2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.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"
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
69 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_dir_ro_inode_operations;
72 static const struct inode_operations btrfs_special_inode_operations;
73 static const struct inode_operations btrfs_file_inode_operations;
74 static const struct address_space_operations btrfs_aops;
75 static const struct address_space_operations btrfs_symlink_aops;
76 static const struct file_operations btrfs_dir_file_operations;
77 static struct extent_io_ops btrfs_extent_io_ops;
79 static struct kmem_cache *btrfs_inode_cachep;
80 static struct kmem_cache *btrfs_delalloc_work_cachep;
81 struct kmem_cache *btrfs_trans_handle_cachep;
82 struct kmem_cache *btrfs_transaction_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
88 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
89 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
90 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
91 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
92 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
93 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
94 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
98 static int btrfs_truncate(struct inode *inode);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
100 static noinline int cow_file_range(struct inode *inode,
101 struct page *locked_page,
102 u64 start, u64 end, int *page_started,
103 unsigned long *nr_written, int unlock);
104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
105 u64 len, u64 orig_start,
106 u64 block_start, u64 block_len,
107 u64 orig_block_len, u64 ram_bytes,
110 static int btrfs_dirty_inode(struct inode *inode);
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode *inode)
115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
120 struct inode *inode, struct inode *dir,
121 const struct qstr *qstr)
125 err = btrfs_init_acl(trans, inode, dir);
127 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_path *path, int extent_inserted,
138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
141 struct page **compressed_pages)
143 struct extent_buffer *leaf;
144 struct page *page = NULL;
147 struct btrfs_file_extent_item *ei;
150 size_t cur_size = size;
151 unsigned long offset;
153 if (compressed_size && compressed_pages)
154 cur_size = compressed_size;
156 inode_add_bytes(inode, size);
158 if (!extent_inserted) {
159 struct btrfs_key key;
162 key.objectid = btrfs_ino(inode);
164 key.type = BTRFS_EXTENT_DATA_KEY;
166 datasize = btrfs_file_extent_calc_inline_size(cur_size);
167 path->leave_spinning = 1;
168 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
178 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
179 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
180 btrfs_set_file_extent_encryption(leaf, ei, 0);
181 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
182 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
183 ptr = btrfs_file_extent_inline_start(ei);
185 if (compress_type != BTRFS_COMPRESS_NONE) {
188 while (compressed_size > 0) {
189 cpage = compressed_pages[i];
190 cur_size = min_t(unsigned long, compressed_size,
193 kaddr = kmap_atomic(cpage);
194 write_extent_buffer(leaf, kaddr, ptr, cur_size);
195 kunmap_atomic(kaddr);
199 compressed_size -= cur_size;
201 btrfs_set_file_extent_compression(leaf, ei,
204 page = find_get_page(inode->i_mapping,
205 start >> PAGE_CACHE_SHIFT);
206 btrfs_set_file_extent_compression(leaf, ei, 0);
207 kaddr = kmap_atomic(page);
208 offset = start & (PAGE_CACHE_SIZE - 1);
209 write_extent_buffer(leaf, kaddr + offset, ptr, size);
210 kunmap_atomic(kaddr);
211 page_cache_release(page);
213 btrfs_mark_buffer_dirty(leaf);
214 btrfs_release_path(path);
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
225 BTRFS_I(inode)->disk_i_size = inode->i_size;
226 ret = btrfs_update_inode(trans, root, inode);
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
239 static noinline int cow_file_range_inline(struct btrfs_root *root,
240 struct inode *inode, u64 start,
241 u64 end, size_t compressed_size,
243 struct page **compressed_pages)
245 struct btrfs_trans_handle *trans;
246 u64 isize = i_size_read(inode);
247 u64 actual_end = min(end + 1, isize);
248 u64 inline_len = actual_end - start;
249 u64 aligned_end = ALIGN(end, root->sectorsize);
250 u64 data_len = inline_len;
252 struct btrfs_path *path;
253 int extent_inserted = 0;
254 u32 extent_item_size;
257 data_len = compressed_size;
260 actual_end > PAGE_CACHE_SIZE ||
261 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
263 (actual_end & (root->sectorsize - 1)) == 0) ||
265 data_len > root->fs_info->max_inline) {
269 path = btrfs_alloc_path();
273 trans = btrfs_join_transaction(root);
275 btrfs_free_path(path);
276 return PTR_ERR(trans);
278 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
280 if (compressed_size && compressed_pages)
281 extent_item_size = btrfs_file_extent_calc_inline_size(
284 extent_item_size = btrfs_file_extent_calc_inline_size(
287 ret = __btrfs_drop_extents(trans, root, inode, path,
288 start, aligned_end, NULL,
289 1, 1, extent_item_size, &extent_inserted);
291 btrfs_abort_transaction(trans, root, ret);
295 if (isize > actual_end)
296 inline_len = min_t(u64, isize, actual_end);
297 ret = insert_inline_extent(trans, path, extent_inserted,
299 inline_len, compressed_size,
300 compress_type, compressed_pages);
301 if (ret && ret != -ENOSPC) {
302 btrfs_abort_transaction(trans, root, ret);
304 } else if (ret == -ENOSPC) {
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
310 btrfs_delalloc_release_metadata(inode, end + 1 - start);
311 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
314 * Don't forget to free the reserved space, as for inlined extent
315 * it won't count as data extent, free them directly here.
316 * And at reserve time, it's always aligned to page size, so
317 * just free one page here.
319 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
320 btrfs_free_path(path);
321 btrfs_end_transaction(trans, root);
325 struct async_extent {
330 unsigned long nr_pages;
332 struct list_head list;
337 struct btrfs_root *root;
338 struct page *locked_page;
341 struct list_head extents;
342 struct btrfs_work work;
345 static noinline int add_async_extent(struct async_cow *cow,
346 u64 start, u64 ram_size,
349 unsigned long nr_pages,
352 struct async_extent *async_extent;
354 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
355 BUG_ON(!async_extent); /* -ENOMEM */
356 async_extent->start = start;
357 async_extent->ram_size = ram_size;
358 async_extent->compressed_size = compressed_size;
359 async_extent->pages = pages;
360 async_extent->nr_pages = nr_pages;
361 async_extent->compress_type = compress_type;
362 list_add_tail(&async_extent->list, &cow->extents);
366 static inline int inode_need_compress(struct inode *inode)
368 struct btrfs_root *root = BTRFS_I(inode)->root;
371 if (btrfs_test_opt(root, FORCE_COMPRESS))
373 /* bad compression ratios */
374 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
376 if (btrfs_test_opt(root, COMPRESS) ||
377 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
378 BTRFS_I(inode)->force_compress)
384 * we create compressed extents in two phases. The first
385 * phase compresses a range of pages that have already been
386 * locked (both pages and state bits are locked).
388 * This is done inside an ordered work queue, and the compression
389 * is spread across many cpus. The actual IO submission is step
390 * two, and the ordered work queue takes care of making sure that
391 * happens in the same order things were put onto the queue by
392 * writepages and friends.
394 * If this code finds it can't get good compression, it puts an
395 * entry onto the work queue to write the uncompressed bytes. This
396 * makes sure that both compressed inodes and uncompressed inodes
397 * are written in the same order that the flusher thread sent them
400 static noinline void compress_file_range(struct inode *inode,
401 struct page *locked_page,
403 struct async_cow *async_cow,
406 struct btrfs_root *root = BTRFS_I(inode)->root;
408 u64 blocksize = root->sectorsize;
410 u64 isize = i_size_read(inode);
412 struct page **pages = NULL;
413 unsigned long nr_pages;
414 unsigned long nr_pages_ret = 0;
415 unsigned long total_compressed = 0;
416 unsigned long total_in = 0;
417 unsigned long max_compressed = 128 * 1024;
418 unsigned long max_uncompressed = 128 * 1024;
421 int compress_type = root->fs_info->compress_type;
424 /* if this is a small write inside eof, kick off a defrag */
425 if ((end - start + 1) < 16 * 1024 &&
426 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
427 btrfs_add_inode_defrag(NULL, inode);
429 actual_end = min_t(u64, isize, end + 1);
432 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
433 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
436 * we don't want to send crud past the end of i_size through
437 * compression, that's just a waste of CPU time. So, if the
438 * end of the file is before the start of our current
439 * requested range of bytes, we bail out to the uncompressed
440 * cleanup code that can deal with all of this.
442 * It isn't really the fastest way to fix things, but this is a
443 * very uncommon corner.
445 if (actual_end <= start)
446 goto cleanup_and_bail_uncompressed;
448 total_compressed = actual_end - start;
451 * skip compression for a small file range(<=blocksize) that
452 * isn't an inline extent, since it dosen't save disk space at all.
454 if (total_compressed <= blocksize &&
455 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
456 goto cleanup_and_bail_uncompressed;
458 /* we want to make sure that amount of ram required to uncompress
459 * an extent is reasonable, so we limit the total size in ram
460 * of a compressed extent to 128k. This is a crucial number
461 * because it also controls how easily we can spread reads across
462 * cpus for decompression.
464 * We also want to make sure the amount of IO required to do
465 * a random read is reasonably small, so we limit the size of
466 * a compressed extent to 128k.
468 total_compressed = min(total_compressed, max_uncompressed);
469 num_bytes = ALIGN(end - start + 1, blocksize);
470 num_bytes = max(blocksize, num_bytes);
475 * we do compression for mount -o compress and when the
476 * inode has not been flagged as nocompress. This flag can
477 * change at any time if we discover bad compression ratios.
479 if (inode_need_compress(inode)) {
481 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
483 /* just bail out to the uncompressed code */
487 if (BTRFS_I(inode)->force_compress)
488 compress_type = BTRFS_I(inode)->force_compress;
491 * we need to call clear_page_dirty_for_io on each
492 * page in the range. Otherwise applications with the file
493 * mmap'd can wander in and change the page contents while
494 * we are compressing them.
496 * If the compression fails for any reason, we set the pages
497 * dirty again later on.
499 extent_range_clear_dirty_for_io(inode, start, end);
501 ret = btrfs_compress_pages(compress_type,
502 inode->i_mapping, start,
503 total_compressed, pages,
504 nr_pages, &nr_pages_ret,
510 unsigned long offset = total_compressed &
511 (PAGE_CACHE_SIZE - 1);
512 struct page *page = pages[nr_pages_ret - 1];
515 /* zero the tail end of the last page, we might be
516 * sending it down to disk
519 kaddr = kmap_atomic(page);
520 memset(kaddr + offset, 0,
521 PAGE_CACHE_SIZE - offset);
522 kunmap_atomic(kaddr);
529 /* lets try to make an inline extent */
530 if (ret || total_in < (actual_end - start)) {
531 /* we didn't compress the entire range, try
532 * to make an uncompressed inline extent.
534 ret = cow_file_range_inline(root, inode, start, end,
537 /* try making a compressed inline extent */
538 ret = cow_file_range_inline(root, inode, start, end,
540 compress_type, pages);
543 unsigned long clear_flags = EXTENT_DELALLOC |
545 unsigned long page_error_op;
547 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
548 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
551 * inline extent creation worked or returned error,
552 * we don't need to create any more async work items.
553 * Unlock and free up our temp pages.
555 extent_clear_unlock_delalloc(inode, start, end, NULL,
556 clear_flags, PAGE_UNLOCK |
567 * we aren't doing an inline extent round the compressed size
568 * up to a block size boundary so the allocator does sane
571 total_compressed = ALIGN(total_compressed, blocksize);
574 * one last check to make sure the compression is really a
575 * win, compare the page count read with the blocks on disk
577 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
578 if (total_compressed >= total_in) {
581 num_bytes = total_in;
584 if (!will_compress && pages) {
586 * the compression code ran but failed to make things smaller,
587 * free any pages it allocated and our page pointer array
589 for (i = 0; i < nr_pages_ret; i++) {
590 WARN_ON(pages[i]->mapping);
591 page_cache_release(pages[i]);
595 total_compressed = 0;
598 /* flag the file so we don't compress in the future */
599 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
600 !(BTRFS_I(inode)->force_compress)) {
601 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
607 /* the async work queues will take care of doing actual
608 * allocation on disk for these compressed pages,
609 * and will submit them to the elevator.
611 add_async_extent(async_cow, start, num_bytes,
612 total_compressed, pages, nr_pages_ret,
615 if (start + num_bytes < end) {
622 cleanup_and_bail_uncompressed:
624 * No compression, but we still need to write the pages in
625 * the file we've been given so far. redirty the locked
626 * page if it corresponds to our extent and set things up
627 * for the async work queue to run cow_file_range to do
628 * the normal delalloc dance
630 if (page_offset(locked_page) >= start &&
631 page_offset(locked_page) <= end) {
632 __set_page_dirty_nobuffers(locked_page);
633 /* unlocked later on in the async handlers */
636 extent_range_redirty_for_io(inode, start, end);
637 add_async_extent(async_cow, start, end - start + 1,
638 0, NULL, 0, BTRFS_COMPRESS_NONE);
645 for (i = 0; i < nr_pages_ret; i++) {
646 WARN_ON(pages[i]->mapping);
647 page_cache_release(pages[i]);
652 static void free_async_extent_pages(struct async_extent *async_extent)
656 if (!async_extent->pages)
659 for (i = 0; i < async_extent->nr_pages; i++) {
660 WARN_ON(async_extent->pages[i]->mapping);
661 page_cache_release(async_extent->pages[i]);
663 kfree(async_extent->pages);
664 async_extent->nr_pages = 0;
665 async_extent->pages = NULL;
669 * phase two of compressed writeback. This is the ordered portion
670 * of the code, which only gets called in the order the work was
671 * queued. We walk all the async extents created by compress_file_range
672 * and send them down to the disk.
674 static noinline void submit_compressed_extents(struct inode *inode,
675 struct async_cow *async_cow)
677 struct async_extent *async_extent;
679 struct btrfs_key ins;
680 struct extent_map *em;
681 struct btrfs_root *root = BTRFS_I(inode)->root;
682 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
683 struct extent_io_tree *io_tree;
687 while (!list_empty(&async_cow->extents)) {
688 async_extent = list_entry(async_cow->extents.next,
689 struct async_extent, list);
690 list_del(&async_extent->list);
692 io_tree = &BTRFS_I(inode)->io_tree;
695 /* did the compression code fall back to uncompressed IO? */
696 if (!async_extent->pages) {
697 int page_started = 0;
698 unsigned long nr_written = 0;
700 lock_extent(io_tree, async_extent->start,
701 async_extent->start +
702 async_extent->ram_size - 1);
704 /* allocate blocks */
705 ret = cow_file_range(inode, async_cow->locked_page,
707 async_extent->start +
708 async_extent->ram_size - 1,
709 &page_started, &nr_written, 0);
714 * if page_started, cow_file_range inserted an
715 * inline extent and took care of all the unlocking
716 * and IO for us. Otherwise, we need to submit
717 * all those pages down to the drive.
719 if (!page_started && !ret)
720 extent_write_locked_range(io_tree,
721 inode, async_extent->start,
722 async_extent->start +
723 async_extent->ram_size - 1,
727 unlock_page(async_cow->locked_page);
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start + async_extent->ram_size - 1);
736 ret = btrfs_reserve_extent(root,
737 async_extent->compressed_size,
738 async_extent->compressed_size,
739 0, alloc_hint, &ins, 1, 1);
741 free_async_extent_pages(async_extent);
743 if (ret == -ENOSPC) {
744 unlock_extent(io_tree, async_extent->start,
745 async_extent->start +
746 async_extent->ram_size - 1);
749 * we need to redirty the pages if we decide to
750 * fallback to uncompressed IO, otherwise we
751 * will not submit these pages down to lower
754 extent_range_redirty_for_io(inode,
756 async_extent->start +
757 async_extent->ram_size - 1);
764 * here we're doing allocation and writeback of the
767 btrfs_drop_extent_cache(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1, 0);
771 em = alloc_extent_map();
774 goto out_free_reserve;
776 em->start = async_extent->start;
777 em->len = async_extent->ram_size;
778 em->orig_start = em->start;
779 em->mod_start = em->start;
780 em->mod_len = em->len;
782 em->block_start = ins.objectid;
783 em->block_len = ins.offset;
784 em->orig_block_len = ins.offset;
785 em->ram_bytes = async_extent->ram_size;
786 em->bdev = root->fs_info->fs_devices->latest_bdev;
787 em->compress_type = async_extent->compress_type;
788 set_bit(EXTENT_FLAG_PINNED, &em->flags);
789 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
793 write_lock(&em_tree->lock);
794 ret = add_extent_mapping(em_tree, em, 1);
795 write_unlock(&em_tree->lock);
796 if (ret != -EEXIST) {
800 btrfs_drop_extent_cache(inode, async_extent->start,
801 async_extent->start +
802 async_extent->ram_size - 1, 0);
806 goto out_free_reserve;
808 ret = btrfs_add_ordered_extent_compress(inode,
811 async_extent->ram_size,
813 BTRFS_ORDERED_COMPRESSED,
814 async_extent->compress_type);
816 btrfs_drop_extent_cache(inode, async_extent->start,
817 async_extent->start +
818 async_extent->ram_size - 1, 0);
819 goto out_free_reserve;
823 * clear dirty, set writeback and unlock the pages.
825 extent_clear_unlock_delalloc(inode, async_extent->start,
826 async_extent->start +
827 async_extent->ram_size - 1,
828 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
829 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
831 ret = btrfs_submit_compressed_write(inode,
833 async_extent->ram_size,
835 ins.offset, async_extent->pages,
836 async_extent->nr_pages);
838 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
839 struct page *p = async_extent->pages[0];
840 const u64 start = async_extent->start;
841 const u64 end = start + async_extent->ram_size - 1;
843 p->mapping = inode->i_mapping;
844 tree->ops->writepage_end_io_hook(p, start, end,
847 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
850 free_async_extent_pages(async_extent);
852 alloc_hint = ins.objectid + ins.offset;
858 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
860 extent_clear_unlock_delalloc(inode, async_extent->start,
861 async_extent->start +
862 async_extent->ram_size - 1,
863 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
864 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
865 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
866 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
868 free_async_extent_pages(async_extent);
873 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
877 struct extent_map *em;
880 read_lock(&em_tree->lock);
881 em = search_extent_mapping(em_tree, start, num_bytes);
884 * if block start isn't an actual block number then find the
885 * first block in this inode and use that as a hint. If that
886 * block is also bogus then just don't worry about it.
888 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
890 em = search_extent_mapping(em_tree, 0, 0);
891 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
892 alloc_hint = em->block_start;
896 alloc_hint = em->block_start;
900 read_unlock(&em_tree->lock);
906 * when extent_io.c finds a delayed allocation range in the file,
907 * the call backs end up in this code. The basic idea is to
908 * allocate extents on disk for the range, and create ordered data structs
909 * in ram to track those extents.
911 * locked_page is the page that writepage had locked already. We use
912 * it to make sure we don't do extra locks or unlocks.
914 * *page_started is set to one if we unlock locked_page and do everything
915 * required to start IO on it. It may be clean and already done with
918 static noinline int cow_file_range(struct inode *inode,
919 struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written,
924 struct btrfs_root *root = BTRFS_I(inode)->root;
927 unsigned long ram_size;
930 u64 blocksize = root->sectorsize;
931 struct btrfs_key ins;
932 struct extent_map *em;
933 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
936 if (btrfs_is_free_space_inode(inode)) {
942 num_bytes = ALIGN(end - start + 1, blocksize);
943 num_bytes = max(blocksize, num_bytes);
944 disk_num_bytes = num_bytes;
946 /* if this is a small write inside eof, kick off defrag */
947 if (num_bytes < 64 * 1024 &&
948 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
949 btrfs_add_inode_defrag(NULL, inode);
952 /* lets try to make an inline extent */
953 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
956 extent_clear_unlock_delalloc(inode, start, end, NULL,
957 EXTENT_LOCKED | EXTENT_DELALLOC |
958 EXTENT_DEFRAG, PAGE_UNLOCK |
959 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
962 *nr_written = *nr_written +
963 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
966 } else if (ret < 0) {
971 BUG_ON(disk_num_bytes >
972 btrfs_super_total_bytes(root->fs_info->super_copy));
974 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
975 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
977 while (disk_num_bytes > 0) {
980 cur_alloc_size = disk_num_bytes;
981 ret = btrfs_reserve_extent(root, cur_alloc_size,
982 root->sectorsize, 0, alloc_hint,
987 em = alloc_extent_map();
993 em->orig_start = em->start;
994 ram_size = ins.offset;
995 em->len = ins.offset;
996 em->mod_start = em->start;
997 em->mod_len = em->len;
999 em->block_start = ins.objectid;
1000 em->block_len = ins.offset;
1001 em->orig_block_len = ins.offset;
1002 em->ram_bytes = ram_size;
1003 em->bdev = root->fs_info->fs_devices->latest_bdev;
1004 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1005 em->generation = -1;
1008 write_lock(&em_tree->lock);
1009 ret = add_extent_mapping(em_tree, em, 1);
1010 write_unlock(&em_tree->lock);
1011 if (ret != -EEXIST) {
1012 free_extent_map(em);
1015 btrfs_drop_extent_cache(inode, start,
1016 start + ram_size - 1, 0);
1021 cur_alloc_size = ins.offset;
1022 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1023 ram_size, cur_alloc_size, 0);
1025 goto out_drop_extent_cache;
1027 if (root->root_key.objectid ==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1029 ret = btrfs_reloc_clone_csums(inode, start,
1032 goto out_drop_extent_cache;
1035 if (disk_num_bytes < cur_alloc_size)
1038 /* we're not doing compressed IO, don't unlock the first
1039 * page (which the caller expects to stay locked), don't
1040 * clear any dirty bits and don't set any writeback bits
1042 * Do set the Private2 bit so we know this page was properly
1043 * setup for writepage
1045 op = unlock ? PAGE_UNLOCK : 0;
1046 op |= PAGE_SET_PRIVATE2;
1048 extent_clear_unlock_delalloc(inode, start,
1049 start + ram_size - 1, locked_page,
1050 EXTENT_LOCKED | EXTENT_DELALLOC,
1052 disk_num_bytes -= cur_alloc_size;
1053 num_bytes -= cur_alloc_size;
1054 alloc_hint = ins.objectid + ins.offset;
1055 start += cur_alloc_size;
1060 out_drop_extent_cache:
1061 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1063 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1065 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1066 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1067 EXTENT_DELALLOC | EXTENT_DEFRAG,
1068 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1069 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1074 * work queue call back to started compression on a file and pages
1076 static noinline void async_cow_start(struct btrfs_work *work)
1078 struct async_cow *async_cow;
1080 async_cow = container_of(work, struct async_cow, work);
1082 compress_file_range(async_cow->inode, async_cow->locked_page,
1083 async_cow->start, async_cow->end, async_cow,
1085 if (num_added == 0) {
1086 btrfs_add_delayed_iput(async_cow->inode);
1087 async_cow->inode = NULL;
1092 * work queue call back to submit previously compressed pages
1094 static noinline void async_cow_submit(struct btrfs_work *work)
1096 struct async_cow *async_cow;
1097 struct btrfs_root *root;
1098 unsigned long nr_pages;
1100 async_cow = container_of(work, struct async_cow, work);
1102 root = async_cow->root;
1103 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1107 * atomic_sub_return implies a barrier for waitqueue_active
1109 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1111 waitqueue_active(&root->fs_info->async_submit_wait))
1112 wake_up(&root->fs_info->async_submit_wait);
1114 if (async_cow->inode)
1115 submit_compressed_extents(async_cow->inode, async_cow);
1118 static noinline void async_cow_free(struct btrfs_work *work)
1120 struct async_cow *async_cow;
1121 async_cow = container_of(work, struct async_cow, work);
1122 if (async_cow->inode)
1123 btrfs_add_delayed_iput(async_cow->inode);
1127 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1128 u64 start, u64 end, int *page_started,
1129 unsigned long *nr_written)
1131 struct async_cow *async_cow;
1132 struct btrfs_root *root = BTRFS_I(inode)->root;
1133 unsigned long nr_pages;
1135 int limit = 10 * 1024 * 1024;
1137 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1138 1, 0, NULL, GFP_NOFS);
1139 while (start < end) {
1140 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1141 BUG_ON(!async_cow); /* -ENOMEM */
1142 async_cow->inode = igrab(inode);
1143 async_cow->root = root;
1144 async_cow->locked_page = locked_page;
1145 async_cow->start = start;
1147 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1148 !btrfs_test_opt(root, FORCE_COMPRESS))
1151 cur_end = min(end, start + 512 * 1024 - 1);
1153 async_cow->end = cur_end;
1154 INIT_LIST_HEAD(&async_cow->extents);
1156 btrfs_init_work(&async_cow->work,
1157 btrfs_delalloc_helper,
1158 async_cow_start, async_cow_submit,
1161 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1163 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1165 btrfs_queue_work(root->fs_info->delalloc_workers,
1168 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1169 wait_event(root->fs_info->async_submit_wait,
1170 (atomic_read(&root->fs_info->async_delalloc_pages) <
1174 while (atomic_read(&root->fs_info->async_submit_draining) &&
1175 atomic_read(&root->fs_info->async_delalloc_pages)) {
1176 wait_event(root->fs_info->async_submit_wait,
1177 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1181 *nr_written += nr_pages;
1182 start = cur_end + 1;
1188 static noinline int csum_exist_in_range(struct btrfs_root *root,
1189 u64 bytenr, u64 num_bytes)
1192 struct btrfs_ordered_sum *sums;
1195 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1196 bytenr + num_bytes - 1, &list, 0);
1197 if (ret == 0 && list_empty(&list))
1200 while (!list_empty(&list)) {
1201 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1202 list_del(&sums->list);
1209 * when nowcow writeback call back. This checks for snapshots or COW copies
1210 * of the extents that exist in the file, and COWs the file as required.
1212 * If no cow copies or snapshots exist, we write directly to the existing
1215 static noinline int run_delalloc_nocow(struct inode *inode,
1216 struct page *locked_page,
1217 u64 start, u64 end, int *page_started, int force,
1218 unsigned long *nr_written)
1220 struct btrfs_root *root = BTRFS_I(inode)->root;
1221 struct btrfs_trans_handle *trans;
1222 struct extent_buffer *leaf;
1223 struct btrfs_path *path;
1224 struct btrfs_file_extent_item *fi;
1225 struct btrfs_key found_key;
1240 u64 ino = btrfs_ino(inode);
1242 path = btrfs_alloc_path();
1244 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1245 EXTENT_LOCKED | EXTENT_DELALLOC |
1246 EXTENT_DO_ACCOUNTING |
1247 EXTENT_DEFRAG, PAGE_UNLOCK |
1249 PAGE_SET_WRITEBACK |
1250 PAGE_END_WRITEBACK);
1254 nolock = btrfs_is_free_space_inode(inode);
1257 trans = btrfs_join_transaction_nolock(root);
1259 trans = btrfs_join_transaction(root);
1261 if (IS_ERR(trans)) {
1262 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1263 EXTENT_LOCKED | EXTENT_DELALLOC |
1264 EXTENT_DO_ACCOUNTING |
1265 EXTENT_DEFRAG, PAGE_UNLOCK |
1267 PAGE_SET_WRITEBACK |
1268 PAGE_END_WRITEBACK);
1269 btrfs_free_path(path);
1270 return PTR_ERR(trans);
1273 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1275 cow_start = (u64)-1;
1278 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1282 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &found_key,
1285 path->slots[0] - 1);
1286 if (found_key.objectid == ino &&
1287 found_key.type == BTRFS_EXTENT_DATA_KEY)
1292 leaf = path->nodes[0];
1293 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1294 ret = btrfs_next_leaf(root, path);
1299 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1307 if (found_key.objectid > ino ||
1308 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1309 found_key.offset > end)
1312 if (found_key.offset > cur_offset) {
1313 extent_end = found_key.offset;
1318 fi = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_file_extent_item);
1320 extent_type = btrfs_file_extent_type(leaf, fi);
1322 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1323 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1324 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1325 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1326 extent_offset = btrfs_file_extent_offset(leaf, fi);
1327 extent_end = found_key.offset +
1328 btrfs_file_extent_num_bytes(leaf, fi);
1330 btrfs_file_extent_disk_num_bytes(leaf, fi);
1331 if (extent_end <= start) {
1335 if (disk_bytenr == 0)
1337 if (btrfs_file_extent_compression(leaf, fi) ||
1338 btrfs_file_extent_encryption(leaf, fi) ||
1339 btrfs_file_extent_other_encoding(leaf, fi))
1341 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1343 if (btrfs_extent_readonly(root, disk_bytenr))
1345 if (btrfs_cross_ref_exist(trans, root, ino,
1347 extent_offset, disk_bytenr))
1349 disk_bytenr += extent_offset;
1350 disk_bytenr += cur_offset - found_key.offset;
1351 num_bytes = min(end + 1, extent_end) - cur_offset;
1353 * if there are pending snapshots for this root,
1354 * we fall into common COW way.
1357 err = btrfs_start_write_no_snapshoting(root);
1362 * force cow if csum exists in the range.
1363 * this ensure that csum for a given extent are
1364 * either valid or do not exist.
1366 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1369 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1370 extent_end = found_key.offset +
1371 btrfs_file_extent_inline_len(leaf,
1372 path->slots[0], fi);
1373 extent_end = ALIGN(extent_end, root->sectorsize);
1378 if (extent_end <= start) {
1380 if (!nolock && nocow)
1381 btrfs_end_write_no_snapshoting(root);
1385 if (cow_start == (u64)-1)
1386 cow_start = cur_offset;
1387 cur_offset = extent_end;
1388 if (cur_offset > end)
1394 btrfs_release_path(path);
1395 if (cow_start != (u64)-1) {
1396 ret = cow_file_range(inode, locked_page,
1397 cow_start, found_key.offset - 1,
1398 page_started, nr_written, 1);
1400 if (!nolock && nocow)
1401 btrfs_end_write_no_snapshoting(root);
1404 cow_start = (u64)-1;
1407 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1408 struct extent_map *em;
1409 struct extent_map_tree *em_tree;
1410 em_tree = &BTRFS_I(inode)->extent_tree;
1411 em = alloc_extent_map();
1412 BUG_ON(!em); /* -ENOMEM */
1413 em->start = cur_offset;
1414 em->orig_start = found_key.offset - extent_offset;
1415 em->len = num_bytes;
1416 em->block_len = num_bytes;
1417 em->block_start = disk_bytenr;
1418 em->orig_block_len = disk_num_bytes;
1419 em->ram_bytes = ram_bytes;
1420 em->bdev = root->fs_info->fs_devices->latest_bdev;
1421 em->mod_start = em->start;
1422 em->mod_len = em->len;
1423 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1424 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1425 em->generation = -1;
1427 write_lock(&em_tree->lock);
1428 ret = add_extent_mapping(em_tree, em, 1);
1429 write_unlock(&em_tree->lock);
1430 if (ret != -EEXIST) {
1431 free_extent_map(em);
1434 btrfs_drop_extent_cache(inode, em->start,
1435 em->start + em->len - 1, 0);
1437 type = BTRFS_ORDERED_PREALLOC;
1439 type = BTRFS_ORDERED_NOCOW;
1442 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1443 num_bytes, num_bytes, type);
1444 BUG_ON(ret); /* -ENOMEM */
1446 if (root->root_key.objectid ==
1447 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1448 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1451 if (!nolock && nocow)
1452 btrfs_end_write_no_snapshoting(root);
1457 extent_clear_unlock_delalloc(inode, cur_offset,
1458 cur_offset + num_bytes - 1,
1459 locked_page, EXTENT_LOCKED |
1460 EXTENT_DELALLOC, PAGE_UNLOCK |
1462 if (!nolock && nocow)
1463 btrfs_end_write_no_snapshoting(root);
1464 cur_offset = extent_end;
1465 if (cur_offset > end)
1468 btrfs_release_path(path);
1470 if (cur_offset <= end && cow_start == (u64)-1) {
1471 cow_start = cur_offset;
1475 if (cow_start != (u64)-1) {
1476 ret = cow_file_range(inode, locked_page, cow_start, end,
1477 page_started, nr_written, 1);
1483 err = btrfs_end_transaction(trans, root);
1487 if (ret && cur_offset < end)
1488 extent_clear_unlock_delalloc(inode, cur_offset, end,
1489 locked_page, EXTENT_LOCKED |
1490 EXTENT_DELALLOC | EXTENT_DEFRAG |
1491 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1493 PAGE_SET_WRITEBACK |
1494 PAGE_END_WRITEBACK);
1495 btrfs_free_path(path);
1499 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1502 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1503 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1507 * @defrag_bytes is a hint value, no spinlock held here,
1508 * if is not zero, it means the file is defragging.
1509 * Force cow if given extent needs to be defragged.
1511 if (BTRFS_I(inode)->defrag_bytes &&
1512 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1513 EXTENT_DEFRAG, 0, NULL))
1520 * extent_io.c call back to do delayed allocation processing
1522 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1523 u64 start, u64 end, int *page_started,
1524 unsigned long *nr_written)
1527 int force_cow = need_force_cow(inode, start, end);
1529 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1530 ret = run_delalloc_nocow(inode, locked_page, start, end,
1531 page_started, 1, nr_written);
1532 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1533 ret = run_delalloc_nocow(inode, locked_page, start, end,
1534 page_started, 0, nr_written);
1535 } else if (!inode_need_compress(inode)) {
1536 ret = cow_file_range(inode, locked_page, start, end,
1537 page_started, nr_written, 1);
1539 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1540 &BTRFS_I(inode)->runtime_flags);
1541 ret = cow_file_range_async(inode, locked_page, start, end,
1542 page_started, nr_written);
1547 static void btrfs_split_extent_hook(struct inode *inode,
1548 struct extent_state *orig, u64 split)
1552 /* not delalloc, ignore it */
1553 if (!(orig->state & EXTENT_DELALLOC))
1556 size = orig->end - orig->start + 1;
1557 if (size > BTRFS_MAX_EXTENT_SIZE) {
1562 * See the explanation in btrfs_merge_extent_hook, the same
1563 * applies here, just in reverse.
1565 new_size = orig->end - split + 1;
1566 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1567 BTRFS_MAX_EXTENT_SIZE);
1568 new_size = split - orig->start;
1569 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1570 BTRFS_MAX_EXTENT_SIZE);
1571 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1572 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1576 spin_lock(&BTRFS_I(inode)->lock);
1577 BTRFS_I(inode)->outstanding_extents++;
1578 spin_unlock(&BTRFS_I(inode)->lock);
1582 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1583 * extents so we can keep track of new extents that are just merged onto old
1584 * extents, such as when we are doing sequential writes, so we can properly
1585 * account for the metadata space we'll need.
1587 static void btrfs_merge_extent_hook(struct inode *inode,
1588 struct extent_state *new,
1589 struct extent_state *other)
1591 u64 new_size, old_size;
1594 /* not delalloc, ignore it */
1595 if (!(other->state & EXTENT_DELALLOC))
1598 if (new->start > other->start)
1599 new_size = new->end - other->start + 1;
1601 new_size = other->end - new->start + 1;
1603 /* we're not bigger than the max, unreserve the space and go */
1604 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1605 spin_lock(&BTRFS_I(inode)->lock);
1606 BTRFS_I(inode)->outstanding_extents--;
1607 spin_unlock(&BTRFS_I(inode)->lock);
1612 * We have to add up either side to figure out how many extents were
1613 * accounted for before we merged into one big extent. If the number of
1614 * extents we accounted for is <= the amount we need for the new range
1615 * then we can return, otherwise drop. Think of it like this
1619 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1620 * need 2 outstanding extents, on one side we have 1 and the other side
1621 * we have 1 so they are == and we can return. But in this case
1623 * [MAX_SIZE+4k][MAX_SIZE+4k]
1625 * Each range on their own accounts for 2 extents, but merged together
1626 * they are only 3 extents worth of accounting, so we need to drop in
1629 old_size = other->end - other->start + 1;
1630 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1631 BTRFS_MAX_EXTENT_SIZE);
1632 old_size = new->end - new->start + 1;
1633 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1634 BTRFS_MAX_EXTENT_SIZE);
1636 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1637 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1640 spin_lock(&BTRFS_I(inode)->lock);
1641 BTRFS_I(inode)->outstanding_extents--;
1642 spin_unlock(&BTRFS_I(inode)->lock);
1645 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1646 struct inode *inode)
1648 spin_lock(&root->delalloc_lock);
1649 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1650 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1651 &root->delalloc_inodes);
1652 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1653 &BTRFS_I(inode)->runtime_flags);
1654 root->nr_delalloc_inodes++;
1655 if (root->nr_delalloc_inodes == 1) {
1656 spin_lock(&root->fs_info->delalloc_root_lock);
1657 BUG_ON(!list_empty(&root->delalloc_root));
1658 list_add_tail(&root->delalloc_root,
1659 &root->fs_info->delalloc_roots);
1660 spin_unlock(&root->fs_info->delalloc_root_lock);
1663 spin_unlock(&root->delalloc_lock);
1666 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1667 struct inode *inode)
1669 spin_lock(&root->delalloc_lock);
1670 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1671 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1672 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1673 &BTRFS_I(inode)->runtime_flags);
1674 root->nr_delalloc_inodes--;
1675 if (!root->nr_delalloc_inodes) {
1676 spin_lock(&root->fs_info->delalloc_root_lock);
1677 BUG_ON(list_empty(&root->delalloc_root));
1678 list_del_init(&root->delalloc_root);
1679 spin_unlock(&root->fs_info->delalloc_root_lock);
1682 spin_unlock(&root->delalloc_lock);
1686 * extent_io.c set_bit_hook, used to track delayed allocation
1687 * bytes in this file, and to maintain the list of inodes that
1688 * have pending delalloc work to be done.
1690 static void btrfs_set_bit_hook(struct inode *inode,
1691 struct extent_state *state, unsigned *bits)
1694 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1697 * set_bit and clear bit hooks normally require _irqsave/restore
1698 * but in this case, we are only testing for the DELALLOC
1699 * bit, which is only set or cleared with irqs on
1701 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1702 struct btrfs_root *root = BTRFS_I(inode)->root;
1703 u64 len = state->end + 1 - state->start;
1704 bool do_list = !btrfs_is_free_space_inode(inode);
1706 if (*bits & EXTENT_FIRST_DELALLOC) {
1707 *bits &= ~EXTENT_FIRST_DELALLOC;
1709 spin_lock(&BTRFS_I(inode)->lock);
1710 BTRFS_I(inode)->outstanding_extents++;
1711 spin_unlock(&BTRFS_I(inode)->lock);
1714 /* For sanity tests */
1715 if (btrfs_test_is_dummy_root(root))
1718 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1719 root->fs_info->delalloc_batch);
1720 spin_lock(&BTRFS_I(inode)->lock);
1721 BTRFS_I(inode)->delalloc_bytes += len;
1722 if (*bits & EXTENT_DEFRAG)
1723 BTRFS_I(inode)->defrag_bytes += len;
1724 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1725 &BTRFS_I(inode)->runtime_flags))
1726 btrfs_add_delalloc_inodes(root, inode);
1727 spin_unlock(&BTRFS_I(inode)->lock);
1732 * extent_io.c clear_bit_hook, see set_bit_hook for why
1734 static void btrfs_clear_bit_hook(struct inode *inode,
1735 struct extent_state *state,
1738 u64 len = state->end + 1 - state->start;
1739 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1740 BTRFS_MAX_EXTENT_SIZE);
1742 spin_lock(&BTRFS_I(inode)->lock);
1743 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1744 BTRFS_I(inode)->defrag_bytes -= len;
1745 spin_unlock(&BTRFS_I(inode)->lock);
1748 * set_bit and clear bit hooks normally require _irqsave/restore
1749 * but in this case, we are only testing for the DELALLOC
1750 * bit, which is only set or cleared with irqs on
1752 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1753 struct btrfs_root *root = BTRFS_I(inode)->root;
1754 bool do_list = !btrfs_is_free_space_inode(inode);
1756 if (*bits & EXTENT_FIRST_DELALLOC) {
1757 *bits &= ~EXTENT_FIRST_DELALLOC;
1758 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1759 spin_lock(&BTRFS_I(inode)->lock);
1760 BTRFS_I(inode)->outstanding_extents -= num_extents;
1761 spin_unlock(&BTRFS_I(inode)->lock);
1765 * We don't reserve metadata space for space cache inodes so we
1766 * don't need to call dellalloc_release_metadata if there is an
1769 if (*bits & EXTENT_DO_ACCOUNTING &&
1770 root != root->fs_info->tree_root)
1771 btrfs_delalloc_release_metadata(inode, len);
1773 /* For sanity tests. */
1774 if (btrfs_test_is_dummy_root(root))
1777 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1778 && do_list && !(state->state & EXTENT_NORESERVE))
1779 btrfs_free_reserved_data_space_noquota(inode,
1782 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1783 root->fs_info->delalloc_batch);
1784 spin_lock(&BTRFS_I(inode)->lock);
1785 BTRFS_I(inode)->delalloc_bytes -= len;
1786 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1787 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1788 &BTRFS_I(inode)->runtime_flags))
1789 btrfs_del_delalloc_inode(root, inode);
1790 spin_unlock(&BTRFS_I(inode)->lock);
1795 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1796 * we don't create bios that span stripes or chunks
1798 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1799 size_t size, struct bio *bio,
1800 unsigned long bio_flags)
1802 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1803 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1808 if (bio_flags & EXTENT_BIO_COMPRESSED)
1811 length = bio->bi_iter.bi_size;
1812 map_length = length;
1813 ret = btrfs_map_block(root->fs_info, rw, logical,
1814 &map_length, NULL, 0);
1815 /* Will always return 0 with map_multi == NULL */
1817 if (map_length < length + size)
1823 * in order to insert checksums into the metadata in large chunks,
1824 * we wait until bio submission time. All the pages in the bio are
1825 * checksummed and sums are attached onto the ordered extent record.
1827 * At IO completion time the cums attached on the ordered extent record
1828 * are inserted into the btree
1830 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1831 struct bio *bio, int mirror_num,
1832 unsigned long bio_flags,
1835 struct btrfs_root *root = BTRFS_I(inode)->root;
1838 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1839 BUG_ON(ret); /* -ENOMEM */
1844 * in order to insert checksums into the metadata in large chunks,
1845 * we wait until bio submission time. All the pages in the bio are
1846 * checksummed and sums are attached onto the ordered extent record.
1848 * At IO completion time the cums attached on the ordered extent record
1849 * are inserted into the btree
1851 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1852 int mirror_num, unsigned long bio_flags,
1855 struct btrfs_root *root = BTRFS_I(inode)->root;
1858 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1860 bio->bi_error = ret;
1867 * extent_io.c submission hook. This does the right thing for csum calculation
1868 * on write, or reading the csums from the tree before a read
1870 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1871 int mirror_num, unsigned long bio_flags,
1874 struct btrfs_root *root = BTRFS_I(inode)->root;
1875 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1878 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1880 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1882 if (btrfs_is_free_space_inode(inode))
1883 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1885 if (!(rw & REQ_WRITE)) {
1886 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1890 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1891 ret = btrfs_submit_compressed_read(inode, bio,
1895 } else if (!skip_sum) {
1896 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1901 } else if (async && !skip_sum) {
1902 /* csum items have already been cloned */
1903 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1905 /* we're doing a write, do the async checksumming */
1906 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1907 inode, rw, bio, mirror_num,
1908 bio_flags, bio_offset,
1909 __btrfs_submit_bio_start,
1910 __btrfs_submit_bio_done);
1912 } else if (!skip_sum) {
1913 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1919 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1923 bio->bi_error = ret;
1930 * given a list of ordered sums record them in the inode. This happens
1931 * at IO completion time based on sums calculated at bio submission time.
1933 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1934 struct inode *inode, u64 file_offset,
1935 struct list_head *list)
1937 struct btrfs_ordered_sum *sum;
1939 list_for_each_entry(sum, list, list) {
1940 trans->adding_csums = 1;
1941 btrfs_csum_file_blocks(trans,
1942 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1943 trans->adding_csums = 0;
1948 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1949 struct extent_state **cached_state)
1951 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1952 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1953 cached_state, GFP_NOFS);
1956 /* see btrfs_writepage_start_hook for details on why this is required */
1957 struct btrfs_writepage_fixup {
1959 struct btrfs_work work;
1962 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1964 struct btrfs_writepage_fixup *fixup;
1965 struct btrfs_ordered_extent *ordered;
1966 struct extent_state *cached_state = NULL;
1968 struct inode *inode;
1973 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1977 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1978 ClearPageChecked(page);
1982 inode = page->mapping->host;
1983 page_start = page_offset(page);
1984 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1986 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1989 /* already ordered? We're done */
1990 if (PagePrivate2(page))
1993 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1995 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1996 page_end, &cached_state, GFP_NOFS);
1998 btrfs_start_ordered_extent(inode, ordered, 1);
1999 btrfs_put_ordered_extent(ordered);
2003 ret = btrfs_delalloc_reserve_space(inode, page_start,
2006 mapping_set_error(page->mapping, ret);
2007 end_extent_writepage(page, ret, page_start, page_end);
2008 ClearPageChecked(page);
2012 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2013 ClearPageChecked(page);
2014 set_page_dirty(page);
2016 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2017 &cached_state, GFP_NOFS);
2020 page_cache_release(page);
2025 * There are a few paths in the higher layers of the kernel that directly
2026 * set the page dirty bit without asking the filesystem if it is a
2027 * good idea. This causes problems because we want to make sure COW
2028 * properly happens and the data=ordered rules are followed.
2030 * In our case any range that doesn't have the ORDERED bit set
2031 * hasn't been properly setup for IO. We kick off an async process
2032 * to fix it up. The async helper will wait for ordered extents, set
2033 * the delalloc bit and make it safe to write the page.
2035 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2037 struct inode *inode = page->mapping->host;
2038 struct btrfs_writepage_fixup *fixup;
2039 struct btrfs_root *root = BTRFS_I(inode)->root;
2041 /* this page is properly in the ordered list */
2042 if (TestClearPagePrivate2(page))
2045 if (PageChecked(page))
2048 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2052 SetPageChecked(page);
2053 page_cache_get(page);
2054 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2055 btrfs_writepage_fixup_worker, NULL, NULL);
2057 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2061 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2062 struct inode *inode, u64 file_pos,
2063 u64 disk_bytenr, u64 disk_num_bytes,
2064 u64 num_bytes, u64 ram_bytes,
2065 u8 compression, u8 encryption,
2066 u16 other_encoding, int extent_type)
2068 struct btrfs_root *root = BTRFS_I(inode)->root;
2069 struct btrfs_file_extent_item *fi;
2070 struct btrfs_path *path;
2071 struct extent_buffer *leaf;
2072 struct btrfs_key ins;
2073 int extent_inserted = 0;
2076 path = btrfs_alloc_path();
2081 * we may be replacing one extent in the tree with another.
2082 * The new extent is pinned in the extent map, and we don't want
2083 * to drop it from the cache until it is completely in the btree.
2085 * So, tell btrfs_drop_extents to leave this extent in the cache.
2086 * the caller is expected to unpin it and allow it to be merged
2089 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2090 file_pos + num_bytes, NULL, 0,
2091 1, sizeof(*fi), &extent_inserted);
2095 if (!extent_inserted) {
2096 ins.objectid = btrfs_ino(inode);
2097 ins.offset = file_pos;
2098 ins.type = BTRFS_EXTENT_DATA_KEY;
2100 path->leave_spinning = 1;
2101 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2106 leaf = path->nodes[0];
2107 fi = btrfs_item_ptr(leaf, path->slots[0],
2108 struct btrfs_file_extent_item);
2109 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2110 btrfs_set_file_extent_type(leaf, fi, extent_type);
2111 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2112 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2113 btrfs_set_file_extent_offset(leaf, fi, 0);
2114 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2115 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2116 btrfs_set_file_extent_compression(leaf, fi, compression);
2117 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2118 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2120 btrfs_mark_buffer_dirty(leaf);
2121 btrfs_release_path(path);
2123 inode_add_bytes(inode, num_bytes);
2125 ins.objectid = disk_bytenr;
2126 ins.offset = disk_num_bytes;
2127 ins.type = BTRFS_EXTENT_ITEM_KEY;
2128 ret = btrfs_alloc_reserved_file_extent(trans, root,
2129 root->root_key.objectid,
2130 btrfs_ino(inode), file_pos, &ins);
2134 * Release the reserved range from inode dirty range map, and
2135 * move it to delayed ref codes, as now accounting only happens at
2136 * commit_transaction() time.
2138 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2139 ret = btrfs_add_delayed_qgroup_reserve(root->fs_info, trans,
2140 root->objectid, disk_bytenr, ram_bytes);
2142 btrfs_free_path(path);
2147 /* snapshot-aware defrag */
2148 struct sa_defrag_extent_backref {
2149 struct rb_node node;
2150 struct old_sa_defrag_extent *old;
2159 struct old_sa_defrag_extent {
2160 struct list_head list;
2161 struct new_sa_defrag_extent *new;
2170 struct new_sa_defrag_extent {
2171 struct rb_root root;
2172 struct list_head head;
2173 struct btrfs_path *path;
2174 struct inode *inode;
2182 static int backref_comp(struct sa_defrag_extent_backref *b1,
2183 struct sa_defrag_extent_backref *b2)
2185 if (b1->root_id < b2->root_id)
2187 else if (b1->root_id > b2->root_id)
2190 if (b1->inum < b2->inum)
2192 else if (b1->inum > b2->inum)
2195 if (b1->file_pos < b2->file_pos)
2197 else if (b1->file_pos > b2->file_pos)
2201 * [------------------------------] ===> (a range of space)
2202 * |<--->| |<---->| =============> (fs/file tree A)
2203 * |<---------------------------->| ===> (fs/file tree B)
2205 * A range of space can refer to two file extents in one tree while
2206 * refer to only one file extent in another tree.
2208 * So we may process a disk offset more than one time(two extents in A)
2209 * and locate at the same extent(one extent in B), then insert two same
2210 * backrefs(both refer to the extent in B).
2215 static void backref_insert(struct rb_root *root,
2216 struct sa_defrag_extent_backref *backref)
2218 struct rb_node **p = &root->rb_node;
2219 struct rb_node *parent = NULL;
2220 struct sa_defrag_extent_backref *entry;
2225 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2227 ret = backref_comp(backref, entry);
2231 p = &(*p)->rb_right;
2234 rb_link_node(&backref->node, parent, p);
2235 rb_insert_color(&backref->node, root);
2239 * Note the backref might has changed, and in this case we just return 0.
2241 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2244 struct btrfs_file_extent_item *extent;
2245 struct btrfs_fs_info *fs_info;
2246 struct old_sa_defrag_extent *old = ctx;
2247 struct new_sa_defrag_extent *new = old->new;
2248 struct btrfs_path *path = new->path;
2249 struct btrfs_key key;
2250 struct btrfs_root *root;
2251 struct sa_defrag_extent_backref *backref;
2252 struct extent_buffer *leaf;
2253 struct inode *inode = new->inode;
2259 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2260 inum == btrfs_ino(inode))
2263 key.objectid = root_id;
2264 key.type = BTRFS_ROOT_ITEM_KEY;
2265 key.offset = (u64)-1;
2267 fs_info = BTRFS_I(inode)->root->fs_info;
2268 root = btrfs_read_fs_root_no_name(fs_info, &key);
2270 if (PTR_ERR(root) == -ENOENT)
2273 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2274 inum, offset, root_id);
2275 return PTR_ERR(root);
2278 key.objectid = inum;
2279 key.type = BTRFS_EXTENT_DATA_KEY;
2280 if (offset > (u64)-1 << 32)
2283 key.offset = offset;
2285 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2286 if (WARN_ON(ret < 0))
2293 leaf = path->nodes[0];
2294 slot = path->slots[0];
2296 if (slot >= btrfs_header_nritems(leaf)) {
2297 ret = btrfs_next_leaf(root, path);
2300 } else if (ret > 0) {
2309 btrfs_item_key_to_cpu(leaf, &key, slot);
2311 if (key.objectid > inum)
2314 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2317 extent = btrfs_item_ptr(leaf, slot,
2318 struct btrfs_file_extent_item);
2320 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2324 * 'offset' refers to the exact key.offset,
2325 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2326 * (key.offset - extent_offset).
2328 if (key.offset != offset)
2331 extent_offset = btrfs_file_extent_offset(leaf, extent);
2332 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2334 if (extent_offset >= old->extent_offset + old->offset +
2335 old->len || extent_offset + num_bytes <=
2336 old->extent_offset + old->offset)
2341 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2347 backref->root_id = root_id;
2348 backref->inum = inum;
2349 backref->file_pos = offset;
2350 backref->num_bytes = num_bytes;
2351 backref->extent_offset = extent_offset;
2352 backref->generation = btrfs_file_extent_generation(leaf, extent);
2354 backref_insert(&new->root, backref);
2357 btrfs_release_path(path);
2362 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2363 struct new_sa_defrag_extent *new)
2365 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2366 struct old_sa_defrag_extent *old, *tmp;
2371 list_for_each_entry_safe(old, tmp, &new->head, list) {
2372 ret = iterate_inodes_from_logical(old->bytenr +
2373 old->extent_offset, fs_info,
2374 path, record_one_backref,
2376 if (ret < 0 && ret != -ENOENT)
2379 /* no backref to be processed for this extent */
2381 list_del(&old->list);
2386 if (list_empty(&new->head))
2392 static int relink_is_mergable(struct extent_buffer *leaf,
2393 struct btrfs_file_extent_item *fi,
2394 struct new_sa_defrag_extent *new)
2396 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2399 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2402 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2405 if (btrfs_file_extent_encryption(leaf, fi) ||
2406 btrfs_file_extent_other_encoding(leaf, fi))
2413 * Note the backref might has changed, and in this case we just return 0.
2415 static noinline int relink_extent_backref(struct btrfs_path *path,
2416 struct sa_defrag_extent_backref *prev,
2417 struct sa_defrag_extent_backref *backref)
2419 struct btrfs_file_extent_item *extent;
2420 struct btrfs_file_extent_item *item;
2421 struct btrfs_ordered_extent *ordered;
2422 struct btrfs_trans_handle *trans;
2423 struct btrfs_fs_info *fs_info;
2424 struct btrfs_root *root;
2425 struct btrfs_key key;
2426 struct extent_buffer *leaf;
2427 struct old_sa_defrag_extent *old = backref->old;
2428 struct new_sa_defrag_extent *new = old->new;
2429 struct inode *src_inode = new->inode;
2430 struct inode *inode;
2431 struct extent_state *cached = NULL;
2440 if (prev && prev->root_id == backref->root_id &&
2441 prev->inum == backref->inum &&
2442 prev->file_pos + prev->num_bytes == backref->file_pos)
2445 /* step 1: get root */
2446 key.objectid = backref->root_id;
2447 key.type = BTRFS_ROOT_ITEM_KEY;
2448 key.offset = (u64)-1;
2450 fs_info = BTRFS_I(src_inode)->root->fs_info;
2451 index = srcu_read_lock(&fs_info->subvol_srcu);
2453 root = btrfs_read_fs_root_no_name(fs_info, &key);
2455 srcu_read_unlock(&fs_info->subvol_srcu, index);
2456 if (PTR_ERR(root) == -ENOENT)
2458 return PTR_ERR(root);
2461 if (btrfs_root_readonly(root)) {
2462 srcu_read_unlock(&fs_info->subvol_srcu, index);
2466 /* step 2: get inode */
2467 key.objectid = backref->inum;
2468 key.type = BTRFS_INODE_ITEM_KEY;
2471 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2472 if (IS_ERR(inode)) {
2473 srcu_read_unlock(&fs_info->subvol_srcu, index);
2477 srcu_read_unlock(&fs_info->subvol_srcu, index);
2479 /* step 3: relink backref */
2480 lock_start = backref->file_pos;
2481 lock_end = backref->file_pos + backref->num_bytes - 1;
2482 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2485 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2487 btrfs_put_ordered_extent(ordered);
2491 trans = btrfs_join_transaction(root);
2492 if (IS_ERR(trans)) {
2493 ret = PTR_ERR(trans);
2497 key.objectid = backref->inum;
2498 key.type = BTRFS_EXTENT_DATA_KEY;
2499 key.offset = backref->file_pos;
2501 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2504 } else if (ret > 0) {
2509 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2510 struct btrfs_file_extent_item);
2512 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2513 backref->generation)
2516 btrfs_release_path(path);
2518 start = backref->file_pos;
2519 if (backref->extent_offset < old->extent_offset + old->offset)
2520 start += old->extent_offset + old->offset -
2521 backref->extent_offset;
2523 len = min(backref->extent_offset + backref->num_bytes,
2524 old->extent_offset + old->offset + old->len);
2525 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2527 ret = btrfs_drop_extents(trans, root, inode, start,
2532 key.objectid = btrfs_ino(inode);
2533 key.type = BTRFS_EXTENT_DATA_KEY;
2536 path->leave_spinning = 1;
2538 struct btrfs_file_extent_item *fi;
2540 struct btrfs_key found_key;
2542 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2547 leaf = path->nodes[0];
2548 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2550 fi = btrfs_item_ptr(leaf, path->slots[0],
2551 struct btrfs_file_extent_item);
2552 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2554 if (extent_len + found_key.offset == start &&
2555 relink_is_mergable(leaf, fi, new)) {
2556 btrfs_set_file_extent_num_bytes(leaf, fi,
2558 btrfs_mark_buffer_dirty(leaf);
2559 inode_add_bytes(inode, len);
2565 btrfs_release_path(path);
2570 ret = btrfs_insert_empty_item(trans, root, path, &key,
2573 btrfs_abort_transaction(trans, root, ret);
2577 leaf = path->nodes[0];
2578 item = btrfs_item_ptr(leaf, path->slots[0],
2579 struct btrfs_file_extent_item);
2580 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2581 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2582 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2583 btrfs_set_file_extent_num_bytes(leaf, item, len);
2584 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2585 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2586 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2587 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2588 btrfs_set_file_extent_encryption(leaf, item, 0);
2589 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2591 btrfs_mark_buffer_dirty(leaf);
2592 inode_add_bytes(inode, len);
2593 btrfs_release_path(path);
2595 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2597 backref->root_id, backref->inum,
2598 new->file_pos, 0); /* start - extent_offset */
2600 btrfs_abort_transaction(trans, root, ret);
2606 btrfs_release_path(path);
2607 path->leave_spinning = 0;
2608 btrfs_end_transaction(trans, root);
2610 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2616 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2618 struct old_sa_defrag_extent *old, *tmp;
2623 list_for_each_entry_safe(old, tmp, &new->head, list) {
2629 static void relink_file_extents(struct new_sa_defrag_extent *new)
2631 struct btrfs_path *path;
2632 struct sa_defrag_extent_backref *backref;
2633 struct sa_defrag_extent_backref *prev = NULL;
2634 struct inode *inode;
2635 struct btrfs_root *root;
2636 struct rb_node *node;
2640 root = BTRFS_I(inode)->root;
2642 path = btrfs_alloc_path();
2646 if (!record_extent_backrefs(path, new)) {
2647 btrfs_free_path(path);
2650 btrfs_release_path(path);
2653 node = rb_first(&new->root);
2656 rb_erase(node, &new->root);
2658 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2660 ret = relink_extent_backref(path, prev, backref);
2673 btrfs_free_path(path);
2675 free_sa_defrag_extent(new);
2677 atomic_dec(&root->fs_info->defrag_running);
2678 wake_up(&root->fs_info->transaction_wait);
2681 static struct new_sa_defrag_extent *
2682 record_old_file_extents(struct inode *inode,
2683 struct btrfs_ordered_extent *ordered)
2685 struct btrfs_root *root = BTRFS_I(inode)->root;
2686 struct btrfs_path *path;
2687 struct btrfs_key key;
2688 struct old_sa_defrag_extent *old;
2689 struct new_sa_defrag_extent *new;
2692 new = kmalloc(sizeof(*new), GFP_NOFS);
2697 new->file_pos = ordered->file_offset;
2698 new->len = ordered->len;
2699 new->bytenr = ordered->start;
2700 new->disk_len = ordered->disk_len;
2701 new->compress_type = ordered->compress_type;
2702 new->root = RB_ROOT;
2703 INIT_LIST_HEAD(&new->head);
2705 path = btrfs_alloc_path();
2709 key.objectid = btrfs_ino(inode);
2710 key.type = BTRFS_EXTENT_DATA_KEY;
2711 key.offset = new->file_pos;
2713 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2716 if (ret > 0 && path->slots[0] > 0)
2719 /* find out all the old extents for the file range */
2721 struct btrfs_file_extent_item *extent;
2722 struct extent_buffer *l;
2731 slot = path->slots[0];
2733 if (slot >= btrfs_header_nritems(l)) {
2734 ret = btrfs_next_leaf(root, path);
2742 btrfs_item_key_to_cpu(l, &key, slot);
2744 if (key.objectid != btrfs_ino(inode))
2746 if (key.type != BTRFS_EXTENT_DATA_KEY)
2748 if (key.offset >= new->file_pos + new->len)
2751 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2753 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2754 if (key.offset + num_bytes < new->file_pos)
2757 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2761 extent_offset = btrfs_file_extent_offset(l, extent);
2763 old = kmalloc(sizeof(*old), GFP_NOFS);
2767 offset = max(new->file_pos, key.offset);
2768 end = min(new->file_pos + new->len, key.offset + num_bytes);
2770 old->bytenr = disk_bytenr;
2771 old->extent_offset = extent_offset;
2772 old->offset = offset - key.offset;
2773 old->len = end - offset;
2776 list_add_tail(&old->list, &new->head);
2782 btrfs_free_path(path);
2783 atomic_inc(&root->fs_info->defrag_running);
2788 btrfs_free_path(path);
2790 free_sa_defrag_extent(new);
2794 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2797 struct btrfs_block_group_cache *cache;
2799 cache = btrfs_lookup_block_group(root->fs_info, start);
2802 spin_lock(&cache->lock);
2803 cache->delalloc_bytes -= len;
2804 spin_unlock(&cache->lock);
2806 btrfs_put_block_group(cache);
2809 /* as ordered data IO finishes, this gets called so we can finish
2810 * an ordered extent if the range of bytes in the file it covers are
2813 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2815 struct inode *inode = ordered_extent->inode;
2816 struct btrfs_root *root = BTRFS_I(inode)->root;
2817 struct btrfs_trans_handle *trans = NULL;
2818 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2819 struct extent_state *cached_state = NULL;
2820 struct new_sa_defrag_extent *new = NULL;
2821 int compress_type = 0;
2823 u64 logical_len = ordered_extent->len;
2825 bool truncated = false;
2827 nolock = btrfs_is_free_space_inode(inode);
2829 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2834 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2835 ordered_extent->file_offset +
2836 ordered_extent->len - 1);
2838 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2840 logical_len = ordered_extent->truncated_len;
2841 /* Truncated the entire extent, don't bother adding */
2846 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2847 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2850 * For mwrite(mmap + memset to write) case, we still reserve
2851 * space for NOCOW range.
2852 * As NOCOW won't cause a new delayed ref, just free the space
2854 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2855 ordered_extent->len);
2856 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2858 trans = btrfs_join_transaction_nolock(root);
2860 trans = btrfs_join_transaction(root);
2861 if (IS_ERR(trans)) {
2862 ret = PTR_ERR(trans);
2866 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2867 ret = btrfs_update_inode_fallback(trans, root, inode);
2868 if (ret) /* -ENOMEM or corruption */
2869 btrfs_abort_transaction(trans, root, ret);
2873 lock_extent_bits(io_tree, ordered_extent->file_offset,
2874 ordered_extent->file_offset + ordered_extent->len - 1,
2877 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2878 ordered_extent->file_offset + ordered_extent->len - 1,
2879 EXTENT_DEFRAG, 1, cached_state);
2881 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2882 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2883 /* the inode is shared */
2884 new = record_old_file_extents(inode, ordered_extent);
2886 clear_extent_bit(io_tree, ordered_extent->file_offset,
2887 ordered_extent->file_offset + ordered_extent->len - 1,
2888 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2892 trans = btrfs_join_transaction_nolock(root);
2894 trans = btrfs_join_transaction(root);
2895 if (IS_ERR(trans)) {
2896 ret = PTR_ERR(trans);
2901 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2903 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2904 compress_type = ordered_extent->compress_type;
2905 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2906 BUG_ON(compress_type);
2907 ret = btrfs_mark_extent_written(trans, inode,
2908 ordered_extent->file_offset,
2909 ordered_extent->file_offset +
2912 BUG_ON(root == root->fs_info->tree_root);
2913 ret = insert_reserved_file_extent(trans, inode,
2914 ordered_extent->file_offset,
2915 ordered_extent->start,
2916 ordered_extent->disk_len,
2917 logical_len, logical_len,
2918 compress_type, 0, 0,
2919 BTRFS_FILE_EXTENT_REG);
2921 btrfs_release_delalloc_bytes(root,
2922 ordered_extent->start,
2923 ordered_extent->disk_len);
2925 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2926 ordered_extent->file_offset, ordered_extent->len,
2929 btrfs_abort_transaction(trans, root, ret);
2933 add_pending_csums(trans, inode, ordered_extent->file_offset,
2934 &ordered_extent->list);
2936 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2937 ret = btrfs_update_inode_fallback(trans, root, inode);
2938 if (ret) { /* -ENOMEM or corruption */
2939 btrfs_abort_transaction(trans, root, ret);
2944 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2945 ordered_extent->file_offset +
2946 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2948 if (root != root->fs_info->tree_root)
2949 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2951 btrfs_end_transaction(trans, root);
2953 if (ret || truncated) {
2957 start = ordered_extent->file_offset + logical_len;
2959 start = ordered_extent->file_offset;
2960 end = ordered_extent->file_offset + ordered_extent->len - 1;
2961 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2963 /* Drop the cache for the part of the extent we didn't write. */
2964 btrfs_drop_extent_cache(inode, start, end, 0);
2967 * If the ordered extent had an IOERR or something else went
2968 * wrong we need to return the space for this ordered extent
2969 * back to the allocator. We only free the extent in the
2970 * truncated case if we didn't write out the extent at all.
2972 if ((ret || !logical_len) &&
2973 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2974 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2975 btrfs_free_reserved_extent(root, ordered_extent->start,
2976 ordered_extent->disk_len, 1);
2981 * This needs to be done to make sure anybody waiting knows we are done
2982 * updating everything for this ordered extent.
2984 btrfs_remove_ordered_extent(inode, ordered_extent);
2986 /* for snapshot-aware defrag */
2989 free_sa_defrag_extent(new);
2990 atomic_dec(&root->fs_info->defrag_running);
2992 relink_file_extents(new);
2997 btrfs_put_ordered_extent(ordered_extent);
2998 /* once for the tree */
2999 btrfs_put_ordered_extent(ordered_extent);
3004 static void finish_ordered_fn(struct btrfs_work *work)
3006 struct btrfs_ordered_extent *ordered_extent;
3007 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3008 btrfs_finish_ordered_io(ordered_extent);
3011 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3012 struct extent_state *state, int uptodate)
3014 struct inode *inode = page->mapping->host;
3015 struct btrfs_root *root = BTRFS_I(inode)->root;
3016 struct btrfs_ordered_extent *ordered_extent = NULL;
3017 struct btrfs_workqueue *wq;
3018 btrfs_work_func_t func;
3020 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3022 ClearPagePrivate2(page);
3023 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3024 end - start + 1, uptodate))
3027 if (btrfs_is_free_space_inode(inode)) {
3028 wq = root->fs_info->endio_freespace_worker;
3029 func = btrfs_freespace_write_helper;
3031 wq = root->fs_info->endio_write_workers;
3032 func = btrfs_endio_write_helper;
3035 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3037 btrfs_queue_work(wq, &ordered_extent->work);
3042 static int __readpage_endio_check(struct inode *inode,
3043 struct btrfs_io_bio *io_bio,
3044 int icsum, struct page *page,
3045 int pgoff, u64 start, size_t len)
3051 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3053 kaddr = kmap_atomic(page);
3054 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3055 btrfs_csum_final(csum, (char *)&csum);
3056 if (csum != csum_expected)
3059 kunmap_atomic(kaddr);
3062 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3063 "csum failed ino %llu off %llu csum %u expected csum %u",
3064 btrfs_ino(inode), start, csum, csum_expected);
3065 memset(kaddr + pgoff, 1, len);
3066 flush_dcache_page(page);
3067 kunmap_atomic(kaddr);
3068 if (csum_expected == 0)
3074 * when reads are done, we need to check csums to verify the data is correct
3075 * if there's a match, we allow the bio to finish. If not, the code in
3076 * extent_io.c will try to find good copies for us.
3078 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3079 u64 phy_offset, struct page *page,
3080 u64 start, u64 end, int mirror)
3082 size_t offset = start - page_offset(page);
3083 struct inode *inode = page->mapping->host;
3084 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3085 struct btrfs_root *root = BTRFS_I(inode)->root;
3087 if (PageChecked(page)) {
3088 ClearPageChecked(page);
3092 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3095 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3096 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3097 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3102 phy_offset >>= inode->i_sb->s_blocksize_bits;
3103 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3104 start, (size_t)(end - start + 1));
3107 struct delayed_iput {
3108 struct list_head list;
3109 struct inode *inode;
3112 /* JDM: If this is fs-wide, why can't we add a pointer to
3113 * btrfs_inode instead and avoid the allocation? */
3114 void btrfs_add_delayed_iput(struct inode *inode)
3116 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3117 struct delayed_iput *delayed;
3119 if (atomic_add_unless(&inode->i_count, -1, 1))
3122 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3123 delayed->inode = inode;
3125 spin_lock(&fs_info->delayed_iput_lock);
3126 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3127 spin_unlock(&fs_info->delayed_iput_lock);
3130 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3133 struct btrfs_fs_info *fs_info = root->fs_info;
3134 struct delayed_iput *delayed;
3137 spin_lock(&fs_info->delayed_iput_lock);
3138 empty = list_empty(&fs_info->delayed_iputs);
3139 spin_unlock(&fs_info->delayed_iput_lock);
3143 down_read(&fs_info->delayed_iput_sem);
3145 spin_lock(&fs_info->delayed_iput_lock);
3146 list_splice_init(&fs_info->delayed_iputs, &list);
3147 spin_unlock(&fs_info->delayed_iput_lock);
3149 while (!list_empty(&list)) {
3150 delayed = list_entry(list.next, struct delayed_iput, list);
3151 list_del(&delayed->list);
3152 iput(delayed->inode);
3156 up_read(&root->fs_info->delayed_iput_sem);
3160 * This is called in transaction commit time. If there are no orphan
3161 * files in the subvolume, it removes orphan item and frees block_rsv
3164 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3165 struct btrfs_root *root)
3167 struct btrfs_block_rsv *block_rsv;
3170 if (atomic_read(&root->orphan_inodes) ||
3171 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3174 spin_lock(&root->orphan_lock);
3175 if (atomic_read(&root->orphan_inodes)) {
3176 spin_unlock(&root->orphan_lock);
3180 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3181 spin_unlock(&root->orphan_lock);
3185 block_rsv = root->orphan_block_rsv;
3186 root->orphan_block_rsv = NULL;
3187 spin_unlock(&root->orphan_lock);
3189 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3190 btrfs_root_refs(&root->root_item) > 0) {
3191 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3192 root->root_key.objectid);
3194 btrfs_abort_transaction(trans, root, ret);
3196 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3201 WARN_ON(block_rsv->size > 0);
3202 btrfs_free_block_rsv(root, block_rsv);
3207 * This creates an orphan entry for the given inode in case something goes
3208 * wrong in the middle of an unlink/truncate.
3210 * NOTE: caller of this function should reserve 5 units of metadata for
3213 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3215 struct btrfs_root *root = BTRFS_I(inode)->root;
3216 struct btrfs_block_rsv *block_rsv = NULL;
3221 if (!root->orphan_block_rsv) {
3222 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3227 spin_lock(&root->orphan_lock);
3228 if (!root->orphan_block_rsv) {
3229 root->orphan_block_rsv = block_rsv;
3230 } else if (block_rsv) {
3231 btrfs_free_block_rsv(root, block_rsv);
3235 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3236 &BTRFS_I(inode)->runtime_flags)) {
3239 * For proper ENOSPC handling, we should do orphan
3240 * cleanup when mounting. But this introduces backward
3241 * compatibility issue.
3243 if (!xchg(&root->orphan_item_inserted, 1))
3249 atomic_inc(&root->orphan_inodes);
3252 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3253 &BTRFS_I(inode)->runtime_flags))
3255 spin_unlock(&root->orphan_lock);
3257 /* grab metadata reservation from transaction handle */
3259 ret = btrfs_orphan_reserve_metadata(trans, inode);
3260 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3263 /* insert an orphan item to track this unlinked/truncated file */
3265 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3267 atomic_dec(&root->orphan_inodes);
3269 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3270 &BTRFS_I(inode)->runtime_flags);
3271 btrfs_orphan_release_metadata(inode);
3273 if (ret != -EEXIST) {
3274 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3275 &BTRFS_I(inode)->runtime_flags);
3276 btrfs_abort_transaction(trans, root, ret);
3283 /* insert an orphan item to track subvolume contains orphan files */
3285 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3286 root->root_key.objectid);
3287 if (ret && ret != -EEXIST) {
3288 btrfs_abort_transaction(trans, root, ret);
3296 * We have done the truncate/delete so we can go ahead and remove the orphan
3297 * item for this particular inode.
3299 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3300 struct inode *inode)
3302 struct btrfs_root *root = BTRFS_I(inode)->root;
3303 int delete_item = 0;
3304 int release_rsv = 0;
3307 spin_lock(&root->orphan_lock);
3308 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3309 &BTRFS_I(inode)->runtime_flags))
3312 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3313 &BTRFS_I(inode)->runtime_flags))
3315 spin_unlock(&root->orphan_lock);
3318 atomic_dec(&root->orphan_inodes);
3320 ret = btrfs_del_orphan_item(trans, root,
3325 btrfs_orphan_release_metadata(inode);
3331 * this cleans up any orphans that may be left on the list from the last use
3334 int btrfs_orphan_cleanup(struct btrfs_root *root)
3336 struct btrfs_path *path;
3337 struct extent_buffer *leaf;
3338 struct btrfs_key key, found_key;
3339 struct btrfs_trans_handle *trans;
3340 struct inode *inode;
3341 u64 last_objectid = 0;
3342 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3344 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3347 path = btrfs_alloc_path();
3354 key.objectid = BTRFS_ORPHAN_OBJECTID;
3355 key.type = BTRFS_ORPHAN_ITEM_KEY;
3356 key.offset = (u64)-1;
3359 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3364 * if ret == 0 means we found what we were searching for, which
3365 * is weird, but possible, so only screw with path if we didn't
3366 * find the key and see if we have stuff that matches
3370 if (path->slots[0] == 0)
3375 /* pull out the item */
3376 leaf = path->nodes[0];
3377 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3379 /* make sure the item matches what we want */
3380 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3382 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3385 /* release the path since we're done with it */
3386 btrfs_release_path(path);
3389 * this is where we are basically btrfs_lookup, without the
3390 * crossing root thing. we store the inode number in the
3391 * offset of the orphan item.
3394 if (found_key.offset == last_objectid) {
3395 btrfs_err(root->fs_info,
3396 "Error removing orphan entry, stopping orphan cleanup");
3401 last_objectid = found_key.offset;
3403 found_key.objectid = found_key.offset;
3404 found_key.type = BTRFS_INODE_ITEM_KEY;
3405 found_key.offset = 0;
3406 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3407 ret = PTR_ERR_OR_ZERO(inode);
3408 if (ret && ret != -ESTALE)
3411 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3412 struct btrfs_root *dead_root;
3413 struct btrfs_fs_info *fs_info = root->fs_info;
3414 int is_dead_root = 0;
3417 * this is an orphan in the tree root. Currently these
3418 * could come from 2 sources:
3419 * a) a snapshot deletion in progress
3420 * b) a free space cache inode
3421 * We need to distinguish those two, as the snapshot
3422 * orphan must not get deleted.
3423 * find_dead_roots already ran before us, so if this
3424 * is a snapshot deletion, we should find the root
3425 * in the dead_roots list
3427 spin_lock(&fs_info->trans_lock);
3428 list_for_each_entry(dead_root, &fs_info->dead_roots,
3430 if (dead_root->root_key.objectid ==
3431 found_key.objectid) {
3436 spin_unlock(&fs_info->trans_lock);
3438 /* prevent this orphan from being found again */
3439 key.offset = found_key.objectid - 1;
3444 * Inode is already gone but the orphan item is still there,
3445 * kill the orphan item.
3447 if (ret == -ESTALE) {
3448 trans = btrfs_start_transaction(root, 1);
3449 if (IS_ERR(trans)) {
3450 ret = PTR_ERR(trans);
3453 btrfs_debug(root->fs_info, "auto deleting %Lu",
3454 found_key.objectid);
3455 ret = btrfs_del_orphan_item(trans, root,
3456 found_key.objectid);
3457 btrfs_end_transaction(trans, root);
3464 * add this inode to the orphan list so btrfs_orphan_del does
3465 * the proper thing when we hit it
3467 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3468 &BTRFS_I(inode)->runtime_flags);
3469 atomic_inc(&root->orphan_inodes);
3471 /* if we have links, this was a truncate, lets do that */
3472 if (inode->i_nlink) {
3473 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3479 /* 1 for the orphan item deletion. */
3480 trans = btrfs_start_transaction(root, 1);
3481 if (IS_ERR(trans)) {
3483 ret = PTR_ERR(trans);
3486 ret = btrfs_orphan_add(trans, inode);
3487 btrfs_end_transaction(trans, root);
3493 ret = btrfs_truncate(inode);
3495 btrfs_orphan_del(NULL, inode);
3500 /* this will do delete_inode and everything for us */
3505 /* release the path since we're done with it */
3506 btrfs_release_path(path);
3508 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3510 if (root->orphan_block_rsv)
3511 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3514 if (root->orphan_block_rsv ||
3515 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3516 trans = btrfs_join_transaction(root);
3518 btrfs_end_transaction(trans, root);
3522 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3524 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3528 btrfs_err(root->fs_info,
3529 "could not do orphan cleanup %d", ret);
3530 btrfs_free_path(path);
3535 * very simple check to peek ahead in the leaf looking for xattrs. If we
3536 * don't find any xattrs, we know there can't be any acls.
3538 * slot is the slot the inode is in, objectid is the objectid of the inode
3540 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3541 int slot, u64 objectid,
3542 int *first_xattr_slot)
3544 u32 nritems = btrfs_header_nritems(leaf);
3545 struct btrfs_key found_key;
3546 static u64 xattr_access = 0;
3547 static u64 xattr_default = 0;
3550 if (!xattr_access) {
3551 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3552 strlen(POSIX_ACL_XATTR_ACCESS));
3553 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3554 strlen(POSIX_ACL_XATTR_DEFAULT));
3558 *first_xattr_slot = -1;
3559 while (slot < nritems) {
3560 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3562 /* we found a different objectid, there must not be acls */
3563 if (found_key.objectid != objectid)
3566 /* we found an xattr, assume we've got an acl */
3567 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3568 if (*first_xattr_slot == -1)
3569 *first_xattr_slot = slot;
3570 if (found_key.offset == xattr_access ||
3571 found_key.offset == xattr_default)
3576 * we found a key greater than an xattr key, there can't
3577 * be any acls later on
3579 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3586 * it goes inode, inode backrefs, xattrs, extents,
3587 * so if there are a ton of hard links to an inode there can
3588 * be a lot of backrefs. Don't waste time searching too hard,
3589 * this is just an optimization
3594 /* we hit the end of the leaf before we found an xattr or
3595 * something larger than an xattr. We have to assume the inode
3598 if (*first_xattr_slot == -1)
3599 *first_xattr_slot = slot;
3604 * read an inode from the btree into the in-memory inode
3606 static void btrfs_read_locked_inode(struct inode *inode)
3608 struct btrfs_path *path;
3609 struct extent_buffer *leaf;
3610 struct btrfs_inode_item *inode_item;
3611 struct btrfs_root *root = BTRFS_I(inode)->root;
3612 struct btrfs_key location;
3617 bool filled = false;
3618 int first_xattr_slot;
3620 ret = btrfs_fill_inode(inode, &rdev);
3624 path = btrfs_alloc_path();
3628 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3630 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3634 leaf = path->nodes[0];
3639 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3640 struct btrfs_inode_item);
3641 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3642 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3643 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3644 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3645 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3647 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3648 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3650 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3651 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3653 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3654 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3656 BTRFS_I(inode)->i_otime.tv_sec =
3657 btrfs_timespec_sec(leaf, &inode_item->otime);
3658 BTRFS_I(inode)->i_otime.tv_nsec =
3659 btrfs_timespec_nsec(leaf, &inode_item->otime);
3661 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3662 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3663 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3665 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3666 inode->i_generation = BTRFS_I(inode)->generation;
3668 rdev = btrfs_inode_rdev(leaf, inode_item);
3670 BTRFS_I(inode)->index_cnt = (u64)-1;
3671 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3675 * If we were modified in the current generation and evicted from memory
3676 * and then re-read we need to do a full sync since we don't have any
3677 * idea about which extents were modified before we were evicted from
3680 * This is required for both inode re-read from disk and delayed inode
3681 * in delayed_nodes_tree.
3683 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3684 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3685 &BTRFS_I(inode)->runtime_flags);
3688 * We don't persist the id of the transaction where an unlink operation
3689 * against the inode was last made. So here we assume the inode might
3690 * have been evicted, and therefore the exact value of last_unlink_trans
3691 * lost, and set it to last_trans to avoid metadata inconsistencies
3692 * between the inode and its parent if the inode is fsync'ed and the log
3693 * replayed. For example, in the scenario:
3696 * ln mydir/foo mydir/bar
3699 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3700 * xfs_io -c fsync mydir/foo
3702 * mount fs, triggers fsync log replay
3704 * We must make sure that when we fsync our inode foo we also log its
3705 * parent inode, otherwise after log replay the parent still has the
3706 * dentry with the "bar" name but our inode foo has a link count of 1
3707 * and doesn't have an inode ref with the name "bar" anymore.
3709 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3710 * but it guarantees correctness at the expense of ocassional full
3711 * transaction commits on fsync if our inode is a directory, or if our
3712 * inode is not a directory, logging its parent unnecessarily.
3714 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3717 if (inode->i_nlink != 1 ||
3718 path->slots[0] >= btrfs_header_nritems(leaf))
3721 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3722 if (location.objectid != btrfs_ino(inode))
3725 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3726 if (location.type == BTRFS_INODE_REF_KEY) {
3727 struct btrfs_inode_ref *ref;
3729 ref = (struct btrfs_inode_ref *)ptr;
3730 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3731 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3732 struct btrfs_inode_extref *extref;
3734 extref = (struct btrfs_inode_extref *)ptr;
3735 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3740 * try to precache a NULL acl entry for files that don't have
3741 * any xattrs or acls
3743 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3744 btrfs_ino(inode), &first_xattr_slot);
3745 if (first_xattr_slot != -1) {
3746 path->slots[0] = first_xattr_slot;
3747 ret = btrfs_load_inode_props(inode, path);
3749 btrfs_err(root->fs_info,
3750 "error loading props for ino %llu (root %llu): %d",
3752 root->root_key.objectid, ret);
3754 btrfs_free_path(path);
3757 cache_no_acl(inode);
3759 switch (inode->i_mode & S_IFMT) {
3761 inode->i_mapping->a_ops = &btrfs_aops;
3762 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3763 inode->i_fop = &btrfs_file_operations;
3764 inode->i_op = &btrfs_file_inode_operations;
3767 inode->i_fop = &btrfs_dir_file_operations;
3768 if (root == root->fs_info->tree_root)
3769 inode->i_op = &btrfs_dir_ro_inode_operations;
3771 inode->i_op = &btrfs_dir_inode_operations;
3774 inode->i_op = &btrfs_symlink_inode_operations;
3775 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3778 inode->i_op = &btrfs_special_inode_operations;
3779 init_special_inode(inode, inode->i_mode, rdev);
3783 btrfs_update_iflags(inode);
3787 btrfs_free_path(path);
3788 make_bad_inode(inode);
3792 * given a leaf and an inode, copy the inode fields into the leaf
3794 static void fill_inode_item(struct btrfs_trans_handle *trans,
3795 struct extent_buffer *leaf,
3796 struct btrfs_inode_item *item,
3797 struct inode *inode)
3799 struct btrfs_map_token token;
3801 btrfs_init_map_token(&token);
3803 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3804 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3805 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3807 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3808 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3810 btrfs_set_token_timespec_sec(leaf, &item->atime,
3811 inode->i_atime.tv_sec, &token);
3812 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3813 inode->i_atime.tv_nsec, &token);
3815 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3816 inode->i_mtime.tv_sec, &token);
3817 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3818 inode->i_mtime.tv_nsec, &token);
3820 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3821 inode->i_ctime.tv_sec, &token);
3822 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3823 inode->i_ctime.tv_nsec, &token);
3825 btrfs_set_token_timespec_sec(leaf, &item->otime,
3826 BTRFS_I(inode)->i_otime.tv_sec, &token);
3827 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3828 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3830 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3832 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3834 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3835 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3836 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3837 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3838 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3842 * copy everything in the in-memory inode into the btree.
3844 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3845 struct btrfs_root *root, struct inode *inode)
3847 struct btrfs_inode_item *inode_item;
3848 struct btrfs_path *path;
3849 struct extent_buffer *leaf;
3852 path = btrfs_alloc_path();
3856 path->leave_spinning = 1;
3857 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3865 leaf = path->nodes[0];
3866 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3867 struct btrfs_inode_item);
3869 fill_inode_item(trans, leaf, inode_item, inode);
3870 btrfs_mark_buffer_dirty(leaf);
3871 btrfs_set_inode_last_trans(trans, inode);
3874 btrfs_free_path(path);
3879 * copy everything in the in-memory inode into the btree.
3881 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3882 struct btrfs_root *root, struct inode *inode)
3887 * If the inode is a free space inode, we can deadlock during commit
3888 * if we put it into the delayed code.
3890 * The data relocation inode should also be directly updated
3893 if (!btrfs_is_free_space_inode(inode)
3894 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3895 && !root->fs_info->log_root_recovering) {
3896 btrfs_update_root_times(trans, root);
3898 ret = btrfs_delayed_update_inode(trans, root, inode);
3900 btrfs_set_inode_last_trans(trans, inode);
3904 return btrfs_update_inode_item(trans, root, inode);
3907 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3908 struct btrfs_root *root,
3909 struct inode *inode)
3913 ret = btrfs_update_inode(trans, root, inode);
3915 return btrfs_update_inode_item(trans, root, inode);
3920 * unlink helper that gets used here in inode.c and in the tree logging
3921 * recovery code. It remove a link in a directory with a given name, and
3922 * also drops the back refs in the inode to the directory
3924 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3925 struct btrfs_root *root,
3926 struct inode *dir, struct inode *inode,
3927 const char *name, int name_len)
3929 struct btrfs_path *path;
3931 struct extent_buffer *leaf;
3932 struct btrfs_dir_item *di;
3933 struct btrfs_key key;
3935 u64 ino = btrfs_ino(inode);
3936 u64 dir_ino = btrfs_ino(dir);
3938 path = btrfs_alloc_path();
3944 path->leave_spinning = 1;
3945 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3946 name, name_len, -1);
3955 leaf = path->nodes[0];
3956 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3957 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3960 btrfs_release_path(path);
3963 * If we don't have dir index, we have to get it by looking up
3964 * the inode ref, since we get the inode ref, remove it directly,
3965 * it is unnecessary to do delayed deletion.
3967 * But if we have dir index, needn't search inode ref to get it.
3968 * Since the inode ref is close to the inode item, it is better
3969 * that we delay to delete it, and just do this deletion when
3970 * we update the inode item.
3972 if (BTRFS_I(inode)->dir_index) {
3973 ret = btrfs_delayed_delete_inode_ref(inode);
3975 index = BTRFS_I(inode)->dir_index;
3980 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3983 btrfs_info(root->fs_info,
3984 "failed to delete reference to %.*s, inode %llu parent %llu",
3985 name_len, name, ino, dir_ino);
3986 btrfs_abort_transaction(trans, root, ret);
3990 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3992 btrfs_abort_transaction(trans, root, ret);
3996 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3998 if (ret != 0 && ret != -ENOENT) {
3999 btrfs_abort_transaction(trans, root, ret);
4003 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4008 btrfs_abort_transaction(trans, root, ret);
4010 btrfs_free_path(path);
4014 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4015 inode_inc_iversion(inode);
4016 inode_inc_iversion(dir);
4017 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4018 ret = btrfs_update_inode(trans, root, dir);
4023 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4024 struct btrfs_root *root,
4025 struct inode *dir, struct inode *inode,
4026 const char *name, int name_len)
4029 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4032 ret = btrfs_update_inode(trans, root, inode);
4038 * helper to start transaction for unlink and rmdir.
4040 * unlink and rmdir are special in btrfs, they do not always free space, so
4041 * if we cannot make our reservations the normal way try and see if there is
4042 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4043 * allow the unlink to occur.
4045 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4047 struct btrfs_trans_handle *trans;
4048 struct btrfs_root *root = BTRFS_I(dir)->root;
4052 * 1 for the possible orphan item
4053 * 1 for the dir item
4054 * 1 for the dir index
4055 * 1 for the inode ref
4058 trans = btrfs_start_transaction(root, 5);
4059 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
4062 if (PTR_ERR(trans) == -ENOSPC) {
4063 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4065 trans = btrfs_start_transaction(root, 0);
4068 ret = btrfs_cond_migrate_bytes(root->fs_info,
4069 &root->fs_info->trans_block_rsv,
4072 btrfs_end_transaction(trans, root);
4073 return ERR_PTR(ret);
4075 trans->block_rsv = &root->fs_info->trans_block_rsv;
4076 trans->bytes_reserved = num_bytes;
4081 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4083 struct btrfs_root *root = BTRFS_I(dir)->root;
4084 struct btrfs_trans_handle *trans;
4085 struct inode *inode = d_inode(dentry);
4088 trans = __unlink_start_trans(dir);
4090 return PTR_ERR(trans);
4092 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4094 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4095 dentry->d_name.name, dentry->d_name.len);
4099 if (inode->i_nlink == 0) {
4100 ret = btrfs_orphan_add(trans, inode);
4106 btrfs_end_transaction(trans, root);
4107 btrfs_btree_balance_dirty(root);
4111 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4112 struct btrfs_root *root,
4113 struct inode *dir, u64 objectid,
4114 const char *name, int name_len)
4116 struct btrfs_path *path;
4117 struct extent_buffer *leaf;
4118 struct btrfs_dir_item *di;
4119 struct btrfs_key key;
4122 u64 dir_ino = btrfs_ino(dir);
4124 path = btrfs_alloc_path();
4128 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4129 name, name_len, -1);
4130 if (IS_ERR_OR_NULL(di)) {
4138 leaf = path->nodes[0];
4139 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4140 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4141 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4143 btrfs_abort_transaction(trans, root, ret);
4146 btrfs_release_path(path);
4148 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4149 objectid, root->root_key.objectid,
4150 dir_ino, &index, name, name_len);
4152 if (ret != -ENOENT) {
4153 btrfs_abort_transaction(trans, root, ret);
4156 di = btrfs_search_dir_index_item(root, path, dir_ino,
4158 if (IS_ERR_OR_NULL(di)) {
4163 btrfs_abort_transaction(trans, root, ret);
4167 leaf = path->nodes[0];
4168 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4169 btrfs_release_path(path);
4172 btrfs_release_path(path);
4174 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4176 btrfs_abort_transaction(trans, root, ret);
4180 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4181 inode_inc_iversion(dir);
4182 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4183 ret = btrfs_update_inode_fallback(trans, root, dir);
4185 btrfs_abort_transaction(trans, root, ret);
4187 btrfs_free_path(path);
4191 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4193 struct inode *inode = d_inode(dentry);
4195 struct btrfs_root *root = BTRFS_I(dir)->root;
4196 struct btrfs_trans_handle *trans;
4198 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4200 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4203 trans = __unlink_start_trans(dir);
4205 return PTR_ERR(trans);
4207 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4208 err = btrfs_unlink_subvol(trans, root, dir,
4209 BTRFS_I(inode)->location.objectid,
4210 dentry->d_name.name,
4211 dentry->d_name.len);
4215 err = btrfs_orphan_add(trans, inode);
4219 /* now the directory is empty */
4220 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4221 dentry->d_name.name, dentry->d_name.len);
4223 btrfs_i_size_write(inode, 0);
4225 btrfs_end_transaction(trans, root);
4226 btrfs_btree_balance_dirty(root);
4231 static int truncate_space_check(struct btrfs_trans_handle *trans,
4232 struct btrfs_root *root,
4237 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4238 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4239 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4241 trans->bytes_reserved += bytes_deleted;
4246 static int truncate_inline_extent(struct inode *inode,
4247 struct btrfs_path *path,
4248 struct btrfs_key *found_key,
4252 struct extent_buffer *leaf = path->nodes[0];
4253 int slot = path->slots[0];
4254 struct btrfs_file_extent_item *fi;
4255 u32 size = (u32)(new_size - found_key->offset);
4256 struct btrfs_root *root = BTRFS_I(inode)->root;
4258 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4260 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4261 loff_t offset = new_size;
4262 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4265 * Zero out the remaining of the last page of our inline extent,
4266 * instead of directly truncating our inline extent here - that
4267 * would be much more complex (decompressing all the data, then
4268 * compressing the truncated data, which might be bigger than
4269 * the size of the inline extent, resize the extent, etc).
4270 * We release the path because to get the page we might need to
4271 * read the extent item from disk (data not in the page cache).
4273 btrfs_release_path(path);
4274 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4277 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4278 size = btrfs_file_extent_calc_inline_size(size);
4279 btrfs_truncate_item(root, path, size, 1);
4281 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4282 inode_sub_bytes(inode, item_end + 1 - new_size);
4288 * this can truncate away extent items, csum items and directory items.
4289 * It starts at a high offset and removes keys until it can't find
4290 * any higher than new_size
4292 * csum items that cross the new i_size are truncated to the new size
4295 * min_type is the minimum key type to truncate down to. If set to 0, this
4296 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4298 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4299 struct btrfs_root *root,
4300 struct inode *inode,
4301 u64 new_size, u32 min_type)
4303 struct btrfs_path *path;
4304 struct extent_buffer *leaf;
4305 struct btrfs_file_extent_item *fi;
4306 struct btrfs_key key;
4307 struct btrfs_key found_key;
4308 u64 extent_start = 0;
4309 u64 extent_num_bytes = 0;
4310 u64 extent_offset = 0;
4312 u64 last_size = new_size;
4313 u32 found_type = (u8)-1;
4316 int pending_del_nr = 0;
4317 int pending_del_slot = 0;
4318 int extent_type = -1;
4321 u64 ino = btrfs_ino(inode);
4322 u64 bytes_deleted = 0;
4324 bool should_throttle = 0;
4325 bool should_end = 0;
4327 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4330 * for non-free space inodes and ref cows, we want to back off from
4333 if (!btrfs_is_free_space_inode(inode) &&
4334 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4337 path = btrfs_alloc_path();
4343 * We want to drop from the next block forward in case this new size is
4344 * not block aligned since we will be keeping the last block of the
4345 * extent just the way it is.
4347 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4348 root == root->fs_info->tree_root)
4349 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4350 root->sectorsize), (u64)-1, 0);
4353 * This function is also used to drop the items in the log tree before
4354 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4355 * it is used to drop the loged items. So we shouldn't kill the delayed
4358 if (min_type == 0 && root == BTRFS_I(inode)->root)
4359 btrfs_kill_delayed_inode_items(inode);
4362 key.offset = (u64)-1;
4367 * with a 16K leaf size and 128MB extents, you can actually queue
4368 * up a huge file in a single leaf. Most of the time that
4369 * bytes_deleted is > 0, it will be huge by the time we get here
4371 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4372 if (btrfs_should_end_transaction(trans, root)) {
4379 path->leave_spinning = 1;
4380 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4387 /* there are no items in the tree for us to truncate, we're
4390 if (path->slots[0] == 0)
4397 leaf = path->nodes[0];
4398 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4399 found_type = found_key.type;
4401 if (found_key.objectid != ino)
4404 if (found_type < min_type)
4407 item_end = found_key.offset;
4408 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4409 fi = btrfs_item_ptr(leaf, path->slots[0],
4410 struct btrfs_file_extent_item);
4411 extent_type = btrfs_file_extent_type(leaf, fi);
4412 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4414 btrfs_file_extent_num_bytes(leaf, fi);
4415 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4416 item_end += btrfs_file_extent_inline_len(leaf,
4417 path->slots[0], fi);
4421 if (found_type > min_type) {
4424 if (item_end < new_size)
4426 if (found_key.offset >= new_size)
4432 /* FIXME, shrink the extent if the ref count is only 1 */
4433 if (found_type != BTRFS_EXTENT_DATA_KEY)
4437 last_size = found_key.offset;
4439 last_size = new_size;
4441 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4443 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4445 u64 orig_num_bytes =
4446 btrfs_file_extent_num_bytes(leaf, fi);
4447 extent_num_bytes = ALIGN(new_size -
4450 btrfs_set_file_extent_num_bytes(leaf, fi,
4452 num_dec = (orig_num_bytes -
4454 if (test_bit(BTRFS_ROOT_REF_COWS,
4457 inode_sub_bytes(inode, num_dec);
4458 btrfs_mark_buffer_dirty(leaf);
4461 btrfs_file_extent_disk_num_bytes(leaf,
4463 extent_offset = found_key.offset -
4464 btrfs_file_extent_offset(leaf, fi);
4466 /* FIXME blocksize != 4096 */
4467 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4468 if (extent_start != 0) {
4470 if (test_bit(BTRFS_ROOT_REF_COWS,
4472 inode_sub_bytes(inode, num_dec);
4475 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4477 * we can't truncate inline items that have had
4481 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4482 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4485 * Need to release path in order to truncate a
4486 * compressed extent. So delete any accumulated
4487 * extent items so far.
4489 if (btrfs_file_extent_compression(leaf, fi) !=
4490 BTRFS_COMPRESS_NONE && pending_del_nr) {
4491 err = btrfs_del_items(trans, root, path,
4495 btrfs_abort_transaction(trans,
4503 err = truncate_inline_extent(inode, path,
4508 btrfs_abort_transaction(trans,
4512 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4514 inode_sub_bytes(inode, item_end + 1 - new_size);
4519 if (!pending_del_nr) {
4520 /* no pending yet, add ourselves */
4521 pending_del_slot = path->slots[0];
4523 } else if (pending_del_nr &&
4524 path->slots[0] + 1 == pending_del_slot) {
4525 /* hop on the pending chunk */
4527 pending_del_slot = path->slots[0];
4534 should_throttle = 0;
4537 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4538 root == root->fs_info->tree_root)) {
4539 btrfs_set_path_blocking(path);
4540 bytes_deleted += extent_num_bytes;
4541 ret = btrfs_free_extent(trans, root, extent_start,
4542 extent_num_bytes, 0,
4543 btrfs_header_owner(leaf),
4544 ino, extent_offset, 0);
4546 if (btrfs_should_throttle_delayed_refs(trans, root))
4547 btrfs_async_run_delayed_refs(root,
4548 trans->delayed_ref_updates * 2, 0);
4550 if (truncate_space_check(trans, root,
4551 extent_num_bytes)) {
4554 if (btrfs_should_throttle_delayed_refs(trans,
4556 should_throttle = 1;
4561 if (found_type == BTRFS_INODE_ITEM_KEY)
4564 if (path->slots[0] == 0 ||
4565 path->slots[0] != pending_del_slot ||
4566 should_throttle || should_end) {
4567 if (pending_del_nr) {
4568 ret = btrfs_del_items(trans, root, path,
4572 btrfs_abort_transaction(trans,
4578 btrfs_release_path(path);
4579 if (should_throttle) {
4580 unsigned long updates = trans->delayed_ref_updates;
4582 trans->delayed_ref_updates = 0;
4583 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4589 * if we failed to refill our space rsv, bail out
4590 * and let the transaction restart
4602 if (pending_del_nr) {
4603 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4606 btrfs_abort_transaction(trans, root, ret);
4609 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4610 btrfs_ordered_update_i_size(inode, last_size, NULL);
4612 btrfs_free_path(path);
4614 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4615 unsigned long updates = trans->delayed_ref_updates;
4617 trans->delayed_ref_updates = 0;
4618 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4627 * btrfs_truncate_page - read, zero a chunk and write a page
4628 * @inode - inode that we're zeroing
4629 * @from - the offset to start zeroing
4630 * @len - the length to zero, 0 to zero the entire range respective to the
4632 * @front - zero up to the offset instead of from the offset on
4634 * This will find the page for the "from" offset and cow the page and zero the
4635 * part we want to zero. This is used with truncate and hole punching.
4637 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4640 struct address_space *mapping = inode->i_mapping;
4641 struct btrfs_root *root = BTRFS_I(inode)->root;
4642 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4643 struct btrfs_ordered_extent *ordered;
4644 struct extent_state *cached_state = NULL;
4646 u32 blocksize = root->sectorsize;
4647 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4648 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4650 gfp_t mask = btrfs_alloc_write_mask(mapping);
4655 if ((offset & (blocksize - 1)) == 0 &&
4656 (!len || ((len & (blocksize - 1)) == 0)))
4658 ret = btrfs_delalloc_reserve_space(inode,
4659 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4664 page = find_or_create_page(mapping, index, mask);
4666 btrfs_delalloc_release_space(inode,
4667 round_down(from, PAGE_CACHE_SIZE),
4673 page_start = page_offset(page);
4674 page_end = page_start + PAGE_CACHE_SIZE - 1;
4676 if (!PageUptodate(page)) {
4677 ret = btrfs_readpage(NULL, page);
4679 if (page->mapping != mapping) {
4681 page_cache_release(page);
4684 if (!PageUptodate(page)) {
4689 wait_on_page_writeback(page);
4691 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4692 set_page_extent_mapped(page);
4694 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4696 unlock_extent_cached(io_tree, page_start, page_end,
4697 &cached_state, GFP_NOFS);
4699 page_cache_release(page);
4700 btrfs_start_ordered_extent(inode, ordered, 1);
4701 btrfs_put_ordered_extent(ordered);
4705 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4706 EXTENT_DIRTY | EXTENT_DELALLOC |
4707 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4708 0, 0, &cached_state, GFP_NOFS);
4710 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4713 unlock_extent_cached(io_tree, page_start, page_end,
4714 &cached_state, GFP_NOFS);
4718 if (offset != PAGE_CACHE_SIZE) {
4720 len = PAGE_CACHE_SIZE - offset;
4723 memset(kaddr, 0, offset);
4725 memset(kaddr + offset, 0, len);
4726 flush_dcache_page(page);
4729 ClearPageChecked(page);
4730 set_page_dirty(page);
4731 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4736 btrfs_delalloc_release_space(inode, page_start,
4739 page_cache_release(page);
4744 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4745 u64 offset, u64 len)
4747 struct btrfs_trans_handle *trans;
4751 * Still need to make sure the inode looks like it's been updated so
4752 * that any holes get logged if we fsync.
4754 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4755 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4756 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4757 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4762 * 1 - for the one we're dropping
4763 * 1 - for the one we're adding
4764 * 1 - for updating the inode.
4766 trans = btrfs_start_transaction(root, 3);
4768 return PTR_ERR(trans);
4770 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4772 btrfs_abort_transaction(trans, root, ret);
4773 btrfs_end_transaction(trans, root);
4777 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4778 0, 0, len, 0, len, 0, 0, 0);
4780 btrfs_abort_transaction(trans, root, ret);
4782 btrfs_update_inode(trans, root, inode);
4783 btrfs_end_transaction(trans, root);
4788 * This function puts in dummy file extents for the area we're creating a hole
4789 * for. So if we are truncating this file to a larger size we need to insert
4790 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4791 * the range between oldsize and size
4793 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4795 struct btrfs_root *root = BTRFS_I(inode)->root;
4796 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4797 struct extent_map *em = NULL;
4798 struct extent_state *cached_state = NULL;
4799 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4800 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4801 u64 block_end = ALIGN(size, root->sectorsize);
4808 * If our size started in the middle of a page we need to zero out the
4809 * rest of the page before we expand the i_size, otherwise we could
4810 * expose stale data.
4812 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4816 if (size <= hole_start)
4820 struct btrfs_ordered_extent *ordered;
4822 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4824 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4825 block_end - hole_start);
4828 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4829 &cached_state, GFP_NOFS);
4830 btrfs_start_ordered_extent(inode, ordered, 1);
4831 btrfs_put_ordered_extent(ordered);
4834 cur_offset = hole_start;
4836 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4837 block_end - cur_offset, 0);
4843 last_byte = min(extent_map_end(em), block_end);
4844 last_byte = ALIGN(last_byte , root->sectorsize);
4845 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4846 struct extent_map *hole_em;
4847 hole_size = last_byte - cur_offset;
4849 err = maybe_insert_hole(root, inode, cur_offset,
4853 btrfs_drop_extent_cache(inode, cur_offset,
4854 cur_offset + hole_size - 1, 0);
4855 hole_em = alloc_extent_map();
4857 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4858 &BTRFS_I(inode)->runtime_flags);
4861 hole_em->start = cur_offset;
4862 hole_em->len = hole_size;
4863 hole_em->orig_start = cur_offset;
4865 hole_em->block_start = EXTENT_MAP_HOLE;
4866 hole_em->block_len = 0;
4867 hole_em->orig_block_len = 0;
4868 hole_em->ram_bytes = hole_size;
4869 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4870 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4871 hole_em->generation = root->fs_info->generation;
4874 write_lock(&em_tree->lock);
4875 err = add_extent_mapping(em_tree, hole_em, 1);
4876 write_unlock(&em_tree->lock);
4879 btrfs_drop_extent_cache(inode, cur_offset,
4883 free_extent_map(hole_em);
4886 free_extent_map(em);
4888 cur_offset = last_byte;
4889 if (cur_offset >= block_end)
4892 free_extent_map(em);
4893 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4898 static int wait_snapshoting_atomic_t(atomic_t *a)
4904 static void wait_for_snapshot_creation(struct btrfs_root *root)
4909 ret = btrfs_start_write_no_snapshoting(root);
4912 wait_on_atomic_t(&root->will_be_snapshoted,
4913 wait_snapshoting_atomic_t,
4914 TASK_UNINTERRUPTIBLE);
4918 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4920 struct btrfs_root *root = BTRFS_I(inode)->root;
4921 struct btrfs_trans_handle *trans;
4922 loff_t oldsize = i_size_read(inode);
4923 loff_t newsize = attr->ia_size;
4924 int mask = attr->ia_valid;
4928 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4929 * special case where we need to update the times despite not having
4930 * these flags set. For all other operations the VFS set these flags
4931 * explicitly if it wants a timestamp update.
4933 if (newsize != oldsize) {
4934 inode_inc_iversion(inode);
4935 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4936 inode->i_ctime = inode->i_mtime =
4937 current_fs_time(inode->i_sb);
4940 if (newsize > oldsize) {
4941 truncate_pagecache(inode, newsize);
4943 * Don't do an expanding truncate while snapshoting is ongoing.
4944 * This is to ensure the snapshot captures a fully consistent
4945 * state of this file - if the snapshot captures this expanding
4946 * truncation, it must capture all writes that happened before
4949 wait_for_snapshot_creation(root);
4950 ret = btrfs_cont_expand(inode, oldsize, newsize);
4952 btrfs_end_write_no_snapshoting(root);
4956 trans = btrfs_start_transaction(root, 1);
4957 if (IS_ERR(trans)) {
4958 btrfs_end_write_no_snapshoting(root);
4959 return PTR_ERR(trans);
4962 i_size_write(inode, newsize);
4963 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4964 ret = btrfs_update_inode(trans, root, inode);
4965 btrfs_end_write_no_snapshoting(root);
4966 btrfs_end_transaction(trans, root);
4970 * We're truncating a file that used to have good data down to
4971 * zero. Make sure it gets into the ordered flush list so that
4972 * any new writes get down to disk quickly.
4975 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4976 &BTRFS_I(inode)->runtime_flags);
4979 * 1 for the orphan item we're going to add
4980 * 1 for the orphan item deletion.
4982 trans = btrfs_start_transaction(root, 2);
4984 return PTR_ERR(trans);
4987 * We need to do this in case we fail at _any_ point during the
4988 * actual truncate. Once we do the truncate_setsize we could
4989 * invalidate pages which forces any outstanding ordered io to
4990 * be instantly completed which will give us extents that need
4991 * to be truncated. If we fail to get an orphan inode down we
4992 * could have left over extents that were never meant to live,
4993 * so we need to garuntee from this point on that everything
4994 * will be consistent.
4996 ret = btrfs_orphan_add(trans, inode);
4997 btrfs_end_transaction(trans, root);
5001 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5002 truncate_setsize(inode, newsize);
5004 /* Disable nonlocked read DIO to avoid the end less truncate */
5005 btrfs_inode_block_unlocked_dio(inode);
5006 inode_dio_wait(inode);
5007 btrfs_inode_resume_unlocked_dio(inode);
5009 ret = btrfs_truncate(inode);
5010 if (ret && inode->i_nlink) {
5014 * failed to truncate, disk_i_size is only adjusted down
5015 * as we remove extents, so it should represent the true
5016 * size of the inode, so reset the in memory size and
5017 * delete our orphan entry.
5019 trans = btrfs_join_transaction(root);
5020 if (IS_ERR(trans)) {
5021 btrfs_orphan_del(NULL, inode);
5024 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5025 err = btrfs_orphan_del(trans, inode);
5027 btrfs_abort_transaction(trans, root, err);
5028 btrfs_end_transaction(trans, root);
5035 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5037 struct inode *inode = d_inode(dentry);
5038 struct btrfs_root *root = BTRFS_I(inode)->root;
5041 if (btrfs_root_readonly(root))
5044 err = inode_change_ok(inode, attr);
5048 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5049 err = btrfs_setsize(inode, attr);
5054 if (attr->ia_valid) {
5055 setattr_copy(inode, attr);
5056 inode_inc_iversion(inode);
5057 err = btrfs_dirty_inode(inode);
5059 if (!err && attr->ia_valid & ATTR_MODE)
5060 err = posix_acl_chmod(inode, inode->i_mode);
5067 * While truncating the inode pages during eviction, we get the VFS calling
5068 * btrfs_invalidatepage() against each page of the inode. This is slow because
5069 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5070 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5071 * extent_state structures over and over, wasting lots of time.
5073 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5074 * those expensive operations on a per page basis and do only the ordered io
5075 * finishing, while we release here the extent_map and extent_state structures,
5076 * without the excessive merging and splitting.
5078 static void evict_inode_truncate_pages(struct inode *inode)
5080 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5081 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5082 struct rb_node *node;
5084 ASSERT(inode->i_state & I_FREEING);
5085 truncate_inode_pages_final(&inode->i_data);
5087 write_lock(&map_tree->lock);
5088 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5089 struct extent_map *em;
5091 node = rb_first(&map_tree->map);
5092 em = rb_entry(node, struct extent_map, rb_node);
5093 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5094 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5095 remove_extent_mapping(map_tree, em);
5096 free_extent_map(em);
5097 if (need_resched()) {
5098 write_unlock(&map_tree->lock);
5100 write_lock(&map_tree->lock);
5103 write_unlock(&map_tree->lock);
5106 * Keep looping until we have no more ranges in the io tree.
5107 * We can have ongoing bios started by readpages (called from readahead)
5108 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5109 * still in progress (unlocked the pages in the bio but did not yet
5110 * unlocked the ranges in the io tree). Therefore this means some
5111 * ranges can still be locked and eviction started because before
5112 * submitting those bios, which are executed by a separate task (work
5113 * queue kthread), inode references (inode->i_count) were not taken
5114 * (which would be dropped in the end io callback of each bio).
5115 * Therefore here we effectively end up waiting for those bios and
5116 * anyone else holding locked ranges without having bumped the inode's
5117 * reference count - if we don't do it, when they access the inode's
5118 * io_tree to unlock a range it may be too late, leading to an
5119 * use-after-free issue.
5121 spin_lock(&io_tree->lock);
5122 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5123 struct extent_state *state;
5124 struct extent_state *cached_state = NULL;
5128 node = rb_first(&io_tree->state);
5129 state = rb_entry(node, struct extent_state, rb_node);
5130 start = state->start;
5132 spin_unlock(&io_tree->lock);
5134 lock_extent_bits(io_tree, start, end, 0, &cached_state);
5137 * If still has DELALLOC flag, the extent didn't reach disk,
5138 * and its reserved space won't be freed by delayed_ref.
5139 * So we need to free its reserved space here.
5140 * (Refer to comment in btrfs_invalidatepage, case 2)
5142 * Note, end is the bytenr of last byte, so we need + 1 here.
5144 if (state->state & EXTENT_DELALLOC)
5145 btrfs_qgroup_free_data(inode, start, end - start + 1);
5147 clear_extent_bit(io_tree, start, end,
5148 EXTENT_LOCKED | EXTENT_DIRTY |
5149 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5150 EXTENT_DEFRAG, 1, 1,
5151 &cached_state, GFP_NOFS);
5154 spin_lock(&io_tree->lock);
5156 spin_unlock(&io_tree->lock);
5159 void btrfs_evict_inode(struct inode *inode)
5161 struct btrfs_trans_handle *trans;
5162 struct btrfs_root *root = BTRFS_I(inode)->root;
5163 struct btrfs_block_rsv *rsv, *global_rsv;
5164 int steal_from_global = 0;
5165 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5168 trace_btrfs_inode_evict(inode);
5170 evict_inode_truncate_pages(inode);
5172 if (inode->i_nlink &&
5173 ((btrfs_root_refs(&root->root_item) != 0 &&
5174 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5175 btrfs_is_free_space_inode(inode)))
5178 if (is_bad_inode(inode)) {
5179 btrfs_orphan_del(NULL, inode);
5182 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5183 if (!special_file(inode->i_mode))
5184 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5186 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5188 if (root->fs_info->log_root_recovering) {
5189 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5190 &BTRFS_I(inode)->runtime_flags));
5194 if (inode->i_nlink > 0) {
5195 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5196 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5200 ret = btrfs_commit_inode_delayed_inode(inode);
5202 btrfs_orphan_del(NULL, inode);
5206 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5208 btrfs_orphan_del(NULL, inode);
5211 rsv->size = min_size;
5213 global_rsv = &root->fs_info->global_block_rsv;
5215 btrfs_i_size_write(inode, 0);
5218 * This is a bit simpler than btrfs_truncate since we've already
5219 * reserved our space for our orphan item in the unlink, so we just
5220 * need to reserve some slack space in case we add bytes and update
5221 * inode item when doing the truncate.
5224 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5225 BTRFS_RESERVE_FLUSH_LIMIT);
5228 * Try and steal from the global reserve since we will
5229 * likely not use this space anyway, we want to try as
5230 * hard as possible to get this to work.
5233 steal_from_global++;
5235 steal_from_global = 0;
5239 * steal_from_global == 0: we reserved stuff, hooray!
5240 * steal_from_global == 1: we didn't reserve stuff, boo!
5241 * steal_from_global == 2: we've committed, still not a lot of
5242 * room but maybe we'll have room in the global reserve this
5244 * steal_from_global == 3: abandon all hope!
5246 if (steal_from_global > 2) {
5247 btrfs_warn(root->fs_info,
5248 "Could not get space for a delete, will truncate on mount %d",
5250 btrfs_orphan_del(NULL, inode);
5251 btrfs_free_block_rsv(root, rsv);
5255 trans = btrfs_join_transaction(root);
5256 if (IS_ERR(trans)) {
5257 btrfs_orphan_del(NULL, inode);
5258 btrfs_free_block_rsv(root, rsv);
5263 * We can't just steal from the global reserve, we need tomake
5264 * sure there is room to do it, if not we need to commit and try
5267 if (steal_from_global) {
5268 if (!btrfs_check_space_for_delayed_refs(trans, root))
5269 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5276 * Couldn't steal from the global reserve, we have too much
5277 * pending stuff built up, commit the transaction and try it
5281 ret = btrfs_commit_transaction(trans, root);
5283 btrfs_orphan_del(NULL, inode);
5284 btrfs_free_block_rsv(root, rsv);
5289 steal_from_global = 0;
5292 trans->block_rsv = rsv;
5294 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5295 if (ret != -ENOSPC && ret != -EAGAIN)
5298 trans->block_rsv = &root->fs_info->trans_block_rsv;
5299 btrfs_end_transaction(trans, root);
5301 btrfs_btree_balance_dirty(root);
5304 btrfs_free_block_rsv(root, rsv);
5307 * Errors here aren't a big deal, it just means we leave orphan items
5308 * in the tree. They will be cleaned up on the next mount.
5311 trans->block_rsv = root->orphan_block_rsv;
5312 btrfs_orphan_del(trans, inode);
5314 btrfs_orphan_del(NULL, inode);
5317 trans->block_rsv = &root->fs_info->trans_block_rsv;
5318 if (!(root == root->fs_info->tree_root ||
5319 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5320 btrfs_return_ino(root, btrfs_ino(inode));
5322 btrfs_end_transaction(trans, root);
5323 btrfs_btree_balance_dirty(root);
5325 btrfs_remove_delayed_node(inode);
5331 * this returns the key found in the dir entry in the location pointer.
5332 * If no dir entries were found, location->objectid is 0.
5334 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5335 struct btrfs_key *location)
5337 const char *name = dentry->d_name.name;
5338 int namelen = dentry->d_name.len;
5339 struct btrfs_dir_item *di;
5340 struct btrfs_path *path;
5341 struct btrfs_root *root = BTRFS_I(dir)->root;
5344 path = btrfs_alloc_path();
5348 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5353 if (IS_ERR_OR_NULL(di))
5356 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5358 btrfs_free_path(path);
5361 location->objectid = 0;
5366 * when we hit a tree root in a directory, the btrfs part of the inode
5367 * needs to be changed to reflect the root directory of the tree root. This
5368 * is kind of like crossing a mount point.
5370 static int fixup_tree_root_location(struct btrfs_root *root,
5372 struct dentry *dentry,
5373 struct btrfs_key *location,
5374 struct btrfs_root **sub_root)
5376 struct btrfs_path *path;
5377 struct btrfs_root *new_root;
5378 struct btrfs_root_ref *ref;
5379 struct extent_buffer *leaf;
5380 struct btrfs_key key;
5384 path = btrfs_alloc_path();
5391 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5392 key.type = BTRFS_ROOT_REF_KEY;
5393 key.offset = location->objectid;
5395 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5403 leaf = path->nodes[0];
5404 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5405 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5406 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5409 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5410 (unsigned long)(ref + 1),
5411 dentry->d_name.len);
5415 btrfs_release_path(path);
5417 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5418 if (IS_ERR(new_root)) {
5419 err = PTR_ERR(new_root);
5423 *sub_root = new_root;
5424 location->objectid = btrfs_root_dirid(&new_root->root_item);
5425 location->type = BTRFS_INODE_ITEM_KEY;
5426 location->offset = 0;
5429 btrfs_free_path(path);
5433 static void inode_tree_add(struct inode *inode)
5435 struct btrfs_root *root = BTRFS_I(inode)->root;
5436 struct btrfs_inode *entry;
5438 struct rb_node *parent;
5439 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5440 u64 ino = btrfs_ino(inode);
5442 if (inode_unhashed(inode))
5445 spin_lock(&root->inode_lock);
5446 p = &root->inode_tree.rb_node;
5449 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5451 if (ino < btrfs_ino(&entry->vfs_inode))
5452 p = &parent->rb_left;
5453 else if (ino > btrfs_ino(&entry->vfs_inode))
5454 p = &parent->rb_right;
5456 WARN_ON(!(entry->vfs_inode.i_state &
5457 (I_WILL_FREE | I_FREEING)));
5458 rb_replace_node(parent, new, &root->inode_tree);
5459 RB_CLEAR_NODE(parent);
5460 spin_unlock(&root->inode_lock);
5464 rb_link_node(new, parent, p);
5465 rb_insert_color(new, &root->inode_tree);
5466 spin_unlock(&root->inode_lock);
5469 static void inode_tree_del(struct inode *inode)
5471 struct btrfs_root *root = BTRFS_I(inode)->root;
5474 spin_lock(&root->inode_lock);
5475 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5476 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5477 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5478 empty = RB_EMPTY_ROOT(&root->inode_tree);
5480 spin_unlock(&root->inode_lock);
5482 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5483 synchronize_srcu(&root->fs_info->subvol_srcu);
5484 spin_lock(&root->inode_lock);
5485 empty = RB_EMPTY_ROOT(&root->inode_tree);
5486 spin_unlock(&root->inode_lock);
5488 btrfs_add_dead_root(root);
5492 void btrfs_invalidate_inodes(struct btrfs_root *root)
5494 struct rb_node *node;
5495 struct rb_node *prev;
5496 struct btrfs_inode *entry;
5497 struct inode *inode;
5500 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5501 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5503 spin_lock(&root->inode_lock);
5505 node = root->inode_tree.rb_node;
5509 entry = rb_entry(node, struct btrfs_inode, rb_node);
5511 if (objectid < btrfs_ino(&entry->vfs_inode))
5512 node = node->rb_left;
5513 else if (objectid > btrfs_ino(&entry->vfs_inode))
5514 node = node->rb_right;
5520 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5521 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5525 prev = rb_next(prev);
5529 entry = rb_entry(node, struct btrfs_inode, rb_node);
5530 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5531 inode = igrab(&entry->vfs_inode);
5533 spin_unlock(&root->inode_lock);
5534 if (atomic_read(&inode->i_count) > 1)
5535 d_prune_aliases(inode);
5537 * btrfs_drop_inode will have it removed from
5538 * the inode cache when its usage count
5543 spin_lock(&root->inode_lock);
5547 if (cond_resched_lock(&root->inode_lock))
5550 node = rb_next(node);
5552 spin_unlock(&root->inode_lock);
5555 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5557 struct btrfs_iget_args *args = p;
5558 inode->i_ino = args->location->objectid;
5559 memcpy(&BTRFS_I(inode)->location, args->location,
5560 sizeof(*args->location));
5561 BTRFS_I(inode)->root = args->root;
5565 static int btrfs_find_actor(struct inode *inode, void *opaque)
5567 struct btrfs_iget_args *args = opaque;
5568 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5569 args->root == BTRFS_I(inode)->root;
5572 static struct inode *btrfs_iget_locked(struct super_block *s,
5573 struct btrfs_key *location,
5574 struct btrfs_root *root)
5576 struct inode *inode;
5577 struct btrfs_iget_args args;
5578 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5580 args.location = location;
5583 inode = iget5_locked(s, hashval, btrfs_find_actor,
5584 btrfs_init_locked_inode,
5589 /* Get an inode object given its location and corresponding root.
5590 * Returns in *is_new if the inode was read from disk
5592 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5593 struct btrfs_root *root, int *new)
5595 struct inode *inode;
5597 inode = btrfs_iget_locked(s, location, root);
5599 return ERR_PTR(-ENOMEM);
5601 if (inode->i_state & I_NEW) {
5602 btrfs_read_locked_inode(inode);
5603 if (!is_bad_inode(inode)) {
5604 inode_tree_add(inode);
5605 unlock_new_inode(inode);
5609 unlock_new_inode(inode);
5611 inode = ERR_PTR(-ESTALE);
5618 static struct inode *new_simple_dir(struct super_block *s,
5619 struct btrfs_key *key,
5620 struct btrfs_root *root)
5622 struct inode *inode = new_inode(s);
5625 return ERR_PTR(-ENOMEM);
5627 BTRFS_I(inode)->root = root;
5628 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5629 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5631 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5632 inode->i_op = &btrfs_dir_ro_inode_operations;
5633 inode->i_fop = &simple_dir_operations;
5634 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5635 inode->i_mtime = CURRENT_TIME;
5636 inode->i_atime = inode->i_mtime;
5637 inode->i_ctime = inode->i_mtime;
5638 BTRFS_I(inode)->i_otime = inode->i_mtime;
5643 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5645 struct inode *inode;
5646 struct btrfs_root *root = BTRFS_I(dir)->root;
5647 struct btrfs_root *sub_root = root;
5648 struct btrfs_key location;
5652 if (dentry->d_name.len > BTRFS_NAME_LEN)
5653 return ERR_PTR(-ENAMETOOLONG);
5655 ret = btrfs_inode_by_name(dir, dentry, &location);
5657 return ERR_PTR(ret);
5659 if (location.objectid == 0)
5660 return ERR_PTR(-ENOENT);
5662 if (location.type == BTRFS_INODE_ITEM_KEY) {
5663 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5667 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5669 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5670 ret = fixup_tree_root_location(root, dir, dentry,
5671 &location, &sub_root);
5674 inode = ERR_PTR(ret);
5676 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5678 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5680 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5682 if (!IS_ERR(inode) && root != sub_root) {
5683 down_read(&root->fs_info->cleanup_work_sem);
5684 if (!(inode->i_sb->s_flags & MS_RDONLY))
5685 ret = btrfs_orphan_cleanup(sub_root);
5686 up_read(&root->fs_info->cleanup_work_sem);
5689 inode = ERR_PTR(ret);
5696 static int btrfs_dentry_delete(const struct dentry *dentry)
5698 struct btrfs_root *root;
5699 struct inode *inode = d_inode(dentry);
5701 if (!inode && !IS_ROOT(dentry))
5702 inode = d_inode(dentry->d_parent);
5705 root = BTRFS_I(inode)->root;
5706 if (btrfs_root_refs(&root->root_item) == 0)
5709 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5715 static void btrfs_dentry_release(struct dentry *dentry)
5717 kfree(dentry->d_fsdata);
5720 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5723 struct inode *inode;
5725 inode = btrfs_lookup_dentry(dir, dentry);
5726 if (IS_ERR(inode)) {
5727 if (PTR_ERR(inode) == -ENOENT)
5730 return ERR_CAST(inode);
5733 return d_splice_alias(inode, dentry);
5736 unsigned char btrfs_filetype_table[] = {
5737 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5740 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5742 struct inode *inode = file_inode(file);
5743 struct btrfs_root *root = BTRFS_I(inode)->root;
5744 struct btrfs_item *item;
5745 struct btrfs_dir_item *di;
5746 struct btrfs_key key;
5747 struct btrfs_key found_key;
5748 struct btrfs_path *path;
5749 struct list_head ins_list;
5750 struct list_head del_list;
5752 struct extent_buffer *leaf;
5754 unsigned char d_type;
5759 int key_type = BTRFS_DIR_INDEX_KEY;
5763 int is_curr = 0; /* ctx->pos points to the current index? */
5765 /* FIXME, use a real flag for deciding about the key type */
5766 if (root->fs_info->tree_root == root)
5767 key_type = BTRFS_DIR_ITEM_KEY;
5769 if (!dir_emit_dots(file, ctx))
5772 path = btrfs_alloc_path();
5778 if (key_type == BTRFS_DIR_INDEX_KEY) {
5779 INIT_LIST_HEAD(&ins_list);
5780 INIT_LIST_HEAD(&del_list);
5781 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5784 key.type = key_type;
5785 key.offset = ctx->pos;
5786 key.objectid = btrfs_ino(inode);
5788 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5793 leaf = path->nodes[0];
5794 slot = path->slots[0];
5795 if (slot >= btrfs_header_nritems(leaf)) {
5796 ret = btrfs_next_leaf(root, path);
5804 item = btrfs_item_nr(slot);
5805 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5807 if (found_key.objectid != key.objectid)
5809 if (found_key.type != key_type)
5811 if (found_key.offset < ctx->pos)
5813 if (key_type == BTRFS_DIR_INDEX_KEY &&
5814 btrfs_should_delete_dir_index(&del_list,
5818 ctx->pos = found_key.offset;
5821 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5823 di_total = btrfs_item_size(leaf, item);
5825 while (di_cur < di_total) {
5826 struct btrfs_key location;
5828 if (verify_dir_item(root, leaf, di))
5831 name_len = btrfs_dir_name_len(leaf, di);
5832 if (name_len <= sizeof(tmp_name)) {
5833 name_ptr = tmp_name;
5835 name_ptr = kmalloc(name_len, GFP_NOFS);
5841 read_extent_buffer(leaf, name_ptr,
5842 (unsigned long)(di + 1), name_len);
5844 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5845 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5848 /* is this a reference to our own snapshot? If so
5851 * In contrast to old kernels, we insert the snapshot's
5852 * dir item and dir index after it has been created, so
5853 * we won't find a reference to our own snapshot. We
5854 * still keep the following code for backward
5857 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5858 location.objectid == root->root_key.objectid) {
5862 over = !dir_emit(ctx, name_ptr, name_len,
5863 location.objectid, d_type);
5866 if (name_ptr != tmp_name)
5871 di_len = btrfs_dir_name_len(leaf, di) +
5872 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5874 di = (struct btrfs_dir_item *)((char *)di + di_len);
5880 if (key_type == BTRFS_DIR_INDEX_KEY) {
5883 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5888 /* Reached end of directory/root. Bump pos past the last item. */
5892 * Stop new entries from being returned after we return the last
5895 * New directory entries are assigned a strictly increasing
5896 * offset. This means that new entries created during readdir
5897 * are *guaranteed* to be seen in the future by that readdir.
5898 * This has broken buggy programs which operate on names as
5899 * they're returned by readdir. Until we re-use freed offsets
5900 * we have this hack to stop new entries from being returned
5901 * under the assumption that they'll never reach this huge
5904 * This is being careful not to overflow 32bit loff_t unless the
5905 * last entry requires it because doing so has broken 32bit apps
5908 if (key_type == BTRFS_DIR_INDEX_KEY) {
5909 if (ctx->pos >= INT_MAX)
5910 ctx->pos = LLONG_MAX;
5917 if (key_type == BTRFS_DIR_INDEX_KEY)
5918 btrfs_put_delayed_items(&ins_list, &del_list);
5919 btrfs_free_path(path);
5923 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5925 struct btrfs_root *root = BTRFS_I(inode)->root;
5926 struct btrfs_trans_handle *trans;
5928 bool nolock = false;
5930 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5933 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5936 if (wbc->sync_mode == WB_SYNC_ALL) {
5938 trans = btrfs_join_transaction_nolock(root);
5940 trans = btrfs_join_transaction(root);
5942 return PTR_ERR(trans);
5943 ret = btrfs_commit_transaction(trans, root);
5949 * This is somewhat expensive, updating the tree every time the
5950 * inode changes. But, it is most likely to find the inode in cache.
5951 * FIXME, needs more benchmarking...there are no reasons other than performance
5952 * to keep or drop this code.
5954 static int btrfs_dirty_inode(struct inode *inode)
5956 struct btrfs_root *root = BTRFS_I(inode)->root;
5957 struct btrfs_trans_handle *trans;
5960 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5963 trans = btrfs_join_transaction(root);
5965 return PTR_ERR(trans);
5967 ret = btrfs_update_inode(trans, root, inode);
5968 if (ret && ret == -ENOSPC) {
5969 /* whoops, lets try again with the full transaction */
5970 btrfs_end_transaction(trans, root);
5971 trans = btrfs_start_transaction(root, 1);
5973 return PTR_ERR(trans);
5975 ret = btrfs_update_inode(trans, root, inode);
5977 btrfs_end_transaction(trans, root);
5978 if (BTRFS_I(inode)->delayed_node)
5979 btrfs_balance_delayed_items(root);
5985 * This is a copy of file_update_time. We need this so we can return error on
5986 * ENOSPC for updating the inode in the case of file write and mmap writes.
5988 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5991 struct btrfs_root *root = BTRFS_I(inode)->root;
5993 if (btrfs_root_readonly(root))
5996 if (flags & S_VERSION)
5997 inode_inc_iversion(inode);
5998 if (flags & S_CTIME)
5999 inode->i_ctime = *now;
6000 if (flags & S_MTIME)
6001 inode->i_mtime = *now;
6002 if (flags & S_ATIME)
6003 inode->i_atime = *now;
6004 return btrfs_dirty_inode(inode);
6008 * find the highest existing sequence number in a directory
6009 * and then set the in-memory index_cnt variable to reflect
6010 * free sequence numbers
6012 static int btrfs_set_inode_index_count(struct inode *inode)
6014 struct btrfs_root *root = BTRFS_I(inode)->root;
6015 struct btrfs_key key, found_key;
6016 struct btrfs_path *path;
6017 struct extent_buffer *leaf;
6020 key.objectid = btrfs_ino(inode);
6021 key.type = BTRFS_DIR_INDEX_KEY;
6022 key.offset = (u64)-1;
6024 path = btrfs_alloc_path();
6028 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6031 /* FIXME: we should be able to handle this */
6037 * MAGIC NUMBER EXPLANATION:
6038 * since we search a directory based on f_pos we have to start at 2
6039 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6040 * else has to start at 2
6042 if (path->slots[0] == 0) {
6043 BTRFS_I(inode)->index_cnt = 2;
6049 leaf = path->nodes[0];
6050 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6052 if (found_key.objectid != btrfs_ino(inode) ||
6053 found_key.type != BTRFS_DIR_INDEX_KEY) {
6054 BTRFS_I(inode)->index_cnt = 2;
6058 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6060 btrfs_free_path(path);
6065 * helper to find a free sequence number in a given directory. This current
6066 * code is very simple, later versions will do smarter things in the btree
6068 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6072 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6073 ret = btrfs_inode_delayed_dir_index_count(dir);
6075 ret = btrfs_set_inode_index_count(dir);
6081 *index = BTRFS_I(dir)->index_cnt;
6082 BTRFS_I(dir)->index_cnt++;
6087 static int btrfs_insert_inode_locked(struct inode *inode)
6089 struct btrfs_iget_args args;
6090 args.location = &BTRFS_I(inode)->location;
6091 args.root = BTRFS_I(inode)->root;
6093 return insert_inode_locked4(inode,
6094 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6095 btrfs_find_actor, &args);
6098 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6099 struct btrfs_root *root,
6101 const char *name, int name_len,
6102 u64 ref_objectid, u64 objectid,
6103 umode_t mode, u64 *index)
6105 struct inode *inode;
6106 struct btrfs_inode_item *inode_item;
6107 struct btrfs_key *location;
6108 struct btrfs_path *path;
6109 struct btrfs_inode_ref *ref;
6110 struct btrfs_key key[2];
6112 int nitems = name ? 2 : 1;
6116 path = btrfs_alloc_path();
6118 return ERR_PTR(-ENOMEM);
6120 inode = new_inode(root->fs_info->sb);
6122 btrfs_free_path(path);
6123 return ERR_PTR(-ENOMEM);
6127 * O_TMPFILE, set link count to 0, so that after this point,
6128 * we fill in an inode item with the correct link count.
6131 set_nlink(inode, 0);
6134 * we have to initialize this early, so we can reclaim the inode
6135 * number if we fail afterwards in this function.
6137 inode->i_ino = objectid;
6140 trace_btrfs_inode_request(dir);
6142 ret = btrfs_set_inode_index(dir, index);
6144 btrfs_free_path(path);
6146 return ERR_PTR(ret);
6152 * index_cnt is ignored for everything but a dir,
6153 * btrfs_get_inode_index_count has an explanation for the magic
6156 BTRFS_I(inode)->index_cnt = 2;
6157 BTRFS_I(inode)->dir_index = *index;
6158 BTRFS_I(inode)->root = root;
6159 BTRFS_I(inode)->generation = trans->transid;
6160 inode->i_generation = BTRFS_I(inode)->generation;
6163 * We could have gotten an inode number from somebody who was fsynced
6164 * and then removed in this same transaction, so let's just set full
6165 * sync since it will be a full sync anyway and this will blow away the
6166 * old info in the log.
6168 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6170 key[0].objectid = objectid;
6171 key[0].type = BTRFS_INODE_ITEM_KEY;
6174 sizes[0] = sizeof(struct btrfs_inode_item);
6178 * Start new inodes with an inode_ref. This is slightly more
6179 * efficient for small numbers of hard links since they will
6180 * be packed into one item. Extended refs will kick in if we
6181 * add more hard links than can fit in the ref item.
6183 key[1].objectid = objectid;
6184 key[1].type = BTRFS_INODE_REF_KEY;
6185 key[1].offset = ref_objectid;
6187 sizes[1] = name_len + sizeof(*ref);
6190 location = &BTRFS_I(inode)->location;
6191 location->objectid = objectid;
6192 location->offset = 0;
6193 location->type = BTRFS_INODE_ITEM_KEY;
6195 ret = btrfs_insert_inode_locked(inode);
6199 path->leave_spinning = 1;
6200 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6204 inode_init_owner(inode, dir, mode);
6205 inode_set_bytes(inode, 0);
6207 inode->i_mtime = CURRENT_TIME;
6208 inode->i_atime = inode->i_mtime;
6209 inode->i_ctime = inode->i_mtime;
6210 BTRFS_I(inode)->i_otime = inode->i_mtime;
6212 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6213 struct btrfs_inode_item);
6214 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6215 sizeof(*inode_item));
6216 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6219 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6220 struct btrfs_inode_ref);
6221 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6222 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6223 ptr = (unsigned long)(ref + 1);
6224 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6227 btrfs_mark_buffer_dirty(path->nodes[0]);
6228 btrfs_free_path(path);
6230 btrfs_inherit_iflags(inode, dir);
6232 if (S_ISREG(mode)) {
6233 if (btrfs_test_opt(root, NODATASUM))
6234 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6235 if (btrfs_test_opt(root, NODATACOW))
6236 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6237 BTRFS_INODE_NODATASUM;
6240 inode_tree_add(inode);
6242 trace_btrfs_inode_new(inode);
6243 btrfs_set_inode_last_trans(trans, inode);
6245 btrfs_update_root_times(trans, root);
6247 ret = btrfs_inode_inherit_props(trans, inode, dir);
6249 btrfs_err(root->fs_info,
6250 "error inheriting props for ino %llu (root %llu): %d",
6251 btrfs_ino(inode), root->root_key.objectid, ret);
6256 unlock_new_inode(inode);
6259 BTRFS_I(dir)->index_cnt--;
6260 btrfs_free_path(path);
6262 return ERR_PTR(ret);
6265 static inline u8 btrfs_inode_type(struct inode *inode)
6267 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6271 * utility function to add 'inode' into 'parent_inode' with
6272 * a give name and a given sequence number.
6273 * if 'add_backref' is true, also insert a backref from the
6274 * inode to the parent directory.
6276 int btrfs_add_link(struct btrfs_trans_handle *trans,
6277 struct inode *parent_inode, struct inode *inode,
6278 const char *name, int name_len, int add_backref, u64 index)
6281 struct btrfs_key key;
6282 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6283 u64 ino = btrfs_ino(inode);
6284 u64 parent_ino = btrfs_ino(parent_inode);
6286 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6287 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6290 key.type = BTRFS_INODE_ITEM_KEY;
6294 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6295 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6296 key.objectid, root->root_key.objectid,
6297 parent_ino, index, name, name_len);
6298 } else if (add_backref) {
6299 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6303 /* Nothing to clean up yet */
6307 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6309 btrfs_inode_type(inode), index);
6310 if (ret == -EEXIST || ret == -EOVERFLOW)
6313 btrfs_abort_transaction(trans, root, ret);
6317 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6319 inode_inc_iversion(parent_inode);
6320 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6321 ret = btrfs_update_inode(trans, root, parent_inode);
6323 btrfs_abort_transaction(trans, root, ret);
6327 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6330 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6331 key.objectid, root->root_key.objectid,
6332 parent_ino, &local_index, name, name_len);
6334 } else if (add_backref) {
6338 err = btrfs_del_inode_ref(trans, root, name, name_len,
6339 ino, parent_ino, &local_index);
6344 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6345 struct inode *dir, struct dentry *dentry,
6346 struct inode *inode, int backref, u64 index)
6348 int err = btrfs_add_link(trans, dir, inode,
6349 dentry->d_name.name, dentry->d_name.len,
6356 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6357 umode_t mode, dev_t rdev)
6359 struct btrfs_trans_handle *trans;
6360 struct btrfs_root *root = BTRFS_I(dir)->root;
6361 struct inode *inode = NULL;
6367 if (!new_valid_dev(rdev))
6371 * 2 for inode item and ref
6373 * 1 for xattr if selinux is on
6375 trans = btrfs_start_transaction(root, 5);
6377 return PTR_ERR(trans);
6379 err = btrfs_find_free_ino(root, &objectid);
6383 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6384 dentry->d_name.len, btrfs_ino(dir), objectid,
6386 if (IS_ERR(inode)) {
6387 err = PTR_ERR(inode);
6392 * If the active LSM wants to access the inode during
6393 * d_instantiate it needs these. Smack checks to see
6394 * if the filesystem supports xattrs by looking at the
6397 inode->i_op = &btrfs_special_inode_operations;
6398 init_special_inode(inode, inode->i_mode, rdev);
6400 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6402 goto out_unlock_inode;
6404 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6406 goto out_unlock_inode;
6408 btrfs_update_inode(trans, root, inode);
6409 unlock_new_inode(inode);
6410 d_instantiate(dentry, inode);
6414 btrfs_end_transaction(trans, root);
6415 btrfs_balance_delayed_items(root);
6416 btrfs_btree_balance_dirty(root);
6418 inode_dec_link_count(inode);
6425 unlock_new_inode(inode);
6430 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6431 umode_t mode, bool excl)
6433 struct btrfs_trans_handle *trans;
6434 struct btrfs_root *root = BTRFS_I(dir)->root;
6435 struct inode *inode = NULL;
6436 int drop_inode_on_err = 0;
6442 * 2 for inode item and ref
6444 * 1 for xattr if selinux is on
6446 trans = btrfs_start_transaction(root, 5);
6448 return PTR_ERR(trans);
6450 err = btrfs_find_free_ino(root, &objectid);
6454 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6455 dentry->d_name.len, btrfs_ino(dir), objectid,
6457 if (IS_ERR(inode)) {
6458 err = PTR_ERR(inode);
6461 drop_inode_on_err = 1;
6463 * If the active LSM wants to access the inode during
6464 * d_instantiate it needs these. Smack checks to see
6465 * if the filesystem supports xattrs by looking at the
6468 inode->i_fop = &btrfs_file_operations;
6469 inode->i_op = &btrfs_file_inode_operations;
6470 inode->i_mapping->a_ops = &btrfs_aops;
6472 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6474 goto out_unlock_inode;
6476 err = btrfs_update_inode(trans, root, inode);
6478 goto out_unlock_inode;
6480 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6482 goto out_unlock_inode;
6484 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6485 unlock_new_inode(inode);
6486 d_instantiate(dentry, inode);
6489 btrfs_end_transaction(trans, root);
6490 if (err && drop_inode_on_err) {
6491 inode_dec_link_count(inode);
6494 btrfs_balance_delayed_items(root);
6495 btrfs_btree_balance_dirty(root);
6499 unlock_new_inode(inode);
6504 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6505 struct dentry *dentry)
6507 struct btrfs_trans_handle *trans;
6508 struct btrfs_root *root = BTRFS_I(dir)->root;
6509 struct inode *inode = d_inode(old_dentry);
6514 /* do not allow sys_link's with other subvols of the same device */
6515 if (root->objectid != BTRFS_I(inode)->root->objectid)
6518 if (inode->i_nlink >= BTRFS_LINK_MAX)
6521 err = btrfs_set_inode_index(dir, &index);
6526 * 2 items for inode and inode ref
6527 * 2 items for dir items
6528 * 1 item for parent inode
6530 trans = btrfs_start_transaction(root, 5);
6531 if (IS_ERR(trans)) {
6532 err = PTR_ERR(trans);
6536 /* There are several dir indexes for this inode, clear the cache. */
6537 BTRFS_I(inode)->dir_index = 0ULL;
6539 inode_inc_iversion(inode);
6540 inode->i_ctime = CURRENT_TIME;
6542 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6544 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6549 struct dentry *parent = dentry->d_parent;
6550 err = btrfs_update_inode(trans, root, inode);
6553 if (inode->i_nlink == 1) {
6555 * If new hard link count is 1, it's a file created
6556 * with open(2) O_TMPFILE flag.
6558 err = btrfs_orphan_del(trans, inode);
6562 d_instantiate(dentry, inode);
6563 btrfs_log_new_name(trans, inode, NULL, parent);
6566 btrfs_end_transaction(trans, root);
6567 btrfs_balance_delayed_items(root);
6570 inode_dec_link_count(inode);
6573 btrfs_btree_balance_dirty(root);
6577 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6579 struct inode *inode = NULL;
6580 struct btrfs_trans_handle *trans;
6581 struct btrfs_root *root = BTRFS_I(dir)->root;
6583 int drop_on_err = 0;
6588 * 2 items for inode and ref
6589 * 2 items for dir items
6590 * 1 for xattr if selinux is on
6592 trans = btrfs_start_transaction(root, 5);
6594 return PTR_ERR(trans);
6596 err = btrfs_find_free_ino(root, &objectid);
6600 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6601 dentry->d_name.len, btrfs_ino(dir), objectid,
6602 S_IFDIR | mode, &index);
6603 if (IS_ERR(inode)) {
6604 err = PTR_ERR(inode);
6609 /* these must be set before we unlock the inode */
6610 inode->i_op = &btrfs_dir_inode_operations;
6611 inode->i_fop = &btrfs_dir_file_operations;
6613 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6615 goto out_fail_inode;
6617 btrfs_i_size_write(inode, 0);
6618 err = btrfs_update_inode(trans, root, inode);
6620 goto out_fail_inode;
6622 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6623 dentry->d_name.len, 0, index);
6625 goto out_fail_inode;
6627 d_instantiate(dentry, inode);
6629 * mkdir is special. We're unlocking after we call d_instantiate
6630 * to avoid a race with nfsd calling d_instantiate.
6632 unlock_new_inode(inode);
6636 btrfs_end_transaction(trans, root);
6638 inode_dec_link_count(inode);
6641 btrfs_balance_delayed_items(root);
6642 btrfs_btree_balance_dirty(root);
6646 unlock_new_inode(inode);
6650 /* Find next extent map of a given extent map, caller needs to ensure locks */
6651 static struct extent_map *next_extent_map(struct extent_map *em)
6653 struct rb_node *next;
6655 next = rb_next(&em->rb_node);
6658 return container_of(next, struct extent_map, rb_node);
6661 static struct extent_map *prev_extent_map(struct extent_map *em)
6663 struct rb_node *prev;
6665 prev = rb_prev(&em->rb_node);
6668 return container_of(prev, struct extent_map, rb_node);
6671 /* helper for btfs_get_extent. Given an existing extent in the tree,
6672 * the existing extent is the nearest extent to map_start,
6673 * and an extent that you want to insert, deal with overlap and insert
6674 * the best fitted new extent into the tree.
6676 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6677 struct extent_map *existing,
6678 struct extent_map *em,
6681 struct extent_map *prev;
6682 struct extent_map *next;
6687 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6689 if (existing->start > map_start) {
6691 prev = prev_extent_map(next);
6694 next = next_extent_map(prev);
6697 start = prev ? extent_map_end(prev) : em->start;
6698 start = max_t(u64, start, em->start);
6699 end = next ? next->start : extent_map_end(em);
6700 end = min_t(u64, end, extent_map_end(em));
6701 start_diff = start - em->start;
6703 em->len = end - start;
6704 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6705 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6706 em->block_start += start_diff;
6707 em->block_len -= start_diff;
6709 return add_extent_mapping(em_tree, em, 0);
6712 static noinline int uncompress_inline(struct btrfs_path *path,
6713 struct inode *inode, struct page *page,
6714 size_t pg_offset, u64 extent_offset,
6715 struct btrfs_file_extent_item *item)
6718 struct extent_buffer *leaf = path->nodes[0];
6721 unsigned long inline_size;
6725 WARN_ON(pg_offset != 0);
6726 compress_type = btrfs_file_extent_compression(leaf, item);
6727 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6728 inline_size = btrfs_file_extent_inline_item_len(leaf,
6729 btrfs_item_nr(path->slots[0]));
6730 tmp = kmalloc(inline_size, GFP_NOFS);
6733 ptr = btrfs_file_extent_inline_start(item);
6735 read_extent_buffer(leaf, tmp, ptr, inline_size);
6737 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6738 ret = btrfs_decompress(compress_type, tmp, page,
6739 extent_offset, inline_size, max_size);
6745 * a bit scary, this does extent mapping from logical file offset to the disk.
6746 * the ugly parts come from merging extents from the disk with the in-ram
6747 * representation. This gets more complex because of the data=ordered code,
6748 * where the in-ram extents might be locked pending data=ordered completion.
6750 * This also copies inline extents directly into the page.
6753 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6754 size_t pg_offset, u64 start, u64 len,
6759 u64 extent_start = 0;
6761 u64 objectid = btrfs_ino(inode);
6763 struct btrfs_path *path = NULL;
6764 struct btrfs_root *root = BTRFS_I(inode)->root;
6765 struct btrfs_file_extent_item *item;
6766 struct extent_buffer *leaf;
6767 struct btrfs_key found_key;
6768 struct extent_map *em = NULL;
6769 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6770 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6771 struct btrfs_trans_handle *trans = NULL;
6772 const bool new_inline = !page || create;
6775 read_lock(&em_tree->lock);
6776 em = lookup_extent_mapping(em_tree, start, len);
6778 em->bdev = root->fs_info->fs_devices->latest_bdev;
6779 read_unlock(&em_tree->lock);
6782 if (em->start > start || em->start + em->len <= start)
6783 free_extent_map(em);
6784 else if (em->block_start == EXTENT_MAP_INLINE && page)
6785 free_extent_map(em);
6789 em = alloc_extent_map();
6794 em->bdev = root->fs_info->fs_devices->latest_bdev;
6795 em->start = EXTENT_MAP_HOLE;
6796 em->orig_start = EXTENT_MAP_HOLE;
6798 em->block_len = (u64)-1;
6801 path = btrfs_alloc_path();
6807 * Chances are we'll be called again, so go ahead and do
6813 ret = btrfs_lookup_file_extent(trans, root, path,
6814 objectid, start, trans != NULL);
6821 if (path->slots[0] == 0)
6826 leaf = path->nodes[0];
6827 item = btrfs_item_ptr(leaf, path->slots[0],
6828 struct btrfs_file_extent_item);
6829 /* are we inside the extent that was found? */
6830 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6831 found_type = found_key.type;
6832 if (found_key.objectid != objectid ||
6833 found_type != BTRFS_EXTENT_DATA_KEY) {
6835 * If we backup past the first extent we want to move forward
6836 * and see if there is an extent in front of us, otherwise we'll
6837 * say there is a hole for our whole search range which can
6844 found_type = btrfs_file_extent_type(leaf, item);
6845 extent_start = found_key.offset;
6846 if (found_type == BTRFS_FILE_EXTENT_REG ||
6847 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6848 extent_end = extent_start +
6849 btrfs_file_extent_num_bytes(leaf, item);
6850 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6852 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6853 extent_end = ALIGN(extent_start + size, root->sectorsize);
6856 if (start >= extent_end) {
6858 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6859 ret = btrfs_next_leaf(root, path);
6866 leaf = path->nodes[0];
6868 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6869 if (found_key.objectid != objectid ||
6870 found_key.type != BTRFS_EXTENT_DATA_KEY)
6872 if (start + len <= found_key.offset)
6874 if (start > found_key.offset)
6877 em->orig_start = start;
6878 em->len = found_key.offset - start;
6882 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6884 if (found_type == BTRFS_FILE_EXTENT_REG ||
6885 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6887 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6891 size_t extent_offset;
6897 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6898 extent_offset = page_offset(page) + pg_offset - extent_start;
6899 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6900 size - extent_offset);
6901 em->start = extent_start + extent_offset;
6902 em->len = ALIGN(copy_size, root->sectorsize);
6903 em->orig_block_len = em->len;
6904 em->orig_start = em->start;
6905 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6906 if (create == 0 && !PageUptodate(page)) {
6907 if (btrfs_file_extent_compression(leaf, item) !=
6908 BTRFS_COMPRESS_NONE) {
6909 ret = uncompress_inline(path, inode, page,
6911 extent_offset, item);
6918 read_extent_buffer(leaf, map + pg_offset, ptr,
6920 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6921 memset(map + pg_offset + copy_size, 0,
6922 PAGE_CACHE_SIZE - pg_offset -
6927 flush_dcache_page(page);
6928 } else if (create && PageUptodate(page)) {
6932 free_extent_map(em);
6935 btrfs_release_path(path);
6936 trans = btrfs_join_transaction(root);
6939 return ERR_CAST(trans);
6943 write_extent_buffer(leaf, map + pg_offset, ptr,
6946 btrfs_mark_buffer_dirty(leaf);
6948 set_extent_uptodate(io_tree, em->start,
6949 extent_map_end(em) - 1, NULL, GFP_NOFS);
6954 em->orig_start = start;
6957 em->block_start = EXTENT_MAP_HOLE;
6958 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6960 btrfs_release_path(path);
6961 if (em->start > start || extent_map_end(em) <= start) {
6962 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6963 em->start, em->len, start, len);
6969 write_lock(&em_tree->lock);
6970 ret = add_extent_mapping(em_tree, em, 0);
6971 /* it is possible that someone inserted the extent into the tree
6972 * while we had the lock dropped. It is also possible that
6973 * an overlapping map exists in the tree
6975 if (ret == -EEXIST) {
6976 struct extent_map *existing;
6980 existing = search_extent_mapping(em_tree, start, len);
6982 * existing will always be non-NULL, since there must be
6983 * extent causing the -EEXIST.
6985 if (start >= extent_map_end(existing) ||
6986 start <= existing->start) {
6988 * The existing extent map is the one nearest to
6989 * the [start, start + len) range which overlaps
6991 err = merge_extent_mapping(em_tree, existing,
6993 free_extent_map(existing);
6995 free_extent_map(em);
6999 free_extent_map(em);
7004 write_unlock(&em_tree->lock);
7007 trace_btrfs_get_extent(root, em);
7009 btrfs_free_path(path);
7011 ret = btrfs_end_transaction(trans, root);
7016 free_extent_map(em);
7017 return ERR_PTR(err);
7019 BUG_ON(!em); /* Error is always set */
7023 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7024 size_t pg_offset, u64 start, u64 len,
7027 struct extent_map *em;
7028 struct extent_map *hole_em = NULL;
7029 u64 range_start = start;
7035 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7042 * - a pre-alloc extent,
7043 * there might actually be delalloc bytes behind it.
7045 if (em->block_start != EXTENT_MAP_HOLE &&
7046 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7052 /* check to see if we've wrapped (len == -1 or similar) */
7061 /* ok, we didn't find anything, lets look for delalloc */
7062 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7063 end, len, EXTENT_DELALLOC, 1);
7064 found_end = range_start + found;
7065 if (found_end < range_start)
7066 found_end = (u64)-1;
7069 * we didn't find anything useful, return
7070 * the original results from get_extent()
7072 if (range_start > end || found_end <= start) {
7078 /* adjust the range_start to make sure it doesn't
7079 * go backwards from the start they passed in
7081 range_start = max(start, range_start);
7082 found = found_end - range_start;
7085 u64 hole_start = start;
7088 em = alloc_extent_map();
7094 * when btrfs_get_extent can't find anything it
7095 * returns one huge hole
7097 * make sure what it found really fits our range, and
7098 * adjust to make sure it is based on the start from
7102 u64 calc_end = extent_map_end(hole_em);
7104 if (calc_end <= start || (hole_em->start > end)) {
7105 free_extent_map(hole_em);
7108 hole_start = max(hole_em->start, start);
7109 hole_len = calc_end - hole_start;
7113 if (hole_em && range_start > hole_start) {
7114 /* our hole starts before our delalloc, so we
7115 * have to return just the parts of the hole
7116 * that go until the delalloc starts
7118 em->len = min(hole_len,
7119 range_start - hole_start);
7120 em->start = hole_start;
7121 em->orig_start = hole_start;
7123 * don't adjust block start at all,
7124 * it is fixed at EXTENT_MAP_HOLE
7126 em->block_start = hole_em->block_start;
7127 em->block_len = hole_len;
7128 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7129 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7131 em->start = range_start;
7133 em->orig_start = range_start;
7134 em->block_start = EXTENT_MAP_DELALLOC;
7135 em->block_len = found;
7137 } else if (hole_em) {
7142 free_extent_map(hole_em);
7144 free_extent_map(em);
7145 return ERR_PTR(err);
7150 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7153 struct btrfs_root *root = BTRFS_I(inode)->root;
7154 struct extent_map *em;
7155 struct btrfs_key ins;
7159 alloc_hint = get_extent_allocation_hint(inode, start, len);
7160 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7161 alloc_hint, &ins, 1, 1);
7163 return ERR_PTR(ret);
7165 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7166 ins.offset, ins.offset, ins.offset, 0);
7168 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7172 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7173 ins.offset, ins.offset, 0);
7175 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7176 free_extent_map(em);
7177 return ERR_PTR(ret);
7184 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7185 * block must be cow'd
7187 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7188 u64 *orig_start, u64 *orig_block_len,
7191 struct btrfs_trans_handle *trans;
7192 struct btrfs_path *path;
7194 struct extent_buffer *leaf;
7195 struct btrfs_root *root = BTRFS_I(inode)->root;
7196 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7197 struct btrfs_file_extent_item *fi;
7198 struct btrfs_key key;
7205 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7207 path = btrfs_alloc_path();
7211 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7216 slot = path->slots[0];
7219 /* can't find the item, must cow */
7226 leaf = path->nodes[0];
7227 btrfs_item_key_to_cpu(leaf, &key, slot);
7228 if (key.objectid != btrfs_ino(inode) ||
7229 key.type != BTRFS_EXTENT_DATA_KEY) {
7230 /* not our file or wrong item type, must cow */
7234 if (key.offset > offset) {
7235 /* Wrong offset, must cow */
7239 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7240 found_type = btrfs_file_extent_type(leaf, fi);
7241 if (found_type != BTRFS_FILE_EXTENT_REG &&
7242 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7243 /* not a regular extent, must cow */
7247 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7250 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7251 if (extent_end <= offset)
7254 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7255 if (disk_bytenr == 0)
7258 if (btrfs_file_extent_compression(leaf, fi) ||
7259 btrfs_file_extent_encryption(leaf, fi) ||
7260 btrfs_file_extent_other_encoding(leaf, fi))
7263 backref_offset = btrfs_file_extent_offset(leaf, fi);
7266 *orig_start = key.offset - backref_offset;
7267 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7268 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7271 if (btrfs_extent_readonly(root, disk_bytenr))
7274 num_bytes = min(offset + *len, extent_end) - offset;
7275 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7278 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7279 ret = test_range_bit(io_tree, offset, range_end,
7280 EXTENT_DELALLOC, 0, NULL);
7287 btrfs_release_path(path);
7290 * look for other files referencing this extent, if we
7291 * find any we must cow
7293 trans = btrfs_join_transaction(root);
7294 if (IS_ERR(trans)) {
7299 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7300 key.offset - backref_offset, disk_bytenr);
7301 btrfs_end_transaction(trans, root);
7308 * adjust disk_bytenr and num_bytes to cover just the bytes
7309 * in this extent we are about to write. If there
7310 * are any csums in that range we have to cow in order
7311 * to keep the csums correct
7313 disk_bytenr += backref_offset;
7314 disk_bytenr += offset - key.offset;
7315 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7318 * all of the above have passed, it is safe to overwrite this extent
7324 btrfs_free_path(path);
7328 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7330 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7332 void **pagep = NULL;
7333 struct page *page = NULL;
7337 start_idx = start >> PAGE_CACHE_SHIFT;
7340 * end is the last byte in the last page. end == start is legal
7342 end_idx = end >> PAGE_CACHE_SHIFT;
7346 /* Most of the code in this while loop is lifted from
7347 * find_get_page. It's been modified to begin searching from a
7348 * page and return just the first page found in that range. If the
7349 * found idx is less than or equal to the end idx then we know that
7350 * a page exists. If no pages are found or if those pages are
7351 * outside of the range then we're fine (yay!) */
7352 while (page == NULL &&
7353 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7354 page = radix_tree_deref_slot(pagep);
7355 if (unlikely(!page))
7358 if (radix_tree_exception(page)) {
7359 if (radix_tree_deref_retry(page)) {
7364 * Otherwise, shmem/tmpfs must be storing a swap entry
7365 * here as an exceptional entry: so return it without
7366 * attempting to raise page count.
7369 break; /* TODO: Is this relevant for this use case? */
7372 if (!page_cache_get_speculative(page)) {
7378 * Has the page moved?
7379 * This is part of the lockless pagecache protocol. See
7380 * include/linux/pagemap.h for details.
7382 if (unlikely(page != *pagep)) {
7383 page_cache_release(page);
7389 if (page->index <= end_idx)
7391 page_cache_release(page);
7398 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7399 struct extent_state **cached_state, int writing)
7401 struct btrfs_ordered_extent *ordered;
7405 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7408 * We're concerned with the entire range that we're going to be
7409 * doing DIO to, so we need to make sure theres no ordered
7410 * extents in this range.
7412 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7413 lockend - lockstart + 1);
7416 * We need to make sure there are no buffered pages in this
7417 * range either, we could have raced between the invalidate in
7418 * generic_file_direct_write and locking the extent. The
7419 * invalidate needs to happen so that reads after a write do not
7424 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7427 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7428 cached_state, GFP_NOFS);
7431 btrfs_start_ordered_extent(inode, ordered, 1);
7432 btrfs_put_ordered_extent(ordered);
7434 /* Screw you mmap */
7435 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7438 ret = filemap_fdatawait_range(inode->i_mapping,
7445 * If we found a page that couldn't be invalidated just
7446 * fall back to buffered.
7448 ret = invalidate_inode_pages2_range(inode->i_mapping,
7449 lockstart >> PAGE_CACHE_SHIFT,
7450 lockend >> PAGE_CACHE_SHIFT);
7461 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7462 u64 len, u64 orig_start,
7463 u64 block_start, u64 block_len,
7464 u64 orig_block_len, u64 ram_bytes,
7467 struct extent_map_tree *em_tree;
7468 struct extent_map *em;
7469 struct btrfs_root *root = BTRFS_I(inode)->root;
7472 em_tree = &BTRFS_I(inode)->extent_tree;
7473 em = alloc_extent_map();
7475 return ERR_PTR(-ENOMEM);
7478 em->orig_start = orig_start;
7479 em->mod_start = start;
7482 em->block_len = block_len;
7483 em->block_start = block_start;
7484 em->bdev = root->fs_info->fs_devices->latest_bdev;
7485 em->orig_block_len = orig_block_len;
7486 em->ram_bytes = ram_bytes;
7487 em->generation = -1;
7488 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7489 if (type == BTRFS_ORDERED_PREALLOC)
7490 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7493 btrfs_drop_extent_cache(inode, em->start,
7494 em->start + em->len - 1, 0);
7495 write_lock(&em_tree->lock);
7496 ret = add_extent_mapping(em_tree, em, 1);
7497 write_unlock(&em_tree->lock);
7498 } while (ret == -EEXIST);
7501 free_extent_map(em);
7502 return ERR_PTR(ret);
7508 struct btrfs_dio_data {
7509 u64 outstanding_extents;
7513 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7514 struct buffer_head *bh_result, int create)
7516 struct extent_map *em;
7517 struct btrfs_root *root = BTRFS_I(inode)->root;
7518 struct extent_state *cached_state = NULL;
7519 struct btrfs_dio_data *dio_data = NULL;
7520 u64 start = iblock << inode->i_blkbits;
7521 u64 lockstart, lockend;
7522 u64 len = bh_result->b_size;
7523 int unlock_bits = EXTENT_LOCKED;
7527 unlock_bits |= EXTENT_DIRTY;
7529 len = min_t(u64, len, root->sectorsize);
7532 lockend = start + len - 1;
7534 if (current->journal_info) {
7536 * Need to pull our outstanding extents and set journal_info to NULL so
7537 * that anything that needs to check if there's a transction doesn't get
7540 dio_data = current->journal_info;
7541 current->journal_info = NULL;
7545 * If this errors out it's because we couldn't invalidate pagecache for
7546 * this range and we need to fallback to buffered.
7548 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7551 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7558 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7559 * io. INLINE is special, and we could probably kludge it in here, but
7560 * it's still buffered so for safety lets just fall back to the generic
7563 * For COMPRESSED we _have_ to read the entire extent in so we can
7564 * decompress it, so there will be buffering required no matter what we
7565 * do, so go ahead and fallback to buffered.
7567 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7568 * to buffered IO. Don't blame me, this is the price we pay for using
7571 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7572 em->block_start == EXTENT_MAP_INLINE) {
7573 free_extent_map(em);
7578 /* Just a good old fashioned hole, return */
7579 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7580 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7581 free_extent_map(em);
7586 * We don't allocate a new extent in the following cases
7588 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7590 * 2) The extent is marked as PREALLOC. We're good to go here and can
7591 * just use the extent.
7595 len = min(len, em->len - (start - em->start));
7596 lockstart = start + len;
7600 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7601 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7602 em->block_start != EXTENT_MAP_HOLE)) {
7604 u64 block_start, orig_start, orig_block_len, ram_bytes;
7606 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7607 type = BTRFS_ORDERED_PREALLOC;
7609 type = BTRFS_ORDERED_NOCOW;
7610 len = min(len, em->len - (start - em->start));
7611 block_start = em->block_start + (start - em->start);
7613 if (can_nocow_extent(inode, start, &len, &orig_start,
7614 &orig_block_len, &ram_bytes) == 1) {
7615 if (type == BTRFS_ORDERED_PREALLOC) {
7616 free_extent_map(em);
7617 em = create_pinned_em(inode, start, len,
7628 ret = btrfs_add_ordered_extent_dio(inode, start,
7629 block_start, len, len, type);
7631 free_extent_map(em);
7639 * this will cow the extent, reset the len in case we changed
7642 len = bh_result->b_size;
7643 free_extent_map(em);
7644 em = btrfs_new_extent_direct(inode, start, len);
7649 len = min(len, em->len - (start - em->start));
7651 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7653 bh_result->b_size = len;
7654 bh_result->b_bdev = em->bdev;
7655 set_buffer_mapped(bh_result);
7657 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7658 set_buffer_new(bh_result);
7661 * Need to update the i_size under the extent lock so buffered
7662 * readers will get the updated i_size when we unlock.
7664 if (start + len > i_size_read(inode))
7665 i_size_write(inode, start + len);
7668 * If we have an outstanding_extents count still set then we're
7669 * within our reservation, otherwise we need to adjust our inode
7670 * counter appropriately.
7672 if (dio_data->outstanding_extents) {
7673 (dio_data->outstanding_extents)--;
7675 spin_lock(&BTRFS_I(inode)->lock);
7676 BTRFS_I(inode)->outstanding_extents++;
7677 spin_unlock(&BTRFS_I(inode)->lock);
7680 btrfs_free_reserved_data_space(inode, start, len);
7681 WARN_ON(dio_data->reserve < len);
7682 dio_data->reserve -= len;
7683 current->journal_info = dio_data;
7687 * In the case of write we need to clear and unlock the entire range,
7688 * in the case of read we need to unlock only the end area that we
7689 * aren't using if there is any left over space.
7691 if (lockstart < lockend) {
7692 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7693 lockend, unlock_bits, 1, 0,
7694 &cached_state, GFP_NOFS);
7696 free_extent_state(cached_state);
7699 free_extent_map(em);
7704 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7705 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7707 current->journal_info = dio_data;
7711 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7712 int rw, int mirror_num)
7714 struct btrfs_root *root = BTRFS_I(inode)->root;
7717 BUG_ON(rw & REQ_WRITE);
7721 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7722 BTRFS_WQ_ENDIO_DIO_REPAIR);
7726 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7732 static int btrfs_check_dio_repairable(struct inode *inode,
7733 struct bio *failed_bio,
7734 struct io_failure_record *failrec,
7739 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7740 failrec->logical, failrec->len);
7741 if (num_copies == 1) {
7743 * we only have a single copy of the data, so don't bother with
7744 * all the retry and error correction code that follows. no
7745 * matter what the error is, it is very likely to persist.
7747 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7748 num_copies, failrec->this_mirror, failed_mirror);
7752 failrec->failed_mirror = failed_mirror;
7753 failrec->this_mirror++;
7754 if (failrec->this_mirror == failed_mirror)
7755 failrec->this_mirror++;
7757 if (failrec->this_mirror > num_copies) {
7758 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7759 num_copies, failrec->this_mirror, failed_mirror);
7766 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7767 struct page *page, u64 start, u64 end,
7768 int failed_mirror, bio_end_io_t *repair_endio,
7771 struct io_failure_record *failrec;
7777 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7779 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7783 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7786 free_io_failure(inode, failrec);
7790 if (failed_bio->bi_vcnt > 1)
7791 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7793 read_mode = READ_SYNC;
7795 isector = start - btrfs_io_bio(failed_bio)->logical;
7796 isector >>= inode->i_sb->s_blocksize_bits;
7797 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7798 0, isector, repair_endio, repair_arg);
7800 free_io_failure(inode, failrec);
7804 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7805 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7806 read_mode, failrec->this_mirror, failrec->in_validation);
7808 ret = submit_dio_repair_bio(inode, bio, read_mode,
7809 failrec->this_mirror);
7811 free_io_failure(inode, failrec);
7818 struct btrfs_retry_complete {
7819 struct completion done;
7820 struct inode *inode;
7825 static void btrfs_retry_endio_nocsum(struct bio *bio)
7827 struct btrfs_retry_complete *done = bio->bi_private;
7828 struct bio_vec *bvec;
7835 bio_for_each_segment_all(bvec, bio, i)
7836 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7838 complete(&done->done);
7842 static int __btrfs_correct_data_nocsum(struct inode *inode,
7843 struct btrfs_io_bio *io_bio)
7845 struct bio_vec *bvec;
7846 struct btrfs_retry_complete done;
7851 start = io_bio->logical;
7854 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7858 init_completion(&done.done);
7860 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7861 start + bvec->bv_len - 1,
7863 btrfs_retry_endio_nocsum, &done);
7867 wait_for_completion(&done.done);
7869 if (!done.uptodate) {
7870 /* We might have another mirror, so try again */
7874 start += bvec->bv_len;
7880 static void btrfs_retry_endio(struct bio *bio)
7882 struct btrfs_retry_complete *done = bio->bi_private;
7883 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7884 struct bio_vec *bvec;
7893 bio_for_each_segment_all(bvec, bio, i) {
7894 ret = __readpage_endio_check(done->inode, io_bio, i,
7896 done->start, bvec->bv_len);
7898 clean_io_failure(done->inode, done->start,
7904 done->uptodate = uptodate;
7906 complete(&done->done);
7910 static int __btrfs_subio_endio_read(struct inode *inode,
7911 struct btrfs_io_bio *io_bio, int err)
7913 struct bio_vec *bvec;
7914 struct btrfs_retry_complete done;
7921 start = io_bio->logical;
7924 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7925 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7926 0, start, bvec->bv_len);
7932 init_completion(&done.done);
7934 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7935 start + bvec->bv_len - 1,
7937 btrfs_retry_endio, &done);
7943 wait_for_completion(&done.done);
7945 if (!done.uptodate) {
7946 /* We might have another mirror, so try again */
7950 offset += bvec->bv_len;
7951 start += bvec->bv_len;
7957 static int btrfs_subio_endio_read(struct inode *inode,
7958 struct btrfs_io_bio *io_bio, int err)
7960 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7964 return __btrfs_correct_data_nocsum(inode, io_bio);
7968 return __btrfs_subio_endio_read(inode, io_bio, err);
7972 static void btrfs_endio_direct_read(struct bio *bio)
7974 struct btrfs_dio_private *dip = bio->bi_private;
7975 struct inode *inode = dip->inode;
7976 struct bio *dio_bio;
7977 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7978 int err = bio->bi_error;
7980 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7981 err = btrfs_subio_endio_read(inode, io_bio, err);
7983 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7984 dip->logical_offset + dip->bytes - 1);
7985 dio_bio = dip->dio_bio;
7989 dio_end_io(dio_bio, bio->bi_error);
7992 io_bio->end_io(io_bio, err);
7996 static void btrfs_endio_direct_write(struct bio *bio)
7998 struct btrfs_dio_private *dip = bio->bi_private;
7999 struct inode *inode = dip->inode;
8000 struct btrfs_root *root = BTRFS_I(inode)->root;
8001 struct btrfs_ordered_extent *ordered = NULL;
8002 u64 ordered_offset = dip->logical_offset;
8003 u64 ordered_bytes = dip->bytes;
8004 struct bio *dio_bio;
8008 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8015 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8016 finish_ordered_fn, NULL, NULL);
8017 btrfs_queue_work(root->fs_info->endio_write_workers,
8021 * our bio might span multiple ordered extents. If we haven't
8022 * completed the accounting for the whole dio, go back and try again
8024 if (ordered_offset < dip->logical_offset + dip->bytes) {
8025 ordered_bytes = dip->logical_offset + dip->bytes -
8030 dio_bio = dip->dio_bio;
8034 dio_end_io(dio_bio, bio->bi_error);
8038 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8039 struct bio *bio, int mirror_num,
8040 unsigned long bio_flags, u64 offset)
8043 struct btrfs_root *root = BTRFS_I(inode)->root;
8044 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8045 BUG_ON(ret); /* -ENOMEM */
8049 static void btrfs_end_dio_bio(struct bio *bio)
8051 struct btrfs_dio_private *dip = bio->bi_private;
8052 int err = bio->bi_error;
8055 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8056 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8057 btrfs_ino(dip->inode), bio->bi_rw,
8058 (unsigned long long)bio->bi_iter.bi_sector,
8059 bio->bi_iter.bi_size, err);
8061 if (dip->subio_endio)
8062 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8068 * before atomic variable goto zero, we must make sure
8069 * dip->errors is perceived to be set.
8071 smp_mb__before_atomic();
8074 /* if there are more bios still pending for this dio, just exit */
8075 if (!atomic_dec_and_test(&dip->pending_bios))
8079 bio_io_error(dip->orig_bio);
8081 dip->dio_bio->bi_error = 0;
8082 bio_endio(dip->orig_bio);
8088 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8089 u64 first_sector, gfp_t gfp_flags)
8092 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8094 bio_associate_current(bio);
8098 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8099 struct inode *inode,
8100 struct btrfs_dio_private *dip,
8104 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8105 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8109 * We load all the csum data we need when we submit
8110 * the first bio to reduce the csum tree search and
8113 if (dip->logical_offset == file_offset) {
8114 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8120 if (bio == dip->orig_bio)
8123 file_offset -= dip->logical_offset;
8124 file_offset >>= inode->i_sb->s_blocksize_bits;
8125 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8130 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8131 int rw, u64 file_offset, int skip_sum,
8134 struct btrfs_dio_private *dip = bio->bi_private;
8135 int write = rw & REQ_WRITE;
8136 struct btrfs_root *root = BTRFS_I(inode)->root;
8140 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8145 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8146 BTRFS_WQ_ENDIO_DATA);
8154 if (write && async_submit) {
8155 ret = btrfs_wq_submit_bio(root->fs_info,
8156 inode, rw, bio, 0, 0,
8158 __btrfs_submit_bio_start_direct_io,
8159 __btrfs_submit_bio_done);
8163 * If we aren't doing async submit, calculate the csum of the
8166 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8170 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8176 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8182 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8185 struct inode *inode = dip->inode;
8186 struct btrfs_root *root = BTRFS_I(inode)->root;
8188 struct bio *orig_bio = dip->orig_bio;
8189 struct bio_vec *bvec = orig_bio->bi_io_vec;
8190 u64 start_sector = orig_bio->bi_iter.bi_sector;
8191 u64 file_offset = dip->logical_offset;
8196 int async_submit = 0;
8198 map_length = orig_bio->bi_iter.bi_size;
8199 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8200 &map_length, NULL, 0);
8204 if (map_length >= orig_bio->bi_iter.bi_size) {
8206 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8210 /* async crcs make it difficult to collect full stripe writes. */
8211 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8216 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8220 bio->bi_private = dip;
8221 bio->bi_end_io = btrfs_end_dio_bio;
8222 btrfs_io_bio(bio)->logical = file_offset;
8223 atomic_inc(&dip->pending_bios);
8225 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8226 if (map_length < submit_len + bvec->bv_len ||
8227 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8228 bvec->bv_offset) < bvec->bv_len) {
8230 * inc the count before we submit the bio so
8231 * we know the end IO handler won't happen before
8232 * we inc the count. Otherwise, the dip might get freed
8233 * before we're done setting it up
8235 atomic_inc(&dip->pending_bios);
8236 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8237 file_offset, skip_sum,
8241 atomic_dec(&dip->pending_bios);
8245 start_sector += submit_len >> 9;
8246 file_offset += submit_len;
8251 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8252 start_sector, GFP_NOFS);
8255 bio->bi_private = dip;
8256 bio->bi_end_io = btrfs_end_dio_bio;
8257 btrfs_io_bio(bio)->logical = file_offset;
8259 map_length = orig_bio->bi_iter.bi_size;
8260 ret = btrfs_map_block(root->fs_info, rw,
8262 &map_length, NULL, 0);
8268 submit_len += bvec->bv_len;
8275 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8284 * before atomic variable goto zero, we must
8285 * make sure dip->errors is perceived to be set.
8287 smp_mb__before_atomic();
8288 if (atomic_dec_and_test(&dip->pending_bios))
8289 bio_io_error(dip->orig_bio);
8291 /* bio_end_io() will handle error, so we needn't return it */
8295 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8296 struct inode *inode, loff_t file_offset)
8298 struct btrfs_dio_private *dip = NULL;
8299 struct bio *io_bio = NULL;
8300 struct btrfs_io_bio *btrfs_bio;
8302 int write = rw & REQ_WRITE;
8305 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8307 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8313 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8319 dip->private = dio_bio->bi_private;
8321 dip->logical_offset = file_offset;
8322 dip->bytes = dio_bio->bi_iter.bi_size;
8323 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8324 io_bio->bi_private = dip;
8325 dip->orig_bio = io_bio;
8326 dip->dio_bio = dio_bio;
8327 atomic_set(&dip->pending_bios, 0);
8328 btrfs_bio = btrfs_io_bio(io_bio);
8329 btrfs_bio->logical = file_offset;
8332 io_bio->bi_end_io = btrfs_endio_direct_write;
8334 io_bio->bi_end_io = btrfs_endio_direct_read;
8335 dip->subio_endio = btrfs_subio_endio_read;
8338 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8342 if (btrfs_bio->end_io)
8343 btrfs_bio->end_io(btrfs_bio, ret);
8347 * If we arrived here it means either we failed to submit the dip
8348 * or we either failed to clone the dio_bio or failed to allocate the
8349 * dip. If we cloned the dio_bio and allocated the dip, we can just
8350 * call bio_endio against our io_bio so that we get proper resource
8351 * cleanup if we fail to submit the dip, otherwise, we must do the
8352 * same as btrfs_endio_direct_[write|read] because we can't call these
8353 * callbacks - they require an allocated dip and a clone of dio_bio.
8355 if (io_bio && dip) {
8356 io_bio->bi_error = -EIO;
8359 * The end io callbacks free our dip, do the final put on io_bio
8360 * and all the cleanup and final put for dio_bio (through
8367 struct btrfs_ordered_extent *ordered;
8369 ordered = btrfs_lookup_ordered_extent(inode,
8371 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8373 * Decrements our ref on the ordered extent and removes
8374 * the ordered extent from the inode's ordered tree,
8375 * doing all the proper resource cleanup such as for the
8376 * reserved space and waking up any waiters for this
8377 * ordered extent (through btrfs_remove_ordered_extent).
8379 btrfs_finish_ordered_io(ordered);
8381 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8382 file_offset + dio_bio->bi_iter.bi_size - 1);
8384 dio_bio->bi_error = -EIO;
8386 * Releases and cleans up our dio_bio, no need to bio_put()
8387 * nor bio_endio()/bio_io_error() against dio_bio.
8389 dio_end_io(dio_bio, ret);
8396 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8397 const struct iov_iter *iter, loff_t offset)
8401 unsigned blocksize_mask = root->sectorsize - 1;
8402 ssize_t retval = -EINVAL;
8404 if (offset & blocksize_mask)
8407 if (iov_iter_alignment(iter) & blocksize_mask)
8410 /* If this is a write we don't need to check anymore */
8411 if (iov_iter_rw(iter) == WRITE)
8414 * Check to make sure we don't have duplicate iov_base's in this
8415 * iovec, if so return EINVAL, otherwise we'll get csum errors
8416 * when reading back.
8418 for (seg = 0; seg < iter->nr_segs; seg++) {
8419 for (i = seg + 1; i < iter->nr_segs; i++) {
8420 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8429 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8432 struct file *file = iocb->ki_filp;
8433 struct inode *inode = file->f_mapping->host;
8434 struct btrfs_root *root = BTRFS_I(inode)->root;
8435 struct btrfs_dio_data dio_data = { 0 };
8439 bool relock = false;
8442 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8445 inode_dio_begin(inode);
8446 smp_mb__after_atomic();
8449 * The generic stuff only does filemap_write_and_wait_range, which
8450 * isn't enough if we've written compressed pages to this area, so
8451 * we need to flush the dirty pages again to make absolutely sure
8452 * that any outstanding dirty pages are on disk.
8454 count = iov_iter_count(iter);
8455 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8456 &BTRFS_I(inode)->runtime_flags))
8457 filemap_fdatawrite_range(inode->i_mapping, offset,
8458 offset + count - 1);
8460 if (iov_iter_rw(iter) == WRITE) {
8462 * If the write DIO is beyond the EOF, we need update
8463 * the isize, but it is protected by i_mutex. So we can
8464 * not unlock the i_mutex at this case.
8466 if (offset + count <= inode->i_size) {
8467 mutex_unlock(&inode->i_mutex);
8470 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8473 dio_data.outstanding_extents = div64_u64(count +
8474 BTRFS_MAX_EXTENT_SIZE - 1,
8475 BTRFS_MAX_EXTENT_SIZE);
8478 * We need to know how many extents we reserved so that we can
8479 * do the accounting properly if we go over the number we
8480 * originally calculated. Abuse current->journal_info for this.
8482 dio_data.reserve = round_up(count, root->sectorsize);
8483 current->journal_info = &dio_data;
8484 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8485 &BTRFS_I(inode)->runtime_flags)) {
8486 inode_dio_end(inode);
8487 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8491 ret = __blockdev_direct_IO(iocb, inode,
8492 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8493 iter, offset, btrfs_get_blocks_direct, NULL,
8494 btrfs_submit_direct, flags);
8495 if (iov_iter_rw(iter) == WRITE) {
8496 current->journal_info = NULL;
8497 if (ret < 0 && ret != -EIOCBQUEUED) {
8498 if (dio_data.reserve)
8499 btrfs_delalloc_release_space(inode, offset,
8501 } else if (ret >= 0 && (size_t)ret < count)
8502 btrfs_delalloc_release_space(inode, offset,
8503 count - (size_t)ret);
8507 inode_dio_end(inode);
8509 mutex_lock(&inode->i_mutex);
8514 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8516 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8517 __u64 start, __u64 len)
8521 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8525 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8528 int btrfs_readpage(struct file *file, struct page *page)
8530 struct extent_io_tree *tree;
8531 tree = &BTRFS_I(page->mapping->host)->io_tree;
8532 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8535 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8537 struct extent_io_tree *tree;
8540 if (current->flags & PF_MEMALLOC) {
8541 redirty_page_for_writepage(wbc, page);
8545 tree = &BTRFS_I(page->mapping->host)->io_tree;
8546 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8549 static int btrfs_writepages(struct address_space *mapping,
8550 struct writeback_control *wbc)
8552 struct extent_io_tree *tree;
8554 tree = &BTRFS_I(mapping->host)->io_tree;
8555 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8559 btrfs_readpages(struct file *file, struct address_space *mapping,
8560 struct list_head *pages, unsigned nr_pages)
8562 struct extent_io_tree *tree;
8563 tree = &BTRFS_I(mapping->host)->io_tree;
8564 return extent_readpages(tree, mapping, pages, nr_pages,
8567 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8569 struct extent_io_tree *tree;
8570 struct extent_map_tree *map;
8573 tree = &BTRFS_I(page->mapping->host)->io_tree;
8574 map = &BTRFS_I(page->mapping->host)->extent_tree;
8575 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8577 ClearPagePrivate(page);
8578 set_page_private(page, 0);
8579 page_cache_release(page);
8584 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8586 if (PageWriteback(page) || PageDirty(page))
8588 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8591 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8592 unsigned int length)
8594 struct inode *inode = page->mapping->host;
8595 struct extent_io_tree *tree;
8596 struct btrfs_ordered_extent *ordered;
8597 struct extent_state *cached_state = NULL;
8598 u64 page_start = page_offset(page);
8599 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8600 int inode_evicting = inode->i_state & I_FREEING;
8603 * we have the page locked, so new writeback can't start,
8604 * and the dirty bit won't be cleared while we are here.
8606 * Wait for IO on this page so that we can safely clear
8607 * the PagePrivate2 bit and do ordered accounting
8609 wait_on_page_writeback(page);
8611 tree = &BTRFS_I(inode)->io_tree;
8613 btrfs_releasepage(page, GFP_NOFS);
8617 if (!inode_evicting)
8618 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8619 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8622 * IO on this page will never be started, so we need
8623 * to account for any ordered extents now
8625 if (!inode_evicting)
8626 clear_extent_bit(tree, page_start, page_end,
8627 EXTENT_DIRTY | EXTENT_DELALLOC |
8628 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8629 EXTENT_DEFRAG, 1, 0, &cached_state,
8632 * whoever cleared the private bit is responsible
8633 * for the finish_ordered_io
8635 if (TestClearPagePrivate2(page)) {
8636 struct btrfs_ordered_inode_tree *tree;
8639 tree = &BTRFS_I(inode)->ordered_tree;
8641 spin_lock_irq(&tree->lock);
8642 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8643 new_len = page_start - ordered->file_offset;
8644 if (new_len < ordered->truncated_len)
8645 ordered->truncated_len = new_len;
8646 spin_unlock_irq(&tree->lock);
8648 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8650 PAGE_CACHE_SIZE, 1))
8651 btrfs_finish_ordered_io(ordered);
8653 btrfs_put_ordered_extent(ordered);
8654 if (!inode_evicting) {
8655 cached_state = NULL;
8656 lock_extent_bits(tree, page_start, page_end, 0,
8662 * Qgroup reserved space handler
8663 * Page here will be either
8664 * 1) Already written to disk
8665 * In this case, its reserved space is released from data rsv map
8666 * and will be freed by delayed_ref handler finally.
8667 * So even we call qgroup_free_data(), it won't decrease reserved
8669 * 2) Not written to disk
8670 * This means the reserved space should be freed here.
8672 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8673 if (!inode_evicting) {
8674 clear_extent_bit(tree, page_start, page_end,
8675 EXTENT_LOCKED | EXTENT_DIRTY |
8676 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8677 EXTENT_DEFRAG, 1, 1,
8678 &cached_state, GFP_NOFS);
8680 __btrfs_releasepage(page, GFP_NOFS);
8683 ClearPageChecked(page);
8684 if (PagePrivate(page)) {
8685 ClearPagePrivate(page);
8686 set_page_private(page, 0);
8687 page_cache_release(page);
8692 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8693 * called from a page fault handler when a page is first dirtied. Hence we must
8694 * be careful to check for EOF conditions here. We set the page up correctly
8695 * for a written page which means we get ENOSPC checking when writing into
8696 * holes and correct delalloc and unwritten extent mapping on filesystems that
8697 * support these features.
8699 * We are not allowed to take the i_mutex here so we have to play games to
8700 * protect against truncate races as the page could now be beyond EOF. Because
8701 * vmtruncate() writes the inode size before removing pages, once we have the
8702 * page lock we can determine safely if the page is beyond EOF. If it is not
8703 * beyond EOF, then the page is guaranteed safe against truncation until we
8706 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8708 struct page *page = vmf->page;
8709 struct inode *inode = file_inode(vma->vm_file);
8710 struct btrfs_root *root = BTRFS_I(inode)->root;
8711 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8712 struct btrfs_ordered_extent *ordered;
8713 struct extent_state *cached_state = NULL;
8715 unsigned long zero_start;
8722 sb_start_pagefault(inode->i_sb);
8723 page_start = page_offset(page);
8724 page_end = page_start + PAGE_CACHE_SIZE - 1;
8726 ret = btrfs_delalloc_reserve_space(inode, page_start,
8729 ret = file_update_time(vma->vm_file);
8735 else /* -ENOSPC, -EIO, etc */
8736 ret = VM_FAULT_SIGBUS;
8742 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8745 size = i_size_read(inode);
8747 if ((page->mapping != inode->i_mapping) ||
8748 (page_start >= size)) {
8749 /* page got truncated out from underneath us */
8752 wait_on_page_writeback(page);
8754 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8755 set_page_extent_mapped(page);
8758 * we can't set the delalloc bits if there are pending ordered
8759 * extents. Drop our locks and wait for them to finish
8761 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8763 unlock_extent_cached(io_tree, page_start, page_end,
8764 &cached_state, GFP_NOFS);
8766 btrfs_start_ordered_extent(inode, ordered, 1);
8767 btrfs_put_ordered_extent(ordered);
8772 * XXX - page_mkwrite gets called every time the page is dirtied, even
8773 * if it was already dirty, so for space accounting reasons we need to
8774 * clear any delalloc bits for the range we are fixing to save. There
8775 * is probably a better way to do this, but for now keep consistent with
8776 * prepare_pages in the normal write path.
8778 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8779 EXTENT_DIRTY | EXTENT_DELALLOC |
8780 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8781 0, 0, &cached_state, GFP_NOFS);
8783 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8786 unlock_extent_cached(io_tree, page_start, page_end,
8787 &cached_state, GFP_NOFS);
8788 ret = VM_FAULT_SIGBUS;
8793 /* page is wholly or partially inside EOF */
8794 if (page_start + PAGE_CACHE_SIZE > size)
8795 zero_start = size & ~PAGE_CACHE_MASK;
8797 zero_start = PAGE_CACHE_SIZE;
8799 if (zero_start != PAGE_CACHE_SIZE) {
8801 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8802 flush_dcache_page(page);
8805 ClearPageChecked(page);
8806 set_page_dirty(page);
8807 SetPageUptodate(page);
8809 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8810 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8811 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8813 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8817 sb_end_pagefault(inode->i_sb);
8818 return VM_FAULT_LOCKED;
8822 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8824 sb_end_pagefault(inode->i_sb);
8828 static int btrfs_truncate(struct inode *inode)
8830 struct btrfs_root *root = BTRFS_I(inode)->root;
8831 struct btrfs_block_rsv *rsv;
8834 struct btrfs_trans_handle *trans;
8835 u64 mask = root->sectorsize - 1;
8836 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8838 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8844 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8845 * 3 things going on here
8847 * 1) We need to reserve space for our orphan item and the space to
8848 * delete our orphan item. Lord knows we don't want to have a dangling
8849 * orphan item because we didn't reserve space to remove it.
8851 * 2) We need to reserve space to update our inode.
8853 * 3) We need to have something to cache all the space that is going to
8854 * be free'd up by the truncate operation, but also have some slack
8855 * space reserved in case it uses space during the truncate (thank you
8856 * very much snapshotting).
8858 * And we need these to all be seperate. The fact is we can use alot of
8859 * space doing the truncate, and we have no earthly idea how much space
8860 * we will use, so we need the truncate reservation to be seperate so it
8861 * doesn't end up using space reserved for updating the inode or
8862 * removing the orphan item. We also need to be able to stop the
8863 * transaction and start a new one, which means we need to be able to
8864 * update the inode several times, and we have no idea of knowing how
8865 * many times that will be, so we can't just reserve 1 item for the
8866 * entirety of the opration, so that has to be done seperately as well.
8867 * Then there is the orphan item, which does indeed need to be held on
8868 * to for the whole operation, and we need nobody to touch this reserved
8869 * space except the orphan code.
8871 * So that leaves us with
8873 * 1) root->orphan_block_rsv - for the orphan deletion.
8874 * 2) rsv - for the truncate reservation, which we will steal from the
8875 * transaction reservation.
8876 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8877 * updating the inode.
8879 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8882 rsv->size = min_size;
8886 * 1 for the truncate slack space
8887 * 1 for updating the inode.
8889 trans = btrfs_start_transaction(root, 2);
8890 if (IS_ERR(trans)) {
8891 err = PTR_ERR(trans);
8895 /* Migrate the slack space for the truncate to our reserve */
8896 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8901 * So if we truncate and then write and fsync we normally would just
8902 * write the extents that changed, which is a problem if we need to
8903 * first truncate that entire inode. So set this flag so we write out
8904 * all of the extents in the inode to the sync log so we're completely
8907 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8908 trans->block_rsv = rsv;
8911 ret = btrfs_truncate_inode_items(trans, root, inode,
8913 BTRFS_EXTENT_DATA_KEY);
8914 if (ret != -ENOSPC && ret != -EAGAIN) {
8919 trans->block_rsv = &root->fs_info->trans_block_rsv;
8920 ret = btrfs_update_inode(trans, root, inode);
8926 btrfs_end_transaction(trans, root);
8927 btrfs_btree_balance_dirty(root);
8929 trans = btrfs_start_transaction(root, 2);
8930 if (IS_ERR(trans)) {
8931 ret = err = PTR_ERR(trans);
8936 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8938 BUG_ON(ret); /* shouldn't happen */
8939 trans->block_rsv = rsv;
8942 if (ret == 0 && inode->i_nlink > 0) {
8943 trans->block_rsv = root->orphan_block_rsv;
8944 ret = btrfs_orphan_del(trans, inode);
8950 trans->block_rsv = &root->fs_info->trans_block_rsv;
8951 ret = btrfs_update_inode(trans, root, inode);
8955 ret = btrfs_end_transaction(trans, root);
8956 btrfs_btree_balance_dirty(root);
8960 btrfs_free_block_rsv(root, rsv);
8969 * create a new subvolume directory/inode (helper for the ioctl).
8971 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8972 struct btrfs_root *new_root,
8973 struct btrfs_root *parent_root,
8976 struct inode *inode;
8980 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8981 new_dirid, new_dirid,
8982 S_IFDIR | (~current_umask() & S_IRWXUGO),
8985 return PTR_ERR(inode);
8986 inode->i_op = &btrfs_dir_inode_operations;
8987 inode->i_fop = &btrfs_dir_file_operations;
8989 set_nlink(inode, 1);
8990 btrfs_i_size_write(inode, 0);
8991 unlock_new_inode(inode);
8993 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8995 btrfs_err(new_root->fs_info,
8996 "error inheriting subvolume %llu properties: %d",
8997 new_root->root_key.objectid, err);
8999 err = btrfs_update_inode(trans, new_root, inode);
9005 struct inode *btrfs_alloc_inode(struct super_block *sb)
9007 struct btrfs_inode *ei;
9008 struct inode *inode;
9010 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9017 ei->last_sub_trans = 0;
9018 ei->logged_trans = 0;
9019 ei->delalloc_bytes = 0;
9020 ei->defrag_bytes = 0;
9021 ei->disk_i_size = 0;
9024 ei->index_cnt = (u64)-1;
9026 ei->last_unlink_trans = 0;
9027 ei->last_log_commit = 0;
9029 spin_lock_init(&ei->lock);
9030 ei->outstanding_extents = 0;
9031 ei->reserved_extents = 0;
9033 ei->runtime_flags = 0;
9034 ei->force_compress = BTRFS_COMPRESS_NONE;
9036 ei->delayed_node = NULL;
9038 ei->i_otime.tv_sec = 0;
9039 ei->i_otime.tv_nsec = 0;
9041 inode = &ei->vfs_inode;
9042 extent_map_tree_init(&ei->extent_tree);
9043 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9044 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9045 ei->io_tree.track_uptodate = 1;
9046 ei->io_failure_tree.track_uptodate = 1;
9047 atomic_set(&ei->sync_writers, 0);
9048 mutex_init(&ei->log_mutex);
9049 mutex_init(&ei->delalloc_mutex);
9050 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9051 INIT_LIST_HEAD(&ei->delalloc_inodes);
9052 RB_CLEAR_NODE(&ei->rb_node);
9057 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9058 void btrfs_test_destroy_inode(struct inode *inode)
9060 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9061 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9065 static void btrfs_i_callback(struct rcu_head *head)
9067 struct inode *inode = container_of(head, struct inode, i_rcu);
9068 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9071 void btrfs_destroy_inode(struct inode *inode)
9073 struct btrfs_ordered_extent *ordered;
9074 struct btrfs_root *root = BTRFS_I(inode)->root;
9076 WARN_ON(!hlist_empty(&inode->i_dentry));
9077 WARN_ON(inode->i_data.nrpages);
9078 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9079 WARN_ON(BTRFS_I(inode)->reserved_extents);
9080 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9081 WARN_ON(BTRFS_I(inode)->csum_bytes);
9082 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9085 * This can happen where we create an inode, but somebody else also
9086 * created the same inode and we need to destroy the one we already
9092 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9093 &BTRFS_I(inode)->runtime_flags)) {
9094 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9096 atomic_dec(&root->orphan_inodes);
9100 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9104 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9105 ordered->file_offset, ordered->len);
9106 btrfs_remove_ordered_extent(inode, ordered);
9107 btrfs_put_ordered_extent(ordered);
9108 btrfs_put_ordered_extent(ordered);
9111 btrfs_qgroup_check_reserved_leak(inode);
9112 inode_tree_del(inode);
9113 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9115 call_rcu(&inode->i_rcu, btrfs_i_callback);
9118 int btrfs_drop_inode(struct inode *inode)
9120 struct btrfs_root *root = BTRFS_I(inode)->root;
9125 /* the snap/subvol tree is on deleting */
9126 if (btrfs_root_refs(&root->root_item) == 0)
9129 return generic_drop_inode(inode);
9132 static void init_once(void *foo)
9134 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9136 inode_init_once(&ei->vfs_inode);
9139 void btrfs_destroy_cachep(void)
9142 * Make sure all delayed rcu free inodes are flushed before we
9146 if (btrfs_inode_cachep)
9147 kmem_cache_destroy(btrfs_inode_cachep);
9148 if (btrfs_trans_handle_cachep)
9149 kmem_cache_destroy(btrfs_trans_handle_cachep);
9150 if (btrfs_transaction_cachep)
9151 kmem_cache_destroy(btrfs_transaction_cachep);
9152 if (btrfs_path_cachep)
9153 kmem_cache_destroy(btrfs_path_cachep);
9154 if (btrfs_free_space_cachep)
9155 kmem_cache_destroy(btrfs_free_space_cachep);
9156 if (btrfs_delalloc_work_cachep)
9157 kmem_cache_destroy(btrfs_delalloc_work_cachep);
9160 int btrfs_init_cachep(void)
9162 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9163 sizeof(struct btrfs_inode), 0,
9164 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9165 if (!btrfs_inode_cachep)
9168 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9169 sizeof(struct btrfs_trans_handle), 0,
9170 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9171 if (!btrfs_trans_handle_cachep)
9174 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9175 sizeof(struct btrfs_transaction), 0,
9176 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9177 if (!btrfs_transaction_cachep)
9180 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9181 sizeof(struct btrfs_path), 0,
9182 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9183 if (!btrfs_path_cachep)
9186 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9187 sizeof(struct btrfs_free_space), 0,
9188 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9189 if (!btrfs_free_space_cachep)
9192 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9193 sizeof(struct btrfs_delalloc_work), 0,
9194 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9196 if (!btrfs_delalloc_work_cachep)
9201 btrfs_destroy_cachep();
9205 static int btrfs_getattr(struct vfsmount *mnt,
9206 struct dentry *dentry, struct kstat *stat)
9209 struct inode *inode = d_inode(dentry);
9210 u32 blocksize = inode->i_sb->s_blocksize;
9212 generic_fillattr(inode, stat);
9213 stat->dev = BTRFS_I(inode)->root->anon_dev;
9214 stat->blksize = PAGE_CACHE_SIZE;
9216 spin_lock(&BTRFS_I(inode)->lock);
9217 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9218 spin_unlock(&BTRFS_I(inode)->lock);
9219 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9220 ALIGN(delalloc_bytes, blocksize)) >> 9;
9224 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9225 struct inode *new_dir, struct dentry *new_dentry)
9227 struct btrfs_trans_handle *trans;
9228 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9229 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9230 struct inode *new_inode = d_inode(new_dentry);
9231 struct inode *old_inode = d_inode(old_dentry);
9232 struct timespec ctime = CURRENT_TIME;
9236 u64 old_ino = btrfs_ino(old_inode);
9238 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9241 /* we only allow rename subvolume link between subvolumes */
9242 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9245 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9246 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9249 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9250 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9254 /* check for collisions, even if the name isn't there */
9255 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9256 new_dentry->d_name.name,
9257 new_dentry->d_name.len);
9260 if (ret == -EEXIST) {
9262 * eexist without a new_inode */
9263 if (WARN_ON(!new_inode)) {
9267 /* maybe -EOVERFLOW */
9274 * we're using rename to replace one file with another. Start IO on it
9275 * now so we don't add too much work to the end of the transaction
9277 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9278 filemap_flush(old_inode->i_mapping);
9280 /* close the racy window with snapshot create/destroy ioctl */
9281 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9282 down_read(&root->fs_info->subvol_sem);
9284 * We want to reserve the absolute worst case amount of items. So if
9285 * both inodes are subvols and we need to unlink them then that would
9286 * require 4 item modifications, but if they are both normal inodes it
9287 * would require 5 item modifications, so we'll assume their normal
9288 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9289 * should cover the worst case number of items we'll modify.
9291 trans = btrfs_start_transaction(root, 11);
9292 if (IS_ERR(trans)) {
9293 ret = PTR_ERR(trans);
9298 btrfs_record_root_in_trans(trans, dest);
9300 ret = btrfs_set_inode_index(new_dir, &index);
9304 BTRFS_I(old_inode)->dir_index = 0ULL;
9305 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9306 /* force full log commit if subvolume involved. */
9307 btrfs_set_log_full_commit(root->fs_info, trans);
9309 ret = btrfs_insert_inode_ref(trans, dest,
9310 new_dentry->d_name.name,
9311 new_dentry->d_name.len,
9313 btrfs_ino(new_dir), index);
9317 * this is an ugly little race, but the rename is required
9318 * to make sure that if we crash, the inode is either at the
9319 * old name or the new one. pinning the log transaction lets
9320 * us make sure we don't allow a log commit to come in after
9321 * we unlink the name but before we add the new name back in.
9323 btrfs_pin_log_trans(root);
9326 inode_inc_iversion(old_dir);
9327 inode_inc_iversion(new_dir);
9328 inode_inc_iversion(old_inode);
9329 old_dir->i_ctime = old_dir->i_mtime = ctime;
9330 new_dir->i_ctime = new_dir->i_mtime = ctime;
9331 old_inode->i_ctime = ctime;
9333 if (old_dentry->d_parent != new_dentry->d_parent)
9334 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9336 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9337 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9338 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9339 old_dentry->d_name.name,
9340 old_dentry->d_name.len);
9342 ret = __btrfs_unlink_inode(trans, root, old_dir,
9343 d_inode(old_dentry),
9344 old_dentry->d_name.name,
9345 old_dentry->d_name.len);
9347 ret = btrfs_update_inode(trans, root, old_inode);
9350 btrfs_abort_transaction(trans, root, ret);
9355 inode_inc_iversion(new_inode);
9356 new_inode->i_ctime = CURRENT_TIME;
9357 if (unlikely(btrfs_ino(new_inode) ==
9358 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9359 root_objectid = BTRFS_I(new_inode)->location.objectid;
9360 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9362 new_dentry->d_name.name,
9363 new_dentry->d_name.len);
9364 BUG_ON(new_inode->i_nlink == 0);
9366 ret = btrfs_unlink_inode(trans, dest, new_dir,
9367 d_inode(new_dentry),
9368 new_dentry->d_name.name,
9369 new_dentry->d_name.len);
9371 if (!ret && new_inode->i_nlink == 0)
9372 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9374 btrfs_abort_transaction(trans, root, ret);
9379 ret = btrfs_add_link(trans, new_dir, old_inode,
9380 new_dentry->d_name.name,
9381 new_dentry->d_name.len, 0, index);
9383 btrfs_abort_transaction(trans, root, ret);
9387 if (old_inode->i_nlink == 1)
9388 BTRFS_I(old_inode)->dir_index = index;
9390 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9391 struct dentry *parent = new_dentry->d_parent;
9392 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9393 btrfs_end_log_trans(root);
9396 btrfs_end_transaction(trans, root);
9398 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9399 up_read(&root->fs_info->subvol_sem);
9404 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9405 struct inode *new_dir, struct dentry *new_dentry,
9408 if (flags & ~RENAME_NOREPLACE)
9411 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9414 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9416 struct btrfs_delalloc_work *delalloc_work;
9417 struct inode *inode;
9419 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9421 inode = delalloc_work->inode;
9422 if (delalloc_work->wait) {
9423 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9425 filemap_flush(inode->i_mapping);
9426 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9427 &BTRFS_I(inode)->runtime_flags))
9428 filemap_flush(inode->i_mapping);
9431 if (delalloc_work->delay_iput)
9432 btrfs_add_delayed_iput(inode);
9435 complete(&delalloc_work->completion);
9438 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9439 int wait, int delay_iput)
9441 struct btrfs_delalloc_work *work;
9443 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9447 init_completion(&work->completion);
9448 INIT_LIST_HEAD(&work->list);
9449 work->inode = inode;
9451 work->delay_iput = delay_iput;
9452 WARN_ON_ONCE(!inode);
9453 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9454 btrfs_run_delalloc_work, NULL, NULL);
9459 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9461 wait_for_completion(&work->completion);
9462 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9466 * some fairly slow code that needs optimization. This walks the list
9467 * of all the inodes with pending delalloc and forces them to disk.
9469 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9472 struct btrfs_inode *binode;
9473 struct inode *inode;
9474 struct btrfs_delalloc_work *work, *next;
9475 struct list_head works;
9476 struct list_head splice;
9479 INIT_LIST_HEAD(&works);
9480 INIT_LIST_HEAD(&splice);
9482 mutex_lock(&root->delalloc_mutex);
9483 spin_lock(&root->delalloc_lock);
9484 list_splice_init(&root->delalloc_inodes, &splice);
9485 while (!list_empty(&splice)) {
9486 binode = list_entry(splice.next, struct btrfs_inode,
9489 list_move_tail(&binode->delalloc_inodes,
9490 &root->delalloc_inodes);
9491 inode = igrab(&binode->vfs_inode);
9493 cond_resched_lock(&root->delalloc_lock);
9496 spin_unlock(&root->delalloc_lock);
9498 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9501 btrfs_add_delayed_iput(inode);
9507 list_add_tail(&work->list, &works);
9508 btrfs_queue_work(root->fs_info->flush_workers,
9511 if (nr != -1 && ret >= nr)
9514 spin_lock(&root->delalloc_lock);
9516 spin_unlock(&root->delalloc_lock);
9519 list_for_each_entry_safe(work, next, &works, list) {
9520 list_del_init(&work->list);
9521 btrfs_wait_and_free_delalloc_work(work);
9524 if (!list_empty_careful(&splice)) {
9525 spin_lock(&root->delalloc_lock);
9526 list_splice_tail(&splice, &root->delalloc_inodes);
9527 spin_unlock(&root->delalloc_lock);
9529 mutex_unlock(&root->delalloc_mutex);
9533 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9537 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9540 ret = __start_delalloc_inodes(root, delay_iput, -1);
9544 * the filemap_flush will queue IO into the worker threads, but
9545 * we have to make sure the IO is actually started and that
9546 * ordered extents get created before we return
9548 atomic_inc(&root->fs_info->async_submit_draining);
9549 while (atomic_read(&root->fs_info->nr_async_submits) ||
9550 atomic_read(&root->fs_info->async_delalloc_pages)) {
9551 wait_event(root->fs_info->async_submit_wait,
9552 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9553 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9555 atomic_dec(&root->fs_info->async_submit_draining);
9559 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9562 struct btrfs_root *root;
9563 struct list_head splice;
9566 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9569 INIT_LIST_HEAD(&splice);
9571 mutex_lock(&fs_info->delalloc_root_mutex);
9572 spin_lock(&fs_info->delalloc_root_lock);
9573 list_splice_init(&fs_info->delalloc_roots, &splice);
9574 while (!list_empty(&splice) && nr) {
9575 root = list_first_entry(&splice, struct btrfs_root,
9577 root = btrfs_grab_fs_root(root);
9579 list_move_tail(&root->delalloc_root,
9580 &fs_info->delalloc_roots);
9581 spin_unlock(&fs_info->delalloc_root_lock);
9583 ret = __start_delalloc_inodes(root, delay_iput, nr);
9584 btrfs_put_fs_root(root);
9592 spin_lock(&fs_info->delalloc_root_lock);
9594 spin_unlock(&fs_info->delalloc_root_lock);
9597 atomic_inc(&fs_info->async_submit_draining);
9598 while (atomic_read(&fs_info->nr_async_submits) ||
9599 atomic_read(&fs_info->async_delalloc_pages)) {
9600 wait_event(fs_info->async_submit_wait,
9601 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9602 atomic_read(&fs_info->async_delalloc_pages) == 0));
9604 atomic_dec(&fs_info->async_submit_draining);
9606 if (!list_empty_careful(&splice)) {
9607 spin_lock(&fs_info->delalloc_root_lock);
9608 list_splice_tail(&splice, &fs_info->delalloc_roots);
9609 spin_unlock(&fs_info->delalloc_root_lock);
9611 mutex_unlock(&fs_info->delalloc_root_mutex);
9615 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9616 const char *symname)
9618 struct btrfs_trans_handle *trans;
9619 struct btrfs_root *root = BTRFS_I(dir)->root;
9620 struct btrfs_path *path;
9621 struct btrfs_key key;
9622 struct inode *inode = NULL;
9630 struct btrfs_file_extent_item *ei;
9631 struct extent_buffer *leaf;
9633 name_len = strlen(symname);
9634 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9635 return -ENAMETOOLONG;
9638 * 2 items for inode item and ref
9639 * 2 items for dir items
9640 * 1 item for xattr if selinux is on
9642 trans = btrfs_start_transaction(root, 5);
9644 return PTR_ERR(trans);
9646 err = btrfs_find_free_ino(root, &objectid);
9650 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9651 dentry->d_name.len, btrfs_ino(dir), objectid,
9652 S_IFLNK|S_IRWXUGO, &index);
9653 if (IS_ERR(inode)) {
9654 err = PTR_ERR(inode);
9659 * If the active LSM wants to access the inode during
9660 * d_instantiate it needs these. Smack checks to see
9661 * if the filesystem supports xattrs by looking at the
9664 inode->i_fop = &btrfs_file_operations;
9665 inode->i_op = &btrfs_file_inode_operations;
9666 inode->i_mapping->a_ops = &btrfs_aops;
9667 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9669 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9671 goto out_unlock_inode;
9673 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9675 goto out_unlock_inode;
9677 path = btrfs_alloc_path();
9680 goto out_unlock_inode;
9682 key.objectid = btrfs_ino(inode);
9684 key.type = BTRFS_EXTENT_DATA_KEY;
9685 datasize = btrfs_file_extent_calc_inline_size(name_len);
9686 err = btrfs_insert_empty_item(trans, root, path, &key,
9689 btrfs_free_path(path);
9690 goto out_unlock_inode;
9692 leaf = path->nodes[0];
9693 ei = btrfs_item_ptr(leaf, path->slots[0],
9694 struct btrfs_file_extent_item);
9695 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9696 btrfs_set_file_extent_type(leaf, ei,
9697 BTRFS_FILE_EXTENT_INLINE);
9698 btrfs_set_file_extent_encryption(leaf, ei, 0);
9699 btrfs_set_file_extent_compression(leaf, ei, 0);
9700 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9701 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9703 ptr = btrfs_file_extent_inline_start(ei);
9704 write_extent_buffer(leaf, symname, ptr, name_len);
9705 btrfs_mark_buffer_dirty(leaf);
9706 btrfs_free_path(path);
9708 inode->i_op = &btrfs_symlink_inode_operations;
9709 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9710 inode_set_bytes(inode, name_len);
9711 btrfs_i_size_write(inode, name_len);
9712 err = btrfs_update_inode(trans, root, inode);
9715 goto out_unlock_inode;
9718 unlock_new_inode(inode);
9719 d_instantiate(dentry, inode);
9722 btrfs_end_transaction(trans, root);
9724 inode_dec_link_count(inode);
9727 btrfs_btree_balance_dirty(root);
9732 unlock_new_inode(inode);
9736 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9737 u64 start, u64 num_bytes, u64 min_size,
9738 loff_t actual_len, u64 *alloc_hint,
9739 struct btrfs_trans_handle *trans)
9741 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9742 struct extent_map *em;
9743 struct btrfs_root *root = BTRFS_I(inode)->root;
9744 struct btrfs_key ins;
9745 u64 cur_offset = start;
9749 bool own_trans = true;
9753 while (num_bytes > 0) {
9755 trans = btrfs_start_transaction(root, 3);
9756 if (IS_ERR(trans)) {
9757 ret = PTR_ERR(trans);
9762 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9763 cur_bytes = max(cur_bytes, min_size);
9764 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9765 *alloc_hint, &ins, 1, 0);
9768 btrfs_end_transaction(trans, root);
9772 ret = insert_reserved_file_extent(trans, inode,
9773 cur_offset, ins.objectid,
9774 ins.offset, ins.offset,
9775 ins.offset, 0, 0, 0,
9776 BTRFS_FILE_EXTENT_PREALLOC);
9778 btrfs_free_reserved_extent(root, ins.objectid,
9780 btrfs_abort_transaction(trans, root, ret);
9782 btrfs_end_transaction(trans, root);
9786 btrfs_drop_extent_cache(inode, cur_offset,
9787 cur_offset + ins.offset -1, 0);
9789 em = alloc_extent_map();
9791 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9792 &BTRFS_I(inode)->runtime_flags);
9796 em->start = cur_offset;
9797 em->orig_start = cur_offset;
9798 em->len = ins.offset;
9799 em->block_start = ins.objectid;
9800 em->block_len = ins.offset;
9801 em->orig_block_len = ins.offset;
9802 em->ram_bytes = ins.offset;
9803 em->bdev = root->fs_info->fs_devices->latest_bdev;
9804 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9805 em->generation = trans->transid;
9808 write_lock(&em_tree->lock);
9809 ret = add_extent_mapping(em_tree, em, 1);
9810 write_unlock(&em_tree->lock);
9813 btrfs_drop_extent_cache(inode, cur_offset,
9814 cur_offset + ins.offset - 1,
9817 free_extent_map(em);
9819 num_bytes -= ins.offset;
9820 cur_offset += ins.offset;
9821 *alloc_hint = ins.objectid + ins.offset;
9823 inode_inc_iversion(inode);
9824 inode->i_ctime = CURRENT_TIME;
9825 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9826 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9827 (actual_len > inode->i_size) &&
9828 (cur_offset > inode->i_size)) {
9829 if (cur_offset > actual_len)
9830 i_size = actual_len;
9832 i_size = cur_offset;
9833 i_size_write(inode, i_size);
9834 btrfs_ordered_update_i_size(inode, i_size, NULL);
9837 ret = btrfs_update_inode(trans, root, inode);
9840 btrfs_abort_transaction(trans, root, ret);
9842 btrfs_end_transaction(trans, root);
9847 btrfs_end_transaction(trans, root);
9852 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9853 u64 start, u64 num_bytes, u64 min_size,
9854 loff_t actual_len, u64 *alloc_hint)
9856 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9857 min_size, actual_len, alloc_hint,
9861 int btrfs_prealloc_file_range_trans(struct inode *inode,
9862 struct btrfs_trans_handle *trans, int mode,
9863 u64 start, u64 num_bytes, u64 min_size,
9864 loff_t actual_len, u64 *alloc_hint)
9866 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9867 min_size, actual_len, alloc_hint, trans);
9870 static int btrfs_set_page_dirty(struct page *page)
9872 return __set_page_dirty_nobuffers(page);
9875 static int btrfs_permission(struct inode *inode, int mask)
9877 struct btrfs_root *root = BTRFS_I(inode)->root;
9878 umode_t mode = inode->i_mode;
9880 if (mask & MAY_WRITE &&
9881 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9882 if (btrfs_root_readonly(root))
9884 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9887 return generic_permission(inode, mask);
9890 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9892 struct btrfs_trans_handle *trans;
9893 struct btrfs_root *root = BTRFS_I(dir)->root;
9894 struct inode *inode = NULL;
9900 * 5 units required for adding orphan entry
9902 trans = btrfs_start_transaction(root, 5);
9904 return PTR_ERR(trans);
9906 ret = btrfs_find_free_ino(root, &objectid);
9910 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9911 btrfs_ino(dir), objectid, mode, &index);
9912 if (IS_ERR(inode)) {
9913 ret = PTR_ERR(inode);
9918 inode->i_fop = &btrfs_file_operations;
9919 inode->i_op = &btrfs_file_inode_operations;
9921 inode->i_mapping->a_ops = &btrfs_aops;
9922 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9924 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9928 ret = btrfs_update_inode(trans, root, inode);
9931 ret = btrfs_orphan_add(trans, inode);
9936 * We set number of links to 0 in btrfs_new_inode(), and here we set
9937 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9940 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9942 set_nlink(inode, 1);
9943 unlock_new_inode(inode);
9944 d_tmpfile(dentry, inode);
9945 mark_inode_dirty(inode);
9948 btrfs_end_transaction(trans, root);
9951 btrfs_balance_delayed_items(root);
9952 btrfs_btree_balance_dirty(root);
9956 unlock_new_inode(inode);
9961 /* Inspired by filemap_check_errors() */
9962 int btrfs_inode_check_errors(struct inode *inode)
9966 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9967 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9969 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9970 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9976 static const struct inode_operations btrfs_dir_inode_operations = {
9977 .getattr = btrfs_getattr,
9978 .lookup = btrfs_lookup,
9979 .create = btrfs_create,
9980 .unlink = btrfs_unlink,
9982 .mkdir = btrfs_mkdir,
9983 .rmdir = btrfs_rmdir,
9984 .rename2 = btrfs_rename2,
9985 .symlink = btrfs_symlink,
9986 .setattr = btrfs_setattr,
9987 .mknod = btrfs_mknod,
9988 .setxattr = btrfs_setxattr,
9989 .getxattr = btrfs_getxattr,
9990 .listxattr = btrfs_listxattr,
9991 .removexattr = btrfs_removexattr,
9992 .permission = btrfs_permission,
9993 .get_acl = btrfs_get_acl,
9994 .set_acl = btrfs_set_acl,
9995 .update_time = btrfs_update_time,
9996 .tmpfile = btrfs_tmpfile,
9998 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9999 .lookup = btrfs_lookup,
10000 .permission = btrfs_permission,
10001 .get_acl = btrfs_get_acl,
10002 .set_acl = btrfs_set_acl,
10003 .update_time = btrfs_update_time,
10006 static const struct file_operations btrfs_dir_file_operations = {
10007 .llseek = generic_file_llseek,
10008 .read = generic_read_dir,
10009 .iterate = btrfs_real_readdir,
10010 .unlocked_ioctl = btrfs_ioctl,
10011 #ifdef CONFIG_COMPAT
10012 .compat_ioctl = btrfs_ioctl,
10014 .release = btrfs_release_file,
10015 .fsync = btrfs_sync_file,
10018 static struct extent_io_ops btrfs_extent_io_ops = {
10019 .fill_delalloc = run_delalloc_range,
10020 .submit_bio_hook = btrfs_submit_bio_hook,
10021 .merge_bio_hook = btrfs_merge_bio_hook,
10022 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10023 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10024 .writepage_start_hook = btrfs_writepage_start_hook,
10025 .set_bit_hook = btrfs_set_bit_hook,
10026 .clear_bit_hook = btrfs_clear_bit_hook,
10027 .merge_extent_hook = btrfs_merge_extent_hook,
10028 .split_extent_hook = btrfs_split_extent_hook,
10032 * btrfs doesn't support the bmap operation because swapfiles
10033 * use bmap to make a mapping of extents in the file. They assume
10034 * these extents won't change over the life of the file and they
10035 * use the bmap result to do IO directly to the drive.
10037 * the btrfs bmap call would return logical addresses that aren't
10038 * suitable for IO and they also will change frequently as COW
10039 * operations happen. So, swapfile + btrfs == corruption.
10041 * For now we're avoiding this by dropping bmap.
10043 static const struct address_space_operations btrfs_aops = {
10044 .readpage = btrfs_readpage,
10045 .writepage = btrfs_writepage,
10046 .writepages = btrfs_writepages,
10047 .readpages = btrfs_readpages,
10048 .direct_IO = btrfs_direct_IO,
10049 .invalidatepage = btrfs_invalidatepage,
10050 .releasepage = btrfs_releasepage,
10051 .set_page_dirty = btrfs_set_page_dirty,
10052 .error_remove_page = generic_error_remove_page,
10055 static const struct address_space_operations btrfs_symlink_aops = {
10056 .readpage = btrfs_readpage,
10057 .writepage = btrfs_writepage,
10058 .invalidatepage = btrfs_invalidatepage,
10059 .releasepage = btrfs_releasepage,
10062 static const struct inode_operations btrfs_file_inode_operations = {
10063 .getattr = btrfs_getattr,
10064 .setattr = btrfs_setattr,
10065 .setxattr = btrfs_setxattr,
10066 .getxattr = btrfs_getxattr,
10067 .listxattr = btrfs_listxattr,
10068 .removexattr = btrfs_removexattr,
10069 .permission = btrfs_permission,
10070 .fiemap = btrfs_fiemap,
10071 .get_acl = btrfs_get_acl,
10072 .set_acl = btrfs_set_acl,
10073 .update_time = btrfs_update_time,
10075 static const struct inode_operations btrfs_special_inode_operations = {
10076 .getattr = btrfs_getattr,
10077 .setattr = btrfs_setattr,
10078 .permission = btrfs_permission,
10079 .setxattr = btrfs_setxattr,
10080 .getxattr = btrfs_getxattr,
10081 .listxattr = btrfs_listxattr,
10082 .removexattr = btrfs_removexattr,
10083 .get_acl = btrfs_get_acl,
10084 .set_acl = btrfs_set_acl,
10085 .update_time = btrfs_update_time,
10087 static const struct inode_operations btrfs_symlink_inode_operations = {
10088 .readlink = generic_readlink,
10089 .follow_link = page_follow_link_light,
10090 .put_link = page_put_link,
10091 .getattr = btrfs_getattr,
10092 .setattr = btrfs_setattr,
10093 .permission = btrfs_permission,
10094 .setxattr = btrfs_setxattr,
10095 .getxattr = btrfs_getxattr,
10096 .listxattr = btrfs_listxattr,
10097 .removexattr = btrfs_removexattr,
10098 .update_time = btrfs_update_time,
10101 const struct dentry_operations btrfs_dentry_operations = {
10102 .d_delete = btrfs_dentry_delete,
10103 .d_release = btrfs_dentry_release,