2885 * unlink and rmdir are special in btrfs, they do not always free space.
2886 * so in enospc case, we should make sure they will free space before
2887 * allowing them to use the global metadata reservation.
2889 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2890 struct dentry *dentry)
2892 struct btrfs_trans_handle *trans;
2893 struct btrfs_root *root = BTRFS_I(dir)->root;
2894 struct btrfs_path *path;
2895 struct btrfs_inode_ref *ref;
2896 struct btrfs_dir_item *di;
2897 struct inode *inode = dentry->d_inode;
2902 u64 ino = btrfs_ino(inode);
2903 u64 dir_ino = btrfs_ino(dir);
2906 * 1 for the possible orphan item
2907 * 1 for the dir item
2908 * 1 for the dir index
2909 * 1 for the inode ref
2910 * 1 for the inode ref in the tree log
2911 * 2 for the dir entries in the log
2914 trans = btrfs_start_transaction(root, 8);
2915 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2918 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2919 return ERR_PTR(-ENOSPC);
2921 /* check if there is someone else holds reference */
2922 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2923 return ERR_PTR(-ENOSPC);
2925 if (atomic_read(&inode->i_count) > 2)
2926 return ERR_PTR(-ENOSPC);
2928 if (xchg(&root->fs_info->enospc_unlink, 1))
2929 return ERR_PTR(-ENOSPC);
2931 path = btrfs_alloc_path();
2933 root->fs_info->enospc_unlink = 0;
2934 return ERR_PTR(-ENOMEM);
2937 /* 1 for the orphan item */
2938 trans = btrfs_start_transaction(root, 1);
2939 if (IS_ERR(trans)) {
2940 btrfs_free_path(path);
2941 root->fs_info->enospc_unlink = 0;
2945 path->skip_locking = 1;
2946 path->search_commit_root = 1;
2948 ret = btrfs_lookup_inode(trans, root, path,
2949 &BTRFS_I(dir)->location, 0);
2955 if (check_path_shared(root, path))
2960 btrfs_release_path(path);
2962 ret = btrfs_lookup_inode(trans, root, path,
2963 &BTRFS_I(inode)->location, 0);
2969 if (check_path_shared(root, path))
2974 btrfs_release_path(path);
2976 if (ret == 0 && S_ISREG(inode->i_mode)) {
2977 ret = btrfs_lookup_file_extent(trans, root, path,
2983 BUG_ON(ret == 0); /* Corruption */
2984 if (check_path_shared(root, path))
2986 btrfs_release_path(path);
2994 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2995 dentry->d_name.name, dentry->d_name.len, 0);
3001 if (check_path_shared(root, path))
3007 btrfs_release_path(path);
3009 ref = btrfs_lookup_inode_ref(trans, root, path,
3010 dentry->d_name.name, dentry->d_name.len,
3016 BUG_ON(!ref); /* Logic error */
3017 if (check_path_shared(root, path))
3019 index = btrfs_inode_ref_index(path->nodes[0], ref);
3020 btrfs_release_path(path);
3023 * This is a commit root search, if we can lookup inode item and other
3024 * relative items in the commit root, it means the transaction of
3025 * dir/file creation has been committed, and the dir index item that we
3026 * delay to insert has also been inserted into the commit root. So
3027 * we needn't worry about the delayed insertion of the dir index item
3030 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3031 dentry->d_name.name, dentry->d_name.len, 0);
3036 BUG_ON(ret == -ENOENT);
3037 if (check_path_shared(root, path))
3042 btrfs_free_path(path);
3043 /* Migrate the orphan reservation over */
3045 err = btrfs_block_rsv_migrate(trans->block_rsv,
3046 &root->fs_info->global_block_rsv,
3047 trans->bytes_reserved);
3050 btrfs_end_transaction(trans, root);
3051 root->fs_info->enospc_unlink = 0;
3052 return ERR_PTR(err);
3055 trans->block_rsv = &root->fs_info->global_block_rsv;
3059 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3060 struct btrfs_root *root)
3062 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3063 btrfs_block_rsv_release(root, trans->block_rsv,
3064 trans->bytes_reserved);
3065 trans->block_rsv = &root->fs_info->trans_block_rsv;
3066 BUG_ON(!root->fs_info->enospc_unlink);
3067 root->fs_info->enospc_unlink = 0;
3069 btrfs_end_transaction(trans, root);
3072 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3074 struct btrfs_root *root = BTRFS_I(dir)->root;
3075 struct btrfs_trans_handle *trans;
3076 struct inode *inode = dentry->d_inode;
3078 unsigned long nr = 0;
3080 trans = __unlink_start_trans(dir, dentry);
3082 return PTR_ERR(trans);
3084 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3086 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3087 dentry->d_name.name, dentry->d_name.len);
3091 if (inode->i_nlink == 0) {
3092 ret = btrfs_orphan_add(trans, inode);
3098 nr = trans->blocks_used;
3099 __unlink_end_trans(trans, root);
3100 btrfs_btree_balance_dirty(root, nr);
3104 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3105 struct btrfs_root *root,
3106 struct inode *dir, u64 objectid,
3107 const char *name, int name_len)
3109 struct btrfs_path *path;
3110 struct extent_buffer *leaf;
3111 struct btrfs_dir_item *di;
3112 struct btrfs_key key;
3115 u64 dir_ino = btrfs_ino(dir);
3117 path = btrfs_alloc_path();
3121 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3122 name, name_len, -1);
3123 if (IS_ERR_OR_NULL(di)) {
3131 leaf = path->nodes[0];
3132 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3133 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3134 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3136 btrfs_abort_transaction(trans, root, ret);
3139 btrfs_release_path(path);
3141 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3142 objectid, root->root_key.objectid,
3143 dir_ino, &index, name, name_len);
3145 if (ret != -ENOENT) {
3146 btrfs_abort_transaction(trans, root, ret);
3149 di = btrfs_search_dir_index_item(root, path, dir_ino,
3151 if (IS_ERR_OR_NULL(di)) {
3156 btrfs_abort_transaction(trans, root, ret);
3160 leaf = path->nodes[0];
3161 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3162 btrfs_release_path(path);
3165 btrfs_release_path(path);
3167 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3169 btrfs_abort_transaction(trans, root, ret);
3173 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3174 inode_inc_iversion(dir);
3175 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3176 ret = btrfs_update_inode(trans, root, dir);
3178 btrfs_abort_transaction(trans, root, ret);
3180 btrfs_free_path(path);
3184 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3186 struct inode *inode = dentry->d_inode;
3188 struct btrfs_root *root = BTRFS_I(dir)->root;
3189 struct btrfs_trans_handle *trans;
3190 unsigned long nr = 0;
3192 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3193 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3196 trans = __unlink_start_trans(dir, dentry);
3198 return PTR_ERR(trans);
3200 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3201 err = btrfs_unlink_subvol(trans, root, dir,
3202 BTRFS_I(inode)->location.objectid,
3203 dentry->d_name.name,
3204 dentry->d_name.len);
3208 err = btrfs_orphan_add(trans, inode);
3212 /* now the directory is empty */
3213 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3214 dentry->d_name.name, dentry->d_name.len);
3216 btrfs_i_size_write(inode, 0);
3218 nr = trans->blocks_used;
3219 __unlink_end_trans(trans, root);
3220 btrfs_btree_balance_dirty(root, nr);
3226 * this can truncate away extent items, csum items and directory items.
3227 * It starts at a high offset and removes keys until it can't find
3228 * any higher than new_size
3230 * csum items that cross the new i_size are truncated to the new size
3233 * min_type is the minimum key type to truncate down to. If set to 0, this
3234 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3236 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3237 struct btrfs_root *root,
3238 struct inode *inode,
3239 u64 new_size, u32 min_type)
3241 struct btrfs_path *path;
3242 struct extent_buffer *leaf;
3243 struct btrfs_file_extent_item *fi;
3244 struct btrfs_key key;
3245 struct btrfs_key found_key;
3246 u64 extent_start = 0;
3247 u64 extent_num_bytes = 0;
3248 u64 extent_offset = 0;
3250 u64 mask = root->sectorsize - 1;
3251 u32 found_type = (u8)-1;
3254 int pending_del_nr = 0;
3255 int pending_del_slot = 0;
3256 int extent_type = -1;
3259 u64 ino = btrfs_ino(inode);
3261 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3263 path = btrfs_alloc_path();
3268 if (root->ref_cows || root == root->fs_info->tree_root)
3269 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3272 * This function is also used to drop the items in the log tree before
3273 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3274 * it is used to drop the loged items. So we shouldn't kill the delayed
3277 if (min_type == 0 && root == BTRFS_I(inode)->root)
3278 btrfs_kill_delayed_inode_items(inode);
3281 key.offset = (u64)-1;
3285 path->leave_spinning = 1;
3286 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3293 /* there are no items in the tree for us to truncate, we're
3296 if (path->slots[0] == 0)
3303 leaf = path->nodes[0];
3304 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3305 found_type = btrfs_key_type(&found_key);
3307 if (found_key.objectid != ino)
3310 if (found_type < min_type)
3313 item_end = found_key.offset;
3314 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3315 fi = btrfs_item_ptr(leaf, path->slots[0],
3316 struct btrfs_file_extent_item);
3317 extent_type = btrfs_file_extent_type(leaf, fi);
3318 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3320 btrfs_file_extent_num_bytes(leaf, fi);
3321 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3322 item_end += btrfs_file_extent_inline_len(leaf,
3327 if (found_type > min_type) {
3330 if (item_end < new_size)
3332 if (found_key.offset >= new_size)
3338 /* FIXME, shrink the extent if the ref count is only 1 */
3339 if (found_type != BTRFS_EXTENT_DATA_KEY)
3342 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3344 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3346 u64 orig_num_bytes =
3347 btrfs_file_extent_num_bytes(leaf, fi);
3348 extent_num_bytes = new_size -
3349 found_key.offset + root->sectorsize - 1;
3350 extent_num_bytes = extent_num_bytes &
3351 ~((u64)root->sectorsize - 1);
3352 btrfs_set_file_extent_num_bytes(leaf, fi,
3354 num_dec = (orig_num_bytes -
3356 if (root->ref_cows && extent_start != 0)
3357 inode_sub_bytes(inode, num_dec);
3358 btrfs_mark_buffer_dirty(leaf);
3361 btrfs_file_extent_disk_num_bytes(leaf,
3363 extent_offset = found_key.offset -
3364 btrfs_file_extent_offset(leaf, fi);
3366 /* FIXME blocksize != 4096 */
3367 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3368 if (extent_start != 0) {
3371 inode_sub_bytes(inode, num_dec);
3374 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3376 * we can't truncate inline items that have had
3380 btrfs_file_extent_compression(leaf, fi) == 0 &&
3381 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3382 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3383 u32 size = new_size - found_key.offset;
3385 if (root->ref_cows) {
3386 inode_sub_bytes(inode, item_end + 1 -
3390 btrfs_file_extent_calc_inline_size(size);
3391 btrfs_truncate_item(trans, root, path,
3393 } else if (root->ref_cows) {
3394 inode_sub_bytes(inode, item_end + 1 -
3400 if (!pending_del_nr) {
3401 /* no pending yet, add ourselves */
3402 pending_del_slot = path->slots[0];
3404 } else if (pending_del_nr &&
3405 path->slots[0] + 1 == pending_del_slot) {
3406 /* hop on the pending chunk */
3408 pending_del_slot = path->slots[0];
3415 if (found_extent && (root->ref_cows ||
3416 root == root->fs_info->tree_root)) {
3417 btrfs_set_path_blocking(path);
3418 ret = btrfs_free_extent(trans, root, extent_start,
3419 extent_num_bytes, 0,
3420 btrfs_header_owner(leaf),
3421 ino, extent_offset, 0);
3425 if (found_type == BTRFS_INODE_ITEM_KEY)
3428 if (path->slots[0] == 0 ||
3429 path->slots[0] != pending_del_slot) {
3430 if (root->ref_cows &&
3431 BTRFS_I(inode)->location.objectid !=
3432 BTRFS_FREE_INO_OBJECTID) {
3436 if (pending_del_nr) {
3437 ret = btrfs_del_items(trans, root, path,
3441 btrfs_abort_transaction(trans,
3447 btrfs_release_path(path);
3454 if (pending_del_nr) {
3455 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3458 btrfs_abort_transaction(trans, root, ret);
3461 btrfs_free_path(path);
3466 * taken from block_truncate_page, but does cow as it zeros out
3467 * any bytes left in the last page in the file.
3469 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3471 struct inode *inode = mapping->host;
3472 struct btrfs_root *root = BTRFS_I(inode)->root;
3473 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3474 struct btrfs_ordered_extent *ordered;
3475 struct extent_state *cached_state = NULL;
3477 u32 blocksize = root->sectorsize;
3478 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3479 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3481 gfp_t mask = btrfs_alloc_write_mask(mapping);
3486 if ((offset & (blocksize - 1)) == 0)
3488 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3494 page = find_or_create_page(mapping, index, mask);
3496 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3500 page_start = page_offset(page);
3501 page_end = page_start + PAGE_CACHE_SIZE - 1;
3503 if (!PageUptodate(page)) {
3504 ret = btrfs_readpage(NULL, page);
3506 if (page->mapping != mapping) {
3508 page_cache_release(page);
3511 if (!PageUptodate(page)) {
3516 wait_on_page_writeback(page);
3518 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3519 set_page_extent_mapped(page);
3521 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3523 unlock_extent_cached(io_tree, page_start, page_end,
3524 &cached_state, GFP_NOFS);
3526 page_cache_release(page);
3527 btrfs_start_ordered_extent(inode, ordered, 1);
3528 btrfs_put_ordered_extent(ordered);
3532 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3533 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3534 0, 0, &cached_state, GFP_NOFS);
3536 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3539 unlock_extent_cached(io_tree, page_start, page_end,
3540 &cached_state, GFP_NOFS);
3545 if (offset != PAGE_CACHE_SIZE) {
3547 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3548 flush_dcache_page(page);
3551 ClearPageChecked(page);
3552 set_page_dirty(page);
3553 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3558 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3560 page_cache_release(page);
3566 * This function puts in dummy file extents for the area we're creating a hole
3567 * for. So if we are truncating this file to a larger size we need to insert
3568 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3569 * the range between oldsize and size
3571 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3573 struct btrfs_trans_handle *trans;
3574 struct btrfs_root *root = BTRFS_I(inode)->root;
3575 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3576 struct extent_map *em = NULL;
3577 struct extent_state *cached_state = NULL;
3578 u64 mask = root->sectorsize - 1;
3579 u64 hole_start = (oldsize + mask) & ~mask;
3580 u64 block_end = (size + mask) & ~mask;
3586 if (size <= hole_start)
3590 struct btrfs_ordered_extent *ordered;
3591 btrfs_wait_ordered_range(inode, hole_start,
3592 block_end - hole_start);
3593 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3595 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3598 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3599 &cached_state, GFP_NOFS);
3600 btrfs_put_ordered_extent(ordered);
3603 cur_offset = hole_start;
3605 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3606 block_end - cur_offset, 0);
3611 last_byte = min(extent_map_end(em), block_end);
3612 last_byte = (last_byte + mask) & ~mask;
3613 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3615 hole_size = last_byte - cur_offset;
3617 trans = btrfs_start_transaction(root, 3);
3618 if (IS_ERR(trans)) {
3619 err = PTR_ERR(trans);
3623 err = btrfs_drop_extents(trans, inode, cur_offset,
3624 cur_offset + hole_size,
3627 btrfs_abort_transaction(trans, root, err);
3628 btrfs_end_transaction(trans, root);
3632 err = btrfs_insert_file_extent(trans, root,
3633 btrfs_ino(inode), cur_offset, 0,
3634 0, hole_size, 0, hole_size,
3637 btrfs_abort_transaction(trans, root, err);
3638 btrfs_end_transaction(trans, root);
3642 btrfs_drop_extent_cache(inode, hole_start,
3645 btrfs_update_inode(trans, root, inode);
3646 btrfs_end_transaction(trans, root);
3648 free_extent_map(em);
3650 cur_offset = last_byte;
3651 if (cur_offset >= block_end)
3655 free_extent_map(em);
3656 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3661 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3663 struct btrfs_root *root = BTRFS_I(inode)->root;
3664 struct btrfs_trans_handle *trans;
3665 loff_t oldsize = i_size_read(inode);
3668 if (newsize == oldsize)
3671 if (newsize > oldsize) {
3672 truncate_pagecache(inode, oldsize, newsize);
3673 ret = btrfs_cont_expand(inode, oldsize, newsize);
3677 trans = btrfs_start_transaction(root, 1);
3679 return PTR_ERR(trans);
3681 i_size_write(inode, newsize);
3682 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3683 ret = btrfs_update_inode(trans, root, inode);
3684 btrfs_end_transaction(trans, root);
3688 * We're truncating a file that used to have good data down to
3689 * zero. Make sure it gets into the ordered flush list so that
3690 * any new writes get down to disk quickly.
3693 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3694 &BTRFS_I(inode)->runtime_flags);
3696 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3697 truncate_setsize(inode, newsize);
3698 ret = btrfs_truncate(inode);
3704 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3706 struct inode *inode = dentry->d_inode;
3707 struct btrfs_root *root = BTRFS_I(inode)->root;
3710 if (btrfs_root_readonly(root))
3713 err = inode_change_ok(inode, attr);
3717 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3718 err = btrfs_setsize(inode, attr->ia_size);
3723 if (attr->ia_valid) {
3724 setattr_copy(inode, attr);
3725 inode_inc_iversion(inode);
3726 err = btrfs_dirty_inode(inode);
3728 if (!err && attr->ia_valid & ATTR_MODE)
3729 err = btrfs_acl_chmod(inode);
3735 void btrfs_evict_inode(struct inode *inode)
3737 struct btrfs_trans_handle *trans;
3738 struct btrfs_root *root = BTRFS_I(inode)->root;
3739 struct btrfs_block_rsv *rsv, *global_rsv;
3740 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3744 trace_btrfs_inode_evict(inode);
3746 truncate_inode_pages(&inode->i_data, 0);
3747 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3748 btrfs_is_free_space_inode(inode)))
3751 if (is_bad_inode(inode)) {
3752 btrfs_orphan_del(NULL, inode);
3755 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3756 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3758 if (root->fs_info->log_root_recovering) {
3759 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3760 &BTRFS_I(inode)->runtime_flags));
3764 if (inode->i_nlink > 0) {
3765 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3769 rsv = btrfs_alloc_block_rsv(root);
3771 btrfs_orphan_del(NULL, inode);
3774 rsv->size = min_size;
3775 global_rsv = &root->fs_info->global_block_rsv;
3777 btrfs_i_size_write(inode, 0);
3780 * This is a bit simpler than btrfs_truncate since
3782 * 1) We've already reserved our space for our orphan item in the
3784 * 2) We're going to delete the inode item, so we don't need to update
3787 * So we just need to reserve some slack space in case we add bytes when
3788 * doing the truncate.
3791 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3794 * Try and steal from the global reserve since we will
3795 * likely not use this space anyway, we want to try as
3796 * hard as possible to get this to work.
3799 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3802 printk(KERN_WARNING "Could not get space for a "
3803 "delete, will truncate on mount %d\n", ret);
3804 btrfs_orphan_del(NULL, inode);
3805 btrfs_free_block_rsv(root, rsv);
3809 trans = btrfs_start_transaction(root, 0);
3810 if (IS_ERR(trans)) {
3811 btrfs_orphan_del(NULL, inode);
3812 btrfs_free_block_rsv(root, rsv);
3816 trans->block_rsv = rsv;
3818 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3822 nr = trans->blocks_used;
3823 btrfs_end_transaction(trans, root);
3825 btrfs_btree_balance_dirty(root, nr);
3828 btrfs_free_block_rsv(root, rsv);
3831 trans->block_rsv = root->orphan_block_rsv;
3832 ret = btrfs_orphan_del(trans, inode);
3836 trans->block_rsv = &root->fs_info->trans_block_rsv;
3837 if (!(root == root->fs_info->tree_root ||
3838 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3839 btrfs_return_ino(root, btrfs_ino(inode));
3841 nr = trans->blocks_used;
3842 btrfs_end_transaction(trans, root);
3843 btrfs_btree_balance_dirty(root, nr);
3850 * this returns the key found in the dir entry in the location pointer.
3851 * If no dir entries were found, location->objectid is 0.
3853 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3854 struct btrfs_key *location)
3856 const char *name = dentry->d_name.name;
3857 int namelen = dentry->d_name.len;
3858 struct btrfs_dir_item *di;
3859 struct btrfs_path *path;
3860 struct btrfs_root *root = BTRFS_I(dir)->root;
3863 path = btrfs_alloc_path();
3867 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3872 if (IS_ERR_OR_NULL(di))
3875 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3877 btrfs_free_path(path);
3880 location->objectid = 0;
3885 * when we hit a tree root in a directory, the btrfs part of the inode
3886 * needs to be changed to reflect the root directory of the tree root. This
3887 * is kind of like crossing a mount point.
3889 static int fixup_tree_root_location(struct btrfs_root *root,
3891 struct dentry *dentry,
3892 struct btrfs_key *location,
3893 struct btrfs_root **sub_root)
3895 struct btrfs_path *path;
3896 struct btrfs_root *new_root;
3897 struct btrfs_root_ref *ref;
3898 struct extent_buffer *leaf;
3902 path = btrfs_alloc_path();
3909 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3910 BTRFS_I(dir)->root->root_key.objectid,
3911 location->objectid);
3918 leaf = path->nodes[0];
3919 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3920 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3921 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3924 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3925 (unsigned long)(ref + 1),
3926 dentry->d_name.len);
3930 btrfs_release_path(path);
3932 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3933 if (IS_ERR(new_root)) {
3934 err = PTR_ERR(new_root);
3938 if (btrfs_root_refs(&new_root->root_item) == 0) {
3943 *sub_root = new_root;
3944 location->objectid = btrfs_root_dirid(&new_root->root_item);
3945 location->type = BTRFS_INODE_ITEM_KEY;
3946 location->offset = 0;
3949 btrfs_free_path(path);
3953 static void inode_tree_add(struct inode *inode)
3955 struct btrfs_root *root = BTRFS_I(inode)->root;
3956 struct btrfs_inode *entry;
3958 struct rb_node *parent;
3959 u64 ino = btrfs_ino(inode);
3961 p = &root->inode_tree.rb_node;
3964 if (inode_unhashed(inode))
3967 spin_lock(&root->inode_lock);
3970 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3972 if (ino < btrfs_ino(&entry->vfs_inode))
3973 p = &parent->rb_left;
3974 else if (ino > btrfs_ino(&entry->vfs_inode))
3975 p = &parent->rb_right;
3977 WARN_ON(!(entry->vfs_inode.i_state &
3978 (I_WILL_FREE | I_FREEING)));
3979 rb_erase(parent, &root->inode_tree);
3980 RB_CLEAR_NODE(parent);
3981 spin_unlock(&root->inode_lock);
3985 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3986 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3987 spin_unlock(&root->inode_lock);
3990 static void inode_tree_del(struct inode *inode)
3992 struct btrfs_root *root = BTRFS_I(inode)->root;
3995 spin_lock(&root->inode_lock);
3996 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3997 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3998 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3999 empty = RB_EMPTY_ROOT(&root->inode_tree);
4001 spin_unlock(&root->inode_lock);
4004 * Free space cache has inodes in the tree root, but the tree root has a
4005 * root_refs of 0, so this could end up dropping the tree root as a
4006 * snapshot, so we need the extra !root->fs_info->tree_root check to
4007 * make sure we don't drop it.
4009 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4010 root != root->fs_info->tree_root) {
4011 synchronize_srcu(&root->fs_info->subvol_srcu);
4012 spin_lock(&root->inode_lock);
4013 empty = RB_EMPTY_ROOT(&root->inode_tree);
4014 spin_unlock(&root->inode_lock);
4016 btrfs_add_dead_root(root);
4020 void btrfs_invalidate_inodes(struct btrfs_root *root)
4022 struct rb_node *node;
4023 struct rb_node *prev;
4024 struct btrfs_inode *entry;
4025 struct inode *inode;
4028 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4030 spin_lock(&root->inode_lock);
4032 node = root->inode_tree.rb_node;
4036 entry = rb_entry(node, struct btrfs_inode, rb_node);
4038 if (objectid < btrfs_ino(&entry->vfs_inode))
4039 node = node->rb_left;
4040 else if (objectid > btrfs_ino(&entry->vfs_inode))
4041 node = node->rb_right;
4047 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4048 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4052 prev = rb_next(prev);
4056 entry = rb_entry(node, struct btrfs_inode, rb_node);
4057 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4058 inode = igrab(&entry->vfs_inode);
4060 spin_unlock(&root->inode_lock);
4061 if (atomic_read(&inode->i_count) > 1)
4062 d_prune_aliases(inode);
4064 * btrfs_drop_inode will have it removed from
4065 * the inode cache when its usage count
4070 spin_lock(&root->inode_lock);
4074 if (cond_resched_lock(&root->inode_lock))
4077 node = rb_next(node);
4079 spin_unlock(&root->inode_lock);
4082 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4084 struct btrfs_iget_args *args = p;
4085 inode->i_ino = args->ino;
4086 BTRFS_I(inode)->root = args->root;
4090 static int btrfs_find_actor(struct inode *inode, void *opaque)
4092 struct btrfs_iget_args *args = opaque;
4093 return args->ino == btrfs_ino(inode) &&
4094 args->root == BTRFS_I(inode)->root;
4097 static struct inode *btrfs_iget_locked(struct super_block *s,
4099 struct btrfs_root *root)
4101 struct inode *inode;
4102 struct btrfs_iget_args args;
4103 args.ino = objectid;
4106 inode = iget5_locked(s, objectid, btrfs_find_actor,
4107 btrfs_init_locked_inode,
4112 /* Get an inode object given its location and corresponding root.
4113 * Returns in *is_new if the inode was read from disk
4115 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4116 struct btrfs_root *root, int *new)
4118 struct inode *inode;
4120 inode = btrfs_iget_locked(s, location->objectid, root);
4122 return ERR_PTR(-ENOMEM);
4124 if (inode->i_state & I_NEW) {
4125 BTRFS_I(inode)->root = root;
4126 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4127 btrfs_read_locked_inode(inode);
4128 if (!is_bad_inode(inode)) {
4129 inode_tree_add(inode);
4130 unlock_new_inode(inode);
4134 unlock_new_inode(inode);
4136 inode = ERR_PTR(-ESTALE);
4143 static struct inode *new_simple_dir(struct super_block *s,
4144 struct btrfs_key *key,
4145 struct btrfs_root *root)
4147 struct inode *inode = new_inode(s);
4150 return ERR_PTR(-ENOMEM);
4152 BTRFS_I(inode)->root = root;
4153 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4154 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4156 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4157 inode->i_op = &btrfs_dir_ro_inode_operations;
4158 inode->i_fop = &simple_dir_operations;
4159 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4160 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4165 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4167 struct inode *inode;
4168 struct btrfs_root *root = BTRFS_I(dir)->root;
4169 struct btrfs_root *sub_root = root;
4170 struct btrfs_key location;
4174 if (dentry->d_name.len > BTRFS_NAME_LEN)
4175 return ERR_PTR(-ENAMETOOLONG);
4177 if (unlikely(d_need_lookup(dentry))) {
4178 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4179 kfree(dentry->d_fsdata);
4180 dentry->d_fsdata = NULL;
4181 /* This thing is hashed, drop it for now */
4184 ret = btrfs_inode_by_name(dir, dentry, &location);
4188 return ERR_PTR(ret);
4190 if (location.objectid == 0)
4193 if (location.type == BTRFS_INODE_ITEM_KEY) {
4194 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4198 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4200 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4201 ret = fixup_tree_root_location(root, dir, dentry,
4202 &location, &sub_root);
4205 inode = ERR_PTR(ret);
4207 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4209 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4211 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4213 if (!IS_ERR(inode) && root != sub_root) {
4214 down_read(&root->fs_info->cleanup_work_sem);
4215 if (!(inode->i_sb->s_flags & MS_RDONLY))
4216 ret = btrfs_orphan_cleanup(sub_root);
4217 up_read(&root->fs_info->cleanup_work_sem);
4219 inode = ERR_PTR(ret);
4225 static int btrfs_dentry_delete(const struct dentry *dentry)
4227 struct btrfs_root *root;
4228 struct inode *inode = dentry->d_inode;
4230 if (!inode && !IS_ROOT(dentry))
4231 inode = dentry->d_parent->d_inode;
4234 root = BTRFS_I(inode)->root;
4235 if (btrfs_root_refs(&root->root_item) == 0)
4238 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4244 static void btrfs_dentry_release(struct dentry *dentry)
4246 if (dentry->d_fsdata)
4247 kfree(dentry->d_fsdata);
4250 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4255 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4256 if (unlikely(d_need_lookup(dentry))) {
4257 spin_lock(&dentry->d_lock);
4258 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4259 spin_unlock(&dentry->d_lock);
4264 unsigned char btrfs_filetype_table[] = {
4265 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4268 static int btrfs_real_readdir(struct file *filp, void *dirent,
4271 struct inode *inode = filp->f_dentry->d_inode;
4272 struct btrfs_root *root = BTRFS_I(inode)->root;
4273 struct btrfs_item *item;
4274 struct btrfs_dir_item *di;
4275 struct btrfs_key key;
4276 struct btrfs_key found_key;
4277 struct btrfs_path *path;
4278 struct list_head ins_list;
4279 struct list_head del_list;
4281 struct extent_buffer *leaf;
4283 unsigned char d_type;
4288 int key_type = BTRFS_DIR_INDEX_KEY;
4292 int is_curr = 0; /* filp->f_pos points to the current index? */
4294 /* FIXME, use a real flag for deciding about the key type */
4295 if (root->fs_info->tree_root == root)
4296 key_type = BTRFS_DIR_ITEM_KEY;
4298 /* special case for "." */
4299 if (filp->f_pos == 0) {
4300 over = filldir(dirent, ".", 1,
4301 filp->f_pos, btrfs_ino(inode), DT_DIR);
4306 /* special case for .., just use the back ref */
4307 if (filp->f_pos == 1) {
4308 u64 pino = parent_ino(filp->f_path.dentry);
4309 over = filldir(dirent, "..", 2,
4310 filp->f_pos, pino, DT_DIR);
4315 path = btrfs_alloc_path();
4321 if (key_type == BTRFS_DIR_INDEX_KEY) {
4322 INIT_LIST_HEAD(&ins_list);
4323 INIT_LIST_HEAD(&del_list);
4324 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4327 btrfs_set_key_type(&key, key_type);
4328 key.offset = filp->f_pos;
4329 key.objectid = btrfs_ino(inode);
4331 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4336 leaf = path->nodes[0];
4337 slot = path->slots[0];
4338 if (slot >= btrfs_header_nritems(leaf)) {
4339 ret = btrfs_next_leaf(root, path);
4347 item = btrfs_item_nr(leaf, slot);
4348 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4350 if (found_key.objectid != key.objectid)
4352 if (btrfs_key_type(&found_key) != key_type)
4354 if (found_key.offset < filp->f_pos)
4356 if (key_type == BTRFS_DIR_INDEX_KEY &&
4357 btrfs_should_delete_dir_index(&del_list,
4361 filp->f_pos = found_key.offset;
4364 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4366 di_total = btrfs_item_size(leaf, item);
4368 while (di_cur < di_total) {
4369 struct btrfs_key location;
4371 if (verify_dir_item(root, leaf, di))
4374 name_len = btrfs_dir_name_len(leaf, di);
4375 if (name_len <= sizeof(tmp_name)) {
4376 name_ptr = tmp_name;
4378 name_ptr = kmalloc(name_len, GFP_NOFS);
4384 read_extent_buffer(leaf, name_ptr,
4385 (unsigned long)(di + 1), name_len);
4387 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4388 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4391 /* is this a reference to our own snapshot? If so
4394 * In contrast to old kernels, we insert the snapshot's
4395 * dir item and dir index after it has been created, so
4396 * we won't find a reference to our own snapshot. We
4397 * still keep the following code for backward
4400 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4401 location.objectid == root->root_key.objectid) {
4405 over = filldir(dirent, name_ptr, name_len,
4406 found_key.offset, location.objectid,
4410 if (name_ptr != tmp_name)
4415 di_len = btrfs_dir_name_len(leaf, di) +
4416 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4418 di = (struct btrfs_dir_item *)((char *)di + di_len);
4424 if (key_type == BTRFS_DIR_INDEX_KEY) {
4427 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4433 /* Reached end of directory/root. Bump pos past the last item. */
4434 if (key_type == BTRFS_DIR_INDEX_KEY)
4436 * 32-bit glibc will use getdents64, but then strtol -
4437 * so the last number we can serve is this.
4439 filp->f_pos = 0x7fffffff;
4445 if (key_type == BTRFS_DIR_INDEX_KEY)
4446 btrfs_put_delayed_items(&ins_list, &del_list);
4447 btrfs_free_path(path);
4451 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4453 struct btrfs_root *root = BTRFS_I(inode)->root;
4454 struct btrfs_trans_handle *trans;
4456 bool nolock = false;
4458 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4461 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4464 if (wbc->sync_mode == WB_SYNC_ALL) {
4466 trans = btrfs_join_transaction_nolock(root);
4468 trans = btrfs_join_transaction(root);
4470 return PTR_ERR(trans);
4472 ret = btrfs_end_transaction_nolock(trans, root);
4474 ret = btrfs_commit_transaction(trans, root);
4480 * This is somewhat expensive, updating the tree every time the
4481 * inode changes. But, it is most likely to find the inode in cache.
4482 * FIXME, needs more benchmarking...there are no reasons other than performance
4483 * to keep or drop this code.
4485 int btrfs_dirty_inode(struct inode *inode)
4487 struct btrfs_root *root = BTRFS_I(inode)->root;
4488 struct btrfs_trans_handle *trans;
4491 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4494 trans = btrfs_join_transaction(root);
4496 return PTR_ERR(trans);
4498 ret = btrfs_update_inode(trans, root, inode);
4499 if (ret && ret == -ENOSPC) {
4500 /* whoops, lets try again with the full transaction */
4501 btrfs_end_transaction(trans, root);
4502 trans = btrfs_start_transaction(root, 1);
4504 return PTR_ERR(trans);
4506 ret = btrfs_update_inode(trans, root, inode);
4508 btrfs_end_transaction(trans, root);
4509 if (BTRFS_I(inode)->delayed_node)
4510 btrfs_balance_delayed_items(root);
4516 * This is a copy of file_update_time. We need this so we can return error on
4517 * ENOSPC for updating the inode in the case of file write and mmap writes.
4519 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4522 struct btrfs_root *root = BTRFS_I(inode)->root;
4524 if (btrfs_root_readonly(root))
4527 if (flags & S_VERSION)
4528 inode_inc_iversion(inode);
4529 if (flags & S_CTIME)
4530 inode->i_ctime = *now;
4531 if (flags & S_MTIME)
4532 inode->i_mtime = *now;
4533 if (flags & S_ATIME)
4534 inode->i_atime = *now;
4535 return btrfs_dirty_inode(inode);
4539 * find the highest existing sequence number in a directory
4540 * and then set the in-memory index_cnt variable to reflect
4541 * free sequence numbers
4543 static int btrfs_set_inode_index_count(struct inode *inode)
4545 struct btrfs_root *root = BTRFS_I(inode)->root;
4546 struct btrfs_key key, found_key;
4547 struct btrfs_path *path;
4548 struct extent_buffer *leaf;
4551 key.objectid = btrfs_ino(inode);
4552 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4553 key.offset = (u64)-1;
4555 path = btrfs_alloc_path();
4559 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4562 /* FIXME: we should be able to handle this */
4568 * MAGIC NUMBER EXPLANATION:
4569 * since we search a directory based on f_pos we have to start at 2
4570 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4571 * else has to start at 2
4573 if (path->slots[0] == 0) {
4574 BTRFS_I(inode)->index_cnt = 2;
4580 leaf = path->nodes[0];
4581 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4583 if (found_key.objectid != btrfs_ino(inode) ||
4584 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4585 BTRFS_I(inode)->index_cnt = 2;
4589 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4591 btrfs_free_path(path);
4596 * helper to find a free sequence number in a given directory. This current
4597 * code is very simple, later versions will do smarter things in the btree
4599 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4603 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4604 ret = btrfs_inode_delayed_dir_index_count(dir);
4606 ret = btrfs_set_inode_index_count(dir);
4612 *index = BTRFS_I(dir)->index_cnt;
4613 BTRFS_I(dir)->index_cnt++;
4618 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4619 struct btrfs_root *root,
4621 const char *name, int name_len,
4622 u64 ref_objectid, u64 objectid,
4623 umode_t mode, u64 *index)
4625 struct inode *inode;
4626 struct btrfs_inode_item *inode_item;
4627 struct btrfs_key *location;
4628 struct btrfs_path *path;
4629 struct btrfs_inode_ref *ref;
4630 struct btrfs_key key[2];
4636 path = btrfs_alloc_path();
4638 return ERR_PTR(-ENOMEM);
4640 inode = new_inode(root->fs_info->sb);
4642 btrfs_free_path(path);
4643 return ERR_PTR(-ENOMEM);
4647 * we have to initialize this early, so we can reclaim the inode
4648 * number if we fail afterwards in this function.
4650 inode->i_ino = objectid;
4653 trace_btrfs_inode_request(dir);
4655 ret = btrfs_set_inode_index(dir, index);
4657 btrfs_free_path(path);
4659 return ERR_PTR(ret);
4663 * index_cnt is ignored for everything but a dir,
4664 * btrfs_get_inode_index_count has an explanation for the magic
4667 BTRFS_I(inode)->index_cnt = 2;
4668 BTRFS_I(inode)->root = root;
4669 BTRFS_I(inode)->generation = trans->transid;
4670 inode->i_generation = BTRFS_I(inode)->generation;
4677 key[0].objectid = objectid;
4678 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4681 key[1].objectid = objectid;
4682 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4683 key[1].offset = ref_objectid;
4685 sizes[0] = sizeof(struct btrfs_inode_item);
4686 sizes[1] = name_len + sizeof(*ref);
4688 path->leave_spinning = 1;
4689 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4693 inode_init_owner(inode, dir, mode);
4694 inode_set_bytes(inode, 0);
4695 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4696 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4697 struct btrfs_inode_item);
4698 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4699 sizeof(*inode_item));
4700 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4702 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4703 struct btrfs_inode_ref);
4704 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4705 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4706 ptr = (unsigned long)(ref + 1);
4707 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4709 btrfs_mark_buffer_dirty(path->nodes[0]);
4710 btrfs_free_path(path);
4712 location = &BTRFS_I(inode)->location;
4713 location->objectid = objectid;
4714 location->offset = 0;
4715 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4717 btrfs_inherit_iflags(inode, dir);
4719 if (S_ISREG(mode)) {
4720 if (btrfs_test_opt(root, NODATASUM))
4721 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4722 if (btrfs_test_opt(root, NODATACOW) ||
4723 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4724 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4727 insert_inode_hash(inode);
4728 inode_tree_add(inode);
4730 trace_btrfs_inode_new(inode);
4731 btrfs_set_inode_last_trans(trans, inode);
4733 btrfs_update_root_times(trans, root);
4738 BTRFS_I(dir)->index_cnt--;
4739 btrfs_free_path(path);
4741 return ERR_PTR(ret);
4744 static inline u8 btrfs_inode_type(struct inode *inode)
4746 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4750 * utility function to add 'inode' into 'parent_inode' with
4751 * a give name and a given sequence number.
4752 * if 'add_backref' is true, also insert a backref from the
4753 * inode to the parent directory.
4755 int btrfs_add_link(struct btrfs_trans_handle *trans,
4756 struct inode *parent_inode, struct inode *inode,
4757 const char *name, int name_len, int add_backref, u64 index)
4760 struct btrfs_key key;
4761 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4762 u64 ino = btrfs_ino(inode);
4763 u64 parent_ino = btrfs_ino(parent_inode);
4765 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4766 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4769 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4773 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4774 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4775 key.objectid, root->root_key.objectid,
4776 parent_ino, index, name, name_len);
4777 } else if (add_backref) {
4778 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4782 /* Nothing to clean up yet */
4786 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4788 btrfs_inode_type(inode), index);
4792 btrfs_abort_transaction(trans, root, ret);
4796 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4798 inode_inc_iversion(parent_inode);
4799 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4800 ret = btrfs_update_inode(trans, root, parent_inode);
4802 btrfs_abort_transaction(trans, root, ret);
4806 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4809 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4810 key.objectid, root->root_key.objectid,
4811 parent_ino, &local_index, name, name_len);
4813 } else if (add_backref) {
4817 err = btrfs_del_inode_ref(trans, root, name, name_len,
4818 ino, parent_ino, &local_index);
4823 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4824 struct inode *dir, struct dentry *dentry,
4825 struct inode *inode, int backref, u64 index)
4827 int err = btrfs_add_link(trans, dir, inode,
4828 dentry->d_name.name, dentry->d_name.len,
4835 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4836 umode_t mode, dev_t rdev)
4838 struct btrfs_trans_handle *trans;
4839 struct btrfs_root *root = BTRFS_I(dir)->root;
4840 struct inode *inode = NULL;
4844 unsigned long nr = 0;
4847 if (!new_valid_dev(rdev))
4851 * 2 for inode item and ref
4853 * 1 for xattr if selinux is on
4855 trans = btrfs_start_transaction(root, 5);
4857 return PTR_ERR(trans);
4859 err = btrfs_find_free_ino(root, &objectid);
4863 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4864 dentry->d_name.len, btrfs_ino(dir), objectid,
4866 if (IS_ERR(inode)) {
4867 err = PTR_ERR(inode);
4871 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4878 * If the active LSM wants to access the inode during
4879 * d_instantiate it needs these. Smack checks to see
4880 * if the filesystem supports xattrs by looking at the
4884 inode->i_op = &btrfs_special_inode_operations;
4885 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4889 init_special_inode(inode, inode->i_mode, rdev);
4890 btrfs_update_inode(trans, root, inode);
4891 d_instantiate(dentry, inode);
4894 nr = trans->blocks_used;
4895 btrfs_end_transaction(trans, root);
4896 btrfs_btree_balance_dirty(root, nr);
4898 inode_dec_link_count(inode);
4904 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4905 umode_t mode, bool excl)
4907 struct btrfs_trans_handle *trans;
4908 struct btrfs_root *root = BTRFS_I(dir)->root;
4909 struct inode *inode = NULL;
4912 unsigned long nr = 0;
4917 * 2 for inode item and ref
4919 * 1 for xattr if selinux is on
4921 trans = btrfs_start_transaction(root, 5);
4923 return PTR_ERR(trans);
4925 err = btrfs_find_free_ino(root, &objectid);
4929 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4930 dentry->d_name.len, btrfs_ino(dir), objectid,
4932 if (IS_ERR(inode)) {
4933 err = PTR_ERR(inode);
4937 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4944 * If the active LSM wants to access the inode during
4945 * d_instantiate it needs these. Smack checks to see
4946 * if the filesystem supports xattrs by looking at the
4949 inode->i_fop = &btrfs_file_operations;
4950 inode->i_op = &btrfs_file_inode_operations;
4952 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4956 inode->i_mapping->a_ops = &btrfs_aops;
4957 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4958 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4959 d_instantiate(dentry, inode);
4962 nr = trans->blocks_used;
4963 btrfs_end_transaction(trans, root);
4965 inode_dec_link_count(inode);
4968 btrfs_btree_balance_dirty(root, nr);
4972 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4973 struct dentry *dentry)
4975 struct btrfs_trans_handle *trans;
4976 struct btrfs_root *root = BTRFS_I(dir)->root;
4977 struct inode *inode = old_dentry->d_inode;
4979 unsigned long nr = 0;
4983 /* do not allow sys_link's with other subvols of the same device */
4984 if (root->objectid != BTRFS_I(inode)->root->objectid)
4987 if (inode->i_nlink == ~0U)
4990 err = btrfs_set_inode_index(dir, &index);
4995 * 2 items for inode and inode ref
4996 * 2 items for dir items
4997 * 1 item for parent inode
4999 trans = btrfs_start_transaction(root, 5);
5000 if (IS_ERR(trans)) {
5001 err = PTR_ERR(trans);
5005 btrfs_inc_nlink(inode);
5006 inode_inc_iversion(inode);
5007 inode->i_ctime = CURRENT_TIME;
5010 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5015 struct dentry *parent = dentry->d_parent;
5016 err = btrfs_update_inode(trans, root, inode);
5019 d_instantiate(dentry, inode);
5020 btrfs_log_new_name(trans, inode, NULL, parent);
5023 nr = trans->blocks_used;
5024 btrfs_end_transaction(trans, root);
5027 inode_dec_link_count(inode);
5030 btrfs_btree_balance_dirty(root, nr);
5034 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5036 struct inode *inode = NULL;
5037 struct btrfs_trans_handle *trans;
5038 struct btrfs_root *root = BTRFS_I(dir)->root;
5040 int drop_on_err = 0;
5043 unsigned long nr = 1;
5046 * 2 items for inode and ref
5047 * 2 items for dir items
5048 * 1 for xattr if selinux is on
5050 trans = btrfs_start_transaction(root, 5);
5052 return PTR_ERR(trans);
5054 err = btrfs_find_free_ino(root, &objectid);
5058 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5059 dentry->d_name.len, btrfs_ino(dir), objectid,
5060 S_IFDIR | mode, &index);
5061 if (IS_ERR(inode)) {
5062 err = PTR_ERR(inode);
5068 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5072 inode->i_op = &btrfs_dir_inode_operations;
5073 inode->i_fop = &btrfs_dir_file_operations;
5075 btrfs_i_size_write(inode, 0);
5076 err = btrfs_update_inode(trans, root, inode);
5080 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5081 dentry->d_name.len, 0, index);
5085 d_instantiate(dentry, inode);
5089 nr = trans->blocks_used;
5090 btrfs_end_transaction(trans, root);
5093 btrfs_btree_balance_dirty(root, nr);
5097 /* helper for btfs_get_extent. Given an existing extent in the tree,
5098 * and an extent that you want to insert, deal with overlap and insert
5099 * the new extent into the tree.
5101 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5102 struct extent_map *existing,
5103 struct extent_map *em,
5104 u64 map_start, u64 map_len)
5108 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5109 start_diff = map_start - em->start;
5110 em->start = map_start;
5112 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5113 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5114 em->block_start += start_diff;
5115 em->block_len -= start_diff;
5117 return add_extent_mapping(em_tree, em);
5120 static noinline int uncompress_inline(struct btrfs_path *path,
5121 struct inode *inode, struct page *page,
5122 size_t pg_offset, u64 extent_offset,
5123 struct btrfs_file_extent_item *item)
5126 struct extent_buffer *leaf = path->nodes[0];
5129 unsigned long inline_size;
5133 WARN_ON(pg_offset != 0);
5134 compress_type = btrfs_file_extent_compression(leaf, item);
5135 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5136 inline_size = btrfs_file_extent_inline_item_len(leaf,
5137 btrfs_item_nr(leaf, path->slots[0]));
5138 tmp = kmalloc(inline_size, GFP_NOFS);
5141 ptr = btrfs_file_extent_inline_start(item);
5143 read_extent_buffer(leaf, tmp, ptr, inline_size);
5145 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5146 ret = btrfs_decompress(compress_type, tmp, page,
5147 extent_offset, inline_size, max_size);
5149 char *kaddr = kmap_atomic(page);
5150 unsigned long copy_size = min_t(u64,
5151 PAGE_CACHE_SIZE - pg_offset,
5152 max_size - extent_offset);
5153 memset(kaddr + pg_offset, 0, copy_size);
5154 kunmap_atomic(kaddr);
5161 * a bit scary, this does extent mapping from logical file offset to the disk.
5162 * the ugly parts come from merging extents from the disk with the in-ram
5163 * representation. This gets more complex because of the data=ordered code,
5164 * where the in-ram extents might be locked pending data=ordered completion.
5166 * This also copies inline extents directly into the page.
5169 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5170 size_t pg_offset, u64 start, u64 len,
5176 u64 extent_start = 0;
5178 u64 objectid = btrfs_ino(inode);
5180 struct btrfs_path *path = NULL;
5181 struct btrfs_root *root = BTRFS_I(inode)->root;
5182 struct btrfs_file_extent_item *item;
5183 struct extent_buffer *leaf;
5184 struct btrfs_key found_key;
5185 struct extent_map *em = NULL;
5186 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5187 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5188 struct btrfs_trans_handle *trans = NULL;
5192 read_lock(&em_tree->lock);
5193 em = lookup_extent_mapping(em_tree, start, len);
5195 em->bdev = root->fs_info->fs_devices->latest_bdev;
5196 read_unlock(&em_tree->lock);
5199 if (em->start > start || em->start + em->len <= start)
5200 free_extent_map(em);
5201 else if (em->block_start == EXTENT_MAP_INLINE && page)
5202 free_extent_map(em);
5206 em = alloc_extent_map();
5211 em->bdev = root->fs_info->fs_devices->latest_bdev;
5212 em->start = EXTENT_MAP_HOLE;
5213 em->orig_start = EXTENT_MAP_HOLE;
5215 em->block_len = (u64)-1;
5218 path = btrfs_alloc_path();
5224 * Chances are we'll be called again, so go ahead and do
5230 ret = btrfs_lookup_file_extent(trans, root, path,
5231 objectid, start, trans != NULL);
5238 if (path->slots[0] == 0)
5243 leaf = path->nodes[0];
5244 item = btrfs_item_ptr(leaf, path->slots[0],
5245 struct btrfs_file_extent_item);
5246 /* are we inside the extent that was found? */
5247 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5248 found_type = btrfs_key_type(&found_key);
5249 if (found_key.objectid != objectid ||
5250 found_type != BTRFS_EXTENT_DATA_KEY) {
5254 found_type = btrfs_file_extent_type(leaf, item);
5255 extent_start = found_key.offset;
5256 compress_type = btrfs_file_extent_compression(leaf, item);
5257 if (found_type == BTRFS_FILE_EXTENT_REG ||
5258 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5259 extent_end = extent_start +
5260 btrfs_file_extent_num_bytes(leaf, item);
5261 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5263 size = btrfs_file_extent_inline_len(leaf, item);
5264 extent_end = (extent_start + size + root->sectorsize - 1) &
5265 ~((u64)root->sectorsize - 1);
5268 if (start >= extent_end) {
5270 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5271 ret = btrfs_next_leaf(root, path);
5278 leaf = path->nodes[0];
5280 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5281 if (found_key.objectid != objectid ||
5282 found_key.type != BTRFS_EXTENT_DATA_KEY)
5284 if (start + len <= found_key.offset)
5287 em->len = found_key.offset - start;
5291 if (found_type == BTRFS_FILE_EXTENT_REG ||
5292 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5293 em->start = extent_start;
5294 em->len = extent_end - extent_start;
5295 em->orig_start = extent_start -
5296 btrfs_file_extent_offset(leaf, item);
5297 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5299 em->block_start = EXTENT_MAP_HOLE;
5302 if (compress_type != BTRFS_COMPRESS_NONE) {
5303 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5304 em->compress_type = compress_type;
5305 em->block_start = bytenr;
5306 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5309 bytenr += btrfs_file_extent_offset(leaf, item);
5310 em->block_start = bytenr;
5311 em->block_len = em->len;
5312 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5313 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5316 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5320 size_t extent_offset;
5323 em->block_start = EXTENT_MAP_INLINE;
5324 if (!page || create) {
5325 em->start = extent_start;
5326 em->len = extent_end - extent_start;
5330 size = btrfs_file_extent_inline_len(leaf, item);
5331 extent_offset = page_offset(page) + pg_offset - extent_start;
5332 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5333 size - extent_offset);
5334 em->start = extent_start + extent_offset;
5335 em->len = (copy_size + root->sectorsize - 1) &
5336 ~((u64)root->sectorsize - 1);
5337 em->orig_start = EXTENT_MAP_INLINE;
5338 if (compress_type) {
5339 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5340 em->compress_type = compress_type;
5342 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5343 if (create == 0 && !PageUptodate(page)) {
5344 if (btrfs_file_extent_compression(leaf, item) !=
5345 BTRFS_COMPRESS_NONE) {
5346 ret = uncompress_inline(path, inode, page,
5348 extent_offset, item);
5349 BUG_ON(ret); /* -ENOMEM */
5352 read_extent_buffer(leaf, map + pg_offset, ptr,
5354 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5355 memset(map + pg_offset + copy_size, 0,
5356 PAGE_CACHE_SIZE - pg_offset -
5361 flush_dcache_page(page);
5362 } else if (create && PageUptodate(page)) {
5366 free_extent_map(em);
5369 btrfs_release_path(path);
5370 trans = btrfs_join_transaction(root);
5373 return ERR_CAST(trans);
5377 write_extent_buffer(leaf, map + pg_offset, ptr,
5380 btrfs_mark_buffer_dirty(leaf);
5382 set_extent_uptodate(io_tree, em->start,
5383 extent_map_end(em) - 1, NULL, GFP_NOFS);
5386 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5393 em->block_start = EXTENT_MAP_HOLE;
5394 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5396 btrfs_release_path(path);
5397 if (em->start > start || extent_map_end(em) <= start) {
5398 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5399 "[%llu %llu]\n", (unsigned long long)em->start,
5400 (unsigned long long)em->len,
5401 (unsigned long long)start,
5402 (unsigned long long)len);
5408 write_lock(&em_tree->lock);
5409 ret = add_extent_mapping(em_tree, em);
5410 /* it is possible that someone inserted the extent into the tree
5411 * while we had the lock dropped. It is also possible that
5412 * an overlapping map exists in the tree
5414 if (ret == -EEXIST) {
5415 struct extent_map *existing;
5419 existing = lookup_extent_mapping(em_tree, start, len);
5420 if (existing && (existing->start > start ||
5421 existing->start + existing->len <= start)) {
5422 free_extent_map(existing);
5426 existing = lookup_extent_mapping(em_tree, em->start,
5429 err = merge_extent_mapping(em_tree, existing,
5432 free_extent_map(existing);
5434 free_extent_map(em);
5439 free_extent_map(em);
5443 free_extent_map(em);
5448 write_unlock(&em_tree->lock);
5451 trace_btrfs_get_extent(root, em);
5454 btrfs_free_path(path);
5456 ret = btrfs_end_transaction(trans, root);
5461 free_extent_map(em);
5462 return ERR_PTR(err);
5464 BUG_ON(!em); /* Error is always set */
5468 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5469 size_t pg_offset, u64 start, u64 len,
5472 struct extent_map *em;
5473 struct extent_map *hole_em = NULL;
5474 u64 range_start = start;
5480 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5485 * if our em maps to a hole, there might
5486 * actually be delalloc bytes behind it
5488 if (em->block_start != EXTENT_MAP_HOLE)
5494 /* check to see if we've wrapped (len == -1 or similar) */
5503 /* ok, we didn't find anything, lets look for delalloc */
5504 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5505 end, len, EXTENT_DELALLOC, 1);
5506 found_end = range_start + found;
5507 if (found_end < range_start)
5508 found_end = (u64)-1;
5511 * we didn't find anything useful, return
5512 * the original results from get_extent()
5514 if (range_start > end || found_end <= start) {
5520 /* adjust the range_start to make sure it doesn't
5521 * go backwards from the start they passed in
5523 range_start = max(start,range_start);
5524 found = found_end - range_start;
5527 u64 hole_start = start;
5530 em = alloc_extent_map();
5536 * when btrfs_get_extent can't find anything it
5537 * returns one huge hole
5539 * make sure what it found really fits our range, and
5540 * adjust to make sure it is based on the start from
5544 u64 calc_end = extent_map_end(hole_em);
5546 if (calc_end <= start || (hole_em->start > end)) {
5547 free_extent_map(hole_em);
5550 hole_start = max(hole_em->start, start);
5551 hole_len = calc_end - hole_start;
5555 if (hole_em && range_start > hole_start) {
5556 /* our hole starts before our delalloc, so we
5557 * have to return just the parts of the hole
5558 * that go until the delalloc starts
5560 em->len = min(hole_len,
5561 range_start - hole_start);
5562 em->start = hole_start;
5563 em->orig_start = hole_start;
5565 * don't adjust block start at all,
5566 * it is fixed at EXTENT_MAP_HOLE
5568 em->block_start = hole_em->block_start;
5569 em->block_len = hole_len;
5571 em->start = range_start;
5573 em->orig_start = range_start;
5574 em->block_start = EXTENT_MAP_DELALLOC;
5575 em->block_len = found;
5577 } else if (hole_em) {
5582 free_extent_map(hole_em);
5584 free_extent_map(em);
5585 return ERR_PTR(err);
5590 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5591 struct extent_map *em,
5594 struct btrfs_root *root = BTRFS_I(inode)->root;
5595 struct btrfs_trans_handle *trans;
5596 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5597 struct btrfs_key ins;
5600 bool insert = false;
5603 * Ok if the extent map we looked up is a hole and is for the exact
5604 * range we want, there is no reason to allocate a new one, however if
5605 * it is not right then we need to free this one and drop the cache for
5608 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5610 free_extent_map(em);
5613 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5616 trans = btrfs_join_transaction(root);
5618 return ERR_CAST(trans);
5620 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5621 btrfs_add_inode_defrag(trans, inode);
5623 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5625 alloc_hint = get_extent_allocation_hint(inode, start, len);
5626 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5627 alloc_hint, &ins, 1);
5634 em = alloc_extent_map();
5636 em = ERR_PTR(-ENOMEM);
5642 em->orig_start = em->start;
5643 em->len = ins.offset;
5645 em->block_start = ins.objectid;
5646 em->block_len = ins.offset;
5647 em->bdev = root->fs_info->fs_devices->latest_bdev;
5650 * We need to do this because if we're using the original em we searched
5651 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5654 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5657 write_lock(&em_tree->lock);
5658 ret = add_extent_mapping(em_tree, em);
5659 write_unlock(&em_tree->lock);
5662 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5665 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5666 ins.offset, ins.offset, 0);
5668 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5672 btrfs_end_transaction(trans, root);
5677 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5678 * block must be cow'd
5680 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5681 struct inode *inode, u64 offset, u64 len)
5683 struct btrfs_path *path;
5685 struct extent_buffer *leaf;
5686 struct btrfs_root *root = BTRFS_I(inode)->root;
5687 struct btrfs_file_extent_item *fi;
5688 struct btrfs_key key;
5696 path = btrfs_alloc_path();
5700 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5705 slot = path->slots[0];
5708 /* can't find the item, must cow */
5715 leaf = path->nodes[0];
5716 btrfs_item_key_to_cpu(leaf, &key, slot);
5717 if (key.objectid != btrfs_ino(inode) ||
5718 key.type != BTRFS_EXTENT_DATA_KEY) {
5719 /* not our file or wrong item type, must cow */
5723 if (key.offset > offset) {
5724 /* Wrong offset, must cow */
5728 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5729 found_type = btrfs_file_extent_type(leaf, fi);
5730 if (found_type != BTRFS_FILE_EXTENT_REG &&
5731 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5732 /* not a regular extent, must cow */
5735 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5736 backref_offset = btrfs_file_extent_offset(leaf, fi);
5738 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5739 if (extent_end < offset + len) {
5740 /* extent doesn't include our full range, must cow */
5744 if (btrfs_extent_readonly(root, disk_bytenr))
5748 * look for other files referencing this extent, if we
5749 * find any we must cow
5751 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5752 key.offset - backref_offset, disk_bytenr))
5756 * adjust disk_bytenr and num_bytes to cover just the bytes
5757 * in this extent we are about to write. If there
5758 * are any csums in that range we have to cow in order
5759 * to keep the csums correct
5761 disk_bytenr += backref_offset;
5762 disk_bytenr += offset - key.offset;
5763 num_bytes = min(offset + len, extent_end) - offset;
5764 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5767 * all of the above have passed, it is safe to overwrite this extent
5772 btrfs_free_path(path);
5776 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5777 struct buffer_head *bh_result, int create)
5779 struct extent_map *em;
5780 struct btrfs_root *root = BTRFS_I(inode)->root;
5781 u64 start = iblock << inode->i_blkbits;
5782 u64 len = bh_result->b_size;
5783 struct btrfs_trans_handle *trans;
5785 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5790 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5791 * io. INLINE is special, and we could probably kludge it in here, but
5792 * it's still buffered so for safety lets just fall back to the generic
5795 * For COMPRESSED we _have_ to read the entire extent in so we can
5796 * decompress it, so there will be buffering required no matter what we
5797 * do, so go ahead and fallback to buffered.
5799 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5800 * to buffered IO. Don't blame me, this is the price we pay for using
5803 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5804 em->block_start == EXTENT_MAP_INLINE) {
5805 free_extent_map(em);
5809 /* Just a good old fashioned hole, return */
5810 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5811 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5812 free_extent_map(em);
5813 /* DIO will do one hole at a time, so just unlock a sector */
5814 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5815 start + root->sectorsize - 1);
5820 * We don't allocate a new extent in the following cases
5822 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5824 * 2) The extent is marked as PREALLOC. We're good to go here and can
5825 * just use the extent.
5829 len = em->len - (start - em->start);
5833 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5834 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5835 em->block_start != EXTENT_MAP_HOLE)) {
5840 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5841 type = BTRFS_ORDERED_PREALLOC;
5843 type = BTRFS_ORDERED_NOCOW;
5844 len = min(len, em->len - (start - em->start));
5845 block_start = em->block_start + (start - em->start);
5848 * we're not going to log anything, but we do need
5849 * to make sure the current transaction stays open
5850 * while we look for nocow cross refs
5852 trans = btrfs_join_transaction(root);
5856 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5857 ret = btrfs_add_ordered_extent_dio(inode, start,
5858 block_start, len, len, type);
5859 btrfs_end_transaction(trans, root);
5861 free_extent_map(em);
5866 btrfs_end_transaction(trans, root);
5870 * this will cow the extent, reset the len in case we changed
5873 len = bh_result->b_size;
5874 em = btrfs_new_extent_direct(inode, em, start, len);
5877 len = min(len, em->len - (start - em->start));
5879 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5880 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5883 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5885 bh_result->b_size = len;
5886 bh_result->b_bdev = em->bdev;
5887 set_buffer_mapped(bh_result);
5889 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5890 set_buffer_new(bh_result);
5893 * Need to update the i_size under the extent lock so buffered
5894 * readers will get the updated i_size when we unlock.
5896 if (start + len > i_size_read(inode))
5897 i_size_write(inode, start + len);
5900 free_extent_map(em);
5905 struct btrfs_dio_private {
5906 struct inode *inode;
5913 /* number of bios pending for this dio */
5914 atomic_t pending_bios;
5919 struct bio *orig_bio;
5922 static void btrfs_endio_direct_read(struct bio *bio, int err)
5924 struct btrfs_dio_private *dip = bio->bi_private;
5925 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5926 struct bio_vec *bvec = bio->bi_io_vec;
5927 struct inode *inode = dip->inode;
5928 struct btrfs_root *root = BTRFS_I(inode)->root;
5930 u32 *private = dip->csums;
5932 start = dip->logical_offset;
5934 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5935 struct page *page = bvec->bv_page;
5938 unsigned long flags;
5940 local_irq_save(flags);
5941 kaddr = kmap_atomic(page);
5942 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5943 csum, bvec->bv_len);
5944 btrfs_csum_final(csum, (char *)&csum);
5945 kunmap_atomic(kaddr);
5946 local_irq_restore(flags);
5948 flush_dcache_page(bvec->bv_page);
5949 if (csum != *private) {
5950 printk(KERN_ERR "btrfs csum failed ino %llu off"
5951 " %llu csum %u private %u\n",
5952 (unsigned long long)btrfs_ino(inode),
5953 (unsigned long long)start,
5959 start += bvec->bv_len;
5962 } while (bvec <= bvec_end);
5964 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5965 dip->logical_offset + dip->bytes - 1);
5966 bio->bi_private = dip->private;
5971 /* If we had a csum failure make sure to clear the uptodate flag */
5973 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5974 dio_end_io(bio, err);
5977 static void btrfs_endio_direct_write(struct bio *bio, int err)
5979 struct btrfs_dio_private *dip = bio->bi_private;
5980 struct inode *inode = dip->inode;
5981 struct btrfs_root *root = BTRFS_I(inode)->root;
5982 struct btrfs_ordered_extent *ordered = NULL;
5983 u64 ordered_offset = dip->logical_offset;
5984 u64 ordered_bytes = dip->bytes;
5990 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5992 ordered_bytes, !err);
5996 ordered->work.func = finish_ordered_fn;
5997 ordered->work.flags = 0;
5998 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6002 * our bio might span multiple ordered extents. If we haven't
6003 * completed the accounting for the whole dio, go back and try again
6005 if (ordered_offset < dip->logical_offset + dip->bytes) {
6006 ordered_bytes = dip->logical_offset + dip->bytes -
6012 bio->bi_private = dip->private;
6016 /* If we had an error make sure to clear the uptodate flag */
6018 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6019 dio_end_io(bio, err);
6022 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6023 struct bio *bio, int mirror_num,
6024 unsigned long bio_flags, u64 offset)
6027 struct btrfs_root *root = BTRFS_I(inode)->root;
6028 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6029 BUG_ON(ret); /* -ENOMEM */
6033 static void btrfs_end_dio_bio(struct bio *bio, int err)
6035 struct btrfs_dio_private *dip = bio->bi_private;
6038 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6039 "sector %#Lx len %u err no %d\n",
6040 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6041 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6045 * before atomic variable goto zero, we must make sure
6046 * dip->errors is perceived to be set.
6048 smp_mb__before_atomic_dec();
6051 /* if there are more bios still pending for this dio, just exit */
6052 if (!atomic_dec_and_test(&dip->pending_bios))
6056 bio_io_error(dip->orig_bio);
6058 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6059 bio_endio(dip->orig_bio, 0);
6065 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6066 u64 first_sector, gfp_t gfp_flags)
6068 int nr_vecs = bio_get_nr_vecs(bdev);
6069 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6072 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6073 int rw, u64 file_offset, int skip_sum,
6074 u32 *csums, int async_submit)
6076 int write = rw & REQ_WRITE;
6077 struct btrfs_root *root = BTRFS_I(inode)->root;
6083 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6091 if (write && async_submit) {
6092 ret = btrfs_wq_submit_bio(root->fs_info,
6093 inode, rw, bio, 0, 0,
6095 __btrfs_submit_bio_start_direct_io,
6096 __btrfs_submit_bio_done);
6100 * If we aren't doing async submit, calculate the csum of the
6103 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6106 } else if (!skip_sum) {
6107 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6108 file_offset, csums);
6114 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6120 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6123 struct inode *inode = dip->inode;
6124 struct btrfs_root *root = BTRFS_I(inode)->root;
6125 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6127 struct bio *orig_bio = dip->orig_bio;
6128 struct bio_vec *bvec = orig_bio->bi_io_vec;
6129 u64 start_sector = orig_bio->bi_sector;
6130 u64 file_offset = dip->logical_offset;
6134 u32 *csums = dip->csums;
6136 int async_submit = 0;
6137 int write = rw & REQ_WRITE;
6139 map_length = orig_bio->bi_size;
6140 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6141 &map_length, NULL, 0);
6147 if (map_length >= orig_bio->bi_size) {
6153 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6156 bio->bi_private = dip;
6157 bio->bi_end_io = btrfs_end_dio_bio;
6158 atomic_inc(&dip->pending_bios);
6160 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6161 if (unlikely(map_length < submit_len + bvec->bv_len ||
6162 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6163 bvec->bv_offset) < bvec->bv_len)) {
6165 * inc the count before we submit the bio so
6166 * we know the end IO handler won't happen before
6167 * we inc the count. Otherwise, the dip might get freed
6168 * before we're done setting it up
6170 atomic_inc(&dip->pending_bios);
6171 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6172 file_offset, skip_sum,
6173 csums, async_submit);
6176 atomic_dec(&dip->pending_bios);
6180 /* Write's use the ordered csums */
6181 if (!write && !skip_sum)
6182 csums = csums + nr_pages;
6183 start_sector += submit_len >> 9;
6184 file_offset += submit_len;
6189 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6190 start_sector, GFP_NOFS);
6193 bio->bi_private = dip;
6194 bio->bi_end_io = btrfs_end_dio_bio;
6196 map_length = orig_bio->bi_size;
6197 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6198 &map_length, NULL, 0);
6204 submit_len += bvec->bv_len;
6211 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6212 csums, async_submit);
6220 * before atomic variable goto zero, we must
6221 * make sure dip->errors is perceived to be set.
6223 smp_mb__before_atomic_dec();
6224 if (atomic_dec_and_test(&dip->pending_bios))
6225 bio_io_error(dip->orig_bio);
6227 /* bio_end_io() will handle error, so we needn't return it */
6231 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6234 struct btrfs_root *root = BTRFS_I(inode)->root;
6235 struct btrfs_dio_private *dip;
6236 struct bio_vec *bvec = bio->bi_io_vec;
6238 int write = rw & REQ_WRITE;
6241 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6243 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6250 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6251 if (!write && !skip_sum) {
6252 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6260 dip->private = bio->bi_private;
6262 dip->logical_offset = file_offset;
6266 dip->bytes += bvec->bv_len;
6268 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6270 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6271 bio->bi_private = dip;
6273 dip->orig_bio = bio;
6274 atomic_set(&dip->pending_bios, 0);
6277 bio->bi_end_io = btrfs_endio_direct_write;
6279 bio->bi_end_io = btrfs_endio_direct_read;
6281 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6286 * If this is a write, we need to clean up the reserved space and kill
6287 * the ordered extent.
6290 struct btrfs_ordered_extent *ordered;
6291 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6292 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6293 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6294 btrfs_free_reserved_extent(root, ordered->start,
6296 btrfs_put_ordered_extent(ordered);
6297 btrfs_put_ordered_extent(ordered);
6299 bio_endio(bio, ret);
6302 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6303 const struct iovec *iov, loff_t offset,
6304 unsigned long nr_segs)
6310 unsigned blocksize_mask = root->sectorsize - 1;
6311 ssize_t retval = -EINVAL;
6312 loff_t end = offset;
6314 if (offset & blocksize_mask)
6317 /* Check the memory alignment. Blocks cannot straddle pages */
6318 for (seg = 0; seg < nr_segs; seg++) {
6319 addr = (unsigned long)iov[seg].iov_base;
6320 size = iov[seg].iov_len;
6322 if ((addr & blocksize_mask) || (size & blocksize_mask))
6325 /* If this is a write we don't need to check anymore */
6330 * Check to make sure we don't have duplicate iov_base's in this
6331 * iovec, if so return EINVAL, otherwise we'll get csum errors
6332 * when reading back.
6334 for (i = seg + 1; i < nr_segs; i++) {
6335 if (iov[seg].iov_base == iov[i].iov_base)
6343 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6344 const struct iovec *iov, loff_t offset,
6345 unsigned long nr_segs)
6347 struct file *file = iocb->ki_filp;
6348 struct inode *inode = file->f_mapping->host;
6349 struct btrfs_ordered_extent *ordered;
6350 struct extent_state *cached_state = NULL;
6351 u64 lockstart, lockend;
6353 int writing = rw & WRITE;
6355 size_t count = iov_length(iov, nr_segs);
6357 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6363 lockend = offset + count - 1;
6366 ret = btrfs_delalloc_reserve_space(inode, count);
6372 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6375 * We're concerned with the entire range that we're going to be
6376 * doing DIO to, so we need to make sure theres no ordered
6377 * extents in this range.
6379 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6380 lockend - lockstart + 1);
6383 * We need to make sure there are no buffered pages in this
6384 * range either, we could have raced between the invalidate in
6385 * generic_file_direct_write and locking the extent. The
6386 * invalidate needs to happen so that reads after a write do not
6389 if (!ordered && (!writing ||
6390 !test_range_bit(&BTRFS_I(inode)->io_tree,
6391 lockstart, lockend, EXTENT_UPTODATE, 0,
6395 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6396 &cached_state, GFP_NOFS);
6399 btrfs_start_ordered_extent(inode, ordered, 1);
6400 btrfs_put_ordered_extent(ordered);
6402 /* Screw you mmap */
6403 ret = filemap_write_and_wait_range(file->f_mapping,
6410 * If we found a page that couldn't be invalidated just
6411 * fall back to buffered.
6413 ret = invalidate_inode_pages2_range(file->f_mapping,
6414 lockstart >> PAGE_CACHE_SHIFT,
6415 lockend >> PAGE_CACHE_SHIFT);
6427 * we don't use btrfs_set_extent_delalloc because we don't want
6428 * the dirty or uptodate bits
6431 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6432 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6433 EXTENT_DELALLOC, NULL, &cached_state,
6436 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6437 lockend, EXTENT_LOCKED | write_bits,
6438 1, 0, &cached_state, GFP_NOFS);
6443 free_extent_state(cached_state);
6444 cached_state = NULL;
6446 ret = __blockdev_direct_IO(rw, iocb, inode,
6447 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6448 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6449 btrfs_submit_direct, 0);
6451 if (ret < 0 && ret != -EIOCBQUEUED) {
6452 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6453 offset + iov_length(iov, nr_segs) - 1,
6454 EXTENT_LOCKED | write_bits, 1, 0,
6455 &cached_state, GFP_NOFS);
6456 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6458 * We're falling back to buffered, unlock the section we didn't
6461 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6462 offset + iov_length(iov, nr_segs) - 1,
6463 EXTENT_LOCKED | write_bits, 1, 0,
6464 &cached_state, GFP_NOFS);
6467 free_extent_state(cached_state);
6471 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6472 __u64 start, __u64 len)
6474 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6477 int btrfs_readpage(struct file *file, struct page *page)
6479 struct extent_io_tree *tree;
6480 tree = &BTRFS_I(page->mapping->host)->io_tree;
6481 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6484 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6486 struct extent_io_tree *tree;
6489 if (current->flags & PF_MEMALLOC) {
6490 redirty_page_for_writepage(wbc, page);
6494 tree = &BTRFS_I(page->mapping->host)->io_tree;
6495 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6498 int btrfs_writepages(struct address_space *mapping,
6499 struct writeback_control *wbc)
6501 struct extent_io_tree *tree;
6503 tree = &BTRFS_I(mapping->host)->io_tree;
6504 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6508 btrfs_readpages(struct file *file, struct address_space *mapping,
6509 struct list_head *pages, unsigned nr_pages)
6511 struct extent_io_tree *tree;
6512 tree = &BTRFS_I(mapping->host)->io_tree;
6513 return extent_readpages(tree, mapping, pages, nr_pages,
6516 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6518 struct extent_io_tree *tree;
6519 struct extent_map_tree *map;
6522 tree = &BTRFS_I(page->mapping->host)->io_tree;
6523 map = &BTRFS_I(page->mapping->host)->extent_tree;
6524 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6526 ClearPagePrivate(page);
6527 set_page_private(page, 0);
6528 page_cache_release(page);
6533 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6535 if (PageWriteback(page) || PageDirty(page))
6537 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6540 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6542 struct inode *inode = page->mapping->host;
6543 struct extent_io_tree *tree;
6544 struct btrfs_ordered_extent *ordered;
6545 struct extent_state *cached_state = NULL;
6546 u64 page_start = page_offset(page);
6547 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6550 * we have the page locked, so new writeback can't start,
6551 * and the dirty bit won't be cleared while we are here.
6553 * Wait for IO on this page so that we can safely clear
6554 * the PagePrivate2 bit and do ordered accounting
6556 wait_on_page_writeback(page);
6558 tree = &BTRFS_I(inode)->io_tree;
6560 btrfs_releasepage(page, GFP_NOFS);
6563 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6564 ordered = btrfs_lookup_ordered_extent(inode,
6568 * IO on this page will never be started, so we need
6569 * to account for any ordered extents now
6571 clear_extent_bit(tree, page_start, page_end,
6572 EXTENT_DIRTY | EXTENT_DELALLOC |
6573 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6574 &cached_state, GFP_NOFS);
6576 * whoever cleared the private bit is responsible
6577 * for the finish_ordered_io
6579 if (TestClearPagePrivate2(page) &&
6580 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6581 PAGE_CACHE_SIZE, 1)) {
6582 btrfs_finish_ordered_io(ordered);
6584 btrfs_put_ordered_extent(ordered);
6585 cached_state = NULL;
6586 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6588 clear_extent_bit(tree, page_start, page_end,
6589 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6590 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6591 __btrfs_releasepage(page, GFP_NOFS);
6593 ClearPageChecked(page);
6594 if (PagePrivate(page)) {
6595 ClearPagePrivate(page);
6596 set_page_private(page, 0);
6597 page_cache_release(page);
6602 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6603 * called from a page fault handler when a page is first dirtied. Hence we must
6604 * be careful to check for EOF conditions here. We set the page up correctly
6605 * for a written page which means we get ENOSPC checking when writing into
6606 * holes and correct delalloc and unwritten extent mapping on filesystems that
6607 * support these features.
6609 * We are not allowed to take the i_mutex here so we have to play games to
6610 * protect against truncate races as the page could now be beyond EOF. Because
6611 * vmtruncate() writes the inode size before removing pages, once we have the
6612 * page lock we can determine safely if the page is beyond EOF. If it is not
6613 * beyond EOF, then the page is guaranteed safe against truncation until we
6616 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6618 struct page *page = vmf->page;
6619 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6620 struct btrfs_root *root = BTRFS_I(inode)->root;
6621 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6622 struct btrfs_ordered_extent *ordered;
6623 struct extent_state *cached_state = NULL;
6625 unsigned long zero_start;
6632 sb_start_pagefault(inode->i_sb);
6633 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6635 ret = file_update_time(vma->vm_file);
6641 else /* -ENOSPC, -EIO, etc */
6642 ret = VM_FAULT_SIGBUS;
6648 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6651 size = i_size_read(inode);
6652 page_start = page_offset(page);
6653 page_end = page_start + PAGE_CACHE_SIZE - 1;
6655 if ((page->mapping != inode->i_mapping) ||
6656 (page_start >= size)) {
6657 /* page got truncated out from underneath us */
6660 wait_on_page_writeback(page);
6662 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6663 set_page_extent_mapped(page);
6666 * we can't set the delalloc bits if there are pending ordered
6667 * extents. Drop our locks and wait for them to finish
6669 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6671 unlock_extent_cached(io_tree, page_start, page_end,
6672 &cached_state, GFP_NOFS);
6674 btrfs_start_ordered_extent(inode, ordered, 1);
6675 btrfs_put_ordered_extent(ordered);
6680 * XXX - page_mkwrite gets called every time the page is dirtied, even
6681 * if it was already dirty, so for space accounting reasons we need to
6682 * clear any delalloc bits for the range we are fixing to save. There
6683 * is probably a better way to do this, but for now keep consistent with
6684 * prepare_pages in the normal write path.
6686 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6687 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6688 0, 0, &cached_state, GFP_NOFS);
6690 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6693 unlock_extent_cached(io_tree, page_start, page_end,
6694 &cached_state, GFP_NOFS);
6695 ret = VM_FAULT_SIGBUS;
6700 /* page is wholly or partially inside EOF */
6701 if (page_start + PAGE_CACHE_SIZE > size)
6702 zero_start = size & ~PAGE_CACHE_MASK;
6704 zero_start = PAGE_CACHE_SIZE;
6706 if (zero_start != PAGE_CACHE_SIZE) {
6708 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6709 flush_dcache_page(page);
6712 ClearPageChecked(page);
6713 set_page_dirty(page);
6714 SetPageUptodate(page);
6716 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6717 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6719 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6723 sb_end_pagefault(inode->i_sb);
6724 return VM_FAULT_LOCKED;
6728 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6730 sb_end_pagefault(inode->i_sb);
6734 static int btrfs_truncate(struct inode *inode)
6736 struct btrfs_root *root = BTRFS_I(inode)->root;
6737 struct btrfs_block_rsv *rsv;
6740 struct btrfs_trans_handle *trans;
6742 u64 mask = root->sectorsize - 1;
6743 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6745 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6749 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6750 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6753 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6754 * 3 things going on here
6756 * 1) We need to reserve space for our orphan item and the space to
6757 * delete our orphan item. Lord knows we don't want to have a dangling
6758 * orphan item because we didn't reserve space to remove it.
6760 * 2) We need to reserve space to update our inode.
6762 * 3) We need to have something to cache all the space that is going to
6763 * be free'd up by the truncate operation, but also have some slack
6764 * space reserved in case it uses space during the truncate (thank you
6765 * very much snapshotting).
6767 * And we need these to all be seperate. The fact is we can use alot of
6768 * space doing the truncate, and we have no earthly idea how much space
6769 * we will use, so we need the truncate reservation to be seperate so it
6770 * doesn't end up using space reserved for updating the inode or
6771 * removing the orphan item. We also need to be able to stop the
6772 * transaction and start a new one, which means we need to be able to
6773 * update the inode several times, and we have no idea of knowing how
6774 * many times that will be, so we can't just reserve 1 item for the
6775 * entirety of the opration, so that has to be done seperately as well.
6776 * Then there is the orphan item, which does indeed need to be held on
6777 * to for the whole operation, and we need nobody to touch this reserved
6778 * space except the orphan code.
6780 * So that leaves us with
6782 * 1) root->orphan_block_rsv - for the orphan deletion.
6783 * 2) rsv - for the truncate reservation, which we will steal from the
6784 * transaction reservation.
6785 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6786 * updating the inode.
6788 rsv = btrfs_alloc_block_rsv(root);
6791 rsv->size = min_size;
6794 * 1 for the truncate slack space
6795 * 1 for the orphan item we're going to add
6796 * 1 for the orphan item deletion
6797 * 1 for updating the inode.
6799 trans = btrfs_start_transaction(root, 4);
6800 if (IS_ERR(trans)) {
6801 err = PTR_ERR(trans);
6805 /* Migrate the slack space for the truncate to our reserve */
6806 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6810 ret = btrfs_orphan_add(trans, inode);
6812 btrfs_end_transaction(trans, root);
6817 * setattr is responsible for setting the ordered_data_close flag,
6818 * but that is only tested during the last file release. That
6819 * could happen well after the next commit, leaving a great big
6820 * window where new writes may get lost if someone chooses to write
6821 * to this file after truncating to zero
6823 * The inode doesn't have any dirty data here, and so if we commit
6824 * this is a noop. If someone immediately starts writing to the inode
6825 * it is very likely we'll catch some of their writes in this
6826 * transaction, and the commit will find this file on the ordered
6827 * data list with good things to send down.
6829 * This is a best effort solution, there is still a window where
6830 * using truncate to replace the contents of the file will
6831 * end up with a zero length file after a crash.
6833 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6834 &BTRFS_I(inode)->runtime_flags))
6835 btrfs_add_ordered_operation(trans, root, inode);
6838 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6841 * This can only happen with the original transaction we
6842 * started above, every other time we shouldn't have a
6843 * transaction started yet.
6852 /* Just need the 1 for updating the inode */
6853 trans = btrfs_start_transaction(root, 1);
6854 if (IS_ERR(trans)) {
6855 ret = err = PTR_ERR(trans);
6861 trans->block_rsv = rsv;
6863 ret = btrfs_truncate_inode_items(trans, root, inode,
6865 BTRFS_EXTENT_DATA_KEY);
6866 if (ret != -EAGAIN) {
6871 trans->block_rsv = &root->fs_info->trans_block_rsv;
6872 ret = btrfs_update_inode(trans, root, inode);
6878 nr = trans->blocks_used;
6879 btrfs_end_transaction(trans, root);
6881 btrfs_btree_balance_dirty(root, nr);
6884 if (ret == 0 && inode->i_nlink > 0) {
6885 trans->block_rsv = root->orphan_block_rsv;
6886 ret = btrfs_orphan_del(trans, inode);
6889 } else if (ret && inode->i_nlink > 0) {
6891 * Failed to do the truncate, remove us from the in memory
6894 ret = btrfs_orphan_del(NULL, inode);
6898 trans->block_rsv = &root->fs_info->trans_block_rsv;
6899 ret = btrfs_update_inode(trans, root, inode);
6903 nr = trans->blocks_used;
6904 ret = btrfs_end_transaction(trans, root);
6905 btrfs_btree_balance_dirty(root, nr);
6909 btrfs_free_block_rsv(root, rsv);
6918 * create a new subvolume directory/inode (helper for the ioctl).
6920 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6921 struct btrfs_root *new_root, u64 new_dirid)
6923 struct inode *inode;
6927 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6928 new_dirid, new_dirid,
6929 S_IFDIR | (~current_umask() & S_IRWXUGO),
6932 return PTR_ERR(inode);
6933 inode->i_op = &btrfs_dir_inode_operations;
6934 inode->i_fop = &btrfs_dir_file_operations;
6936 set_nlink(inode, 1);
6937 btrfs_i_size_write(inode, 0);
6939 err = btrfs_update_inode(trans, new_root, inode);
6945 struct inode *btrfs_alloc_inode(struct super_block *sb)
6947 struct btrfs_inode *ei;
6948 struct inode *inode;
6950 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6957 ei->last_sub_trans = 0;
6958 ei->logged_trans = 0;
6959 ei->delalloc_bytes = 0;
6960 ei->disk_i_size = 0;
6963 ei->index_cnt = (u64)-1;
6964 ei->last_unlink_trans = 0;
6966 spin_lock_init(&ei->lock);
6967 ei->outstanding_extents = 0;
6968 ei->reserved_extents = 0;
6970 ei->runtime_flags = 0;
6971 ei->force_compress = BTRFS_COMPRESS_NONE;
6973 ei->delayed_node = NULL;
6975 inode = &ei->vfs_inode;
6976 extent_map_tree_init(&ei->extent_tree);
6977 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6978 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6979 ei->io_tree.track_uptodate = 1;
6980 ei->io_failure_tree.track_uptodate = 1;
6981 mutex_init(&ei->log_mutex);
6982 mutex_init(&ei->delalloc_mutex);
6983 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6984 INIT_LIST_HEAD(&ei->delalloc_inodes);
6985 INIT_LIST_HEAD(&ei->ordered_operations);
6986 RB_CLEAR_NODE(&ei->rb_node);
6991 static void btrfs_i_callback(struct rcu_head *head)
6993 struct inode *inode = container_of(head, struct inode, i_rcu);
6994 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6997 void btrfs_destroy_inode(struct inode *inode)
6999 struct btrfs_ordered_extent *ordered;
7000 struct btrfs_root *root = BTRFS_I(inode)->root;
7002 WARN_ON(!hlist_empty(&inode->i_dentry));
7003 WARN_ON(inode->i_data.nrpages);
7004 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7005 WARN_ON(BTRFS_I(inode)->reserved_extents);
7006 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7007 WARN_ON(BTRFS_I(inode)->csum_bytes);
7010 * This can happen where we create an inode, but somebody else also
7011 * created the same inode and we need to destroy the one we already
7018 * Make sure we're properly removed from the ordered operation
7022 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7023 spin_lock(&root->fs_info->ordered_extent_lock);
7024 list_del_init(&BTRFS_I(inode)->ordered_operations);
7025 spin_unlock(&root->fs_info->ordered_extent_lock);
7028 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7029 &BTRFS_I(inode)->runtime_flags)) {
7030 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7031 (unsigned long long)btrfs_ino(inode));
7032 atomic_dec(&root->orphan_inodes);
7036 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7040 printk(KERN_ERR "btrfs found ordered "
7041 "extent %llu %llu on inode cleanup\n",
7042 (unsigned long long)ordered->file_offset,
7043 (unsigned long long)ordered->len);
7044 btrfs_remove_ordered_extent(inode, ordered);
7045 btrfs_put_ordered_extent(ordered);
7046 btrfs_put_ordered_extent(ordered);
7049 inode_tree_del(inode);
7050 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7052 btrfs_remove_delayed_node(inode);
7053 call_rcu(&inode->i_rcu, btrfs_i_callback);
7056 int btrfs_drop_inode(struct inode *inode)
7058 struct btrfs_root *root = BTRFS_I(inode)->root;
7060 if (btrfs_root_refs(&root->root_item) == 0 &&
7061 !btrfs_is_free_space_inode(inode))
7064 return generic_drop_inode(inode);
7067 static void init_once(void *foo)
7069 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7071 inode_init_once(&ei->vfs_inode);
7074 void btrfs_destroy_cachep(void)
7076 if (btrfs_inode_cachep)
7077 kmem_cache_destroy(btrfs_inode_cachep);
7078 if (btrfs_trans_handle_cachep)
7079 kmem_cache_destroy(btrfs_trans_handle_cachep);
7080 if (btrfs_transaction_cachep)
7081 kmem_cache_destroy(btrfs_transaction_cachep);
7082 if (btrfs_path_cachep)
7083 kmem_cache_destroy(btrfs_path_cachep);
7084 if (btrfs_free_space_cachep)
7085 kmem_cache_destroy(btrfs_free_space_cachep);
7088 int btrfs_init_cachep(void)
7090 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7091 sizeof(struct btrfs_inode), 0,
7092 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7093 if (!btrfs_inode_cachep)
7096 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7097 sizeof(struct btrfs_trans_handle), 0,
7098 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7099 if (!btrfs_trans_handle_cachep)
7102 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7103 sizeof(struct btrfs_transaction), 0,
7104 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7105 if (!btrfs_transaction_cachep)
7108 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7109 sizeof(struct btrfs_path), 0,
7110 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7111 if (!btrfs_path_cachep)
7114 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7115 sizeof(struct btrfs_free_space), 0,
7116 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7117 if (!btrfs_free_space_cachep)
7122 btrfs_destroy_cachep();
7126 static int btrfs_getattr(struct vfsmount *mnt,
7127 struct dentry *dentry, struct kstat *stat)
7129 struct inode *inode = dentry->d_inode;
7130 u32 blocksize = inode->i_sb->s_blocksize;
7132 generic_fillattr(inode, stat);
7133 stat->dev = BTRFS_I(inode)->root->anon_dev;
7134 stat->blksize = PAGE_CACHE_SIZE;
7135 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7136 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7141 * If a file is moved, it will inherit the cow and compression flags of the new
7144 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7146 struct btrfs_inode *b_dir = BTRFS_I(dir);
7147 struct btrfs_inode *b_inode = BTRFS_I(inode);
7149 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7150 b_inode->flags |= BTRFS_INODE_NODATACOW;
7152 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7154 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7155 b_inode->flags |= BTRFS_INODE_COMPRESS;
7156 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7158 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7159 BTRFS_INODE_NOCOMPRESS);
7163 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7164 struct inode *new_dir, struct dentry *new_dentry)
7166 struct btrfs_trans_handle *trans;
7167 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7168 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7169 struct inode *new_inode = new_dentry->d_inode;
7170 struct inode *old_inode = old_dentry->d_inode;
7171 struct timespec ctime = CURRENT_TIME;
7175 u64 old_ino = btrfs_ino(old_inode);
7177 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7180 /* we only allow rename subvolume link between subvolumes */
7181 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7184 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7185 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7188 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7189 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7192 * we're using rename to replace one file with another.
7193 * and the replacement file is large. Start IO on it now so
7194 * we don't add too much work to the end of the transaction
7196 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7197 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7198 filemap_flush(old_inode->i_mapping);
7200 /* close the racy window with snapshot create/destroy ioctl */
7201 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7202 down_read(&root->fs_info->subvol_sem);
7204 * We want to reserve the absolute worst case amount of items. So if
7205 * both inodes are subvols and we need to unlink them then that would
7206 * require 4 item modifications, but if they are both normal inodes it
7207 * would require 5 item modifications, so we'll assume their normal
7208 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7209 * should cover the worst case number of items we'll modify.
7211 trans = btrfs_start_transaction(root, 20);
7212 if (IS_ERR(trans)) {
7213 ret = PTR_ERR(trans);
7218 btrfs_record_root_in_trans(trans, dest);
7220 ret = btrfs_set_inode_index(new_dir, &index);
7224 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7225 /* force full log commit if subvolume involved. */
7226 root->fs_info->last_trans_log_full_commit = trans->transid;
7228 ret = btrfs_insert_inode_ref(trans, dest,
7229 new_dentry->d_name.name,
7230 new_dentry->d_name.len,
7232 btrfs_ino(new_dir), index);
7236 * this is an ugly little race, but the rename is required
7237 * to make sure that if we crash, the inode is either at the
7238 * old name or the new one. pinning the log transaction lets
7239 * us make sure we don't allow a log commit to come in after
7240 * we unlink the name but before we add the new name back in.
7242 btrfs_pin_log_trans(root);
7245 * make sure the inode gets flushed if it is replacing
7248 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7249 btrfs_add_ordered_operation(trans, root, old_inode);
7251 inode_inc_iversion(old_dir);
7252 inode_inc_iversion(new_dir);
7253 inode_inc_iversion(old_inode);
7254 old_dir->i_ctime = old_dir->i_mtime = ctime;
7255 new_dir->i_ctime = new_dir->i_mtime = ctime;
7256 old_inode->i_ctime = ctime;
7258 if (old_dentry->d_parent != new_dentry->d_parent)
7259 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7261 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7262 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7263 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7264 old_dentry->d_name.name,
7265 old_dentry->d_name.len);
7267 ret = __btrfs_unlink_inode(trans, root, old_dir,
7268 old_dentry->d_inode,
7269 old_dentry->d_name.name,
7270 old_dentry->d_name.len);
7272 ret = btrfs_update_inode(trans, root, old_inode);
7275 btrfs_abort_transaction(trans, root, ret);
7280 inode_inc_iversion(new_inode);
7281 new_inode->i_ctime = CURRENT_TIME;
7282 if (unlikely(btrfs_ino(new_inode) ==
7283 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7284 root_objectid = BTRFS_I(new_inode)->location.objectid;
7285 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7287 new_dentry->d_name.name,
7288 new_dentry->d_name.len);
7289 BUG_ON(new_inode->i_nlink == 0);
7291 ret = btrfs_unlink_inode(trans, dest, new_dir,
7292 new_dentry->d_inode,
7293 new_dentry->d_name.name,
7294 new_dentry->d_name.len);
7296 if (!ret && new_inode->i_nlink == 0) {
7297 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7301 btrfs_abort_transaction(trans, root, ret);
7306 fixup_inode_flags(new_dir, old_inode);
7308 ret = btrfs_add_link(trans, new_dir, old_inode,
7309 new_dentry->d_name.name,
7310 new_dentry->d_name.len, 0, index);
7312 btrfs_abort_transaction(trans, root, ret);
7316 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7317 struct dentry *parent = new_dentry->d_parent;
7318 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7319 btrfs_end_log_trans(root);
7322 btrfs_end_transaction(trans, root);
7324 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7325 up_read(&root->fs_info->subvol_sem);
7331 * some fairly slow code that needs optimization. This walks the list
7332 * of all the inodes with pending delalloc and forces them to disk.
7334 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7336 struct list_head *head = &root->fs_info->delalloc_inodes;
7337 struct btrfs_inode *binode;
7338 struct inode *inode;
7340 if (root->fs_info->sb->s_flags & MS_RDONLY)
7343 spin_lock(&root->fs_info->delalloc_lock);
7344 while (!list_empty(head)) {
7345 binode = list_entry(head->next, struct btrfs_inode,
7347 inode = igrab(&binode->vfs_inode);
7349 list_del_init(&binode->delalloc_inodes);
7350 spin_unlock(&root->fs_info->delalloc_lock);
7352 filemap_flush(inode->i_mapping);
7354 btrfs_add_delayed_iput(inode);
7359 spin_lock(&root->fs_info->delalloc_lock);
7361 spin_unlock(&root->fs_info->delalloc_lock);
7363 /* the filemap_flush will queue IO into the worker threads, but
7364 * we have to make sure the IO is actually started and that
7365 * ordered extents get created before we return
7367 atomic_inc(&root->fs_info->async_submit_draining);
7368 while (atomic_read(&root->fs_info->nr_async_submits) ||
7369 atomic_read(&root->fs_info->async_delalloc_pages)) {
7370 wait_event(root->fs_info->async_submit_wait,
7371 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7372 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7374 atomic_dec(&root->fs_info->async_submit_draining);
7378 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7379 const char *symname)
7381 struct btrfs_trans_handle *trans;
7382 struct btrfs_root *root = BTRFS_I(dir)->root;
7383 struct btrfs_path *path;
7384 struct btrfs_key key;
7385 struct inode *inode = NULL;
7393 struct btrfs_file_extent_item *ei;
7394 struct extent_buffer *leaf;
7395 unsigned long nr = 0;
7397 name_len = strlen(symname) + 1;
7398 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7399 return -ENAMETOOLONG;
7402 * 2 items for inode item and ref
7403 * 2 items for dir items
7404 * 1 item for xattr if selinux is on
7406 trans = btrfs_start_transaction(root, 5);
7408 return PTR_ERR(trans);
7410 err = btrfs_find_free_ino(root, &objectid);
7414 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7415 dentry->d_name.len, btrfs_ino(dir), objectid,
7416 S_IFLNK|S_IRWXUGO, &index);
7417 if (IS_ERR(inode)) {
7418 err = PTR_ERR(inode);
7422 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7429 * If the active LSM wants to access the inode during
7430 * d_instantiate it needs these. Smack checks to see
7431 * if the filesystem supports xattrs by looking at the
7434 inode->i_fop = &btrfs_file_operations;
7435 inode->i_op = &btrfs_file_inode_operations;
7437 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7441 inode->i_mapping->a_ops = &btrfs_aops;
7442 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7443 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7448 path = btrfs_alloc_path();
7454 key.objectid = btrfs_ino(inode);
7456 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7457 datasize = btrfs_file_extent_calc_inline_size(name_len);
7458 err = btrfs_insert_empty_item(trans, root, path, &key,
7462 btrfs_free_path(path);
7465 leaf = path->nodes[0];
7466 ei = btrfs_item_ptr(leaf, path->slots[0],
7467 struct btrfs_file_extent_item);
7468 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7469 btrfs_set_file_extent_type(leaf, ei,
7470 BTRFS_FILE_EXTENT_INLINE);
7471 btrfs_set_file_extent_encryption(leaf, ei, 0);
7472 btrfs_set_file_extent_compression(leaf, ei, 0);
7473 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7474 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7476 ptr = btrfs_file_extent_inline_start(ei);
7477 write_extent_buffer(leaf, symname, ptr, name_len);
7478 btrfs_mark_buffer_dirty(leaf);
7479 btrfs_free_path(path);
7481 inode->i_op = &btrfs_symlink_inode_operations;
7482 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7483 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7484 inode_set_bytes(inode, name_len);
7485 btrfs_i_size_write(inode, name_len - 1);
7486 err = btrfs_update_inode(trans, root, inode);
7492 d_instantiate(dentry, inode);
7493 nr = trans->blocks_used;
7494 btrfs_end_transaction(trans, root);
7496 inode_dec_link_count(inode);
7499 btrfs_btree_balance_dirty(root, nr);
7503 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7504 u64 start, u64 num_bytes, u64 min_size,
7505 loff_t actual_len, u64 *alloc_hint,
7506 struct btrfs_trans_handle *trans)
7508 struct btrfs_root *root = BTRFS_I(inode)->root;
7509 struct btrfs_key ins;
7510 u64 cur_offset = start;
7513 bool own_trans = true;
7517 while (num_bytes > 0) {
7519 trans = btrfs_start_transaction(root, 3);
7520 if (IS_ERR(trans)) {
7521 ret = PTR_ERR(trans);
7526 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7527 0, *alloc_hint, &ins, 1);
7530 btrfs_end_transaction(trans, root);
7534 ret = insert_reserved_file_extent(trans, inode,
7535 cur_offset, ins.objectid,
7536 ins.offset, ins.offset,
7537 ins.offset, 0, 0, 0,
7538 BTRFS_FILE_EXTENT_PREALLOC);
7540 btrfs_abort_transaction(trans, root, ret);
7542 btrfs_end_transaction(trans, root);
7545 btrfs_drop_extent_cache(inode, cur_offset,
7546 cur_offset + ins.offset -1, 0);
7548 num_bytes -= ins.offset;
7549 cur_offset += ins.offset;
7550 *alloc_hint = ins.objectid + ins.offset;
7552 inode_inc_iversion(inode);
7553 inode->i_ctime = CURRENT_TIME;
7554 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7555 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7556 (actual_len > inode->i_size) &&
7557 (cur_offset > inode->i_size)) {
7558 if (cur_offset > actual_len)
7559 i_size = actual_len;
7561 i_size = cur_offset;
7562 i_size_write(inode, i_size);
7563 btrfs_ordered_update_i_size(inode, i_size, NULL);
7566 ret = btrfs_update_inode(trans, root, inode);
7569 btrfs_abort_transaction(trans, root, ret);
7571 btrfs_end_transaction(trans, root);
7576 btrfs_end_transaction(trans, root);
7581 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7582 u64 start, u64 num_bytes, u64 min_size,
7583 loff_t actual_len, u64 *alloc_hint)
7585 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7586 min_size, actual_len, alloc_hint,
7590 int btrfs_prealloc_file_range_trans(struct inode *inode,
7591 struct btrfs_trans_handle *trans, int mode,
7592 u64 start, u64 num_bytes, u64 min_size,
7593 loff_t actual_len, u64 *alloc_hint)
7595 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7596 min_size, actual_len, alloc_hint, trans);
7599 static int btrfs_set_page_dirty(struct page *page)
7601 return __set_page_dirty_nobuffers(page);
7604 static int btrfs_permission(struct inode *inode, int mask)
7606 struct btrfs_root *root = BTRFS_I(inode)->root;
7607 umode_t mode = inode->i_mode;
7609 if (mask & MAY_WRITE &&
7610 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7611 if (btrfs_root_readonly(root))
7613 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7616 return generic_permission(inode, mask);
7619 static const struct inode_operations btrfs_dir_inode_operations = {
7620 .getattr = btrfs_getattr,
7621 .lookup = btrfs_lookup,
7622 .create = btrfs_create,
7623 .unlink = btrfs_unlink,
7625 .mkdir = btrfs_mkdir,
7626 .rmdir = btrfs_rmdir,
7627 .rename = btrfs_rename,
7628 .symlink = btrfs_symlink,
7629 .setattr = btrfs_setattr,
7630 .mknod = btrfs_mknod,
7631 .setxattr = btrfs_setxattr,
7632 .getxattr = btrfs_getxattr,
7633 .listxattr = btrfs_listxattr,
7634 .removexattr = btrfs_removexattr,
7635 .permission = btrfs_permission,
7636 .get_acl = btrfs_get_acl,
7638 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7639 .lookup = btrfs_lookup,
7640 .permission = btrfs_permission,
7641 .get_acl = btrfs_get_acl,
7644 static const struct file_operations btrfs_dir_file_operations = {
7645 .llseek = generic_file_llseek,
7646 .read = generic_read_dir,
7647 .readdir = btrfs_real_readdir,
7648 .unlocked_ioctl = btrfs_ioctl,
7649 #ifdef CONFIG_COMPAT
7650 .compat_ioctl = btrfs_ioctl,
7652 .release = btrfs_release_file,
7653 .fsync = btrfs_sync_file,
7656 static struct extent_io_ops btrfs_extent_io_ops = {
7657 .fill_delalloc = run_delalloc_range,
7658 .submit_bio_hook = btrfs_submit_bio_hook,
7659 .merge_bio_hook = btrfs_merge_bio_hook,
7660 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7661 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7662 .writepage_start_hook = btrfs_writepage_start_hook,
7663 .set_bit_hook = btrfs_set_bit_hook,
7664 .clear_bit_hook = btrfs_clear_bit_hook,
7665 .merge_extent_hook = btrfs_merge_extent_hook,
7666 .split_extent_hook = btrfs_split_extent_hook,
7670 * btrfs doesn't support the bmap operation because swapfiles
7671 * use bmap to make a mapping of extents in the file. They assume
7672 * these extents won't change over the life of the file and they
7673 * use the bmap result to do IO directly to the drive.
7675 * the btrfs bmap call would return logical addresses that aren't
7676 * suitable for IO and they also will change frequently as COW
7677 * operations happen. So, swapfile + btrfs == corruption.
7679 * For now we're avoiding this by dropping bmap.
7681 static const struct address_space_operations btrfs_aops = {
7682 .readpage = btrfs_readpage,
7683 .writepage = btrfs_writepage,
7684 .writepages = btrfs_writepages,
7685 .readpages = btrfs_readpages,
7686 .direct_IO = btrfs_direct_IO,
7687 .invalidatepage = btrfs_invalidatepage,
7688 .releasepage = btrfs_releasepage,
7689 .set_page_dirty = btrfs_set_page_dirty,
7690 .error_remove_page = generic_error_remove_page,
7693 static const struct address_space_operations btrfs_symlink_aops = {
7694 .readpage = btrfs_readpage,
7695 .writepage = btrfs_writepage,
7696 .invalidatepage = btrfs_invalidatepage,
7697 .releasepage = btrfs_releasepage,
7700 static const struct inode_operations btrfs_file_inode_operations = {
7701 .getattr = btrfs_getattr,
7702 .setattr = btrfs_setattr,
7703 .setxattr = btrfs_setxattr,
7704 .getxattr = btrfs_getxattr,
7705 .listxattr = btrfs_listxattr,
7706 .removexattr = btrfs_removexattr,
7707 .permission = btrfs_permission,
7708 .fiemap = btrfs_fiemap,
7709 .get_acl = btrfs_get_acl,
7710 .update_time = btrfs_update_time,
7712 static const struct inode_operations btrfs_special_inode_operations = {
7713 .getattr = btrfs_getattr,
7714 .setattr = btrfs_setattr,
7715 .permission = btrfs_permission,
7716 .setxattr = btrfs_setxattr,
7717 .getxattr = btrfs_getxattr,
7718 .listxattr = btrfs_listxattr,
7719 .removexattr = btrfs_removexattr,
7720 .get_acl = btrfs_get_acl,
7721 .update_time = btrfs_update_time,
7723 static const struct inode_operations btrfs_symlink_inode_operations = {
7724 .readlink = generic_readlink,
7725 .follow_link = page_follow_link_light,
7726 .put_link = page_put_link,
7727 .getattr = btrfs_getattr,
7728 .setattr = btrfs_setattr,
7729 .permission = btrfs_permission,
7730 .setxattr = btrfs_setxattr,
7731 .getxattr = btrfs_getxattr,
7732 .listxattr = btrfs_listxattr,
7733 .removexattr = btrfs_removexattr,
7734 .get_acl = btrfs_get_acl,
7735 .update_time = btrfs_update_time,
7738 const struct dentry_operations btrfs_dentry_operations = {
7739 .d_delete = btrfs_dentry_delete,
7740 .d_release = btrfs_dentry_release,