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 the flusher thread sent them
330 static noinline int compress_file_range(struct inode *inode,
331 struct page *locked_page,
333 struct async_cow *async_cow,
336 struct btrfs_root *root = BTRFS_I(inode)->root;
337 struct btrfs_trans_handle *trans;
339 u64 blocksize = root->sectorsize;
341 u64 isize = i_size_read(inode);
343 struct page **pages = NULL;
344 unsigned long nr_pages;
345 unsigned long nr_pages_ret = 0;
346 unsigned long total_compressed = 0;
347 unsigned long total_in = 0;
348 unsigned long max_compressed = 128 * 1024;
349 unsigned long max_uncompressed = 128 * 1024;
352 int compress_type = root->fs_info->compress_type;
354 /* if this is a small write inside eof, kick off a defrag */
355 if ((end - start + 1) < 16 * 1024 &&
356 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
357 btrfs_add_inode_defrag(NULL, inode);
359 actual_end = min_t(u64, isize, end + 1);
362 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
363 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
366 * we don't want to send crud past the end of i_size through
367 * compression, that's just a waste of CPU time. So, if the
368 * end of the file is before the start of our current
369 * requested range of bytes, we bail out to the uncompressed
370 * cleanup code that can deal with all of this.
372 * It isn't really the fastest way to fix things, but this is a
373 * very uncommon corner.
375 if (actual_end <= start)
376 goto cleanup_and_bail_uncompressed;
378 total_compressed = actual_end - start;
380 /* we want to make sure that amount of ram required to uncompress
381 * an extent is reasonable, so we limit the total size in ram
382 * of a compressed extent to 128k. This is a crucial number
383 * because it also controls how easily we can spread reads across
384 * cpus for decompression.
386 * We also want to make sure the amount of IO required to do
387 * a random read is reasonably small, so we limit the size of
388 * a compressed extent to 128k.
390 total_compressed = min(total_compressed, max_uncompressed);
391 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
392 num_bytes = max(blocksize, num_bytes);
397 * we do compression for mount -o compress and when the
398 * inode has not been flagged as nocompress. This flag can
399 * change at any time if we discover bad compression ratios.
401 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
402 (btrfs_test_opt(root, COMPRESS) ||
403 (BTRFS_I(inode)->force_compress) ||
404 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
406 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
408 /* just bail out to the uncompressed code */
412 if (BTRFS_I(inode)->force_compress)
413 compress_type = BTRFS_I(inode)->force_compress;
415 ret = btrfs_compress_pages(compress_type,
416 inode->i_mapping, start,
417 total_compressed, pages,
418 nr_pages, &nr_pages_ret,
424 unsigned long offset = total_compressed &
425 (PAGE_CACHE_SIZE - 1);
426 struct page *page = pages[nr_pages_ret - 1];
429 /* zero the tail end of the last page, we might be
430 * sending it down to disk
433 kaddr = kmap_atomic(page);
434 memset(kaddr + offset, 0,
435 PAGE_CACHE_SIZE - offset);
436 kunmap_atomic(kaddr);
443 trans = btrfs_join_transaction(root);
445 ret = PTR_ERR(trans);
447 goto cleanup_and_out;
449 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
451 /* lets try to make an inline extent */
452 if (ret || total_in < (actual_end - start)) {
453 /* we didn't compress the entire range, try
454 * to make an uncompressed inline extent.
456 ret = cow_file_range_inline(trans, root, inode,
457 start, end, 0, 0, NULL);
459 /* try making a compressed inline extent */
460 ret = cow_file_range_inline(trans, root, inode,
463 compress_type, pages);
467 * inline extent creation worked or returned error,
468 * we don't need to create any more async work items.
469 * Unlock and free up our temp pages.
471 extent_clear_unlock_delalloc(inode,
472 &BTRFS_I(inode)->io_tree,
474 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
475 EXTENT_CLEAR_DELALLOC |
476 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
478 btrfs_end_transaction(trans, root);
481 btrfs_end_transaction(trans, root);
486 * we aren't doing an inline extent round the compressed size
487 * up to a block size boundary so the allocator does sane
490 total_compressed = (total_compressed + blocksize - 1) &
494 * one last check to make sure the compression is really a
495 * win, compare the page count read with the blocks on disk
497 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
498 ~(PAGE_CACHE_SIZE - 1);
499 if (total_compressed >= total_in) {
502 num_bytes = total_in;
505 if (!will_compress && pages) {
507 * the compression code ran but failed to make things smaller,
508 * free any pages it allocated and our page pointer array
510 for (i = 0; i < nr_pages_ret; i++) {
511 WARN_ON(pages[i]->mapping);
512 page_cache_release(pages[i]);
516 total_compressed = 0;
519 /* flag the file so we don't compress in the future */
520 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
521 !(BTRFS_I(inode)->force_compress)) {
522 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
528 /* the async work queues will take care of doing actual
529 * allocation on disk for these compressed pages,
530 * and will submit them to the elevator.
532 add_async_extent(async_cow, start, num_bytes,
533 total_compressed, pages, nr_pages_ret,
536 if (start + num_bytes < end) {
543 cleanup_and_bail_uncompressed:
545 * No compression, but we still need to write the pages in
546 * the file we've been given so far. redirty the locked
547 * page if it corresponds to our extent and set things up
548 * for the async work queue to run cow_file_range to do
549 * the normal delalloc dance
551 if (page_offset(locked_page) >= start &&
552 page_offset(locked_page) <= end) {
553 __set_page_dirty_nobuffers(locked_page);
554 /* unlocked later on in the async handlers */
556 add_async_extent(async_cow, start, end - start + 1,
557 0, NULL, 0, BTRFS_COMPRESS_NONE);
565 for (i = 0; i < nr_pages_ret; i++) {
566 WARN_ON(pages[i]->mapping);
567 page_cache_release(pages[i]);
574 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
576 EXTENT_CLEAR_UNLOCK_PAGE |
578 EXTENT_CLEAR_DELALLOC |
579 EXTENT_SET_WRITEBACK |
580 EXTENT_END_WRITEBACK);
581 if (!trans || IS_ERR(trans))
582 btrfs_error(root->fs_info, ret, "Failed to join transaction");
584 btrfs_abort_transaction(trans, root, ret);
589 * phase two of compressed writeback. This is the ordered portion
590 * of the code, which only gets called in the order the work was
591 * queued. We walk all the async extents created by compress_file_range
592 * and send them down to the disk.
594 static noinline int submit_compressed_extents(struct inode *inode,
595 struct async_cow *async_cow)
597 struct async_extent *async_extent;
599 struct btrfs_trans_handle *trans;
600 struct btrfs_key ins;
601 struct extent_map *em;
602 struct btrfs_root *root = BTRFS_I(inode)->root;
603 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
604 struct extent_io_tree *io_tree;
607 if (list_empty(&async_cow->extents))
611 while (!list_empty(&async_cow->extents)) {
612 async_extent = list_entry(async_cow->extents.next,
613 struct async_extent, list);
614 list_del(&async_extent->list);
616 io_tree = &BTRFS_I(inode)->io_tree;
619 /* did the compression code fall back to uncompressed IO? */
620 if (!async_extent->pages) {
621 int page_started = 0;
622 unsigned long nr_written = 0;
624 lock_extent(io_tree, async_extent->start,
625 async_extent->start +
626 async_extent->ram_size - 1);
628 /* allocate blocks */
629 ret = cow_file_range(inode, async_cow->locked_page,
631 async_extent->start +
632 async_extent->ram_size - 1,
633 &page_started, &nr_written, 0);
638 * if page_started, cow_file_range inserted an
639 * inline extent and took care of all the unlocking
640 * and IO for us. Otherwise, we need to submit
641 * all those pages down to the drive.
643 if (!page_started && !ret)
644 extent_write_locked_range(io_tree,
645 inode, async_extent->start,
646 async_extent->start +
647 async_extent->ram_size - 1,
655 lock_extent(io_tree, async_extent->start,
656 async_extent->start + async_extent->ram_size - 1);
658 trans = btrfs_join_transaction(root);
660 ret = PTR_ERR(trans);
662 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
663 ret = btrfs_reserve_extent(trans, root,
664 async_extent->compressed_size,
665 async_extent->compressed_size,
666 0, alloc_hint, &ins, 1);
668 btrfs_abort_transaction(trans, root, ret);
669 btrfs_end_transaction(trans, root);
674 for (i = 0; i < async_extent->nr_pages; i++) {
675 WARN_ON(async_extent->pages[i]->mapping);
676 page_cache_release(async_extent->pages[i]);
678 kfree(async_extent->pages);
679 async_extent->nr_pages = 0;
680 async_extent->pages = NULL;
681 unlock_extent(io_tree, async_extent->start,
682 async_extent->start +
683 async_extent->ram_size - 1);
686 goto out_free; /* JDM: Requeue? */
690 * here we're doing allocation and writeback of the
693 btrfs_drop_extent_cache(inode, async_extent->start,
694 async_extent->start +
695 async_extent->ram_size - 1, 0);
697 em = alloc_extent_map();
698 BUG_ON(!em); /* -ENOMEM */
699 em->start = async_extent->start;
700 em->len = async_extent->ram_size;
701 em->orig_start = em->start;
703 em->block_start = ins.objectid;
704 em->block_len = ins.offset;
705 em->bdev = root->fs_info->fs_devices->latest_bdev;
706 em->compress_type = async_extent->compress_type;
707 set_bit(EXTENT_FLAG_PINNED, &em->flags);
708 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
711 write_lock(&em_tree->lock);
712 ret = add_extent_mapping(em_tree, em);
713 write_unlock(&em_tree->lock);
714 if (ret != -EEXIST) {
718 btrfs_drop_extent_cache(inode, async_extent->start,
719 async_extent->start +
720 async_extent->ram_size - 1, 0);
723 ret = btrfs_add_ordered_extent_compress(inode,
726 async_extent->ram_size,
728 BTRFS_ORDERED_COMPRESSED,
729 async_extent->compress_type);
730 BUG_ON(ret); /* -ENOMEM */
733 * clear dirty, set writeback and unlock the pages.
735 extent_clear_unlock_delalloc(inode,
736 &BTRFS_I(inode)->io_tree,
738 async_extent->start +
739 async_extent->ram_size - 1,
740 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
741 EXTENT_CLEAR_UNLOCK |
742 EXTENT_CLEAR_DELALLOC |
743 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
745 ret = btrfs_submit_compressed_write(inode,
747 async_extent->ram_size,
749 ins.offset, async_extent->pages,
750 async_extent->nr_pages);
752 BUG_ON(ret); /* -ENOMEM */
753 alloc_hint = ins.objectid + ins.offset;
765 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
768 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
769 struct extent_map *em;
772 read_lock(&em_tree->lock);
773 em = search_extent_mapping(em_tree, start, num_bytes);
776 * if block start isn't an actual block number then find the
777 * first block in this inode and use that as a hint. If that
778 * block is also bogus then just don't worry about it.
780 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
782 em = search_extent_mapping(em_tree, 0, 0);
783 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
784 alloc_hint = em->block_start;
788 alloc_hint = em->block_start;
792 read_unlock(&em_tree->lock);
798 * when extent_io.c finds a delayed allocation range in the file,
799 * the call backs end up in this code. The basic idea is to
800 * allocate extents on disk for the range, and create ordered data structs
801 * in ram to track those extents.
803 * locked_page is the page that writepage had locked already. We use
804 * it to make sure we don't do extra locks or unlocks.
806 * *page_started is set to one if we unlock locked_page and do everything
807 * required to start IO on it. It may be clean and already done with
810 static noinline int cow_file_range(struct inode *inode,
811 struct page *locked_page,
812 u64 start, u64 end, int *page_started,
813 unsigned long *nr_written,
816 struct btrfs_root *root = BTRFS_I(inode)->root;
817 struct btrfs_trans_handle *trans;
820 unsigned long ram_size;
823 u64 blocksize = root->sectorsize;
824 struct btrfs_key ins;
825 struct extent_map *em;
826 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
829 BUG_ON(btrfs_is_free_space_inode(inode));
830 trans = btrfs_join_transaction(root);
832 extent_clear_unlock_delalloc(inode,
833 &BTRFS_I(inode)->io_tree,
834 start, end, locked_page,
835 EXTENT_CLEAR_UNLOCK_PAGE |
836 EXTENT_CLEAR_UNLOCK |
837 EXTENT_CLEAR_DELALLOC |
839 EXTENT_SET_WRITEBACK |
840 EXTENT_END_WRITEBACK);
841 return PTR_ERR(trans);
843 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
845 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
846 num_bytes = max(blocksize, num_bytes);
847 disk_num_bytes = num_bytes;
850 /* if this is a small write inside eof, kick off defrag */
851 if (num_bytes < 64 * 1024 &&
852 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
853 btrfs_add_inode_defrag(trans, inode);
856 /* lets try to make an inline extent */
857 ret = cow_file_range_inline(trans, root, inode,
858 start, end, 0, 0, NULL);
860 extent_clear_unlock_delalloc(inode,
861 &BTRFS_I(inode)->io_tree,
863 EXTENT_CLEAR_UNLOCK_PAGE |
864 EXTENT_CLEAR_UNLOCK |
865 EXTENT_CLEAR_DELALLOC |
867 EXTENT_SET_WRITEBACK |
868 EXTENT_END_WRITEBACK);
870 *nr_written = *nr_written +
871 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
874 } else if (ret < 0) {
875 btrfs_abort_transaction(trans, root, ret);
880 BUG_ON(disk_num_bytes >
881 btrfs_super_total_bytes(root->fs_info->super_copy));
883 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
884 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
886 while (disk_num_bytes > 0) {
889 cur_alloc_size = disk_num_bytes;
890 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
891 root->sectorsize, 0, alloc_hint,
894 btrfs_abort_transaction(trans, root, ret);
898 em = alloc_extent_map();
899 BUG_ON(!em); /* -ENOMEM */
901 em->orig_start = em->start;
902 ram_size = ins.offset;
903 em->len = ins.offset;
905 em->block_start = ins.objectid;
906 em->block_len = ins.offset;
907 em->bdev = root->fs_info->fs_devices->latest_bdev;
908 set_bit(EXTENT_FLAG_PINNED, &em->flags);
911 write_lock(&em_tree->lock);
912 ret = add_extent_mapping(em_tree, em);
913 write_unlock(&em_tree->lock);
914 if (ret != -EEXIST) {
918 btrfs_drop_extent_cache(inode, start,
919 start + ram_size - 1, 0);
922 cur_alloc_size = ins.offset;
923 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
924 ram_size, cur_alloc_size, 0);
925 BUG_ON(ret); /* -ENOMEM */
927 if (root->root_key.objectid ==
928 BTRFS_DATA_RELOC_TREE_OBJECTID) {
929 ret = btrfs_reloc_clone_csums(inode, start,
932 btrfs_abort_transaction(trans, root, ret);
937 if (disk_num_bytes < cur_alloc_size)
940 /* we're not doing compressed IO, don't unlock the first
941 * page (which the caller expects to stay locked), don't
942 * clear any dirty bits and don't set any writeback bits
944 * Do set the Private2 bit so we know this page was properly
945 * setup for writepage
947 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
948 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
951 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
952 start, start + ram_size - 1,
954 disk_num_bytes -= cur_alloc_size;
955 num_bytes -= cur_alloc_size;
956 alloc_hint = ins.objectid + ins.offset;
957 start += cur_alloc_size;
961 btrfs_end_transaction(trans, root);
965 extent_clear_unlock_delalloc(inode,
966 &BTRFS_I(inode)->io_tree,
967 start, end, locked_page,
968 EXTENT_CLEAR_UNLOCK_PAGE |
969 EXTENT_CLEAR_UNLOCK |
970 EXTENT_CLEAR_DELALLOC |
972 EXTENT_SET_WRITEBACK |
973 EXTENT_END_WRITEBACK);
979 * work queue call back to started compression on a file and pages
981 static noinline void async_cow_start(struct btrfs_work *work)
983 struct async_cow *async_cow;
985 async_cow = container_of(work, struct async_cow, work);
987 compress_file_range(async_cow->inode, async_cow->locked_page,
988 async_cow->start, async_cow->end, async_cow,
990 if (num_added == 0) {
991 btrfs_add_delayed_iput(async_cow->inode);
992 async_cow->inode = NULL;
997 * work queue call back to submit previously compressed pages
999 static noinline void async_cow_submit(struct btrfs_work *work)
1001 struct async_cow *async_cow;
1002 struct btrfs_root *root;
1003 unsigned long nr_pages;
1005 async_cow = container_of(work, struct async_cow, work);
1007 root = async_cow->root;
1008 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1011 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1013 if (atomic_read(&root->fs_info->async_delalloc_pages) <
1015 waitqueue_active(&root->fs_info->async_submit_wait))
1016 wake_up(&root->fs_info->async_submit_wait);
1018 if (async_cow->inode)
1019 submit_compressed_extents(async_cow->inode, async_cow);
1022 static noinline void async_cow_free(struct btrfs_work *work)
1024 struct async_cow *async_cow;
1025 async_cow = container_of(work, struct async_cow, work);
1026 if (async_cow->inode)
1027 btrfs_add_delayed_iput(async_cow->inode);
1031 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1032 u64 start, u64 end, int *page_started,
1033 unsigned long *nr_written)
1035 struct async_cow *async_cow;
1036 struct btrfs_root *root = BTRFS_I(inode)->root;
1037 unsigned long nr_pages;
1039 int limit = 10 * 1024 * 1024;
1041 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1042 1, 0, NULL, GFP_NOFS);
1043 while (start < end) {
1044 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1045 BUG_ON(!async_cow); /* -ENOMEM */
1046 async_cow->inode = igrab(inode);
1047 async_cow->root = root;
1048 async_cow->locked_page = locked_page;
1049 async_cow->start = start;
1051 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1054 cur_end = min(end, start + 512 * 1024 - 1);
1056 async_cow->end = cur_end;
1057 INIT_LIST_HEAD(&async_cow->extents);
1059 async_cow->work.func = async_cow_start;
1060 async_cow->work.ordered_func = async_cow_submit;
1061 async_cow->work.ordered_free = async_cow_free;
1062 async_cow->work.flags = 0;
1064 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1066 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1068 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1071 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1072 wait_event(root->fs_info->async_submit_wait,
1073 (atomic_read(&root->fs_info->async_delalloc_pages) <
1077 while (atomic_read(&root->fs_info->async_submit_draining) &&
1078 atomic_read(&root->fs_info->async_delalloc_pages)) {
1079 wait_event(root->fs_info->async_submit_wait,
1080 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1084 *nr_written += nr_pages;
1085 start = cur_end + 1;
1091 static noinline int csum_exist_in_range(struct btrfs_root *root,
1092 u64 bytenr, u64 num_bytes)
1095 struct btrfs_ordered_sum *sums;
1098 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1099 bytenr + num_bytes - 1, &list, 0);
1100 if (ret == 0 && list_empty(&list))
1103 while (!list_empty(&list)) {
1104 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1105 list_del(&sums->list);
1112 * when nowcow writeback call back. This checks for snapshots or COW copies
1113 * of the extents that exist in the file, and COWs the file as required.
1115 * If no cow copies or snapshots exist, we write directly to the existing
1118 static noinline int run_delalloc_nocow(struct inode *inode,
1119 struct page *locked_page,
1120 u64 start, u64 end, int *page_started, int force,
1121 unsigned long *nr_written)
1123 struct btrfs_root *root = BTRFS_I(inode)->root;
1124 struct btrfs_trans_handle *trans;
1125 struct extent_buffer *leaf;
1126 struct btrfs_path *path;
1127 struct btrfs_file_extent_item *fi;
1128 struct btrfs_key found_key;
1141 u64 ino = btrfs_ino(inode);
1143 path = btrfs_alloc_path();
1145 extent_clear_unlock_delalloc(inode,
1146 &BTRFS_I(inode)->io_tree,
1147 start, end, locked_page,
1148 EXTENT_CLEAR_UNLOCK_PAGE |
1149 EXTENT_CLEAR_UNLOCK |
1150 EXTENT_CLEAR_DELALLOC |
1151 EXTENT_CLEAR_DIRTY |
1152 EXTENT_SET_WRITEBACK |
1153 EXTENT_END_WRITEBACK);
1157 nolock = btrfs_is_free_space_inode(inode);
1160 trans = btrfs_join_transaction_nolock(root);
1162 trans = btrfs_join_transaction(root);
1164 if (IS_ERR(trans)) {
1165 extent_clear_unlock_delalloc(inode,
1166 &BTRFS_I(inode)->io_tree,
1167 start, end, locked_page,
1168 EXTENT_CLEAR_UNLOCK_PAGE |
1169 EXTENT_CLEAR_UNLOCK |
1170 EXTENT_CLEAR_DELALLOC |
1171 EXTENT_CLEAR_DIRTY |
1172 EXTENT_SET_WRITEBACK |
1173 EXTENT_END_WRITEBACK);
1174 btrfs_free_path(path);
1175 return PTR_ERR(trans);
1178 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1180 cow_start = (u64)-1;
1183 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1186 btrfs_abort_transaction(trans, root, ret);
1189 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1190 leaf = path->nodes[0];
1191 btrfs_item_key_to_cpu(leaf, &found_key,
1192 path->slots[0] - 1);
1193 if (found_key.objectid == ino &&
1194 found_key.type == BTRFS_EXTENT_DATA_KEY)
1199 leaf = path->nodes[0];
1200 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1201 ret = btrfs_next_leaf(root, path);
1203 btrfs_abort_transaction(trans, root, ret);
1208 leaf = path->nodes[0];
1214 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1216 if (found_key.objectid > ino ||
1217 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1218 found_key.offset > end)
1221 if (found_key.offset > cur_offset) {
1222 extent_end = found_key.offset;
1227 fi = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_file_extent_item);
1229 extent_type = btrfs_file_extent_type(leaf, fi);
1231 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1232 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1233 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1234 extent_offset = btrfs_file_extent_offset(leaf, fi);
1235 extent_end = found_key.offset +
1236 btrfs_file_extent_num_bytes(leaf, fi);
1237 if (extent_end <= start) {
1241 if (disk_bytenr == 0)
1243 if (btrfs_file_extent_compression(leaf, fi) ||
1244 btrfs_file_extent_encryption(leaf, fi) ||
1245 btrfs_file_extent_other_encoding(leaf, fi))
1247 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1249 if (btrfs_extent_readonly(root, disk_bytenr))
1251 if (btrfs_cross_ref_exist(trans, root, ino,
1253 extent_offset, disk_bytenr))
1255 disk_bytenr += extent_offset;
1256 disk_bytenr += cur_offset - found_key.offset;
1257 num_bytes = min(end + 1, extent_end) - cur_offset;
1259 * force cow if csum exists in the range.
1260 * this ensure that csum for a given extent are
1261 * either valid or do not exist.
1263 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1266 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1267 extent_end = found_key.offset +
1268 btrfs_file_extent_inline_len(leaf, fi);
1269 extent_end = ALIGN(extent_end, root->sectorsize);
1274 if (extent_end <= start) {
1279 if (cow_start == (u64)-1)
1280 cow_start = cur_offset;
1281 cur_offset = extent_end;
1282 if (cur_offset > end)
1288 btrfs_release_path(path);
1289 if (cow_start != (u64)-1) {
1290 ret = cow_file_range(inode, locked_page, cow_start,
1291 found_key.offset - 1, page_started,
1294 btrfs_abort_transaction(trans, root, ret);
1297 cow_start = (u64)-1;
1300 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1301 struct extent_map *em;
1302 struct extent_map_tree *em_tree;
1303 em_tree = &BTRFS_I(inode)->extent_tree;
1304 em = alloc_extent_map();
1305 BUG_ON(!em); /* -ENOMEM */
1306 em->start = cur_offset;
1307 em->orig_start = em->start;
1308 em->len = num_bytes;
1309 em->block_len = num_bytes;
1310 em->block_start = disk_bytenr;
1311 em->bdev = root->fs_info->fs_devices->latest_bdev;
1312 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1314 write_lock(&em_tree->lock);
1315 ret = add_extent_mapping(em_tree, em);
1316 write_unlock(&em_tree->lock);
1317 if (ret != -EEXIST) {
1318 free_extent_map(em);
1321 btrfs_drop_extent_cache(inode, em->start,
1322 em->start + em->len - 1, 0);
1324 type = BTRFS_ORDERED_PREALLOC;
1326 type = BTRFS_ORDERED_NOCOW;
1329 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1330 num_bytes, num_bytes, type);
1331 BUG_ON(ret); /* -ENOMEM */
1333 if (root->root_key.objectid ==
1334 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1335 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1338 btrfs_abort_transaction(trans, root, ret);
1343 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1344 cur_offset, cur_offset + num_bytes - 1,
1345 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1346 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1347 EXTENT_SET_PRIVATE2);
1348 cur_offset = extent_end;
1349 if (cur_offset > end)
1352 btrfs_release_path(path);
1354 if (cur_offset <= end && cow_start == (u64)-1) {
1355 cow_start = cur_offset;
1359 if (cow_start != (u64)-1) {
1360 ret = cow_file_range(inode, locked_page, cow_start, end,
1361 page_started, nr_written, 1);
1363 btrfs_abort_transaction(trans, root, ret);
1370 err = btrfs_end_transaction_nolock(trans, root);
1372 err = btrfs_end_transaction(trans, root);
1377 if (ret && cur_offset < end)
1378 extent_clear_unlock_delalloc(inode,
1379 &BTRFS_I(inode)->io_tree,
1380 cur_offset, end, locked_page,
1381 EXTENT_CLEAR_UNLOCK_PAGE |
1382 EXTENT_CLEAR_UNLOCK |
1383 EXTENT_CLEAR_DELALLOC |
1384 EXTENT_CLEAR_DIRTY |
1385 EXTENT_SET_WRITEBACK |
1386 EXTENT_END_WRITEBACK);
1388 btrfs_free_path(path);
1393 * extent_io.c call back to do delayed allocation processing
1395 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1396 u64 start, u64 end, int *page_started,
1397 unsigned long *nr_written)
1400 struct btrfs_root *root = BTRFS_I(inode)->root;
1402 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1403 ret = run_delalloc_nocow(inode, locked_page, start, end,
1404 page_started, 1, nr_written);
1405 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1406 ret = run_delalloc_nocow(inode, locked_page, start, end,
1407 page_started, 0, nr_written);
1408 } else if (!btrfs_test_opt(root, COMPRESS) &&
1409 !(BTRFS_I(inode)->force_compress) &&
1410 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1411 ret = cow_file_range(inode, locked_page, start, end,
1412 page_started, nr_written, 1);
1414 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1415 &BTRFS_I(inode)->runtime_flags);
1416 ret = cow_file_range_async(inode, locked_page, start, end,
1417 page_started, nr_written);
1422 static void btrfs_split_extent_hook(struct inode *inode,
1423 struct extent_state *orig, u64 split)
1425 /* not delalloc, ignore it */
1426 if (!(orig->state & EXTENT_DELALLOC))
1429 spin_lock(&BTRFS_I(inode)->lock);
1430 BTRFS_I(inode)->outstanding_extents++;
1431 spin_unlock(&BTRFS_I(inode)->lock);
1435 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1436 * extents so we can keep track of new extents that are just merged onto old
1437 * extents, such as when we are doing sequential writes, so we can properly
1438 * account for the metadata space we'll need.
1440 static void btrfs_merge_extent_hook(struct inode *inode,
1441 struct extent_state *new,
1442 struct extent_state *other)
1444 /* not delalloc, ignore it */
1445 if (!(other->state & EXTENT_DELALLOC))
1448 spin_lock(&BTRFS_I(inode)->lock);
1449 BTRFS_I(inode)->outstanding_extents--;
1450 spin_unlock(&BTRFS_I(inode)->lock);
1454 * extent_io.c set_bit_hook, used to track delayed allocation
1455 * bytes in this file, and to maintain the list of inodes that
1456 * have pending delalloc work to be done.
1458 static void btrfs_set_bit_hook(struct inode *inode,
1459 struct extent_state *state, int *bits)
1463 * set_bit and clear bit hooks normally require _irqsave/restore
1464 * but in this case, we are only testing for the DELALLOC
1465 * bit, which is only set or cleared with irqs on
1467 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1468 struct btrfs_root *root = BTRFS_I(inode)->root;
1469 u64 len = state->end + 1 - state->start;
1470 bool do_list = !btrfs_is_free_space_inode(inode);
1472 if (*bits & EXTENT_FIRST_DELALLOC) {
1473 *bits &= ~EXTENT_FIRST_DELALLOC;
1475 spin_lock(&BTRFS_I(inode)->lock);
1476 BTRFS_I(inode)->outstanding_extents++;
1477 spin_unlock(&BTRFS_I(inode)->lock);
1480 spin_lock(&root->fs_info->delalloc_lock);
1481 BTRFS_I(inode)->delalloc_bytes += len;
1482 root->fs_info->delalloc_bytes += len;
1483 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1484 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1485 &root->fs_info->delalloc_inodes);
1487 spin_unlock(&root->fs_info->delalloc_lock);
1492 * extent_io.c clear_bit_hook, see set_bit_hook for why
1494 static void btrfs_clear_bit_hook(struct inode *inode,
1495 struct extent_state *state, int *bits)
1498 * set_bit and clear bit hooks normally require _irqsave/restore
1499 * but in this case, we are only testing for the DELALLOC
1500 * bit, which is only set or cleared with irqs on
1502 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1503 struct btrfs_root *root = BTRFS_I(inode)->root;
1504 u64 len = state->end + 1 - state->start;
1505 bool do_list = !btrfs_is_free_space_inode(inode);
1507 if (*bits & EXTENT_FIRST_DELALLOC) {
1508 *bits &= ~EXTENT_FIRST_DELALLOC;
1509 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1510 spin_lock(&BTRFS_I(inode)->lock);
1511 BTRFS_I(inode)->outstanding_extents--;
1512 spin_unlock(&BTRFS_I(inode)->lock);
1515 if (*bits & EXTENT_DO_ACCOUNTING)
1516 btrfs_delalloc_release_metadata(inode, len);
1518 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1520 btrfs_free_reserved_data_space(inode, len);
1522 spin_lock(&root->fs_info->delalloc_lock);
1523 root->fs_info->delalloc_bytes -= len;
1524 BTRFS_I(inode)->delalloc_bytes -= len;
1526 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1527 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1528 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1530 spin_unlock(&root->fs_info->delalloc_lock);
1535 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1536 * we don't create bios that span stripes or chunks
1538 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1539 size_t size, struct bio *bio,
1540 unsigned long bio_flags)
1542 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1543 struct btrfs_mapping_tree *map_tree;
1544 u64 logical = (u64)bio->bi_sector << 9;
1549 if (bio_flags & EXTENT_BIO_COMPRESSED)
1552 length = bio->bi_size;
1553 map_tree = &root->fs_info->mapping_tree;
1554 map_length = length;
1555 ret = btrfs_map_block(map_tree, READ, logical,
1556 &map_length, NULL, 0);
1557 /* Will always return 0 or 1 with map_multi == NULL */
1559 if (map_length < length + size)
1565 * in order to insert checksums into the metadata in large chunks,
1566 * we wait until bio submission time. All the pages in the bio are
1567 * checksummed and sums are attached onto the ordered extent record.
1569 * At IO completion time the cums attached on the ordered extent record
1570 * are inserted into the btree
1572 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1573 struct bio *bio, int mirror_num,
1574 unsigned long bio_flags,
1577 struct btrfs_root *root = BTRFS_I(inode)->root;
1580 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1581 BUG_ON(ret); /* -ENOMEM */
1586 * in order to insert checksums into the metadata in large chunks,
1587 * we wait until bio submission time. All the pages in the bio are
1588 * checksummed and sums are attached onto the ordered extent record.
1590 * At IO completion time the cums attached on the ordered extent record
1591 * are inserted into the btree
1593 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1594 int mirror_num, unsigned long bio_flags,
1597 struct btrfs_root *root = BTRFS_I(inode)->root;
1598 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1602 * extent_io.c submission hook. This does the right thing for csum calculation
1603 * on write, or reading the csums from the tree before a read
1605 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1606 int mirror_num, unsigned long bio_flags,
1609 struct btrfs_root *root = BTRFS_I(inode)->root;
1614 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1616 if (btrfs_is_free_space_inode(inode))
1619 if (!(rw & REQ_WRITE)) {
1620 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1624 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1625 return btrfs_submit_compressed_read(inode, bio,
1626 mirror_num, bio_flags);
1627 } else if (!skip_sum) {
1628 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1633 } else if (!skip_sum) {
1634 /* csum items have already been cloned */
1635 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1637 /* we're doing a write, do the async checksumming */
1638 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1639 inode, rw, bio, mirror_num,
1640 bio_flags, bio_offset,
1641 __btrfs_submit_bio_start,
1642 __btrfs_submit_bio_done);
1646 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1650 * given a list of ordered sums record them in the inode. This happens
1651 * at IO completion time based on sums calculated at bio submission time.
1653 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1654 struct inode *inode, u64 file_offset,
1655 struct list_head *list)
1657 struct btrfs_ordered_sum *sum;
1659 list_for_each_entry(sum, list, list) {
1660 btrfs_csum_file_blocks(trans,
1661 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1666 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1667 struct extent_state **cached_state)
1669 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1671 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1672 cached_state, GFP_NOFS);
1675 /* see btrfs_writepage_start_hook for details on why this is required */
1676 struct btrfs_writepage_fixup {
1678 struct btrfs_work work;
1681 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1683 struct btrfs_writepage_fixup *fixup;
1684 struct btrfs_ordered_extent *ordered;
1685 struct extent_state *cached_state = NULL;
1687 struct inode *inode;
1692 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1696 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1697 ClearPageChecked(page);
1701 inode = page->mapping->host;
1702 page_start = page_offset(page);
1703 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1705 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1708 /* already ordered? We're done */
1709 if (PagePrivate2(page))
1712 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1714 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1715 page_end, &cached_state, GFP_NOFS);
1717 btrfs_start_ordered_extent(inode, ordered, 1);
1718 btrfs_put_ordered_extent(ordered);
1722 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1724 mapping_set_error(page->mapping, ret);
1725 end_extent_writepage(page, ret, page_start, page_end);
1726 ClearPageChecked(page);
1730 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1731 ClearPageChecked(page);
1732 set_page_dirty(page);
1734 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1735 &cached_state, GFP_NOFS);
1738 page_cache_release(page);
1743 * There are a few paths in the higher layers of the kernel that directly
1744 * set the page dirty bit without asking the filesystem if it is a
1745 * good idea. This causes problems because we want to make sure COW
1746 * properly happens and the data=ordered rules are followed.
1748 * In our case any range that doesn't have the ORDERED bit set
1749 * hasn't been properly setup for IO. We kick off an async process
1750 * to fix it up. The async helper will wait for ordered extents, set
1751 * the delalloc bit and make it safe to write the page.
1753 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1755 struct inode *inode = page->mapping->host;
1756 struct btrfs_writepage_fixup *fixup;
1757 struct btrfs_root *root = BTRFS_I(inode)->root;
1759 /* this page is properly in the ordered list */
1760 if (TestClearPagePrivate2(page))
1763 if (PageChecked(page))
1766 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1770 SetPageChecked(page);
1771 page_cache_get(page);
1772 fixup->work.func = btrfs_writepage_fixup_worker;
1774 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1778 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1779 struct inode *inode, u64 file_pos,
1780 u64 disk_bytenr, u64 disk_num_bytes,
1781 u64 num_bytes, u64 ram_bytes,
1782 u8 compression, u8 encryption,
1783 u16 other_encoding, int extent_type)
1785 struct btrfs_root *root = BTRFS_I(inode)->root;
1786 struct btrfs_file_extent_item *fi;
1787 struct btrfs_path *path;
1788 struct extent_buffer *leaf;
1789 struct btrfs_key ins;
1793 path = btrfs_alloc_path();
1797 path->leave_spinning = 1;
1800 * we may be replacing one extent in the tree with another.
1801 * The new extent is pinned in the extent map, and we don't want
1802 * to drop it from the cache until it is completely in the btree.
1804 * So, tell btrfs_drop_extents to leave this extent in the cache.
1805 * the caller is expected to unpin it and allow it to be merged
1808 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1813 ins.objectid = btrfs_ino(inode);
1814 ins.offset = file_pos;
1815 ins.type = BTRFS_EXTENT_DATA_KEY;
1816 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1819 leaf = path->nodes[0];
1820 fi = btrfs_item_ptr(leaf, path->slots[0],
1821 struct btrfs_file_extent_item);
1822 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1823 btrfs_set_file_extent_type(leaf, fi, extent_type);
1824 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1825 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1826 btrfs_set_file_extent_offset(leaf, fi, 0);
1827 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1828 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1829 btrfs_set_file_extent_compression(leaf, fi, compression);
1830 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1831 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1833 btrfs_unlock_up_safe(path, 1);
1834 btrfs_set_lock_blocking(leaf);
1836 btrfs_mark_buffer_dirty(leaf);
1838 inode_add_bytes(inode, num_bytes);
1840 ins.objectid = disk_bytenr;
1841 ins.offset = disk_num_bytes;
1842 ins.type = BTRFS_EXTENT_ITEM_KEY;
1843 ret = btrfs_alloc_reserved_file_extent(trans, root,
1844 root->root_key.objectid,
1845 btrfs_ino(inode), file_pos, &ins);
1847 btrfs_free_path(path);
1853 * helper function for btrfs_finish_ordered_io, this
1854 * just reads in some of the csum leaves to prime them into ram
1855 * before we start the transaction. It limits the amount of btree
1856 * reads required while inside the transaction.
1858 /* as ordered data IO finishes, this gets called so we can finish
1859 * an ordered extent if the range of bytes in the file it covers are
1862 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1864 struct inode *inode = ordered_extent->inode;
1865 struct btrfs_root *root = BTRFS_I(inode)->root;
1866 struct btrfs_trans_handle *trans = NULL;
1867 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1868 struct extent_state *cached_state = NULL;
1869 int compress_type = 0;
1873 nolock = btrfs_is_free_space_inode(inode);
1875 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1880 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1881 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1882 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1885 trans = btrfs_join_transaction_nolock(root);
1887 trans = btrfs_join_transaction(root);
1889 return PTR_ERR(trans);
1890 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1891 ret = btrfs_update_inode_fallback(trans, root, inode);
1892 if (ret) /* -ENOMEM or corruption */
1893 btrfs_abort_transaction(trans, root, ret);
1898 lock_extent_bits(io_tree, ordered_extent->file_offset,
1899 ordered_extent->file_offset + ordered_extent->len - 1,
1903 trans = btrfs_join_transaction_nolock(root);
1905 trans = btrfs_join_transaction(root);
1906 if (IS_ERR(trans)) {
1907 ret = PTR_ERR(trans);
1911 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1913 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1914 compress_type = ordered_extent->compress_type;
1915 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1916 BUG_ON(compress_type);
1917 ret = btrfs_mark_extent_written(trans, inode,
1918 ordered_extent->file_offset,
1919 ordered_extent->file_offset +
1920 ordered_extent->len);
1922 BUG_ON(root == root->fs_info->tree_root);
1923 ret = insert_reserved_file_extent(trans, inode,
1924 ordered_extent->file_offset,
1925 ordered_extent->start,
1926 ordered_extent->disk_len,
1927 ordered_extent->len,
1928 ordered_extent->len,
1929 compress_type, 0, 0,
1930 BTRFS_FILE_EXTENT_REG);
1931 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1932 ordered_extent->file_offset,
1933 ordered_extent->len);
1937 btrfs_abort_transaction(trans, root, ret);
1941 add_pending_csums(trans, inode, ordered_extent->file_offset,
1942 &ordered_extent->list);
1944 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1945 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1946 ret = btrfs_update_inode_fallback(trans, root, inode);
1947 if (ret) { /* -ENOMEM or corruption */
1948 btrfs_abort_transaction(trans, root, ret);
1954 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1955 ordered_extent->file_offset +
1956 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1958 if (root != root->fs_info->tree_root)
1959 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1962 btrfs_end_transaction_nolock(trans, root);
1964 btrfs_end_transaction(trans, root);
1968 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1969 ordered_extent->file_offset +
1970 ordered_extent->len - 1, NULL, GFP_NOFS);
1973 * This needs to be dont to make sure anybody waiting knows we are done
1974 * upating everything for this ordered extent.
1976 btrfs_remove_ordered_extent(inode, ordered_extent);
1979 btrfs_put_ordered_extent(ordered_extent);
1980 /* once for the tree */
1981 btrfs_put_ordered_extent(ordered_extent);
1986 static void finish_ordered_fn(struct btrfs_work *work)
1988 struct btrfs_ordered_extent *ordered_extent;
1989 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1990 btrfs_finish_ordered_io(ordered_extent);
1993 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1994 struct extent_state *state, int uptodate)
1996 struct inode *inode = page->mapping->host;
1997 struct btrfs_root *root = BTRFS_I(inode)->root;
1998 struct btrfs_ordered_extent *ordered_extent = NULL;
1999 struct btrfs_workers *workers;
2001 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2003 ClearPagePrivate2(page);
2004 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2005 end - start + 1, uptodate))
2008 ordered_extent->work.func = finish_ordered_fn;
2009 ordered_extent->work.flags = 0;
2011 if (btrfs_is_free_space_inode(inode))
2012 workers = &root->fs_info->endio_freespace_worker;
2014 workers = &root->fs_info->endio_write_workers;
2015 btrfs_queue_worker(workers, &ordered_extent->work);
2021 * when reads are done, we need to check csums to verify the data is correct
2022 * if there's a match, we allow the bio to finish. If not, the code in
2023 * extent_io.c will try to find good copies for us.
2025 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2026 struct extent_state *state, int mirror)
2028 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2029 struct inode *inode = page->mapping->host;
2030 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2032 u64 private = ~(u32)0;
2034 struct btrfs_root *root = BTRFS_I(inode)->root;
2037 if (PageChecked(page)) {
2038 ClearPageChecked(page);
2042 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2045 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2046 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2047 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2052 if (state && state->start == start) {
2053 private = state->private;
2056 ret = get_state_private(io_tree, start, &private);
2058 kaddr = kmap_atomic(page);
2062 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2063 btrfs_csum_final(csum, (char *)&csum);
2064 if (csum != private)
2067 kunmap_atomic(kaddr);
2072 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2074 (unsigned long long)btrfs_ino(page->mapping->host),
2075 (unsigned long long)start, csum,
2076 (unsigned long long)private);
2077 memset(kaddr + offset, 1, end - start + 1);
2078 flush_dcache_page(page);
2079 kunmap_atomic(kaddr);
2085 struct delayed_iput {
2086 struct list_head list;
2087 struct inode *inode;
2090 /* JDM: If this is fs-wide, why can't we add a pointer to
2091 * btrfs_inode instead and avoid the allocation? */
2092 void btrfs_add_delayed_iput(struct inode *inode)
2094 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2095 struct delayed_iput *delayed;
2097 if (atomic_add_unless(&inode->i_count, -1, 1))
2100 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2101 delayed->inode = inode;
2103 spin_lock(&fs_info->delayed_iput_lock);
2104 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2105 spin_unlock(&fs_info->delayed_iput_lock);
2108 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2111 struct btrfs_fs_info *fs_info = root->fs_info;
2112 struct delayed_iput *delayed;
2115 spin_lock(&fs_info->delayed_iput_lock);
2116 empty = list_empty(&fs_info->delayed_iputs);
2117 spin_unlock(&fs_info->delayed_iput_lock);
2121 down_read(&root->fs_info->cleanup_work_sem);
2122 spin_lock(&fs_info->delayed_iput_lock);
2123 list_splice_init(&fs_info->delayed_iputs, &list);
2124 spin_unlock(&fs_info->delayed_iput_lock);
2126 while (!list_empty(&list)) {
2127 delayed = list_entry(list.next, struct delayed_iput, list);
2128 list_del(&delayed->list);
2129 iput(delayed->inode);
2132 up_read(&root->fs_info->cleanup_work_sem);
2135 enum btrfs_orphan_cleanup_state {
2136 ORPHAN_CLEANUP_STARTED = 1,
2137 ORPHAN_CLEANUP_DONE = 2,
2141 * This is called in transaction commit time. If there are no orphan
2142 * files in the subvolume, it removes orphan item and frees block_rsv
2145 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2146 struct btrfs_root *root)
2148 struct btrfs_block_rsv *block_rsv;
2151 if (atomic_read(&root->orphan_inodes) ||
2152 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2155 spin_lock(&root->orphan_lock);
2156 if (atomic_read(&root->orphan_inodes)) {
2157 spin_unlock(&root->orphan_lock);
2161 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2162 spin_unlock(&root->orphan_lock);
2166 block_rsv = root->orphan_block_rsv;
2167 root->orphan_block_rsv = NULL;
2168 spin_unlock(&root->orphan_lock);
2170 if (root->orphan_item_inserted &&
2171 btrfs_root_refs(&root->root_item) > 0) {
2172 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2173 root->root_key.objectid);
2175 root->orphan_item_inserted = 0;
2179 WARN_ON(block_rsv->size > 0);
2180 btrfs_free_block_rsv(root, block_rsv);
2185 * This creates an orphan entry for the given inode in case something goes
2186 * wrong in the middle of an unlink/truncate.
2188 * NOTE: caller of this function should reserve 5 units of metadata for
2191 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2193 struct btrfs_root *root = BTRFS_I(inode)->root;
2194 struct btrfs_block_rsv *block_rsv = NULL;
2199 if (!root->orphan_block_rsv) {
2200 block_rsv = btrfs_alloc_block_rsv(root);
2205 spin_lock(&root->orphan_lock);
2206 if (!root->orphan_block_rsv) {
2207 root->orphan_block_rsv = block_rsv;
2208 } else if (block_rsv) {
2209 btrfs_free_block_rsv(root, block_rsv);
2213 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2214 &BTRFS_I(inode)->runtime_flags)) {
2217 * For proper ENOSPC handling, we should do orphan
2218 * cleanup when mounting. But this introduces backward
2219 * compatibility issue.
2221 if (!xchg(&root->orphan_item_inserted, 1))
2227 atomic_dec(&root->orphan_inodes);
2230 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2231 &BTRFS_I(inode)->runtime_flags))
2233 spin_unlock(&root->orphan_lock);
2235 /* grab metadata reservation from transaction handle */
2237 ret = btrfs_orphan_reserve_metadata(trans, inode);
2238 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2241 /* insert an orphan item to track this unlinked/truncated file */
2243 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2244 if (ret && ret != -EEXIST) {
2245 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2246 &BTRFS_I(inode)->runtime_flags);
2247 btrfs_abort_transaction(trans, root, ret);
2253 /* insert an orphan item to track subvolume contains orphan files */
2255 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2256 root->root_key.objectid);
2257 if (ret && ret != -EEXIST) {
2258 btrfs_abort_transaction(trans, root, ret);
2266 * We have done the truncate/delete so we can go ahead and remove the orphan
2267 * item for this particular inode.
2269 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2271 struct btrfs_root *root = BTRFS_I(inode)->root;
2272 int delete_item = 0;
2273 int release_rsv = 0;
2276 spin_lock(&root->orphan_lock);
2277 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2278 &BTRFS_I(inode)->runtime_flags))
2281 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2282 &BTRFS_I(inode)->runtime_flags))
2284 spin_unlock(&root->orphan_lock);
2286 if (trans && delete_item) {
2287 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2288 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2292 btrfs_orphan_release_metadata(inode);
2293 atomic_dec(&root->orphan_inodes);
2300 * this cleans up any orphans that may be left on the list from the last use
2303 int btrfs_orphan_cleanup(struct btrfs_root *root)
2305 struct btrfs_path *path;
2306 struct extent_buffer *leaf;
2307 struct btrfs_key key, found_key;
2308 struct btrfs_trans_handle *trans;
2309 struct inode *inode;
2310 u64 last_objectid = 0;
2311 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2313 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2316 path = btrfs_alloc_path();
2323 key.objectid = BTRFS_ORPHAN_OBJECTID;
2324 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2325 key.offset = (u64)-1;
2328 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2333 * if ret == 0 means we found what we were searching for, which
2334 * is weird, but possible, so only screw with path if we didn't
2335 * find the key and see if we have stuff that matches
2339 if (path->slots[0] == 0)
2344 /* pull out the item */
2345 leaf = path->nodes[0];
2346 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2348 /* make sure the item matches what we want */
2349 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2351 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2354 /* release the path since we're done with it */
2355 btrfs_release_path(path);
2358 * this is where we are basically btrfs_lookup, without the
2359 * crossing root thing. we store the inode number in the
2360 * offset of the orphan item.
2363 if (found_key.offset == last_objectid) {
2364 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2365 "stopping orphan cleanup\n");
2370 last_objectid = found_key.offset;
2372 found_key.objectid = found_key.offset;
2373 found_key.type = BTRFS_INODE_ITEM_KEY;
2374 found_key.offset = 0;
2375 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2376 ret = PTR_RET(inode);
2377 if (ret && ret != -ESTALE)
2380 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2381 struct btrfs_root *dead_root;
2382 struct btrfs_fs_info *fs_info = root->fs_info;
2383 int is_dead_root = 0;
2386 * this is an orphan in the tree root. Currently these
2387 * could come from 2 sources:
2388 * a) a snapshot deletion in progress
2389 * b) a free space cache inode
2390 * We need to distinguish those two, as the snapshot
2391 * orphan must not get deleted.
2392 * find_dead_roots already ran before us, so if this
2393 * is a snapshot deletion, we should find the root
2394 * in the dead_roots list
2396 spin_lock(&fs_info->trans_lock);
2397 list_for_each_entry(dead_root, &fs_info->dead_roots,
2399 if (dead_root->root_key.objectid ==
2400 found_key.objectid) {
2405 spin_unlock(&fs_info->trans_lock);
2407 /* prevent this orphan from being found again */
2408 key.offset = found_key.objectid - 1;
2413 * Inode is already gone but the orphan item is still there,
2414 * kill the orphan item.
2416 if (ret == -ESTALE) {
2417 trans = btrfs_start_transaction(root, 1);
2418 if (IS_ERR(trans)) {
2419 ret = PTR_ERR(trans);
2422 printk(KERN_ERR "auto deleting %Lu\n",
2423 found_key.objectid);
2424 ret = btrfs_del_orphan_item(trans, root,
2425 found_key.objectid);
2426 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2427 btrfs_end_transaction(trans, root);
2432 * add this inode to the orphan list so btrfs_orphan_del does
2433 * the proper thing when we hit it
2435 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2436 &BTRFS_I(inode)->runtime_flags);
2438 /* if we have links, this was a truncate, lets do that */
2439 if (inode->i_nlink) {
2440 if (!S_ISREG(inode->i_mode)) {
2446 ret = btrfs_truncate(inode);
2451 /* this will do delete_inode and everything for us */
2456 /* release the path since we're done with it */
2457 btrfs_release_path(path);
2459 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2461 if (root->orphan_block_rsv)
2462 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2465 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2466 trans = btrfs_join_transaction(root);
2468 btrfs_end_transaction(trans, root);
2472 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2474 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2478 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2479 btrfs_free_path(path);
2484 * very simple check to peek ahead in the leaf looking for xattrs. If we
2485 * don't find any xattrs, we know there can't be any acls.
2487 * slot is the slot the inode is in, objectid is the objectid of the inode
2489 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2490 int slot, u64 objectid)
2492 u32 nritems = btrfs_header_nritems(leaf);
2493 struct btrfs_key found_key;
2497 while (slot < nritems) {
2498 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2500 /* we found a different objectid, there must not be acls */
2501 if (found_key.objectid != objectid)
2504 /* we found an xattr, assume we've got an acl */
2505 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2509 * we found a key greater than an xattr key, there can't
2510 * be any acls later on
2512 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2519 * it goes inode, inode backrefs, xattrs, extents,
2520 * so if there are a ton of hard links to an inode there can
2521 * be a lot of backrefs. Don't waste time searching too hard,
2522 * this is just an optimization
2527 /* we hit the end of the leaf before we found an xattr or
2528 * something larger than an xattr. We have to assume the inode
2535 * read an inode from the btree into the in-memory inode
2537 static void btrfs_read_locked_inode(struct inode *inode)
2539 struct btrfs_path *path;
2540 struct extent_buffer *leaf;
2541 struct btrfs_inode_item *inode_item;
2542 struct btrfs_timespec *tspec;
2543 struct btrfs_root *root = BTRFS_I(inode)->root;
2544 struct btrfs_key location;
2548 bool filled = false;
2550 ret = btrfs_fill_inode(inode, &rdev);
2554 path = btrfs_alloc_path();
2558 path->leave_spinning = 1;
2559 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2561 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2565 leaf = path->nodes[0];
2570 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2571 struct btrfs_inode_item);
2572 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2573 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2574 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2575 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2576 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2578 tspec = btrfs_inode_atime(inode_item);
2579 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2580 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2582 tspec = btrfs_inode_mtime(inode_item);
2583 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2584 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2586 tspec = btrfs_inode_ctime(inode_item);
2587 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2588 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2590 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2591 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2592 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2593 inode->i_generation = BTRFS_I(inode)->generation;
2595 rdev = btrfs_inode_rdev(leaf, inode_item);
2597 BTRFS_I(inode)->index_cnt = (u64)-1;
2598 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2601 * try to precache a NULL acl entry for files that don't have
2602 * any xattrs or acls
2604 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2607 cache_no_acl(inode);
2609 btrfs_free_path(path);
2611 switch (inode->i_mode & S_IFMT) {
2613 inode->i_mapping->a_ops = &btrfs_aops;
2614 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2615 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2616 inode->i_fop = &btrfs_file_operations;
2617 inode->i_op = &btrfs_file_inode_operations;
2620 inode->i_fop = &btrfs_dir_file_operations;
2621 if (root == root->fs_info->tree_root)
2622 inode->i_op = &btrfs_dir_ro_inode_operations;
2624 inode->i_op = &btrfs_dir_inode_operations;
2627 inode->i_op = &btrfs_symlink_inode_operations;
2628 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2629 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2632 inode->i_op = &btrfs_special_inode_operations;
2633 init_special_inode(inode, inode->i_mode, rdev);
2637 btrfs_update_iflags(inode);
2641 btrfs_free_path(path);
2642 make_bad_inode(inode);
2646 * given a leaf and an inode, copy the inode fields into the leaf
2648 static void fill_inode_item(struct btrfs_trans_handle *trans,
2649 struct extent_buffer *leaf,
2650 struct btrfs_inode_item *item,
2651 struct inode *inode)
2653 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2654 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2655 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2656 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2657 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2659 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2660 inode->i_atime.tv_sec);
2661 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2662 inode->i_atime.tv_nsec);
2664 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2665 inode->i_mtime.tv_sec);
2666 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2667 inode->i_mtime.tv_nsec);
2669 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2670 inode->i_ctime.tv_sec);
2671 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2672 inode->i_ctime.tv_nsec);
2674 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2675 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2676 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2677 btrfs_set_inode_transid(leaf, item, trans->transid);
2678 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2679 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2680 btrfs_set_inode_block_group(leaf, item, 0);
2684 * copy everything in the in-memory inode into the btree.
2686 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2687 struct btrfs_root *root, struct inode *inode)
2689 struct btrfs_inode_item *inode_item;
2690 struct btrfs_path *path;
2691 struct extent_buffer *leaf;
2694 path = btrfs_alloc_path();
2698 path->leave_spinning = 1;
2699 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2707 btrfs_unlock_up_safe(path, 1);
2708 leaf = path->nodes[0];
2709 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2710 struct btrfs_inode_item);
2712 fill_inode_item(trans, leaf, inode_item, inode);
2713 btrfs_mark_buffer_dirty(leaf);
2714 btrfs_set_inode_last_trans(trans, inode);
2717 btrfs_free_path(path);
2722 * copy everything in the in-memory inode into the btree.
2724 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2725 struct btrfs_root *root, struct inode *inode)
2730 * If the inode is a free space inode, we can deadlock during commit
2731 * if we put it into the delayed code.
2733 * The data relocation inode should also be directly updated
2736 if (!btrfs_is_free_space_inode(inode)
2737 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2738 btrfs_update_root_times(trans, root);
2740 ret = btrfs_delayed_update_inode(trans, root, inode);
2742 btrfs_set_inode_last_trans(trans, inode);
2746 return btrfs_update_inode_item(trans, root, inode);
2749 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2750 struct btrfs_root *root, struct inode *inode)
2754 ret = btrfs_update_inode(trans, root, inode);
2756 return btrfs_update_inode_item(trans, root, inode);
2761 * unlink helper that gets used here in inode.c and in the tree logging
2762 * recovery code. It remove a link in a directory with a given name, and
2763 * also drops the back refs in the inode to the directory
2765 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2766 struct btrfs_root *root,
2767 struct inode *dir, struct inode *inode,
2768 const char *name, int name_len)
2770 struct btrfs_path *path;
2772 struct extent_buffer *leaf;
2773 struct btrfs_dir_item *di;
2774 struct btrfs_key key;
2776 u64 ino = btrfs_ino(inode);
2777 u64 dir_ino = btrfs_ino(dir);
2779 path = btrfs_alloc_path();
2785 path->leave_spinning = 1;
2786 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2787 name, name_len, -1);
2796 leaf = path->nodes[0];
2797 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2798 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2801 btrfs_release_path(path);
2803 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2806 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2807 "inode %llu parent %llu\n", name_len, name,
2808 (unsigned long long)ino, (unsigned long long)dir_ino);
2809 btrfs_abort_transaction(trans, root, ret);
2813 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2815 btrfs_abort_transaction(trans, root, ret);
2819 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2821 if (ret != 0 && ret != -ENOENT) {
2822 btrfs_abort_transaction(trans, root, ret);
2826 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2831 btrfs_free_path(path);
2835 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2836 inode_inc_iversion(inode);
2837 inode_inc_iversion(dir);
2838 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2839 ret = btrfs_update_inode(trans, root, dir);
2844 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2845 struct btrfs_root *root,
2846 struct inode *dir, struct inode *inode,
2847 const char *name, int name_len)
2850 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2852 btrfs_drop_nlink(inode);
2853 ret = btrfs_update_inode(trans, root, inode);
2859 /* helper to check if there is any shared block in the path */
2860 static int check_path_shared(struct btrfs_root *root,
2861 struct btrfs_path *path)
2863 struct extent_buffer *eb;
2867 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2870 if (!path->nodes[level])
2872 eb = path->nodes[level];
2873 if (!btrfs_block_can_be_shared(root, eb))
2875 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2884 * helper to start transaction for unlink and rmdir.
2886 * unlink and rmdir are special in btrfs, they do not always free space.
2887 * so in enospc case, we should make sure they will free space before
2888 * allowing them to use the global metadata reservation.
2890 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2891 struct dentry *dentry)
2893 struct btrfs_trans_handle *trans;
2894 struct btrfs_root *root = BTRFS_I(dir)->root;
2895 struct btrfs_path *path;
2896 struct btrfs_inode_ref *ref;
2897 struct btrfs_dir_item *di;
2898 struct inode *inode = dentry->d_inode;
2903 u64 ino = btrfs_ino(inode);
2904 u64 dir_ino = btrfs_ino(dir);
2907 * 1 for the possible orphan item
2908 * 1 for the dir item
2909 * 1 for the dir index
2910 * 1 for the inode ref
2911 * 1 for the inode ref in the tree log
2912 * 2 for the dir entries in the log
2915 trans = btrfs_start_transaction(root, 8);
2916 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2919 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2920 return ERR_PTR(-ENOSPC);
2922 /* check if there is someone else holds reference */
2923 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2924 return ERR_PTR(-ENOSPC);
2926 if (atomic_read(&inode->i_count) > 2)
2927 return ERR_PTR(-ENOSPC);
2929 if (xchg(&root->fs_info->enospc_unlink, 1))
2930 return ERR_PTR(-ENOSPC);
2932 path = btrfs_alloc_path();
2934 root->fs_info->enospc_unlink = 0;
2935 return ERR_PTR(-ENOMEM);
2938 /* 1 for the orphan item */
2939 trans = btrfs_start_transaction(root, 1);
2940 if (IS_ERR(trans)) {
2941 btrfs_free_path(path);
2942 root->fs_info->enospc_unlink = 0;
2946 path->skip_locking = 1;
2947 path->search_commit_root = 1;
2949 ret = btrfs_lookup_inode(trans, root, path,
2950 &BTRFS_I(dir)->location, 0);
2956 if (check_path_shared(root, path))
2961 btrfs_release_path(path);
2963 ret = btrfs_lookup_inode(trans, root, path,
2964 &BTRFS_I(inode)->location, 0);
2970 if (check_path_shared(root, path))
2975 btrfs_release_path(path);
2977 if (ret == 0 && S_ISREG(inode->i_mode)) {
2978 ret = btrfs_lookup_file_extent(trans, root, path,
2984 BUG_ON(ret == 0); /* Corruption */
2985 if (check_path_shared(root, path))
2987 btrfs_release_path(path);
2995 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2996 dentry->d_name.name, dentry->d_name.len, 0);
3002 if (check_path_shared(root, path))
3008 btrfs_release_path(path);
3010 ref = btrfs_lookup_inode_ref(trans, root, path,
3011 dentry->d_name.name, dentry->d_name.len,
3017 BUG_ON(!ref); /* Logic error */
3018 if (check_path_shared(root, path))
3020 index = btrfs_inode_ref_index(path->nodes[0], ref);
3021 btrfs_release_path(path);
3024 * This is a commit root search, if we can lookup inode item and other
3025 * relative items in the commit root, it means the transaction of
3026 * dir/file creation has been committed, and the dir index item that we
3027 * delay to insert has also been inserted into the commit root. So
3028 * we needn't worry about the delayed insertion of the dir index item
3031 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3032 dentry->d_name.name, dentry->d_name.len, 0);
3037 BUG_ON(ret == -ENOENT);
3038 if (check_path_shared(root, path))
3043 btrfs_free_path(path);
3044 /* Migrate the orphan reservation over */
3046 err = btrfs_block_rsv_migrate(trans->block_rsv,
3047 &root->fs_info->global_block_rsv,
3048 trans->bytes_reserved);
3051 btrfs_end_transaction(trans, root);
3052 root->fs_info->enospc_unlink = 0;
3053 return ERR_PTR(err);
3056 trans->block_rsv = &root->fs_info->global_block_rsv;
3060 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3061 struct btrfs_root *root)
3063 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3064 btrfs_block_rsv_release(root, trans->block_rsv,
3065 trans->bytes_reserved);
3066 trans->block_rsv = &root->fs_info->trans_block_rsv;
3067 BUG_ON(!root->fs_info->enospc_unlink);
3068 root->fs_info->enospc_unlink = 0;
3070 btrfs_end_transaction(trans, root);
3073 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3075 struct btrfs_root *root = BTRFS_I(dir)->root;
3076 struct btrfs_trans_handle *trans;
3077 struct inode *inode = dentry->d_inode;
3079 unsigned long nr = 0;
3081 trans = __unlink_start_trans(dir, dentry);
3083 return PTR_ERR(trans);
3085 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3087 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3088 dentry->d_name.name, dentry->d_name.len);
3092 if (inode->i_nlink == 0) {
3093 ret = btrfs_orphan_add(trans, inode);
3099 nr = trans->blocks_used;
3100 __unlink_end_trans(trans, root);
3101 btrfs_btree_balance_dirty(root, nr);
3105 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3106 struct btrfs_root *root,
3107 struct inode *dir, u64 objectid,
3108 const char *name, int name_len)
3110 struct btrfs_path *path;
3111 struct extent_buffer *leaf;
3112 struct btrfs_dir_item *di;
3113 struct btrfs_key key;
3116 u64 dir_ino = btrfs_ino(dir);
3118 path = btrfs_alloc_path();
3122 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3123 name, name_len, -1);
3124 if (IS_ERR_OR_NULL(di)) {
3132 leaf = path->nodes[0];
3133 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3134 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3135 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3137 btrfs_abort_transaction(trans, root, ret);
3140 btrfs_release_path(path);
3142 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3143 objectid, root->root_key.objectid,
3144 dir_ino, &index, name, name_len);
3146 if (ret != -ENOENT) {
3147 btrfs_abort_transaction(trans, root, ret);
3150 di = btrfs_search_dir_index_item(root, path, dir_ino,
3152 if (IS_ERR_OR_NULL(di)) {
3157 btrfs_abort_transaction(trans, root, ret);
3161 leaf = path->nodes[0];
3162 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3163 btrfs_release_path(path);
3166 btrfs_release_path(path);
3168 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3170 btrfs_abort_transaction(trans, root, ret);
3174 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3175 inode_inc_iversion(dir);
3176 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3177 ret = btrfs_update_inode(trans, root, dir);
3179 btrfs_abort_transaction(trans, root, ret);
3181 btrfs_free_path(path);
3185 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3187 struct inode *inode = dentry->d_inode;
3189 struct btrfs_root *root = BTRFS_I(dir)->root;
3190 struct btrfs_trans_handle *trans;
3191 unsigned long nr = 0;
3193 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3194 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3197 trans = __unlink_start_trans(dir, dentry);
3199 return PTR_ERR(trans);
3201 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3202 err = btrfs_unlink_subvol(trans, root, dir,
3203 BTRFS_I(inode)->location.objectid,
3204 dentry->d_name.name,
3205 dentry->d_name.len);
3209 err = btrfs_orphan_add(trans, inode);
3213 /* now the directory is empty */
3214 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3215 dentry->d_name.name, dentry->d_name.len);
3217 btrfs_i_size_write(inode, 0);
3219 nr = trans->blocks_used;
3220 __unlink_end_trans(trans, root);
3221 btrfs_btree_balance_dirty(root, nr);
3227 * this can truncate away extent items, csum items and directory items.
3228 * It starts at a high offset and removes keys until it can't find
3229 * any higher than new_size
3231 * csum items that cross the new i_size are truncated to the new size
3234 * min_type is the minimum key type to truncate down to. If set to 0, this
3235 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3237 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3238 struct btrfs_root *root,
3239 struct inode *inode,
3240 u64 new_size, u32 min_type)
3242 struct btrfs_path *path;
3243 struct extent_buffer *leaf;
3244 struct btrfs_file_extent_item *fi;
3245 struct btrfs_key key;
3246 struct btrfs_key found_key;
3247 u64 extent_start = 0;
3248 u64 extent_num_bytes = 0;
3249 u64 extent_offset = 0;
3251 u64 mask = root->sectorsize - 1;
3252 u32 found_type = (u8)-1;
3255 int pending_del_nr = 0;
3256 int pending_del_slot = 0;
3257 int extent_type = -1;
3260 u64 ino = btrfs_ino(inode);
3262 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3264 path = btrfs_alloc_path();
3269 if (root->ref_cows || root == root->fs_info->tree_root)
3270 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3273 * This function is also used to drop the items in the log tree before
3274 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3275 * it is used to drop the loged items. So we shouldn't kill the delayed
3278 if (min_type == 0 && root == BTRFS_I(inode)->root)
3279 btrfs_kill_delayed_inode_items(inode);
3282 key.offset = (u64)-1;
3286 path->leave_spinning = 1;
3287 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3294 /* there are no items in the tree for us to truncate, we're
3297 if (path->slots[0] == 0)
3304 leaf = path->nodes[0];
3305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3306 found_type = btrfs_key_type(&found_key);
3308 if (found_key.objectid != ino)
3311 if (found_type < min_type)
3314 item_end = found_key.offset;
3315 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3316 fi = btrfs_item_ptr(leaf, path->slots[0],
3317 struct btrfs_file_extent_item);
3318 extent_type = btrfs_file_extent_type(leaf, fi);
3319 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3321 btrfs_file_extent_num_bytes(leaf, fi);
3322 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3323 item_end += btrfs_file_extent_inline_len(leaf,
3328 if (found_type > min_type) {
3331 if (item_end < new_size)
3333 if (found_key.offset >= new_size)
3339 /* FIXME, shrink the extent if the ref count is only 1 */
3340 if (found_type != BTRFS_EXTENT_DATA_KEY)
3343 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3345 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3347 u64 orig_num_bytes =
3348 btrfs_file_extent_num_bytes(leaf, fi);
3349 extent_num_bytes = new_size -
3350 found_key.offset + root->sectorsize - 1;
3351 extent_num_bytes = extent_num_bytes &
3352 ~((u64)root->sectorsize - 1);
3353 btrfs_set_file_extent_num_bytes(leaf, fi,
3355 num_dec = (orig_num_bytes -
3357 if (root->ref_cows && extent_start != 0)
3358 inode_sub_bytes(inode, num_dec);
3359 btrfs_mark_buffer_dirty(leaf);
3362 btrfs_file_extent_disk_num_bytes(leaf,
3364 extent_offset = found_key.offset -
3365 btrfs_file_extent_offset(leaf, fi);
3367 /* FIXME blocksize != 4096 */
3368 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3369 if (extent_start != 0) {
3372 inode_sub_bytes(inode, num_dec);
3375 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3377 * we can't truncate inline items that have had
3381 btrfs_file_extent_compression(leaf, fi) == 0 &&
3382 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3383 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3384 u32 size = new_size - found_key.offset;
3386 if (root->ref_cows) {
3387 inode_sub_bytes(inode, item_end + 1 -
3391 btrfs_file_extent_calc_inline_size(size);
3392 btrfs_truncate_item(trans, root, path,
3394 } else if (root->ref_cows) {
3395 inode_sub_bytes(inode, item_end + 1 -
3401 if (!pending_del_nr) {
3402 /* no pending yet, add ourselves */
3403 pending_del_slot = path->slots[0];
3405 } else if (pending_del_nr &&
3406 path->slots[0] + 1 == pending_del_slot) {
3407 /* hop on the pending chunk */
3409 pending_del_slot = path->slots[0];
3416 if (found_extent && (root->ref_cows ||
3417 root == root->fs_info->tree_root)) {
3418 btrfs_set_path_blocking(path);
3419 ret = btrfs_free_extent(trans, root, extent_start,
3420 extent_num_bytes, 0,
3421 btrfs_header_owner(leaf),
3422 ino, extent_offset, 0);
3426 if (found_type == BTRFS_INODE_ITEM_KEY)
3429 if (path->slots[0] == 0 ||
3430 path->slots[0] != pending_del_slot) {
3431 if (root->ref_cows &&
3432 BTRFS_I(inode)->location.objectid !=
3433 BTRFS_FREE_INO_OBJECTID) {
3437 if (pending_del_nr) {
3438 ret = btrfs_del_items(trans, root, path,
3442 btrfs_abort_transaction(trans,
3448 btrfs_release_path(path);
3455 if (pending_del_nr) {
3456 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3459 btrfs_abort_transaction(trans, root, ret);
3462 btrfs_free_path(path);
3467 * taken from block_truncate_page, but does cow as it zeros out
3468 * any bytes left in the last page in the file.
3470 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3472 struct inode *inode = mapping->host;
3473 struct btrfs_root *root = BTRFS_I(inode)->root;
3474 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3475 struct btrfs_ordered_extent *ordered;
3476 struct extent_state *cached_state = NULL;
3478 u32 blocksize = root->sectorsize;
3479 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3480 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3482 gfp_t mask = btrfs_alloc_write_mask(mapping);
3487 if ((offset & (blocksize - 1)) == 0)
3489 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3495 page = find_or_create_page(mapping, index, mask);
3497 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3501 page_start = page_offset(page);
3502 page_end = page_start + PAGE_CACHE_SIZE - 1;
3504 if (!PageUptodate(page)) {
3505 ret = btrfs_readpage(NULL, page);
3507 if (page->mapping != mapping) {
3509 page_cache_release(page);
3512 if (!PageUptodate(page)) {
3517 wait_on_page_writeback(page);
3519 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3520 set_page_extent_mapped(page);
3522 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3524 unlock_extent_cached(io_tree, page_start, page_end,
3525 &cached_state, GFP_NOFS);
3527 page_cache_release(page);
3528 btrfs_start_ordered_extent(inode, ordered, 1);
3529 btrfs_put_ordered_extent(ordered);
3533 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3534 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3535 0, 0, &cached_state, GFP_NOFS);
3537 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3540 unlock_extent_cached(io_tree, page_start, page_end,
3541 &cached_state, GFP_NOFS);
3546 if (offset != PAGE_CACHE_SIZE) {
3548 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3549 flush_dcache_page(page);
3552 ClearPageChecked(page);
3553 set_page_dirty(page);
3554 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3559 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3561 page_cache_release(page);
3567 * This function puts in dummy file extents for the area we're creating a hole
3568 * for. So if we are truncating this file to a larger size we need to insert
3569 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3570 * the range between oldsize and size
3572 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3574 struct btrfs_trans_handle *trans;
3575 struct btrfs_root *root = BTRFS_I(inode)->root;
3576 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3577 struct extent_map *em = NULL;
3578 struct extent_state *cached_state = NULL;
3579 u64 mask = root->sectorsize - 1;
3580 u64 hole_start = (oldsize + mask) & ~mask;
3581 u64 block_end = (size + mask) & ~mask;
3587 if (size <= hole_start)
3591 struct btrfs_ordered_extent *ordered;
3592 btrfs_wait_ordered_range(inode, hole_start,
3593 block_end - hole_start);
3594 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3596 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3599 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3600 &cached_state, GFP_NOFS);
3601 btrfs_put_ordered_extent(ordered);
3604 cur_offset = hole_start;
3606 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3607 block_end - cur_offset, 0);
3612 last_byte = min(extent_map_end(em), block_end);
3613 last_byte = (last_byte + mask) & ~mask;
3614 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3616 hole_size = last_byte - cur_offset;
3618 trans = btrfs_start_transaction(root, 3);
3619 if (IS_ERR(trans)) {
3620 err = PTR_ERR(trans);
3624 err = btrfs_drop_extents(trans, inode, cur_offset,
3625 cur_offset + hole_size,
3628 btrfs_abort_transaction(trans, root, err);
3629 btrfs_end_transaction(trans, root);
3633 err = btrfs_insert_file_extent(trans, root,
3634 btrfs_ino(inode), cur_offset, 0,
3635 0, hole_size, 0, hole_size,
3638 btrfs_abort_transaction(trans, root, err);
3639 btrfs_end_transaction(trans, root);
3643 btrfs_drop_extent_cache(inode, hole_start,
3646 btrfs_update_inode(trans, root, inode);
3647 btrfs_end_transaction(trans, root);
3649 free_extent_map(em);
3651 cur_offset = last_byte;
3652 if (cur_offset >= block_end)
3656 free_extent_map(em);
3657 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3662 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3664 struct btrfs_root *root = BTRFS_I(inode)->root;
3665 struct btrfs_trans_handle *trans;
3666 loff_t oldsize = i_size_read(inode);
3669 if (newsize == oldsize)
3672 if (newsize > oldsize) {
3673 truncate_pagecache(inode, oldsize, newsize);
3674 ret = btrfs_cont_expand(inode, oldsize, newsize);
3678 trans = btrfs_start_transaction(root, 1);
3680 return PTR_ERR(trans);
3682 i_size_write(inode, newsize);
3683 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3684 ret = btrfs_update_inode(trans, root, inode);
3685 btrfs_end_transaction(trans, root);
3689 * We're truncating a file that used to have good data down to
3690 * zero. Make sure it gets into the ordered flush list so that
3691 * any new writes get down to disk quickly.
3694 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3695 &BTRFS_I(inode)->runtime_flags);
3697 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3698 truncate_setsize(inode, newsize);
3699 ret = btrfs_truncate(inode);
3705 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3707 struct inode *inode = dentry->d_inode;
3708 struct btrfs_root *root = BTRFS_I(inode)->root;
3711 if (btrfs_root_readonly(root))
3714 err = inode_change_ok(inode, attr);
3718 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3719 err = btrfs_setsize(inode, attr->ia_size);
3724 if (attr->ia_valid) {
3725 setattr_copy(inode, attr);
3726 inode_inc_iversion(inode);
3727 err = btrfs_dirty_inode(inode);
3729 if (!err && attr->ia_valid & ATTR_MODE)
3730 err = btrfs_acl_chmod(inode);
3736 void btrfs_evict_inode(struct inode *inode)
3738 struct btrfs_trans_handle *trans;
3739 struct btrfs_root *root = BTRFS_I(inode)->root;
3740 struct btrfs_block_rsv *rsv, *global_rsv;
3741 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3745 trace_btrfs_inode_evict(inode);
3747 truncate_inode_pages(&inode->i_data, 0);
3748 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3749 btrfs_is_free_space_inode(inode)))
3752 if (is_bad_inode(inode)) {
3753 btrfs_orphan_del(NULL, inode);
3756 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3757 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3759 if (root->fs_info->log_root_recovering) {
3760 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3761 &BTRFS_I(inode)->runtime_flags));
3765 if (inode->i_nlink > 0) {
3766 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3770 rsv = btrfs_alloc_block_rsv(root);
3772 btrfs_orphan_del(NULL, inode);
3775 rsv->size = min_size;
3776 global_rsv = &root->fs_info->global_block_rsv;
3778 btrfs_i_size_write(inode, 0);
3781 * This is a bit simpler than btrfs_truncate since
3783 * 1) We've already reserved our space for our orphan item in the
3785 * 2) We're going to delete the inode item, so we don't need to update
3788 * So we just need to reserve some slack space in case we add bytes when
3789 * doing the truncate.
3792 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3795 * Try and steal from the global reserve since we will
3796 * likely not use this space anyway, we want to try as
3797 * hard as possible to get this to work.
3800 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3803 printk(KERN_WARNING "Could not get space for a "
3804 "delete, will truncate on mount %d\n", ret);
3805 btrfs_orphan_del(NULL, inode);
3806 btrfs_free_block_rsv(root, rsv);
3810 trans = btrfs_start_transaction(root, 0);
3811 if (IS_ERR(trans)) {
3812 btrfs_orphan_del(NULL, inode);
3813 btrfs_free_block_rsv(root, rsv);
3817 trans->block_rsv = rsv;
3819 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3823 nr = trans->blocks_used;
3824 btrfs_end_transaction(trans, root);
3826 btrfs_btree_balance_dirty(root, nr);
3829 btrfs_free_block_rsv(root, rsv);
3832 trans->block_rsv = root->orphan_block_rsv;
3833 ret = btrfs_orphan_del(trans, inode);
3837 trans->block_rsv = &root->fs_info->trans_block_rsv;
3838 if (!(root == root->fs_info->tree_root ||
3839 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3840 btrfs_return_ino(root, btrfs_ino(inode));
3842 nr = trans->blocks_used;
3843 btrfs_end_transaction(trans, root);
3844 btrfs_btree_balance_dirty(root, nr);
3851 * this returns the key found in the dir entry in the location pointer.
3852 * If no dir entries were found, location->objectid is 0.
3854 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3855 struct btrfs_key *location)
3857 const char *name = dentry->d_name.name;
3858 int namelen = dentry->d_name.len;
3859 struct btrfs_dir_item *di;
3860 struct btrfs_path *path;
3861 struct btrfs_root *root = BTRFS_I(dir)->root;
3864 path = btrfs_alloc_path();
3868 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3873 if (IS_ERR_OR_NULL(di))
3876 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3878 btrfs_free_path(path);
3881 location->objectid = 0;
3886 * when we hit a tree root in a directory, the btrfs part of the inode
3887 * needs to be changed to reflect the root directory of the tree root. This
3888 * is kind of like crossing a mount point.
3890 static int fixup_tree_root_location(struct btrfs_root *root,
3892 struct dentry *dentry,
3893 struct btrfs_key *location,
3894 struct btrfs_root **sub_root)
3896 struct btrfs_path *path;
3897 struct btrfs_root *new_root;
3898 struct btrfs_root_ref *ref;
3899 struct extent_buffer *leaf;
3903 path = btrfs_alloc_path();
3910 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3911 BTRFS_I(dir)->root->root_key.objectid,
3912 location->objectid);
3919 leaf = path->nodes[0];
3920 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3921 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3922 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3925 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3926 (unsigned long)(ref + 1),
3927 dentry->d_name.len);
3931 btrfs_release_path(path);
3933 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3934 if (IS_ERR(new_root)) {
3935 err = PTR_ERR(new_root);
3939 if (btrfs_root_refs(&new_root->root_item) == 0) {
3944 *sub_root = new_root;
3945 location->objectid = btrfs_root_dirid(&new_root->root_item);
3946 location->type = BTRFS_INODE_ITEM_KEY;
3947 location->offset = 0;
3950 btrfs_free_path(path);
3954 static void inode_tree_add(struct inode *inode)
3956 struct btrfs_root *root = BTRFS_I(inode)->root;
3957 struct btrfs_inode *entry;
3959 struct rb_node *parent;
3960 u64 ino = btrfs_ino(inode);
3962 p = &root->inode_tree.rb_node;
3965 if (inode_unhashed(inode))
3968 spin_lock(&root->inode_lock);
3971 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3973 if (ino < btrfs_ino(&entry->vfs_inode))
3974 p = &parent->rb_left;
3975 else if (ino > btrfs_ino(&entry->vfs_inode))
3976 p = &parent->rb_right;
3978 WARN_ON(!(entry->vfs_inode.i_state &
3979 (I_WILL_FREE | I_FREEING)));
3980 rb_erase(parent, &root->inode_tree);
3981 RB_CLEAR_NODE(parent);
3982 spin_unlock(&root->inode_lock);
3986 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3987 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3988 spin_unlock(&root->inode_lock);
3991 static void inode_tree_del(struct inode *inode)
3993 struct btrfs_root *root = BTRFS_I(inode)->root;
3996 spin_lock(&root->inode_lock);
3997 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3998 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3999 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4000 empty = RB_EMPTY_ROOT(&root->inode_tree);
4002 spin_unlock(&root->inode_lock);
4005 * Free space cache has inodes in the tree root, but the tree root has a
4006 * root_refs of 0, so this could end up dropping the tree root as a
4007 * snapshot, so we need the extra !root->fs_info->tree_root check to
4008 * make sure we don't drop it.
4010 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4011 root != root->fs_info->tree_root) {
4012 synchronize_srcu(&root->fs_info->subvol_srcu);
4013 spin_lock(&root->inode_lock);
4014 empty = RB_EMPTY_ROOT(&root->inode_tree);
4015 spin_unlock(&root->inode_lock);
4017 btrfs_add_dead_root(root);
4021 void btrfs_invalidate_inodes(struct btrfs_root *root)
4023 struct rb_node *node;
4024 struct rb_node *prev;
4025 struct btrfs_inode *entry;
4026 struct inode *inode;
4029 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4031 spin_lock(&root->inode_lock);
4033 node = root->inode_tree.rb_node;
4037 entry = rb_entry(node, struct btrfs_inode, rb_node);
4039 if (objectid < btrfs_ino(&entry->vfs_inode))
4040 node = node->rb_left;
4041 else if (objectid > btrfs_ino(&entry->vfs_inode))
4042 node = node->rb_right;
4048 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4049 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4053 prev = rb_next(prev);
4057 entry = rb_entry(node, struct btrfs_inode, rb_node);
4058 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4059 inode = igrab(&entry->vfs_inode);
4061 spin_unlock(&root->inode_lock);
4062 if (atomic_read(&inode->i_count) > 1)
4063 d_prune_aliases(inode);
4065 * btrfs_drop_inode will have it removed from
4066 * the inode cache when its usage count
4071 spin_lock(&root->inode_lock);
4075 if (cond_resched_lock(&root->inode_lock))
4078 node = rb_next(node);
4080 spin_unlock(&root->inode_lock);
4083 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4085 struct btrfs_iget_args *args = p;
4086 inode->i_ino = args->ino;
4087 BTRFS_I(inode)->root = args->root;
4091 static int btrfs_find_actor(struct inode *inode, void *opaque)
4093 struct btrfs_iget_args *args = opaque;
4094 return args->ino == btrfs_ino(inode) &&
4095 args->root == BTRFS_I(inode)->root;
4098 static struct inode *btrfs_iget_locked(struct super_block *s,
4100 struct btrfs_root *root)
4102 struct inode *inode;
4103 struct btrfs_iget_args args;
4104 args.ino = objectid;
4107 inode = iget5_locked(s, objectid, btrfs_find_actor,
4108 btrfs_init_locked_inode,
4113 /* Get an inode object given its location and corresponding root.
4114 * Returns in *is_new if the inode was read from disk
4116 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4117 struct btrfs_root *root, int *new)
4119 struct inode *inode;
4121 inode = btrfs_iget_locked(s, location->objectid, root);
4123 return ERR_PTR(-ENOMEM);
4125 if (inode->i_state & I_NEW) {
4126 BTRFS_I(inode)->root = root;
4127 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4128 btrfs_read_locked_inode(inode);
4129 if (!is_bad_inode(inode)) {
4130 inode_tree_add(inode);
4131 unlock_new_inode(inode);
4135 unlock_new_inode(inode);
4137 inode = ERR_PTR(-ESTALE);
4144 static struct inode *new_simple_dir(struct super_block *s,
4145 struct btrfs_key *key,
4146 struct btrfs_root *root)
4148 struct inode *inode = new_inode(s);
4151 return ERR_PTR(-ENOMEM);
4153 BTRFS_I(inode)->root = root;
4154 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4155 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4157 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4158 inode->i_op = &btrfs_dir_ro_inode_operations;
4159 inode->i_fop = &simple_dir_operations;
4160 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4161 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4166 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4168 struct inode *inode;
4169 struct btrfs_root *root = BTRFS_I(dir)->root;
4170 struct btrfs_root *sub_root = root;
4171 struct btrfs_key location;
4175 if (dentry->d_name.len > BTRFS_NAME_LEN)
4176 return ERR_PTR(-ENAMETOOLONG);
4178 if (unlikely(d_need_lookup(dentry))) {
4179 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4180 kfree(dentry->d_fsdata);
4181 dentry->d_fsdata = NULL;
4182 /* This thing is hashed, drop it for now */
4185 ret = btrfs_inode_by_name(dir, dentry, &location);
4189 return ERR_PTR(ret);
4191 if (location.objectid == 0)
4194 if (location.type == BTRFS_INODE_ITEM_KEY) {
4195 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4199 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4201 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4202 ret = fixup_tree_root_location(root, dir, dentry,
4203 &location, &sub_root);
4206 inode = ERR_PTR(ret);
4208 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4210 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4212 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4214 if (!IS_ERR(inode) && root != sub_root) {
4215 down_read(&root->fs_info->cleanup_work_sem);
4216 if (!(inode->i_sb->s_flags & MS_RDONLY))
4217 ret = btrfs_orphan_cleanup(sub_root);
4218 up_read(&root->fs_info->cleanup_work_sem);
4220 inode = ERR_PTR(ret);
4226 static int btrfs_dentry_delete(const struct dentry *dentry)
4228 struct btrfs_root *root;
4229 struct inode *inode = dentry->d_inode;
4231 if (!inode && !IS_ROOT(dentry))
4232 inode = dentry->d_parent->d_inode;
4235 root = BTRFS_I(inode)->root;
4236 if (btrfs_root_refs(&root->root_item) == 0)
4239 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4245 static void btrfs_dentry_release(struct dentry *dentry)
4247 if (dentry->d_fsdata)
4248 kfree(dentry->d_fsdata);
4251 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4256 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4257 if (unlikely(d_need_lookup(dentry))) {
4258 spin_lock(&dentry->d_lock);
4259 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4260 spin_unlock(&dentry->d_lock);
4265 unsigned char btrfs_filetype_table[] = {
4266 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4269 static int btrfs_real_readdir(struct file *filp, void *dirent,
4272 struct inode *inode = filp->f_dentry->d_inode;
4273 struct btrfs_root *root = BTRFS_I(inode)->root;
4274 struct btrfs_item *item;
4275 struct btrfs_dir_item *di;
4276 struct btrfs_key key;
4277 struct btrfs_key found_key;
4278 struct btrfs_path *path;
4279 struct list_head ins_list;
4280 struct list_head del_list;
4282 struct extent_buffer *leaf;
4284 unsigned char d_type;
4289 int key_type = BTRFS_DIR_INDEX_KEY;
4293 int is_curr = 0; /* filp->f_pos points to the current index? */
4295 /* FIXME, use a real flag for deciding about the key type */
4296 if (root->fs_info->tree_root == root)
4297 key_type = BTRFS_DIR_ITEM_KEY;
4299 /* special case for "." */
4300 if (filp->f_pos == 0) {
4301 over = filldir(dirent, ".", 1,
4302 filp->f_pos, btrfs_ino(inode), DT_DIR);
4307 /* special case for .., just use the back ref */
4308 if (filp->f_pos == 1) {
4309 u64 pino = parent_ino(filp->f_path.dentry);
4310 over = filldir(dirent, "..", 2,
4311 filp->f_pos, pino, DT_DIR);
4316 path = btrfs_alloc_path();
4322 if (key_type == BTRFS_DIR_INDEX_KEY) {
4323 INIT_LIST_HEAD(&ins_list);
4324 INIT_LIST_HEAD(&del_list);
4325 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4328 btrfs_set_key_type(&key, key_type);
4329 key.offset = filp->f_pos;
4330 key.objectid = btrfs_ino(inode);
4332 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4337 leaf = path->nodes[0];
4338 slot = path->slots[0];
4339 if (slot >= btrfs_header_nritems(leaf)) {
4340 ret = btrfs_next_leaf(root, path);
4348 item = btrfs_item_nr(leaf, slot);
4349 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4351 if (found_key.objectid != key.objectid)
4353 if (btrfs_key_type(&found_key) != key_type)
4355 if (found_key.offset < filp->f_pos)
4357 if (key_type == BTRFS_DIR_INDEX_KEY &&
4358 btrfs_should_delete_dir_index(&del_list,
4362 filp->f_pos = found_key.offset;
4365 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4367 di_total = btrfs_item_size(leaf, item);
4369 while (di_cur < di_total) {
4370 struct btrfs_key location;
4372 if (verify_dir_item(root, leaf, di))
4375 name_len = btrfs_dir_name_len(leaf, di);
4376 if (name_len <= sizeof(tmp_name)) {
4377 name_ptr = tmp_name;
4379 name_ptr = kmalloc(name_len, GFP_NOFS);
4385 read_extent_buffer(leaf, name_ptr,
4386 (unsigned long)(di + 1), name_len);
4388 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4389 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4392 /* is this a reference to our own snapshot? If so
4395 * In contrast to old kernels, we insert the snapshot's
4396 * dir item and dir index after it has been created, so
4397 * we won't find a reference to our own snapshot. We
4398 * still keep the following code for backward
4401 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4402 location.objectid == root->root_key.objectid) {
4406 over = filldir(dirent, name_ptr, name_len,
4407 found_key.offset, location.objectid,
4411 if (name_ptr != tmp_name)
4416 di_len = btrfs_dir_name_len(leaf, di) +
4417 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4419 di = (struct btrfs_dir_item *)((char *)di + di_len);
4425 if (key_type == BTRFS_DIR_INDEX_KEY) {
4428 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4434 /* Reached end of directory/root. Bump pos past the last item. */
4435 if (key_type == BTRFS_DIR_INDEX_KEY)
4437 * 32-bit glibc will use getdents64, but then strtol -
4438 * so the last number we can serve is this.
4440 filp->f_pos = 0x7fffffff;
4446 if (key_type == BTRFS_DIR_INDEX_KEY)
4447 btrfs_put_delayed_items(&ins_list, &del_list);
4448 btrfs_free_path(path);
4452 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4454 struct btrfs_root *root = BTRFS_I(inode)->root;
4455 struct btrfs_trans_handle *trans;
4457 bool nolock = false;
4459 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4462 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4465 if (wbc->sync_mode == WB_SYNC_ALL) {
4467 trans = btrfs_join_transaction_nolock(root);
4469 trans = btrfs_join_transaction(root);
4471 return PTR_ERR(trans);
4473 ret = btrfs_end_transaction_nolock(trans, root);
4475 ret = btrfs_commit_transaction(trans, root);
4481 * This is somewhat expensive, updating the tree every time the
4482 * inode changes. But, it is most likely to find the inode in cache.
4483 * FIXME, needs more benchmarking...there are no reasons other than performance
4484 * to keep or drop this code.
4486 int btrfs_dirty_inode(struct inode *inode)
4488 struct btrfs_root *root = BTRFS_I(inode)->root;
4489 struct btrfs_trans_handle *trans;
4492 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4495 trans = btrfs_join_transaction(root);
4497 return PTR_ERR(trans);
4499 ret = btrfs_update_inode(trans, root, inode);
4500 if (ret && ret == -ENOSPC) {
4501 /* whoops, lets try again with the full transaction */
4502 btrfs_end_transaction(trans, root);
4503 trans = btrfs_start_transaction(root, 1);
4505 return PTR_ERR(trans);
4507 ret = btrfs_update_inode(trans, root, inode);
4509 btrfs_end_transaction(trans, root);
4510 if (BTRFS_I(inode)->delayed_node)
4511 btrfs_balance_delayed_items(root);
4517 * This is a copy of file_update_time. We need this so we can return error on
4518 * ENOSPC for updating the inode in the case of file write and mmap writes.
4520 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4523 struct btrfs_root *root = BTRFS_I(inode)->root;
4525 if (btrfs_root_readonly(root))
4528 if (flags & S_VERSION)
4529 inode_inc_iversion(inode);
4530 if (flags & S_CTIME)
4531 inode->i_ctime = *now;
4532 if (flags & S_MTIME)
4533 inode->i_mtime = *now;
4534 if (flags & S_ATIME)
4535 inode->i_atime = *now;
4536 return btrfs_dirty_inode(inode);
4540 * find the highest existing sequence number in a directory
4541 * and then set the in-memory index_cnt variable to reflect
4542 * free sequence numbers
4544 static int btrfs_set_inode_index_count(struct inode *inode)
4546 struct btrfs_root *root = BTRFS_I(inode)->root;
4547 struct btrfs_key key, found_key;
4548 struct btrfs_path *path;
4549 struct extent_buffer *leaf;
4552 key.objectid = btrfs_ino(inode);
4553 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4554 key.offset = (u64)-1;
4556 path = btrfs_alloc_path();
4560 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4563 /* FIXME: we should be able to handle this */
4569 * MAGIC NUMBER EXPLANATION:
4570 * since we search a directory based on f_pos we have to start at 2
4571 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4572 * else has to start at 2
4574 if (path->slots[0] == 0) {
4575 BTRFS_I(inode)->index_cnt = 2;
4581 leaf = path->nodes[0];
4582 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4584 if (found_key.objectid != btrfs_ino(inode) ||
4585 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4586 BTRFS_I(inode)->index_cnt = 2;
4590 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4592 btrfs_free_path(path);
4597 * helper to find a free sequence number in a given directory. This current
4598 * code is very simple, later versions will do smarter things in the btree
4600 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4604 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4605 ret = btrfs_inode_delayed_dir_index_count(dir);
4607 ret = btrfs_set_inode_index_count(dir);
4613 *index = BTRFS_I(dir)->index_cnt;
4614 BTRFS_I(dir)->index_cnt++;
4619 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4620 struct btrfs_root *root,
4622 const char *name, int name_len,
4623 u64 ref_objectid, u64 objectid,
4624 umode_t mode, u64 *index)
4626 struct inode *inode;
4627 struct btrfs_inode_item *inode_item;
4628 struct btrfs_key *location;
4629 struct btrfs_path *path;
4630 struct btrfs_inode_ref *ref;
4631 struct btrfs_key key[2];
4637 path = btrfs_alloc_path();
4639 return ERR_PTR(-ENOMEM);
4641 inode = new_inode(root->fs_info->sb);
4643 btrfs_free_path(path);
4644 return ERR_PTR(-ENOMEM);
4648 * we have to initialize this early, so we can reclaim the inode
4649 * number if we fail afterwards in this function.
4651 inode->i_ino = objectid;
4654 trace_btrfs_inode_request(dir);
4656 ret = btrfs_set_inode_index(dir, index);
4658 btrfs_free_path(path);
4660 return ERR_PTR(ret);
4664 * index_cnt is ignored for everything but a dir,
4665 * btrfs_get_inode_index_count has an explanation for the magic
4668 BTRFS_I(inode)->index_cnt = 2;
4669 BTRFS_I(inode)->root = root;
4670 BTRFS_I(inode)->generation = trans->transid;
4671 inode->i_generation = BTRFS_I(inode)->generation;
4678 key[0].objectid = objectid;
4679 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4682 key[1].objectid = objectid;
4683 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4684 key[1].offset = ref_objectid;
4686 sizes[0] = sizeof(struct btrfs_inode_item);
4687 sizes[1] = name_len + sizeof(*ref);
4689 path->leave_spinning = 1;
4690 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4694 inode_init_owner(inode, dir, mode);
4695 inode_set_bytes(inode, 0);
4696 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4697 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4698 struct btrfs_inode_item);
4699 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4700 sizeof(*inode_item));
4701 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4703 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4704 struct btrfs_inode_ref);
4705 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4706 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4707 ptr = (unsigned long)(ref + 1);
4708 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4710 btrfs_mark_buffer_dirty(path->nodes[0]);
4711 btrfs_free_path(path);
4713 location = &BTRFS_I(inode)->location;
4714 location->objectid = objectid;
4715 location->offset = 0;
4716 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4718 btrfs_inherit_iflags(inode, dir);
4720 if (S_ISREG(mode)) {
4721 if (btrfs_test_opt(root, NODATASUM))
4722 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4723 if (btrfs_test_opt(root, NODATACOW) ||
4724 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4725 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4728 insert_inode_hash(inode);
4729 inode_tree_add(inode);
4731 trace_btrfs_inode_new(inode);
4732 btrfs_set_inode_last_trans(trans, inode);
4734 btrfs_update_root_times(trans, root);
4739 BTRFS_I(dir)->index_cnt--;
4740 btrfs_free_path(path);
4742 return ERR_PTR(ret);
4745 static inline u8 btrfs_inode_type(struct inode *inode)
4747 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4751 * utility function to add 'inode' into 'parent_inode' with
4752 * a give name and a given sequence number.
4753 * if 'add_backref' is true, also insert a backref from the
4754 * inode to the parent directory.
4756 int btrfs_add_link(struct btrfs_trans_handle *trans,
4757 struct inode *parent_inode, struct inode *inode,
4758 const char *name, int name_len, int add_backref, u64 index)
4761 struct btrfs_key key;
4762 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4763 u64 ino = btrfs_ino(inode);
4764 u64 parent_ino = btrfs_ino(parent_inode);
4766 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4767 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4770 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4774 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4775 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4776 key.objectid, root->root_key.objectid,
4777 parent_ino, index, name, name_len);
4778 } else if (add_backref) {
4779 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4783 /* Nothing to clean up yet */
4787 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4789 btrfs_inode_type(inode), index);
4793 btrfs_abort_transaction(trans, root, ret);
4797 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4799 inode_inc_iversion(parent_inode);
4800 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4801 ret = btrfs_update_inode(trans, root, parent_inode);
4803 btrfs_abort_transaction(trans, root, ret);
4807 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4810 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4811 key.objectid, root->root_key.objectid,
4812 parent_ino, &local_index, name, name_len);
4814 } else if (add_backref) {
4818 err = btrfs_del_inode_ref(trans, root, name, name_len,
4819 ino, parent_ino, &local_index);
4824 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4825 struct inode *dir, struct dentry *dentry,
4826 struct inode *inode, int backref, u64 index)
4828 int err = btrfs_add_link(trans, dir, inode,
4829 dentry->d_name.name, dentry->d_name.len,
4836 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4837 umode_t mode, dev_t rdev)
4839 struct btrfs_trans_handle *trans;
4840 struct btrfs_root *root = BTRFS_I(dir)->root;
4841 struct inode *inode = NULL;
4845 unsigned long nr = 0;
4848 if (!new_valid_dev(rdev))
4852 * 2 for inode item and ref
4854 * 1 for xattr if selinux is on
4856 trans = btrfs_start_transaction(root, 5);
4858 return PTR_ERR(trans);
4860 err = btrfs_find_free_ino(root, &objectid);
4864 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4865 dentry->d_name.len, btrfs_ino(dir), objectid,
4867 if (IS_ERR(inode)) {
4868 err = PTR_ERR(inode);
4872 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4879 * If the active LSM wants to access the inode during
4880 * d_instantiate it needs these. Smack checks to see
4881 * if the filesystem supports xattrs by looking at the
4885 inode->i_op = &btrfs_special_inode_operations;
4886 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4890 init_special_inode(inode, inode->i_mode, rdev);
4891 btrfs_update_inode(trans, root, inode);
4892 d_instantiate(dentry, inode);
4895 nr = trans->blocks_used;
4896 btrfs_end_transaction(trans, root);
4897 btrfs_btree_balance_dirty(root, nr);
4899 inode_dec_link_count(inode);
4905 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4906 umode_t mode, bool excl)
4908 struct btrfs_trans_handle *trans;
4909 struct btrfs_root *root = BTRFS_I(dir)->root;
4910 struct inode *inode = NULL;
4913 unsigned long nr = 0;
4918 * 2 for inode item and ref
4920 * 1 for xattr if selinux is on
4922 trans = btrfs_start_transaction(root, 5);
4924 return PTR_ERR(trans);
4926 err = btrfs_find_free_ino(root, &objectid);
4930 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4931 dentry->d_name.len, btrfs_ino(dir), objectid,
4933 if (IS_ERR(inode)) {
4934 err = PTR_ERR(inode);
4938 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4945 * If the active LSM wants to access the inode during
4946 * d_instantiate it needs these. Smack checks to see
4947 * if the filesystem supports xattrs by looking at the
4950 inode->i_fop = &btrfs_file_operations;
4951 inode->i_op = &btrfs_file_inode_operations;
4953 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4957 inode->i_mapping->a_ops = &btrfs_aops;
4958 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4959 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4960 d_instantiate(dentry, inode);
4963 nr = trans->blocks_used;
4964 btrfs_end_transaction(trans, root);
4966 inode_dec_link_count(inode);
4969 btrfs_btree_balance_dirty(root, nr);
4973 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4974 struct dentry *dentry)
4976 struct btrfs_trans_handle *trans;
4977 struct btrfs_root *root = BTRFS_I(dir)->root;
4978 struct inode *inode = old_dentry->d_inode;
4980 unsigned long nr = 0;
4984 /* do not allow sys_link's with other subvols of the same device */
4985 if (root->objectid != BTRFS_I(inode)->root->objectid)
4988 if (inode->i_nlink == ~0U)
4991 err = btrfs_set_inode_index(dir, &index);
4996 * 2 items for inode and inode ref
4997 * 2 items for dir items
4998 * 1 item for parent inode
5000 trans = btrfs_start_transaction(root, 5);
5001 if (IS_ERR(trans)) {
5002 err = PTR_ERR(trans);
5006 btrfs_inc_nlink(inode);
5007 inode_inc_iversion(inode);
5008 inode->i_ctime = CURRENT_TIME;
5011 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5016 struct dentry *parent = dentry->d_parent;
5017 err = btrfs_update_inode(trans, root, inode);
5020 d_instantiate(dentry, inode);
5021 btrfs_log_new_name(trans, inode, NULL, parent);
5024 nr = trans->blocks_used;
5025 btrfs_end_transaction(trans, root);
5028 inode_dec_link_count(inode);
5031 btrfs_btree_balance_dirty(root, nr);
5035 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5037 struct inode *inode = NULL;
5038 struct btrfs_trans_handle *trans;
5039 struct btrfs_root *root = BTRFS_I(dir)->root;
5041 int drop_on_err = 0;
5044 unsigned long nr = 1;
5047 * 2 items for inode and ref
5048 * 2 items for dir items
5049 * 1 for xattr if selinux is on
5051 trans = btrfs_start_transaction(root, 5);
5053 return PTR_ERR(trans);
5055 err = btrfs_find_free_ino(root, &objectid);
5059 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5060 dentry->d_name.len, btrfs_ino(dir), objectid,
5061 S_IFDIR | mode, &index);
5062 if (IS_ERR(inode)) {
5063 err = PTR_ERR(inode);
5069 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5073 inode->i_op = &btrfs_dir_inode_operations;
5074 inode->i_fop = &btrfs_dir_file_operations;
5076 btrfs_i_size_write(inode, 0);
5077 err = btrfs_update_inode(trans, root, inode);
5081 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5082 dentry->d_name.len, 0, index);
5086 d_instantiate(dentry, inode);
5090 nr = trans->blocks_used;
5091 btrfs_end_transaction(trans, root);
5094 btrfs_btree_balance_dirty(root, nr);
5098 /* helper for btfs_get_extent. Given an existing extent in the tree,
5099 * and an extent that you want to insert, deal with overlap and insert
5100 * the new extent into the tree.
5102 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5103 struct extent_map *existing,
5104 struct extent_map *em,
5105 u64 map_start, u64 map_len)
5109 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5110 start_diff = map_start - em->start;
5111 em->start = map_start;
5113 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5114 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5115 em->block_start += start_diff;
5116 em->block_len -= start_diff;
5118 return add_extent_mapping(em_tree, em);
5121 static noinline int uncompress_inline(struct btrfs_path *path,
5122 struct inode *inode, struct page *page,
5123 size_t pg_offset, u64 extent_offset,
5124 struct btrfs_file_extent_item *item)
5127 struct extent_buffer *leaf = path->nodes[0];
5130 unsigned long inline_size;
5134 WARN_ON(pg_offset != 0);
5135 compress_type = btrfs_file_extent_compression(leaf, item);
5136 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5137 inline_size = btrfs_file_extent_inline_item_len(leaf,
5138 btrfs_item_nr(leaf, path->slots[0]));
5139 tmp = kmalloc(inline_size, GFP_NOFS);
5142 ptr = btrfs_file_extent_inline_start(item);
5144 read_extent_buffer(leaf, tmp, ptr, inline_size);
5146 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5147 ret = btrfs_decompress(compress_type, tmp, page,
5148 extent_offset, inline_size, max_size);
5150 char *kaddr = kmap_atomic(page);
5151 unsigned long copy_size = min_t(u64,
5152 PAGE_CACHE_SIZE - pg_offset,
5153 max_size - extent_offset);
5154 memset(kaddr + pg_offset, 0, copy_size);
5155 kunmap_atomic(kaddr);
5162 * a bit scary, this does extent mapping from logical file offset to the disk.
5163 * the ugly parts come from merging extents from the disk with the in-ram
5164 * representation. This gets more complex because of the data=ordered code,
5165 * where the in-ram extents might be locked pending data=ordered completion.
5167 * This also copies inline extents directly into the page.
5170 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5171 size_t pg_offset, u64 start, u64 len,
5177 u64 extent_start = 0;
5179 u64 objectid = btrfs_ino(inode);
5181 struct btrfs_path *path = NULL;
5182 struct btrfs_root *root = BTRFS_I(inode)->root;
5183 struct btrfs_file_extent_item *item;
5184 struct extent_buffer *leaf;
5185 struct btrfs_key found_key;
5186 struct extent_map *em = NULL;
5187 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5188 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5189 struct btrfs_trans_handle *trans = NULL;
5193 read_lock(&em_tree->lock);
5194 em = lookup_extent_mapping(em_tree, start, len);
5196 em->bdev = root->fs_info->fs_devices->latest_bdev;
5197 read_unlock(&em_tree->lock);
5200 if (em->start > start || em->start + em->len <= start)
5201 free_extent_map(em);
5202 else if (em->block_start == EXTENT_MAP_INLINE && page)
5203 free_extent_map(em);
5207 em = alloc_extent_map();
5212 em->bdev = root->fs_info->fs_devices->latest_bdev;
5213 em->start = EXTENT_MAP_HOLE;
5214 em->orig_start = EXTENT_MAP_HOLE;
5216 em->block_len = (u64)-1;
5219 path = btrfs_alloc_path();
5225 * Chances are we'll be called again, so go ahead and do
5231 ret = btrfs_lookup_file_extent(trans, root, path,
5232 objectid, start, trans != NULL);
5239 if (path->slots[0] == 0)
5244 leaf = path->nodes[0];
5245 item = btrfs_item_ptr(leaf, path->slots[0],
5246 struct btrfs_file_extent_item);
5247 /* are we inside the extent that was found? */
5248 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5249 found_type = btrfs_key_type(&found_key);
5250 if (found_key.objectid != objectid ||
5251 found_type != BTRFS_EXTENT_DATA_KEY) {
5255 found_type = btrfs_file_extent_type(leaf, item);
5256 extent_start = found_key.offset;
5257 compress_type = btrfs_file_extent_compression(leaf, item);
5258 if (found_type == BTRFS_FILE_EXTENT_REG ||
5259 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5260 extent_end = extent_start +
5261 btrfs_file_extent_num_bytes(leaf, item);
5262 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5264 size = btrfs_file_extent_inline_len(leaf, item);
5265 extent_end = (extent_start + size + root->sectorsize - 1) &
5266 ~((u64)root->sectorsize - 1);
5269 if (start >= extent_end) {
5271 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5272 ret = btrfs_next_leaf(root, path);
5279 leaf = path->nodes[0];
5281 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5282 if (found_key.objectid != objectid ||
5283 found_key.type != BTRFS_EXTENT_DATA_KEY)
5285 if (start + len <= found_key.offset)
5288 em->len = found_key.offset - start;
5292 if (found_type == BTRFS_FILE_EXTENT_REG ||
5293 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5294 em->start = extent_start;
5295 em->len = extent_end - extent_start;
5296 em->orig_start = extent_start -
5297 btrfs_file_extent_offset(leaf, item);
5298 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5300 em->block_start = EXTENT_MAP_HOLE;
5303 if (compress_type != BTRFS_COMPRESS_NONE) {
5304 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5305 em->compress_type = compress_type;
5306 em->block_start = bytenr;
5307 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5310 bytenr += btrfs_file_extent_offset(leaf, item);
5311 em->block_start = bytenr;
5312 em->block_len = em->len;
5313 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5314 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5317 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5321 size_t extent_offset;
5324 em->block_start = EXTENT_MAP_INLINE;
5325 if (!page || create) {
5326 em->start = extent_start;
5327 em->len = extent_end - extent_start;
5331 size = btrfs_file_extent_inline_len(leaf, item);
5332 extent_offset = page_offset(page) + pg_offset - extent_start;
5333 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5334 size - extent_offset);
5335 em->start = extent_start + extent_offset;
5336 em->len = (copy_size + root->sectorsize - 1) &
5337 ~((u64)root->sectorsize - 1);
5338 em->orig_start = EXTENT_MAP_INLINE;
5339 if (compress_type) {
5340 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5341 em->compress_type = compress_type;
5343 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5344 if (create == 0 && !PageUptodate(page)) {
5345 if (btrfs_file_extent_compression(leaf, item) !=
5346 BTRFS_COMPRESS_NONE) {
5347 ret = uncompress_inline(path, inode, page,
5349 extent_offset, item);
5350 BUG_ON(ret); /* -ENOMEM */
5353 read_extent_buffer(leaf, map + pg_offset, ptr,
5355 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5356 memset(map + pg_offset + copy_size, 0,
5357 PAGE_CACHE_SIZE - pg_offset -
5362 flush_dcache_page(page);
5363 } else if (create && PageUptodate(page)) {
5367 free_extent_map(em);
5370 btrfs_release_path(path);
5371 trans = btrfs_join_transaction(root);
5374 return ERR_CAST(trans);
5378 write_extent_buffer(leaf, map + pg_offset, ptr,
5381 btrfs_mark_buffer_dirty(leaf);
5383 set_extent_uptodate(io_tree, em->start,
5384 extent_map_end(em) - 1, NULL, GFP_NOFS);
5387 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5394 em->block_start = EXTENT_MAP_HOLE;
5395 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5397 btrfs_release_path(path);
5398 if (em->start > start || extent_map_end(em) <= start) {
5399 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5400 "[%llu %llu]\n", (unsigned long long)em->start,
5401 (unsigned long long)em->len,
5402 (unsigned long long)start,
5403 (unsigned long long)len);
5409 write_lock(&em_tree->lock);
5410 ret = add_extent_mapping(em_tree, em);
5411 /* it is possible that someone inserted the extent into the tree
5412 * while we had the lock dropped. It is also possible that
5413 * an overlapping map exists in the tree
5415 if (ret == -EEXIST) {
5416 struct extent_map *existing;
5420 existing = lookup_extent_mapping(em_tree, start, len);
5421 if (existing && (existing->start > start ||
5422 existing->start + existing->len <= start)) {
5423 free_extent_map(existing);
5427 existing = lookup_extent_mapping(em_tree, em->start,
5430 err = merge_extent_mapping(em_tree, existing,
5433 free_extent_map(existing);
5435 free_extent_map(em);
5440 free_extent_map(em);
5444 free_extent_map(em);
5449 write_unlock(&em_tree->lock);
5452 trace_btrfs_get_extent(root, em);
5455 btrfs_free_path(path);
5457 ret = btrfs_end_transaction(trans, root);
5462 free_extent_map(em);
5463 return ERR_PTR(err);
5465 BUG_ON(!em); /* Error is always set */
5469 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5470 size_t pg_offset, u64 start, u64 len,
5473 struct extent_map *em;
5474 struct extent_map *hole_em = NULL;
5475 u64 range_start = start;
5481 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5486 * if our em maps to a hole, there might
5487 * actually be delalloc bytes behind it
5489 if (em->block_start != EXTENT_MAP_HOLE)
5495 /* check to see if we've wrapped (len == -1 or similar) */
5504 /* ok, we didn't find anything, lets look for delalloc */
5505 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5506 end, len, EXTENT_DELALLOC, 1);
5507 found_end = range_start + found;
5508 if (found_end < range_start)
5509 found_end = (u64)-1;
5512 * we didn't find anything useful, return
5513 * the original results from get_extent()
5515 if (range_start > end || found_end <= start) {
5521 /* adjust the range_start to make sure it doesn't
5522 * go backwards from the start they passed in
5524 range_start = max(start,range_start);
5525 found = found_end - range_start;
5528 u64 hole_start = start;
5531 em = alloc_extent_map();
5537 * when btrfs_get_extent can't find anything it
5538 * returns one huge hole
5540 * make sure what it found really fits our range, and
5541 * adjust to make sure it is based on the start from
5545 u64 calc_end = extent_map_end(hole_em);
5547 if (calc_end <= start || (hole_em->start > end)) {
5548 free_extent_map(hole_em);
5551 hole_start = max(hole_em->start, start);
5552 hole_len = calc_end - hole_start;
5556 if (hole_em && range_start > hole_start) {
5557 /* our hole starts before our delalloc, so we
5558 * have to return just the parts of the hole
5559 * that go until the delalloc starts
5561 em->len = min(hole_len,
5562 range_start - hole_start);
5563 em->start = hole_start;
5564 em->orig_start = hole_start;
5566 * don't adjust block start at all,
5567 * it is fixed at EXTENT_MAP_HOLE
5569 em->block_start = hole_em->block_start;
5570 em->block_len = hole_len;
5572 em->start = range_start;
5574 em->orig_start = range_start;
5575 em->block_start = EXTENT_MAP_DELALLOC;
5576 em->block_len = found;
5578 } else if (hole_em) {
5583 free_extent_map(hole_em);
5585 free_extent_map(em);
5586 return ERR_PTR(err);
5591 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5592 struct extent_map *em,
5595 struct btrfs_root *root = BTRFS_I(inode)->root;
5596 struct btrfs_trans_handle *trans;
5597 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5598 struct btrfs_key ins;
5601 bool insert = false;
5604 * Ok if the extent map we looked up is a hole and is for the exact
5605 * range we want, there is no reason to allocate a new one, however if
5606 * it is not right then we need to free this one and drop the cache for
5609 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5611 free_extent_map(em);
5614 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5617 trans = btrfs_join_transaction(root);
5619 return ERR_CAST(trans);
5621 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5622 btrfs_add_inode_defrag(trans, inode);
5624 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5626 alloc_hint = get_extent_allocation_hint(inode, start, len);
5627 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5628 alloc_hint, &ins, 1);
5635 em = alloc_extent_map();
5637 em = ERR_PTR(-ENOMEM);
5643 em->orig_start = em->start;
5644 em->len = ins.offset;
5646 em->block_start = ins.objectid;
5647 em->block_len = ins.offset;
5648 em->bdev = root->fs_info->fs_devices->latest_bdev;
5651 * We need to do this because if we're using the original em we searched
5652 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5655 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5658 write_lock(&em_tree->lock);
5659 ret = add_extent_mapping(em_tree, em);
5660 write_unlock(&em_tree->lock);
5663 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5666 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5667 ins.offset, ins.offset, 0);
5669 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5673 btrfs_end_transaction(trans, root);
5678 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5679 * block must be cow'd
5681 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5682 struct inode *inode, u64 offset, u64 len)
5684 struct btrfs_path *path;
5686 struct extent_buffer *leaf;
5687 struct btrfs_root *root = BTRFS_I(inode)->root;
5688 struct btrfs_file_extent_item *fi;
5689 struct btrfs_key key;
5697 path = btrfs_alloc_path();
5701 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5706 slot = path->slots[0];
5709 /* can't find the item, must cow */
5716 leaf = path->nodes[0];
5717 btrfs_item_key_to_cpu(leaf, &key, slot);
5718 if (key.objectid != btrfs_ino(inode) ||
5719 key.type != BTRFS_EXTENT_DATA_KEY) {
5720 /* not our file or wrong item type, must cow */
5724 if (key.offset > offset) {
5725 /* Wrong offset, must cow */
5729 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5730 found_type = btrfs_file_extent_type(leaf, fi);
5731 if (found_type != BTRFS_FILE_EXTENT_REG &&
5732 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5733 /* not a regular extent, must cow */
5736 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5737 backref_offset = btrfs_file_extent_offset(leaf, fi);
5739 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5740 if (extent_end < offset + len) {
5741 /* extent doesn't include our full range, must cow */
5745 if (btrfs_extent_readonly(root, disk_bytenr))
5749 * look for other files referencing this extent, if we
5750 * find any we must cow
5752 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5753 key.offset - backref_offset, disk_bytenr))
5757 * adjust disk_bytenr and num_bytes to cover just the bytes
5758 * in this extent we are about to write. If there
5759 * are any csums in that range we have to cow in order
5760 * to keep the csums correct
5762 disk_bytenr += backref_offset;
5763 disk_bytenr += offset - key.offset;
5764 num_bytes = min(offset + len, extent_end) - offset;
5765 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5768 * all of the above have passed, it is safe to overwrite this extent
5773 btrfs_free_path(path);
5777 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5778 struct buffer_head *bh_result, int create)
5780 struct extent_map *em;
5781 struct btrfs_root *root = BTRFS_I(inode)->root;
5782 u64 start = iblock << inode->i_blkbits;
5783 u64 len = bh_result->b_size;
5784 struct btrfs_trans_handle *trans;
5786 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5791 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5792 * io. INLINE is special, and we could probably kludge it in here, but
5793 * it's still buffered so for safety lets just fall back to the generic
5796 * For COMPRESSED we _have_ to read the entire extent in so we can
5797 * decompress it, so there will be buffering required no matter what we
5798 * do, so go ahead and fallback to buffered.
5800 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5801 * to buffered IO. Don't blame me, this is the price we pay for using
5804 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5805 em->block_start == EXTENT_MAP_INLINE) {
5806 free_extent_map(em);
5810 /* Just a good old fashioned hole, return */
5811 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5812 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5813 free_extent_map(em);
5814 /* DIO will do one hole at a time, so just unlock a sector */
5815 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5816 start + root->sectorsize - 1);
5821 * We don't allocate a new extent in the following cases
5823 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5825 * 2) The extent is marked as PREALLOC. We're good to go here and can
5826 * just use the extent.
5830 len = em->len - (start - em->start);
5834 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5835 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5836 em->block_start != EXTENT_MAP_HOLE)) {
5841 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5842 type = BTRFS_ORDERED_PREALLOC;
5844 type = BTRFS_ORDERED_NOCOW;
5845 len = min(len, em->len - (start - em->start));
5846 block_start = em->block_start + (start - em->start);
5849 * we're not going to log anything, but we do need
5850 * to make sure the current transaction stays open
5851 * while we look for nocow cross refs
5853 trans = btrfs_join_transaction(root);
5857 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5858 ret = btrfs_add_ordered_extent_dio(inode, start,
5859 block_start, len, len, type);
5860 btrfs_end_transaction(trans, root);
5862 free_extent_map(em);
5867 btrfs_end_transaction(trans, root);
5871 * this will cow the extent, reset the len in case we changed
5874 len = bh_result->b_size;
5875 em = btrfs_new_extent_direct(inode, em, start, len);
5878 len = min(len, em->len - (start - em->start));
5880 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5881 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5884 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5886 bh_result->b_size = len;
5887 bh_result->b_bdev = em->bdev;
5888 set_buffer_mapped(bh_result);
5890 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5891 set_buffer_new(bh_result);
5894 * Need to update the i_size under the extent lock so buffered
5895 * readers will get the updated i_size when we unlock.
5897 if (start + len > i_size_read(inode))
5898 i_size_write(inode, start + len);
5901 free_extent_map(em);
5906 struct btrfs_dio_private {
5907 struct inode *inode;
5914 /* number of bios pending for this dio */
5915 atomic_t pending_bios;
5920 struct bio *orig_bio;
5923 static void btrfs_endio_direct_read(struct bio *bio, int err)
5925 struct btrfs_dio_private *dip = bio->bi_private;
5926 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5927 struct bio_vec *bvec = bio->bi_io_vec;
5928 struct inode *inode = dip->inode;
5929 struct btrfs_root *root = BTRFS_I(inode)->root;
5931 u32 *private = dip->csums;
5933 start = dip->logical_offset;
5935 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5936 struct page *page = bvec->bv_page;
5939 unsigned long flags;
5941 local_irq_save(flags);
5942 kaddr = kmap_atomic(page);
5943 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5944 csum, bvec->bv_len);
5945 btrfs_csum_final(csum, (char *)&csum);
5946 kunmap_atomic(kaddr);
5947 local_irq_restore(flags);
5949 flush_dcache_page(bvec->bv_page);
5950 if (csum != *private) {
5951 printk(KERN_ERR "btrfs csum failed ino %llu off"
5952 " %llu csum %u private %u\n",
5953 (unsigned long long)btrfs_ino(inode),
5954 (unsigned long long)start,
5960 start += bvec->bv_len;
5963 } while (bvec <= bvec_end);
5965 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5966 dip->logical_offset + dip->bytes - 1);
5967 bio->bi_private = dip->private;
5972 /* If we had a csum failure make sure to clear the uptodate flag */
5974 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5975 dio_end_io(bio, err);
5978 static void btrfs_endio_direct_write(struct bio *bio, int err)
5980 struct btrfs_dio_private *dip = bio->bi_private;
5981 struct inode *inode = dip->inode;
5982 struct btrfs_root *root = BTRFS_I(inode)->root;
5983 struct btrfs_ordered_extent *ordered = NULL;
5984 u64 ordered_offset = dip->logical_offset;
5985 u64 ordered_bytes = dip->bytes;
5991 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5993 ordered_bytes, !err);
5997 ordered->work.func = finish_ordered_fn;
5998 ordered->work.flags = 0;
5999 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6003 * our bio might span multiple ordered extents. If we haven't
6004 * completed the accounting for the whole dio, go back and try again
6006 if (ordered_offset < dip->logical_offset + dip->bytes) {
6007 ordered_bytes = dip->logical_offset + dip->bytes -
6013 bio->bi_private = dip->private;
6017 /* If we had an error make sure to clear the uptodate flag */
6019 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6020 dio_end_io(bio, err);
6023 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6024 struct bio *bio, int mirror_num,
6025 unsigned long bio_flags, u64 offset)
6028 struct btrfs_root *root = BTRFS_I(inode)->root;
6029 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6030 BUG_ON(ret); /* -ENOMEM */
6034 static void btrfs_end_dio_bio(struct bio *bio, int err)
6036 struct btrfs_dio_private *dip = bio->bi_private;
6039 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6040 "sector %#Lx len %u err no %d\n",
6041 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6042 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6046 * before atomic variable goto zero, we must make sure
6047 * dip->errors is perceived to be set.
6049 smp_mb__before_atomic_dec();
6052 /* if there are more bios still pending for this dio, just exit */
6053 if (!atomic_dec_and_test(&dip->pending_bios))
6057 bio_io_error(dip->orig_bio);
6059 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6060 bio_endio(dip->orig_bio, 0);
6066 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6067 u64 first_sector, gfp_t gfp_flags)
6069 int nr_vecs = bio_get_nr_vecs(bdev);
6070 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6073 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6074 int rw, u64 file_offset, int skip_sum,
6075 u32 *csums, int async_submit)
6077 int write = rw & REQ_WRITE;
6078 struct btrfs_root *root = BTRFS_I(inode)->root;
6084 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6092 if (write && async_submit) {
6093 ret = btrfs_wq_submit_bio(root->fs_info,
6094 inode, rw, bio, 0, 0,
6096 __btrfs_submit_bio_start_direct_io,
6097 __btrfs_submit_bio_done);
6101 * If we aren't doing async submit, calculate the csum of the
6104 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6107 } else if (!skip_sum) {
6108 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6109 file_offset, csums);
6115 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6121 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6124 struct inode *inode = dip->inode;
6125 struct btrfs_root *root = BTRFS_I(inode)->root;
6126 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6128 struct bio *orig_bio = dip->orig_bio;
6129 struct bio_vec *bvec = orig_bio->bi_io_vec;
6130 u64 start_sector = orig_bio->bi_sector;
6131 u64 file_offset = dip->logical_offset;
6135 u32 *csums = dip->csums;
6137 int async_submit = 0;
6138 int write = rw & REQ_WRITE;
6140 map_length = orig_bio->bi_size;
6141 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6142 &map_length, NULL, 0);
6148 if (map_length >= orig_bio->bi_size) {
6154 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6157 bio->bi_private = dip;
6158 bio->bi_end_io = btrfs_end_dio_bio;
6159 atomic_inc(&dip->pending_bios);
6161 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6162 if (unlikely(map_length < submit_len + bvec->bv_len ||
6163 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6164 bvec->bv_offset) < bvec->bv_len)) {
6166 * inc the count before we submit the bio so
6167 * we know the end IO handler won't happen before
6168 * we inc the count. Otherwise, the dip might get freed
6169 * before we're done setting it up
6171 atomic_inc(&dip->pending_bios);
6172 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6173 file_offset, skip_sum,
6174 csums, async_submit);
6177 atomic_dec(&dip->pending_bios);
6181 /* Write's use the ordered csums */
6182 if (!write && !skip_sum)
6183 csums = csums + nr_pages;
6184 start_sector += submit_len >> 9;
6185 file_offset += submit_len;
6190 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6191 start_sector, GFP_NOFS);
6194 bio->bi_private = dip;
6195 bio->bi_end_io = btrfs_end_dio_bio;
6197 map_length = orig_bio->bi_size;
6198 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6199 &map_length, NULL, 0);
6205 submit_len += bvec->bv_len;
6212 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6213 csums, async_submit);
6221 * before atomic variable goto zero, we must
6222 * make sure dip->errors is perceived to be set.
6224 smp_mb__before_atomic_dec();
6225 if (atomic_dec_and_test(&dip->pending_bios))
6226 bio_io_error(dip->orig_bio);
6228 /* bio_end_io() will handle error, so we needn't return it */
6232 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6235 struct btrfs_root *root = BTRFS_I(inode)->root;
6236 struct btrfs_dio_private *dip;
6237 struct bio_vec *bvec = bio->bi_io_vec;
6239 int write = rw & REQ_WRITE;
6242 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6244 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6251 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6252 if (!write && !skip_sum) {
6253 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6261 dip->private = bio->bi_private;
6263 dip->logical_offset = file_offset;
6267 dip->bytes += bvec->bv_len;
6269 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6271 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6272 bio->bi_private = dip;
6274 dip->orig_bio = bio;
6275 atomic_set(&dip->pending_bios, 0);
6278 bio->bi_end_io = btrfs_endio_direct_write;
6280 bio->bi_end_io = btrfs_endio_direct_read;
6282 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6287 * If this is a write, we need to clean up the reserved space and kill
6288 * the ordered extent.
6291 struct btrfs_ordered_extent *ordered;
6292 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6293 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6294 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6295 btrfs_free_reserved_extent(root, ordered->start,
6297 btrfs_put_ordered_extent(ordered);
6298 btrfs_put_ordered_extent(ordered);
6300 bio_endio(bio, ret);
6303 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6304 const struct iovec *iov, loff_t offset,
6305 unsigned long nr_segs)
6311 unsigned blocksize_mask = root->sectorsize - 1;
6312 ssize_t retval = -EINVAL;
6313 loff_t end = offset;
6315 if (offset & blocksize_mask)
6318 /* Check the memory alignment. Blocks cannot straddle pages */
6319 for (seg = 0; seg < nr_segs; seg++) {
6320 addr = (unsigned long)iov[seg].iov_base;
6321 size = iov[seg].iov_len;
6323 if ((addr & blocksize_mask) || (size & blocksize_mask))
6326 /* If this is a write we don't need to check anymore */
6331 * Check to make sure we don't have duplicate iov_base's in this
6332 * iovec, if so return EINVAL, otherwise we'll get csum errors
6333 * when reading back.
6335 for (i = seg + 1; i < nr_segs; i++) {
6336 if (iov[seg].iov_base == iov[i].iov_base)
6344 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6345 const struct iovec *iov, loff_t offset,
6346 unsigned long nr_segs)
6348 struct file *file = iocb->ki_filp;
6349 struct inode *inode = file->f_mapping->host;
6350 struct btrfs_ordered_extent *ordered;
6351 struct extent_state *cached_state = NULL;
6352 u64 lockstart, lockend;
6354 int writing = rw & WRITE;
6356 size_t count = iov_length(iov, nr_segs);
6358 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6364 lockend = offset + count - 1;
6367 ret = btrfs_delalloc_reserve_space(inode, count);
6373 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6376 * We're concerned with the entire range that we're going to be
6377 * doing DIO to, so we need to make sure theres no ordered
6378 * extents in this range.
6380 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6381 lockend - lockstart + 1);
6384 * We need to make sure there are no buffered pages in this
6385 * range either, we could have raced between the invalidate in
6386 * generic_file_direct_write and locking the extent. The
6387 * invalidate needs to happen so that reads after a write do not
6390 if (!ordered && (!writing ||
6391 !test_range_bit(&BTRFS_I(inode)->io_tree,
6392 lockstart, lockend, EXTENT_UPTODATE, 0,
6396 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6397 &cached_state, GFP_NOFS);
6400 btrfs_start_ordered_extent(inode, ordered, 1);
6401 btrfs_put_ordered_extent(ordered);
6403 /* Screw you mmap */
6404 ret = filemap_write_and_wait_range(file->f_mapping,
6411 * If we found a page that couldn't be invalidated just
6412 * fall back to buffered.
6414 ret = invalidate_inode_pages2_range(file->f_mapping,
6415 lockstart >> PAGE_CACHE_SHIFT,
6416 lockend >> PAGE_CACHE_SHIFT);
6428 * we don't use btrfs_set_extent_delalloc because we don't want
6429 * the dirty or uptodate bits
6432 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6433 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6434 EXTENT_DELALLOC, NULL, &cached_state,
6437 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6438 lockend, EXTENT_LOCKED | write_bits,
6439 1, 0, &cached_state, GFP_NOFS);
6444 free_extent_state(cached_state);
6445 cached_state = NULL;
6447 ret = __blockdev_direct_IO(rw, iocb, inode,
6448 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6449 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6450 btrfs_submit_direct, 0);
6452 if (ret < 0 && ret != -EIOCBQUEUED) {
6453 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6454 offset + iov_length(iov, nr_segs) - 1,
6455 EXTENT_LOCKED | write_bits, 1, 0,
6456 &cached_state, GFP_NOFS);
6457 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6459 * We're falling back to buffered, unlock the section we didn't
6462 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6463 offset + iov_length(iov, nr_segs) - 1,
6464 EXTENT_LOCKED | write_bits, 1, 0,
6465 &cached_state, GFP_NOFS);
6468 free_extent_state(cached_state);
6472 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6473 __u64 start, __u64 len)
6475 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6478 int btrfs_readpage(struct file *file, struct page *page)
6480 struct extent_io_tree *tree;
6481 tree = &BTRFS_I(page->mapping->host)->io_tree;
6482 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6485 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6487 struct extent_io_tree *tree;
6490 if (current->flags & PF_MEMALLOC) {
6491 redirty_page_for_writepage(wbc, page);
6495 tree = &BTRFS_I(page->mapping->host)->io_tree;
6496 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6499 int btrfs_writepages(struct address_space *mapping,
6500 struct writeback_control *wbc)
6502 struct extent_io_tree *tree;
6504 tree = &BTRFS_I(mapping->host)->io_tree;
6505 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6509 btrfs_readpages(struct file *file, struct address_space *mapping,
6510 struct list_head *pages, unsigned nr_pages)
6512 struct extent_io_tree *tree;
6513 tree = &BTRFS_I(mapping->host)->io_tree;
6514 return extent_readpages(tree, mapping, pages, nr_pages,
6517 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6519 struct extent_io_tree *tree;
6520 struct extent_map_tree *map;
6523 tree = &BTRFS_I(page->mapping->host)->io_tree;
6524 map = &BTRFS_I(page->mapping->host)->extent_tree;
6525 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6527 ClearPagePrivate(page);
6528 set_page_private(page, 0);
6529 page_cache_release(page);
6534 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6536 if (PageWriteback(page) || PageDirty(page))
6538 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6541 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6543 struct inode *inode = page->mapping->host;
6544 struct extent_io_tree *tree;
6545 struct btrfs_ordered_extent *ordered;
6546 struct extent_state *cached_state = NULL;
6547 u64 page_start = page_offset(page);
6548 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6551 * we have the page locked, so new writeback can't start,
6552 * and the dirty bit won't be cleared while we are here.
6554 * Wait for IO on this page so that we can safely clear
6555 * the PagePrivate2 bit and do ordered accounting
6557 wait_on_page_writeback(page);
6559 tree = &BTRFS_I(inode)->io_tree;
6561 btrfs_releasepage(page, GFP_NOFS);
6564 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6565 ordered = btrfs_lookup_ordered_extent(inode,
6569 * IO on this page will never be started, so we need
6570 * to account for any ordered extents now
6572 clear_extent_bit(tree, page_start, page_end,
6573 EXTENT_DIRTY | EXTENT_DELALLOC |
6574 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6575 &cached_state, GFP_NOFS);
6577 * whoever cleared the private bit is responsible
6578 * for the finish_ordered_io
6580 if (TestClearPagePrivate2(page) &&
6581 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6582 PAGE_CACHE_SIZE, 1)) {
6583 btrfs_finish_ordered_io(ordered);
6585 btrfs_put_ordered_extent(ordered);
6586 cached_state = NULL;
6587 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6589 clear_extent_bit(tree, page_start, page_end,
6590 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6591 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6592 __btrfs_releasepage(page, GFP_NOFS);
6594 ClearPageChecked(page);
6595 if (PagePrivate(page)) {
6596 ClearPagePrivate(page);
6597 set_page_private(page, 0);
6598 page_cache_release(page);
6603 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6604 * called from a page fault handler when a page is first dirtied. Hence we must
6605 * be careful to check for EOF conditions here. We set the page up correctly
6606 * for a written page which means we get ENOSPC checking when writing into
6607 * holes and correct delalloc and unwritten extent mapping on filesystems that
6608 * support these features.
6610 * We are not allowed to take the i_mutex here so we have to play games to
6611 * protect against truncate races as the page could now be beyond EOF. Because
6612 * vmtruncate() writes the inode size before removing pages, once we have the
6613 * page lock we can determine safely if the page is beyond EOF. If it is not
6614 * beyond EOF, then the page is guaranteed safe against truncation until we
6617 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6619 struct page *page = vmf->page;
6620 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6621 struct btrfs_root *root = BTRFS_I(inode)->root;
6622 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6623 struct btrfs_ordered_extent *ordered;
6624 struct extent_state *cached_state = NULL;
6626 unsigned long zero_start;
6633 sb_start_pagefault(inode->i_sb);
6634 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6636 ret = file_update_time(vma->vm_file);
6642 else /* -ENOSPC, -EIO, etc */
6643 ret = VM_FAULT_SIGBUS;
6649 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6652 size = i_size_read(inode);
6653 page_start = page_offset(page);
6654 page_end = page_start + PAGE_CACHE_SIZE - 1;
6656 if ((page->mapping != inode->i_mapping) ||
6657 (page_start >= size)) {
6658 /* page got truncated out from underneath us */
6661 wait_on_page_writeback(page);
6663 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6664 set_page_extent_mapped(page);
6667 * we can't set the delalloc bits if there are pending ordered
6668 * extents. Drop our locks and wait for them to finish
6670 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6672 unlock_extent_cached(io_tree, page_start, page_end,
6673 &cached_state, GFP_NOFS);
6675 btrfs_start_ordered_extent(inode, ordered, 1);
6676 btrfs_put_ordered_extent(ordered);
6681 * XXX - page_mkwrite gets called every time the page is dirtied, even
6682 * if it was already dirty, so for space accounting reasons we need to
6683 * clear any delalloc bits for the range we are fixing to save. There
6684 * is probably a better way to do this, but for now keep consistent with
6685 * prepare_pages in the normal write path.
6687 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6688 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6689 0, 0, &cached_state, GFP_NOFS);
6691 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6694 unlock_extent_cached(io_tree, page_start, page_end,
6695 &cached_state, GFP_NOFS);
6696 ret = VM_FAULT_SIGBUS;
6701 /* page is wholly or partially inside EOF */
6702 if (page_start + PAGE_CACHE_SIZE > size)
6703 zero_start = size & ~PAGE_CACHE_MASK;
6705 zero_start = PAGE_CACHE_SIZE;
6707 if (zero_start != PAGE_CACHE_SIZE) {
6709 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6710 flush_dcache_page(page);
6713 ClearPageChecked(page);
6714 set_page_dirty(page);
6715 SetPageUptodate(page);
6717 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6718 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6720 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6724 sb_end_pagefault(inode->i_sb);
6725 return VM_FAULT_LOCKED;
6729 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6731 sb_end_pagefault(inode->i_sb);
6735 static int btrfs_truncate(struct inode *inode)
6737 struct btrfs_root *root = BTRFS_I(inode)->root;
6738 struct btrfs_block_rsv *rsv;
6741 struct btrfs_trans_handle *trans;
6743 u64 mask = root->sectorsize - 1;
6744 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6746 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6750 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6751 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6754 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6755 * 3 things going on here
6757 * 1) We need to reserve space for our orphan item and the space to
6758 * delete our orphan item. Lord knows we don't want to have a dangling
6759 * orphan item because we didn't reserve space to remove it.
6761 * 2) We need to reserve space to update our inode.
6763 * 3) We need to have something to cache all the space that is going to
6764 * be free'd up by the truncate operation, but also have some slack
6765 * space reserved in case it uses space during the truncate (thank you
6766 * very much snapshotting).
6768 * And we need these to all be seperate. The fact is we can use alot of
6769 * space doing the truncate, and we have no earthly idea how much space
6770 * we will use, so we need the truncate reservation to be seperate so it
6771 * doesn't end up using space reserved for updating the inode or
6772 * removing the orphan item. We also need to be able to stop the
6773 * transaction and start a new one, which means we need to be able to
6774 * update the inode several times, and we have no idea of knowing how
6775 * many times that will be, so we can't just reserve 1 item for the
6776 * entirety of the opration, so that has to be done seperately as well.
6777 * Then there is the orphan item, which does indeed need to be held on
6778 * to for the whole operation, and we need nobody to touch this reserved
6779 * space except the orphan code.
6781 * So that leaves us with
6783 * 1) root->orphan_block_rsv - for the orphan deletion.
6784 * 2) rsv - for the truncate reservation, which we will steal from the
6785 * transaction reservation.
6786 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6787 * updating the inode.
6789 rsv = btrfs_alloc_block_rsv(root);
6792 rsv->size = min_size;
6795 * 1 for the truncate slack space
6796 * 1 for the orphan item we're going to add
6797 * 1 for the orphan item deletion
6798 * 1 for updating the inode.
6800 trans = btrfs_start_transaction(root, 4);
6801 if (IS_ERR(trans)) {
6802 err = PTR_ERR(trans);
6806 /* Migrate the slack space for the truncate to our reserve */
6807 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6811 ret = btrfs_orphan_add(trans, inode);
6813 btrfs_end_transaction(trans, root);
6818 * setattr is responsible for setting the ordered_data_close flag,
6819 * but that is only tested during the last file release. That
6820 * could happen well after the next commit, leaving a great big
6821 * window where new writes may get lost if someone chooses to write
6822 * to this file after truncating to zero
6824 * The inode doesn't have any dirty data here, and so if we commit
6825 * this is a noop. If someone immediately starts writing to the inode
6826 * it is very likely we'll catch some of their writes in this
6827 * transaction, and the commit will find this file on the ordered
6828 * data list with good things to send down.
6830 * This is a best effort solution, there is still a window where
6831 * using truncate to replace the contents of the file will
6832 * end up with a zero length file after a crash.
6834 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6835 &BTRFS_I(inode)->runtime_flags))
6836 btrfs_add_ordered_operation(trans, root, inode);
6839 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6842 * This can only happen with the original transaction we
6843 * started above, every other time we shouldn't have a
6844 * transaction started yet.
6853 /* Just need the 1 for updating the inode */
6854 trans = btrfs_start_transaction(root, 1);
6855 if (IS_ERR(trans)) {
6856 ret = err = PTR_ERR(trans);
6862 trans->block_rsv = rsv;
6864 ret = btrfs_truncate_inode_items(trans, root, inode,
6866 BTRFS_EXTENT_DATA_KEY);
6867 if (ret != -EAGAIN) {
6872 trans->block_rsv = &root->fs_info->trans_block_rsv;
6873 ret = btrfs_update_inode(trans, root, inode);
6879 nr = trans->blocks_used;
6880 btrfs_end_transaction(trans, root);
6882 btrfs_btree_balance_dirty(root, nr);
6885 if (ret == 0 && inode->i_nlink > 0) {
6886 trans->block_rsv = root->orphan_block_rsv;
6887 ret = btrfs_orphan_del(trans, inode);
6890 } else if (ret && inode->i_nlink > 0) {
6892 * Failed to do the truncate, remove us from the in memory
6895 ret = btrfs_orphan_del(NULL, inode);
6899 trans->block_rsv = &root->fs_info->trans_block_rsv;
6900 ret = btrfs_update_inode(trans, root, inode);
6904 nr = trans->blocks_used;
6905 ret = btrfs_end_transaction(trans, root);
6906 btrfs_btree_balance_dirty(root, nr);
6910 btrfs_free_block_rsv(root, rsv);
6919 * create a new subvolume directory/inode (helper for the ioctl).
6921 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6922 struct btrfs_root *new_root, u64 new_dirid)
6924 struct inode *inode;
6928 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6929 new_dirid, new_dirid,
6930 S_IFDIR | (~current_umask() & S_IRWXUGO),
6933 return PTR_ERR(inode);
6934 inode->i_op = &btrfs_dir_inode_operations;
6935 inode->i_fop = &btrfs_dir_file_operations;
6937 set_nlink(inode, 1);
6938 btrfs_i_size_write(inode, 0);
6940 err = btrfs_update_inode(trans, new_root, inode);
6946 struct inode *btrfs_alloc_inode(struct super_block *sb)
6948 struct btrfs_inode *ei;
6949 struct inode *inode;
6951 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6958 ei->last_sub_trans = 0;
6959 ei->logged_trans = 0;
6960 ei->delalloc_bytes = 0;
6961 ei->disk_i_size = 0;
6964 ei->index_cnt = (u64)-1;
6965 ei->last_unlink_trans = 0;
6967 spin_lock_init(&ei->lock);
6968 ei->outstanding_extents = 0;
6969 ei->reserved_extents = 0;
6971 ei->runtime_flags = 0;
6972 ei->force_compress = BTRFS_COMPRESS_NONE;
6974 ei->delayed_node = NULL;
6976 inode = &ei->vfs_inode;
6977 extent_map_tree_init(&ei->extent_tree);
6978 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6979 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6980 ei->io_tree.track_uptodate = 1;
6981 ei->io_failure_tree.track_uptodate = 1;
6982 mutex_init(&ei->log_mutex);
6983 mutex_init(&ei->delalloc_mutex);
6984 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6985 INIT_LIST_HEAD(&ei->delalloc_inodes);
6986 INIT_LIST_HEAD(&ei->ordered_operations);
6987 RB_CLEAR_NODE(&ei->rb_node);
6992 static void btrfs_i_callback(struct rcu_head *head)
6994 struct inode *inode = container_of(head, struct inode, i_rcu);
6995 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6998 void btrfs_destroy_inode(struct inode *inode)
7000 struct btrfs_ordered_extent *ordered;
7001 struct btrfs_root *root = BTRFS_I(inode)->root;
7003 WARN_ON(!hlist_empty(&inode->i_dentry));
7004 WARN_ON(inode->i_data.nrpages);
7005 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7006 WARN_ON(BTRFS_I(inode)->reserved_extents);
7007 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7008 WARN_ON(BTRFS_I(inode)->csum_bytes);
7011 * This can happen where we create an inode, but somebody else also
7012 * created the same inode and we need to destroy the one we already
7019 * Make sure we're properly removed from the ordered operation
7023 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7024 spin_lock(&root->fs_info->ordered_extent_lock);
7025 list_del_init(&BTRFS_I(inode)->ordered_operations);
7026 spin_unlock(&root->fs_info->ordered_extent_lock);
7029 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7030 &BTRFS_I(inode)->runtime_flags)) {
7031 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7032 (unsigned long long)btrfs_ino(inode));
7033 atomic_dec(&root->orphan_inodes);
7037 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7041 printk(KERN_ERR "btrfs found ordered "
7042 "extent %llu %llu on inode cleanup\n",
7043 (unsigned long long)ordered->file_offset,
7044 (unsigned long long)ordered->len);
7045 btrfs_remove_ordered_extent(inode, ordered);
7046 btrfs_put_ordered_extent(ordered);
7047 btrfs_put_ordered_extent(ordered);
7050 inode_tree_del(inode);
7051 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7053 btrfs_remove_delayed_node(inode);
7054 call_rcu(&inode->i_rcu, btrfs_i_callback);
7057 int btrfs_drop_inode(struct inode *inode)
7059 struct btrfs_root *root = BTRFS_I(inode)->root;
7061 if (btrfs_root_refs(&root->root_item) == 0 &&
7062 !btrfs_is_free_space_inode(inode))
7065 return generic_drop_inode(inode);
7068 static void init_once(void *foo)
7070 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7072 inode_init_once(&ei->vfs_inode);
7075 void btrfs_destroy_cachep(void)
7077 if (btrfs_inode_cachep)
7078 kmem_cache_destroy(btrfs_inode_cachep);
7079 if (btrfs_trans_handle_cachep)
7080 kmem_cache_destroy(btrfs_trans_handle_cachep);
7081 if (btrfs_transaction_cachep)
7082 kmem_cache_destroy(btrfs_transaction_cachep);
7083 if (btrfs_path_cachep)
7084 kmem_cache_destroy(btrfs_path_cachep);
7085 if (btrfs_free_space_cachep)
7086 kmem_cache_destroy(btrfs_free_space_cachep);
7089 int btrfs_init_cachep(void)
7091 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7092 sizeof(struct btrfs_inode), 0,
7093 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7094 if (!btrfs_inode_cachep)
7097 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7098 sizeof(struct btrfs_trans_handle), 0,
7099 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7100 if (!btrfs_trans_handle_cachep)
7103 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7104 sizeof(struct btrfs_transaction), 0,
7105 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7106 if (!btrfs_transaction_cachep)
7109 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7110 sizeof(struct btrfs_path), 0,
7111 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7112 if (!btrfs_path_cachep)
7115 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7116 sizeof(struct btrfs_free_space), 0,
7117 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7118 if (!btrfs_free_space_cachep)
7123 btrfs_destroy_cachep();
7127 static int btrfs_getattr(struct vfsmount *mnt,
7128 struct dentry *dentry, struct kstat *stat)
7130 struct inode *inode = dentry->d_inode;
7131 u32 blocksize = inode->i_sb->s_blocksize;
7133 generic_fillattr(inode, stat);
7134 stat->dev = BTRFS_I(inode)->root->anon_dev;
7135 stat->blksize = PAGE_CACHE_SIZE;
7136 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7137 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7142 * If a file is moved, it will inherit the cow and compression flags of the new
7145 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7147 struct btrfs_inode *b_dir = BTRFS_I(dir);
7148 struct btrfs_inode *b_inode = BTRFS_I(inode);
7150 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7151 b_inode->flags |= BTRFS_INODE_NODATACOW;
7153 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7155 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7156 b_inode->flags |= BTRFS_INODE_COMPRESS;
7157 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7159 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7160 BTRFS_INODE_NOCOMPRESS);
7164 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7165 struct inode *new_dir, struct dentry *new_dentry)
7167 struct btrfs_trans_handle *trans;
7168 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7169 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7170 struct inode *new_inode = new_dentry->d_inode;
7171 struct inode *old_inode = old_dentry->d_inode;
7172 struct timespec ctime = CURRENT_TIME;
7176 u64 old_ino = btrfs_ino(old_inode);
7178 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7181 /* we only allow rename subvolume link between subvolumes */
7182 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7185 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7186 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7189 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7190 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7193 * we're using rename to replace one file with another.
7194 * and the replacement file is large. Start IO on it now so
7195 * we don't add too much work to the end of the transaction
7197 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7198 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7199 filemap_flush(old_inode->i_mapping);
7201 /* close the racy window with snapshot create/destroy ioctl */
7202 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7203 down_read(&root->fs_info->subvol_sem);
7205 * We want to reserve the absolute worst case amount of items. So if
7206 * both inodes are subvols and we need to unlink them then that would
7207 * require 4 item modifications, but if they are both normal inodes it
7208 * would require 5 item modifications, so we'll assume their normal
7209 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7210 * should cover the worst case number of items we'll modify.
7212 trans = btrfs_start_transaction(root, 20);
7213 if (IS_ERR(trans)) {
7214 ret = PTR_ERR(trans);
7219 btrfs_record_root_in_trans(trans, dest);
7221 ret = btrfs_set_inode_index(new_dir, &index);
7225 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7226 /* force full log commit if subvolume involved. */
7227 root->fs_info->last_trans_log_full_commit = trans->transid;
7229 ret = btrfs_insert_inode_ref(trans, dest,
7230 new_dentry->d_name.name,
7231 new_dentry->d_name.len,
7233 btrfs_ino(new_dir), index);
7237 * this is an ugly little race, but the rename is required
7238 * to make sure that if we crash, the inode is either at the
7239 * old name or the new one. pinning the log transaction lets
7240 * us make sure we don't allow a log commit to come in after
7241 * we unlink the name but before we add the new name back in.
7243 btrfs_pin_log_trans(root);
7246 * make sure the inode gets flushed if it is replacing
7249 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7250 btrfs_add_ordered_operation(trans, root, old_inode);
7252 inode_inc_iversion(old_dir);
7253 inode_inc_iversion(new_dir);
7254 inode_inc_iversion(old_inode);
7255 old_dir->i_ctime = old_dir->i_mtime = ctime;
7256 new_dir->i_ctime = new_dir->i_mtime = ctime;
7257 old_inode->i_ctime = ctime;
7259 if (old_dentry->d_parent != new_dentry->d_parent)
7260 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7262 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7263 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7264 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7265 old_dentry->d_name.name,
7266 old_dentry->d_name.len);
7268 ret = __btrfs_unlink_inode(trans, root, old_dir,
7269 old_dentry->d_inode,
7270 old_dentry->d_name.name,
7271 old_dentry->d_name.len);
7273 ret = btrfs_update_inode(trans, root, old_inode);
7276 btrfs_abort_transaction(trans, root, ret);
7281 inode_inc_iversion(new_inode);
7282 new_inode->i_ctime = CURRENT_TIME;
7283 if (unlikely(btrfs_ino(new_inode) ==
7284 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7285 root_objectid = BTRFS_I(new_inode)->location.objectid;
7286 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7288 new_dentry->d_name.name,
7289 new_dentry->d_name.len);
7290 BUG_ON(new_inode->i_nlink == 0);
7292 ret = btrfs_unlink_inode(trans, dest, new_dir,
7293 new_dentry->d_inode,
7294 new_dentry->d_name.name,
7295 new_dentry->d_name.len);
7297 if (!ret && new_inode->i_nlink == 0) {
7298 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7302 btrfs_abort_transaction(trans, root, ret);
7307 fixup_inode_flags(new_dir, old_inode);
7309 ret = btrfs_add_link(trans, new_dir, old_inode,
7310 new_dentry->d_name.name,
7311 new_dentry->d_name.len, 0, index);
7313 btrfs_abort_transaction(trans, root, ret);
7317 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7318 struct dentry *parent = new_dentry->d_parent;
7319 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7320 btrfs_end_log_trans(root);
7323 btrfs_end_transaction(trans, root);
7325 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7326 up_read(&root->fs_info->subvol_sem);
7332 * some fairly slow code that needs optimization. This walks the list
7333 * of all the inodes with pending delalloc and forces them to disk.
7335 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7337 struct list_head *head = &root->fs_info->delalloc_inodes;
7338 struct btrfs_inode *binode;
7339 struct inode *inode;
7341 if (root->fs_info->sb->s_flags & MS_RDONLY)
7344 spin_lock(&root->fs_info->delalloc_lock);
7345 while (!list_empty(head)) {
7346 binode = list_entry(head->next, struct btrfs_inode,
7348 inode = igrab(&binode->vfs_inode);
7350 list_del_init(&binode->delalloc_inodes);
7351 spin_unlock(&root->fs_info->delalloc_lock);
7353 filemap_flush(inode->i_mapping);
7355 btrfs_add_delayed_iput(inode);
7360 spin_lock(&root->fs_info->delalloc_lock);
7362 spin_unlock(&root->fs_info->delalloc_lock);
7364 /* the filemap_flush will queue IO into the worker threads, but
7365 * we have to make sure the IO is actually started and that
7366 * ordered extents get created before we return
7368 atomic_inc(&root->fs_info->async_submit_draining);
7369 while (atomic_read(&root->fs_info->nr_async_submits) ||
7370 atomic_read(&root->fs_info->async_delalloc_pages)) {
7371 wait_event(root->fs_info->async_submit_wait,
7372 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7373 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7375 atomic_dec(&root->fs_info->async_submit_draining);
7379 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7380 const char *symname)
7382 struct btrfs_trans_handle *trans;
7383 struct btrfs_root *root = BTRFS_I(dir)->root;
7384 struct btrfs_path *path;
7385 struct btrfs_key key;
7386 struct inode *inode = NULL;
7394 struct btrfs_file_extent_item *ei;
7395 struct extent_buffer *leaf;
7396 unsigned long nr = 0;
7398 name_len = strlen(symname) + 1;
7399 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7400 return -ENAMETOOLONG;
7403 * 2 items for inode item and ref
7404 * 2 items for dir items
7405 * 1 item for xattr if selinux is on
7407 trans = btrfs_start_transaction(root, 5);
7409 return PTR_ERR(trans);
7411 err = btrfs_find_free_ino(root, &objectid);
7415 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7416 dentry->d_name.len, btrfs_ino(dir), objectid,
7417 S_IFLNK|S_IRWXUGO, &index);
7418 if (IS_ERR(inode)) {
7419 err = PTR_ERR(inode);
7423 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7430 * If the active LSM wants to access the inode during
7431 * d_instantiate it needs these. Smack checks to see
7432 * if the filesystem supports xattrs by looking at the
7435 inode->i_fop = &btrfs_file_operations;
7436 inode->i_op = &btrfs_file_inode_operations;
7438 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7442 inode->i_mapping->a_ops = &btrfs_aops;
7443 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7444 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7449 path = btrfs_alloc_path();
7455 key.objectid = btrfs_ino(inode);
7457 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7458 datasize = btrfs_file_extent_calc_inline_size(name_len);
7459 err = btrfs_insert_empty_item(trans, root, path, &key,
7463 btrfs_free_path(path);
7466 leaf = path->nodes[0];
7467 ei = btrfs_item_ptr(leaf, path->slots[0],
7468 struct btrfs_file_extent_item);
7469 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7470 btrfs_set_file_extent_type(leaf, ei,
7471 BTRFS_FILE_EXTENT_INLINE);
7472 btrfs_set_file_extent_encryption(leaf, ei, 0);
7473 btrfs_set_file_extent_compression(leaf, ei, 0);
7474 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7475 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7477 ptr = btrfs_file_extent_inline_start(ei);
7478 write_extent_buffer(leaf, symname, ptr, name_len);
7479 btrfs_mark_buffer_dirty(leaf);
7480 btrfs_free_path(path);
7482 inode->i_op = &btrfs_symlink_inode_operations;
7483 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7484 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7485 inode_set_bytes(inode, name_len);
7486 btrfs_i_size_write(inode, name_len - 1);
7487 err = btrfs_update_inode(trans, root, inode);
7493 d_instantiate(dentry, inode);
7494 nr = trans->blocks_used;
7495 btrfs_end_transaction(trans, root);
7497 inode_dec_link_count(inode);
7500 btrfs_btree_balance_dirty(root, nr);
7504 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7505 u64 start, u64 num_bytes, u64 min_size,
7506 loff_t actual_len, u64 *alloc_hint,
7507 struct btrfs_trans_handle *trans)
7509 struct btrfs_root *root = BTRFS_I(inode)->root;
7510 struct btrfs_key ins;
7511 u64 cur_offset = start;
7514 bool own_trans = true;
7518 while (num_bytes > 0) {
7520 trans = btrfs_start_transaction(root, 3);
7521 if (IS_ERR(trans)) {
7522 ret = PTR_ERR(trans);
7527 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7528 0, *alloc_hint, &ins, 1);
7531 btrfs_end_transaction(trans, root);
7535 ret = insert_reserved_file_extent(trans, inode,
7536 cur_offset, ins.objectid,
7537 ins.offset, ins.offset,
7538 ins.offset, 0, 0, 0,
7539 BTRFS_FILE_EXTENT_PREALLOC);
7541 btrfs_abort_transaction(trans, root, ret);
7543 btrfs_end_transaction(trans, root);
7546 btrfs_drop_extent_cache(inode, cur_offset,
7547 cur_offset + ins.offset -1, 0);
7549 num_bytes -= ins.offset;
7550 cur_offset += ins.offset;
7551 *alloc_hint = ins.objectid + ins.offset;
7553 inode_inc_iversion(inode);
7554 inode->i_ctime = CURRENT_TIME;
7555 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7556 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7557 (actual_len > inode->i_size) &&
7558 (cur_offset > inode->i_size)) {
7559 if (cur_offset > actual_len)
7560 i_size = actual_len;
7562 i_size = cur_offset;
7563 i_size_write(inode, i_size);
7564 btrfs_ordered_update_i_size(inode, i_size, NULL);
7567 ret = btrfs_update_inode(trans, root, inode);
7570 btrfs_abort_transaction(trans, root, ret);
7572 btrfs_end_transaction(trans, root);
7577 btrfs_end_transaction(trans, root);
7582 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7583 u64 start, u64 num_bytes, u64 min_size,
7584 loff_t actual_len, u64 *alloc_hint)
7586 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7587 min_size, actual_len, alloc_hint,
7591 int btrfs_prealloc_file_range_trans(struct inode *inode,
7592 struct btrfs_trans_handle *trans, int mode,
7593 u64 start, u64 num_bytes, u64 min_size,
7594 loff_t actual_len, u64 *alloc_hint)
7596 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7597 min_size, actual_len, alloc_hint, trans);
7600 static int btrfs_set_page_dirty(struct page *page)
7602 return __set_page_dirty_nobuffers(page);
7605 static int btrfs_permission(struct inode *inode, int mask)
7607 struct btrfs_root *root = BTRFS_I(inode)->root;
7608 umode_t mode = inode->i_mode;
7610 if (mask & MAY_WRITE &&
7611 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7612 if (btrfs_root_readonly(root))
7614 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7617 return generic_permission(inode, mask);
7620 static const struct inode_operations btrfs_dir_inode_operations = {
7621 .getattr = btrfs_getattr,
7622 .lookup = btrfs_lookup,
7623 .create = btrfs_create,
7624 .unlink = btrfs_unlink,
7626 .mkdir = btrfs_mkdir,
7627 .rmdir = btrfs_rmdir,
7628 .rename = btrfs_rename,
7629 .symlink = btrfs_symlink,
7630 .setattr = btrfs_setattr,
7631 .mknod = btrfs_mknod,
7632 .setxattr = btrfs_setxattr,
7633 .getxattr = btrfs_getxattr,
7634 .listxattr = btrfs_listxattr,
7635 .removexattr = btrfs_removexattr,
7636 .permission = btrfs_permission,
7637 .get_acl = btrfs_get_acl,
7639 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7640 .lookup = btrfs_lookup,
7641 .permission = btrfs_permission,
7642 .get_acl = btrfs_get_acl,
7645 static const struct file_operations btrfs_dir_file_operations = {
7646 .llseek = generic_file_llseek,
7647 .read = generic_read_dir,
7648 .readdir = btrfs_real_readdir,
7649 .unlocked_ioctl = btrfs_ioctl,
7650 #ifdef CONFIG_COMPAT
7651 .compat_ioctl = btrfs_ioctl,
7653 .release = btrfs_release_file,
7654 .fsync = btrfs_sync_file,
7657 static struct extent_io_ops btrfs_extent_io_ops = {
7658 .fill_delalloc = run_delalloc_range,
7659 .submit_bio_hook = btrfs_submit_bio_hook,
7660 .merge_bio_hook = btrfs_merge_bio_hook,
7661 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7662 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7663 .writepage_start_hook = btrfs_writepage_start_hook,
7664 .set_bit_hook = btrfs_set_bit_hook,
7665 .clear_bit_hook = btrfs_clear_bit_hook,
7666 .merge_extent_hook = btrfs_merge_extent_hook,
7667 .split_extent_hook = btrfs_split_extent_hook,
7671 * btrfs doesn't support the bmap operation because swapfiles
7672 * use bmap to make a mapping of extents in the file. They assume
7673 * these extents won't change over the life of the file and they
7674 * use the bmap result to do IO directly to the drive.
7676 * the btrfs bmap call would return logical addresses that aren't
7677 * suitable for IO and they also will change frequently as COW
7678 * operations happen. So, swapfile + btrfs == corruption.
7680 * For now we're avoiding this by dropping bmap.
7682 static const struct address_space_operations btrfs_aops = {
7683 .readpage = btrfs_readpage,
7684 .writepage = btrfs_writepage,
7685 .writepages = btrfs_writepages,
7686 .readpages = btrfs_readpages,
7687 .direct_IO = btrfs_direct_IO,
7688 .invalidatepage = btrfs_invalidatepage,
7689 .releasepage = btrfs_releasepage,
7690 .set_page_dirty = btrfs_set_page_dirty,
7691 .error_remove_page = generic_error_remove_page,
7694 static const struct address_space_operations btrfs_symlink_aops = {
7695 .readpage = btrfs_readpage,
7696 .writepage = btrfs_writepage,
7697 .invalidatepage = btrfs_invalidatepage,
7698 .releasepage = btrfs_releasepage,
7701 static const struct inode_operations btrfs_file_inode_operations = {
7702 .getattr = btrfs_getattr,
7703 .setattr = btrfs_setattr,
7704 .setxattr = btrfs_setxattr,
7705 .getxattr = btrfs_getxattr,
7706 .listxattr = btrfs_listxattr,
7707 .removexattr = btrfs_removexattr,
7708 .permission = btrfs_permission,
7709 .fiemap = btrfs_fiemap,
7710 .get_acl = btrfs_get_acl,
7711 .update_time = btrfs_update_time,
7713 static const struct inode_operations btrfs_special_inode_operations = {
7714 .getattr = btrfs_getattr,
7715 .setattr = btrfs_setattr,
7716 .permission = btrfs_permission,
7717 .setxattr = btrfs_setxattr,
7718 .getxattr = btrfs_getxattr,
7719 .listxattr = btrfs_listxattr,
7720 .removexattr = btrfs_removexattr,
7721 .get_acl = btrfs_get_acl,
7722 .update_time = btrfs_update_time,
7724 static const struct inode_operations btrfs_symlink_inode_operations = {
7725 .readlink = generic_readlink,
7726 .follow_link = page_follow_link_light,
7727 .put_link = page_put_link,
7728 .getattr = btrfs_getattr,
7729 .setattr = btrfs_setattr,
7730 .permission = btrfs_permission,
7731 .setxattr = btrfs_setxattr,
7732 .getxattr = btrfs_getxattr,
7733 .listxattr = btrfs_listxattr,
7734 .removexattr = btrfs_removexattr,
7735 .get_acl = btrfs_get_acl,
7736 .update_time = btrfs_update_time,
7739 const struct dentry_operations btrfs_dentry_operations = {
7740 .d_delete = btrfs_dentry_delete,
7741 .d_release = btrfs_dentry_release,