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(root, 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 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 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 * 1042;
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(root, 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 ret = cow_file_range_async(inode, locked_page, start, end,
1414 page_started, nr_written);
1418 static void btrfs_split_extent_hook(struct inode *inode,
1419 struct extent_state *orig, u64 split)
1421 /* not delalloc, ignore it */
1422 if (!(orig->state & EXTENT_DELALLOC))
1425 spin_lock(&BTRFS_I(inode)->lock);
1426 BTRFS_I(inode)->outstanding_extents++;
1427 spin_unlock(&BTRFS_I(inode)->lock);
1431 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1432 * extents so we can keep track of new extents that are just merged onto old
1433 * extents, such as when we are doing sequential writes, so we can properly
1434 * account for the metadata space we'll need.
1436 static void btrfs_merge_extent_hook(struct inode *inode,
1437 struct extent_state *new,
1438 struct extent_state *other)
1440 /* not delalloc, ignore it */
1441 if (!(other->state & EXTENT_DELALLOC))
1444 spin_lock(&BTRFS_I(inode)->lock);
1445 BTRFS_I(inode)->outstanding_extents--;
1446 spin_unlock(&BTRFS_I(inode)->lock);
1450 * extent_io.c set_bit_hook, used to track delayed allocation
1451 * bytes in this file, and to maintain the list of inodes that
1452 * have pending delalloc work to be done.
1454 static void btrfs_set_bit_hook(struct inode *inode,
1455 struct extent_state *state, int *bits)
1459 * set_bit and clear bit hooks normally require _irqsave/restore
1460 * but in this case, we are only testing for the DELALLOC
1461 * bit, which is only set or cleared with irqs on
1463 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1464 struct btrfs_root *root = BTRFS_I(inode)->root;
1465 u64 len = state->end + 1 - state->start;
1466 bool do_list = !btrfs_is_free_space_inode(root, inode);
1468 if (*bits & EXTENT_FIRST_DELALLOC) {
1469 *bits &= ~EXTENT_FIRST_DELALLOC;
1471 spin_lock(&BTRFS_I(inode)->lock);
1472 BTRFS_I(inode)->outstanding_extents++;
1473 spin_unlock(&BTRFS_I(inode)->lock);
1476 spin_lock(&root->fs_info->delalloc_lock);
1477 BTRFS_I(inode)->delalloc_bytes += len;
1478 root->fs_info->delalloc_bytes += len;
1479 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1480 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1481 &root->fs_info->delalloc_inodes);
1483 spin_unlock(&root->fs_info->delalloc_lock);
1488 * extent_io.c clear_bit_hook, see set_bit_hook for why
1490 static void btrfs_clear_bit_hook(struct inode *inode,
1491 struct extent_state *state, int *bits)
1494 * set_bit and clear bit hooks normally require _irqsave/restore
1495 * but in this case, we are only testing for the DELALLOC
1496 * bit, which is only set or cleared with irqs on
1498 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1499 struct btrfs_root *root = BTRFS_I(inode)->root;
1500 u64 len = state->end + 1 - state->start;
1501 bool do_list = !btrfs_is_free_space_inode(root, inode);
1503 if (*bits & EXTENT_FIRST_DELALLOC) {
1504 *bits &= ~EXTENT_FIRST_DELALLOC;
1505 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1506 spin_lock(&BTRFS_I(inode)->lock);
1507 BTRFS_I(inode)->outstanding_extents--;
1508 spin_unlock(&BTRFS_I(inode)->lock);
1511 if (*bits & EXTENT_DO_ACCOUNTING)
1512 btrfs_delalloc_release_metadata(inode, len);
1514 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1516 btrfs_free_reserved_data_space(inode, len);
1518 spin_lock(&root->fs_info->delalloc_lock);
1519 root->fs_info->delalloc_bytes -= len;
1520 BTRFS_I(inode)->delalloc_bytes -= len;
1522 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1523 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1524 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1526 spin_unlock(&root->fs_info->delalloc_lock);
1531 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1532 * we don't create bios that span stripes or chunks
1534 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1535 size_t size, struct bio *bio,
1536 unsigned long bio_flags)
1538 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1539 struct btrfs_mapping_tree *map_tree;
1540 u64 logical = (u64)bio->bi_sector << 9;
1545 if (bio_flags & EXTENT_BIO_COMPRESSED)
1548 length = bio->bi_size;
1549 map_tree = &root->fs_info->mapping_tree;
1550 map_length = length;
1551 ret = btrfs_map_block(map_tree, READ, logical,
1552 &map_length, NULL, 0);
1553 /* Will always return 0 or 1 with map_multi == NULL */
1555 if (map_length < length + size)
1561 * in order to insert checksums into the metadata in large chunks,
1562 * we wait until bio submission time. All the pages in the bio are
1563 * checksummed and sums are attached onto the ordered extent record.
1565 * At IO completion time the cums attached on the ordered extent record
1566 * are inserted into the btree
1568 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1569 struct bio *bio, int mirror_num,
1570 unsigned long bio_flags,
1573 struct btrfs_root *root = BTRFS_I(inode)->root;
1576 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1577 BUG_ON(ret); /* -ENOMEM */
1582 * in order to insert checksums into the metadata in large chunks,
1583 * we wait until bio submission time. All the pages in the bio are
1584 * checksummed and sums are attached onto the ordered extent record.
1586 * At IO completion time the cums attached on the ordered extent record
1587 * are inserted into the btree
1589 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1590 int mirror_num, unsigned long bio_flags,
1593 struct btrfs_root *root = BTRFS_I(inode)->root;
1594 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1598 * extent_io.c submission hook. This does the right thing for csum calculation
1599 * on write, or reading the csums from the tree before a read
1601 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1602 int mirror_num, unsigned long bio_flags,
1605 struct btrfs_root *root = BTRFS_I(inode)->root;
1610 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1612 if (btrfs_is_free_space_inode(root, inode))
1615 if (!(rw & REQ_WRITE)) {
1616 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1620 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1621 return btrfs_submit_compressed_read(inode, bio,
1622 mirror_num, bio_flags);
1623 } else if (!skip_sum) {
1624 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1629 } else if (!skip_sum) {
1630 /* csum items have already been cloned */
1631 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1633 /* we're doing a write, do the async checksumming */
1634 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1635 inode, rw, bio, mirror_num,
1636 bio_flags, bio_offset,
1637 __btrfs_submit_bio_start,
1638 __btrfs_submit_bio_done);
1642 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1646 * given a list of ordered sums record them in the inode. This happens
1647 * at IO completion time based on sums calculated at bio submission time.
1649 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1650 struct inode *inode, u64 file_offset,
1651 struct list_head *list)
1653 struct btrfs_ordered_sum *sum;
1655 list_for_each_entry(sum, list, list) {
1656 btrfs_csum_file_blocks(trans,
1657 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1662 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1663 struct extent_state **cached_state)
1665 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1667 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1668 cached_state, GFP_NOFS);
1671 /* see btrfs_writepage_start_hook for details on why this is required */
1672 struct btrfs_writepage_fixup {
1674 struct btrfs_work work;
1677 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1679 struct btrfs_writepage_fixup *fixup;
1680 struct btrfs_ordered_extent *ordered;
1681 struct extent_state *cached_state = NULL;
1683 struct inode *inode;
1688 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1692 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1693 ClearPageChecked(page);
1697 inode = page->mapping->host;
1698 page_start = page_offset(page);
1699 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1701 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1704 /* already ordered? We're done */
1705 if (PagePrivate2(page))
1708 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1710 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1711 page_end, &cached_state, GFP_NOFS);
1713 btrfs_start_ordered_extent(inode, ordered, 1);
1714 btrfs_put_ordered_extent(ordered);
1718 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1720 mapping_set_error(page->mapping, ret);
1721 end_extent_writepage(page, ret, page_start, page_end);
1722 ClearPageChecked(page);
1726 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1727 ClearPageChecked(page);
1728 set_page_dirty(page);
1730 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1731 &cached_state, GFP_NOFS);
1734 page_cache_release(page);
1739 * There are a few paths in the higher layers of the kernel that directly
1740 * set the page dirty bit without asking the filesystem if it is a
1741 * good idea. This causes problems because we want to make sure COW
1742 * properly happens and the data=ordered rules are followed.
1744 * In our case any range that doesn't have the ORDERED bit set
1745 * hasn't been properly setup for IO. We kick off an async process
1746 * to fix it up. The async helper will wait for ordered extents, set
1747 * the delalloc bit and make it safe to write the page.
1749 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1751 struct inode *inode = page->mapping->host;
1752 struct btrfs_writepage_fixup *fixup;
1753 struct btrfs_root *root = BTRFS_I(inode)->root;
1755 /* this page is properly in the ordered list */
1756 if (TestClearPagePrivate2(page))
1759 if (PageChecked(page))
1762 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1766 SetPageChecked(page);
1767 page_cache_get(page);
1768 fixup->work.func = btrfs_writepage_fixup_worker;
1770 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1774 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1775 struct inode *inode, u64 file_pos,
1776 u64 disk_bytenr, u64 disk_num_bytes,
1777 u64 num_bytes, u64 ram_bytes,
1778 u8 compression, u8 encryption,
1779 u16 other_encoding, int extent_type)
1781 struct btrfs_root *root = BTRFS_I(inode)->root;
1782 struct btrfs_file_extent_item *fi;
1783 struct btrfs_path *path;
1784 struct extent_buffer *leaf;
1785 struct btrfs_key ins;
1789 path = btrfs_alloc_path();
1793 path->leave_spinning = 1;
1796 * we may be replacing one extent in the tree with another.
1797 * The new extent is pinned in the extent map, and we don't want
1798 * to drop it from the cache until it is completely in the btree.
1800 * So, tell btrfs_drop_extents to leave this extent in the cache.
1801 * the caller is expected to unpin it and allow it to be merged
1804 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1809 ins.objectid = btrfs_ino(inode);
1810 ins.offset = file_pos;
1811 ins.type = BTRFS_EXTENT_DATA_KEY;
1812 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1815 leaf = path->nodes[0];
1816 fi = btrfs_item_ptr(leaf, path->slots[0],
1817 struct btrfs_file_extent_item);
1818 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1819 btrfs_set_file_extent_type(leaf, fi, extent_type);
1820 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1821 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1822 btrfs_set_file_extent_offset(leaf, fi, 0);
1823 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1824 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1825 btrfs_set_file_extent_compression(leaf, fi, compression);
1826 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1827 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1829 btrfs_unlock_up_safe(path, 1);
1830 btrfs_set_lock_blocking(leaf);
1832 btrfs_mark_buffer_dirty(leaf);
1834 inode_add_bytes(inode, num_bytes);
1836 ins.objectid = disk_bytenr;
1837 ins.offset = disk_num_bytes;
1838 ins.type = BTRFS_EXTENT_ITEM_KEY;
1839 ret = btrfs_alloc_reserved_file_extent(trans, root,
1840 root->root_key.objectid,
1841 btrfs_ino(inode), file_pos, &ins);
1843 btrfs_free_path(path);
1849 * helper function for btrfs_finish_ordered_io, this
1850 * just reads in some of the csum leaves to prime them into ram
1851 * before we start the transaction. It limits the amount of btree
1852 * reads required while inside the transaction.
1854 /* as ordered data IO finishes, this gets called so we can finish
1855 * an ordered extent if the range of bytes in the file it covers are
1858 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1860 struct inode *inode = ordered_extent->inode;
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
1862 struct btrfs_trans_handle *trans = NULL;
1863 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1864 struct extent_state *cached_state = NULL;
1865 int compress_type = 0;
1869 nolock = btrfs_is_free_space_inode(root, inode);
1871 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1876 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1877 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1878 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1881 trans = btrfs_join_transaction_nolock(root);
1883 trans = btrfs_join_transaction(root);
1885 return PTR_ERR(trans);
1886 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1887 ret = btrfs_update_inode_fallback(trans, root, inode);
1888 if (ret) /* -ENOMEM or corruption */
1889 btrfs_abort_transaction(trans, root, ret);
1894 lock_extent_bits(io_tree, ordered_extent->file_offset,
1895 ordered_extent->file_offset + ordered_extent->len - 1,
1899 trans = btrfs_join_transaction_nolock(root);
1901 trans = btrfs_join_transaction(root);
1902 if (IS_ERR(trans)) {
1903 ret = PTR_ERR(trans);
1907 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1909 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1910 compress_type = ordered_extent->compress_type;
1911 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1912 BUG_ON(compress_type);
1913 ret = btrfs_mark_extent_written(trans, inode,
1914 ordered_extent->file_offset,
1915 ordered_extent->file_offset +
1916 ordered_extent->len);
1918 BUG_ON(root == root->fs_info->tree_root);
1919 ret = insert_reserved_file_extent(trans, inode,
1920 ordered_extent->file_offset,
1921 ordered_extent->start,
1922 ordered_extent->disk_len,
1923 ordered_extent->len,
1924 ordered_extent->len,
1925 compress_type, 0, 0,
1926 BTRFS_FILE_EXTENT_REG);
1927 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1928 ordered_extent->file_offset,
1929 ordered_extent->len);
1933 btrfs_abort_transaction(trans, root, ret);
1937 add_pending_csums(trans, inode, ordered_extent->file_offset,
1938 &ordered_extent->list);
1940 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1941 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1942 ret = btrfs_update_inode_fallback(trans, root, inode);
1943 if (ret) { /* -ENOMEM or corruption */
1944 btrfs_abort_transaction(trans, root, ret);
1950 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1951 ordered_extent->file_offset +
1952 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1954 if (root != root->fs_info->tree_root)
1955 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1958 btrfs_end_transaction_nolock(trans, root);
1960 btrfs_end_transaction(trans, root);
1964 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1965 ordered_extent->file_offset +
1966 ordered_extent->len - 1, NULL, GFP_NOFS);
1969 * This needs to be dont to make sure anybody waiting knows we are done
1970 * upating everything for this ordered extent.
1972 btrfs_remove_ordered_extent(inode, ordered_extent);
1975 btrfs_put_ordered_extent(ordered_extent);
1976 /* once for the tree */
1977 btrfs_put_ordered_extent(ordered_extent);
1982 static void finish_ordered_fn(struct btrfs_work *work)
1984 struct btrfs_ordered_extent *ordered_extent;
1985 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1986 btrfs_finish_ordered_io(ordered_extent);
1989 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1990 struct extent_state *state, int uptodate)
1992 struct inode *inode = page->mapping->host;
1993 struct btrfs_root *root = BTRFS_I(inode)->root;
1994 struct btrfs_ordered_extent *ordered_extent = NULL;
1995 struct btrfs_workers *workers;
1997 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1999 ClearPagePrivate2(page);
2000 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2001 end - start + 1, uptodate))
2004 ordered_extent->work.func = finish_ordered_fn;
2005 ordered_extent->work.flags = 0;
2007 if (btrfs_is_free_space_inode(root, inode))
2008 workers = &root->fs_info->endio_freespace_worker;
2010 workers = &root->fs_info->endio_write_workers;
2011 btrfs_queue_worker(workers, &ordered_extent->work);
2017 * when reads are done, we need to check csums to verify the data is correct
2018 * if there's a match, we allow the bio to finish. If not, the code in
2019 * extent_io.c will try to find good copies for us.
2021 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2022 struct extent_state *state, int mirror)
2024 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2025 struct inode *inode = page->mapping->host;
2026 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2028 u64 private = ~(u32)0;
2030 struct btrfs_root *root = BTRFS_I(inode)->root;
2033 if (PageChecked(page)) {
2034 ClearPageChecked(page);
2038 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2041 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2042 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2043 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2048 if (state && state->start == start) {
2049 private = state->private;
2052 ret = get_state_private(io_tree, start, &private);
2054 kaddr = kmap_atomic(page);
2058 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2059 btrfs_csum_final(csum, (char *)&csum);
2060 if (csum != private)
2063 kunmap_atomic(kaddr);
2068 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2070 (unsigned long long)btrfs_ino(page->mapping->host),
2071 (unsigned long long)start, csum,
2072 (unsigned long long)private);
2073 memset(kaddr + offset, 1, end - start + 1);
2074 flush_dcache_page(page);
2075 kunmap_atomic(kaddr);
2081 struct delayed_iput {
2082 struct list_head list;
2083 struct inode *inode;
2086 /* JDM: If this is fs-wide, why can't we add a pointer to
2087 * btrfs_inode instead and avoid the allocation? */
2088 void btrfs_add_delayed_iput(struct inode *inode)
2090 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2091 struct delayed_iput *delayed;
2093 if (atomic_add_unless(&inode->i_count, -1, 1))
2096 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2097 delayed->inode = inode;
2099 spin_lock(&fs_info->delayed_iput_lock);
2100 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2101 spin_unlock(&fs_info->delayed_iput_lock);
2104 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2107 struct btrfs_fs_info *fs_info = root->fs_info;
2108 struct delayed_iput *delayed;
2111 spin_lock(&fs_info->delayed_iput_lock);
2112 empty = list_empty(&fs_info->delayed_iputs);
2113 spin_unlock(&fs_info->delayed_iput_lock);
2117 down_read(&root->fs_info->cleanup_work_sem);
2118 spin_lock(&fs_info->delayed_iput_lock);
2119 list_splice_init(&fs_info->delayed_iputs, &list);
2120 spin_unlock(&fs_info->delayed_iput_lock);
2122 while (!list_empty(&list)) {
2123 delayed = list_entry(list.next, struct delayed_iput, list);
2124 list_del(&delayed->list);
2125 iput(delayed->inode);
2128 up_read(&root->fs_info->cleanup_work_sem);
2131 enum btrfs_orphan_cleanup_state {
2132 ORPHAN_CLEANUP_STARTED = 1,
2133 ORPHAN_CLEANUP_DONE = 2,
2137 * This is called in transaction commit time. If there are no orphan
2138 * files in the subvolume, it removes orphan item and frees block_rsv
2141 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2142 struct btrfs_root *root)
2144 struct btrfs_block_rsv *block_rsv;
2147 if (atomic_read(&root->orphan_inodes) ||
2148 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2151 spin_lock(&root->orphan_lock);
2152 if (atomic_read(&root->orphan_inodes)) {
2153 spin_unlock(&root->orphan_lock);
2157 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2158 spin_unlock(&root->orphan_lock);
2162 block_rsv = root->orphan_block_rsv;
2163 root->orphan_block_rsv = NULL;
2164 spin_unlock(&root->orphan_lock);
2166 if (root->orphan_item_inserted &&
2167 btrfs_root_refs(&root->root_item) > 0) {
2168 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2169 root->root_key.objectid);
2171 root->orphan_item_inserted = 0;
2175 WARN_ON(block_rsv->size > 0);
2176 btrfs_free_block_rsv(root, block_rsv);
2181 * This creates an orphan entry for the given inode in case something goes
2182 * wrong in the middle of an unlink/truncate.
2184 * NOTE: caller of this function should reserve 5 units of metadata for
2187 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2189 struct btrfs_root *root = BTRFS_I(inode)->root;
2190 struct btrfs_block_rsv *block_rsv = NULL;
2195 if (!root->orphan_block_rsv) {
2196 block_rsv = btrfs_alloc_block_rsv(root);
2201 spin_lock(&root->orphan_lock);
2202 if (!root->orphan_block_rsv) {
2203 root->orphan_block_rsv = block_rsv;
2204 } else if (block_rsv) {
2205 btrfs_free_block_rsv(root, block_rsv);
2209 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2210 &BTRFS_I(inode)->runtime_flags)) {
2213 * For proper ENOSPC handling, we should do orphan
2214 * cleanup when mounting. But this introduces backward
2215 * compatibility issue.
2217 if (!xchg(&root->orphan_item_inserted, 1))
2223 atomic_dec(&root->orphan_inodes);
2226 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2227 &BTRFS_I(inode)->runtime_flags))
2229 spin_unlock(&root->orphan_lock);
2231 /* grab metadata reservation from transaction handle */
2233 ret = btrfs_orphan_reserve_metadata(trans, inode);
2234 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2237 /* insert an orphan item to track this unlinked/truncated file */
2239 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2240 if (ret && ret != -EEXIST) {
2241 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2242 &BTRFS_I(inode)->runtime_flags);
2243 btrfs_abort_transaction(trans, root, ret);
2249 /* insert an orphan item to track subvolume contains orphan files */
2251 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2252 root->root_key.objectid);
2253 if (ret && ret != -EEXIST) {
2254 btrfs_abort_transaction(trans, root, ret);
2262 * We have done the truncate/delete so we can go ahead and remove the orphan
2263 * item for this particular inode.
2265 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2267 struct btrfs_root *root = BTRFS_I(inode)->root;
2268 int delete_item = 0;
2269 int release_rsv = 0;
2272 spin_lock(&root->orphan_lock);
2273 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2274 &BTRFS_I(inode)->runtime_flags))
2277 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2278 &BTRFS_I(inode)->runtime_flags))
2280 spin_unlock(&root->orphan_lock);
2282 if (trans && delete_item) {
2283 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2284 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2288 btrfs_orphan_release_metadata(inode);
2289 atomic_dec(&root->orphan_inodes);
2296 * this cleans up any orphans that may be left on the list from the last use
2299 int btrfs_orphan_cleanup(struct btrfs_root *root)
2301 struct btrfs_path *path;
2302 struct extent_buffer *leaf;
2303 struct btrfs_key key, found_key;
2304 struct btrfs_trans_handle *trans;
2305 struct inode *inode;
2306 u64 last_objectid = 0;
2307 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2309 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2312 path = btrfs_alloc_path();
2319 key.objectid = BTRFS_ORPHAN_OBJECTID;
2320 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2321 key.offset = (u64)-1;
2324 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2329 * if ret == 0 means we found what we were searching for, which
2330 * is weird, but possible, so only screw with path if we didn't
2331 * find the key and see if we have stuff that matches
2335 if (path->slots[0] == 0)
2340 /* pull out the item */
2341 leaf = path->nodes[0];
2342 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2344 /* make sure the item matches what we want */
2345 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2347 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2350 /* release the path since we're done with it */
2351 btrfs_release_path(path);
2354 * this is where we are basically btrfs_lookup, without the
2355 * crossing root thing. we store the inode number in the
2356 * offset of the orphan item.
2359 if (found_key.offset == last_objectid) {
2360 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2361 "stopping orphan cleanup\n");
2366 last_objectid = found_key.offset;
2368 found_key.objectid = found_key.offset;
2369 found_key.type = BTRFS_INODE_ITEM_KEY;
2370 found_key.offset = 0;
2371 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2372 ret = PTR_RET(inode);
2373 if (ret && ret != -ESTALE)
2376 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2377 struct btrfs_root *dead_root;
2378 struct btrfs_fs_info *fs_info = root->fs_info;
2379 int is_dead_root = 0;
2382 * this is an orphan in the tree root. Currently these
2383 * could come from 2 sources:
2384 * a) a snapshot deletion in progress
2385 * b) a free space cache inode
2386 * We need to distinguish those two, as the snapshot
2387 * orphan must not get deleted.
2388 * find_dead_roots already ran before us, so if this
2389 * is a snapshot deletion, we should find the root
2390 * in the dead_roots list
2392 spin_lock(&fs_info->trans_lock);
2393 list_for_each_entry(dead_root, &fs_info->dead_roots,
2395 if (dead_root->root_key.objectid ==
2396 found_key.objectid) {
2401 spin_unlock(&fs_info->trans_lock);
2403 /* prevent this orphan from being found again */
2404 key.offset = found_key.objectid - 1;
2409 * Inode is already gone but the orphan item is still there,
2410 * kill the orphan item.
2412 if (ret == -ESTALE) {
2413 trans = btrfs_start_transaction(root, 1);
2414 if (IS_ERR(trans)) {
2415 ret = PTR_ERR(trans);
2418 printk(KERN_ERR "auto deleting %Lu\n",
2419 found_key.objectid);
2420 ret = btrfs_del_orphan_item(trans, root,
2421 found_key.objectid);
2422 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2423 btrfs_end_transaction(trans, root);
2428 * add this inode to the orphan list so btrfs_orphan_del does
2429 * the proper thing when we hit it
2431 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2432 &BTRFS_I(inode)->runtime_flags);
2434 /* if we have links, this was a truncate, lets do that */
2435 if (inode->i_nlink) {
2436 if (!S_ISREG(inode->i_mode)) {
2442 ret = btrfs_truncate(inode);
2447 /* this will do delete_inode and everything for us */
2452 /* release the path since we're done with it */
2453 btrfs_release_path(path);
2455 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2457 if (root->orphan_block_rsv)
2458 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2461 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2462 trans = btrfs_join_transaction(root);
2464 btrfs_end_transaction(trans, root);
2468 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2470 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2474 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2475 btrfs_free_path(path);
2480 * very simple check to peek ahead in the leaf looking for xattrs. If we
2481 * don't find any xattrs, we know there can't be any acls.
2483 * slot is the slot the inode is in, objectid is the objectid of the inode
2485 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2486 int slot, u64 objectid)
2488 u32 nritems = btrfs_header_nritems(leaf);
2489 struct btrfs_key found_key;
2493 while (slot < nritems) {
2494 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2496 /* we found a different objectid, there must not be acls */
2497 if (found_key.objectid != objectid)
2500 /* we found an xattr, assume we've got an acl */
2501 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2505 * we found a key greater than an xattr key, there can't
2506 * be any acls later on
2508 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2515 * it goes inode, inode backrefs, xattrs, extents,
2516 * so if there are a ton of hard links to an inode there can
2517 * be a lot of backrefs. Don't waste time searching too hard,
2518 * this is just an optimization
2523 /* we hit the end of the leaf before we found an xattr or
2524 * something larger than an xattr. We have to assume the inode
2531 * read an inode from the btree into the in-memory inode
2533 static void btrfs_read_locked_inode(struct inode *inode)
2535 struct btrfs_path *path;
2536 struct extent_buffer *leaf;
2537 struct btrfs_inode_item *inode_item;
2538 struct btrfs_timespec *tspec;
2539 struct btrfs_root *root = BTRFS_I(inode)->root;
2540 struct btrfs_key location;
2544 bool filled = false;
2546 ret = btrfs_fill_inode(inode, &rdev);
2550 path = btrfs_alloc_path();
2554 path->leave_spinning = 1;
2555 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2557 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2561 leaf = path->nodes[0];
2566 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2567 struct btrfs_inode_item);
2568 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2569 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2570 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2571 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2572 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2574 tspec = btrfs_inode_atime(inode_item);
2575 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2576 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2578 tspec = btrfs_inode_mtime(inode_item);
2579 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2580 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2582 tspec = btrfs_inode_ctime(inode_item);
2583 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2584 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2586 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2587 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2588 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2589 inode->i_generation = BTRFS_I(inode)->generation;
2591 rdev = btrfs_inode_rdev(leaf, inode_item);
2593 BTRFS_I(inode)->index_cnt = (u64)-1;
2594 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2597 * try to precache a NULL acl entry for files that don't have
2598 * any xattrs or acls
2600 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2603 cache_no_acl(inode);
2605 btrfs_free_path(path);
2607 switch (inode->i_mode & S_IFMT) {
2609 inode->i_mapping->a_ops = &btrfs_aops;
2610 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2611 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2612 inode->i_fop = &btrfs_file_operations;
2613 inode->i_op = &btrfs_file_inode_operations;
2616 inode->i_fop = &btrfs_dir_file_operations;
2617 if (root == root->fs_info->tree_root)
2618 inode->i_op = &btrfs_dir_ro_inode_operations;
2620 inode->i_op = &btrfs_dir_inode_operations;
2623 inode->i_op = &btrfs_symlink_inode_operations;
2624 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2625 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2628 inode->i_op = &btrfs_special_inode_operations;
2629 init_special_inode(inode, inode->i_mode, rdev);
2633 btrfs_update_iflags(inode);
2637 btrfs_free_path(path);
2638 make_bad_inode(inode);
2642 * given a leaf and an inode, copy the inode fields into the leaf
2644 static void fill_inode_item(struct btrfs_trans_handle *trans,
2645 struct extent_buffer *leaf,
2646 struct btrfs_inode_item *item,
2647 struct inode *inode)
2649 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2650 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2651 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2652 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2653 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2655 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2656 inode->i_atime.tv_sec);
2657 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2658 inode->i_atime.tv_nsec);
2660 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2661 inode->i_mtime.tv_sec);
2662 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2663 inode->i_mtime.tv_nsec);
2665 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2666 inode->i_ctime.tv_sec);
2667 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2668 inode->i_ctime.tv_nsec);
2670 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2671 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2672 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2673 btrfs_set_inode_transid(leaf, item, trans->transid);
2674 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2675 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2676 btrfs_set_inode_block_group(leaf, item, 0);
2680 * copy everything in the in-memory inode into the btree.
2682 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root, struct inode *inode)
2685 struct btrfs_inode_item *inode_item;
2686 struct btrfs_path *path;
2687 struct extent_buffer *leaf;
2690 path = btrfs_alloc_path();
2694 path->leave_spinning = 1;
2695 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2703 btrfs_unlock_up_safe(path, 1);
2704 leaf = path->nodes[0];
2705 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2706 struct btrfs_inode_item);
2708 fill_inode_item(trans, leaf, inode_item, inode);
2709 btrfs_mark_buffer_dirty(leaf);
2710 btrfs_set_inode_last_trans(trans, inode);
2713 btrfs_free_path(path);
2718 * copy everything in the in-memory inode into the btree.
2720 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2721 struct btrfs_root *root, struct inode *inode)
2726 * If the inode is a free space inode, we can deadlock during commit
2727 * if we put it into the delayed code.
2729 * The data relocation inode should also be directly updated
2732 if (!btrfs_is_free_space_inode(root, inode)
2733 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2734 ret = btrfs_delayed_update_inode(trans, root, inode);
2736 btrfs_set_inode_last_trans(trans, inode);
2740 return btrfs_update_inode_item(trans, root, inode);
2743 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2744 struct btrfs_root *root, struct inode *inode)
2748 ret = btrfs_update_inode(trans, root, inode);
2750 return btrfs_update_inode_item(trans, root, inode);
2755 * unlink helper that gets used here in inode.c and in the tree logging
2756 * recovery code. It remove a link in a directory with a given name, and
2757 * also drops the back refs in the inode to the directory
2759 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2760 struct btrfs_root *root,
2761 struct inode *dir, struct inode *inode,
2762 const char *name, int name_len)
2764 struct btrfs_path *path;
2766 struct extent_buffer *leaf;
2767 struct btrfs_dir_item *di;
2768 struct btrfs_key key;
2770 u64 ino = btrfs_ino(inode);
2771 u64 dir_ino = btrfs_ino(dir);
2773 path = btrfs_alloc_path();
2779 path->leave_spinning = 1;
2780 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2781 name, name_len, -1);
2790 leaf = path->nodes[0];
2791 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2792 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2795 btrfs_release_path(path);
2797 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2800 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2801 "inode %llu parent %llu\n", name_len, name,
2802 (unsigned long long)ino, (unsigned long long)dir_ino);
2803 btrfs_abort_transaction(trans, root, ret);
2807 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2809 btrfs_abort_transaction(trans, root, ret);
2813 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2815 if (ret != 0 && ret != -ENOENT) {
2816 btrfs_abort_transaction(trans, root, ret);
2820 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2825 btrfs_free_path(path);
2829 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2830 inode_inc_iversion(inode);
2831 inode_inc_iversion(dir);
2832 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2833 btrfs_update_inode(trans, root, dir);
2838 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2839 struct btrfs_root *root,
2840 struct inode *dir, struct inode *inode,
2841 const char *name, int name_len)
2844 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2846 btrfs_drop_nlink(inode);
2847 ret = btrfs_update_inode(trans, root, inode);
2853 /* helper to check if there is any shared block in the path */
2854 static int check_path_shared(struct btrfs_root *root,
2855 struct btrfs_path *path)
2857 struct extent_buffer *eb;
2861 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2864 if (!path->nodes[level])
2866 eb = path->nodes[level];
2867 if (!btrfs_block_can_be_shared(root, eb))
2869 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2878 * helper to start transaction for unlink and rmdir.
2880 * unlink and rmdir are special in btrfs, they do not always free space.
2881 * so in enospc case, we should make sure they will free space before
2882 * allowing them to use the global metadata reservation.
2884 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2885 struct dentry *dentry)
2887 struct btrfs_trans_handle *trans;
2888 struct btrfs_root *root = BTRFS_I(dir)->root;
2889 struct btrfs_path *path;
2890 struct btrfs_inode_ref *ref;
2891 struct btrfs_dir_item *di;
2892 struct inode *inode = dentry->d_inode;
2897 u64 ino = btrfs_ino(inode);
2898 u64 dir_ino = btrfs_ino(dir);
2901 * 1 for the possible orphan item
2902 * 1 for the dir item
2903 * 1 for the dir index
2904 * 1 for the inode ref
2905 * 1 for the inode ref in the tree log
2906 * 2 for the dir entries in the log
2909 trans = btrfs_start_transaction(root, 8);
2910 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2913 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2914 return ERR_PTR(-ENOSPC);
2916 /* check if there is someone else holds reference */
2917 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2918 return ERR_PTR(-ENOSPC);
2920 if (atomic_read(&inode->i_count) > 2)
2921 return ERR_PTR(-ENOSPC);
2923 if (xchg(&root->fs_info->enospc_unlink, 1))
2924 return ERR_PTR(-ENOSPC);
2926 path = btrfs_alloc_path();
2928 root->fs_info->enospc_unlink = 0;
2929 return ERR_PTR(-ENOMEM);
2932 /* 1 for the orphan item */
2933 trans = btrfs_start_transaction(root, 1);
2934 if (IS_ERR(trans)) {
2935 btrfs_free_path(path);
2936 root->fs_info->enospc_unlink = 0;
2940 path->skip_locking = 1;
2941 path->search_commit_root = 1;
2943 ret = btrfs_lookup_inode(trans, root, path,
2944 &BTRFS_I(dir)->location, 0);
2950 if (check_path_shared(root, path))
2955 btrfs_release_path(path);
2957 ret = btrfs_lookup_inode(trans, root, path,
2958 &BTRFS_I(inode)->location, 0);
2964 if (check_path_shared(root, path))
2969 btrfs_release_path(path);
2971 if (ret == 0 && S_ISREG(inode->i_mode)) {
2972 ret = btrfs_lookup_file_extent(trans, root, path,
2978 BUG_ON(ret == 0); /* Corruption */
2979 if (check_path_shared(root, path))
2981 btrfs_release_path(path);
2989 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2990 dentry->d_name.name, dentry->d_name.len, 0);
2996 if (check_path_shared(root, path))
3002 btrfs_release_path(path);
3004 ref = btrfs_lookup_inode_ref(trans, root, path,
3005 dentry->d_name.name, dentry->d_name.len,
3011 BUG_ON(!ref); /* Logic error */
3012 if (check_path_shared(root, path))
3014 index = btrfs_inode_ref_index(path->nodes[0], ref);
3015 btrfs_release_path(path);
3018 * This is a commit root search, if we can lookup inode item and other
3019 * relative items in the commit root, it means the transaction of
3020 * dir/file creation has been committed, and the dir index item that we
3021 * delay to insert has also been inserted into the commit root. So
3022 * we needn't worry about the delayed insertion of the dir index item
3025 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3026 dentry->d_name.name, dentry->d_name.len, 0);
3031 BUG_ON(ret == -ENOENT);
3032 if (check_path_shared(root, path))
3037 btrfs_free_path(path);
3038 /* Migrate the orphan reservation over */
3040 err = btrfs_block_rsv_migrate(trans->block_rsv,
3041 &root->fs_info->global_block_rsv,
3042 trans->bytes_reserved);
3045 btrfs_end_transaction(trans, root);
3046 root->fs_info->enospc_unlink = 0;
3047 return ERR_PTR(err);
3050 trans->block_rsv = &root->fs_info->global_block_rsv;
3054 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3055 struct btrfs_root *root)
3057 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3058 btrfs_block_rsv_release(root, trans->block_rsv,
3059 trans->bytes_reserved);
3060 trans->block_rsv = &root->fs_info->trans_block_rsv;
3061 BUG_ON(!root->fs_info->enospc_unlink);
3062 root->fs_info->enospc_unlink = 0;
3064 btrfs_end_transaction(trans, root);
3067 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3069 struct btrfs_root *root = BTRFS_I(dir)->root;
3070 struct btrfs_trans_handle *trans;
3071 struct inode *inode = dentry->d_inode;
3073 unsigned long nr = 0;
3075 trans = __unlink_start_trans(dir, dentry);
3077 return PTR_ERR(trans);
3079 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3081 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3082 dentry->d_name.name, dentry->d_name.len);
3086 if (inode->i_nlink == 0) {
3087 ret = btrfs_orphan_add(trans, inode);
3093 nr = trans->blocks_used;
3094 __unlink_end_trans(trans, root);
3095 btrfs_btree_balance_dirty(root, nr);
3099 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3100 struct btrfs_root *root,
3101 struct inode *dir, u64 objectid,
3102 const char *name, int name_len)
3104 struct btrfs_path *path;
3105 struct extent_buffer *leaf;
3106 struct btrfs_dir_item *di;
3107 struct btrfs_key key;
3110 u64 dir_ino = btrfs_ino(dir);
3112 path = btrfs_alloc_path();
3116 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3117 name, name_len, -1);
3118 if (IS_ERR_OR_NULL(di)) {
3126 leaf = path->nodes[0];
3127 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3128 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3129 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3131 btrfs_abort_transaction(trans, root, ret);
3134 btrfs_release_path(path);
3136 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3137 objectid, root->root_key.objectid,
3138 dir_ino, &index, name, name_len);
3140 if (ret != -ENOENT) {
3141 btrfs_abort_transaction(trans, root, ret);
3144 di = btrfs_search_dir_index_item(root, path, dir_ino,
3146 if (IS_ERR_OR_NULL(di)) {
3151 btrfs_abort_transaction(trans, root, ret);
3155 leaf = path->nodes[0];
3156 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3157 btrfs_release_path(path);
3160 btrfs_release_path(path);
3162 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3164 btrfs_abort_transaction(trans, root, ret);
3168 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3169 inode_inc_iversion(dir);
3170 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3171 ret = btrfs_update_inode(trans, root, dir);
3173 btrfs_abort_transaction(trans, root, ret);
3175 btrfs_free_path(path);
3179 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3181 struct inode *inode = dentry->d_inode;
3183 struct btrfs_root *root = BTRFS_I(dir)->root;
3184 struct btrfs_trans_handle *trans;
3185 unsigned long nr = 0;
3187 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3188 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3191 trans = __unlink_start_trans(dir, dentry);
3193 return PTR_ERR(trans);
3195 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3196 err = btrfs_unlink_subvol(trans, root, dir,
3197 BTRFS_I(inode)->location.objectid,
3198 dentry->d_name.name,
3199 dentry->d_name.len);
3203 err = btrfs_orphan_add(trans, inode);
3207 /* now the directory is empty */
3208 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3209 dentry->d_name.name, dentry->d_name.len);
3211 btrfs_i_size_write(inode, 0);
3213 nr = trans->blocks_used;
3214 __unlink_end_trans(trans, root);
3215 btrfs_btree_balance_dirty(root, nr);
3221 * this can truncate away extent items, csum items and directory items.
3222 * It starts at a high offset and removes keys until it can't find
3223 * any higher than new_size
3225 * csum items that cross the new i_size are truncated to the new size
3228 * min_type is the minimum key type to truncate down to. If set to 0, this
3229 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3231 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3232 struct btrfs_root *root,
3233 struct inode *inode,
3234 u64 new_size, u32 min_type)
3236 struct btrfs_path *path;
3237 struct extent_buffer *leaf;
3238 struct btrfs_file_extent_item *fi;
3239 struct btrfs_key key;
3240 struct btrfs_key found_key;
3241 u64 extent_start = 0;
3242 u64 extent_num_bytes = 0;
3243 u64 extent_offset = 0;
3245 u64 mask = root->sectorsize - 1;
3246 u32 found_type = (u8)-1;
3249 int pending_del_nr = 0;
3250 int pending_del_slot = 0;
3251 int extent_type = -1;
3254 u64 ino = btrfs_ino(inode);
3256 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3258 path = btrfs_alloc_path();
3263 if (root->ref_cows || root == root->fs_info->tree_root)
3264 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3267 * This function is also used to drop the items in the log tree before
3268 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3269 * it is used to drop the loged items. So we shouldn't kill the delayed
3272 if (min_type == 0 && root == BTRFS_I(inode)->root)
3273 btrfs_kill_delayed_inode_items(inode);
3276 key.offset = (u64)-1;
3280 path->leave_spinning = 1;
3281 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3288 /* there are no items in the tree for us to truncate, we're
3291 if (path->slots[0] == 0)
3298 leaf = path->nodes[0];
3299 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3300 found_type = btrfs_key_type(&found_key);
3302 if (found_key.objectid != ino)
3305 if (found_type < min_type)
3308 item_end = found_key.offset;
3309 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3310 fi = btrfs_item_ptr(leaf, path->slots[0],
3311 struct btrfs_file_extent_item);
3312 extent_type = btrfs_file_extent_type(leaf, fi);
3313 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3315 btrfs_file_extent_num_bytes(leaf, fi);
3316 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3317 item_end += btrfs_file_extent_inline_len(leaf,
3322 if (found_type > min_type) {
3325 if (item_end < new_size)
3327 if (found_key.offset >= new_size)
3333 /* FIXME, shrink the extent if the ref count is only 1 */
3334 if (found_type != BTRFS_EXTENT_DATA_KEY)
3337 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3339 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3341 u64 orig_num_bytes =
3342 btrfs_file_extent_num_bytes(leaf, fi);
3343 extent_num_bytes = new_size -
3344 found_key.offset + root->sectorsize - 1;
3345 extent_num_bytes = extent_num_bytes &
3346 ~((u64)root->sectorsize - 1);
3347 btrfs_set_file_extent_num_bytes(leaf, fi,
3349 num_dec = (orig_num_bytes -
3351 if (root->ref_cows && extent_start != 0)
3352 inode_sub_bytes(inode, num_dec);
3353 btrfs_mark_buffer_dirty(leaf);
3356 btrfs_file_extent_disk_num_bytes(leaf,
3358 extent_offset = found_key.offset -
3359 btrfs_file_extent_offset(leaf, fi);
3361 /* FIXME blocksize != 4096 */
3362 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3363 if (extent_start != 0) {
3366 inode_sub_bytes(inode, num_dec);
3369 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3371 * we can't truncate inline items that have had
3375 btrfs_file_extent_compression(leaf, fi) == 0 &&
3376 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3377 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3378 u32 size = new_size - found_key.offset;
3380 if (root->ref_cows) {
3381 inode_sub_bytes(inode, item_end + 1 -
3385 btrfs_file_extent_calc_inline_size(size);
3386 btrfs_truncate_item(trans, root, path,
3388 } else if (root->ref_cows) {
3389 inode_sub_bytes(inode, item_end + 1 -
3395 if (!pending_del_nr) {
3396 /* no pending yet, add ourselves */
3397 pending_del_slot = path->slots[0];
3399 } else if (pending_del_nr &&
3400 path->slots[0] + 1 == pending_del_slot) {
3401 /* hop on the pending chunk */
3403 pending_del_slot = path->slots[0];
3410 if (found_extent && (root->ref_cows ||
3411 root == root->fs_info->tree_root)) {
3412 btrfs_set_path_blocking(path);
3413 ret = btrfs_free_extent(trans, root, extent_start,
3414 extent_num_bytes, 0,
3415 btrfs_header_owner(leaf),
3416 ino, extent_offset, 0);
3420 if (found_type == BTRFS_INODE_ITEM_KEY)
3423 if (path->slots[0] == 0 ||
3424 path->slots[0] != pending_del_slot) {
3425 if (root->ref_cows &&
3426 BTRFS_I(inode)->location.objectid !=
3427 BTRFS_FREE_INO_OBJECTID) {
3431 if (pending_del_nr) {
3432 ret = btrfs_del_items(trans, root, path,
3436 btrfs_abort_transaction(trans,
3442 btrfs_release_path(path);
3449 if (pending_del_nr) {
3450 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3453 btrfs_abort_transaction(trans, root, ret);
3456 btrfs_free_path(path);
3461 * taken from block_truncate_page, but does cow as it zeros out
3462 * any bytes left in the last page in the file.
3464 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3466 struct inode *inode = mapping->host;
3467 struct btrfs_root *root = BTRFS_I(inode)->root;
3468 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3469 struct btrfs_ordered_extent *ordered;
3470 struct extent_state *cached_state = NULL;
3472 u32 blocksize = root->sectorsize;
3473 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3474 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3476 gfp_t mask = btrfs_alloc_write_mask(mapping);
3481 if ((offset & (blocksize - 1)) == 0)
3483 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3489 page = find_or_create_page(mapping, index, mask);
3491 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3495 page_start = page_offset(page);
3496 page_end = page_start + PAGE_CACHE_SIZE - 1;
3498 if (!PageUptodate(page)) {
3499 ret = btrfs_readpage(NULL, page);
3501 if (page->mapping != mapping) {
3503 page_cache_release(page);
3506 if (!PageUptodate(page)) {
3511 wait_on_page_writeback(page);
3513 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3514 set_page_extent_mapped(page);
3516 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3518 unlock_extent_cached(io_tree, page_start, page_end,
3519 &cached_state, GFP_NOFS);
3521 page_cache_release(page);
3522 btrfs_start_ordered_extent(inode, ordered, 1);
3523 btrfs_put_ordered_extent(ordered);
3527 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3528 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3529 0, 0, &cached_state, GFP_NOFS);
3531 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3534 unlock_extent_cached(io_tree, page_start, page_end,
3535 &cached_state, GFP_NOFS);
3540 if (offset != PAGE_CACHE_SIZE) {
3542 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3543 flush_dcache_page(page);
3546 ClearPageChecked(page);
3547 set_page_dirty(page);
3548 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3553 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3555 page_cache_release(page);
3561 * This function puts in dummy file extents for the area we're creating a hole
3562 * for. So if we are truncating this file to a larger size we need to insert
3563 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3564 * the range between oldsize and size
3566 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3568 struct btrfs_trans_handle *trans;
3569 struct btrfs_root *root = BTRFS_I(inode)->root;
3570 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3571 struct extent_map *em = NULL;
3572 struct extent_state *cached_state = NULL;
3573 u64 mask = root->sectorsize - 1;
3574 u64 hole_start = (oldsize + mask) & ~mask;
3575 u64 block_end = (size + mask) & ~mask;
3581 if (size <= hole_start)
3585 struct btrfs_ordered_extent *ordered;
3586 btrfs_wait_ordered_range(inode, hole_start,
3587 block_end - hole_start);
3588 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3590 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3593 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3594 &cached_state, GFP_NOFS);
3595 btrfs_put_ordered_extent(ordered);
3598 cur_offset = hole_start;
3600 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3601 block_end - cur_offset, 0);
3606 last_byte = min(extent_map_end(em), block_end);
3607 last_byte = (last_byte + mask) & ~mask;
3608 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3610 hole_size = last_byte - cur_offset;
3612 trans = btrfs_start_transaction(root, 3);
3613 if (IS_ERR(trans)) {
3614 err = PTR_ERR(trans);
3618 err = btrfs_drop_extents(trans, inode, cur_offset,
3619 cur_offset + hole_size,
3622 btrfs_abort_transaction(trans, root, err);
3623 btrfs_end_transaction(trans, root);
3627 err = btrfs_insert_file_extent(trans, root,
3628 btrfs_ino(inode), cur_offset, 0,
3629 0, hole_size, 0, hole_size,
3632 btrfs_abort_transaction(trans, root, err);
3633 btrfs_end_transaction(trans, root);
3637 btrfs_drop_extent_cache(inode, hole_start,
3640 btrfs_update_inode(trans, root, inode);
3641 btrfs_end_transaction(trans, root);
3643 free_extent_map(em);
3645 cur_offset = last_byte;
3646 if (cur_offset >= block_end)
3650 free_extent_map(em);
3651 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3656 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3658 struct btrfs_root *root = BTRFS_I(inode)->root;
3659 struct btrfs_trans_handle *trans;
3660 loff_t oldsize = i_size_read(inode);
3663 if (newsize == oldsize)
3666 if (newsize > oldsize) {
3667 truncate_pagecache(inode, oldsize, newsize);
3668 ret = btrfs_cont_expand(inode, oldsize, newsize);
3672 trans = btrfs_start_transaction(root, 1);
3674 return PTR_ERR(trans);
3676 i_size_write(inode, newsize);
3677 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3678 ret = btrfs_update_inode(trans, root, inode);
3679 btrfs_end_transaction(trans, root);
3683 * We're truncating a file that used to have good data down to
3684 * zero. Make sure it gets into the ordered flush list so that
3685 * any new writes get down to disk quickly.
3688 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3689 &BTRFS_I(inode)->runtime_flags);
3691 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3692 truncate_setsize(inode, newsize);
3693 ret = btrfs_truncate(inode);
3699 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3701 struct inode *inode = dentry->d_inode;
3702 struct btrfs_root *root = BTRFS_I(inode)->root;
3705 if (btrfs_root_readonly(root))
3708 err = inode_change_ok(inode, attr);
3712 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3713 err = btrfs_setsize(inode, attr->ia_size);
3718 if (attr->ia_valid) {
3719 setattr_copy(inode, attr);
3720 inode_inc_iversion(inode);
3721 err = btrfs_dirty_inode(inode);
3723 if (!err && attr->ia_valid & ATTR_MODE)
3724 err = btrfs_acl_chmod(inode);
3730 void btrfs_evict_inode(struct inode *inode)
3732 struct btrfs_trans_handle *trans;
3733 struct btrfs_root *root = BTRFS_I(inode)->root;
3734 struct btrfs_block_rsv *rsv, *global_rsv;
3735 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3739 trace_btrfs_inode_evict(inode);
3741 truncate_inode_pages(&inode->i_data, 0);
3742 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3743 btrfs_is_free_space_inode(root, inode)))
3746 if (is_bad_inode(inode)) {
3747 btrfs_orphan_del(NULL, inode);
3750 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3751 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3753 if (root->fs_info->log_root_recovering) {
3754 BUG_ON(!test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3755 &BTRFS_I(inode)->runtime_flags));
3759 if (inode->i_nlink > 0) {
3760 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3764 rsv = btrfs_alloc_block_rsv(root);
3766 btrfs_orphan_del(NULL, inode);
3769 rsv->size = min_size;
3770 global_rsv = &root->fs_info->global_block_rsv;
3772 btrfs_i_size_write(inode, 0);
3775 * This is a bit simpler than btrfs_truncate since
3777 * 1) We've already reserved our space for our orphan item in the
3779 * 2) We're going to delete the inode item, so we don't need to update
3782 * So we just need to reserve some slack space in case we add bytes when
3783 * doing the truncate.
3786 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3789 * Try and steal from the global reserve since we will
3790 * likely not use this space anyway, we want to try as
3791 * hard as possible to get this to work.
3794 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3797 printk(KERN_WARNING "Could not get space for a "
3798 "delete, will truncate on mount %d\n", ret);
3799 btrfs_orphan_del(NULL, inode);
3800 btrfs_free_block_rsv(root, rsv);
3804 trans = btrfs_start_transaction(root, 0);
3805 if (IS_ERR(trans)) {
3806 btrfs_orphan_del(NULL, inode);
3807 btrfs_free_block_rsv(root, rsv);
3811 trans->block_rsv = rsv;
3813 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3817 nr = trans->blocks_used;
3818 btrfs_end_transaction(trans, root);
3820 btrfs_btree_balance_dirty(root, nr);
3823 btrfs_free_block_rsv(root, rsv);
3826 trans->block_rsv = root->orphan_block_rsv;
3827 ret = btrfs_orphan_del(trans, inode);
3831 trans->block_rsv = &root->fs_info->trans_block_rsv;
3832 if (!(root == root->fs_info->tree_root ||
3833 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3834 btrfs_return_ino(root, btrfs_ino(inode));
3836 nr = trans->blocks_used;
3837 btrfs_end_transaction(trans, root);
3838 btrfs_btree_balance_dirty(root, nr);
3840 end_writeback(inode);
3845 * this returns the key found in the dir entry in the location pointer.
3846 * If no dir entries were found, location->objectid is 0.
3848 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3849 struct btrfs_key *location)
3851 const char *name = dentry->d_name.name;
3852 int namelen = dentry->d_name.len;
3853 struct btrfs_dir_item *di;
3854 struct btrfs_path *path;
3855 struct btrfs_root *root = BTRFS_I(dir)->root;
3858 path = btrfs_alloc_path();
3862 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3867 if (IS_ERR_OR_NULL(di))
3870 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3872 btrfs_free_path(path);
3875 location->objectid = 0;
3880 * when we hit a tree root in a directory, the btrfs part of the inode
3881 * needs to be changed to reflect the root directory of the tree root. This
3882 * is kind of like crossing a mount point.
3884 static int fixup_tree_root_location(struct btrfs_root *root,
3886 struct dentry *dentry,
3887 struct btrfs_key *location,
3888 struct btrfs_root **sub_root)
3890 struct btrfs_path *path;
3891 struct btrfs_root *new_root;
3892 struct btrfs_root_ref *ref;
3893 struct extent_buffer *leaf;
3897 path = btrfs_alloc_path();
3904 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3905 BTRFS_I(dir)->root->root_key.objectid,
3906 location->objectid);
3913 leaf = path->nodes[0];
3914 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3915 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3916 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3919 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3920 (unsigned long)(ref + 1),
3921 dentry->d_name.len);
3925 btrfs_release_path(path);
3927 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3928 if (IS_ERR(new_root)) {
3929 err = PTR_ERR(new_root);
3933 if (btrfs_root_refs(&new_root->root_item) == 0) {
3938 *sub_root = new_root;
3939 location->objectid = btrfs_root_dirid(&new_root->root_item);
3940 location->type = BTRFS_INODE_ITEM_KEY;
3941 location->offset = 0;
3944 btrfs_free_path(path);
3948 static void inode_tree_add(struct inode *inode)
3950 struct btrfs_root *root = BTRFS_I(inode)->root;
3951 struct btrfs_inode *entry;
3953 struct rb_node *parent;
3954 u64 ino = btrfs_ino(inode);
3956 p = &root->inode_tree.rb_node;
3959 if (inode_unhashed(inode))
3962 spin_lock(&root->inode_lock);
3965 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3967 if (ino < btrfs_ino(&entry->vfs_inode))
3968 p = &parent->rb_left;
3969 else if (ino > btrfs_ino(&entry->vfs_inode))
3970 p = &parent->rb_right;
3972 WARN_ON(!(entry->vfs_inode.i_state &
3973 (I_WILL_FREE | I_FREEING)));
3974 rb_erase(parent, &root->inode_tree);
3975 RB_CLEAR_NODE(parent);
3976 spin_unlock(&root->inode_lock);
3980 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3981 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3982 spin_unlock(&root->inode_lock);
3985 static void inode_tree_del(struct inode *inode)
3987 struct btrfs_root *root = BTRFS_I(inode)->root;
3990 spin_lock(&root->inode_lock);
3991 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3992 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3993 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3994 empty = RB_EMPTY_ROOT(&root->inode_tree);
3996 spin_unlock(&root->inode_lock);
3999 * Free space cache has inodes in the tree root, but the tree root has a
4000 * root_refs of 0, so this could end up dropping the tree root as a
4001 * snapshot, so we need the extra !root->fs_info->tree_root check to
4002 * make sure we don't drop it.
4004 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4005 root != root->fs_info->tree_root) {
4006 synchronize_srcu(&root->fs_info->subvol_srcu);
4007 spin_lock(&root->inode_lock);
4008 empty = RB_EMPTY_ROOT(&root->inode_tree);
4009 spin_unlock(&root->inode_lock);
4011 btrfs_add_dead_root(root);
4015 void btrfs_invalidate_inodes(struct btrfs_root *root)
4017 struct rb_node *node;
4018 struct rb_node *prev;
4019 struct btrfs_inode *entry;
4020 struct inode *inode;
4023 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4025 spin_lock(&root->inode_lock);
4027 node = root->inode_tree.rb_node;
4031 entry = rb_entry(node, struct btrfs_inode, rb_node);
4033 if (objectid < btrfs_ino(&entry->vfs_inode))
4034 node = node->rb_left;
4035 else if (objectid > btrfs_ino(&entry->vfs_inode))
4036 node = node->rb_right;
4042 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4043 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4047 prev = rb_next(prev);
4051 entry = rb_entry(node, struct btrfs_inode, rb_node);
4052 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4053 inode = igrab(&entry->vfs_inode);
4055 spin_unlock(&root->inode_lock);
4056 if (atomic_read(&inode->i_count) > 1)
4057 d_prune_aliases(inode);
4059 * btrfs_drop_inode will have it removed from
4060 * the inode cache when its usage count
4065 spin_lock(&root->inode_lock);
4069 if (cond_resched_lock(&root->inode_lock))
4072 node = rb_next(node);
4074 spin_unlock(&root->inode_lock);
4077 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4079 struct btrfs_iget_args *args = p;
4080 inode->i_ino = args->ino;
4081 BTRFS_I(inode)->root = args->root;
4082 btrfs_set_inode_space_info(args->root, inode);
4086 static int btrfs_find_actor(struct inode *inode, void *opaque)
4088 struct btrfs_iget_args *args = opaque;
4089 return args->ino == btrfs_ino(inode) &&
4090 args->root == BTRFS_I(inode)->root;
4093 static struct inode *btrfs_iget_locked(struct super_block *s,
4095 struct btrfs_root *root)
4097 struct inode *inode;
4098 struct btrfs_iget_args args;
4099 args.ino = objectid;
4102 inode = iget5_locked(s, objectid, btrfs_find_actor,
4103 btrfs_init_locked_inode,
4108 /* Get an inode object given its location and corresponding root.
4109 * Returns in *is_new if the inode was read from disk
4111 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4112 struct btrfs_root *root, int *new)
4114 struct inode *inode;
4116 inode = btrfs_iget_locked(s, location->objectid, root);
4118 return ERR_PTR(-ENOMEM);
4120 if (inode->i_state & I_NEW) {
4121 BTRFS_I(inode)->root = root;
4122 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4123 btrfs_read_locked_inode(inode);
4124 if (!is_bad_inode(inode)) {
4125 inode_tree_add(inode);
4126 unlock_new_inode(inode);
4130 unlock_new_inode(inode);
4132 inode = ERR_PTR(-ESTALE);
4139 static struct inode *new_simple_dir(struct super_block *s,
4140 struct btrfs_key *key,
4141 struct btrfs_root *root)
4143 struct inode *inode = new_inode(s);
4146 return ERR_PTR(-ENOMEM);
4148 BTRFS_I(inode)->root = root;
4149 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4150 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4152 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4153 inode->i_op = &btrfs_dir_ro_inode_operations;
4154 inode->i_fop = &simple_dir_operations;
4155 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4156 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4161 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4163 struct inode *inode;
4164 struct btrfs_root *root = BTRFS_I(dir)->root;
4165 struct btrfs_root *sub_root = root;
4166 struct btrfs_key location;
4170 if (dentry->d_name.len > BTRFS_NAME_LEN)
4171 return ERR_PTR(-ENAMETOOLONG);
4173 if (unlikely(d_need_lookup(dentry))) {
4174 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4175 kfree(dentry->d_fsdata);
4176 dentry->d_fsdata = NULL;
4177 /* This thing is hashed, drop it for now */
4180 ret = btrfs_inode_by_name(dir, dentry, &location);
4184 return ERR_PTR(ret);
4186 if (location.objectid == 0)
4189 if (location.type == BTRFS_INODE_ITEM_KEY) {
4190 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4194 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4196 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4197 ret = fixup_tree_root_location(root, dir, dentry,
4198 &location, &sub_root);
4201 inode = ERR_PTR(ret);
4203 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4205 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4207 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4209 if (!IS_ERR(inode) && root != sub_root) {
4210 down_read(&root->fs_info->cleanup_work_sem);
4211 if (!(inode->i_sb->s_flags & MS_RDONLY))
4212 ret = btrfs_orphan_cleanup(sub_root);
4213 up_read(&root->fs_info->cleanup_work_sem);
4215 inode = ERR_PTR(ret);
4221 static int btrfs_dentry_delete(const struct dentry *dentry)
4223 struct btrfs_root *root;
4224 struct inode *inode = dentry->d_inode;
4226 if (!inode && !IS_ROOT(dentry))
4227 inode = dentry->d_parent->d_inode;
4230 root = BTRFS_I(inode)->root;
4231 if (btrfs_root_refs(&root->root_item) == 0)
4234 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4240 static void btrfs_dentry_release(struct dentry *dentry)
4242 if (dentry->d_fsdata)
4243 kfree(dentry->d_fsdata);
4246 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4247 struct nameidata *nd)
4251 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4252 if (unlikely(d_need_lookup(dentry))) {
4253 spin_lock(&dentry->d_lock);
4254 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4255 spin_unlock(&dentry->d_lock);
4260 unsigned char btrfs_filetype_table[] = {
4261 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4264 static int btrfs_real_readdir(struct file *filp, void *dirent,
4267 struct inode *inode = filp->f_dentry->d_inode;
4268 struct btrfs_root *root = BTRFS_I(inode)->root;
4269 struct btrfs_item *item;
4270 struct btrfs_dir_item *di;
4271 struct btrfs_key key;
4272 struct btrfs_key found_key;
4273 struct btrfs_path *path;
4274 struct list_head ins_list;
4275 struct list_head del_list;
4277 struct extent_buffer *leaf;
4279 unsigned char d_type;
4284 int key_type = BTRFS_DIR_INDEX_KEY;
4288 int is_curr = 0; /* filp->f_pos points to the current index? */
4290 /* FIXME, use a real flag for deciding about the key type */
4291 if (root->fs_info->tree_root == root)
4292 key_type = BTRFS_DIR_ITEM_KEY;
4294 /* special case for "." */
4295 if (filp->f_pos == 0) {
4296 over = filldir(dirent, ".", 1,
4297 filp->f_pos, btrfs_ino(inode), DT_DIR);
4302 /* special case for .., just use the back ref */
4303 if (filp->f_pos == 1) {
4304 u64 pino = parent_ino(filp->f_path.dentry);
4305 over = filldir(dirent, "..", 2,
4306 filp->f_pos, pino, DT_DIR);
4311 path = btrfs_alloc_path();
4317 if (key_type == BTRFS_DIR_INDEX_KEY) {
4318 INIT_LIST_HEAD(&ins_list);
4319 INIT_LIST_HEAD(&del_list);
4320 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4323 btrfs_set_key_type(&key, key_type);
4324 key.offset = filp->f_pos;
4325 key.objectid = btrfs_ino(inode);
4327 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4332 leaf = path->nodes[0];
4333 slot = path->slots[0];
4334 if (slot >= btrfs_header_nritems(leaf)) {
4335 ret = btrfs_next_leaf(root, path);
4343 item = btrfs_item_nr(leaf, slot);
4344 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4346 if (found_key.objectid != key.objectid)
4348 if (btrfs_key_type(&found_key) != key_type)
4350 if (found_key.offset < filp->f_pos)
4352 if (key_type == BTRFS_DIR_INDEX_KEY &&
4353 btrfs_should_delete_dir_index(&del_list,
4357 filp->f_pos = found_key.offset;
4360 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4362 di_total = btrfs_item_size(leaf, item);
4364 while (di_cur < di_total) {
4365 struct btrfs_key location;
4367 if (verify_dir_item(root, leaf, di))
4370 name_len = btrfs_dir_name_len(leaf, di);
4371 if (name_len <= sizeof(tmp_name)) {
4372 name_ptr = tmp_name;
4374 name_ptr = kmalloc(name_len, GFP_NOFS);
4380 read_extent_buffer(leaf, name_ptr,
4381 (unsigned long)(di + 1), name_len);
4383 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4384 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4387 /* is this a reference to our own snapshot? If so
4390 * In contrast to old kernels, we insert the snapshot's
4391 * dir item and dir index after it has been created, so
4392 * we won't find a reference to our own snapshot. We
4393 * still keep the following code for backward
4396 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4397 location.objectid == root->root_key.objectid) {
4401 over = filldir(dirent, name_ptr, name_len,
4402 found_key.offset, location.objectid,
4406 if (name_ptr != tmp_name)
4411 di_len = btrfs_dir_name_len(leaf, di) +
4412 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4414 di = (struct btrfs_dir_item *)((char *)di + di_len);
4420 if (key_type == BTRFS_DIR_INDEX_KEY) {
4423 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4429 /* Reached end of directory/root. Bump pos past the last item. */
4430 if (key_type == BTRFS_DIR_INDEX_KEY)
4432 * 32-bit glibc will use getdents64, but then strtol -
4433 * so the last number we can serve is this.
4435 filp->f_pos = 0x7fffffff;
4441 if (key_type == BTRFS_DIR_INDEX_KEY)
4442 btrfs_put_delayed_items(&ins_list, &del_list);
4443 btrfs_free_path(path);
4447 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4449 struct btrfs_root *root = BTRFS_I(inode)->root;
4450 struct btrfs_trans_handle *trans;
4452 bool nolock = false;
4454 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4457 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4460 if (wbc->sync_mode == WB_SYNC_ALL) {
4462 trans = btrfs_join_transaction_nolock(root);
4464 trans = btrfs_join_transaction(root);
4466 return PTR_ERR(trans);
4468 ret = btrfs_end_transaction_nolock(trans, root);
4470 ret = btrfs_commit_transaction(trans, root);
4476 * This is somewhat expensive, updating the tree every time the
4477 * inode changes. But, it is most likely to find the inode in cache.
4478 * FIXME, needs more benchmarking...there are no reasons other than performance
4479 * to keep or drop this code.
4481 int btrfs_dirty_inode(struct inode *inode)
4483 struct btrfs_root *root = BTRFS_I(inode)->root;
4484 struct btrfs_trans_handle *trans;
4487 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4490 trans = btrfs_join_transaction(root);
4492 return PTR_ERR(trans);
4494 ret = btrfs_update_inode(trans, root, inode);
4495 if (ret && ret == -ENOSPC) {
4496 /* whoops, lets try again with the full transaction */
4497 btrfs_end_transaction(trans, root);
4498 trans = btrfs_start_transaction(root, 1);
4500 return PTR_ERR(trans);
4502 ret = btrfs_update_inode(trans, root, inode);
4504 btrfs_end_transaction(trans, root);
4505 if (BTRFS_I(inode)->delayed_node)
4506 btrfs_balance_delayed_items(root);
4512 * This is a copy of file_update_time. We need this so we can return error on
4513 * ENOSPC for updating the inode in the case of file write and mmap writes.
4515 int btrfs_update_time(struct file *file)
4517 struct inode *inode = file->f_path.dentry->d_inode;
4518 struct timespec now;
4520 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4522 /* First try to exhaust all avenues to not sync */
4523 if (IS_NOCMTIME(inode))
4526 now = current_fs_time(inode->i_sb);
4527 if (!timespec_equal(&inode->i_mtime, &now))
4530 if (!timespec_equal(&inode->i_ctime, &now))
4533 if (IS_I_VERSION(inode))
4534 sync_it |= S_VERSION;
4539 /* Finally allowed to write? Takes lock. */
4540 if (mnt_want_write_file(file))
4543 /* Only change inode inside the lock region */
4544 if (sync_it & S_VERSION)
4545 inode_inc_iversion(inode);
4546 if (sync_it & S_CTIME)
4547 inode->i_ctime = now;
4548 if (sync_it & S_MTIME)
4549 inode->i_mtime = now;
4550 ret = btrfs_dirty_inode(inode);
4552 mark_inode_dirty_sync(inode);
4553 mnt_drop_write(file->f_path.mnt);
4558 * find the highest existing sequence number in a directory
4559 * and then set the in-memory index_cnt variable to reflect
4560 * free sequence numbers
4562 static int btrfs_set_inode_index_count(struct inode *inode)
4564 struct btrfs_root *root = BTRFS_I(inode)->root;
4565 struct btrfs_key key, found_key;
4566 struct btrfs_path *path;
4567 struct extent_buffer *leaf;
4570 key.objectid = btrfs_ino(inode);
4571 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4572 key.offset = (u64)-1;
4574 path = btrfs_alloc_path();
4578 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4581 /* FIXME: we should be able to handle this */
4587 * MAGIC NUMBER EXPLANATION:
4588 * since we search a directory based on f_pos we have to start at 2
4589 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4590 * else has to start at 2
4592 if (path->slots[0] == 0) {
4593 BTRFS_I(inode)->index_cnt = 2;
4599 leaf = path->nodes[0];
4600 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4602 if (found_key.objectid != btrfs_ino(inode) ||
4603 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4604 BTRFS_I(inode)->index_cnt = 2;
4608 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4610 btrfs_free_path(path);
4615 * helper to find a free sequence number in a given directory. This current
4616 * code is very simple, later versions will do smarter things in the btree
4618 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4622 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4623 ret = btrfs_inode_delayed_dir_index_count(dir);
4625 ret = btrfs_set_inode_index_count(dir);
4631 *index = BTRFS_I(dir)->index_cnt;
4632 BTRFS_I(dir)->index_cnt++;
4637 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4638 struct btrfs_root *root,
4640 const char *name, int name_len,
4641 u64 ref_objectid, u64 objectid,
4642 umode_t mode, u64 *index)
4644 struct inode *inode;
4645 struct btrfs_inode_item *inode_item;
4646 struct btrfs_key *location;
4647 struct btrfs_path *path;
4648 struct btrfs_inode_ref *ref;
4649 struct btrfs_key key[2];
4655 path = btrfs_alloc_path();
4657 return ERR_PTR(-ENOMEM);
4659 inode = new_inode(root->fs_info->sb);
4661 btrfs_free_path(path);
4662 return ERR_PTR(-ENOMEM);
4666 * we have to initialize this early, so we can reclaim the inode
4667 * number if we fail afterwards in this function.
4669 inode->i_ino = objectid;
4672 trace_btrfs_inode_request(dir);
4674 ret = btrfs_set_inode_index(dir, index);
4676 btrfs_free_path(path);
4678 return ERR_PTR(ret);
4682 * index_cnt is ignored for everything but a dir,
4683 * btrfs_get_inode_index_count has an explanation for the magic
4686 BTRFS_I(inode)->index_cnt = 2;
4687 BTRFS_I(inode)->root = root;
4688 BTRFS_I(inode)->generation = trans->transid;
4689 inode->i_generation = BTRFS_I(inode)->generation;
4690 btrfs_set_inode_space_info(root, inode);
4697 key[0].objectid = objectid;
4698 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4701 key[1].objectid = objectid;
4702 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4703 key[1].offset = ref_objectid;
4705 sizes[0] = sizeof(struct btrfs_inode_item);
4706 sizes[1] = name_len + sizeof(*ref);
4708 path->leave_spinning = 1;
4709 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4713 inode_init_owner(inode, dir, mode);
4714 inode_set_bytes(inode, 0);
4715 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4716 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4717 struct btrfs_inode_item);
4718 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4720 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4721 struct btrfs_inode_ref);
4722 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4723 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4724 ptr = (unsigned long)(ref + 1);
4725 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4727 btrfs_mark_buffer_dirty(path->nodes[0]);
4728 btrfs_free_path(path);
4730 location = &BTRFS_I(inode)->location;
4731 location->objectid = objectid;
4732 location->offset = 0;
4733 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4735 btrfs_inherit_iflags(inode, dir);
4737 if (S_ISREG(mode)) {
4738 if (btrfs_test_opt(root, NODATASUM))
4739 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4740 if (btrfs_test_opt(root, NODATACOW) ||
4741 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4742 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4745 insert_inode_hash(inode);
4746 inode_tree_add(inode);
4748 trace_btrfs_inode_new(inode);
4749 btrfs_set_inode_last_trans(trans, inode);
4754 BTRFS_I(dir)->index_cnt--;
4755 btrfs_free_path(path);
4757 return ERR_PTR(ret);
4760 static inline u8 btrfs_inode_type(struct inode *inode)
4762 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4766 * utility function to add 'inode' into 'parent_inode' with
4767 * a give name and a given sequence number.
4768 * if 'add_backref' is true, also insert a backref from the
4769 * inode to the parent directory.
4771 int btrfs_add_link(struct btrfs_trans_handle *trans,
4772 struct inode *parent_inode, struct inode *inode,
4773 const char *name, int name_len, int add_backref, u64 index)
4776 struct btrfs_key key;
4777 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4778 u64 ino = btrfs_ino(inode);
4779 u64 parent_ino = btrfs_ino(parent_inode);
4781 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4782 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4785 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4789 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4790 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4791 key.objectid, root->root_key.objectid,
4792 parent_ino, index, name, name_len);
4793 } else if (add_backref) {
4794 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4798 /* Nothing to clean up yet */
4802 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4804 btrfs_inode_type(inode), index);
4808 btrfs_abort_transaction(trans, root, ret);
4812 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4814 inode_inc_iversion(parent_inode);
4815 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4816 ret = btrfs_update_inode(trans, root, parent_inode);
4818 btrfs_abort_transaction(trans, root, ret);
4822 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4825 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4826 key.objectid, root->root_key.objectid,
4827 parent_ino, &local_index, name, name_len);
4829 } else if (add_backref) {
4833 err = btrfs_del_inode_ref(trans, root, name, name_len,
4834 ino, parent_ino, &local_index);
4839 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4840 struct inode *dir, struct dentry *dentry,
4841 struct inode *inode, int backref, u64 index)
4843 int err = btrfs_add_link(trans, dir, inode,
4844 dentry->d_name.name, dentry->d_name.len,
4851 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4852 umode_t mode, dev_t rdev)
4854 struct btrfs_trans_handle *trans;
4855 struct btrfs_root *root = BTRFS_I(dir)->root;
4856 struct inode *inode = NULL;
4860 unsigned long nr = 0;
4863 if (!new_valid_dev(rdev))
4867 * 2 for inode item and ref
4869 * 1 for xattr if selinux is on
4871 trans = btrfs_start_transaction(root, 5);
4873 return PTR_ERR(trans);
4875 err = btrfs_find_free_ino(root, &objectid);
4879 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4880 dentry->d_name.len, btrfs_ino(dir), objectid,
4882 if (IS_ERR(inode)) {
4883 err = PTR_ERR(inode);
4887 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4894 * If the active LSM wants to access the inode during
4895 * d_instantiate it needs these. Smack checks to see
4896 * if the filesystem supports xattrs by looking at the
4900 inode->i_op = &btrfs_special_inode_operations;
4901 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4905 init_special_inode(inode, inode->i_mode, rdev);
4906 btrfs_update_inode(trans, root, inode);
4907 d_instantiate(dentry, inode);
4910 nr = trans->blocks_used;
4911 btrfs_end_transaction(trans, root);
4912 btrfs_btree_balance_dirty(root, nr);
4914 inode_dec_link_count(inode);
4920 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4921 umode_t mode, struct nameidata *nd)
4923 struct btrfs_trans_handle *trans;
4924 struct btrfs_root *root = BTRFS_I(dir)->root;
4925 struct inode *inode = NULL;
4928 unsigned long nr = 0;
4933 * 2 for inode item and ref
4935 * 1 for xattr if selinux is on
4937 trans = btrfs_start_transaction(root, 5);
4939 return PTR_ERR(trans);
4941 err = btrfs_find_free_ino(root, &objectid);
4945 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4946 dentry->d_name.len, btrfs_ino(dir), objectid,
4948 if (IS_ERR(inode)) {
4949 err = PTR_ERR(inode);
4953 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4960 * If the active LSM wants to access the inode during
4961 * d_instantiate it needs these. Smack checks to see
4962 * if the filesystem supports xattrs by looking at the
4965 inode->i_fop = &btrfs_file_operations;
4966 inode->i_op = &btrfs_file_inode_operations;
4968 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4972 inode->i_mapping->a_ops = &btrfs_aops;
4973 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4974 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4975 d_instantiate(dentry, inode);
4978 nr = trans->blocks_used;
4979 btrfs_end_transaction(trans, root);
4981 inode_dec_link_count(inode);
4984 btrfs_btree_balance_dirty(root, nr);
4988 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4989 struct dentry *dentry)
4991 struct btrfs_trans_handle *trans;
4992 struct btrfs_root *root = BTRFS_I(dir)->root;
4993 struct inode *inode = old_dentry->d_inode;
4995 unsigned long nr = 0;
4999 /* do not allow sys_link's with other subvols of the same device */
5000 if (root->objectid != BTRFS_I(inode)->root->objectid)
5003 if (inode->i_nlink == ~0U)
5006 err = btrfs_set_inode_index(dir, &index);
5011 * 2 items for inode and inode ref
5012 * 2 items for dir items
5013 * 1 item for parent inode
5015 trans = btrfs_start_transaction(root, 5);
5016 if (IS_ERR(trans)) {
5017 err = PTR_ERR(trans);
5021 btrfs_inc_nlink(inode);
5022 inode_inc_iversion(inode);
5023 inode->i_ctime = CURRENT_TIME;
5026 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5031 struct dentry *parent = dentry->d_parent;
5032 err = btrfs_update_inode(trans, root, inode);
5035 d_instantiate(dentry, inode);
5036 btrfs_log_new_name(trans, inode, NULL, parent);
5039 nr = trans->blocks_used;
5040 btrfs_end_transaction(trans, root);
5043 inode_dec_link_count(inode);
5046 btrfs_btree_balance_dirty(root, nr);
5050 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5052 struct inode *inode = NULL;
5053 struct btrfs_trans_handle *trans;
5054 struct btrfs_root *root = BTRFS_I(dir)->root;
5056 int drop_on_err = 0;
5059 unsigned long nr = 1;
5062 * 2 items for inode and ref
5063 * 2 items for dir items
5064 * 1 for xattr if selinux is on
5066 trans = btrfs_start_transaction(root, 5);
5068 return PTR_ERR(trans);
5070 err = btrfs_find_free_ino(root, &objectid);
5074 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5075 dentry->d_name.len, btrfs_ino(dir), objectid,
5076 S_IFDIR | mode, &index);
5077 if (IS_ERR(inode)) {
5078 err = PTR_ERR(inode);
5084 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5088 inode->i_op = &btrfs_dir_inode_operations;
5089 inode->i_fop = &btrfs_dir_file_operations;
5091 btrfs_i_size_write(inode, 0);
5092 err = btrfs_update_inode(trans, root, inode);
5096 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5097 dentry->d_name.len, 0, index);
5101 d_instantiate(dentry, inode);
5105 nr = trans->blocks_used;
5106 btrfs_end_transaction(trans, root);
5109 btrfs_btree_balance_dirty(root, nr);
5113 /* helper for btfs_get_extent. Given an existing extent in the tree,
5114 * and an extent that you want to insert, deal with overlap and insert
5115 * the new extent into the tree.
5117 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5118 struct extent_map *existing,
5119 struct extent_map *em,
5120 u64 map_start, u64 map_len)
5124 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5125 start_diff = map_start - em->start;
5126 em->start = map_start;
5128 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5129 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5130 em->block_start += start_diff;
5131 em->block_len -= start_diff;
5133 return add_extent_mapping(em_tree, em);
5136 static noinline int uncompress_inline(struct btrfs_path *path,
5137 struct inode *inode, struct page *page,
5138 size_t pg_offset, u64 extent_offset,
5139 struct btrfs_file_extent_item *item)
5142 struct extent_buffer *leaf = path->nodes[0];
5145 unsigned long inline_size;
5149 WARN_ON(pg_offset != 0);
5150 compress_type = btrfs_file_extent_compression(leaf, item);
5151 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5152 inline_size = btrfs_file_extent_inline_item_len(leaf,
5153 btrfs_item_nr(leaf, path->slots[0]));
5154 tmp = kmalloc(inline_size, GFP_NOFS);
5157 ptr = btrfs_file_extent_inline_start(item);
5159 read_extent_buffer(leaf, tmp, ptr, inline_size);
5161 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5162 ret = btrfs_decompress(compress_type, tmp, page,
5163 extent_offset, inline_size, max_size);
5165 char *kaddr = kmap_atomic(page);
5166 unsigned long copy_size = min_t(u64,
5167 PAGE_CACHE_SIZE - pg_offset,
5168 max_size - extent_offset);
5169 memset(kaddr + pg_offset, 0, copy_size);
5170 kunmap_atomic(kaddr);
5177 * a bit scary, this does extent mapping from logical file offset to the disk.
5178 * the ugly parts come from merging extents from the disk with the in-ram
5179 * representation. This gets more complex because of the data=ordered code,
5180 * where the in-ram extents might be locked pending data=ordered completion.
5182 * This also copies inline extents directly into the page.
5185 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5186 size_t pg_offset, u64 start, u64 len,
5192 u64 extent_start = 0;
5194 u64 objectid = btrfs_ino(inode);
5196 struct btrfs_path *path = NULL;
5197 struct btrfs_root *root = BTRFS_I(inode)->root;
5198 struct btrfs_file_extent_item *item;
5199 struct extent_buffer *leaf;
5200 struct btrfs_key found_key;
5201 struct extent_map *em = NULL;
5202 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5203 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5204 struct btrfs_trans_handle *trans = NULL;
5208 read_lock(&em_tree->lock);
5209 em = lookup_extent_mapping(em_tree, start, len);
5211 em->bdev = root->fs_info->fs_devices->latest_bdev;
5212 read_unlock(&em_tree->lock);
5215 if (em->start > start || em->start + em->len <= start)
5216 free_extent_map(em);
5217 else if (em->block_start == EXTENT_MAP_INLINE && page)
5218 free_extent_map(em);
5222 em = alloc_extent_map();
5227 em->bdev = root->fs_info->fs_devices->latest_bdev;
5228 em->start = EXTENT_MAP_HOLE;
5229 em->orig_start = EXTENT_MAP_HOLE;
5231 em->block_len = (u64)-1;
5234 path = btrfs_alloc_path();
5240 * Chances are we'll be called again, so go ahead and do
5246 ret = btrfs_lookup_file_extent(trans, root, path,
5247 objectid, start, trans != NULL);
5254 if (path->slots[0] == 0)
5259 leaf = path->nodes[0];
5260 item = btrfs_item_ptr(leaf, path->slots[0],
5261 struct btrfs_file_extent_item);
5262 /* are we inside the extent that was found? */
5263 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5264 found_type = btrfs_key_type(&found_key);
5265 if (found_key.objectid != objectid ||
5266 found_type != BTRFS_EXTENT_DATA_KEY) {
5270 found_type = btrfs_file_extent_type(leaf, item);
5271 extent_start = found_key.offset;
5272 compress_type = btrfs_file_extent_compression(leaf, item);
5273 if (found_type == BTRFS_FILE_EXTENT_REG ||
5274 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5275 extent_end = extent_start +
5276 btrfs_file_extent_num_bytes(leaf, item);
5277 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5279 size = btrfs_file_extent_inline_len(leaf, item);
5280 extent_end = (extent_start + size + root->sectorsize - 1) &
5281 ~((u64)root->sectorsize - 1);
5284 if (start >= extent_end) {
5286 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5287 ret = btrfs_next_leaf(root, path);
5294 leaf = path->nodes[0];
5296 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5297 if (found_key.objectid != objectid ||
5298 found_key.type != BTRFS_EXTENT_DATA_KEY)
5300 if (start + len <= found_key.offset)
5303 em->len = found_key.offset - start;
5307 if (found_type == BTRFS_FILE_EXTENT_REG ||
5308 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5309 em->start = extent_start;
5310 em->len = extent_end - extent_start;
5311 em->orig_start = extent_start -
5312 btrfs_file_extent_offset(leaf, item);
5313 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5315 em->block_start = EXTENT_MAP_HOLE;
5318 if (compress_type != BTRFS_COMPRESS_NONE) {
5319 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5320 em->compress_type = compress_type;
5321 em->block_start = bytenr;
5322 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5325 bytenr += btrfs_file_extent_offset(leaf, item);
5326 em->block_start = bytenr;
5327 em->block_len = em->len;
5328 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5329 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5332 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5336 size_t extent_offset;
5339 em->block_start = EXTENT_MAP_INLINE;
5340 if (!page || create) {
5341 em->start = extent_start;
5342 em->len = extent_end - extent_start;
5346 size = btrfs_file_extent_inline_len(leaf, item);
5347 extent_offset = page_offset(page) + pg_offset - extent_start;
5348 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5349 size - extent_offset);
5350 em->start = extent_start + extent_offset;
5351 em->len = (copy_size + root->sectorsize - 1) &
5352 ~((u64)root->sectorsize - 1);
5353 em->orig_start = EXTENT_MAP_INLINE;
5354 if (compress_type) {
5355 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5356 em->compress_type = compress_type;
5358 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5359 if (create == 0 && !PageUptodate(page)) {
5360 if (btrfs_file_extent_compression(leaf, item) !=
5361 BTRFS_COMPRESS_NONE) {
5362 ret = uncompress_inline(path, inode, page,
5364 extent_offset, item);
5365 BUG_ON(ret); /* -ENOMEM */
5368 read_extent_buffer(leaf, map + pg_offset, ptr,
5370 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5371 memset(map + pg_offset + copy_size, 0,
5372 PAGE_CACHE_SIZE - pg_offset -
5377 flush_dcache_page(page);
5378 } else if (create && PageUptodate(page)) {
5382 free_extent_map(em);
5385 btrfs_release_path(path);
5386 trans = btrfs_join_transaction(root);
5389 return ERR_CAST(trans);
5393 write_extent_buffer(leaf, map + pg_offset, ptr,
5396 btrfs_mark_buffer_dirty(leaf);
5398 set_extent_uptodate(io_tree, em->start,
5399 extent_map_end(em) - 1, NULL, GFP_NOFS);
5402 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5409 em->block_start = EXTENT_MAP_HOLE;
5410 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5412 btrfs_release_path(path);
5413 if (em->start > start || extent_map_end(em) <= start) {
5414 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5415 "[%llu %llu]\n", (unsigned long long)em->start,
5416 (unsigned long long)em->len,
5417 (unsigned long long)start,
5418 (unsigned long long)len);
5424 write_lock(&em_tree->lock);
5425 ret = add_extent_mapping(em_tree, em);
5426 /* it is possible that someone inserted the extent into the tree
5427 * while we had the lock dropped. It is also possible that
5428 * an overlapping map exists in the tree
5430 if (ret == -EEXIST) {
5431 struct extent_map *existing;
5435 existing = lookup_extent_mapping(em_tree, start, len);
5436 if (existing && (existing->start > start ||
5437 existing->start + existing->len <= start)) {
5438 free_extent_map(existing);
5442 existing = lookup_extent_mapping(em_tree, em->start,
5445 err = merge_extent_mapping(em_tree, existing,
5448 free_extent_map(existing);
5450 free_extent_map(em);
5455 free_extent_map(em);
5459 free_extent_map(em);
5464 write_unlock(&em_tree->lock);
5467 trace_btrfs_get_extent(root, em);
5470 btrfs_free_path(path);
5472 ret = btrfs_end_transaction(trans, root);
5477 free_extent_map(em);
5478 return ERR_PTR(err);
5480 BUG_ON(!em); /* Error is always set */
5484 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5485 size_t pg_offset, u64 start, u64 len,
5488 struct extent_map *em;
5489 struct extent_map *hole_em = NULL;
5490 u64 range_start = start;
5496 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5501 * if our em maps to a hole, there might
5502 * actually be delalloc bytes behind it
5504 if (em->block_start != EXTENT_MAP_HOLE)
5510 /* check to see if we've wrapped (len == -1 or similar) */
5519 /* ok, we didn't find anything, lets look for delalloc */
5520 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5521 end, len, EXTENT_DELALLOC, 1);
5522 found_end = range_start + found;
5523 if (found_end < range_start)
5524 found_end = (u64)-1;
5527 * we didn't find anything useful, return
5528 * the original results from get_extent()
5530 if (range_start > end || found_end <= start) {
5536 /* adjust the range_start to make sure it doesn't
5537 * go backwards from the start they passed in
5539 range_start = max(start,range_start);
5540 found = found_end - range_start;
5543 u64 hole_start = start;
5546 em = alloc_extent_map();
5552 * when btrfs_get_extent can't find anything it
5553 * returns one huge hole
5555 * make sure what it found really fits our range, and
5556 * adjust to make sure it is based on the start from
5560 u64 calc_end = extent_map_end(hole_em);
5562 if (calc_end <= start || (hole_em->start > end)) {
5563 free_extent_map(hole_em);
5566 hole_start = max(hole_em->start, start);
5567 hole_len = calc_end - hole_start;
5571 if (hole_em && range_start > hole_start) {
5572 /* our hole starts before our delalloc, so we
5573 * have to return just the parts of the hole
5574 * that go until the delalloc starts
5576 em->len = min(hole_len,
5577 range_start - hole_start);
5578 em->start = hole_start;
5579 em->orig_start = hole_start;
5581 * don't adjust block start at all,
5582 * it is fixed at EXTENT_MAP_HOLE
5584 em->block_start = hole_em->block_start;
5585 em->block_len = hole_len;
5587 em->start = range_start;
5589 em->orig_start = range_start;
5590 em->block_start = EXTENT_MAP_DELALLOC;
5591 em->block_len = found;
5593 } else if (hole_em) {
5598 free_extent_map(hole_em);
5600 free_extent_map(em);
5601 return ERR_PTR(err);
5606 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5607 struct extent_map *em,
5610 struct btrfs_root *root = BTRFS_I(inode)->root;
5611 struct btrfs_trans_handle *trans;
5612 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5613 struct btrfs_key ins;
5616 bool insert = false;
5619 * Ok if the extent map we looked up is a hole and is for the exact
5620 * range we want, there is no reason to allocate a new one, however if
5621 * it is not right then we need to free this one and drop the cache for
5624 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5626 free_extent_map(em);
5629 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5632 trans = btrfs_join_transaction(root);
5634 return ERR_CAST(trans);
5636 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5637 btrfs_add_inode_defrag(trans, inode);
5639 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5641 alloc_hint = get_extent_allocation_hint(inode, start, len);
5642 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5643 alloc_hint, &ins, 1);
5650 em = alloc_extent_map();
5652 em = ERR_PTR(-ENOMEM);
5658 em->orig_start = em->start;
5659 em->len = ins.offset;
5661 em->block_start = ins.objectid;
5662 em->block_len = ins.offset;
5663 em->bdev = root->fs_info->fs_devices->latest_bdev;
5666 * We need to do this because if we're using the original em we searched
5667 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5670 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5673 write_lock(&em_tree->lock);
5674 ret = add_extent_mapping(em_tree, em);
5675 write_unlock(&em_tree->lock);
5678 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5681 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5682 ins.offset, ins.offset, 0);
5684 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5688 btrfs_end_transaction(trans, root);
5693 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5694 * block must be cow'd
5696 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5697 struct inode *inode, u64 offset, u64 len)
5699 struct btrfs_path *path;
5701 struct extent_buffer *leaf;
5702 struct btrfs_root *root = BTRFS_I(inode)->root;
5703 struct btrfs_file_extent_item *fi;
5704 struct btrfs_key key;
5712 path = btrfs_alloc_path();
5716 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5721 slot = path->slots[0];
5724 /* can't find the item, must cow */
5731 leaf = path->nodes[0];
5732 btrfs_item_key_to_cpu(leaf, &key, slot);
5733 if (key.objectid != btrfs_ino(inode) ||
5734 key.type != BTRFS_EXTENT_DATA_KEY) {
5735 /* not our file or wrong item type, must cow */
5739 if (key.offset > offset) {
5740 /* Wrong offset, must cow */
5744 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5745 found_type = btrfs_file_extent_type(leaf, fi);
5746 if (found_type != BTRFS_FILE_EXTENT_REG &&
5747 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5748 /* not a regular extent, must cow */
5751 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5752 backref_offset = btrfs_file_extent_offset(leaf, fi);
5754 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5755 if (extent_end < offset + len) {
5756 /* extent doesn't include our full range, must cow */
5760 if (btrfs_extent_readonly(root, disk_bytenr))
5764 * look for other files referencing this extent, if we
5765 * find any we must cow
5767 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5768 key.offset - backref_offset, disk_bytenr))
5772 * adjust disk_bytenr and num_bytes to cover just the bytes
5773 * in this extent we are about to write. If there
5774 * are any csums in that range we have to cow in order
5775 * to keep the csums correct
5777 disk_bytenr += backref_offset;
5778 disk_bytenr += offset - key.offset;
5779 num_bytes = min(offset + len, extent_end) - offset;
5780 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5783 * all of the above have passed, it is safe to overwrite this extent
5788 btrfs_free_path(path);
5792 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5793 struct buffer_head *bh_result, int create)
5795 struct extent_map *em;
5796 struct btrfs_root *root = BTRFS_I(inode)->root;
5797 u64 start = iblock << inode->i_blkbits;
5798 u64 len = bh_result->b_size;
5799 struct btrfs_trans_handle *trans;
5801 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5806 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5807 * io. INLINE is special, and we could probably kludge it in here, but
5808 * it's still buffered so for safety lets just fall back to the generic
5811 * For COMPRESSED we _have_ to read the entire extent in so we can
5812 * decompress it, so there will be buffering required no matter what we
5813 * do, so go ahead and fallback to buffered.
5815 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5816 * to buffered IO. Don't blame me, this is the price we pay for using
5819 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5820 em->block_start == EXTENT_MAP_INLINE) {
5821 free_extent_map(em);
5825 /* Just a good old fashioned hole, return */
5826 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5827 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5828 free_extent_map(em);
5829 /* DIO will do one hole at a time, so just unlock a sector */
5830 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5831 start + root->sectorsize - 1);
5836 * We don't allocate a new extent in the following cases
5838 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5840 * 2) The extent is marked as PREALLOC. We're good to go here and can
5841 * just use the extent.
5845 len = em->len - (start - em->start);
5849 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5850 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5851 em->block_start != EXTENT_MAP_HOLE)) {
5856 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5857 type = BTRFS_ORDERED_PREALLOC;
5859 type = BTRFS_ORDERED_NOCOW;
5860 len = min(len, em->len - (start - em->start));
5861 block_start = em->block_start + (start - em->start);
5864 * we're not going to log anything, but we do need
5865 * to make sure the current transaction stays open
5866 * while we look for nocow cross refs
5868 trans = btrfs_join_transaction(root);
5872 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5873 ret = btrfs_add_ordered_extent_dio(inode, start,
5874 block_start, len, len, type);
5875 btrfs_end_transaction(trans, root);
5877 free_extent_map(em);
5882 btrfs_end_transaction(trans, root);
5886 * this will cow the extent, reset the len in case we changed
5889 len = bh_result->b_size;
5890 em = btrfs_new_extent_direct(inode, em, start, len);
5893 len = min(len, em->len - (start - em->start));
5895 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5896 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5899 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5901 bh_result->b_size = len;
5902 bh_result->b_bdev = em->bdev;
5903 set_buffer_mapped(bh_result);
5904 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5905 set_buffer_new(bh_result);
5907 free_extent_map(em);
5912 struct btrfs_dio_private {
5913 struct inode *inode;
5920 /* number of bios pending for this dio */
5921 atomic_t pending_bios;
5926 struct bio *orig_bio;
5929 static void btrfs_endio_direct_read(struct bio *bio, int err)
5931 struct btrfs_dio_private *dip = bio->bi_private;
5932 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5933 struct bio_vec *bvec = bio->bi_io_vec;
5934 struct inode *inode = dip->inode;
5935 struct btrfs_root *root = BTRFS_I(inode)->root;
5937 u32 *private = dip->csums;
5939 start = dip->logical_offset;
5941 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5942 struct page *page = bvec->bv_page;
5945 unsigned long flags;
5947 local_irq_save(flags);
5948 kaddr = kmap_atomic(page);
5949 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5950 csum, bvec->bv_len);
5951 btrfs_csum_final(csum, (char *)&csum);
5952 kunmap_atomic(kaddr);
5953 local_irq_restore(flags);
5955 flush_dcache_page(bvec->bv_page);
5956 if (csum != *private) {
5957 printk(KERN_ERR "btrfs csum failed ino %llu off"
5958 " %llu csum %u private %u\n",
5959 (unsigned long long)btrfs_ino(inode),
5960 (unsigned long long)start,
5966 start += bvec->bv_len;
5969 } while (bvec <= bvec_end);
5971 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5972 dip->logical_offset + dip->bytes - 1);
5973 bio->bi_private = dip->private;
5978 /* If we had a csum failure make sure to clear the uptodate flag */
5980 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5981 dio_end_io(bio, err);
5984 static void btrfs_endio_direct_write(struct bio *bio, int err)
5986 struct btrfs_dio_private *dip = bio->bi_private;
5987 struct inode *inode = dip->inode;
5988 struct btrfs_root *root = BTRFS_I(inode)->root;
5989 struct btrfs_ordered_extent *ordered = NULL;
5990 u64 ordered_offset = dip->logical_offset;
5991 u64 ordered_bytes = dip->bytes;
5997 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5999 ordered_bytes, !err);
6003 ordered->work.func = finish_ordered_fn;
6004 ordered->work.flags = 0;
6005 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6009 * our bio might span multiple ordered extents. If we haven't
6010 * completed the accounting for the whole dio, go back and try again
6012 if (ordered_offset < dip->logical_offset + dip->bytes) {
6013 ordered_bytes = dip->logical_offset + dip->bytes -
6019 bio->bi_private = dip->private;
6023 /* If we had an error make sure to clear the uptodate flag */
6025 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6026 dio_end_io(bio, err);
6029 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6030 struct bio *bio, int mirror_num,
6031 unsigned long bio_flags, u64 offset)
6034 struct btrfs_root *root = BTRFS_I(inode)->root;
6035 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6036 BUG_ON(ret); /* -ENOMEM */
6040 static void btrfs_end_dio_bio(struct bio *bio, int err)
6042 struct btrfs_dio_private *dip = bio->bi_private;
6045 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6046 "sector %#Lx len %u err no %d\n",
6047 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6048 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6052 * before atomic variable goto zero, we must make sure
6053 * dip->errors is perceived to be set.
6055 smp_mb__before_atomic_dec();
6058 /* if there are more bios still pending for this dio, just exit */
6059 if (!atomic_dec_and_test(&dip->pending_bios))
6063 bio_io_error(dip->orig_bio);
6065 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6066 bio_endio(dip->orig_bio, 0);
6072 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6073 u64 first_sector, gfp_t gfp_flags)
6075 int nr_vecs = bio_get_nr_vecs(bdev);
6076 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6079 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6080 int rw, u64 file_offset, int skip_sum,
6081 u32 *csums, int async_submit)
6083 int write = rw & REQ_WRITE;
6084 struct btrfs_root *root = BTRFS_I(inode)->root;
6090 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6098 if (write && async_submit) {
6099 ret = btrfs_wq_submit_bio(root->fs_info,
6100 inode, rw, bio, 0, 0,
6102 __btrfs_submit_bio_start_direct_io,
6103 __btrfs_submit_bio_done);
6107 * If we aren't doing async submit, calculate the csum of the
6110 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6113 } else if (!skip_sum) {
6114 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6115 file_offset, csums);
6121 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6127 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6130 struct inode *inode = dip->inode;
6131 struct btrfs_root *root = BTRFS_I(inode)->root;
6132 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6134 struct bio *orig_bio = dip->orig_bio;
6135 struct bio_vec *bvec = orig_bio->bi_io_vec;
6136 u64 start_sector = orig_bio->bi_sector;
6137 u64 file_offset = dip->logical_offset;
6141 u32 *csums = dip->csums;
6143 int async_submit = 0;
6144 int write = rw & REQ_WRITE;
6146 map_length = orig_bio->bi_size;
6147 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6148 &map_length, NULL, 0);
6154 if (map_length >= orig_bio->bi_size) {
6160 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6163 bio->bi_private = dip;
6164 bio->bi_end_io = btrfs_end_dio_bio;
6165 atomic_inc(&dip->pending_bios);
6167 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6168 if (unlikely(map_length < submit_len + bvec->bv_len ||
6169 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6170 bvec->bv_offset) < bvec->bv_len)) {
6172 * inc the count before we submit the bio so
6173 * we know the end IO handler won't happen before
6174 * we inc the count. Otherwise, the dip might get freed
6175 * before we're done setting it up
6177 atomic_inc(&dip->pending_bios);
6178 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6179 file_offset, skip_sum,
6180 csums, async_submit);
6183 atomic_dec(&dip->pending_bios);
6187 /* Write's use the ordered csums */
6188 if (!write && !skip_sum)
6189 csums = csums + nr_pages;
6190 start_sector += submit_len >> 9;
6191 file_offset += submit_len;
6196 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6197 start_sector, GFP_NOFS);
6200 bio->bi_private = dip;
6201 bio->bi_end_io = btrfs_end_dio_bio;
6203 map_length = orig_bio->bi_size;
6204 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6205 &map_length, NULL, 0);
6211 submit_len += bvec->bv_len;
6218 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6219 csums, async_submit);
6227 * before atomic variable goto zero, we must
6228 * make sure dip->errors is perceived to be set.
6230 smp_mb__before_atomic_dec();
6231 if (atomic_dec_and_test(&dip->pending_bios))
6232 bio_io_error(dip->orig_bio);
6234 /* bio_end_io() will handle error, so we needn't return it */
6238 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6241 struct btrfs_root *root = BTRFS_I(inode)->root;
6242 struct btrfs_dio_private *dip;
6243 struct bio_vec *bvec = bio->bi_io_vec;
6245 int write = rw & REQ_WRITE;
6248 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6250 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6257 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6258 if (!write && !skip_sum) {
6259 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6267 dip->private = bio->bi_private;
6269 dip->logical_offset = file_offset;
6273 dip->bytes += bvec->bv_len;
6275 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6277 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6278 bio->bi_private = dip;
6280 dip->orig_bio = bio;
6281 atomic_set(&dip->pending_bios, 0);
6284 bio->bi_end_io = btrfs_endio_direct_write;
6286 bio->bi_end_io = btrfs_endio_direct_read;
6288 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6293 * If this is a write, we need to clean up the reserved space and kill
6294 * the ordered extent.
6297 struct btrfs_ordered_extent *ordered;
6298 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6299 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6300 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6301 btrfs_free_reserved_extent(root, ordered->start,
6303 btrfs_put_ordered_extent(ordered);
6304 btrfs_put_ordered_extent(ordered);
6306 bio_endio(bio, ret);
6309 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6310 const struct iovec *iov, loff_t offset,
6311 unsigned long nr_segs)
6317 unsigned blocksize_mask = root->sectorsize - 1;
6318 ssize_t retval = -EINVAL;
6319 loff_t end = offset;
6321 if (offset & blocksize_mask)
6324 /* Check the memory alignment. Blocks cannot straddle pages */
6325 for (seg = 0; seg < nr_segs; seg++) {
6326 addr = (unsigned long)iov[seg].iov_base;
6327 size = iov[seg].iov_len;
6329 if ((addr & blocksize_mask) || (size & blocksize_mask))
6332 /* If this is a write we don't need to check anymore */
6337 * Check to make sure we don't have duplicate iov_base's in this
6338 * iovec, if so return EINVAL, otherwise we'll get csum errors
6339 * when reading back.
6341 for (i = seg + 1; i < nr_segs; i++) {
6342 if (iov[seg].iov_base == iov[i].iov_base)
6350 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6351 const struct iovec *iov, loff_t offset,
6352 unsigned long nr_segs)
6354 struct file *file = iocb->ki_filp;
6355 struct inode *inode = file->f_mapping->host;
6356 struct btrfs_ordered_extent *ordered;
6357 struct extent_state *cached_state = NULL;
6358 u64 lockstart, lockend;
6360 int writing = rw & WRITE;
6362 size_t count = iov_length(iov, nr_segs);
6364 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6370 lockend = offset + count - 1;
6373 ret = btrfs_delalloc_reserve_space(inode, count);
6379 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6382 * We're concerned with the entire range that we're going to be
6383 * doing DIO to, so we need to make sure theres no ordered
6384 * extents in this range.
6386 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6387 lockend - lockstart + 1);
6390 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6391 &cached_state, GFP_NOFS);
6392 btrfs_start_ordered_extent(inode, ordered, 1);
6393 btrfs_put_ordered_extent(ordered);
6398 * we don't use btrfs_set_extent_delalloc because we don't want
6399 * the dirty or uptodate bits
6402 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6403 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6404 EXTENT_DELALLOC, NULL, &cached_state,
6407 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6408 lockend, EXTENT_LOCKED | write_bits,
6409 1, 0, &cached_state, GFP_NOFS);
6414 free_extent_state(cached_state);
6415 cached_state = NULL;
6417 ret = __blockdev_direct_IO(rw, iocb, inode,
6418 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6419 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6420 btrfs_submit_direct, 0);
6422 if (ret < 0 && ret != -EIOCBQUEUED) {
6423 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6424 offset + iov_length(iov, nr_segs) - 1,
6425 EXTENT_LOCKED | write_bits, 1, 0,
6426 &cached_state, GFP_NOFS);
6427 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6429 * We're falling back to buffered, unlock the section we didn't
6432 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6433 offset + iov_length(iov, nr_segs) - 1,
6434 EXTENT_LOCKED | write_bits, 1, 0,
6435 &cached_state, GFP_NOFS);
6438 free_extent_state(cached_state);
6442 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6443 __u64 start, __u64 len)
6445 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6448 int btrfs_readpage(struct file *file, struct page *page)
6450 struct extent_io_tree *tree;
6451 tree = &BTRFS_I(page->mapping->host)->io_tree;
6452 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6455 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6457 struct extent_io_tree *tree;
6460 if (current->flags & PF_MEMALLOC) {
6461 redirty_page_for_writepage(wbc, page);
6465 tree = &BTRFS_I(page->mapping->host)->io_tree;
6466 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6469 int btrfs_writepages(struct address_space *mapping,
6470 struct writeback_control *wbc)
6472 struct extent_io_tree *tree;
6474 tree = &BTRFS_I(mapping->host)->io_tree;
6475 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6479 btrfs_readpages(struct file *file, struct address_space *mapping,
6480 struct list_head *pages, unsigned nr_pages)
6482 struct extent_io_tree *tree;
6483 tree = &BTRFS_I(mapping->host)->io_tree;
6484 return extent_readpages(tree, mapping, pages, nr_pages,
6487 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6489 struct extent_io_tree *tree;
6490 struct extent_map_tree *map;
6493 tree = &BTRFS_I(page->mapping->host)->io_tree;
6494 map = &BTRFS_I(page->mapping->host)->extent_tree;
6495 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6497 ClearPagePrivate(page);
6498 set_page_private(page, 0);
6499 page_cache_release(page);
6504 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6506 if (PageWriteback(page) || PageDirty(page))
6508 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6511 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6513 struct inode *inode = page->mapping->host;
6514 struct extent_io_tree *tree;
6515 struct btrfs_ordered_extent *ordered;
6516 struct extent_state *cached_state = NULL;
6517 u64 page_start = page_offset(page);
6518 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6521 * we have the page locked, so new writeback can't start,
6522 * and the dirty bit won't be cleared while we are here.
6524 * Wait for IO on this page so that we can safely clear
6525 * the PagePrivate2 bit and do ordered accounting
6527 wait_on_page_writeback(page);
6529 tree = &BTRFS_I(inode)->io_tree;
6531 btrfs_releasepage(page, GFP_NOFS);
6534 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6535 ordered = btrfs_lookup_ordered_extent(inode,
6539 * IO on this page will never be started, so we need
6540 * to account for any ordered extents now
6542 clear_extent_bit(tree, page_start, page_end,
6543 EXTENT_DIRTY | EXTENT_DELALLOC |
6544 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6545 &cached_state, GFP_NOFS);
6547 * whoever cleared the private bit is responsible
6548 * for the finish_ordered_io
6550 if (TestClearPagePrivate2(page) &&
6551 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6552 PAGE_CACHE_SIZE, 1)) {
6553 btrfs_finish_ordered_io(ordered);
6555 btrfs_put_ordered_extent(ordered);
6556 cached_state = NULL;
6557 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6559 clear_extent_bit(tree, page_start, page_end,
6560 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6561 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6562 __btrfs_releasepage(page, GFP_NOFS);
6564 ClearPageChecked(page);
6565 if (PagePrivate(page)) {
6566 ClearPagePrivate(page);
6567 set_page_private(page, 0);
6568 page_cache_release(page);
6573 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6574 * called from a page fault handler when a page is first dirtied. Hence we must
6575 * be careful to check for EOF conditions here. We set the page up correctly
6576 * for a written page which means we get ENOSPC checking when writing into
6577 * holes and correct delalloc and unwritten extent mapping on filesystems that
6578 * support these features.
6580 * We are not allowed to take the i_mutex here so we have to play games to
6581 * protect against truncate races as the page could now be beyond EOF. Because
6582 * vmtruncate() writes the inode size before removing pages, once we have the
6583 * page lock we can determine safely if the page is beyond EOF. If it is not
6584 * beyond EOF, then the page is guaranteed safe against truncation until we
6587 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6589 struct page *page = vmf->page;
6590 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6591 struct btrfs_root *root = BTRFS_I(inode)->root;
6592 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6593 struct btrfs_ordered_extent *ordered;
6594 struct extent_state *cached_state = NULL;
6596 unsigned long zero_start;
6603 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6605 ret = btrfs_update_time(vma->vm_file);
6611 else /* -ENOSPC, -EIO, etc */
6612 ret = VM_FAULT_SIGBUS;
6618 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6621 size = i_size_read(inode);
6622 page_start = page_offset(page);
6623 page_end = page_start + PAGE_CACHE_SIZE - 1;
6625 if ((page->mapping != inode->i_mapping) ||
6626 (page_start >= size)) {
6627 /* page got truncated out from underneath us */
6630 wait_on_page_writeback(page);
6632 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6633 set_page_extent_mapped(page);
6636 * we can't set the delalloc bits if there are pending ordered
6637 * extents. Drop our locks and wait for them to finish
6639 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6641 unlock_extent_cached(io_tree, page_start, page_end,
6642 &cached_state, GFP_NOFS);
6644 btrfs_start_ordered_extent(inode, ordered, 1);
6645 btrfs_put_ordered_extent(ordered);
6650 * XXX - page_mkwrite gets called every time the page is dirtied, even
6651 * if it was already dirty, so for space accounting reasons we need to
6652 * clear any delalloc bits for the range we are fixing to save. There
6653 * is probably a better way to do this, but for now keep consistent with
6654 * prepare_pages in the normal write path.
6656 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6657 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6658 0, 0, &cached_state, GFP_NOFS);
6660 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6663 unlock_extent_cached(io_tree, page_start, page_end,
6664 &cached_state, GFP_NOFS);
6665 ret = VM_FAULT_SIGBUS;
6670 /* page is wholly or partially inside EOF */
6671 if (page_start + PAGE_CACHE_SIZE > size)
6672 zero_start = size & ~PAGE_CACHE_MASK;
6674 zero_start = PAGE_CACHE_SIZE;
6676 if (zero_start != PAGE_CACHE_SIZE) {
6678 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6679 flush_dcache_page(page);
6682 ClearPageChecked(page);
6683 set_page_dirty(page);
6684 SetPageUptodate(page);
6686 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6687 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6689 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6693 return VM_FAULT_LOCKED;
6696 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6701 static int btrfs_truncate(struct inode *inode)
6703 struct btrfs_root *root = BTRFS_I(inode)->root;
6704 struct btrfs_block_rsv *rsv;
6707 struct btrfs_trans_handle *trans;
6709 u64 mask = root->sectorsize - 1;
6710 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6712 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6716 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6717 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6720 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6721 * 3 things going on here
6723 * 1) We need to reserve space for our orphan item and the space to
6724 * delete our orphan item. Lord knows we don't want to have a dangling
6725 * orphan item because we didn't reserve space to remove it.
6727 * 2) We need to reserve space to update our inode.
6729 * 3) We need to have something to cache all the space that is going to
6730 * be free'd up by the truncate operation, but also have some slack
6731 * space reserved in case it uses space during the truncate (thank you
6732 * very much snapshotting).
6734 * And we need these to all be seperate. The fact is we can use alot of
6735 * space doing the truncate, and we have no earthly idea how much space
6736 * we will use, so we need the truncate reservation to be seperate so it
6737 * doesn't end up using space reserved for updating the inode or
6738 * removing the orphan item. We also need to be able to stop the
6739 * transaction and start a new one, which means we need to be able to
6740 * update the inode several times, and we have no idea of knowing how
6741 * many times that will be, so we can't just reserve 1 item for the
6742 * entirety of the opration, so that has to be done seperately as well.
6743 * Then there is the orphan item, which does indeed need to be held on
6744 * to for the whole operation, and we need nobody to touch this reserved
6745 * space except the orphan code.
6747 * So that leaves us with
6749 * 1) root->orphan_block_rsv - for the orphan deletion.
6750 * 2) rsv - for the truncate reservation, which we will steal from the
6751 * transaction reservation.
6752 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6753 * updating the inode.
6755 rsv = btrfs_alloc_block_rsv(root);
6758 rsv->size = min_size;
6761 * 1 for the truncate slack space
6762 * 1 for the orphan item we're going to add
6763 * 1 for the orphan item deletion
6764 * 1 for updating the inode.
6766 trans = btrfs_start_transaction(root, 4);
6767 if (IS_ERR(trans)) {
6768 err = PTR_ERR(trans);
6772 /* Migrate the slack space for the truncate to our reserve */
6773 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6777 ret = btrfs_orphan_add(trans, inode);
6779 btrfs_end_transaction(trans, root);
6784 * setattr is responsible for setting the ordered_data_close flag,
6785 * but that is only tested during the last file release. That
6786 * could happen well after the next commit, leaving a great big
6787 * window where new writes may get lost if someone chooses to write
6788 * to this file after truncating to zero
6790 * The inode doesn't have any dirty data here, and so if we commit
6791 * this is a noop. If someone immediately starts writing to the inode
6792 * it is very likely we'll catch some of their writes in this
6793 * transaction, and the commit will find this file on the ordered
6794 * data list with good things to send down.
6796 * This is a best effort solution, there is still a window where
6797 * using truncate to replace the contents of the file will
6798 * end up with a zero length file after a crash.
6800 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6801 &BTRFS_I(inode)->runtime_flags))
6802 btrfs_add_ordered_operation(trans, root, inode);
6805 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6808 * This can only happen with the original transaction we
6809 * started above, every other time we shouldn't have a
6810 * transaction started yet.
6819 /* Just need the 1 for updating the inode */
6820 trans = btrfs_start_transaction(root, 1);
6821 if (IS_ERR(trans)) {
6822 ret = err = PTR_ERR(trans);
6828 trans->block_rsv = rsv;
6830 ret = btrfs_truncate_inode_items(trans, root, inode,
6832 BTRFS_EXTENT_DATA_KEY);
6833 if (ret != -EAGAIN) {
6838 trans->block_rsv = &root->fs_info->trans_block_rsv;
6839 ret = btrfs_update_inode(trans, root, inode);
6845 nr = trans->blocks_used;
6846 btrfs_end_transaction(trans, root);
6848 btrfs_btree_balance_dirty(root, nr);
6851 if (ret == 0 && inode->i_nlink > 0) {
6852 trans->block_rsv = root->orphan_block_rsv;
6853 ret = btrfs_orphan_del(trans, inode);
6856 } else if (ret && inode->i_nlink > 0) {
6858 * Failed to do the truncate, remove us from the in memory
6861 ret = btrfs_orphan_del(NULL, inode);
6865 trans->block_rsv = &root->fs_info->trans_block_rsv;
6866 ret = btrfs_update_inode(trans, root, inode);
6870 nr = trans->blocks_used;
6871 ret = btrfs_end_transaction(trans, root);
6872 btrfs_btree_balance_dirty(root, nr);
6876 btrfs_free_block_rsv(root, rsv);
6885 * create a new subvolume directory/inode (helper for the ioctl).
6887 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6888 struct btrfs_root *new_root, u64 new_dirid)
6890 struct inode *inode;
6894 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6895 new_dirid, new_dirid,
6896 S_IFDIR | (~current_umask() & S_IRWXUGO),
6899 return PTR_ERR(inode);
6900 inode->i_op = &btrfs_dir_inode_operations;
6901 inode->i_fop = &btrfs_dir_file_operations;
6903 set_nlink(inode, 1);
6904 btrfs_i_size_write(inode, 0);
6906 err = btrfs_update_inode(trans, new_root, inode);
6912 struct inode *btrfs_alloc_inode(struct super_block *sb)
6914 struct btrfs_inode *ei;
6915 struct inode *inode;
6917 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6922 ei->space_info = NULL;
6925 ei->last_sub_trans = 0;
6926 ei->logged_trans = 0;
6927 ei->delalloc_bytes = 0;
6928 ei->disk_i_size = 0;
6931 ei->index_cnt = (u64)-1;
6932 ei->last_unlink_trans = 0;
6934 spin_lock_init(&ei->lock);
6935 ei->outstanding_extents = 0;
6936 ei->reserved_extents = 0;
6938 ei->runtime_flags = 0;
6939 ei->force_compress = BTRFS_COMPRESS_NONE;
6941 ei->delayed_node = NULL;
6943 inode = &ei->vfs_inode;
6944 extent_map_tree_init(&ei->extent_tree);
6945 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6946 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6947 ei->io_tree.track_uptodate = 1;
6948 ei->io_failure_tree.track_uptodate = 1;
6949 mutex_init(&ei->log_mutex);
6950 mutex_init(&ei->delalloc_mutex);
6951 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6952 INIT_LIST_HEAD(&ei->delalloc_inodes);
6953 INIT_LIST_HEAD(&ei->ordered_operations);
6954 RB_CLEAR_NODE(&ei->rb_node);
6959 static void btrfs_i_callback(struct rcu_head *head)
6961 struct inode *inode = container_of(head, struct inode, i_rcu);
6962 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6965 void btrfs_destroy_inode(struct inode *inode)
6967 struct btrfs_ordered_extent *ordered;
6968 struct btrfs_root *root = BTRFS_I(inode)->root;
6970 WARN_ON(!list_empty(&inode->i_dentry));
6971 WARN_ON(inode->i_data.nrpages);
6972 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6973 WARN_ON(BTRFS_I(inode)->reserved_extents);
6974 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6975 WARN_ON(BTRFS_I(inode)->csum_bytes);
6978 * This can happen where we create an inode, but somebody else also
6979 * created the same inode and we need to destroy the one we already
6986 * Make sure we're properly removed from the ordered operation
6990 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6991 spin_lock(&root->fs_info->ordered_extent_lock);
6992 list_del_init(&BTRFS_I(inode)->ordered_operations);
6993 spin_unlock(&root->fs_info->ordered_extent_lock);
6996 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
6997 &BTRFS_I(inode)->runtime_flags)) {
6998 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6999 (unsigned long long)btrfs_ino(inode));
7000 atomic_dec(&root->orphan_inodes);
7004 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7008 printk(KERN_ERR "btrfs found ordered "
7009 "extent %llu %llu on inode cleanup\n",
7010 (unsigned long long)ordered->file_offset,
7011 (unsigned long long)ordered->len);
7012 btrfs_remove_ordered_extent(inode, ordered);
7013 btrfs_put_ordered_extent(ordered);
7014 btrfs_put_ordered_extent(ordered);
7017 inode_tree_del(inode);
7018 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7020 btrfs_remove_delayed_node(inode);
7021 call_rcu(&inode->i_rcu, btrfs_i_callback);
7024 int btrfs_drop_inode(struct inode *inode)
7026 struct btrfs_root *root = BTRFS_I(inode)->root;
7028 if (btrfs_root_refs(&root->root_item) == 0 &&
7029 !btrfs_is_free_space_inode(root, inode))
7032 return generic_drop_inode(inode);
7035 static void init_once(void *foo)
7037 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7039 inode_init_once(&ei->vfs_inode);
7042 void btrfs_destroy_cachep(void)
7044 if (btrfs_inode_cachep)
7045 kmem_cache_destroy(btrfs_inode_cachep);
7046 if (btrfs_trans_handle_cachep)
7047 kmem_cache_destroy(btrfs_trans_handle_cachep);
7048 if (btrfs_transaction_cachep)
7049 kmem_cache_destroy(btrfs_transaction_cachep);
7050 if (btrfs_path_cachep)
7051 kmem_cache_destroy(btrfs_path_cachep);
7052 if (btrfs_free_space_cachep)
7053 kmem_cache_destroy(btrfs_free_space_cachep);
7056 int btrfs_init_cachep(void)
7058 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7059 sizeof(struct btrfs_inode), 0,
7060 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7061 if (!btrfs_inode_cachep)
7064 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7065 sizeof(struct btrfs_trans_handle), 0,
7066 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7067 if (!btrfs_trans_handle_cachep)
7070 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7071 sizeof(struct btrfs_transaction), 0,
7072 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7073 if (!btrfs_transaction_cachep)
7076 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7077 sizeof(struct btrfs_path), 0,
7078 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7079 if (!btrfs_path_cachep)
7082 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7083 sizeof(struct btrfs_free_space), 0,
7084 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7085 if (!btrfs_free_space_cachep)
7090 btrfs_destroy_cachep();
7094 static int btrfs_getattr(struct vfsmount *mnt,
7095 struct dentry *dentry, struct kstat *stat)
7097 struct inode *inode = dentry->d_inode;
7098 u32 blocksize = inode->i_sb->s_blocksize;
7100 generic_fillattr(inode, stat);
7101 stat->dev = BTRFS_I(inode)->root->anon_dev;
7102 stat->blksize = PAGE_CACHE_SIZE;
7103 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7104 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7109 * If a file is moved, it will inherit the cow and compression flags of the new
7112 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7114 struct btrfs_inode *b_dir = BTRFS_I(dir);
7115 struct btrfs_inode *b_inode = BTRFS_I(inode);
7117 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7118 b_inode->flags |= BTRFS_INODE_NODATACOW;
7120 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7122 if (b_dir->flags & BTRFS_INODE_COMPRESS)
7123 b_inode->flags |= BTRFS_INODE_COMPRESS;
7125 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
7128 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7129 struct inode *new_dir, struct dentry *new_dentry)
7131 struct btrfs_trans_handle *trans;
7132 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7133 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7134 struct inode *new_inode = new_dentry->d_inode;
7135 struct inode *old_inode = old_dentry->d_inode;
7136 struct timespec ctime = CURRENT_TIME;
7140 u64 old_ino = btrfs_ino(old_inode);
7142 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7145 /* we only allow rename subvolume link between subvolumes */
7146 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7149 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7150 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7153 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7154 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7157 * we're using rename to replace one file with another.
7158 * and the replacement file is large. Start IO on it now so
7159 * we don't add too much work to the end of the transaction
7161 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7162 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7163 filemap_flush(old_inode->i_mapping);
7165 /* close the racy window with snapshot create/destroy ioctl */
7166 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7167 down_read(&root->fs_info->subvol_sem);
7169 * We want to reserve the absolute worst case amount of items. So if
7170 * both inodes are subvols and we need to unlink them then that would
7171 * require 4 item modifications, but if they are both normal inodes it
7172 * would require 5 item modifications, so we'll assume their normal
7173 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7174 * should cover the worst case number of items we'll modify.
7176 trans = btrfs_start_transaction(root, 20);
7177 if (IS_ERR(trans)) {
7178 ret = PTR_ERR(trans);
7183 btrfs_record_root_in_trans(trans, dest);
7185 ret = btrfs_set_inode_index(new_dir, &index);
7189 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7190 /* force full log commit if subvolume involved. */
7191 root->fs_info->last_trans_log_full_commit = trans->transid;
7193 ret = btrfs_insert_inode_ref(trans, dest,
7194 new_dentry->d_name.name,
7195 new_dentry->d_name.len,
7197 btrfs_ino(new_dir), index);
7201 * this is an ugly little race, but the rename is required
7202 * to make sure that if we crash, the inode is either at the
7203 * old name or the new one. pinning the log transaction lets
7204 * us make sure we don't allow a log commit to come in after
7205 * we unlink the name but before we add the new name back in.
7207 btrfs_pin_log_trans(root);
7210 * make sure the inode gets flushed if it is replacing
7213 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7214 btrfs_add_ordered_operation(trans, root, old_inode);
7216 inode_inc_iversion(old_dir);
7217 inode_inc_iversion(new_dir);
7218 inode_inc_iversion(old_inode);
7219 old_dir->i_ctime = old_dir->i_mtime = ctime;
7220 new_dir->i_ctime = new_dir->i_mtime = ctime;
7221 old_inode->i_ctime = ctime;
7223 if (old_dentry->d_parent != new_dentry->d_parent)
7224 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7226 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7227 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7228 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7229 old_dentry->d_name.name,
7230 old_dentry->d_name.len);
7232 ret = __btrfs_unlink_inode(trans, root, old_dir,
7233 old_dentry->d_inode,
7234 old_dentry->d_name.name,
7235 old_dentry->d_name.len);
7237 ret = btrfs_update_inode(trans, root, old_inode);
7240 btrfs_abort_transaction(trans, root, ret);
7245 inode_inc_iversion(new_inode);
7246 new_inode->i_ctime = CURRENT_TIME;
7247 if (unlikely(btrfs_ino(new_inode) ==
7248 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7249 root_objectid = BTRFS_I(new_inode)->location.objectid;
7250 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7252 new_dentry->d_name.name,
7253 new_dentry->d_name.len);
7254 BUG_ON(new_inode->i_nlink == 0);
7256 ret = btrfs_unlink_inode(trans, dest, new_dir,
7257 new_dentry->d_inode,
7258 new_dentry->d_name.name,
7259 new_dentry->d_name.len);
7261 if (!ret && new_inode->i_nlink == 0) {
7262 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7266 btrfs_abort_transaction(trans, root, ret);
7271 fixup_inode_flags(new_dir, old_inode);
7273 ret = btrfs_add_link(trans, new_dir, old_inode,
7274 new_dentry->d_name.name,
7275 new_dentry->d_name.len, 0, index);
7277 btrfs_abort_transaction(trans, root, ret);
7281 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7282 struct dentry *parent = new_dentry->d_parent;
7283 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7284 btrfs_end_log_trans(root);
7287 btrfs_end_transaction(trans, root);
7289 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7290 up_read(&root->fs_info->subvol_sem);
7296 * some fairly slow code that needs optimization. This walks the list
7297 * of all the inodes with pending delalloc and forces them to disk.
7299 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7301 struct list_head *head = &root->fs_info->delalloc_inodes;
7302 struct btrfs_inode *binode;
7303 struct inode *inode;
7305 if (root->fs_info->sb->s_flags & MS_RDONLY)
7308 spin_lock(&root->fs_info->delalloc_lock);
7309 while (!list_empty(head)) {
7310 binode = list_entry(head->next, struct btrfs_inode,
7312 inode = igrab(&binode->vfs_inode);
7314 list_del_init(&binode->delalloc_inodes);
7315 spin_unlock(&root->fs_info->delalloc_lock);
7317 filemap_flush(inode->i_mapping);
7319 btrfs_add_delayed_iput(inode);
7324 spin_lock(&root->fs_info->delalloc_lock);
7326 spin_unlock(&root->fs_info->delalloc_lock);
7328 /* the filemap_flush will queue IO into the worker threads, but
7329 * we have to make sure the IO is actually started and that
7330 * ordered extents get created before we return
7332 atomic_inc(&root->fs_info->async_submit_draining);
7333 while (atomic_read(&root->fs_info->nr_async_submits) ||
7334 atomic_read(&root->fs_info->async_delalloc_pages)) {
7335 wait_event(root->fs_info->async_submit_wait,
7336 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7337 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7339 atomic_dec(&root->fs_info->async_submit_draining);
7343 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7344 const char *symname)
7346 struct btrfs_trans_handle *trans;
7347 struct btrfs_root *root = BTRFS_I(dir)->root;
7348 struct btrfs_path *path;
7349 struct btrfs_key key;
7350 struct inode *inode = NULL;
7358 struct btrfs_file_extent_item *ei;
7359 struct extent_buffer *leaf;
7360 unsigned long nr = 0;
7362 name_len = strlen(symname) + 1;
7363 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7364 return -ENAMETOOLONG;
7367 * 2 items for inode item and ref
7368 * 2 items for dir items
7369 * 1 item for xattr if selinux is on
7371 trans = btrfs_start_transaction(root, 5);
7373 return PTR_ERR(trans);
7375 err = btrfs_find_free_ino(root, &objectid);
7379 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7380 dentry->d_name.len, btrfs_ino(dir), objectid,
7381 S_IFLNK|S_IRWXUGO, &index);
7382 if (IS_ERR(inode)) {
7383 err = PTR_ERR(inode);
7387 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7394 * If the active LSM wants to access the inode during
7395 * d_instantiate it needs these. Smack checks to see
7396 * if the filesystem supports xattrs by looking at the
7399 inode->i_fop = &btrfs_file_operations;
7400 inode->i_op = &btrfs_file_inode_operations;
7402 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7406 inode->i_mapping->a_ops = &btrfs_aops;
7407 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7408 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7413 path = btrfs_alloc_path();
7419 key.objectid = btrfs_ino(inode);
7421 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7422 datasize = btrfs_file_extent_calc_inline_size(name_len);
7423 err = btrfs_insert_empty_item(trans, root, path, &key,
7427 btrfs_free_path(path);
7430 leaf = path->nodes[0];
7431 ei = btrfs_item_ptr(leaf, path->slots[0],
7432 struct btrfs_file_extent_item);
7433 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7434 btrfs_set_file_extent_type(leaf, ei,
7435 BTRFS_FILE_EXTENT_INLINE);
7436 btrfs_set_file_extent_encryption(leaf, ei, 0);
7437 btrfs_set_file_extent_compression(leaf, ei, 0);
7438 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7439 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7441 ptr = btrfs_file_extent_inline_start(ei);
7442 write_extent_buffer(leaf, symname, ptr, name_len);
7443 btrfs_mark_buffer_dirty(leaf);
7444 btrfs_free_path(path);
7446 inode->i_op = &btrfs_symlink_inode_operations;
7447 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7448 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7449 inode_set_bytes(inode, name_len);
7450 btrfs_i_size_write(inode, name_len - 1);
7451 err = btrfs_update_inode(trans, root, inode);
7457 d_instantiate(dentry, inode);
7458 nr = trans->blocks_used;
7459 btrfs_end_transaction(trans, root);
7461 inode_dec_link_count(inode);
7464 btrfs_btree_balance_dirty(root, nr);
7468 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7469 u64 start, u64 num_bytes, u64 min_size,
7470 loff_t actual_len, u64 *alloc_hint,
7471 struct btrfs_trans_handle *trans)
7473 struct btrfs_root *root = BTRFS_I(inode)->root;
7474 struct btrfs_key ins;
7475 u64 cur_offset = start;
7478 bool own_trans = true;
7482 while (num_bytes > 0) {
7484 trans = btrfs_start_transaction(root, 3);
7485 if (IS_ERR(trans)) {
7486 ret = PTR_ERR(trans);
7491 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7492 0, *alloc_hint, &ins, 1);
7495 btrfs_end_transaction(trans, root);
7499 ret = insert_reserved_file_extent(trans, inode,
7500 cur_offset, ins.objectid,
7501 ins.offset, ins.offset,
7502 ins.offset, 0, 0, 0,
7503 BTRFS_FILE_EXTENT_PREALLOC);
7505 btrfs_abort_transaction(trans, root, ret);
7507 btrfs_end_transaction(trans, root);
7510 btrfs_drop_extent_cache(inode, cur_offset,
7511 cur_offset + ins.offset -1, 0);
7513 num_bytes -= ins.offset;
7514 cur_offset += ins.offset;
7515 *alloc_hint = ins.objectid + ins.offset;
7517 inode_inc_iversion(inode);
7518 inode->i_ctime = CURRENT_TIME;
7519 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7520 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7521 (actual_len > inode->i_size) &&
7522 (cur_offset > inode->i_size)) {
7523 if (cur_offset > actual_len)
7524 i_size = actual_len;
7526 i_size = cur_offset;
7527 i_size_write(inode, i_size);
7528 btrfs_ordered_update_i_size(inode, i_size, NULL);
7531 ret = btrfs_update_inode(trans, root, inode);
7534 btrfs_abort_transaction(trans, root, ret);
7536 btrfs_end_transaction(trans, root);
7541 btrfs_end_transaction(trans, root);
7546 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7547 u64 start, u64 num_bytes, u64 min_size,
7548 loff_t actual_len, u64 *alloc_hint)
7550 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7551 min_size, actual_len, alloc_hint,
7555 int btrfs_prealloc_file_range_trans(struct inode *inode,
7556 struct btrfs_trans_handle *trans, int mode,
7557 u64 start, u64 num_bytes, u64 min_size,
7558 loff_t actual_len, u64 *alloc_hint)
7560 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7561 min_size, actual_len, alloc_hint, trans);
7564 static int btrfs_set_page_dirty(struct page *page)
7566 return __set_page_dirty_nobuffers(page);
7569 static int btrfs_permission(struct inode *inode, int mask)
7571 struct btrfs_root *root = BTRFS_I(inode)->root;
7572 umode_t mode = inode->i_mode;
7574 if (mask & MAY_WRITE &&
7575 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7576 if (btrfs_root_readonly(root))
7578 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7581 return generic_permission(inode, mask);
7584 static const struct inode_operations btrfs_dir_inode_operations = {
7585 .getattr = btrfs_getattr,
7586 .lookup = btrfs_lookup,
7587 .create = btrfs_create,
7588 .unlink = btrfs_unlink,
7590 .mkdir = btrfs_mkdir,
7591 .rmdir = btrfs_rmdir,
7592 .rename = btrfs_rename,
7593 .symlink = btrfs_symlink,
7594 .setattr = btrfs_setattr,
7595 .mknod = btrfs_mknod,
7596 .setxattr = btrfs_setxattr,
7597 .getxattr = btrfs_getxattr,
7598 .listxattr = btrfs_listxattr,
7599 .removexattr = btrfs_removexattr,
7600 .permission = btrfs_permission,
7601 .get_acl = btrfs_get_acl,
7603 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7604 .lookup = btrfs_lookup,
7605 .permission = btrfs_permission,
7606 .get_acl = btrfs_get_acl,
7609 static const struct file_operations btrfs_dir_file_operations = {
7610 .llseek = generic_file_llseek,
7611 .read = generic_read_dir,
7612 .readdir = btrfs_real_readdir,
7613 .unlocked_ioctl = btrfs_ioctl,
7614 #ifdef CONFIG_COMPAT
7615 .compat_ioctl = btrfs_ioctl,
7617 .release = btrfs_release_file,
7618 .fsync = btrfs_sync_file,
7621 static struct extent_io_ops btrfs_extent_io_ops = {
7622 .fill_delalloc = run_delalloc_range,
7623 .submit_bio_hook = btrfs_submit_bio_hook,
7624 .merge_bio_hook = btrfs_merge_bio_hook,
7625 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7626 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7627 .writepage_start_hook = btrfs_writepage_start_hook,
7628 .set_bit_hook = btrfs_set_bit_hook,
7629 .clear_bit_hook = btrfs_clear_bit_hook,
7630 .merge_extent_hook = btrfs_merge_extent_hook,
7631 .split_extent_hook = btrfs_split_extent_hook,
7635 * btrfs doesn't support the bmap operation because swapfiles
7636 * use bmap to make a mapping of extents in the file. They assume
7637 * these extents won't change over the life of the file and they
7638 * use the bmap result to do IO directly to the drive.
7640 * the btrfs bmap call would return logical addresses that aren't
7641 * suitable for IO and they also will change frequently as COW
7642 * operations happen. So, swapfile + btrfs == corruption.
7644 * For now we're avoiding this by dropping bmap.
7646 static const struct address_space_operations btrfs_aops = {
7647 .readpage = btrfs_readpage,
7648 .writepage = btrfs_writepage,
7649 .writepages = btrfs_writepages,
7650 .readpages = btrfs_readpages,
7651 .direct_IO = btrfs_direct_IO,
7652 .invalidatepage = btrfs_invalidatepage,
7653 .releasepage = btrfs_releasepage,
7654 .set_page_dirty = btrfs_set_page_dirty,
7655 .error_remove_page = generic_error_remove_page,
7658 static const struct address_space_operations btrfs_symlink_aops = {
7659 .readpage = btrfs_readpage,
7660 .writepage = btrfs_writepage,
7661 .invalidatepage = btrfs_invalidatepage,
7662 .releasepage = btrfs_releasepage,
7665 static const struct inode_operations btrfs_file_inode_operations = {
7666 .getattr = btrfs_getattr,
7667 .setattr = btrfs_setattr,
7668 .setxattr = btrfs_setxattr,
7669 .getxattr = btrfs_getxattr,
7670 .listxattr = btrfs_listxattr,
7671 .removexattr = btrfs_removexattr,
7672 .permission = btrfs_permission,
7673 .fiemap = btrfs_fiemap,
7674 .get_acl = btrfs_get_acl,
7676 static const struct inode_operations btrfs_special_inode_operations = {
7677 .getattr = btrfs_getattr,
7678 .setattr = btrfs_setattr,
7679 .permission = btrfs_permission,
7680 .setxattr = btrfs_setxattr,
7681 .getxattr = btrfs_getxattr,
7682 .listxattr = btrfs_listxattr,
7683 .removexattr = btrfs_removexattr,
7684 .get_acl = btrfs_get_acl,
7686 static const struct inode_operations btrfs_symlink_inode_operations = {
7687 .readlink = generic_readlink,
7688 .follow_link = page_follow_link_light,
7689 .put_link = page_put_link,
7690 .getattr = btrfs_getattr,
7691 .setattr = btrfs_setattr,
7692 .permission = btrfs_permission,
7693 .setxattr = btrfs_setxattr,
7694 .getxattr = btrfs_getxattr,
7695 .listxattr = btrfs_listxattr,
7696 .removexattr = btrfs_removexattr,
7697 .get_acl = btrfs_get_acl,
7700 const struct dentry_operations btrfs_dentry_operations = {
7701 .d_delete = btrfs_dentry_delete,
7702 .d_release = btrfs_dentry_release,