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/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "ordered-data.h"
56 #include "compression.h"
58 #include "free-space-cache.h"
59 #include "inode-map.h"
63 struct btrfs_iget_args {
65 struct btrfs_root *root;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_dir_ro_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct address_space_operations btrfs_symlink_aops;
75 static const struct file_operations btrfs_dir_file_operations;
76 static struct extent_io_ops btrfs_extent_io_ops;
78 static struct kmem_cache *btrfs_inode_cachep;
79 static struct kmem_cache *btrfs_delalloc_work_cachep;
80 struct kmem_cache *btrfs_trans_handle_cachep;
81 struct kmem_cache *btrfs_transaction_cachep;
82 struct kmem_cache *btrfs_path_cachep;
83 struct kmem_cache *btrfs_free_space_cachep;
86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
97 static int btrfs_truncate(struct inode *inode);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
99 static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
106 u64 orig_block_len, u64 ram_bytes,
109 static int btrfs_dirty_inode(struct inode *inode);
111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
117 err = btrfs_init_acl(trans, inode, dir);
119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
132 struct page **compressed_pages)
134 struct btrfs_key key;
135 struct btrfs_path *path;
136 struct extent_buffer *leaf;
137 struct page *page = NULL;
140 struct btrfs_file_extent_item *ei;
143 size_t cur_size = size;
145 unsigned long offset;
147 if (compressed_size && compressed_pages)
148 cur_size = compressed_size;
150 path = btrfs_alloc_path();
154 path->leave_spinning = 1;
156 key.objectid = btrfs_ino(inode);
158 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
159 datasize = btrfs_file_extent_calc_inline_size(cur_size);
161 inode_add_bytes(inode, size);
162 ret = btrfs_insert_empty_item(trans, root, path, &key,
168 leaf = path->nodes[0];
169 ei = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_file_extent_item);
171 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
172 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
173 btrfs_set_file_extent_encryption(leaf, ei, 0);
174 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
175 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
176 ptr = btrfs_file_extent_inline_start(ei);
178 if (compress_type != BTRFS_COMPRESS_NONE) {
181 while (compressed_size > 0) {
182 cpage = compressed_pages[i];
183 cur_size = min_t(unsigned long, compressed_size,
186 kaddr = kmap_atomic(cpage);
187 write_extent_buffer(leaf, kaddr, ptr, cur_size);
188 kunmap_atomic(kaddr);
192 compressed_size -= cur_size;
194 btrfs_set_file_extent_compression(leaf, ei,
197 page = find_get_page(inode->i_mapping,
198 start >> PAGE_CACHE_SHIFT);
199 btrfs_set_file_extent_compression(leaf, ei, 0);
200 kaddr = kmap_atomic(page);
201 offset = start & (PAGE_CACHE_SIZE - 1);
202 write_extent_buffer(leaf, kaddr + offset, ptr, size);
203 kunmap_atomic(kaddr);
204 page_cache_release(page);
206 btrfs_mark_buffer_dirty(leaf);
207 btrfs_free_path(path);
210 * we're an inline extent, so nobody can
211 * extend the file past i_size without locking
212 * a page we already have locked.
214 * We must do any isize and inode updates
215 * before we unlock the pages. Otherwise we
216 * could end up racing with unlink.
218 BTRFS_I(inode)->disk_i_size = inode->i_size;
219 ret = btrfs_update_inode(trans, root, inode);
223 btrfs_free_path(path);
229 * conditionally insert an inline extent into the file. This
230 * does the checks required to make sure the data is small enough
231 * to fit as an inline extent.
233 static noinline int cow_file_range_inline(struct btrfs_root *root,
234 struct inode *inode, u64 start,
235 u64 end, size_t compressed_size,
237 struct page **compressed_pages)
239 struct btrfs_trans_handle *trans;
240 u64 isize = i_size_read(inode);
241 u64 actual_end = min(end + 1, isize);
242 u64 inline_len = actual_end - start;
243 u64 aligned_end = ALIGN(end, root->sectorsize);
244 u64 data_len = inline_len;
248 data_len = compressed_size;
251 actual_end >= PAGE_CACHE_SIZE ||
252 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
254 (actual_end & (root->sectorsize - 1)) == 0) ||
256 data_len > root->fs_info->max_inline) {
260 trans = btrfs_join_transaction(root);
262 return PTR_ERR(trans);
263 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
265 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
267 btrfs_abort_transaction(trans, root, ret);
271 if (isize > actual_end)
272 inline_len = min_t(u64, isize, actual_end);
273 ret = insert_inline_extent(trans, root, inode, start,
274 inline_len, compressed_size,
275 compress_type, compressed_pages);
276 if (ret && ret != -ENOSPC) {
277 btrfs_abort_transaction(trans, root, ret);
279 } else if (ret == -ENOSPC) {
284 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
285 btrfs_delalloc_release_metadata(inode, end + 1 - start);
286 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
288 btrfs_end_transaction(trans, root);
292 struct async_extent {
297 unsigned long nr_pages;
299 struct list_head list;
304 struct btrfs_root *root;
305 struct page *locked_page;
308 struct list_head extents;
309 struct btrfs_work work;
312 static noinline int add_async_extent(struct async_cow *cow,
313 u64 start, u64 ram_size,
316 unsigned long nr_pages,
319 struct async_extent *async_extent;
321 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
322 BUG_ON(!async_extent); /* -ENOMEM */
323 async_extent->start = start;
324 async_extent->ram_size = ram_size;
325 async_extent->compressed_size = compressed_size;
326 async_extent->pages = pages;
327 async_extent->nr_pages = nr_pages;
328 async_extent->compress_type = compress_type;
329 list_add_tail(&async_extent->list, &cow->extents);
334 * we create compressed extents in two phases. The first
335 * phase compresses a range of pages that have already been
336 * locked (both pages and state bits are locked).
338 * This is done inside an ordered work queue, and the compression
339 * is spread across many cpus. The actual IO submission is step
340 * two, and the ordered work queue takes care of making sure that
341 * happens in the same order things were put onto the queue by
342 * writepages and friends.
344 * If this code finds it can't get good compression, it puts an
345 * entry onto the work queue to write the uncompressed bytes. This
346 * makes sure that both compressed inodes and uncompressed inodes
347 * are written in the same order that the flusher thread sent them
350 static noinline int compress_file_range(struct inode *inode,
351 struct page *locked_page,
353 struct async_cow *async_cow,
356 struct btrfs_root *root = BTRFS_I(inode)->root;
358 u64 blocksize = root->sectorsize;
360 u64 isize = i_size_read(inode);
362 struct page **pages = NULL;
363 unsigned long nr_pages;
364 unsigned long nr_pages_ret = 0;
365 unsigned long total_compressed = 0;
366 unsigned long total_in = 0;
367 unsigned long max_compressed = 128 * 1024;
368 unsigned long max_uncompressed = 128 * 1024;
371 int compress_type = root->fs_info->compress_type;
374 /* if this is a small write inside eof, kick off a defrag */
375 if ((end - start + 1) < 16 * 1024 &&
376 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
377 btrfs_add_inode_defrag(NULL, inode);
379 actual_end = min_t(u64, isize, end + 1);
382 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
383 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
386 * we don't want to send crud past the end of i_size through
387 * compression, that's just a waste of CPU time. So, if the
388 * end of the file is before the start of our current
389 * requested range of bytes, we bail out to the uncompressed
390 * cleanup code that can deal with all of this.
392 * It isn't really the fastest way to fix things, but this is a
393 * very uncommon corner.
395 if (actual_end <= start)
396 goto cleanup_and_bail_uncompressed;
398 total_compressed = actual_end - start;
400 /* we want to make sure that amount of ram required to uncompress
401 * an extent is reasonable, so we limit the total size in ram
402 * of a compressed extent to 128k. This is a crucial number
403 * because it also controls how easily we can spread reads across
404 * cpus for decompression.
406 * We also want to make sure the amount of IO required to do
407 * a random read is reasonably small, so we limit the size of
408 * a compressed extent to 128k.
410 total_compressed = min(total_compressed, max_uncompressed);
411 num_bytes = ALIGN(end - start + 1, blocksize);
412 num_bytes = max(blocksize, num_bytes);
417 * we do compression for mount -o compress and when the
418 * inode has not been flagged as nocompress. This flag can
419 * change at any time if we discover bad compression ratios.
421 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
422 (btrfs_test_opt(root, COMPRESS) ||
423 (BTRFS_I(inode)->force_compress) ||
424 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
426 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
428 /* just bail out to the uncompressed code */
432 if (BTRFS_I(inode)->force_compress)
433 compress_type = BTRFS_I(inode)->force_compress;
436 * we need to call clear_page_dirty_for_io on each
437 * page in the range. Otherwise applications with the file
438 * mmap'd can wander in and change the page contents while
439 * we are compressing them.
441 * If the compression fails for any reason, we set the pages
442 * dirty again later on.
444 extent_range_clear_dirty_for_io(inode, start, end);
446 ret = btrfs_compress_pages(compress_type,
447 inode->i_mapping, start,
448 total_compressed, pages,
449 nr_pages, &nr_pages_ret,
455 unsigned long offset = total_compressed &
456 (PAGE_CACHE_SIZE - 1);
457 struct page *page = pages[nr_pages_ret - 1];
460 /* zero the tail end of the last page, we might be
461 * sending it down to disk
464 kaddr = kmap_atomic(page);
465 memset(kaddr + offset, 0,
466 PAGE_CACHE_SIZE - offset);
467 kunmap_atomic(kaddr);
474 /* lets try to make an inline extent */
475 if (ret || total_in < (actual_end - start)) {
476 /* we didn't compress the entire range, try
477 * to make an uncompressed inline extent.
479 ret = cow_file_range_inline(root, inode, start, end,
482 /* try making a compressed inline extent */
483 ret = cow_file_range_inline(root, inode, start, end,
485 compress_type, pages);
488 unsigned long clear_flags = EXTENT_DELALLOC |
490 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
493 * inline extent creation worked or returned error,
494 * we don't need to create any more async work items.
495 * Unlock and free up our temp pages.
497 extent_clear_unlock_delalloc(inode, start, end, NULL,
498 clear_flags, PAGE_UNLOCK |
508 * we aren't doing an inline extent round the compressed size
509 * up to a block size boundary so the allocator does sane
512 total_compressed = ALIGN(total_compressed, blocksize);
515 * one last check to make sure the compression is really a
516 * win, compare the page count read with the blocks on disk
518 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
519 if (total_compressed >= total_in) {
522 num_bytes = total_in;
525 if (!will_compress && pages) {
527 * the compression code ran but failed to make things smaller,
528 * free any pages it allocated and our page pointer array
530 for (i = 0; i < nr_pages_ret; i++) {
531 WARN_ON(pages[i]->mapping);
532 page_cache_release(pages[i]);
536 total_compressed = 0;
539 /* flag the file so we don't compress in the future */
540 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
541 !(BTRFS_I(inode)->force_compress)) {
542 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
548 /* the async work queues will take care of doing actual
549 * allocation on disk for these compressed pages,
550 * and will submit them to the elevator.
552 add_async_extent(async_cow, start, num_bytes,
553 total_compressed, pages, nr_pages_ret,
556 if (start + num_bytes < end) {
563 cleanup_and_bail_uncompressed:
565 * No compression, but we still need to write the pages in
566 * the file we've been given so far. redirty the locked
567 * page if it corresponds to our extent and set things up
568 * for the async work queue to run cow_file_range to do
569 * the normal delalloc dance
571 if (page_offset(locked_page) >= start &&
572 page_offset(locked_page) <= end) {
573 __set_page_dirty_nobuffers(locked_page);
574 /* unlocked later on in the async handlers */
577 extent_range_redirty_for_io(inode, start, end);
578 add_async_extent(async_cow, start, end - start + 1,
579 0, NULL, 0, BTRFS_COMPRESS_NONE);
587 for (i = 0; i < nr_pages_ret; i++) {
588 WARN_ON(pages[i]->mapping);
589 page_cache_release(pages[i]);
597 * phase two of compressed writeback. This is the ordered portion
598 * of the code, which only gets called in the order the work was
599 * queued. We walk all the async extents created by compress_file_range
600 * and send them down to the disk.
602 static noinline int submit_compressed_extents(struct inode *inode,
603 struct async_cow *async_cow)
605 struct async_extent *async_extent;
607 struct btrfs_key ins;
608 struct extent_map *em;
609 struct btrfs_root *root = BTRFS_I(inode)->root;
610 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
611 struct extent_io_tree *io_tree;
614 if (list_empty(&async_cow->extents))
618 while (!list_empty(&async_cow->extents)) {
619 async_extent = list_entry(async_cow->extents.next,
620 struct async_extent, list);
621 list_del(&async_extent->list);
623 io_tree = &BTRFS_I(inode)->io_tree;
626 /* did the compression code fall back to uncompressed IO? */
627 if (!async_extent->pages) {
628 int page_started = 0;
629 unsigned long nr_written = 0;
631 lock_extent(io_tree, async_extent->start,
632 async_extent->start +
633 async_extent->ram_size - 1);
635 /* allocate blocks */
636 ret = cow_file_range(inode, async_cow->locked_page,
638 async_extent->start +
639 async_extent->ram_size - 1,
640 &page_started, &nr_written, 0);
645 * if page_started, cow_file_range inserted an
646 * inline extent and took care of all the unlocking
647 * and IO for us. Otherwise, we need to submit
648 * all those pages down to the drive.
650 if (!page_started && !ret)
651 extent_write_locked_range(io_tree,
652 inode, async_extent->start,
653 async_extent->start +
654 async_extent->ram_size - 1,
658 unlock_page(async_cow->locked_page);
664 lock_extent(io_tree, async_extent->start,
665 async_extent->start + async_extent->ram_size - 1);
667 ret = btrfs_reserve_extent(root,
668 async_extent->compressed_size,
669 async_extent->compressed_size,
670 0, alloc_hint, &ins, 1);
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;
682 if (ret == -ENOSPC) {
683 unlock_extent(io_tree, async_extent->start,
684 async_extent->start +
685 async_extent->ram_size - 1);
692 * here we're doing allocation and writeback of the
695 btrfs_drop_extent_cache(inode, async_extent->start,
696 async_extent->start +
697 async_extent->ram_size - 1, 0);
699 em = alloc_extent_map();
702 goto out_free_reserve;
704 em->start = async_extent->start;
705 em->len = async_extent->ram_size;
706 em->orig_start = em->start;
707 em->mod_start = em->start;
708 em->mod_len = em->len;
710 em->block_start = ins.objectid;
711 em->block_len = ins.offset;
712 em->orig_block_len = ins.offset;
713 em->ram_bytes = async_extent->ram_size;
714 em->bdev = root->fs_info->fs_devices->latest_bdev;
715 em->compress_type = async_extent->compress_type;
716 set_bit(EXTENT_FLAG_PINNED, &em->flags);
717 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
721 write_lock(&em_tree->lock);
722 ret = add_extent_mapping(em_tree, em, 1);
723 write_unlock(&em_tree->lock);
724 if (ret != -EEXIST) {
728 btrfs_drop_extent_cache(inode, async_extent->start,
729 async_extent->start +
730 async_extent->ram_size - 1, 0);
734 goto out_free_reserve;
736 ret = btrfs_add_ordered_extent_compress(inode,
739 async_extent->ram_size,
741 BTRFS_ORDERED_COMPRESSED,
742 async_extent->compress_type);
744 goto out_free_reserve;
747 * clear dirty, set writeback and unlock the pages.
749 extent_clear_unlock_delalloc(inode, async_extent->start,
750 async_extent->start +
751 async_extent->ram_size - 1,
752 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
753 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
755 ret = btrfs_submit_compressed_write(inode,
757 async_extent->ram_size,
759 ins.offset, async_extent->pages,
760 async_extent->nr_pages);
761 alloc_hint = ins.objectid + ins.offset;
771 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
773 extent_clear_unlock_delalloc(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1,
776 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
777 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
778 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
779 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
784 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
787 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
788 struct extent_map *em;
791 read_lock(&em_tree->lock);
792 em = search_extent_mapping(em_tree, start, num_bytes);
795 * if block start isn't an actual block number then find the
796 * first block in this inode and use that as a hint. If that
797 * block is also bogus then just don't worry about it.
799 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
801 em = search_extent_mapping(em_tree, 0, 0);
802 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
803 alloc_hint = em->block_start;
807 alloc_hint = em->block_start;
811 read_unlock(&em_tree->lock);
817 * when extent_io.c finds a delayed allocation range in the file,
818 * the call backs end up in this code. The basic idea is to
819 * allocate extents on disk for the range, and create ordered data structs
820 * in ram to track those extents.
822 * locked_page is the page that writepage had locked already. We use
823 * it to make sure we don't do extra locks or unlocks.
825 * *page_started is set to one if we unlock locked_page and do everything
826 * required to start IO on it. It may be clean and already done with
829 static noinline int cow_file_range(struct inode *inode,
830 struct page *locked_page,
831 u64 start, u64 end, int *page_started,
832 unsigned long *nr_written,
835 struct btrfs_root *root = BTRFS_I(inode)->root;
838 unsigned long ram_size;
841 u64 blocksize = root->sectorsize;
842 struct btrfs_key ins;
843 struct extent_map *em;
844 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
847 BUG_ON(btrfs_is_free_space_inode(inode));
849 num_bytes = ALIGN(end - start + 1, blocksize);
850 num_bytes = max(blocksize, num_bytes);
851 disk_num_bytes = num_bytes;
853 /* if this is a small write inside eof, kick off defrag */
854 if (num_bytes < 64 * 1024 &&
855 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
856 btrfs_add_inode_defrag(NULL, inode);
859 /* lets try to make an inline extent */
860 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
863 extent_clear_unlock_delalloc(inode, start, end, NULL,
864 EXTENT_LOCKED | EXTENT_DELALLOC |
865 EXTENT_DEFRAG, PAGE_UNLOCK |
866 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
869 *nr_written = *nr_written +
870 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
873 } else if (ret < 0) {
878 BUG_ON(disk_num_bytes >
879 btrfs_super_total_bytes(root->fs_info->super_copy));
881 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
882 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
884 while (disk_num_bytes > 0) {
887 cur_alloc_size = disk_num_bytes;
888 ret = btrfs_reserve_extent(root, cur_alloc_size,
889 root->sectorsize, 0, alloc_hint,
894 em = alloc_extent_map();
900 em->orig_start = em->start;
901 ram_size = ins.offset;
902 em->len = ins.offset;
903 em->mod_start = em->start;
904 em->mod_len = em->len;
906 em->block_start = ins.objectid;
907 em->block_len = ins.offset;
908 em->orig_block_len = ins.offset;
909 em->ram_bytes = ram_size;
910 em->bdev = root->fs_info->fs_devices->latest_bdev;
911 set_bit(EXTENT_FLAG_PINNED, &em->flags);
915 write_lock(&em_tree->lock);
916 ret = add_extent_mapping(em_tree, em, 1);
917 write_unlock(&em_tree->lock);
918 if (ret != -EEXIST) {
922 btrfs_drop_extent_cache(inode, start,
923 start + ram_size - 1, 0);
928 cur_alloc_size = ins.offset;
929 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
930 ram_size, cur_alloc_size, 0);
934 if (root->root_key.objectid ==
935 BTRFS_DATA_RELOC_TREE_OBJECTID) {
936 ret = btrfs_reloc_clone_csums(inode, start,
942 if (disk_num_bytes < cur_alloc_size)
945 /* we're not doing compressed IO, don't unlock the first
946 * page (which the caller expects to stay locked), don't
947 * clear any dirty bits and don't set any writeback bits
949 * Do set the Private2 bit so we know this page was properly
950 * setup for writepage
952 op = unlock ? PAGE_UNLOCK : 0;
953 op |= PAGE_SET_PRIVATE2;
955 extent_clear_unlock_delalloc(inode, start,
956 start + ram_size - 1, locked_page,
957 EXTENT_LOCKED | EXTENT_DELALLOC,
959 disk_num_bytes -= cur_alloc_size;
960 num_bytes -= cur_alloc_size;
961 alloc_hint = ins.objectid + ins.offset;
962 start += cur_alloc_size;
968 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
970 extent_clear_unlock_delalloc(inode, start, end, locked_page,
971 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
972 EXTENT_DELALLOC | EXTENT_DEFRAG,
973 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
974 PAGE_SET_WRITEBACK | PAGE_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 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1013 waitqueue_active(&root->fs_info->async_submit_wait))
1014 wake_up(&root->fs_info->async_submit_wait);
1016 if (async_cow->inode)
1017 submit_compressed_extents(async_cow->inode, async_cow);
1020 static noinline void async_cow_free(struct btrfs_work *work)
1022 struct async_cow *async_cow;
1023 async_cow = container_of(work, struct async_cow, work);
1024 if (async_cow->inode)
1025 btrfs_add_delayed_iput(async_cow->inode);
1029 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1030 u64 start, u64 end, int *page_started,
1031 unsigned long *nr_written)
1033 struct async_cow *async_cow;
1034 struct btrfs_root *root = BTRFS_I(inode)->root;
1035 unsigned long nr_pages;
1037 int limit = 10 * 1024 * 1024;
1039 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1040 1, 0, NULL, GFP_NOFS);
1041 while (start < end) {
1042 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1043 BUG_ON(!async_cow); /* -ENOMEM */
1044 async_cow->inode = igrab(inode);
1045 async_cow->root = root;
1046 async_cow->locked_page = locked_page;
1047 async_cow->start = start;
1049 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1052 cur_end = min(end, start + 512 * 1024 - 1);
1054 async_cow->end = cur_end;
1055 INIT_LIST_HEAD(&async_cow->extents);
1057 async_cow->work.func = async_cow_start;
1058 async_cow->work.ordered_func = async_cow_submit;
1059 async_cow->work.ordered_free = async_cow_free;
1060 async_cow->work.flags = 0;
1062 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1064 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1066 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1069 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1070 wait_event(root->fs_info->async_submit_wait,
1071 (atomic_read(&root->fs_info->async_delalloc_pages) <
1075 while (atomic_read(&root->fs_info->async_submit_draining) &&
1076 atomic_read(&root->fs_info->async_delalloc_pages)) {
1077 wait_event(root->fs_info->async_submit_wait,
1078 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1082 *nr_written += nr_pages;
1083 start = cur_end + 1;
1089 static noinline int csum_exist_in_range(struct btrfs_root *root,
1090 u64 bytenr, u64 num_bytes)
1093 struct btrfs_ordered_sum *sums;
1096 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1097 bytenr + num_bytes - 1, &list, 0);
1098 if (ret == 0 && list_empty(&list))
1101 while (!list_empty(&list)) {
1102 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1103 list_del(&sums->list);
1110 * when nowcow writeback call back. This checks for snapshots or COW copies
1111 * of the extents that exist in the file, and COWs the file as required.
1113 * If no cow copies or snapshots exist, we write directly to the existing
1116 static noinline int run_delalloc_nocow(struct inode *inode,
1117 struct page *locked_page,
1118 u64 start, u64 end, int *page_started, int force,
1119 unsigned long *nr_written)
1121 struct btrfs_root *root = BTRFS_I(inode)->root;
1122 struct btrfs_trans_handle *trans;
1123 struct extent_buffer *leaf;
1124 struct btrfs_path *path;
1125 struct btrfs_file_extent_item *fi;
1126 struct btrfs_key found_key;
1141 u64 ino = btrfs_ino(inode);
1143 path = btrfs_alloc_path();
1145 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1146 EXTENT_LOCKED | EXTENT_DELALLOC |
1147 EXTENT_DO_ACCOUNTING |
1148 EXTENT_DEFRAG, PAGE_UNLOCK |
1150 PAGE_SET_WRITEBACK |
1151 PAGE_END_WRITEBACK);
1155 nolock = btrfs_is_free_space_inode(inode);
1158 trans = btrfs_join_transaction_nolock(root);
1160 trans = btrfs_join_transaction(root);
1162 if (IS_ERR(trans)) {
1163 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1164 EXTENT_LOCKED | EXTENT_DELALLOC |
1165 EXTENT_DO_ACCOUNTING |
1166 EXTENT_DEFRAG, PAGE_UNLOCK |
1168 PAGE_SET_WRITEBACK |
1169 PAGE_END_WRITEBACK);
1170 btrfs_free_path(path);
1171 return PTR_ERR(trans);
1174 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1176 cow_start = (u64)-1;
1179 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1182 btrfs_abort_transaction(trans, root, ret);
1185 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1186 leaf = path->nodes[0];
1187 btrfs_item_key_to_cpu(leaf, &found_key,
1188 path->slots[0] - 1);
1189 if (found_key.objectid == ino &&
1190 found_key.type == BTRFS_EXTENT_DATA_KEY)
1195 leaf = path->nodes[0];
1196 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1197 ret = btrfs_next_leaf(root, path);
1199 btrfs_abort_transaction(trans, root, ret);
1204 leaf = path->nodes[0];
1210 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1212 if (found_key.objectid > ino ||
1213 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1214 found_key.offset > end)
1217 if (found_key.offset > cur_offset) {
1218 extent_end = found_key.offset;
1223 fi = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_file_extent_item);
1225 extent_type = btrfs_file_extent_type(leaf, fi);
1227 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1228 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1229 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1230 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1231 extent_offset = btrfs_file_extent_offset(leaf, fi);
1232 extent_end = found_key.offset +
1233 btrfs_file_extent_num_bytes(leaf, fi);
1235 btrfs_file_extent_disk_num_bytes(leaf, fi);
1236 if (extent_end <= start) {
1240 if (disk_bytenr == 0)
1242 if (btrfs_file_extent_compression(leaf, fi) ||
1243 btrfs_file_extent_encryption(leaf, fi) ||
1244 btrfs_file_extent_other_encoding(leaf, fi))
1246 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1248 if (btrfs_extent_readonly(root, disk_bytenr))
1250 if (btrfs_cross_ref_exist(trans, root, ino,
1252 extent_offset, disk_bytenr))
1254 disk_bytenr += extent_offset;
1255 disk_bytenr += cur_offset - found_key.offset;
1256 num_bytes = min(end + 1, extent_end) - cur_offset;
1258 * force cow if csum exists in the range.
1259 * this ensure that csum for a given extent are
1260 * either valid or do not exist.
1262 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1265 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266 extent_end = found_key.offset +
1267 btrfs_file_extent_inline_len(leaf, fi);
1268 extent_end = ALIGN(extent_end, root->sectorsize);
1273 if (extent_end <= start) {
1278 if (cow_start == (u64)-1)
1279 cow_start = cur_offset;
1280 cur_offset = extent_end;
1281 if (cur_offset > end)
1287 btrfs_release_path(path);
1288 if (cow_start != (u64)-1) {
1289 ret = cow_file_range(inode, locked_page,
1290 cow_start, found_key.offset - 1,
1291 page_started, nr_written, 1);
1293 btrfs_abort_transaction(trans, root, ret);
1296 cow_start = (u64)-1;
1299 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300 struct extent_map *em;
1301 struct extent_map_tree *em_tree;
1302 em_tree = &BTRFS_I(inode)->extent_tree;
1303 em = alloc_extent_map();
1304 BUG_ON(!em); /* -ENOMEM */
1305 em->start = cur_offset;
1306 em->orig_start = found_key.offset - extent_offset;
1307 em->len = num_bytes;
1308 em->block_len = num_bytes;
1309 em->block_start = disk_bytenr;
1310 em->orig_block_len = disk_num_bytes;
1311 em->ram_bytes = ram_bytes;
1312 em->bdev = root->fs_info->fs_devices->latest_bdev;
1313 em->mod_start = em->start;
1314 em->mod_len = em->len;
1315 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1316 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1317 em->generation = -1;
1319 write_lock(&em_tree->lock);
1320 ret = add_extent_mapping(em_tree, em, 1);
1321 write_unlock(&em_tree->lock);
1322 if (ret != -EEXIST) {
1323 free_extent_map(em);
1326 btrfs_drop_extent_cache(inode, em->start,
1327 em->start + em->len - 1, 0);
1329 type = BTRFS_ORDERED_PREALLOC;
1331 type = BTRFS_ORDERED_NOCOW;
1334 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1335 num_bytes, num_bytes, type);
1336 BUG_ON(ret); /* -ENOMEM */
1338 if (root->root_key.objectid ==
1339 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1340 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1343 btrfs_abort_transaction(trans, root, ret);
1348 extent_clear_unlock_delalloc(inode, cur_offset,
1349 cur_offset + num_bytes - 1,
1350 locked_page, EXTENT_LOCKED |
1351 EXTENT_DELALLOC, PAGE_UNLOCK |
1353 cur_offset = extent_end;
1354 if (cur_offset > end)
1357 btrfs_release_path(path);
1359 if (cur_offset <= end && cow_start == (u64)-1) {
1360 cow_start = cur_offset;
1364 if (cow_start != (u64)-1) {
1365 ret = cow_file_range(inode, locked_page, cow_start, end,
1366 page_started, nr_written, 1);
1368 btrfs_abort_transaction(trans, root, ret);
1374 err = btrfs_end_transaction(trans, root);
1378 if (ret && cur_offset < end)
1379 extent_clear_unlock_delalloc(inode, cur_offset, end,
1380 locked_page, EXTENT_LOCKED |
1381 EXTENT_DELALLOC | EXTENT_DEFRAG |
1382 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1384 PAGE_SET_WRITEBACK |
1385 PAGE_END_WRITEBACK);
1386 btrfs_free_path(path);
1391 * extent_io.c call back to do delayed allocation processing
1393 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1394 u64 start, u64 end, int *page_started,
1395 unsigned long *nr_written)
1398 struct btrfs_root *root = BTRFS_I(inode)->root;
1400 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1401 ret = run_delalloc_nocow(inode, locked_page, start, end,
1402 page_started, 1, nr_written);
1403 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1404 ret = run_delalloc_nocow(inode, locked_page, start, end,
1405 page_started, 0, nr_written);
1406 } else if (!btrfs_test_opt(root, COMPRESS) &&
1407 !(BTRFS_I(inode)->force_compress) &&
1408 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1409 ret = cow_file_range(inode, locked_page, start, end,
1410 page_started, nr_written, 1);
1412 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1413 &BTRFS_I(inode)->runtime_flags);
1414 ret = cow_file_range_async(inode, locked_page, start, end,
1415 page_started, nr_written);
1420 static void btrfs_split_extent_hook(struct inode *inode,
1421 struct extent_state *orig, u64 split)
1423 /* not delalloc, ignore it */
1424 if (!(orig->state & EXTENT_DELALLOC))
1427 spin_lock(&BTRFS_I(inode)->lock);
1428 BTRFS_I(inode)->outstanding_extents++;
1429 spin_unlock(&BTRFS_I(inode)->lock);
1433 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1434 * extents so we can keep track of new extents that are just merged onto old
1435 * extents, such as when we are doing sequential writes, so we can properly
1436 * account for the metadata space we'll need.
1438 static void btrfs_merge_extent_hook(struct inode *inode,
1439 struct extent_state *new,
1440 struct extent_state *other)
1442 /* not delalloc, ignore it */
1443 if (!(other->state & EXTENT_DELALLOC))
1446 spin_lock(&BTRFS_I(inode)->lock);
1447 BTRFS_I(inode)->outstanding_extents--;
1448 spin_unlock(&BTRFS_I(inode)->lock);
1451 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1452 struct inode *inode)
1454 spin_lock(&root->delalloc_lock);
1455 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1456 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1457 &root->delalloc_inodes);
1458 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1459 &BTRFS_I(inode)->runtime_flags);
1460 root->nr_delalloc_inodes++;
1461 if (root->nr_delalloc_inodes == 1) {
1462 spin_lock(&root->fs_info->delalloc_root_lock);
1463 BUG_ON(!list_empty(&root->delalloc_root));
1464 list_add_tail(&root->delalloc_root,
1465 &root->fs_info->delalloc_roots);
1466 spin_unlock(&root->fs_info->delalloc_root_lock);
1469 spin_unlock(&root->delalloc_lock);
1472 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1473 struct inode *inode)
1475 spin_lock(&root->delalloc_lock);
1476 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1477 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1478 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1479 &BTRFS_I(inode)->runtime_flags);
1480 root->nr_delalloc_inodes--;
1481 if (!root->nr_delalloc_inodes) {
1482 spin_lock(&root->fs_info->delalloc_root_lock);
1483 BUG_ON(list_empty(&root->delalloc_root));
1484 list_del_init(&root->delalloc_root);
1485 spin_unlock(&root->fs_info->delalloc_root_lock);
1488 spin_unlock(&root->delalloc_lock);
1492 * extent_io.c set_bit_hook, used to track delayed allocation
1493 * bytes in this file, and to maintain the list of inodes that
1494 * have pending delalloc work to be done.
1496 static void btrfs_set_bit_hook(struct inode *inode,
1497 struct extent_state *state, unsigned long *bits)
1501 * set_bit and clear bit hooks normally require _irqsave/restore
1502 * but in this case, we are only testing for the DELALLOC
1503 * bit, which is only set or cleared with irqs on
1505 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1506 struct btrfs_root *root = BTRFS_I(inode)->root;
1507 u64 len = state->end + 1 - state->start;
1508 bool do_list = !btrfs_is_free_space_inode(inode);
1510 if (*bits & EXTENT_FIRST_DELALLOC) {
1511 *bits &= ~EXTENT_FIRST_DELALLOC;
1513 spin_lock(&BTRFS_I(inode)->lock);
1514 BTRFS_I(inode)->outstanding_extents++;
1515 spin_unlock(&BTRFS_I(inode)->lock);
1518 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1519 root->fs_info->delalloc_batch);
1520 spin_lock(&BTRFS_I(inode)->lock);
1521 BTRFS_I(inode)->delalloc_bytes += len;
1522 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1523 &BTRFS_I(inode)->runtime_flags))
1524 btrfs_add_delalloc_inodes(root, inode);
1525 spin_unlock(&BTRFS_I(inode)->lock);
1530 * extent_io.c clear_bit_hook, see set_bit_hook for why
1532 static void btrfs_clear_bit_hook(struct inode *inode,
1533 struct extent_state *state,
1534 unsigned long *bits)
1537 * set_bit and clear bit hooks normally require _irqsave/restore
1538 * but in this case, we are only testing for the DELALLOC
1539 * bit, which is only set or cleared with irqs on
1541 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1542 struct btrfs_root *root = BTRFS_I(inode)->root;
1543 u64 len = state->end + 1 - state->start;
1544 bool do_list = !btrfs_is_free_space_inode(inode);
1546 if (*bits & EXTENT_FIRST_DELALLOC) {
1547 *bits &= ~EXTENT_FIRST_DELALLOC;
1548 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1549 spin_lock(&BTRFS_I(inode)->lock);
1550 BTRFS_I(inode)->outstanding_extents--;
1551 spin_unlock(&BTRFS_I(inode)->lock);
1554 if (*bits & EXTENT_DO_ACCOUNTING)
1555 btrfs_delalloc_release_metadata(inode, len);
1557 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1558 && do_list && !(state->state & EXTENT_NORESERVE))
1559 btrfs_free_reserved_data_space(inode, len);
1561 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1562 root->fs_info->delalloc_batch);
1563 spin_lock(&BTRFS_I(inode)->lock);
1564 BTRFS_I(inode)->delalloc_bytes -= len;
1565 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1566 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1567 &BTRFS_I(inode)->runtime_flags))
1568 btrfs_del_delalloc_inode(root, inode);
1569 spin_unlock(&BTRFS_I(inode)->lock);
1574 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1575 * we don't create bios that span stripes or chunks
1577 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1578 size_t size, struct bio *bio,
1579 unsigned long bio_flags)
1581 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1582 u64 logical = (u64)bio->bi_sector << 9;
1587 if (bio_flags & EXTENT_BIO_COMPRESSED)
1590 length = bio->bi_size;
1591 map_length = length;
1592 ret = btrfs_map_block(root->fs_info, rw, logical,
1593 &map_length, NULL, 0);
1594 /* Will always return 0 with map_multi == NULL */
1596 if (map_length < length + size)
1602 * in order to insert checksums into the metadata in large chunks,
1603 * we wait until bio submission time. All the pages in the bio are
1604 * checksummed and sums are attached onto the ordered extent record.
1606 * At IO completion time the cums attached on the ordered extent record
1607 * are inserted into the btree
1609 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1610 struct bio *bio, int mirror_num,
1611 unsigned long bio_flags,
1614 struct btrfs_root *root = BTRFS_I(inode)->root;
1617 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1618 BUG_ON(ret); /* -ENOMEM */
1623 * in order to insert checksums into the metadata in large chunks,
1624 * we wait until bio submission time. All the pages in the bio are
1625 * checksummed and sums are attached onto the ordered extent record.
1627 * At IO completion time the cums attached on the ordered extent record
1628 * are inserted into the btree
1630 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1631 int mirror_num, unsigned long bio_flags,
1634 struct btrfs_root *root = BTRFS_I(inode)->root;
1637 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1639 bio_endio(bio, ret);
1644 * extent_io.c submission hook. This does the right thing for csum calculation
1645 * on write, or reading the csums from the tree before a read
1647 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1648 int mirror_num, unsigned long bio_flags,
1651 struct btrfs_root *root = BTRFS_I(inode)->root;
1655 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1657 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1659 if (btrfs_is_free_space_inode(inode))
1662 if (!(rw & REQ_WRITE)) {
1663 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1667 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1668 ret = btrfs_submit_compressed_read(inode, bio,
1672 } else if (!skip_sum) {
1673 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1678 } else if (async && !skip_sum) {
1679 /* csum items have already been cloned */
1680 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1682 /* we're doing a write, do the async checksumming */
1683 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1684 inode, rw, bio, mirror_num,
1685 bio_flags, bio_offset,
1686 __btrfs_submit_bio_start,
1687 __btrfs_submit_bio_done);
1689 } else if (!skip_sum) {
1690 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1696 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1700 bio_endio(bio, ret);
1705 * given a list of ordered sums record them in the inode. This happens
1706 * at IO completion time based on sums calculated at bio submission time.
1708 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1709 struct inode *inode, u64 file_offset,
1710 struct list_head *list)
1712 struct btrfs_ordered_sum *sum;
1714 list_for_each_entry(sum, list, list) {
1715 trans->adding_csums = 1;
1716 btrfs_csum_file_blocks(trans,
1717 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1718 trans->adding_csums = 0;
1723 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1724 struct extent_state **cached_state)
1726 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1727 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1728 cached_state, GFP_NOFS);
1731 /* see btrfs_writepage_start_hook for details on why this is required */
1732 struct btrfs_writepage_fixup {
1734 struct btrfs_work work;
1737 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1739 struct btrfs_writepage_fixup *fixup;
1740 struct btrfs_ordered_extent *ordered;
1741 struct extent_state *cached_state = NULL;
1743 struct inode *inode;
1748 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1752 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1753 ClearPageChecked(page);
1757 inode = page->mapping->host;
1758 page_start = page_offset(page);
1759 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1761 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1764 /* already ordered? We're done */
1765 if (PagePrivate2(page))
1768 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1770 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1771 page_end, &cached_state, GFP_NOFS);
1773 btrfs_start_ordered_extent(inode, ordered, 1);
1774 btrfs_put_ordered_extent(ordered);
1778 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1780 mapping_set_error(page->mapping, ret);
1781 end_extent_writepage(page, ret, page_start, page_end);
1782 ClearPageChecked(page);
1786 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1787 ClearPageChecked(page);
1788 set_page_dirty(page);
1790 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1791 &cached_state, GFP_NOFS);
1794 page_cache_release(page);
1799 * There are a few paths in the higher layers of the kernel that directly
1800 * set the page dirty bit without asking the filesystem if it is a
1801 * good idea. This causes problems because we want to make sure COW
1802 * properly happens and the data=ordered rules are followed.
1804 * In our case any range that doesn't have the ORDERED bit set
1805 * hasn't been properly setup for IO. We kick off an async process
1806 * to fix it up. The async helper will wait for ordered extents, set
1807 * the delalloc bit and make it safe to write the page.
1809 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1811 struct inode *inode = page->mapping->host;
1812 struct btrfs_writepage_fixup *fixup;
1813 struct btrfs_root *root = BTRFS_I(inode)->root;
1815 /* this page is properly in the ordered list */
1816 if (TestClearPagePrivate2(page))
1819 if (PageChecked(page))
1822 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1826 SetPageChecked(page);
1827 page_cache_get(page);
1828 fixup->work.func = btrfs_writepage_fixup_worker;
1830 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1834 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1835 struct inode *inode, u64 file_pos,
1836 u64 disk_bytenr, u64 disk_num_bytes,
1837 u64 num_bytes, u64 ram_bytes,
1838 u8 compression, u8 encryption,
1839 u16 other_encoding, int extent_type)
1841 struct btrfs_root *root = BTRFS_I(inode)->root;
1842 struct btrfs_file_extent_item *fi;
1843 struct btrfs_path *path;
1844 struct extent_buffer *leaf;
1845 struct btrfs_key ins;
1848 path = btrfs_alloc_path();
1852 path->leave_spinning = 1;
1855 * we may be replacing one extent in the tree with another.
1856 * The new extent is pinned in the extent map, and we don't want
1857 * to drop it from the cache until it is completely in the btree.
1859 * So, tell btrfs_drop_extents to leave this extent in the cache.
1860 * the caller is expected to unpin it and allow it to be merged
1863 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1864 file_pos + num_bytes, 0);
1868 ins.objectid = btrfs_ino(inode);
1869 ins.offset = file_pos;
1870 ins.type = BTRFS_EXTENT_DATA_KEY;
1871 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1874 leaf = path->nodes[0];
1875 fi = btrfs_item_ptr(leaf, path->slots[0],
1876 struct btrfs_file_extent_item);
1877 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1878 btrfs_set_file_extent_type(leaf, fi, extent_type);
1879 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1880 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1881 btrfs_set_file_extent_offset(leaf, fi, 0);
1882 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1883 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1884 btrfs_set_file_extent_compression(leaf, fi, compression);
1885 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1886 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1888 btrfs_mark_buffer_dirty(leaf);
1889 btrfs_release_path(path);
1891 inode_add_bytes(inode, num_bytes);
1893 ins.objectid = disk_bytenr;
1894 ins.offset = disk_num_bytes;
1895 ins.type = BTRFS_EXTENT_ITEM_KEY;
1896 ret = btrfs_alloc_reserved_file_extent(trans, root,
1897 root->root_key.objectid,
1898 btrfs_ino(inode), file_pos, &ins);
1900 btrfs_free_path(path);
1905 /* snapshot-aware defrag */
1906 struct sa_defrag_extent_backref {
1907 struct rb_node node;
1908 struct old_sa_defrag_extent *old;
1917 struct old_sa_defrag_extent {
1918 struct list_head list;
1919 struct new_sa_defrag_extent *new;
1928 struct new_sa_defrag_extent {
1929 struct rb_root root;
1930 struct list_head head;
1931 struct btrfs_path *path;
1932 struct inode *inode;
1940 static int backref_comp(struct sa_defrag_extent_backref *b1,
1941 struct sa_defrag_extent_backref *b2)
1943 if (b1->root_id < b2->root_id)
1945 else if (b1->root_id > b2->root_id)
1948 if (b1->inum < b2->inum)
1950 else if (b1->inum > b2->inum)
1953 if (b1->file_pos < b2->file_pos)
1955 else if (b1->file_pos > b2->file_pos)
1959 * [------------------------------] ===> (a range of space)
1960 * |<--->| |<---->| =============> (fs/file tree A)
1961 * |<---------------------------->| ===> (fs/file tree B)
1963 * A range of space can refer to two file extents in one tree while
1964 * refer to only one file extent in another tree.
1966 * So we may process a disk offset more than one time(two extents in A)
1967 * and locate at the same extent(one extent in B), then insert two same
1968 * backrefs(both refer to the extent in B).
1973 static void backref_insert(struct rb_root *root,
1974 struct sa_defrag_extent_backref *backref)
1976 struct rb_node **p = &root->rb_node;
1977 struct rb_node *parent = NULL;
1978 struct sa_defrag_extent_backref *entry;
1983 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1985 ret = backref_comp(backref, entry);
1989 p = &(*p)->rb_right;
1992 rb_link_node(&backref->node, parent, p);
1993 rb_insert_color(&backref->node, root);
1997 * Note the backref might has changed, and in this case we just return 0.
1999 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2002 struct btrfs_file_extent_item *extent;
2003 struct btrfs_fs_info *fs_info;
2004 struct old_sa_defrag_extent *old = ctx;
2005 struct new_sa_defrag_extent *new = old->new;
2006 struct btrfs_path *path = new->path;
2007 struct btrfs_key key;
2008 struct btrfs_root *root;
2009 struct sa_defrag_extent_backref *backref;
2010 struct extent_buffer *leaf;
2011 struct inode *inode = new->inode;
2017 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2018 inum == btrfs_ino(inode))
2021 key.objectid = root_id;
2022 key.type = BTRFS_ROOT_ITEM_KEY;
2023 key.offset = (u64)-1;
2025 fs_info = BTRFS_I(inode)->root->fs_info;
2026 root = btrfs_read_fs_root_no_name(fs_info, &key);
2028 if (PTR_ERR(root) == -ENOENT)
2031 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2032 inum, offset, root_id);
2033 return PTR_ERR(root);
2036 key.objectid = inum;
2037 key.type = BTRFS_EXTENT_DATA_KEY;
2038 if (offset > (u64)-1 << 32)
2041 key.offset = offset;
2043 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2053 leaf = path->nodes[0];
2054 slot = path->slots[0];
2056 if (slot >= btrfs_header_nritems(leaf)) {
2057 ret = btrfs_next_leaf(root, path);
2060 } else if (ret > 0) {
2069 btrfs_item_key_to_cpu(leaf, &key, slot);
2071 if (key.objectid > inum)
2074 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2077 extent = btrfs_item_ptr(leaf, slot,
2078 struct btrfs_file_extent_item);
2080 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2084 * 'offset' refers to the exact key.offset,
2085 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2086 * (key.offset - extent_offset).
2088 if (key.offset != offset)
2091 extent_offset = btrfs_file_extent_offset(leaf, extent);
2092 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2094 if (extent_offset >= old->extent_offset + old->offset +
2095 old->len || extent_offset + num_bytes <=
2096 old->extent_offset + old->offset)
2101 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2107 backref->root_id = root_id;
2108 backref->inum = inum;
2109 backref->file_pos = offset;
2110 backref->num_bytes = num_bytes;
2111 backref->extent_offset = extent_offset;
2112 backref->generation = btrfs_file_extent_generation(leaf, extent);
2114 backref_insert(&new->root, backref);
2117 btrfs_release_path(path);
2122 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2123 struct new_sa_defrag_extent *new)
2125 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2126 struct old_sa_defrag_extent *old, *tmp;
2131 list_for_each_entry_safe(old, tmp, &new->head, list) {
2132 ret = iterate_inodes_from_logical(old->bytenr +
2133 old->extent_offset, fs_info,
2134 path, record_one_backref,
2136 BUG_ON(ret < 0 && ret != -ENOENT);
2138 /* no backref to be processed for this extent */
2140 list_del(&old->list);
2145 if (list_empty(&new->head))
2151 static int relink_is_mergable(struct extent_buffer *leaf,
2152 struct btrfs_file_extent_item *fi,
2153 struct new_sa_defrag_extent *new)
2155 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2158 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2161 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2164 if (btrfs_file_extent_encryption(leaf, fi) ||
2165 btrfs_file_extent_other_encoding(leaf, fi))
2172 * Note the backref might has changed, and in this case we just return 0.
2174 static noinline int relink_extent_backref(struct btrfs_path *path,
2175 struct sa_defrag_extent_backref *prev,
2176 struct sa_defrag_extent_backref *backref)
2178 struct btrfs_file_extent_item *extent;
2179 struct btrfs_file_extent_item *item;
2180 struct btrfs_ordered_extent *ordered;
2181 struct btrfs_trans_handle *trans;
2182 struct btrfs_fs_info *fs_info;
2183 struct btrfs_root *root;
2184 struct btrfs_key key;
2185 struct extent_buffer *leaf;
2186 struct old_sa_defrag_extent *old = backref->old;
2187 struct new_sa_defrag_extent *new = old->new;
2188 struct inode *src_inode = new->inode;
2189 struct inode *inode;
2190 struct extent_state *cached = NULL;
2199 if (prev && prev->root_id == backref->root_id &&
2200 prev->inum == backref->inum &&
2201 prev->file_pos + prev->num_bytes == backref->file_pos)
2204 /* step 1: get root */
2205 key.objectid = backref->root_id;
2206 key.type = BTRFS_ROOT_ITEM_KEY;
2207 key.offset = (u64)-1;
2209 fs_info = BTRFS_I(src_inode)->root->fs_info;
2210 index = srcu_read_lock(&fs_info->subvol_srcu);
2212 root = btrfs_read_fs_root_no_name(fs_info, &key);
2214 srcu_read_unlock(&fs_info->subvol_srcu, index);
2215 if (PTR_ERR(root) == -ENOENT)
2217 return PTR_ERR(root);
2220 /* step 2: get inode */
2221 key.objectid = backref->inum;
2222 key.type = BTRFS_INODE_ITEM_KEY;
2225 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2226 if (IS_ERR(inode)) {
2227 srcu_read_unlock(&fs_info->subvol_srcu, index);
2231 srcu_read_unlock(&fs_info->subvol_srcu, index);
2233 /* step 3: relink backref */
2234 lock_start = backref->file_pos;
2235 lock_end = backref->file_pos + backref->num_bytes - 1;
2236 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2239 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2241 btrfs_put_ordered_extent(ordered);
2245 trans = btrfs_join_transaction(root);
2246 if (IS_ERR(trans)) {
2247 ret = PTR_ERR(trans);
2251 key.objectid = backref->inum;
2252 key.type = BTRFS_EXTENT_DATA_KEY;
2253 key.offset = backref->file_pos;
2255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2258 } else if (ret > 0) {
2263 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2264 struct btrfs_file_extent_item);
2266 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2267 backref->generation)
2270 btrfs_release_path(path);
2272 start = backref->file_pos;
2273 if (backref->extent_offset < old->extent_offset + old->offset)
2274 start += old->extent_offset + old->offset -
2275 backref->extent_offset;
2277 len = min(backref->extent_offset + backref->num_bytes,
2278 old->extent_offset + old->offset + old->len);
2279 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2281 ret = btrfs_drop_extents(trans, root, inode, start,
2286 key.objectid = btrfs_ino(inode);
2287 key.type = BTRFS_EXTENT_DATA_KEY;
2290 path->leave_spinning = 1;
2292 struct btrfs_file_extent_item *fi;
2294 struct btrfs_key found_key;
2296 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2301 leaf = path->nodes[0];
2302 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2304 fi = btrfs_item_ptr(leaf, path->slots[0],
2305 struct btrfs_file_extent_item);
2306 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2308 if (extent_len + found_key.offset == start &&
2309 relink_is_mergable(leaf, fi, new)) {
2310 btrfs_set_file_extent_num_bytes(leaf, fi,
2312 btrfs_mark_buffer_dirty(leaf);
2313 inode_add_bytes(inode, len);
2319 btrfs_release_path(path);
2324 ret = btrfs_insert_empty_item(trans, root, path, &key,
2327 btrfs_abort_transaction(trans, root, ret);
2331 leaf = path->nodes[0];
2332 item = btrfs_item_ptr(leaf, path->slots[0],
2333 struct btrfs_file_extent_item);
2334 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2335 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2336 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2337 btrfs_set_file_extent_num_bytes(leaf, item, len);
2338 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2339 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2340 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2341 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2342 btrfs_set_file_extent_encryption(leaf, item, 0);
2343 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2345 btrfs_mark_buffer_dirty(leaf);
2346 inode_add_bytes(inode, len);
2347 btrfs_release_path(path);
2349 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2351 backref->root_id, backref->inum,
2352 new->file_pos, 0); /* start - extent_offset */
2354 btrfs_abort_transaction(trans, root, ret);
2360 btrfs_release_path(path);
2361 path->leave_spinning = 0;
2362 btrfs_end_transaction(trans, root);
2364 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2370 static void relink_file_extents(struct new_sa_defrag_extent *new)
2372 struct btrfs_path *path;
2373 struct old_sa_defrag_extent *old, *tmp;
2374 struct sa_defrag_extent_backref *backref;
2375 struct sa_defrag_extent_backref *prev = NULL;
2376 struct inode *inode;
2377 struct btrfs_root *root;
2378 struct rb_node *node;
2382 root = BTRFS_I(inode)->root;
2384 path = btrfs_alloc_path();
2388 if (!record_extent_backrefs(path, new)) {
2389 btrfs_free_path(path);
2392 btrfs_release_path(path);
2395 node = rb_first(&new->root);
2398 rb_erase(node, &new->root);
2400 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2402 ret = relink_extent_backref(path, prev, backref);
2415 btrfs_free_path(path);
2417 list_for_each_entry_safe(old, tmp, &new->head, list) {
2418 list_del(&old->list);
2422 atomic_dec(&root->fs_info->defrag_running);
2423 wake_up(&root->fs_info->transaction_wait);
2428 static struct new_sa_defrag_extent *
2429 record_old_file_extents(struct inode *inode,
2430 struct btrfs_ordered_extent *ordered)
2432 struct btrfs_root *root = BTRFS_I(inode)->root;
2433 struct btrfs_path *path;
2434 struct btrfs_key key;
2435 struct old_sa_defrag_extent *old, *tmp;
2436 struct new_sa_defrag_extent *new;
2439 new = kmalloc(sizeof(*new), GFP_NOFS);
2444 new->file_pos = ordered->file_offset;
2445 new->len = ordered->len;
2446 new->bytenr = ordered->start;
2447 new->disk_len = ordered->disk_len;
2448 new->compress_type = ordered->compress_type;
2449 new->root = RB_ROOT;
2450 INIT_LIST_HEAD(&new->head);
2452 path = btrfs_alloc_path();
2456 key.objectid = btrfs_ino(inode);
2457 key.type = BTRFS_EXTENT_DATA_KEY;
2458 key.offset = new->file_pos;
2460 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2463 if (ret > 0 && path->slots[0] > 0)
2466 /* find out all the old extents for the file range */
2468 struct btrfs_file_extent_item *extent;
2469 struct extent_buffer *l;
2478 slot = path->slots[0];
2480 if (slot >= btrfs_header_nritems(l)) {
2481 ret = btrfs_next_leaf(root, path);
2489 btrfs_item_key_to_cpu(l, &key, slot);
2491 if (key.objectid != btrfs_ino(inode))
2493 if (key.type != BTRFS_EXTENT_DATA_KEY)
2495 if (key.offset >= new->file_pos + new->len)
2498 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2500 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2501 if (key.offset + num_bytes < new->file_pos)
2504 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2508 extent_offset = btrfs_file_extent_offset(l, extent);
2510 old = kmalloc(sizeof(*old), GFP_NOFS);
2514 offset = max(new->file_pos, key.offset);
2515 end = min(new->file_pos + new->len, key.offset + num_bytes);
2517 old->bytenr = disk_bytenr;
2518 old->extent_offset = extent_offset;
2519 old->offset = offset - key.offset;
2520 old->len = end - offset;
2523 list_add_tail(&old->list, &new->head);
2529 btrfs_free_path(path);
2530 atomic_inc(&root->fs_info->defrag_running);
2535 list_for_each_entry_safe(old, tmp, &new->head, list) {
2536 list_del(&old->list);
2540 btrfs_free_path(path);
2547 * helper function for btrfs_finish_ordered_io, this
2548 * just reads in some of the csum leaves to prime them into ram
2549 * before we start the transaction. It limits the amount of btree
2550 * reads required while inside the transaction.
2552 /* as ordered data IO finishes, this gets called so we can finish
2553 * an ordered extent if the range of bytes in the file it covers are
2556 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2558 struct inode *inode = ordered_extent->inode;
2559 struct btrfs_root *root = BTRFS_I(inode)->root;
2560 struct btrfs_trans_handle *trans = NULL;
2561 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2562 struct extent_state *cached_state = NULL;
2563 struct new_sa_defrag_extent *new = NULL;
2564 int compress_type = 0;
2566 u64 logical_len = ordered_extent->len;
2568 bool truncated = false;
2570 nolock = btrfs_is_free_space_inode(inode);
2572 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2577 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2579 logical_len = ordered_extent->truncated_len;
2580 /* Truncated the entire extent, don't bother adding */
2585 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2586 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2587 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2589 trans = btrfs_join_transaction_nolock(root);
2591 trans = btrfs_join_transaction(root);
2592 if (IS_ERR(trans)) {
2593 ret = PTR_ERR(trans);
2597 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2598 ret = btrfs_update_inode_fallback(trans, root, inode);
2599 if (ret) /* -ENOMEM or corruption */
2600 btrfs_abort_transaction(trans, root, ret);
2604 lock_extent_bits(io_tree, ordered_extent->file_offset,
2605 ordered_extent->file_offset + ordered_extent->len - 1,
2608 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2609 ordered_extent->file_offset + ordered_extent->len - 1,
2610 EXTENT_DEFRAG, 1, cached_state);
2612 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2613 if (last_snapshot >= BTRFS_I(inode)->generation)
2614 /* the inode is shared */
2615 new = record_old_file_extents(inode, ordered_extent);
2617 clear_extent_bit(io_tree, ordered_extent->file_offset,
2618 ordered_extent->file_offset + ordered_extent->len - 1,
2619 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2623 trans = btrfs_join_transaction_nolock(root);
2625 trans = btrfs_join_transaction(root);
2626 if (IS_ERR(trans)) {
2627 ret = PTR_ERR(trans);
2631 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2633 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2634 compress_type = ordered_extent->compress_type;
2635 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2636 BUG_ON(compress_type);
2637 ret = btrfs_mark_extent_written(trans, inode,
2638 ordered_extent->file_offset,
2639 ordered_extent->file_offset +
2642 BUG_ON(root == root->fs_info->tree_root);
2643 ret = insert_reserved_file_extent(trans, inode,
2644 ordered_extent->file_offset,
2645 ordered_extent->start,
2646 ordered_extent->disk_len,
2647 logical_len, logical_len,
2648 compress_type, 0, 0,
2649 BTRFS_FILE_EXTENT_REG);
2651 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2652 ordered_extent->file_offset, ordered_extent->len,
2655 btrfs_abort_transaction(trans, root, ret);
2659 add_pending_csums(trans, inode, ordered_extent->file_offset,
2660 &ordered_extent->list);
2662 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2663 ret = btrfs_update_inode_fallback(trans, root, inode);
2664 if (ret) { /* -ENOMEM or corruption */
2665 btrfs_abort_transaction(trans, root, ret);
2670 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2671 ordered_extent->file_offset +
2672 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2674 if (root != root->fs_info->tree_root)
2675 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2677 btrfs_end_transaction(trans, root);
2679 if (ret || truncated) {
2683 start = ordered_extent->file_offset + logical_len;
2685 start = ordered_extent->file_offset;
2686 end = ordered_extent->file_offset + ordered_extent->len - 1;
2687 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2689 /* Drop the cache for the part of the extent we didn't write. */
2690 btrfs_drop_extent_cache(inode, start, end, 0);
2693 * If the ordered extent had an IOERR or something else went
2694 * wrong we need to return the space for this ordered extent
2695 * back to the allocator. We only free the extent in the
2696 * truncated case if we didn't write out the extent at all.
2698 if ((ret || !logical_len) &&
2699 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2700 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2701 btrfs_free_reserved_extent(root, ordered_extent->start,
2702 ordered_extent->disk_len);
2707 * This needs to be done to make sure anybody waiting knows we are done
2708 * updating everything for this ordered extent.
2710 btrfs_remove_ordered_extent(inode, ordered_extent);
2712 /* for snapshot-aware defrag */
2714 relink_file_extents(new);
2717 btrfs_put_ordered_extent(ordered_extent);
2718 /* once for the tree */
2719 btrfs_put_ordered_extent(ordered_extent);
2724 static void finish_ordered_fn(struct btrfs_work *work)
2726 struct btrfs_ordered_extent *ordered_extent;
2727 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2728 btrfs_finish_ordered_io(ordered_extent);
2731 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2732 struct extent_state *state, int uptodate)
2734 struct inode *inode = page->mapping->host;
2735 struct btrfs_root *root = BTRFS_I(inode)->root;
2736 struct btrfs_ordered_extent *ordered_extent = NULL;
2737 struct btrfs_workers *workers;
2739 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2741 ClearPagePrivate2(page);
2742 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2743 end - start + 1, uptodate))
2746 ordered_extent->work.func = finish_ordered_fn;
2747 ordered_extent->work.flags = 0;
2749 if (btrfs_is_free_space_inode(inode))
2750 workers = &root->fs_info->endio_freespace_worker;
2752 workers = &root->fs_info->endio_write_workers;
2753 btrfs_queue_worker(workers, &ordered_extent->work);
2759 * when reads are done, we need to check csums to verify the data is correct
2760 * if there's a match, we allow the bio to finish. If not, the code in
2761 * extent_io.c will try to find good copies for us.
2763 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2764 u64 phy_offset, struct page *page,
2765 u64 start, u64 end, int mirror)
2767 size_t offset = start - page_offset(page);
2768 struct inode *inode = page->mapping->host;
2769 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2771 struct btrfs_root *root = BTRFS_I(inode)->root;
2774 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2775 DEFAULT_RATELIMIT_BURST);
2777 if (PageChecked(page)) {
2778 ClearPageChecked(page);
2782 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2785 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2786 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2787 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2792 phy_offset >>= inode->i_sb->s_blocksize_bits;
2793 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2795 kaddr = kmap_atomic(page);
2796 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2797 btrfs_csum_final(csum, (char *)&csum);
2798 if (csum != csum_expected)
2801 kunmap_atomic(kaddr);
2806 if (__ratelimit(&_rs))
2807 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2808 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2809 memset(kaddr + offset, 1, end - start + 1);
2810 flush_dcache_page(page);
2811 kunmap_atomic(kaddr);
2812 if (csum_expected == 0)
2817 struct delayed_iput {
2818 struct list_head list;
2819 struct inode *inode;
2822 /* JDM: If this is fs-wide, why can't we add a pointer to
2823 * btrfs_inode instead and avoid the allocation? */
2824 void btrfs_add_delayed_iput(struct inode *inode)
2826 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2827 struct delayed_iput *delayed;
2829 if (atomic_add_unless(&inode->i_count, -1, 1))
2832 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2833 delayed->inode = inode;
2835 spin_lock(&fs_info->delayed_iput_lock);
2836 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2837 spin_unlock(&fs_info->delayed_iput_lock);
2840 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2843 struct btrfs_fs_info *fs_info = root->fs_info;
2844 struct delayed_iput *delayed;
2847 spin_lock(&fs_info->delayed_iput_lock);
2848 empty = list_empty(&fs_info->delayed_iputs);
2849 spin_unlock(&fs_info->delayed_iput_lock);
2853 spin_lock(&fs_info->delayed_iput_lock);
2854 list_splice_init(&fs_info->delayed_iputs, &list);
2855 spin_unlock(&fs_info->delayed_iput_lock);
2857 while (!list_empty(&list)) {
2858 delayed = list_entry(list.next, struct delayed_iput, list);
2859 list_del(&delayed->list);
2860 iput(delayed->inode);
2866 * This is called in transaction commit time. If there are no orphan
2867 * files in the subvolume, it removes orphan item and frees block_rsv
2870 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2871 struct btrfs_root *root)
2873 struct btrfs_block_rsv *block_rsv;
2876 if (atomic_read(&root->orphan_inodes) ||
2877 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2880 spin_lock(&root->orphan_lock);
2881 if (atomic_read(&root->orphan_inodes)) {
2882 spin_unlock(&root->orphan_lock);
2886 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2887 spin_unlock(&root->orphan_lock);
2891 block_rsv = root->orphan_block_rsv;
2892 root->orphan_block_rsv = NULL;
2893 spin_unlock(&root->orphan_lock);
2895 if (root->orphan_item_inserted &&
2896 btrfs_root_refs(&root->root_item) > 0) {
2897 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2898 root->root_key.objectid);
2900 btrfs_abort_transaction(trans, root, ret);
2902 root->orphan_item_inserted = 0;
2906 WARN_ON(block_rsv->size > 0);
2907 btrfs_free_block_rsv(root, block_rsv);
2912 * This creates an orphan entry for the given inode in case something goes
2913 * wrong in the middle of an unlink/truncate.
2915 * NOTE: caller of this function should reserve 5 units of metadata for
2918 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2920 struct btrfs_root *root = BTRFS_I(inode)->root;
2921 struct btrfs_block_rsv *block_rsv = NULL;
2926 if (!root->orphan_block_rsv) {
2927 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2932 spin_lock(&root->orphan_lock);
2933 if (!root->orphan_block_rsv) {
2934 root->orphan_block_rsv = block_rsv;
2935 } else if (block_rsv) {
2936 btrfs_free_block_rsv(root, block_rsv);
2940 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2941 &BTRFS_I(inode)->runtime_flags)) {
2944 * For proper ENOSPC handling, we should do orphan
2945 * cleanup when mounting. But this introduces backward
2946 * compatibility issue.
2948 if (!xchg(&root->orphan_item_inserted, 1))
2954 atomic_inc(&root->orphan_inodes);
2957 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2958 &BTRFS_I(inode)->runtime_flags))
2960 spin_unlock(&root->orphan_lock);
2962 /* grab metadata reservation from transaction handle */
2964 ret = btrfs_orphan_reserve_metadata(trans, inode);
2965 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2968 /* insert an orphan item to track this unlinked/truncated file */
2970 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2973 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2974 &BTRFS_I(inode)->runtime_flags);
2975 btrfs_orphan_release_metadata(inode);
2977 if (ret != -EEXIST) {
2978 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2979 &BTRFS_I(inode)->runtime_flags);
2980 btrfs_abort_transaction(trans, root, ret);
2987 /* insert an orphan item to track subvolume contains orphan files */
2989 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2990 root->root_key.objectid);
2991 if (ret && ret != -EEXIST) {
2992 btrfs_abort_transaction(trans, root, ret);
3000 * We have done the truncate/delete so we can go ahead and remove the orphan
3001 * item for this particular inode.
3003 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3004 struct inode *inode)
3006 struct btrfs_root *root = BTRFS_I(inode)->root;
3007 int delete_item = 0;
3008 int release_rsv = 0;
3011 spin_lock(&root->orphan_lock);
3012 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3013 &BTRFS_I(inode)->runtime_flags))
3016 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3017 &BTRFS_I(inode)->runtime_flags))
3019 spin_unlock(&root->orphan_lock);
3021 if (trans && delete_item)
3022 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3025 btrfs_orphan_release_metadata(inode);
3026 atomic_dec(&root->orphan_inodes);
3033 * this cleans up any orphans that may be left on the list from the last use
3036 int btrfs_orphan_cleanup(struct btrfs_root *root)
3038 struct btrfs_path *path;
3039 struct extent_buffer *leaf;
3040 struct btrfs_key key, found_key;
3041 struct btrfs_trans_handle *trans;
3042 struct inode *inode;
3043 u64 last_objectid = 0;
3044 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3046 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3049 path = btrfs_alloc_path();
3056 key.objectid = BTRFS_ORPHAN_OBJECTID;
3057 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3058 key.offset = (u64)-1;
3061 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3066 * if ret == 0 means we found what we were searching for, which
3067 * is weird, but possible, so only screw with path if we didn't
3068 * find the key and see if we have stuff that matches
3072 if (path->slots[0] == 0)
3077 /* pull out the item */
3078 leaf = path->nodes[0];
3079 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3081 /* make sure the item matches what we want */
3082 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3084 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3087 /* release the path since we're done with it */
3088 btrfs_release_path(path);
3091 * this is where we are basically btrfs_lookup, without the
3092 * crossing root thing. we store the inode number in the
3093 * offset of the orphan item.
3096 if (found_key.offset == last_objectid) {
3097 btrfs_err(root->fs_info,
3098 "Error removing orphan entry, stopping orphan cleanup");
3103 last_objectid = found_key.offset;
3105 found_key.objectid = found_key.offset;
3106 found_key.type = BTRFS_INODE_ITEM_KEY;
3107 found_key.offset = 0;
3108 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3109 ret = PTR_ERR_OR_ZERO(inode);
3110 if (ret && ret != -ESTALE)
3113 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3114 struct btrfs_root *dead_root;
3115 struct btrfs_fs_info *fs_info = root->fs_info;
3116 int is_dead_root = 0;
3119 * this is an orphan in the tree root. Currently these
3120 * could come from 2 sources:
3121 * a) a snapshot deletion in progress
3122 * b) a free space cache inode
3123 * We need to distinguish those two, as the snapshot
3124 * orphan must not get deleted.
3125 * find_dead_roots already ran before us, so if this
3126 * is a snapshot deletion, we should find the root
3127 * in the dead_roots list
3129 spin_lock(&fs_info->trans_lock);
3130 list_for_each_entry(dead_root, &fs_info->dead_roots,
3132 if (dead_root->root_key.objectid ==
3133 found_key.objectid) {
3138 spin_unlock(&fs_info->trans_lock);
3140 /* prevent this orphan from being found again */
3141 key.offset = found_key.objectid - 1;
3146 * Inode is already gone but the orphan item is still there,
3147 * kill the orphan item.
3149 if (ret == -ESTALE) {
3150 trans = btrfs_start_transaction(root, 1);
3151 if (IS_ERR(trans)) {
3152 ret = PTR_ERR(trans);
3155 btrfs_debug(root->fs_info, "auto deleting %Lu",
3156 found_key.objectid);
3157 ret = btrfs_del_orphan_item(trans, root,
3158 found_key.objectid);
3159 btrfs_end_transaction(trans, root);
3166 * add this inode to the orphan list so btrfs_orphan_del does
3167 * the proper thing when we hit it
3169 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3170 &BTRFS_I(inode)->runtime_flags);
3171 atomic_inc(&root->orphan_inodes);
3173 /* if we have links, this was a truncate, lets do that */
3174 if (inode->i_nlink) {
3175 if (!S_ISREG(inode->i_mode)) {
3182 /* 1 for the orphan item deletion. */
3183 trans = btrfs_start_transaction(root, 1);
3184 if (IS_ERR(trans)) {
3186 ret = PTR_ERR(trans);
3189 ret = btrfs_orphan_add(trans, inode);
3190 btrfs_end_transaction(trans, root);
3196 ret = btrfs_truncate(inode);
3198 btrfs_orphan_del(NULL, inode);
3203 /* this will do delete_inode and everything for us */
3208 /* release the path since we're done with it */
3209 btrfs_release_path(path);
3211 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3213 if (root->orphan_block_rsv)
3214 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3217 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3218 trans = btrfs_join_transaction(root);
3220 btrfs_end_transaction(trans, root);
3224 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3226 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3230 btrfs_crit(root->fs_info,
3231 "could not do orphan cleanup %d", ret);
3232 btrfs_free_path(path);
3237 * very simple check to peek ahead in the leaf looking for xattrs. If we
3238 * don't find any xattrs, we know there can't be any acls.
3240 * slot is the slot the inode is in, objectid is the objectid of the inode
3242 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3243 int slot, u64 objectid)
3245 u32 nritems = btrfs_header_nritems(leaf);
3246 struct btrfs_key found_key;
3247 static u64 xattr_access = 0;
3248 static u64 xattr_default = 0;
3251 if (!xattr_access) {
3252 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3253 strlen(POSIX_ACL_XATTR_ACCESS));
3254 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3255 strlen(POSIX_ACL_XATTR_DEFAULT));
3259 while (slot < nritems) {
3260 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3262 /* we found a different objectid, there must not be acls */
3263 if (found_key.objectid != objectid)
3266 /* we found an xattr, assume we've got an acl */
3267 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3268 if (found_key.offset == xattr_access ||
3269 found_key.offset == xattr_default)
3274 * we found a key greater than an xattr key, there can't
3275 * be any acls later on
3277 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3284 * it goes inode, inode backrefs, xattrs, extents,
3285 * so if there are a ton of hard links to an inode there can
3286 * be a lot of backrefs. Don't waste time searching too hard,
3287 * this is just an optimization
3292 /* we hit the end of the leaf before we found an xattr or
3293 * something larger than an xattr. We have to assume the inode
3300 * read an inode from the btree into the in-memory inode
3302 static void btrfs_read_locked_inode(struct inode *inode)
3304 struct btrfs_path *path;
3305 struct extent_buffer *leaf;
3306 struct btrfs_inode_item *inode_item;
3307 struct btrfs_timespec *tspec;
3308 struct btrfs_root *root = BTRFS_I(inode)->root;
3309 struct btrfs_key location;
3313 bool filled = false;
3315 ret = btrfs_fill_inode(inode, &rdev);
3319 path = btrfs_alloc_path();
3323 path->leave_spinning = 1;
3324 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3326 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3330 leaf = path->nodes[0];
3335 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3336 struct btrfs_inode_item);
3337 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3338 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3339 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3340 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3341 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3343 tspec = btrfs_inode_atime(inode_item);
3344 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3345 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3347 tspec = btrfs_inode_mtime(inode_item);
3348 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3349 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3351 tspec = btrfs_inode_ctime(inode_item);
3352 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3353 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3355 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3356 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3357 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3360 * If we were modified in the current generation and evicted from memory
3361 * and then re-read we need to do a full sync since we don't have any
3362 * idea about which extents were modified before we were evicted from
3365 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3366 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3367 &BTRFS_I(inode)->runtime_flags);
3369 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3370 inode->i_generation = BTRFS_I(inode)->generation;
3372 rdev = btrfs_inode_rdev(leaf, inode_item);
3374 BTRFS_I(inode)->index_cnt = (u64)-1;
3375 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3378 * try to precache a NULL acl entry for files that don't have
3379 * any xattrs or acls
3381 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3384 cache_no_acl(inode);
3386 btrfs_free_path(path);
3388 switch (inode->i_mode & S_IFMT) {
3390 inode->i_mapping->a_ops = &btrfs_aops;
3391 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3392 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3393 inode->i_fop = &btrfs_file_operations;
3394 inode->i_op = &btrfs_file_inode_operations;
3397 inode->i_fop = &btrfs_dir_file_operations;
3398 if (root == root->fs_info->tree_root)
3399 inode->i_op = &btrfs_dir_ro_inode_operations;
3401 inode->i_op = &btrfs_dir_inode_operations;
3404 inode->i_op = &btrfs_symlink_inode_operations;
3405 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3406 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3409 inode->i_op = &btrfs_special_inode_operations;
3410 init_special_inode(inode, inode->i_mode, rdev);
3414 btrfs_update_iflags(inode);
3418 btrfs_free_path(path);
3419 make_bad_inode(inode);
3423 * given a leaf and an inode, copy the inode fields into the leaf
3425 static void fill_inode_item(struct btrfs_trans_handle *trans,
3426 struct extent_buffer *leaf,
3427 struct btrfs_inode_item *item,
3428 struct inode *inode)
3430 struct btrfs_map_token token;
3432 btrfs_init_map_token(&token);
3434 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3435 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3436 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3438 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3439 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3441 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3442 inode->i_atime.tv_sec, &token);
3443 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3444 inode->i_atime.tv_nsec, &token);
3446 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3447 inode->i_mtime.tv_sec, &token);
3448 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3449 inode->i_mtime.tv_nsec, &token);
3451 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3452 inode->i_ctime.tv_sec, &token);
3453 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3454 inode->i_ctime.tv_nsec, &token);
3456 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3458 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3460 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3461 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3462 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3463 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3464 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3468 * copy everything in the in-memory inode into the btree.
3470 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3471 struct btrfs_root *root, struct inode *inode)
3473 struct btrfs_inode_item *inode_item;
3474 struct btrfs_path *path;
3475 struct extent_buffer *leaf;
3478 path = btrfs_alloc_path();
3482 path->leave_spinning = 1;
3483 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3491 btrfs_unlock_up_safe(path, 1);
3492 leaf = path->nodes[0];
3493 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3494 struct btrfs_inode_item);
3496 fill_inode_item(trans, leaf, inode_item, inode);
3497 btrfs_mark_buffer_dirty(leaf);
3498 btrfs_set_inode_last_trans(trans, inode);
3501 btrfs_free_path(path);
3506 * copy everything in the in-memory inode into the btree.
3508 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3509 struct btrfs_root *root, struct inode *inode)
3514 * If the inode is a free space inode, we can deadlock during commit
3515 * if we put it into the delayed code.
3517 * The data relocation inode should also be directly updated
3520 if (!btrfs_is_free_space_inode(inode)
3521 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3522 btrfs_update_root_times(trans, root);
3524 ret = btrfs_delayed_update_inode(trans, root, inode);
3526 btrfs_set_inode_last_trans(trans, inode);
3530 return btrfs_update_inode_item(trans, root, inode);
3533 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3534 struct btrfs_root *root,
3535 struct inode *inode)
3539 ret = btrfs_update_inode(trans, root, inode);
3541 return btrfs_update_inode_item(trans, root, inode);
3546 * unlink helper that gets used here in inode.c and in the tree logging
3547 * recovery code. It remove a link in a directory with a given name, and
3548 * also drops the back refs in the inode to the directory
3550 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3551 struct btrfs_root *root,
3552 struct inode *dir, struct inode *inode,
3553 const char *name, int name_len)
3555 struct btrfs_path *path;
3557 struct extent_buffer *leaf;
3558 struct btrfs_dir_item *di;
3559 struct btrfs_key key;
3561 u64 ino = btrfs_ino(inode);
3562 u64 dir_ino = btrfs_ino(dir);
3564 path = btrfs_alloc_path();
3570 path->leave_spinning = 1;
3571 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3572 name, name_len, -1);
3581 leaf = path->nodes[0];
3582 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3583 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3586 btrfs_release_path(path);
3588 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3591 btrfs_info(root->fs_info,
3592 "failed to delete reference to %.*s, inode %llu parent %llu",
3593 name_len, name, ino, dir_ino);
3594 btrfs_abort_transaction(trans, root, ret);
3598 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3600 btrfs_abort_transaction(trans, root, ret);
3604 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3606 if (ret != 0 && ret != -ENOENT) {
3607 btrfs_abort_transaction(trans, root, ret);
3611 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3616 btrfs_abort_transaction(trans, root, ret);
3618 btrfs_free_path(path);
3622 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3623 inode_inc_iversion(inode);
3624 inode_inc_iversion(dir);
3625 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3626 ret = btrfs_update_inode(trans, root, dir);
3631 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3632 struct btrfs_root *root,
3633 struct inode *dir, struct inode *inode,
3634 const char *name, int name_len)
3637 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3639 btrfs_drop_nlink(inode);
3640 ret = btrfs_update_inode(trans, root, inode);
3646 * helper to start transaction for unlink and rmdir.
3648 * unlink and rmdir are special in btrfs, they do not always free space, so
3649 * if we cannot make our reservations the normal way try and see if there is
3650 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3651 * allow the unlink to occur.
3653 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3655 struct btrfs_trans_handle *trans;
3656 struct btrfs_root *root = BTRFS_I(dir)->root;
3660 * 1 for the possible orphan item
3661 * 1 for the dir item
3662 * 1 for the dir index
3663 * 1 for the inode ref
3666 trans = btrfs_start_transaction(root, 5);
3667 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3670 if (PTR_ERR(trans) == -ENOSPC) {
3671 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3673 trans = btrfs_start_transaction(root, 0);
3676 ret = btrfs_cond_migrate_bytes(root->fs_info,
3677 &root->fs_info->trans_block_rsv,
3680 btrfs_end_transaction(trans, root);
3681 return ERR_PTR(ret);
3683 trans->block_rsv = &root->fs_info->trans_block_rsv;
3684 trans->bytes_reserved = num_bytes;
3689 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3691 struct btrfs_root *root = BTRFS_I(dir)->root;
3692 struct btrfs_trans_handle *trans;
3693 struct inode *inode = dentry->d_inode;
3696 trans = __unlink_start_trans(dir);
3698 return PTR_ERR(trans);
3700 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3702 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3703 dentry->d_name.name, dentry->d_name.len);
3707 if (inode->i_nlink == 0) {
3708 ret = btrfs_orphan_add(trans, inode);
3714 btrfs_end_transaction(trans, root);
3715 btrfs_btree_balance_dirty(root);
3719 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3720 struct btrfs_root *root,
3721 struct inode *dir, u64 objectid,
3722 const char *name, int name_len)
3724 struct btrfs_path *path;
3725 struct extent_buffer *leaf;
3726 struct btrfs_dir_item *di;
3727 struct btrfs_key key;
3730 u64 dir_ino = btrfs_ino(dir);
3732 path = btrfs_alloc_path();
3736 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3737 name, name_len, -1);
3738 if (IS_ERR_OR_NULL(di)) {
3746 leaf = path->nodes[0];
3747 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3748 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3749 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3751 btrfs_abort_transaction(trans, root, ret);
3754 btrfs_release_path(path);
3756 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3757 objectid, root->root_key.objectid,
3758 dir_ino, &index, name, name_len);
3760 if (ret != -ENOENT) {
3761 btrfs_abort_transaction(trans, root, ret);
3764 di = btrfs_search_dir_index_item(root, path, dir_ino,
3766 if (IS_ERR_OR_NULL(di)) {
3771 btrfs_abort_transaction(trans, root, ret);
3775 leaf = path->nodes[0];
3776 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3777 btrfs_release_path(path);
3780 btrfs_release_path(path);
3782 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3784 btrfs_abort_transaction(trans, root, ret);
3788 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3789 inode_inc_iversion(dir);
3790 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3791 ret = btrfs_update_inode_fallback(trans, root, dir);
3793 btrfs_abort_transaction(trans, root, ret);
3795 btrfs_free_path(path);
3799 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3801 struct inode *inode = dentry->d_inode;
3803 struct btrfs_root *root = BTRFS_I(dir)->root;
3804 struct btrfs_trans_handle *trans;
3806 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3808 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3811 trans = __unlink_start_trans(dir);
3813 return PTR_ERR(trans);
3815 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3816 err = btrfs_unlink_subvol(trans, root, dir,
3817 BTRFS_I(inode)->location.objectid,
3818 dentry->d_name.name,
3819 dentry->d_name.len);
3823 err = btrfs_orphan_add(trans, inode);
3827 /* now the directory is empty */
3828 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3829 dentry->d_name.name, dentry->d_name.len);
3831 btrfs_i_size_write(inode, 0);
3833 btrfs_end_transaction(trans, root);
3834 btrfs_btree_balance_dirty(root);
3840 * this can truncate away extent items, csum items and directory items.
3841 * It starts at a high offset and removes keys until it can't find
3842 * any higher than new_size
3844 * csum items that cross the new i_size are truncated to the new size
3847 * min_type is the minimum key type to truncate down to. If set to 0, this
3848 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3850 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3851 struct btrfs_root *root,
3852 struct inode *inode,
3853 u64 new_size, u32 min_type)
3855 struct btrfs_path *path;
3856 struct extent_buffer *leaf;
3857 struct btrfs_file_extent_item *fi;
3858 struct btrfs_key key;
3859 struct btrfs_key found_key;
3860 u64 extent_start = 0;
3861 u64 extent_num_bytes = 0;
3862 u64 extent_offset = 0;
3864 u64 last_size = (u64)-1;
3865 u32 found_type = (u8)-1;
3868 int pending_del_nr = 0;
3869 int pending_del_slot = 0;
3870 int extent_type = -1;
3873 u64 ino = btrfs_ino(inode);
3875 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3877 path = btrfs_alloc_path();
3883 * We want to drop from the next block forward in case this new size is
3884 * not block aligned since we will be keeping the last block of the
3885 * extent just the way it is.
3887 if (root->ref_cows || root == root->fs_info->tree_root)
3888 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3889 root->sectorsize), (u64)-1, 0);
3892 * This function is also used to drop the items in the log tree before
3893 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3894 * it is used to drop the loged items. So we shouldn't kill the delayed
3897 if (min_type == 0 && root == BTRFS_I(inode)->root)
3898 btrfs_kill_delayed_inode_items(inode);
3901 key.offset = (u64)-1;
3905 path->leave_spinning = 1;
3906 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3913 /* there are no items in the tree for us to truncate, we're
3916 if (path->slots[0] == 0)
3923 leaf = path->nodes[0];
3924 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3925 found_type = btrfs_key_type(&found_key);
3927 if (found_key.objectid != ino)
3930 if (found_type < min_type)
3933 item_end = found_key.offset;
3934 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3935 fi = btrfs_item_ptr(leaf, path->slots[0],
3936 struct btrfs_file_extent_item);
3937 extent_type = btrfs_file_extent_type(leaf, fi);
3938 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3940 btrfs_file_extent_num_bytes(leaf, fi);
3941 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3942 item_end += btrfs_file_extent_inline_len(leaf,
3947 if (found_type > min_type) {
3950 if (item_end < new_size)
3952 if (found_key.offset >= new_size)
3958 /* FIXME, shrink the extent if the ref count is only 1 */
3959 if (found_type != BTRFS_EXTENT_DATA_KEY)
3963 last_size = found_key.offset;
3965 last_size = new_size;
3967 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3969 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3971 u64 orig_num_bytes =
3972 btrfs_file_extent_num_bytes(leaf, fi);
3973 extent_num_bytes = ALIGN(new_size -
3976 btrfs_set_file_extent_num_bytes(leaf, fi,
3978 num_dec = (orig_num_bytes -
3980 if (root->ref_cows && extent_start != 0)
3981 inode_sub_bytes(inode, num_dec);
3982 btrfs_mark_buffer_dirty(leaf);
3985 btrfs_file_extent_disk_num_bytes(leaf,
3987 extent_offset = found_key.offset -
3988 btrfs_file_extent_offset(leaf, fi);
3990 /* FIXME blocksize != 4096 */
3991 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3992 if (extent_start != 0) {
3995 inode_sub_bytes(inode, num_dec);
3998 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4000 * we can't truncate inline items that have had
4004 btrfs_file_extent_compression(leaf, fi) == 0 &&
4005 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4006 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4007 u32 size = new_size - found_key.offset;
4009 if (root->ref_cows) {
4010 inode_sub_bytes(inode, item_end + 1 -
4014 btrfs_file_extent_calc_inline_size(size);
4015 btrfs_truncate_item(root, path, size, 1);
4016 } else if (root->ref_cows) {
4017 inode_sub_bytes(inode, item_end + 1 -
4023 if (!pending_del_nr) {
4024 /* no pending yet, add ourselves */
4025 pending_del_slot = path->slots[0];
4027 } else if (pending_del_nr &&
4028 path->slots[0] + 1 == pending_del_slot) {
4029 /* hop on the pending chunk */
4031 pending_del_slot = path->slots[0];
4038 if (found_extent && (root->ref_cows ||
4039 root == root->fs_info->tree_root)) {
4040 btrfs_set_path_blocking(path);
4041 ret = btrfs_free_extent(trans, root, extent_start,
4042 extent_num_bytes, 0,
4043 btrfs_header_owner(leaf),
4044 ino, extent_offset, 0);
4048 if (found_type == BTRFS_INODE_ITEM_KEY)
4051 if (path->slots[0] == 0 ||
4052 path->slots[0] != pending_del_slot) {
4053 if (pending_del_nr) {
4054 ret = btrfs_del_items(trans, root, path,
4058 btrfs_abort_transaction(trans,
4064 btrfs_release_path(path);
4071 if (pending_del_nr) {
4072 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4075 btrfs_abort_transaction(trans, root, ret);
4078 if (last_size != (u64)-1)
4079 btrfs_ordered_update_i_size(inode, last_size, NULL);
4080 btrfs_free_path(path);
4085 * btrfs_truncate_page - read, zero a chunk and write a page
4086 * @inode - inode that we're zeroing
4087 * @from - the offset to start zeroing
4088 * @len - the length to zero, 0 to zero the entire range respective to the
4090 * @front - zero up to the offset instead of from the offset on
4092 * This will find the page for the "from" offset and cow the page and zero the
4093 * part we want to zero. This is used with truncate and hole punching.
4095 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4098 struct address_space *mapping = inode->i_mapping;
4099 struct btrfs_root *root = BTRFS_I(inode)->root;
4100 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4101 struct btrfs_ordered_extent *ordered;
4102 struct extent_state *cached_state = NULL;
4104 u32 blocksize = root->sectorsize;
4105 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4106 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4108 gfp_t mask = btrfs_alloc_write_mask(mapping);
4113 if ((offset & (blocksize - 1)) == 0 &&
4114 (!len || ((len & (blocksize - 1)) == 0)))
4116 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4121 page = find_or_create_page(mapping, index, mask);
4123 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4128 page_start = page_offset(page);
4129 page_end = page_start + PAGE_CACHE_SIZE - 1;
4131 if (!PageUptodate(page)) {
4132 ret = btrfs_readpage(NULL, page);
4134 if (page->mapping != mapping) {
4136 page_cache_release(page);
4139 if (!PageUptodate(page)) {
4144 wait_on_page_writeback(page);
4146 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4147 set_page_extent_mapped(page);
4149 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4151 unlock_extent_cached(io_tree, page_start, page_end,
4152 &cached_state, GFP_NOFS);
4154 page_cache_release(page);
4155 btrfs_start_ordered_extent(inode, ordered, 1);
4156 btrfs_put_ordered_extent(ordered);
4160 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4161 EXTENT_DIRTY | EXTENT_DELALLOC |
4162 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4163 0, 0, &cached_state, GFP_NOFS);
4165 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4168 unlock_extent_cached(io_tree, page_start, page_end,
4169 &cached_state, GFP_NOFS);
4173 if (offset != PAGE_CACHE_SIZE) {
4175 len = PAGE_CACHE_SIZE - offset;
4178 memset(kaddr, 0, offset);
4180 memset(kaddr + offset, 0, len);
4181 flush_dcache_page(page);
4184 ClearPageChecked(page);
4185 set_page_dirty(page);
4186 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4191 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4193 page_cache_release(page);
4199 * This function puts in dummy file extents for the area we're creating a hole
4200 * for. So if we are truncating this file to a larger size we need to insert
4201 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4202 * the range between oldsize and size
4204 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4206 struct btrfs_trans_handle *trans;
4207 struct btrfs_root *root = BTRFS_I(inode)->root;
4208 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4209 struct extent_map *em = NULL;
4210 struct extent_state *cached_state = NULL;
4211 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4212 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4213 u64 block_end = ALIGN(size, root->sectorsize);
4220 * If our size started in the middle of a page we need to zero out the
4221 * rest of the page before we expand the i_size, otherwise we could
4222 * expose stale data.
4224 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4228 if (size <= hole_start)
4232 struct btrfs_ordered_extent *ordered;
4233 btrfs_wait_ordered_range(inode, hole_start,
4234 block_end - hole_start);
4235 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4237 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4240 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4241 &cached_state, GFP_NOFS);
4242 btrfs_put_ordered_extent(ordered);
4245 cur_offset = hole_start;
4247 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4248 block_end - cur_offset, 0);
4254 last_byte = min(extent_map_end(em), block_end);
4255 last_byte = ALIGN(last_byte , root->sectorsize);
4256 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4257 struct extent_map *hole_em;
4258 hole_size = last_byte - cur_offset;
4260 trans = btrfs_start_transaction(root, 3);
4261 if (IS_ERR(trans)) {
4262 err = PTR_ERR(trans);
4266 err = btrfs_drop_extents(trans, root, inode,
4268 cur_offset + hole_size, 1);
4270 btrfs_abort_transaction(trans, root, err);
4271 btrfs_end_transaction(trans, root);
4275 err = btrfs_insert_file_extent(trans, root,
4276 btrfs_ino(inode), cur_offset, 0,
4277 0, hole_size, 0, hole_size,
4280 btrfs_abort_transaction(trans, root, err);
4281 btrfs_end_transaction(trans, root);
4285 btrfs_drop_extent_cache(inode, cur_offset,
4286 cur_offset + hole_size - 1, 0);
4287 hole_em = alloc_extent_map();
4289 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4290 &BTRFS_I(inode)->runtime_flags);
4293 hole_em->start = cur_offset;
4294 hole_em->len = hole_size;
4295 hole_em->orig_start = cur_offset;
4297 hole_em->block_start = EXTENT_MAP_HOLE;
4298 hole_em->block_len = 0;
4299 hole_em->orig_block_len = 0;
4300 hole_em->ram_bytes = hole_size;
4301 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4302 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4303 hole_em->generation = trans->transid;
4306 write_lock(&em_tree->lock);
4307 err = add_extent_mapping(em_tree, hole_em, 1);
4308 write_unlock(&em_tree->lock);
4311 btrfs_drop_extent_cache(inode, cur_offset,
4315 free_extent_map(hole_em);
4317 btrfs_update_inode(trans, root, inode);
4318 btrfs_end_transaction(trans, root);
4320 free_extent_map(em);
4322 cur_offset = last_byte;
4323 if (cur_offset >= block_end)
4327 free_extent_map(em);
4328 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4333 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4335 struct btrfs_root *root = BTRFS_I(inode)->root;
4336 struct btrfs_trans_handle *trans;
4337 loff_t oldsize = i_size_read(inode);
4338 loff_t newsize = attr->ia_size;
4339 int mask = attr->ia_valid;
4343 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4344 * special case where we need to update the times despite not having
4345 * these flags set. For all other operations the VFS set these flags
4346 * explicitly if it wants a timestamp update.
4348 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4349 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4351 if (newsize > oldsize) {
4352 truncate_pagecache(inode, newsize);
4353 ret = btrfs_cont_expand(inode, oldsize, newsize);
4357 trans = btrfs_start_transaction(root, 1);
4359 return PTR_ERR(trans);
4361 i_size_write(inode, newsize);
4362 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4363 ret = btrfs_update_inode(trans, root, inode);
4364 btrfs_end_transaction(trans, root);
4368 * We're truncating a file that used to have good data down to
4369 * zero. Make sure it gets into the ordered flush list so that
4370 * any new writes get down to disk quickly.
4373 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4374 &BTRFS_I(inode)->runtime_flags);
4377 * 1 for the orphan item we're going to add
4378 * 1 for the orphan item deletion.
4380 trans = btrfs_start_transaction(root, 2);
4382 return PTR_ERR(trans);
4385 * We need to do this in case we fail at _any_ point during the
4386 * actual truncate. Once we do the truncate_setsize we could
4387 * invalidate pages which forces any outstanding ordered io to
4388 * be instantly completed which will give us extents that need
4389 * to be truncated. If we fail to get an orphan inode down we
4390 * could have left over extents that were never meant to live,
4391 * so we need to garuntee from this point on that everything
4392 * will be consistent.
4394 ret = btrfs_orphan_add(trans, inode);
4395 btrfs_end_transaction(trans, root);
4399 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4400 truncate_setsize(inode, newsize);
4402 /* Disable nonlocked read DIO to avoid the end less truncate */
4403 btrfs_inode_block_unlocked_dio(inode);
4404 inode_dio_wait(inode);
4405 btrfs_inode_resume_unlocked_dio(inode);
4407 ret = btrfs_truncate(inode);
4408 if (ret && inode->i_nlink) {
4412 * failed to truncate, disk_i_size is only adjusted down
4413 * as we remove extents, so it should represent the true
4414 * size of the inode, so reset the in memory size and
4415 * delete our orphan entry.
4417 trans = btrfs_join_transaction(root);
4418 if (IS_ERR(trans)) {
4419 btrfs_orphan_del(NULL, inode);
4422 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4423 err = btrfs_orphan_del(trans, inode);
4425 btrfs_abort_transaction(trans, root, err);
4426 btrfs_end_transaction(trans, root);
4433 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4435 struct inode *inode = dentry->d_inode;
4436 struct btrfs_root *root = BTRFS_I(inode)->root;
4439 if (btrfs_root_readonly(root))
4442 err = inode_change_ok(inode, attr);
4446 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4447 err = btrfs_setsize(inode, attr);
4452 if (attr->ia_valid) {
4453 setattr_copy(inode, attr);
4454 inode_inc_iversion(inode);
4455 err = btrfs_dirty_inode(inode);
4457 if (!err && attr->ia_valid & ATTR_MODE)
4458 err = btrfs_acl_chmod(inode);
4464 void btrfs_evict_inode(struct inode *inode)
4466 struct btrfs_trans_handle *trans;
4467 struct btrfs_root *root = BTRFS_I(inode)->root;
4468 struct btrfs_block_rsv *rsv, *global_rsv;
4469 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4472 trace_btrfs_inode_evict(inode);
4474 truncate_inode_pages(&inode->i_data, 0);
4475 if (inode->i_nlink &&
4476 ((btrfs_root_refs(&root->root_item) != 0 &&
4477 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4478 btrfs_is_free_space_inode(inode)))
4481 if (is_bad_inode(inode)) {
4482 btrfs_orphan_del(NULL, inode);
4485 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4486 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4488 if (root->fs_info->log_root_recovering) {
4489 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4490 &BTRFS_I(inode)->runtime_flags));
4494 if (inode->i_nlink > 0) {
4495 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4496 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4500 ret = btrfs_commit_inode_delayed_inode(inode);
4502 btrfs_orphan_del(NULL, inode);
4506 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4508 btrfs_orphan_del(NULL, inode);
4511 rsv->size = min_size;
4513 global_rsv = &root->fs_info->global_block_rsv;
4515 btrfs_i_size_write(inode, 0);
4518 * This is a bit simpler than btrfs_truncate since we've already
4519 * reserved our space for our orphan item in the unlink, so we just
4520 * need to reserve some slack space in case we add bytes and update
4521 * inode item when doing the truncate.
4524 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4525 BTRFS_RESERVE_FLUSH_LIMIT);
4528 * Try and steal from the global reserve since we will
4529 * likely not use this space anyway, we want to try as
4530 * hard as possible to get this to work.
4533 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4536 btrfs_warn(root->fs_info,
4537 "Could not get space for a delete, will truncate on mount %d",
4539 btrfs_orphan_del(NULL, inode);
4540 btrfs_free_block_rsv(root, rsv);
4544 trans = btrfs_join_transaction(root);
4545 if (IS_ERR(trans)) {
4546 btrfs_orphan_del(NULL, inode);
4547 btrfs_free_block_rsv(root, rsv);
4551 trans->block_rsv = rsv;
4553 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4557 trans->block_rsv = &root->fs_info->trans_block_rsv;
4558 btrfs_end_transaction(trans, root);
4560 btrfs_btree_balance_dirty(root);
4563 btrfs_free_block_rsv(root, rsv);
4566 * Errors here aren't a big deal, it just means we leave orphan items
4567 * in the tree. They will be cleaned up on the next mount.
4570 trans->block_rsv = root->orphan_block_rsv;
4571 btrfs_orphan_del(trans, inode);
4573 btrfs_orphan_del(NULL, inode);
4576 trans->block_rsv = &root->fs_info->trans_block_rsv;
4577 if (!(root == root->fs_info->tree_root ||
4578 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4579 btrfs_return_ino(root, btrfs_ino(inode));
4581 btrfs_end_transaction(trans, root);
4582 btrfs_btree_balance_dirty(root);
4584 btrfs_remove_delayed_node(inode);
4590 * this returns the key found in the dir entry in the location pointer.
4591 * If no dir entries were found, location->objectid is 0.
4593 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4594 struct btrfs_key *location)
4596 const char *name = dentry->d_name.name;
4597 int namelen = dentry->d_name.len;
4598 struct btrfs_dir_item *di;
4599 struct btrfs_path *path;
4600 struct btrfs_root *root = BTRFS_I(dir)->root;
4603 path = btrfs_alloc_path();
4607 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4612 if (IS_ERR_OR_NULL(di))
4615 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4617 btrfs_free_path(path);
4620 location->objectid = 0;
4625 * when we hit a tree root in a directory, the btrfs part of the inode
4626 * needs to be changed to reflect the root directory of the tree root. This
4627 * is kind of like crossing a mount point.
4629 static int fixup_tree_root_location(struct btrfs_root *root,
4631 struct dentry *dentry,
4632 struct btrfs_key *location,
4633 struct btrfs_root **sub_root)
4635 struct btrfs_path *path;
4636 struct btrfs_root *new_root;
4637 struct btrfs_root_ref *ref;
4638 struct extent_buffer *leaf;
4642 path = btrfs_alloc_path();
4649 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4650 BTRFS_I(dir)->root->root_key.objectid,
4651 location->objectid);
4658 leaf = path->nodes[0];
4659 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4660 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4661 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4664 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4665 (unsigned long)(ref + 1),
4666 dentry->d_name.len);
4670 btrfs_release_path(path);
4672 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4673 if (IS_ERR(new_root)) {
4674 err = PTR_ERR(new_root);
4678 *sub_root = new_root;
4679 location->objectid = btrfs_root_dirid(&new_root->root_item);
4680 location->type = BTRFS_INODE_ITEM_KEY;
4681 location->offset = 0;
4684 btrfs_free_path(path);
4688 static void inode_tree_add(struct inode *inode)
4690 struct btrfs_root *root = BTRFS_I(inode)->root;
4691 struct btrfs_inode *entry;
4693 struct rb_node *parent;
4694 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4695 u64 ino = btrfs_ino(inode);
4697 if (inode_unhashed(inode))
4700 spin_lock(&root->inode_lock);
4701 p = &root->inode_tree.rb_node;
4704 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4706 if (ino < btrfs_ino(&entry->vfs_inode))
4707 p = &parent->rb_left;
4708 else if (ino > btrfs_ino(&entry->vfs_inode))
4709 p = &parent->rb_right;
4711 WARN_ON(!(entry->vfs_inode.i_state &
4712 (I_WILL_FREE | I_FREEING)));
4713 rb_replace_node(parent, new, &root->inode_tree);
4714 RB_CLEAR_NODE(parent);
4715 spin_unlock(&root->inode_lock);
4719 rb_link_node(new, parent, p);
4720 rb_insert_color(new, &root->inode_tree);
4721 spin_unlock(&root->inode_lock);
4724 static void inode_tree_del(struct inode *inode)
4726 struct btrfs_root *root = BTRFS_I(inode)->root;
4729 spin_lock(&root->inode_lock);
4730 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4731 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4732 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4733 empty = RB_EMPTY_ROOT(&root->inode_tree);
4735 spin_unlock(&root->inode_lock);
4737 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4738 synchronize_srcu(&root->fs_info->subvol_srcu);
4739 spin_lock(&root->inode_lock);
4740 empty = RB_EMPTY_ROOT(&root->inode_tree);
4741 spin_unlock(&root->inode_lock);
4743 btrfs_add_dead_root(root);
4747 void btrfs_invalidate_inodes(struct btrfs_root *root)
4749 struct rb_node *node;
4750 struct rb_node *prev;
4751 struct btrfs_inode *entry;
4752 struct inode *inode;
4755 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4757 spin_lock(&root->inode_lock);
4759 node = root->inode_tree.rb_node;
4763 entry = rb_entry(node, struct btrfs_inode, rb_node);
4765 if (objectid < btrfs_ino(&entry->vfs_inode))
4766 node = node->rb_left;
4767 else if (objectid > btrfs_ino(&entry->vfs_inode))
4768 node = node->rb_right;
4774 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4775 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4779 prev = rb_next(prev);
4783 entry = rb_entry(node, struct btrfs_inode, rb_node);
4784 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4785 inode = igrab(&entry->vfs_inode);
4787 spin_unlock(&root->inode_lock);
4788 if (atomic_read(&inode->i_count) > 1)
4789 d_prune_aliases(inode);
4791 * btrfs_drop_inode will have it removed from
4792 * the inode cache when its usage count
4797 spin_lock(&root->inode_lock);
4801 if (cond_resched_lock(&root->inode_lock))
4804 node = rb_next(node);
4806 spin_unlock(&root->inode_lock);
4809 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4811 struct btrfs_iget_args *args = p;
4812 inode->i_ino = args->ino;
4813 BTRFS_I(inode)->root = args->root;
4817 static int btrfs_find_actor(struct inode *inode, void *opaque)
4819 struct btrfs_iget_args *args = opaque;
4820 return args->ino == btrfs_ino(inode) &&
4821 args->root == BTRFS_I(inode)->root;
4824 static struct inode *btrfs_iget_locked(struct super_block *s,
4826 struct btrfs_root *root)
4828 struct inode *inode;
4829 struct btrfs_iget_args args;
4830 args.ino = objectid;
4833 inode = iget5_locked(s, objectid, btrfs_find_actor,
4834 btrfs_init_locked_inode,
4839 /* Get an inode object given its location and corresponding root.
4840 * Returns in *is_new if the inode was read from disk
4842 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4843 struct btrfs_root *root, int *new)
4845 struct inode *inode;
4847 inode = btrfs_iget_locked(s, location->objectid, root);
4849 return ERR_PTR(-ENOMEM);
4851 if (inode->i_state & I_NEW) {
4852 BTRFS_I(inode)->root = root;
4853 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4854 btrfs_read_locked_inode(inode);
4855 if (!is_bad_inode(inode)) {
4856 inode_tree_add(inode);
4857 unlock_new_inode(inode);
4861 unlock_new_inode(inode);
4863 inode = ERR_PTR(-ESTALE);
4870 static struct inode *new_simple_dir(struct super_block *s,
4871 struct btrfs_key *key,
4872 struct btrfs_root *root)
4874 struct inode *inode = new_inode(s);
4877 return ERR_PTR(-ENOMEM);
4879 BTRFS_I(inode)->root = root;
4880 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4881 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4883 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4884 inode->i_op = &btrfs_dir_ro_inode_operations;
4885 inode->i_fop = &simple_dir_operations;
4886 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4887 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4892 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4894 struct inode *inode;
4895 struct btrfs_root *root = BTRFS_I(dir)->root;
4896 struct btrfs_root *sub_root = root;
4897 struct btrfs_key location;
4901 if (dentry->d_name.len > BTRFS_NAME_LEN)
4902 return ERR_PTR(-ENAMETOOLONG);
4904 ret = btrfs_inode_by_name(dir, dentry, &location);
4906 return ERR_PTR(ret);
4908 if (location.objectid == 0)
4911 if (location.type == BTRFS_INODE_ITEM_KEY) {
4912 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4916 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4918 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4919 ret = fixup_tree_root_location(root, dir, dentry,
4920 &location, &sub_root);
4923 inode = ERR_PTR(ret);
4925 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4927 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4929 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4931 if (!IS_ERR(inode) && root != sub_root) {
4932 down_read(&root->fs_info->cleanup_work_sem);
4933 if (!(inode->i_sb->s_flags & MS_RDONLY))
4934 ret = btrfs_orphan_cleanup(sub_root);
4935 up_read(&root->fs_info->cleanup_work_sem);
4938 inode = ERR_PTR(ret);
4945 static int btrfs_dentry_delete(const struct dentry *dentry)
4947 struct btrfs_root *root;
4948 struct inode *inode = dentry->d_inode;
4950 if (!inode && !IS_ROOT(dentry))
4951 inode = dentry->d_parent->d_inode;
4954 root = BTRFS_I(inode)->root;
4955 if (btrfs_root_refs(&root->root_item) == 0)
4958 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4964 static void btrfs_dentry_release(struct dentry *dentry)
4966 if (dentry->d_fsdata)
4967 kfree(dentry->d_fsdata);
4970 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4975 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4979 unsigned char btrfs_filetype_table[] = {
4980 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4983 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
4985 struct inode *inode = file_inode(file);
4986 struct btrfs_root *root = BTRFS_I(inode)->root;
4987 struct btrfs_item *item;
4988 struct btrfs_dir_item *di;
4989 struct btrfs_key key;
4990 struct btrfs_key found_key;
4991 struct btrfs_path *path;
4992 struct list_head ins_list;
4993 struct list_head del_list;
4995 struct extent_buffer *leaf;
4997 unsigned char d_type;
5002 int key_type = BTRFS_DIR_INDEX_KEY;
5006 int is_curr = 0; /* ctx->pos points to the current index? */
5008 /* FIXME, use a real flag for deciding about the key type */
5009 if (root->fs_info->tree_root == root)
5010 key_type = BTRFS_DIR_ITEM_KEY;
5012 if (!dir_emit_dots(file, ctx))
5015 path = btrfs_alloc_path();
5021 if (key_type == BTRFS_DIR_INDEX_KEY) {
5022 INIT_LIST_HEAD(&ins_list);
5023 INIT_LIST_HEAD(&del_list);
5024 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5027 btrfs_set_key_type(&key, key_type);
5028 key.offset = ctx->pos;
5029 key.objectid = btrfs_ino(inode);
5031 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5036 leaf = path->nodes[0];
5037 slot = path->slots[0];
5038 if (slot >= btrfs_header_nritems(leaf)) {
5039 ret = btrfs_next_leaf(root, path);
5047 item = btrfs_item_nr(slot);
5048 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5050 if (found_key.objectid != key.objectid)
5052 if (btrfs_key_type(&found_key) != key_type)
5054 if (found_key.offset < ctx->pos)
5056 if (key_type == BTRFS_DIR_INDEX_KEY &&
5057 btrfs_should_delete_dir_index(&del_list,
5061 ctx->pos = found_key.offset;
5064 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5066 di_total = btrfs_item_size(leaf, item);
5068 while (di_cur < di_total) {
5069 struct btrfs_key location;
5071 if (verify_dir_item(root, leaf, di))
5074 name_len = btrfs_dir_name_len(leaf, di);
5075 if (name_len <= sizeof(tmp_name)) {
5076 name_ptr = tmp_name;
5078 name_ptr = kmalloc(name_len, GFP_NOFS);
5084 read_extent_buffer(leaf, name_ptr,
5085 (unsigned long)(di + 1), name_len);
5087 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5088 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5091 /* is this a reference to our own snapshot? If so
5094 * In contrast to old kernels, we insert the snapshot's
5095 * dir item and dir index after it has been created, so
5096 * we won't find a reference to our own snapshot. We
5097 * still keep the following code for backward
5100 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5101 location.objectid == root->root_key.objectid) {
5105 over = !dir_emit(ctx, name_ptr, name_len,
5106 location.objectid, d_type);
5109 if (name_ptr != tmp_name)
5114 di_len = btrfs_dir_name_len(leaf, di) +
5115 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5117 di = (struct btrfs_dir_item *)((char *)di + di_len);
5123 if (key_type == BTRFS_DIR_INDEX_KEY) {
5126 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5131 /* Reached end of directory/root. Bump pos past the last item. */
5135 * Stop new entries from being returned after we return the last
5138 * New directory entries are assigned a strictly increasing
5139 * offset. This means that new entries created during readdir
5140 * are *guaranteed* to be seen in the future by that readdir.
5141 * This has broken buggy programs which operate on names as
5142 * they're returned by readdir. Until we re-use freed offsets
5143 * we have this hack to stop new entries from being returned
5144 * under the assumption that they'll never reach this huge
5147 * This is being careful not to overflow 32bit loff_t unless the
5148 * last entry requires it because doing so has broken 32bit apps
5151 if (key_type == BTRFS_DIR_INDEX_KEY) {
5152 if (ctx->pos >= INT_MAX)
5153 ctx->pos = LLONG_MAX;
5160 if (key_type == BTRFS_DIR_INDEX_KEY)
5161 btrfs_put_delayed_items(&ins_list, &del_list);
5162 btrfs_free_path(path);
5166 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5168 struct btrfs_root *root = BTRFS_I(inode)->root;
5169 struct btrfs_trans_handle *trans;
5171 bool nolock = false;
5173 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5176 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5179 if (wbc->sync_mode == WB_SYNC_ALL) {
5181 trans = btrfs_join_transaction_nolock(root);
5183 trans = btrfs_join_transaction(root);
5185 return PTR_ERR(trans);
5186 ret = btrfs_commit_transaction(trans, root);
5192 * This is somewhat expensive, updating the tree every time the
5193 * inode changes. But, it is most likely to find the inode in cache.
5194 * FIXME, needs more benchmarking...there are no reasons other than performance
5195 * to keep or drop this code.
5197 static int btrfs_dirty_inode(struct inode *inode)
5199 struct btrfs_root *root = BTRFS_I(inode)->root;
5200 struct btrfs_trans_handle *trans;
5203 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5206 trans = btrfs_join_transaction(root);
5208 return PTR_ERR(trans);
5210 ret = btrfs_update_inode(trans, root, inode);
5211 if (ret && ret == -ENOSPC) {
5212 /* whoops, lets try again with the full transaction */
5213 btrfs_end_transaction(trans, root);
5214 trans = btrfs_start_transaction(root, 1);
5216 return PTR_ERR(trans);
5218 ret = btrfs_update_inode(trans, root, inode);
5220 btrfs_end_transaction(trans, root);
5221 if (BTRFS_I(inode)->delayed_node)
5222 btrfs_balance_delayed_items(root);
5228 * This is a copy of file_update_time. We need this so we can return error on
5229 * ENOSPC for updating the inode in the case of file write and mmap writes.
5231 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5234 struct btrfs_root *root = BTRFS_I(inode)->root;
5236 if (btrfs_root_readonly(root))
5239 if (flags & S_VERSION)
5240 inode_inc_iversion(inode);
5241 if (flags & S_CTIME)
5242 inode->i_ctime = *now;
5243 if (flags & S_MTIME)
5244 inode->i_mtime = *now;
5245 if (flags & S_ATIME)
5246 inode->i_atime = *now;
5247 return btrfs_dirty_inode(inode);
5251 * find the highest existing sequence number in a directory
5252 * and then set the in-memory index_cnt variable to reflect
5253 * free sequence numbers
5255 static int btrfs_set_inode_index_count(struct inode *inode)
5257 struct btrfs_root *root = BTRFS_I(inode)->root;
5258 struct btrfs_key key, found_key;
5259 struct btrfs_path *path;
5260 struct extent_buffer *leaf;
5263 key.objectid = btrfs_ino(inode);
5264 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5265 key.offset = (u64)-1;
5267 path = btrfs_alloc_path();
5271 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5274 /* FIXME: we should be able to handle this */
5280 * MAGIC NUMBER EXPLANATION:
5281 * since we search a directory based on f_pos we have to start at 2
5282 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5283 * else has to start at 2
5285 if (path->slots[0] == 0) {
5286 BTRFS_I(inode)->index_cnt = 2;
5292 leaf = path->nodes[0];
5293 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5295 if (found_key.objectid != btrfs_ino(inode) ||
5296 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5297 BTRFS_I(inode)->index_cnt = 2;
5301 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5303 btrfs_free_path(path);
5308 * helper to find a free sequence number in a given directory. This current
5309 * code is very simple, later versions will do smarter things in the btree
5311 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5315 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5316 ret = btrfs_inode_delayed_dir_index_count(dir);
5318 ret = btrfs_set_inode_index_count(dir);
5324 *index = BTRFS_I(dir)->index_cnt;
5325 BTRFS_I(dir)->index_cnt++;
5330 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5331 struct btrfs_root *root,
5333 const char *name, int name_len,
5334 u64 ref_objectid, u64 objectid,
5335 umode_t mode, u64 *index)
5337 struct inode *inode;
5338 struct btrfs_inode_item *inode_item;
5339 struct btrfs_key *location;
5340 struct btrfs_path *path;
5341 struct btrfs_inode_ref *ref;
5342 struct btrfs_key key[2];
5348 path = btrfs_alloc_path();
5350 return ERR_PTR(-ENOMEM);
5352 inode = new_inode(root->fs_info->sb);
5354 btrfs_free_path(path);
5355 return ERR_PTR(-ENOMEM);
5359 * we have to initialize this early, so we can reclaim the inode
5360 * number if we fail afterwards in this function.
5362 inode->i_ino = objectid;
5365 trace_btrfs_inode_request(dir);
5367 ret = btrfs_set_inode_index(dir, index);
5369 btrfs_free_path(path);
5371 return ERR_PTR(ret);
5375 * index_cnt is ignored for everything but a dir,
5376 * btrfs_get_inode_index_count has an explanation for the magic
5379 BTRFS_I(inode)->index_cnt = 2;
5380 BTRFS_I(inode)->root = root;
5381 BTRFS_I(inode)->generation = trans->transid;
5382 inode->i_generation = BTRFS_I(inode)->generation;
5385 * We could have gotten an inode number from somebody who was fsynced
5386 * and then removed in this same transaction, so let's just set full
5387 * sync since it will be a full sync anyway and this will blow away the
5388 * old info in the log.
5390 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5397 key[0].objectid = objectid;
5398 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5402 * Start new inodes with an inode_ref. This is slightly more
5403 * efficient for small numbers of hard links since they will
5404 * be packed into one item. Extended refs will kick in if we
5405 * add more hard links than can fit in the ref item.
5407 key[1].objectid = objectid;
5408 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5409 key[1].offset = ref_objectid;
5411 sizes[0] = sizeof(struct btrfs_inode_item);
5412 sizes[1] = name_len + sizeof(*ref);
5414 path->leave_spinning = 1;
5415 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5419 inode_init_owner(inode, dir, mode);
5420 inode_set_bytes(inode, 0);
5421 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5422 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5423 struct btrfs_inode_item);
5424 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5425 sizeof(*inode_item));
5426 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5428 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5429 struct btrfs_inode_ref);
5430 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5431 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5432 ptr = (unsigned long)(ref + 1);
5433 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5435 btrfs_mark_buffer_dirty(path->nodes[0]);
5436 btrfs_free_path(path);
5438 location = &BTRFS_I(inode)->location;
5439 location->objectid = objectid;
5440 location->offset = 0;
5441 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5443 btrfs_inherit_iflags(inode, dir);
5445 if (S_ISREG(mode)) {
5446 if (btrfs_test_opt(root, NODATASUM))
5447 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5448 if (btrfs_test_opt(root, NODATACOW))
5449 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5450 BTRFS_INODE_NODATASUM;
5453 insert_inode_hash(inode);
5454 inode_tree_add(inode);
5456 trace_btrfs_inode_new(inode);
5457 btrfs_set_inode_last_trans(trans, inode);
5459 btrfs_update_root_times(trans, root);
5464 BTRFS_I(dir)->index_cnt--;
5465 btrfs_free_path(path);
5467 return ERR_PTR(ret);
5470 static inline u8 btrfs_inode_type(struct inode *inode)
5472 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5476 * utility function to add 'inode' into 'parent_inode' with
5477 * a give name and a given sequence number.
5478 * if 'add_backref' is true, also insert a backref from the
5479 * inode to the parent directory.
5481 int btrfs_add_link(struct btrfs_trans_handle *trans,
5482 struct inode *parent_inode, struct inode *inode,
5483 const char *name, int name_len, int add_backref, u64 index)
5486 struct btrfs_key key;
5487 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5488 u64 ino = btrfs_ino(inode);
5489 u64 parent_ino = btrfs_ino(parent_inode);
5491 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5492 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5495 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5499 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5500 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5501 key.objectid, root->root_key.objectid,
5502 parent_ino, index, name, name_len);
5503 } else if (add_backref) {
5504 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5508 /* Nothing to clean up yet */
5512 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5514 btrfs_inode_type(inode), index);
5515 if (ret == -EEXIST || ret == -EOVERFLOW)
5518 btrfs_abort_transaction(trans, root, ret);
5522 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5524 inode_inc_iversion(parent_inode);
5525 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5526 ret = btrfs_update_inode(trans, root, parent_inode);
5528 btrfs_abort_transaction(trans, root, ret);
5532 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5535 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5536 key.objectid, root->root_key.objectid,
5537 parent_ino, &local_index, name, name_len);
5539 } else if (add_backref) {
5543 err = btrfs_del_inode_ref(trans, root, name, name_len,
5544 ino, parent_ino, &local_index);
5549 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5550 struct inode *dir, struct dentry *dentry,
5551 struct inode *inode, int backref, u64 index)
5553 int err = btrfs_add_link(trans, dir, inode,
5554 dentry->d_name.name, dentry->d_name.len,
5561 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5562 umode_t mode, dev_t rdev)
5564 struct btrfs_trans_handle *trans;
5565 struct btrfs_root *root = BTRFS_I(dir)->root;
5566 struct inode *inode = NULL;
5572 if (!new_valid_dev(rdev))
5576 * 2 for inode item and ref
5578 * 1 for xattr if selinux is on
5580 trans = btrfs_start_transaction(root, 5);
5582 return PTR_ERR(trans);
5584 err = btrfs_find_free_ino(root, &objectid);
5588 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5589 dentry->d_name.len, btrfs_ino(dir), objectid,
5591 if (IS_ERR(inode)) {
5592 err = PTR_ERR(inode);
5596 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5603 * If the active LSM wants to access the inode during
5604 * d_instantiate it needs these. Smack checks to see
5605 * if the filesystem supports xattrs by looking at the
5609 inode->i_op = &btrfs_special_inode_operations;
5610 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5614 init_special_inode(inode, inode->i_mode, rdev);
5615 btrfs_update_inode(trans, root, inode);
5616 d_instantiate(dentry, inode);
5619 btrfs_end_transaction(trans, root);
5620 btrfs_btree_balance_dirty(root);
5622 inode_dec_link_count(inode);
5628 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5629 umode_t mode, bool excl)
5631 struct btrfs_trans_handle *trans;
5632 struct btrfs_root *root = BTRFS_I(dir)->root;
5633 struct inode *inode = NULL;
5634 int drop_inode_on_err = 0;
5640 * 2 for inode item and ref
5642 * 1 for xattr if selinux is on
5644 trans = btrfs_start_transaction(root, 5);
5646 return PTR_ERR(trans);
5648 err = btrfs_find_free_ino(root, &objectid);
5652 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5653 dentry->d_name.len, btrfs_ino(dir), objectid,
5655 if (IS_ERR(inode)) {
5656 err = PTR_ERR(inode);
5659 drop_inode_on_err = 1;
5661 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5665 err = btrfs_update_inode(trans, root, inode);
5670 * If the active LSM wants to access the inode during
5671 * d_instantiate it needs these. Smack checks to see
5672 * if the filesystem supports xattrs by looking at the
5675 inode->i_fop = &btrfs_file_operations;
5676 inode->i_op = &btrfs_file_inode_operations;
5678 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5682 inode->i_mapping->a_ops = &btrfs_aops;
5683 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5684 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5685 d_instantiate(dentry, inode);
5688 btrfs_end_transaction(trans, root);
5689 if (err && drop_inode_on_err) {
5690 inode_dec_link_count(inode);
5693 btrfs_btree_balance_dirty(root);
5697 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5698 struct dentry *dentry)
5700 struct btrfs_trans_handle *trans;
5701 struct btrfs_root *root = BTRFS_I(dir)->root;
5702 struct inode *inode = old_dentry->d_inode;
5707 /* do not allow sys_link's with other subvols of the same device */
5708 if (root->objectid != BTRFS_I(inode)->root->objectid)
5711 if (inode->i_nlink >= BTRFS_LINK_MAX)
5714 err = btrfs_set_inode_index(dir, &index);
5719 * 2 items for inode and inode ref
5720 * 2 items for dir items
5721 * 1 item for parent inode
5723 trans = btrfs_start_transaction(root, 5);
5724 if (IS_ERR(trans)) {
5725 err = PTR_ERR(trans);
5729 btrfs_inc_nlink(inode);
5730 inode_inc_iversion(inode);
5731 inode->i_ctime = CURRENT_TIME;
5733 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5735 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5740 struct dentry *parent = dentry->d_parent;
5741 err = btrfs_update_inode(trans, root, inode);
5744 d_instantiate(dentry, inode);
5745 btrfs_log_new_name(trans, inode, NULL, parent);
5748 btrfs_end_transaction(trans, root);
5751 inode_dec_link_count(inode);
5754 btrfs_btree_balance_dirty(root);
5758 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5760 struct inode *inode = NULL;
5761 struct btrfs_trans_handle *trans;
5762 struct btrfs_root *root = BTRFS_I(dir)->root;
5764 int drop_on_err = 0;
5769 * 2 items for inode and ref
5770 * 2 items for dir items
5771 * 1 for xattr if selinux is on
5773 trans = btrfs_start_transaction(root, 5);
5775 return PTR_ERR(trans);
5777 err = btrfs_find_free_ino(root, &objectid);
5781 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5782 dentry->d_name.len, btrfs_ino(dir), objectid,
5783 S_IFDIR | mode, &index);
5784 if (IS_ERR(inode)) {
5785 err = PTR_ERR(inode);
5791 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5795 inode->i_op = &btrfs_dir_inode_operations;
5796 inode->i_fop = &btrfs_dir_file_operations;
5798 btrfs_i_size_write(inode, 0);
5799 err = btrfs_update_inode(trans, root, inode);
5803 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5804 dentry->d_name.len, 0, index);
5808 d_instantiate(dentry, inode);
5812 btrfs_end_transaction(trans, root);
5815 btrfs_btree_balance_dirty(root);
5819 /* helper for btfs_get_extent. Given an existing extent in the tree,
5820 * and an extent that you want to insert, deal with overlap and insert
5821 * the new extent into the tree.
5823 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5824 struct extent_map *existing,
5825 struct extent_map *em,
5826 u64 map_start, u64 map_len)
5830 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5831 start_diff = map_start - em->start;
5832 em->start = map_start;
5834 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5835 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5836 em->block_start += start_diff;
5837 em->block_len -= start_diff;
5839 return add_extent_mapping(em_tree, em, 0);
5842 static noinline int uncompress_inline(struct btrfs_path *path,
5843 struct inode *inode, struct page *page,
5844 size_t pg_offset, u64 extent_offset,
5845 struct btrfs_file_extent_item *item)
5848 struct extent_buffer *leaf = path->nodes[0];
5851 unsigned long inline_size;
5855 WARN_ON(pg_offset != 0);
5856 compress_type = btrfs_file_extent_compression(leaf, item);
5857 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5858 inline_size = btrfs_file_extent_inline_item_len(leaf,
5859 btrfs_item_nr(path->slots[0]));
5860 tmp = kmalloc(inline_size, GFP_NOFS);
5863 ptr = btrfs_file_extent_inline_start(item);
5865 read_extent_buffer(leaf, tmp, ptr, inline_size);
5867 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5868 ret = btrfs_decompress(compress_type, tmp, page,
5869 extent_offset, inline_size, max_size);
5871 char *kaddr = kmap_atomic(page);
5872 unsigned long copy_size = min_t(u64,
5873 PAGE_CACHE_SIZE - pg_offset,
5874 max_size - extent_offset);
5875 memset(kaddr + pg_offset, 0, copy_size);
5876 kunmap_atomic(kaddr);
5883 * a bit scary, this does extent mapping from logical file offset to the disk.
5884 * the ugly parts come from merging extents from the disk with the in-ram
5885 * representation. This gets more complex because of the data=ordered code,
5886 * where the in-ram extents might be locked pending data=ordered completion.
5888 * This also copies inline extents directly into the page.
5891 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5892 size_t pg_offset, u64 start, u64 len,
5898 u64 extent_start = 0;
5900 u64 objectid = btrfs_ino(inode);
5902 struct btrfs_path *path = NULL;
5903 struct btrfs_root *root = BTRFS_I(inode)->root;
5904 struct btrfs_file_extent_item *item;
5905 struct extent_buffer *leaf;
5906 struct btrfs_key found_key;
5907 struct extent_map *em = NULL;
5908 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5909 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5910 struct btrfs_trans_handle *trans = NULL;
5914 read_lock(&em_tree->lock);
5915 em = lookup_extent_mapping(em_tree, start, len);
5917 em->bdev = root->fs_info->fs_devices->latest_bdev;
5918 read_unlock(&em_tree->lock);
5921 if (em->start > start || em->start + em->len <= start)
5922 free_extent_map(em);
5923 else if (em->block_start == EXTENT_MAP_INLINE && page)
5924 free_extent_map(em);
5928 em = alloc_extent_map();
5933 em->bdev = root->fs_info->fs_devices->latest_bdev;
5934 em->start = EXTENT_MAP_HOLE;
5935 em->orig_start = EXTENT_MAP_HOLE;
5937 em->block_len = (u64)-1;
5940 path = btrfs_alloc_path();
5946 * Chances are we'll be called again, so go ahead and do
5952 ret = btrfs_lookup_file_extent(trans, root, path,
5953 objectid, start, trans != NULL);
5960 if (path->slots[0] == 0)
5965 leaf = path->nodes[0];
5966 item = btrfs_item_ptr(leaf, path->slots[0],
5967 struct btrfs_file_extent_item);
5968 /* are we inside the extent that was found? */
5969 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5970 found_type = btrfs_key_type(&found_key);
5971 if (found_key.objectid != objectid ||
5972 found_type != BTRFS_EXTENT_DATA_KEY) {
5976 found_type = btrfs_file_extent_type(leaf, item);
5977 extent_start = found_key.offset;
5978 compress_type = btrfs_file_extent_compression(leaf, item);
5979 if (found_type == BTRFS_FILE_EXTENT_REG ||
5980 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5981 extent_end = extent_start +
5982 btrfs_file_extent_num_bytes(leaf, item);
5983 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5985 size = btrfs_file_extent_inline_len(leaf, item);
5986 extent_end = ALIGN(extent_start + size, root->sectorsize);
5989 if (start >= extent_end) {
5991 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5992 ret = btrfs_next_leaf(root, path);
5999 leaf = path->nodes[0];
6001 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6002 if (found_key.objectid != objectid ||
6003 found_key.type != BTRFS_EXTENT_DATA_KEY)
6005 if (start + len <= found_key.offset)
6008 em->orig_start = start;
6009 em->len = found_key.offset - start;
6013 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6014 if (found_type == BTRFS_FILE_EXTENT_REG ||
6015 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6016 em->start = extent_start;
6017 em->len = extent_end - extent_start;
6018 em->orig_start = extent_start -
6019 btrfs_file_extent_offset(leaf, item);
6020 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6022 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6024 em->block_start = EXTENT_MAP_HOLE;
6027 if (compress_type != BTRFS_COMPRESS_NONE) {
6028 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6029 em->compress_type = compress_type;
6030 em->block_start = bytenr;
6031 em->block_len = em->orig_block_len;
6033 bytenr += btrfs_file_extent_offset(leaf, item);
6034 em->block_start = bytenr;
6035 em->block_len = em->len;
6036 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6037 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6040 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6044 size_t extent_offset;
6047 em->block_start = EXTENT_MAP_INLINE;
6048 if (!page || create) {
6049 em->start = extent_start;
6050 em->len = extent_end - extent_start;
6054 size = btrfs_file_extent_inline_len(leaf, item);
6055 extent_offset = page_offset(page) + pg_offset - extent_start;
6056 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6057 size - extent_offset);
6058 em->start = extent_start + extent_offset;
6059 em->len = ALIGN(copy_size, root->sectorsize);
6060 em->orig_block_len = em->len;
6061 em->orig_start = em->start;
6062 if (compress_type) {
6063 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6064 em->compress_type = compress_type;
6066 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6067 if (create == 0 && !PageUptodate(page)) {
6068 if (btrfs_file_extent_compression(leaf, item) !=
6069 BTRFS_COMPRESS_NONE) {
6070 ret = uncompress_inline(path, inode, page,
6072 extent_offset, item);
6073 BUG_ON(ret); /* -ENOMEM */
6076 read_extent_buffer(leaf, map + pg_offset, ptr,
6078 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6079 memset(map + pg_offset + copy_size, 0,
6080 PAGE_CACHE_SIZE - pg_offset -
6085 flush_dcache_page(page);
6086 } else if (create && PageUptodate(page)) {
6090 free_extent_map(em);
6093 btrfs_release_path(path);
6094 trans = btrfs_join_transaction(root);
6097 return ERR_CAST(trans);
6101 write_extent_buffer(leaf, map + pg_offset, ptr,
6104 btrfs_mark_buffer_dirty(leaf);
6106 set_extent_uptodate(io_tree, em->start,
6107 extent_map_end(em) - 1, NULL, GFP_NOFS);
6110 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6114 em->orig_start = start;
6117 em->block_start = EXTENT_MAP_HOLE;
6118 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6120 btrfs_release_path(path);
6121 if (em->start > start || extent_map_end(em) <= start) {
6122 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6123 em->start, em->len, start, len);
6129 write_lock(&em_tree->lock);
6130 ret = add_extent_mapping(em_tree, em, 0);
6131 /* it is possible that someone inserted the extent into the tree
6132 * while we had the lock dropped. It is also possible that
6133 * an overlapping map exists in the tree
6135 if (ret == -EEXIST) {
6136 struct extent_map *existing;
6140 existing = lookup_extent_mapping(em_tree, start, len);
6141 if (existing && (existing->start > start ||
6142 existing->start + existing->len <= start)) {
6143 free_extent_map(existing);
6147 existing = lookup_extent_mapping(em_tree, em->start,
6150 err = merge_extent_mapping(em_tree, existing,
6153 free_extent_map(existing);
6155 free_extent_map(em);
6160 free_extent_map(em);
6164 free_extent_map(em);
6169 write_unlock(&em_tree->lock);
6173 trace_btrfs_get_extent(root, em);
6176 btrfs_free_path(path);
6178 ret = btrfs_end_transaction(trans, root);
6183 free_extent_map(em);
6184 return ERR_PTR(err);
6186 BUG_ON(!em); /* Error is always set */
6190 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6191 size_t pg_offset, u64 start, u64 len,
6194 struct extent_map *em;
6195 struct extent_map *hole_em = NULL;
6196 u64 range_start = start;
6202 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6209 * - a pre-alloc extent,
6210 * there might actually be delalloc bytes behind it.
6212 if (em->block_start != EXTENT_MAP_HOLE &&
6213 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6219 /* check to see if we've wrapped (len == -1 or similar) */
6228 /* ok, we didn't find anything, lets look for delalloc */
6229 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6230 end, len, EXTENT_DELALLOC, 1);
6231 found_end = range_start + found;
6232 if (found_end < range_start)
6233 found_end = (u64)-1;
6236 * we didn't find anything useful, return
6237 * the original results from get_extent()
6239 if (range_start > end || found_end <= start) {
6245 /* adjust the range_start to make sure it doesn't
6246 * go backwards from the start they passed in
6248 range_start = max(start,range_start);
6249 found = found_end - range_start;
6252 u64 hole_start = start;
6255 em = alloc_extent_map();
6261 * when btrfs_get_extent can't find anything it
6262 * returns one huge hole
6264 * make sure what it found really fits our range, and
6265 * adjust to make sure it is based on the start from
6269 u64 calc_end = extent_map_end(hole_em);
6271 if (calc_end <= start || (hole_em->start > end)) {
6272 free_extent_map(hole_em);
6275 hole_start = max(hole_em->start, start);
6276 hole_len = calc_end - hole_start;
6280 if (hole_em && range_start > hole_start) {
6281 /* our hole starts before our delalloc, so we
6282 * have to return just the parts of the hole
6283 * that go until the delalloc starts
6285 em->len = min(hole_len,
6286 range_start - hole_start);
6287 em->start = hole_start;
6288 em->orig_start = hole_start;
6290 * don't adjust block start at all,
6291 * it is fixed at EXTENT_MAP_HOLE
6293 em->block_start = hole_em->block_start;
6294 em->block_len = hole_len;
6295 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6296 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6298 em->start = range_start;
6300 em->orig_start = range_start;
6301 em->block_start = EXTENT_MAP_DELALLOC;
6302 em->block_len = found;
6304 } else if (hole_em) {
6309 free_extent_map(hole_em);
6311 free_extent_map(em);
6312 return ERR_PTR(err);
6317 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6320 struct btrfs_root *root = BTRFS_I(inode)->root;
6321 struct extent_map *em;
6322 struct btrfs_key ins;
6326 alloc_hint = get_extent_allocation_hint(inode, start, len);
6327 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6328 alloc_hint, &ins, 1);
6330 return ERR_PTR(ret);
6332 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6333 ins.offset, ins.offset, ins.offset, 0);
6335 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6339 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6340 ins.offset, ins.offset, 0);
6342 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6343 free_extent_map(em);
6344 return ERR_PTR(ret);
6351 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6352 * block must be cow'd
6354 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6355 u64 *orig_start, u64 *orig_block_len,
6358 struct btrfs_trans_handle *trans;
6359 struct btrfs_path *path;
6361 struct extent_buffer *leaf;
6362 struct btrfs_root *root = BTRFS_I(inode)->root;
6363 struct btrfs_file_extent_item *fi;
6364 struct btrfs_key key;
6371 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6372 path = btrfs_alloc_path();
6376 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6381 slot = path->slots[0];
6384 /* can't find the item, must cow */
6391 leaf = path->nodes[0];
6392 btrfs_item_key_to_cpu(leaf, &key, slot);
6393 if (key.objectid != btrfs_ino(inode) ||
6394 key.type != BTRFS_EXTENT_DATA_KEY) {
6395 /* not our file or wrong item type, must cow */
6399 if (key.offset > offset) {
6400 /* Wrong offset, must cow */
6404 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6405 found_type = btrfs_file_extent_type(leaf, fi);
6406 if (found_type != BTRFS_FILE_EXTENT_REG &&
6407 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6408 /* not a regular extent, must cow */
6412 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6415 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6416 if (disk_bytenr == 0)
6419 if (btrfs_file_extent_compression(leaf, fi) ||
6420 btrfs_file_extent_encryption(leaf, fi) ||
6421 btrfs_file_extent_other_encoding(leaf, fi))
6424 backref_offset = btrfs_file_extent_offset(leaf, fi);
6427 *orig_start = key.offset - backref_offset;
6428 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6429 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6432 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6434 if (btrfs_extent_readonly(root, disk_bytenr))
6436 btrfs_release_path(path);
6439 * look for other files referencing this extent, if we
6440 * find any we must cow
6442 trans = btrfs_join_transaction(root);
6443 if (IS_ERR(trans)) {
6448 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6449 key.offset - backref_offset, disk_bytenr);
6450 btrfs_end_transaction(trans, root);
6457 * adjust disk_bytenr and num_bytes to cover just the bytes
6458 * in this extent we are about to write. If there
6459 * are any csums in that range we have to cow in order
6460 * to keep the csums correct
6462 disk_bytenr += backref_offset;
6463 disk_bytenr += offset - key.offset;
6464 num_bytes = min(offset + *len, extent_end) - offset;
6465 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6468 * all of the above have passed, it is safe to overwrite this extent
6474 btrfs_free_path(path);
6478 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6479 struct extent_state **cached_state, int writing)
6481 struct btrfs_ordered_extent *ordered;
6485 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6488 * We're concerned with the entire range that we're going to be
6489 * doing DIO to, so we need to make sure theres no ordered
6490 * extents in this range.
6492 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6493 lockend - lockstart + 1);
6496 * We need to make sure there are no buffered pages in this
6497 * range either, we could have raced between the invalidate in
6498 * generic_file_direct_write and locking the extent. The
6499 * invalidate needs to happen so that reads after a write do not
6502 if (!ordered && (!writing ||
6503 !test_range_bit(&BTRFS_I(inode)->io_tree,
6504 lockstart, lockend, EXTENT_UPTODATE, 0,
6508 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6509 cached_state, GFP_NOFS);
6512 btrfs_start_ordered_extent(inode, ordered, 1);
6513 btrfs_put_ordered_extent(ordered);
6515 /* Screw you mmap */
6516 ret = filemap_write_and_wait_range(inode->i_mapping,
6523 * If we found a page that couldn't be invalidated just
6524 * fall back to buffered.
6526 ret = invalidate_inode_pages2_range(inode->i_mapping,
6527 lockstart >> PAGE_CACHE_SHIFT,
6528 lockend >> PAGE_CACHE_SHIFT);
6539 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6540 u64 len, u64 orig_start,
6541 u64 block_start, u64 block_len,
6542 u64 orig_block_len, u64 ram_bytes,
6545 struct extent_map_tree *em_tree;
6546 struct extent_map *em;
6547 struct btrfs_root *root = BTRFS_I(inode)->root;
6550 em_tree = &BTRFS_I(inode)->extent_tree;
6551 em = alloc_extent_map();
6553 return ERR_PTR(-ENOMEM);
6556 em->orig_start = orig_start;
6557 em->mod_start = start;
6560 em->block_len = block_len;
6561 em->block_start = block_start;
6562 em->bdev = root->fs_info->fs_devices->latest_bdev;
6563 em->orig_block_len = orig_block_len;
6564 em->ram_bytes = ram_bytes;
6565 em->generation = -1;
6566 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6567 if (type == BTRFS_ORDERED_PREALLOC)
6568 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6571 btrfs_drop_extent_cache(inode, em->start,
6572 em->start + em->len - 1, 0);
6573 write_lock(&em_tree->lock);
6574 ret = add_extent_mapping(em_tree, em, 1);
6575 write_unlock(&em_tree->lock);
6576 } while (ret == -EEXIST);
6579 free_extent_map(em);
6580 return ERR_PTR(ret);
6587 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6588 struct buffer_head *bh_result, int create)
6590 struct extent_map *em;
6591 struct btrfs_root *root = BTRFS_I(inode)->root;
6592 struct extent_state *cached_state = NULL;
6593 u64 start = iblock << inode->i_blkbits;
6594 u64 lockstart, lockend;
6595 u64 len = bh_result->b_size;
6596 int unlock_bits = EXTENT_LOCKED;
6600 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6602 len = min_t(u64, len, root->sectorsize);
6605 lockend = start + len - 1;
6608 * If this errors out it's because we couldn't invalidate pagecache for
6609 * this range and we need to fallback to buffered.
6611 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6614 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6621 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6622 * io. INLINE is special, and we could probably kludge it in here, but
6623 * it's still buffered so for safety lets just fall back to the generic
6626 * For COMPRESSED we _have_ to read the entire extent in so we can
6627 * decompress it, so there will be buffering required no matter what we
6628 * do, so go ahead and fallback to buffered.
6630 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6631 * to buffered IO. Don't blame me, this is the price we pay for using
6634 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6635 em->block_start == EXTENT_MAP_INLINE) {
6636 free_extent_map(em);
6641 /* Just a good old fashioned hole, return */
6642 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6643 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6644 free_extent_map(em);
6649 * We don't allocate a new extent in the following cases
6651 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6653 * 2) The extent is marked as PREALLOC. We're good to go here and can
6654 * just use the extent.
6658 len = min(len, em->len - (start - em->start));
6659 lockstart = start + len;
6663 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6664 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6665 em->block_start != EXTENT_MAP_HOLE)) {
6668 u64 block_start, orig_start, orig_block_len, ram_bytes;
6670 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6671 type = BTRFS_ORDERED_PREALLOC;
6673 type = BTRFS_ORDERED_NOCOW;
6674 len = min(len, em->len - (start - em->start));
6675 block_start = em->block_start + (start - em->start);
6677 if (can_nocow_extent(inode, start, &len, &orig_start,
6678 &orig_block_len, &ram_bytes) == 1) {
6679 if (type == BTRFS_ORDERED_PREALLOC) {
6680 free_extent_map(em);
6681 em = create_pinned_em(inode, start, len,
6690 ret = btrfs_add_ordered_extent_dio(inode, start,
6691 block_start, len, len, type);
6693 free_extent_map(em);
6701 * this will cow the extent, reset the len in case we changed
6704 len = bh_result->b_size;
6705 free_extent_map(em);
6706 em = btrfs_new_extent_direct(inode, start, len);
6711 len = min(len, em->len - (start - em->start));
6713 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6715 bh_result->b_size = len;
6716 bh_result->b_bdev = em->bdev;
6717 set_buffer_mapped(bh_result);
6719 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6720 set_buffer_new(bh_result);
6723 * Need to update the i_size under the extent lock so buffered
6724 * readers will get the updated i_size when we unlock.
6726 if (start + len > i_size_read(inode))
6727 i_size_write(inode, start + len);
6729 spin_lock(&BTRFS_I(inode)->lock);
6730 BTRFS_I(inode)->outstanding_extents++;
6731 spin_unlock(&BTRFS_I(inode)->lock);
6733 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6734 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6735 &cached_state, GFP_NOFS);
6740 * In the case of write we need to clear and unlock the entire range,
6741 * in the case of read we need to unlock only the end area that we
6742 * aren't using if there is any left over space.
6744 if (lockstart < lockend) {
6745 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6746 lockend, unlock_bits, 1, 0,
6747 &cached_state, GFP_NOFS);
6749 free_extent_state(cached_state);
6752 free_extent_map(em);
6757 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6758 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6762 static void btrfs_endio_direct_read(struct bio *bio, int err)
6764 struct btrfs_dio_private *dip = bio->bi_private;
6765 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6766 struct bio_vec *bvec = bio->bi_io_vec;
6767 struct inode *inode = dip->inode;
6768 struct btrfs_root *root = BTRFS_I(inode)->root;
6769 struct bio *dio_bio;
6770 u32 *csums = (u32 *)dip->csum;
6774 start = dip->logical_offset;
6776 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6777 struct page *page = bvec->bv_page;
6780 unsigned long flags;
6782 local_irq_save(flags);
6783 kaddr = kmap_atomic(page);
6784 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6785 csum, bvec->bv_len);
6786 btrfs_csum_final(csum, (char *)&csum);
6787 kunmap_atomic(kaddr);
6788 local_irq_restore(flags);
6790 flush_dcache_page(bvec->bv_page);
6791 if (csum != csums[index]) {
6792 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6793 btrfs_ino(inode), start, csum,
6799 start += bvec->bv_len;
6802 } while (bvec <= bvec_end);
6804 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6805 dip->logical_offset + dip->bytes - 1);
6806 dio_bio = dip->dio_bio;
6810 /* If we had a csum failure make sure to clear the uptodate flag */
6812 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6813 dio_end_io(dio_bio, err);
6817 static void btrfs_endio_direct_write(struct bio *bio, int err)
6819 struct btrfs_dio_private *dip = bio->bi_private;
6820 struct inode *inode = dip->inode;
6821 struct btrfs_root *root = BTRFS_I(inode)->root;
6822 struct btrfs_ordered_extent *ordered = NULL;
6823 u64 ordered_offset = dip->logical_offset;
6824 u64 ordered_bytes = dip->bytes;
6825 struct bio *dio_bio;
6831 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6833 ordered_bytes, !err);
6837 ordered->work.func = finish_ordered_fn;
6838 ordered->work.flags = 0;
6839 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6843 * our bio might span multiple ordered extents. If we haven't
6844 * completed the accounting for the whole dio, go back and try again
6846 if (ordered_offset < dip->logical_offset + dip->bytes) {
6847 ordered_bytes = dip->logical_offset + dip->bytes -
6853 dio_bio = dip->dio_bio;
6857 /* If we had an error make sure to clear the uptodate flag */
6859 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6860 dio_end_io(dio_bio, err);
6864 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6865 struct bio *bio, int mirror_num,
6866 unsigned long bio_flags, u64 offset)
6869 struct btrfs_root *root = BTRFS_I(inode)->root;
6870 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6871 BUG_ON(ret); /* -ENOMEM */
6875 static void btrfs_end_dio_bio(struct bio *bio, int err)
6877 struct btrfs_dio_private *dip = bio->bi_private;
6880 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6881 "sector %#Lx len %u err no %d\n",
6882 btrfs_ino(dip->inode), bio->bi_rw,
6883 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6887 * before atomic variable goto zero, we must make sure
6888 * dip->errors is perceived to be set.
6890 smp_mb__before_atomic_dec();
6893 /* if there are more bios still pending for this dio, just exit */
6894 if (!atomic_dec_and_test(&dip->pending_bios))
6898 bio_io_error(dip->orig_bio);
6900 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6901 bio_endio(dip->orig_bio, 0);
6907 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6908 u64 first_sector, gfp_t gfp_flags)
6910 int nr_vecs = bio_get_nr_vecs(bdev);
6911 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6914 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6915 int rw, u64 file_offset, int skip_sum,
6918 struct btrfs_dio_private *dip = bio->bi_private;
6919 int write = rw & REQ_WRITE;
6920 struct btrfs_root *root = BTRFS_I(inode)->root;
6924 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6929 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6937 if (write && async_submit) {
6938 ret = btrfs_wq_submit_bio(root->fs_info,
6939 inode, rw, bio, 0, 0,
6941 __btrfs_submit_bio_start_direct_io,
6942 __btrfs_submit_bio_done);
6946 * If we aren't doing async submit, calculate the csum of the
6949 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6952 } else if (!skip_sum) {
6953 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
6960 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6966 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6969 struct inode *inode = dip->inode;
6970 struct btrfs_root *root = BTRFS_I(inode)->root;
6972 struct bio *orig_bio = dip->orig_bio;
6973 struct bio_vec *bvec = orig_bio->bi_io_vec;
6974 u64 start_sector = orig_bio->bi_sector;
6975 u64 file_offset = dip->logical_offset;
6980 int async_submit = 0;
6982 map_length = orig_bio->bi_size;
6983 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
6984 &map_length, NULL, 0);
6990 if (map_length >= orig_bio->bi_size) {
6995 /* async crcs make it difficult to collect full stripe writes. */
6996 if (btrfs_get_alloc_profile(root, 1) &
6997 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7002 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7005 bio->bi_private = dip;
7006 bio->bi_end_io = btrfs_end_dio_bio;
7007 atomic_inc(&dip->pending_bios);
7009 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7010 if (unlikely(map_length < submit_len + bvec->bv_len ||
7011 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7012 bvec->bv_offset) < bvec->bv_len)) {
7014 * inc the count before we submit the bio so
7015 * we know the end IO handler won't happen before
7016 * we inc the count. Otherwise, the dip might get freed
7017 * before we're done setting it up
7019 atomic_inc(&dip->pending_bios);
7020 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7021 file_offset, skip_sum,
7025 atomic_dec(&dip->pending_bios);
7029 start_sector += submit_len >> 9;
7030 file_offset += submit_len;
7035 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7036 start_sector, GFP_NOFS);
7039 bio->bi_private = dip;
7040 bio->bi_end_io = btrfs_end_dio_bio;
7042 map_length = orig_bio->bi_size;
7043 ret = btrfs_map_block(root->fs_info, rw,
7045 &map_length, NULL, 0);
7051 submit_len += bvec->bv_len;
7058 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7067 * before atomic variable goto zero, we must
7068 * make sure dip->errors is perceived to be set.
7070 smp_mb__before_atomic_dec();
7071 if (atomic_dec_and_test(&dip->pending_bios))
7072 bio_io_error(dip->orig_bio);
7074 /* bio_end_io() will handle error, so we needn't return it */
7078 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7079 struct inode *inode, loff_t file_offset)
7081 struct btrfs_root *root = BTRFS_I(inode)->root;
7082 struct btrfs_dio_private *dip;
7086 int write = rw & REQ_WRITE;
7090 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7092 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7098 if (!skip_sum && !write) {
7099 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7100 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7101 sum_len *= csum_size;
7106 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7112 dip->private = dio_bio->bi_private;
7114 dip->logical_offset = file_offset;
7115 dip->bytes = dio_bio->bi_size;
7116 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7117 io_bio->bi_private = dip;
7119 dip->orig_bio = io_bio;
7120 dip->dio_bio = dio_bio;
7121 atomic_set(&dip->pending_bios, 0);
7124 io_bio->bi_end_io = btrfs_endio_direct_write;
7126 io_bio->bi_end_io = btrfs_endio_direct_read;
7128 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7137 * If this is a write, we need to clean up the reserved space and kill
7138 * the ordered extent.
7141 struct btrfs_ordered_extent *ordered;
7142 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7143 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7144 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7145 btrfs_free_reserved_extent(root, ordered->start,
7147 btrfs_put_ordered_extent(ordered);
7148 btrfs_put_ordered_extent(ordered);
7150 bio_endio(dio_bio, ret);
7153 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7154 const struct iovec *iov, loff_t offset,
7155 unsigned long nr_segs)
7161 unsigned blocksize_mask = root->sectorsize - 1;
7162 ssize_t retval = -EINVAL;
7163 loff_t end = offset;
7165 if (offset & blocksize_mask)
7168 /* Check the memory alignment. Blocks cannot straddle pages */
7169 for (seg = 0; seg < nr_segs; seg++) {
7170 addr = (unsigned long)iov[seg].iov_base;
7171 size = iov[seg].iov_len;
7173 if ((addr & blocksize_mask) || (size & blocksize_mask))
7176 /* If this is a write we don't need to check anymore */
7181 * Check to make sure we don't have duplicate iov_base's in this
7182 * iovec, if so return EINVAL, otherwise we'll get csum errors
7183 * when reading back.
7185 for (i = seg + 1; i < nr_segs; i++) {
7186 if (iov[seg].iov_base == iov[i].iov_base)
7195 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7196 const struct iovec *iov, loff_t offset,
7197 unsigned long nr_segs)
7199 struct file *file = iocb->ki_filp;
7200 struct inode *inode = file->f_mapping->host;
7204 bool relock = false;
7207 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7211 atomic_inc(&inode->i_dio_count);
7212 smp_mb__after_atomic_inc();
7215 * The generic stuff only does filemap_write_and_wait_range, which isn't
7216 * enough if we've written compressed pages to this area, so we need to
7217 * call btrfs_wait_ordered_range to make absolutely sure that any
7218 * outstanding dirty pages are on disk.
7220 count = iov_length(iov, nr_segs);
7221 btrfs_wait_ordered_range(inode, offset, count);
7225 * If the write DIO is beyond the EOF, we need update
7226 * the isize, but it is protected by i_mutex. So we can
7227 * not unlock the i_mutex at this case.
7229 if (offset + count <= inode->i_size) {
7230 mutex_unlock(&inode->i_mutex);
7233 ret = btrfs_delalloc_reserve_space(inode, count);
7236 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7237 &BTRFS_I(inode)->runtime_flags))) {
7238 inode_dio_done(inode);
7239 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7243 ret = __blockdev_direct_IO(rw, iocb, inode,
7244 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7245 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7246 btrfs_submit_direct, flags);
7248 if (ret < 0 && ret != -EIOCBQUEUED)
7249 btrfs_delalloc_release_space(inode, count);
7250 else if (ret >= 0 && (size_t)ret < count)
7251 btrfs_delalloc_release_space(inode,
7252 count - (size_t)ret);
7254 btrfs_delalloc_release_metadata(inode, 0);
7258 inode_dio_done(inode);
7260 mutex_lock(&inode->i_mutex);
7265 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7267 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7268 __u64 start, __u64 len)
7272 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7276 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7279 int btrfs_readpage(struct file *file, struct page *page)
7281 struct extent_io_tree *tree;
7282 tree = &BTRFS_I(page->mapping->host)->io_tree;
7283 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7286 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7288 struct extent_io_tree *tree;
7291 if (current->flags & PF_MEMALLOC) {
7292 redirty_page_for_writepage(wbc, page);
7296 tree = &BTRFS_I(page->mapping->host)->io_tree;
7297 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7300 static int btrfs_writepages(struct address_space *mapping,
7301 struct writeback_control *wbc)
7303 struct extent_io_tree *tree;
7305 tree = &BTRFS_I(mapping->host)->io_tree;
7306 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7310 btrfs_readpages(struct file *file, struct address_space *mapping,
7311 struct list_head *pages, unsigned nr_pages)
7313 struct extent_io_tree *tree;
7314 tree = &BTRFS_I(mapping->host)->io_tree;
7315 return extent_readpages(tree, mapping, pages, nr_pages,
7318 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7320 struct extent_io_tree *tree;
7321 struct extent_map_tree *map;
7324 tree = &BTRFS_I(page->mapping->host)->io_tree;
7325 map = &BTRFS_I(page->mapping->host)->extent_tree;
7326 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7328 ClearPagePrivate(page);
7329 set_page_private(page, 0);
7330 page_cache_release(page);
7335 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7337 if (PageWriteback(page) || PageDirty(page))
7339 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7342 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7343 unsigned int length)
7345 struct inode *inode = page->mapping->host;
7346 struct extent_io_tree *tree;
7347 struct btrfs_ordered_extent *ordered;
7348 struct extent_state *cached_state = NULL;
7349 u64 page_start = page_offset(page);
7350 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7353 * we have the page locked, so new writeback can't start,
7354 * and the dirty bit won't be cleared while we are here.
7356 * Wait for IO on this page so that we can safely clear
7357 * the PagePrivate2 bit and do ordered accounting
7359 wait_on_page_writeback(page);
7361 tree = &BTRFS_I(inode)->io_tree;
7363 btrfs_releasepage(page, GFP_NOFS);
7366 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7367 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7370 * IO on this page will never be started, so we need
7371 * to account for any ordered extents now
7373 clear_extent_bit(tree, page_start, page_end,
7374 EXTENT_DIRTY | EXTENT_DELALLOC |
7375 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7376 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7378 * whoever cleared the private bit is responsible
7379 * for the finish_ordered_io
7381 if (TestClearPagePrivate2(page)) {
7382 struct btrfs_ordered_inode_tree *tree;
7385 tree = &BTRFS_I(inode)->ordered_tree;
7387 spin_lock_irq(&tree->lock);
7388 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7389 new_len = page_start - ordered->file_offset;
7390 if (new_len < ordered->truncated_len)
7391 ordered->truncated_len = new_len;
7392 spin_unlock_irq(&tree->lock);
7394 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7396 PAGE_CACHE_SIZE, 1))
7397 btrfs_finish_ordered_io(ordered);
7399 btrfs_put_ordered_extent(ordered);
7400 cached_state = NULL;
7401 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7403 clear_extent_bit(tree, page_start, page_end,
7404 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7405 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7406 &cached_state, GFP_NOFS);
7407 __btrfs_releasepage(page, GFP_NOFS);
7409 ClearPageChecked(page);
7410 if (PagePrivate(page)) {
7411 ClearPagePrivate(page);
7412 set_page_private(page, 0);
7413 page_cache_release(page);
7418 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7419 * called from a page fault handler when a page is first dirtied. Hence we must
7420 * be careful to check for EOF conditions here. We set the page up correctly
7421 * for a written page which means we get ENOSPC checking when writing into
7422 * holes and correct delalloc and unwritten extent mapping on filesystems that
7423 * support these features.
7425 * We are not allowed to take the i_mutex here so we have to play games to
7426 * protect against truncate races as the page could now be beyond EOF. Because
7427 * vmtruncate() writes the inode size before removing pages, once we have the
7428 * page lock we can determine safely if the page is beyond EOF. If it is not
7429 * beyond EOF, then the page is guaranteed safe against truncation until we
7432 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7434 struct page *page = vmf->page;
7435 struct inode *inode = file_inode(vma->vm_file);
7436 struct btrfs_root *root = BTRFS_I(inode)->root;
7437 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7438 struct btrfs_ordered_extent *ordered;
7439 struct extent_state *cached_state = NULL;
7441 unsigned long zero_start;
7448 sb_start_pagefault(inode->i_sb);
7449 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7451 ret = file_update_time(vma->vm_file);
7457 else /* -ENOSPC, -EIO, etc */
7458 ret = VM_FAULT_SIGBUS;
7464 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7467 size = i_size_read(inode);
7468 page_start = page_offset(page);
7469 page_end = page_start + PAGE_CACHE_SIZE - 1;
7471 if ((page->mapping != inode->i_mapping) ||
7472 (page_start >= size)) {
7473 /* page got truncated out from underneath us */
7476 wait_on_page_writeback(page);
7478 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7479 set_page_extent_mapped(page);
7482 * we can't set the delalloc bits if there are pending ordered
7483 * extents. Drop our locks and wait for them to finish
7485 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7487 unlock_extent_cached(io_tree, page_start, page_end,
7488 &cached_state, GFP_NOFS);
7490 btrfs_start_ordered_extent(inode, ordered, 1);
7491 btrfs_put_ordered_extent(ordered);
7496 * XXX - page_mkwrite gets called every time the page is dirtied, even
7497 * if it was already dirty, so for space accounting reasons we need to
7498 * clear any delalloc bits for the range we are fixing to save. There
7499 * is probably a better way to do this, but for now keep consistent with
7500 * prepare_pages in the normal write path.
7502 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7503 EXTENT_DIRTY | EXTENT_DELALLOC |
7504 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7505 0, 0, &cached_state, GFP_NOFS);
7507 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7510 unlock_extent_cached(io_tree, page_start, page_end,
7511 &cached_state, GFP_NOFS);
7512 ret = VM_FAULT_SIGBUS;
7517 /* page is wholly or partially inside EOF */
7518 if (page_start + PAGE_CACHE_SIZE > size)
7519 zero_start = size & ~PAGE_CACHE_MASK;
7521 zero_start = PAGE_CACHE_SIZE;
7523 if (zero_start != PAGE_CACHE_SIZE) {
7525 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7526 flush_dcache_page(page);
7529 ClearPageChecked(page);
7530 set_page_dirty(page);
7531 SetPageUptodate(page);
7533 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7534 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7535 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7537 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7541 sb_end_pagefault(inode->i_sb);
7542 return VM_FAULT_LOCKED;
7546 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7548 sb_end_pagefault(inode->i_sb);
7552 static int btrfs_truncate(struct inode *inode)
7554 struct btrfs_root *root = BTRFS_I(inode)->root;
7555 struct btrfs_block_rsv *rsv;
7558 struct btrfs_trans_handle *trans;
7559 u64 mask = root->sectorsize - 1;
7560 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7562 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7565 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7566 * 3 things going on here
7568 * 1) We need to reserve space for our orphan item and the space to
7569 * delete our orphan item. Lord knows we don't want to have a dangling
7570 * orphan item because we didn't reserve space to remove it.
7572 * 2) We need to reserve space to update our inode.
7574 * 3) We need to have something to cache all the space that is going to
7575 * be free'd up by the truncate operation, but also have some slack
7576 * space reserved in case it uses space during the truncate (thank you
7577 * very much snapshotting).
7579 * And we need these to all be seperate. The fact is we can use alot of
7580 * space doing the truncate, and we have no earthly idea how much space
7581 * we will use, so we need the truncate reservation to be seperate so it
7582 * doesn't end up using space reserved for updating the inode or
7583 * removing the orphan item. We also need to be able to stop the
7584 * transaction and start a new one, which means we need to be able to
7585 * update the inode several times, and we have no idea of knowing how
7586 * many times that will be, so we can't just reserve 1 item for the
7587 * entirety of the opration, so that has to be done seperately as well.
7588 * Then there is the orphan item, which does indeed need to be held on
7589 * to for the whole operation, and we need nobody to touch this reserved
7590 * space except the orphan code.
7592 * So that leaves us with
7594 * 1) root->orphan_block_rsv - for the orphan deletion.
7595 * 2) rsv - for the truncate reservation, which we will steal from the
7596 * transaction reservation.
7597 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7598 * updating the inode.
7600 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7603 rsv->size = min_size;
7607 * 1 for the truncate slack space
7608 * 1 for updating the inode.
7610 trans = btrfs_start_transaction(root, 2);
7611 if (IS_ERR(trans)) {
7612 err = PTR_ERR(trans);
7616 /* Migrate the slack space for the truncate to our reserve */
7617 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7622 * setattr is responsible for setting the ordered_data_close flag,
7623 * but that is only tested during the last file release. That
7624 * could happen well after the next commit, leaving a great big
7625 * window where new writes may get lost if someone chooses to write
7626 * to this file after truncating to zero
7628 * The inode doesn't have any dirty data here, and so if we commit
7629 * this is a noop. If someone immediately starts writing to the inode
7630 * it is very likely we'll catch some of their writes in this
7631 * transaction, and the commit will find this file on the ordered
7632 * data list with good things to send down.
7634 * This is a best effort solution, there is still a window where
7635 * using truncate to replace the contents of the file will
7636 * end up with a zero length file after a crash.
7638 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7639 &BTRFS_I(inode)->runtime_flags))
7640 btrfs_add_ordered_operation(trans, root, inode);
7643 * So if we truncate and then write and fsync we normally would just
7644 * write the extents that changed, which is a problem if we need to
7645 * first truncate that entire inode. So set this flag so we write out
7646 * all of the extents in the inode to the sync log so we're completely
7649 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7650 trans->block_rsv = rsv;
7653 ret = btrfs_truncate_inode_items(trans, root, inode,
7655 BTRFS_EXTENT_DATA_KEY);
7656 if (ret != -ENOSPC) {
7661 trans->block_rsv = &root->fs_info->trans_block_rsv;
7662 ret = btrfs_update_inode(trans, root, inode);
7668 btrfs_end_transaction(trans, root);
7669 btrfs_btree_balance_dirty(root);
7671 trans = btrfs_start_transaction(root, 2);
7672 if (IS_ERR(trans)) {
7673 ret = err = PTR_ERR(trans);
7678 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7680 BUG_ON(ret); /* shouldn't happen */
7681 trans->block_rsv = rsv;
7684 if (ret == 0 && inode->i_nlink > 0) {
7685 trans->block_rsv = root->orphan_block_rsv;
7686 ret = btrfs_orphan_del(trans, inode);
7692 trans->block_rsv = &root->fs_info->trans_block_rsv;
7693 ret = btrfs_update_inode(trans, root, inode);
7697 ret = btrfs_end_transaction(trans, root);
7698 btrfs_btree_balance_dirty(root);
7702 btrfs_free_block_rsv(root, rsv);
7711 * create a new subvolume directory/inode (helper for the ioctl).
7713 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7714 struct btrfs_root *new_root, u64 new_dirid)
7716 struct inode *inode;
7720 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7721 new_dirid, new_dirid,
7722 S_IFDIR | (~current_umask() & S_IRWXUGO),
7725 return PTR_ERR(inode);
7726 inode->i_op = &btrfs_dir_inode_operations;
7727 inode->i_fop = &btrfs_dir_file_operations;
7729 set_nlink(inode, 1);
7730 btrfs_i_size_write(inode, 0);
7732 err = btrfs_update_inode(trans, new_root, inode);
7738 struct inode *btrfs_alloc_inode(struct super_block *sb)
7740 struct btrfs_inode *ei;
7741 struct inode *inode;
7743 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7750 ei->last_sub_trans = 0;
7751 ei->logged_trans = 0;
7752 ei->delalloc_bytes = 0;
7753 ei->disk_i_size = 0;
7756 ei->index_cnt = (u64)-1;
7757 ei->last_unlink_trans = 0;
7758 ei->last_log_commit = 0;
7760 spin_lock_init(&ei->lock);
7761 ei->outstanding_extents = 0;
7762 ei->reserved_extents = 0;
7764 ei->runtime_flags = 0;
7765 ei->force_compress = BTRFS_COMPRESS_NONE;
7767 ei->delayed_node = NULL;
7769 inode = &ei->vfs_inode;
7770 extent_map_tree_init(&ei->extent_tree);
7771 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7772 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7773 ei->io_tree.track_uptodate = 1;
7774 ei->io_failure_tree.track_uptodate = 1;
7775 atomic_set(&ei->sync_writers, 0);
7776 mutex_init(&ei->log_mutex);
7777 mutex_init(&ei->delalloc_mutex);
7778 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7779 INIT_LIST_HEAD(&ei->delalloc_inodes);
7780 INIT_LIST_HEAD(&ei->ordered_operations);
7781 RB_CLEAR_NODE(&ei->rb_node);
7786 static void btrfs_i_callback(struct rcu_head *head)
7788 struct inode *inode = container_of(head, struct inode, i_rcu);
7789 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7792 void btrfs_destroy_inode(struct inode *inode)
7794 struct btrfs_ordered_extent *ordered;
7795 struct btrfs_root *root = BTRFS_I(inode)->root;
7797 WARN_ON(!hlist_empty(&inode->i_dentry));
7798 WARN_ON(inode->i_data.nrpages);
7799 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7800 WARN_ON(BTRFS_I(inode)->reserved_extents);
7801 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7802 WARN_ON(BTRFS_I(inode)->csum_bytes);
7805 * This can happen where we create an inode, but somebody else also
7806 * created the same inode and we need to destroy the one we already
7813 * Make sure we're properly removed from the ordered operation
7817 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7818 spin_lock(&root->fs_info->ordered_root_lock);
7819 list_del_init(&BTRFS_I(inode)->ordered_operations);
7820 spin_unlock(&root->fs_info->ordered_root_lock);
7823 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7824 &BTRFS_I(inode)->runtime_flags)) {
7825 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7827 atomic_dec(&root->orphan_inodes);
7831 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7835 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7836 ordered->file_offset, ordered->len);
7837 btrfs_remove_ordered_extent(inode, ordered);
7838 btrfs_put_ordered_extent(ordered);
7839 btrfs_put_ordered_extent(ordered);
7842 inode_tree_del(inode);
7843 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7845 call_rcu(&inode->i_rcu, btrfs_i_callback);
7848 int btrfs_drop_inode(struct inode *inode)
7850 struct btrfs_root *root = BTRFS_I(inode)->root;
7855 /* the snap/subvol tree is on deleting */
7856 if (btrfs_root_refs(&root->root_item) == 0)
7859 return generic_drop_inode(inode);
7862 static void init_once(void *foo)
7864 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7866 inode_init_once(&ei->vfs_inode);
7869 void btrfs_destroy_cachep(void)
7872 * Make sure all delayed rcu free inodes are flushed before we
7876 if (btrfs_inode_cachep)
7877 kmem_cache_destroy(btrfs_inode_cachep);
7878 if (btrfs_trans_handle_cachep)
7879 kmem_cache_destroy(btrfs_trans_handle_cachep);
7880 if (btrfs_transaction_cachep)
7881 kmem_cache_destroy(btrfs_transaction_cachep);
7882 if (btrfs_path_cachep)
7883 kmem_cache_destroy(btrfs_path_cachep);
7884 if (btrfs_free_space_cachep)
7885 kmem_cache_destroy(btrfs_free_space_cachep);
7886 if (btrfs_delalloc_work_cachep)
7887 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7890 int btrfs_init_cachep(void)
7892 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7893 sizeof(struct btrfs_inode), 0,
7894 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7895 if (!btrfs_inode_cachep)
7898 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7899 sizeof(struct btrfs_trans_handle), 0,
7900 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7901 if (!btrfs_trans_handle_cachep)
7904 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7905 sizeof(struct btrfs_transaction), 0,
7906 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7907 if (!btrfs_transaction_cachep)
7910 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7911 sizeof(struct btrfs_path), 0,
7912 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7913 if (!btrfs_path_cachep)
7916 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7917 sizeof(struct btrfs_free_space), 0,
7918 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7919 if (!btrfs_free_space_cachep)
7922 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7923 sizeof(struct btrfs_delalloc_work), 0,
7924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7926 if (!btrfs_delalloc_work_cachep)
7931 btrfs_destroy_cachep();
7935 static int btrfs_getattr(struct vfsmount *mnt,
7936 struct dentry *dentry, struct kstat *stat)
7939 struct inode *inode = dentry->d_inode;
7940 u32 blocksize = inode->i_sb->s_blocksize;
7942 generic_fillattr(inode, stat);
7943 stat->dev = BTRFS_I(inode)->root->anon_dev;
7944 stat->blksize = PAGE_CACHE_SIZE;
7946 spin_lock(&BTRFS_I(inode)->lock);
7947 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7948 spin_unlock(&BTRFS_I(inode)->lock);
7949 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7950 ALIGN(delalloc_bytes, blocksize)) >> 9;
7954 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7955 struct inode *new_dir, struct dentry *new_dentry)
7957 struct btrfs_trans_handle *trans;
7958 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7959 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7960 struct inode *new_inode = new_dentry->d_inode;
7961 struct inode *old_inode = old_dentry->d_inode;
7962 struct timespec ctime = CURRENT_TIME;
7966 u64 old_ino = btrfs_ino(old_inode);
7968 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7971 /* we only allow rename subvolume link between subvolumes */
7972 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7975 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7976 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7979 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7980 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7984 /* check for collisions, even if the name isn't there */
7985 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
7986 new_dentry->d_name.name,
7987 new_dentry->d_name.len);
7990 if (ret == -EEXIST) {
7992 * eexist without a new_inode */
7998 /* maybe -EOVERFLOW */
8005 * we're using rename to replace one file with another.
8006 * and the replacement file is large. Start IO on it now so
8007 * we don't add too much work to the end of the transaction
8009 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8010 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8011 filemap_flush(old_inode->i_mapping);
8013 /* close the racy window with snapshot create/destroy ioctl */
8014 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8015 down_read(&root->fs_info->subvol_sem);
8017 * We want to reserve the absolute worst case amount of items. So if
8018 * both inodes are subvols and we need to unlink them then that would
8019 * require 4 item modifications, but if they are both normal inodes it
8020 * would require 5 item modifications, so we'll assume their normal
8021 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8022 * should cover the worst case number of items we'll modify.
8024 trans = btrfs_start_transaction(root, 11);
8025 if (IS_ERR(trans)) {
8026 ret = PTR_ERR(trans);
8031 btrfs_record_root_in_trans(trans, dest);
8033 ret = btrfs_set_inode_index(new_dir, &index);
8037 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8038 /* force full log commit if subvolume involved. */
8039 root->fs_info->last_trans_log_full_commit = trans->transid;
8041 ret = btrfs_insert_inode_ref(trans, dest,
8042 new_dentry->d_name.name,
8043 new_dentry->d_name.len,
8045 btrfs_ino(new_dir), index);
8049 * this is an ugly little race, but the rename is required
8050 * to make sure that if we crash, the inode is either at the
8051 * old name or the new one. pinning the log transaction lets
8052 * us make sure we don't allow a log commit to come in after
8053 * we unlink the name but before we add the new name back in.
8055 btrfs_pin_log_trans(root);
8058 * make sure the inode gets flushed if it is replacing
8061 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8062 btrfs_add_ordered_operation(trans, root, old_inode);
8064 inode_inc_iversion(old_dir);
8065 inode_inc_iversion(new_dir);
8066 inode_inc_iversion(old_inode);
8067 old_dir->i_ctime = old_dir->i_mtime = ctime;
8068 new_dir->i_ctime = new_dir->i_mtime = ctime;
8069 old_inode->i_ctime = ctime;
8071 if (old_dentry->d_parent != new_dentry->d_parent)
8072 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8074 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8075 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8076 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8077 old_dentry->d_name.name,
8078 old_dentry->d_name.len);
8080 ret = __btrfs_unlink_inode(trans, root, old_dir,
8081 old_dentry->d_inode,
8082 old_dentry->d_name.name,
8083 old_dentry->d_name.len);
8085 ret = btrfs_update_inode(trans, root, old_inode);
8088 btrfs_abort_transaction(trans, root, ret);
8093 inode_inc_iversion(new_inode);
8094 new_inode->i_ctime = CURRENT_TIME;
8095 if (unlikely(btrfs_ino(new_inode) ==
8096 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8097 root_objectid = BTRFS_I(new_inode)->location.objectid;
8098 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8100 new_dentry->d_name.name,
8101 new_dentry->d_name.len);
8102 BUG_ON(new_inode->i_nlink == 0);
8104 ret = btrfs_unlink_inode(trans, dest, new_dir,
8105 new_dentry->d_inode,
8106 new_dentry->d_name.name,
8107 new_dentry->d_name.len);
8109 if (!ret && new_inode->i_nlink == 0)
8110 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8112 btrfs_abort_transaction(trans, root, ret);
8117 ret = btrfs_add_link(trans, new_dir, old_inode,
8118 new_dentry->d_name.name,
8119 new_dentry->d_name.len, 0, index);
8121 btrfs_abort_transaction(trans, root, ret);
8125 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8126 struct dentry *parent = new_dentry->d_parent;
8127 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8128 btrfs_end_log_trans(root);
8131 btrfs_end_transaction(trans, root);
8133 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8134 up_read(&root->fs_info->subvol_sem);
8139 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8141 struct btrfs_delalloc_work *delalloc_work;
8143 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8145 if (delalloc_work->wait)
8146 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8148 filemap_flush(delalloc_work->inode->i_mapping);
8150 if (delalloc_work->delay_iput)
8151 btrfs_add_delayed_iput(delalloc_work->inode);
8153 iput(delalloc_work->inode);
8154 complete(&delalloc_work->completion);
8157 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8158 int wait, int delay_iput)
8160 struct btrfs_delalloc_work *work;
8162 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8166 init_completion(&work->completion);
8167 INIT_LIST_HEAD(&work->list);
8168 work->inode = inode;
8170 work->delay_iput = delay_iput;
8171 work->work.func = btrfs_run_delalloc_work;
8176 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8178 wait_for_completion(&work->completion);
8179 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8183 * some fairly slow code that needs optimization. This walks the list
8184 * of all the inodes with pending delalloc and forces them to disk.
8186 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8188 struct btrfs_inode *binode;
8189 struct inode *inode;
8190 struct btrfs_delalloc_work *work, *next;
8191 struct list_head works;
8192 struct list_head splice;
8195 INIT_LIST_HEAD(&works);
8196 INIT_LIST_HEAD(&splice);
8198 spin_lock(&root->delalloc_lock);
8199 list_splice_init(&root->delalloc_inodes, &splice);
8200 while (!list_empty(&splice)) {
8201 binode = list_entry(splice.next, struct btrfs_inode,
8204 list_move_tail(&binode->delalloc_inodes,
8205 &root->delalloc_inodes);
8206 inode = igrab(&binode->vfs_inode);
8208 cond_resched_lock(&root->delalloc_lock);
8211 spin_unlock(&root->delalloc_lock);
8213 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8214 if (unlikely(!work)) {
8216 btrfs_add_delayed_iput(inode);
8222 list_add_tail(&work->list, &works);
8223 btrfs_queue_worker(&root->fs_info->flush_workers,
8227 spin_lock(&root->delalloc_lock);
8229 spin_unlock(&root->delalloc_lock);
8231 list_for_each_entry_safe(work, next, &works, list) {
8232 list_del_init(&work->list);
8233 btrfs_wait_and_free_delalloc_work(work);
8237 list_for_each_entry_safe(work, next, &works, list) {
8238 list_del_init(&work->list);
8239 btrfs_wait_and_free_delalloc_work(work);
8242 if (!list_empty_careful(&splice)) {
8243 spin_lock(&root->delalloc_lock);
8244 list_splice_tail(&splice, &root->delalloc_inodes);
8245 spin_unlock(&root->delalloc_lock);
8250 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8254 if (root->fs_info->sb->s_flags & MS_RDONLY)
8257 ret = __start_delalloc_inodes(root, delay_iput);
8259 * the filemap_flush will queue IO into the worker threads, but
8260 * we have to make sure the IO is actually started and that
8261 * ordered extents get created before we return
8263 atomic_inc(&root->fs_info->async_submit_draining);
8264 while (atomic_read(&root->fs_info->nr_async_submits) ||
8265 atomic_read(&root->fs_info->async_delalloc_pages)) {
8266 wait_event(root->fs_info->async_submit_wait,
8267 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8268 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8270 atomic_dec(&root->fs_info->async_submit_draining);
8274 int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8277 struct btrfs_root *root;
8278 struct list_head splice;
8281 if (fs_info->sb->s_flags & MS_RDONLY)
8284 INIT_LIST_HEAD(&splice);
8286 spin_lock(&fs_info->delalloc_root_lock);
8287 list_splice_init(&fs_info->delalloc_roots, &splice);
8288 while (!list_empty(&splice)) {
8289 root = list_first_entry(&splice, struct btrfs_root,
8291 root = btrfs_grab_fs_root(root);
8293 list_move_tail(&root->delalloc_root,
8294 &fs_info->delalloc_roots);
8295 spin_unlock(&fs_info->delalloc_root_lock);
8297 ret = __start_delalloc_inodes(root, delay_iput);
8298 btrfs_put_fs_root(root);
8302 spin_lock(&fs_info->delalloc_root_lock);
8304 spin_unlock(&fs_info->delalloc_root_lock);
8306 atomic_inc(&fs_info->async_submit_draining);
8307 while (atomic_read(&fs_info->nr_async_submits) ||
8308 atomic_read(&fs_info->async_delalloc_pages)) {
8309 wait_event(fs_info->async_submit_wait,
8310 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8311 atomic_read(&fs_info->async_delalloc_pages) == 0));
8313 atomic_dec(&fs_info->async_submit_draining);
8316 if (!list_empty_careful(&splice)) {
8317 spin_lock(&fs_info->delalloc_root_lock);
8318 list_splice_tail(&splice, &fs_info->delalloc_roots);
8319 spin_unlock(&fs_info->delalloc_root_lock);
8324 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8325 const char *symname)
8327 struct btrfs_trans_handle *trans;
8328 struct btrfs_root *root = BTRFS_I(dir)->root;
8329 struct btrfs_path *path;
8330 struct btrfs_key key;
8331 struct inode *inode = NULL;
8339 struct btrfs_file_extent_item *ei;
8340 struct extent_buffer *leaf;
8342 name_len = strlen(symname);
8343 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8344 return -ENAMETOOLONG;
8347 * 2 items for inode item and ref
8348 * 2 items for dir items
8349 * 1 item for xattr if selinux is on
8351 trans = btrfs_start_transaction(root, 5);
8353 return PTR_ERR(trans);
8355 err = btrfs_find_free_ino(root, &objectid);
8359 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8360 dentry->d_name.len, btrfs_ino(dir), objectid,
8361 S_IFLNK|S_IRWXUGO, &index);
8362 if (IS_ERR(inode)) {
8363 err = PTR_ERR(inode);
8367 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8374 * If the active LSM wants to access the inode during
8375 * d_instantiate it needs these. Smack checks to see
8376 * if the filesystem supports xattrs by looking at the
8379 inode->i_fop = &btrfs_file_operations;
8380 inode->i_op = &btrfs_file_inode_operations;
8382 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8386 inode->i_mapping->a_ops = &btrfs_aops;
8387 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8388 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8393 path = btrfs_alloc_path();
8399 key.objectid = btrfs_ino(inode);
8401 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8402 datasize = btrfs_file_extent_calc_inline_size(name_len);
8403 err = btrfs_insert_empty_item(trans, root, path, &key,
8407 btrfs_free_path(path);
8410 leaf = path->nodes[0];
8411 ei = btrfs_item_ptr(leaf, path->slots[0],
8412 struct btrfs_file_extent_item);
8413 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8414 btrfs_set_file_extent_type(leaf, ei,
8415 BTRFS_FILE_EXTENT_INLINE);
8416 btrfs_set_file_extent_encryption(leaf, ei, 0);
8417 btrfs_set_file_extent_compression(leaf, ei, 0);
8418 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8419 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8421 ptr = btrfs_file_extent_inline_start(ei);
8422 write_extent_buffer(leaf, symname, ptr, name_len);
8423 btrfs_mark_buffer_dirty(leaf);
8424 btrfs_free_path(path);
8426 inode->i_op = &btrfs_symlink_inode_operations;
8427 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8428 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8429 inode_set_bytes(inode, name_len);
8430 btrfs_i_size_write(inode, name_len);
8431 err = btrfs_update_inode(trans, root, inode);
8437 d_instantiate(dentry, inode);
8438 btrfs_end_transaction(trans, root);
8440 inode_dec_link_count(inode);
8443 btrfs_btree_balance_dirty(root);
8447 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8448 u64 start, u64 num_bytes, u64 min_size,
8449 loff_t actual_len, u64 *alloc_hint,
8450 struct btrfs_trans_handle *trans)
8452 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8453 struct extent_map *em;
8454 struct btrfs_root *root = BTRFS_I(inode)->root;
8455 struct btrfs_key ins;
8456 u64 cur_offset = start;
8460 bool own_trans = true;
8464 while (num_bytes > 0) {
8466 trans = btrfs_start_transaction(root, 3);
8467 if (IS_ERR(trans)) {
8468 ret = PTR_ERR(trans);
8473 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8474 cur_bytes = max(cur_bytes, min_size);
8475 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8476 *alloc_hint, &ins, 1);
8479 btrfs_end_transaction(trans, root);
8483 ret = insert_reserved_file_extent(trans, inode,
8484 cur_offset, ins.objectid,
8485 ins.offset, ins.offset,
8486 ins.offset, 0, 0, 0,
8487 BTRFS_FILE_EXTENT_PREALLOC);
8489 btrfs_abort_transaction(trans, root, ret);
8491 btrfs_end_transaction(trans, root);
8494 btrfs_drop_extent_cache(inode, cur_offset,
8495 cur_offset + ins.offset -1, 0);
8497 em = alloc_extent_map();
8499 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8500 &BTRFS_I(inode)->runtime_flags);
8504 em->start = cur_offset;
8505 em->orig_start = cur_offset;
8506 em->len = ins.offset;
8507 em->block_start = ins.objectid;
8508 em->block_len = ins.offset;
8509 em->orig_block_len = ins.offset;
8510 em->ram_bytes = ins.offset;
8511 em->bdev = root->fs_info->fs_devices->latest_bdev;
8512 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8513 em->generation = trans->transid;
8516 write_lock(&em_tree->lock);
8517 ret = add_extent_mapping(em_tree, em, 1);
8518 write_unlock(&em_tree->lock);
8521 btrfs_drop_extent_cache(inode, cur_offset,
8522 cur_offset + ins.offset - 1,
8525 free_extent_map(em);
8527 num_bytes -= ins.offset;
8528 cur_offset += ins.offset;
8529 *alloc_hint = ins.objectid + ins.offset;
8531 inode_inc_iversion(inode);
8532 inode->i_ctime = CURRENT_TIME;
8533 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8534 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8535 (actual_len > inode->i_size) &&
8536 (cur_offset > inode->i_size)) {
8537 if (cur_offset > actual_len)
8538 i_size = actual_len;
8540 i_size = cur_offset;
8541 i_size_write(inode, i_size);
8542 btrfs_ordered_update_i_size(inode, i_size, NULL);
8545 ret = btrfs_update_inode(trans, root, inode);
8548 btrfs_abort_transaction(trans, root, ret);
8550 btrfs_end_transaction(trans, root);
8555 btrfs_end_transaction(trans, root);
8560 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8561 u64 start, u64 num_bytes, u64 min_size,
8562 loff_t actual_len, u64 *alloc_hint)
8564 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8565 min_size, actual_len, alloc_hint,
8569 int btrfs_prealloc_file_range_trans(struct inode *inode,
8570 struct btrfs_trans_handle *trans, int mode,
8571 u64 start, u64 num_bytes, u64 min_size,
8572 loff_t actual_len, u64 *alloc_hint)
8574 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8575 min_size, actual_len, alloc_hint, trans);
8578 static int btrfs_set_page_dirty(struct page *page)
8580 return __set_page_dirty_nobuffers(page);
8583 static int btrfs_permission(struct inode *inode, int mask)
8585 struct btrfs_root *root = BTRFS_I(inode)->root;
8586 umode_t mode = inode->i_mode;
8588 if (mask & MAY_WRITE &&
8589 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8590 if (btrfs_root_readonly(root))
8592 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8595 return generic_permission(inode, mask);
8598 static const struct inode_operations btrfs_dir_inode_operations = {
8599 .getattr = btrfs_getattr,
8600 .lookup = btrfs_lookup,
8601 .create = btrfs_create,
8602 .unlink = btrfs_unlink,
8604 .mkdir = btrfs_mkdir,
8605 .rmdir = btrfs_rmdir,
8606 .rename = btrfs_rename,
8607 .symlink = btrfs_symlink,
8608 .setattr = btrfs_setattr,
8609 .mknod = btrfs_mknod,
8610 .setxattr = btrfs_setxattr,
8611 .getxattr = btrfs_getxattr,
8612 .listxattr = btrfs_listxattr,
8613 .removexattr = btrfs_removexattr,
8614 .permission = btrfs_permission,
8615 .get_acl = btrfs_get_acl,
8616 .update_time = btrfs_update_time,
8618 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8619 .lookup = btrfs_lookup,
8620 .permission = btrfs_permission,
8621 .get_acl = btrfs_get_acl,
8622 .update_time = btrfs_update_time,
8625 static const struct file_operations btrfs_dir_file_operations = {
8626 .llseek = generic_file_llseek,
8627 .read = generic_read_dir,
8628 .iterate = btrfs_real_readdir,
8629 .unlocked_ioctl = btrfs_ioctl,
8630 #ifdef CONFIG_COMPAT
8631 .compat_ioctl = btrfs_ioctl,
8633 .release = btrfs_release_file,
8634 .fsync = btrfs_sync_file,
8637 static struct extent_io_ops btrfs_extent_io_ops = {
8638 .fill_delalloc = run_delalloc_range,
8639 .submit_bio_hook = btrfs_submit_bio_hook,
8640 .merge_bio_hook = btrfs_merge_bio_hook,
8641 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8642 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8643 .writepage_start_hook = btrfs_writepage_start_hook,
8644 .set_bit_hook = btrfs_set_bit_hook,
8645 .clear_bit_hook = btrfs_clear_bit_hook,
8646 .merge_extent_hook = btrfs_merge_extent_hook,
8647 .split_extent_hook = btrfs_split_extent_hook,
8651 * btrfs doesn't support the bmap operation because swapfiles
8652 * use bmap to make a mapping of extents in the file. They assume
8653 * these extents won't change over the life of the file and they
8654 * use the bmap result to do IO directly to the drive.
8656 * the btrfs bmap call would return logical addresses that aren't
8657 * suitable for IO and they also will change frequently as COW
8658 * operations happen. So, swapfile + btrfs == corruption.
8660 * For now we're avoiding this by dropping bmap.
8662 static const struct address_space_operations btrfs_aops = {
8663 .readpage = btrfs_readpage,
8664 .writepage = btrfs_writepage,
8665 .writepages = btrfs_writepages,
8666 .readpages = btrfs_readpages,
8667 .direct_IO = btrfs_direct_IO,
8668 .invalidatepage = btrfs_invalidatepage,
8669 .releasepage = btrfs_releasepage,
8670 .set_page_dirty = btrfs_set_page_dirty,
8671 .error_remove_page = generic_error_remove_page,
8674 static const struct address_space_operations btrfs_symlink_aops = {
8675 .readpage = btrfs_readpage,
8676 .writepage = btrfs_writepage,
8677 .invalidatepage = btrfs_invalidatepage,
8678 .releasepage = btrfs_releasepage,
8681 static const struct inode_operations btrfs_file_inode_operations = {
8682 .getattr = btrfs_getattr,
8683 .setattr = btrfs_setattr,
8684 .setxattr = btrfs_setxattr,
8685 .getxattr = btrfs_getxattr,
8686 .listxattr = btrfs_listxattr,
8687 .removexattr = btrfs_removexattr,
8688 .permission = btrfs_permission,
8689 .fiemap = btrfs_fiemap,
8690 .get_acl = btrfs_get_acl,
8691 .update_time = btrfs_update_time,
8693 static const struct inode_operations btrfs_special_inode_operations = {
8694 .getattr = btrfs_getattr,
8695 .setattr = btrfs_setattr,
8696 .permission = btrfs_permission,
8697 .setxattr = btrfs_setxattr,
8698 .getxattr = btrfs_getxattr,
8699 .listxattr = btrfs_listxattr,
8700 .removexattr = btrfs_removexattr,
8701 .get_acl = btrfs_get_acl,
8702 .update_time = btrfs_update_time,
8704 static const struct inode_operations btrfs_symlink_inode_operations = {
8705 .readlink = generic_readlink,
8706 .follow_link = page_follow_link_light,
8707 .put_link = page_put_link,
8708 .getattr = btrfs_getattr,
8709 .setattr = btrfs_setattr,
8710 .permission = btrfs_permission,
8711 .setxattr = btrfs_setxattr,
8712 .getxattr = btrfs_getxattr,
8713 .listxattr = btrfs_listxattr,
8714 .removexattr = btrfs_removexattr,
8715 .get_acl = btrfs_get_acl,
8716 .update_time = btrfs_update_time,
8719 const struct dentry_operations btrfs_dentry_operations = {
8720 .d_delete = btrfs_dentry_delete,
8721 .d_release = btrfs_dentry_release,