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;
591 * The async work queues will take care of doing actual
592 * allocation on disk for these compressed pages, and
593 * will submit them to the elevator.
595 add_async_extent(async_cow, start, num_bytes,
596 total_compressed, pages, nr_pages_ret,
599 if (start + num_bytes < end) {
610 * the compression code ran but failed to make things smaller,
611 * free any pages it allocated and our page pointer array
613 for (i = 0; i < nr_pages_ret; i++) {
614 WARN_ON(pages[i]->mapping);
619 total_compressed = 0;
622 /* flag the file so we don't compress in the future */
623 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
624 !(BTRFS_I(inode)->force_compress)) {
625 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
628 cleanup_and_bail_uncompressed:
630 * No compression, but we still need to write the pages in the file
631 * we've been given so far. redirty the locked page if it corresponds
632 * to our extent and set things up for the async work queue to run
633 * cow_file_range to do the normal delalloc dance.
635 if (page_offset(locked_page) >= start &&
636 page_offset(locked_page) <= end)
637 __set_page_dirty_nobuffers(locked_page);
638 /* unlocked later on in the async handlers */
641 extent_range_redirty_for_io(inode, start, end);
642 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
643 BTRFS_COMPRESS_NONE);
649 for (i = 0; i < nr_pages_ret; i++) {
650 WARN_ON(pages[i]->mapping);
656 static void free_async_extent_pages(struct async_extent *async_extent)
660 if (!async_extent->pages)
663 for (i = 0; i < async_extent->nr_pages; i++) {
664 WARN_ON(async_extent->pages[i]->mapping);
665 put_page(async_extent->pages[i]);
667 kfree(async_extent->pages);
668 async_extent->nr_pages = 0;
669 async_extent->pages = NULL;
673 * phase two of compressed writeback. This is the ordered portion
674 * of the code, which only gets called in the order the work was
675 * queued. We walk all the async extents created by compress_file_range
676 * and send them down to the disk.
678 static noinline void submit_compressed_extents(struct inode *inode,
679 struct async_cow *async_cow)
681 struct async_extent *async_extent;
683 struct btrfs_key ins;
684 struct extent_map *em;
685 struct btrfs_root *root = BTRFS_I(inode)->root;
686 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
687 struct extent_io_tree *io_tree;
691 while (!list_empty(&async_cow->extents)) {
692 async_extent = list_entry(async_cow->extents.next,
693 struct async_extent, list);
694 list_del(&async_extent->list);
696 io_tree = &BTRFS_I(inode)->io_tree;
699 /* did the compression code fall back to uncompressed IO? */
700 if (!async_extent->pages) {
701 int page_started = 0;
702 unsigned long nr_written = 0;
704 lock_extent(io_tree, async_extent->start,
705 async_extent->start +
706 async_extent->ram_size - 1);
708 /* allocate blocks */
709 ret = cow_file_range(inode, async_cow->locked_page,
711 async_extent->start +
712 async_extent->ram_size - 1,
713 &page_started, &nr_written, 0);
718 * if page_started, cow_file_range inserted an
719 * inline extent and took care of all the unlocking
720 * and IO for us. Otherwise, we need to submit
721 * all those pages down to the drive.
723 if (!page_started && !ret)
724 extent_write_locked_range(io_tree,
725 inode, async_extent->start,
726 async_extent->start +
727 async_extent->ram_size - 1,
731 unlock_page(async_cow->locked_page);
737 lock_extent(io_tree, async_extent->start,
738 async_extent->start + async_extent->ram_size - 1);
740 ret = btrfs_reserve_extent(root,
741 async_extent->compressed_size,
742 async_extent->compressed_size,
743 0, alloc_hint, &ins, 1, 1);
745 free_async_extent_pages(async_extent);
747 if (ret == -ENOSPC) {
748 unlock_extent(io_tree, async_extent->start,
749 async_extent->start +
750 async_extent->ram_size - 1);
753 * we need to redirty the pages if we decide to
754 * fallback to uncompressed IO, otherwise we
755 * will not submit these pages down to lower
758 extent_range_redirty_for_io(inode,
760 async_extent->start +
761 async_extent->ram_size - 1);
768 * here we're doing allocation and writeback of the
771 btrfs_drop_extent_cache(inode, async_extent->start,
772 async_extent->start +
773 async_extent->ram_size - 1, 0);
775 em = alloc_extent_map();
778 goto out_free_reserve;
780 em->start = async_extent->start;
781 em->len = async_extent->ram_size;
782 em->orig_start = em->start;
783 em->mod_start = em->start;
784 em->mod_len = em->len;
786 em->block_start = ins.objectid;
787 em->block_len = ins.offset;
788 em->orig_block_len = ins.offset;
789 em->ram_bytes = async_extent->ram_size;
790 em->bdev = root->fs_info->fs_devices->latest_bdev;
791 em->compress_type = async_extent->compress_type;
792 set_bit(EXTENT_FLAG_PINNED, &em->flags);
793 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
797 write_lock(&em_tree->lock);
798 ret = add_extent_mapping(em_tree, em, 1);
799 write_unlock(&em_tree->lock);
800 if (ret != -EEXIST) {
804 btrfs_drop_extent_cache(inode, async_extent->start,
805 async_extent->start +
806 async_extent->ram_size - 1, 0);
810 goto out_free_reserve;
812 ret = btrfs_add_ordered_extent_compress(inode,
815 async_extent->ram_size,
817 BTRFS_ORDERED_COMPRESSED,
818 async_extent->compress_type);
820 btrfs_drop_extent_cache(inode, async_extent->start,
821 async_extent->start +
822 async_extent->ram_size - 1, 0);
823 goto out_free_reserve;
825 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
828 * clear dirty, set writeback and unlock the pages.
830 extent_clear_unlock_delalloc(inode, async_extent->start,
831 async_extent->start +
832 async_extent->ram_size - 1,
833 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
834 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
836 ret = btrfs_submit_compressed_write(inode,
838 async_extent->ram_size,
840 ins.offset, async_extent->pages,
841 async_extent->nr_pages);
843 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
844 struct page *p = async_extent->pages[0];
845 const u64 start = async_extent->start;
846 const u64 end = start + async_extent->ram_size - 1;
848 p->mapping = inode->i_mapping;
849 tree->ops->writepage_end_io_hook(p, start, end,
852 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
855 free_async_extent_pages(async_extent);
857 alloc_hint = ins.objectid + ins.offset;
863 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
864 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
866 extent_clear_unlock_delalloc(inode, async_extent->start,
867 async_extent->start +
868 async_extent->ram_size - 1,
869 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
870 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
872 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
874 free_async_extent_pages(async_extent);
879 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 struct extent_map *em;
886 read_lock(&em_tree->lock);
887 em = search_extent_mapping(em_tree, start, num_bytes);
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
894 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
896 em = search_extent_mapping(em_tree, 0, 0);
897 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898 alloc_hint = em->block_start;
902 alloc_hint = em->block_start;
906 read_unlock(&em_tree->lock);
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
924 static noinline int cow_file_range(struct inode *inode,
925 struct page *locked_page,
926 u64 start, u64 end, int *page_started,
927 unsigned long *nr_written,
930 struct btrfs_root *root = BTRFS_I(inode)->root;
933 unsigned long ram_size;
936 u64 blocksize = root->sectorsize;
937 struct btrfs_key ins;
938 struct extent_map *em;
939 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
942 if (btrfs_is_free_space_inode(inode)) {
948 num_bytes = ALIGN(end - start + 1, blocksize);
949 num_bytes = max(blocksize, num_bytes);
950 disk_num_bytes = num_bytes;
952 /* if this is a small write inside eof, kick off defrag */
953 if (num_bytes < SZ_64K &&
954 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
955 btrfs_add_inode_defrag(NULL, inode);
958 /* lets try to make an inline extent */
959 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
962 extent_clear_unlock_delalloc(inode, start, end, NULL,
963 EXTENT_LOCKED | EXTENT_DELALLOC |
964 EXTENT_DEFRAG, PAGE_UNLOCK |
965 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
968 *nr_written = *nr_written +
969 (end - start + PAGE_SIZE) / PAGE_SIZE;
972 } else if (ret < 0) {
977 BUG_ON(disk_num_bytes >
978 btrfs_super_total_bytes(root->fs_info->super_copy));
980 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
981 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
983 while (disk_num_bytes > 0) {
986 cur_alloc_size = disk_num_bytes;
987 ret = btrfs_reserve_extent(root, cur_alloc_size,
988 root->sectorsize, 0, alloc_hint,
993 em = alloc_extent_map();
999 em->orig_start = em->start;
1000 ram_size = ins.offset;
1001 em->len = ins.offset;
1002 em->mod_start = em->start;
1003 em->mod_len = em->len;
1005 em->block_start = ins.objectid;
1006 em->block_len = ins.offset;
1007 em->orig_block_len = ins.offset;
1008 em->ram_bytes = ram_size;
1009 em->bdev = root->fs_info->fs_devices->latest_bdev;
1010 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1011 em->generation = -1;
1014 write_lock(&em_tree->lock);
1015 ret = add_extent_mapping(em_tree, em, 1);
1016 write_unlock(&em_tree->lock);
1017 if (ret != -EEXIST) {
1018 free_extent_map(em);
1021 btrfs_drop_extent_cache(inode, start,
1022 start + ram_size - 1, 0);
1027 cur_alloc_size = ins.offset;
1028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029 ram_size, cur_alloc_size, 0);
1031 goto out_drop_extent_cache;
1033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1038 goto out_drop_extent_cache;
1041 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1043 if (disk_num_bytes < cur_alloc_size)
1046 /* we're not doing compressed IO, don't unlock the first
1047 * page (which the caller expects to stay locked), don't
1048 * clear any dirty bits and don't set any writeback bits
1050 * Do set the Private2 bit so we know this page was properly
1051 * setup for writepage
1053 op = unlock ? PAGE_UNLOCK : 0;
1054 op |= PAGE_SET_PRIVATE2;
1056 extent_clear_unlock_delalloc(inode, start,
1057 start + ram_size - 1, locked_page,
1058 EXTENT_LOCKED | EXTENT_DELALLOC,
1060 disk_num_bytes -= cur_alloc_size;
1061 num_bytes -= cur_alloc_size;
1062 alloc_hint = ins.objectid + ins.offset;
1063 start += cur_alloc_size;
1068 out_drop_extent_cache:
1069 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1071 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1072 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1074 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1075 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1076 EXTENT_DELALLOC | EXTENT_DEFRAG,
1077 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1078 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1083 * work queue call back to started compression on a file and pages
1085 static noinline void async_cow_start(struct btrfs_work *work)
1087 struct async_cow *async_cow;
1089 async_cow = container_of(work, struct async_cow, work);
1091 compress_file_range(async_cow->inode, async_cow->locked_page,
1092 async_cow->start, async_cow->end, async_cow,
1094 if (num_added == 0) {
1095 btrfs_add_delayed_iput(async_cow->inode);
1096 async_cow->inode = NULL;
1101 * work queue call back to submit previously compressed pages
1103 static noinline void async_cow_submit(struct btrfs_work *work)
1105 struct async_cow *async_cow;
1106 struct btrfs_root *root;
1107 unsigned long nr_pages;
1109 async_cow = container_of(work, struct async_cow, work);
1111 root = async_cow->root;
1112 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1116 * atomic_sub_return implies a barrier for waitqueue_active
1118 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1120 waitqueue_active(&root->fs_info->async_submit_wait))
1121 wake_up(&root->fs_info->async_submit_wait);
1123 if (async_cow->inode)
1124 submit_compressed_extents(async_cow->inode, async_cow);
1127 static noinline void async_cow_free(struct btrfs_work *work)
1129 struct async_cow *async_cow;
1130 async_cow = container_of(work, struct async_cow, work);
1131 if (async_cow->inode)
1132 btrfs_add_delayed_iput(async_cow->inode);
1136 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1137 u64 start, u64 end, int *page_started,
1138 unsigned long *nr_written)
1140 struct async_cow *async_cow;
1141 struct btrfs_root *root = BTRFS_I(inode)->root;
1142 unsigned long nr_pages;
1144 int limit = 10 * SZ_1M;
1146 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1147 1, 0, NULL, GFP_NOFS);
1148 while (start < end) {
1149 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1150 BUG_ON(!async_cow); /* -ENOMEM */
1151 async_cow->inode = igrab(inode);
1152 async_cow->root = root;
1153 async_cow->locked_page = locked_page;
1154 async_cow->start = start;
1156 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1157 !btrfs_test_opt(root, FORCE_COMPRESS))
1160 cur_end = min(end, start + SZ_512K - 1);
1162 async_cow->end = cur_end;
1163 INIT_LIST_HEAD(&async_cow->extents);
1165 btrfs_init_work(&async_cow->work,
1166 btrfs_delalloc_helper,
1167 async_cow_start, async_cow_submit,
1170 nr_pages = (cur_end - start + PAGE_SIZE) >>
1172 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1174 btrfs_queue_work(root->fs_info->delalloc_workers,
1177 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1178 wait_event(root->fs_info->async_submit_wait,
1179 (atomic_read(&root->fs_info->async_delalloc_pages) <
1183 while (atomic_read(&root->fs_info->async_submit_draining) &&
1184 atomic_read(&root->fs_info->async_delalloc_pages)) {
1185 wait_event(root->fs_info->async_submit_wait,
1186 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1190 *nr_written += nr_pages;
1191 start = cur_end + 1;
1197 static noinline int csum_exist_in_range(struct btrfs_root *root,
1198 u64 bytenr, u64 num_bytes)
1201 struct btrfs_ordered_sum *sums;
1204 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1205 bytenr + num_bytes - 1, &list, 0);
1206 if (ret == 0 && list_empty(&list))
1209 while (!list_empty(&list)) {
1210 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1211 list_del(&sums->list);
1218 * when nowcow writeback call back. This checks for snapshots or COW copies
1219 * of the extents that exist in the file, and COWs the file as required.
1221 * If no cow copies or snapshots exist, we write directly to the existing
1224 static noinline int run_delalloc_nocow(struct inode *inode,
1225 struct page *locked_page,
1226 u64 start, u64 end, int *page_started, int force,
1227 unsigned long *nr_written)
1229 struct btrfs_root *root = BTRFS_I(inode)->root;
1230 struct btrfs_trans_handle *trans;
1231 struct extent_buffer *leaf;
1232 struct btrfs_path *path;
1233 struct btrfs_file_extent_item *fi;
1234 struct btrfs_key found_key;
1249 u64 ino = btrfs_ino(inode);
1251 path = btrfs_alloc_path();
1253 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1254 EXTENT_LOCKED | EXTENT_DELALLOC |
1255 EXTENT_DO_ACCOUNTING |
1256 EXTENT_DEFRAG, PAGE_UNLOCK |
1258 PAGE_SET_WRITEBACK |
1259 PAGE_END_WRITEBACK);
1263 nolock = btrfs_is_free_space_inode(inode);
1266 trans = btrfs_join_transaction_nolock(root);
1268 trans = btrfs_join_transaction(root);
1270 if (IS_ERR(trans)) {
1271 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1272 EXTENT_LOCKED | EXTENT_DELALLOC |
1273 EXTENT_DO_ACCOUNTING |
1274 EXTENT_DEFRAG, PAGE_UNLOCK |
1276 PAGE_SET_WRITEBACK |
1277 PAGE_END_WRITEBACK);
1278 btrfs_free_path(path);
1279 return PTR_ERR(trans);
1282 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1284 cow_start = (u64)-1;
1287 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1291 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1292 leaf = path->nodes[0];
1293 btrfs_item_key_to_cpu(leaf, &found_key,
1294 path->slots[0] - 1);
1295 if (found_key.objectid == ino &&
1296 found_key.type == BTRFS_EXTENT_DATA_KEY)
1301 leaf = path->nodes[0];
1302 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1303 ret = btrfs_next_leaf(root, path);
1308 leaf = path->nodes[0];
1314 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1316 if (found_key.objectid > ino)
1318 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1319 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1323 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1324 found_key.offset > end)
1327 if (found_key.offset > cur_offset) {
1328 extent_end = found_key.offset;
1333 fi = btrfs_item_ptr(leaf, path->slots[0],
1334 struct btrfs_file_extent_item);
1335 extent_type = btrfs_file_extent_type(leaf, fi);
1337 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1338 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1339 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1340 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1341 extent_offset = btrfs_file_extent_offset(leaf, fi);
1342 extent_end = found_key.offset +
1343 btrfs_file_extent_num_bytes(leaf, fi);
1345 btrfs_file_extent_disk_num_bytes(leaf, fi);
1346 if (extent_end <= start) {
1350 if (disk_bytenr == 0)
1352 if (btrfs_file_extent_compression(leaf, fi) ||
1353 btrfs_file_extent_encryption(leaf, fi) ||
1354 btrfs_file_extent_other_encoding(leaf, fi))
1356 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1358 if (btrfs_extent_readonly(root, disk_bytenr))
1360 if (btrfs_cross_ref_exist(trans, root, ino,
1362 extent_offset, disk_bytenr))
1364 disk_bytenr += extent_offset;
1365 disk_bytenr += cur_offset - found_key.offset;
1366 num_bytes = min(end + 1, extent_end) - cur_offset;
1368 * if there are pending snapshots for this root,
1369 * we fall into common COW way.
1372 err = btrfs_start_write_no_snapshoting(root);
1377 * force cow if csum exists in the range.
1378 * this ensure that csum for a given extent are
1379 * either valid or do not exist.
1381 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1383 if (!btrfs_inc_nocow_writers(root->fs_info,
1387 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1388 extent_end = found_key.offset +
1389 btrfs_file_extent_inline_len(leaf,
1390 path->slots[0], fi);
1391 extent_end = ALIGN(extent_end, root->sectorsize);
1396 if (extent_end <= start) {
1398 if (!nolock && nocow)
1399 btrfs_end_write_no_snapshoting(root);
1401 btrfs_dec_nocow_writers(root->fs_info,
1406 if (cow_start == (u64)-1)
1407 cow_start = cur_offset;
1408 cur_offset = extent_end;
1409 if (cur_offset > end)
1415 btrfs_release_path(path);
1416 if (cow_start != (u64)-1) {
1417 ret = cow_file_range(inode, locked_page,
1418 cow_start, found_key.offset - 1,
1419 page_started, nr_written, 1);
1421 if (!nolock && nocow)
1422 btrfs_end_write_no_snapshoting(root);
1424 btrfs_dec_nocow_writers(root->fs_info,
1428 cow_start = (u64)-1;
1431 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1432 struct extent_map *em;
1433 struct extent_map_tree *em_tree;
1434 em_tree = &BTRFS_I(inode)->extent_tree;
1435 em = alloc_extent_map();
1436 BUG_ON(!em); /* -ENOMEM */
1437 em->start = cur_offset;
1438 em->orig_start = found_key.offset - extent_offset;
1439 em->len = num_bytes;
1440 em->block_len = num_bytes;
1441 em->block_start = disk_bytenr;
1442 em->orig_block_len = disk_num_bytes;
1443 em->ram_bytes = ram_bytes;
1444 em->bdev = root->fs_info->fs_devices->latest_bdev;
1445 em->mod_start = em->start;
1446 em->mod_len = em->len;
1447 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1448 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1449 em->generation = -1;
1451 write_lock(&em_tree->lock);
1452 ret = add_extent_mapping(em_tree, em, 1);
1453 write_unlock(&em_tree->lock);
1454 if (ret != -EEXIST) {
1455 free_extent_map(em);
1458 btrfs_drop_extent_cache(inode, em->start,
1459 em->start + em->len - 1, 0);
1461 type = BTRFS_ORDERED_PREALLOC;
1463 type = BTRFS_ORDERED_NOCOW;
1466 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1467 num_bytes, num_bytes, type);
1469 btrfs_dec_nocow_writers(root->fs_info, disk_bytenr);
1470 BUG_ON(ret); /* -ENOMEM */
1472 if (root->root_key.objectid ==
1473 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1474 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1477 if (!nolock && nocow)
1478 btrfs_end_write_no_snapshoting(root);
1483 extent_clear_unlock_delalloc(inode, cur_offset,
1484 cur_offset + num_bytes - 1,
1485 locked_page, EXTENT_LOCKED |
1486 EXTENT_DELALLOC, PAGE_UNLOCK |
1488 if (!nolock && nocow)
1489 btrfs_end_write_no_snapshoting(root);
1490 cur_offset = extent_end;
1491 if (cur_offset > end)
1494 btrfs_release_path(path);
1496 if (cur_offset <= end && cow_start == (u64)-1) {
1497 cow_start = cur_offset;
1501 if (cow_start != (u64)-1) {
1502 ret = cow_file_range(inode, locked_page, cow_start, end,
1503 page_started, nr_written, 1);
1509 err = btrfs_end_transaction(trans, root);
1513 if (ret && cur_offset < end)
1514 extent_clear_unlock_delalloc(inode, cur_offset, end,
1515 locked_page, EXTENT_LOCKED |
1516 EXTENT_DELALLOC | EXTENT_DEFRAG |
1517 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1519 PAGE_SET_WRITEBACK |
1520 PAGE_END_WRITEBACK);
1521 btrfs_free_path(path);
1525 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1528 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1529 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1533 * @defrag_bytes is a hint value, no spinlock held here,
1534 * if is not zero, it means the file is defragging.
1535 * Force cow if given extent needs to be defragged.
1537 if (BTRFS_I(inode)->defrag_bytes &&
1538 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1539 EXTENT_DEFRAG, 0, NULL))
1546 * extent_io.c call back to do delayed allocation processing
1548 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1549 u64 start, u64 end, int *page_started,
1550 unsigned long *nr_written)
1553 int force_cow = need_force_cow(inode, start, end);
1555 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1556 ret = run_delalloc_nocow(inode, locked_page, start, end,
1557 page_started, 1, nr_written);
1558 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1559 ret = run_delalloc_nocow(inode, locked_page, start, end,
1560 page_started, 0, nr_written);
1561 } else if (!inode_need_compress(inode)) {
1562 ret = cow_file_range(inode, locked_page, start, end,
1563 page_started, nr_written, 1);
1565 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1566 &BTRFS_I(inode)->runtime_flags);
1567 ret = cow_file_range_async(inode, locked_page, start, end,
1568 page_started, nr_written);
1573 static void btrfs_split_extent_hook(struct inode *inode,
1574 struct extent_state *orig, u64 split)
1578 /* not delalloc, ignore it */
1579 if (!(orig->state & EXTENT_DELALLOC))
1582 size = orig->end - orig->start + 1;
1583 if (size > BTRFS_MAX_EXTENT_SIZE) {
1588 * See the explanation in btrfs_merge_extent_hook, the same
1589 * applies here, just in reverse.
1591 new_size = orig->end - split + 1;
1592 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1593 BTRFS_MAX_EXTENT_SIZE);
1594 new_size = split - orig->start;
1595 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1596 BTRFS_MAX_EXTENT_SIZE);
1597 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1598 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1602 spin_lock(&BTRFS_I(inode)->lock);
1603 BTRFS_I(inode)->outstanding_extents++;
1604 spin_unlock(&BTRFS_I(inode)->lock);
1608 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1609 * extents so we can keep track of new extents that are just merged onto old
1610 * extents, such as when we are doing sequential writes, so we can properly
1611 * account for the metadata space we'll need.
1613 static void btrfs_merge_extent_hook(struct inode *inode,
1614 struct extent_state *new,
1615 struct extent_state *other)
1617 u64 new_size, old_size;
1620 /* not delalloc, ignore it */
1621 if (!(other->state & EXTENT_DELALLOC))
1624 if (new->start > other->start)
1625 new_size = new->end - other->start + 1;
1627 new_size = other->end - new->start + 1;
1629 /* we're not bigger than the max, unreserve the space and go */
1630 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1631 spin_lock(&BTRFS_I(inode)->lock);
1632 BTRFS_I(inode)->outstanding_extents--;
1633 spin_unlock(&BTRFS_I(inode)->lock);
1638 * We have to add up either side to figure out how many extents were
1639 * accounted for before we merged into one big extent. If the number of
1640 * extents we accounted for is <= the amount we need for the new range
1641 * then we can return, otherwise drop. Think of it like this
1645 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1646 * need 2 outstanding extents, on one side we have 1 and the other side
1647 * we have 1 so they are == and we can return. But in this case
1649 * [MAX_SIZE+4k][MAX_SIZE+4k]
1651 * Each range on their own accounts for 2 extents, but merged together
1652 * they are only 3 extents worth of accounting, so we need to drop in
1655 old_size = other->end - other->start + 1;
1656 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1657 BTRFS_MAX_EXTENT_SIZE);
1658 old_size = new->end - new->start + 1;
1659 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1660 BTRFS_MAX_EXTENT_SIZE);
1662 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1663 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1666 spin_lock(&BTRFS_I(inode)->lock);
1667 BTRFS_I(inode)->outstanding_extents--;
1668 spin_unlock(&BTRFS_I(inode)->lock);
1671 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1672 struct inode *inode)
1674 spin_lock(&root->delalloc_lock);
1675 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1676 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1677 &root->delalloc_inodes);
1678 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1679 &BTRFS_I(inode)->runtime_flags);
1680 root->nr_delalloc_inodes++;
1681 if (root->nr_delalloc_inodes == 1) {
1682 spin_lock(&root->fs_info->delalloc_root_lock);
1683 BUG_ON(!list_empty(&root->delalloc_root));
1684 list_add_tail(&root->delalloc_root,
1685 &root->fs_info->delalloc_roots);
1686 spin_unlock(&root->fs_info->delalloc_root_lock);
1689 spin_unlock(&root->delalloc_lock);
1692 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1693 struct inode *inode)
1695 spin_lock(&root->delalloc_lock);
1696 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1697 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1698 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1699 &BTRFS_I(inode)->runtime_flags);
1700 root->nr_delalloc_inodes--;
1701 if (!root->nr_delalloc_inodes) {
1702 spin_lock(&root->fs_info->delalloc_root_lock);
1703 BUG_ON(list_empty(&root->delalloc_root));
1704 list_del_init(&root->delalloc_root);
1705 spin_unlock(&root->fs_info->delalloc_root_lock);
1708 spin_unlock(&root->delalloc_lock);
1712 * extent_io.c set_bit_hook, used to track delayed allocation
1713 * bytes in this file, and to maintain the list of inodes that
1714 * have pending delalloc work to be done.
1716 static void btrfs_set_bit_hook(struct inode *inode,
1717 struct extent_state *state, unsigned *bits)
1720 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1723 * set_bit and clear bit hooks normally require _irqsave/restore
1724 * but in this case, we are only testing for the DELALLOC
1725 * bit, which is only set or cleared with irqs on
1727 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1728 struct btrfs_root *root = BTRFS_I(inode)->root;
1729 u64 len = state->end + 1 - state->start;
1730 bool do_list = !btrfs_is_free_space_inode(inode);
1732 if (*bits & EXTENT_FIRST_DELALLOC) {
1733 *bits &= ~EXTENT_FIRST_DELALLOC;
1735 spin_lock(&BTRFS_I(inode)->lock);
1736 BTRFS_I(inode)->outstanding_extents++;
1737 spin_unlock(&BTRFS_I(inode)->lock);
1740 /* For sanity tests */
1741 if (btrfs_test_is_dummy_root(root))
1744 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1745 root->fs_info->delalloc_batch);
1746 spin_lock(&BTRFS_I(inode)->lock);
1747 BTRFS_I(inode)->delalloc_bytes += len;
1748 if (*bits & EXTENT_DEFRAG)
1749 BTRFS_I(inode)->defrag_bytes += len;
1750 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1751 &BTRFS_I(inode)->runtime_flags))
1752 btrfs_add_delalloc_inodes(root, inode);
1753 spin_unlock(&BTRFS_I(inode)->lock);
1758 * extent_io.c clear_bit_hook, see set_bit_hook for why
1760 static void btrfs_clear_bit_hook(struct inode *inode,
1761 struct extent_state *state,
1764 u64 len = state->end + 1 - state->start;
1765 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1766 BTRFS_MAX_EXTENT_SIZE);
1768 spin_lock(&BTRFS_I(inode)->lock);
1769 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1770 BTRFS_I(inode)->defrag_bytes -= len;
1771 spin_unlock(&BTRFS_I(inode)->lock);
1774 * set_bit and clear bit hooks normally require _irqsave/restore
1775 * but in this case, we are only testing for the DELALLOC
1776 * bit, which is only set or cleared with irqs on
1778 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1779 struct btrfs_root *root = BTRFS_I(inode)->root;
1780 bool do_list = !btrfs_is_free_space_inode(inode);
1782 if (*bits & EXTENT_FIRST_DELALLOC) {
1783 *bits &= ~EXTENT_FIRST_DELALLOC;
1784 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1785 spin_lock(&BTRFS_I(inode)->lock);
1786 BTRFS_I(inode)->outstanding_extents -= num_extents;
1787 spin_unlock(&BTRFS_I(inode)->lock);
1791 * We don't reserve metadata space for space cache inodes so we
1792 * don't need to call dellalloc_release_metadata if there is an
1795 if (*bits & EXTENT_DO_ACCOUNTING &&
1796 root != root->fs_info->tree_root)
1797 btrfs_delalloc_release_metadata(inode, len);
1799 /* For sanity tests. */
1800 if (btrfs_test_is_dummy_root(root))
1803 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1804 && do_list && !(state->state & EXTENT_NORESERVE))
1805 btrfs_free_reserved_data_space_noquota(inode,
1808 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1809 root->fs_info->delalloc_batch);
1810 spin_lock(&BTRFS_I(inode)->lock);
1811 BTRFS_I(inode)->delalloc_bytes -= len;
1812 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1813 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1814 &BTRFS_I(inode)->runtime_flags))
1815 btrfs_del_delalloc_inode(root, inode);
1816 spin_unlock(&BTRFS_I(inode)->lock);
1821 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1822 * we don't create bios that span stripes or chunks
1824 * return 1 if page cannot be merged to bio
1825 * return 0 if page can be merged to bio
1826 * return error otherwise
1828 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1829 size_t size, struct bio *bio,
1830 unsigned long bio_flags)
1832 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1833 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1838 if (bio_flags & EXTENT_BIO_COMPRESSED)
1841 length = bio->bi_iter.bi_size;
1842 map_length = length;
1843 ret = btrfs_map_block(root->fs_info, rw, logical,
1844 &map_length, NULL, 0);
1847 if (map_length < length + size)
1853 * in order to insert checksums into the metadata in large chunks,
1854 * we wait until bio submission time. All the pages in the bio are
1855 * checksummed and sums are attached onto the ordered extent record.
1857 * At IO completion time the cums attached on the ordered extent record
1858 * are inserted into the btree
1860 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1861 struct bio *bio, int mirror_num,
1862 unsigned long bio_flags,
1865 struct btrfs_root *root = BTRFS_I(inode)->root;
1868 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1869 BUG_ON(ret); /* -ENOMEM */
1874 * in order to insert checksums into the metadata in large chunks,
1875 * we wait until bio submission time. All the pages in the bio are
1876 * checksummed and sums are attached onto the ordered extent record.
1878 * At IO completion time the cums attached on the ordered extent record
1879 * are inserted into the btree
1881 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1882 int mirror_num, unsigned long bio_flags,
1885 struct btrfs_root *root = BTRFS_I(inode)->root;
1888 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1890 bio->bi_error = ret;
1897 * extent_io.c submission hook. This does the right thing for csum calculation
1898 * on write, or reading the csums from the tree before a read
1900 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1901 int mirror_num, unsigned long bio_flags,
1904 struct btrfs_root *root = BTRFS_I(inode)->root;
1905 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1908 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1910 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1912 if (btrfs_is_free_space_inode(inode))
1913 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1915 if (!(rw & REQ_WRITE)) {
1916 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1920 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1921 ret = btrfs_submit_compressed_read(inode, bio,
1925 } else if (!skip_sum) {
1926 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1931 } else if (async && !skip_sum) {
1932 /* csum items have already been cloned */
1933 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1935 /* we're doing a write, do the async checksumming */
1936 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1937 inode, rw, bio, mirror_num,
1938 bio_flags, bio_offset,
1939 __btrfs_submit_bio_start,
1940 __btrfs_submit_bio_done);
1942 } else if (!skip_sum) {
1943 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1949 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1953 bio->bi_error = ret;
1960 * given a list of ordered sums record them in the inode. This happens
1961 * at IO completion time based on sums calculated at bio submission time.
1963 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1964 struct inode *inode, u64 file_offset,
1965 struct list_head *list)
1967 struct btrfs_ordered_sum *sum;
1969 list_for_each_entry(sum, list, list) {
1970 trans->adding_csums = 1;
1971 btrfs_csum_file_blocks(trans,
1972 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1973 trans->adding_csums = 0;
1978 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1979 struct extent_state **cached_state)
1981 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1982 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1986 /* see btrfs_writepage_start_hook for details on why this is required */
1987 struct btrfs_writepage_fixup {
1989 struct btrfs_work work;
1992 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1994 struct btrfs_writepage_fixup *fixup;
1995 struct btrfs_ordered_extent *ordered;
1996 struct extent_state *cached_state = NULL;
1998 struct inode *inode;
2003 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2007 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2008 ClearPageChecked(page);
2012 inode = page->mapping->host;
2013 page_start = page_offset(page);
2014 page_end = page_offset(page) + PAGE_SIZE - 1;
2016 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2019 /* already ordered? We're done */
2020 if (PagePrivate2(page))
2023 ordered = btrfs_lookup_ordered_range(inode, page_start,
2026 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2027 page_end, &cached_state, GFP_NOFS);
2029 btrfs_start_ordered_extent(inode, ordered, 1);
2030 btrfs_put_ordered_extent(ordered);
2034 ret = btrfs_delalloc_reserve_space(inode, page_start,
2037 mapping_set_error(page->mapping, ret);
2038 end_extent_writepage(page, ret, page_start, page_end);
2039 ClearPageChecked(page);
2043 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2044 ClearPageChecked(page);
2045 set_page_dirty(page);
2047 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2048 &cached_state, GFP_NOFS);
2056 * There are a few paths in the higher layers of the kernel that directly
2057 * set the page dirty bit without asking the filesystem if it is a
2058 * good idea. This causes problems because we want to make sure COW
2059 * properly happens and the data=ordered rules are followed.
2061 * In our case any range that doesn't have the ORDERED bit set
2062 * hasn't been properly setup for IO. We kick off an async process
2063 * to fix it up. The async helper will wait for ordered extents, set
2064 * the delalloc bit and make it safe to write the page.
2066 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2068 struct inode *inode = page->mapping->host;
2069 struct btrfs_writepage_fixup *fixup;
2070 struct btrfs_root *root = BTRFS_I(inode)->root;
2072 /* this page is properly in the ordered list */
2073 if (TestClearPagePrivate2(page))
2076 if (PageChecked(page))
2079 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2083 SetPageChecked(page);
2085 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2086 btrfs_writepage_fixup_worker, NULL, NULL);
2088 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2092 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2093 struct inode *inode, u64 file_pos,
2094 u64 disk_bytenr, u64 disk_num_bytes,
2095 u64 num_bytes, u64 ram_bytes,
2096 u8 compression, u8 encryption,
2097 u16 other_encoding, int extent_type)
2099 struct btrfs_root *root = BTRFS_I(inode)->root;
2100 struct btrfs_file_extent_item *fi;
2101 struct btrfs_path *path;
2102 struct extent_buffer *leaf;
2103 struct btrfs_key ins;
2104 int extent_inserted = 0;
2107 path = btrfs_alloc_path();
2112 * we may be replacing one extent in the tree with another.
2113 * The new extent is pinned in the extent map, and we don't want
2114 * to drop it from the cache until it is completely in the btree.
2116 * So, tell btrfs_drop_extents to leave this extent in the cache.
2117 * the caller is expected to unpin it and allow it to be merged
2120 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2121 file_pos + num_bytes, NULL, 0,
2122 1, sizeof(*fi), &extent_inserted);
2126 if (!extent_inserted) {
2127 ins.objectid = btrfs_ino(inode);
2128 ins.offset = file_pos;
2129 ins.type = BTRFS_EXTENT_DATA_KEY;
2131 path->leave_spinning = 1;
2132 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2137 leaf = path->nodes[0];
2138 fi = btrfs_item_ptr(leaf, path->slots[0],
2139 struct btrfs_file_extent_item);
2140 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2141 btrfs_set_file_extent_type(leaf, fi, extent_type);
2142 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2143 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2144 btrfs_set_file_extent_offset(leaf, fi, 0);
2145 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2146 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2147 btrfs_set_file_extent_compression(leaf, fi, compression);
2148 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2149 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2151 btrfs_mark_buffer_dirty(leaf);
2152 btrfs_release_path(path);
2154 inode_add_bytes(inode, num_bytes);
2156 ins.objectid = disk_bytenr;
2157 ins.offset = disk_num_bytes;
2158 ins.type = BTRFS_EXTENT_ITEM_KEY;
2159 ret = btrfs_alloc_reserved_file_extent(trans, root,
2160 root->root_key.objectid,
2161 btrfs_ino(inode), file_pos,
2164 * Release the reserved range from inode dirty range map, as it is
2165 * already moved into delayed_ref_head
2167 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2169 btrfs_free_path(path);
2174 /* snapshot-aware defrag */
2175 struct sa_defrag_extent_backref {
2176 struct rb_node node;
2177 struct old_sa_defrag_extent *old;
2186 struct old_sa_defrag_extent {
2187 struct list_head list;
2188 struct new_sa_defrag_extent *new;
2197 struct new_sa_defrag_extent {
2198 struct rb_root root;
2199 struct list_head head;
2200 struct btrfs_path *path;
2201 struct inode *inode;
2209 static int backref_comp(struct sa_defrag_extent_backref *b1,
2210 struct sa_defrag_extent_backref *b2)
2212 if (b1->root_id < b2->root_id)
2214 else if (b1->root_id > b2->root_id)
2217 if (b1->inum < b2->inum)
2219 else if (b1->inum > b2->inum)
2222 if (b1->file_pos < b2->file_pos)
2224 else if (b1->file_pos > b2->file_pos)
2228 * [------------------------------] ===> (a range of space)
2229 * |<--->| |<---->| =============> (fs/file tree A)
2230 * |<---------------------------->| ===> (fs/file tree B)
2232 * A range of space can refer to two file extents in one tree while
2233 * refer to only one file extent in another tree.
2235 * So we may process a disk offset more than one time(two extents in A)
2236 * and locate at the same extent(one extent in B), then insert two same
2237 * backrefs(both refer to the extent in B).
2242 static void backref_insert(struct rb_root *root,
2243 struct sa_defrag_extent_backref *backref)
2245 struct rb_node **p = &root->rb_node;
2246 struct rb_node *parent = NULL;
2247 struct sa_defrag_extent_backref *entry;
2252 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2254 ret = backref_comp(backref, entry);
2258 p = &(*p)->rb_right;
2261 rb_link_node(&backref->node, parent, p);
2262 rb_insert_color(&backref->node, root);
2266 * Note the backref might has changed, and in this case we just return 0.
2268 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2271 struct btrfs_file_extent_item *extent;
2272 struct btrfs_fs_info *fs_info;
2273 struct old_sa_defrag_extent *old = ctx;
2274 struct new_sa_defrag_extent *new = old->new;
2275 struct btrfs_path *path = new->path;
2276 struct btrfs_key key;
2277 struct btrfs_root *root;
2278 struct sa_defrag_extent_backref *backref;
2279 struct extent_buffer *leaf;
2280 struct inode *inode = new->inode;
2286 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2287 inum == btrfs_ino(inode))
2290 key.objectid = root_id;
2291 key.type = BTRFS_ROOT_ITEM_KEY;
2292 key.offset = (u64)-1;
2294 fs_info = BTRFS_I(inode)->root->fs_info;
2295 root = btrfs_read_fs_root_no_name(fs_info, &key);
2297 if (PTR_ERR(root) == -ENOENT)
2300 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2301 inum, offset, root_id);
2302 return PTR_ERR(root);
2305 key.objectid = inum;
2306 key.type = BTRFS_EXTENT_DATA_KEY;
2307 if (offset > (u64)-1 << 32)
2310 key.offset = offset;
2312 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2313 if (WARN_ON(ret < 0))
2320 leaf = path->nodes[0];
2321 slot = path->slots[0];
2323 if (slot >= btrfs_header_nritems(leaf)) {
2324 ret = btrfs_next_leaf(root, path);
2327 } else if (ret > 0) {
2336 btrfs_item_key_to_cpu(leaf, &key, slot);
2338 if (key.objectid > inum)
2341 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2344 extent = btrfs_item_ptr(leaf, slot,
2345 struct btrfs_file_extent_item);
2347 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2351 * 'offset' refers to the exact key.offset,
2352 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2353 * (key.offset - extent_offset).
2355 if (key.offset != offset)
2358 extent_offset = btrfs_file_extent_offset(leaf, extent);
2359 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2361 if (extent_offset >= old->extent_offset + old->offset +
2362 old->len || extent_offset + num_bytes <=
2363 old->extent_offset + old->offset)
2368 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2374 backref->root_id = root_id;
2375 backref->inum = inum;
2376 backref->file_pos = offset;
2377 backref->num_bytes = num_bytes;
2378 backref->extent_offset = extent_offset;
2379 backref->generation = btrfs_file_extent_generation(leaf, extent);
2381 backref_insert(&new->root, backref);
2384 btrfs_release_path(path);
2389 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2390 struct new_sa_defrag_extent *new)
2392 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2393 struct old_sa_defrag_extent *old, *tmp;
2398 list_for_each_entry_safe(old, tmp, &new->head, list) {
2399 ret = iterate_inodes_from_logical(old->bytenr +
2400 old->extent_offset, fs_info,
2401 path, record_one_backref,
2403 if (ret < 0 && ret != -ENOENT)
2406 /* no backref to be processed for this extent */
2408 list_del(&old->list);
2413 if (list_empty(&new->head))
2419 static int relink_is_mergable(struct extent_buffer *leaf,
2420 struct btrfs_file_extent_item *fi,
2421 struct new_sa_defrag_extent *new)
2423 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2426 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2429 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2432 if (btrfs_file_extent_encryption(leaf, fi) ||
2433 btrfs_file_extent_other_encoding(leaf, fi))
2440 * Note the backref might has changed, and in this case we just return 0.
2442 static noinline int relink_extent_backref(struct btrfs_path *path,
2443 struct sa_defrag_extent_backref *prev,
2444 struct sa_defrag_extent_backref *backref)
2446 struct btrfs_file_extent_item *extent;
2447 struct btrfs_file_extent_item *item;
2448 struct btrfs_ordered_extent *ordered;
2449 struct btrfs_trans_handle *trans;
2450 struct btrfs_fs_info *fs_info;
2451 struct btrfs_root *root;
2452 struct btrfs_key key;
2453 struct extent_buffer *leaf;
2454 struct old_sa_defrag_extent *old = backref->old;
2455 struct new_sa_defrag_extent *new = old->new;
2456 struct inode *src_inode = new->inode;
2457 struct inode *inode;
2458 struct extent_state *cached = NULL;
2467 if (prev && prev->root_id == backref->root_id &&
2468 prev->inum == backref->inum &&
2469 prev->file_pos + prev->num_bytes == backref->file_pos)
2472 /* step 1: get root */
2473 key.objectid = backref->root_id;
2474 key.type = BTRFS_ROOT_ITEM_KEY;
2475 key.offset = (u64)-1;
2477 fs_info = BTRFS_I(src_inode)->root->fs_info;
2478 index = srcu_read_lock(&fs_info->subvol_srcu);
2480 root = btrfs_read_fs_root_no_name(fs_info, &key);
2482 srcu_read_unlock(&fs_info->subvol_srcu, index);
2483 if (PTR_ERR(root) == -ENOENT)
2485 return PTR_ERR(root);
2488 if (btrfs_root_readonly(root)) {
2489 srcu_read_unlock(&fs_info->subvol_srcu, index);
2493 /* step 2: get inode */
2494 key.objectid = backref->inum;
2495 key.type = BTRFS_INODE_ITEM_KEY;
2498 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2499 if (IS_ERR(inode)) {
2500 srcu_read_unlock(&fs_info->subvol_srcu, index);
2504 srcu_read_unlock(&fs_info->subvol_srcu, index);
2506 /* step 3: relink backref */
2507 lock_start = backref->file_pos;
2508 lock_end = backref->file_pos + backref->num_bytes - 1;
2509 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2512 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2514 btrfs_put_ordered_extent(ordered);
2518 trans = btrfs_join_transaction(root);
2519 if (IS_ERR(trans)) {
2520 ret = PTR_ERR(trans);
2524 key.objectid = backref->inum;
2525 key.type = BTRFS_EXTENT_DATA_KEY;
2526 key.offset = backref->file_pos;
2528 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2531 } else if (ret > 0) {
2536 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2537 struct btrfs_file_extent_item);
2539 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2540 backref->generation)
2543 btrfs_release_path(path);
2545 start = backref->file_pos;
2546 if (backref->extent_offset < old->extent_offset + old->offset)
2547 start += old->extent_offset + old->offset -
2548 backref->extent_offset;
2550 len = min(backref->extent_offset + backref->num_bytes,
2551 old->extent_offset + old->offset + old->len);
2552 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2554 ret = btrfs_drop_extents(trans, root, inode, start,
2559 key.objectid = btrfs_ino(inode);
2560 key.type = BTRFS_EXTENT_DATA_KEY;
2563 path->leave_spinning = 1;
2565 struct btrfs_file_extent_item *fi;
2567 struct btrfs_key found_key;
2569 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2574 leaf = path->nodes[0];
2575 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2577 fi = btrfs_item_ptr(leaf, path->slots[0],
2578 struct btrfs_file_extent_item);
2579 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2581 if (extent_len + found_key.offset == start &&
2582 relink_is_mergable(leaf, fi, new)) {
2583 btrfs_set_file_extent_num_bytes(leaf, fi,
2585 btrfs_mark_buffer_dirty(leaf);
2586 inode_add_bytes(inode, len);
2592 btrfs_release_path(path);
2597 ret = btrfs_insert_empty_item(trans, root, path, &key,
2600 btrfs_abort_transaction(trans, root, ret);
2604 leaf = path->nodes[0];
2605 item = btrfs_item_ptr(leaf, path->slots[0],
2606 struct btrfs_file_extent_item);
2607 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2608 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2609 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2610 btrfs_set_file_extent_num_bytes(leaf, item, len);
2611 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2612 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2613 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2614 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2615 btrfs_set_file_extent_encryption(leaf, item, 0);
2616 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2618 btrfs_mark_buffer_dirty(leaf);
2619 inode_add_bytes(inode, len);
2620 btrfs_release_path(path);
2622 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2624 backref->root_id, backref->inum,
2625 new->file_pos); /* start - extent_offset */
2627 btrfs_abort_transaction(trans, root, ret);
2633 btrfs_release_path(path);
2634 path->leave_spinning = 0;
2635 btrfs_end_transaction(trans, root);
2637 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2643 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2645 struct old_sa_defrag_extent *old, *tmp;
2650 list_for_each_entry_safe(old, tmp, &new->head, list) {
2656 static void relink_file_extents(struct new_sa_defrag_extent *new)
2658 struct btrfs_path *path;
2659 struct sa_defrag_extent_backref *backref;
2660 struct sa_defrag_extent_backref *prev = NULL;
2661 struct inode *inode;
2662 struct btrfs_root *root;
2663 struct rb_node *node;
2667 root = BTRFS_I(inode)->root;
2669 path = btrfs_alloc_path();
2673 if (!record_extent_backrefs(path, new)) {
2674 btrfs_free_path(path);
2677 btrfs_release_path(path);
2680 node = rb_first(&new->root);
2683 rb_erase(node, &new->root);
2685 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2687 ret = relink_extent_backref(path, prev, backref);
2700 btrfs_free_path(path);
2702 free_sa_defrag_extent(new);
2704 atomic_dec(&root->fs_info->defrag_running);
2705 wake_up(&root->fs_info->transaction_wait);
2708 static struct new_sa_defrag_extent *
2709 record_old_file_extents(struct inode *inode,
2710 struct btrfs_ordered_extent *ordered)
2712 struct btrfs_root *root = BTRFS_I(inode)->root;
2713 struct btrfs_path *path;
2714 struct btrfs_key key;
2715 struct old_sa_defrag_extent *old;
2716 struct new_sa_defrag_extent *new;
2719 new = kmalloc(sizeof(*new), GFP_NOFS);
2724 new->file_pos = ordered->file_offset;
2725 new->len = ordered->len;
2726 new->bytenr = ordered->start;
2727 new->disk_len = ordered->disk_len;
2728 new->compress_type = ordered->compress_type;
2729 new->root = RB_ROOT;
2730 INIT_LIST_HEAD(&new->head);
2732 path = btrfs_alloc_path();
2736 key.objectid = btrfs_ino(inode);
2737 key.type = BTRFS_EXTENT_DATA_KEY;
2738 key.offset = new->file_pos;
2740 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2743 if (ret > 0 && path->slots[0] > 0)
2746 /* find out all the old extents for the file range */
2748 struct btrfs_file_extent_item *extent;
2749 struct extent_buffer *l;
2758 slot = path->slots[0];
2760 if (slot >= btrfs_header_nritems(l)) {
2761 ret = btrfs_next_leaf(root, path);
2769 btrfs_item_key_to_cpu(l, &key, slot);
2771 if (key.objectid != btrfs_ino(inode))
2773 if (key.type != BTRFS_EXTENT_DATA_KEY)
2775 if (key.offset >= new->file_pos + new->len)
2778 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2780 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2781 if (key.offset + num_bytes < new->file_pos)
2784 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2788 extent_offset = btrfs_file_extent_offset(l, extent);
2790 old = kmalloc(sizeof(*old), GFP_NOFS);
2794 offset = max(new->file_pos, key.offset);
2795 end = min(new->file_pos + new->len, key.offset + num_bytes);
2797 old->bytenr = disk_bytenr;
2798 old->extent_offset = extent_offset;
2799 old->offset = offset - key.offset;
2800 old->len = end - offset;
2803 list_add_tail(&old->list, &new->head);
2809 btrfs_free_path(path);
2810 atomic_inc(&root->fs_info->defrag_running);
2815 btrfs_free_path(path);
2817 free_sa_defrag_extent(new);
2821 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2824 struct btrfs_block_group_cache *cache;
2826 cache = btrfs_lookup_block_group(root->fs_info, start);
2829 spin_lock(&cache->lock);
2830 cache->delalloc_bytes -= len;
2831 spin_unlock(&cache->lock);
2833 btrfs_put_block_group(cache);
2836 /* as ordered data IO finishes, this gets called so we can finish
2837 * an ordered extent if the range of bytes in the file it covers are
2840 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2842 struct inode *inode = ordered_extent->inode;
2843 struct btrfs_root *root = BTRFS_I(inode)->root;
2844 struct btrfs_trans_handle *trans = NULL;
2845 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2846 struct extent_state *cached_state = NULL;
2847 struct new_sa_defrag_extent *new = NULL;
2848 int compress_type = 0;
2850 u64 logical_len = ordered_extent->len;
2852 bool truncated = false;
2854 nolock = btrfs_is_free_space_inode(inode);
2856 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2861 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2862 ordered_extent->file_offset +
2863 ordered_extent->len - 1);
2865 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2867 logical_len = ordered_extent->truncated_len;
2868 /* Truncated the entire extent, don't bother adding */
2873 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2874 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2877 * For mwrite(mmap + memset to write) case, we still reserve
2878 * space for NOCOW range.
2879 * As NOCOW won't cause a new delayed ref, just free the space
2881 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2882 ordered_extent->len);
2883 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2885 trans = btrfs_join_transaction_nolock(root);
2887 trans = btrfs_join_transaction(root);
2888 if (IS_ERR(trans)) {
2889 ret = PTR_ERR(trans);
2893 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2894 ret = btrfs_update_inode_fallback(trans, root, inode);
2895 if (ret) /* -ENOMEM or corruption */
2896 btrfs_abort_transaction(trans, root, ret);
2900 lock_extent_bits(io_tree, ordered_extent->file_offset,
2901 ordered_extent->file_offset + ordered_extent->len - 1,
2904 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2905 ordered_extent->file_offset + ordered_extent->len - 1,
2906 EXTENT_DEFRAG, 1, cached_state);
2908 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2909 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2910 /* the inode is shared */
2911 new = record_old_file_extents(inode, ordered_extent);
2913 clear_extent_bit(io_tree, ordered_extent->file_offset,
2914 ordered_extent->file_offset + ordered_extent->len - 1,
2915 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2919 trans = btrfs_join_transaction_nolock(root);
2921 trans = btrfs_join_transaction(root);
2922 if (IS_ERR(trans)) {
2923 ret = PTR_ERR(trans);
2928 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2930 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2931 compress_type = ordered_extent->compress_type;
2932 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2933 BUG_ON(compress_type);
2934 ret = btrfs_mark_extent_written(trans, inode,
2935 ordered_extent->file_offset,
2936 ordered_extent->file_offset +
2939 BUG_ON(root == root->fs_info->tree_root);
2940 ret = insert_reserved_file_extent(trans, inode,
2941 ordered_extent->file_offset,
2942 ordered_extent->start,
2943 ordered_extent->disk_len,
2944 logical_len, logical_len,
2945 compress_type, 0, 0,
2946 BTRFS_FILE_EXTENT_REG);
2948 btrfs_release_delalloc_bytes(root,
2949 ordered_extent->start,
2950 ordered_extent->disk_len);
2952 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2953 ordered_extent->file_offset, ordered_extent->len,
2956 btrfs_abort_transaction(trans, root, ret);
2960 add_pending_csums(trans, inode, ordered_extent->file_offset,
2961 &ordered_extent->list);
2963 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2964 ret = btrfs_update_inode_fallback(trans, root, inode);
2965 if (ret) { /* -ENOMEM or corruption */
2966 btrfs_abort_transaction(trans, root, ret);
2971 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2972 ordered_extent->file_offset +
2973 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2975 if (root != root->fs_info->tree_root)
2976 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2978 btrfs_end_transaction(trans, root);
2980 if (ret || truncated) {
2984 start = ordered_extent->file_offset + logical_len;
2986 start = ordered_extent->file_offset;
2987 end = ordered_extent->file_offset + ordered_extent->len - 1;
2988 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2990 /* Drop the cache for the part of the extent we didn't write. */
2991 btrfs_drop_extent_cache(inode, start, end, 0);
2994 * If the ordered extent had an IOERR or something else went
2995 * wrong we need to return the space for this ordered extent
2996 * back to the allocator. We only free the extent in the
2997 * truncated case if we didn't write out the extent at all.
2999 if ((ret || !logical_len) &&
3000 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3001 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3002 btrfs_free_reserved_extent(root, ordered_extent->start,
3003 ordered_extent->disk_len, 1);
3008 * This needs to be done to make sure anybody waiting knows we are done
3009 * updating everything for this ordered extent.
3011 btrfs_remove_ordered_extent(inode, ordered_extent);
3013 /* for snapshot-aware defrag */
3016 free_sa_defrag_extent(new);
3017 atomic_dec(&root->fs_info->defrag_running);
3019 relink_file_extents(new);
3024 btrfs_put_ordered_extent(ordered_extent);
3025 /* once for the tree */
3026 btrfs_put_ordered_extent(ordered_extent);
3031 static void finish_ordered_fn(struct btrfs_work *work)
3033 struct btrfs_ordered_extent *ordered_extent;
3034 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3035 btrfs_finish_ordered_io(ordered_extent);
3038 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3039 struct extent_state *state, int uptodate)
3041 struct inode *inode = page->mapping->host;
3042 struct btrfs_root *root = BTRFS_I(inode)->root;
3043 struct btrfs_ordered_extent *ordered_extent = NULL;
3044 struct btrfs_workqueue *wq;
3045 btrfs_work_func_t func;
3047 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3049 ClearPagePrivate2(page);
3050 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3051 end - start + 1, uptodate))
3054 if (btrfs_is_free_space_inode(inode)) {
3055 wq = root->fs_info->endio_freespace_worker;
3056 func = btrfs_freespace_write_helper;
3058 wq = root->fs_info->endio_write_workers;
3059 func = btrfs_endio_write_helper;
3062 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3064 btrfs_queue_work(wq, &ordered_extent->work);
3069 static int __readpage_endio_check(struct inode *inode,
3070 struct btrfs_io_bio *io_bio,
3071 int icsum, struct page *page,
3072 int pgoff, u64 start, size_t len)
3078 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3080 kaddr = kmap_atomic(page);
3081 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3082 btrfs_csum_final(csum, (char *)&csum);
3083 if (csum != csum_expected)
3086 kunmap_atomic(kaddr);
3089 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3090 "csum failed ino %llu off %llu csum %u expected csum %u",
3091 btrfs_ino(inode), start, csum, csum_expected);
3092 memset(kaddr + pgoff, 1, len);
3093 flush_dcache_page(page);
3094 kunmap_atomic(kaddr);
3095 if (csum_expected == 0)
3101 * when reads are done, we need to check csums to verify the data is correct
3102 * if there's a match, we allow the bio to finish. If not, the code in
3103 * extent_io.c will try to find good copies for us.
3105 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3106 u64 phy_offset, struct page *page,
3107 u64 start, u64 end, int mirror)
3109 size_t offset = start - page_offset(page);
3110 struct inode *inode = page->mapping->host;
3111 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3112 struct btrfs_root *root = BTRFS_I(inode)->root;
3114 if (PageChecked(page)) {
3115 ClearPageChecked(page);
3119 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3122 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3123 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3124 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3128 phy_offset >>= inode->i_sb->s_blocksize_bits;
3129 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3130 start, (size_t)(end - start + 1));
3133 void btrfs_add_delayed_iput(struct inode *inode)
3135 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3136 struct btrfs_inode *binode = BTRFS_I(inode);
3138 if (atomic_add_unless(&inode->i_count, -1, 1))
3141 spin_lock(&fs_info->delayed_iput_lock);
3142 if (binode->delayed_iput_count == 0) {
3143 ASSERT(list_empty(&binode->delayed_iput));
3144 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3146 binode->delayed_iput_count++;
3148 spin_unlock(&fs_info->delayed_iput_lock);
3151 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3153 struct btrfs_fs_info *fs_info = root->fs_info;
3155 spin_lock(&fs_info->delayed_iput_lock);
3156 while (!list_empty(&fs_info->delayed_iputs)) {
3157 struct btrfs_inode *inode;
3159 inode = list_first_entry(&fs_info->delayed_iputs,
3160 struct btrfs_inode, delayed_iput);
3161 if (inode->delayed_iput_count) {
3162 inode->delayed_iput_count--;
3163 list_move_tail(&inode->delayed_iput,
3164 &fs_info->delayed_iputs);
3166 list_del_init(&inode->delayed_iput);
3168 spin_unlock(&fs_info->delayed_iput_lock);
3169 iput(&inode->vfs_inode);
3170 spin_lock(&fs_info->delayed_iput_lock);
3172 spin_unlock(&fs_info->delayed_iput_lock);
3176 * This is called in transaction commit time. If there are no orphan
3177 * files in the subvolume, it removes orphan item and frees block_rsv
3180 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3181 struct btrfs_root *root)
3183 struct btrfs_block_rsv *block_rsv;
3186 if (atomic_read(&root->orphan_inodes) ||
3187 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3190 spin_lock(&root->orphan_lock);
3191 if (atomic_read(&root->orphan_inodes)) {
3192 spin_unlock(&root->orphan_lock);
3196 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3197 spin_unlock(&root->orphan_lock);
3201 block_rsv = root->orphan_block_rsv;
3202 root->orphan_block_rsv = NULL;
3203 spin_unlock(&root->orphan_lock);
3205 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3206 btrfs_root_refs(&root->root_item) > 0) {
3207 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3208 root->root_key.objectid);
3210 btrfs_abort_transaction(trans, root, ret);
3212 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3217 WARN_ON(block_rsv->size > 0);
3218 btrfs_free_block_rsv(root, block_rsv);
3223 * This creates an orphan entry for the given inode in case something goes
3224 * wrong in the middle of an unlink/truncate.
3226 * NOTE: caller of this function should reserve 5 units of metadata for
3229 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3231 struct btrfs_root *root = BTRFS_I(inode)->root;
3232 struct btrfs_block_rsv *block_rsv = NULL;
3237 if (!root->orphan_block_rsv) {
3238 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3243 spin_lock(&root->orphan_lock);
3244 if (!root->orphan_block_rsv) {
3245 root->orphan_block_rsv = block_rsv;
3246 } else if (block_rsv) {
3247 btrfs_free_block_rsv(root, block_rsv);
3251 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3252 &BTRFS_I(inode)->runtime_flags)) {
3255 * For proper ENOSPC handling, we should do orphan
3256 * cleanup when mounting. But this introduces backward
3257 * compatibility issue.
3259 if (!xchg(&root->orphan_item_inserted, 1))
3265 atomic_inc(&root->orphan_inodes);
3268 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3269 &BTRFS_I(inode)->runtime_flags))
3271 spin_unlock(&root->orphan_lock);
3273 /* grab metadata reservation from transaction handle */
3275 ret = btrfs_orphan_reserve_metadata(trans, inode);
3278 atomic_dec(&root->orphan_inodes);
3279 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3280 &BTRFS_I(inode)->runtime_flags);
3282 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3283 &BTRFS_I(inode)->runtime_flags);
3288 /* insert an orphan item to track this unlinked/truncated file */
3290 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3292 atomic_dec(&root->orphan_inodes);
3294 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3295 &BTRFS_I(inode)->runtime_flags);
3296 btrfs_orphan_release_metadata(inode);
3298 if (ret != -EEXIST) {
3299 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3300 &BTRFS_I(inode)->runtime_flags);
3301 btrfs_abort_transaction(trans, root, ret);
3308 /* insert an orphan item to track subvolume contains orphan files */
3310 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3311 root->root_key.objectid);
3312 if (ret && ret != -EEXIST) {
3313 btrfs_abort_transaction(trans, root, ret);
3321 * We have done the truncate/delete so we can go ahead and remove the orphan
3322 * item for this particular inode.
3324 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3325 struct inode *inode)
3327 struct btrfs_root *root = BTRFS_I(inode)->root;
3328 int delete_item = 0;
3329 int release_rsv = 0;
3332 spin_lock(&root->orphan_lock);
3333 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3334 &BTRFS_I(inode)->runtime_flags))
3337 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3338 &BTRFS_I(inode)->runtime_flags))
3340 spin_unlock(&root->orphan_lock);
3343 atomic_dec(&root->orphan_inodes);
3345 ret = btrfs_del_orphan_item(trans, root,
3350 btrfs_orphan_release_metadata(inode);
3356 * this cleans up any orphans that may be left on the list from the last use
3359 int btrfs_orphan_cleanup(struct btrfs_root *root)
3361 struct btrfs_path *path;
3362 struct extent_buffer *leaf;
3363 struct btrfs_key key, found_key;
3364 struct btrfs_trans_handle *trans;
3365 struct inode *inode;
3366 u64 last_objectid = 0;
3367 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3369 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3372 path = btrfs_alloc_path();
3377 path->reada = READA_BACK;
3379 key.objectid = BTRFS_ORPHAN_OBJECTID;
3380 key.type = BTRFS_ORPHAN_ITEM_KEY;
3381 key.offset = (u64)-1;
3384 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3389 * if ret == 0 means we found what we were searching for, which
3390 * is weird, but possible, so only screw with path if we didn't
3391 * find the key and see if we have stuff that matches
3395 if (path->slots[0] == 0)
3400 /* pull out the item */
3401 leaf = path->nodes[0];
3402 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3404 /* make sure the item matches what we want */
3405 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3407 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3410 /* release the path since we're done with it */
3411 btrfs_release_path(path);
3414 * this is where we are basically btrfs_lookup, without the
3415 * crossing root thing. we store the inode number in the
3416 * offset of the orphan item.
3419 if (found_key.offset == last_objectid) {
3420 btrfs_err(root->fs_info,
3421 "Error removing orphan entry, stopping orphan cleanup");
3426 last_objectid = found_key.offset;
3428 found_key.objectid = found_key.offset;
3429 found_key.type = BTRFS_INODE_ITEM_KEY;
3430 found_key.offset = 0;
3431 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3432 ret = PTR_ERR_OR_ZERO(inode);
3433 if (ret && ret != -ESTALE)
3436 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3437 struct btrfs_root *dead_root;
3438 struct btrfs_fs_info *fs_info = root->fs_info;
3439 int is_dead_root = 0;
3442 * this is an orphan in the tree root. Currently these
3443 * could come from 2 sources:
3444 * a) a snapshot deletion in progress
3445 * b) a free space cache inode
3446 * We need to distinguish those two, as the snapshot
3447 * orphan must not get deleted.
3448 * find_dead_roots already ran before us, so if this
3449 * is a snapshot deletion, we should find the root
3450 * in the dead_roots list
3452 spin_lock(&fs_info->trans_lock);
3453 list_for_each_entry(dead_root, &fs_info->dead_roots,
3455 if (dead_root->root_key.objectid ==
3456 found_key.objectid) {
3461 spin_unlock(&fs_info->trans_lock);
3463 /* prevent this orphan from being found again */
3464 key.offset = found_key.objectid - 1;
3469 * Inode is already gone but the orphan item is still there,
3470 * kill the orphan item.
3472 if (ret == -ESTALE) {
3473 trans = btrfs_start_transaction(root, 1);
3474 if (IS_ERR(trans)) {
3475 ret = PTR_ERR(trans);
3478 btrfs_debug(root->fs_info, "auto deleting %Lu",
3479 found_key.objectid);
3480 ret = btrfs_del_orphan_item(trans, root,
3481 found_key.objectid);
3482 btrfs_end_transaction(trans, root);
3489 * add this inode to the orphan list so btrfs_orphan_del does
3490 * the proper thing when we hit it
3492 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3493 &BTRFS_I(inode)->runtime_flags);
3494 atomic_inc(&root->orphan_inodes);
3496 /* if we have links, this was a truncate, lets do that */
3497 if (inode->i_nlink) {
3498 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3504 /* 1 for the orphan item deletion. */
3505 trans = btrfs_start_transaction(root, 1);
3506 if (IS_ERR(trans)) {
3508 ret = PTR_ERR(trans);
3511 ret = btrfs_orphan_add(trans, inode);
3512 btrfs_end_transaction(trans, root);
3518 ret = btrfs_truncate(inode);
3520 btrfs_orphan_del(NULL, inode);
3525 /* this will do delete_inode and everything for us */
3530 /* release the path since we're done with it */
3531 btrfs_release_path(path);
3533 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3535 if (root->orphan_block_rsv)
3536 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3539 if (root->orphan_block_rsv ||
3540 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3541 trans = btrfs_join_transaction(root);
3543 btrfs_end_transaction(trans, root);
3547 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3549 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3553 btrfs_err(root->fs_info,
3554 "could not do orphan cleanup %d", ret);
3555 btrfs_free_path(path);
3560 * very simple check to peek ahead in the leaf looking for xattrs. If we
3561 * don't find any xattrs, we know there can't be any acls.
3563 * slot is the slot the inode is in, objectid is the objectid of the inode
3565 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3566 int slot, u64 objectid,
3567 int *first_xattr_slot)
3569 u32 nritems = btrfs_header_nritems(leaf);
3570 struct btrfs_key found_key;
3571 static u64 xattr_access = 0;
3572 static u64 xattr_default = 0;
3575 if (!xattr_access) {
3576 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3577 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3578 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3579 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3583 *first_xattr_slot = -1;
3584 while (slot < nritems) {
3585 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3587 /* we found a different objectid, there must not be acls */
3588 if (found_key.objectid != objectid)
3591 /* we found an xattr, assume we've got an acl */
3592 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3593 if (*first_xattr_slot == -1)
3594 *first_xattr_slot = slot;
3595 if (found_key.offset == xattr_access ||
3596 found_key.offset == xattr_default)
3601 * we found a key greater than an xattr key, there can't
3602 * be any acls later on
3604 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3611 * it goes inode, inode backrefs, xattrs, extents,
3612 * so if there are a ton of hard links to an inode there can
3613 * be a lot of backrefs. Don't waste time searching too hard,
3614 * this is just an optimization
3619 /* we hit the end of the leaf before we found an xattr or
3620 * something larger than an xattr. We have to assume the inode
3623 if (*first_xattr_slot == -1)
3624 *first_xattr_slot = slot;
3629 * read an inode from the btree into the in-memory inode
3631 static void btrfs_read_locked_inode(struct inode *inode)
3633 struct btrfs_path *path;
3634 struct extent_buffer *leaf;
3635 struct btrfs_inode_item *inode_item;
3636 struct btrfs_root *root = BTRFS_I(inode)->root;
3637 struct btrfs_key location;
3642 bool filled = false;
3643 int first_xattr_slot;
3645 ret = btrfs_fill_inode(inode, &rdev);
3649 path = btrfs_alloc_path();
3653 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3655 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3659 leaf = path->nodes[0];
3664 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3665 struct btrfs_inode_item);
3666 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3667 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3668 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3669 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3670 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3672 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3673 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3675 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3676 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3678 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3679 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3681 BTRFS_I(inode)->i_otime.tv_sec =
3682 btrfs_timespec_sec(leaf, &inode_item->otime);
3683 BTRFS_I(inode)->i_otime.tv_nsec =
3684 btrfs_timespec_nsec(leaf, &inode_item->otime);
3686 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3687 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3688 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3690 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3691 inode->i_generation = BTRFS_I(inode)->generation;
3693 rdev = btrfs_inode_rdev(leaf, inode_item);
3695 BTRFS_I(inode)->index_cnt = (u64)-1;
3696 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3700 * If we were modified in the current generation and evicted from memory
3701 * and then re-read we need to do a full sync since we don't have any
3702 * idea about which extents were modified before we were evicted from
3705 * This is required for both inode re-read from disk and delayed inode
3706 * in delayed_nodes_tree.
3708 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3709 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3710 &BTRFS_I(inode)->runtime_flags);
3713 * We don't persist the id of the transaction where an unlink operation
3714 * against the inode was last made. So here we assume the inode might
3715 * have been evicted, and therefore the exact value of last_unlink_trans
3716 * lost, and set it to last_trans to avoid metadata inconsistencies
3717 * between the inode and its parent if the inode is fsync'ed and the log
3718 * replayed. For example, in the scenario:
3721 * ln mydir/foo mydir/bar
3724 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3725 * xfs_io -c fsync mydir/foo
3727 * mount fs, triggers fsync log replay
3729 * We must make sure that when we fsync our inode foo we also log its
3730 * parent inode, otherwise after log replay the parent still has the
3731 * dentry with the "bar" name but our inode foo has a link count of 1
3732 * and doesn't have an inode ref with the name "bar" anymore.
3734 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3735 * but it guarantees correctness at the expense of occasional full
3736 * transaction commits on fsync if our inode is a directory, or if our
3737 * inode is not a directory, logging its parent unnecessarily.
3739 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3742 if (inode->i_nlink != 1 ||
3743 path->slots[0] >= btrfs_header_nritems(leaf))
3746 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3747 if (location.objectid != btrfs_ino(inode))
3750 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3751 if (location.type == BTRFS_INODE_REF_KEY) {
3752 struct btrfs_inode_ref *ref;
3754 ref = (struct btrfs_inode_ref *)ptr;
3755 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3756 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3757 struct btrfs_inode_extref *extref;
3759 extref = (struct btrfs_inode_extref *)ptr;
3760 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3765 * try to precache a NULL acl entry for files that don't have
3766 * any xattrs or acls
3768 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3769 btrfs_ino(inode), &first_xattr_slot);
3770 if (first_xattr_slot != -1) {
3771 path->slots[0] = first_xattr_slot;
3772 ret = btrfs_load_inode_props(inode, path);
3774 btrfs_err(root->fs_info,
3775 "error loading props for ino %llu (root %llu): %d",
3777 root->root_key.objectid, ret);
3779 btrfs_free_path(path);
3782 cache_no_acl(inode);
3784 switch (inode->i_mode & S_IFMT) {
3786 inode->i_mapping->a_ops = &btrfs_aops;
3787 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3788 inode->i_fop = &btrfs_file_operations;
3789 inode->i_op = &btrfs_file_inode_operations;
3792 inode->i_fop = &btrfs_dir_file_operations;
3793 if (root == root->fs_info->tree_root)
3794 inode->i_op = &btrfs_dir_ro_inode_operations;
3796 inode->i_op = &btrfs_dir_inode_operations;
3799 inode->i_op = &btrfs_symlink_inode_operations;
3800 inode_nohighmem(inode);
3801 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3804 inode->i_op = &btrfs_special_inode_operations;
3805 init_special_inode(inode, inode->i_mode, rdev);
3809 btrfs_update_iflags(inode);
3813 btrfs_free_path(path);
3814 make_bad_inode(inode);
3818 * given a leaf and an inode, copy the inode fields into the leaf
3820 static void fill_inode_item(struct btrfs_trans_handle *trans,
3821 struct extent_buffer *leaf,
3822 struct btrfs_inode_item *item,
3823 struct inode *inode)
3825 struct btrfs_map_token token;
3827 btrfs_init_map_token(&token);
3829 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3830 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3831 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3833 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3834 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3836 btrfs_set_token_timespec_sec(leaf, &item->atime,
3837 inode->i_atime.tv_sec, &token);
3838 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3839 inode->i_atime.tv_nsec, &token);
3841 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3842 inode->i_mtime.tv_sec, &token);
3843 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3844 inode->i_mtime.tv_nsec, &token);
3846 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3847 inode->i_ctime.tv_sec, &token);
3848 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3849 inode->i_ctime.tv_nsec, &token);
3851 btrfs_set_token_timespec_sec(leaf, &item->otime,
3852 BTRFS_I(inode)->i_otime.tv_sec, &token);
3853 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3854 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3856 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3858 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3860 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3861 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3862 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3863 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3864 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3868 * copy everything in the in-memory inode into the btree.
3870 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3871 struct btrfs_root *root, struct inode *inode)
3873 struct btrfs_inode_item *inode_item;
3874 struct btrfs_path *path;
3875 struct extent_buffer *leaf;
3878 path = btrfs_alloc_path();
3882 path->leave_spinning = 1;
3883 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3891 leaf = path->nodes[0];
3892 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3893 struct btrfs_inode_item);
3895 fill_inode_item(trans, leaf, inode_item, inode);
3896 btrfs_mark_buffer_dirty(leaf);
3897 btrfs_set_inode_last_trans(trans, inode);
3900 btrfs_free_path(path);
3905 * copy everything in the in-memory inode into the btree.
3907 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3908 struct btrfs_root *root, struct inode *inode)
3913 * If the inode is a free space inode, we can deadlock during commit
3914 * if we put it into the delayed code.
3916 * The data relocation inode should also be directly updated
3919 if (!btrfs_is_free_space_inode(inode)
3920 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3921 && !root->fs_info->log_root_recovering) {
3922 btrfs_update_root_times(trans, root);
3924 ret = btrfs_delayed_update_inode(trans, root, inode);
3926 btrfs_set_inode_last_trans(trans, inode);
3930 return btrfs_update_inode_item(trans, root, inode);
3933 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3934 struct btrfs_root *root,
3935 struct inode *inode)
3939 ret = btrfs_update_inode(trans, root, inode);
3941 return btrfs_update_inode_item(trans, root, inode);
3946 * unlink helper that gets used here in inode.c and in the tree logging
3947 * recovery code. It remove a link in a directory with a given name, and
3948 * also drops the back refs in the inode to the directory
3950 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3951 struct btrfs_root *root,
3952 struct inode *dir, struct inode *inode,
3953 const char *name, int name_len)
3955 struct btrfs_path *path;
3957 struct extent_buffer *leaf;
3958 struct btrfs_dir_item *di;
3959 struct btrfs_key key;
3961 u64 ino = btrfs_ino(inode);
3962 u64 dir_ino = btrfs_ino(dir);
3964 path = btrfs_alloc_path();
3970 path->leave_spinning = 1;
3971 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3972 name, name_len, -1);
3981 leaf = path->nodes[0];
3982 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3983 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3986 btrfs_release_path(path);
3989 * If we don't have dir index, we have to get it by looking up
3990 * the inode ref, since we get the inode ref, remove it directly,
3991 * it is unnecessary to do delayed deletion.
3993 * But if we have dir index, needn't search inode ref to get it.
3994 * Since the inode ref is close to the inode item, it is better
3995 * that we delay to delete it, and just do this deletion when
3996 * we update the inode item.
3998 if (BTRFS_I(inode)->dir_index) {
3999 ret = btrfs_delayed_delete_inode_ref(inode);
4001 index = BTRFS_I(inode)->dir_index;
4006 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4009 btrfs_info(root->fs_info,
4010 "failed to delete reference to %.*s, inode %llu parent %llu",
4011 name_len, name, ino, dir_ino);
4012 btrfs_abort_transaction(trans, root, ret);
4016 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4018 btrfs_abort_transaction(trans, root, ret);
4022 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4024 if (ret != 0 && ret != -ENOENT) {
4025 btrfs_abort_transaction(trans, root, ret);
4029 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4034 btrfs_abort_transaction(trans, root, ret);
4036 btrfs_free_path(path);
4040 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4041 inode_inc_iversion(inode);
4042 inode_inc_iversion(dir);
4043 inode->i_ctime = dir->i_mtime =
4044 dir->i_ctime = current_fs_time(inode->i_sb);
4045 ret = btrfs_update_inode(trans, root, dir);
4050 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4051 struct btrfs_root *root,
4052 struct inode *dir, struct inode *inode,
4053 const char *name, int name_len)
4056 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4059 ret = btrfs_update_inode(trans, root, inode);
4065 * helper to start transaction for unlink and rmdir.
4067 * unlink and rmdir are special in btrfs, they do not always free space, so
4068 * if we cannot make our reservations the normal way try and see if there is
4069 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4070 * allow the unlink to occur.
4072 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4074 struct btrfs_root *root = BTRFS_I(dir)->root;
4077 * 1 for the possible orphan item
4078 * 1 for the dir item
4079 * 1 for the dir index
4080 * 1 for the inode ref
4083 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4086 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4088 struct btrfs_root *root = BTRFS_I(dir)->root;
4089 struct btrfs_trans_handle *trans;
4090 struct inode *inode = d_inode(dentry);
4093 trans = __unlink_start_trans(dir);
4095 return PTR_ERR(trans);
4097 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4099 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4100 dentry->d_name.name, dentry->d_name.len);
4104 if (inode->i_nlink == 0) {
4105 ret = btrfs_orphan_add(trans, inode);
4111 btrfs_end_transaction(trans, root);
4112 btrfs_btree_balance_dirty(root);
4116 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4117 struct btrfs_root *root,
4118 struct inode *dir, u64 objectid,
4119 const char *name, int name_len)
4121 struct btrfs_path *path;
4122 struct extent_buffer *leaf;
4123 struct btrfs_dir_item *di;
4124 struct btrfs_key key;
4127 u64 dir_ino = btrfs_ino(dir);
4129 path = btrfs_alloc_path();
4133 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4134 name, name_len, -1);
4135 if (IS_ERR_OR_NULL(di)) {
4143 leaf = path->nodes[0];
4144 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4145 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4146 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4148 btrfs_abort_transaction(trans, root, ret);
4151 btrfs_release_path(path);
4153 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4154 objectid, root->root_key.objectid,
4155 dir_ino, &index, name, name_len);
4157 if (ret != -ENOENT) {
4158 btrfs_abort_transaction(trans, root, ret);
4161 di = btrfs_search_dir_index_item(root, path, dir_ino,
4163 if (IS_ERR_OR_NULL(di)) {
4168 btrfs_abort_transaction(trans, root, ret);
4172 leaf = path->nodes[0];
4173 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4174 btrfs_release_path(path);
4177 btrfs_release_path(path);
4179 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4181 btrfs_abort_transaction(trans, root, ret);
4185 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4186 inode_inc_iversion(dir);
4187 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4188 ret = btrfs_update_inode_fallback(trans, root, dir);
4190 btrfs_abort_transaction(trans, root, ret);
4192 btrfs_free_path(path);
4196 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4198 struct inode *inode = d_inode(dentry);
4200 struct btrfs_root *root = BTRFS_I(dir)->root;
4201 struct btrfs_trans_handle *trans;
4203 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4205 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4208 trans = __unlink_start_trans(dir);
4210 return PTR_ERR(trans);
4212 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4213 err = btrfs_unlink_subvol(trans, root, dir,
4214 BTRFS_I(inode)->location.objectid,
4215 dentry->d_name.name,
4216 dentry->d_name.len);
4220 err = btrfs_orphan_add(trans, inode);
4224 /* now the directory is empty */
4225 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4226 dentry->d_name.name, dentry->d_name.len);
4228 btrfs_i_size_write(inode, 0);
4230 btrfs_end_transaction(trans, root);
4231 btrfs_btree_balance_dirty(root);
4236 static int truncate_space_check(struct btrfs_trans_handle *trans,
4237 struct btrfs_root *root,
4243 * This is only used to apply pressure to the enospc system, we don't
4244 * intend to use this reservation at all.
4246 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4247 bytes_deleted *= root->nodesize;
4248 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4249 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4251 trace_btrfs_space_reservation(root->fs_info, "transaction",
4254 trans->bytes_reserved += bytes_deleted;
4260 static int truncate_inline_extent(struct inode *inode,
4261 struct btrfs_path *path,
4262 struct btrfs_key *found_key,
4266 struct extent_buffer *leaf = path->nodes[0];
4267 int slot = path->slots[0];
4268 struct btrfs_file_extent_item *fi;
4269 u32 size = (u32)(new_size - found_key->offset);
4270 struct btrfs_root *root = BTRFS_I(inode)->root;
4272 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4274 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4275 loff_t offset = new_size;
4276 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4279 * Zero out the remaining of the last page of our inline extent,
4280 * instead of directly truncating our inline extent here - that
4281 * would be much more complex (decompressing all the data, then
4282 * compressing the truncated data, which might be bigger than
4283 * the size of the inline extent, resize the extent, etc).
4284 * We release the path because to get the page we might need to
4285 * read the extent item from disk (data not in the page cache).
4287 btrfs_release_path(path);
4288 return btrfs_truncate_block(inode, offset, page_end - offset,
4292 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4293 size = btrfs_file_extent_calc_inline_size(size);
4294 btrfs_truncate_item(root, path, size, 1);
4296 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4297 inode_sub_bytes(inode, item_end + 1 - new_size);
4303 * this can truncate away extent items, csum items and directory items.
4304 * It starts at a high offset and removes keys until it can't find
4305 * any higher than new_size
4307 * csum items that cross the new i_size are truncated to the new size
4310 * min_type is the minimum key type to truncate down to. If set to 0, this
4311 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4313 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4314 struct btrfs_root *root,
4315 struct inode *inode,
4316 u64 new_size, u32 min_type)
4318 struct btrfs_path *path;
4319 struct extent_buffer *leaf;
4320 struct btrfs_file_extent_item *fi;
4321 struct btrfs_key key;
4322 struct btrfs_key found_key;
4323 u64 extent_start = 0;
4324 u64 extent_num_bytes = 0;
4325 u64 extent_offset = 0;
4327 u64 last_size = new_size;
4328 u32 found_type = (u8)-1;
4331 int pending_del_nr = 0;
4332 int pending_del_slot = 0;
4333 int extent_type = -1;
4336 u64 ino = btrfs_ino(inode);
4337 u64 bytes_deleted = 0;
4339 bool should_throttle = 0;
4340 bool should_end = 0;
4342 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4345 * for non-free space inodes and ref cows, we want to back off from
4348 if (!btrfs_is_free_space_inode(inode) &&
4349 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4352 path = btrfs_alloc_path();
4355 path->reada = READA_BACK;
4358 * We want to drop from the next block forward in case this new size is
4359 * not block aligned since we will be keeping the last block of the
4360 * extent just the way it is.
4362 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4363 root == root->fs_info->tree_root)
4364 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4365 root->sectorsize), (u64)-1, 0);
4368 * This function is also used to drop the items in the log tree before
4369 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4370 * it is used to drop the loged items. So we shouldn't kill the delayed
4373 if (min_type == 0 && root == BTRFS_I(inode)->root)
4374 btrfs_kill_delayed_inode_items(inode);
4377 key.offset = (u64)-1;
4382 * with a 16K leaf size and 128MB extents, you can actually queue
4383 * up a huge file in a single leaf. Most of the time that
4384 * bytes_deleted is > 0, it will be huge by the time we get here
4386 if (be_nice && bytes_deleted > SZ_32M) {
4387 if (btrfs_should_end_transaction(trans, root)) {
4394 path->leave_spinning = 1;
4395 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4402 /* there are no items in the tree for us to truncate, we're
4405 if (path->slots[0] == 0)
4412 leaf = path->nodes[0];
4413 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4414 found_type = found_key.type;
4416 if (found_key.objectid != ino)
4419 if (found_type < min_type)
4422 item_end = found_key.offset;
4423 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4424 fi = btrfs_item_ptr(leaf, path->slots[0],
4425 struct btrfs_file_extent_item);
4426 extent_type = btrfs_file_extent_type(leaf, fi);
4427 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4429 btrfs_file_extent_num_bytes(leaf, fi);
4430 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4431 item_end += btrfs_file_extent_inline_len(leaf,
4432 path->slots[0], fi);
4436 if (found_type > min_type) {
4439 if (item_end < new_size)
4441 if (found_key.offset >= new_size)
4447 /* FIXME, shrink the extent if the ref count is only 1 */
4448 if (found_type != BTRFS_EXTENT_DATA_KEY)
4452 last_size = found_key.offset;
4454 last_size = new_size;
4456 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4458 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4460 u64 orig_num_bytes =
4461 btrfs_file_extent_num_bytes(leaf, fi);
4462 extent_num_bytes = ALIGN(new_size -
4465 btrfs_set_file_extent_num_bytes(leaf, fi,
4467 num_dec = (orig_num_bytes -
4469 if (test_bit(BTRFS_ROOT_REF_COWS,
4472 inode_sub_bytes(inode, num_dec);
4473 btrfs_mark_buffer_dirty(leaf);
4476 btrfs_file_extent_disk_num_bytes(leaf,
4478 extent_offset = found_key.offset -
4479 btrfs_file_extent_offset(leaf, fi);
4481 /* FIXME blocksize != 4096 */
4482 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4483 if (extent_start != 0) {
4485 if (test_bit(BTRFS_ROOT_REF_COWS,
4487 inode_sub_bytes(inode, num_dec);
4490 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4492 * we can't truncate inline items that have had
4496 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4497 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4500 * Need to release path in order to truncate a
4501 * compressed extent. So delete any accumulated
4502 * extent items so far.
4504 if (btrfs_file_extent_compression(leaf, fi) !=
4505 BTRFS_COMPRESS_NONE && pending_del_nr) {
4506 err = btrfs_del_items(trans, root, path,
4510 btrfs_abort_transaction(trans,
4518 err = truncate_inline_extent(inode, path,
4523 btrfs_abort_transaction(trans,
4527 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4529 inode_sub_bytes(inode, item_end + 1 - new_size);
4534 if (!pending_del_nr) {
4535 /* no pending yet, add ourselves */
4536 pending_del_slot = path->slots[0];
4538 } else if (pending_del_nr &&
4539 path->slots[0] + 1 == pending_del_slot) {
4540 /* hop on the pending chunk */
4542 pending_del_slot = path->slots[0];
4549 should_throttle = 0;
4552 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4553 root == root->fs_info->tree_root)) {
4554 btrfs_set_path_blocking(path);
4555 bytes_deleted += extent_num_bytes;
4556 ret = btrfs_free_extent(trans, root, extent_start,
4557 extent_num_bytes, 0,
4558 btrfs_header_owner(leaf),
4559 ino, extent_offset);
4561 if (btrfs_should_throttle_delayed_refs(trans, root))
4562 btrfs_async_run_delayed_refs(root,
4564 trans->delayed_ref_updates * 2, 0);
4566 if (truncate_space_check(trans, root,
4567 extent_num_bytes)) {
4570 if (btrfs_should_throttle_delayed_refs(trans,
4572 should_throttle = 1;
4577 if (found_type == BTRFS_INODE_ITEM_KEY)
4580 if (path->slots[0] == 0 ||
4581 path->slots[0] != pending_del_slot ||
4582 should_throttle || should_end) {
4583 if (pending_del_nr) {
4584 ret = btrfs_del_items(trans, root, path,
4588 btrfs_abort_transaction(trans,
4594 btrfs_release_path(path);
4595 if (should_throttle) {
4596 unsigned long updates = trans->delayed_ref_updates;
4598 trans->delayed_ref_updates = 0;
4599 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4605 * if we failed to refill our space rsv, bail out
4606 * and let the transaction restart
4618 if (pending_del_nr) {
4619 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4622 btrfs_abort_transaction(trans, root, ret);
4625 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4626 btrfs_ordered_update_i_size(inode, last_size, NULL);
4628 btrfs_free_path(path);
4630 if (be_nice && bytes_deleted > SZ_32M) {
4631 unsigned long updates = trans->delayed_ref_updates;
4633 trans->delayed_ref_updates = 0;
4634 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4643 * btrfs_truncate_block - read, zero a chunk and write a block
4644 * @inode - inode that we're zeroing
4645 * @from - the offset to start zeroing
4646 * @len - the length to zero, 0 to zero the entire range respective to the
4648 * @front - zero up to the offset instead of from the offset on
4650 * This will find the block for the "from" offset and cow the block and zero the
4651 * part we want to zero. This is used with truncate and hole punching.
4653 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4656 struct address_space *mapping = inode->i_mapping;
4657 struct btrfs_root *root = BTRFS_I(inode)->root;
4658 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4659 struct btrfs_ordered_extent *ordered;
4660 struct extent_state *cached_state = NULL;
4662 u32 blocksize = root->sectorsize;
4663 pgoff_t index = from >> PAGE_SHIFT;
4664 unsigned offset = from & (blocksize - 1);
4666 gfp_t mask = btrfs_alloc_write_mask(mapping);
4671 if ((offset & (blocksize - 1)) == 0 &&
4672 (!len || ((len & (blocksize - 1)) == 0)))
4675 ret = btrfs_delalloc_reserve_space(inode,
4676 round_down(from, blocksize), blocksize);
4681 page = find_or_create_page(mapping, index, mask);
4683 btrfs_delalloc_release_space(inode,
4684 round_down(from, blocksize),
4690 block_start = round_down(from, blocksize);
4691 block_end = block_start + blocksize - 1;
4693 if (!PageUptodate(page)) {
4694 ret = btrfs_readpage(NULL, page);
4696 if (page->mapping != mapping) {
4701 if (!PageUptodate(page)) {
4706 wait_on_page_writeback(page);
4708 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4709 set_page_extent_mapped(page);
4711 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4713 unlock_extent_cached(io_tree, block_start, block_end,
4714 &cached_state, GFP_NOFS);
4717 btrfs_start_ordered_extent(inode, ordered, 1);
4718 btrfs_put_ordered_extent(ordered);
4722 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4723 EXTENT_DIRTY | EXTENT_DELALLOC |
4724 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4725 0, 0, &cached_state, GFP_NOFS);
4727 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4730 unlock_extent_cached(io_tree, block_start, block_end,
4731 &cached_state, GFP_NOFS);
4735 if (offset != blocksize) {
4737 len = blocksize - offset;
4740 memset(kaddr + (block_start - page_offset(page)),
4743 memset(kaddr + (block_start - page_offset(page)) + offset,
4745 flush_dcache_page(page);
4748 ClearPageChecked(page);
4749 set_page_dirty(page);
4750 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4755 btrfs_delalloc_release_space(inode, block_start,
4763 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4764 u64 offset, u64 len)
4766 struct btrfs_trans_handle *trans;
4770 * Still need to make sure the inode looks like it's been updated so
4771 * that any holes get logged if we fsync.
4773 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4774 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4775 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4776 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4781 * 1 - for the one we're dropping
4782 * 1 - for the one we're adding
4783 * 1 - for updating the inode.
4785 trans = btrfs_start_transaction(root, 3);
4787 return PTR_ERR(trans);
4789 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4791 btrfs_abort_transaction(trans, root, ret);
4792 btrfs_end_transaction(trans, root);
4796 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4797 0, 0, len, 0, len, 0, 0, 0);
4799 btrfs_abort_transaction(trans, root, ret);
4801 btrfs_update_inode(trans, root, inode);
4802 btrfs_end_transaction(trans, root);
4807 * This function puts in dummy file extents for the area we're creating a hole
4808 * for. So if we are truncating this file to a larger size we need to insert
4809 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4810 * the range between oldsize and size
4812 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4814 struct btrfs_root *root = BTRFS_I(inode)->root;
4815 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4816 struct extent_map *em = NULL;
4817 struct extent_state *cached_state = NULL;
4818 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4819 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4820 u64 block_end = ALIGN(size, root->sectorsize);
4827 * If our size started in the middle of a block we need to zero out the
4828 * rest of the block before we expand the i_size, otherwise we could
4829 * expose stale data.
4831 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4835 if (size <= hole_start)
4839 struct btrfs_ordered_extent *ordered;
4841 lock_extent_bits(io_tree, hole_start, block_end - 1,
4843 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4844 block_end - hole_start);
4847 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4848 &cached_state, GFP_NOFS);
4849 btrfs_start_ordered_extent(inode, ordered, 1);
4850 btrfs_put_ordered_extent(ordered);
4853 cur_offset = hole_start;
4855 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4856 block_end - cur_offset, 0);
4862 last_byte = min(extent_map_end(em), block_end);
4863 last_byte = ALIGN(last_byte , root->sectorsize);
4864 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4865 struct extent_map *hole_em;
4866 hole_size = last_byte - cur_offset;
4868 err = maybe_insert_hole(root, inode, cur_offset,
4872 btrfs_drop_extent_cache(inode, cur_offset,
4873 cur_offset + hole_size - 1, 0);
4874 hole_em = alloc_extent_map();
4876 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4877 &BTRFS_I(inode)->runtime_flags);
4880 hole_em->start = cur_offset;
4881 hole_em->len = hole_size;
4882 hole_em->orig_start = cur_offset;
4884 hole_em->block_start = EXTENT_MAP_HOLE;
4885 hole_em->block_len = 0;
4886 hole_em->orig_block_len = 0;
4887 hole_em->ram_bytes = hole_size;
4888 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4889 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4890 hole_em->generation = root->fs_info->generation;
4893 write_lock(&em_tree->lock);
4894 err = add_extent_mapping(em_tree, hole_em, 1);
4895 write_unlock(&em_tree->lock);
4898 btrfs_drop_extent_cache(inode, cur_offset,
4902 free_extent_map(hole_em);
4905 free_extent_map(em);
4907 cur_offset = last_byte;
4908 if (cur_offset >= block_end)
4911 free_extent_map(em);
4912 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4917 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4919 struct btrfs_root *root = BTRFS_I(inode)->root;
4920 struct btrfs_trans_handle *trans;
4921 loff_t oldsize = i_size_read(inode);
4922 loff_t newsize = attr->ia_size;
4923 int mask = attr->ia_valid;
4927 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4928 * special case where we need to update the times despite not having
4929 * these flags set. For all other operations the VFS set these flags
4930 * explicitly if it wants a timestamp update.
4932 if (newsize != oldsize) {
4933 inode_inc_iversion(inode);
4934 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4935 inode->i_ctime = inode->i_mtime =
4936 current_fs_time(inode->i_sb);
4939 if (newsize > oldsize) {
4941 * Don't do an expanding truncate while snapshoting is ongoing.
4942 * This is to ensure the snapshot captures a fully consistent
4943 * state of this file - if the snapshot captures this expanding
4944 * truncation, it must capture all writes that happened before
4947 btrfs_wait_for_snapshot_creation(root);
4948 ret = btrfs_cont_expand(inode, oldsize, newsize);
4950 btrfs_end_write_no_snapshoting(root);
4954 trans = btrfs_start_transaction(root, 1);
4955 if (IS_ERR(trans)) {
4956 btrfs_end_write_no_snapshoting(root);
4957 return PTR_ERR(trans);
4960 i_size_write(inode, newsize);
4961 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4962 pagecache_isize_extended(inode, oldsize, newsize);
4963 ret = btrfs_update_inode(trans, root, inode);
4964 btrfs_end_write_no_snapshoting(root);
4965 btrfs_end_transaction(trans, root);
4969 * We're truncating a file that used to have good data down to
4970 * zero. Make sure it gets into the ordered flush list so that
4971 * any new writes get down to disk quickly.
4974 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4975 &BTRFS_I(inode)->runtime_flags);
4978 * 1 for the orphan item we're going to add
4979 * 1 for the orphan item deletion.
4981 trans = btrfs_start_transaction(root, 2);
4983 return PTR_ERR(trans);
4986 * We need to do this in case we fail at _any_ point during the
4987 * actual truncate. Once we do the truncate_setsize we could
4988 * invalidate pages which forces any outstanding ordered io to
4989 * be instantly completed which will give us extents that need
4990 * to be truncated. If we fail to get an orphan inode down we
4991 * could have left over extents that were never meant to live,
4992 * so we need to guarantee from this point on that everything
4993 * will be consistent.
4995 ret = btrfs_orphan_add(trans, inode);
4996 btrfs_end_transaction(trans, root);
5000 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5001 truncate_setsize(inode, newsize);
5003 /* Disable nonlocked read DIO to avoid the end less truncate */
5004 btrfs_inode_block_unlocked_dio(inode);
5005 inode_dio_wait(inode);
5006 btrfs_inode_resume_unlocked_dio(inode);
5008 ret = btrfs_truncate(inode);
5009 if (ret && inode->i_nlink) {
5013 * failed to truncate, disk_i_size is only adjusted down
5014 * as we remove extents, so it should represent the true
5015 * size of the inode, so reset the in memory size and
5016 * delete our orphan entry.
5018 trans = btrfs_join_transaction(root);
5019 if (IS_ERR(trans)) {
5020 btrfs_orphan_del(NULL, inode);
5023 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5024 err = btrfs_orphan_del(trans, inode);
5026 btrfs_abort_transaction(trans, root, err);
5027 btrfs_end_transaction(trans, root);
5034 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5036 struct inode *inode = d_inode(dentry);
5037 struct btrfs_root *root = BTRFS_I(inode)->root;
5040 if (btrfs_root_readonly(root))
5043 err = inode_change_ok(inode, attr);
5047 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5048 err = btrfs_setsize(inode, attr);
5053 if (attr->ia_valid) {
5054 setattr_copy(inode, attr);
5055 inode_inc_iversion(inode);
5056 err = btrfs_dirty_inode(inode);
5058 if (!err && attr->ia_valid & ATTR_MODE)
5059 err = posix_acl_chmod(inode, inode->i_mode);
5066 * While truncating the inode pages during eviction, we get the VFS calling
5067 * btrfs_invalidatepage() against each page of the inode. This is slow because
5068 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5069 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5070 * extent_state structures over and over, wasting lots of time.
5072 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5073 * those expensive operations on a per page basis and do only the ordered io
5074 * finishing, while we release here the extent_map and extent_state structures,
5075 * without the excessive merging and splitting.
5077 static void evict_inode_truncate_pages(struct inode *inode)
5079 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5080 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5081 struct rb_node *node;
5083 ASSERT(inode->i_state & I_FREEING);
5084 truncate_inode_pages_final(&inode->i_data);
5086 write_lock(&map_tree->lock);
5087 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5088 struct extent_map *em;
5090 node = rb_first(&map_tree->map);
5091 em = rb_entry(node, struct extent_map, rb_node);
5092 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5093 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5094 remove_extent_mapping(map_tree, em);
5095 free_extent_map(em);
5096 if (need_resched()) {
5097 write_unlock(&map_tree->lock);
5099 write_lock(&map_tree->lock);
5102 write_unlock(&map_tree->lock);
5105 * Keep looping until we have no more ranges in the io tree.
5106 * We can have ongoing bios started by readpages (called from readahead)
5107 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5108 * still in progress (unlocked the pages in the bio but did not yet
5109 * unlocked the ranges in the io tree). Therefore this means some
5110 * ranges can still be locked and eviction started because before
5111 * submitting those bios, which are executed by a separate task (work
5112 * queue kthread), inode references (inode->i_count) were not taken
5113 * (which would be dropped in the end io callback of each bio).
5114 * Therefore here we effectively end up waiting for those bios and
5115 * anyone else holding locked ranges without having bumped the inode's
5116 * reference count - if we don't do it, when they access the inode's
5117 * io_tree to unlock a range it may be too late, leading to an
5118 * use-after-free issue.
5120 spin_lock(&io_tree->lock);
5121 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5122 struct extent_state *state;
5123 struct extent_state *cached_state = NULL;
5127 node = rb_first(&io_tree->state);
5128 state = rb_entry(node, struct extent_state, rb_node);
5129 start = state->start;
5131 spin_unlock(&io_tree->lock);
5133 lock_extent_bits(io_tree, start, end, &cached_state);
5136 * If still has DELALLOC flag, the extent didn't reach disk,
5137 * and its reserved space won't be freed by delayed_ref.
5138 * So we need to free its reserved space here.
5139 * (Refer to comment in btrfs_invalidatepage, case 2)
5141 * Note, end is the bytenr of last byte, so we need + 1 here.
5143 if (state->state & EXTENT_DELALLOC)
5144 btrfs_qgroup_free_data(inode, start, end - start + 1);
5146 clear_extent_bit(io_tree, start, end,
5147 EXTENT_LOCKED | EXTENT_DIRTY |
5148 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5149 EXTENT_DEFRAG, 1, 1,
5150 &cached_state, GFP_NOFS);
5153 spin_lock(&io_tree->lock);
5155 spin_unlock(&io_tree->lock);
5158 void btrfs_evict_inode(struct inode *inode)
5160 struct btrfs_trans_handle *trans;
5161 struct btrfs_root *root = BTRFS_I(inode)->root;
5162 struct btrfs_block_rsv *rsv, *global_rsv;
5163 int steal_from_global = 0;
5167 trace_btrfs_inode_evict(inode);
5170 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5174 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5176 evict_inode_truncate_pages(inode);
5178 if (inode->i_nlink &&
5179 ((btrfs_root_refs(&root->root_item) != 0 &&
5180 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5181 btrfs_is_free_space_inode(inode)))
5184 if (is_bad_inode(inode)) {
5185 btrfs_orphan_del(NULL, inode);
5188 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5189 if (!special_file(inode->i_mode))
5190 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5192 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5194 if (root->fs_info->log_root_recovering) {
5195 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5196 &BTRFS_I(inode)->runtime_flags));
5200 if (inode->i_nlink > 0) {
5201 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5202 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5206 ret = btrfs_commit_inode_delayed_inode(inode);
5208 btrfs_orphan_del(NULL, inode);
5212 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5214 btrfs_orphan_del(NULL, inode);
5217 rsv->size = min_size;
5219 global_rsv = &root->fs_info->global_block_rsv;
5221 btrfs_i_size_write(inode, 0);
5224 * This is a bit simpler than btrfs_truncate since we've already
5225 * reserved our space for our orphan item in the unlink, so we just
5226 * need to reserve some slack space in case we add bytes and update
5227 * inode item when doing the truncate.
5230 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5231 BTRFS_RESERVE_FLUSH_LIMIT);
5234 * Try and steal from the global reserve since we will
5235 * likely not use this space anyway, we want to try as
5236 * hard as possible to get this to work.
5239 steal_from_global++;
5241 steal_from_global = 0;
5245 * steal_from_global == 0: we reserved stuff, hooray!
5246 * steal_from_global == 1: we didn't reserve stuff, boo!
5247 * steal_from_global == 2: we've committed, still not a lot of
5248 * room but maybe we'll have room in the global reserve this
5250 * steal_from_global == 3: abandon all hope!
5252 if (steal_from_global > 2) {
5253 btrfs_warn(root->fs_info,
5254 "Could not get space for a delete, will truncate on mount %d",
5256 btrfs_orphan_del(NULL, inode);
5257 btrfs_free_block_rsv(root, rsv);
5261 trans = btrfs_join_transaction(root);
5262 if (IS_ERR(trans)) {
5263 btrfs_orphan_del(NULL, inode);
5264 btrfs_free_block_rsv(root, rsv);
5269 * We can't just steal from the global reserve, we need to make
5270 * sure there is room to do it, if not we need to commit and try
5273 if (steal_from_global) {
5274 if (!btrfs_check_space_for_delayed_refs(trans, root))
5275 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5282 * Couldn't steal from the global reserve, we have too much
5283 * pending stuff built up, commit the transaction and try it
5287 ret = btrfs_commit_transaction(trans, root);
5289 btrfs_orphan_del(NULL, inode);
5290 btrfs_free_block_rsv(root, rsv);
5295 steal_from_global = 0;
5298 trans->block_rsv = rsv;
5300 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5301 if (ret != -ENOSPC && ret != -EAGAIN)
5304 trans->block_rsv = &root->fs_info->trans_block_rsv;
5305 btrfs_end_transaction(trans, root);
5307 btrfs_btree_balance_dirty(root);
5310 btrfs_free_block_rsv(root, rsv);
5313 * Errors here aren't a big deal, it just means we leave orphan items
5314 * in the tree. They will be cleaned up on the next mount.
5317 trans->block_rsv = root->orphan_block_rsv;
5318 btrfs_orphan_del(trans, inode);
5320 btrfs_orphan_del(NULL, inode);
5323 trans->block_rsv = &root->fs_info->trans_block_rsv;
5324 if (!(root == root->fs_info->tree_root ||
5325 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5326 btrfs_return_ino(root, btrfs_ino(inode));
5328 btrfs_end_transaction(trans, root);
5329 btrfs_btree_balance_dirty(root);
5331 btrfs_remove_delayed_node(inode);
5336 * this returns the key found in the dir entry in the location pointer.
5337 * If no dir entries were found, location->objectid is 0.
5339 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5340 struct btrfs_key *location)
5342 const char *name = dentry->d_name.name;
5343 int namelen = dentry->d_name.len;
5344 struct btrfs_dir_item *di;
5345 struct btrfs_path *path;
5346 struct btrfs_root *root = BTRFS_I(dir)->root;
5349 path = btrfs_alloc_path();
5353 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5358 if (IS_ERR_OR_NULL(di))
5361 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5363 btrfs_free_path(path);
5366 location->objectid = 0;
5371 * when we hit a tree root in a directory, the btrfs part of the inode
5372 * needs to be changed to reflect the root directory of the tree root. This
5373 * is kind of like crossing a mount point.
5375 static int fixup_tree_root_location(struct btrfs_root *root,
5377 struct dentry *dentry,
5378 struct btrfs_key *location,
5379 struct btrfs_root **sub_root)
5381 struct btrfs_path *path;
5382 struct btrfs_root *new_root;
5383 struct btrfs_root_ref *ref;
5384 struct extent_buffer *leaf;
5385 struct btrfs_key key;
5389 path = btrfs_alloc_path();
5396 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5397 key.type = BTRFS_ROOT_REF_KEY;
5398 key.offset = location->objectid;
5400 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5408 leaf = path->nodes[0];
5409 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5410 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5411 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5414 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5415 (unsigned long)(ref + 1),
5416 dentry->d_name.len);
5420 btrfs_release_path(path);
5422 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5423 if (IS_ERR(new_root)) {
5424 err = PTR_ERR(new_root);
5428 *sub_root = new_root;
5429 location->objectid = btrfs_root_dirid(&new_root->root_item);
5430 location->type = BTRFS_INODE_ITEM_KEY;
5431 location->offset = 0;
5434 btrfs_free_path(path);
5438 static void inode_tree_add(struct inode *inode)
5440 struct btrfs_root *root = BTRFS_I(inode)->root;
5441 struct btrfs_inode *entry;
5443 struct rb_node *parent;
5444 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5445 u64 ino = btrfs_ino(inode);
5447 if (inode_unhashed(inode))
5450 spin_lock(&root->inode_lock);
5451 p = &root->inode_tree.rb_node;
5454 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5456 if (ino < btrfs_ino(&entry->vfs_inode))
5457 p = &parent->rb_left;
5458 else if (ino > btrfs_ino(&entry->vfs_inode))
5459 p = &parent->rb_right;
5461 WARN_ON(!(entry->vfs_inode.i_state &
5462 (I_WILL_FREE | I_FREEING)));
5463 rb_replace_node(parent, new, &root->inode_tree);
5464 RB_CLEAR_NODE(parent);
5465 spin_unlock(&root->inode_lock);
5469 rb_link_node(new, parent, p);
5470 rb_insert_color(new, &root->inode_tree);
5471 spin_unlock(&root->inode_lock);
5474 static void inode_tree_del(struct inode *inode)
5476 struct btrfs_root *root = BTRFS_I(inode)->root;
5479 spin_lock(&root->inode_lock);
5480 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5481 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5482 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5483 empty = RB_EMPTY_ROOT(&root->inode_tree);
5485 spin_unlock(&root->inode_lock);
5487 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5488 synchronize_srcu(&root->fs_info->subvol_srcu);
5489 spin_lock(&root->inode_lock);
5490 empty = RB_EMPTY_ROOT(&root->inode_tree);
5491 spin_unlock(&root->inode_lock);
5493 btrfs_add_dead_root(root);
5497 void btrfs_invalidate_inodes(struct btrfs_root *root)
5499 struct rb_node *node;
5500 struct rb_node *prev;
5501 struct btrfs_inode *entry;
5502 struct inode *inode;
5505 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5506 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5508 spin_lock(&root->inode_lock);
5510 node = root->inode_tree.rb_node;
5514 entry = rb_entry(node, struct btrfs_inode, rb_node);
5516 if (objectid < btrfs_ino(&entry->vfs_inode))
5517 node = node->rb_left;
5518 else if (objectid > btrfs_ino(&entry->vfs_inode))
5519 node = node->rb_right;
5525 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5526 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5530 prev = rb_next(prev);
5534 entry = rb_entry(node, struct btrfs_inode, rb_node);
5535 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5536 inode = igrab(&entry->vfs_inode);
5538 spin_unlock(&root->inode_lock);
5539 if (atomic_read(&inode->i_count) > 1)
5540 d_prune_aliases(inode);
5542 * btrfs_drop_inode will have it removed from
5543 * the inode cache when its usage count
5548 spin_lock(&root->inode_lock);
5552 if (cond_resched_lock(&root->inode_lock))
5555 node = rb_next(node);
5557 spin_unlock(&root->inode_lock);
5560 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5562 struct btrfs_iget_args *args = p;
5563 inode->i_ino = args->location->objectid;
5564 memcpy(&BTRFS_I(inode)->location, args->location,
5565 sizeof(*args->location));
5566 BTRFS_I(inode)->root = args->root;
5570 static int btrfs_find_actor(struct inode *inode, void *opaque)
5572 struct btrfs_iget_args *args = opaque;
5573 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5574 args->root == BTRFS_I(inode)->root;
5577 static struct inode *btrfs_iget_locked(struct super_block *s,
5578 struct btrfs_key *location,
5579 struct btrfs_root *root)
5581 struct inode *inode;
5582 struct btrfs_iget_args args;
5583 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5585 args.location = location;
5588 inode = iget5_locked(s, hashval, btrfs_find_actor,
5589 btrfs_init_locked_inode,
5594 /* Get an inode object given its location and corresponding root.
5595 * Returns in *is_new if the inode was read from disk
5597 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5598 struct btrfs_root *root, int *new)
5600 struct inode *inode;
5602 inode = btrfs_iget_locked(s, location, root);
5604 return ERR_PTR(-ENOMEM);
5606 if (inode->i_state & I_NEW) {
5607 btrfs_read_locked_inode(inode);
5608 if (!is_bad_inode(inode)) {
5609 inode_tree_add(inode);
5610 unlock_new_inode(inode);
5614 unlock_new_inode(inode);
5616 inode = ERR_PTR(-ESTALE);
5623 static struct inode *new_simple_dir(struct super_block *s,
5624 struct btrfs_key *key,
5625 struct btrfs_root *root)
5627 struct inode *inode = new_inode(s);
5630 return ERR_PTR(-ENOMEM);
5632 BTRFS_I(inode)->root = root;
5633 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5634 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5636 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5637 inode->i_op = &btrfs_dir_ro_inode_operations;
5638 inode->i_fop = &simple_dir_operations;
5639 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5640 inode->i_mtime = current_fs_time(inode->i_sb);
5641 inode->i_atime = inode->i_mtime;
5642 inode->i_ctime = inode->i_mtime;
5643 BTRFS_I(inode)->i_otime = inode->i_mtime;
5648 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5650 struct inode *inode;
5651 struct btrfs_root *root = BTRFS_I(dir)->root;
5652 struct btrfs_root *sub_root = root;
5653 struct btrfs_key location;
5657 if (dentry->d_name.len > BTRFS_NAME_LEN)
5658 return ERR_PTR(-ENAMETOOLONG);
5660 ret = btrfs_inode_by_name(dir, dentry, &location);
5662 return ERR_PTR(ret);
5664 if (location.objectid == 0)
5665 return ERR_PTR(-ENOENT);
5667 if (location.type == BTRFS_INODE_ITEM_KEY) {
5668 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5672 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5674 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5675 ret = fixup_tree_root_location(root, dir, dentry,
5676 &location, &sub_root);
5679 inode = ERR_PTR(ret);
5681 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5683 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5685 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5687 if (!IS_ERR(inode) && root != sub_root) {
5688 down_read(&root->fs_info->cleanup_work_sem);
5689 if (!(inode->i_sb->s_flags & MS_RDONLY))
5690 ret = btrfs_orphan_cleanup(sub_root);
5691 up_read(&root->fs_info->cleanup_work_sem);
5694 inode = ERR_PTR(ret);
5701 static int btrfs_dentry_delete(const struct dentry *dentry)
5703 struct btrfs_root *root;
5704 struct inode *inode = d_inode(dentry);
5706 if (!inode && !IS_ROOT(dentry))
5707 inode = d_inode(dentry->d_parent);
5710 root = BTRFS_I(inode)->root;
5711 if (btrfs_root_refs(&root->root_item) == 0)
5714 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5720 static void btrfs_dentry_release(struct dentry *dentry)
5722 kfree(dentry->d_fsdata);
5725 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5728 struct inode *inode;
5730 inode = btrfs_lookup_dentry(dir, dentry);
5731 if (IS_ERR(inode)) {
5732 if (PTR_ERR(inode) == -ENOENT)
5735 return ERR_CAST(inode);
5738 return d_splice_alias(inode, dentry);
5741 unsigned char btrfs_filetype_table[] = {
5742 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5745 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5747 struct inode *inode = file_inode(file);
5748 struct btrfs_root *root = BTRFS_I(inode)->root;
5749 struct btrfs_item *item;
5750 struct btrfs_dir_item *di;
5751 struct btrfs_key key;
5752 struct btrfs_key found_key;
5753 struct btrfs_path *path;
5754 struct list_head ins_list;
5755 struct list_head del_list;
5757 struct extent_buffer *leaf;
5759 unsigned char d_type;
5764 int key_type = BTRFS_DIR_INDEX_KEY;
5768 int is_curr = 0; /* ctx->pos points to the current index? */
5772 /* FIXME, use a real flag for deciding about the key type */
5773 if (root->fs_info->tree_root == root)
5774 key_type = BTRFS_DIR_ITEM_KEY;
5776 if (!dir_emit_dots(file, ctx))
5779 path = btrfs_alloc_path();
5783 path->reada = READA_FORWARD;
5785 if (key_type == BTRFS_DIR_INDEX_KEY) {
5786 INIT_LIST_HEAD(&ins_list);
5787 INIT_LIST_HEAD(&del_list);
5788 put = btrfs_readdir_get_delayed_items(inode, &ins_list,
5792 key.type = key_type;
5793 key.offset = ctx->pos;
5794 key.objectid = btrfs_ino(inode);
5796 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5802 leaf = path->nodes[0];
5803 slot = path->slots[0];
5804 if (slot >= btrfs_header_nritems(leaf)) {
5805 ret = btrfs_next_leaf(root, path);
5813 item = btrfs_item_nr(slot);
5814 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5816 if (found_key.objectid != key.objectid)
5818 if (found_key.type != key_type)
5820 if (found_key.offset < ctx->pos)
5822 if (key_type == BTRFS_DIR_INDEX_KEY &&
5823 btrfs_should_delete_dir_index(&del_list,
5827 ctx->pos = found_key.offset;
5830 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5832 di_total = btrfs_item_size(leaf, item);
5834 while (di_cur < di_total) {
5835 struct btrfs_key location;
5837 if (verify_dir_item(root, leaf, di))
5840 name_len = btrfs_dir_name_len(leaf, di);
5841 if (name_len <= sizeof(tmp_name)) {
5842 name_ptr = tmp_name;
5844 name_ptr = kmalloc(name_len, GFP_KERNEL);
5850 read_extent_buffer(leaf, name_ptr,
5851 (unsigned long)(di + 1), name_len);
5853 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5854 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5857 /* is this a reference to our own snapshot? If so
5860 * In contrast to old kernels, we insert the snapshot's
5861 * dir item and dir index after it has been created, so
5862 * we won't find a reference to our own snapshot. We
5863 * still keep the following code for backward
5866 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5867 location.objectid == root->root_key.objectid) {
5871 over = !dir_emit(ctx, name_ptr, name_len,
5872 location.objectid, d_type);
5875 if (name_ptr != tmp_name)
5881 di_len = btrfs_dir_name_len(leaf, di) +
5882 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5884 di = (struct btrfs_dir_item *)((char *)di + di_len);
5890 if (key_type == BTRFS_DIR_INDEX_KEY) {
5893 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5899 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5900 * it was was set to the termination value in previous call. We assume
5901 * that "." and ".." were emitted if we reach this point and set the
5902 * termination value as well for an empty directory.
5904 if (ctx->pos > 2 && !emitted)
5907 /* Reached end of directory/root. Bump pos past the last item. */
5911 * Stop new entries from being returned after we return the last
5914 * New directory entries are assigned a strictly increasing
5915 * offset. This means that new entries created during readdir
5916 * are *guaranteed* to be seen in the future by that readdir.
5917 * This has broken buggy programs which operate on names as
5918 * they're returned by readdir. Until we re-use freed offsets
5919 * we have this hack to stop new entries from being returned
5920 * under the assumption that they'll never reach this huge
5923 * This is being careful not to overflow 32bit loff_t unless the
5924 * last entry requires it because doing so has broken 32bit apps
5927 if (key_type == BTRFS_DIR_INDEX_KEY) {
5928 if (ctx->pos >= INT_MAX)
5929 ctx->pos = LLONG_MAX;
5937 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5938 btrfs_free_path(path);
5942 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5944 struct btrfs_root *root = BTRFS_I(inode)->root;
5945 struct btrfs_trans_handle *trans;
5947 bool nolock = false;
5949 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5952 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5955 if (wbc->sync_mode == WB_SYNC_ALL) {
5957 trans = btrfs_join_transaction_nolock(root);
5959 trans = btrfs_join_transaction(root);
5961 return PTR_ERR(trans);
5962 ret = btrfs_commit_transaction(trans, root);
5968 * This is somewhat expensive, updating the tree every time the
5969 * inode changes. But, it is most likely to find the inode in cache.
5970 * FIXME, needs more benchmarking...there are no reasons other than performance
5971 * to keep or drop this code.
5973 static int btrfs_dirty_inode(struct inode *inode)
5975 struct btrfs_root *root = BTRFS_I(inode)->root;
5976 struct btrfs_trans_handle *trans;
5979 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5982 trans = btrfs_join_transaction(root);
5984 return PTR_ERR(trans);
5986 ret = btrfs_update_inode(trans, root, inode);
5987 if (ret && ret == -ENOSPC) {
5988 /* whoops, lets try again with the full transaction */
5989 btrfs_end_transaction(trans, root);
5990 trans = btrfs_start_transaction(root, 1);
5992 return PTR_ERR(trans);
5994 ret = btrfs_update_inode(trans, root, inode);
5996 btrfs_end_transaction(trans, root);
5997 if (BTRFS_I(inode)->delayed_node)
5998 btrfs_balance_delayed_items(root);
6004 * This is a copy of file_update_time. We need this so we can return error on
6005 * ENOSPC for updating the inode in the case of file write and mmap writes.
6007 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6010 struct btrfs_root *root = BTRFS_I(inode)->root;
6012 if (btrfs_root_readonly(root))
6015 if (flags & S_VERSION)
6016 inode_inc_iversion(inode);
6017 if (flags & S_CTIME)
6018 inode->i_ctime = *now;
6019 if (flags & S_MTIME)
6020 inode->i_mtime = *now;
6021 if (flags & S_ATIME)
6022 inode->i_atime = *now;
6023 return btrfs_dirty_inode(inode);
6027 * find the highest existing sequence number in a directory
6028 * and then set the in-memory index_cnt variable to reflect
6029 * free sequence numbers
6031 static int btrfs_set_inode_index_count(struct inode *inode)
6033 struct btrfs_root *root = BTRFS_I(inode)->root;
6034 struct btrfs_key key, found_key;
6035 struct btrfs_path *path;
6036 struct extent_buffer *leaf;
6039 key.objectid = btrfs_ino(inode);
6040 key.type = BTRFS_DIR_INDEX_KEY;
6041 key.offset = (u64)-1;
6043 path = btrfs_alloc_path();
6047 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6050 /* FIXME: we should be able to handle this */
6056 * MAGIC NUMBER EXPLANATION:
6057 * since we search a directory based on f_pos we have to start at 2
6058 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6059 * else has to start at 2
6061 if (path->slots[0] == 0) {
6062 BTRFS_I(inode)->index_cnt = 2;
6068 leaf = path->nodes[0];
6069 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6071 if (found_key.objectid != btrfs_ino(inode) ||
6072 found_key.type != BTRFS_DIR_INDEX_KEY) {
6073 BTRFS_I(inode)->index_cnt = 2;
6077 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6079 btrfs_free_path(path);
6084 * helper to find a free sequence number in a given directory. This current
6085 * code is very simple, later versions will do smarter things in the btree
6087 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6091 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6092 ret = btrfs_inode_delayed_dir_index_count(dir);
6094 ret = btrfs_set_inode_index_count(dir);
6100 *index = BTRFS_I(dir)->index_cnt;
6101 BTRFS_I(dir)->index_cnt++;
6106 static int btrfs_insert_inode_locked(struct inode *inode)
6108 struct btrfs_iget_args args;
6109 args.location = &BTRFS_I(inode)->location;
6110 args.root = BTRFS_I(inode)->root;
6112 return insert_inode_locked4(inode,
6113 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6114 btrfs_find_actor, &args);
6117 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6118 struct btrfs_root *root,
6120 const char *name, int name_len,
6121 u64 ref_objectid, u64 objectid,
6122 umode_t mode, u64 *index)
6124 struct inode *inode;
6125 struct btrfs_inode_item *inode_item;
6126 struct btrfs_key *location;
6127 struct btrfs_path *path;
6128 struct btrfs_inode_ref *ref;
6129 struct btrfs_key key[2];
6131 int nitems = name ? 2 : 1;
6135 path = btrfs_alloc_path();
6137 return ERR_PTR(-ENOMEM);
6139 inode = new_inode(root->fs_info->sb);
6141 btrfs_free_path(path);
6142 return ERR_PTR(-ENOMEM);
6146 * O_TMPFILE, set link count to 0, so that after this point,
6147 * we fill in an inode item with the correct link count.
6150 set_nlink(inode, 0);
6153 * we have to initialize this early, so we can reclaim the inode
6154 * number if we fail afterwards in this function.
6156 inode->i_ino = objectid;
6159 trace_btrfs_inode_request(dir);
6161 ret = btrfs_set_inode_index(dir, index);
6163 btrfs_free_path(path);
6165 return ERR_PTR(ret);
6171 * index_cnt is ignored for everything but a dir,
6172 * btrfs_get_inode_index_count has an explanation for the magic
6175 BTRFS_I(inode)->index_cnt = 2;
6176 BTRFS_I(inode)->dir_index = *index;
6177 BTRFS_I(inode)->root = root;
6178 BTRFS_I(inode)->generation = trans->transid;
6179 inode->i_generation = BTRFS_I(inode)->generation;
6182 * We could have gotten an inode number from somebody who was fsynced
6183 * and then removed in this same transaction, so let's just set full
6184 * sync since it will be a full sync anyway and this will blow away the
6185 * old info in the log.
6187 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6189 key[0].objectid = objectid;
6190 key[0].type = BTRFS_INODE_ITEM_KEY;
6193 sizes[0] = sizeof(struct btrfs_inode_item);
6197 * Start new inodes with an inode_ref. This is slightly more
6198 * efficient for small numbers of hard links since they will
6199 * be packed into one item. Extended refs will kick in if we
6200 * add more hard links than can fit in the ref item.
6202 key[1].objectid = objectid;
6203 key[1].type = BTRFS_INODE_REF_KEY;
6204 key[1].offset = ref_objectid;
6206 sizes[1] = name_len + sizeof(*ref);
6209 location = &BTRFS_I(inode)->location;
6210 location->objectid = objectid;
6211 location->offset = 0;
6212 location->type = BTRFS_INODE_ITEM_KEY;
6214 ret = btrfs_insert_inode_locked(inode);
6218 path->leave_spinning = 1;
6219 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6223 inode_init_owner(inode, dir, mode);
6224 inode_set_bytes(inode, 0);
6226 inode->i_mtime = current_fs_time(inode->i_sb);
6227 inode->i_atime = inode->i_mtime;
6228 inode->i_ctime = inode->i_mtime;
6229 BTRFS_I(inode)->i_otime = inode->i_mtime;
6231 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6232 struct btrfs_inode_item);
6233 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6234 sizeof(*inode_item));
6235 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6238 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6239 struct btrfs_inode_ref);
6240 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6241 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6242 ptr = (unsigned long)(ref + 1);
6243 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6246 btrfs_mark_buffer_dirty(path->nodes[0]);
6247 btrfs_free_path(path);
6249 btrfs_inherit_iflags(inode, dir);
6251 if (S_ISREG(mode)) {
6252 if (btrfs_test_opt(root, NODATASUM))
6253 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6254 if (btrfs_test_opt(root, NODATACOW))
6255 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6256 BTRFS_INODE_NODATASUM;
6259 inode_tree_add(inode);
6261 trace_btrfs_inode_new(inode);
6262 btrfs_set_inode_last_trans(trans, inode);
6264 btrfs_update_root_times(trans, root);
6266 ret = btrfs_inode_inherit_props(trans, inode, dir);
6268 btrfs_err(root->fs_info,
6269 "error inheriting props for ino %llu (root %llu): %d",
6270 btrfs_ino(inode), root->root_key.objectid, ret);
6275 unlock_new_inode(inode);
6278 BTRFS_I(dir)->index_cnt--;
6279 btrfs_free_path(path);
6281 return ERR_PTR(ret);
6284 static inline u8 btrfs_inode_type(struct inode *inode)
6286 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6290 * utility function to add 'inode' into 'parent_inode' with
6291 * a give name and a given sequence number.
6292 * if 'add_backref' is true, also insert a backref from the
6293 * inode to the parent directory.
6295 int btrfs_add_link(struct btrfs_trans_handle *trans,
6296 struct inode *parent_inode, struct inode *inode,
6297 const char *name, int name_len, int add_backref, u64 index)
6300 struct btrfs_key key;
6301 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6302 u64 ino = btrfs_ino(inode);
6303 u64 parent_ino = btrfs_ino(parent_inode);
6305 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6306 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6309 key.type = BTRFS_INODE_ITEM_KEY;
6313 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6314 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6315 key.objectid, root->root_key.objectid,
6316 parent_ino, index, name, name_len);
6317 } else if (add_backref) {
6318 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6322 /* Nothing to clean up yet */
6326 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6328 btrfs_inode_type(inode), index);
6329 if (ret == -EEXIST || ret == -EOVERFLOW)
6332 btrfs_abort_transaction(trans, root, ret);
6336 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6338 inode_inc_iversion(parent_inode);
6339 parent_inode->i_mtime = parent_inode->i_ctime =
6340 current_fs_time(parent_inode->i_sb);
6341 ret = btrfs_update_inode(trans, root, parent_inode);
6343 btrfs_abort_transaction(trans, root, ret);
6347 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6350 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6351 key.objectid, root->root_key.objectid,
6352 parent_ino, &local_index, name, name_len);
6354 } else if (add_backref) {
6358 err = btrfs_del_inode_ref(trans, root, name, name_len,
6359 ino, parent_ino, &local_index);
6364 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6365 struct inode *dir, struct dentry *dentry,
6366 struct inode *inode, int backref, u64 index)
6368 int err = btrfs_add_link(trans, dir, inode,
6369 dentry->d_name.name, dentry->d_name.len,
6376 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6377 umode_t mode, dev_t rdev)
6379 struct btrfs_trans_handle *trans;
6380 struct btrfs_root *root = BTRFS_I(dir)->root;
6381 struct inode *inode = NULL;
6388 * 2 for inode item and ref
6390 * 1 for xattr if selinux is on
6392 trans = btrfs_start_transaction(root, 5);
6394 return PTR_ERR(trans);
6396 err = btrfs_find_free_ino(root, &objectid);
6400 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6401 dentry->d_name.len, btrfs_ino(dir), objectid,
6403 if (IS_ERR(inode)) {
6404 err = PTR_ERR(inode);
6409 * If the active LSM wants to access the inode during
6410 * d_instantiate it needs these. Smack checks to see
6411 * if the filesystem supports xattrs by looking at the
6414 inode->i_op = &btrfs_special_inode_operations;
6415 init_special_inode(inode, inode->i_mode, rdev);
6417 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6419 goto out_unlock_inode;
6421 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6423 goto out_unlock_inode;
6425 btrfs_update_inode(trans, root, inode);
6426 unlock_new_inode(inode);
6427 d_instantiate(dentry, inode);
6431 btrfs_end_transaction(trans, root);
6432 btrfs_balance_delayed_items(root);
6433 btrfs_btree_balance_dirty(root);
6435 inode_dec_link_count(inode);
6442 unlock_new_inode(inode);
6447 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6448 umode_t mode, bool excl)
6450 struct btrfs_trans_handle *trans;
6451 struct btrfs_root *root = BTRFS_I(dir)->root;
6452 struct inode *inode = NULL;
6453 int drop_inode_on_err = 0;
6459 * 2 for inode item and ref
6461 * 1 for xattr if selinux is on
6463 trans = btrfs_start_transaction(root, 5);
6465 return PTR_ERR(trans);
6467 err = btrfs_find_free_ino(root, &objectid);
6471 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6472 dentry->d_name.len, btrfs_ino(dir), objectid,
6474 if (IS_ERR(inode)) {
6475 err = PTR_ERR(inode);
6478 drop_inode_on_err = 1;
6480 * If the active LSM wants to access the inode during
6481 * d_instantiate it needs these. Smack checks to see
6482 * if the filesystem supports xattrs by looking at the
6485 inode->i_fop = &btrfs_file_operations;
6486 inode->i_op = &btrfs_file_inode_operations;
6487 inode->i_mapping->a_ops = &btrfs_aops;
6489 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6491 goto out_unlock_inode;
6493 err = btrfs_update_inode(trans, root, inode);
6495 goto out_unlock_inode;
6497 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6499 goto out_unlock_inode;
6501 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6502 unlock_new_inode(inode);
6503 d_instantiate(dentry, inode);
6506 btrfs_end_transaction(trans, root);
6507 if (err && drop_inode_on_err) {
6508 inode_dec_link_count(inode);
6511 btrfs_balance_delayed_items(root);
6512 btrfs_btree_balance_dirty(root);
6516 unlock_new_inode(inode);
6521 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6522 struct dentry *dentry)
6524 struct btrfs_trans_handle *trans = NULL;
6525 struct btrfs_root *root = BTRFS_I(dir)->root;
6526 struct inode *inode = d_inode(old_dentry);
6531 /* do not allow sys_link's with other subvols of the same device */
6532 if (root->objectid != BTRFS_I(inode)->root->objectid)
6535 if (inode->i_nlink >= BTRFS_LINK_MAX)
6538 err = btrfs_set_inode_index(dir, &index);
6543 * 2 items for inode and inode ref
6544 * 2 items for dir items
6545 * 1 item for parent inode
6547 trans = btrfs_start_transaction(root, 5);
6548 if (IS_ERR(trans)) {
6549 err = PTR_ERR(trans);
6554 /* There are several dir indexes for this inode, clear the cache. */
6555 BTRFS_I(inode)->dir_index = 0ULL;
6557 inode_inc_iversion(inode);
6558 inode->i_ctime = current_fs_time(inode->i_sb);
6560 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6562 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6567 struct dentry *parent = dentry->d_parent;
6568 err = btrfs_update_inode(trans, root, inode);
6571 if (inode->i_nlink == 1) {
6573 * If new hard link count is 1, it's a file created
6574 * with open(2) O_TMPFILE flag.
6576 err = btrfs_orphan_del(trans, inode);
6580 d_instantiate(dentry, inode);
6581 btrfs_log_new_name(trans, inode, NULL, parent);
6584 btrfs_balance_delayed_items(root);
6587 btrfs_end_transaction(trans, root);
6589 inode_dec_link_count(inode);
6592 btrfs_btree_balance_dirty(root);
6596 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6598 struct inode *inode = NULL;
6599 struct btrfs_trans_handle *trans;
6600 struct btrfs_root *root = BTRFS_I(dir)->root;
6602 int drop_on_err = 0;
6607 * 2 items for inode and ref
6608 * 2 items for dir items
6609 * 1 for xattr if selinux is on
6611 trans = btrfs_start_transaction(root, 5);
6613 return PTR_ERR(trans);
6615 err = btrfs_find_free_ino(root, &objectid);
6619 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6620 dentry->d_name.len, btrfs_ino(dir), objectid,
6621 S_IFDIR | mode, &index);
6622 if (IS_ERR(inode)) {
6623 err = PTR_ERR(inode);
6628 /* these must be set before we unlock the inode */
6629 inode->i_op = &btrfs_dir_inode_operations;
6630 inode->i_fop = &btrfs_dir_file_operations;
6632 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6634 goto out_fail_inode;
6636 btrfs_i_size_write(inode, 0);
6637 err = btrfs_update_inode(trans, root, inode);
6639 goto out_fail_inode;
6641 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6642 dentry->d_name.len, 0, index);
6644 goto out_fail_inode;
6646 d_instantiate(dentry, inode);
6648 * mkdir is special. We're unlocking after we call d_instantiate
6649 * to avoid a race with nfsd calling d_instantiate.
6651 unlock_new_inode(inode);
6655 btrfs_end_transaction(trans, root);
6657 inode_dec_link_count(inode);
6660 btrfs_balance_delayed_items(root);
6661 btrfs_btree_balance_dirty(root);
6665 unlock_new_inode(inode);
6669 /* Find next extent map of a given extent map, caller needs to ensure locks */
6670 static struct extent_map *next_extent_map(struct extent_map *em)
6672 struct rb_node *next;
6674 next = rb_next(&em->rb_node);
6677 return container_of(next, struct extent_map, rb_node);
6680 static struct extent_map *prev_extent_map(struct extent_map *em)
6682 struct rb_node *prev;
6684 prev = rb_prev(&em->rb_node);
6687 return container_of(prev, struct extent_map, rb_node);
6690 /* helper for btfs_get_extent. Given an existing extent in the tree,
6691 * the existing extent is the nearest extent to map_start,
6692 * and an extent that you want to insert, deal with overlap and insert
6693 * the best fitted new extent into the tree.
6695 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6696 struct extent_map *existing,
6697 struct extent_map *em,
6700 struct extent_map *prev;
6701 struct extent_map *next;
6706 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6708 if (existing->start > map_start) {
6710 prev = prev_extent_map(next);
6713 next = next_extent_map(prev);
6716 start = prev ? extent_map_end(prev) : em->start;
6717 start = max_t(u64, start, em->start);
6718 end = next ? next->start : extent_map_end(em);
6719 end = min_t(u64, end, extent_map_end(em));
6720 start_diff = start - em->start;
6722 em->len = end - start;
6723 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6724 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6725 em->block_start += start_diff;
6726 em->block_len -= start_diff;
6728 return add_extent_mapping(em_tree, em, 0);
6731 static noinline int uncompress_inline(struct btrfs_path *path,
6733 size_t pg_offset, u64 extent_offset,
6734 struct btrfs_file_extent_item *item)
6737 struct extent_buffer *leaf = path->nodes[0];
6740 unsigned long inline_size;
6744 WARN_ON(pg_offset != 0);
6745 compress_type = btrfs_file_extent_compression(leaf, item);
6746 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6747 inline_size = btrfs_file_extent_inline_item_len(leaf,
6748 btrfs_item_nr(path->slots[0]));
6749 tmp = kmalloc(inline_size, GFP_NOFS);
6752 ptr = btrfs_file_extent_inline_start(item);
6754 read_extent_buffer(leaf, tmp, ptr, inline_size);
6756 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6757 ret = btrfs_decompress(compress_type, tmp, page,
6758 extent_offset, inline_size, max_size);
6764 * a bit scary, this does extent mapping from logical file offset to the disk.
6765 * the ugly parts come from merging extents from the disk with the in-ram
6766 * representation. This gets more complex because of the data=ordered code,
6767 * where the in-ram extents might be locked pending data=ordered completion.
6769 * This also copies inline extents directly into the page.
6772 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6773 size_t pg_offset, u64 start, u64 len,
6778 u64 extent_start = 0;
6780 u64 objectid = btrfs_ino(inode);
6782 struct btrfs_path *path = NULL;
6783 struct btrfs_root *root = BTRFS_I(inode)->root;
6784 struct btrfs_file_extent_item *item;
6785 struct extent_buffer *leaf;
6786 struct btrfs_key found_key;
6787 struct extent_map *em = NULL;
6788 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6789 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6790 struct btrfs_trans_handle *trans = NULL;
6791 const bool new_inline = !page || create;
6794 read_lock(&em_tree->lock);
6795 em = lookup_extent_mapping(em_tree, start, len);
6797 em->bdev = root->fs_info->fs_devices->latest_bdev;
6798 read_unlock(&em_tree->lock);
6801 if (em->start > start || em->start + em->len <= start)
6802 free_extent_map(em);
6803 else if (em->block_start == EXTENT_MAP_INLINE && page)
6804 free_extent_map(em);
6808 em = alloc_extent_map();
6813 em->bdev = root->fs_info->fs_devices->latest_bdev;
6814 em->start = EXTENT_MAP_HOLE;
6815 em->orig_start = EXTENT_MAP_HOLE;
6817 em->block_len = (u64)-1;
6820 path = btrfs_alloc_path();
6826 * Chances are we'll be called again, so go ahead and do
6829 path->reada = READA_FORWARD;
6832 ret = btrfs_lookup_file_extent(trans, root, path,
6833 objectid, start, trans != NULL);
6840 if (path->slots[0] == 0)
6845 leaf = path->nodes[0];
6846 item = btrfs_item_ptr(leaf, path->slots[0],
6847 struct btrfs_file_extent_item);
6848 /* are we inside the extent that was found? */
6849 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6850 found_type = found_key.type;
6851 if (found_key.objectid != objectid ||
6852 found_type != BTRFS_EXTENT_DATA_KEY) {
6854 * If we backup past the first extent we want to move forward
6855 * and see if there is an extent in front of us, otherwise we'll
6856 * say there is a hole for our whole search range which can
6863 found_type = btrfs_file_extent_type(leaf, item);
6864 extent_start = found_key.offset;
6865 if (found_type == BTRFS_FILE_EXTENT_REG ||
6866 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6867 extent_end = extent_start +
6868 btrfs_file_extent_num_bytes(leaf, item);
6869 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6871 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6872 extent_end = ALIGN(extent_start + size, root->sectorsize);
6875 if (start >= extent_end) {
6877 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6878 ret = btrfs_next_leaf(root, path);
6885 leaf = path->nodes[0];
6887 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6888 if (found_key.objectid != objectid ||
6889 found_key.type != BTRFS_EXTENT_DATA_KEY)
6891 if (start + len <= found_key.offset)
6893 if (start > found_key.offset)
6896 em->orig_start = start;
6897 em->len = found_key.offset - start;
6901 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6903 if (found_type == BTRFS_FILE_EXTENT_REG ||
6904 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6906 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6910 size_t extent_offset;
6916 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6917 extent_offset = page_offset(page) + pg_offset - extent_start;
6918 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6919 size - extent_offset);
6920 em->start = extent_start + extent_offset;
6921 em->len = ALIGN(copy_size, root->sectorsize);
6922 em->orig_block_len = em->len;
6923 em->orig_start = em->start;
6924 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6925 if (create == 0 && !PageUptodate(page)) {
6926 if (btrfs_file_extent_compression(leaf, item) !=
6927 BTRFS_COMPRESS_NONE) {
6928 ret = uncompress_inline(path, page, pg_offset,
6929 extent_offset, item);
6936 read_extent_buffer(leaf, map + pg_offset, ptr,
6938 if (pg_offset + copy_size < PAGE_SIZE) {
6939 memset(map + pg_offset + copy_size, 0,
6940 PAGE_SIZE - pg_offset -
6945 flush_dcache_page(page);
6946 } else if (create && PageUptodate(page)) {
6950 free_extent_map(em);
6953 btrfs_release_path(path);
6954 trans = btrfs_join_transaction(root);
6957 return ERR_CAST(trans);
6961 write_extent_buffer(leaf, map + pg_offset, ptr,
6964 btrfs_mark_buffer_dirty(leaf);
6966 set_extent_uptodate(io_tree, em->start,
6967 extent_map_end(em) - 1, NULL, GFP_NOFS);
6972 em->orig_start = start;
6975 em->block_start = EXTENT_MAP_HOLE;
6976 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6978 btrfs_release_path(path);
6979 if (em->start > start || extent_map_end(em) <= start) {
6980 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6981 em->start, em->len, start, len);
6987 write_lock(&em_tree->lock);
6988 ret = add_extent_mapping(em_tree, em, 0);
6989 /* it is possible that someone inserted the extent into the tree
6990 * while we had the lock dropped. It is also possible that
6991 * an overlapping map exists in the tree
6993 if (ret == -EEXIST) {
6994 struct extent_map *existing;
6998 existing = search_extent_mapping(em_tree, start, len);
7000 * existing will always be non-NULL, since there must be
7001 * extent causing the -EEXIST.
7003 if (existing->start == em->start &&
7004 extent_map_end(existing) == extent_map_end(em) &&
7005 em->block_start == existing->block_start) {
7007 * these two extents are the same, it happens
7008 * with inlines especially
7010 free_extent_map(em);
7014 } else if (start >= extent_map_end(existing) ||
7015 start <= existing->start) {
7017 * The existing extent map is the one nearest to
7018 * the [start, start + len) range which overlaps
7020 err = merge_extent_mapping(em_tree, existing,
7022 free_extent_map(existing);
7024 free_extent_map(em);
7028 free_extent_map(em);
7033 write_unlock(&em_tree->lock);
7036 trace_btrfs_get_extent(root, em);
7038 btrfs_free_path(path);
7040 ret = btrfs_end_transaction(trans, root);
7045 free_extent_map(em);
7046 return ERR_PTR(err);
7048 BUG_ON(!em); /* Error is always set */
7052 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7053 size_t pg_offset, u64 start, u64 len,
7056 struct extent_map *em;
7057 struct extent_map *hole_em = NULL;
7058 u64 range_start = start;
7064 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7071 * - a pre-alloc extent,
7072 * there might actually be delalloc bytes behind it.
7074 if (em->block_start != EXTENT_MAP_HOLE &&
7075 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7081 /* check to see if we've wrapped (len == -1 or similar) */
7090 /* ok, we didn't find anything, lets look for delalloc */
7091 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7092 end, len, EXTENT_DELALLOC, 1);
7093 found_end = range_start + found;
7094 if (found_end < range_start)
7095 found_end = (u64)-1;
7098 * we didn't find anything useful, return
7099 * the original results from get_extent()
7101 if (range_start > end || found_end <= start) {
7107 /* adjust the range_start to make sure it doesn't
7108 * go backwards from the start they passed in
7110 range_start = max(start, range_start);
7111 found = found_end - range_start;
7114 u64 hole_start = start;
7117 em = alloc_extent_map();
7123 * when btrfs_get_extent can't find anything it
7124 * returns one huge hole
7126 * make sure what it found really fits our range, and
7127 * adjust to make sure it is based on the start from
7131 u64 calc_end = extent_map_end(hole_em);
7133 if (calc_end <= start || (hole_em->start > end)) {
7134 free_extent_map(hole_em);
7137 hole_start = max(hole_em->start, start);
7138 hole_len = calc_end - hole_start;
7142 if (hole_em && range_start > hole_start) {
7143 /* our hole starts before our delalloc, so we
7144 * have to return just the parts of the hole
7145 * that go until the delalloc starts
7147 em->len = min(hole_len,
7148 range_start - hole_start);
7149 em->start = hole_start;
7150 em->orig_start = hole_start;
7152 * don't adjust block start at all,
7153 * it is fixed at EXTENT_MAP_HOLE
7155 em->block_start = hole_em->block_start;
7156 em->block_len = hole_len;
7157 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7158 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7160 em->start = range_start;
7162 em->orig_start = range_start;
7163 em->block_start = EXTENT_MAP_DELALLOC;
7164 em->block_len = found;
7166 } else if (hole_em) {
7171 free_extent_map(hole_em);
7173 free_extent_map(em);
7174 return ERR_PTR(err);
7179 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7182 const u64 orig_start,
7183 const u64 block_start,
7184 const u64 block_len,
7185 const u64 orig_block_len,
7186 const u64 ram_bytes,
7189 struct extent_map *em = NULL;
7192 down_read(&BTRFS_I(inode)->dio_sem);
7193 if (type != BTRFS_ORDERED_NOCOW) {
7194 em = create_pinned_em(inode, start, len, orig_start,
7195 block_start, block_len, orig_block_len,
7200 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7201 len, block_len, type);
7204 free_extent_map(em);
7205 btrfs_drop_extent_cache(inode, start,
7206 start + len - 1, 0);
7211 up_read(&BTRFS_I(inode)->dio_sem);
7216 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7219 struct btrfs_root *root = BTRFS_I(inode)->root;
7220 struct extent_map *em;
7221 struct btrfs_key ins;
7225 alloc_hint = get_extent_allocation_hint(inode, start, len);
7226 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7227 alloc_hint, &ins, 1, 1);
7229 return ERR_PTR(ret);
7231 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7232 ins.objectid, ins.offset, ins.offset,
7234 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7236 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7242 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7243 * block must be cow'd
7245 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7246 u64 *orig_start, u64 *orig_block_len,
7249 struct btrfs_trans_handle *trans;
7250 struct btrfs_path *path;
7252 struct extent_buffer *leaf;
7253 struct btrfs_root *root = BTRFS_I(inode)->root;
7254 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7255 struct btrfs_file_extent_item *fi;
7256 struct btrfs_key key;
7263 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7265 path = btrfs_alloc_path();
7269 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7274 slot = path->slots[0];
7277 /* can't find the item, must cow */
7284 leaf = path->nodes[0];
7285 btrfs_item_key_to_cpu(leaf, &key, slot);
7286 if (key.objectid != btrfs_ino(inode) ||
7287 key.type != BTRFS_EXTENT_DATA_KEY) {
7288 /* not our file or wrong item type, must cow */
7292 if (key.offset > offset) {
7293 /* Wrong offset, must cow */
7297 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7298 found_type = btrfs_file_extent_type(leaf, fi);
7299 if (found_type != BTRFS_FILE_EXTENT_REG &&
7300 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7301 /* not a regular extent, must cow */
7305 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7308 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7309 if (extent_end <= offset)
7312 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7313 if (disk_bytenr == 0)
7316 if (btrfs_file_extent_compression(leaf, fi) ||
7317 btrfs_file_extent_encryption(leaf, fi) ||
7318 btrfs_file_extent_other_encoding(leaf, fi))
7321 backref_offset = btrfs_file_extent_offset(leaf, fi);
7324 *orig_start = key.offset - backref_offset;
7325 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7326 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7329 if (btrfs_extent_readonly(root, disk_bytenr))
7332 num_bytes = min(offset + *len, extent_end) - offset;
7333 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7336 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7337 ret = test_range_bit(io_tree, offset, range_end,
7338 EXTENT_DELALLOC, 0, NULL);
7345 btrfs_release_path(path);
7348 * look for other files referencing this extent, if we
7349 * find any we must cow
7351 trans = btrfs_join_transaction(root);
7352 if (IS_ERR(trans)) {
7357 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7358 key.offset - backref_offset, disk_bytenr);
7359 btrfs_end_transaction(trans, root);
7366 * adjust disk_bytenr and num_bytes to cover just the bytes
7367 * in this extent we are about to write. If there
7368 * are any csums in that range we have to cow in order
7369 * to keep the csums correct
7371 disk_bytenr += backref_offset;
7372 disk_bytenr += offset - key.offset;
7373 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7376 * all of the above have passed, it is safe to overwrite this extent
7382 btrfs_free_path(path);
7386 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7388 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7390 void **pagep = NULL;
7391 struct page *page = NULL;
7395 start_idx = start >> PAGE_SHIFT;
7398 * end is the last byte in the last page. end == start is legal
7400 end_idx = end >> PAGE_SHIFT;
7404 /* Most of the code in this while loop is lifted from
7405 * find_get_page. It's been modified to begin searching from a
7406 * page and return just the first page found in that range. If the
7407 * found idx is less than or equal to the end idx then we know that
7408 * a page exists. If no pages are found or if those pages are
7409 * outside of the range then we're fine (yay!) */
7410 while (page == NULL &&
7411 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7412 page = radix_tree_deref_slot(pagep);
7413 if (unlikely(!page))
7416 if (radix_tree_exception(page)) {
7417 if (radix_tree_deref_retry(page)) {
7422 * Otherwise, shmem/tmpfs must be storing a swap entry
7423 * here as an exceptional entry: so return it without
7424 * attempting to raise page count.
7427 break; /* TODO: Is this relevant for this use case? */
7430 if (!page_cache_get_speculative(page)) {
7436 * Has the page moved?
7437 * This is part of the lockless pagecache protocol. See
7438 * include/linux/pagemap.h for details.
7440 if (unlikely(page != *pagep)) {
7447 if (page->index <= end_idx)
7456 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7457 struct extent_state **cached_state, int writing)
7459 struct btrfs_ordered_extent *ordered;
7463 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7466 * We're concerned with the entire range that we're going to be
7467 * doing DIO to, so we need to make sure there's no ordered
7468 * extents in this range.
7470 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7471 lockend - lockstart + 1);
7474 * We need to make sure there are no buffered pages in this
7475 * range either, we could have raced between the invalidate in
7476 * generic_file_direct_write and locking the extent. The
7477 * invalidate needs to happen so that reads after a write do not
7482 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7485 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7486 cached_state, GFP_NOFS);
7490 * If we are doing a DIO read and the ordered extent we
7491 * found is for a buffered write, we can not wait for it
7492 * to complete and retry, because if we do so we can
7493 * deadlock with concurrent buffered writes on page
7494 * locks. This happens only if our DIO read covers more
7495 * than one extent map, if at this point has already
7496 * created an ordered extent for a previous extent map
7497 * and locked its range in the inode's io tree, and a
7498 * concurrent write against that previous extent map's
7499 * range and this range started (we unlock the ranges
7500 * in the io tree only when the bios complete and
7501 * buffered writes always lock pages before attempting
7502 * to lock range in the io tree).
7505 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7506 btrfs_start_ordered_extent(inode, ordered, 1);
7509 btrfs_put_ordered_extent(ordered);
7512 * We could trigger writeback for this range (and wait
7513 * for it to complete) and then invalidate the pages for
7514 * this range (through invalidate_inode_pages2_range()),
7515 * but that can lead us to a deadlock with a concurrent
7516 * call to readpages() (a buffered read or a defrag call
7517 * triggered a readahead) on a page lock due to an
7518 * ordered dio extent we created before but did not have
7519 * yet a corresponding bio submitted (whence it can not
7520 * complete), which makes readpages() wait for that
7521 * ordered extent to complete while holding a lock on
7536 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7537 u64 len, u64 orig_start,
7538 u64 block_start, u64 block_len,
7539 u64 orig_block_len, u64 ram_bytes,
7542 struct extent_map_tree *em_tree;
7543 struct extent_map *em;
7544 struct btrfs_root *root = BTRFS_I(inode)->root;
7547 em_tree = &BTRFS_I(inode)->extent_tree;
7548 em = alloc_extent_map();
7550 return ERR_PTR(-ENOMEM);
7553 em->orig_start = orig_start;
7554 em->mod_start = start;
7557 em->block_len = block_len;
7558 em->block_start = block_start;
7559 em->bdev = root->fs_info->fs_devices->latest_bdev;
7560 em->orig_block_len = orig_block_len;
7561 em->ram_bytes = ram_bytes;
7562 em->generation = -1;
7563 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7564 if (type == BTRFS_ORDERED_PREALLOC)
7565 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7568 btrfs_drop_extent_cache(inode, em->start,
7569 em->start + em->len - 1, 0);
7570 write_lock(&em_tree->lock);
7571 ret = add_extent_mapping(em_tree, em, 1);
7572 write_unlock(&em_tree->lock);
7573 } while (ret == -EEXIST);
7576 free_extent_map(em);
7577 return ERR_PTR(ret);
7583 static void adjust_dio_outstanding_extents(struct inode *inode,
7584 struct btrfs_dio_data *dio_data,
7587 unsigned num_extents;
7589 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7590 BTRFS_MAX_EXTENT_SIZE);
7592 * If we have an outstanding_extents count still set then we're
7593 * within our reservation, otherwise we need to adjust our inode
7594 * counter appropriately.
7596 if (dio_data->outstanding_extents) {
7597 dio_data->outstanding_extents -= num_extents;
7599 spin_lock(&BTRFS_I(inode)->lock);
7600 BTRFS_I(inode)->outstanding_extents += num_extents;
7601 spin_unlock(&BTRFS_I(inode)->lock);
7605 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7606 struct buffer_head *bh_result, int create)
7608 struct extent_map *em;
7609 struct btrfs_root *root = BTRFS_I(inode)->root;
7610 struct extent_state *cached_state = NULL;
7611 struct btrfs_dio_data *dio_data = NULL;
7612 u64 start = iblock << inode->i_blkbits;
7613 u64 lockstart, lockend;
7614 u64 len = bh_result->b_size;
7615 int unlock_bits = EXTENT_LOCKED;
7619 unlock_bits |= EXTENT_DIRTY;
7621 len = min_t(u64, len, root->sectorsize);
7624 lockend = start + len - 1;
7626 if (current->journal_info) {
7628 * Need to pull our outstanding extents and set journal_info to NULL so
7629 * that anything that needs to check if there's a transaction doesn't get
7632 dio_data = current->journal_info;
7633 current->journal_info = NULL;
7637 * If this errors out it's because we couldn't invalidate pagecache for
7638 * this range and we need to fallback to buffered.
7640 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7646 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7653 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7654 * io. INLINE is special, and we could probably kludge it in here, but
7655 * it's still buffered so for safety lets just fall back to the generic
7658 * For COMPRESSED we _have_ to read the entire extent in so we can
7659 * decompress it, so there will be buffering required no matter what we
7660 * do, so go ahead and fallback to buffered.
7662 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7663 * to buffered IO. Don't blame me, this is the price we pay for using
7666 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7667 em->block_start == EXTENT_MAP_INLINE) {
7668 free_extent_map(em);
7673 /* Just a good old fashioned hole, return */
7674 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7675 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7676 free_extent_map(em);
7681 * We don't allocate a new extent in the following cases
7683 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7685 * 2) The extent is marked as PREALLOC. We're good to go here and can
7686 * just use the extent.
7690 len = min(len, em->len - (start - em->start));
7691 lockstart = start + len;
7695 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7696 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7697 em->block_start != EXTENT_MAP_HOLE)) {
7699 u64 block_start, orig_start, orig_block_len, ram_bytes;
7701 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7702 type = BTRFS_ORDERED_PREALLOC;
7704 type = BTRFS_ORDERED_NOCOW;
7705 len = min(len, em->len - (start - em->start));
7706 block_start = em->block_start + (start - em->start);
7708 if (can_nocow_extent(inode, start, &len, &orig_start,
7709 &orig_block_len, &ram_bytes) == 1 &&
7710 btrfs_inc_nocow_writers(root->fs_info, block_start)) {
7711 struct extent_map *em2;
7713 em2 = btrfs_create_dio_extent(inode, start, len,
7714 orig_start, block_start,
7715 len, orig_block_len,
7717 btrfs_dec_nocow_writers(root->fs_info, block_start);
7718 if (type == BTRFS_ORDERED_PREALLOC) {
7719 free_extent_map(em);
7722 if (em2 && IS_ERR(em2)) {
7731 * this will cow the extent, reset the len in case we changed
7734 len = bh_result->b_size;
7735 free_extent_map(em);
7736 em = btrfs_new_extent_direct(inode, start, len);
7741 len = min(len, em->len - (start - em->start));
7743 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7745 bh_result->b_size = len;
7746 bh_result->b_bdev = em->bdev;
7747 set_buffer_mapped(bh_result);
7749 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7750 set_buffer_new(bh_result);
7753 * Need to update the i_size under the extent lock so buffered
7754 * readers will get the updated i_size when we unlock.
7756 if (start + len > i_size_read(inode))
7757 i_size_write(inode, start + len);
7759 adjust_dio_outstanding_extents(inode, dio_data, len);
7760 btrfs_free_reserved_data_space(inode, start, len);
7761 WARN_ON(dio_data->reserve < len);
7762 dio_data->reserve -= len;
7763 dio_data->unsubmitted_oe_range_end = start + len;
7764 current->journal_info = dio_data;
7768 * In the case of write we need to clear and unlock the entire range,
7769 * in the case of read we need to unlock only the end area that we
7770 * aren't using if there is any left over space.
7772 if (lockstart < lockend) {
7773 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7774 lockend, unlock_bits, 1, 0,
7775 &cached_state, GFP_NOFS);
7777 free_extent_state(cached_state);
7780 free_extent_map(em);
7785 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7786 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7789 current->journal_info = dio_data;
7791 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7792 * write less data then expected, so that we don't underflow our inode's
7793 * outstanding extents counter.
7795 if (create && dio_data)
7796 adjust_dio_outstanding_extents(inode, dio_data, len);
7801 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7802 int rw, int mirror_num)
7804 struct btrfs_root *root = BTRFS_I(inode)->root;
7807 BUG_ON(rw & REQ_WRITE);
7811 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7812 BTRFS_WQ_ENDIO_DIO_REPAIR);
7816 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7822 static int btrfs_check_dio_repairable(struct inode *inode,
7823 struct bio *failed_bio,
7824 struct io_failure_record *failrec,
7829 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7830 failrec->logical, failrec->len);
7831 if (num_copies == 1) {
7833 * we only have a single copy of the data, so don't bother with
7834 * all the retry and error correction code that follows. no
7835 * matter what the error is, it is very likely to persist.
7837 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7838 num_copies, failrec->this_mirror, failed_mirror);
7842 failrec->failed_mirror = failed_mirror;
7843 failrec->this_mirror++;
7844 if (failrec->this_mirror == failed_mirror)
7845 failrec->this_mirror++;
7847 if (failrec->this_mirror > num_copies) {
7848 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7849 num_copies, failrec->this_mirror, failed_mirror);
7856 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7857 struct page *page, unsigned int pgoff,
7858 u64 start, u64 end, int failed_mirror,
7859 bio_end_io_t *repair_endio, void *repair_arg)
7861 struct io_failure_record *failrec;
7867 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7869 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7873 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7876 free_io_failure(inode, failrec);
7880 if ((failed_bio->bi_vcnt > 1)
7881 || (failed_bio->bi_io_vec->bv_len
7882 > BTRFS_I(inode)->root->sectorsize))
7883 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7885 read_mode = READ_SYNC;
7887 isector = start - btrfs_io_bio(failed_bio)->logical;
7888 isector >>= inode->i_sb->s_blocksize_bits;
7889 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7890 pgoff, isector, repair_endio, repair_arg);
7892 free_io_failure(inode, failrec);
7896 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7897 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7898 read_mode, failrec->this_mirror, failrec->in_validation);
7900 ret = submit_dio_repair_bio(inode, bio, read_mode,
7901 failrec->this_mirror);
7903 free_io_failure(inode, failrec);
7910 struct btrfs_retry_complete {
7911 struct completion done;
7912 struct inode *inode;
7917 static void btrfs_retry_endio_nocsum(struct bio *bio)
7919 struct btrfs_retry_complete *done = bio->bi_private;
7920 struct inode *inode;
7921 struct bio_vec *bvec;
7927 ASSERT(bio->bi_vcnt == 1);
7928 inode = bio->bi_io_vec->bv_page->mapping->host;
7929 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7932 bio_for_each_segment_all(bvec, bio, i)
7933 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7935 complete(&done->done);
7939 static int __btrfs_correct_data_nocsum(struct inode *inode,
7940 struct btrfs_io_bio *io_bio)
7942 struct btrfs_fs_info *fs_info;
7943 struct bio_vec *bvec;
7944 struct btrfs_retry_complete done;
7952 fs_info = BTRFS_I(inode)->root->fs_info;
7953 sectorsize = BTRFS_I(inode)->root->sectorsize;
7955 start = io_bio->logical;
7958 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7959 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7960 pgoff = bvec->bv_offset;
7962 next_block_or_try_again:
7965 init_completion(&done.done);
7967 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7968 pgoff, start, start + sectorsize - 1,
7970 btrfs_retry_endio_nocsum, &done);
7974 wait_for_completion(&done.done);
7976 if (!done.uptodate) {
7977 /* We might have another mirror, so try again */
7978 goto next_block_or_try_again;
7981 start += sectorsize;
7984 pgoff += sectorsize;
7985 goto next_block_or_try_again;
7992 static void btrfs_retry_endio(struct bio *bio)
7994 struct btrfs_retry_complete *done = bio->bi_private;
7995 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7996 struct inode *inode;
7997 struct bio_vec *bvec;
8008 start = done->start;
8010 ASSERT(bio->bi_vcnt == 1);
8011 inode = bio->bi_io_vec->bv_page->mapping->host;
8012 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
8014 bio_for_each_segment_all(bvec, bio, i) {
8015 ret = __readpage_endio_check(done->inode, io_bio, i,
8016 bvec->bv_page, bvec->bv_offset,
8017 done->start, bvec->bv_len);
8019 clean_io_failure(done->inode, done->start,
8020 bvec->bv_page, bvec->bv_offset);
8025 done->uptodate = uptodate;
8027 complete(&done->done);
8031 static int __btrfs_subio_endio_read(struct inode *inode,
8032 struct btrfs_io_bio *io_bio, int err)
8034 struct btrfs_fs_info *fs_info;
8035 struct bio_vec *bvec;
8036 struct btrfs_retry_complete done;
8046 fs_info = BTRFS_I(inode)->root->fs_info;
8047 sectorsize = BTRFS_I(inode)->root->sectorsize;
8050 start = io_bio->logical;
8053 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8054 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8056 pgoff = bvec->bv_offset;
8058 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8059 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8060 bvec->bv_page, pgoff, start,
8067 init_completion(&done.done);
8069 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8070 pgoff, start, start + sectorsize - 1,
8072 btrfs_retry_endio, &done);
8078 wait_for_completion(&done.done);
8080 if (!done.uptodate) {
8081 /* We might have another mirror, so try again */
8085 offset += sectorsize;
8086 start += sectorsize;
8091 pgoff += sectorsize;
8099 static int btrfs_subio_endio_read(struct inode *inode,
8100 struct btrfs_io_bio *io_bio, int err)
8102 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8106 return __btrfs_correct_data_nocsum(inode, io_bio);
8110 return __btrfs_subio_endio_read(inode, io_bio, err);
8114 static void btrfs_endio_direct_read(struct bio *bio)
8116 struct btrfs_dio_private *dip = bio->bi_private;
8117 struct inode *inode = dip->inode;
8118 struct bio *dio_bio;
8119 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8120 int err = bio->bi_error;
8122 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8123 err = btrfs_subio_endio_read(inode, io_bio, err);
8125 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8126 dip->logical_offset + dip->bytes - 1);
8127 dio_bio = dip->dio_bio;
8131 dio_bio->bi_error = bio->bi_error;
8132 dio_end_io(dio_bio, bio->bi_error);
8135 io_bio->end_io(io_bio, err);
8139 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8144 struct btrfs_root *root = BTRFS_I(inode)->root;
8145 struct btrfs_ordered_extent *ordered = NULL;
8146 u64 ordered_offset = offset;
8147 u64 ordered_bytes = bytes;
8151 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8158 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8159 finish_ordered_fn, NULL, NULL);
8160 btrfs_queue_work(root->fs_info->endio_write_workers,
8164 * our bio might span multiple ordered extents. If we haven't
8165 * completed the accounting for the whole dio, go back and try again
8167 if (ordered_offset < offset + bytes) {
8168 ordered_bytes = offset + bytes - ordered_offset;
8174 static void btrfs_endio_direct_write(struct bio *bio)
8176 struct btrfs_dio_private *dip = bio->bi_private;
8177 struct bio *dio_bio = dip->dio_bio;
8179 btrfs_endio_direct_write_update_ordered(dip->inode,
8180 dip->logical_offset,
8186 dio_bio->bi_error = bio->bi_error;
8187 dio_end_io(dio_bio, bio->bi_error);
8191 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8192 struct bio *bio, int mirror_num,
8193 unsigned long bio_flags, u64 offset)
8196 struct btrfs_root *root = BTRFS_I(inode)->root;
8197 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8198 BUG_ON(ret); /* -ENOMEM */
8202 static void btrfs_end_dio_bio(struct bio *bio)
8204 struct btrfs_dio_private *dip = bio->bi_private;
8205 int err = bio->bi_error;
8208 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8209 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8210 btrfs_ino(dip->inode), bio->bi_rw,
8211 (unsigned long long)bio->bi_iter.bi_sector,
8212 bio->bi_iter.bi_size, err);
8214 if (dip->subio_endio)
8215 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8221 * before atomic variable goto zero, we must make sure
8222 * dip->errors is perceived to be set.
8224 smp_mb__before_atomic();
8227 /* if there are more bios still pending for this dio, just exit */
8228 if (!atomic_dec_and_test(&dip->pending_bios))
8232 bio_io_error(dip->orig_bio);
8234 dip->dio_bio->bi_error = 0;
8235 bio_endio(dip->orig_bio);
8241 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8242 u64 first_sector, gfp_t gfp_flags)
8245 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8247 bio_associate_current(bio);
8251 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8252 struct inode *inode,
8253 struct btrfs_dio_private *dip,
8257 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8258 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8262 * We load all the csum data we need when we submit
8263 * the first bio to reduce the csum tree search and
8266 if (dip->logical_offset == file_offset) {
8267 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8273 if (bio == dip->orig_bio)
8276 file_offset -= dip->logical_offset;
8277 file_offset >>= inode->i_sb->s_blocksize_bits;
8278 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8283 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8284 int rw, u64 file_offset, int skip_sum,
8287 struct btrfs_dio_private *dip = bio->bi_private;
8288 int write = rw & REQ_WRITE;
8289 struct btrfs_root *root = BTRFS_I(inode)->root;
8293 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8298 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8299 BTRFS_WQ_ENDIO_DATA);
8307 if (write && async_submit) {
8308 ret = btrfs_wq_submit_bio(root->fs_info,
8309 inode, rw, bio, 0, 0,
8311 __btrfs_submit_bio_start_direct_io,
8312 __btrfs_submit_bio_done);
8316 * If we aren't doing async submit, calculate the csum of the
8319 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8323 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8329 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8335 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8338 struct inode *inode = dip->inode;
8339 struct btrfs_root *root = BTRFS_I(inode)->root;
8341 struct bio *orig_bio = dip->orig_bio;
8342 struct bio_vec *bvec = orig_bio->bi_io_vec;
8343 u64 start_sector = orig_bio->bi_iter.bi_sector;
8344 u64 file_offset = dip->logical_offset;
8347 u32 blocksize = root->sectorsize;
8348 int async_submit = 0;
8353 map_length = orig_bio->bi_iter.bi_size;
8354 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8355 &map_length, NULL, 0);
8359 if (map_length >= orig_bio->bi_iter.bi_size) {
8361 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8365 /* async crcs make it difficult to collect full stripe writes. */
8366 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8371 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8375 bio->bi_private = dip;
8376 bio->bi_end_io = btrfs_end_dio_bio;
8377 btrfs_io_bio(bio)->logical = file_offset;
8378 atomic_inc(&dip->pending_bios);
8380 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8381 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8384 if (unlikely(map_length < submit_len + blocksize ||
8385 bio_add_page(bio, bvec->bv_page, blocksize,
8386 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8388 * inc the count before we submit the bio so
8389 * we know the end IO handler won't happen before
8390 * we inc the count. Otherwise, the dip might get freed
8391 * before we're done setting it up
8393 atomic_inc(&dip->pending_bios);
8394 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8395 file_offset, skip_sum,
8399 atomic_dec(&dip->pending_bios);
8403 start_sector += submit_len >> 9;
8404 file_offset += submit_len;
8408 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8409 start_sector, GFP_NOFS);
8412 bio->bi_private = dip;
8413 bio->bi_end_io = btrfs_end_dio_bio;
8414 btrfs_io_bio(bio)->logical = file_offset;
8416 map_length = orig_bio->bi_iter.bi_size;
8417 ret = btrfs_map_block(root->fs_info, rw,
8419 &map_length, NULL, 0);
8427 submit_len += blocksize;
8437 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8446 * before atomic variable goto zero, we must
8447 * make sure dip->errors is perceived to be set.
8449 smp_mb__before_atomic();
8450 if (atomic_dec_and_test(&dip->pending_bios))
8451 bio_io_error(dip->orig_bio);
8453 /* bio_end_io() will handle error, so we needn't return it */
8457 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8458 struct inode *inode, loff_t file_offset)
8460 struct btrfs_dio_private *dip = NULL;
8461 struct bio *io_bio = NULL;
8462 struct btrfs_io_bio *btrfs_bio;
8464 int write = rw & REQ_WRITE;
8467 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8469 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8475 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8481 dip->private = dio_bio->bi_private;
8483 dip->logical_offset = file_offset;
8484 dip->bytes = dio_bio->bi_iter.bi_size;
8485 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8486 io_bio->bi_private = dip;
8487 dip->orig_bio = io_bio;
8488 dip->dio_bio = dio_bio;
8489 atomic_set(&dip->pending_bios, 0);
8490 btrfs_bio = btrfs_io_bio(io_bio);
8491 btrfs_bio->logical = file_offset;
8494 io_bio->bi_end_io = btrfs_endio_direct_write;
8496 io_bio->bi_end_io = btrfs_endio_direct_read;
8497 dip->subio_endio = btrfs_subio_endio_read;
8501 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8502 * even if we fail to submit a bio, because in such case we do the
8503 * corresponding error handling below and it must not be done a second
8504 * time by btrfs_direct_IO().
8507 struct btrfs_dio_data *dio_data = current->journal_info;
8509 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8511 dio_data->unsubmitted_oe_range_start =
8512 dio_data->unsubmitted_oe_range_end;
8515 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8519 if (btrfs_bio->end_io)
8520 btrfs_bio->end_io(btrfs_bio, ret);
8524 * If we arrived here it means either we failed to submit the dip
8525 * or we either failed to clone the dio_bio or failed to allocate the
8526 * dip. If we cloned the dio_bio and allocated the dip, we can just
8527 * call bio_endio against our io_bio so that we get proper resource
8528 * cleanup if we fail to submit the dip, otherwise, we must do the
8529 * same as btrfs_endio_direct_[write|read] because we can't call these
8530 * callbacks - they require an allocated dip and a clone of dio_bio.
8532 if (io_bio && dip) {
8533 io_bio->bi_error = -EIO;
8536 * The end io callbacks free our dip, do the final put on io_bio
8537 * and all the cleanup and final put for dio_bio (through
8544 btrfs_endio_direct_write_update_ordered(inode,
8546 dio_bio->bi_iter.bi_size,
8549 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8550 file_offset + dio_bio->bi_iter.bi_size - 1);
8552 dio_bio->bi_error = -EIO;
8554 * Releases and cleans up our dio_bio, no need to bio_put()
8555 * nor bio_endio()/bio_io_error() against dio_bio.
8557 dio_end_io(dio_bio, ret);
8564 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8565 const struct iov_iter *iter, loff_t offset)
8569 unsigned blocksize_mask = root->sectorsize - 1;
8570 ssize_t retval = -EINVAL;
8572 if (offset & blocksize_mask)
8575 if (iov_iter_alignment(iter) & blocksize_mask)
8578 /* If this is a write we don't need to check anymore */
8579 if (iov_iter_rw(iter) == WRITE)
8582 * Check to make sure we don't have duplicate iov_base's in this
8583 * iovec, if so return EINVAL, otherwise we'll get csum errors
8584 * when reading back.
8586 for (seg = 0; seg < iter->nr_segs; seg++) {
8587 for (i = seg + 1; i < iter->nr_segs; i++) {
8588 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8597 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8599 struct file *file = iocb->ki_filp;
8600 struct inode *inode = file->f_mapping->host;
8601 struct btrfs_root *root = BTRFS_I(inode)->root;
8602 struct btrfs_dio_data dio_data = { 0 };
8603 loff_t offset = iocb->ki_pos;
8607 bool relock = false;
8610 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8613 inode_dio_begin(inode);
8614 smp_mb__after_atomic();
8617 * The generic stuff only does filemap_write_and_wait_range, which
8618 * isn't enough if we've written compressed pages to this area, so
8619 * we need to flush the dirty pages again to make absolutely sure
8620 * that any outstanding dirty pages are on disk.
8622 count = iov_iter_count(iter);
8623 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8624 &BTRFS_I(inode)->runtime_flags))
8625 filemap_fdatawrite_range(inode->i_mapping, offset,
8626 offset + count - 1);
8628 if (iov_iter_rw(iter) == WRITE) {
8630 * If the write DIO is beyond the EOF, we need update
8631 * the isize, but it is protected by i_mutex. So we can
8632 * not unlock the i_mutex at this case.
8634 if (offset + count <= inode->i_size) {
8635 inode_unlock(inode);
8638 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8641 dio_data.outstanding_extents = div64_u64(count +
8642 BTRFS_MAX_EXTENT_SIZE - 1,
8643 BTRFS_MAX_EXTENT_SIZE);
8646 * We need to know how many extents we reserved so that we can
8647 * do the accounting properly if we go over the number we
8648 * originally calculated. Abuse current->journal_info for this.
8650 dio_data.reserve = round_up(count, root->sectorsize);
8651 dio_data.unsubmitted_oe_range_start = (u64)offset;
8652 dio_data.unsubmitted_oe_range_end = (u64)offset;
8653 current->journal_info = &dio_data;
8654 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8655 &BTRFS_I(inode)->runtime_flags)) {
8656 inode_dio_end(inode);
8657 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8661 ret = __blockdev_direct_IO(iocb, inode,
8662 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8663 iter, btrfs_get_blocks_direct, NULL,
8664 btrfs_submit_direct, flags);
8665 if (iov_iter_rw(iter) == WRITE) {
8666 current->journal_info = NULL;
8667 if (ret < 0 && ret != -EIOCBQUEUED) {
8668 if (dio_data.reserve)
8669 btrfs_delalloc_release_space(inode, offset,
8672 * On error we might have left some ordered extents
8673 * without submitting corresponding bios for them, so
8674 * cleanup them up to avoid other tasks getting them
8675 * and waiting for them to complete forever.
8677 if (dio_data.unsubmitted_oe_range_start <
8678 dio_data.unsubmitted_oe_range_end)
8679 btrfs_endio_direct_write_update_ordered(inode,
8680 dio_data.unsubmitted_oe_range_start,
8681 dio_data.unsubmitted_oe_range_end -
8682 dio_data.unsubmitted_oe_range_start,
8684 } else if (ret >= 0 && (size_t)ret < count)
8685 btrfs_delalloc_release_space(inode, offset,
8686 count - (size_t)ret);
8690 inode_dio_end(inode);
8697 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8699 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8700 __u64 start, __u64 len)
8704 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8708 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8711 int btrfs_readpage(struct file *file, struct page *page)
8713 struct extent_io_tree *tree;
8714 tree = &BTRFS_I(page->mapping->host)->io_tree;
8715 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8718 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8720 struct extent_io_tree *tree;
8721 struct inode *inode = page->mapping->host;
8724 if (current->flags & PF_MEMALLOC) {
8725 redirty_page_for_writepage(wbc, page);
8731 * If we are under memory pressure we will call this directly from the
8732 * VM, we need to make sure we have the inode referenced for the ordered
8733 * extent. If not just return like we didn't do anything.
8735 if (!igrab(inode)) {
8736 redirty_page_for_writepage(wbc, page);
8737 return AOP_WRITEPAGE_ACTIVATE;
8739 tree = &BTRFS_I(page->mapping->host)->io_tree;
8740 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8741 btrfs_add_delayed_iput(inode);
8745 static int btrfs_writepages(struct address_space *mapping,
8746 struct writeback_control *wbc)
8748 struct extent_io_tree *tree;
8750 tree = &BTRFS_I(mapping->host)->io_tree;
8751 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8755 btrfs_readpages(struct file *file, struct address_space *mapping,
8756 struct list_head *pages, unsigned nr_pages)
8758 struct extent_io_tree *tree;
8759 tree = &BTRFS_I(mapping->host)->io_tree;
8760 return extent_readpages(tree, mapping, pages, nr_pages,
8763 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8765 struct extent_io_tree *tree;
8766 struct extent_map_tree *map;
8769 tree = &BTRFS_I(page->mapping->host)->io_tree;
8770 map = &BTRFS_I(page->mapping->host)->extent_tree;
8771 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8773 ClearPagePrivate(page);
8774 set_page_private(page, 0);
8780 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8782 if (PageWriteback(page) || PageDirty(page))
8784 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8787 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8788 unsigned int length)
8790 struct inode *inode = page->mapping->host;
8791 struct extent_io_tree *tree;
8792 struct btrfs_ordered_extent *ordered;
8793 struct extent_state *cached_state = NULL;
8794 u64 page_start = page_offset(page);
8795 u64 page_end = page_start + PAGE_SIZE - 1;
8798 int inode_evicting = inode->i_state & I_FREEING;
8801 * we have the page locked, so new writeback can't start,
8802 * and the dirty bit won't be cleared while we are here.
8804 * Wait for IO on this page so that we can safely clear
8805 * the PagePrivate2 bit and do ordered accounting
8807 wait_on_page_writeback(page);
8809 tree = &BTRFS_I(inode)->io_tree;
8811 btrfs_releasepage(page, GFP_NOFS);
8815 if (!inode_evicting)
8816 lock_extent_bits(tree, page_start, page_end, &cached_state);
8819 ordered = btrfs_lookup_ordered_range(inode, start,
8820 page_end - start + 1);
8822 end = min(page_end, ordered->file_offset + ordered->len - 1);
8824 * IO on this page will never be started, so we need
8825 * to account for any ordered extents now
8827 if (!inode_evicting)
8828 clear_extent_bit(tree, start, end,
8829 EXTENT_DIRTY | EXTENT_DELALLOC |
8830 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8831 EXTENT_DEFRAG, 1, 0, &cached_state,
8834 * whoever cleared the private bit is responsible
8835 * for the finish_ordered_io
8837 if (TestClearPagePrivate2(page)) {
8838 struct btrfs_ordered_inode_tree *tree;
8841 tree = &BTRFS_I(inode)->ordered_tree;
8843 spin_lock_irq(&tree->lock);
8844 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8845 new_len = start - ordered->file_offset;
8846 if (new_len < ordered->truncated_len)
8847 ordered->truncated_len = new_len;
8848 spin_unlock_irq(&tree->lock);
8850 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8852 end - start + 1, 1))
8853 btrfs_finish_ordered_io(ordered);
8855 btrfs_put_ordered_extent(ordered);
8856 if (!inode_evicting) {
8857 cached_state = NULL;
8858 lock_extent_bits(tree, start, end,
8863 if (start < page_end)
8868 * Qgroup reserved space handler
8869 * Page here will be either
8870 * 1) Already written to disk
8871 * In this case, its reserved space is released from data rsv map
8872 * and will be freed by delayed_ref handler finally.
8873 * So even we call qgroup_free_data(), it won't decrease reserved
8875 * 2) Not written to disk
8876 * This means the reserved space should be freed here.
8878 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8879 if (!inode_evicting) {
8880 clear_extent_bit(tree, page_start, page_end,
8881 EXTENT_LOCKED | EXTENT_DIRTY |
8882 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8883 EXTENT_DEFRAG, 1, 1,
8884 &cached_state, GFP_NOFS);
8886 __btrfs_releasepage(page, GFP_NOFS);
8889 ClearPageChecked(page);
8890 if (PagePrivate(page)) {
8891 ClearPagePrivate(page);
8892 set_page_private(page, 0);
8898 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8899 * called from a page fault handler when a page is first dirtied. Hence we must
8900 * be careful to check for EOF conditions here. We set the page up correctly
8901 * for a written page which means we get ENOSPC checking when writing into
8902 * holes and correct delalloc and unwritten extent mapping on filesystems that
8903 * support these features.
8905 * We are not allowed to take the i_mutex here so we have to play games to
8906 * protect against truncate races as the page could now be beyond EOF. Because
8907 * vmtruncate() writes the inode size before removing pages, once we have the
8908 * page lock we can determine safely if the page is beyond EOF. If it is not
8909 * beyond EOF, then the page is guaranteed safe against truncation until we
8912 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8914 struct page *page = vmf->page;
8915 struct inode *inode = file_inode(vma->vm_file);
8916 struct btrfs_root *root = BTRFS_I(inode)->root;
8917 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8918 struct btrfs_ordered_extent *ordered;
8919 struct extent_state *cached_state = NULL;
8921 unsigned long zero_start;
8930 reserved_space = PAGE_SIZE;
8932 sb_start_pagefault(inode->i_sb);
8933 page_start = page_offset(page);
8934 page_end = page_start + PAGE_SIZE - 1;
8938 * Reserving delalloc space after obtaining the page lock can lead to
8939 * deadlock. For example, if a dirty page is locked by this function
8940 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8941 * dirty page write out, then the btrfs_writepage() function could
8942 * end up waiting indefinitely to get a lock on the page currently
8943 * being processed by btrfs_page_mkwrite() function.
8945 ret = btrfs_delalloc_reserve_space(inode, page_start,
8948 ret = file_update_time(vma->vm_file);
8954 else /* -ENOSPC, -EIO, etc */
8955 ret = VM_FAULT_SIGBUS;
8961 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8964 size = i_size_read(inode);
8966 if ((page->mapping != inode->i_mapping) ||
8967 (page_start >= size)) {
8968 /* page got truncated out from underneath us */
8971 wait_on_page_writeback(page);
8973 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8974 set_page_extent_mapped(page);
8977 * we can't set the delalloc bits if there are pending ordered
8978 * extents. Drop our locks and wait for them to finish
8980 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8982 unlock_extent_cached(io_tree, page_start, page_end,
8983 &cached_state, GFP_NOFS);
8985 btrfs_start_ordered_extent(inode, ordered, 1);
8986 btrfs_put_ordered_extent(ordered);
8990 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8991 reserved_space = round_up(size - page_start, root->sectorsize);
8992 if (reserved_space < PAGE_SIZE) {
8993 end = page_start + reserved_space - 1;
8994 spin_lock(&BTRFS_I(inode)->lock);
8995 BTRFS_I(inode)->outstanding_extents++;
8996 spin_unlock(&BTRFS_I(inode)->lock);
8997 btrfs_delalloc_release_space(inode, page_start,
8998 PAGE_SIZE - reserved_space);
9003 * XXX - page_mkwrite gets called every time the page is dirtied, even
9004 * if it was already dirty, so for space accounting reasons we need to
9005 * clear any delalloc bits for the range we are fixing to save. There
9006 * is probably a better way to do this, but for now keep consistent with
9007 * prepare_pages in the normal write path.
9009 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9010 EXTENT_DIRTY | EXTENT_DELALLOC |
9011 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9012 0, 0, &cached_state, GFP_NOFS);
9014 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9017 unlock_extent_cached(io_tree, page_start, page_end,
9018 &cached_state, GFP_NOFS);
9019 ret = VM_FAULT_SIGBUS;
9024 /* page is wholly or partially inside EOF */
9025 if (page_start + PAGE_SIZE > size)
9026 zero_start = size & ~PAGE_MASK;
9028 zero_start = PAGE_SIZE;
9030 if (zero_start != PAGE_SIZE) {
9032 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9033 flush_dcache_page(page);
9036 ClearPageChecked(page);
9037 set_page_dirty(page);
9038 SetPageUptodate(page);
9040 BTRFS_I(inode)->last_trans = root->fs_info->generation;
9041 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9042 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9044 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9048 sb_end_pagefault(inode->i_sb);
9049 return VM_FAULT_LOCKED;
9053 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9055 sb_end_pagefault(inode->i_sb);
9059 static int btrfs_truncate(struct inode *inode)
9061 struct btrfs_root *root = BTRFS_I(inode)->root;
9062 struct btrfs_block_rsv *rsv;
9065 struct btrfs_trans_handle *trans;
9066 u64 mask = root->sectorsize - 1;
9067 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9069 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9075 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9076 * 3 things going on here
9078 * 1) We need to reserve space for our orphan item and the space to
9079 * delete our orphan item. Lord knows we don't want to have a dangling
9080 * orphan item because we didn't reserve space to remove it.
9082 * 2) We need to reserve space to update our inode.
9084 * 3) We need to have something to cache all the space that is going to
9085 * be free'd up by the truncate operation, but also have some slack
9086 * space reserved in case it uses space during the truncate (thank you
9087 * very much snapshotting).
9089 * And we need these to all be separate. The fact is we can use a lot of
9090 * space doing the truncate, and we have no earthly idea how much space
9091 * we will use, so we need the truncate reservation to be separate so it
9092 * doesn't end up using space reserved for updating the inode or
9093 * removing the orphan item. We also need to be able to stop the
9094 * transaction and start a new one, which means we need to be able to
9095 * update the inode several times, and we have no idea of knowing how
9096 * many times that will be, so we can't just reserve 1 item for the
9097 * entirety of the operation, so that has to be done separately as well.
9098 * Then there is the orphan item, which does indeed need to be held on
9099 * to for the whole operation, and we need nobody to touch this reserved
9100 * space except the orphan code.
9102 * So that leaves us with
9104 * 1) root->orphan_block_rsv - for the orphan deletion.
9105 * 2) rsv - for the truncate reservation, which we will steal from the
9106 * transaction reservation.
9107 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9108 * updating the inode.
9110 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9113 rsv->size = min_size;
9117 * 1 for the truncate slack space
9118 * 1 for updating the inode.
9120 trans = btrfs_start_transaction(root, 2);
9121 if (IS_ERR(trans)) {
9122 err = PTR_ERR(trans);
9126 /* Migrate the slack space for the truncate to our reserve */
9127 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9132 * So if we truncate and then write and fsync we normally would just
9133 * write the extents that changed, which is a problem if we need to
9134 * first truncate that entire inode. So set this flag so we write out
9135 * all of the extents in the inode to the sync log so we're completely
9138 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9139 trans->block_rsv = rsv;
9142 ret = btrfs_truncate_inode_items(trans, root, inode,
9144 BTRFS_EXTENT_DATA_KEY);
9145 if (ret != -ENOSPC && ret != -EAGAIN) {
9150 trans->block_rsv = &root->fs_info->trans_block_rsv;
9151 ret = btrfs_update_inode(trans, root, inode);
9157 btrfs_end_transaction(trans, root);
9158 btrfs_btree_balance_dirty(root);
9160 trans = btrfs_start_transaction(root, 2);
9161 if (IS_ERR(trans)) {
9162 ret = err = PTR_ERR(trans);
9167 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9169 BUG_ON(ret); /* shouldn't happen */
9170 trans->block_rsv = rsv;
9173 if (ret == 0 && inode->i_nlink > 0) {
9174 trans->block_rsv = root->orphan_block_rsv;
9175 ret = btrfs_orphan_del(trans, inode);
9181 trans->block_rsv = &root->fs_info->trans_block_rsv;
9182 ret = btrfs_update_inode(trans, root, inode);
9186 ret = btrfs_end_transaction(trans, root);
9187 btrfs_btree_balance_dirty(root);
9190 btrfs_free_block_rsv(root, rsv);
9199 * create a new subvolume directory/inode (helper for the ioctl).
9201 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9202 struct btrfs_root *new_root,
9203 struct btrfs_root *parent_root,
9206 struct inode *inode;
9210 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9211 new_dirid, new_dirid,
9212 S_IFDIR | (~current_umask() & S_IRWXUGO),
9215 return PTR_ERR(inode);
9216 inode->i_op = &btrfs_dir_inode_operations;
9217 inode->i_fop = &btrfs_dir_file_operations;
9219 set_nlink(inode, 1);
9220 btrfs_i_size_write(inode, 0);
9221 unlock_new_inode(inode);
9223 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9225 btrfs_err(new_root->fs_info,
9226 "error inheriting subvolume %llu properties: %d",
9227 new_root->root_key.objectid, err);
9229 err = btrfs_update_inode(trans, new_root, inode);
9235 struct inode *btrfs_alloc_inode(struct super_block *sb)
9237 struct btrfs_inode *ei;
9238 struct inode *inode;
9240 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9247 ei->last_sub_trans = 0;
9248 ei->logged_trans = 0;
9249 ei->delalloc_bytes = 0;
9250 ei->defrag_bytes = 0;
9251 ei->disk_i_size = 0;
9254 ei->index_cnt = (u64)-1;
9256 ei->last_unlink_trans = 0;
9257 ei->last_log_commit = 0;
9258 ei->delayed_iput_count = 0;
9260 spin_lock_init(&ei->lock);
9261 ei->outstanding_extents = 0;
9262 ei->reserved_extents = 0;
9264 ei->runtime_flags = 0;
9265 ei->force_compress = BTRFS_COMPRESS_NONE;
9267 ei->delayed_node = NULL;
9269 ei->i_otime.tv_sec = 0;
9270 ei->i_otime.tv_nsec = 0;
9272 inode = &ei->vfs_inode;
9273 extent_map_tree_init(&ei->extent_tree);
9274 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9275 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9276 ei->io_tree.track_uptodate = 1;
9277 ei->io_failure_tree.track_uptodate = 1;
9278 atomic_set(&ei->sync_writers, 0);
9279 mutex_init(&ei->log_mutex);
9280 mutex_init(&ei->delalloc_mutex);
9281 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9282 INIT_LIST_HEAD(&ei->delalloc_inodes);
9283 INIT_LIST_HEAD(&ei->delayed_iput);
9284 RB_CLEAR_NODE(&ei->rb_node);
9285 init_rwsem(&ei->dio_sem);
9290 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9291 void btrfs_test_destroy_inode(struct inode *inode)
9293 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9294 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9298 static void btrfs_i_callback(struct rcu_head *head)
9300 struct inode *inode = container_of(head, struct inode, i_rcu);
9301 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9304 void btrfs_destroy_inode(struct inode *inode)
9306 struct btrfs_ordered_extent *ordered;
9307 struct btrfs_root *root = BTRFS_I(inode)->root;
9309 WARN_ON(!hlist_empty(&inode->i_dentry));
9310 WARN_ON(inode->i_data.nrpages);
9311 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9312 WARN_ON(BTRFS_I(inode)->reserved_extents);
9313 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9314 WARN_ON(BTRFS_I(inode)->csum_bytes);
9315 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9318 * This can happen where we create an inode, but somebody else also
9319 * created the same inode and we need to destroy the one we already
9325 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9326 &BTRFS_I(inode)->runtime_flags)) {
9327 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9329 atomic_dec(&root->orphan_inodes);
9333 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9337 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9338 ordered->file_offset, ordered->len);
9339 btrfs_remove_ordered_extent(inode, ordered);
9340 btrfs_put_ordered_extent(ordered);
9341 btrfs_put_ordered_extent(ordered);
9344 btrfs_qgroup_check_reserved_leak(inode);
9345 inode_tree_del(inode);
9346 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9348 call_rcu(&inode->i_rcu, btrfs_i_callback);
9351 int btrfs_drop_inode(struct inode *inode)
9353 struct btrfs_root *root = BTRFS_I(inode)->root;
9358 /* the snap/subvol tree is on deleting */
9359 if (btrfs_root_refs(&root->root_item) == 0)
9362 return generic_drop_inode(inode);
9365 static void init_once(void *foo)
9367 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9369 inode_init_once(&ei->vfs_inode);
9372 void btrfs_destroy_cachep(void)
9375 * Make sure all delayed rcu free inodes are flushed before we
9379 kmem_cache_destroy(btrfs_inode_cachep);
9380 kmem_cache_destroy(btrfs_trans_handle_cachep);
9381 kmem_cache_destroy(btrfs_transaction_cachep);
9382 kmem_cache_destroy(btrfs_path_cachep);
9383 kmem_cache_destroy(btrfs_free_space_cachep);
9386 int btrfs_init_cachep(void)
9388 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9389 sizeof(struct btrfs_inode), 0,
9390 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9392 if (!btrfs_inode_cachep)
9395 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9396 sizeof(struct btrfs_trans_handle), 0,
9397 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9398 if (!btrfs_trans_handle_cachep)
9401 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9402 sizeof(struct btrfs_transaction), 0,
9403 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9404 if (!btrfs_transaction_cachep)
9407 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9408 sizeof(struct btrfs_path), 0,
9409 SLAB_MEM_SPREAD, NULL);
9410 if (!btrfs_path_cachep)
9413 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9414 sizeof(struct btrfs_free_space), 0,
9415 SLAB_MEM_SPREAD, NULL);
9416 if (!btrfs_free_space_cachep)
9421 btrfs_destroy_cachep();
9425 static int btrfs_getattr(struct vfsmount *mnt,
9426 struct dentry *dentry, struct kstat *stat)
9429 struct inode *inode = d_inode(dentry);
9430 u32 blocksize = inode->i_sb->s_blocksize;
9432 generic_fillattr(inode, stat);
9433 stat->dev = BTRFS_I(inode)->root->anon_dev;
9435 spin_lock(&BTRFS_I(inode)->lock);
9436 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9437 spin_unlock(&BTRFS_I(inode)->lock);
9438 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9439 ALIGN(delalloc_bytes, blocksize)) >> 9;
9443 static int btrfs_rename_exchange(struct inode *old_dir,
9444 struct dentry *old_dentry,
9445 struct inode *new_dir,
9446 struct dentry *new_dentry)
9448 struct btrfs_trans_handle *trans;
9449 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9450 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9451 struct inode *new_inode = new_dentry->d_inode;
9452 struct inode *old_inode = old_dentry->d_inode;
9453 struct timespec ctime = CURRENT_TIME;
9454 struct dentry *parent;
9455 u64 old_ino = btrfs_ino(old_inode);
9456 u64 new_ino = btrfs_ino(new_inode);
9461 bool root_log_pinned = false;
9462 bool dest_log_pinned = false;
9464 /* we only allow rename subvolume link between subvolumes */
9465 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9468 /* close the race window with snapshot create/destroy ioctl */
9469 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9470 down_read(&root->fs_info->subvol_sem);
9471 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9472 down_read(&dest->fs_info->subvol_sem);
9475 * We want to reserve the absolute worst case amount of items. So if
9476 * both inodes are subvols and we need to unlink them then that would
9477 * require 4 item modifications, but if they are both normal inodes it
9478 * would require 5 item modifications, so we'll assume their normal
9479 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9480 * should cover the worst case number of items we'll modify.
9482 trans = btrfs_start_transaction(root, 12);
9483 if (IS_ERR(trans)) {
9484 ret = PTR_ERR(trans);
9489 * We need to find a free sequence number both in the source and
9490 * in the destination directory for the exchange.
9492 ret = btrfs_set_inode_index(new_dir, &old_idx);
9495 ret = btrfs_set_inode_index(old_dir, &new_idx);
9499 BTRFS_I(old_inode)->dir_index = 0ULL;
9500 BTRFS_I(new_inode)->dir_index = 0ULL;
9502 /* Reference for the source. */
9503 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9504 /* force full log commit if subvolume involved. */
9505 btrfs_set_log_full_commit(root->fs_info, trans);
9507 btrfs_pin_log_trans(root);
9508 root_log_pinned = true;
9509 ret = btrfs_insert_inode_ref(trans, dest,
9510 new_dentry->d_name.name,
9511 new_dentry->d_name.len,
9513 btrfs_ino(new_dir), old_idx);
9518 /* And now for the dest. */
9519 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9520 /* force full log commit if subvolume involved. */
9521 btrfs_set_log_full_commit(dest->fs_info, trans);
9523 btrfs_pin_log_trans(dest);
9524 dest_log_pinned = true;
9525 ret = btrfs_insert_inode_ref(trans, root,
9526 old_dentry->d_name.name,
9527 old_dentry->d_name.len,
9529 btrfs_ino(old_dir), new_idx);
9534 /* Update inode version and ctime/mtime. */
9535 inode_inc_iversion(old_dir);
9536 inode_inc_iversion(new_dir);
9537 inode_inc_iversion(old_inode);
9538 inode_inc_iversion(new_inode);
9539 old_dir->i_ctime = old_dir->i_mtime = ctime;
9540 new_dir->i_ctime = new_dir->i_mtime = ctime;
9541 old_inode->i_ctime = ctime;
9542 new_inode->i_ctime = ctime;
9544 if (old_dentry->d_parent != new_dentry->d_parent) {
9545 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9546 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9549 /* src is a subvolume */
9550 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9551 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9552 ret = btrfs_unlink_subvol(trans, root, old_dir,
9554 old_dentry->d_name.name,
9555 old_dentry->d_name.len);
9556 } else { /* src is an inode */
9557 ret = __btrfs_unlink_inode(trans, root, old_dir,
9558 old_dentry->d_inode,
9559 old_dentry->d_name.name,
9560 old_dentry->d_name.len);
9562 ret = btrfs_update_inode(trans, root, old_inode);
9565 btrfs_abort_transaction(trans, root, ret);
9569 /* dest is a subvolume */
9570 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9571 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9572 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9574 new_dentry->d_name.name,
9575 new_dentry->d_name.len);
9576 } else { /* dest is an inode */
9577 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9578 new_dentry->d_inode,
9579 new_dentry->d_name.name,
9580 new_dentry->d_name.len);
9582 ret = btrfs_update_inode(trans, dest, new_inode);
9585 btrfs_abort_transaction(trans, root, ret);
9589 ret = btrfs_add_link(trans, new_dir, old_inode,
9590 new_dentry->d_name.name,
9591 new_dentry->d_name.len, 0, old_idx);
9593 btrfs_abort_transaction(trans, root, ret);
9597 ret = btrfs_add_link(trans, old_dir, new_inode,
9598 old_dentry->d_name.name,
9599 old_dentry->d_name.len, 0, new_idx);
9601 btrfs_abort_transaction(trans, root, ret);
9605 if (old_inode->i_nlink == 1)
9606 BTRFS_I(old_inode)->dir_index = old_idx;
9607 if (new_inode->i_nlink == 1)
9608 BTRFS_I(new_inode)->dir_index = new_idx;
9610 if (root_log_pinned) {
9611 parent = new_dentry->d_parent;
9612 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9613 btrfs_end_log_trans(root);
9614 root_log_pinned = false;
9616 if (dest_log_pinned) {
9617 parent = old_dentry->d_parent;
9618 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9619 btrfs_end_log_trans(dest);
9620 dest_log_pinned = false;
9624 * If we have pinned a log and an error happened, we unpin tasks
9625 * trying to sync the log and force them to fallback to a transaction
9626 * commit if the log currently contains any of the inodes involved in
9627 * this rename operation (to ensure we do not persist a log with an
9628 * inconsistent state for any of these inodes or leading to any
9629 * inconsistencies when replayed). If the transaction was aborted, the
9630 * abortion reason is propagated to userspace when attempting to commit
9631 * the transaction. If the log does not contain any of these inodes, we
9632 * allow the tasks to sync it.
9634 if (ret && (root_log_pinned || dest_log_pinned)) {
9635 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9636 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9637 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9639 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9640 btrfs_set_log_full_commit(root->fs_info, trans);
9642 if (root_log_pinned) {
9643 btrfs_end_log_trans(root);
9644 root_log_pinned = false;
9646 if (dest_log_pinned) {
9647 btrfs_end_log_trans(dest);
9648 dest_log_pinned = false;
9651 ret = btrfs_end_transaction(trans, root);
9653 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9654 up_read(&dest->fs_info->subvol_sem);
9655 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9656 up_read(&root->fs_info->subvol_sem);
9661 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9662 struct btrfs_root *root,
9664 struct dentry *dentry)
9667 struct inode *inode;
9671 ret = btrfs_find_free_ino(root, &objectid);
9675 inode = btrfs_new_inode(trans, root, dir,
9676 dentry->d_name.name,
9680 S_IFCHR | WHITEOUT_MODE,
9683 if (IS_ERR(inode)) {
9684 ret = PTR_ERR(inode);
9688 inode->i_op = &btrfs_special_inode_operations;
9689 init_special_inode(inode, inode->i_mode,
9692 ret = btrfs_init_inode_security(trans, inode, dir,
9697 ret = btrfs_add_nondir(trans, dir, dentry,
9702 ret = btrfs_update_inode(trans, root, inode);
9704 unlock_new_inode(inode);
9706 inode_dec_link_count(inode);
9712 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9713 struct inode *new_dir, struct dentry *new_dentry,
9716 struct btrfs_trans_handle *trans;
9717 unsigned int trans_num_items;
9718 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9719 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9720 struct inode *new_inode = d_inode(new_dentry);
9721 struct inode *old_inode = d_inode(old_dentry);
9725 u64 old_ino = btrfs_ino(old_inode);
9726 bool log_pinned = false;
9728 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9731 /* we only allow rename subvolume link between subvolumes */
9732 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9735 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9736 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9739 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9740 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9744 /* check for collisions, even if the name isn't there */
9745 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9746 new_dentry->d_name.name,
9747 new_dentry->d_name.len);
9750 if (ret == -EEXIST) {
9752 * eexist without a new_inode */
9753 if (WARN_ON(!new_inode)) {
9757 /* maybe -EOVERFLOW */
9764 * we're using rename to replace one file with another. Start IO on it
9765 * now so we don't add too much work to the end of the transaction
9767 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9768 filemap_flush(old_inode->i_mapping);
9770 /* close the racy window with snapshot create/destroy ioctl */
9771 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9772 down_read(&root->fs_info->subvol_sem);
9774 * We want to reserve the absolute worst case amount of items. So if
9775 * both inodes are subvols and we need to unlink them then that would
9776 * require 4 item modifications, but if they are both normal inodes it
9777 * would require 5 item modifications, so we'll assume they are normal
9778 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9779 * should cover the worst case number of items we'll modify.
9780 * If our rename has the whiteout flag, we need more 5 units for the
9781 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9782 * when selinux is enabled).
9784 trans_num_items = 11;
9785 if (flags & RENAME_WHITEOUT)
9786 trans_num_items += 5;
9787 trans = btrfs_start_transaction(root, trans_num_items);
9788 if (IS_ERR(trans)) {
9789 ret = PTR_ERR(trans);
9794 btrfs_record_root_in_trans(trans, dest);
9796 ret = btrfs_set_inode_index(new_dir, &index);
9800 BTRFS_I(old_inode)->dir_index = 0ULL;
9801 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9802 /* force full log commit if subvolume involved. */
9803 btrfs_set_log_full_commit(root->fs_info, trans);
9805 btrfs_pin_log_trans(root);
9807 ret = btrfs_insert_inode_ref(trans, dest,
9808 new_dentry->d_name.name,
9809 new_dentry->d_name.len,
9811 btrfs_ino(new_dir), index);
9816 inode_inc_iversion(old_dir);
9817 inode_inc_iversion(new_dir);
9818 inode_inc_iversion(old_inode);
9819 old_dir->i_ctime = old_dir->i_mtime =
9820 new_dir->i_ctime = new_dir->i_mtime =
9821 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9823 if (old_dentry->d_parent != new_dentry->d_parent)
9824 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9826 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9827 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9828 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9829 old_dentry->d_name.name,
9830 old_dentry->d_name.len);
9832 ret = __btrfs_unlink_inode(trans, root, old_dir,
9833 d_inode(old_dentry),
9834 old_dentry->d_name.name,
9835 old_dentry->d_name.len);
9837 ret = btrfs_update_inode(trans, root, old_inode);
9840 btrfs_abort_transaction(trans, root, ret);
9845 inode_inc_iversion(new_inode);
9846 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9847 if (unlikely(btrfs_ino(new_inode) ==
9848 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9849 root_objectid = BTRFS_I(new_inode)->location.objectid;
9850 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9852 new_dentry->d_name.name,
9853 new_dentry->d_name.len);
9854 BUG_ON(new_inode->i_nlink == 0);
9856 ret = btrfs_unlink_inode(trans, dest, new_dir,
9857 d_inode(new_dentry),
9858 new_dentry->d_name.name,
9859 new_dentry->d_name.len);
9861 if (!ret && new_inode->i_nlink == 0)
9862 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9864 btrfs_abort_transaction(trans, root, ret);
9869 ret = btrfs_add_link(trans, new_dir, old_inode,
9870 new_dentry->d_name.name,
9871 new_dentry->d_name.len, 0, index);
9873 btrfs_abort_transaction(trans, root, ret);
9877 if (old_inode->i_nlink == 1)
9878 BTRFS_I(old_inode)->dir_index = index;
9881 struct dentry *parent = new_dentry->d_parent;
9883 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9884 btrfs_end_log_trans(root);
9888 if (flags & RENAME_WHITEOUT) {
9889 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9893 btrfs_abort_transaction(trans, root, ret);
9899 * If we have pinned the log and an error happened, we unpin tasks
9900 * trying to sync the log and force them to fallback to a transaction
9901 * commit if the log currently contains any of the inodes involved in
9902 * this rename operation (to ensure we do not persist a log with an
9903 * inconsistent state for any of these inodes or leading to any
9904 * inconsistencies when replayed). If the transaction was aborted, the
9905 * abortion reason is propagated to userspace when attempting to commit
9906 * the transaction. If the log does not contain any of these inodes, we
9907 * allow the tasks to sync it.
9909 if (ret && log_pinned) {
9910 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9911 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9912 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9914 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9915 btrfs_set_log_full_commit(root->fs_info, trans);
9917 btrfs_end_log_trans(root);
9920 btrfs_end_transaction(trans, root);
9922 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9923 up_read(&root->fs_info->subvol_sem);
9928 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9929 struct inode *new_dir, struct dentry *new_dentry,
9932 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9935 if (flags & RENAME_EXCHANGE)
9936 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9939 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9942 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9944 struct btrfs_delalloc_work *delalloc_work;
9945 struct inode *inode;
9947 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9949 inode = delalloc_work->inode;
9950 filemap_flush(inode->i_mapping);
9951 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9952 &BTRFS_I(inode)->runtime_flags))
9953 filemap_flush(inode->i_mapping);
9955 if (delalloc_work->delay_iput)
9956 btrfs_add_delayed_iput(inode);
9959 complete(&delalloc_work->completion);
9962 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9965 struct btrfs_delalloc_work *work;
9967 work = kmalloc(sizeof(*work), GFP_NOFS);
9971 init_completion(&work->completion);
9972 INIT_LIST_HEAD(&work->list);
9973 work->inode = inode;
9974 work->delay_iput = delay_iput;
9975 WARN_ON_ONCE(!inode);
9976 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9977 btrfs_run_delalloc_work, NULL, NULL);
9982 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9984 wait_for_completion(&work->completion);
9989 * some fairly slow code that needs optimization. This walks the list
9990 * of all the inodes with pending delalloc and forces them to disk.
9992 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9995 struct btrfs_inode *binode;
9996 struct inode *inode;
9997 struct btrfs_delalloc_work *work, *next;
9998 struct list_head works;
9999 struct list_head splice;
10002 INIT_LIST_HEAD(&works);
10003 INIT_LIST_HEAD(&splice);
10005 mutex_lock(&root->delalloc_mutex);
10006 spin_lock(&root->delalloc_lock);
10007 list_splice_init(&root->delalloc_inodes, &splice);
10008 while (!list_empty(&splice)) {
10009 binode = list_entry(splice.next, struct btrfs_inode,
10012 list_move_tail(&binode->delalloc_inodes,
10013 &root->delalloc_inodes);
10014 inode = igrab(&binode->vfs_inode);
10016 cond_resched_lock(&root->delalloc_lock);
10019 spin_unlock(&root->delalloc_lock);
10021 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10024 btrfs_add_delayed_iput(inode);
10030 list_add_tail(&work->list, &works);
10031 btrfs_queue_work(root->fs_info->flush_workers,
10034 if (nr != -1 && ret >= nr)
10037 spin_lock(&root->delalloc_lock);
10039 spin_unlock(&root->delalloc_lock);
10042 list_for_each_entry_safe(work, next, &works, list) {
10043 list_del_init(&work->list);
10044 btrfs_wait_and_free_delalloc_work(work);
10047 if (!list_empty_careful(&splice)) {
10048 spin_lock(&root->delalloc_lock);
10049 list_splice_tail(&splice, &root->delalloc_inodes);
10050 spin_unlock(&root->delalloc_lock);
10052 mutex_unlock(&root->delalloc_mutex);
10056 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10060 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
10063 ret = __start_delalloc_inodes(root, delay_iput, -1);
10067 * the filemap_flush will queue IO into the worker threads, but
10068 * we have to make sure the IO is actually started and that
10069 * ordered extents get created before we return
10071 atomic_inc(&root->fs_info->async_submit_draining);
10072 while (atomic_read(&root->fs_info->nr_async_submits) ||
10073 atomic_read(&root->fs_info->async_delalloc_pages)) {
10074 wait_event(root->fs_info->async_submit_wait,
10075 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
10076 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
10078 atomic_dec(&root->fs_info->async_submit_draining);
10082 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10085 struct btrfs_root *root;
10086 struct list_head splice;
10089 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10092 INIT_LIST_HEAD(&splice);
10094 mutex_lock(&fs_info->delalloc_root_mutex);
10095 spin_lock(&fs_info->delalloc_root_lock);
10096 list_splice_init(&fs_info->delalloc_roots, &splice);
10097 while (!list_empty(&splice) && nr) {
10098 root = list_first_entry(&splice, struct btrfs_root,
10100 root = btrfs_grab_fs_root(root);
10102 list_move_tail(&root->delalloc_root,
10103 &fs_info->delalloc_roots);
10104 spin_unlock(&fs_info->delalloc_root_lock);
10106 ret = __start_delalloc_inodes(root, delay_iput, nr);
10107 btrfs_put_fs_root(root);
10115 spin_lock(&fs_info->delalloc_root_lock);
10117 spin_unlock(&fs_info->delalloc_root_lock);
10120 atomic_inc(&fs_info->async_submit_draining);
10121 while (atomic_read(&fs_info->nr_async_submits) ||
10122 atomic_read(&fs_info->async_delalloc_pages)) {
10123 wait_event(fs_info->async_submit_wait,
10124 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10125 atomic_read(&fs_info->async_delalloc_pages) == 0));
10127 atomic_dec(&fs_info->async_submit_draining);
10129 if (!list_empty_careful(&splice)) {
10130 spin_lock(&fs_info->delalloc_root_lock);
10131 list_splice_tail(&splice, &fs_info->delalloc_roots);
10132 spin_unlock(&fs_info->delalloc_root_lock);
10134 mutex_unlock(&fs_info->delalloc_root_mutex);
10138 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10139 const char *symname)
10141 struct btrfs_trans_handle *trans;
10142 struct btrfs_root *root = BTRFS_I(dir)->root;
10143 struct btrfs_path *path;
10144 struct btrfs_key key;
10145 struct inode *inode = NULL;
10147 int drop_inode = 0;
10153 struct btrfs_file_extent_item *ei;
10154 struct extent_buffer *leaf;
10156 name_len = strlen(symname);
10157 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
10158 return -ENAMETOOLONG;
10161 * 2 items for inode item and ref
10162 * 2 items for dir items
10163 * 1 item for updating parent inode item
10164 * 1 item for the inline extent item
10165 * 1 item for xattr if selinux is on
10167 trans = btrfs_start_transaction(root, 7);
10169 return PTR_ERR(trans);
10171 err = btrfs_find_free_ino(root, &objectid);
10175 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10176 dentry->d_name.len, btrfs_ino(dir), objectid,
10177 S_IFLNK|S_IRWXUGO, &index);
10178 if (IS_ERR(inode)) {
10179 err = PTR_ERR(inode);
10184 * If the active LSM wants to access the inode during
10185 * d_instantiate it needs these. Smack checks to see
10186 * if the filesystem supports xattrs by looking at the
10189 inode->i_fop = &btrfs_file_operations;
10190 inode->i_op = &btrfs_file_inode_operations;
10191 inode->i_mapping->a_ops = &btrfs_aops;
10192 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10194 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10196 goto out_unlock_inode;
10198 path = btrfs_alloc_path();
10201 goto out_unlock_inode;
10203 key.objectid = btrfs_ino(inode);
10205 key.type = BTRFS_EXTENT_DATA_KEY;
10206 datasize = btrfs_file_extent_calc_inline_size(name_len);
10207 err = btrfs_insert_empty_item(trans, root, path, &key,
10210 btrfs_free_path(path);
10211 goto out_unlock_inode;
10213 leaf = path->nodes[0];
10214 ei = btrfs_item_ptr(leaf, path->slots[0],
10215 struct btrfs_file_extent_item);
10216 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10217 btrfs_set_file_extent_type(leaf, ei,
10218 BTRFS_FILE_EXTENT_INLINE);
10219 btrfs_set_file_extent_encryption(leaf, ei, 0);
10220 btrfs_set_file_extent_compression(leaf, ei, 0);
10221 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10222 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10224 ptr = btrfs_file_extent_inline_start(ei);
10225 write_extent_buffer(leaf, symname, ptr, name_len);
10226 btrfs_mark_buffer_dirty(leaf);
10227 btrfs_free_path(path);
10229 inode->i_op = &btrfs_symlink_inode_operations;
10230 inode_nohighmem(inode);
10231 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10232 inode_set_bytes(inode, name_len);
10233 btrfs_i_size_write(inode, name_len);
10234 err = btrfs_update_inode(trans, root, inode);
10236 * Last step, add directory indexes for our symlink inode. This is the
10237 * last step to avoid extra cleanup of these indexes if an error happens
10241 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10244 goto out_unlock_inode;
10247 unlock_new_inode(inode);
10248 d_instantiate(dentry, inode);
10251 btrfs_end_transaction(trans, root);
10253 inode_dec_link_count(inode);
10256 btrfs_btree_balance_dirty(root);
10261 unlock_new_inode(inode);
10265 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10266 u64 start, u64 num_bytes, u64 min_size,
10267 loff_t actual_len, u64 *alloc_hint,
10268 struct btrfs_trans_handle *trans)
10270 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10271 struct extent_map *em;
10272 struct btrfs_root *root = BTRFS_I(inode)->root;
10273 struct btrfs_key ins;
10274 u64 cur_offset = start;
10277 u64 last_alloc = (u64)-1;
10279 bool own_trans = true;
10283 while (num_bytes > 0) {
10285 trans = btrfs_start_transaction(root, 3);
10286 if (IS_ERR(trans)) {
10287 ret = PTR_ERR(trans);
10292 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10293 cur_bytes = max(cur_bytes, min_size);
10295 * If we are severely fragmented we could end up with really
10296 * small allocations, so if the allocator is returning small
10297 * chunks lets make its job easier by only searching for those
10300 cur_bytes = min(cur_bytes, last_alloc);
10301 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
10302 *alloc_hint, &ins, 1, 0);
10305 btrfs_end_transaction(trans, root);
10308 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
10310 last_alloc = ins.offset;
10311 ret = insert_reserved_file_extent(trans, inode,
10312 cur_offset, ins.objectid,
10313 ins.offset, ins.offset,
10314 ins.offset, 0, 0, 0,
10315 BTRFS_FILE_EXTENT_PREALLOC);
10317 btrfs_free_reserved_extent(root, ins.objectid,
10319 btrfs_abort_transaction(trans, root, ret);
10321 btrfs_end_transaction(trans, root);
10325 btrfs_drop_extent_cache(inode, cur_offset,
10326 cur_offset + ins.offset -1, 0);
10328 em = alloc_extent_map();
10330 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10331 &BTRFS_I(inode)->runtime_flags);
10335 em->start = cur_offset;
10336 em->orig_start = cur_offset;
10337 em->len = ins.offset;
10338 em->block_start = ins.objectid;
10339 em->block_len = ins.offset;
10340 em->orig_block_len = ins.offset;
10341 em->ram_bytes = ins.offset;
10342 em->bdev = root->fs_info->fs_devices->latest_bdev;
10343 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10344 em->generation = trans->transid;
10347 write_lock(&em_tree->lock);
10348 ret = add_extent_mapping(em_tree, em, 1);
10349 write_unlock(&em_tree->lock);
10350 if (ret != -EEXIST)
10352 btrfs_drop_extent_cache(inode, cur_offset,
10353 cur_offset + ins.offset - 1,
10356 free_extent_map(em);
10358 num_bytes -= ins.offset;
10359 cur_offset += ins.offset;
10360 *alloc_hint = ins.objectid + ins.offset;
10362 inode_inc_iversion(inode);
10363 inode->i_ctime = current_fs_time(inode->i_sb);
10364 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10365 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10366 (actual_len > inode->i_size) &&
10367 (cur_offset > inode->i_size)) {
10368 if (cur_offset > actual_len)
10369 i_size = actual_len;
10371 i_size = cur_offset;
10372 i_size_write(inode, i_size);
10373 btrfs_ordered_update_i_size(inode, i_size, NULL);
10376 ret = btrfs_update_inode(trans, root, inode);
10379 btrfs_abort_transaction(trans, root, ret);
10381 btrfs_end_transaction(trans, root);
10386 btrfs_end_transaction(trans, root);
10391 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10392 u64 start, u64 num_bytes, u64 min_size,
10393 loff_t actual_len, u64 *alloc_hint)
10395 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10396 min_size, actual_len, alloc_hint,
10400 int btrfs_prealloc_file_range_trans(struct inode *inode,
10401 struct btrfs_trans_handle *trans, int mode,
10402 u64 start, u64 num_bytes, u64 min_size,
10403 loff_t actual_len, u64 *alloc_hint)
10405 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10406 min_size, actual_len, alloc_hint, trans);
10409 static int btrfs_set_page_dirty(struct page *page)
10411 return __set_page_dirty_nobuffers(page);
10414 static int btrfs_permission(struct inode *inode, int mask)
10416 struct btrfs_root *root = BTRFS_I(inode)->root;
10417 umode_t mode = inode->i_mode;
10419 if (mask & MAY_WRITE &&
10420 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10421 if (btrfs_root_readonly(root))
10423 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10426 return generic_permission(inode, mask);
10429 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10431 struct btrfs_trans_handle *trans;
10432 struct btrfs_root *root = BTRFS_I(dir)->root;
10433 struct inode *inode = NULL;
10439 * 5 units required for adding orphan entry
10441 trans = btrfs_start_transaction(root, 5);
10443 return PTR_ERR(trans);
10445 ret = btrfs_find_free_ino(root, &objectid);
10449 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10450 btrfs_ino(dir), objectid, mode, &index);
10451 if (IS_ERR(inode)) {
10452 ret = PTR_ERR(inode);
10457 inode->i_fop = &btrfs_file_operations;
10458 inode->i_op = &btrfs_file_inode_operations;
10460 inode->i_mapping->a_ops = &btrfs_aops;
10461 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10463 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10467 ret = btrfs_update_inode(trans, root, inode);
10470 ret = btrfs_orphan_add(trans, inode);
10475 * We set number of links to 0 in btrfs_new_inode(), and here we set
10476 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10479 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10481 set_nlink(inode, 1);
10482 unlock_new_inode(inode);
10483 d_tmpfile(dentry, inode);
10484 mark_inode_dirty(inode);
10487 btrfs_end_transaction(trans, root);
10490 btrfs_balance_delayed_items(root);
10491 btrfs_btree_balance_dirty(root);
10495 unlock_new_inode(inode);
10500 /* Inspired by filemap_check_errors() */
10501 int btrfs_inode_check_errors(struct inode *inode)
10505 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10506 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10508 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10509 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10515 static const struct inode_operations btrfs_dir_inode_operations = {
10516 .getattr = btrfs_getattr,
10517 .lookup = btrfs_lookup,
10518 .create = btrfs_create,
10519 .unlink = btrfs_unlink,
10520 .link = btrfs_link,
10521 .mkdir = btrfs_mkdir,
10522 .rmdir = btrfs_rmdir,
10523 .rename2 = btrfs_rename2,
10524 .symlink = btrfs_symlink,
10525 .setattr = btrfs_setattr,
10526 .mknod = btrfs_mknod,
10527 .setxattr = generic_setxattr,
10528 .getxattr = generic_getxattr,
10529 .listxattr = btrfs_listxattr,
10530 .removexattr = generic_removexattr,
10531 .permission = btrfs_permission,
10532 .get_acl = btrfs_get_acl,
10533 .set_acl = btrfs_set_acl,
10534 .update_time = btrfs_update_time,
10535 .tmpfile = btrfs_tmpfile,
10537 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10538 .lookup = btrfs_lookup,
10539 .permission = btrfs_permission,
10540 .get_acl = btrfs_get_acl,
10541 .set_acl = btrfs_set_acl,
10542 .update_time = btrfs_update_time,
10545 static const struct file_operations btrfs_dir_file_operations = {
10546 .llseek = generic_file_llseek,
10547 .read = generic_read_dir,
10548 .iterate_shared = btrfs_real_readdir,
10549 .unlocked_ioctl = btrfs_ioctl,
10550 #ifdef CONFIG_COMPAT
10551 .compat_ioctl = btrfs_compat_ioctl,
10553 .release = btrfs_release_file,
10554 .fsync = btrfs_sync_file,
10557 static const struct extent_io_ops btrfs_extent_io_ops = {
10558 .fill_delalloc = run_delalloc_range,
10559 .submit_bio_hook = btrfs_submit_bio_hook,
10560 .merge_bio_hook = btrfs_merge_bio_hook,
10561 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10562 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10563 .writepage_start_hook = btrfs_writepage_start_hook,
10564 .set_bit_hook = btrfs_set_bit_hook,
10565 .clear_bit_hook = btrfs_clear_bit_hook,
10566 .merge_extent_hook = btrfs_merge_extent_hook,
10567 .split_extent_hook = btrfs_split_extent_hook,
10571 * btrfs doesn't support the bmap operation because swapfiles
10572 * use bmap to make a mapping of extents in the file. They assume
10573 * these extents won't change over the life of the file and they
10574 * use the bmap result to do IO directly to the drive.
10576 * the btrfs bmap call would return logical addresses that aren't
10577 * suitable for IO and they also will change frequently as COW
10578 * operations happen. So, swapfile + btrfs == corruption.
10580 * For now we're avoiding this by dropping bmap.
10582 static const struct address_space_operations btrfs_aops = {
10583 .readpage = btrfs_readpage,
10584 .writepage = btrfs_writepage,
10585 .writepages = btrfs_writepages,
10586 .readpages = btrfs_readpages,
10587 .direct_IO = btrfs_direct_IO,
10588 .invalidatepage = btrfs_invalidatepage,
10589 .releasepage = btrfs_releasepage,
10590 .set_page_dirty = btrfs_set_page_dirty,
10591 .error_remove_page = generic_error_remove_page,
10594 static const struct address_space_operations btrfs_symlink_aops = {
10595 .readpage = btrfs_readpage,
10596 .writepage = btrfs_writepage,
10597 .invalidatepage = btrfs_invalidatepage,
10598 .releasepage = btrfs_releasepage,
10601 static const struct inode_operations btrfs_file_inode_operations = {
10602 .getattr = btrfs_getattr,
10603 .setattr = btrfs_setattr,
10604 .setxattr = generic_setxattr,
10605 .getxattr = generic_getxattr,
10606 .listxattr = btrfs_listxattr,
10607 .removexattr = generic_removexattr,
10608 .permission = btrfs_permission,
10609 .fiemap = btrfs_fiemap,
10610 .get_acl = btrfs_get_acl,
10611 .set_acl = btrfs_set_acl,
10612 .update_time = btrfs_update_time,
10614 static const struct inode_operations btrfs_special_inode_operations = {
10615 .getattr = btrfs_getattr,
10616 .setattr = btrfs_setattr,
10617 .permission = btrfs_permission,
10618 .setxattr = generic_setxattr,
10619 .getxattr = generic_getxattr,
10620 .listxattr = btrfs_listxattr,
10621 .removexattr = generic_removexattr,
10622 .get_acl = btrfs_get_acl,
10623 .set_acl = btrfs_set_acl,
10624 .update_time = btrfs_update_time,
10626 static const struct inode_operations btrfs_symlink_inode_operations = {
10627 .readlink = generic_readlink,
10628 .get_link = page_get_link,
10629 .getattr = btrfs_getattr,
10630 .setattr = btrfs_setattr,
10631 .permission = btrfs_permission,
10632 .setxattr = generic_setxattr,
10633 .getxattr = generic_getxattr,
10634 .listxattr = btrfs_listxattr,
10635 .removexattr = generic_removexattr,
10636 .update_time = btrfs_update_time,
10639 const struct dentry_operations btrfs_dentry_operations = {
10640 .d_delete = btrfs_dentry_delete,
10641 .d_release = btrfs_dentry_release,