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 struct btrfs_dio_data {
70 u64 outstanding_extents;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
76 static const struct inode_operations btrfs_dir_inode_operations;
77 static const struct inode_operations btrfs_symlink_inode_operations;
78 static const struct inode_operations btrfs_dir_ro_inode_operations;
79 static const struct inode_operations btrfs_special_inode_operations;
80 static const struct inode_operations btrfs_file_inode_operations;
81 static const struct address_space_operations btrfs_aops;
82 static const struct address_space_operations btrfs_symlink_aops;
83 static const struct file_operations btrfs_dir_file_operations;
84 static const struct extent_io_ops btrfs_extent_io_ops;
86 static struct kmem_cache *btrfs_inode_cachep;
87 struct kmem_cache *btrfs_trans_handle_cachep;
88 struct kmem_cache *btrfs_transaction_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, int *page_started,
109 unsigned long *nr_written, int unlock);
110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
113 u64 orig_block_len, u64 ram_bytes,
116 static int btrfs_dirty_inode(struct inode *inode);
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode *inode)
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
131 err = btrfs_init_acl(trans, inode, dir);
133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
143 struct btrfs_path *path, int extent_inserted,
144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
147 struct page **compressed_pages)
149 struct extent_buffer *leaf;
150 struct page *page = NULL;
153 struct btrfs_file_extent_item *ei;
156 size_t cur_size = size;
157 unsigned long offset;
159 if (compressed_size && compressed_pages)
160 cur_size = compressed_size;
162 inode_add_bytes(inode, size);
164 if (!extent_inserted) {
165 struct btrfs_key key;
168 key.objectid = btrfs_ino(inode);
170 key.type = BTRFS_EXTENT_DATA_KEY;
172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
191 if (compress_type != BTRFS_COMPRESS_NONE) {
194 while (compressed_size > 0) {
195 cpage = compressed_pages[i];
196 cur_size = min_t(unsigned long, compressed_size,
199 kaddr = kmap_atomic(cpage);
200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
201 kunmap_atomic(kaddr);
205 compressed_size -= cur_size;
207 btrfs_set_file_extent_compression(leaf, ei,
210 page = find_get_page(inode->i_mapping,
211 start >> PAGE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
219 btrfs_mark_buffer_dirty(leaf);
220 btrfs_release_path(path);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode)->disk_i_size = inode->i_size;
232 ret = btrfs_update_inode(trans, root, inode);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
249 struct page **compressed_pages)
251 struct btrfs_trans_handle *trans;
252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
255 u64 aligned_end = ALIGN(end, root->sectorsize);
256 u64 data_len = inline_len;
258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
263 data_len = compressed_size;
266 actual_end > root->sectorsize ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
269 (actual_end & (root->sectorsize - 1)) == 0) ||
271 data_len > root->fs_info->max_inline) {
275 path = btrfs_alloc_path();
279 trans = btrfs_join_transaction(root);
281 btrfs_free_path(path);
282 return PTR_ERR(trans);
284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
290 extent_item_size = btrfs_file_extent_calc_inline_size(
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
297 btrfs_abort_transaction(trans, root, ret);
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
303 ret = insert_inline_extent(trans, path, extent_inserted,
305 inline_len, compressed_size,
306 compress_type, compressed_pages);
307 if (ret && ret != -ENOSPC) {
308 btrfs_abort_transaction(trans, root, ret);
310 } else if (ret == -ENOSPC) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
326 btrfs_free_path(path);
327 btrfs_end_transaction(trans, root);
331 struct async_extent {
336 unsigned long nr_pages;
338 struct list_head list;
343 struct btrfs_root *root;
344 struct page *locked_page;
347 struct list_head extents;
348 struct btrfs_work work;
351 static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
355 unsigned long nr_pages,
358 struct async_extent *async_extent;
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
361 BUG_ON(!async_extent); /* -ENOMEM */
362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
367 async_extent->compress_type = compress_type;
368 list_add_tail(&async_extent->list, &cow->extents);
372 static inline int inode_need_compress(struct inode *inode)
374 struct btrfs_root *root = BTRFS_I(inode)->root;
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
406 static noinline void compress_file_range(struct inode *inode,
407 struct page *locked_page,
409 struct async_cow *async_cow,
412 struct btrfs_root *root = BTRFS_I(inode)->root;
414 u64 blocksize = root->sectorsize;
416 u64 isize = i_size_read(inode);
418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
427 int compress_type = root->fs_info->compress_type;
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end - start + 1) < SZ_16K &&
432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
435 actual_end = min_t(u64, isize, end + 1);
438 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
454 total_compressed = actual_end - start;
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it doesn't save disk space at all.
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed = min(total_compressed, max_uncompressed);
475 num_bytes = ALIGN(end - start + 1, blocksize);
476 num_bytes = max(blocksize, num_bytes);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode)) {
487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode, start, end);
507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
516 unsigned long offset = total_compressed &
518 struct page *page = pages[nr_pages_ret - 1];
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
525 kaddr = kmap_atomic(page);
526 memset(kaddr + offset, 0,
528 kunmap_atomic(kaddr);
535 /* lets try to make an inline extent */
536 if (ret || total_in < (actual_end - start)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
540 ret = cow_file_range_inline(root, inode, start, end,
543 /* try making a compressed inline extent */
544 ret = cow_file_range_inline(root, inode, start, end,
546 compress_type, pages);
549 unsigned long clear_flags = EXTENT_DELALLOC |
551 unsigned long page_error_op;
553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
561 extent_clear_unlock_delalloc(inode, start, end, NULL,
562 clear_flags, PAGE_UNLOCK |
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
577 total_compressed = ALIGN(total_compressed, blocksize);
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
583 total_in = ALIGN(total_in, PAGE_SIZE);
584 if (total_compressed >= total_in) {
587 num_bytes = total_in;
590 if (!will_compress && pages) {
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
595 for (i = 0; i < nr_pages_ret; i++) {
596 WARN_ON(pages[i]->mapping);
601 total_compressed = 0;
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
617 add_async_extent(async_cow, start, num_bytes,
618 total_compressed, pages, nr_pages_ret,
621 if (start + num_bytes < end) {
628 cleanup_and_bail_uncompressed:
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
642 extent_range_redirty_for_io(inode, start, end);
643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
658 static void free_async_extent_pages(struct async_extent *async_extent)
662 if (!async_extent->pages)
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 put_page(async_extent->pages[i]);
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline void submit_compressed_extents(struct inode *inode,
681 struct async_cow *async_cow)
683 struct async_extent *async_extent;
685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
693 while (!list_empty(&async_cow->extents)) {
694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
698 io_tree = &BTRFS_I(inode)->io_tree;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
706 lock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
710 /* allocate blocks */
711 ret = cow_file_range(inode, async_cow->locked_page,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &page_started, &nr_written, 0);
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
725 if (!page_started && !ret)
726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1,
733 unlock_page(async_cow->locked_page);
739 lock_extent(io_tree, async_extent->start,
740 async_extent->start + async_extent->ram_size - 1);
742 ret = btrfs_reserve_extent(root,
743 async_extent->compressed_size,
744 async_extent->compressed_size,
745 0, alloc_hint, &ins, 1, 1);
747 free_async_extent_pages(async_extent);
749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
760 extent_range_redirty_for_io(inode,
762 async_extent->start +
763 async_extent->ram_size - 1);
770 * here we're doing allocation and writeback of the
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
777 em = alloc_extent_map();
780 goto out_free_reserve;
782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
784 em->orig_start = em->start;
785 em->mod_start = em->start;
786 em->mod_len = em->len;
788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
790 em->orig_block_len = ins.offset;
791 em->ram_bytes = async_extent->ram_size;
792 em->bdev = root->fs_info->fs_devices->latest_bdev;
793 em->compress_type = async_extent->compress_type;
794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
799 write_lock(&em_tree->lock);
800 ret = add_extent_mapping(em_tree, em, 1);
801 write_unlock(&em_tree->lock);
802 if (ret != -EEXIST) {
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
812 goto out_free_reserve;
814 ret = btrfs_add_ordered_extent_compress(inode,
817 async_extent->ram_size,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
825 goto out_free_reserve;
827 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
830 * clear dirty, set writeback and unlock the pages.
832 extent_clear_unlock_delalloc(inode, async_extent->start,
833 async_extent->start +
834 async_extent->ram_size - 1,
835 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
836 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
838 ret = btrfs_submit_compressed_write(inode,
840 async_extent->ram_size,
842 ins.offset, async_extent->pages,
843 async_extent->nr_pages);
845 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
846 struct page *p = async_extent->pages[0];
847 const u64 start = async_extent->start;
848 const u64 end = start + async_extent->ram_size - 1;
850 p->mapping = inode->i_mapping;
851 tree->ops->writepage_end_io_hook(p, start, end,
854 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
857 free_async_extent_pages(async_extent);
859 alloc_hint = ins.objectid + ins.offset;
865 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
866 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
868 extent_clear_unlock_delalloc(inode, async_extent->start,
869 async_extent->start +
870 async_extent->ram_size - 1,
871 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
872 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
873 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
874 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
876 free_async_extent_pages(async_extent);
881 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
884 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
885 struct extent_map *em;
888 read_lock(&em_tree->lock);
889 em = search_extent_mapping(em_tree, start, num_bytes);
892 * if block start isn't an actual block number then find the
893 * first block in this inode and use that as a hint. If that
894 * block is also bogus then just don't worry about it.
896 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
898 em = search_extent_mapping(em_tree, 0, 0);
899 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
900 alloc_hint = em->block_start;
904 alloc_hint = em->block_start;
908 read_unlock(&em_tree->lock);
914 * when extent_io.c finds a delayed allocation range in the file,
915 * the call backs end up in this code. The basic idea is to
916 * allocate extents on disk for the range, and create ordered data structs
917 * in ram to track those extents.
919 * locked_page is the page that writepage had locked already. We use
920 * it to make sure we don't do extra locks or unlocks.
922 * *page_started is set to one if we unlock locked_page and do everything
923 * required to start IO on it. It may be clean and already done with
926 static noinline int cow_file_range(struct inode *inode,
927 struct page *locked_page,
928 u64 start, u64 end, int *page_started,
929 unsigned long *nr_written,
932 struct btrfs_root *root = BTRFS_I(inode)->root;
935 unsigned long ram_size;
938 u64 blocksize = root->sectorsize;
939 struct btrfs_key ins;
940 struct extent_map *em;
941 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
944 if (btrfs_is_free_space_inode(inode)) {
950 num_bytes = ALIGN(end - start + 1, blocksize);
951 num_bytes = max(blocksize, num_bytes);
952 disk_num_bytes = num_bytes;
954 /* if this is a small write inside eof, kick off defrag */
955 if (num_bytes < SZ_64K &&
956 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
957 btrfs_add_inode_defrag(NULL, inode);
960 /* lets try to make an inline extent */
961 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
964 extent_clear_unlock_delalloc(inode, start, end, NULL,
965 EXTENT_LOCKED | EXTENT_DELALLOC |
966 EXTENT_DEFRAG, PAGE_UNLOCK |
967 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
970 *nr_written = *nr_written +
971 (end - start + PAGE_SIZE) / PAGE_SIZE;
974 } else if (ret < 0) {
979 BUG_ON(disk_num_bytes >
980 btrfs_super_total_bytes(root->fs_info->super_copy));
982 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
983 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
985 while (disk_num_bytes > 0) {
988 cur_alloc_size = disk_num_bytes;
989 ret = btrfs_reserve_extent(root, cur_alloc_size,
990 root->sectorsize, 0, alloc_hint,
995 em = alloc_extent_map();
1001 em->orig_start = em->start;
1002 ram_size = ins.offset;
1003 em->len = ins.offset;
1004 em->mod_start = em->start;
1005 em->mod_len = em->len;
1007 em->block_start = ins.objectid;
1008 em->block_len = ins.offset;
1009 em->orig_block_len = ins.offset;
1010 em->ram_bytes = ram_size;
1011 em->bdev = root->fs_info->fs_devices->latest_bdev;
1012 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1013 em->generation = -1;
1016 write_lock(&em_tree->lock);
1017 ret = add_extent_mapping(em_tree, em, 1);
1018 write_unlock(&em_tree->lock);
1019 if (ret != -EEXIST) {
1020 free_extent_map(em);
1023 btrfs_drop_extent_cache(inode, start,
1024 start + ram_size - 1, 0);
1029 cur_alloc_size = ins.offset;
1030 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1031 ram_size, cur_alloc_size, 0);
1033 goto out_drop_extent_cache;
1035 if (root->root_key.objectid ==
1036 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1037 ret = btrfs_reloc_clone_csums(inode, start,
1040 goto out_drop_extent_cache;
1043 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1045 if (disk_num_bytes < cur_alloc_size)
1048 /* we're not doing compressed IO, don't unlock the first
1049 * page (which the caller expects to stay locked), don't
1050 * clear any dirty bits and don't set any writeback bits
1052 * Do set the Private2 bit so we know this page was properly
1053 * setup for writepage
1055 op = unlock ? PAGE_UNLOCK : 0;
1056 op |= PAGE_SET_PRIVATE2;
1058 extent_clear_unlock_delalloc(inode, start,
1059 start + ram_size - 1, locked_page,
1060 EXTENT_LOCKED | EXTENT_DELALLOC,
1062 disk_num_bytes -= cur_alloc_size;
1063 num_bytes -= cur_alloc_size;
1064 alloc_hint = ins.objectid + ins.offset;
1065 start += cur_alloc_size;
1070 out_drop_extent_cache:
1071 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1073 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1074 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1076 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1077 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1078 EXTENT_DELALLOC | EXTENT_DEFRAG,
1079 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1080 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1085 * work queue call back to started compression on a file and pages
1087 static noinline void async_cow_start(struct btrfs_work *work)
1089 struct async_cow *async_cow;
1091 async_cow = container_of(work, struct async_cow, work);
1093 compress_file_range(async_cow->inode, async_cow->locked_page,
1094 async_cow->start, async_cow->end, async_cow,
1096 if (num_added == 0) {
1097 btrfs_add_delayed_iput(async_cow->inode);
1098 async_cow->inode = NULL;
1103 * work queue call back to submit previously compressed pages
1105 static noinline void async_cow_submit(struct btrfs_work *work)
1107 struct async_cow *async_cow;
1108 struct btrfs_root *root;
1109 unsigned long nr_pages;
1111 async_cow = container_of(work, struct async_cow, work);
1113 root = async_cow->root;
1114 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1118 * atomic_sub_return implies a barrier for waitqueue_active
1120 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1122 waitqueue_active(&root->fs_info->async_submit_wait))
1123 wake_up(&root->fs_info->async_submit_wait);
1125 if (async_cow->inode)
1126 submit_compressed_extents(async_cow->inode, async_cow);
1129 static noinline void async_cow_free(struct btrfs_work *work)
1131 struct async_cow *async_cow;
1132 async_cow = container_of(work, struct async_cow, work);
1133 if (async_cow->inode)
1134 btrfs_add_delayed_iput(async_cow->inode);
1138 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1139 u64 start, u64 end, int *page_started,
1140 unsigned long *nr_written)
1142 struct async_cow *async_cow;
1143 struct btrfs_root *root = BTRFS_I(inode)->root;
1144 unsigned long nr_pages;
1146 int limit = 10 * SZ_1M;
1148 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1149 1, 0, NULL, GFP_NOFS);
1150 while (start < end) {
1151 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1152 BUG_ON(!async_cow); /* -ENOMEM */
1153 async_cow->inode = igrab(inode);
1154 async_cow->root = root;
1155 async_cow->locked_page = locked_page;
1156 async_cow->start = start;
1158 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1159 !btrfs_test_opt(root, FORCE_COMPRESS))
1162 cur_end = min(end, start + SZ_512K - 1);
1164 async_cow->end = cur_end;
1165 INIT_LIST_HEAD(&async_cow->extents);
1167 btrfs_init_work(&async_cow->work,
1168 btrfs_delalloc_helper,
1169 async_cow_start, async_cow_submit,
1172 nr_pages = (cur_end - start + PAGE_SIZE) >>
1174 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1176 btrfs_queue_work(root->fs_info->delalloc_workers,
1179 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1180 wait_event(root->fs_info->async_submit_wait,
1181 (atomic_read(&root->fs_info->async_delalloc_pages) <
1185 while (atomic_read(&root->fs_info->async_submit_draining) &&
1186 atomic_read(&root->fs_info->async_delalloc_pages)) {
1187 wait_event(root->fs_info->async_submit_wait,
1188 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1192 *nr_written += nr_pages;
1193 start = cur_end + 1;
1199 static noinline int csum_exist_in_range(struct btrfs_root *root,
1200 u64 bytenr, u64 num_bytes)
1203 struct btrfs_ordered_sum *sums;
1206 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1207 bytenr + num_bytes - 1, &list, 0);
1208 if (ret == 0 && list_empty(&list))
1211 while (!list_empty(&list)) {
1212 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1213 list_del(&sums->list);
1220 * when nowcow writeback call back. This checks for snapshots or COW copies
1221 * of the extents that exist in the file, and COWs the file as required.
1223 * If no cow copies or snapshots exist, we write directly to the existing
1226 static noinline int run_delalloc_nocow(struct inode *inode,
1227 struct page *locked_page,
1228 u64 start, u64 end, int *page_started, int force,
1229 unsigned long *nr_written)
1231 struct btrfs_root *root = BTRFS_I(inode)->root;
1232 struct btrfs_trans_handle *trans;
1233 struct extent_buffer *leaf;
1234 struct btrfs_path *path;
1235 struct btrfs_file_extent_item *fi;
1236 struct btrfs_key found_key;
1251 u64 ino = btrfs_ino(inode);
1253 path = btrfs_alloc_path();
1255 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1256 EXTENT_LOCKED | EXTENT_DELALLOC |
1257 EXTENT_DO_ACCOUNTING |
1258 EXTENT_DEFRAG, PAGE_UNLOCK |
1260 PAGE_SET_WRITEBACK |
1261 PAGE_END_WRITEBACK);
1265 nolock = btrfs_is_free_space_inode(inode);
1268 trans = btrfs_join_transaction_nolock(root);
1270 trans = btrfs_join_transaction(root);
1272 if (IS_ERR(trans)) {
1273 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1274 EXTENT_LOCKED | EXTENT_DELALLOC |
1275 EXTENT_DO_ACCOUNTING |
1276 EXTENT_DEFRAG, PAGE_UNLOCK |
1278 PAGE_SET_WRITEBACK |
1279 PAGE_END_WRITEBACK);
1280 btrfs_free_path(path);
1281 return PTR_ERR(trans);
1284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1286 cow_start = (u64)-1;
1289 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1293 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1294 leaf = path->nodes[0];
1295 btrfs_item_key_to_cpu(leaf, &found_key,
1296 path->slots[0] - 1);
1297 if (found_key.objectid == ino &&
1298 found_key.type == BTRFS_EXTENT_DATA_KEY)
1303 leaf = path->nodes[0];
1304 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1305 ret = btrfs_next_leaf(root, path);
1310 leaf = path->nodes[0];
1316 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1318 if (found_key.objectid > ino)
1320 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1321 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1325 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1326 found_key.offset > end)
1329 if (found_key.offset > cur_offset) {
1330 extent_end = found_key.offset;
1335 fi = btrfs_item_ptr(leaf, path->slots[0],
1336 struct btrfs_file_extent_item);
1337 extent_type = btrfs_file_extent_type(leaf, fi);
1339 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1340 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1341 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1342 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1343 extent_offset = btrfs_file_extent_offset(leaf, fi);
1344 extent_end = found_key.offset +
1345 btrfs_file_extent_num_bytes(leaf, fi);
1347 btrfs_file_extent_disk_num_bytes(leaf, fi);
1348 if (extent_end <= start) {
1352 if (disk_bytenr == 0)
1354 if (btrfs_file_extent_compression(leaf, fi) ||
1355 btrfs_file_extent_encryption(leaf, fi) ||
1356 btrfs_file_extent_other_encoding(leaf, fi))
1358 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1360 if (btrfs_extent_readonly(root, disk_bytenr))
1362 if (btrfs_cross_ref_exist(trans, root, ino,
1364 extent_offset, disk_bytenr))
1366 disk_bytenr += extent_offset;
1367 disk_bytenr += cur_offset - found_key.offset;
1368 num_bytes = min(end + 1, extent_end) - cur_offset;
1370 * if there are pending snapshots for this root,
1371 * we fall into common COW way.
1374 err = btrfs_start_write_no_snapshoting(root);
1379 * force cow if csum exists in the range.
1380 * this ensure that csum for a given extent are
1381 * either valid or do not exist.
1383 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1385 if (!btrfs_inc_nocow_writers(root->fs_info,
1389 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1390 extent_end = found_key.offset +
1391 btrfs_file_extent_inline_len(leaf,
1392 path->slots[0], fi);
1393 extent_end = ALIGN(extent_end, root->sectorsize);
1398 if (extent_end <= start) {
1400 if (!nolock && nocow)
1401 btrfs_end_write_no_snapshoting(root);
1403 btrfs_dec_nocow_writers(root->fs_info,
1408 if (cow_start == (u64)-1)
1409 cow_start = cur_offset;
1410 cur_offset = extent_end;
1411 if (cur_offset > end)
1417 btrfs_release_path(path);
1418 if (cow_start != (u64)-1) {
1419 ret = cow_file_range(inode, locked_page,
1420 cow_start, found_key.offset - 1,
1421 page_started, nr_written, 1);
1423 if (!nolock && nocow)
1424 btrfs_end_write_no_snapshoting(root);
1426 btrfs_dec_nocow_writers(root->fs_info,
1430 cow_start = (u64)-1;
1433 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1434 struct extent_map *em;
1435 struct extent_map_tree *em_tree;
1436 em_tree = &BTRFS_I(inode)->extent_tree;
1437 em = alloc_extent_map();
1438 BUG_ON(!em); /* -ENOMEM */
1439 em->start = cur_offset;
1440 em->orig_start = found_key.offset - extent_offset;
1441 em->len = num_bytes;
1442 em->block_len = num_bytes;
1443 em->block_start = disk_bytenr;
1444 em->orig_block_len = disk_num_bytes;
1445 em->ram_bytes = ram_bytes;
1446 em->bdev = root->fs_info->fs_devices->latest_bdev;
1447 em->mod_start = em->start;
1448 em->mod_len = em->len;
1449 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1450 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1451 em->generation = -1;
1453 write_lock(&em_tree->lock);
1454 ret = add_extent_mapping(em_tree, em, 1);
1455 write_unlock(&em_tree->lock);
1456 if (ret != -EEXIST) {
1457 free_extent_map(em);
1460 btrfs_drop_extent_cache(inode, em->start,
1461 em->start + em->len - 1, 0);
1463 type = BTRFS_ORDERED_PREALLOC;
1465 type = BTRFS_ORDERED_NOCOW;
1468 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1469 num_bytes, num_bytes, type);
1471 btrfs_dec_nocow_writers(root->fs_info, disk_bytenr);
1472 BUG_ON(ret); /* -ENOMEM */
1474 if (root->root_key.objectid ==
1475 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1476 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1479 if (!nolock && nocow)
1480 btrfs_end_write_no_snapshoting(root);
1485 extent_clear_unlock_delalloc(inode, cur_offset,
1486 cur_offset + num_bytes - 1,
1487 locked_page, EXTENT_LOCKED |
1488 EXTENT_DELALLOC, PAGE_UNLOCK |
1490 if (!nolock && nocow)
1491 btrfs_end_write_no_snapshoting(root);
1492 cur_offset = extent_end;
1493 if (cur_offset > end)
1496 btrfs_release_path(path);
1498 if (cur_offset <= end && cow_start == (u64)-1) {
1499 cow_start = cur_offset;
1503 if (cow_start != (u64)-1) {
1504 ret = cow_file_range(inode, locked_page, cow_start, end,
1505 page_started, nr_written, 1);
1511 err = btrfs_end_transaction(trans, root);
1515 if (ret && cur_offset < end)
1516 extent_clear_unlock_delalloc(inode, cur_offset, end,
1517 locked_page, EXTENT_LOCKED |
1518 EXTENT_DELALLOC | EXTENT_DEFRAG |
1519 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1521 PAGE_SET_WRITEBACK |
1522 PAGE_END_WRITEBACK);
1523 btrfs_free_path(path);
1527 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1530 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1531 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1535 * @defrag_bytes is a hint value, no spinlock held here,
1536 * if is not zero, it means the file is defragging.
1537 * Force cow if given extent needs to be defragged.
1539 if (BTRFS_I(inode)->defrag_bytes &&
1540 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1541 EXTENT_DEFRAG, 0, NULL))
1548 * extent_io.c call back to do delayed allocation processing
1550 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1551 u64 start, u64 end, int *page_started,
1552 unsigned long *nr_written)
1555 int force_cow = need_force_cow(inode, start, end);
1557 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1558 ret = run_delalloc_nocow(inode, locked_page, start, end,
1559 page_started, 1, nr_written);
1560 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1561 ret = run_delalloc_nocow(inode, locked_page, start, end,
1562 page_started, 0, nr_written);
1563 } else if (!inode_need_compress(inode)) {
1564 ret = cow_file_range(inode, locked_page, start, end,
1565 page_started, nr_written, 1);
1567 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1568 &BTRFS_I(inode)->runtime_flags);
1569 ret = cow_file_range_async(inode, locked_page, start, end,
1570 page_started, nr_written);
1575 static void btrfs_split_extent_hook(struct inode *inode,
1576 struct extent_state *orig, u64 split)
1580 /* not delalloc, ignore it */
1581 if (!(orig->state & EXTENT_DELALLOC))
1584 size = orig->end - orig->start + 1;
1585 if (size > BTRFS_MAX_EXTENT_SIZE) {
1590 * See the explanation in btrfs_merge_extent_hook, the same
1591 * applies here, just in reverse.
1593 new_size = orig->end - split + 1;
1594 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1595 BTRFS_MAX_EXTENT_SIZE);
1596 new_size = split - orig->start;
1597 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1598 BTRFS_MAX_EXTENT_SIZE);
1599 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1600 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1604 spin_lock(&BTRFS_I(inode)->lock);
1605 BTRFS_I(inode)->outstanding_extents++;
1606 spin_unlock(&BTRFS_I(inode)->lock);
1610 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1611 * extents so we can keep track of new extents that are just merged onto old
1612 * extents, such as when we are doing sequential writes, so we can properly
1613 * account for the metadata space we'll need.
1615 static void btrfs_merge_extent_hook(struct inode *inode,
1616 struct extent_state *new,
1617 struct extent_state *other)
1619 u64 new_size, old_size;
1622 /* not delalloc, ignore it */
1623 if (!(other->state & EXTENT_DELALLOC))
1626 if (new->start > other->start)
1627 new_size = new->end - other->start + 1;
1629 new_size = other->end - new->start + 1;
1631 /* we're not bigger than the max, unreserve the space and go */
1632 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1633 spin_lock(&BTRFS_I(inode)->lock);
1634 BTRFS_I(inode)->outstanding_extents--;
1635 spin_unlock(&BTRFS_I(inode)->lock);
1640 * We have to add up either side to figure out how many extents were
1641 * accounted for before we merged into one big extent. If the number of
1642 * extents we accounted for is <= the amount we need for the new range
1643 * then we can return, otherwise drop. Think of it like this
1647 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1648 * need 2 outstanding extents, on one side we have 1 and the other side
1649 * we have 1 so they are == and we can return. But in this case
1651 * [MAX_SIZE+4k][MAX_SIZE+4k]
1653 * Each range on their own accounts for 2 extents, but merged together
1654 * they are only 3 extents worth of accounting, so we need to drop in
1657 old_size = other->end - other->start + 1;
1658 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1659 BTRFS_MAX_EXTENT_SIZE);
1660 old_size = new->end - new->start + 1;
1661 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1662 BTRFS_MAX_EXTENT_SIZE);
1664 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1665 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1668 spin_lock(&BTRFS_I(inode)->lock);
1669 BTRFS_I(inode)->outstanding_extents--;
1670 spin_unlock(&BTRFS_I(inode)->lock);
1673 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1674 struct inode *inode)
1676 spin_lock(&root->delalloc_lock);
1677 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1678 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1679 &root->delalloc_inodes);
1680 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1681 &BTRFS_I(inode)->runtime_flags);
1682 root->nr_delalloc_inodes++;
1683 if (root->nr_delalloc_inodes == 1) {
1684 spin_lock(&root->fs_info->delalloc_root_lock);
1685 BUG_ON(!list_empty(&root->delalloc_root));
1686 list_add_tail(&root->delalloc_root,
1687 &root->fs_info->delalloc_roots);
1688 spin_unlock(&root->fs_info->delalloc_root_lock);
1691 spin_unlock(&root->delalloc_lock);
1694 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1695 struct inode *inode)
1697 spin_lock(&root->delalloc_lock);
1698 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1699 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1700 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1701 &BTRFS_I(inode)->runtime_flags);
1702 root->nr_delalloc_inodes--;
1703 if (!root->nr_delalloc_inodes) {
1704 spin_lock(&root->fs_info->delalloc_root_lock);
1705 BUG_ON(list_empty(&root->delalloc_root));
1706 list_del_init(&root->delalloc_root);
1707 spin_unlock(&root->fs_info->delalloc_root_lock);
1710 spin_unlock(&root->delalloc_lock);
1714 * extent_io.c set_bit_hook, used to track delayed allocation
1715 * bytes in this file, and to maintain the list of inodes that
1716 * have pending delalloc work to be done.
1718 static void btrfs_set_bit_hook(struct inode *inode,
1719 struct extent_state *state, unsigned *bits)
1722 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1725 * set_bit and clear bit hooks normally require _irqsave/restore
1726 * but in this case, we are only testing for the DELALLOC
1727 * bit, which is only set or cleared with irqs on
1729 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1730 struct btrfs_root *root = BTRFS_I(inode)->root;
1731 u64 len = state->end + 1 - state->start;
1732 bool do_list = !btrfs_is_free_space_inode(inode);
1734 if (*bits & EXTENT_FIRST_DELALLOC) {
1735 *bits &= ~EXTENT_FIRST_DELALLOC;
1737 spin_lock(&BTRFS_I(inode)->lock);
1738 BTRFS_I(inode)->outstanding_extents++;
1739 spin_unlock(&BTRFS_I(inode)->lock);
1742 /* For sanity tests */
1743 if (btrfs_test_is_dummy_root(root))
1746 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1747 root->fs_info->delalloc_batch);
1748 spin_lock(&BTRFS_I(inode)->lock);
1749 BTRFS_I(inode)->delalloc_bytes += len;
1750 if (*bits & EXTENT_DEFRAG)
1751 BTRFS_I(inode)->defrag_bytes += len;
1752 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1753 &BTRFS_I(inode)->runtime_flags))
1754 btrfs_add_delalloc_inodes(root, inode);
1755 spin_unlock(&BTRFS_I(inode)->lock);
1760 * extent_io.c clear_bit_hook, see set_bit_hook for why
1762 static void btrfs_clear_bit_hook(struct inode *inode,
1763 struct extent_state *state,
1766 u64 len = state->end + 1 - state->start;
1767 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1768 BTRFS_MAX_EXTENT_SIZE);
1770 spin_lock(&BTRFS_I(inode)->lock);
1771 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1772 BTRFS_I(inode)->defrag_bytes -= len;
1773 spin_unlock(&BTRFS_I(inode)->lock);
1776 * set_bit and clear bit hooks normally require _irqsave/restore
1777 * but in this case, we are only testing for the DELALLOC
1778 * bit, which is only set or cleared with irqs on
1780 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1781 struct btrfs_root *root = BTRFS_I(inode)->root;
1782 bool do_list = !btrfs_is_free_space_inode(inode);
1784 if (*bits & EXTENT_FIRST_DELALLOC) {
1785 *bits &= ~EXTENT_FIRST_DELALLOC;
1786 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1787 spin_lock(&BTRFS_I(inode)->lock);
1788 BTRFS_I(inode)->outstanding_extents -= num_extents;
1789 spin_unlock(&BTRFS_I(inode)->lock);
1793 * We don't reserve metadata space for space cache inodes so we
1794 * don't need to call dellalloc_release_metadata if there is an
1797 if (*bits & EXTENT_DO_ACCOUNTING &&
1798 root != root->fs_info->tree_root)
1799 btrfs_delalloc_release_metadata(inode, len);
1801 /* For sanity tests. */
1802 if (btrfs_test_is_dummy_root(root))
1805 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1806 && do_list && !(state->state & EXTENT_NORESERVE))
1807 btrfs_free_reserved_data_space_noquota(inode,
1810 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1811 root->fs_info->delalloc_batch);
1812 spin_lock(&BTRFS_I(inode)->lock);
1813 BTRFS_I(inode)->delalloc_bytes -= len;
1814 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1815 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1816 &BTRFS_I(inode)->runtime_flags))
1817 btrfs_del_delalloc_inode(root, inode);
1818 spin_unlock(&BTRFS_I(inode)->lock);
1823 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1824 * we don't create bios that span stripes or chunks
1826 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1827 size_t size, struct bio *bio,
1828 unsigned long bio_flags)
1830 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1831 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1836 if (bio_flags & EXTENT_BIO_COMPRESSED)
1839 length = bio->bi_iter.bi_size;
1840 map_length = length;
1841 ret = btrfs_map_block(root->fs_info, rw, logical,
1842 &map_length, NULL, 0);
1843 /* Will always return 0 with map_multi == NULL */
1845 if (map_length < length + size)
1851 * in order to insert checksums into the metadata in large chunks,
1852 * we wait until bio submission time. All the pages in the bio are
1853 * checksummed and sums are attached onto the ordered extent record.
1855 * At IO completion time the cums attached on the ordered extent record
1856 * are inserted into the btree
1858 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1859 struct bio *bio, int mirror_num,
1860 unsigned long bio_flags,
1863 struct btrfs_root *root = BTRFS_I(inode)->root;
1866 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1867 BUG_ON(ret); /* -ENOMEM */
1872 * in order to insert checksums into the metadata in large chunks,
1873 * we wait until bio submission time. All the pages in the bio are
1874 * checksummed and sums are attached onto the ordered extent record.
1876 * At IO completion time the cums attached on the ordered extent record
1877 * are inserted into the btree
1879 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1880 int mirror_num, unsigned long bio_flags,
1883 struct btrfs_root *root = BTRFS_I(inode)->root;
1886 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1888 bio->bi_error = ret;
1895 * extent_io.c submission hook. This does the right thing for csum calculation
1896 * on write, or reading the csums from the tree before a read
1898 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1899 int mirror_num, unsigned long bio_flags,
1902 struct btrfs_root *root = BTRFS_I(inode)->root;
1903 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1906 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1908 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1910 if (btrfs_is_free_space_inode(inode))
1911 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1913 if (!(rw & REQ_WRITE)) {
1914 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1918 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1919 ret = btrfs_submit_compressed_read(inode, bio,
1923 } else if (!skip_sum) {
1924 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1929 } else if (async && !skip_sum) {
1930 /* csum items have already been cloned */
1931 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1933 /* we're doing a write, do the async checksumming */
1934 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1935 inode, rw, bio, mirror_num,
1936 bio_flags, bio_offset,
1937 __btrfs_submit_bio_start,
1938 __btrfs_submit_bio_done);
1940 } else if (!skip_sum) {
1941 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1947 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1951 bio->bi_error = ret;
1958 * given a list of ordered sums record them in the inode. This happens
1959 * at IO completion time based on sums calculated at bio submission time.
1961 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1962 struct inode *inode, u64 file_offset,
1963 struct list_head *list)
1965 struct btrfs_ordered_sum *sum;
1967 list_for_each_entry(sum, list, list) {
1968 trans->adding_csums = 1;
1969 btrfs_csum_file_blocks(trans,
1970 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1971 trans->adding_csums = 0;
1976 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1977 struct extent_state **cached_state)
1979 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1980 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1984 /* see btrfs_writepage_start_hook for details on why this is required */
1985 struct btrfs_writepage_fixup {
1987 struct btrfs_work work;
1990 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1992 struct btrfs_writepage_fixup *fixup;
1993 struct btrfs_ordered_extent *ordered;
1994 struct extent_state *cached_state = NULL;
1996 struct inode *inode;
2001 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2005 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2006 ClearPageChecked(page);
2010 inode = page->mapping->host;
2011 page_start = page_offset(page);
2012 page_end = page_offset(page) + PAGE_SIZE - 1;
2014 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2017 /* already ordered? We're done */
2018 if (PagePrivate2(page))
2021 ordered = btrfs_lookup_ordered_range(inode, page_start,
2024 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2025 page_end, &cached_state, GFP_NOFS);
2027 btrfs_start_ordered_extent(inode, ordered, 1);
2028 btrfs_put_ordered_extent(ordered);
2032 ret = btrfs_delalloc_reserve_space(inode, page_start,
2035 mapping_set_error(page->mapping, ret);
2036 end_extent_writepage(page, ret, page_start, page_end);
2037 ClearPageChecked(page);
2041 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2042 ClearPageChecked(page);
2043 set_page_dirty(page);
2045 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2046 &cached_state, GFP_NOFS);
2054 * There are a few paths in the higher layers of the kernel that directly
2055 * set the page dirty bit without asking the filesystem if it is a
2056 * good idea. This causes problems because we want to make sure COW
2057 * properly happens and the data=ordered rules are followed.
2059 * In our case any range that doesn't have the ORDERED bit set
2060 * hasn't been properly setup for IO. We kick off an async process
2061 * to fix it up. The async helper will wait for ordered extents, set
2062 * the delalloc bit and make it safe to write the page.
2064 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2066 struct inode *inode = page->mapping->host;
2067 struct btrfs_writepage_fixup *fixup;
2068 struct btrfs_root *root = BTRFS_I(inode)->root;
2070 /* this page is properly in the ordered list */
2071 if (TestClearPagePrivate2(page))
2074 if (PageChecked(page))
2077 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2081 SetPageChecked(page);
2083 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2084 btrfs_writepage_fixup_worker, NULL, NULL);
2086 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2090 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2091 struct inode *inode, u64 file_pos,
2092 u64 disk_bytenr, u64 disk_num_bytes,
2093 u64 num_bytes, u64 ram_bytes,
2094 u8 compression, u8 encryption,
2095 u16 other_encoding, int extent_type)
2097 struct btrfs_root *root = BTRFS_I(inode)->root;
2098 struct btrfs_file_extent_item *fi;
2099 struct btrfs_path *path;
2100 struct extent_buffer *leaf;
2101 struct btrfs_key ins;
2102 int extent_inserted = 0;
2105 path = btrfs_alloc_path();
2110 * we may be replacing one extent in the tree with another.
2111 * The new extent is pinned in the extent map, and we don't want
2112 * to drop it from the cache until it is completely in the btree.
2114 * So, tell btrfs_drop_extents to leave this extent in the cache.
2115 * the caller is expected to unpin it and allow it to be merged
2118 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2119 file_pos + num_bytes, NULL, 0,
2120 1, sizeof(*fi), &extent_inserted);
2124 if (!extent_inserted) {
2125 ins.objectid = btrfs_ino(inode);
2126 ins.offset = file_pos;
2127 ins.type = BTRFS_EXTENT_DATA_KEY;
2129 path->leave_spinning = 1;
2130 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2135 leaf = path->nodes[0];
2136 fi = btrfs_item_ptr(leaf, path->slots[0],
2137 struct btrfs_file_extent_item);
2138 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2139 btrfs_set_file_extent_type(leaf, fi, extent_type);
2140 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2141 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2142 btrfs_set_file_extent_offset(leaf, fi, 0);
2143 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2144 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2145 btrfs_set_file_extent_compression(leaf, fi, compression);
2146 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2147 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2149 btrfs_mark_buffer_dirty(leaf);
2150 btrfs_release_path(path);
2152 inode_add_bytes(inode, num_bytes);
2154 ins.objectid = disk_bytenr;
2155 ins.offset = disk_num_bytes;
2156 ins.type = BTRFS_EXTENT_ITEM_KEY;
2157 ret = btrfs_alloc_reserved_file_extent(trans, root,
2158 root->root_key.objectid,
2159 btrfs_ino(inode), file_pos,
2162 * Release the reserved range from inode dirty range map, as it is
2163 * already moved into delayed_ref_head
2165 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2167 btrfs_free_path(path);
2172 /* snapshot-aware defrag */
2173 struct sa_defrag_extent_backref {
2174 struct rb_node node;
2175 struct old_sa_defrag_extent *old;
2184 struct old_sa_defrag_extent {
2185 struct list_head list;
2186 struct new_sa_defrag_extent *new;
2195 struct new_sa_defrag_extent {
2196 struct rb_root root;
2197 struct list_head head;
2198 struct btrfs_path *path;
2199 struct inode *inode;
2207 static int backref_comp(struct sa_defrag_extent_backref *b1,
2208 struct sa_defrag_extent_backref *b2)
2210 if (b1->root_id < b2->root_id)
2212 else if (b1->root_id > b2->root_id)
2215 if (b1->inum < b2->inum)
2217 else if (b1->inum > b2->inum)
2220 if (b1->file_pos < b2->file_pos)
2222 else if (b1->file_pos > b2->file_pos)
2226 * [------------------------------] ===> (a range of space)
2227 * |<--->| |<---->| =============> (fs/file tree A)
2228 * |<---------------------------->| ===> (fs/file tree B)
2230 * A range of space can refer to two file extents in one tree while
2231 * refer to only one file extent in another tree.
2233 * So we may process a disk offset more than one time(two extents in A)
2234 * and locate at the same extent(one extent in B), then insert two same
2235 * backrefs(both refer to the extent in B).
2240 static void backref_insert(struct rb_root *root,
2241 struct sa_defrag_extent_backref *backref)
2243 struct rb_node **p = &root->rb_node;
2244 struct rb_node *parent = NULL;
2245 struct sa_defrag_extent_backref *entry;
2250 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2252 ret = backref_comp(backref, entry);
2256 p = &(*p)->rb_right;
2259 rb_link_node(&backref->node, parent, p);
2260 rb_insert_color(&backref->node, root);
2264 * Note the backref might has changed, and in this case we just return 0.
2266 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2269 struct btrfs_file_extent_item *extent;
2270 struct btrfs_fs_info *fs_info;
2271 struct old_sa_defrag_extent *old = ctx;
2272 struct new_sa_defrag_extent *new = old->new;
2273 struct btrfs_path *path = new->path;
2274 struct btrfs_key key;
2275 struct btrfs_root *root;
2276 struct sa_defrag_extent_backref *backref;
2277 struct extent_buffer *leaf;
2278 struct inode *inode = new->inode;
2284 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2285 inum == btrfs_ino(inode))
2288 key.objectid = root_id;
2289 key.type = BTRFS_ROOT_ITEM_KEY;
2290 key.offset = (u64)-1;
2292 fs_info = BTRFS_I(inode)->root->fs_info;
2293 root = btrfs_read_fs_root_no_name(fs_info, &key);
2295 if (PTR_ERR(root) == -ENOENT)
2298 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2299 inum, offset, root_id);
2300 return PTR_ERR(root);
2303 key.objectid = inum;
2304 key.type = BTRFS_EXTENT_DATA_KEY;
2305 if (offset > (u64)-1 << 32)
2308 key.offset = offset;
2310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2311 if (WARN_ON(ret < 0))
2318 leaf = path->nodes[0];
2319 slot = path->slots[0];
2321 if (slot >= btrfs_header_nritems(leaf)) {
2322 ret = btrfs_next_leaf(root, path);
2325 } else if (ret > 0) {
2334 btrfs_item_key_to_cpu(leaf, &key, slot);
2336 if (key.objectid > inum)
2339 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2342 extent = btrfs_item_ptr(leaf, slot,
2343 struct btrfs_file_extent_item);
2345 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2349 * 'offset' refers to the exact key.offset,
2350 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2351 * (key.offset - extent_offset).
2353 if (key.offset != offset)
2356 extent_offset = btrfs_file_extent_offset(leaf, extent);
2357 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2359 if (extent_offset >= old->extent_offset + old->offset +
2360 old->len || extent_offset + num_bytes <=
2361 old->extent_offset + old->offset)
2366 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2372 backref->root_id = root_id;
2373 backref->inum = inum;
2374 backref->file_pos = offset;
2375 backref->num_bytes = num_bytes;
2376 backref->extent_offset = extent_offset;
2377 backref->generation = btrfs_file_extent_generation(leaf, extent);
2379 backref_insert(&new->root, backref);
2382 btrfs_release_path(path);
2387 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2388 struct new_sa_defrag_extent *new)
2390 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2391 struct old_sa_defrag_extent *old, *tmp;
2396 list_for_each_entry_safe(old, tmp, &new->head, list) {
2397 ret = iterate_inodes_from_logical(old->bytenr +
2398 old->extent_offset, fs_info,
2399 path, record_one_backref,
2401 if (ret < 0 && ret != -ENOENT)
2404 /* no backref to be processed for this extent */
2406 list_del(&old->list);
2411 if (list_empty(&new->head))
2417 static int relink_is_mergable(struct extent_buffer *leaf,
2418 struct btrfs_file_extent_item *fi,
2419 struct new_sa_defrag_extent *new)
2421 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2424 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2427 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2430 if (btrfs_file_extent_encryption(leaf, fi) ||
2431 btrfs_file_extent_other_encoding(leaf, fi))
2438 * Note the backref might has changed, and in this case we just return 0.
2440 static noinline int relink_extent_backref(struct btrfs_path *path,
2441 struct sa_defrag_extent_backref *prev,
2442 struct sa_defrag_extent_backref *backref)
2444 struct btrfs_file_extent_item *extent;
2445 struct btrfs_file_extent_item *item;
2446 struct btrfs_ordered_extent *ordered;
2447 struct btrfs_trans_handle *trans;
2448 struct btrfs_fs_info *fs_info;
2449 struct btrfs_root *root;
2450 struct btrfs_key key;
2451 struct extent_buffer *leaf;
2452 struct old_sa_defrag_extent *old = backref->old;
2453 struct new_sa_defrag_extent *new = old->new;
2454 struct inode *src_inode = new->inode;
2455 struct inode *inode;
2456 struct extent_state *cached = NULL;
2465 if (prev && prev->root_id == backref->root_id &&
2466 prev->inum == backref->inum &&
2467 prev->file_pos + prev->num_bytes == backref->file_pos)
2470 /* step 1: get root */
2471 key.objectid = backref->root_id;
2472 key.type = BTRFS_ROOT_ITEM_KEY;
2473 key.offset = (u64)-1;
2475 fs_info = BTRFS_I(src_inode)->root->fs_info;
2476 index = srcu_read_lock(&fs_info->subvol_srcu);
2478 root = btrfs_read_fs_root_no_name(fs_info, &key);
2480 srcu_read_unlock(&fs_info->subvol_srcu, index);
2481 if (PTR_ERR(root) == -ENOENT)
2483 return PTR_ERR(root);
2486 if (btrfs_root_readonly(root)) {
2487 srcu_read_unlock(&fs_info->subvol_srcu, index);
2491 /* step 2: get inode */
2492 key.objectid = backref->inum;
2493 key.type = BTRFS_INODE_ITEM_KEY;
2496 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2497 if (IS_ERR(inode)) {
2498 srcu_read_unlock(&fs_info->subvol_srcu, index);
2502 srcu_read_unlock(&fs_info->subvol_srcu, index);
2504 /* step 3: relink backref */
2505 lock_start = backref->file_pos;
2506 lock_end = backref->file_pos + backref->num_bytes - 1;
2507 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2510 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2512 btrfs_put_ordered_extent(ordered);
2516 trans = btrfs_join_transaction(root);
2517 if (IS_ERR(trans)) {
2518 ret = PTR_ERR(trans);
2522 key.objectid = backref->inum;
2523 key.type = BTRFS_EXTENT_DATA_KEY;
2524 key.offset = backref->file_pos;
2526 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2529 } else if (ret > 0) {
2534 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2535 struct btrfs_file_extent_item);
2537 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2538 backref->generation)
2541 btrfs_release_path(path);
2543 start = backref->file_pos;
2544 if (backref->extent_offset < old->extent_offset + old->offset)
2545 start += old->extent_offset + old->offset -
2546 backref->extent_offset;
2548 len = min(backref->extent_offset + backref->num_bytes,
2549 old->extent_offset + old->offset + old->len);
2550 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2552 ret = btrfs_drop_extents(trans, root, inode, start,
2557 key.objectid = btrfs_ino(inode);
2558 key.type = BTRFS_EXTENT_DATA_KEY;
2561 path->leave_spinning = 1;
2563 struct btrfs_file_extent_item *fi;
2565 struct btrfs_key found_key;
2567 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2572 leaf = path->nodes[0];
2573 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2575 fi = btrfs_item_ptr(leaf, path->slots[0],
2576 struct btrfs_file_extent_item);
2577 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2579 if (extent_len + found_key.offset == start &&
2580 relink_is_mergable(leaf, fi, new)) {
2581 btrfs_set_file_extent_num_bytes(leaf, fi,
2583 btrfs_mark_buffer_dirty(leaf);
2584 inode_add_bytes(inode, len);
2590 btrfs_release_path(path);
2595 ret = btrfs_insert_empty_item(trans, root, path, &key,
2598 btrfs_abort_transaction(trans, root, ret);
2602 leaf = path->nodes[0];
2603 item = btrfs_item_ptr(leaf, path->slots[0],
2604 struct btrfs_file_extent_item);
2605 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2606 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2607 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2608 btrfs_set_file_extent_num_bytes(leaf, item, len);
2609 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2610 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2611 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2612 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2613 btrfs_set_file_extent_encryption(leaf, item, 0);
2614 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2616 btrfs_mark_buffer_dirty(leaf);
2617 inode_add_bytes(inode, len);
2618 btrfs_release_path(path);
2620 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2622 backref->root_id, backref->inum,
2623 new->file_pos); /* start - extent_offset */
2625 btrfs_abort_transaction(trans, root, ret);
2631 btrfs_release_path(path);
2632 path->leave_spinning = 0;
2633 btrfs_end_transaction(trans, root);
2635 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2641 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2643 struct old_sa_defrag_extent *old, *tmp;
2648 list_for_each_entry_safe(old, tmp, &new->head, list) {
2654 static void relink_file_extents(struct new_sa_defrag_extent *new)
2656 struct btrfs_path *path;
2657 struct sa_defrag_extent_backref *backref;
2658 struct sa_defrag_extent_backref *prev = NULL;
2659 struct inode *inode;
2660 struct btrfs_root *root;
2661 struct rb_node *node;
2665 root = BTRFS_I(inode)->root;
2667 path = btrfs_alloc_path();
2671 if (!record_extent_backrefs(path, new)) {
2672 btrfs_free_path(path);
2675 btrfs_release_path(path);
2678 node = rb_first(&new->root);
2681 rb_erase(node, &new->root);
2683 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2685 ret = relink_extent_backref(path, prev, backref);
2698 btrfs_free_path(path);
2700 free_sa_defrag_extent(new);
2702 atomic_dec(&root->fs_info->defrag_running);
2703 wake_up(&root->fs_info->transaction_wait);
2706 static struct new_sa_defrag_extent *
2707 record_old_file_extents(struct inode *inode,
2708 struct btrfs_ordered_extent *ordered)
2710 struct btrfs_root *root = BTRFS_I(inode)->root;
2711 struct btrfs_path *path;
2712 struct btrfs_key key;
2713 struct old_sa_defrag_extent *old;
2714 struct new_sa_defrag_extent *new;
2717 new = kmalloc(sizeof(*new), GFP_NOFS);
2722 new->file_pos = ordered->file_offset;
2723 new->len = ordered->len;
2724 new->bytenr = ordered->start;
2725 new->disk_len = ordered->disk_len;
2726 new->compress_type = ordered->compress_type;
2727 new->root = RB_ROOT;
2728 INIT_LIST_HEAD(&new->head);
2730 path = btrfs_alloc_path();
2734 key.objectid = btrfs_ino(inode);
2735 key.type = BTRFS_EXTENT_DATA_KEY;
2736 key.offset = new->file_pos;
2738 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2741 if (ret > 0 && path->slots[0] > 0)
2744 /* find out all the old extents for the file range */
2746 struct btrfs_file_extent_item *extent;
2747 struct extent_buffer *l;
2756 slot = path->slots[0];
2758 if (slot >= btrfs_header_nritems(l)) {
2759 ret = btrfs_next_leaf(root, path);
2767 btrfs_item_key_to_cpu(l, &key, slot);
2769 if (key.objectid != btrfs_ino(inode))
2771 if (key.type != BTRFS_EXTENT_DATA_KEY)
2773 if (key.offset >= new->file_pos + new->len)
2776 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2778 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2779 if (key.offset + num_bytes < new->file_pos)
2782 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2786 extent_offset = btrfs_file_extent_offset(l, extent);
2788 old = kmalloc(sizeof(*old), GFP_NOFS);
2792 offset = max(new->file_pos, key.offset);
2793 end = min(new->file_pos + new->len, key.offset + num_bytes);
2795 old->bytenr = disk_bytenr;
2796 old->extent_offset = extent_offset;
2797 old->offset = offset - key.offset;
2798 old->len = end - offset;
2801 list_add_tail(&old->list, &new->head);
2807 btrfs_free_path(path);
2808 atomic_inc(&root->fs_info->defrag_running);
2813 btrfs_free_path(path);
2815 free_sa_defrag_extent(new);
2819 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2822 struct btrfs_block_group_cache *cache;
2824 cache = btrfs_lookup_block_group(root->fs_info, start);
2827 spin_lock(&cache->lock);
2828 cache->delalloc_bytes -= len;
2829 spin_unlock(&cache->lock);
2831 btrfs_put_block_group(cache);
2834 /* as ordered data IO finishes, this gets called so we can finish
2835 * an ordered extent if the range of bytes in the file it covers are
2838 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2840 struct inode *inode = ordered_extent->inode;
2841 struct btrfs_root *root = BTRFS_I(inode)->root;
2842 struct btrfs_trans_handle *trans = NULL;
2843 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2844 struct extent_state *cached_state = NULL;
2845 struct new_sa_defrag_extent *new = NULL;
2846 int compress_type = 0;
2848 u64 logical_len = ordered_extent->len;
2850 bool truncated = false;
2852 nolock = btrfs_is_free_space_inode(inode);
2854 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2859 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2860 ordered_extent->file_offset +
2861 ordered_extent->len - 1);
2863 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2865 logical_len = ordered_extent->truncated_len;
2866 /* Truncated the entire extent, don't bother adding */
2871 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2872 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2875 * For mwrite(mmap + memset to write) case, we still reserve
2876 * space for NOCOW range.
2877 * As NOCOW won't cause a new delayed ref, just free the space
2879 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2880 ordered_extent->len);
2881 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2883 trans = btrfs_join_transaction_nolock(root);
2885 trans = btrfs_join_transaction(root);
2886 if (IS_ERR(trans)) {
2887 ret = PTR_ERR(trans);
2891 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2892 ret = btrfs_update_inode_fallback(trans, root, inode);
2893 if (ret) /* -ENOMEM or corruption */
2894 btrfs_abort_transaction(trans, root, ret);
2898 lock_extent_bits(io_tree, ordered_extent->file_offset,
2899 ordered_extent->file_offset + ordered_extent->len - 1,
2902 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2903 ordered_extent->file_offset + ordered_extent->len - 1,
2904 EXTENT_DEFRAG, 1, cached_state);
2906 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2907 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2908 /* the inode is shared */
2909 new = record_old_file_extents(inode, ordered_extent);
2911 clear_extent_bit(io_tree, ordered_extent->file_offset,
2912 ordered_extent->file_offset + ordered_extent->len - 1,
2913 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2917 trans = btrfs_join_transaction_nolock(root);
2919 trans = btrfs_join_transaction(root);
2920 if (IS_ERR(trans)) {
2921 ret = PTR_ERR(trans);
2926 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2928 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2929 compress_type = ordered_extent->compress_type;
2930 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2931 BUG_ON(compress_type);
2932 ret = btrfs_mark_extent_written(trans, inode,
2933 ordered_extent->file_offset,
2934 ordered_extent->file_offset +
2937 BUG_ON(root == root->fs_info->tree_root);
2938 ret = insert_reserved_file_extent(trans, inode,
2939 ordered_extent->file_offset,
2940 ordered_extent->start,
2941 ordered_extent->disk_len,
2942 logical_len, logical_len,
2943 compress_type, 0, 0,
2944 BTRFS_FILE_EXTENT_REG);
2946 btrfs_release_delalloc_bytes(root,
2947 ordered_extent->start,
2948 ordered_extent->disk_len);
2950 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2951 ordered_extent->file_offset, ordered_extent->len,
2954 btrfs_abort_transaction(trans, root, ret);
2958 add_pending_csums(trans, inode, ordered_extent->file_offset,
2959 &ordered_extent->list);
2961 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2962 ret = btrfs_update_inode_fallback(trans, root, inode);
2963 if (ret) { /* -ENOMEM or corruption */
2964 btrfs_abort_transaction(trans, root, ret);
2969 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2970 ordered_extent->file_offset +
2971 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2973 if (root != root->fs_info->tree_root)
2974 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2976 btrfs_end_transaction(trans, root);
2978 if (ret || truncated) {
2982 start = ordered_extent->file_offset + logical_len;
2984 start = ordered_extent->file_offset;
2985 end = ordered_extent->file_offset + ordered_extent->len - 1;
2986 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2988 /* Drop the cache for the part of the extent we didn't write. */
2989 btrfs_drop_extent_cache(inode, start, end, 0);
2992 * If the ordered extent had an IOERR or something else went
2993 * wrong we need to return the space for this ordered extent
2994 * back to the allocator. We only free the extent in the
2995 * truncated case if we didn't write out the extent at all.
2997 if ((ret || !logical_len) &&
2998 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2999 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3000 btrfs_free_reserved_extent(root, ordered_extent->start,
3001 ordered_extent->disk_len, 1);
3006 * This needs to be done to make sure anybody waiting knows we are done
3007 * updating everything for this ordered extent.
3009 btrfs_remove_ordered_extent(inode, ordered_extent);
3011 /* for snapshot-aware defrag */
3014 free_sa_defrag_extent(new);
3015 atomic_dec(&root->fs_info->defrag_running);
3017 relink_file_extents(new);
3022 btrfs_put_ordered_extent(ordered_extent);
3023 /* once for the tree */
3024 btrfs_put_ordered_extent(ordered_extent);
3029 static void finish_ordered_fn(struct btrfs_work *work)
3031 struct btrfs_ordered_extent *ordered_extent;
3032 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3033 btrfs_finish_ordered_io(ordered_extent);
3036 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3037 struct extent_state *state, int uptodate)
3039 struct inode *inode = page->mapping->host;
3040 struct btrfs_root *root = BTRFS_I(inode)->root;
3041 struct btrfs_ordered_extent *ordered_extent = NULL;
3042 struct btrfs_workqueue *wq;
3043 btrfs_work_func_t func;
3045 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3047 ClearPagePrivate2(page);
3048 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3049 end - start + 1, uptodate))
3052 if (btrfs_is_free_space_inode(inode)) {
3053 wq = root->fs_info->endio_freespace_worker;
3054 func = btrfs_freespace_write_helper;
3056 wq = root->fs_info->endio_write_workers;
3057 func = btrfs_endio_write_helper;
3060 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3062 btrfs_queue_work(wq, &ordered_extent->work);
3067 static int __readpage_endio_check(struct inode *inode,
3068 struct btrfs_io_bio *io_bio,
3069 int icsum, struct page *page,
3070 int pgoff, u64 start, size_t len)
3076 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3078 kaddr = kmap_atomic(page);
3079 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3080 btrfs_csum_final(csum, (char *)&csum);
3081 if (csum != csum_expected)
3084 kunmap_atomic(kaddr);
3087 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3088 "csum failed ino %llu off %llu csum %u expected csum %u",
3089 btrfs_ino(inode), start, csum, csum_expected);
3090 memset(kaddr + pgoff, 1, len);
3091 flush_dcache_page(page);
3092 kunmap_atomic(kaddr);
3093 if (csum_expected == 0)
3099 * when reads are done, we need to check csums to verify the data is correct
3100 * if there's a match, we allow the bio to finish. If not, the code in
3101 * extent_io.c will try to find good copies for us.
3103 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3104 u64 phy_offset, struct page *page,
3105 u64 start, u64 end, int mirror)
3107 size_t offset = start - page_offset(page);
3108 struct inode *inode = page->mapping->host;
3109 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3110 struct btrfs_root *root = BTRFS_I(inode)->root;
3112 if (PageChecked(page)) {
3113 ClearPageChecked(page);
3117 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3120 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3121 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3122 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3126 phy_offset >>= inode->i_sb->s_blocksize_bits;
3127 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3128 start, (size_t)(end - start + 1));
3131 void btrfs_add_delayed_iput(struct inode *inode)
3133 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3134 struct btrfs_inode *binode = BTRFS_I(inode);
3136 if (atomic_add_unless(&inode->i_count, -1, 1))
3139 spin_lock(&fs_info->delayed_iput_lock);
3140 if (binode->delayed_iput_count == 0) {
3141 ASSERT(list_empty(&binode->delayed_iput));
3142 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3144 binode->delayed_iput_count++;
3146 spin_unlock(&fs_info->delayed_iput_lock);
3149 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3151 struct btrfs_fs_info *fs_info = root->fs_info;
3153 spin_lock(&fs_info->delayed_iput_lock);
3154 while (!list_empty(&fs_info->delayed_iputs)) {
3155 struct btrfs_inode *inode;
3157 inode = list_first_entry(&fs_info->delayed_iputs,
3158 struct btrfs_inode, delayed_iput);
3159 if (inode->delayed_iput_count) {
3160 inode->delayed_iput_count--;
3161 list_move_tail(&inode->delayed_iput,
3162 &fs_info->delayed_iputs);
3164 list_del_init(&inode->delayed_iput);
3166 spin_unlock(&fs_info->delayed_iput_lock);
3167 iput(&inode->vfs_inode);
3168 spin_lock(&fs_info->delayed_iput_lock);
3170 spin_unlock(&fs_info->delayed_iput_lock);
3174 * This is called in transaction commit time. If there are no orphan
3175 * files in the subvolume, it removes orphan item and frees block_rsv
3178 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3179 struct btrfs_root *root)
3181 struct btrfs_block_rsv *block_rsv;
3184 if (atomic_read(&root->orphan_inodes) ||
3185 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3188 spin_lock(&root->orphan_lock);
3189 if (atomic_read(&root->orphan_inodes)) {
3190 spin_unlock(&root->orphan_lock);
3194 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3195 spin_unlock(&root->orphan_lock);
3199 block_rsv = root->orphan_block_rsv;
3200 root->orphan_block_rsv = NULL;
3201 spin_unlock(&root->orphan_lock);
3203 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3204 btrfs_root_refs(&root->root_item) > 0) {
3205 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3206 root->root_key.objectid);
3208 btrfs_abort_transaction(trans, root, ret);
3210 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3215 WARN_ON(block_rsv->size > 0);
3216 btrfs_free_block_rsv(root, block_rsv);
3221 * This creates an orphan entry for the given inode in case something goes
3222 * wrong in the middle of an unlink/truncate.
3224 * NOTE: caller of this function should reserve 5 units of metadata for
3227 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3229 struct btrfs_root *root = BTRFS_I(inode)->root;
3230 struct btrfs_block_rsv *block_rsv = NULL;
3235 if (!root->orphan_block_rsv) {
3236 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3241 spin_lock(&root->orphan_lock);
3242 if (!root->orphan_block_rsv) {
3243 root->orphan_block_rsv = block_rsv;
3244 } else if (block_rsv) {
3245 btrfs_free_block_rsv(root, block_rsv);
3249 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3250 &BTRFS_I(inode)->runtime_flags)) {
3253 * For proper ENOSPC handling, we should do orphan
3254 * cleanup when mounting. But this introduces backward
3255 * compatibility issue.
3257 if (!xchg(&root->orphan_item_inserted, 1))
3263 atomic_inc(&root->orphan_inodes);
3266 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3267 &BTRFS_I(inode)->runtime_flags))
3269 spin_unlock(&root->orphan_lock);
3271 /* grab metadata reservation from transaction handle */
3273 ret = btrfs_orphan_reserve_metadata(trans, inode);
3276 atomic_dec(&root->orphan_inodes);
3277 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3278 &BTRFS_I(inode)->runtime_flags);
3280 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3281 &BTRFS_I(inode)->runtime_flags);
3286 /* insert an orphan item to track this unlinked/truncated file */
3288 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3290 atomic_dec(&root->orphan_inodes);
3292 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3293 &BTRFS_I(inode)->runtime_flags);
3294 btrfs_orphan_release_metadata(inode);
3296 if (ret != -EEXIST) {
3297 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3298 &BTRFS_I(inode)->runtime_flags);
3299 btrfs_abort_transaction(trans, root, ret);
3306 /* insert an orphan item to track subvolume contains orphan files */
3308 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3309 root->root_key.objectid);
3310 if (ret && ret != -EEXIST) {
3311 btrfs_abort_transaction(trans, root, ret);
3319 * We have done the truncate/delete so we can go ahead and remove the orphan
3320 * item for this particular inode.
3322 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3323 struct inode *inode)
3325 struct btrfs_root *root = BTRFS_I(inode)->root;
3326 int delete_item = 0;
3327 int release_rsv = 0;
3330 spin_lock(&root->orphan_lock);
3331 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3332 &BTRFS_I(inode)->runtime_flags))
3335 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3336 &BTRFS_I(inode)->runtime_flags))
3338 spin_unlock(&root->orphan_lock);
3341 atomic_dec(&root->orphan_inodes);
3343 ret = btrfs_del_orphan_item(trans, root,
3348 btrfs_orphan_release_metadata(inode);
3354 * this cleans up any orphans that may be left on the list from the last use
3357 int btrfs_orphan_cleanup(struct btrfs_root *root)
3359 struct btrfs_path *path;
3360 struct extent_buffer *leaf;
3361 struct btrfs_key key, found_key;
3362 struct btrfs_trans_handle *trans;
3363 struct inode *inode;
3364 u64 last_objectid = 0;
3365 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3367 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3370 path = btrfs_alloc_path();
3375 path->reada = READA_BACK;
3377 key.objectid = BTRFS_ORPHAN_OBJECTID;
3378 key.type = BTRFS_ORPHAN_ITEM_KEY;
3379 key.offset = (u64)-1;
3382 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3387 * if ret == 0 means we found what we were searching for, which
3388 * is weird, but possible, so only screw with path if we didn't
3389 * find the key and see if we have stuff that matches
3393 if (path->slots[0] == 0)
3398 /* pull out the item */
3399 leaf = path->nodes[0];
3400 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3402 /* make sure the item matches what we want */
3403 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3405 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3408 /* release the path since we're done with it */
3409 btrfs_release_path(path);
3412 * this is where we are basically btrfs_lookup, without the
3413 * crossing root thing. we store the inode number in the
3414 * offset of the orphan item.
3417 if (found_key.offset == last_objectid) {
3418 btrfs_err(root->fs_info,
3419 "Error removing orphan entry, stopping orphan cleanup");
3424 last_objectid = found_key.offset;
3426 found_key.objectid = found_key.offset;
3427 found_key.type = BTRFS_INODE_ITEM_KEY;
3428 found_key.offset = 0;
3429 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3430 ret = PTR_ERR_OR_ZERO(inode);
3431 if (ret && ret != -ENOENT)
3434 if (ret == -ENOENT && root == root->fs_info->tree_root) {
3435 struct btrfs_root *dead_root;
3436 struct btrfs_fs_info *fs_info = root->fs_info;
3437 int is_dead_root = 0;
3440 * this is an orphan in the tree root. Currently these
3441 * could come from 2 sources:
3442 * a) a snapshot deletion in progress
3443 * b) a free space cache inode
3444 * We need to distinguish those two, as the snapshot
3445 * orphan must not get deleted.
3446 * find_dead_roots already ran before us, so if this
3447 * is a snapshot deletion, we should find the root
3448 * in the dead_roots list
3450 spin_lock(&fs_info->trans_lock);
3451 list_for_each_entry(dead_root, &fs_info->dead_roots,
3453 if (dead_root->root_key.objectid ==
3454 found_key.objectid) {
3459 spin_unlock(&fs_info->trans_lock);
3461 /* prevent this orphan from being found again */
3462 key.offset = found_key.objectid - 1;
3467 * Inode is already gone but the orphan item is still there,
3468 * kill the orphan item.
3470 if (ret == -ENOENT) {
3471 trans = btrfs_start_transaction(root, 1);
3472 if (IS_ERR(trans)) {
3473 ret = PTR_ERR(trans);
3476 btrfs_debug(root->fs_info, "auto deleting %Lu",
3477 found_key.objectid);
3478 ret = btrfs_del_orphan_item(trans, root,
3479 found_key.objectid);
3480 btrfs_end_transaction(trans, root);
3487 * add this inode to the orphan list so btrfs_orphan_del does
3488 * the proper thing when we hit it
3490 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3491 &BTRFS_I(inode)->runtime_flags);
3492 atomic_inc(&root->orphan_inodes);
3494 /* if we have links, this was a truncate, lets do that */
3495 if (inode->i_nlink) {
3496 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3502 /* 1 for the orphan item deletion. */
3503 trans = btrfs_start_transaction(root, 1);
3504 if (IS_ERR(trans)) {
3506 ret = PTR_ERR(trans);
3509 ret = btrfs_orphan_add(trans, inode);
3510 btrfs_end_transaction(trans, root);
3516 ret = btrfs_truncate(inode);
3518 btrfs_orphan_del(NULL, inode);
3523 /* this will do delete_inode and everything for us */
3528 /* release the path since we're done with it */
3529 btrfs_release_path(path);
3531 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3533 if (root->orphan_block_rsv)
3534 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3537 if (root->orphan_block_rsv ||
3538 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3539 trans = btrfs_join_transaction(root);
3541 btrfs_end_transaction(trans, root);
3545 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3547 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3551 btrfs_err(root->fs_info,
3552 "could not do orphan cleanup %d", ret);
3553 btrfs_free_path(path);
3558 * very simple check to peek ahead in the leaf looking for xattrs. If we
3559 * don't find any xattrs, we know there can't be any acls.
3561 * slot is the slot the inode is in, objectid is the objectid of the inode
3563 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3564 int slot, u64 objectid,
3565 int *first_xattr_slot)
3567 u32 nritems = btrfs_header_nritems(leaf);
3568 struct btrfs_key found_key;
3569 static u64 xattr_access = 0;
3570 static u64 xattr_default = 0;
3573 if (!xattr_access) {
3574 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3575 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3576 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3577 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3581 *first_xattr_slot = -1;
3582 while (slot < nritems) {
3583 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3585 /* we found a different objectid, there must not be acls */
3586 if (found_key.objectid != objectid)
3589 /* we found an xattr, assume we've got an acl */
3590 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3591 if (*first_xattr_slot == -1)
3592 *first_xattr_slot = slot;
3593 if (found_key.offset == xattr_access ||
3594 found_key.offset == xattr_default)
3599 * we found a key greater than an xattr key, there can't
3600 * be any acls later on
3602 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3609 * it goes inode, inode backrefs, xattrs, extents,
3610 * so if there are a ton of hard links to an inode there can
3611 * be a lot of backrefs. Don't waste time searching too hard,
3612 * this is just an optimization
3617 /* we hit the end of the leaf before we found an xattr or
3618 * something larger than an xattr. We have to assume the inode
3621 if (*first_xattr_slot == -1)
3622 *first_xattr_slot = slot;
3627 * read an inode from the btree into the in-memory inode
3629 static int btrfs_read_locked_inode(struct inode *inode)
3631 struct btrfs_path *path;
3632 struct extent_buffer *leaf;
3633 struct btrfs_inode_item *inode_item;
3634 struct btrfs_root *root = BTRFS_I(inode)->root;
3635 struct btrfs_key location;
3640 bool filled = false;
3641 int first_xattr_slot;
3643 ret = btrfs_fill_inode(inode, &rdev);
3647 path = btrfs_alloc_path();
3653 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3655 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3662 leaf = path->nodes[0];
3667 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3668 struct btrfs_inode_item);
3669 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3670 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3671 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3672 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3673 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3675 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3676 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3678 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3679 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3681 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3682 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3684 BTRFS_I(inode)->i_otime.tv_sec =
3685 btrfs_timespec_sec(leaf, &inode_item->otime);
3686 BTRFS_I(inode)->i_otime.tv_nsec =
3687 btrfs_timespec_nsec(leaf, &inode_item->otime);
3689 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3690 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3691 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3693 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3694 inode->i_generation = BTRFS_I(inode)->generation;
3696 rdev = btrfs_inode_rdev(leaf, inode_item);
3698 BTRFS_I(inode)->index_cnt = (u64)-1;
3699 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3703 * If we were modified in the current generation and evicted from memory
3704 * and then re-read we need to do a full sync since we don't have any
3705 * idea about which extents were modified before we were evicted from
3708 * This is required for both inode re-read from disk and delayed inode
3709 * in delayed_nodes_tree.
3711 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3712 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3713 &BTRFS_I(inode)->runtime_flags);
3716 * We don't persist the id of the transaction where an unlink operation
3717 * against the inode was last made. So here we assume the inode might
3718 * have been evicted, and therefore the exact value of last_unlink_trans
3719 * lost, and set it to last_trans to avoid metadata inconsistencies
3720 * between the inode and its parent if the inode is fsync'ed and the log
3721 * replayed. For example, in the scenario:
3724 * ln mydir/foo mydir/bar
3727 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3728 * xfs_io -c fsync mydir/foo
3730 * mount fs, triggers fsync log replay
3732 * We must make sure that when we fsync our inode foo we also log its
3733 * parent inode, otherwise after log replay the parent still has the
3734 * dentry with the "bar" name but our inode foo has a link count of 1
3735 * and doesn't have an inode ref with the name "bar" anymore.
3737 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3738 * but it guarantees correctness at the expense of occasional full
3739 * transaction commits on fsync if our inode is a directory, or if our
3740 * inode is not a directory, logging its parent unnecessarily.
3742 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3745 if (inode->i_nlink != 1 ||
3746 path->slots[0] >= btrfs_header_nritems(leaf))
3749 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3750 if (location.objectid != btrfs_ino(inode))
3753 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3754 if (location.type == BTRFS_INODE_REF_KEY) {
3755 struct btrfs_inode_ref *ref;
3757 ref = (struct btrfs_inode_ref *)ptr;
3758 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3759 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3760 struct btrfs_inode_extref *extref;
3762 extref = (struct btrfs_inode_extref *)ptr;
3763 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3768 * try to precache a NULL acl entry for files that don't have
3769 * any xattrs or acls
3771 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3772 btrfs_ino(inode), &first_xattr_slot);
3773 if (first_xattr_slot != -1) {
3774 path->slots[0] = first_xattr_slot;
3775 ret = btrfs_load_inode_props(inode, path);
3777 btrfs_err(root->fs_info,
3778 "error loading props for ino %llu (root %llu): %d",
3780 root->root_key.objectid, ret);
3782 btrfs_free_path(path);
3785 cache_no_acl(inode);
3787 switch (inode->i_mode & S_IFMT) {
3789 inode->i_mapping->a_ops = &btrfs_aops;
3790 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3791 inode->i_fop = &btrfs_file_operations;
3792 inode->i_op = &btrfs_file_inode_operations;
3795 inode->i_fop = &btrfs_dir_file_operations;
3796 if (root == root->fs_info->tree_root)
3797 inode->i_op = &btrfs_dir_ro_inode_operations;
3799 inode->i_op = &btrfs_dir_inode_operations;
3802 inode->i_op = &btrfs_symlink_inode_operations;
3803 inode_nohighmem(inode);
3804 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3807 inode->i_op = &btrfs_special_inode_operations;
3808 init_special_inode(inode, inode->i_mode, rdev);
3812 btrfs_update_iflags(inode);
3816 btrfs_free_path(path);
3817 make_bad_inode(inode);
3822 * given a leaf and an inode, copy the inode fields into the leaf
3824 static void fill_inode_item(struct btrfs_trans_handle *trans,
3825 struct extent_buffer *leaf,
3826 struct btrfs_inode_item *item,
3827 struct inode *inode)
3829 struct btrfs_map_token token;
3831 btrfs_init_map_token(&token);
3833 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3834 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3835 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3837 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3838 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3840 btrfs_set_token_timespec_sec(leaf, &item->atime,
3841 inode->i_atime.tv_sec, &token);
3842 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3843 inode->i_atime.tv_nsec, &token);
3845 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3846 inode->i_mtime.tv_sec, &token);
3847 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3848 inode->i_mtime.tv_nsec, &token);
3850 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3851 inode->i_ctime.tv_sec, &token);
3852 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3853 inode->i_ctime.tv_nsec, &token);
3855 btrfs_set_token_timespec_sec(leaf, &item->otime,
3856 BTRFS_I(inode)->i_otime.tv_sec, &token);
3857 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3858 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3860 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3862 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3864 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3865 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3866 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3867 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3868 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3872 * copy everything in the in-memory inode into the btree.
3874 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3875 struct btrfs_root *root, struct inode *inode)
3877 struct btrfs_inode_item *inode_item;
3878 struct btrfs_path *path;
3879 struct extent_buffer *leaf;
3882 path = btrfs_alloc_path();
3886 path->leave_spinning = 1;
3887 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3895 leaf = path->nodes[0];
3896 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3897 struct btrfs_inode_item);
3899 fill_inode_item(trans, leaf, inode_item, inode);
3900 btrfs_mark_buffer_dirty(leaf);
3901 btrfs_set_inode_last_trans(trans, inode);
3904 btrfs_free_path(path);
3909 * copy everything in the in-memory inode into the btree.
3911 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3912 struct btrfs_root *root, struct inode *inode)
3917 * If the inode is a free space inode, we can deadlock during commit
3918 * if we put it into the delayed code.
3920 * The data relocation inode should also be directly updated
3923 if (!btrfs_is_free_space_inode(inode)
3924 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3925 && !root->fs_info->log_root_recovering) {
3926 btrfs_update_root_times(trans, root);
3928 ret = btrfs_delayed_update_inode(trans, root, inode);
3930 btrfs_set_inode_last_trans(trans, inode);
3934 return btrfs_update_inode_item(trans, root, inode);
3937 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3938 struct btrfs_root *root,
3939 struct inode *inode)
3943 ret = btrfs_update_inode(trans, root, inode);
3945 return btrfs_update_inode_item(trans, root, inode);
3950 * unlink helper that gets used here in inode.c and in the tree logging
3951 * recovery code. It remove a link in a directory with a given name, and
3952 * also drops the back refs in the inode to the directory
3954 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3955 struct btrfs_root *root,
3956 struct inode *dir, struct inode *inode,
3957 const char *name, int name_len)
3959 struct btrfs_path *path;
3961 struct extent_buffer *leaf;
3962 struct btrfs_dir_item *di;
3963 struct btrfs_key key;
3965 u64 ino = btrfs_ino(inode);
3966 u64 dir_ino = btrfs_ino(dir);
3968 path = btrfs_alloc_path();
3974 path->leave_spinning = 1;
3975 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3976 name, name_len, -1);
3985 leaf = path->nodes[0];
3986 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3987 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3990 btrfs_release_path(path);
3993 * If we don't have dir index, we have to get it by looking up
3994 * the inode ref, since we get the inode ref, remove it directly,
3995 * it is unnecessary to do delayed deletion.
3997 * But if we have dir index, needn't search inode ref to get it.
3998 * Since the inode ref is close to the inode item, it is better
3999 * that we delay to delete it, and just do this deletion when
4000 * we update the inode item.
4002 if (BTRFS_I(inode)->dir_index) {
4003 ret = btrfs_delayed_delete_inode_ref(inode);
4005 index = BTRFS_I(inode)->dir_index;
4010 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4013 btrfs_info(root->fs_info,
4014 "failed to delete reference to %.*s, inode %llu parent %llu",
4015 name_len, name, ino, dir_ino);
4016 btrfs_abort_transaction(trans, root, ret);
4020 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4022 btrfs_abort_transaction(trans, root, ret);
4026 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4028 if (ret != 0 && ret != -ENOENT) {
4029 btrfs_abort_transaction(trans, root, ret);
4033 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4038 btrfs_abort_transaction(trans, root, ret);
4040 btrfs_free_path(path);
4044 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4045 inode_inc_iversion(inode);
4046 inode_inc_iversion(dir);
4047 inode->i_ctime = dir->i_mtime =
4048 dir->i_ctime = current_fs_time(inode->i_sb);
4049 ret = btrfs_update_inode(trans, root, dir);
4054 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4055 struct btrfs_root *root,
4056 struct inode *dir, struct inode *inode,
4057 const char *name, int name_len)
4060 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4063 ret = btrfs_update_inode(trans, root, inode);
4069 * helper to start transaction for unlink and rmdir.
4071 * unlink and rmdir are special in btrfs, they do not always free space, so
4072 * if we cannot make our reservations the normal way try and see if there is
4073 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4074 * allow the unlink to occur.
4076 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4078 struct btrfs_root *root = BTRFS_I(dir)->root;
4081 * 1 for the possible orphan item
4082 * 1 for the dir item
4083 * 1 for the dir index
4084 * 1 for the inode ref
4087 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4090 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4092 struct btrfs_root *root = BTRFS_I(dir)->root;
4093 struct btrfs_trans_handle *trans;
4094 struct inode *inode = d_inode(dentry);
4097 trans = __unlink_start_trans(dir);
4099 return PTR_ERR(trans);
4101 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4103 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4104 dentry->d_name.name, dentry->d_name.len);
4108 if (inode->i_nlink == 0) {
4109 ret = btrfs_orphan_add(trans, inode);
4115 btrfs_end_transaction(trans, root);
4116 btrfs_btree_balance_dirty(root);
4120 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4121 struct btrfs_root *root,
4122 struct inode *dir, u64 objectid,
4123 const char *name, int name_len)
4125 struct btrfs_path *path;
4126 struct extent_buffer *leaf;
4127 struct btrfs_dir_item *di;
4128 struct btrfs_key key;
4131 u64 dir_ino = btrfs_ino(dir);
4133 path = btrfs_alloc_path();
4137 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4138 name, name_len, -1);
4139 if (IS_ERR_OR_NULL(di)) {
4147 leaf = path->nodes[0];
4148 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4149 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4150 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4152 btrfs_abort_transaction(trans, root, ret);
4155 btrfs_release_path(path);
4157 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4158 objectid, root->root_key.objectid,
4159 dir_ino, &index, name, name_len);
4161 if (ret != -ENOENT) {
4162 btrfs_abort_transaction(trans, root, ret);
4165 di = btrfs_search_dir_index_item(root, path, dir_ino,
4167 if (IS_ERR_OR_NULL(di)) {
4172 btrfs_abort_transaction(trans, root, ret);
4176 leaf = path->nodes[0];
4177 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4178 btrfs_release_path(path);
4181 btrfs_release_path(path);
4183 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4185 btrfs_abort_transaction(trans, root, ret);
4189 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4190 inode_inc_iversion(dir);
4191 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4192 ret = btrfs_update_inode_fallback(trans, root, dir);
4194 btrfs_abort_transaction(trans, root, ret);
4196 btrfs_free_path(path);
4200 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4202 struct inode *inode = d_inode(dentry);
4204 struct btrfs_root *root = BTRFS_I(dir)->root;
4205 struct btrfs_trans_handle *trans;
4206 u64 last_unlink_trans;
4208 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4210 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4213 trans = __unlink_start_trans(dir);
4215 return PTR_ERR(trans);
4217 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4218 err = btrfs_unlink_subvol(trans, root, dir,
4219 BTRFS_I(inode)->location.objectid,
4220 dentry->d_name.name,
4221 dentry->d_name.len);
4225 err = btrfs_orphan_add(trans, inode);
4229 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4231 /* now the directory is empty */
4232 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4233 dentry->d_name.name, dentry->d_name.len);
4235 btrfs_i_size_write(inode, 0);
4237 * Propagate the last_unlink_trans value of the deleted dir to
4238 * its parent directory. This is to prevent an unrecoverable
4239 * log tree in the case we do something like this:
4241 * 2) create snapshot under dir foo
4242 * 3) delete the snapshot
4245 * 6) fsync foo or some file inside foo
4247 if (last_unlink_trans >= trans->transid)
4248 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4251 btrfs_end_transaction(trans, root);
4252 btrfs_btree_balance_dirty(root);
4257 static int truncate_space_check(struct btrfs_trans_handle *trans,
4258 struct btrfs_root *root,
4264 * This is only used to apply pressure to the enospc system, we don't
4265 * intend to use this reservation at all.
4267 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4268 bytes_deleted *= root->nodesize;
4269 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4270 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4272 trace_btrfs_space_reservation(root->fs_info, "transaction",
4275 trans->bytes_reserved += bytes_deleted;
4281 static int truncate_inline_extent(struct inode *inode,
4282 struct btrfs_path *path,
4283 struct btrfs_key *found_key,
4287 struct extent_buffer *leaf = path->nodes[0];
4288 int slot = path->slots[0];
4289 struct btrfs_file_extent_item *fi;
4290 u32 size = (u32)(new_size - found_key->offset);
4291 struct btrfs_root *root = BTRFS_I(inode)->root;
4293 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4295 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4296 loff_t offset = new_size;
4297 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4300 * Zero out the remaining of the last page of our inline extent,
4301 * instead of directly truncating our inline extent here - that
4302 * would be much more complex (decompressing all the data, then
4303 * compressing the truncated data, which might be bigger than
4304 * the size of the inline extent, resize the extent, etc).
4305 * We release the path because to get the page we might need to
4306 * read the extent item from disk (data not in the page cache).
4308 btrfs_release_path(path);
4309 return btrfs_truncate_block(inode, offset, page_end - offset,
4313 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4314 size = btrfs_file_extent_calc_inline_size(size);
4315 btrfs_truncate_item(root, path, size, 1);
4317 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4318 inode_sub_bytes(inode, item_end + 1 - new_size);
4324 * this can truncate away extent items, csum items and directory items.
4325 * It starts at a high offset and removes keys until it can't find
4326 * any higher than new_size
4328 * csum items that cross the new i_size are truncated to the new size
4331 * min_type is the minimum key type to truncate down to. If set to 0, this
4332 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4334 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4335 struct btrfs_root *root,
4336 struct inode *inode,
4337 u64 new_size, u32 min_type)
4339 struct btrfs_path *path;
4340 struct extent_buffer *leaf;
4341 struct btrfs_file_extent_item *fi;
4342 struct btrfs_key key;
4343 struct btrfs_key found_key;
4344 u64 extent_start = 0;
4345 u64 extent_num_bytes = 0;
4346 u64 extent_offset = 0;
4348 u64 last_size = new_size;
4349 u32 found_type = (u8)-1;
4352 int pending_del_nr = 0;
4353 int pending_del_slot = 0;
4354 int extent_type = -1;
4357 u64 ino = btrfs_ino(inode);
4358 u64 bytes_deleted = 0;
4360 bool should_throttle = 0;
4361 bool should_end = 0;
4363 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4366 * for non-free space inodes and ref cows, we want to back off from
4369 if (!btrfs_is_free_space_inode(inode) &&
4370 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4373 path = btrfs_alloc_path();
4376 path->reada = READA_BACK;
4379 * We want to drop from the next block forward in case this new size is
4380 * not block aligned since we will be keeping the last block of the
4381 * extent just the way it is.
4383 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4384 root == root->fs_info->tree_root)
4385 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4386 root->sectorsize), (u64)-1, 0);
4389 * This function is also used to drop the items in the log tree before
4390 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4391 * it is used to drop the loged items. So we shouldn't kill the delayed
4394 if (min_type == 0 && root == BTRFS_I(inode)->root)
4395 btrfs_kill_delayed_inode_items(inode);
4398 key.offset = (u64)-1;
4403 * with a 16K leaf size and 128MB extents, you can actually queue
4404 * up a huge file in a single leaf. Most of the time that
4405 * bytes_deleted is > 0, it will be huge by the time we get here
4407 if (be_nice && bytes_deleted > SZ_32M) {
4408 if (btrfs_should_end_transaction(trans, root)) {
4415 path->leave_spinning = 1;
4416 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4423 /* there are no items in the tree for us to truncate, we're
4426 if (path->slots[0] == 0)
4433 leaf = path->nodes[0];
4434 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4435 found_type = found_key.type;
4437 if (found_key.objectid != ino)
4440 if (found_type < min_type)
4443 item_end = found_key.offset;
4444 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4445 fi = btrfs_item_ptr(leaf, path->slots[0],
4446 struct btrfs_file_extent_item);
4447 extent_type = btrfs_file_extent_type(leaf, fi);
4448 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4450 btrfs_file_extent_num_bytes(leaf, fi);
4451 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4452 item_end += btrfs_file_extent_inline_len(leaf,
4453 path->slots[0], fi);
4457 if (found_type > min_type) {
4460 if (item_end < new_size)
4462 if (found_key.offset >= new_size)
4468 /* FIXME, shrink the extent if the ref count is only 1 */
4469 if (found_type != BTRFS_EXTENT_DATA_KEY)
4473 last_size = found_key.offset;
4475 last_size = new_size;
4477 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4479 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4481 u64 orig_num_bytes =
4482 btrfs_file_extent_num_bytes(leaf, fi);
4483 extent_num_bytes = ALIGN(new_size -
4486 btrfs_set_file_extent_num_bytes(leaf, fi,
4488 num_dec = (orig_num_bytes -
4490 if (test_bit(BTRFS_ROOT_REF_COWS,
4493 inode_sub_bytes(inode, num_dec);
4494 btrfs_mark_buffer_dirty(leaf);
4497 btrfs_file_extent_disk_num_bytes(leaf,
4499 extent_offset = found_key.offset -
4500 btrfs_file_extent_offset(leaf, fi);
4502 /* FIXME blocksize != 4096 */
4503 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4504 if (extent_start != 0) {
4506 if (test_bit(BTRFS_ROOT_REF_COWS,
4508 inode_sub_bytes(inode, num_dec);
4511 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4513 * we can't truncate inline items that have had
4517 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4518 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4521 * Need to release path in order to truncate a
4522 * compressed extent. So delete any accumulated
4523 * extent items so far.
4525 if (btrfs_file_extent_compression(leaf, fi) !=
4526 BTRFS_COMPRESS_NONE && pending_del_nr) {
4527 err = btrfs_del_items(trans, root, path,
4531 btrfs_abort_transaction(trans,
4539 err = truncate_inline_extent(inode, path,
4544 btrfs_abort_transaction(trans,
4548 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4550 inode_sub_bytes(inode, item_end + 1 - new_size);
4555 if (!pending_del_nr) {
4556 /* no pending yet, add ourselves */
4557 pending_del_slot = path->slots[0];
4559 } else if (pending_del_nr &&
4560 path->slots[0] + 1 == pending_del_slot) {
4561 /* hop on the pending chunk */
4563 pending_del_slot = path->slots[0];
4570 should_throttle = 0;
4573 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4574 root == root->fs_info->tree_root)) {
4575 btrfs_set_path_blocking(path);
4576 bytes_deleted += extent_num_bytes;
4577 ret = btrfs_free_extent(trans, root, extent_start,
4578 extent_num_bytes, 0,
4579 btrfs_header_owner(leaf),
4580 ino, extent_offset);
4582 if (btrfs_should_throttle_delayed_refs(trans, root))
4583 btrfs_async_run_delayed_refs(root,
4585 trans->delayed_ref_updates * 2, 0);
4587 if (truncate_space_check(trans, root,
4588 extent_num_bytes)) {
4591 if (btrfs_should_throttle_delayed_refs(trans,
4593 should_throttle = 1;
4598 if (found_type == BTRFS_INODE_ITEM_KEY)
4601 if (path->slots[0] == 0 ||
4602 path->slots[0] != pending_del_slot ||
4603 should_throttle || should_end) {
4604 if (pending_del_nr) {
4605 ret = btrfs_del_items(trans, root, path,
4609 btrfs_abort_transaction(trans,
4615 btrfs_release_path(path);
4616 if (should_throttle) {
4617 unsigned long updates = trans->delayed_ref_updates;
4619 trans->delayed_ref_updates = 0;
4620 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4626 * if we failed to refill our space rsv, bail out
4627 * and let the transaction restart
4639 if (pending_del_nr) {
4640 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4643 btrfs_abort_transaction(trans, root, ret);
4646 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4647 btrfs_ordered_update_i_size(inode, last_size, NULL);
4649 btrfs_free_path(path);
4651 if (be_nice && bytes_deleted > SZ_32M) {
4652 unsigned long updates = trans->delayed_ref_updates;
4654 trans->delayed_ref_updates = 0;
4655 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4664 * btrfs_truncate_block - read, zero a chunk and write a block
4665 * @inode - inode that we're zeroing
4666 * @from - the offset to start zeroing
4667 * @len - the length to zero, 0 to zero the entire range respective to the
4669 * @front - zero up to the offset instead of from the offset on
4671 * This will find the block for the "from" offset and cow the block and zero the
4672 * part we want to zero. This is used with truncate and hole punching.
4674 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4677 struct address_space *mapping = inode->i_mapping;
4678 struct btrfs_root *root = BTRFS_I(inode)->root;
4679 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4680 struct btrfs_ordered_extent *ordered;
4681 struct extent_state *cached_state = NULL;
4683 u32 blocksize = root->sectorsize;
4684 pgoff_t index = from >> PAGE_SHIFT;
4685 unsigned offset = from & (blocksize - 1);
4687 gfp_t mask = btrfs_alloc_write_mask(mapping);
4692 if ((offset & (blocksize - 1)) == 0 &&
4693 (!len || ((len & (blocksize - 1)) == 0)))
4696 ret = btrfs_delalloc_reserve_space(inode,
4697 round_down(from, blocksize), blocksize);
4702 page = find_or_create_page(mapping, index, mask);
4704 btrfs_delalloc_release_space(inode,
4705 round_down(from, blocksize),
4711 block_start = round_down(from, blocksize);
4712 block_end = block_start + blocksize - 1;
4714 if (!PageUptodate(page)) {
4715 ret = btrfs_readpage(NULL, page);
4717 if (page->mapping != mapping) {
4722 if (!PageUptodate(page)) {
4727 wait_on_page_writeback(page);
4729 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4730 set_page_extent_mapped(page);
4732 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4734 unlock_extent_cached(io_tree, block_start, block_end,
4735 &cached_state, GFP_NOFS);
4738 btrfs_start_ordered_extent(inode, ordered, 1);
4739 btrfs_put_ordered_extent(ordered);
4743 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4744 EXTENT_DIRTY | EXTENT_DELALLOC |
4745 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4746 0, 0, &cached_state, GFP_NOFS);
4748 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4751 unlock_extent_cached(io_tree, block_start, block_end,
4752 &cached_state, GFP_NOFS);
4756 if (offset != blocksize) {
4758 len = blocksize - offset;
4761 memset(kaddr + (block_start - page_offset(page)),
4764 memset(kaddr + (block_start - page_offset(page)) + offset,
4766 flush_dcache_page(page);
4769 ClearPageChecked(page);
4770 set_page_dirty(page);
4771 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4776 btrfs_delalloc_release_space(inode, block_start,
4784 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4785 u64 offset, u64 len)
4787 struct btrfs_trans_handle *trans;
4791 * Still need to make sure the inode looks like it's been updated so
4792 * that any holes get logged if we fsync.
4794 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4795 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4796 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4797 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4802 * 1 - for the one we're dropping
4803 * 1 - for the one we're adding
4804 * 1 - for updating the inode.
4806 trans = btrfs_start_transaction(root, 3);
4808 return PTR_ERR(trans);
4810 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4812 btrfs_abort_transaction(trans, root, ret);
4813 btrfs_end_transaction(trans, root);
4817 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4818 0, 0, len, 0, len, 0, 0, 0);
4820 btrfs_abort_transaction(trans, root, ret);
4822 btrfs_update_inode(trans, root, inode);
4823 btrfs_end_transaction(trans, root);
4828 * This function puts in dummy file extents for the area we're creating a hole
4829 * for. So if we are truncating this file to a larger size we need to insert
4830 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4831 * the range between oldsize and size
4833 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4835 struct btrfs_root *root = BTRFS_I(inode)->root;
4836 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4837 struct extent_map *em = NULL;
4838 struct extent_state *cached_state = NULL;
4839 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4840 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4841 u64 block_end = ALIGN(size, root->sectorsize);
4848 * If our size started in the middle of a block we need to zero out the
4849 * rest of the block before we expand the i_size, otherwise we could
4850 * expose stale data.
4852 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4856 if (size <= hole_start)
4860 struct btrfs_ordered_extent *ordered;
4862 lock_extent_bits(io_tree, hole_start, block_end - 1,
4864 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4865 block_end - hole_start);
4868 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4869 &cached_state, GFP_NOFS);
4870 btrfs_start_ordered_extent(inode, ordered, 1);
4871 btrfs_put_ordered_extent(ordered);
4874 cur_offset = hole_start;
4876 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4877 block_end - cur_offset, 0);
4883 last_byte = min(extent_map_end(em), block_end);
4884 last_byte = ALIGN(last_byte , root->sectorsize);
4885 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4886 struct extent_map *hole_em;
4887 hole_size = last_byte - cur_offset;
4889 err = maybe_insert_hole(root, inode, cur_offset,
4893 btrfs_drop_extent_cache(inode, cur_offset,
4894 cur_offset + hole_size - 1, 0);
4895 hole_em = alloc_extent_map();
4897 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4898 &BTRFS_I(inode)->runtime_flags);
4901 hole_em->start = cur_offset;
4902 hole_em->len = hole_size;
4903 hole_em->orig_start = cur_offset;
4905 hole_em->block_start = EXTENT_MAP_HOLE;
4906 hole_em->block_len = 0;
4907 hole_em->orig_block_len = 0;
4908 hole_em->ram_bytes = hole_size;
4909 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4910 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4911 hole_em->generation = root->fs_info->generation;
4914 write_lock(&em_tree->lock);
4915 err = add_extent_mapping(em_tree, hole_em, 1);
4916 write_unlock(&em_tree->lock);
4919 btrfs_drop_extent_cache(inode, cur_offset,
4923 free_extent_map(hole_em);
4926 free_extent_map(em);
4928 cur_offset = last_byte;
4929 if (cur_offset >= block_end)
4932 free_extent_map(em);
4933 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4938 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4940 struct btrfs_root *root = BTRFS_I(inode)->root;
4941 struct btrfs_trans_handle *trans;
4942 loff_t oldsize = i_size_read(inode);
4943 loff_t newsize = attr->ia_size;
4944 int mask = attr->ia_valid;
4948 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4949 * special case where we need to update the times despite not having
4950 * these flags set. For all other operations the VFS set these flags
4951 * explicitly if it wants a timestamp update.
4953 if (newsize != oldsize) {
4954 inode_inc_iversion(inode);
4955 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4956 inode->i_ctime = inode->i_mtime =
4957 current_fs_time(inode->i_sb);
4960 if (newsize > oldsize) {
4962 * Don't do an expanding truncate while snapshoting is ongoing.
4963 * This is to ensure the snapshot captures a fully consistent
4964 * state of this file - if the snapshot captures this expanding
4965 * truncation, it must capture all writes that happened before
4968 btrfs_wait_for_snapshot_creation(root);
4969 ret = btrfs_cont_expand(inode, oldsize, newsize);
4971 btrfs_end_write_no_snapshoting(root);
4975 trans = btrfs_start_transaction(root, 1);
4976 if (IS_ERR(trans)) {
4977 btrfs_end_write_no_snapshoting(root);
4978 return PTR_ERR(trans);
4981 i_size_write(inode, newsize);
4982 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4983 pagecache_isize_extended(inode, oldsize, newsize);
4984 ret = btrfs_update_inode(trans, root, inode);
4985 btrfs_end_write_no_snapshoting(root);
4986 btrfs_end_transaction(trans, root);
4990 * We're truncating a file that used to have good data down to
4991 * zero. Make sure it gets into the ordered flush list so that
4992 * any new writes get down to disk quickly.
4995 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4996 &BTRFS_I(inode)->runtime_flags);
4999 * 1 for the orphan item we're going to add
5000 * 1 for the orphan item deletion.
5002 trans = btrfs_start_transaction(root, 2);
5004 return PTR_ERR(trans);
5007 * We need to do this in case we fail at _any_ point during the
5008 * actual truncate. Once we do the truncate_setsize we could
5009 * invalidate pages which forces any outstanding ordered io to
5010 * be instantly completed which will give us extents that need
5011 * to be truncated. If we fail to get an orphan inode down we
5012 * could have left over extents that were never meant to live,
5013 * so we need to guarantee from this point on that everything
5014 * will be consistent.
5016 ret = btrfs_orphan_add(trans, inode);
5017 btrfs_end_transaction(trans, root);
5021 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5022 truncate_setsize(inode, newsize);
5024 /* Disable nonlocked read DIO to avoid the end less truncate */
5025 btrfs_inode_block_unlocked_dio(inode);
5026 inode_dio_wait(inode);
5027 btrfs_inode_resume_unlocked_dio(inode);
5029 ret = btrfs_truncate(inode);
5030 if (ret && inode->i_nlink) {
5034 * failed to truncate, disk_i_size is only adjusted down
5035 * as we remove extents, so it should represent the true
5036 * size of the inode, so reset the in memory size and
5037 * delete our orphan entry.
5039 trans = btrfs_join_transaction(root);
5040 if (IS_ERR(trans)) {
5041 btrfs_orphan_del(NULL, inode);
5044 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5045 err = btrfs_orphan_del(trans, inode);
5047 btrfs_abort_transaction(trans, root, err);
5048 btrfs_end_transaction(trans, root);
5055 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5057 struct inode *inode = d_inode(dentry);
5058 struct btrfs_root *root = BTRFS_I(inode)->root;
5061 if (btrfs_root_readonly(root))
5064 err = inode_change_ok(inode, attr);
5068 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5069 err = btrfs_setsize(inode, attr);
5074 if (attr->ia_valid) {
5075 setattr_copy(inode, attr);
5076 inode_inc_iversion(inode);
5077 err = btrfs_dirty_inode(inode);
5079 if (!err && attr->ia_valid & ATTR_MODE)
5080 err = posix_acl_chmod(inode, inode->i_mode);
5087 * While truncating the inode pages during eviction, we get the VFS calling
5088 * btrfs_invalidatepage() against each page of the inode. This is slow because
5089 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5090 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5091 * extent_state structures over and over, wasting lots of time.
5093 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5094 * those expensive operations on a per page basis and do only the ordered io
5095 * finishing, while we release here the extent_map and extent_state structures,
5096 * without the excessive merging and splitting.
5098 static void evict_inode_truncate_pages(struct inode *inode)
5100 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5101 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5102 struct rb_node *node;
5104 ASSERT(inode->i_state & I_FREEING);
5105 truncate_inode_pages_final(&inode->i_data);
5107 write_lock(&map_tree->lock);
5108 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5109 struct extent_map *em;
5111 node = rb_first(&map_tree->map);
5112 em = rb_entry(node, struct extent_map, rb_node);
5113 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5114 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5115 remove_extent_mapping(map_tree, em);
5116 free_extent_map(em);
5117 if (need_resched()) {
5118 write_unlock(&map_tree->lock);
5120 write_lock(&map_tree->lock);
5123 write_unlock(&map_tree->lock);
5126 * Keep looping until we have no more ranges in the io tree.
5127 * We can have ongoing bios started by readpages (called from readahead)
5128 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5129 * still in progress (unlocked the pages in the bio but did not yet
5130 * unlocked the ranges in the io tree). Therefore this means some
5131 * ranges can still be locked and eviction started because before
5132 * submitting those bios, which are executed by a separate task (work
5133 * queue kthread), inode references (inode->i_count) were not taken
5134 * (which would be dropped in the end io callback of each bio).
5135 * Therefore here we effectively end up waiting for those bios and
5136 * anyone else holding locked ranges without having bumped the inode's
5137 * reference count - if we don't do it, when they access the inode's
5138 * io_tree to unlock a range it may be too late, leading to an
5139 * use-after-free issue.
5141 spin_lock(&io_tree->lock);
5142 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5143 struct extent_state *state;
5144 struct extent_state *cached_state = NULL;
5148 node = rb_first(&io_tree->state);
5149 state = rb_entry(node, struct extent_state, rb_node);
5150 start = state->start;
5152 spin_unlock(&io_tree->lock);
5154 lock_extent_bits(io_tree, start, end, &cached_state);
5157 * If still has DELALLOC flag, the extent didn't reach disk,
5158 * and its reserved space won't be freed by delayed_ref.
5159 * So we need to free its reserved space here.
5160 * (Refer to comment in btrfs_invalidatepage, case 2)
5162 * Note, end is the bytenr of last byte, so we need + 1 here.
5164 if (state->state & EXTENT_DELALLOC)
5165 btrfs_qgroup_free_data(inode, start, end - start + 1);
5167 clear_extent_bit(io_tree, start, end,
5168 EXTENT_LOCKED | EXTENT_DIRTY |
5169 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5170 EXTENT_DEFRAG, 1, 1,
5171 &cached_state, GFP_NOFS);
5174 spin_lock(&io_tree->lock);
5176 spin_unlock(&io_tree->lock);
5179 void btrfs_evict_inode(struct inode *inode)
5181 struct btrfs_trans_handle *trans;
5182 struct btrfs_root *root = BTRFS_I(inode)->root;
5183 struct btrfs_block_rsv *rsv, *global_rsv;
5184 int steal_from_global = 0;
5185 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5188 trace_btrfs_inode_evict(inode);
5190 evict_inode_truncate_pages(inode);
5192 if (inode->i_nlink &&
5193 ((btrfs_root_refs(&root->root_item) != 0 &&
5194 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5195 btrfs_is_free_space_inode(inode)))
5198 if (is_bad_inode(inode)) {
5199 btrfs_orphan_del(NULL, inode);
5202 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5203 if (!special_file(inode->i_mode))
5204 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5206 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5208 if (root->fs_info->log_root_recovering) {
5209 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5210 &BTRFS_I(inode)->runtime_flags));
5214 if (inode->i_nlink > 0) {
5215 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5216 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5220 ret = btrfs_commit_inode_delayed_inode(inode);
5222 btrfs_orphan_del(NULL, inode);
5226 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5228 btrfs_orphan_del(NULL, inode);
5231 rsv->size = min_size;
5233 global_rsv = &root->fs_info->global_block_rsv;
5235 btrfs_i_size_write(inode, 0);
5238 * This is a bit simpler than btrfs_truncate since we've already
5239 * reserved our space for our orphan item in the unlink, so we just
5240 * need to reserve some slack space in case we add bytes and update
5241 * inode item when doing the truncate.
5244 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5245 BTRFS_RESERVE_FLUSH_LIMIT);
5248 * Try and steal from the global reserve since we will
5249 * likely not use this space anyway, we want to try as
5250 * hard as possible to get this to work.
5253 steal_from_global++;
5255 steal_from_global = 0;
5259 * steal_from_global == 0: we reserved stuff, hooray!
5260 * steal_from_global == 1: we didn't reserve stuff, boo!
5261 * steal_from_global == 2: we've committed, still not a lot of
5262 * room but maybe we'll have room in the global reserve this
5264 * steal_from_global == 3: abandon all hope!
5266 if (steal_from_global > 2) {
5267 btrfs_warn(root->fs_info,
5268 "Could not get space for a delete, will truncate on mount %d",
5270 btrfs_orphan_del(NULL, inode);
5271 btrfs_free_block_rsv(root, rsv);
5275 trans = btrfs_join_transaction(root);
5276 if (IS_ERR(trans)) {
5277 btrfs_orphan_del(NULL, inode);
5278 btrfs_free_block_rsv(root, rsv);
5283 * We can't just steal from the global reserve, we need to make
5284 * sure there is room to do it, if not we need to commit and try
5287 if (steal_from_global) {
5288 if (!btrfs_check_space_for_delayed_refs(trans, root))
5289 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5296 * Couldn't steal from the global reserve, we have too much
5297 * pending stuff built up, commit the transaction and try it
5301 ret = btrfs_commit_transaction(trans, root);
5303 btrfs_orphan_del(NULL, inode);
5304 btrfs_free_block_rsv(root, rsv);
5309 steal_from_global = 0;
5312 trans->block_rsv = rsv;
5314 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5315 if (ret != -ENOSPC && ret != -EAGAIN)
5318 trans->block_rsv = &root->fs_info->trans_block_rsv;
5319 btrfs_end_transaction(trans, root);
5321 btrfs_btree_balance_dirty(root);
5324 btrfs_free_block_rsv(root, rsv);
5327 * Errors here aren't a big deal, it just means we leave orphan items
5328 * in the tree. They will be cleaned up on the next mount.
5331 trans->block_rsv = root->orphan_block_rsv;
5332 btrfs_orphan_del(trans, inode);
5334 btrfs_orphan_del(NULL, inode);
5337 trans->block_rsv = &root->fs_info->trans_block_rsv;
5338 if (!(root == root->fs_info->tree_root ||
5339 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5340 btrfs_return_ino(root, btrfs_ino(inode));
5342 btrfs_end_transaction(trans, root);
5343 btrfs_btree_balance_dirty(root);
5345 btrfs_remove_delayed_node(inode);
5350 * this returns the key found in the dir entry in the location pointer.
5351 * If no dir entries were found, location->objectid is 0.
5353 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5354 struct btrfs_key *location)
5356 const char *name = dentry->d_name.name;
5357 int namelen = dentry->d_name.len;
5358 struct btrfs_dir_item *di;
5359 struct btrfs_path *path;
5360 struct btrfs_root *root = BTRFS_I(dir)->root;
5363 path = btrfs_alloc_path();
5367 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5372 if (IS_ERR_OR_NULL(di))
5375 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5377 btrfs_free_path(path);
5380 location->objectid = 0;
5385 * when we hit a tree root in a directory, the btrfs part of the inode
5386 * needs to be changed to reflect the root directory of the tree root. This
5387 * is kind of like crossing a mount point.
5389 static int fixup_tree_root_location(struct btrfs_root *root,
5391 struct dentry *dentry,
5392 struct btrfs_key *location,
5393 struct btrfs_root **sub_root)
5395 struct btrfs_path *path;
5396 struct btrfs_root *new_root;
5397 struct btrfs_root_ref *ref;
5398 struct extent_buffer *leaf;
5399 struct btrfs_key key;
5403 path = btrfs_alloc_path();
5410 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5411 key.type = BTRFS_ROOT_REF_KEY;
5412 key.offset = location->objectid;
5414 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5422 leaf = path->nodes[0];
5423 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5424 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5425 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5428 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5429 (unsigned long)(ref + 1),
5430 dentry->d_name.len);
5434 btrfs_release_path(path);
5436 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5437 if (IS_ERR(new_root)) {
5438 err = PTR_ERR(new_root);
5442 *sub_root = new_root;
5443 location->objectid = btrfs_root_dirid(&new_root->root_item);
5444 location->type = BTRFS_INODE_ITEM_KEY;
5445 location->offset = 0;
5448 btrfs_free_path(path);
5452 static void inode_tree_add(struct inode *inode)
5454 struct btrfs_root *root = BTRFS_I(inode)->root;
5455 struct btrfs_inode *entry;
5457 struct rb_node *parent;
5458 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5459 u64 ino = btrfs_ino(inode);
5461 if (inode_unhashed(inode))
5464 spin_lock(&root->inode_lock);
5465 p = &root->inode_tree.rb_node;
5468 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5470 if (ino < btrfs_ino(&entry->vfs_inode))
5471 p = &parent->rb_left;
5472 else if (ino > btrfs_ino(&entry->vfs_inode))
5473 p = &parent->rb_right;
5475 WARN_ON(!(entry->vfs_inode.i_state &
5476 (I_WILL_FREE | I_FREEING)));
5477 rb_replace_node(parent, new, &root->inode_tree);
5478 RB_CLEAR_NODE(parent);
5479 spin_unlock(&root->inode_lock);
5483 rb_link_node(new, parent, p);
5484 rb_insert_color(new, &root->inode_tree);
5485 spin_unlock(&root->inode_lock);
5488 static void inode_tree_del(struct inode *inode)
5490 struct btrfs_root *root = BTRFS_I(inode)->root;
5493 spin_lock(&root->inode_lock);
5494 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5495 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5496 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5497 empty = RB_EMPTY_ROOT(&root->inode_tree);
5499 spin_unlock(&root->inode_lock);
5501 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5502 synchronize_srcu(&root->fs_info->subvol_srcu);
5503 spin_lock(&root->inode_lock);
5504 empty = RB_EMPTY_ROOT(&root->inode_tree);
5505 spin_unlock(&root->inode_lock);
5507 btrfs_add_dead_root(root);
5511 void btrfs_invalidate_inodes(struct btrfs_root *root)
5513 struct rb_node *node;
5514 struct rb_node *prev;
5515 struct btrfs_inode *entry;
5516 struct inode *inode;
5519 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5520 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5522 spin_lock(&root->inode_lock);
5524 node = root->inode_tree.rb_node;
5528 entry = rb_entry(node, struct btrfs_inode, rb_node);
5530 if (objectid < btrfs_ino(&entry->vfs_inode))
5531 node = node->rb_left;
5532 else if (objectid > btrfs_ino(&entry->vfs_inode))
5533 node = node->rb_right;
5539 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5540 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5544 prev = rb_next(prev);
5548 entry = rb_entry(node, struct btrfs_inode, rb_node);
5549 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5550 inode = igrab(&entry->vfs_inode);
5552 spin_unlock(&root->inode_lock);
5553 if (atomic_read(&inode->i_count) > 1)
5554 d_prune_aliases(inode);
5556 * btrfs_drop_inode will have it removed from
5557 * the inode cache when its usage count
5562 spin_lock(&root->inode_lock);
5566 if (cond_resched_lock(&root->inode_lock))
5569 node = rb_next(node);
5571 spin_unlock(&root->inode_lock);
5574 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5576 struct btrfs_iget_args *args = p;
5577 inode->i_ino = args->location->objectid;
5578 memcpy(&BTRFS_I(inode)->location, args->location,
5579 sizeof(*args->location));
5580 BTRFS_I(inode)->root = args->root;
5584 static int btrfs_find_actor(struct inode *inode, void *opaque)
5586 struct btrfs_iget_args *args = opaque;
5587 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5588 args->root == BTRFS_I(inode)->root;
5591 static struct inode *btrfs_iget_locked(struct super_block *s,
5592 struct btrfs_key *location,
5593 struct btrfs_root *root)
5595 struct inode *inode;
5596 struct btrfs_iget_args args;
5597 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5599 args.location = location;
5602 inode = iget5_locked(s, hashval, btrfs_find_actor,
5603 btrfs_init_locked_inode,
5608 /* Get an inode object given its location and corresponding root.
5609 * Returns in *is_new if the inode was read from disk
5611 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5612 struct btrfs_root *root, int *new)
5614 struct inode *inode;
5616 inode = btrfs_iget_locked(s, location, root);
5618 return ERR_PTR(-ENOMEM);
5620 if (inode->i_state & I_NEW) {
5623 ret = btrfs_read_locked_inode(inode);
5624 if (!is_bad_inode(inode)) {
5625 inode_tree_add(inode);
5626 unlock_new_inode(inode);
5630 unlock_new_inode(inode);
5633 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5640 static struct inode *new_simple_dir(struct super_block *s,
5641 struct btrfs_key *key,
5642 struct btrfs_root *root)
5644 struct inode *inode = new_inode(s);
5647 return ERR_PTR(-ENOMEM);
5649 BTRFS_I(inode)->root = root;
5650 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5651 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5653 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5654 inode->i_op = &btrfs_dir_ro_inode_operations;
5655 inode->i_fop = &simple_dir_operations;
5656 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5657 inode->i_mtime = current_fs_time(inode->i_sb);
5658 inode->i_atime = inode->i_mtime;
5659 inode->i_ctime = inode->i_mtime;
5660 BTRFS_I(inode)->i_otime = inode->i_mtime;
5665 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5667 struct inode *inode;
5668 struct btrfs_root *root = BTRFS_I(dir)->root;
5669 struct btrfs_root *sub_root = root;
5670 struct btrfs_key location;
5674 if (dentry->d_name.len > BTRFS_NAME_LEN)
5675 return ERR_PTR(-ENAMETOOLONG);
5677 ret = btrfs_inode_by_name(dir, dentry, &location);
5679 return ERR_PTR(ret);
5681 if (location.objectid == 0)
5682 return ERR_PTR(-ENOENT);
5684 if (location.type == BTRFS_INODE_ITEM_KEY) {
5685 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5689 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5691 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5692 ret = fixup_tree_root_location(root, dir, dentry,
5693 &location, &sub_root);
5696 inode = ERR_PTR(ret);
5698 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5700 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5702 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5704 if (!IS_ERR(inode) && root != sub_root) {
5705 down_read(&root->fs_info->cleanup_work_sem);
5706 if (!(inode->i_sb->s_flags & MS_RDONLY))
5707 ret = btrfs_orphan_cleanup(sub_root);
5708 up_read(&root->fs_info->cleanup_work_sem);
5711 inode = ERR_PTR(ret);
5718 static int btrfs_dentry_delete(const struct dentry *dentry)
5720 struct btrfs_root *root;
5721 struct inode *inode = d_inode(dentry);
5723 if (!inode && !IS_ROOT(dentry))
5724 inode = d_inode(dentry->d_parent);
5727 root = BTRFS_I(inode)->root;
5728 if (btrfs_root_refs(&root->root_item) == 0)
5731 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5737 static void btrfs_dentry_release(struct dentry *dentry)
5739 kfree(dentry->d_fsdata);
5742 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5745 struct inode *inode;
5747 inode = btrfs_lookup_dentry(dir, dentry);
5748 if (IS_ERR(inode)) {
5749 if (PTR_ERR(inode) == -ENOENT)
5752 return ERR_CAST(inode);
5755 return d_splice_alias(inode, dentry);
5758 unsigned char btrfs_filetype_table[] = {
5759 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5762 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5764 struct inode *inode = file_inode(file);
5765 struct btrfs_root *root = BTRFS_I(inode)->root;
5766 struct btrfs_item *item;
5767 struct btrfs_dir_item *di;
5768 struct btrfs_key key;
5769 struct btrfs_key found_key;
5770 struct btrfs_path *path;
5771 struct list_head ins_list;
5772 struct list_head del_list;
5774 struct extent_buffer *leaf;
5776 unsigned char d_type;
5781 int key_type = BTRFS_DIR_INDEX_KEY;
5785 int is_curr = 0; /* ctx->pos points to the current index? */
5789 /* FIXME, use a real flag for deciding about the key type */
5790 if (root->fs_info->tree_root == root)
5791 key_type = BTRFS_DIR_ITEM_KEY;
5793 if (!dir_emit_dots(file, ctx))
5796 path = btrfs_alloc_path();
5800 path->reada = READA_FORWARD;
5802 if (key_type == BTRFS_DIR_INDEX_KEY) {
5803 INIT_LIST_HEAD(&ins_list);
5804 INIT_LIST_HEAD(&del_list);
5805 put = btrfs_readdir_get_delayed_items(inode, &ins_list,
5809 key.type = key_type;
5810 key.offset = ctx->pos;
5811 key.objectid = btrfs_ino(inode);
5813 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5819 leaf = path->nodes[0];
5820 slot = path->slots[0];
5821 if (slot >= btrfs_header_nritems(leaf)) {
5822 ret = btrfs_next_leaf(root, path);
5830 item = btrfs_item_nr(slot);
5831 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5833 if (found_key.objectid != key.objectid)
5835 if (found_key.type != key_type)
5837 if (found_key.offset < ctx->pos)
5839 if (key_type == BTRFS_DIR_INDEX_KEY &&
5840 btrfs_should_delete_dir_index(&del_list,
5844 ctx->pos = found_key.offset;
5847 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5849 di_total = btrfs_item_size(leaf, item);
5851 while (di_cur < di_total) {
5852 struct btrfs_key location;
5854 if (verify_dir_item(root, leaf, di))
5857 name_len = btrfs_dir_name_len(leaf, di);
5858 if (name_len <= sizeof(tmp_name)) {
5859 name_ptr = tmp_name;
5861 name_ptr = kmalloc(name_len, GFP_KERNEL);
5867 read_extent_buffer(leaf, name_ptr,
5868 (unsigned long)(di + 1), name_len);
5870 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5871 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5874 /* is this a reference to our own snapshot? If so
5877 * In contrast to old kernels, we insert the snapshot's
5878 * dir item and dir index after it has been created, so
5879 * we won't find a reference to our own snapshot. We
5880 * still keep the following code for backward
5883 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5884 location.objectid == root->root_key.objectid) {
5888 over = !dir_emit(ctx, name_ptr, name_len,
5889 location.objectid, d_type);
5892 if (name_ptr != tmp_name)
5898 di_len = btrfs_dir_name_len(leaf, di) +
5899 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5901 di = (struct btrfs_dir_item *)((char *)di + di_len);
5907 if (key_type == BTRFS_DIR_INDEX_KEY) {
5910 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5916 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5917 * it was was set to the termination value in previous call. We assume
5918 * that "." and ".." were emitted if we reach this point and set the
5919 * termination value as well for an empty directory.
5921 if (ctx->pos > 2 && !emitted)
5924 /* Reached end of directory/root. Bump pos past the last item. */
5928 * Stop new entries from being returned after we return the last
5931 * New directory entries are assigned a strictly increasing
5932 * offset. This means that new entries created during readdir
5933 * are *guaranteed* to be seen in the future by that readdir.
5934 * This has broken buggy programs which operate on names as
5935 * they're returned by readdir. Until we re-use freed offsets
5936 * we have this hack to stop new entries from being returned
5937 * under the assumption that they'll never reach this huge
5940 * This is being careful not to overflow 32bit loff_t unless the
5941 * last entry requires it because doing so has broken 32bit apps
5944 if (key_type == BTRFS_DIR_INDEX_KEY) {
5945 if (ctx->pos >= INT_MAX)
5946 ctx->pos = LLONG_MAX;
5954 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5955 btrfs_free_path(path);
5959 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5961 struct btrfs_root *root = BTRFS_I(inode)->root;
5962 struct btrfs_trans_handle *trans;
5964 bool nolock = false;
5966 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5969 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5972 if (wbc->sync_mode == WB_SYNC_ALL) {
5974 trans = btrfs_join_transaction_nolock(root);
5976 trans = btrfs_join_transaction(root);
5978 return PTR_ERR(trans);
5979 ret = btrfs_commit_transaction(trans, root);
5985 * This is somewhat expensive, updating the tree every time the
5986 * inode changes. But, it is most likely to find the inode in cache.
5987 * FIXME, needs more benchmarking...there are no reasons other than performance
5988 * to keep or drop this code.
5990 static int btrfs_dirty_inode(struct inode *inode)
5992 struct btrfs_root *root = BTRFS_I(inode)->root;
5993 struct btrfs_trans_handle *trans;
5996 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5999 trans = btrfs_join_transaction(root);
6001 return PTR_ERR(trans);
6003 ret = btrfs_update_inode(trans, root, inode);
6004 if (ret && ret == -ENOSPC) {
6005 /* whoops, lets try again with the full transaction */
6006 btrfs_end_transaction(trans, root);
6007 trans = btrfs_start_transaction(root, 1);
6009 return PTR_ERR(trans);
6011 ret = btrfs_update_inode(trans, root, inode);
6013 btrfs_end_transaction(trans, root);
6014 if (BTRFS_I(inode)->delayed_node)
6015 btrfs_balance_delayed_items(root);
6021 * This is a copy of file_update_time. We need this so we can return error on
6022 * ENOSPC for updating the inode in the case of file write and mmap writes.
6024 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6027 struct btrfs_root *root = BTRFS_I(inode)->root;
6029 if (btrfs_root_readonly(root))
6032 if (flags & S_VERSION)
6033 inode_inc_iversion(inode);
6034 if (flags & S_CTIME)
6035 inode->i_ctime = *now;
6036 if (flags & S_MTIME)
6037 inode->i_mtime = *now;
6038 if (flags & S_ATIME)
6039 inode->i_atime = *now;
6040 return btrfs_dirty_inode(inode);
6044 * find the highest existing sequence number in a directory
6045 * and then set the in-memory index_cnt variable to reflect
6046 * free sequence numbers
6048 static int btrfs_set_inode_index_count(struct inode *inode)
6050 struct btrfs_root *root = BTRFS_I(inode)->root;
6051 struct btrfs_key key, found_key;
6052 struct btrfs_path *path;
6053 struct extent_buffer *leaf;
6056 key.objectid = btrfs_ino(inode);
6057 key.type = BTRFS_DIR_INDEX_KEY;
6058 key.offset = (u64)-1;
6060 path = btrfs_alloc_path();
6064 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6067 /* FIXME: we should be able to handle this */
6073 * MAGIC NUMBER EXPLANATION:
6074 * since we search a directory based on f_pos we have to start at 2
6075 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6076 * else has to start at 2
6078 if (path->slots[0] == 0) {
6079 BTRFS_I(inode)->index_cnt = 2;
6085 leaf = path->nodes[0];
6086 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6088 if (found_key.objectid != btrfs_ino(inode) ||
6089 found_key.type != BTRFS_DIR_INDEX_KEY) {
6090 BTRFS_I(inode)->index_cnt = 2;
6094 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6096 btrfs_free_path(path);
6101 * helper to find a free sequence number in a given directory. This current
6102 * code is very simple, later versions will do smarter things in the btree
6104 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6108 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6109 ret = btrfs_inode_delayed_dir_index_count(dir);
6111 ret = btrfs_set_inode_index_count(dir);
6117 *index = BTRFS_I(dir)->index_cnt;
6118 BTRFS_I(dir)->index_cnt++;
6123 static int btrfs_insert_inode_locked(struct inode *inode)
6125 struct btrfs_iget_args args;
6126 args.location = &BTRFS_I(inode)->location;
6127 args.root = BTRFS_I(inode)->root;
6129 return insert_inode_locked4(inode,
6130 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6131 btrfs_find_actor, &args);
6134 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6135 struct btrfs_root *root,
6137 const char *name, int name_len,
6138 u64 ref_objectid, u64 objectid,
6139 umode_t mode, u64 *index)
6141 struct inode *inode;
6142 struct btrfs_inode_item *inode_item;
6143 struct btrfs_key *location;
6144 struct btrfs_path *path;
6145 struct btrfs_inode_ref *ref;
6146 struct btrfs_key key[2];
6148 int nitems = name ? 2 : 1;
6152 path = btrfs_alloc_path();
6154 return ERR_PTR(-ENOMEM);
6156 inode = new_inode(root->fs_info->sb);
6158 btrfs_free_path(path);
6159 return ERR_PTR(-ENOMEM);
6163 * O_TMPFILE, set link count to 0, so that after this point,
6164 * we fill in an inode item with the correct link count.
6167 set_nlink(inode, 0);
6170 * we have to initialize this early, so we can reclaim the inode
6171 * number if we fail afterwards in this function.
6173 inode->i_ino = objectid;
6176 trace_btrfs_inode_request(dir);
6178 ret = btrfs_set_inode_index(dir, index);
6180 btrfs_free_path(path);
6182 return ERR_PTR(ret);
6188 * index_cnt is ignored for everything but a dir,
6189 * btrfs_get_inode_index_count has an explanation for the magic
6192 BTRFS_I(inode)->index_cnt = 2;
6193 BTRFS_I(inode)->dir_index = *index;
6194 BTRFS_I(inode)->root = root;
6195 BTRFS_I(inode)->generation = trans->transid;
6196 inode->i_generation = BTRFS_I(inode)->generation;
6199 * We could have gotten an inode number from somebody who was fsynced
6200 * and then removed in this same transaction, so let's just set full
6201 * sync since it will be a full sync anyway and this will blow away the
6202 * old info in the log.
6204 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6206 key[0].objectid = objectid;
6207 key[0].type = BTRFS_INODE_ITEM_KEY;
6210 sizes[0] = sizeof(struct btrfs_inode_item);
6214 * Start new inodes with an inode_ref. This is slightly more
6215 * efficient for small numbers of hard links since they will
6216 * be packed into one item. Extended refs will kick in if we
6217 * add more hard links than can fit in the ref item.
6219 key[1].objectid = objectid;
6220 key[1].type = BTRFS_INODE_REF_KEY;
6221 key[1].offset = ref_objectid;
6223 sizes[1] = name_len + sizeof(*ref);
6226 location = &BTRFS_I(inode)->location;
6227 location->objectid = objectid;
6228 location->offset = 0;
6229 location->type = BTRFS_INODE_ITEM_KEY;
6231 ret = btrfs_insert_inode_locked(inode);
6235 path->leave_spinning = 1;
6236 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6240 inode_init_owner(inode, dir, mode);
6241 inode_set_bytes(inode, 0);
6243 inode->i_mtime = current_fs_time(inode->i_sb);
6244 inode->i_atime = inode->i_mtime;
6245 inode->i_ctime = inode->i_mtime;
6246 BTRFS_I(inode)->i_otime = inode->i_mtime;
6248 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6249 struct btrfs_inode_item);
6250 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6251 sizeof(*inode_item));
6252 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6255 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6256 struct btrfs_inode_ref);
6257 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6258 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6259 ptr = (unsigned long)(ref + 1);
6260 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6263 btrfs_mark_buffer_dirty(path->nodes[0]);
6264 btrfs_free_path(path);
6266 btrfs_inherit_iflags(inode, dir);
6268 if (S_ISREG(mode)) {
6269 if (btrfs_test_opt(root, NODATASUM))
6270 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6271 if (btrfs_test_opt(root, NODATACOW))
6272 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6273 BTRFS_INODE_NODATASUM;
6276 inode_tree_add(inode);
6278 trace_btrfs_inode_new(inode);
6279 btrfs_set_inode_last_trans(trans, inode);
6281 btrfs_update_root_times(trans, root);
6283 ret = btrfs_inode_inherit_props(trans, inode, dir);
6285 btrfs_err(root->fs_info,
6286 "error inheriting props for ino %llu (root %llu): %d",
6287 btrfs_ino(inode), root->root_key.objectid, ret);
6292 unlock_new_inode(inode);
6295 BTRFS_I(dir)->index_cnt--;
6296 btrfs_free_path(path);
6298 return ERR_PTR(ret);
6301 static inline u8 btrfs_inode_type(struct inode *inode)
6303 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6307 * utility function to add 'inode' into 'parent_inode' with
6308 * a give name and a given sequence number.
6309 * if 'add_backref' is true, also insert a backref from the
6310 * inode to the parent directory.
6312 int btrfs_add_link(struct btrfs_trans_handle *trans,
6313 struct inode *parent_inode, struct inode *inode,
6314 const char *name, int name_len, int add_backref, u64 index)
6317 struct btrfs_key key;
6318 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6319 u64 ino = btrfs_ino(inode);
6320 u64 parent_ino = btrfs_ino(parent_inode);
6322 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6323 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6326 key.type = BTRFS_INODE_ITEM_KEY;
6330 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6331 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6332 key.objectid, root->root_key.objectid,
6333 parent_ino, index, name, name_len);
6334 } else if (add_backref) {
6335 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6339 /* Nothing to clean up yet */
6343 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6345 btrfs_inode_type(inode), index);
6346 if (ret == -EEXIST || ret == -EOVERFLOW)
6349 btrfs_abort_transaction(trans, root, ret);
6353 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6355 inode_inc_iversion(parent_inode);
6356 parent_inode->i_mtime = parent_inode->i_ctime =
6357 current_fs_time(parent_inode->i_sb);
6358 ret = btrfs_update_inode(trans, root, parent_inode);
6360 btrfs_abort_transaction(trans, root, ret);
6364 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6367 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6368 key.objectid, root->root_key.objectid,
6369 parent_ino, &local_index, name, name_len);
6371 } else if (add_backref) {
6375 err = btrfs_del_inode_ref(trans, root, name, name_len,
6376 ino, parent_ino, &local_index);
6381 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6382 struct inode *dir, struct dentry *dentry,
6383 struct inode *inode, int backref, u64 index)
6385 int err = btrfs_add_link(trans, dir, inode,
6386 dentry->d_name.name, dentry->d_name.len,
6393 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6394 umode_t mode, dev_t rdev)
6396 struct btrfs_trans_handle *trans;
6397 struct btrfs_root *root = BTRFS_I(dir)->root;
6398 struct inode *inode = NULL;
6405 * 2 for inode item and ref
6407 * 1 for xattr if selinux is on
6409 trans = btrfs_start_transaction(root, 5);
6411 return PTR_ERR(trans);
6413 err = btrfs_find_free_ino(root, &objectid);
6417 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6418 dentry->d_name.len, btrfs_ino(dir), objectid,
6420 if (IS_ERR(inode)) {
6421 err = PTR_ERR(inode);
6426 * If the active LSM wants to access the inode during
6427 * d_instantiate it needs these. Smack checks to see
6428 * if the filesystem supports xattrs by looking at the
6431 inode->i_op = &btrfs_special_inode_operations;
6432 init_special_inode(inode, inode->i_mode, rdev);
6434 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6436 goto out_unlock_inode;
6438 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6440 goto out_unlock_inode;
6442 btrfs_update_inode(trans, root, inode);
6443 unlock_new_inode(inode);
6444 d_instantiate(dentry, inode);
6448 btrfs_end_transaction(trans, root);
6449 btrfs_balance_delayed_items(root);
6450 btrfs_btree_balance_dirty(root);
6452 inode_dec_link_count(inode);
6459 unlock_new_inode(inode);
6464 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6465 umode_t mode, bool excl)
6467 struct btrfs_trans_handle *trans;
6468 struct btrfs_root *root = BTRFS_I(dir)->root;
6469 struct inode *inode = NULL;
6470 int drop_inode_on_err = 0;
6476 * 2 for inode item and ref
6478 * 1 for xattr if selinux is on
6480 trans = btrfs_start_transaction(root, 5);
6482 return PTR_ERR(trans);
6484 err = btrfs_find_free_ino(root, &objectid);
6488 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6489 dentry->d_name.len, btrfs_ino(dir), objectid,
6491 if (IS_ERR(inode)) {
6492 err = PTR_ERR(inode);
6495 drop_inode_on_err = 1;
6497 * If the active LSM wants to access the inode during
6498 * d_instantiate it needs these. Smack checks to see
6499 * if the filesystem supports xattrs by looking at the
6502 inode->i_fop = &btrfs_file_operations;
6503 inode->i_op = &btrfs_file_inode_operations;
6504 inode->i_mapping->a_ops = &btrfs_aops;
6506 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6508 goto out_unlock_inode;
6510 err = btrfs_update_inode(trans, root, inode);
6512 goto out_unlock_inode;
6514 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6516 goto out_unlock_inode;
6518 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6519 unlock_new_inode(inode);
6520 d_instantiate(dentry, inode);
6523 btrfs_end_transaction(trans, root);
6524 if (err && drop_inode_on_err) {
6525 inode_dec_link_count(inode);
6528 btrfs_balance_delayed_items(root);
6529 btrfs_btree_balance_dirty(root);
6533 unlock_new_inode(inode);
6538 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6539 struct dentry *dentry)
6541 struct btrfs_trans_handle *trans = NULL;
6542 struct btrfs_root *root = BTRFS_I(dir)->root;
6543 struct inode *inode = d_inode(old_dentry);
6548 /* do not allow sys_link's with other subvols of the same device */
6549 if (root->objectid != BTRFS_I(inode)->root->objectid)
6552 if (inode->i_nlink >= BTRFS_LINK_MAX)
6555 err = btrfs_set_inode_index(dir, &index);
6560 * 2 items for inode and inode ref
6561 * 2 items for dir items
6562 * 1 item for parent inode
6564 trans = btrfs_start_transaction(root, 5);
6565 if (IS_ERR(trans)) {
6566 err = PTR_ERR(trans);
6571 /* There are several dir indexes for this inode, clear the cache. */
6572 BTRFS_I(inode)->dir_index = 0ULL;
6574 inode_inc_iversion(inode);
6575 inode->i_ctime = current_fs_time(inode->i_sb);
6577 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6579 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6584 struct dentry *parent = dentry->d_parent;
6585 err = btrfs_update_inode(trans, root, inode);
6588 if (inode->i_nlink == 1) {
6590 * If new hard link count is 1, it's a file created
6591 * with open(2) O_TMPFILE flag.
6593 err = btrfs_orphan_del(trans, inode);
6597 d_instantiate(dentry, inode);
6598 btrfs_log_new_name(trans, inode, NULL, parent);
6601 btrfs_balance_delayed_items(root);
6604 btrfs_end_transaction(trans, root);
6606 inode_dec_link_count(inode);
6609 btrfs_btree_balance_dirty(root);
6613 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6615 struct inode *inode = NULL;
6616 struct btrfs_trans_handle *trans;
6617 struct btrfs_root *root = BTRFS_I(dir)->root;
6619 int drop_on_err = 0;
6624 * 2 items for inode and ref
6625 * 2 items for dir items
6626 * 1 for xattr if selinux is on
6628 trans = btrfs_start_transaction(root, 5);
6630 return PTR_ERR(trans);
6632 err = btrfs_find_free_ino(root, &objectid);
6636 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6637 dentry->d_name.len, btrfs_ino(dir), objectid,
6638 S_IFDIR | mode, &index);
6639 if (IS_ERR(inode)) {
6640 err = PTR_ERR(inode);
6645 /* these must be set before we unlock the inode */
6646 inode->i_op = &btrfs_dir_inode_operations;
6647 inode->i_fop = &btrfs_dir_file_operations;
6649 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6651 goto out_fail_inode;
6653 btrfs_i_size_write(inode, 0);
6654 err = btrfs_update_inode(trans, root, inode);
6656 goto out_fail_inode;
6658 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6659 dentry->d_name.len, 0, index);
6661 goto out_fail_inode;
6663 d_instantiate(dentry, inode);
6665 * mkdir is special. We're unlocking after we call d_instantiate
6666 * to avoid a race with nfsd calling d_instantiate.
6668 unlock_new_inode(inode);
6672 btrfs_end_transaction(trans, root);
6674 inode_dec_link_count(inode);
6677 btrfs_balance_delayed_items(root);
6678 btrfs_btree_balance_dirty(root);
6682 unlock_new_inode(inode);
6686 /* Find next extent map of a given extent map, caller needs to ensure locks */
6687 static struct extent_map *next_extent_map(struct extent_map *em)
6689 struct rb_node *next;
6691 next = rb_next(&em->rb_node);
6694 return container_of(next, struct extent_map, rb_node);
6697 static struct extent_map *prev_extent_map(struct extent_map *em)
6699 struct rb_node *prev;
6701 prev = rb_prev(&em->rb_node);
6704 return container_of(prev, struct extent_map, rb_node);
6707 /* helper for btfs_get_extent. Given an existing extent in the tree,
6708 * the existing extent is the nearest extent to map_start,
6709 * and an extent that you want to insert, deal with overlap and insert
6710 * the best fitted new extent into the tree.
6712 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6713 struct extent_map *existing,
6714 struct extent_map *em,
6717 struct extent_map *prev;
6718 struct extent_map *next;
6723 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6725 if (existing->start > map_start) {
6727 prev = prev_extent_map(next);
6730 next = next_extent_map(prev);
6733 start = prev ? extent_map_end(prev) : em->start;
6734 start = max_t(u64, start, em->start);
6735 end = next ? next->start : extent_map_end(em);
6736 end = min_t(u64, end, extent_map_end(em));
6737 start_diff = start - em->start;
6739 em->len = end - start;
6740 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6741 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6742 em->block_start += start_diff;
6743 em->block_len -= start_diff;
6745 return add_extent_mapping(em_tree, em, 0);
6748 static noinline int uncompress_inline(struct btrfs_path *path,
6750 size_t pg_offset, u64 extent_offset,
6751 struct btrfs_file_extent_item *item)
6754 struct extent_buffer *leaf = path->nodes[0];
6757 unsigned long inline_size;
6761 WARN_ON(pg_offset != 0);
6762 compress_type = btrfs_file_extent_compression(leaf, item);
6763 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6764 inline_size = btrfs_file_extent_inline_item_len(leaf,
6765 btrfs_item_nr(path->slots[0]));
6766 tmp = kmalloc(inline_size, GFP_NOFS);
6769 ptr = btrfs_file_extent_inline_start(item);
6771 read_extent_buffer(leaf, tmp, ptr, inline_size);
6773 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6774 ret = btrfs_decompress(compress_type, tmp, page,
6775 extent_offset, inline_size, max_size);
6781 * a bit scary, this does extent mapping from logical file offset to the disk.
6782 * the ugly parts come from merging extents from the disk with the in-ram
6783 * representation. This gets more complex because of the data=ordered code,
6784 * where the in-ram extents might be locked pending data=ordered completion.
6786 * This also copies inline extents directly into the page.
6789 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6790 size_t pg_offset, u64 start, u64 len,
6795 u64 extent_start = 0;
6797 u64 objectid = btrfs_ino(inode);
6799 struct btrfs_path *path = NULL;
6800 struct btrfs_root *root = BTRFS_I(inode)->root;
6801 struct btrfs_file_extent_item *item;
6802 struct extent_buffer *leaf;
6803 struct btrfs_key found_key;
6804 struct extent_map *em = NULL;
6805 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6806 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6807 struct btrfs_trans_handle *trans = NULL;
6808 const bool new_inline = !page || create;
6811 read_lock(&em_tree->lock);
6812 em = lookup_extent_mapping(em_tree, start, len);
6814 em->bdev = root->fs_info->fs_devices->latest_bdev;
6815 read_unlock(&em_tree->lock);
6818 if (em->start > start || em->start + em->len <= start)
6819 free_extent_map(em);
6820 else if (em->block_start == EXTENT_MAP_INLINE && page)
6821 free_extent_map(em);
6825 em = alloc_extent_map();
6830 em->bdev = root->fs_info->fs_devices->latest_bdev;
6831 em->start = EXTENT_MAP_HOLE;
6832 em->orig_start = EXTENT_MAP_HOLE;
6834 em->block_len = (u64)-1;
6837 path = btrfs_alloc_path();
6843 * Chances are we'll be called again, so go ahead and do
6846 path->reada = READA_FORWARD;
6849 ret = btrfs_lookup_file_extent(trans, root, path,
6850 objectid, start, trans != NULL);
6857 if (path->slots[0] == 0)
6862 leaf = path->nodes[0];
6863 item = btrfs_item_ptr(leaf, path->slots[0],
6864 struct btrfs_file_extent_item);
6865 /* are we inside the extent that was found? */
6866 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6867 found_type = found_key.type;
6868 if (found_key.objectid != objectid ||
6869 found_type != BTRFS_EXTENT_DATA_KEY) {
6871 * If we backup past the first extent we want to move forward
6872 * and see if there is an extent in front of us, otherwise we'll
6873 * say there is a hole for our whole search range which can
6880 found_type = btrfs_file_extent_type(leaf, item);
6881 extent_start = found_key.offset;
6882 if (found_type == BTRFS_FILE_EXTENT_REG ||
6883 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6884 extent_end = extent_start +
6885 btrfs_file_extent_num_bytes(leaf, item);
6886 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6888 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6889 extent_end = ALIGN(extent_start + size, root->sectorsize);
6892 if (start >= extent_end) {
6894 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6895 ret = btrfs_next_leaf(root, path);
6902 leaf = path->nodes[0];
6904 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6905 if (found_key.objectid != objectid ||
6906 found_key.type != BTRFS_EXTENT_DATA_KEY)
6908 if (start + len <= found_key.offset)
6910 if (start > found_key.offset)
6913 em->orig_start = start;
6914 em->len = found_key.offset - start;
6918 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6920 if (found_type == BTRFS_FILE_EXTENT_REG ||
6921 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6923 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6927 size_t extent_offset;
6933 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6934 extent_offset = page_offset(page) + pg_offset - extent_start;
6935 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6936 size - extent_offset);
6937 em->start = extent_start + extent_offset;
6938 em->len = ALIGN(copy_size, root->sectorsize);
6939 em->orig_block_len = em->len;
6940 em->orig_start = em->start;
6941 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6942 if (create == 0 && !PageUptodate(page)) {
6943 if (btrfs_file_extent_compression(leaf, item) !=
6944 BTRFS_COMPRESS_NONE) {
6945 ret = uncompress_inline(path, page, pg_offset,
6946 extent_offset, item);
6953 read_extent_buffer(leaf, map + pg_offset, ptr,
6955 if (pg_offset + copy_size < PAGE_SIZE) {
6956 memset(map + pg_offset + copy_size, 0,
6957 PAGE_SIZE - pg_offset -
6962 flush_dcache_page(page);
6963 } else if (create && PageUptodate(page)) {
6967 free_extent_map(em);
6970 btrfs_release_path(path);
6971 trans = btrfs_join_transaction(root);
6974 return ERR_CAST(trans);
6978 write_extent_buffer(leaf, map + pg_offset, ptr,
6981 btrfs_mark_buffer_dirty(leaf);
6983 set_extent_uptodate(io_tree, em->start,
6984 extent_map_end(em) - 1, NULL, GFP_NOFS);
6989 em->orig_start = start;
6992 em->block_start = EXTENT_MAP_HOLE;
6993 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6995 btrfs_release_path(path);
6996 if (em->start > start || extent_map_end(em) <= start) {
6997 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6998 em->start, em->len, start, len);
7004 write_lock(&em_tree->lock);
7005 ret = add_extent_mapping(em_tree, em, 0);
7006 /* it is possible that someone inserted the extent into the tree
7007 * while we had the lock dropped. It is also possible that
7008 * an overlapping map exists in the tree
7010 if (ret == -EEXIST) {
7011 struct extent_map *existing;
7015 existing = search_extent_mapping(em_tree, start, len);
7017 * existing will always be non-NULL, since there must be
7018 * extent causing the -EEXIST.
7020 if (existing->start == em->start &&
7021 extent_map_end(existing) == extent_map_end(em) &&
7022 em->block_start == existing->block_start) {
7024 * these two extents are the same, it happens
7025 * with inlines especially
7027 free_extent_map(em);
7031 } else if (start >= extent_map_end(existing) ||
7032 start <= existing->start) {
7034 * The existing extent map is the one nearest to
7035 * the [start, start + len) range which overlaps
7037 err = merge_extent_mapping(em_tree, existing,
7039 free_extent_map(existing);
7041 free_extent_map(em);
7045 free_extent_map(em);
7050 write_unlock(&em_tree->lock);
7053 trace_btrfs_get_extent(root, em);
7055 btrfs_free_path(path);
7057 ret = btrfs_end_transaction(trans, root);
7062 free_extent_map(em);
7063 return ERR_PTR(err);
7065 BUG_ON(!em); /* Error is always set */
7069 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7070 size_t pg_offset, u64 start, u64 len,
7073 struct extent_map *em;
7074 struct extent_map *hole_em = NULL;
7075 u64 range_start = start;
7081 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7088 * - a pre-alloc extent,
7089 * there might actually be delalloc bytes behind it.
7091 if (em->block_start != EXTENT_MAP_HOLE &&
7092 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7098 /* check to see if we've wrapped (len == -1 or similar) */
7107 /* ok, we didn't find anything, lets look for delalloc */
7108 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7109 end, len, EXTENT_DELALLOC, 1);
7110 found_end = range_start + found;
7111 if (found_end < range_start)
7112 found_end = (u64)-1;
7115 * we didn't find anything useful, return
7116 * the original results from get_extent()
7118 if (range_start > end || found_end <= start) {
7124 /* adjust the range_start to make sure it doesn't
7125 * go backwards from the start they passed in
7127 range_start = max(start, range_start);
7128 found = found_end - range_start;
7131 u64 hole_start = start;
7134 em = alloc_extent_map();
7140 * when btrfs_get_extent can't find anything it
7141 * returns one huge hole
7143 * make sure what it found really fits our range, and
7144 * adjust to make sure it is based on the start from
7148 u64 calc_end = extent_map_end(hole_em);
7150 if (calc_end <= start || (hole_em->start > end)) {
7151 free_extent_map(hole_em);
7154 hole_start = max(hole_em->start, start);
7155 hole_len = calc_end - hole_start;
7159 if (hole_em && range_start > hole_start) {
7160 /* our hole starts before our delalloc, so we
7161 * have to return just the parts of the hole
7162 * that go until the delalloc starts
7164 em->len = min(hole_len,
7165 range_start - hole_start);
7166 em->start = hole_start;
7167 em->orig_start = hole_start;
7169 * don't adjust block start at all,
7170 * it is fixed at EXTENT_MAP_HOLE
7172 em->block_start = hole_em->block_start;
7173 em->block_len = hole_len;
7174 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7175 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7177 em->start = range_start;
7179 em->orig_start = range_start;
7180 em->block_start = EXTENT_MAP_DELALLOC;
7181 em->block_len = found;
7183 } else if (hole_em) {
7188 free_extent_map(hole_em);
7190 free_extent_map(em);
7191 return ERR_PTR(err);
7196 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7199 const u64 orig_start,
7200 const u64 block_start,
7201 const u64 block_len,
7202 const u64 orig_block_len,
7203 const u64 ram_bytes,
7206 struct extent_map *em = NULL;
7209 down_read(&BTRFS_I(inode)->dio_sem);
7210 if (type != BTRFS_ORDERED_NOCOW) {
7211 em = create_pinned_em(inode, start, len, orig_start,
7212 block_start, block_len, orig_block_len,
7217 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7218 len, block_len, type);
7221 free_extent_map(em);
7222 btrfs_drop_extent_cache(inode, start,
7223 start + len - 1, 0);
7228 up_read(&BTRFS_I(inode)->dio_sem);
7233 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7236 struct btrfs_root *root = BTRFS_I(inode)->root;
7237 struct extent_map *em;
7238 struct btrfs_key ins;
7242 alloc_hint = get_extent_allocation_hint(inode, start, len);
7243 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7244 alloc_hint, &ins, 1, 1);
7246 return ERR_PTR(ret);
7248 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7249 ins.objectid, ins.offset, ins.offset,
7251 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7253 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7259 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7260 * block must be cow'd
7262 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7263 u64 *orig_start, u64 *orig_block_len,
7266 struct btrfs_trans_handle *trans;
7267 struct btrfs_path *path;
7269 struct extent_buffer *leaf;
7270 struct btrfs_root *root = BTRFS_I(inode)->root;
7271 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7272 struct btrfs_file_extent_item *fi;
7273 struct btrfs_key key;
7280 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7282 path = btrfs_alloc_path();
7286 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7291 slot = path->slots[0];
7294 /* can't find the item, must cow */
7301 leaf = path->nodes[0];
7302 btrfs_item_key_to_cpu(leaf, &key, slot);
7303 if (key.objectid != btrfs_ino(inode) ||
7304 key.type != BTRFS_EXTENT_DATA_KEY) {
7305 /* not our file or wrong item type, must cow */
7309 if (key.offset > offset) {
7310 /* Wrong offset, must cow */
7314 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7315 found_type = btrfs_file_extent_type(leaf, fi);
7316 if (found_type != BTRFS_FILE_EXTENT_REG &&
7317 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7318 /* not a regular extent, must cow */
7322 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7325 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7326 if (extent_end <= offset)
7329 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7330 if (disk_bytenr == 0)
7333 if (btrfs_file_extent_compression(leaf, fi) ||
7334 btrfs_file_extent_encryption(leaf, fi) ||
7335 btrfs_file_extent_other_encoding(leaf, fi))
7338 backref_offset = btrfs_file_extent_offset(leaf, fi);
7341 *orig_start = key.offset - backref_offset;
7342 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7343 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7346 if (btrfs_extent_readonly(root, disk_bytenr))
7349 num_bytes = min(offset + *len, extent_end) - offset;
7350 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7353 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7354 ret = test_range_bit(io_tree, offset, range_end,
7355 EXTENT_DELALLOC, 0, NULL);
7362 btrfs_release_path(path);
7365 * look for other files referencing this extent, if we
7366 * find any we must cow
7368 trans = btrfs_join_transaction(root);
7369 if (IS_ERR(trans)) {
7374 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7375 key.offset - backref_offset, disk_bytenr);
7376 btrfs_end_transaction(trans, root);
7383 * adjust disk_bytenr and num_bytes to cover just the bytes
7384 * in this extent we are about to write. If there
7385 * are any csums in that range we have to cow in order
7386 * to keep the csums correct
7388 disk_bytenr += backref_offset;
7389 disk_bytenr += offset - key.offset;
7390 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7393 * all of the above have passed, it is safe to overwrite this extent
7399 btrfs_free_path(path);
7403 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7405 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7407 void **pagep = NULL;
7408 struct page *page = NULL;
7412 start_idx = start >> PAGE_SHIFT;
7415 * end is the last byte in the last page. end == start is legal
7417 end_idx = end >> PAGE_SHIFT;
7421 /* Most of the code in this while loop is lifted from
7422 * find_get_page. It's been modified to begin searching from a
7423 * page and return just the first page found in that range. If the
7424 * found idx is less than or equal to the end idx then we know that
7425 * a page exists. If no pages are found or if those pages are
7426 * outside of the range then we're fine (yay!) */
7427 while (page == NULL &&
7428 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7429 page = radix_tree_deref_slot(pagep);
7430 if (unlikely(!page))
7433 if (radix_tree_exception(page)) {
7434 if (radix_tree_deref_retry(page)) {
7439 * Otherwise, shmem/tmpfs must be storing a swap entry
7440 * here as an exceptional entry: so return it without
7441 * attempting to raise page count.
7444 break; /* TODO: Is this relevant for this use case? */
7447 if (!page_cache_get_speculative(page)) {
7453 * Has the page moved?
7454 * This is part of the lockless pagecache protocol. See
7455 * include/linux/pagemap.h for details.
7457 if (unlikely(page != *pagep)) {
7464 if (page->index <= end_idx)
7473 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7474 struct extent_state **cached_state, int writing)
7476 struct btrfs_ordered_extent *ordered;
7480 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7483 * We're concerned with the entire range that we're going to be
7484 * doing DIO to, so we need to make sure there's no ordered
7485 * extents in this range.
7487 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7488 lockend - lockstart + 1);
7491 * We need to make sure there are no buffered pages in this
7492 * range either, we could have raced between the invalidate in
7493 * generic_file_direct_write and locking the extent. The
7494 * invalidate needs to happen so that reads after a write do not
7499 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7502 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7503 cached_state, GFP_NOFS);
7507 * If we are doing a DIO read and the ordered extent we
7508 * found is for a buffered write, we can not wait for it
7509 * to complete and retry, because if we do so we can
7510 * deadlock with concurrent buffered writes on page
7511 * locks. This happens only if our DIO read covers more
7512 * than one extent map, if at this point has already
7513 * created an ordered extent for a previous extent map
7514 * and locked its range in the inode's io tree, and a
7515 * concurrent write against that previous extent map's
7516 * range and this range started (we unlock the ranges
7517 * in the io tree only when the bios complete and
7518 * buffered writes always lock pages before attempting
7519 * to lock range in the io tree).
7522 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7523 btrfs_start_ordered_extent(inode, ordered, 1);
7526 btrfs_put_ordered_extent(ordered);
7529 * We could trigger writeback for this range (and wait
7530 * for it to complete) and then invalidate the pages for
7531 * this range (through invalidate_inode_pages2_range()),
7532 * but that can lead us to a deadlock with a concurrent
7533 * call to readpages() (a buffered read or a defrag call
7534 * triggered a readahead) on a page lock due to an
7535 * ordered dio extent we created before but did not have
7536 * yet a corresponding bio submitted (whence it can not
7537 * complete), which makes readpages() wait for that
7538 * ordered extent to complete while holding a lock on
7553 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7554 u64 len, u64 orig_start,
7555 u64 block_start, u64 block_len,
7556 u64 orig_block_len, u64 ram_bytes,
7559 struct extent_map_tree *em_tree;
7560 struct extent_map *em;
7561 struct btrfs_root *root = BTRFS_I(inode)->root;
7564 em_tree = &BTRFS_I(inode)->extent_tree;
7565 em = alloc_extent_map();
7567 return ERR_PTR(-ENOMEM);
7570 em->orig_start = orig_start;
7571 em->mod_start = start;
7574 em->block_len = block_len;
7575 em->block_start = block_start;
7576 em->bdev = root->fs_info->fs_devices->latest_bdev;
7577 em->orig_block_len = orig_block_len;
7578 em->ram_bytes = ram_bytes;
7579 em->generation = -1;
7580 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7581 if (type == BTRFS_ORDERED_PREALLOC)
7582 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7585 btrfs_drop_extent_cache(inode, em->start,
7586 em->start + em->len - 1, 0);
7587 write_lock(&em_tree->lock);
7588 ret = add_extent_mapping(em_tree, em, 1);
7589 write_unlock(&em_tree->lock);
7590 } while (ret == -EEXIST);
7593 free_extent_map(em);
7594 return ERR_PTR(ret);
7600 static void adjust_dio_outstanding_extents(struct inode *inode,
7601 struct btrfs_dio_data *dio_data,
7604 unsigned num_extents;
7606 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7607 BTRFS_MAX_EXTENT_SIZE);
7609 * If we have an outstanding_extents count still set then we're
7610 * within our reservation, otherwise we need to adjust our inode
7611 * counter appropriately.
7613 if (dio_data->outstanding_extents) {
7614 dio_data->outstanding_extents -= num_extents;
7616 spin_lock(&BTRFS_I(inode)->lock);
7617 BTRFS_I(inode)->outstanding_extents += num_extents;
7618 spin_unlock(&BTRFS_I(inode)->lock);
7622 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7623 struct buffer_head *bh_result, int create)
7625 struct extent_map *em;
7626 struct btrfs_root *root = BTRFS_I(inode)->root;
7627 struct extent_state *cached_state = NULL;
7628 struct btrfs_dio_data *dio_data = NULL;
7629 u64 start = iblock << inode->i_blkbits;
7630 u64 lockstart, lockend;
7631 u64 len = bh_result->b_size;
7632 int unlock_bits = EXTENT_LOCKED;
7636 unlock_bits |= EXTENT_DIRTY;
7638 len = min_t(u64, len, root->sectorsize);
7641 lockend = start + len - 1;
7643 if (current->journal_info) {
7645 * Need to pull our outstanding extents and set journal_info to NULL so
7646 * that anything that needs to check if there's a transaction doesn't get
7649 dio_data = current->journal_info;
7650 current->journal_info = NULL;
7654 * If this errors out it's because we couldn't invalidate pagecache for
7655 * this range and we need to fallback to buffered.
7657 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7663 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7670 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7671 * io. INLINE is special, and we could probably kludge it in here, but
7672 * it's still buffered so for safety lets just fall back to the generic
7675 * For COMPRESSED we _have_ to read the entire extent in so we can
7676 * decompress it, so there will be buffering required no matter what we
7677 * do, so go ahead and fallback to buffered.
7679 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7680 * to buffered IO. Don't blame me, this is the price we pay for using
7683 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7684 em->block_start == EXTENT_MAP_INLINE) {
7685 free_extent_map(em);
7690 /* Just a good old fashioned hole, return */
7691 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7692 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7693 free_extent_map(em);
7698 * We don't allocate a new extent in the following cases
7700 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7702 * 2) The extent is marked as PREALLOC. We're good to go here and can
7703 * just use the extent.
7707 len = min(len, em->len - (start - em->start));
7708 lockstart = start + len;
7712 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7713 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7714 em->block_start != EXTENT_MAP_HOLE)) {
7716 u64 block_start, orig_start, orig_block_len, ram_bytes;
7718 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7719 type = BTRFS_ORDERED_PREALLOC;
7721 type = BTRFS_ORDERED_NOCOW;
7722 len = min(len, em->len - (start - em->start));
7723 block_start = em->block_start + (start - em->start);
7725 if (can_nocow_extent(inode, start, &len, &orig_start,
7726 &orig_block_len, &ram_bytes) == 1 &&
7727 btrfs_inc_nocow_writers(root->fs_info, block_start)) {
7728 struct extent_map *em2;
7730 em2 = btrfs_create_dio_extent(inode, start, len,
7731 orig_start, block_start,
7732 len, orig_block_len,
7734 btrfs_dec_nocow_writers(root->fs_info, block_start);
7735 if (type == BTRFS_ORDERED_PREALLOC) {
7736 free_extent_map(em);
7739 if (em2 && IS_ERR(em2)) {
7748 * this will cow the extent, reset the len in case we changed
7751 len = bh_result->b_size;
7752 free_extent_map(em);
7753 em = btrfs_new_extent_direct(inode, start, len);
7758 len = min(len, em->len - (start - em->start));
7760 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7762 bh_result->b_size = len;
7763 bh_result->b_bdev = em->bdev;
7764 set_buffer_mapped(bh_result);
7766 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7767 set_buffer_new(bh_result);
7770 * Need to update the i_size under the extent lock so buffered
7771 * readers will get the updated i_size when we unlock.
7773 if (start + len > i_size_read(inode))
7774 i_size_write(inode, start + len);
7776 adjust_dio_outstanding_extents(inode, dio_data, len);
7777 btrfs_free_reserved_data_space(inode, start, len);
7778 WARN_ON(dio_data->reserve < len);
7779 dio_data->reserve -= len;
7780 dio_data->unsubmitted_oe_range_end = start + len;
7781 current->journal_info = dio_data;
7785 * In the case of write we need to clear and unlock the entire range,
7786 * in the case of read we need to unlock only the end area that we
7787 * aren't using if there is any left over space.
7789 if (lockstart < lockend) {
7790 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7791 lockend, unlock_bits, 1, 0,
7792 &cached_state, GFP_NOFS);
7794 free_extent_state(cached_state);
7797 free_extent_map(em);
7802 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7803 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7806 current->journal_info = dio_data;
7808 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7809 * write less data then expected, so that we don't underflow our inode's
7810 * outstanding extents counter.
7812 if (create && dio_data)
7813 adjust_dio_outstanding_extents(inode, dio_data, len);
7818 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7819 int rw, int mirror_num)
7821 struct btrfs_root *root = BTRFS_I(inode)->root;
7824 BUG_ON(rw & REQ_WRITE);
7828 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7829 BTRFS_WQ_ENDIO_DIO_REPAIR);
7833 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7839 static int btrfs_check_dio_repairable(struct inode *inode,
7840 struct bio *failed_bio,
7841 struct io_failure_record *failrec,
7846 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7847 failrec->logical, failrec->len);
7848 if (num_copies == 1) {
7850 * we only have a single copy of the data, so don't bother with
7851 * all the retry and error correction code that follows. no
7852 * matter what the error is, it is very likely to persist.
7854 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7855 num_copies, failrec->this_mirror, failed_mirror);
7859 failrec->failed_mirror = failed_mirror;
7860 failrec->this_mirror++;
7861 if (failrec->this_mirror == failed_mirror)
7862 failrec->this_mirror++;
7864 if (failrec->this_mirror > num_copies) {
7865 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7866 num_copies, failrec->this_mirror, failed_mirror);
7873 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7874 struct page *page, unsigned int pgoff,
7875 u64 start, u64 end, int failed_mirror,
7876 bio_end_io_t *repair_endio, void *repair_arg)
7878 struct io_failure_record *failrec;
7884 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7886 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7890 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7893 free_io_failure(inode, failrec);
7897 if ((failed_bio->bi_vcnt > 1)
7898 || (failed_bio->bi_io_vec->bv_len
7899 > BTRFS_I(inode)->root->sectorsize))
7900 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7902 read_mode = READ_SYNC;
7904 isector = start - btrfs_io_bio(failed_bio)->logical;
7905 isector >>= inode->i_sb->s_blocksize_bits;
7906 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7907 pgoff, isector, repair_endio, repair_arg);
7909 free_io_failure(inode, failrec);
7913 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7914 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7915 read_mode, failrec->this_mirror, failrec->in_validation);
7917 ret = submit_dio_repair_bio(inode, bio, read_mode,
7918 failrec->this_mirror);
7920 free_io_failure(inode, failrec);
7927 struct btrfs_retry_complete {
7928 struct completion done;
7929 struct inode *inode;
7934 static void btrfs_retry_endio_nocsum(struct bio *bio)
7936 struct btrfs_retry_complete *done = bio->bi_private;
7937 struct inode *inode;
7938 struct bio_vec *bvec;
7944 ASSERT(bio->bi_vcnt == 1);
7945 inode = bio->bi_io_vec->bv_page->mapping->host;
7946 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7949 bio_for_each_segment_all(bvec, bio, i)
7950 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7952 complete(&done->done);
7956 static int __btrfs_correct_data_nocsum(struct inode *inode,
7957 struct btrfs_io_bio *io_bio)
7959 struct btrfs_fs_info *fs_info;
7960 struct bio_vec *bvec;
7961 struct btrfs_retry_complete done;
7969 fs_info = BTRFS_I(inode)->root->fs_info;
7970 sectorsize = BTRFS_I(inode)->root->sectorsize;
7972 start = io_bio->logical;
7975 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7976 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7977 pgoff = bvec->bv_offset;
7979 next_block_or_try_again:
7982 init_completion(&done.done);
7984 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7985 pgoff, start, start + sectorsize - 1,
7987 btrfs_retry_endio_nocsum, &done);
7991 wait_for_completion(&done.done);
7993 if (!done.uptodate) {
7994 /* We might have another mirror, so try again */
7995 goto next_block_or_try_again;
7998 start += sectorsize;
8001 pgoff += sectorsize;
8002 goto next_block_or_try_again;
8009 static void btrfs_retry_endio(struct bio *bio)
8011 struct btrfs_retry_complete *done = bio->bi_private;
8012 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8013 struct inode *inode;
8014 struct bio_vec *bvec;
8025 start = done->start;
8027 ASSERT(bio->bi_vcnt == 1);
8028 inode = bio->bi_io_vec->bv_page->mapping->host;
8029 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
8031 bio_for_each_segment_all(bvec, bio, i) {
8032 ret = __readpage_endio_check(done->inode, io_bio, i,
8033 bvec->bv_page, bvec->bv_offset,
8034 done->start, bvec->bv_len);
8036 clean_io_failure(done->inode, done->start,
8037 bvec->bv_page, bvec->bv_offset);
8042 done->uptodate = uptodate;
8044 complete(&done->done);
8048 static int __btrfs_subio_endio_read(struct inode *inode,
8049 struct btrfs_io_bio *io_bio, int err)
8051 struct btrfs_fs_info *fs_info;
8052 struct bio_vec *bvec;
8053 struct btrfs_retry_complete done;
8063 fs_info = BTRFS_I(inode)->root->fs_info;
8064 sectorsize = BTRFS_I(inode)->root->sectorsize;
8067 start = io_bio->logical;
8070 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8071 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8073 pgoff = bvec->bv_offset;
8075 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8076 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8077 bvec->bv_page, pgoff, start,
8084 init_completion(&done.done);
8086 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8087 pgoff, start, start + sectorsize - 1,
8089 btrfs_retry_endio, &done);
8095 wait_for_completion(&done.done);
8097 if (!done.uptodate) {
8098 /* We might have another mirror, so try again */
8102 offset += sectorsize;
8103 start += sectorsize;
8108 pgoff += sectorsize;
8116 static int btrfs_subio_endio_read(struct inode *inode,
8117 struct btrfs_io_bio *io_bio, int err)
8119 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8123 return __btrfs_correct_data_nocsum(inode, io_bio);
8127 return __btrfs_subio_endio_read(inode, io_bio, err);
8131 static void btrfs_endio_direct_read(struct bio *bio)
8133 struct btrfs_dio_private *dip = bio->bi_private;
8134 struct inode *inode = dip->inode;
8135 struct bio *dio_bio;
8136 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8137 int err = bio->bi_error;
8139 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8140 err = btrfs_subio_endio_read(inode, io_bio, err);
8142 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8143 dip->logical_offset + dip->bytes - 1);
8144 dio_bio = dip->dio_bio;
8148 dio_bio->bi_error = bio->bi_error;
8149 dio_end_io(dio_bio, bio->bi_error);
8152 io_bio->end_io(io_bio, err);
8156 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8161 struct btrfs_root *root = BTRFS_I(inode)->root;
8162 struct btrfs_ordered_extent *ordered = NULL;
8163 u64 ordered_offset = offset;
8164 u64 ordered_bytes = bytes;
8168 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8175 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8176 finish_ordered_fn, NULL, NULL);
8177 btrfs_queue_work(root->fs_info->endio_write_workers,
8181 * our bio might span multiple ordered extents. If we haven't
8182 * completed the accounting for the whole dio, go back and try again
8184 if (ordered_offset < offset + bytes) {
8185 ordered_bytes = offset + bytes - ordered_offset;
8191 static void btrfs_endio_direct_write(struct bio *bio)
8193 struct btrfs_dio_private *dip = bio->bi_private;
8194 struct bio *dio_bio = dip->dio_bio;
8196 btrfs_endio_direct_write_update_ordered(dip->inode,
8197 dip->logical_offset,
8203 dio_bio->bi_error = bio->bi_error;
8204 dio_end_io(dio_bio, bio->bi_error);
8208 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8209 struct bio *bio, int mirror_num,
8210 unsigned long bio_flags, u64 offset)
8213 struct btrfs_root *root = BTRFS_I(inode)->root;
8214 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8215 BUG_ON(ret); /* -ENOMEM */
8219 static void btrfs_end_dio_bio(struct bio *bio)
8221 struct btrfs_dio_private *dip = bio->bi_private;
8222 int err = bio->bi_error;
8225 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8226 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8227 btrfs_ino(dip->inode), bio->bi_rw,
8228 (unsigned long long)bio->bi_iter.bi_sector,
8229 bio->bi_iter.bi_size, err);
8231 if (dip->subio_endio)
8232 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8238 * before atomic variable goto zero, we must make sure
8239 * dip->errors is perceived to be set.
8241 smp_mb__before_atomic();
8244 /* if there are more bios still pending for this dio, just exit */
8245 if (!atomic_dec_and_test(&dip->pending_bios))
8249 bio_io_error(dip->orig_bio);
8251 dip->dio_bio->bi_error = 0;
8252 bio_endio(dip->orig_bio);
8258 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8259 u64 first_sector, gfp_t gfp_flags)
8262 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8264 bio_associate_current(bio);
8268 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8269 struct inode *inode,
8270 struct btrfs_dio_private *dip,
8274 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8275 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8279 * We load all the csum data we need when we submit
8280 * the first bio to reduce the csum tree search and
8283 if (dip->logical_offset == file_offset) {
8284 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8290 if (bio == dip->orig_bio)
8293 file_offset -= dip->logical_offset;
8294 file_offset >>= inode->i_sb->s_blocksize_bits;
8295 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8300 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8301 int rw, u64 file_offset, int skip_sum,
8304 struct btrfs_dio_private *dip = bio->bi_private;
8305 int write = rw & REQ_WRITE;
8306 struct btrfs_root *root = BTRFS_I(inode)->root;
8310 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8315 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8316 BTRFS_WQ_ENDIO_DATA);
8324 if (write && async_submit) {
8325 ret = btrfs_wq_submit_bio(root->fs_info,
8326 inode, rw, bio, 0, 0,
8328 __btrfs_submit_bio_start_direct_io,
8329 __btrfs_submit_bio_done);
8333 * If we aren't doing async submit, calculate the csum of the
8336 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8340 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8346 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8352 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8355 struct inode *inode = dip->inode;
8356 struct btrfs_root *root = BTRFS_I(inode)->root;
8358 struct bio *orig_bio = dip->orig_bio;
8359 struct bio_vec *bvec = orig_bio->bi_io_vec;
8360 u64 start_sector = orig_bio->bi_iter.bi_sector;
8361 u64 file_offset = dip->logical_offset;
8364 u32 blocksize = root->sectorsize;
8365 int async_submit = 0;
8370 map_length = orig_bio->bi_iter.bi_size;
8371 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8372 &map_length, NULL, 0);
8376 if (map_length >= orig_bio->bi_iter.bi_size) {
8378 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8382 /* async crcs make it difficult to collect full stripe writes. */
8383 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8388 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8392 bio->bi_private = dip;
8393 bio->bi_end_io = btrfs_end_dio_bio;
8394 btrfs_io_bio(bio)->logical = file_offset;
8395 atomic_inc(&dip->pending_bios);
8397 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8398 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8401 if (unlikely(map_length < submit_len + blocksize ||
8402 bio_add_page(bio, bvec->bv_page, blocksize,
8403 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8405 * inc the count before we submit the bio so
8406 * we know the end IO handler won't happen before
8407 * we inc the count. Otherwise, the dip might get freed
8408 * before we're done setting it up
8410 atomic_inc(&dip->pending_bios);
8411 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8412 file_offset, skip_sum,
8416 atomic_dec(&dip->pending_bios);
8420 start_sector += submit_len >> 9;
8421 file_offset += submit_len;
8425 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8426 start_sector, GFP_NOFS);
8429 bio->bi_private = dip;
8430 bio->bi_end_io = btrfs_end_dio_bio;
8431 btrfs_io_bio(bio)->logical = file_offset;
8433 map_length = orig_bio->bi_iter.bi_size;
8434 ret = btrfs_map_block(root->fs_info, rw,
8436 &map_length, NULL, 0);
8444 submit_len += blocksize;
8454 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8463 * before atomic variable goto zero, we must
8464 * make sure dip->errors is perceived to be set.
8466 smp_mb__before_atomic();
8467 if (atomic_dec_and_test(&dip->pending_bios))
8468 bio_io_error(dip->orig_bio);
8470 /* bio_end_io() will handle error, so we needn't return it */
8474 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8475 struct inode *inode, loff_t file_offset)
8477 struct btrfs_dio_private *dip = NULL;
8478 struct bio *io_bio = NULL;
8479 struct btrfs_io_bio *btrfs_bio;
8481 int write = rw & REQ_WRITE;
8484 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8486 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8492 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8498 dip->private = dio_bio->bi_private;
8500 dip->logical_offset = file_offset;
8501 dip->bytes = dio_bio->bi_iter.bi_size;
8502 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8503 io_bio->bi_private = dip;
8504 dip->orig_bio = io_bio;
8505 dip->dio_bio = dio_bio;
8506 atomic_set(&dip->pending_bios, 0);
8507 btrfs_bio = btrfs_io_bio(io_bio);
8508 btrfs_bio->logical = file_offset;
8511 io_bio->bi_end_io = btrfs_endio_direct_write;
8513 io_bio->bi_end_io = btrfs_endio_direct_read;
8514 dip->subio_endio = btrfs_subio_endio_read;
8518 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8519 * even if we fail to submit a bio, because in such case we do the
8520 * corresponding error handling below and it must not be done a second
8521 * time by btrfs_direct_IO().
8524 struct btrfs_dio_data *dio_data = current->journal_info;
8526 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8528 dio_data->unsubmitted_oe_range_start =
8529 dio_data->unsubmitted_oe_range_end;
8532 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8536 if (btrfs_bio->end_io)
8537 btrfs_bio->end_io(btrfs_bio, ret);
8541 * If we arrived here it means either we failed to submit the dip
8542 * or we either failed to clone the dio_bio or failed to allocate the
8543 * dip. If we cloned the dio_bio and allocated the dip, we can just
8544 * call bio_endio against our io_bio so that we get proper resource
8545 * cleanup if we fail to submit the dip, otherwise, we must do the
8546 * same as btrfs_endio_direct_[write|read] because we can't call these
8547 * callbacks - they require an allocated dip and a clone of dio_bio.
8549 if (io_bio && dip) {
8550 io_bio->bi_error = -EIO;
8553 * The end io callbacks free our dip, do the final put on io_bio
8554 * and all the cleanup and final put for dio_bio (through
8561 btrfs_endio_direct_write_update_ordered(inode,
8563 dio_bio->bi_iter.bi_size,
8566 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8567 file_offset + dio_bio->bi_iter.bi_size - 1);
8569 dio_bio->bi_error = -EIO;
8571 * Releases and cleans up our dio_bio, no need to bio_put()
8572 * nor bio_endio()/bio_io_error() against dio_bio.
8574 dio_end_io(dio_bio, ret);
8581 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8582 const struct iov_iter *iter, loff_t offset)
8586 unsigned blocksize_mask = root->sectorsize - 1;
8587 ssize_t retval = -EINVAL;
8589 if (offset & blocksize_mask)
8592 if (iov_iter_alignment(iter) & blocksize_mask)
8595 /* If this is a write we don't need to check anymore */
8596 if (iov_iter_rw(iter) == WRITE)
8599 * Check to make sure we don't have duplicate iov_base's in this
8600 * iovec, if so return EINVAL, otherwise we'll get csum errors
8601 * when reading back.
8603 for (seg = 0; seg < iter->nr_segs; seg++) {
8604 for (i = seg + 1; i < iter->nr_segs; i++) {
8605 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8614 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8616 struct file *file = iocb->ki_filp;
8617 struct inode *inode = file->f_mapping->host;
8618 struct btrfs_root *root = BTRFS_I(inode)->root;
8619 struct btrfs_dio_data dio_data = { 0 };
8620 loff_t offset = iocb->ki_pos;
8624 bool relock = false;
8627 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8630 inode_dio_begin(inode);
8631 smp_mb__after_atomic();
8634 * The generic stuff only does filemap_write_and_wait_range, which
8635 * isn't enough if we've written compressed pages to this area, so
8636 * we need to flush the dirty pages again to make absolutely sure
8637 * that any outstanding dirty pages are on disk.
8639 count = iov_iter_count(iter);
8640 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8641 &BTRFS_I(inode)->runtime_flags))
8642 filemap_fdatawrite_range(inode->i_mapping, offset,
8643 offset + count - 1);
8645 if (iov_iter_rw(iter) == WRITE) {
8647 * If the write DIO is beyond the EOF, we need update
8648 * the isize, but it is protected by i_mutex. So we can
8649 * not unlock the i_mutex at this case.
8651 if (offset + count <= inode->i_size) {
8652 inode_unlock(inode);
8655 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8658 dio_data.outstanding_extents = div64_u64(count +
8659 BTRFS_MAX_EXTENT_SIZE - 1,
8660 BTRFS_MAX_EXTENT_SIZE);
8663 * We need to know how many extents we reserved so that we can
8664 * do the accounting properly if we go over the number we
8665 * originally calculated. Abuse current->journal_info for this.
8667 dio_data.reserve = round_up(count, root->sectorsize);
8668 dio_data.unsubmitted_oe_range_start = (u64)offset;
8669 dio_data.unsubmitted_oe_range_end = (u64)offset;
8670 current->journal_info = &dio_data;
8671 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8672 &BTRFS_I(inode)->runtime_flags)) {
8673 inode_dio_end(inode);
8674 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8678 ret = __blockdev_direct_IO(iocb, inode,
8679 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8680 iter, btrfs_get_blocks_direct, NULL,
8681 btrfs_submit_direct, flags);
8682 if (iov_iter_rw(iter) == WRITE) {
8683 current->journal_info = NULL;
8684 if (ret < 0 && ret != -EIOCBQUEUED) {
8685 if (dio_data.reserve)
8686 btrfs_delalloc_release_space(inode, offset,
8689 * On error we might have left some ordered extents
8690 * without submitting corresponding bios for them, so
8691 * cleanup them up to avoid other tasks getting them
8692 * and waiting for them to complete forever.
8694 if (dio_data.unsubmitted_oe_range_start <
8695 dio_data.unsubmitted_oe_range_end)
8696 btrfs_endio_direct_write_update_ordered(inode,
8697 dio_data.unsubmitted_oe_range_start,
8698 dio_data.unsubmitted_oe_range_end -
8699 dio_data.unsubmitted_oe_range_start,
8701 } else if (ret >= 0 && (size_t)ret < count)
8702 btrfs_delalloc_release_space(inode, offset,
8703 count - (size_t)ret);
8707 inode_dio_end(inode);
8714 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8716 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8717 __u64 start, __u64 len)
8721 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8725 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8728 int btrfs_readpage(struct file *file, struct page *page)
8730 struct extent_io_tree *tree;
8731 tree = &BTRFS_I(page->mapping->host)->io_tree;
8732 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8735 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8737 struct extent_io_tree *tree;
8738 struct inode *inode = page->mapping->host;
8741 if (current->flags & PF_MEMALLOC) {
8742 redirty_page_for_writepage(wbc, page);
8748 * If we are under memory pressure we will call this directly from the
8749 * VM, we need to make sure we have the inode referenced for the ordered
8750 * extent. If not just return like we didn't do anything.
8752 if (!igrab(inode)) {
8753 redirty_page_for_writepage(wbc, page);
8754 return AOP_WRITEPAGE_ACTIVATE;
8756 tree = &BTRFS_I(page->mapping->host)->io_tree;
8757 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8758 btrfs_add_delayed_iput(inode);
8762 static int btrfs_writepages(struct address_space *mapping,
8763 struct writeback_control *wbc)
8765 struct extent_io_tree *tree;
8767 tree = &BTRFS_I(mapping->host)->io_tree;
8768 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8772 btrfs_readpages(struct file *file, struct address_space *mapping,
8773 struct list_head *pages, unsigned nr_pages)
8775 struct extent_io_tree *tree;
8776 tree = &BTRFS_I(mapping->host)->io_tree;
8777 return extent_readpages(tree, mapping, pages, nr_pages,
8780 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8782 struct extent_io_tree *tree;
8783 struct extent_map_tree *map;
8786 tree = &BTRFS_I(page->mapping->host)->io_tree;
8787 map = &BTRFS_I(page->mapping->host)->extent_tree;
8788 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8790 ClearPagePrivate(page);
8791 set_page_private(page, 0);
8797 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8799 if (PageWriteback(page) || PageDirty(page))
8801 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8804 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8805 unsigned int length)
8807 struct inode *inode = page->mapping->host;
8808 struct extent_io_tree *tree;
8809 struct btrfs_ordered_extent *ordered;
8810 struct extent_state *cached_state = NULL;
8811 u64 page_start = page_offset(page);
8812 u64 page_end = page_start + PAGE_SIZE - 1;
8815 int inode_evicting = inode->i_state & I_FREEING;
8818 * we have the page locked, so new writeback can't start,
8819 * and the dirty bit won't be cleared while we are here.
8821 * Wait for IO on this page so that we can safely clear
8822 * the PagePrivate2 bit and do ordered accounting
8824 wait_on_page_writeback(page);
8826 tree = &BTRFS_I(inode)->io_tree;
8828 btrfs_releasepage(page, GFP_NOFS);
8832 if (!inode_evicting)
8833 lock_extent_bits(tree, page_start, page_end, &cached_state);
8836 ordered = btrfs_lookup_ordered_range(inode, start,
8837 page_end - start + 1);
8839 end = min(page_end, ordered->file_offset + ordered->len - 1);
8841 * IO on this page will never be started, so we need
8842 * to account for any ordered extents now
8844 if (!inode_evicting)
8845 clear_extent_bit(tree, start, end,
8846 EXTENT_DIRTY | EXTENT_DELALLOC |
8847 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8848 EXTENT_DEFRAG, 1, 0, &cached_state,
8851 * whoever cleared the private bit is responsible
8852 * for the finish_ordered_io
8854 if (TestClearPagePrivate2(page)) {
8855 struct btrfs_ordered_inode_tree *tree;
8858 tree = &BTRFS_I(inode)->ordered_tree;
8860 spin_lock_irq(&tree->lock);
8861 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8862 new_len = start - ordered->file_offset;
8863 if (new_len < ordered->truncated_len)
8864 ordered->truncated_len = new_len;
8865 spin_unlock_irq(&tree->lock);
8867 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8869 end - start + 1, 1))
8870 btrfs_finish_ordered_io(ordered);
8872 btrfs_put_ordered_extent(ordered);
8873 if (!inode_evicting) {
8874 cached_state = NULL;
8875 lock_extent_bits(tree, start, end,
8880 if (start < page_end)
8885 * Qgroup reserved space handler
8886 * Page here will be either
8887 * 1) Already written to disk
8888 * In this case, its reserved space is released from data rsv map
8889 * and will be freed by delayed_ref handler finally.
8890 * So even we call qgroup_free_data(), it won't decrease reserved
8892 * 2) Not written to disk
8893 * This means the reserved space should be freed here.
8895 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8896 if (!inode_evicting) {
8897 clear_extent_bit(tree, page_start, page_end,
8898 EXTENT_LOCKED | EXTENT_DIRTY |
8899 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8900 EXTENT_DEFRAG, 1, 1,
8901 &cached_state, GFP_NOFS);
8903 __btrfs_releasepage(page, GFP_NOFS);
8906 ClearPageChecked(page);
8907 if (PagePrivate(page)) {
8908 ClearPagePrivate(page);
8909 set_page_private(page, 0);
8915 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8916 * called from a page fault handler when a page is first dirtied. Hence we must
8917 * be careful to check for EOF conditions here. We set the page up correctly
8918 * for a written page which means we get ENOSPC checking when writing into
8919 * holes and correct delalloc and unwritten extent mapping on filesystems that
8920 * support these features.
8922 * We are not allowed to take the i_mutex here so we have to play games to
8923 * protect against truncate races as the page could now be beyond EOF. Because
8924 * vmtruncate() writes the inode size before removing pages, once we have the
8925 * page lock we can determine safely if the page is beyond EOF. If it is not
8926 * beyond EOF, then the page is guaranteed safe against truncation until we
8929 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8931 struct page *page = vmf->page;
8932 struct inode *inode = file_inode(vma->vm_file);
8933 struct btrfs_root *root = BTRFS_I(inode)->root;
8934 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8935 struct btrfs_ordered_extent *ordered;
8936 struct extent_state *cached_state = NULL;
8938 unsigned long zero_start;
8947 reserved_space = PAGE_SIZE;
8949 sb_start_pagefault(inode->i_sb);
8950 page_start = page_offset(page);
8951 page_end = page_start + PAGE_SIZE - 1;
8955 * Reserving delalloc space after obtaining the page lock can lead to
8956 * deadlock. For example, if a dirty page is locked by this function
8957 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8958 * dirty page write out, then the btrfs_writepage() function could
8959 * end up waiting indefinitely to get a lock on the page currently
8960 * being processed by btrfs_page_mkwrite() function.
8962 ret = btrfs_delalloc_reserve_space(inode, page_start,
8965 ret = file_update_time(vma->vm_file);
8971 else /* -ENOSPC, -EIO, etc */
8972 ret = VM_FAULT_SIGBUS;
8978 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8981 size = i_size_read(inode);
8983 if ((page->mapping != inode->i_mapping) ||
8984 (page_start >= size)) {
8985 /* page got truncated out from underneath us */
8988 wait_on_page_writeback(page);
8990 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8991 set_page_extent_mapped(page);
8994 * we can't set the delalloc bits if there are pending ordered
8995 * extents. Drop our locks and wait for them to finish
8997 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8999 unlock_extent_cached(io_tree, page_start, page_end,
9000 &cached_state, GFP_NOFS);
9002 btrfs_start_ordered_extent(inode, ordered, 1);
9003 btrfs_put_ordered_extent(ordered);
9007 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9008 reserved_space = round_up(size - page_start, root->sectorsize);
9009 if (reserved_space < PAGE_SIZE) {
9010 end = page_start + reserved_space - 1;
9011 spin_lock(&BTRFS_I(inode)->lock);
9012 BTRFS_I(inode)->outstanding_extents++;
9013 spin_unlock(&BTRFS_I(inode)->lock);
9014 btrfs_delalloc_release_space(inode, page_start,
9015 PAGE_SIZE - reserved_space);
9020 * XXX - page_mkwrite gets called every time the page is dirtied, even
9021 * if it was already dirty, so for space accounting reasons we need to
9022 * clear any delalloc bits for the range we are fixing to save. There
9023 * is probably a better way to do this, but for now keep consistent with
9024 * prepare_pages in the normal write path.
9026 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9027 EXTENT_DIRTY | EXTENT_DELALLOC |
9028 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9029 0, 0, &cached_state, GFP_NOFS);
9031 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9034 unlock_extent_cached(io_tree, page_start, page_end,
9035 &cached_state, GFP_NOFS);
9036 ret = VM_FAULT_SIGBUS;
9041 /* page is wholly or partially inside EOF */
9042 if (page_start + PAGE_SIZE > size)
9043 zero_start = size & ~PAGE_MASK;
9045 zero_start = PAGE_SIZE;
9047 if (zero_start != PAGE_SIZE) {
9049 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9050 flush_dcache_page(page);
9053 ClearPageChecked(page);
9054 set_page_dirty(page);
9055 SetPageUptodate(page);
9057 BTRFS_I(inode)->last_trans = root->fs_info->generation;
9058 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9059 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9061 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9065 sb_end_pagefault(inode->i_sb);
9066 return VM_FAULT_LOCKED;
9070 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9072 sb_end_pagefault(inode->i_sb);
9076 static int btrfs_truncate(struct inode *inode)
9078 struct btrfs_root *root = BTRFS_I(inode)->root;
9079 struct btrfs_block_rsv *rsv;
9082 struct btrfs_trans_handle *trans;
9083 u64 mask = root->sectorsize - 1;
9084 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9086 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9092 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9093 * 3 things going on here
9095 * 1) We need to reserve space for our orphan item and the space to
9096 * delete our orphan item. Lord knows we don't want to have a dangling
9097 * orphan item because we didn't reserve space to remove it.
9099 * 2) We need to reserve space to update our inode.
9101 * 3) We need to have something to cache all the space that is going to
9102 * be free'd up by the truncate operation, but also have some slack
9103 * space reserved in case it uses space during the truncate (thank you
9104 * very much snapshotting).
9106 * And we need these to all be separate. The fact is we can use a lot of
9107 * space doing the truncate, and we have no earthly idea how much space
9108 * we will use, so we need the truncate reservation to be separate so it
9109 * doesn't end up using space reserved for updating the inode or
9110 * removing the orphan item. We also need to be able to stop the
9111 * transaction and start a new one, which means we need to be able to
9112 * update the inode several times, and we have no idea of knowing how
9113 * many times that will be, so we can't just reserve 1 item for the
9114 * entirety of the operation, so that has to be done separately as well.
9115 * Then there is the orphan item, which does indeed need to be held on
9116 * to for the whole operation, and we need nobody to touch this reserved
9117 * space except the orphan code.
9119 * So that leaves us with
9121 * 1) root->orphan_block_rsv - for the orphan deletion.
9122 * 2) rsv - for the truncate reservation, which we will steal from the
9123 * transaction reservation.
9124 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9125 * updating the inode.
9127 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9130 rsv->size = min_size;
9134 * 1 for the truncate slack space
9135 * 1 for updating the inode.
9137 trans = btrfs_start_transaction(root, 2);
9138 if (IS_ERR(trans)) {
9139 err = PTR_ERR(trans);
9143 /* Migrate the slack space for the truncate to our reserve */
9144 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9149 * So if we truncate and then write and fsync we normally would just
9150 * write the extents that changed, which is a problem if we need to
9151 * first truncate that entire inode. So set this flag so we write out
9152 * all of the extents in the inode to the sync log so we're completely
9155 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9156 trans->block_rsv = rsv;
9159 ret = btrfs_truncate_inode_items(trans, root, inode,
9161 BTRFS_EXTENT_DATA_KEY);
9162 if (ret != -ENOSPC && ret != -EAGAIN) {
9167 trans->block_rsv = &root->fs_info->trans_block_rsv;
9168 ret = btrfs_update_inode(trans, root, inode);
9174 btrfs_end_transaction(trans, root);
9175 btrfs_btree_balance_dirty(root);
9177 trans = btrfs_start_transaction(root, 2);
9178 if (IS_ERR(trans)) {
9179 ret = err = PTR_ERR(trans);
9184 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9186 BUG_ON(ret); /* shouldn't happen */
9187 trans->block_rsv = rsv;
9190 if (ret == 0 && inode->i_nlink > 0) {
9191 trans->block_rsv = root->orphan_block_rsv;
9192 ret = btrfs_orphan_del(trans, inode);
9198 trans->block_rsv = &root->fs_info->trans_block_rsv;
9199 ret = btrfs_update_inode(trans, root, inode);
9203 ret = btrfs_end_transaction(trans, root);
9204 btrfs_btree_balance_dirty(root);
9207 btrfs_free_block_rsv(root, rsv);
9216 * create a new subvolume directory/inode (helper for the ioctl).
9218 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9219 struct btrfs_root *new_root,
9220 struct btrfs_root *parent_root,
9223 struct inode *inode;
9227 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9228 new_dirid, new_dirid,
9229 S_IFDIR | (~current_umask() & S_IRWXUGO),
9232 return PTR_ERR(inode);
9233 inode->i_op = &btrfs_dir_inode_operations;
9234 inode->i_fop = &btrfs_dir_file_operations;
9236 set_nlink(inode, 1);
9237 btrfs_i_size_write(inode, 0);
9238 unlock_new_inode(inode);
9240 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9242 btrfs_err(new_root->fs_info,
9243 "error inheriting subvolume %llu properties: %d",
9244 new_root->root_key.objectid, err);
9246 err = btrfs_update_inode(trans, new_root, inode);
9252 struct inode *btrfs_alloc_inode(struct super_block *sb)
9254 struct btrfs_inode *ei;
9255 struct inode *inode;
9257 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9264 ei->last_sub_trans = 0;
9265 ei->logged_trans = 0;
9266 ei->delalloc_bytes = 0;
9267 ei->defrag_bytes = 0;
9268 ei->disk_i_size = 0;
9271 ei->index_cnt = (u64)-1;
9273 ei->last_unlink_trans = 0;
9274 ei->last_log_commit = 0;
9275 ei->delayed_iput_count = 0;
9277 spin_lock_init(&ei->lock);
9278 ei->outstanding_extents = 0;
9279 ei->reserved_extents = 0;
9281 ei->runtime_flags = 0;
9282 ei->force_compress = BTRFS_COMPRESS_NONE;
9284 ei->delayed_node = NULL;
9286 ei->i_otime.tv_sec = 0;
9287 ei->i_otime.tv_nsec = 0;
9289 inode = &ei->vfs_inode;
9290 extent_map_tree_init(&ei->extent_tree);
9291 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9292 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9293 ei->io_tree.track_uptodate = 1;
9294 ei->io_failure_tree.track_uptodate = 1;
9295 atomic_set(&ei->sync_writers, 0);
9296 mutex_init(&ei->log_mutex);
9297 mutex_init(&ei->delalloc_mutex);
9298 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9299 INIT_LIST_HEAD(&ei->delalloc_inodes);
9300 INIT_LIST_HEAD(&ei->delayed_iput);
9301 RB_CLEAR_NODE(&ei->rb_node);
9302 init_rwsem(&ei->dio_sem);
9307 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9308 void btrfs_test_destroy_inode(struct inode *inode)
9310 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9311 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9315 static void btrfs_i_callback(struct rcu_head *head)
9317 struct inode *inode = container_of(head, struct inode, i_rcu);
9318 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9321 void btrfs_destroy_inode(struct inode *inode)
9323 struct btrfs_ordered_extent *ordered;
9324 struct btrfs_root *root = BTRFS_I(inode)->root;
9326 WARN_ON(!hlist_empty(&inode->i_dentry));
9327 WARN_ON(inode->i_data.nrpages);
9328 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9329 WARN_ON(BTRFS_I(inode)->reserved_extents);
9330 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9331 WARN_ON(BTRFS_I(inode)->csum_bytes);
9332 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9335 * This can happen where we create an inode, but somebody else also
9336 * created the same inode and we need to destroy the one we already
9342 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9343 &BTRFS_I(inode)->runtime_flags)) {
9344 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9346 atomic_dec(&root->orphan_inodes);
9350 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9354 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9355 ordered->file_offset, ordered->len);
9356 btrfs_remove_ordered_extent(inode, ordered);
9357 btrfs_put_ordered_extent(ordered);
9358 btrfs_put_ordered_extent(ordered);
9361 btrfs_qgroup_check_reserved_leak(inode);
9362 inode_tree_del(inode);
9363 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9365 call_rcu(&inode->i_rcu, btrfs_i_callback);
9368 int btrfs_drop_inode(struct inode *inode)
9370 struct btrfs_root *root = BTRFS_I(inode)->root;
9375 /* the snap/subvol tree is on deleting */
9376 if (btrfs_root_refs(&root->root_item) == 0)
9379 return generic_drop_inode(inode);
9382 static void init_once(void *foo)
9384 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9386 inode_init_once(&ei->vfs_inode);
9389 void btrfs_destroy_cachep(void)
9392 * Make sure all delayed rcu free inodes are flushed before we
9396 kmem_cache_destroy(btrfs_inode_cachep);
9397 kmem_cache_destroy(btrfs_trans_handle_cachep);
9398 kmem_cache_destroy(btrfs_transaction_cachep);
9399 kmem_cache_destroy(btrfs_path_cachep);
9400 kmem_cache_destroy(btrfs_free_space_cachep);
9403 int btrfs_init_cachep(void)
9405 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9406 sizeof(struct btrfs_inode), 0,
9407 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9409 if (!btrfs_inode_cachep)
9412 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9413 sizeof(struct btrfs_trans_handle), 0,
9414 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9415 if (!btrfs_trans_handle_cachep)
9418 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9419 sizeof(struct btrfs_transaction), 0,
9420 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9421 if (!btrfs_transaction_cachep)
9424 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9425 sizeof(struct btrfs_path), 0,
9426 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9427 if (!btrfs_path_cachep)
9430 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9431 sizeof(struct btrfs_free_space), 0,
9432 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9433 if (!btrfs_free_space_cachep)
9438 btrfs_destroy_cachep();
9442 static int btrfs_getattr(struct vfsmount *mnt,
9443 struct dentry *dentry, struct kstat *stat)
9446 struct inode *inode = d_inode(dentry);
9447 u32 blocksize = inode->i_sb->s_blocksize;
9449 generic_fillattr(inode, stat);
9450 stat->dev = BTRFS_I(inode)->root->anon_dev;
9452 spin_lock(&BTRFS_I(inode)->lock);
9453 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9454 spin_unlock(&BTRFS_I(inode)->lock);
9455 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9456 ALIGN(delalloc_bytes, blocksize)) >> 9;
9460 static int btrfs_rename_exchange(struct inode *old_dir,
9461 struct dentry *old_dentry,
9462 struct inode *new_dir,
9463 struct dentry *new_dentry)
9465 struct btrfs_trans_handle *trans;
9466 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9467 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9468 struct inode *new_inode = new_dentry->d_inode;
9469 struct inode *old_inode = old_dentry->d_inode;
9470 struct timespec ctime = CURRENT_TIME;
9471 struct dentry *parent;
9472 u64 old_ino = btrfs_ino(old_inode);
9473 u64 new_ino = btrfs_ino(new_inode);
9478 bool root_log_pinned = false;
9479 bool dest_log_pinned = false;
9481 /* we only allow rename subvolume link between subvolumes */
9482 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9485 /* close the race window with snapshot create/destroy ioctl */
9486 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9487 down_read(&root->fs_info->subvol_sem);
9488 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9489 down_read(&dest->fs_info->subvol_sem);
9492 * We want to reserve the absolute worst case amount of items. So if
9493 * both inodes are subvols and we need to unlink them then that would
9494 * require 4 item modifications, but if they are both normal inodes it
9495 * would require 5 item modifications, so we'll assume their normal
9496 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9497 * should cover the worst case number of items we'll modify.
9499 trans = btrfs_start_transaction(root, 12);
9500 if (IS_ERR(trans)) {
9501 ret = PTR_ERR(trans);
9506 * We need to find a free sequence number both in the source and
9507 * in the destination directory for the exchange.
9509 ret = btrfs_set_inode_index(new_dir, &old_idx);
9512 ret = btrfs_set_inode_index(old_dir, &new_idx);
9516 BTRFS_I(old_inode)->dir_index = 0ULL;
9517 BTRFS_I(new_inode)->dir_index = 0ULL;
9519 /* Reference for the source. */
9520 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9521 /* force full log commit if subvolume involved. */
9522 btrfs_set_log_full_commit(root->fs_info, trans);
9524 btrfs_pin_log_trans(root);
9525 root_log_pinned = true;
9526 ret = btrfs_insert_inode_ref(trans, dest,
9527 new_dentry->d_name.name,
9528 new_dentry->d_name.len,
9530 btrfs_ino(new_dir), old_idx);
9535 /* And now for the dest. */
9536 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9537 /* force full log commit if subvolume involved. */
9538 btrfs_set_log_full_commit(dest->fs_info, trans);
9540 btrfs_pin_log_trans(dest);
9541 dest_log_pinned = true;
9542 ret = btrfs_insert_inode_ref(trans, root,
9543 old_dentry->d_name.name,
9544 old_dentry->d_name.len,
9546 btrfs_ino(old_dir), new_idx);
9551 /* Update inode version and ctime/mtime. */
9552 inode_inc_iversion(old_dir);
9553 inode_inc_iversion(new_dir);
9554 inode_inc_iversion(old_inode);
9555 inode_inc_iversion(new_inode);
9556 old_dir->i_ctime = old_dir->i_mtime = ctime;
9557 new_dir->i_ctime = new_dir->i_mtime = ctime;
9558 old_inode->i_ctime = ctime;
9559 new_inode->i_ctime = ctime;
9561 if (old_dentry->d_parent != new_dentry->d_parent) {
9562 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9563 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9566 /* src is a subvolume */
9567 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9568 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9569 ret = btrfs_unlink_subvol(trans, root, old_dir,
9571 old_dentry->d_name.name,
9572 old_dentry->d_name.len);
9573 } else { /* src is an inode */
9574 ret = __btrfs_unlink_inode(trans, root, old_dir,
9575 old_dentry->d_inode,
9576 old_dentry->d_name.name,
9577 old_dentry->d_name.len);
9579 ret = btrfs_update_inode(trans, root, old_inode);
9582 btrfs_abort_transaction(trans, root, ret);
9586 /* dest is a subvolume */
9587 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9588 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9589 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9591 new_dentry->d_name.name,
9592 new_dentry->d_name.len);
9593 } else { /* dest is an inode */
9594 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9595 new_dentry->d_inode,
9596 new_dentry->d_name.name,
9597 new_dentry->d_name.len);
9599 ret = btrfs_update_inode(trans, dest, new_inode);
9602 btrfs_abort_transaction(trans, root, ret);
9606 ret = btrfs_add_link(trans, new_dir, old_inode,
9607 new_dentry->d_name.name,
9608 new_dentry->d_name.len, 0, old_idx);
9610 btrfs_abort_transaction(trans, root, ret);
9614 ret = btrfs_add_link(trans, old_dir, new_inode,
9615 old_dentry->d_name.name,
9616 old_dentry->d_name.len, 0, new_idx);
9618 btrfs_abort_transaction(trans, root, ret);
9622 if (old_inode->i_nlink == 1)
9623 BTRFS_I(old_inode)->dir_index = old_idx;
9624 if (new_inode->i_nlink == 1)
9625 BTRFS_I(new_inode)->dir_index = new_idx;
9627 if (root_log_pinned) {
9628 parent = new_dentry->d_parent;
9629 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9630 btrfs_end_log_trans(root);
9631 root_log_pinned = false;
9633 if (dest_log_pinned) {
9634 parent = old_dentry->d_parent;
9635 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9636 btrfs_end_log_trans(dest);
9637 dest_log_pinned = false;
9641 * If we have pinned a log and an error happened, we unpin tasks
9642 * trying to sync the log and force them to fallback to a transaction
9643 * commit if the log currently contains any of the inodes involved in
9644 * this rename operation (to ensure we do not persist a log with an
9645 * inconsistent state for any of these inodes or leading to any
9646 * inconsistencies when replayed). If the transaction was aborted, the
9647 * abortion reason is propagated to userspace when attempting to commit
9648 * the transaction. If the log does not contain any of these inodes, we
9649 * allow the tasks to sync it.
9651 if (ret && (root_log_pinned || dest_log_pinned)) {
9652 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9653 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9654 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9656 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9657 btrfs_set_log_full_commit(root->fs_info, trans);
9659 if (root_log_pinned) {
9660 btrfs_end_log_trans(root);
9661 root_log_pinned = false;
9663 if (dest_log_pinned) {
9664 btrfs_end_log_trans(dest);
9665 dest_log_pinned = false;
9668 ret = btrfs_end_transaction(trans, root);
9670 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9671 up_read(&dest->fs_info->subvol_sem);
9672 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9673 up_read(&root->fs_info->subvol_sem);
9678 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9679 struct btrfs_root *root,
9681 struct dentry *dentry)
9684 struct inode *inode;
9688 ret = btrfs_find_free_ino(root, &objectid);
9692 inode = btrfs_new_inode(trans, root, dir,
9693 dentry->d_name.name,
9697 S_IFCHR | WHITEOUT_MODE,
9700 if (IS_ERR(inode)) {
9701 ret = PTR_ERR(inode);
9705 inode->i_op = &btrfs_special_inode_operations;
9706 init_special_inode(inode, inode->i_mode,
9709 ret = btrfs_init_inode_security(trans, inode, dir,
9714 ret = btrfs_add_nondir(trans, dir, dentry,
9719 ret = btrfs_update_inode(trans, root, inode);
9721 unlock_new_inode(inode);
9723 inode_dec_link_count(inode);
9729 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9730 struct inode *new_dir, struct dentry *new_dentry,
9733 struct btrfs_trans_handle *trans;
9734 unsigned int trans_num_items;
9735 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9736 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9737 struct inode *new_inode = d_inode(new_dentry);
9738 struct inode *old_inode = d_inode(old_dentry);
9742 u64 old_ino = btrfs_ino(old_inode);
9743 bool log_pinned = false;
9745 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9748 /* we only allow rename subvolume link between subvolumes */
9749 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9752 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9753 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9756 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9757 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9761 /* check for collisions, even if the name isn't there */
9762 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9763 new_dentry->d_name.name,
9764 new_dentry->d_name.len);
9767 if (ret == -EEXIST) {
9769 * eexist without a new_inode */
9770 if (WARN_ON(!new_inode)) {
9774 /* maybe -EOVERFLOW */
9781 * we're using rename to replace one file with another. Start IO on it
9782 * now so we don't add too much work to the end of the transaction
9784 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9785 filemap_flush(old_inode->i_mapping);
9787 /* close the racy window with snapshot create/destroy ioctl */
9788 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9789 down_read(&root->fs_info->subvol_sem);
9791 * We want to reserve the absolute worst case amount of items. So if
9792 * both inodes are subvols and we need to unlink them then that would
9793 * require 4 item modifications, but if they are both normal inodes it
9794 * would require 5 item modifications, so we'll assume they are normal
9795 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9796 * should cover the worst case number of items we'll modify.
9797 * If our rename has the whiteout flag, we need more 5 units for the
9798 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9799 * when selinux is enabled).
9801 trans_num_items = 11;
9802 if (flags & RENAME_WHITEOUT)
9803 trans_num_items += 5;
9804 trans = btrfs_start_transaction(root, trans_num_items);
9805 if (IS_ERR(trans)) {
9806 ret = PTR_ERR(trans);
9811 btrfs_record_root_in_trans(trans, dest);
9813 ret = btrfs_set_inode_index(new_dir, &index);
9817 BTRFS_I(old_inode)->dir_index = 0ULL;
9818 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9819 /* force full log commit if subvolume involved. */
9820 btrfs_set_log_full_commit(root->fs_info, trans);
9822 btrfs_pin_log_trans(root);
9824 ret = btrfs_insert_inode_ref(trans, dest,
9825 new_dentry->d_name.name,
9826 new_dentry->d_name.len,
9828 btrfs_ino(new_dir), index);
9833 inode_inc_iversion(old_dir);
9834 inode_inc_iversion(new_dir);
9835 inode_inc_iversion(old_inode);
9836 old_dir->i_ctime = old_dir->i_mtime =
9837 new_dir->i_ctime = new_dir->i_mtime =
9838 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9840 if (old_dentry->d_parent != new_dentry->d_parent)
9841 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9843 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9844 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9845 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9846 old_dentry->d_name.name,
9847 old_dentry->d_name.len);
9849 ret = __btrfs_unlink_inode(trans, root, old_dir,
9850 d_inode(old_dentry),
9851 old_dentry->d_name.name,
9852 old_dentry->d_name.len);
9854 ret = btrfs_update_inode(trans, root, old_inode);
9857 btrfs_abort_transaction(trans, root, ret);
9862 inode_inc_iversion(new_inode);
9863 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9864 if (unlikely(btrfs_ino(new_inode) ==
9865 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9866 root_objectid = BTRFS_I(new_inode)->location.objectid;
9867 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9869 new_dentry->d_name.name,
9870 new_dentry->d_name.len);
9871 BUG_ON(new_inode->i_nlink == 0);
9873 ret = btrfs_unlink_inode(trans, dest, new_dir,
9874 d_inode(new_dentry),
9875 new_dentry->d_name.name,
9876 new_dentry->d_name.len);
9878 if (!ret && new_inode->i_nlink == 0)
9879 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9881 btrfs_abort_transaction(trans, root, ret);
9886 ret = btrfs_add_link(trans, new_dir, old_inode,
9887 new_dentry->d_name.name,
9888 new_dentry->d_name.len, 0, index);
9890 btrfs_abort_transaction(trans, root, ret);
9894 if (old_inode->i_nlink == 1)
9895 BTRFS_I(old_inode)->dir_index = index;
9898 struct dentry *parent = new_dentry->d_parent;
9900 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9901 btrfs_end_log_trans(root);
9905 if (flags & RENAME_WHITEOUT) {
9906 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9910 btrfs_abort_transaction(trans, root, ret);
9916 * If we have pinned the log and an error happened, we unpin tasks
9917 * trying to sync the log and force them to fallback to a transaction
9918 * commit if the log currently contains any of the inodes involved in
9919 * this rename operation (to ensure we do not persist a log with an
9920 * inconsistent state for any of these inodes or leading to any
9921 * inconsistencies when replayed). If the transaction was aborted, the
9922 * abortion reason is propagated to userspace when attempting to commit
9923 * the transaction. If the log does not contain any of these inodes, we
9924 * allow the tasks to sync it.
9926 if (ret && log_pinned) {
9927 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9928 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9929 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9931 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9932 btrfs_set_log_full_commit(root->fs_info, trans);
9934 btrfs_end_log_trans(root);
9937 btrfs_end_transaction(trans, root);
9939 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9940 up_read(&root->fs_info->subvol_sem);
9945 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9946 struct inode *new_dir, struct dentry *new_dentry,
9949 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9952 if (flags & RENAME_EXCHANGE)
9953 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9956 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9959 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9961 struct btrfs_delalloc_work *delalloc_work;
9962 struct inode *inode;
9964 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9966 inode = delalloc_work->inode;
9967 filemap_flush(inode->i_mapping);
9968 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9969 &BTRFS_I(inode)->runtime_flags))
9970 filemap_flush(inode->i_mapping);
9972 if (delalloc_work->delay_iput)
9973 btrfs_add_delayed_iput(inode);
9976 complete(&delalloc_work->completion);
9979 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9982 struct btrfs_delalloc_work *work;
9984 work = kmalloc(sizeof(*work), GFP_NOFS);
9988 init_completion(&work->completion);
9989 INIT_LIST_HEAD(&work->list);
9990 work->inode = inode;
9991 work->delay_iput = delay_iput;
9992 WARN_ON_ONCE(!inode);
9993 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9994 btrfs_run_delalloc_work, NULL, NULL);
9999 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10001 wait_for_completion(&work->completion);
10006 * some fairly slow code that needs optimization. This walks the list
10007 * of all the inodes with pending delalloc and forces them to disk.
10009 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10012 struct btrfs_inode *binode;
10013 struct inode *inode;
10014 struct btrfs_delalloc_work *work, *next;
10015 struct list_head works;
10016 struct list_head splice;
10019 INIT_LIST_HEAD(&works);
10020 INIT_LIST_HEAD(&splice);
10022 mutex_lock(&root->delalloc_mutex);
10023 spin_lock(&root->delalloc_lock);
10024 list_splice_init(&root->delalloc_inodes, &splice);
10025 while (!list_empty(&splice)) {
10026 binode = list_entry(splice.next, struct btrfs_inode,
10029 list_move_tail(&binode->delalloc_inodes,
10030 &root->delalloc_inodes);
10031 inode = igrab(&binode->vfs_inode);
10033 cond_resched_lock(&root->delalloc_lock);
10036 spin_unlock(&root->delalloc_lock);
10038 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10041 btrfs_add_delayed_iput(inode);
10047 list_add_tail(&work->list, &works);
10048 btrfs_queue_work(root->fs_info->flush_workers,
10051 if (nr != -1 && ret >= nr)
10054 spin_lock(&root->delalloc_lock);
10056 spin_unlock(&root->delalloc_lock);
10059 list_for_each_entry_safe(work, next, &works, list) {
10060 list_del_init(&work->list);
10061 btrfs_wait_and_free_delalloc_work(work);
10064 if (!list_empty_careful(&splice)) {
10065 spin_lock(&root->delalloc_lock);
10066 list_splice_tail(&splice, &root->delalloc_inodes);
10067 spin_unlock(&root->delalloc_lock);
10069 mutex_unlock(&root->delalloc_mutex);
10073 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10077 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
10080 ret = __start_delalloc_inodes(root, delay_iput, -1);
10084 * the filemap_flush will queue IO into the worker threads, but
10085 * we have to make sure the IO is actually started and that
10086 * ordered extents get created before we return
10088 atomic_inc(&root->fs_info->async_submit_draining);
10089 while (atomic_read(&root->fs_info->nr_async_submits) ||
10090 atomic_read(&root->fs_info->async_delalloc_pages)) {
10091 wait_event(root->fs_info->async_submit_wait,
10092 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
10093 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
10095 atomic_dec(&root->fs_info->async_submit_draining);
10099 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10102 struct btrfs_root *root;
10103 struct list_head splice;
10106 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10109 INIT_LIST_HEAD(&splice);
10111 mutex_lock(&fs_info->delalloc_root_mutex);
10112 spin_lock(&fs_info->delalloc_root_lock);
10113 list_splice_init(&fs_info->delalloc_roots, &splice);
10114 while (!list_empty(&splice) && nr) {
10115 root = list_first_entry(&splice, struct btrfs_root,
10117 root = btrfs_grab_fs_root(root);
10119 list_move_tail(&root->delalloc_root,
10120 &fs_info->delalloc_roots);
10121 spin_unlock(&fs_info->delalloc_root_lock);
10123 ret = __start_delalloc_inodes(root, delay_iput, nr);
10124 btrfs_put_fs_root(root);
10132 spin_lock(&fs_info->delalloc_root_lock);
10134 spin_unlock(&fs_info->delalloc_root_lock);
10137 atomic_inc(&fs_info->async_submit_draining);
10138 while (atomic_read(&fs_info->nr_async_submits) ||
10139 atomic_read(&fs_info->async_delalloc_pages)) {
10140 wait_event(fs_info->async_submit_wait,
10141 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10142 atomic_read(&fs_info->async_delalloc_pages) == 0));
10144 atomic_dec(&fs_info->async_submit_draining);
10146 if (!list_empty_careful(&splice)) {
10147 spin_lock(&fs_info->delalloc_root_lock);
10148 list_splice_tail(&splice, &fs_info->delalloc_roots);
10149 spin_unlock(&fs_info->delalloc_root_lock);
10151 mutex_unlock(&fs_info->delalloc_root_mutex);
10155 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10156 const char *symname)
10158 struct btrfs_trans_handle *trans;
10159 struct btrfs_root *root = BTRFS_I(dir)->root;
10160 struct btrfs_path *path;
10161 struct btrfs_key key;
10162 struct inode *inode = NULL;
10164 int drop_inode = 0;
10170 struct btrfs_file_extent_item *ei;
10171 struct extent_buffer *leaf;
10173 name_len = strlen(symname);
10174 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
10175 return -ENAMETOOLONG;
10178 * 2 items for inode item and ref
10179 * 2 items for dir items
10180 * 1 item for updating parent inode item
10181 * 1 item for the inline extent item
10182 * 1 item for xattr if selinux is on
10184 trans = btrfs_start_transaction(root, 7);
10186 return PTR_ERR(trans);
10188 err = btrfs_find_free_ino(root, &objectid);
10192 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10193 dentry->d_name.len, btrfs_ino(dir), objectid,
10194 S_IFLNK|S_IRWXUGO, &index);
10195 if (IS_ERR(inode)) {
10196 err = PTR_ERR(inode);
10201 * If the active LSM wants to access the inode during
10202 * d_instantiate it needs these. Smack checks to see
10203 * if the filesystem supports xattrs by looking at the
10206 inode->i_fop = &btrfs_file_operations;
10207 inode->i_op = &btrfs_file_inode_operations;
10208 inode->i_mapping->a_ops = &btrfs_aops;
10209 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10211 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10213 goto out_unlock_inode;
10215 path = btrfs_alloc_path();
10218 goto out_unlock_inode;
10220 key.objectid = btrfs_ino(inode);
10222 key.type = BTRFS_EXTENT_DATA_KEY;
10223 datasize = btrfs_file_extent_calc_inline_size(name_len);
10224 err = btrfs_insert_empty_item(trans, root, path, &key,
10227 btrfs_free_path(path);
10228 goto out_unlock_inode;
10230 leaf = path->nodes[0];
10231 ei = btrfs_item_ptr(leaf, path->slots[0],
10232 struct btrfs_file_extent_item);
10233 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10234 btrfs_set_file_extent_type(leaf, ei,
10235 BTRFS_FILE_EXTENT_INLINE);
10236 btrfs_set_file_extent_encryption(leaf, ei, 0);
10237 btrfs_set_file_extent_compression(leaf, ei, 0);
10238 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10239 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10241 ptr = btrfs_file_extent_inline_start(ei);
10242 write_extent_buffer(leaf, symname, ptr, name_len);
10243 btrfs_mark_buffer_dirty(leaf);
10244 btrfs_free_path(path);
10246 inode->i_op = &btrfs_symlink_inode_operations;
10247 inode_nohighmem(inode);
10248 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10249 inode_set_bytes(inode, name_len);
10250 btrfs_i_size_write(inode, name_len);
10251 err = btrfs_update_inode(trans, root, inode);
10253 * Last step, add directory indexes for our symlink inode. This is the
10254 * last step to avoid extra cleanup of these indexes if an error happens
10258 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10261 goto out_unlock_inode;
10264 unlock_new_inode(inode);
10265 d_instantiate(dentry, inode);
10268 btrfs_end_transaction(trans, root);
10270 inode_dec_link_count(inode);
10273 btrfs_btree_balance_dirty(root);
10278 unlock_new_inode(inode);
10282 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10283 u64 start, u64 num_bytes, u64 min_size,
10284 loff_t actual_len, u64 *alloc_hint,
10285 struct btrfs_trans_handle *trans)
10287 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10288 struct extent_map *em;
10289 struct btrfs_root *root = BTRFS_I(inode)->root;
10290 struct btrfs_key ins;
10291 u64 cur_offset = start;
10294 u64 last_alloc = (u64)-1;
10296 bool own_trans = true;
10300 while (num_bytes > 0) {
10302 trans = btrfs_start_transaction(root, 3);
10303 if (IS_ERR(trans)) {
10304 ret = PTR_ERR(trans);
10309 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10310 cur_bytes = max(cur_bytes, min_size);
10312 * If we are severely fragmented we could end up with really
10313 * small allocations, so if the allocator is returning small
10314 * chunks lets make its job easier by only searching for those
10317 cur_bytes = min(cur_bytes, last_alloc);
10318 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
10319 *alloc_hint, &ins, 1, 0);
10322 btrfs_end_transaction(trans, root);
10325 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
10327 last_alloc = ins.offset;
10328 ret = insert_reserved_file_extent(trans, inode,
10329 cur_offset, ins.objectid,
10330 ins.offset, ins.offset,
10331 ins.offset, 0, 0, 0,
10332 BTRFS_FILE_EXTENT_PREALLOC);
10334 btrfs_free_reserved_extent(root, ins.objectid,
10336 btrfs_abort_transaction(trans, root, ret);
10338 btrfs_end_transaction(trans, root);
10342 btrfs_drop_extent_cache(inode, cur_offset,
10343 cur_offset + ins.offset -1, 0);
10345 em = alloc_extent_map();
10347 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10348 &BTRFS_I(inode)->runtime_flags);
10352 em->start = cur_offset;
10353 em->orig_start = cur_offset;
10354 em->len = ins.offset;
10355 em->block_start = ins.objectid;
10356 em->block_len = ins.offset;
10357 em->orig_block_len = ins.offset;
10358 em->ram_bytes = ins.offset;
10359 em->bdev = root->fs_info->fs_devices->latest_bdev;
10360 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10361 em->generation = trans->transid;
10364 write_lock(&em_tree->lock);
10365 ret = add_extent_mapping(em_tree, em, 1);
10366 write_unlock(&em_tree->lock);
10367 if (ret != -EEXIST)
10369 btrfs_drop_extent_cache(inode, cur_offset,
10370 cur_offset + ins.offset - 1,
10373 free_extent_map(em);
10375 num_bytes -= ins.offset;
10376 cur_offset += ins.offset;
10377 *alloc_hint = ins.objectid + ins.offset;
10379 inode_inc_iversion(inode);
10380 inode->i_ctime = current_fs_time(inode->i_sb);
10381 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10382 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10383 (actual_len > inode->i_size) &&
10384 (cur_offset > inode->i_size)) {
10385 if (cur_offset > actual_len)
10386 i_size = actual_len;
10388 i_size = cur_offset;
10389 i_size_write(inode, i_size);
10390 btrfs_ordered_update_i_size(inode, i_size, NULL);
10393 ret = btrfs_update_inode(trans, root, inode);
10396 btrfs_abort_transaction(trans, root, ret);
10398 btrfs_end_transaction(trans, root);
10403 btrfs_end_transaction(trans, root);
10408 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10409 u64 start, u64 num_bytes, u64 min_size,
10410 loff_t actual_len, u64 *alloc_hint)
10412 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10413 min_size, actual_len, alloc_hint,
10417 int btrfs_prealloc_file_range_trans(struct inode *inode,
10418 struct btrfs_trans_handle *trans, int mode,
10419 u64 start, u64 num_bytes, u64 min_size,
10420 loff_t actual_len, u64 *alloc_hint)
10422 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10423 min_size, actual_len, alloc_hint, trans);
10426 static int btrfs_set_page_dirty(struct page *page)
10428 return __set_page_dirty_nobuffers(page);
10431 static int btrfs_permission(struct inode *inode, int mask)
10433 struct btrfs_root *root = BTRFS_I(inode)->root;
10434 umode_t mode = inode->i_mode;
10436 if (mask & MAY_WRITE &&
10437 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10438 if (btrfs_root_readonly(root))
10440 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10443 return generic_permission(inode, mask);
10446 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10448 struct btrfs_trans_handle *trans;
10449 struct btrfs_root *root = BTRFS_I(dir)->root;
10450 struct inode *inode = NULL;
10456 * 5 units required for adding orphan entry
10458 trans = btrfs_start_transaction(root, 5);
10460 return PTR_ERR(trans);
10462 ret = btrfs_find_free_ino(root, &objectid);
10466 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10467 btrfs_ino(dir), objectid, mode, &index);
10468 if (IS_ERR(inode)) {
10469 ret = PTR_ERR(inode);
10474 inode->i_fop = &btrfs_file_operations;
10475 inode->i_op = &btrfs_file_inode_operations;
10477 inode->i_mapping->a_ops = &btrfs_aops;
10478 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10480 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10484 ret = btrfs_update_inode(trans, root, inode);
10487 ret = btrfs_orphan_add(trans, inode);
10492 * We set number of links to 0 in btrfs_new_inode(), and here we set
10493 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10496 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10498 set_nlink(inode, 1);
10499 unlock_new_inode(inode);
10500 d_tmpfile(dentry, inode);
10501 mark_inode_dirty(inode);
10504 btrfs_end_transaction(trans, root);
10507 btrfs_balance_delayed_items(root);
10508 btrfs_btree_balance_dirty(root);
10512 unlock_new_inode(inode);
10517 /* Inspired by filemap_check_errors() */
10518 int btrfs_inode_check_errors(struct inode *inode)
10522 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10523 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10525 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10526 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10532 static const struct inode_operations btrfs_dir_inode_operations = {
10533 .getattr = btrfs_getattr,
10534 .lookup = btrfs_lookup,
10535 .create = btrfs_create,
10536 .unlink = btrfs_unlink,
10537 .link = btrfs_link,
10538 .mkdir = btrfs_mkdir,
10539 .rmdir = btrfs_rmdir,
10540 .rename2 = btrfs_rename2,
10541 .symlink = btrfs_symlink,
10542 .setattr = btrfs_setattr,
10543 .mknod = btrfs_mknod,
10544 .setxattr = generic_setxattr,
10545 .getxattr = generic_getxattr,
10546 .listxattr = btrfs_listxattr,
10547 .removexattr = generic_removexattr,
10548 .permission = btrfs_permission,
10549 .get_acl = btrfs_get_acl,
10550 .set_acl = btrfs_set_acl,
10551 .update_time = btrfs_update_time,
10552 .tmpfile = btrfs_tmpfile,
10554 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10555 .lookup = btrfs_lookup,
10556 .permission = btrfs_permission,
10557 .get_acl = btrfs_get_acl,
10558 .set_acl = btrfs_set_acl,
10559 .update_time = btrfs_update_time,
10562 static const struct file_operations btrfs_dir_file_operations = {
10563 .llseek = generic_file_llseek,
10564 .read = generic_read_dir,
10565 .iterate_shared = btrfs_real_readdir,
10566 .unlocked_ioctl = btrfs_ioctl,
10567 #ifdef CONFIG_COMPAT
10568 .compat_ioctl = btrfs_compat_ioctl,
10570 .release = btrfs_release_file,
10571 .fsync = btrfs_sync_file,
10574 static const struct extent_io_ops btrfs_extent_io_ops = {
10575 .fill_delalloc = run_delalloc_range,
10576 .submit_bio_hook = btrfs_submit_bio_hook,
10577 .merge_bio_hook = btrfs_merge_bio_hook,
10578 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10579 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10580 .writepage_start_hook = btrfs_writepage_start_hook,
10581 .set_bit_hook = btrfs_set_bit_hook,
10582 .clear_bit_hook = btrfs_clear_bit_hook,
10583 .merge_extent_hook = btrfs_merge_extent_hook,
10584 .split_extent_hook = btrfs_split_extent_hook,
10588 * btrfs doesn't support the bmap operation because swapfiles
10589 * use bmap to make a mapping of extents in the file. They assume
10590 * these extents won't change over the life of the file and they
10591 * use the bmap result to do IO directly to the drive.
10593 * the btrfs bmap call would return logical addresses that aren't
10594 * suitable for IO and they also will change frequently as COW
10595 * operations happen. So, swapfile + btrfs == corruption.
10597 * For now we're avoiding this by dropping bmap.
10599 static const struct address_space_operations btrfs_aops = {
10600 .readpage = btrfs_readpage,
10601 .writepage = btrfs_writepage,
10602 .writepages = btrfs_writepages,
10603 .readpages = btrfs_readpages,
10604 .direct_IO = btrfs_direct_IO,
10605 .invalidatepage = btrfs_invalidatepage,
10606 .releasepage = btrfs_releasepage,
10607 .set_page_dirty = btrfs_set_page_dirty,
10608 .error_remove_page = generic_error_remove_page,
10611 static const struct address_space_operations btrfs_symlink_aops = {
10612 .readpage = btrfs_readpage,
10613 .writepage = btrfs_writepage,
10614 .invalidatepage = btrfs_invalidatepage,
10615 .releasepage = btrfs_releasepage,
10618 static const struct inode_operations btrfs_file_inode_operations = {
10619 .getattr = btrfs_getattr,
10620 .setattr = btrfs_setattr,
10621 .setxattr = generic_setxattr,
10622 .getxattr = generic_getxattr,
10623 .listxattr = btrfs_listxattr,
10624 .removexattr = generic_removexattr,
10625 .permission = btrfs_permission,
10626 .fiemap = btrfs_fiemap,
10627 .get_acl = btrfs_get_acl,
10628 .set_acl = btrfs_set_acl,
10629 .update_time = btrfs_update_time,
10631 static const struct inode_operations btrfs_special_inode_operations = {
10632 .getattr = btrfs_getattr,
10633 .setattr = btrfs_setattr,
10634 .permission = btrfs_permission,
10635 .setxattr = generic_setxattr,
10636 .getxattr = generic_getxattr,
10637 .listxattr = btrfs_listxattr,
10638 .removexattr = generic_removexattr,
10639 .get_acl = btrfs_get_acl,
10640 .set_acl = btrfs_set_acl,
10641 .update_time = btrfs_update_time,
10643 static const struct inode_operations btrfs_symlink_inode_operations = {
10644 .readlink = generic_readlink,
10645 .get_link = page_get_link,
10646 .getattr = btrfs_getattr,
10647 .setattr = btrfs_setattr,
10648 .permission = btrfs_permission,
10649 .setxattr = generic_setxattr,
10650 .getxattr = generic_getxattr,
10651 .listxattr = btrfs_listxattr,
10652 .removexattr = generic_removexattr,
10653 .update_time = btrfs_update_time,
10656 const struct dentry_operations btrfs_dentry_operations = {
10657 .d_delete = btrfs_dentry_delete,
10658 .d_release = btrfs_dentry_release,