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>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
62 struct btrfs_iget_args {
64 struct btrfs_root *root;
67 static const struct inode_operations btrfs_dir_inode_operations;
68 static const struct inode_operations btrfs_symlink_inode_operations;
69 static const struct inode_operations btrfs_dir_ro_inode_operations;
70 static const struct inode_operations btrfs_special_inode_operations;
71 static const struct inode_operations btrfs_file_inode_operations;
72 static const struct address_space_operations btrfs_aops;
73 static const struct address_space_operations btrfs_symlink_aops;
74 static const struct file_operations btrfs_dir_file_operations;
75 static struct extent_io_ops btrfs_extent_io_ops;
77 static struct kmem_cache *btrfs_inode_cachep;
78 static struct kmem_cache *btrfs_delalloc_work_cachep;
79 struct kmem_cache *btrfs_trans_handle_cachep;
80 struct kmem_cache *btrfs_transaction_cachep;
81 struct kmem_cache *btrfs_path_cachep;
82 struct kmem_cache *btrfs_free_space_cachep;
85 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
86 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
87 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
88 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
89 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
90 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
91 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
92 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
95 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
96 static int btrfs_truncate(struct inode *inode);
97 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
98 static noinline int cow_file_range(struct inode *inode,
99 struct page *locked_page,
100 u64 start, u64 end, int *page_started,
101 unsigned long *nr_written, int unlock);
102 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
103 u64 len, u64 orig_start,
104 u64 block_start, u64 block_len,
105 u64 orig_block_len, u64 ram_bytes,
108 static int btrfs_dirty_inode(struct inode *inode);
110 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
111 struct inode *inode, struct inode *dir,
112 const struct qstr *qstr)
116 err = btrfs_init_acl(trans, inode, dir);
118 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
127 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
128 struct btrfs_root *root, struct inode *inode,
129 u64 start, size_t size, size_t compressed_size,
131 struct page **compressed_pages)
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
144 unsigned long offset;
146 if (compressed_size && compressed_pages)
147 cur_size = compressed_size;
149 path = btrfs_alloc_path();
153 path->leave_spinning = 1;
155 key.objectid = btrfs_ino(inode);
157 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
160 inode_add_bytes(inode, size);
161 ret = btrfs_insert_empty_item(trans, root, path, &key,
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
177 if (compress_type != BTRFS_COMPRESS_NONE) {
180 while (compressed_size > 0) {
181 cpage = compressed_pages[i];
182 cur_size = min_t(unsigned long, compressed_size,
185 kaddr = kmap_atomic(cpage);
186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 kunmap_atomic(kaddr);
191 compressed_size -= cur_size;
193 btrfs_set_file_extent_compression(leaf, ei,
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
199 kaddr = kmap_atomic(page);
200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 kunmap_atomic(kaddr);
203 page_cache_release(page);
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_free_path(path);
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
217 BTRFS_I(inode)->disk_i_size = inode->i_size;
218 ret = btrfs_update_inode(trans, root, inode);
222 btrfs_free_path(path);
228 * conditionally insert an inline extent into the file. This
229 * does the checks required to make sure the data is small enough
230 * to fit as an inline extent.
232 static noinline int cow_file_range_inline(struct btrfs_root *root,
233 struct inode *inode, u64 start,
234 u64 end, size_t compressed_size,
236 struct page **compressed_pages)
238 struct btrfs_trans_handle *trans;
239 u64 isize = i_size_read(inode);
240 u64 actual_end = min(end + 1, isize);
241 u64 inline_len = actual_end - start;
242 u64 aligned_end = ALIGN(end, root->sectorsize);
243 u64 data_len = inline_len;
247 data_len = compressed_size;
250 actual_end >= PAGE_CACHE_SIZE ||
251 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
253 (actual_end & (root->sectorsize - 1)) == 0) ||
255 data_len > root->fs_info->max_inline) {
259 trans = btrfs_join_transaction(root);
261 return PTR_ERR(trans);
262 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
264 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
266 btrfs_abort_transaction(trans, root, ret);
270 if (isize > actual_end)
271 inline_len = min_t(u64, isize, actual_end);
272 ret = insert_inline_extent(trans, root, inode, start,
273 inline_len, compressed_size,
274 compress_type, compressed_pages);
275 if (ret && ret != -ENOSPC) {
276 btrfs_abort_transaction(trans, root, ret);
278 } else if (ret == -ENOSPC) {
283 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
284 btrfs_delalloc_release_metadata(inode, end + 1 - start);
285 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
287 btrfs_end_transaction(trans, root);
291 struct async_extent {
296 unsigned long nr_pages;
298 struct list_head list;
303 struct btrfs_root *root;
304 struct page *locked_page;
307 struct list_head extents;
308 struct btrfs_work work;
311 static noinline int add_async_extent(struct async_cow *cow,
312 u64 start, u64 ram_size,
315 unsigned long nr_pages,
318 struct async_extent *async_extent;
320 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
321 BUG_ON(!async_extent); /* -ENOMEM */
322 async_extent->start = start;
323 async_extent->ram_size = ram_size;
324 async_extent->compressed_size = compressed_size;
325 async_extent->pages = pages;
326 async_extent->nr_pages = nr_pages;
327 async_extent->compress_type = compress_type;
328 list_add_tail(&async_extent->list, &cow->extents);
333 * we create compressed extents in two phases. The first
334 * phase compresses a range of pages that have already been
335 * locked (both pages and state bits are locked).
337 * This is done inside an ordered work queue, and the compression
338 * is spread across many cpus. The actual IO submission is step
339 * two, and the ordered work queue takes care of making sure that
340 * happens in the same order things were put onto the queue by
341 * writepages and friends.
343 * If this code finds it can't get good compression, it puts an
344 * entry onto the work queue to write the uncompressed bytes. This
345 * makes sure that both compressed inodes and uncompressed inodes
346 * are written in the same order that the flusher thread sent them
349 static noinline int compress_file_range(struct inode *inode,
350 struct page *locked_page,
352 struct async_cow *async_cow,
355 struct btrfs_root *root = BTRFS_I(inode)->root;
357 u64 blocksize = root->sectorsize;
359 u64 isize = i_size_read(inode);
361 struct page **pages = NULL;
362 unsigned long nr_pages;
363 unsigned long nr_pages_ret = 0;
364 unsigned long total_compressed = 0;
365 unsigned long total_in = 0;
366 unsigned long max_compressed = 128 * 1024;
367 unsigned long max_uncompressed = 128 * 1024;
370 int compress_type = root->fs_info->compress_type;
373 /* if this is a small write inside eof, kick off a defrag */
374 if ((end - start + 1) < 16 * 1024 &&
375 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
376 btrfs_add_inode_defrag(NULL, inode);
378 actual_end = min_t(u64, isize, end + 1);
381 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
382 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
385 * we don't want to send crud past the end of i_size through
386 * compression, that's just a waste of CPU time. So, if the
387 * end of the file is before the start of our current
388 * requested range of bytes, we bail out to the uncompressed
389 * cleanup code that can deal with all of this.
391 * It isn't really the fastest way to fix things, but this is a
392 * very uncommon corner.
394 if (actual_end <= start)
395 goto cleanup_and_bail_uncompressed;
397 total_compressed = actual_end - start;
399 /* we want to make sure that amount of ram required to uncompress
400 * an extent is reasonable, so we limit the total size in ram
401 * of a compressed extent to 128k. This is a crucial number
402 * because it also controls how easily we can spread reads across
403 * cpus for decompression.
405 * We also want to make sure the amount of IO required to do
406 * a random read is reasonably small, so we limit the size of
407 * a compressed extent to 128k.
409 total_compressed = min(total_compressed, max_uncompressed);
410 num_bytes = ALIGN(end - start + 1, blocksize);
411 num_bytes = max(blocksize, num_bytes);
416 * we do compression for mount -o compress and when the
417 * inode has not been flagged as nocompress. This flag can
418 * change at any time if we discover bad compression ratios.
420 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
421 (btrfs_test_opt(root, COMPRESS) ||
422 (BTRFS_I(inode)->force_compress) ||
423 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
425 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
427 /* just bail out to the uncompressed code */
431 if (BTRFS_I(inode)->force_compress)
432 compress_type = BTRFS_I(inode)->force_compress;
435 * we need to call clear_page_dirty_for_io on each
436 * page in the range. Otherwise applications with the file
437 * mmap'd can wander in and change the page contents while
438 * we are compressing them.
440 * If the compression fails for any reason, we set the pages
441 * dirty again later on.
443 extent_range_clear_dirty_for_io(inode, start, end);
445 ret = btrfs_compress_pages(compress_type,
446 inode->i_mapping, start,
447 total_compressed, pages,
448 nr_pages, &nr_pages_ret,
454 unsigned long offset = total_compressed &
455 (PAGE_CACHE_SIZE - 1);
456 struct page *page = pages[nr_pages_ret - 1];
459 /* zero the tail end of the last page, we might be
460 * sending it down to disk
463 kaddr = kmap_atomic(page);
464 memset(kaddr + offset, 0,
465 PAGE_CACHE_SIZE - offset);
466 kunmap_atomic(kaddr);
473 /* lets try to make an inline extent */
474 if (ret || total_in < (actual_end - start)) {
475 /* we didn't compress the entire range, try
476 * to make an uncompressed inline extent.
478 ret = cow_file_range_inline(root, inode, start, end,
481 /* try making a compressed inline extent */
482 ret = cow_file_range_inline(root, inode, start, end,
484 compress_type, pages);
487 unsigned long clear_flags = EXTENT_DELALLOC |
489 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
492 * inline extent creation worked or returned error,
493 * we don't need to create any more async work items.
494 * Unlock and free up our temp pages.
496 extent_clear_unlock_delalloc(inode, start, end, NULL,
497 clear_flags, PAGE_UNLOCK |
507 * we aren't doing an inline extent round the compressed size
508 * up to a block size boundary so the allocator does sane
511 total_compressed = ALIGN(total_compressed, blocksize);
514 * one last check to make sure the compression is really a
515 * win, compare the page count read with the blocks on disk
517 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
518 if (total_compressed >= total_in) {
521 num_bytes = total_in;
524 if (!will_compress && pages) {
526 * the compression code ran but failed to make things smaller,
527 * free any pages it allocated and our page pointer array
529 for (i = 0; i < nr_pages_ret; i++) {
530 WARN_ON(pages[i]->mapping);
531 page_cache_release(pages[i]);
535 total_compressed = 0;
538 /* flag the file so we don't compress in the future */
539 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
540 !(BTRFS_I(inode)->force_compress)) {
541 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
547 /* the async work queues will take care of doing actual
548 * allocation on disk for these compressed pages,
549 * and will submit them to the elevator.
551 add_async_extent(async_cow, start, num_bytes,
552 total_compressed, pages, nr_pages_ret,
555 if (start + num_bytes < end) {
562 cleanup_and_bail_uncompressed:
564 * No compression, but we still need to write the pages in
565 * the file we've been given so far. redirty the locked
566 * page if it corresponds to our extent and set things up
567 * for the async work queue to run cow_file_range to do
568 * the normal delalloc dance
570 if (page_offset(locked_page) >= start &&
571 page_offset(locked_page) <= end) {
572 __set_page_dirty_nobuffers(locked_page);
573 /* unlocked later on in the async handlers */
576 extent_range_redirty_for_io(inode, start, end);
577 add_async_extent(async_cow, start, end - start + 1,
578 0, NULL, 0, BTRFS_COMPRESS_NONE);
586 for (i = 0; i < nr_pages_ret; i++) {
587 WARN_ON(pages[i]->mapping);
588 page_cache_release(pages[i]);
596 * phase two of compressed writeback. This is the ordered portion
597 * of the code, which only gets called in the order the work was
598 * queued. We walk all the async extents created by compress_file_range
599 * and send them down to the disk.
601 static noinline int submit_compressed_extents(struct inode *inode,
602 struct async_cow *async_cow)
604 struct async_extent *async_extent;
606 struct btrfs_key ins;
607 struct extent_map *em;
608 struct btrfs_root *root = BTRFS_I(inode)->root;
609 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
610 struct extent_io_tree *io_tree;
613 if (list_empty(&async_cow->extents))
617 while (!list_empty(&async_cow->extents)) {
618 async_extent = list_entry(async_cow->extents.next,
619 struct async_extent, list);
620 list_del(&async_extent->list);
622 io_tree = &BTRFS_I(inode)->io_tree;
625 /* did the compression code fall back to uncompressed IO? */
626 if (!async_extent->pages) {
627 int page_started = 0;
628 unsigned long nr_written = 0;
630 lock_extent(io_tree, async_extent->start,
631 async_extent->start +
632 async_extent->ram_size - 1);
634 /* allocate blocks */
635 ret = cow_file_range(inode, async_cow->locked_page,
637 async_extent->start +
638 async_extent->ram_size - 1,
639 &page_started, &nr_written, 0);
644 * if page_started, cow_file_range inserted an
645 * inline extent and took care of all the unlocking
646 * and IO for us. Otherwise, we need to submit
647 * all those pages down to the drive.
649 if (!page_started && !ret)
650 extent_write_locked_range(io_tree,
651 inode, async_extent->start,
652 async_extent->start +
653 async_extent->ram_size - 1,
657 unlock_page(async_cow->locked_page);
663 lock_extent(io_tree, async_extent->start,
664 async_extent->start + async_extent->ram_size - 1);
666 ret = btrfs_reserve_extent(root,
667 async_extent->compressed_size,
668 async_extent->compressed_size,
669 0, alloc_hint, &ins, 1);
673 for (i = 0; i < async_extent->nr_pages; i++) {
674 WARN_ON(async_extent->pages[i]->mapping);
675 page_cache_release(async_extent->pages[i]);
677 kfree(async_extent->pages);
678 async_extent->nr_pages = 0;
679 async_extent->pages = NULL;
681 if (ret == -ENOSPC) {
682 unlock_extent(io_tree, async_extent->start,
683 async_extent->start +
684 async_extent->ram_size - 1);
691 * here we're doing allocation and writeback of the
694 btrfs_drop_extent_cache(inode, async_extent->start,
695 async_extent->start +
696 async_extent->ram_size - 1, 0);
698 em = alloc_extent_map();
701 goto out_free_reserve;
703 em->start = async_extent->start;
704 em->len = async_extent->ram_size;
705 em->orig_start = em->start;
706 em->mod_start = em->start;
707 em->mod_len = em->len;
709 em->block_start = ins.objectid;
710 em->block_len = ins.offset;
711 em->orig_block_len = ins.offset;
712 em->ram_bytes = async_extent->ram_size;
713 em->bdev = root->fs_info->fs_devices->latest_bdev;
714 em->compress_type = async_extent->compress_type;
715 set_bit(EXTENT_FLAG_PINNED, &em->flags);
716 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
720 write_lock(&em_tree->lock);
721 ret = add_extent_mapping(em_tree, em, 1);
722 write_unlock(&em_tree->lock);
723 if (ret != -EEXIST) {
727 btrfs_drop_extent_cache(inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1, 0);
733 goto out_free_reserve;
735 ret = btrfs_add_ordered_extent_compress(inode,
738 async_extent->ram_size,
740 BTRFS_ORDERED_COMPRESSED,
741 async_extent->compress_type);
743 goto out_free_reserve;
746 * clear dirty, set writeback and unlock the pages.
748 extent_clear_unlock_delalloc(inode, async_extent->start,
749 async_extent->start +
750 async_extent->ram_size - 1,
751 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
752 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
754 ret = btrfs_submit_compressed_write(inode,
756 async_extent->ram_size,
758 ins.offset, async_extent->pages,
759 async_extent->nr_pages);
760 alloc_hint = ins.objectid + ins.offset;
770 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
772 extent_clear_unlock_delalloc(inode, async_extent->start,
773 async_extent->start +
774 async_extent->ram_size - 1,
775 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
776 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
777 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
778 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
783 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
786 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
787 struct extent_map *em;
790 read_lock(&em_tree->lock);
791 em = search_extent_mapping(em_tree, start, num_bytes);
794 * if block start isn't an actual block number then find the
795 * first block in this inode and use that as a hint. If that
796 * block is also bogus then just don't worry about it.
798 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
800 em = search_extent_mapping(em_tree, 0, 0);
801 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
802 alloc_hint = em->block_start;
806 alloc_hint = em->block_start;
810 read_unlock(&em_tree->lock);
816 * when extent_io.c finds a delayed allocation range in the file,
817 * the call backs end up in this code. The basic idea is to
818 * allocate extents on disk for the range, and create ordered data structs
819 * in ram to track those extents.
821 * locked_page is the page that writepage had locked already. We use
822 * it to make sure we don't do extra locks or unlocks.
824 * *page_started is set to one if we unlock locked_page and do everything
825 * required to start IO on it. It may be clean and already done with
828 static noinline int cow_file_range(struct inode *inode,
829 struct page *locked_page,
830 u64 start, u64 end, int *page_started,
831 unsigned long *nr_written,
834 struct btrfs_root *root = BTRFS_I(inode)->root;
837 unsigned long ram_size;
840 u64 blocksize = root->sectorsize;
841 struct btrfs_key ins;
842 struct extent_map *em;
843 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
846 if (btrfs_is_free_space_inode(inode)) {
851 num_bytes = ALIGN(end - start + 1, blocksize);
852 num_bytes = max(blocksize, num_bytes);
853 disk_num_bytes = num_bytes;
855 /* if this is a small write inside eof, kick off defrag */
856 if (num_bytes < 64 * 1024 &&
857 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
858 btrfs_add_inode_defrag(NULL, inode);
861 /* lets try to make an inline extent */
862 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
865 extent_clear_unlock_delalloc(inode, start, end, NULL,
866 EXTENT_LOCKED | EXTENT_DELALLOC |
867 EXTENT_DEFRAG, PAGE_UNLOCK |
868 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
871 *nr_written = *nr_written +
872 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
875 } else if (ret < 0) {
880 BUG_ON(disk_num_bytes >
881 btrfs_super_total_bytes(root->fs_info->super_copy));
883 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
884 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
886 while (disk_num_bytes > 0) {
889 cur_alloc_size = disk_num_bytes;
890 ret = btrfs_reserve_extent(root, cur_alloc_size,
891 root->sectorsize, 0, alloc_hint,
896 em = alloc_extent_map();
902 em->orig_start = em->start;
903 ram_size = ins.offset;
904 em->len = ins.offset;
905 em->mod_start = em->start;
906 em->mod_len = em->len;
908 em->block_start = ins.objectid;
909 em->block_len = ins.offset;
910 em->orig_block_len = ins.offset;
911 em->ram_bytes = ram_size;
912 em->bdev = root->fs_info->fs_devices->latest_bdev;
913 set_bit(EXTENT_FLAG_PINNED, &em->flags);
917 write_lock(&em_tree->lock);
918 ret = add_extent_mapping(em_tree, em, 1);
919 write_unlock(&em_tree->lock);
920 if (ret != -EEXIST) {
924 btrfs_drop_extent_cache(inode, start,
925 start + ram_size - 1, 0);
930 cur_alloc_size = ins.offset;
931 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
932 ram_size, cur_alloc_size, 0);
936 if (root->root_key.objectid ==
937 BTRFS_DATA_RELOC_TREE_OBJECTID) {
938 ret = btrfs_reloc_clone_csums(inode, start,
944 if (disk_num_bytes < cur_alloc_size)
947 /* we're not doing compressed IO, don't unlock the first
948 * page (which the caller expects to stay locked), don't
949 * clear any dirty bits and don't set any writeback bits
951 * Do set the Private2 bit so we know this page was properly
952 * setup for writepage
954 op = unlock ? PAGE_UNLOCK : 0;
955 op |= PAGE_SET_PRIVATE2;
957 extent_clear_unlock_delalloc(inode, start,
958 start + ram_size - 1, locked_page,
959 EXTENT_LOCKED | EXTENT_DELALLOC,
961 disk_num_bytes -= cur_alloc_size;
962 num_bytes -= cur_alloc_size;
963 alloc_hint = ins.objectid + ins.offset;
964 start += cur_alloc_size;
970 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
972 extent_clear_unlock_delalloc(inode, start, end, locked_page,
973 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
974 EXTENT_DELALLOC | EXTENT_DEFRAG,
975 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
976 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
981 * work queue call back to started compression on a file and pages
983 static noinline void async_cow_start(struct btrfs_work *work)
985 struct async_cow *async_cow;
987 async_cow = container_of(work, struct async_cow, work);
989 compress_file_range(async_cow->inode, async_cow->locked_page,
990 async_cow->start, async_cow->end, async_cow,
992 if (num_added == 0) {
993 btrfs_add_delayed_iput(async_cow->inode);
994 async_cow->inode = NULL;
999 * work queue call back to submit previously compressed pages
1001 static noinline void async_cow_submit(struct btrfs_work *work)
1003 struct async_cow *async_cow;
1004 struct btrfs_root *root;
1005 unsigned long nr_pages;
1007 async_cow = container_of(work, struct async_cow, work);
1009 root = async_cow->root;
1010 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1013 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1015 waitqueue_active(&root->fs_info->async_submit_wait))
1016 wake_up(&root->fs_info->async_submit_wait);
1018 if (async_cow->inode)
1019 submit_compressed_extents(async_cow->inode, async_cow);
1022 static noinline void async_cow_free(struct btrfs_work *work)
1024 struct async_cow *async_cow;
1025 async_cow = container_of(work, struct async_cow, work);
1026 if (async_cow->inode)
1027 btrfs_add_delayed_iput(async_cow->inode);
1031 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1032 u64 start, u64 end, int *page_started,
1033 unsigned long *nr_written)
1035 struct async_cow *async_cow;
1036 struct btrfs_root *root = BTRFS_I(inode)->root;
1037 unsigned long nr_pages;
1039 int limit = 10 * 1024 * 1024;
1041 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1042 1, 0, NULL, GFP_NOFS);
1043 while (start < end) {
1044 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1045 BUG_ON(!async_cow); /* -ENOMEM */
1046 async_cow->inode = igrab(inode);
1047 async_cow->root = root;
1048 async_cow->locked_page = locked_page;
1049 async_cow->start = start;
1051 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1054 cur_end = min(end, start + 512 * 1024 - 1);
1056 async_cow->end = cur_end;
1057 INIT_LIST_HEAD(&async_cow->extents);
1059 async_cow->work.func = async_cow_start;
1060 async_cow->work.ordered_func = async_cow_submit;
1061 async_cow->work.ordered_free = async_cow_free;
1062 async_cow->work.flags = 0;
1064 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1066 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1068 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1071 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1072 wait_event(root->fs_info->async_submit_wait,
1073 (atomic_read(&root->fs_info->async_delalloc_pages) <
1077 while (atomic_read(&root->fs_info->async_submit_draining) &&
1078 atomic_read(&root->fs_info->async_delalloc_pages)) {
1079 wait_event(root->fs_info->async_submit_wait,
1080 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1084 *nr_written += nr_pages;
1085 start = cur_end + 1;
1091 static noinline int csum_exist_in_range(struct btrfs_root *root,
1092 u64 bytenr, u64 num_bytes)
1095 struct btrfs_ordered_sum *sums;
1098 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1099 bytenr + num_bytes - 1, &list, 0);
1100 if (ret == 0 && list_empty(&list))
1103 while (!list_empty(&list)) {
1104 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1105 list_del(&sums->list);
1112 * when nowcow writeback call back. This checks for snapshots or COW copies
1113 * of the extents that exist in the file, and COWs the file as required.
1115 * If no cow copies or snapshots exist, we write directly to the existing
1118 static noinline int run_delalloc_nocow(struct inode *inode,
1119 struct page *locked_page,
1120 u64 start, u64 end, int *page_started, int force,
1121 unsigned long *nr_written)
1123 struct btrfs_root *root = BTRFS_I(inode)->root;
1124 struct btrfs_trans_handle *trans;
1125 struct extent_buffer *leaf;
1126 struct btrfs_path *path;
1127 struct btrfs_file_extent_item *fi;
1128 struct btrfs_key found_key;
1143 u64 ino = btrfs_ino(inode);
1145 path = btrfs_alloc_path();
1147 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1148 EXTENT_LOCKED | EXTENT_DELALLOC |
1149 EXTENT_DO_ACCOUNTING |
1150 EXTENT_DEFRAG, PAGE_UNLOCK |
1152 PAGE_SET_WRITEBACK |
1153 PAGE_END_WRITEBACK);
1157 nolock = btrfs_is_free_space_inode(inode);
1160 trans = btrfs_join_transaction_nolock(root);
1162 trans = btrfs_join_transaction(root);
1164 if (IS_ERR(trans)) {
1165 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1166 EXTENT_LOCKED | EXTENT_DELALLOC |
1167 EXTENT_DO_ACCOUNTING |
1168 EXTENT_DEFRAG, PAGE_UNLOCK |
1170 PAGE_SET_WRITEBACK |
1171 PAGE_END_WRITEBACK);
1172 btrfs_free_path(path);
1173 return PTR_ERR(trans);
1176 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1178 cow_start = (u64)-1;
1181 ret = btrfs_lookup_file_extent(trans, root, path, ino,
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);
1202 leaf = path->nodes[0];
1208 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1210 if (found_key.objectid > ino ||
1211 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1212 found_key.offset > end)
1215 if (found_key.offset > cur_offset) {
1216 extent_end = found_key.offset;
1221 fi = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_file_extent_item);
1223 extent_type = btrfs_file_extent_type(leaf, fi);
1225 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1226 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1227 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1228 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1229 extent_offset = btrfs_file_extent_offset(leaf, fi);
1230 extent_end = found_key.offset +
1231 btrfs_file_extent_num_bytes(leaf, fi);
1233 btrfs_file_extent_disk_num_bytes(leaf, fi);
1234 if (extent_end <= start) {
1238 if (disk_bytenr == 0)
1240 if (btrfs_file_extent_compression(leaf, fi) ||
1241 btrfs_file_extent_encryption(leaf, fi) ||
1242 btrfs_file_extent_other_encoding(leaf, fi))
1244 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1246 if (btrfs_extent_readonly(root, disk_bytenr))
1248 if (btrfs_cross_ref_exist(trans, root, ino,
1250 extent_offset, disk_bytenr))
1252 disk_bytenr += extent_offset;
1253 disk_bytenr += cur_offset - found_key.offset;
1254 num_bytes = min(end + 1, extent_end) - cur_offset;
1256 * force cow if csum exists in the range.
1257 * this ensure that csum for a given extent are
1258 * either valid or do not exist.
1260 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1263 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1264 extent_end = found_key.offset +
1265 btrfs_file_extent_inline_len(leaf, fi);
1266 extent_end = ALIGN(extent_end, root->sectorsize);
1271 if (extent_end <= start) {
1276 if (cow_start == (u64)-1)
1277 cow_start = cur_offset;
1278 cur_offset = extent_end;
1279 if (cur_offset > end)
1285 btrfs_release_path(path);
1286 if (cow_start != (u64)-1) {
1287 ret = cow_file_range(inode, locked_page,
1288 cow_start, found_key.offset - 1,
1289 page_started, nr_written, 1);
1292 cow_start = (u64)-1;
1295 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1296 struct extent_map *em;
1297 struct extent_map_tree *em_tree;
1298 em_tree = &BTRFS_I(inode)->extent_tree;
1299 em = alloc_extent_map();
1300 BUG_ON(!em); /* -ENOMEM */
1301 em->start = cur_offset;
1302 em->orig_start = found_key.offset - extent_offset;
1303 em->len = num_bytes;
1304 em->block_len = num_bytes;
1305 em->block_start = disk_bytenr;
1306 em->orig_block_len = disk_num_bytes;
1307 em->ram_bytes = ram_bytes;
1308 em->bdev = root->fs_info->fs_devices->latest_bdev;
1309 em->mod_start = em->start;
1310 em->mod_len = em->len;
1311 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1312 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1313 em->generation = -1;
1315 write_lock(&em_tree->lock);
1316 ret = add_extent_mapping(em_tree, em, 1);
1317 write_unlock(&em_tree->lock);
1318 if (ret != -EEXIST) {
1319 free_extent_map(em);
1322 btrfs_drop_extent_cache(inode, em->start,
1323 em->start + em->len - 1, 0);
1325 type = BTRFS_ORDERED_PREALLOC;
1327 type = BTRFS_ORDERED_NOCOW;
1330 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1331 num_bytes, num_bytes, type);
1332 BUG_ON(ret); /* -ENOMEM */
1334 if (root->root_key.objectid ==
1335 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1336 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1342 extent_clear_unlock_delalloc(inode, cur_offset,
1343 cur_offset + num_bytes - 1,
1344 locked_page, EXTENT_LOCKED |
1345 EXTENT_DELALLOC, PAGE_UNLOCK |
1347 cur_offset = extent_end;
1348 if (cur_offset > end)
1351 btrfs_release_path(path);
1353 if (cur_offset <= end && cow_start == (u64)-1) {
1354 cow_start = cur_offset;
1358 if (cow_start != (u64)-1) {
1359 ret = cow_file_range(inode, locked_page, cow_start, end,
1360 page_started, nr_written, 1);
1366 err = btrfs_end_transaction(trans, root);
1370 if (ret && cur_offset < end)
1371 extent_clear_unlock_delalloc(inode, cur_offset, end,
1372 locked_page, EXTENT_LOCKED |
1373 EXTENT_DELALLOC | EXTENT_DEFRAG |
1374 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1376 PAGE_SET_WRITEBACK |
1377 PAGE_END_WRITEBACK);
1378 btrfs_free_path(path);
1383 * extent_io.c call back to do delayed allocation processing
1385 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1386 u64 start, u64 end, int *page_started,
1387 unsigned long *nr_written)
1390 struct btrfs_root *root = BTRFS_I(inode)->root;
1392 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1393 ret = run_delalloc_nocow(inode, locked_page, start, end,
1394 page_started, 1, nr_written);
1395 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1396 ret = run_delalloc_nocow(inode, locked_page, start, end,
1397 page_started, 0, nr_written);
1398 } else if (!btrfs_test_opt(root, COMPRESS) &&
1399 !(BTRFS_I(inode)->force_compress) &&
1400 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1401 ret = cow_file_range(inode, locked_page, start, end,
1402 page_started, nr_written, 1);
1404 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1405 &BTRFS_I(inode)->runtime_flags);
1406 ret = cow_file_range_async(inode, locked_page, start, end,
1407 page_started, nr_written);
1412 static void btrfs_split_extent_hook(struct inode *inode,
1413 struct extent_state *orig, u64 split)
1415 /* not delalloc, ignore it */
1416 if (!(orig->state & EXTENT_DELALLOC))
1419 spin_lock(&BTRFS_I(inode)->lock);
1420 BTRFS_I(inode)->outstanding_extents++;
1421 spin_unlock(&BTRFS_I(inode)->lock);
1425 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1426 * extents so we can keep track of new extents that are just merged onto old
1427 * extents, such as when we are doing sequential writes, so we can properly
1428 * account for the metadata space we'll need.
1430 static void btrfs_merge_extent_hook(struct inode *inode,
1431 struct extent_state *new,
1432 struct extent_state *other)
1434 /* not delalloc, ignore it */
1435 if (!(other->state & EXTENT_DELALLOC))
1438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents--;
1440 spin_unlock(&BTRFS_I(inode)->lock);
1443 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1444 struct inode *inode)
1446 spin_lock(&root->delalloc_lock);
1447 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1448 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1449 &root->delalloc_inodes);
1450 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1451 &BTRFS_I(inode)->runtime_flags);
1452 root->nr_delalloc_inodes++;
1453 if (root->nr_delalloc_inodes == 1) {
1454 spin_lock(&root->fs_info->delalloc_root_lock);
1455 BUG_ON(!list_empty(&root->delalloc_root));
1456 list_add_tail(&root->delalloc_root,
1457 &root->fs_info->delalloc_roots);
1458 spin_unlock(&root->fs_info->delalloc_root_lock);
1461 spin_unlock(&root->delalloc_lock);
1464 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1465 struct inode *inode)
1467 spin_lock(&root->delalloc_lock);
1468 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1469 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1470 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1471 &BTRFS_I(inode)->runtime_flags);
1472 root->nr_delalloc_inodes--;
1473 if (!root->nr_delalloc_inodes) {
1474 spin_lock(&root->fs_info->delalloc_root_lock);
1475 BUG_ON(list_empty(&root->delalloc_root));
1476 list_del_init(&root->delalloc_root);
1477 spin_unlock(&root->fs_info->delalloc_root_lock);
1480 spin_unlock(&root->delalloc_lock);
1484 * extent_io.c set_bit_hook, used to track delayed allocation
1485 * bytes in this file, and to maintain the list of inodes that
1486 * have pending delalloc work to be done.
1488 static void btrfs_set_bit_hook(struct inode *inode,
1489 struct extent_state *state, unsigned long *bits)
1493 * set_bit and clear bit hooks normally require _irqsave/restore
1494 * but in this case, we are only testing for the DELALLOC
1495 * bit, which is only set or cleared with irqs on
1497 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1498 struct btrfs_root *root = BTRFS_I(inode)->root;
1499 u64 len = state->end + 1 - state->start;
1500 bool do_list = !btrfs_is_free_space_inode(inode);
1502 if (*bits & EXTENT_FIRST_DELALLOC) {
1503 *bits &= ~EXTENT_FIRST_DELALLOC;
1505 spin_lock(&BTRFS_I(inode)->lock);
1506 BTRFS_I(inode)->outstanding_extents++;
1507 spin_unlock(&BTRFS_I(inode)->lock);
1510 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1511 root->fs_info->delalloc_batch);
1512 spin_lock(&BTRFS_I(inode)->lock);
1513 BTRFS_I(inode)->delalloc_bytes += len;
1514 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1515 &BTRFS_I(inode)->runtime_flags))
1516 btrfs_add_delalloc_inodes(root, inode);
1517 spin_unlock(&BTRFS_I(inode)->lock);
1522 * extent_io.c clear_bit_hook, see set_bit_hook for why
1524 static void btrfs_clear_bit_hook(struct inode *inode,
1525 struct extent_state *state,
1526 unsigned long *bits)
1529 * set_bit and clear bit hooks normally require _irqsave/restore
1530 * but in this case, we are only testing for the DELALLOC
1531 * bit, which is only set or cleared with irqs on
1533 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1534 struct btrfs_root *root = BTRFS_I(inode)->root;
1535 u64 len = state->end + 1 - state->start;
1536 bool do_list = !btrfs_is_free_space_inode(inode);
1538 if (*bits & EXTENT_FIRST_DELALLOC) {
1539 *bits &= ~EXTENT_FIRST_DELALLOC;
1540 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541 spin_lock(&BTRFS_I(inode)->lock);
1542 BTRFS_I(inode)->outstanding_extents--;
1543 spin_unlock(&BTRFS_I(inode)->lock);
1547 * We don't reserve metadata space for space cache inodes so we
1548 * don't need to call dellalloc_release_metadata if there is an
1551 if (*bits & EXTENT_DO_ACCOUNTING &&
1552 root != root->fs_info->tree_root)
1553 btrfs_delalloc_release_metadata(inode, len);
1555 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1556 && do_list && !(state->state & EXTENT_NORESERVE))
1557 btrfs_free_reserved_data_space(inode, len);
1559 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1560 root->fs_info->delalloc_batch);
1561 spin_lock(&BTRFS_I(inode)->lock);
1562 BTRFS_I(inode)->delalloc_bytes -= len;
1563 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1564 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1565 &BTRFS_I(inode)->runtime_flags))
1566 btrfs_del_delalloc_inode(root, inode);
1567 spin_unlock(&BTRFS_I(inode)->lock);
1572 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1573 * we don't create bios that span stripes or chunks
1575 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1576 size_t size, struct bio *bio,
1577 unsigned long bio_flags)
1579 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1580 u64 logical = (u64)bio->bi_sector << 9;
1585 if (bio_flags & EXTENT_BIO_COMPRESSED)
1588 length = bio->bi_size;
1589 map_length = length;
1590 ret = btrfs_map_block(root->fs_info, rw, logical,
1591 &map_length, NULL, 0);
1592 /* Will always return 0 with map_multi == NULL */
1594 if (map_length < length + size)
1600 * in order to insert checksums into the metadata in large chunks,
1601 * we wait until bio submission time. All the pages in the bio are
1602 * checksummed and sums are attached onto the ordered extent record.
1604 * At IO completion time the cums attached on the ordered extent record
1605 * are inserted into the btree
1607 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1608 struct bio *bio, int mirror_num,
1609 unsigned long bio_flags,
1612 struct btrfs_root *root = BTRFS_I(inode)->root;
1615 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1616 BUG_ON(ret); /* -ENOMEM */
1621 * in order to insert checksums into the metadata in large chunks,
1622 * we wait until bio submission time. All the pages in the bio are
1623 * checksummed and sums are attached onto the ordered extent record.
1625 * At IO completion time the cums attached on the ordered extent record
1626 * are inserted into the btree
1628 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1629 int mirror_num, unsigned long bio_flags,
1632 struct btrfs_root *root = BTRFS_I(inode)->root;
1635 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1637 bio_endio(bio, ret);
1642 * extent_io.c submission hook. This does the right thing for csum calculation
1643 * on write, or reading the csums from the tree before a read
1645 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1646 int mirror_num, unsigned long bio_flags,
1649 struct btrfs_root *root = BTRFS_I(inode)->root;
1653 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1655 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1657 if (btrfs_is_free_space_inode(inode))
1660 if (!(rw & REQ_WRITE)) {
1661 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1665 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1666 ret = btrfs_submit_compressed_read(inode, bio,
1670 } else if (!skip_sum) {
1671 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1676 } else if (async && !skip_sum) {
1677 /* csum items have already been cloned */
1678 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1680 /* we're doing a write, do the async checksumming */
1681 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1682 inode, rw, bio, mirror_num,
1683 bio_flags, bio_offset,
1684 __btrfs_submit_bio_start,
1685 __btrfs_submit_bio_done);
1687 } else if (!skip_sum) {
1688 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1694 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1698 bio_endio(bio, ret);
1703 * given a list of ordered sums record them in the inode. This happens
1704 * at IO completion time based on sums calculated at bio submission time.
1706 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1707 struct inode *inode, u64 file_offset,
1708 struct list_head *list)
1710 struct btrfs_ordered_sum *sum;
1712 list_for_each_entry(sum, list, list) {
1713 trans->adding_csums = 1;
1714 btrfs_csum_file_blocks(trans,
1715 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1716 trans->adding_csums = 0;
1721 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1722 struct extent_state **cached_state)
1724 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1725 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1726 cached_state, GFP_NOFS);
1729 /* see btrfs_writepage_start_hook for details on why this is required */
1730 struct btrfs_writepage_fixup {
1732 struct btrfs_work work;
1735 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1737 struct btrfs_writepage_fixup *fixup;
1738 struct btrfs_ordered_extent *ordered;
1739 struct extent_state *cached_state = NULL;
1741 struct inode *inode;
1746 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1750 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1751 ClearPageChecked(page);
1755 inode = page->mapping->host;
1756 page_start = page_offset(page);
1757 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1759 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1762 /* already ordered? We're done */
1763 if (PagePrivate2(page))
1766 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1768 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1769 page_end, &cached_state, GFP_NOFS);
1771 btrfs_start_ordered_extent(inode, ordered, 1);
1772 btrfs_put_ordered_extent(ordered);
1776 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1778 mapping_set_error(page->mapping, ret);
1779 end_extent_writepage(page, ret, page_start, page_end);
1780 ClearPageChecked(page);
1784 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1785 ClearPageChecked(page);
1786 set_page_dirty(page);
1788 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1789 &cached_state, GFP_NOFS);
1792 page_cache_release(page);
1797 * There are a few paths in the higher layers of the kernel that directly
1798 * set the page dirty bit without asking the filesystem if it is a
1799 * good idea. This causes problems because we want to make sure COW
1800 * properly happens and the data=ordered rules are followed.
1802 * In our case any range that doesn't have the ORDERED bit set
1803 * hasn't been properly setup for IO. We kick off an async process
1804 * to fix it up. The async helper will wait for ordered extents, set
1805 * the delalloc bit and make it safe to write the page.
1807 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1809 struct inode *inode = page->mapping->host;
1810 struct btrfs_writepage_fixup *fixup;
1811 struct btrfs_root *root = BTRFS_I(inode)->root;
1813 /* this page is properly in the ordered list */
1814 if (TestClearPagePrivate2(page))
1817 if (PageChecked(page))
1820 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1824 SetPageChecked(page);
1825 page_cache_get(page);
1826 fixup->work.func = btrfs_writepage_fixup_worker;
1828 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1832 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1833 struct inode *inode, u64 file_pos,
1834 u64 disk_bytenr, u64 disk_num_bytes,
1835 u64 num_bytes, u64 ram_bytes,
1836 u8 compression, u8 encryption,
1837 u16 other_encoding, int extent_type)
1839 struct btrfs_root *root = BTRFS_I(inode)->root;
1840 struct btrfs_file_extent_item *fi;
1841 struct btrfs_path *path;
1842 struct extent_buffer *leaf;
1843 struct btrfs_key ins;
1846 path = btrfs_alloc_path();
1850 path->leave_spinning = 1;
1853 * we may be replacing one extent in the tree with another.
1854 * The new extent is pinned in the extent map, and we don't want
1855 * to drop it from the cache until it is completely in the btree.
1857 * So, tell btrfs_drop_extents to leave this extent in the cache.
1858 * the caller is expected to unpin it and allow it to be merged
1861 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1862 file_pos + num_bytes, 0);
1866 ins.objectid = btrfs_ino(inode);
1867 ins.offset = file_pos;
1868 ins.type = BTRFS_EXTENT_DATA_KEY;
1869 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1872 leaf = path->nodes[0];
1873 fi = btrfs_item_ptr(leaf, path->slots[0],
1874 struct btrfs_file_extent_item);
1875 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1876 btrfs_set_file_extent_type(leaf, fi, extent_type);
1877 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1878 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1879 btrfs_set_file_extent_offset(leaf, fi, 0);
1880 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1881 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1882 btrfs_set_file_extent_compression(leaf, fi, compression);
1883 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1884 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1886 btrfs_mark_buffer_dirty(leaf);
1887 btrfs_release_path(path);
1889 inode_add_bytes(inode, num_bytes);
1891 ins.objectid = disk_bytenr;
1892 ins.offset = disk_num_bytes;
1893 ins.type = BTRFS_EXTENT_ITEM_KEY;
1894 ret = btrfs_alloc_reserved_file_extent(trans, root,
1895 root->root_key.objectid,
1896 btrfs_ino(inode), file_pos, &ins);
1898 btrfs_free_path(path);
1903 /* snapshot-aware defrag */
1904 struct sa_defrag_extent_backref {
1905 struct rb_node node;
1906 struct old_sa_defrag_extent *old;
1915 struct old_sa_defrag_extent {
1916 struct list_head list;
1917 struct new_sa_defrag_extent *new;
1926 struct new_sa_defrag_extent {
1927 struct rb_root root;
1928 struct list_head head;
1929 struct btrfs_path *path;
1930 struct inode *inode;
1938 static int backref_comp(struct sa_defrag_extent_backref *b1,
1939 struct sa_defrag_extent_backref *b2)
1941 if (b1->root_id < b2->root_id)
1943 else if (b1->root_id > b2->root_id)
1946 if (b1->inum < b2->inum)
1948 else if (b1->inum > b2->inum)
1951 if (b1->file_pos < b2->file_pos)
1953 else if (b1->file_pos > b2->file_pos)
1957 * [------------------------------] ===> (a range of space)
1958 * |<--->| |<---->| =============> (fs/file tree A)
1959 * |<---------------------------->| ===> (fs/file tree B)
1961 * A range of space can refer to two file extents in one tree while
1962 * refer to only one file extent in another tree.
1964 * So we may process a disk offset more than one time(two extents in A)
1965 * and locate at the same extent(one extent in B), then insert two same
1966 * backrefs(both refer to the extent in B).
1971 static void backref_insert(struct rb_root *root,
1972 struct sa_defrag_extent_backref *backref)
1974 struct rb_node **p = &root->rb_node;
1975 struct rb_node *parent = NULL;
1976 struct sa_defrag_extent_backref *entry;
1981 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1983 ret = backref_comp(backref, entry);
1987 p = &(*p)->rb_right;
1990 rb_link_node(&backref->node, parent, p);
1991 rb_insert_color(&backref->node, root);
1995 * Note the backref might has changed, and in this case we just return 0.
1997 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2000 struct btrfs_file_extent_item *extent;
2001 struct btrfs_fs_info *fs_info;
2002 struct old_sa_defrag_extent *old = ctx;
2003 struct new_sa_defrag_extent *new = old->new;
2004 struct btrfs_path *path = new->path;
2005 struct btrfs_key key;
2006 struct btrfs_root *root;
2007 struct sa_defrag_extent_backref *backref;
2008 struct extent_buffer *leaf;
2009 struct inode *inode = new->inode;
2015 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2016 inum == btrfs_ino(inode))
2019 key.objectid = root_id;
2020 key.type = BTRFS_ROOT_ITEM_KEY;
2021 key.offset = (u64)-1;
2023 fs_info = BTRFS_I(inode)->root->fs_info;
2024 root = btrfs_read_fs_root_no_name(fs_info, &key);
2026 if (PTR_ERR(root) == -ENOENT)
2029 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2030 inum, offset, root_id);
2031 return PTR_ERR(root);
2034 key.objectid = inum;
2035 key.type = BTRFS_EXTENT_DATA_KEY;
2036 if (offset > (u64)-1 << 32)
2039 key.offset = offset;
2041 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2042 if (WARN_ON(ret < 0))
2049 leaf = path->nodes[0];
2050 slot = path->slots[0];
2052 if (slot >= btrfs_header_nritems(leaf)) {
2053 ret = btrfs_next_leaf(root, path);
2056 } else if (ret > 0) {
2065 btrfs_item_key_to_cpu(leaf, &key, slot);
2067 if (key.objectid > inum)
2070 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2073 extent = btrfs_item_ptr(leaf, slot,
2074 struct btrfs_file_extent_item);
2076 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2080 * 'offset' refers to the exact key.offset,
2081 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2082 * (key.offset - extent_offset).
2084 if (key.offset != offset)
2087 extent_offset = btrfs_file_extent_offset(leaf, extent);
2088 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2090 if (extent_offset >= old->extent_offset + old->offset +
2091 old->len || extent_offset + num_bytes <=
2092 old->extent_offset + old->offset)
2097 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2103 backref->root_id = root_id;
2104 backref->inum = inum;
2105 backref->file_pos = offset;
2106 backref->num_bytes = num_bytes;
2107 backref->extent_offset = extent_offset;
2108 backref->generation = btrfs_file_extent_generation(leaf, extent);
2110 backref_insert(&new->root, backref);
2113 btrfs_release_path(path);
2118 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2119 struct new_sa_defrag_extent *new)
2121 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2122 struct old_sa_defrag_extent *old, *tmp;
2127 list_for_each_entry_safe(old, tmp, &new->head, list) {
2128 ret = iterate_inodes_from_logical(old->bytenr +
2129 old->extent_offset, fs_info,
2130 path, record_one_backref,
2132 BUG_ON(ret < 0 && ret != -ENOENT);
2134 /* no backref to be processed for this extent */
2136 list_del(&old->list);
2141 if (list_empty(&new->head))
2147 static int relink_is_mergable(struct extent_buffer *leaf,
2148 struct btrfs_file_extent_item *fi,
2149 struct new_sa_defrag_extent *new)
2151 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2154 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2157 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2160 if (btrfs_file_extent_encryption(leaf, fi) ||
2161 btrfs_file_extent_other_encoding(leaf, fi))
2168 * Note the backref might has changed, and in this case we just return 0.
2170 static noinline int relink_extent_backref(struct btrfs_path *path,
2171 struct sa_defrag_extent_backref *prev,
2172 struct sa_defrag_extent_backref *backref)
2174 struct btrfs_file_extent_item *extent;
2175 struct btrfs_file_extent_item *item;
2176 struct btrfs_ordered_extent *ordered;
2177 struct btrfs_trans_handle *trans;
2178 struct btrfs_fs_info *fs_info;
2179 struct btrfs_root *root;
2180 struct btrfs_key key;
2181 struct extent_buffer *leaf;
2182 struct old_sa_defrag_extent *old = backref->old;
2183 struct new_sa_defrag_extent *new = old->new;
2184 struct inode *src_inode = new->inode;
2185 struct inode *inode;
2186 struct extent_state *cached = NULL;
2195 if (prev && prev->root_id == backref->root_id &&
2196 prev->inum == backref->inum &&
2197 prev->file_pos + prev->num_bytes == backref->file_pos)
2200 /* step 1: get root */
2201 key.objectid = backref->root_id;
2202 key.type = BTRFS_ROOT_ITEM_KEY;
2203 key.offset = (u64)-1;
2205 fs_info = BTRFS_I(src_inode)->root->fs_info;
2206 index = srcu_read_lock(&fs_info->subvol_srcu);
2208 root = btrfs_read_fs_root_no_name(fs_info, &key);
2210 srcu_read_unlock(&fs_info->subvol_srcu, index);
2211 if (PTR_ERR(root) == -ENOENT)
2213 return PTR_ERR(root);
2216 /* step 2: get inode */
2217 key.objectid = backref->inum;
2218 key.type = BTRFS_INODE_ITEM_KEY;
2221 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2222 if (IS_ERR(inode)) {
2223 srcu_read_unlock(&fs_info->subvol_srcu, index);
2227 srcu_read_unlock(&fs_info->subvol_srcu, index);
2229 /* step 3: relink backref */
2230 lock_start = backref->file_pos;
2231 lock_end = backref->file_pos + backref->num_bytes - 1;
2232 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2235 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2237 btrfs_put_ordered_extent(ordered);
2241 trans = btrfs_join_transaction(root);
2242 if (IS_ERR(trans)) {
2243 ret = PTR_ERR(trans);
2247 key.objectid = backref->inum;
2248 key.type = BTRFS_EXTENT_DATA_KEY;
2249 key.offset = backref->file_pos;
2251 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2254 } else if (ret > 0) {
2259 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2260 struct btrfs_file_extent_item);
2262 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2263 backref->generation)
2266 btrfs_release_path(path);
2268 start = backref->file_pos;
2269 if (backref->extent_offset < old->extent_offset + old->offset)
2270 start += old->extent_offset + old->offset -
2271 backref->extent_offset;
2273 len = min(backref->extent_offset + backref->num_bytes,
2274 old->extent_offset + old->offset + old->len);
2275 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2277 ret = btrfs_drop_extents(trans, root, inode, start,
2282 key.objectid = btrfs_ino(inode);
2283 key.type = BTRFS_EXTENT_DATA_KEY;
2286 path->leave_spinning = 1;
2288 struct btrfs_file_extent_item *fi;
2290 struct btrfs_key found_key;
2292 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2297 leaf = path->nodes[0];
2298 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2300 fi = btrfs_item_ptr(leaf, path->slots[0],
2301 struct btrfs_file_extent_item);
2302 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2304 if (extent_len + found_key.offset == start &&
2305 relink_is_mergable(leaf, fi, new)) {
2306 btrfs_set_file_extent_num_bytes(leaf, fi,
2308 btrfs_mark_buffer_dirty(leaf);
2309 inode_add_bytes(inode, len);
2315 btrfs_release_path(path);
2320 ret = btrfs_insert_empty_item(trans, root, path, &key,
2323 btrfs_abort_transaction(trans, root, ret);
2327 leaf = path->nodes[0];
2328 item = btrfs_item_ptr(leaf, path->slots[0],
2329 struct btrfs_file_extent_item);
2330 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2331 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2332 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2333 btrfs_set_file_extent_num_bytes(leaf, item, len);
2334 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2335 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2336 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2337 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2338 btrfs_set_file_extent_encryption(leaf, item, 0);
2339 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2341 btrfs_mark_buffer_dirty(leaf);
2342 inode_add_bytes(inode, len);
2343 btrfs_release_path(path);
2345 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2347 backref->root_id, backref->inum,
2348 new->file_pos, 0); /* start - extent_offset */
2350 btrfs_abort_transaction(trans, root, ret);
2356 btrfs_release_path(path);
2357 path->leave_spinning = 0;
2358 btrfs_end_transaction(trans, root);
2360 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2366 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2368 struct old_sa_defrag_extent *old, *tmp;
2373 list_for_each_entry_safe(old, tmp, &new->head, list) {
2374 list_del(&old->list);
2380 static void relink_file_extents(struct new_sa_defrag_extent *new)
2382 struct btrfs_path *path;
2383 struct sa_defrag_extent_backref *backref;
2384 struct sa_defrag_extent_backref *prev = NULL;
2385 struct inode *inode;
2386 struct btrfs_root *root;
2387 struct rb_node *node;
2391 root = BTRFS_I(inode)->root;
2393 path = btrfs_alloc_path();
2397 if (!record_extent_backrefs(path, new)) {
2398 btrfs_free_path(path);
2401 btrfs_release_path(path);
2404 node = rb_first(&new->root);
2407 rb_erase(node, &new->root);
2409 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2411 ret = relink_extent_backref(path, prev, backref);
2424 btrfs_free_path(path);
2426 free_sa_defrag_extent(new);
2428 atomic_dec(&root->fs_info->defrag_running);
2429 wake_up(&root->fs_info->transaction_wait);
2432 static struct new_sa_defrag_extent *
2433 record_old_file_extents(struct inode *inode,
2434 struct btrfs_ordered_extent *ordered)
2436 struct btrfs_root *root = BTRFS_I(inode)->root;
2437 struct btrfs_path *path;
2438 struct btrfs_key key;
2439 struct old_sa_defrag_extent *old;
2440 struct new_sa_defrag_extent *new;
2443 new = kmalloc(sizeof(*new), GFP_NOFS);
2448 new->file_pos = ordered->file_offset;
2449 new->len = ordered->len;
2450 new->bytenr = ordered->start;
2451 new->disk_len = ordered->disk_len;
2452 new->compress_type = ordered->compress_type;
2453 new->root = RB_ROOT;
2454 INIT_LIST_HEAD(&new->head);
2456 path = btrfs_alloc_path();
2460 key.objectid = btrfs_ino(inode);
2461 key.type = BTRFS_EXTENT_DATA_KEY;
2462 key.offset = new->file_pos;
2464 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2467 if (ret > 0 && path->slots[0] > 0)
2470 /* find out all the old extents for the file range */
2472 struct btrfs_file_extent_item *extent;
2473 struct extent_buffer *l;
2482 slot = path->slots[0];
2484 if (slot >= btrfs_header_nritems(l)) {
2485 ret = btrfs_next_leaf(root, path);
2493 btrfs_item_key_to_cpu(l, &key, slot);
2495 if (key.objectid != btrfs_ino(inode))
2497 if (key.type != BTRFS_EXTENT_DATA_KEY)
2499 if (key.offset >= new->file_pos + new->len)
2502 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2504 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2505 if (key.offset + num_bytes < new->file_pos)
2508 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2512 extent_offset = btrfs_file_extent_offset(l, extent);
2514 old = kmalloc(sizeof(*old), GFP_NOFS);
2518 offset = max(new->file_pos, key.offset);
2519 end = min(new->file_pos + new->len, key.offset + num_bytes);
2521 old->bytenr = disk_bytenr;
2522 old->extent_offset = extent_offset;
2523 old->offset = offset - key.offset;
2524 old->len = end - offset;
2527 list_add_tail(&old->list, &new->head);
2533 btrfs_free_path(path);
2534 atomic_inc(&root->fs_info->defrag_running);
2539 btrfs_free_path(path);
2541 free_sa_defrag_extent(new);
2546 * helper function for btrfs_finish_ordered_io, this
2547 * just reads in some of the csum leaves to prime them into ram
2548 * before we start the transaction. It limits the amount of btree
2549 * reads required while inside the transaction.
2551 /* as ordered data IO finishes, this gets called so we can finish
2552 * an ordered extent if the range of bytes in the file it covers are
2555 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2557 struct inode *inode = ordered_extent->inode;
2558 struct btrfs_root *root = BTRFS_I(inode)->root;
2559 struct btrfs_trans_handle *trans = NULL;
2560 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2561 struct extent_state *cached_state = NULL;
2562 struct new_sa_defrag_extent *new = NULL;
2563 int compress_type = 0;
2565 u64 logical_len = ordered_extent->len;
2567 bool truncated = false;
2569 nolock = btrfs_is_free_space_inode(inode);
2571 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2576 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2578 logical_len = ordered_extent->truncated_len;
2579 /* Truncated the entire extent, don't bother adding */
2584 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2585 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2586 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2588 trans = btrfs_join_transaction_nolock(root);
2590 trans = btrfs_join_transaction(root);
2591 if (IS_ERR(trans)) {
2592 ret = PTR_ERR(trans);
2596 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2597 ret = btrfs_update_inode_fallback(trans, root, inode);
2598 if (ret) /* -ENOMEM or corruption */
2599 btrfs_abort_transaction(trans, root, ret);
2603 lock_extent_bits(io_tree, ordered_extent->file_offset,
2604 ordered_extent->file_offset + ordered_extent->len - 1,
2607 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2608 ordered_extent->file_offset + ordered_extent->len - 1,
2609 EXTENT_DEFRAG, 1, cached_state);
2611 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2612 if (last_snapshot >= BTRFS_I(inode)->generation)
2613 /* the inode is shared */
2614 new = record_old_file_extents(inode, ordered_extent);
2616 clear_extent_bit(io_tree, ordered_extent->file_offset,
2617 ordered_extent->file_offset + ordered_extent->len - 1,
2618 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2622 trans = btrfs_join_transaction_nolock(root);
2624 trans = btrfs_join_transaction(root);
2625 if (IS_ERR(trans)) {
2626 ret = PTR_ERR(trans);
2630 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2632 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2633 compress_type = ordered_extent->compress_type;
2634 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2635 BUG_ON(compress_type);
2636 ret = btrfs_mark_extent_written(trans, inode,
2637 ordered_extent->file_offset,
2638 ordered_extent->file_offset +
2641 BUG_ON(root == root->fs_info->tree_root);
2642 ret = insert_reserved_file_extent(trans, inode,
2643 ordered_extent->file_offset,
2644 ordered_extent->start,
2645 ordered_extent->disk_len,
2646 logical_len, logical_len,
2647 compress_type, 0, 0,
2648 BTRFS_FILE_EXTENT_REG);
2650 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2651 ordered_extent->file_offset, ordered_extent->len,
2654 btrfs_abort_transaction(trans, root, ret);
2658 add_pending_csums(trans, inode, ordered_extent->file_offset,
2659 &ordered_extent->list);
2661 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2662 ret = btrfs_update_inode_fallback(trans, root, inode);
2663 if (ret) { /* -ENOMEM or corruption */
2664 btrfs_abort_transaction(trans, root, ret);
2669 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2670 ordered_extent->file_offset +
2671 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2673 if (root != root->fs_info->tree_root)
2674 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2676 btrfs_end_transaction(trans, root);
2678 if (ret || truncated) {
2682 start = ordered_extent->file_offset + logical_len;
2684 start = ordered_extent->file_offset;
2685 end = ordered_extent->file_offset + ordered_extent->len - 1;
2686 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2688 /* Drop the cache for the part of the extent we didn't write. */
2689 btrfs_drop_extent_cache(inode, start, end, 0);
2692 * If the ordered extent had an IOERR or something else went
2693 * wrong we need to return the space for this ordered extent
2694 * back to the allocator. We only free the extent in the
2695 * truncated case if we didn't write out the extent at all.
2697 if ((ret || !logical_len) &&
2698 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2699 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2700 btrfs_free_reserved_extent(root, ordered_extent->start,
2701 ordered_extent->disk_len);
2706 * This needs to be done to make sure anybody waiting knows we are done
2707 * updating everything for this ordered extent.
2709 btrfs_remove_ordered_extent(inode, ordered_extent);
2711 /* for snapshot-aware defrag */
2714 free_sa_defrag_extent(new);
2715 atomic_dec(&root->fs_info->defrag_running);
2717 relink_file_extents(new);
2722 btrfs_put_ordered_extent(ordered_extent);
2723 /* once for the tree */
2724 btrfs_put_ordered_extent(ordered_extent);
2729 static void finish_ordered_fn(struct btrfs_work *work)
2731 struct btrfs_ordered_extent *ordered_extent;
2732 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2733 btrfs_finish_ordered_io(ordered_extent);
2736 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2737 struct extent_state *state, int uptodate)
2739 struct inode *inode = page->mapping->host;
2740 struct btrfs_root *root = BTRFS_I(inode)->root;
2741 struct btrfs_ordered_extent *ordered_extent = NULL;
2742 struct btrfs_workers *workers;
2744 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2746 ClearPagePrivate2(page);
2747 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2748 end - start + 1, uptodate))
2751 ordered_extent->work.func = finish_ordered_fn;
2752 ordered_extent->work.flags = 0;
2754 if (btrfs_is_free_space_inode(inode))
2755 workers = &root->fs_info->endio_freespace_worker;
2757 workers = &root->fs_info->endio_write_workers;
2758 btrfs_queue_worker(workers, &ordered_extent->work);
2764 * when reads are done, we need to check csums to verify the data is correct
2765 * if there's a match, we allow the bio to finish. If not, the code in
2766 * extent_io.c will try to find good copies for us.
2768 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2769 u64 phy_offset, struct page *page,
2770 u64 start, u64 end, int mirror)
2772 size_t offset = start - page_offset(page);
2773 struct inode *inode = page->mapping->host;
2774 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2776 struct btrfs_root *root = BTRFS_I(inode)->root;
2779 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2780 DEFAULT_RATELIMIT_BURST);
2782 if (PageChecked(page)) {
2783 ClearPageChecked(page);
2787 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2790 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2791 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2792 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2797 phy_offset >>= inode->i_sb->s_blocksize_bits;
2798 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2800 kaddr = kmap_atomic(page);
2801 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2802 btrfs_csum_final(csum, (char *)&csum);
2803 if (csum != csum_expected)
2806 kunmap_atomic(kaddr);
2811 if (__ratelimit(&_rs))
2812 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2813 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2814 memset(kaddr + offset, 1, end - start + 1);
2815 flush_dcache_page(page);
2816 kunmap_atomic(kaddr);
2817 if (csum_expected == 0)
2822 struct delayed_iput {
2823 struct list_head list;
2824 struct inode *inode;
2827 /* JDM: If this is fs-wide, why can't we add a pointer to
2828 * btrfs_inode instead and avoid the allocation? */
2829 void btrfs_add_delayed_iput(struct inode *inode)
2831 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2832 struct delayed_iput *delayed;
2834 if (atomic_add_unless(&inode->i_count, -1, 1))
2837 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2838 delayed->inode = inode;
2840 spin_lock(&fs_info->delayed_iput_lock);
2841 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2842 spin_unlock(&fs_info->delayed_iput_lock);
2845 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2848 struct btrfs_fs_info *fs_info = root->fs_info;
2849 struct delayed_iput *delayed;
2852 spin_lock(&fs_info->delayed_iput_lock);
2853 empty = list_empty(&fs_info->delayed_iputs);
2854 spin_unlock(&fs_info->delayed_iput_lock);
2858 spin_lock(&fs_info->delayed_iput_lock);
2859 list_splice_init(&fs_info->delayed_iputs, &list);
2860 spin_unlock(&fs_info->delayed_iput_lock);
2862 while (!list_empty(&list)) {
2863 delayed = list_entry(list.next, struct delayed_iput, list);
2864 list_del(&delayed->list);
2865 iput(delayed->inode);
2871 * This is called in transaction commit time. If there are no orphan
2872 * files in the subvolume, it removes orphan item and frees block_rsv
2875 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2876 struct btrfs_root *root)
2878 struct btrfs_block_rsv *block_rsv;
2881 if (atomic_read(&root->orphan_inodes) ||
2882 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2885 spin_lock(&root->orphan_lock);
2886 if (atomic_read(&root->orphan_inodes)) {
2887 spin_unlock(&root->orphan_lock);
2891 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2892 spin_unlock(&root->orphan_lock);
2896 block_rsv = root->orphan_block_rsv;
2897 root->orphan_block_rsv = NULL;
2898 spin_unlock(&root->orphan_lock);
2900 if (root->orphan_item_inserted &&
2901 btrfs_root_refs(&root->root_item) > 0) {
2902 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2903 root->root_key.objectid);
2905 btrfs_abort_transaction(trans, root, ret);
2907 root->orphan_item_inserted = 0;
2911 WARN_ON(block_rsv->size > 0);
2912 btrfs_free_block_rsv(root, block_rsv);
2917 * This creates an orphan entry for the given inode in case something goes
2918 * wrong in the middle of an unlink/truncate.
2920 * NOTE: caller of this function should reserve 5 units of metadata for
2923 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2925 struct btrfs_root *root = BTRFS_I(inode)->root;
2926 struct btrfs_block_rsv *block_rsv = NULL;
2931 if (!root->orphan_block_rsv) {
2932 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2937 spin_lock(&root->orphan_lock);
2938 if (!root->orphan_block_rsv) {
2939 root->orphan_block_rsv = block_rsv;
2940 } else if (block_rsv) {
2941 btrfs_free_block_rsv(root, block_rsv);
2945 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2946 &BTRFS_I(inode)->runtime_flags)) {
2949 * For proper ENOSPC handling, we should do orphan
2950 * cleanup when mounting. But this introduces backward
2951 * compatibility issue.
2953 if (!xchg(&root->orphan_item_inserted, 1))
2959 atomic_inc(&root->orphan_inodes);
2962 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2963 &BTRFS_I(inode)->runtime_flags))
2965 spin_unlock(&root->orphan_lock);
2967 /* grab metadata reservation from transaction handle */
2969 ret = btrfs_orphan_reserve_metadata(trans, inode);
2970 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2973 /* insert an orphan item to track this unlinked/truncated file */
2975 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2977 atomic_dec(&root->orphan_inodes);
2979 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2980 &BTRFS_I(inode)->runtime_flags);
2981 btrfs_orphan_release_metadata(inode);
2983 if (ret != -EEXIST) {
2984 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2985 &BTRFS_I(inode)->runtime_flags);
2986 btrfs_abort_transaction(trans, root, ret);
2993 /* insert an orphan item to track subvolume contains orphan files */
2995 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2996 root->root_key.objectid);
2997 if (ret && ret != -EEXIST) {
2998 btrfs_abort_transaction(trans, root, ret);
3006 * We have done the truncate/delete so we can go ahead and remove the orphan
3007 * item for this particular inode.
3009 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3010 struct inode *inode)
3012 struct btrfs_root *root = BTRFS_I(inode)->root;
3013 int delete_item = 0;
3014 int release_rsv = 0;
3017 spin_lock(&root->orphan_lock);
3018 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3019 &BTRFS_I(inode)->runtime_flags))
3022 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3023 &BTRFS_I(inode)->runtime_flags))
3025 spin_unlock(&root->orphan_lock);
3028 atomic_dec(&root->orphan_inodes);
3030 ret = btrfs_del_orphan_item(trans, root,
3035 btrfs_orphan_release_metadata(inode);
3041 * this cleans up any orphans that may be left on the list from the last use
3044 int btrfs_orphan_cleanup(struct btrfs_root *root)
3046 struct btrfs_path *path;
3047 struct extent_buffer *leaf;
3048 struct btrfs_key key, found_key;
3049 struct btrfs_trans_handle *trans;
3050 struct inode *inode;
3051 u64 last_objectid = 0;
3052 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3054 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3057 path = btrfs_alloc_path();
3064 key.objectid = BTRFS_ORPHAN_OBJECTID;
3065 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3066 key.offset = (u64)-1;
3069 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3074 * if ret == 0 means we found what we were searching for, which
3075 * is weird, but possible, so only screw with path if we didn't
3076 * find the key and see if we have stuff that matches
3080 if (path->slots[0] == 0)
3085 /* pull out the item */
3086 leaf = path->nodes[0];
3087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3089 /* make sure the item matches what we want */
3090 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3092 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3095 /* release the path since we're done with it */
3096 btrfs_release_path(path);
3099 * this is where we are basically btrfs_lookup, without the
3100 * crossing root thing. we store the inode number in the
3101 * offset of the orphan item.
3104 if (found_key.offset == last_objectid) {
3105 btrfs_err(root->fs_info,
3106 "Error removing orphan entry, stopping orphan cleanup");
3111 last_objectid = found_key.offset;
3113 found_key.objectid = found_key.offset;
3114 found_key.type = BTRFS_INODE_ITEM_KEY;
3115 found_key.offset = 0;
3116 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3117 ret = PTR_ERR_OR_ZERO(inode);
3118 if (ret && ret != -ESTALE)
3121 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3122 struct btrfs_root *dead_root;
3123 struct btrfs_fs_info *fs_info = root->fs_info;
3124 int is_dead_root = 0;
3127 * this is an orphan in the tree root. Currently these
3128 * could come from 2 sources:
3129 * a) a snapshot deletion in progress
3130 * b) a free space cache inode
3131 * We need to distinguish those two, as the snapshot
3132 * orphan must not get deleted.
3133 * find_dead_roots already ran before us, so if this
3134 * is a snapshot deletion, we should find the root
3135 * in the dead_roots list
3137 spin_lock(&fs_info->trans_lock);
3138 list_for_each_entry(dead_root, &fs_info->dead_roots,
3140 if (dead_root->root_key.objectid ==
3141 found_key.objectid) {
3146 spin_unlock(&fs_info->trans_lock);
3148 /* prevent this orphan from being found again */
3149 key.offset = found_key.objectid - 1;
3154 * Inode is already gone but the orphan item is still there,
3155 * kill the orphan item.
3157 if (ret == -ESTALE) {
3158 trans = btrfs_start_transaction(root, 1);
3159 if (IS_ERR(trans)) {
3160 ret = PTR_ERR(trans);
3163 btrfs_debug(root->fs_info, "auto deleting %Lu",
3164 found_key.objectid);
3165 ret = btrfs_del_orphan_item(trans, root,
3166 found_key.objectid);
3167 btrfs_end_transaction(trans, root);
3174 * add this inode to the orphan list so btrfs_orphan_del does
3175 * the proper thing when we hit it
3177 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3178 &BTRFS_I(inode)->runtime_flags);
3179 atomic_inc(&root->orphan_inodes);
3181 /* if we have links, this was a truncate, lets do that */
3182 if (inode->i_nlink) {
3183 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3189 /* 1 for the orphan item deletion. */
3190 trans = btrfs_start_transaction(root, 1);
3191 if (IS_ERR(trans)) {
3193 ret = PTR_ERR(trans);
3196 ret = btrfs_orphan_add(trans, inode);
3197 btrfs_end_transaction(trans, root);
3203 ret = btrfs_truncate(inode);
3205 btrfs_orphan_del(NULL, inode);
3210 /* this will do delete_inode and everything for us */
3215 /* release the path since we're done with it */
3216 btrfs_release_path(path);
3218 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3220 if (root->orphan_block_rsv)
3221 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3224 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3225 trans = btrfs_join_transaction(root);
3227 btrfs_end_transaction(trans, root);
3231 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3233 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3237 btrfs_crit(root->fs_info,
3238 "could not do orphan cleanup %d", ret);
3239 btrfs_free_path(path);
3244 * very simple check to peek ahead in the leaf looking for xattrs. If we
3245 * don't find any xattrs, we know there can't be any acls.
3247 * slot is the slot the inode is in, objectid is the objectid of the inode
3249 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3250 int slot, u64 objectid)
3252 u32 nritems = btrfs_header_nritems(leaf);
3253 struct btrfs_key found_key;
3254 static u64 xattr_access = 0;
3255 static u64 xattr_default = 0;
3258 if (!xattr_access) {
3259 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3260 strlen(POSIX_ACL_XATTR_ACCESS));
3261 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3262 strlen(POSIX_ACL_XATTR_DEFAULT));
3266 while (slot < nritems) {
3267 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3269 /* we found a different objectid, there must not be acls */
3270 if (found_key.objectid != objectid)
3273 /* we found an xattr, assume we've got an acl */
3274 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3275 if (found_key.offset == xattr_access ||
3276 found_key.offset == xattr_default)
3281 * we found a key greater than an xattr key, there can't
3282 * be any acls later on
3284 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3291 * it goes inode, inode backrefs, xattrs, extents,
3292 * so if there are a ton of hard links to an inode there can
3293 * be a lot of backrefs. Don't waste time searching too hard,
3294 * this is just an optimization
3299 /* we hit the end of the leaf before we found an xattr or
3300 * something larger than an xattr. We have to assume the inode
3307 * read an inode from the btree into the in-memory inode
3309 static void btrfs_read_locked_inode(struct inode *inode)
3311 struct btrfs_path *path;
3312 struct extent_buffer *leaf;
3313 struct btrfs_inode_item *inode_item;
3314 struct btrfs_timespec *tspec;
3315 struct btrfs_root *root = BTRFS_I(inode)->root;
3316 struct btrfs_key location;
3320 bool filled = false;
3322 ret = btrfs_fill_inode(inode, &rdev);
3326 path = btrfs_alloc_path();
3330 path->leave_spinning = 1;
3331 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3333 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3337 leaf = path->nodes[0];
3342 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3343 struct btrfs_inode_item);
3344 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3345 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3346 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3347 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3348 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3350 tspec = btrfs_inode_atime(inode_item);
3351 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3352 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3354 tspec = btrfs_inode_mtime(inode_item);
3355 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3356 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3358 tspec = btrfs_inode_ctime(inode_item);
3359 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3360 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3362 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3363 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3364 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3367 * If we were modified in the current generation and evicted from memory
3368 * and then re-read we need to do a full sync since we don't have any
3369 * idea about which extents were modified before we were evicted from
3372 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3373 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3374 &BTRFS_I(inode)->runtime_flags);
3376 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3377 inode->i_generation = BTRFS_I(inode)->generation;
3379 rdev = btrfs_inode_rdev(leaf, inode_item);
3381 BTRFS_I(inode)->index_cnt = (u64)-1;
3382 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3385 * try to precache a NULL acl entry for files that don't have
3386 * any xattrs or acls
3388 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3391 cache_no_acl(inode);
3393 btrfs_free_path(path);
3395 switch (inode->i_mode & S_IFMT) {
3397 inode->i_mapping->a_ops = &btrfs_aops;
3398 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3399 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3400 inode->i_fop = &btrfs_file_operations;
3401 inode->i_op = &btrfs_file_inode_operations;
3404 inode->i_fop = &btrfs_dir_file_operations;
3405 if (root == root->fs_info->tree_root)
3406 inode->i_op = &btrfs_dir_ro_inode_operations;
3408 inode->i_op = &btrfs_dir_inode_operations;
3411 inode->i_op = &btrfs_symlink_inode_operations;
3412 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3413 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3416 inode->i_op = &btrfs_special_inode_operations;
3417 init_special_inode(inode, inode->i_mode, rdev);
3421 btrfs_update_iflags(inode);
3425 btrfs_free_path(path);
3426 make_bad_inode(inode);
3430 * given a leaf and an inode, copy the inode fields into the leaf
3432 static void fill_inode_item(struct btrfs_trans_handle *trans,
3433 struct extent_buffer *leaf,
3434 struct btrfs_inode_item *item,
3435 struct inode *inode)
3437 struct btrfs_map_token token;
3439 btrfs_init_map_token(&token);
3441 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3442 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3443 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3445 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3446 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3448 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3449 inode->i_atime.tv_sec, &token);
3450 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3451 inode->i_atime.tv_nsec, &token);
3453 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3454 inode->i_mtime.tv_sec, &token);
3455 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3456 inode->i_mtime.tv_nsec, &token);
3458 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3459 inode->i_ctime.tv_sec, &token);
3460 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3461 inode->i_ctime.tv_nsec, &token);
3463 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3465 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3467 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3468 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3469 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3470 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3471 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3475 * copy everything in the in-memory inode into the btree.
3477 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3478 struct btrfs_root *root, struct inode *inode)
3480 struct btrfs_inode_item *inode_item;
3481 struct btrfs_path *path;
3482 struct extent_buffer *leaf;
3485 path = btrfs_alloc_path();
3489 path->leave_spinning = 1;
3490 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3498 btrfs_unlock_up_safe(path, 1);
3499 leaf = path->nodes[0];
3500 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3501 struct btrfs_inode_item);
3503 fill_inode_item(trans, leaf, inode_item, inode);
3504 btrfs_mark_buffer_dirty(leaf);
3505 btrfs_set_inode_last_trans(trans, inode);
3508 btrfs_free_path(path);
3513 * copy everything in the in-memory inode into the btree.
3515 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3516 struct btrfs_root *root, struct inode *inode)
3521 * If the inode is a free space inode, we can deadlock during commit
3522 * if we put it into the delayed code.
3524 * The data relocation inode should also be directly updated
3527 if (!btrfs_is_free_space_inode(inode)
3528 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3529 btrfs_update_root_times(trans, root);
3531 ret = btrfs_delayed_update_inode(trans, root, inode);
3533 btrfs_set_inode_last_trans(trans, inode);
3537 return btrfs_update_inode_item(trans, root, inode);
3540 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *root,
3542 struct inode *inode)
3546 ret = btrfs_update_inode(trans, root, inode);
3548 return btrfs_update_inode_item(trans, root, inode);
3553 * unlink helper that gets used here in inode.c and in the tree logging
3554 * recovery code. It remove a link in a directory with a given name, and
3555 * also drops the back refs in the inode to the directory
3557 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3558 struct btrfs_root *root,
3559 struct inode *dir, struct inode *inode,
3560 const char *name, int name_len)
3562 struct btrfs_path *path;
3564 struct extent_buffer *leaf;
3565 struct btrfs_dir_item *di;
3566 struct btrfs_key key;
3568 u64 ino = btrfs_ino(inode);
3569 u64 dir_ino = btrfs_ino(dir);
3571 path = btrfs_alloc_path();
3577 path->leave_spinning = 1;
3578 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3579 name, name_len, -1);
3588 leaf = path->nodes[0];
3589 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3590 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3593 btrfs_release_path(path);
3595 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3598 btrfs_info(root->fs_info,
3599 "failed to delete reference to %.*s, inode %llu parent %llu",
3600 name_len, name, ino, dir_ino);
3601 btrfs_abort_transaction(trans, root, ret);
3605 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3607 btrfs_abort_transaction(trans, root, ret);
3611 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3613 if (ret != 0 && ret != -ENOENT) {
3614 btrfs_abort_transaction(trans, root, ret);
3618 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3623 btrfs_abort_transaction(trans, root, ret);
3625 btrfs_free_path(path);
3629 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3630 inode_inc_iversion(inode);
3631 inode_inc_iversion(dir);
3632 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3633 ret = btrfs_update_inode(trans, root, dir);
3638 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3639 struct btrfs_root *root,
3640 struct inode *dir, struct inode *inode,
3641 const char *name, int name_len)
3644 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3647 ret = btrfs_update_inode(trans, root, inode);
3653 * helper to start transaction for unlink and rmdir.
3655 * unlink and rmdir are special in btrfs, they do not always free space, so
3656 * if we cannot make our reservations the normal way try and see if there is
3657 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3658 * allow the unlink to occur.
3660 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3662 struct btrfs_trans_handle *trans;
3663 struct btrfs_root *root = BTRFS_I(dir)->root;
3667 * 1 for the possible orphan item
3668 * 1 for the dir item
3669 * 1 for the dir index
3670 * 1 for the inode ref
3673 trans = btrfs_start_transaction(root, 5);
3674 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3677 if (PTR_ERR(trans) == -ENOSPC) {
3678 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3680 trans = btrfs_start_transaction(root, 0);
3683 ret = btrfs_cond_migrate_bytes(root->fs_info,
3684 &root->fs_info->trans_block_rsv,
3687 btrfs_end_transaction(trans, root);
3688 return ERR_PTR(ret);
3690 trans->block_rsv = &root->fs_info->trans_block_rsv;
3691 trans->bytes_reserved = num_bytes;
3696 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3698 struct btrfs_root *root = BTRFS_I(dir)->root;
3699 struct btrfs_trans_handle *trans;
3700 struct inode *inode = dentry->d_inode;
3703 trans = __unlink_start_trans(dir);
3705 return PTR_ERR(trans);
3707 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3709 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3710 dentry->d_name.name, dentry->d_name.len);
3714 if (inode->i_nlink == 0) {
3715 ret = btrfs_orphan_add(trans, inode);
3721 btrfs_end_transaction(trans, root);
3722 btrfs_btree_balance_dirty(root);
3726 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3727 struct btrfs_root *root,
3728 struct inode *dir, u64 objectid,
3729 const char *name, int name_len)
3731 struct btrfs_path *path;
3732 struct extent_buffer *leaf;
3733 struct btrfs_dir_item *di;
3734 struct btrfs_key key;
3737 u64 dir_ino = btrfs_ino(dir);
3739 path = btrfs_alloc_path();
3743 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3744 name, name_len, -1);
3745 if (IS_ERR_OR_NULL(di)) {
3753 leaf = path->nodes[0];
3754 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3755 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3756 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3758 btrfs_abort_transaction(trans, root, ret);
3761 btrfs_release_path(path);
3763 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3764 objectid, root->root_key.objectid,
3765 dir_ino, &index, name, name_len);
3767 if (ret != -ENOENT) {
3768 btrfs_abort_transaction(trans, root, ret);
3771 di = btrfs_search_dir_index_item(root, path, dir_ino,
3773 if (IS_ERR_OR_NULL(di)) {
3778 btrfs_abort_transaction(trans, root, ret);
3782 leaf = path->nodes[0];
3783 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3784 btrfs_release_path(path);
3787 btrfs_release_path(path);
3789 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3791 btrfs_abort_transaction(trans, root, ret);
3795 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3796 inode_inc_iversion(dir);
3797 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3798 ret = btrfs_update_inode_fallback(trans, root, dir);
3800 btrfs_abort_transaction(trans, root, ret);
3802 btrfs_free_path(path);
3806 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3808 struct inode *inode = dentry->d_inode;
3810 struct btrfs_root *root = BTRFS_I(dir)->root;
3811 struct btrfs_trans_handle *trans;
3813 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3815 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3818 trans = __unlink_start_trans(dir);
3820 return PTR_ERR(trans);
3822 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3823 err = btrfs_unlink_subvol(trans, root, dir,
3824 BTRFS_I(inode)->location.objectid,
3825 dentry->d_name.name,
3826 dentry->d_name.len);
3830 err = btrfs_orphan_add(trans, inode);
3834 /* now the directory is empty */
3835 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3836 dentry->d_name.name, dentry->d_name.len);
3838 btrfs_i_size_write(inode, 0);
3840 btrfs_end_transaction(trans, root);
3841 btrfs_btree_balance_dirty(root);
3847 * this can truncate away extent items, csum items and directory items.
3848 * It starts at a high offset and removes keys until it can't find
3849 * any higher than new_size
3851 * csum items that cross the new i_size are truncated to the new size
3854 * min_type is the minimum key type to truncate down to. If set to 0, this
3855 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3857 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3858 struct btrfs_root *root,
3859 struct inode *inode,
3860 u64 new_size, u32 min_type)
3862 struct btrfs_path *path;
3863 struct extent_buffer *leaf;
3864 struct btrfs_file_extent_item *fi;
3865 struct btrfs_key key;
3866 struct btrfs_key found_key;
3867 u64 extent_start = 0;
3868 u64 extent_num_bytes = 0;
3869 u64 extent_offset = 0;
3871 u64 last_size = (u64)-1;
3872 u32 found_type = (u8)-1;
3875 int pending_del_nr = 0;
3876 int pending_del_slot = 0;
3877 int extent_type = -1;
3880 u64 ino = btrfs_ino(inode);
3882 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3884 path = btrfs_alloc_path();
3890 * We want to drop from the next block forward in case this new size is
3891 * not block aligned since we will be keeping the last block of the
3892 * extent just the way it is.
3894 if (root->ref_cows || root == root->fs_info->tree_root)
3895 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3896 root->sectorsize), (u64)-1, 0);
3899 * This function is also used to drop the items in the log tree before
3900 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3901 * it is used to drop the loged items. So we shouldn't kill the delayed
3904 if (min_type == 0 && root == BTRFS_I(inode)->root)
3905 btrfs_kill_delayed_inode_items(inode);
3908 key.offset = (u64)-1;
3912 path->leave_spinning = 1;
3913 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3920 /* there are no items in the tree for us to truncate, we're
3923 if (path->slots[0] == 0)
3930 leaf = path->nodes[0];
3931 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3932 found_type = btrfs_key_type(&found_key);
3934 if (found_key.objectid != ino)
3937 if (found_type < min_type)
3940 item_end = found_key.offset;
3941 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3942 fi = btrfs_item_ptr(leaf, path->slots[0],
3943 struct btrfs_file_extent_item);
3944 extent_type = btrfs_file_extent_type(leaf, fi);
3945 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3947 btrfs_file_extent_num_bytes(leaf, fi);
3948 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3949 item_end += btrfs_file_extent_inline_len(leaf,
3954 if (found_type > min_type) {
3957 if (item_end < new_size)
3959 if (found_key.offset >= new_size)
3965 /* FIXME, shrink the extent if the ref count is only 1 */
3966 if (found_type != BTRFS_EXTENT_DATA_KEY)
3970 last_size = found_key.offset;
3972 last_size = new_size;
3974 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3976 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3978 u64 orig_num_bytes =
3979 btrfs_file_extent_num_bytes(leaf, fi);
3980 extent_num_bytes = ALIGN(new_size -
3983 btrfs_set_file_extent_num_bytes(leaf, fi,
3985 num_dec = (orig_num_bytes -
3987 if (root->ref_cows && extent_start != 0)
3988 inode_sub_bytes(inode, num_dec);
3989 btrfs_mark_buffer_dirty(leaf);
3992 btrfs_file_extent_disk_num_bytes(leaf,
3994 extent_offset = found_key.offset -
3995 btrfs_file_extent_offset(leaf, fi);
3997 /* FIXME blocksize != 4096 */
3998 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3999 if (extent_start != 0) {
4002 inode_sub_bytes(inode, num_dec);
4005 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4007 * we can't truncate inline items that have had
4011 btrfs_file_extent_compression(leaf, fi) == 0 &&
4012 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4013 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4014 u32 size = new_size - found_key.offset;
4016 if (root->ref_cows) {
4017 inode_sub_bytes(inode, item_end + 1 -
4021 btrfs_file_extent_calc_inline_size(size);
4022 btrfs_truncate_item(root, path, size, 1);
4023 } else if (root->ref_cows) {
4024 inode_sub_bytes(inode, item_end + 1 -
4030 if (!pending_del_nr) {
4031 /* no pending yet, add ourselves */
4032 pending_del_slot = path->slots[0];
4034 } else if (pending_del_nr &&
4035 path->slots[0] + 1 == pending_del_slot) {
4036 /* hop on the pending chunk */
4038 pending_del_slot = path->slots[0];
4045 if (found_extent && (root->ref_cows ||
4046 root == root->fs_info->tree_root)) {
4047 btrfs_set_path_blocking(path);
4048 ret = btrfs_free_extent(trans, root, extent_start,
4049 extent_num_bytes, 0,
4050 btrfs_header_owner(leaf),
4051 ino, extent_offset, 0);
4055 if (found_type == BTRFS_INODE_ITEM_KEY)
4058 if (path->slots[0] == 0 ||
4059 path->slots[0] != pending_del_slot) {
4060 if (pending_del_nr) {
4061 ret = btrfs_del_items(trans, root, path,
4065 btrfs_abort_transaction(trans,
4071 btrfs_release_path(path);
4078 if (pending_del_nr) {
4079 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4082 btrfs_abort_transaction(trans, root, ret);
4085 if (last_size != (u64)-1)
4086 btrfs_ordered_update_i_size(inode, last_size, NULL);
4087 btrfs_free_path(path);
4092 * btrfs_truncate_page - read, zero a chunk and write a page
4093 * @inode - inode that we're zeroing
4094 * @from - the offset to start zeroing
4095 * @len - the length to zero, 0 to zero the entire range respective to the
4097 * @front - zero up to the offset instead of from the offset on
4099 * This will find the page for the "from" offset and cow the page and zero the
4100 * part we want to zero. This is used with truncate and hole punching.
4102 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4105 struct address_space *mapping = inode->i_mapping;
4106 struct btrfs_root *root = BTRFS_I(inode)->root;
4107 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4108 struct btrfs_ordered_extent *ordered;
4109 struct extent_state *cached_state = NULL;
4111 u32 blocksize = root->sectorsize;
4112 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4113 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4115 gfp_t mask = btrfs_alloc_write_mask(mapping);
4120 if ((offset & (blocksize - 1)) == 0 &&
4121 (!len || ((len & (blocksize - 1)) == 0)))
4123 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4128 page = find_or_create_page(mapping, index, mask);
4130 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4135 page_start = page_offset(page);
4136 page_end = page_start + PAGE_CACHE_SIZE - 1;
4138 if (!PageUptodate(page)) {
4139 ret = btrfs_readpage(NULL, page);
4141 if (page->mapping != mapping) {
4143 page_cache_release(page);
4146 if (!PageUptodate(page)) {
4151 wait_on_page_writeback(page);
4153 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4154 set_page_extent_mapped(page);
4156 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4158 unlock_extent_cached(io_tree, page_start, page_end,
4159 &cached_state, GFP_NOFS);
4161 page_cache_release(page);
4162 btrfs_start_ordered_extent(inode, ordered, 1);
4163 btrfs_put_ordered_extent(ordered);
4167 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4168 EXTENT_DIRTY | EXTENT_DELALLOC |
4169 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4170 0, 0, &cached_state, GFP_NOFS);
4172 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4175 unlock_extent_cached(io_tree, page_start, page_end,
4176 &cached_state, GFP_NOFS);
4180 if (offset != PAGE_CACHE_SIZE) {
4182 len = PAGE_CACHE_SIZE - offset;
4185 memset(kaddr, 0, offset);
4187 memset(kaddr + offset, 0, len);
4188 flush_dcache_page(page);
4191 ClearPageChecked(page);
4192 set_page_dirty(page);
4193 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4198 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4200 page_cache_release(page);
4206 * This function puts in dummy file extents for the area we're creating a hole
4207 * for. So if we are truncating this file to a larger size we need to insert
4208 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4209 * the range between oldsize and size
4211 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4213 struct btrfs_trans_handle *trans;
4214 struct btrfs_root *root = BTRFS_I(inode)->root;
4215 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4216 struct extent_map *em = NULL;
4217 struct extent_state *cached_state = NULL;
4218 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4219 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4220 u64 block_end = ALIGN(size, root->sectorsize);
4227 * If our size started in the middle of a page we need to zero out the
4228 * rest of the page before we expand the i_size, otherwise we could
4229 * expose stale data.
4231 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4235 if (size <= hole_start)
4239 struct btrfs_ordered_extent *ordered;
4241 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4243 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4244 block_end - hole_start);
4247 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4248 &cached_state, GFP_NOFS);
4249 btrfs_start_ordered_extent(inode, ordered, 1);
4250 btrfs_put_ordered_extent(ordered);
4253 cur_offset = hole_start;
4255 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4256 block_end - cur_offset, 0);
4262 last_byte = min(extent_map_end(em), block_end);
4263 last_byte = ALIGN(last_byte , root->sectorsize);
4264 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4265 struct extent_map *hole_em;
4266 hole_size = last_byte - cur_offset;
4268 trans = btrfs_start_transaction(root, 3);
4269 if (IS_ERR(trans)) {
4270 err = PTR_ERR(trans);
4274 err = btrfs_drop_extents(trans, root, inode,
4276 cur_offset + hole_size, 1);
4278 btrfs_abort_transaction(trans, root, err);
4279 btrfs_end_transaction(trans, root);
4283 err = btrfs_insert_file_extent(trans, root,
4284 btrfs_ino(inode), cur_offset, 0,
4285 0, hole_size, 0, hole_size,
4288 btrfs_abort_transaction(trans, root, err);
4289 btrfs_end_transaction(trans, root);
4293 btrfs_drop_extent_cache(inode, cur_offset,
4294 cur_offset + hole_size - 1, 0);
4295 hole_em = alloc_extent_map();
4297 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4298 &BTRFS_I(inode)->runtime_flags);
4301 hole_em->start = cur_offset;
4302 hole_em->len = hole_size;
4303 hole_em->orig_start = cur_offset;
4305 hole_em->block_start = EXTENT_MAP_HOLE;
4306 hole_em->block_len = 0;
4307 hole_em->orig_block_len = 0;
4308 hole_em->ram_bytes = hole_size;
4309 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4310 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4311 hole_em->generation = trans->transid;
4314 write_lock(&em_tree->lock);
4315 err = add_extent_mapping(em_tree, hole_em, 1);
4316 write_unlock(&em_tree->lock);
4319 btrfs_drop_extent_cache(inode, cur_offset,
4323 free_extent_map(hole_em);
4325 btrfs_update_inode(trans, root, inode);
4326 btrfs_end_transaction(trans, root);
4328 free_extent_map(em);
4330 cur_offset = last_byte;
4331 if (cur_offset >= block_end)
4335 free_extent_map(em);
4336 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4341 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4343 struct btrfs_root *root = BTRFS_I(inode)->root;
4344 struct btrfs_trans_handle *trans;
4345 loff_t oldsize = i_size_read(inode);
4346 loff_t newsize = attr->ia_size;
4347 int mask = attr->ia_valid;
4351 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4352 * special case where we need to update the times despite not having
4353 * these flags set. For all other operations the VFS set these flags
4354 * explicitly if it wants a timestamp update.
4356 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4357 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4359 if (newsize > oldsize) {
4360 truncate_pagecache(inode, newsize);
4361 ret = btrfs_cont_expand(inode, oldsize, newsize);
4365 trans = btrfs_start_transaction(root, 1);
4367 return PTR_ERR(trans);
4369 i_size_write(inode, newsize);
4370 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4371 ret = btrfs_update_inode(trans, root, inode);
4372 btrfs_end_transaction(trans, root);
4376 * We're truncating a file that used to have good data down to
4377 * zero. Make sure it gets into the ordered flush list so that
4378 * any new writes get down to disk quickly.
4381 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4382 &BTRFS_I(inode)->runtime_flags);
4385 * 1 for the orphan item we're going to add
4386 * 1 for the orphan item deletion.
4388 trans = btrfs_start_transaction(root, 2);
4390 return PTR_ERR(trans);
4393 * We need to do this in case we fail at _any_ point during the
4394 * actual truncate. Once we do the truncate_setsize we could
4395 * invalidate pages which forces any outstanding ordered io to
4396 * be instantly completed which will give us extents that need
4397 * to be truncated. If we fail to get an orphan inode down we
4398 * could have left over extents that were never meant to live,
4399 * so we need to garuntee from this point on that everything
4400 * will be consistent.
4402 ret = btrfs_orphan_add(trans, inode);
4403 btrfs_end_transaction(trans, root);
4407 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4408 truncate_setsize(inode, newsize);
4410 /* Disable nonlocked read DIO to avoid the end less truncate */
4411 btrfs_inode_block_unlocked_dio(inode);
4412 inode_dio_wait(inode);
4413 btrfs_inode_resume_unlocked_dio(inode);
4415 ret = btrfs_truncate(inode);
4416 if (ret && inode->i_nlink) {
4420 * failed to truncate, disk_i_size is only adjusted down
4421 * as we remove extents, so it should represent the true
4422 * size of the inode, so reset the in memory size and
4423 * delete our orphan entry.
4425 trans = btrfs_join_transaction(root);
4426 if (IS_ERR(trans)) {
4427 btrfs_orphan_del(NULL, inode);
4430 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4431 err = btrfs_orphan_del(trans, inode);
4433 btrfs_abort_transaction(trans, root, err);
4434 btrfs_end_transaction(trans, root);
4441 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4443 struct inode *inode = dentry->d_inode;
4444 struct btrfs_root *root = BTRFS_I(inode)->root;
4447 if (btrfs_root_readonly(root))
4450 err = inode_change_ok(inode, attr);
4454 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4455 err = btrfs_setsize(inode, attr);
4460 if (attr->ia_valid) {
4461 setattr_copy(inode, attr);
4462 inode_inc_iversion(inode);
4463 err = btrfs_dirty_inode(inode);
4465 if (!err && attr->ia_valid & ATTR_MODE)
4466 err = btrfs_acl_chmod(inode);
4472 void btrfs_evict_inode(struct inode *inode)
4474 struct btrfs_trans_handle *trans;
4475 struct btrfs_root *root = BTRFS_I(inode)->root;
4476 struct btrfs_block_rsv *rsv, *global_rsv;
4477 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4480 trace_btrfs_inode_evict(inode);
4482 truncate_inode_pages(&inode->i_data, 0);
4483 if (inode->i_nlink &&
4484 ((btrfs_root_refs(&root->root_item) != 0 &&
4485 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4486 btrfs_is_free_space_inode(inode)))
4489 if (is_bad_inode(inode)) {
4490 btrfs_orphan_del(NULL, inode);
4493 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4494 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4496 if (root->fs_info->log_root_recovering) {
4497 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4498 &BTRFS_I(inode)->runtime_flags));
4502 if (inode->i_nlink > 0) {
4503 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4504 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4508 ret = btrfs_commit_inode_delayed_inode(inode);
4510 btrfs_orphan_del(NULL, inode);
4514 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4516 btrfs_orphan_del(NULL, inode);
4519 rsv->size = min_size;
4521 global_rsv = &root->fs_info->global_block_rsv;
4523 btrfs_i_size_write(inode, 0);
4526 * This is a bit simpler than btrfs_truncate since we've already
4527 * reserved our space for our orphan item in the unlink, so we just
4528 * need to reserve some slack space in case we add bytes and update
4529 * inode item when doing the truncate.
4532 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4533 BTRFS_RESERVE_FLUSH_LIMIT);
4536 * Try and steal from the global reserve since we will
4537 * likely not use this space anyway, we want to try as
4538 * hard as possible to get this to work.
4541 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4544 btrfs_warn(root->fs_info,
4545 "Could not get space for a delete, will truncate on mount %d",
4547 btrfs_orphan_del(NULL, inode);
4548 btrfs_free_block_rsv(root, rsv);
4552 trans = btrfs_join_transaction(root);
4553 if (IS_ERR(trans)) {
4554 btrfs_orphan_del(NULL, inode);
4555 btrfs_free_block_rsv(root, rsv);
4559 trans->block_rsv = rsv;
4561 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4565 trans->block_rsv = &root->fs_info->trans_block_rsv;
4566 btrfs_end_transaction(trans, root);
4568 btrfs_btree_balance_dirty(root);
4571 btrfs_free_block_rsv(root, rsv);
4574 * Errors here aren't a big deal, it just means we leave orphan items
4575 * in the tree. They will be cleaned up on the next mount.
4578 trans->block_rsv = root->orphan_block_rsv;
4579 btrfs_orphan_del(trans, inode);
4581 btrfs_orphan_del(NULL, inode);
4584 trans->block_rsv = &root->fs_info->trans_block_rsv;
4585 if (!(root == root->fs_info->tree_root ||
4586 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4587 btrfs_return_ino(root, btrfs_ino(inode));
4589 btrfs_end_transaction(trans, root);
4590 btrfs_btree_balance_dirty(root);
4592 btrfs_remove_delayed_node(inode);
4598 * this returns the key found in the dir entry in the location pointer.
4599 * If no dir entries were found, location->objectid is 0.
4601 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4602 struct btrfs_key *location)
4604 const char *name = dentry->d_name.name;
4605 int namelen = dentry->d_name.len;
4606 struct btrfs_dir_item *di;
4607 struct btrfs_path *path;
4608 struct btrfs_root *root = BTRFS_I(dir)->root;
4611 path = btrfs_alloc_path();
4615 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4620 if (IS_ERR_OR_NULL(di))
4623 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4625 btrfs_free_path(path);
4628 location->objectid = 0;
4633 * when we hit a tree root in a directory, the btrfs part of the inode
4634 * needs to be changed to reflect the root directory of the tree root. This
4635 * is kind of like crossing a mount point.
4637 static int fixup_tree_root_location(struct btrfs_root *root,
4639 struct dentry *dentry,
4640 struct btrfs_key *location,
4641 struct btrfs_root **sub_root)
4643 struct btrfs_path *path;
4644 struct btrfs_root *new_root;
4645 struct btrfs_root_ref *ref;
4646 struct extent_buffer *leaf;
4650 path = btrfs_alloc_path();
4657 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4658 BTRFS_I(dir)->root->root_key.objectid,
4659 location->objectid);
4666 leaf = path->nodes[0];
4667 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4668 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4669 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4672 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4673 (unsigned long)(ref + 1),
4674 dentry->d_name.len);
4678 btrfs_release_path(path);
4680 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4681 if (IS_ERR(new_root)) {
4682 err = PTR_ERR(new_root);
4686 *sub_root = new_root;
4687 location->objectid = btrfs_root_dirid(&new_root->root_item);
4688 location->type = BTRFS_INODE_ITEM_KEY;
4689 location->offset = 0;
4692 btrfs_free_path(path);
4696 static void inode_tree_add(struct inode *inode)
4698 struct btrfs_root *root = BTRFS_I(inode)->root;
4699 struct btrfs_inode *entry;
4701 struct rb_node *parent;
4702 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4703 u64 ino = btrfs_ino(inode);
4705 if (inode_unhashed(inode))
4708 spin_lock(&root->inode_lock);
4709 p = &root->inode_tree.rb_node;
4712 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4714 if (ino < btrfs_ino(&entry->vfs_inode))
4715 p = &parent->rb_left;
4716 else if (ino > btrfs_ino(&entry->vfs_inode))
4717 p = &parent->rb_right;
4719 WARN_ON(!(entry->vfs_inode.i_state &
4720 (I_WILL_FREE | I_FREEING)));
4721 rb_replace_node(parent, new, &root->inode_tree);
4722 RB_CLEAR_NODE(parent);
4723 spin_unlock(&root->inode_lock);
4727 rb_link_node(new, parent, p);
4728 rb_insert_color(new, &root->inode_tree);
4729 spin_unlock(&root->inode_lock);
4732 static void inode_tree_del(struct inode *inode)
4734 struct btrfs_root *root = BTRFS_I(inode)->root;
4737 spin_lock(&root->inode_lock);
4738 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4739 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4740 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4741 empty = RB_EMPTY_ROOT(&root->inode_tree);
4743 spin_unlock(&root->inode_lock);
4745 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4746 synchronize_srcu(&root->fs_info->subvol_srcu);
4747 spin_lock(&root->inode_lock);
4748 empty = RB_EMPTY_ROOT(&root->inode_tree);
4749 spin_unlock(&root->inode_lock);
4751 btrfs_add_dead_root(root);
4755 void btrfs_invalidate_inodes(struct btrfs_root *root)
4757 struct rb_node *node;
4758 struct rb_node *prev;
4759 struct btrfs_inode *entry;
4760 struct inode *inode;
4763 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4765 spin_lock(&root->inode_lock);
4767 node = root->inode_tree.rb_node;
4771 entry = rb_entry(node, struct btrfs_inode, rb_node);
4773 if (objectid < btrfs_ino(&entry->vfs_inode))
4774 node = node->rb_left;
4775 else if (objectid > btrfs_ino(&entry->vfs_inode))
4776 node = node->rb_right;
4782 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4783 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4787 prev = rb_next(prev);
4791 entry = rb_entry(node, struct btrfs_inode, rb_node);
4792 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4793 inode = igrab(&entry->vfs_inode);
4795 spin_unlock(&root->inode_lock);
4796 if (atomic_read(&inode->i_count) > 1)
4797 d_prune_aliases(inode);
4799 * btrfs_drop_inode will have it removed from
4800 * the inode cache when its usage count
4805 spin_lock(&root->inode_lock);
4809 if (cond_resched_lock(&root->inode_lock))
4812 node = rb_next(node);
4814 spin_unlock(&root->inode_lock);
4817 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4819 struct btrfs_iget_args *args = p;
4820 inode->i_ino = args->ino;
4821 BTRFS_I(inode)->root = args->root;
4825 static int btrfs_find_actor(struct inode *inode, void *opaque)
4827 struct btrfs_iget_args *args = opaque;
4828 return args->ino == btrfs_ino(inode) &&
4829 args->root == BTRFS_I(inode)->root;
4832 static struct inode *btrfs_iget_locked(struct super_block *s,
4834 struct btrfs_root *root)
4836 struct inode *inode;
4837 struct btrfs_iget_args args;
4838 unsigned long hashval = btrfs_inode_hash(objectid, root);
4840 args.ino = objectid;
4843 inode = iget5_locked(s, hashval, btrfs_find_actor,
4844 btrfs_init_locked_inode,
4849 /* Get an inode object given its location and corresponding root.
4850 * Returns in *is_new if the inode was read from disk
4852 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4853 struct btrfs_root *root, int *new)
4855 struct inode *inode;
4857 inode = btrfs_iget_locked(s, location->objectid, root);
4859 return ERR_PTR(-ENOMEM);
4861 if (inode->i_state & I_NEW) {
4862 BTRFS_I(inode)->root = root;
4863 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4864 btrfs_read_locked_inode(inode);
4865 if (!is_bad_inode(inode)) {
4866 inode_tree_add(inode);
4867 unlock_new_inode(inode);
4871 unlock_new_inode(inode);
4873 inode = ERR_PTR(-ESTALE);
4880 static struct inode *new_simple_dir(struct super_block *s,
4881 struct btrfs_key *key,
4882 struct btrfs_root *root)
4884 struct inode *inode = new_inode(s);
4887 return ERR_PTR(-ENOMEM);
4889 BTRFS_I(inode)->root = root;
4890 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4891 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4893 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4894 inode->i_op = &btrfs_dir_ro_inode_operations;
4895 inode->i_fop = &simple_dir_operations;
4896 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4897 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4902 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4904 struct inode *inode;
4905 struct btrfs_root *root = BTRFS_I(dir)->root;
4906 struct btrfs_root *sub_root = root;
4907 struct btrfs_key location;
4911 if (dentry->d_name.len > BTRFS_NAME_LEN)
4912 return ERR_PTR(-ENAMETOOLONG);
4914 ret = btrfs_inode_by_name(dir, dentry, &location);
4916 return ERR_PTR(ret);
4918 if (location.objectid == 0)
4921 if (location.type == BTRFS_INODE_ITEM_KEY) {
4922 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4926 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4928 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4929 ret = fixup_tree_root_location(root, dir, dentry,
4930 &location, &sub_root);
4933 inode = ERR_PTR(ret);
4935 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4937 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4939 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4941 if (!IS_ERR(inode) && root != sub_root) {
4942 down_read(&root->fs_info->cleanup_work_sem);
4943 if (!(inode->i_sb->s_flags & MS_RDONLY))
4944 ret = btrfs_orphan_cleanup(sub_root);
4945 up_read(&root->fs_info->cleanup_work_sem);
4948 inode = ERR_PTR(ret);
4955 static int btrfs_dentry_delete(const struct dentry *dentry)
4957 struct btrfs_root *root;
4958 struct inode *inode = dentry->d_inode;
4960 if (!inode && !IS_ROOT(dentry))
4961 inode = dentry->d_parent->d_inode;
4964 root = BTRFS_I(inode)->root;
4965 if (btrfs_root_refs(&root->root_item) == 0)
4968 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4974 static void btrfs_dentry_release(struct dentry *dentry)
4976 if (dentry->d_fsdata)
4977 kfree(dentry->d_fsdata);
4980 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4985 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4989 unsigned char btrfs_filetype_table[] = {
4990 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4993 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
4995 struct inode *inode = file_inode(file);
4996 struct btrfs_root *root = BTRFS_I(inode)->root;
4997 struct btrfs_item *item;
4998 struct btrfs_dir_item *di;
4999 struct btrfs_key key;
5000 struct btrfs_key found_key;
5001 struct btrfs_path *path;
5002 struct list_head ins_list;
5003 struct list_head del_list;
5005 struct extent_buffer *leaf;
5007 unsigned char d_type;
5012 int key_type = BTRFS_DIR_INDEX_KEY;
5016 int is_curr = 0; /* ctx->pos points to the current index? */
5018 /* FIXME, use a real flag for deciding about the key type */
5019 if (root->fs_info->tree_root == root)
5020 key_type = BTRFS_DIR_ITEM_KEY;
5022 if (!dir_emit_dots(file, ctx))
5025 path = btrfs_alloc_path();
5031 if (key_type == BTRFS_DIR_INDEX_KEY) {
5032 INIT_LIST_HEAD(&ins_list);
5033 INIT_LIST_HEAD(&del_list);
5034 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5037 btrfs_set_key_type(&key, key_type);
5038 key.offset = ctx->pos;
5039 key.objectid = btrfs_ino(inode);
5041 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5046 leaf = path->nodes[0];
5047 slot = path->slots[0];
5048 if (slot >= btrfs_header_nritems(leaf)) {
5049 ret = btrfs_next_leaf(root, path);
5057 item = btrfs_item_nr(slot);
5058 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5060 if (found_key.objectid != key.objectid)
5062 if (btrfs_key_type(&found_key) != key_type)
5064 if (found_key.offset < ctx->pos)
5066 if (key_type == BTRFS_DIR_INDEX_KEY &&
5067 btrfs_should_delete_dir_index(&del_list,
5071 ctx->pos = found_key.offset;
5074 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5076 di_total = btrfs_item_size(leaf, item);
5078 while (di_cur < di_total) {
5079 struct btrfs_key location;
5081 if (verify_dir_item(root, leaf, di))
5084 name_len = btrfs_dir_name_len(leaf, di);
5085 if (name_len <= sizeof(tmp_name)) {
5086 name_ptr = tmp_name;
5088 name_ptr = kmalloc(name_len, GFP_NOFS);
5094 read_extent_buffer(leaf, name_ptr,
5095 (unsigned long)(di + 1), name_len);
5097 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5098 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5101 /* is this a reference to our own snapshot? If so
5104 * In contrast to old kernels, we insert the snapshot's
5105 * dir item and dir index after it has been created, so
5106 * we won't find a reference to our own snapshot. We
5107 * still keep the following code for backward
5110 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5111 location.objectid == root->root_key.objectid) {
5115 over = !dir_emit(ctx, name_ptr, name_len,
5116 location.objectid, d_type);
5119 if (name_ptr != tmp_name)
5124 di_len = btrfs_dir_name_len(leaf, di) +
5125 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5127 di = (struct btrfs_dir_item *)((char *)di + di_len);
5133 if (key_type == BTRFS_DIR_INDEX_KEY) {
5136 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5141 /* Reached end of directory/root. Bump pos past the last item. */
5145 * Stop new entries from being returned after we return the last
5148 * New directory entries are assigned a strictly increasing
5149 * offset. This means that new entries created during readdir
5150 * are *guaranteed* to be seen in the future by that readdir.
5151 * This has broken buggy programs which operate on names as
5152 * they're returned by readdir. Until we re-use freed offsets
5153 * we have this hack to stop new entries from being returned
5154 * under the assumption that they'll never reach this huge
5157 * This is being careful not to overflow 32bit loff_t unless the
5158 * last entry requires it because doing so has broken 32bit apps
5161 if (key_type == BTRFS_DIR_INDEX_KEY) {
5162 if (ctx->pos >= INT_MAX)
5163 ctx->pos = LLONG_MAX;
5170 if (key_type == BTRFS_DIR_INDEX_KEY)
5171 btrfs_put_delayed_items(&ins_list, &del_list);
5172 btrfs_free_path(path);
5176 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5178 struct btrfs_root *root = BTRFS_I(inode)->root;
5179 struct btrfs_trans_handle *trans;
5181 bool nolock = false;
5183 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5186 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5189 if (wbc->sync_mode == WB_SYNC_ALL) {
5191 trans = btrfs_join_transaction_nolock(root);
5193 trans = btrfs_join_transaction(root);
5195 return PTR_ERR(trans);
5196 ret = btrfs_commit_transaction(trans, root);
5202 * This is somewhat expensive, updating the tree every time the
5203 * inode changes. But, it is most likely to find the inode in cache.
5204 * FIXME, needs more benchmarking...there are no reasons other than performance
5205 * to keep or drop this code.
5207 static int btrfs_dirty_inode(struct inode *inode)
5209 struct btrfs_root *root = BTRFS_I(inode)->root;
5210 struct btrfs_trans_handle *trans;
5213 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5216 trans = btrfs_join_transaction(root);
5218 return PTR_ERR(trans);
5220 ret = btrfs_update_inode(trans, root, inode);
5221 if (ret && ret == -ENOSPC) {
5222 /* whoops, lets try again with the full transaction */
5223 btrfs_end_transaction(trans, root);
5224 trans = btrfs_start_transaction(root, 1);
5226 return PTR_ERR(trans);
5228 ret = btrfs_update_inode(trans, root, inode);
5230 btrfs_end_transaction(trans, root);
5231 if (BTRFS_I(inode)->delayed_node)
5232 btrfs_balance_delayed_items(root);
5238 * This is a copy of file_update_time. We need this so we can return error on
5239 * ENOSPC for updating the inode in the case of file write and mmap writes.
5241 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5244 struct btrfs_root *root = BTRFS_I(inode)->root;
5246 if (btrfs_root_readonly(root))
5249 if (flags & S_VERSION)
5250 inode_inc_iversion(inode);
5251 if (flags & S_CTIME)
5252 inode->i_ctime = *now;
5253 if (flags & S_MTIME)
5254 inode->i_mtime = *now;
5255 if (flags & S_ATIME)
5256 inode->i_atime = *now;
5257 return btrfs_dirty_inode(inode);
5261 * find the highest existing sequence number in a directory
5262 * and then set the in-memory index_cnt variable to reflect
5263 * free sequence numbers
5265 static int btrfs_set_inode_index_count(struct inode *inode)
5267 struct btrfs_root *root = BTRFS_I(inode)->root;
5268 struct btrfs_key key, found_key;
5269 struct btrfs_path *path;
5270 struct extent_buffer *leaf;
5273 key.objectid = btrfs_ino(inode);
5274 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5275 key.offset = (u64)-1;
5277 path = btrfs_alloc_path();
5281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5284 /* FIXME: we should be able to handle this */
5290 * MAGIC NUMBER EXPLANATION:
5291 * since we search a directory based on f_pos we have to start at 2
5292 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5293 * else has to start at 2
5295 if (path->slots[0] == 0) {
5296 BTRFS_I(inode)->index_cnt = 2;
5302 leaf = path->nodes[0];
5303 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5305 if (found_key.objectid != btrfs_ino(inode) ||
5306 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5307 BTRFS_I(inode)->index_cnt = 2;
5311 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5313 btrfs_free_path(path);
5318 * helper to find a free sequence number in a given directory. This current
5319 * code is very simple, later versions will do smarter things in the btree
5321 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5325 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5326 ret = btrfs_inode_delayed_dir_index_count(dir);
5328 ret = btrfs_set_inode_index_count(dir);
5334 *index = BTRFS_I(dir)->index_cnt;
5335 BTRFS_I(dir)->index_cnt++;
5340 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5341 struct btrfs_root *root,
5343 const char *name, int name_len,
5344 u64 ref_objectid, u64 objectid,
5345 umode_t mode, u64 *index)
5347 struct inode *inode;
5348 struct btrfs_inode_item *inode_item;
5349 struct btrfs_key *location;
5350 struct btrfs_path *path;
5351 struct btrfs_inode_ref *ref;
5352 struct btrfs_key key[2];
5358 path = btrfs_alloc_path();
5360 return ERR_PTR(-ENOMEM);
5362 inode = new_inode(root->fs_info->sb);
5364 btrfs_free_path(path);
5365 return ERR_PTR(-ENOMEM);
5369 * we have to initialize this early, so we can reclaim the inode
5370 * number if we fail afterwards in this function.
5372 inode->i_ino = objectid;
5375 trace_btrfs_inode_request(dir);
5377 ret = btrfs_set_inode_index(dir, index);
5379 btrfs_free_path(path);
5381 return ERR_PTR(ret);
5385 * index_cnt is ignored for everything but a dir,
5386 * btrfs_get_inode_index_count has an explanation for the magic
5389 BTRFS_I(inode)->index_cnt = 2;
5390 BTRFS_I(inode)->root = root;
5391 BTRFS_I(inode)->generation = trans->transid;
5392 inode->i_generation = BTRFS_I(inode)->generation;
5395 * We could have gotten an inode number from somebody who was fsynced
5396 * and then removed in this same transaction, so let's just set full
5397 * sync since it will be a full sync anyway and this will blow away the
5398 * old info in the log.
5400 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5407 key[0].objectid = objectid;
5408 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5412 * Start new inodes with an inode_ref. This is slightly more
5413 * efficient for small numbers of hard links since they will
5414 * be packed into one item. Extended refs will kick in if we
5415 * add more hard links than can fit in the ref item.
5417 key[1].objectid = objectid;
5418 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5419 key[1].offset = ref_objectid;
5421 sizes[0] = sizeof(struct btrfs_inode_item);
5422 sizes[1] = name_len + sizeof(*ref);
5424 path->leave_spinning = 1;
5425 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5429 inode_init_owner(inode, dir, mode);
5430 inode_set_bytes(inode, 0);
5431 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5432 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5433 struct btrfs_inode_item);
5434 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5435 sizeof(*inode_item));
5436 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5438 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5439 struct btrfs_inode_ref);
5440 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5441 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5442 ptr = (unsigned long)(ref + 1);
5443 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5445 btrfs_mark_buffer_dirty(path->nodes[0]);
5446 btrfs_free_path(path);
5448 location = &BTRFS_I(inode)->location;
5449 location->objectid = objectid;
5450 location->offset = 0;
5451 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5453 btrfs_inherit_iflags(inode, dir);
5455 if (S_ISREG(mode)) {
5456 if (btrfs_test_opt(root, NODATASUM))
5457 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5458 if (btrfs_test_opt(root, NODATACOW))
5459 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5460 BTRFS_INODE_NODATASUM;
5463 btrfs_insert_inode_hash(inode);
5464 inode_tree_add(inode);
5466 trace_btrfs_inode_new(inode);
5467 btrfs_set_inode_last_trans(trans, inode);
5469 btrfs_update_root_times(trans, root);
5474 BTRFS_I(dir)->index_cnt--;
5475 btrfs_free_path(path);
5477 return ERR_PTR(ret);
5480 static inline u8 btrfs_inode_type(struct inode *inode)
5482 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5486 * utility function to add 'inode' into 'parent_inode' with
5487 * a give name and a given sequence number.
5488 * if 'add_backref' is true, also insert a backref from the
5489 * inode to the parent directory.
5491 int btrfs_add_link(struct btrfs_trans_handle *trans,
5492 struct inode *parent_inode, struct inode *inode,
5493 const char *name, int name_len, int add_backref, u64 index)
5496 struct btrfs_key key;
5497 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5498 u64 ino = btrfs_ino(inode);
5499 u64 parent_ino = btrfs_ino(parent_inode);
5501 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5502 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5505 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5509 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5510 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5511 key.objectid, root->root_key.objectid,
5512 parent_ino, index, name, name_len);
5513 } else if (add_backref) {
5514 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5518 /* Nothing to clean up yet */
5522 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5524 btrfs_inode_type(inode), index);
5525 if (ret == -EEXIST || ret == -EOVERFLOW)
5528 btrfs_abort_transaction(trans, root, ret);
5532 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5534 inode_inc_iversion(parent_inode);
5535 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5536 ret = btrfs_update_inode(trans, root, parent_inode);
5538 btrfs_abort_transaction(trans, root, ret);
5542 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5545 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5546 key.objectid, root->root_key.objectid,
5547 parent_ino, &local_index, name, name_len);
5549 } else if (add_backref) {
5553 err = btrfs_del_inode_ref(trans, root, name, name_len,
5554 ino, parent_ino, &local_index);
5559 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5560 struct inode *dir, struct dentry *dentry,
5561 struct inode *inode, int backref, u64 index)
5563 int err = btrfs_add_link(trans, dir, inode,
5564 dentry->d_name.name, dentry->d_name.len,
5571 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5572 umode_t mode, dev_t rdev)
5574 struct btrfs_trans_handle *trans;
5575 struct btrfs_root *root = BTRFS_I(dir)->root;
5576 struct inode *inode = NULL;
5582 if (!new_valid_dev(rdev))
5586 * 2 for inode item and ref
5588 * 1 for xattr if selinux is on
5590 trans = btrfs_start_transaction(root, 5);
5592 return PTR_ERR(trans);
5594 err = btrfs_find_free_ino(root, &objectid);
5598 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5599 dentry->d_name.len, btrfs_ino(dir), objectid,
5601 if (IS_ERR(inode)) {
5602 err = PTR_ERR(inode);
5606 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5613 * If the active LSM wants to access the inode during
5614 * d_instantiate it needs these. Smack checks to see
5615 * if the filesystem supports xattrs by looking at the
5619 inode->i_op = &btrfs_special_inode_operations;
5620 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5624 init_special_inode(inode, inode->i_mode, rdev);
5625 btrfs_update_inode(trans, root, inode);
5626 d_instantiate(dentry, inode);
5629 btrfs_end_transaction(trans, root);
5630 btrfs_btree_balance_dirty(root);
5632 inode_dec_link_count(inode);
5638 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5639 umode_t mode, bool excl)
5641 struct btrfs_trans_handle *trans;
5642 struct btrfs_root *root = BTRFS_I(dir)->root;
5643 struct inode *inode = NULL;
5644 int drop_inode_on_err = 0;
5650 * 2 for inode item and ref
5652 * 1 for xattr if selinux is on
5654 trans = btrfs_start_transaction(root, 5);
5656 return PTR_ERR(trans);
5658 err = btrfs_find_free_ino(root, &objectid);
5662 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5663 dentry->d_name.len, btrfs_ino(dir), objectid,
5665 if (IS_ERR(inode)) {
5666 err = PTR_ERR(inode);
5669 drop_inode_on_err = 1;
5671 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5675 err = btrfs_update_inode(trans, root, inode);
5680 * If the active LSM wants to access the inode during
5681 * d_instantiate it needs these. Smack checks to see
5682 * if the filesystem supports xattrs by looking at the
5685 inode->i_fop = &btrfs_file_operations;
5686 inode->i_op = &btrfs_file_inode_operations;
5688 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5692 inode->i_mapping->a_ops = &btrfs_aops;
5693 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5694 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5695 d_instantiate(dentry, inode);
5698 btrfs_end_transaction(trans, root);
5699 if (err && drop_inode_on_err) {
5700 inode_dec_link_count(inode);
5703 btrfs_btree_balance_dirty(root);
5707 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5708 struct dentry *dentry)
5710 struct btrfs_trans_handle *trans;
5711 struct btrfs_root *root = BTRFS_I(dir)->root;
5712 struct inode *inode = old_dentry->d_inode;
5717 /* do not allow sys_link's with other subvols of the same device */
5718 if (root->objectid != BTRFS_I(inode)->root->objectid)
5721 if (inode->i_nlink >= BTRFS_LINK_MAX)
5724 err = btrfs_set_inode_index(dir, &index);
5729 * 2 items for inode and inode ref
5730 * 2 items for dir items
5731 * 1 item for parent inode
5733 trans = btrfs_start_transaction(root, 5);
5734 if (IS_ERR(trans)) {
5735 err = PTR_ERR(trans);
5740 inode_inc_iversion(inode);
5741 inode->i_ctime = CURRENT_TIME;
5743 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5745 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5750 struct dentry *parent = dentry->d_parent;
5751 err = btrfs_update_inode(trans, root, inode);
5754 d_instantiate(dentry, inode);
5755 btrfs_log_new_name(trans, inode, NULL, parent);
5758 btrfs_end_transaction(trans, root);
5761 inode_dec_link_count(inode);
5764 btrfs_btree_balance_dirty(root);
5768 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5770 struct inode *inode = NULL;
5771 struct btrfs_trans_handle *trans;
5772 struct btrfs_root *root = BTRFS_I(dir)->root;
5774 int drop_on_err = 0;
5779 * 2 items for inode and ref
5780 * 2 items for dir items
5781 * 1 for xattr if selinux is on
5783 trans = btrfs_start_transaction(root, 5);
5785 return PTR_ERR(trans);
5787 err = btrfs_find_free_ino(root, &objectid);
5791 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5792 dentry->d_name.len, btrfs_ino(dir), objectid,
5793 S_IFDIR | mode, &index);
5794 if (IS_ERR(inode)) {
5795 err = PTR_ERR(inode);
5801 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5805 inode->i_op = &btrfs_dir_inode_operations;
5806 inode->i_fop = &btrfs_dir_file_operations;
5808 btrfs_i_size_write(inode, 0);
5809 err = btrfs_update_inode(trans, root, inode);
5813 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5814 dentry->d_name.len, 0, index);
5818 d_instantiate(dentry, inode);
5822 btrfs_end_transaction(trans, root);
5825 btrfs_btree_balance_dirty(root);
5829 /* helper for btfs_get_extent. Given an existing extent in the tree,
5830 * and an extent that you want to insert, deal with overlap and insert
5831 * the new extent into the tree.
5833 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5834 struct extent_map *existing,
5835 struct extent_map *em,
5836 u64 map_start, u64 map_len)
5840 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5841 start_diff = map_start - em->start;
5842 em->start = map_start;
5844 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5845 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5846 em->block_start += start_diff;
5847 em->block_len -= start_diff;
5849 return add_extent_mapping(em_tree, em, 0);
5852 static noinline int uncompress_inline(struct btrfs_path *path,
5853 struct inode *inode, struct page *page,
5854 size_t pg_offset, u64 extent_offset,
5855 struct btrfs_file_extent_item *item)
5858 struct extent_buffer *leaf = path->nodes[0];
5861 unsigned long inline_size;
5865 WARN_ON(pg_offset != 0);
5866 compress_type = btrfs_file_extent_compression(leaf, item);
5867 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5868 inline_size = btrfs_file_extent_inline_item_len(leaf,
5869 btrfs_item_nr(path->slots[0]));
5870 tmp = kmalloc(inline_size, GFP_NOFS);
5873 ptr = btrfs_file_extent_inline_start(item);
5875 read_extent_buffer(leaf, tmp, ptr, inline_size);
5877 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5878 ret = btrfs_decompress(compress_type, tmp, page,
5879 extent_offset, inline_size, max_size);
5881 char *kaddr = kmap_atomic(page);
5882 unsigned long copy_size = min_t(u64,
5883 PAGE_CACHE_SIZE - pg_offset,
5884 max_size - extent_offset);
5885 memset(kaddr + pg_offset, 0, copy_size);
5886 kunmap_atomic(kaddr);
5893 * a bit scary, this does extent mapping from logical file offset to the disk.
5894 * the ugly parts come from merging extents from the disk with the in-ram
5895 * representation. This gets more complex because of the data=ordered code,
5896 * where the in-ram extents might be locked pending data=ordered completion.
5898 * This also copies inline extents directly into the page.
5901 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5902 size_t pg_offset, u64 start, u64 len,
5908 u64 extent_start = 0;
5910 u64 objectid = btrfs_ino(inode);
5912 struct btrfs_path *path = NULL;
5913 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 struct btrfs_file_extent_item *item;
5915 struct extent_buffer *leaf;
5916 struct btrfs_key found_key;
5917 struct extent_map *em = NULL;
5918 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5919 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5920 struct btrfs_trans_handle *trans = NULL;
5924 read_lock(&em_tree->lock);
5925 em = lookup_extent_mapping(em_tree, start, len);
5927 em->bdev = root->fs_info->fs_devices->latest_bdev;
5928 read_unlock(&em_tree->lock);
5931 if (em->start > start || em->start + em->len <= start)
5932 free_extent_map(em);
5933 else if (em->block_start == EXTENT_MAP_INLINE && page)
5934 free_extent_map(em);
5938 em = alloc_extent_map();
5943 em->bdev = root->fs_info->fs_devices->latest_bdev;
5944 em->start = EXTENT_MAP_HOLE;
5945 em->orig_start = EXTENT_MAP_HOLE;
5947 em->block_len = (u64)-1;
5950 path = btrfs_alloc_path();
5956 * Chances are we'll be called again, so go ahead and do
5962 ret = btrfs_lookup_file_extent(trans, root, path,
5963 objectid, start, trans != NULL);
5970 if (path->slots[0] == 0)
5975 leaf = path->nodes[0];
5976 item = btrfs_item_ptr(leaf, path->slots[0],
5977 struct btrfs_file_extent_item);
5978 /* are we inside the extent that was found? */
5979 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5980 found_type = btrfs_key_type(&found_key);
5981 if (found_key.objectid != objectid ||
5982 found_type != BTRFS_EXTENT_DATA_KEY) {
5984 * If we backup past the first extent we want to move forward
5985 * and see if there is an extent in front of us, otherwise we'll
5986 * say there is a hole for our whole search range which can
5993 found_type = btrfs_file_extent_type(leaf, item);
5994 extent_start = found_key.offset;
5995 compress_type = btrfs_file_extent_compression(leaf, item);
5996 if (found_type == BTRFS_FILE_EXTENT_REG ||
5997 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5998 extent_end = extent_start +
5999 btrfs_file_extent_num_bytes(leaf, item);
6000 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6002 size = btrfs_file_extent_inline_len(leaf, item);
6003 extent_end = ALIGN(extent_start + size, root->sectorsize);
6006 if (start >= extent_end) {
6008 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6009 ret = btrfs_next_leaf(root, path);
6016 leaf = path->nodes[0];
6018 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6019 if (found_key.objectid != objectid ||
6020 found_key.type != BTRFS_EXTENT_DATA_KEY)
6022 if (start + len <= found_key.offset)
6025 em->orig_start = start;
6026 em->len = found_key.offset - start;
6030 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6031 if (found_type == BTRFS_FILE_EXTENT_REG ||
6032 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6033 em->start = extent_start;
6034 em->len = extent_end - extent_start;
6035 em->orig_start = extent_start -
6036 btrfs_file_extent_offset(leaf, item);
6037 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6039 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6041 em->block_start = EXTENT_MAP_HOLE;
6044 if (compress_type != BTRFS_COMPRESS_NONE) {
6045 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6046 em->compress_type = compress_type;
6047 em->block_start = bytenr;
6048 em->block_len = em->orig_block_len;
6050 bytenr += btrfs_file_extent_offset(leaf, item);
6051 em->block_start = bytenr;
6052 em->block_len = em->len;
6053 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6054 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6057 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6061 size_t extent_offset;
6064 em->block_start = EXTENT_MAP_INLINE;
6065 if (!page || create) {
6066 em->start = extent_start;
6067 em->len = extent_end - extent_start;
6071 size = btrfs_file_extent_inline_len(leaf, item);
6072 extent_offset = page_offset(page) + pg_offset - extent_start;
6073 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6074 size - extent_offset);
6075 em->start = extent_start + extent_offset;
6076 em->len = ALIGN(copy_size, root->sectorsize);
6077 em->orig_block_len = em->len;
6078 em->orig_start = em->start;
6079 if (compress_type) {
6080 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6081 em->compress_type = compress_type;
6083 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6084 if (create == 0 && !PageUptodate(page)) {
6085 if (btrfs_file_extent_compression(leaf, item) !=
6086 BTRFS_COMPRESS_NONE) {
6087 ret = uncompress_inline(path, inode, page,
6089 extent_offset, item);
6090 BUG_ON(ret); /* -ENOMEM */
6093 read_extent_buffer(leaf, map + pg_offset, ptr,
6095 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6096 memset(map + pg_offset + copy_size, 0,
6097 PAGE_CACHE_SIZE - pg_offset -
6102 flush_dcache_page(page);
6103 } else if (create && PageUptodate(page)) {
6107 free_extent_map(em);
6110 btrfs_release_path(path);
6111 trans = btrfs_join_transaction(root);
6114 return ERR_CAST(trans);
6118 write_extent_buffer(leaf, map + pg_offset, ptr,
6121 btrfs_mark_buffer_dirty(leaf);
6123 set_extent_uptodate(io_tree, em->start,
6124 extent_map_end(em) - 1, NULL, GFP_NOFS);
6127 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6131 em->orig_start = start;
6134 em->block_start = EXTENT_MAP_HOLE;
6135 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6137 btrfs_release_path(path);
6138 if (em->start > start || extent_map_end(em) <= start) {
6139 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6140 em->start, em->len, start, len);
6146 write_lock(&em_tree->lock);
6147 ret = add_extent_mapping(em_tree, em, 0);
6148 /* it is possible that someone inserted the extent into the tree
6149 * while we had the lock dropped. It is also possible that
6150 * an overlapping map exists in the tree
6152 if (ret == -EEXIST) {
6153 struct extent_map *existing;
6157 existing = lookup_extent_mapping(em_tree, start, len);
6158 if (existing && (existing->start > start ||
6159 existing->start + existing->len <= start)) {
6160 free_extent_map(existing);
6164 existing = lookup_extent_mapping(em_tree, em->start,
6167 err = merge_extent_mapping(em_tree, existing,
6170 free_extent_map(existing);
6172 free_extent_map(em);
6177 free_extent_map(em);
6181 free_extent_map(em);
6186 write_unlock(&em_tree->lock);
6190 trace_btrfs_get_extent(root, em);
6193 btrfs_free_path(path);
6195 ret = btrfs_end_transaction(trans, root);
6200 free_extent_map(em);
6201 return ERR_PTR(err);
6203 BUG_ON(!em); /* Error is always set */
6207 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6208 size_t pg_offset, u64 start, u64 len,
6211 struct extent_map *em;
6212 struct extent_map *hole_em = NULL;
6213 u64 range_start = start;
6219 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6226 * - a pre-alloc extent,
6227 * there might actually be delalloc bytes behind it.
6229 if (em->block_start != EXTENT_MAP_HOLE &&
6230 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6236 /* check to see if we've wrapped (len == -1 or similar) */
6245 /* ok, we didn't find anything, lets look for delalloc */
6246 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6247 end, len, EXTENT_DELALLOC, 1);
6248 found_end = range_start + found;
6249 if (found_end < range_start)
6250 found_end = (u64)-1;
6253 * we didn't find anything useful, return
6254 * the original results from get_extent()
6256 if (range_start > end || found_end <= start) {
6262 /* adjust the range_start to make sure it doesn't
6263 * go backwards from the start they passed in
6265 range_start = max(start, range_start);
6266 found = found_end - range_start;
6269 u64 hole_start = start;
6272 em = alloc_extent_map();
6278 * when btrfs_get_extent can't find anything it
6279 * returns one huge hole
6281 * make sure what it found really fits our range, and
6282 * adjust to make sure it is based on the start from
6286 u64 calc_end = extent_map_end(hole_em);
6288 if (calc_end <= start || (hole_em->start > end)) {
6289 free_extent_map(hole_em);
6292 hole_start = max(hole_em->start, start);
6293 hole_len = calc_end - hole_start;
6297 if (hole_em && range_start > hole_start) {
6298 /* our hole starts before our delalloc, so we
6299 * have to return just the parts of the hole
6300 * that go until the delalloc starts
6302 em->len = min(hole_len,
6303 range_start - hole_start);
6304 em->start = hole_start;
6305 em->orig_start = hole_start;
6307 * don't adjust block start at all,
6308 * it is fixed at EXTENT_MAP_HOLE
6310 em->block_start = hole_em->block_start;
6311 em->block_len = hole_len;
6312 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6313 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6315 em->start = range_start;
6317 em->orig_start = range_start;
6318 em->block_start = EXTENT_MAP_DELALLOC;
6319 em->block_len = found;
6321 } else if (hole_em) {
6326 free_extent_map(hole_em);
6328 free_extent_map(em);
6329 return ERR_PTR(err);
6334 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6337 struct btrfs_root *root = BTRFS_I(inode)->root;
6338 struct extent_map *em;
6339 struct btrfs_key ins;
6343 alloc_hint = get_extent_allocation_hint(inode, start, len);
6344 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6345 alloc_hint, &ins, 1);
6347 return ERR_PTR(ret);
6349 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6350 ins.offset, ins.offset, ins.offset, 0);
6352 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6356 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6357 ins.offset, ins.offset, 0);
6359 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6360 free_extent_map(em);
6361 return ERR_PTR(ret);
6368 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6369 * block must be cow'd
6371 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6372 u64 *orig_start, u64 *orig_block_len,
6375 struct btrfs_trans_handle *trans;
6376 struct btrfs_path *path;
6378 struct extent_buffer *leaf;
6379 struct btrfs_root *root = BTRFS_I(inode)->root;
6380 struct btrfs_file_extent_item *fi;
6381 struct btrfs_key key;
6388 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6389 path = btrfs_alloc_path();
6393 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6398 slot = path->slots[0];
6401 /* can't find the item, must cow */
6408 leaf = path->nodes[0];
6409 btrfs_item_key_to_cpu(leaf, &key, slot);
6410 if (key.objectid != btrfs_ino(inode) ||
6411 key.type != BTRFS_EXTENT_DATA_KEY) {
6412 /* not our file or wrong item type, must cow */
6416 if (key.offset > offset) {
6417 /* Wrong offset, must cow */
6421 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6422 found_type = btrfs_file_extent_type(leaf, fi);
6423 if (found_type != BTRFS_FILE_EXTENT_REG &&
6424 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6425 /* not a regular extent, must cow */
6429 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6432 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6433 if (disk_bytenr == 0)
6436 if (btrfs_file_extent_compression(leaf, fi) ||
6437 btrfs_file_extent_encryption(leaf, fi) ||
6438 btrfs_file_extent_other_encoding(leaf, fi))
6441 backref_offset = btrfs_file_extent_offset(leaf, fi);
6444 *orig_start = key.offset - backref_offset;
6445 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6446 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6449 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6451 if (btrfs_extent_readonly(root, disk_bytenr))
6453 btrfs_release_path(path);
6456 * look for other files referencing this extent, if we
6457 * find any we must cow
6459 trans = btrfs_join_transaction(root);
6460 if (IS_ERR(trans)) {
6465 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6466 key.offset - backref_offset, disk_bytenr);
6467 btrfs_end_transaction(trans, root);
6474 * adjust disk_bytenr and num_bytes to cover just the bytes
6475 * in this extent we are about to write. If there
6476 * are any csums in that range we have to cow in order
6477 * to keep the csums correct
6479 disk_bytenr += backref_offset;
6480 disk_bytenr += offset - key.offset;
6481 num_bytes = min(offset + *len, extent_end) - offset;
6482 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6485 * all of the above have passed, it is safe to overwrite this extent
6491 btrfs_free_path(path);
6495 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6496 struct extent_state **cached_state, int writing)
6498 struct btrfs_ordered_extent *ordered;
6502 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6505 * We're concerned with the entire range that we're going to be
6506 * doing DIO to, so we need to make sure theres no ordered
6507 * extents in this range.
6509 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6510 lockend - lockstart + 1);
6513 * We need to make sure there are no buffered pages in this
6514 * range either, we could have raced between the invalidate in
6515 * generic_file_direct_write and locking the extent. The
6516 * invalidate needs to happen so that reads after a write do not
6519 if (!ordered && (!writing ||
6520 !test_range_bit(&BTRFS_I(inode)->io_tree,
6521 lockstart, lockend, EXTENT_UPTODATE, 0,
6525 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6526 cached_state, GFP_NOFS);
6529 btrfs_start_ordered_extent(inode, ordered, 1);
6530 btrfs_put_ordered_extent(ordered);
6532 /* Screw you mmap */
6533 ret = filemap_write_and_wait_range(inode->i_mapping,
6540 * If we found a page that couldn't be invalidated just
6541 * fall back to buffered.
6543 ret = invalidate_inode_pages2_range(inode->i_mapping,
6544 lockstart >> PAGE_CACHE_SHIFT,
6545 lockend >> PAGE_CACHE_SHIFT);
6556 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6557 u64 len, u64 orig_start,
6558 u64 block_start, u64 block_len,
6559 u64 orig_block_len, u64 ram_bytes,
6562 struct extent_map_tree *em_tree;
6563 struct extent_map *em;
6564 struct btrfs_root *root = BTRFS_I(inode)->root;
6567 em_tree = &BTRFS_I(inode)->extent_tree;
6568 em = alloc_extent_map();
6570 return ERR_PTR(-ENOMEM);
6573 em->orig_start = orig_start;
6574 em->mod_start = start;
6577 em->block_len = block_len;
6578 em->block_start = block_start;
6579 em->bdev = root->fs_info->fs_devices->latest_bdev;
6580 em->orig_block_len = orig_block_len;
6581 em->ram_bytes = ram_bytes;
6582 em->generation = -1;
6583 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6584 if (type == BTRFS_ORDERED_PREALLOC)
6585 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6588 btrfs_drop_extent_cache(inode, em->start,
6589 em->start + em->len - 1, 0);
6590 write_lock(&em_tree->lock);
6591 ret = add_extent_mapping(em_tree, em, 1);
6592 write_unlock(&em_tree->lock);
6593 } while (ret == -EEXIST);
6596 free_extent_map(em);
6597 return ERR_PTR(ret);
6604 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6605 struct buffer_head *bh_result, int create)
6607 struct extent_map *em;
6608 struct btrfs_root *root = BTRFS_I(inode)->root;
6609 struct extent_state *cached_state = NULL;
6610 u64 start = iblock << inode->i_blkbits;
6611 u64 lockstart, lockend;
6612 u64 len = bh_result->b_size;
6613 int unlock_bits = EXTENT_LOCKED;
6617 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6619 len = min_t(u64, len, root->sectorsize);
6622 lockend = start + len - 1;
6625 * If this errors out it's because we couldn't invalidate pagecache for
6626 * this range and we need to fallback to buffered.
6628 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6631 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6638 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6639 * io. INLINE is special, and we could probably kludge it in here, but
6640 * it's still buffered so for safety lets just fall back to the generic
6643 * For COMPRESSED we _have_ to read the entire extent in so we can
6644 * decompress it, so there will be buffering required no matter what we
6645 * do, so go ahead and fallback to buffered.
6647 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6648 * to buffered IO. Don't blame me, this is the price we pay for using
6651 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6652 em->block_start == EXTENT_MAP_INLINE) {
6653 free_extent_map(em);
6658 /* Just a good old fashioned hole, return */
6659 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6660 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6661 free_extent_map(em);
6666 * We don't allocate a new extent in the following cases
6668 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6670 * 2) The extent is marked as PREALLOC. We're good to go here and can
6671 * just use the extent.
6675 len = min(len, em->len - (start - em->start));
6676 lockstart = start + len;
6680 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6681 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6682 em->block_start != EXTENT_MAP_HOLE)) {
6685 u64 block_start, orig_start, orig_block_len, ram_bytes;
6687 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6688 type = BTRFS_ORDERED_PREALLOC;
6690 type = BTRFS_ORDERED_NOCOW;
6691 len = min(len, em->len - (start - em->start));
6692 block_start = em->block_start + (start - em->start);
6694 if (can_nocow_extent(inode, start, &len, &orig_start,
6695 &orig_block_len, &ram_bytes) == 1) {
6696 if (type == BTRFS_ORDERED_PREALLOC) {
6697 free_extent_map(em);
6698 em = create_pinned_em(inode, start, len,
6707 ret = btrfs_add_ordered_extent_dio(inode, start,
6708 block_start, len, len, type);
6710 free_extent_map(em);
6718 * this will cow the extent, reset the len in case we changed
6721 len = bh_result->b_size;
6722 free_extent_map(em);
6723 em = btrfs_new_extent_direct(inode, start, len);
6728 len = min(len, em->len - (start - em->start));
6730 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6732 bh_result->b_size = len;
6733 bh_result->b_bdev = em->bdev;
6734 set_buffer_mapped(bh_result);
6736 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6737 set_buffer_new(bh_result);
6740 * Need to update the i_size under the extent lock so buffered
6741 * readers will get the updated i_size when we unlock.
6743 if (start + len > i_size_read(inode))
6744 i_size_write(inode, start + len);
6746 spin_lock(&BTRFS_I(inode)->lock);
6747 BTRFS_I(inode)->outstanding_extents++;
6748 spin_unlock(&BTRFS_I(inode)->lock);
6750 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6751 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6752 &cached_state, GFP_NOFS);
6757 * In the case of write we need to clear and unlock the entire range,
6758 * in the case of read we need to unlock only the end area that we
6759 * aren't using if there is any left over space.
6761 if (lockstart < lockend) {
6762 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6763 lockend, unlock_bits, 1, 0,
6764 &cached_state, GFP_NOFS);
6766 free_extent_state(cached_state);
6769 free_extent_map(em);
6774 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6775 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6779 static void btrfs_endio_direct_read(struct bio *bio, int err)
6781 struct btrfs_dio_private *dip = bio->bi_private;
6782 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6783 struct bio_vec *bvec = bio->bi_io_vec;
6784 struct inode *inode = dip->inode;
6785 struct btrfs_root *root = BTRFS_I(inode)->root;
6786 struct bio *dio_bio;
6787 u32 *csums = (u32 *)dip->csum;
6791 start = dip->logical_offset;
6793 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6794 struct page *page = bvec->bv_page;
6797 unsigned long flags;
6799 local_irq_save(flags);
6800 kaddr = kmap_atomic(page);
6801 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6802 csum, bvec->bv_len);
6803 btrfs_csum_final(csum, (char *)&csum);
6804 kunmap_atomic(kaddr);
6805 local_irq_restore(flags);
6807 flush_dcache_page(bvec->bv_page);
6808 if (csum != csums[index]) {
6809 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6810 btrfs_ino(inode), start, csum,
6816 start += bvec->bv_len;
6819 } while (bvec <= bvec_end);
6821 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6822 dip->logical_offset + dip->bytes - 1);
6823 dio_bio = dip->dio_bio;
6827 /* If we had a csum failure make sure to clear the uptodate flag */
6829 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6830 dio_end_io(dio_bio, err);
6834 static void btrfs_endio_direct_write(struct bio *bio, int err)
6836 struct btrfs_dio_private *dip = bio->bi_private;
6837 struct inode *inode = dip->inode;
6838 struct btrfs_root *root = BTRFS_I(inode)->root;
6839 struct btrfs_ordered_extent *ordered = NULL;
6840 u64 ordered_offset = dip->logical_offset;
6841 u64 ordered_bytes = dip->bytes;
6842 struct bio *dio_bio;
6848 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6850 ordered_bytes, !err);
6854 ordered->work.func = finish_ordered_fn;
6855 ordered->work.flags = 0;
6856 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6860 * our bio might span multiple ordered extents. If we haven't
6861 * completed the accounting for the whole dio, go back and try again
6863 if (ordered_offset < dip->logical_offset + dip->bytes) {
6864 ordered_bytes = dip->logical_offset + dip->bytes -
6870 dio_bio = dip->dio_bio;
6874 /* If we had an error make sure to clear the uptodate flag */
6876 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6877 dio_end_io(dio_bio, err);
6881 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6882 struct bio *bio, int mirror_num,
6883 unsigned long bio_flags, u64 offset)
6886 struct btrfs_root *root = BTRFS_I(inode)->root;
6887 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6888 BUG_ON(ret); /* -ENOMEM */
6892 static void btrfs_end_dio_bio(struct bio *bio, int err)
6894 struct btrfs_dio_private *dip = bio->bi_private;
6897 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6898 "sector %#Lx len %u err no %d\n",
6899 btrfs_ino(dip->inode), bio->bi_rw,
6900 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6904 * before atomic variable goto zero, we must make sure
6905 * dip->errors is perceived to be set.
6907 smp_mb__before_atomic_dec();
6910 /* if there are more bios still pending for this dio, just exit */
6911 if (!atomic_dec_and_test(&dip->pending_bios))
6915 bio_io_error(dip->orig_bio);
6917 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6918 bio_endio(dip->orig_bio, 0);
6924 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6925 u64 first_sector, gfp_t gfp_flags)
6927 int nr_vecs = bio_get_nr_vecs(bdev);
6928 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6931 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6932 int rw, u64 file_offset, int skip_sum,
6935 struct btrfs_dio_private *dip = bio->bi_private;
6936 int write = rw & REQ_WRITE;
6937 struct btrfs_root *root = BTRFS_I(inode)->root;
6941 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6946 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6954 if (write && async_submit) {
6955 ret = btrfs_wq_submit_bio(root->fs_info,
6956 inode, rw, bio, 0, 0,
6958 __btrfs_submit_bio_start_direct_io,
6959 __btrfs_submit_bio_done);
6963 * If we aren't doing async submit, calculate the csum of the
6966 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6969 } else if (!skip_sum) {
6970 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
6977 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6983 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6986 struct inode *inode = dip->inode;
6987 struct btrfs_root *root = BTRFS_I(inode)->root;
6989 struct bio *orig_bio = dip->orig_bio;
6990 struct bio_vec *bvec = orig_bio->bi_io_vec;
6991 u64 start_sector = orig_bio->bi_sector;
6992 u64 file_offset = dip->logical_offset;
6997 int async_submit = 0;
6999 map_length = orig_bio->bi_size;
7000 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7001 &map_length, NULL, 0);
7007 if (map_length >= orig_bio->bi_size) {
7012 /* async crcs make it difficult to collect full stripe writes. */
7013 if (btrfs_get_alloc_profile(root, 1) &
7014 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7019 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7022 bio->bi_private = dip;
7023 bio->bi_end_io = btrfs_end_dio_bio;
7024 atomic_inc(&dip->pending_bios);
7026 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7027 if (unlikely(map_length < submit_len + bvec->bv_len ||
7028 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7029 bvec->bv_offset) < bvec->bv_len)) {
7031 * inc the count before we submit the bio so
7032 * we know the end IO handler won't happen before
7033 * we inc the count. Otherwise, the dip might get freed
7034 * before we're done setting it up
7036 atomic_inc(&dip->pending_bios);
7037 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7038 file_offset, skip_sum,
7042 atomic_dec(&dip->pending_bios);
7046 start_sector += submit_len >> 9;
7047 file_offset += submit_len;
7052 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7053 start_sector, GFP_NOFS);
7056 bio->bi_private = dip;
7057 bio->bi_end_io = btrfs_end_dio_bio;
7059 map_length = orig_bio->bi_size;
7060 ret = btrfs_map_block(root->fs_info, rw,
7062 &map_length, NULL, 0);
7068 submit_len += bvec->bv_len;
7075 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7084 * before atomic variable goto zero, we must
7085 * make sure dip->errors is perceived to be set.
7087 smp_mb__before_atomic_dec();
7088 if (atomic_dec_and_test(&dip->pending_bios))
7089 bio_io_error(dip->orig_bio);
7091 /* bio_end_io() will handle error, so we needn't return it */
7095 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7096 struct inode *inode, loff_t file_offset)
7098 struct btrfs_root *root = BTRFS_I(inode)->root;
7099 struct btrfs_dio_private *dip;
7103 int write = rw & REQ_WRITE;
7107 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7109 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7115 if (!skip_sum && !write) {
7116 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7117 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7118 sum_len *= csum_size;
7123 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7129 dip->private = dio_bio->bi_private;
7131 dip->logical_offset = file_offset;
7132 dip->bytes = dio_bio->bi_size;
7133 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7134 io_bio->bi_private = dip;
7136 dip->orig_bio = io_bio;
7137 dip->dio_bio = dio_bio;
7138 atomic_set(&dip->pending_bios, 0);
7141 io_bio->bi_end_io = btrfs_endio_direct_write;
7143 io_bio->bi_end_io = btrfs_endio_direct_read;
7145 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7154 * If this is a write, we need to clean up the reserved space and kill
7155 * the ordered extent.
7158 struct btrfs_ordered_extent *ordered;
7159 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7160 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7161 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7162 btrfs_free_reserved_extent(root, ordered->start,
7164 btrfs_put_ordered_extent(ordered);
7165 btrfs_put_ordered_extent(ordered);
7167 bio_endio(dio_bio, ret);
7170 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7171 const struct iovec *iov, loff_t offset,
7172 unsigned long nr_segs)
7178 unsigned blocksize_mask = root->sectorsize - 1;
7179 ssize_t retval = -EINVAL;
7180 loff_t end = offset;
7182 if (offset & blocksize_mask)
7185 /* Check the memory alignment. Blocks cannot straddle pages */
7186 for (seg = 0; seg < nr_segs; seg++) {
7187 addr = (unsigned long)iov[seg].iov_base;
7188 size = iov[seg].iov_len;
7190 if ((addr & blocksize_mask) || (size & blocksize_mask))
7193 /* If this is a write we don't need to check anymore */
7198 * Check to make sure we don't have duplicate iov_base's in this
7199 * iovec, if so return EINVAL, otherwise we'll get csum errors
7200 * when reading back.
7202 for (i = seg + 1; i < nr_segs; i++) {
7203 if (iov[seg].iov_base == iov[i].iov_base)
7212 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7213 const struct iovec *iov, loff_t offset,
7214 unsigned long nr_segs)
7216 struct file *file = iocb->ki_filp;
7217 struct inode *inode = file->f_mapping->host;
7221 bool relock = false;
7224 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7228 atomic_inc(&inode->i_dio_count);
7229 smp_mb__after_atomic_inc();
7232 * The generic stuff only does filemap_write_and_wait_range, which isn't
7233 * enough if we've written compressed pages to this area, so we need to
7234 * call btrfs_wait_ordered_range to make absolutely sure that any
7235 * outstanding dirty pages are on disk.
7237 count = iov_length(iov, nr_segs);
7238 ret = btrfs_wait_ordered_range(inode, offset, count);
7244 * If the write DIO is beyond the EOF, we need update
7245 * the isize, but it is protected by i_mutex. So we can
7246 * not unlock the i_mutex at this case.
7248 if (offset + count <= inode->i_size) {
7249 mutex_unlock(&inode->i_mutex);
7252 ret = btrfs_delalloc_reserve_space(inode, count);
7255 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7256 &BTRFS_I(inode)->runtime_flags))) {
7257 inode_dio_done(inode);
7258 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7262 ret = __blockdev_direct_IO(rw, iocb, inode,
7263 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7264 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7265 btrfs_submit_direct, flags);
7267 if (ret < 0 && ret != -EIOCBQUEUED)
7268 btrfs_delalloc_release_space(inode, count);
7269 else if (ret >= 0 && (size_t)ret < count)
7270 btrfs_delalloc_release_space(inode,
7271 count - (size_t)ret);
7273 btrfs_delalloc_release_metadata(inode, 0);
7277 inode_dio_done(inode);
7279 mutex_lock(&inode->i_mutex);
7284 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7286 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7287 __u64 start, __u64 len)
7291 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7295 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7298 int btrfs_readpage(struct file *file, struct page *page)
7300 struct extent_io_tree *tree;
7301 tree = &BTRFS_I(page->mapping->host)->io_tree;
7302 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7305 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7307 struct extent_io_tree *tree;
7310 if (current->flags & PF_MEMALLOC) {
7311 redirty_page_for_writepage(wbc, page);
7315 tree = &BTRFS_I(page->mapping->host)->io_tree;
7316 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7319 static int btrfs_writepages(struct address_space *mapping,
7320 struct writeback_control *wbc)
7322 struct extent_io_tree *tree;
7324 tree = &BTRFS_I(mapping->host)->io_tree;
7325 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7329 btrfs_readpages(struct file *file, struct address_space *mapping,
7330 struct list_head *pages, unsigned nr_pages)
7332 struct extent_io_tree *tree;
7333 tree = &BTRFS_I(mapping->host)->io_tree;
7334 return extent_readpages(tree, mapping, pages, nr_pages,
7337 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7339 struct extent_io_tree *tree;
7340 struct extent_map_tree *map;
7343 tree = &BTRFS_I(page->mapping->host)->io_tree;
7344 map = &BTRFS_I(page->mapping->host)->extent_tree;
7345 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7347 ClearPagePrivate(page);
7348 set_page_private(page, 0);
7349 page_cache_release(page);
7354 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7356 if (PageWriteback(page) || PageDirty(page))
7358 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7361 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7362 unsigned int length)
7364 struct inode *inode = page->mapping->host;
7365 struct extent_io_tree *tree;
7366 struct btrfs_ordered_extent *ordered;
7367 struct extent_state *cached_state = NULL;
7368 u64 page_start = page_offset(page);
7369 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7372 * we have the page locked, so new writeback can't start,
7373 * and the dirty bit won't be cleared while we are here.
7375 * Wait for IO on this page so that we can safely clear
7376 * the PagePrivate2 bit and do ordered accounting
7378 wait_on_page_writeback(page);
7380 tree = &BTRFS_I(inode)->io_tree;
7382 btrfs_releasepage(page, GFP_NOFS);
7385 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7386 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7389 * IO on this page will never be started, so we need
7390 * to account for any ordered extents now
7392 clear_extent_bit(tree, page_start, page_end,
7393 EXTENT_DIRTY | EXTENT_DELALLOC |
7394 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7395 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7397 * whoever cleared the private bit is responsible
7398 * for the finish_ordered_io
7400 if (TestClearPagePrivate2(page)) {
7401 struct btrfs_ordered_inode_tree *tree;
7404 tree = &BTRFS_I(inode)->ordered_tree;
7406 spin_lock_irq(&tree->lock);
7407 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7408 new_len = page_start - ordered->file_offset;
7409 if (new_len < ordered->truncated_len)
7410 ordered->truncated_len = new_len;
7411 spin_unlock_irq(&tree->lock);
7413 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7415 PAGE_CACHE_SIZE, 1))
7416 btrfs_finish_ordered_io(ordered);
7418 btrfs_put_ordered_extent(ordered);
7419 cached_state = NULL;
7420 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7422 clear_extent_bit(tree, page_start, page_end,
7423 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7424 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7425 &cached_state, GFP_NOFS);
7426 __btrfs_releasepage(page, GFP_NOFS);
7428 ClearPageChecked(page);
7429 if (PagePrivate(page)) {
7430 ClearPagePrivate(page);
7431 set_page_private(page, 0);
7432 page_cache_release(page);
7437 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7438 * called from a page fault handler when a page is first dirtied. Hence we must
7439 * be careful to check for EOF conditions here. We set the page up correctly
7440 * for a written page which means we get ENOSPC checking when writing into
7441 * holes and correct delalloc and unwritten extent mapping on filesystems that
7442 * support these features.
7444 * We are not allowed to take the i_mutex here so we have to play games to
7445 * protect against truncate races as the page could now be beyond EOF. Because
7446 * vmtruncate() writes the inode size before removing pages, once we have the
7447 * page lock we can determine safely if the page is beyond EOF. If it is not
7448 * beyond EOF, then the page is guaranteed safe against truncation until we
7451 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7453 struct page *page = vmf->page;
7454 struct inode *inode = file_inode(vma->vm_file);
7455 struct btrfs_root *root = BTRFS_I(inode)->root;
7456 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7457 struct btrfs_ordered_extent *ordered;
7458 struct extent_state *cached_state = NULL;
7460 unsigned long zero_start;
7467 sb_start_pagefault(inode->i_sb);
7468 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7470 ret = file_update_time(vma->vm_file);
7476 else /* -ENOSPC, -EIO, etc */
7477 ret = VM_FAULT_SIGBUS;
7483 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7486 size = i_size_read(inode);
7487 page_start = page_offset(page);
7488 page_end = page_start + PAGE_CACHE_SIZE - 1;
7490 if ((page->mapping != inode->i_mapping) ||
7491 (page_start >= size)) {
7492 /* page got truncated out from underneath us */
7495 wait_on_page_writeback(page);
7497 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7498 set_page_extent_mapped(page);
7501 * we can't set the delalloc bits if there are pending ordered
7502 * extents. Drop our locks and wait for them to finish
7504 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7506 unlock_extent_cached(io_tree, page_start, page_end,
7507 &cached_state, GFP_NOFS);
7509 btrfs_start_ordered_extent(inode, ordered, 1);
7510 btrfs_put_ordered_extent(ordered);
7515 * XXX - page_mkwrite gets called every time the page is dirtied, even
7516 * if it was already dirty, so for space accounting reasons we need to
7517 * clear any delalloc bits for the range we are fixing to save. There
7518 * is probably a better way to do this, but for now keep consistent with
7519 * prepare_pages in the normal write path.
7521 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7522 EXTENT_DIRTY | EXTENT_DELALLOC |
7523 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7524 0, 0, &cached_state, GFP_NOFS);
7526 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7529 unlock_extent_cached(io_tree, page_start, page_end,
7530 &cached_state, GFP_NOFS);
7531 ret = VM_FAULT_SIGBUS;
7536 /* page is wholly or partially inside EOF */
7537 if (page_start + PAGE_CACHE_SIZE > size)
7538 zero_start = size & ~PAGE_CACHE_MASK;
7540 zero_start = PAGE_CACHE_SIZE;
7542 if (zero_start != PAGE_CACHE_SIZE) {
7544 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7545 flush_dcache_page(page);
7548 ClearPageChecked(page);
7549 set_page_dirty(page);
7550 SetPageUptodate(page);
7552 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7553 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7554 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7556 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7560 sb_end_pagefault(inode->i_sb);
7561 return VM_FAULT_LOCKED;
7565 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7567 sb_end_pagefault(inode->i_sb);
7571 static int btrfs_truncate(struct inode *inode)
7573 struct btrfs_root *root = BTRFS_I(inode)->root;
7574 struct btrfs_block_rsv *rsv;
7577 struct btrfs_trans_handle *trans;
7578 u64 mask = root->sectorsize - 1;
7579 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7581 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7587 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7588 * 3 things going on here
7590 * 1) We need to reserve space for our orphan item and the space to
7591 * delete our orphan item. Lord knows we don't want to have a dangling
7592 * orphan item because we didn't reserve space to remove it.
7594 * 2) We need to reserve space to update our inode.
7596 * 3) We need to have something to cache all the space that is going to
7597 * be free'd up by the truncate operation, but also have some slack
7598 * space reserved in case it uses space during the truncate (thank you
7599 * very much snapshotting).
7601 * And we need these to all be seperate. The fact is we can use alot of
7602 * space doing the truncate, and we have no earthly idea how much space
7603 * we will use, so we need the truncate reservation to be seperate so it
7604 * doesn't end up using space reserved for updating the inode or
7605 * removing the orphan item. We also need to be able to stop the
7606 * transaction and start a new one, which means we need to be able to
7607 * update the inode several times, and we have no idea of knowing how
7608 * many times that will be, so we can't just reserve 1 item for the
7609 * entirety of the opration, so that has to be done seperately as well.
7610 * Then there is the orphan item, which does indeed need to be held on
7611 * to for the whole operation, and we need nobody to touch this reserved
7612 * space except the orphan code.
7614 * So that leaves us with
7616 * 1) root->orphan_block_rsv - for the orphan deletion.
7617 * 2) rsv - for the truncate reservation, which we will steal from the
7618 * transaction reservation.
7619 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7620 * updating the inode.
7622 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7625 rsv->size = min_size;
7629 * 1 for the truncate slack space
7630 * 1 for updating the inode.
7632 trans = btrfs_start_transaction(root, 2);
7633 if (IS_ERR(trans)) {
7634 err = PTR_ERR(trans);
7638 /* Migrate the slack space for the truncate to our reserve */
7639 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7644 * setattr is responsible for setting the ordered_data_close flag,
7645 * but that is only tested during the last file release. That
7646 * could happen well after the next commit, leaving a great big
7647 * window where new writes may get lost if someone chooses to write
7648 * to this file after truncating to zero
7650 * The inode doesn't have any dirty data here, and so if we commit
7651 * this is a noop. If someone immediately starts writing to the inode
7652 * it is very likely we'll catch some of their writes in this
7653 * transaction, and the commit will find this file on the ordered
7654 * data list with good things to send down.
7656 * This is a best effort solution, there is still a window where
7657 * using truncate to replace the contents of the file will
7658 * end up with a zero length file after a crash.
7660 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7661 &BTRFS_I(inode)->runtime_flags))
7662 btrfs_add_ordered_operation(trans, root, inode);
7665 * So if we truncate and then write and fsync we normally would just
7666 * write the extents that changed, which is a problem if we need to
7667 * first truncate that entire inode. So set this flag so we write out
7668 * all of the extents in the inode to the sync log so we're completely
7671 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7672 trans->block_rsv = rsv;
7675 ret = btrfs_truncate_inode_items(trans, root, inode,
7677 BTRFS_EXTENT_DATA_KEY);
7678 if (ret != -ENOSPC) {
7683 trans->block_rsv = &root->fs_info->trans_block_rsv;
7684 ret = btrfs_update_inode(trans, root, inode);
7690 btrfs_end_transaction(trans, root);
7691 btrfs_btree_balance_dirty(root);
7693 trans = btrfs_start_transaction(root, 2);
7694 if (IS_ERR(trans)) {
7695 ret = err = PTR_ERR(trans);
7700 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7702 BUG_ON(ret); /* shouldn't happen */
7703 trans->block_rsv = rsv;
7706 if (ret == 0 && inode->i_nlink > 0) {
7707 trans->block_rsv = root->orphan_block_rsv;
7708 ret = btrfs_orphan_del(trans, inode);
7714 trans->block_rsv = &root->fs_info->trans_block_rsv;
7715 ret = btrfs_update_inode(trans, root, inode);
7719 ret = btrfs_end_transaction(trans, root);
7720 btrfs_btree_balance_dirty(root);
7724 btrfs_free_block_rsv(root, rsv);
7733 * create a new subvolume directory/inode (helper for the ioctl).
7735 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7736 struct btrfs_root *new_root, u64 new_dirid)
7738 struct inode *inode;
7742 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7743 new_dirid, new_dirid,
7744 S_IFDIR | (~current_umask() & S_IRWXUGO),
7747 return PTR_ERR(inode);
7748 inode->i_op = &btrfs_dir_inode_operations;
7749 inode->i_fop = &btrfs_dir_file_operations;
7751 set_nlink(inode, 1);
7752 btrfs_i_size_write(inode, 0);
7754 err = btrfs_update_inode(trans, new_root, inode);
7760 struct inode *btrfs_alloc_inode(struct super_block *sb)
7762 struct btrfs_inode *ei;
7763 struct inode *inode;
7765 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7772 ei->last_sub_trans = 0;
7773 ei->logged_trans = 0;
7774 ei->delalloc_bytes = 0;
7775 ei->disk_i_size = 0;
7778 ei->index_cnt = (u64)-1;
7779 ei->last_unlink_trans = 0;
7780 ei->last_log_commit = 0;
7782 spin_lock_init(&ei->lock);
7783 ei->outstanding_extents = 0;
7784 ei->reserved_extents = 0;
7786 ei->runtime_flags = 0;
7787 ei->force_compress = BTRFS_COMPRESS_NONE;
7789 ei->delayed_node = NULL;
7791 inode = &ei->vfs_inode;
7792 extent_map_tree_init(&ei->extent_tree);
7793 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7794 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7795 ei->io_tree.track_uptodate = 1;
7796 ei->io_failure_tree.track_uptodate = 1;
7797 atomic_set(&ei->sync_writers, 0);
7798 mutex_init(&ei->log_mutex);
7799 mutex_init(&ei->delalloc_mutex);
7800 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7801 INIT_LIST_HEAD(&ei->delalloc_inodes);
7802 INIT_LIST_HEAD(&ei->ordered_operations);
7803 RB_CLEAR_NODE(&ei->rb_node);
7808 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7809 void btrfs_test_destroy_inode(struct inode *inode)
7811 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7812 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7816 static void btrfs_i_callback(struct rcu_head *head)
7818 struct inode *inode = container_of(head, struct inode, i_rcu);
7819 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7822 void btrfs_destroy_inode(struct inode *inode)
7824 struct btrfs_ordered_extent *ordered;
7825 struct btrfs_root *root = BTRFS_I(inode)->root;
7827 WARN_ON(!hlist_empty(&inode->i_dentry));
7828 WARN_ON(inode->i_data.nrpages);
7829 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7830 WARN_ON(BTRFS_I(inode)->reserved_extents);
7831 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7832 WARN_ON(BTRFS_I(inode)->csum_bytes);
7835 * This can happen where we create an inode, but somebody else also
7836 * created the same inode and we need to destroy the one we already
7843 * Make sure we're properly removed from the ordered operation
7847 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7848 spin_lock(&root->fs_info->ordered_root_lock);
7849 list_del_init(&BTRFS_I(inode)->ordered_operations);
7850 spin_unlock(&root->fs_info->ordered_root_lock);
7853 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7854 &BTRFS_I(inode)->runtime_flags)) {
7855 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7857 atomic_dec(&root->orphan_inodes);
7861 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7865 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7866 ordered->file_offset, ordered->len);
7867 btrfs_remove_ordered_extent(inode, ordered);
7868 btrfs_put_ordered_extent(ordered);
7869 btrfs_put_ordered_extent(ordered);
7872 inode_tree_del(inode);
7873 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7875 call_rcu(&inode->i_rcu, btrfs_i_callback);
7878 int btrfs_drop_inode(struct inode *inode)
7880 struct btrfs_root *root = BTRFS_I(inode)->root;
7885 /* the snap/subvol tree is on deleting */
7886 if (btrfs_root_refs(&root->root_item) == 0)
7889 return generic_drop_inode(inode);
7892 static void init_once(void *foo)
7894 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7896 inode_init_once(&ei->vfs_inode);
7899 void btrfs_destroy_cachep(void)
7902 * Make sure all delayed rcu free inodes are flushed before we
7906 if (btrfs_inode_cachep)
7907 kmem_cache_destroy(btrfs_inode_cachep);
7908 if (btrfs_trans_handle_cachep)
7909 kmem_cache_destroy(btrfs_trans_handle_cachep);
7910 if (btrfs_transaction_cachep)
7911 kmem_cache_destroy(btrfs_transaction_cachep);
7912 if (btrfs_path_cachep)
7913 kmem_cache_destroy(btrfs_path_cachep);
7914 if (btrfs_free_space_cachep)
7915 kmem_cache_destroy(btrfs_free_space_cachep);
7916 if (btrfs_delalloc_work_cachep)
7917 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7920 int btrfs_init_cachep(void)
7922 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7923 sizeof(struct btrfs_inode), 0,
7924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7925 if (!btrfs_inode_cachep)
7928 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7929 sizeof(struct btrfs_trans_handle), 0,
7930 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7931 if (!btrfs_trans_handle_cachep)
7934 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7935 sizeof(struct btrfs_transaction), 0,
7936 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7937 if (!btrfs_transaction_cachep)
7940 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7941 sizeof(struct btrfs_path), 0,
7942 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7943 if (!btrfs_path_cachep)
7946 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7947 sizeof(struct btrfs_free_space), 0,
7948 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7949 if (!btrfs_free_space_cachep)
7952 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7953 sizeof(struct btrfs_delalloc_work), 0,
7954 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7956 if (!btrfs_delalloc_work_cachep)
7961 btrfs_destroy_cachep();
7965 static int btrfs_getattr(struct vfsmount *mnt,
7966 struct dentry *dentry, struct kstat *stat)
7969 struct inode *inode = dentry->d_inode;
7970 u32 blocksize = inode->i_sb->s_blocksize;
7972 generic_fillattr(inode, stat);
7973 stat->dev = BTRFS_I(inode)->root->anon_dev;
7974 stat->blksize = PAGE_CACHE_SIZE;
7976 spin_lock(&BTRFS_I(inode)->lock);
7977 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7978 spin_unlock(&BTRFS_I(inode)->lock);
7979 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7980 ALIGN(delalloc_bytes, blocksize)) >> 9;
7984 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7985 struct inode *new_dir, struct dentry *new_dentry)
7987 struct btrfs_trans_handle *trans;
7988 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7989 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7990 struct inode *new_inode = new_dentry->d_inode;
7991 struct inode *old_inode = old_dentry->d_inode;
7992 struct timespec ctime = CURRENT_TIME;
7996 u64 old_ino = btrfs_ino(old_inode);
7998 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8001 /* we only allow rename subvolume link between subvolumes */
8002 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8005 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8006 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8009 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8010 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8014 /* check for collisions, even if the name isn't there */
8015 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8016 new_dentry->d_name.name,
8017 new_dentry->d_name.len);
8020 if (ret == -EEXIST) {
8022 * eexist without a new_inode */
8023 if (WARN_ON(!new_inode)) {
8027 /* maybe -EOVERFLOW */
8034 * we're using rename to replace one file with another.
8035 * and the replacement file is large. Start IO on it now so
8036 * we don't add too much work to the end of the transaction
8038 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8039 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8040 filemap_flush(old_inode->i_mapping);
8042 /* close the racy window with snapshot create/destroy ioctl */
8043 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8044 down_read(&root->fs_info->subvol_sem);
8046 * We want to reserve the absolute worst case amount of items. So if
8047 * both inodes are subvols and we need to unlink them then that would
8048 * require 4 item modifications, but if they are both normal inodes it
8049 * would require 5 item modifications, so we'll assume their normal
8050 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8051 * should cover the worst case number of items we'll modify.
8053 trans = btrfs_start_transaction(root, 11);
8054 if (IS_ERR(trans)) {
8055 ret = PTR_ERR(trans);
8060 btrfs_record_root_in_trans(trans, dest);
8062 ret = btrfs_set_inode_index(new_dir, &index);
8066 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8067 /* force full log commit if subvolume involved. */
8068 root->fs_info->last_trans_log_full_commit = trans->transid;
8070 ret = btrfs_insert_inode_ref(trans, dest,
8071 new_dentry->d_name.name,
8072 new_dentry->d_name.len,
8074 btrfs_ino(new_dir), index);
8078 * this is an ugly little race, but the rename is required
8079 * to make sure that if we crash, the inode is either at the
8080 * old name or the new one. pinning the log transaction lets
8081 * us make sure we don't allow a log commit to come in after
8082 * we unlink the name but before we add the new name back in.
8084 btrfs_pin_log_trans(root);
8087 * make sure the inode gets flushed if it is replacing
8090 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8091 btrfs_add_ordered_operation(trans, root, old_inode);
8093 inode_inc_iversion(old_dir);
8094 inode_inc_iversion(new_dir);
8095 inode_inc_iversion(old_inode);
8096 old_dir->i_ctime = old_dir->i_mtime = ctime;
8097 new_dir->i_ctime = new_dir->i_mtime = ctime;
8098 old_inode->i_ctime = ctime;
8100 if (old_dentry->d_parent != new_dentry->d_parent)
8101 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8103 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8104 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8105 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8106 old_dentry->d_name.name,
8107 old_dentry->d_name.len);
8109 ret = __btrfs_unlink_inode(trans, root, old_dir,
8110 old_dentry->d_inode,
8111 old_dentry->d_name.name,
8112 old_dentry->d_name.len);
8114 ret = btrfs_update_inode(trans, root, old_inode);
8117 btrfs_abort_transaction(trans, root, ret);
8122 inode_inc_iversion(new_inode);
8123 new_inode->i_ctime = CURRENT_TIME;
8124 if (unlikely(btrfs_ino(new_inode) ==
8125 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8126 root_objectid = BTRFS_I(new_inode)->location.objectid;
8127 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8129 new_dentry->d_name.name,
8130 new_dentry->d_name.len);
8131 BUG_ON(new_inode->i_nlink == 0);
8133 ret = btrfs_unlink_inode(trans, dest, new_dir,
8134 new_dentry->d_inode,
8135 new_dentry->d_name.name,
8136 new_dentry->d_name.len);
8138 if (!ret && new_inode->i_nlink == 0)
8139 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8141 btrfs_abort_transaction(trans, root, ret);
8146 ret = btrfs_add_link(trans, new_dir, old_inode,
8147 new_dentry->d_name.name,
8148 new_dentry->d_name.len, 0, index);
8150 btrfs_abort_transaction(trans, root, ret);
8154 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8155 struct dentry *parent = new_dentry->d_parent;
8156 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8157 btrfs_end_log_trans(root);
8160 btrfs_end_transaction(trans, root);
8162 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8163 up_read(&root->fs_info->subvol_sem);
8168 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8170 struct btrfs_delalloc_work *delalloc_work;
8171 struct inode *inode;
8173 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8175 inode = delalloc_work->inode;
8176 if (delalloc_work->wait) {
8177 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8179 filemap_flush(inode->i_mapping);
8180 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8181 &BTRFS_I(inode)->runtime_flags))
8182 filemap_flush(inode->i_mapping);
8185 if (delalloc_work->delay_iput)
8186 btrfs_add_delayed_iput(inode);
8189 complete(&delalloc_work->completion);
8192 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8193 int wait, int delay_iput)
8195 struct btrfs_delalloc_work *work;
8197 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8201 init_completion(&work->completion);
8202 INIT_LIST_HEAD(&work->list);
8203 work->inode = inode;
8205 work->delay_iput = delay_iput;
8206 work->work.func = btrfs_run_delalloc_work;
8211 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8213 wait_for_completion(&work->completion);
8214 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8218 * some fairly slow code that needs optimization. This walks the list
8219 * of all the inodes with pending delalloc and forces them to disk.
8221 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8223 struct btrfs_inode *binode;
8224 struct inode *inode;
8225 struct btrfs_delalloc_work *work, *next;
8226 struct list_head works;
8227 struct list_head splice;
8230 INIT_LIST_HEAD(&works);
8231 INIT_LIST_HEAD(&splice);
8233 spin_lock(&root->delalloc_lock);
8234 list_splice_init(&root->delalloc_inodes, &splice);
8235 while (!list_empty(&splice)) {
8236 binode = list_entry(splice.next, struct btrfs_inode,
8239 list_move_tail(&binode->delalloc_inodes,
8240 &root->delalloc_inodes);
8241 inode = igrab(&binode->vfs_inode);
8243 cond_resched_lock(&root->delalloc_lock);
8246 spin_unlock(&root->delalloc_lock);
8248 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8249 if (unlikely(!work)) {
8251 btrfs_add_delayed_iput(inode);
8257 list_add_tail(&work->list, &works);
8258 btrfs_queue_worker(&root->fs_info->flush_workers,
8262 spin_lock(&root->delalloc_lock);
8264 spin_unlock(&root->delalloc_lock);
8266 list_for_each_entry_safe(work, next, &works, list) {
8267 list_del_init(&work->list);
8268 btrfs_wait_and_free_delalloc_work(work);
8272 list_for_each_entry_safe(work, next, &works, list) {
8273 list_del_init(&work->list);
8274 btrfs_wait_and_free_delalloc_work(work);
8277 if (!list_empty_careful(&splice)) {
8278 spin_lock(&root->delalloc_lock);
8279 list_splice_tail(&splice, &root->delalloc_inodes);
8280 spin_unlock(&root->delalloc_lock);
8285 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8289 if (root->fs_info->sb->s_flags & MS_RDONLY)
8292 ret = __start_delalloc_inodes(root, delay_iput);
8294 * the filemap_flush will queue IO into the worker threads, but
8295 * we have to make sure the IO is actually started and that
8296 * ordered extents get created before we return
8298 atomic_inc(&root->fs_info->async_submit_draining);
8299 while (atomic_read(&root->fs_info->nr_async_submits) ||
8300 atomic_read(&root->fs_info->async_delalloc_pages)) {
8301 wait_event(root->fs_info->async_submit_wait,
8302 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8303 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8305 atomic_dec(&root->fs_info->async_submit_draining);
8309 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8311 struct btrfs_root *root;
8312 struct list_head splice;
8315 if (fs_info->sb->s_flags & MS_RDONLY)
8318 INIT_LIST_HEAD(&splice);
8320 spin_lock(&fs_info->delalloc_root_lock);
8321 list_splice_init(&fs_info->delalloc_roots, &splice);
8322 while (!list_empty(&splice)) {
8323 root = list_first_entry(&splice, struct btrfs_root,
8325 root = btrfs_grab_fs_root(root);
8327 list_move_tail(&root->delalloc_root,
8328 &fs_info->delalloc_roots);
8329 spin_unlock(&fs_info->delalloc_root_lock);
8331 ret = __start_delalloc_inodes(root, delay_iput);
8332 btrfs_put_fs_root(root);
8336 spin_lock(&fs_info->delalloc_root_lock);
8338 spin_unlock(&fs_info->delalloc_root_lock);
8340 atomic_inc(&fs_info->async_submit_draining);
8341 while (atomic_read(&fs_info->nr_async_submits) ||
8342 atomic_read(&fs_info->async_delalloc_pages)) {
8343 wait_event(fs_info->async_submit_wait,
8344 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8345 atomic_read(&fs_info->async_delalloc_pages) == 0));
8347 atomic_dec(&fs_info->async_submit_draining);
8350 if (!list_empty_careful(&splice)) {
8351 spin_lock(&fs_info->delalloc_root_lock);
8352 list_splice_tail(&splice, &fs_info->delalloc_roots);
8353 spin_unlock(&fs_info->delalloc_root_lock);
8358 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8359 const char *symname)
8361 struct btrfs_trans_handle *trans;
8362 struct btrfs_root *root = BTRFS_I(dir)->root;
8363 struct btrfs_path *path;
8364 struct btrfs_key key;
8365 struct inode *inode = NULL;
8373 struct btrfs_file_extent_item *ei;
8374 struct extent_buffer *leaf;
8376 name_len = strlen(symname);
8377 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8378 return -ENAMETOOLONG;
8381 * 2 items for inode item and ref
8382 * 2 items for dir items
8383 * 1 item for xattr if selinux is on
8385 trans = btrfs_start_transaction(root, 5);
8387 return PTR_ERR(trans);
8389 err = btrfs_find_free_ino(root, &objectid);
8393 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8394 dentry->d_name.len, btrfs_ino(dir), objectid,
8395 S_IFLNK|S_IRWXUGO, &index);
8396 if (IS_ERR(inode)) {
8397 err = PTR_ERR(inode);
8401 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8408 * If the active LSM wants to access the inode during
8409 * d_instantiate it needs these. Smack checks to see
8410 * if the filesystem supports xattrs by looking at the
8413 inode->i_fop = &btrfs_file_operations;
8414 inode->i_op = &btrfs_file_inode_operations;
8416 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8420 inode->i_mapping->a_ops = &btrfs_aops;
8421 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8422 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8427 path = btrfs_alloc_path();
8433 key.objectid = btrfs_ino(inode);
8435 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8436 datasize = btrfs_file_extent_calc_inline_size(name_len);
8437 err = btrfs_insert_empty_item(trans, root, path, &key,
8441 btrfs_free_path(path);
8444 leaf = path->nodes[0];
8445 ei = btrfs_item_ptr(leaf, path->slots[0],
8446 struct btrfs_file_extent_item);
8447 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8448 btrfs_set_file_extent_type(leaf, ei,
8449 BTRFS_FILE_EXTENT_INLINE);
8450 btrfs_set_file_extent_encryption(leaf, ei, 0);
8451 btrfs_set_file_extent_compression(leaf, ei, 0);
8452 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8453 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8455 ptr = btrfs_file_extent_inline_start(ei);
8456 write_extent_buffer(leaf, symname, ptr, name_len);
8457 btrfs_mark_buffer_dirty(leaf);
8458 btrfs_free_path(path);
8460 inode->i_op = &btrfs_symlink_inode_operations;
8461 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8462 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8463 inode_set_bytes(inode, name_len);
8464 btrfs_i_size_write(inode, name_len);
8465 err = btrfs_update_inode(trans, root, inode);
8471 d_instantiate(dentry, inode);
8472 btrfs_end_transaction(trans, root);
8474 inode_dec_link_count(inode);
8477 btrfs_btree_balance_dirty(root);
8481 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8482 u64 start, u64 num_bytes, u64 min_size,
8483 loff_t actual_len, u64 *alloc_hint,
8484 struct btrfs_trans_handle *trans)
8486 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8487 struct extent_map *em;
8488 struct btrfs_root *root = BTRFS_I(inode)->root;
8489 struct btrfs_key ins;
8490 u64 cur_offset = start;
8494 bool own_trans = true;
8498 while (num_bytes > 0) {
8500 trans = btrfs_start_transaction(root, 3);
8501 if (IS_ERR(trans)) {
8502 ret = PTR_ERR(trans);
8507 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8508 cur_bytes = max(cur_bytes, min_size);
8509 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8510 *alloc_hint, &ins, 1);
8513 btrfs_end_transaction(trans, root);
8517 ret = insert_reserved_file_extent(trans, inode,
8518 cur_offset, ins.objectid,
8519 ins.offset, ins.offset,
8520 ins.offset, 0, 0, 0,
8521 BTRFS_FILE_EXTENT_PREALLOC);
8523 btrfs_free_reserved_extent(root, ins.objectid,
8525 btrfs_abort_transaction(trans, root, ret);
8527 btrfs_end_transaction(trans, root);
8530 btrfs_drop_extent_cache(inode, cur_offset,
8531 cur_offset + ins.offset -1, 0);
8533 em = alloc_extent_map();
8535 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8536 &BTRFS_I(inode)->runtime_flags);
8540 em->start = cur_offset;
8541 em->orig_start = cur_offset;
8542 em->len = ins.offset;
8543 em->block_start = ins.objectid;
8544 em->block_len = ins.offset;
8545 em->orig_block_len = ins.offset;
8546 em->ram_bytes = ins.offset;
8547 em->bdev = root->fs_info->fs_devices->latest_bdev;
8548 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8549 em->generation = trans->transid;
8552 write_lock(&em_tree->lock);
8553 ret = add_extent_mapping(em_tree, em, 1);
8554 write_unlock(&em_tree->lock);
8557 btrfs_drop_extent_cache(inode, cur_offset,
8558 cur_offset + ins.offset - 1,
8561 free_extent_map(em);
8563 num_bytes -= ins.offset;
8564 cur_offset += ins.offset;
8565 *alloc_hint = ins.objectid + ins.offset;
8567 inode_inc_iversion(inode);
8568 inode->i_ctime = CURRENT_TIME;
8569 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8570 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8571 (actual_len > inode->i_size) &&
8572 (cur_offset > inode->i_size)) {
8573 if (cur_offset > actual_len)
8574 i_size = actual_len;
8576 i_size = cur_offset;
8577 i_size_write(inode, i_size);
8578 btrfs_ordered_update_i_size(inode, i_size, NULL);
8581 ret = btrfs_update_inode(trans, root, inode);
8584 btrfs_abort_transaction(trans, root, ret);
8586 btrfs_end_transaction(trans, root);
8591 btrfs_end_transaction(trans, root);
8596 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8597 u64 start, u64 num_bytes, u64 min_size,
8598 loff_t actual_len, u64 *alloc_hint)
8600 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8601 min_size, actual_len, alloc_hint,
8605 int btrfs_prealloc_file_range_trans(struct inode *inode,
8606 struct btrfs_trans_handle *trans, int mode,
8607 u64 start, u64 num_bytes, u64 min_size,
8608 loff_t actual_len, u64 *alloc_hint)
8610 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8611 min_size, actual_len, alloc_hint, trans);
8614 static int btrfs_set_page_dirty(struct page *page)
8616 return __set_page_dirty_nobuffers(page);
8619 static int btrfs_permission(struct inode *inode, int mask)
8621 struct btrfs_root *root = BTRFS_I(inode)->root;
8622 umode_t mode = inode->i_mode;
8624 if (mask & MAY_WRITE &&
8625 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8626 if (btrfs_root_readonly(root))
8628 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8631 return generic_permission(inode, mask);
8634 static const struct inode_operations btrfs_dir_inode_operations = {
8635 .getattr = btrfs_getattr,
8636 .lookup = btrfs_lookup,
8637 .create = btrfs_create,
8638 .unlink = btrfs_unlink,
8640 .mkdir = btrfs_mkdir,
8641 .rmdir = btrfs_rmdir,
8642 .rename = btrfs_rename,
8643 .symlink = btrfs_symlink,
8644 .setattr = btrfs_setattr,
8645 .mknod = btrfs_mknod,
8646 .setxattr = btrfs_setxattr,
8647 .getxattr = btrfs_getxattr,
8648 .listxattr = btrfs_listxattr,
8649 .removexattr = btrfs_removexattr,
8650 .permission = btrfs_permission,
8651 .get_acl = btrfs_get_acl,
8652 .update_time = btrfs_update_time,
8654 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8655 .lookup = btrfs_lookup,
8656 .permission = btrfs_permission,
8657 .get_acl = btrfs_get_acl,
8658 .update_time = btrfs_update_time,
8661 static const struct file_operations btrfs_dir_file_operations = {
8662 .llseek = generic_file_llseek,
8663 .read = generic_read_dir,
8664 .iterate = btrfs_real_readdir,
8665 .unlocked_ioctl = btrfs_ioctl,
8666 #ifdef CONFIG_COMPAT
8667 .compat_ioctl = btrfs_ioctl,
8669 .release = btrfs_release_file,
8670 .fsync = btrfs_sync_file,
8673 static struct extent_io_ops btrfs_extent_io_ops = {
8674 .fill_delalloc = run_delalloc_range,
8675 .submit_bio_hook = btrfs_submit_bio_hook,
8676 .merge_bio_hook = btrfs_merge_bio_hook,
8677 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8678 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8679 .writepage_start_hook = btrfs_writepage_start_hook,
8680 .set_bit_hook = btrfs_set_bit_hook,
8681 .clear_bit_hook = btrfs_clear_bit_hook,
8682 .merge_extent_hook = btrfs_merge_extent_hook,
8683 .split_extent_hook = btrfs_split_extent_hook,
8687 * btrfs doesn't support the bmap operation because swapfiles
8688 * use bmap to make a mapping of extents in the file. They assume
8689 * these extents won't change over the life of the file and they
8690 * use the bmap result to do IO directly to the drive.
8692 * the btrfs bmap call would return logical addresses that aren't
8693 * suitable for IO and they also will change frequently as COW
8694 * operations happen. So, swapfile + btrfs == corruption.
8696 * For now we're avoiding this by dropping bmap.
8698 static const struct address_space_operations btrfs_aops = {
8699 .readpage = btrfs_readpage,
8700 .writepage = btrfs_writepage,
8701 .writepages = btrfs_writepages,
8702 .readpages = btrfs_readpages,
8703 .direct_IO = btrfs_direct_IO,
8704 .invalidatepage = btrfs_invalidatepage,
8705 .releasepage = btrfs_releasepage,
8706 .set_page_dirty = btrfs_set_page_dirty,
8707 .error_remove_page = generic_error_remove_page,
8710 static const struct address_space_operations btrfs_symlink_aops = {
8711 .readpage = btrfs_readpage,
8712 .writepage = btrfs_writepage,
8713 .invalidatepage = btrfs_invalidatepage,
8714 .releasepage = btrfs_releasepage,
8717 static const struct inode_operations btrfs_file_inode_operations = {
8718 .getattr = btrfs_getattr,
8719 .setattr = btrfs_setattr,
8720 .setxattr = btrfs_setxattr,
8721 .getxattr = btrfs_getxattr,
8722 .listxattr = btrfs_listxattr,
8723 .removexattr = btrfs_removexattr,
8724 .permission = btrfs_permission,
8725 .fiemap = btrfs_fiemap,
8726 .get_acl = btrfs_get_acl,
8727 .update_time = btrfs_update_time,
8729 static const struct inode_operations btrfs_special_inode_operations = {
8730 .getattr = btrfs_getattr,
8731 .setattr = btrfs_setattr,
8732 .permission = btrfs_permission,
8733 .setxattr = btrfs_setxattr,
8734 .getxattr = btrfs_getxattr,
8735 .listxattr = btrfs_listxattr,
8736 .removexattr = btrfs_removexattr,
8737 .get_acl = btrfs_get_acl,
8738 .update_time = btrfs_update_time,
8740 static const struct inode_operations btrfs_symlink_inode_operations = {
8741 .readlink = generic_readlink,
8742 .follow_link = page_follow_link_light,
8743 .put_link = page_put_link,
8744 .getattr = btrfs_getattr,
8745 .setattr = btrfs_setattr,
8746 .permission = btrfs_permission,
8747 .setxattr = btrfs_setxattr,
8748 .getxattr = btrfs_getxattr,
8749 .listxattr = btrfs_listxattr,
8750 .removexattr = btrfs_removexattr,
8751 .get_acl = btrfs_get_acl,
8752 .update_time = btrfs_update_time,
8755 const struct dentry_operations btrfs_dentry_operations = {
8756 .d_delete = btrfs_dentry_delete,
8757 .d_release = btrfs_dentry_release,