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/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
54 #include "compression.h"
56 #include "free-space-cache.h"
57 #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;
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_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, u64 delalloc_end,
109 int *page_started, unsigned long *nr_written,
110 int unlock, struct btrfs_dedupe_hash *hash);
111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
112 u64 orig_start, u64 block_start,
113 u64 block_len, u64 orig_block_len,
114 u64 ram_bytes, int compress_type,
117 static void __endio_write_update_ordered(struct inode *inode,
118 const u64 offset, const u64 bytes,
119 const bool uptodate);
122 * Cleanup all submitted ordered extents in specified range to handle errors
123 * from the fill_dellaloc() callback.
125 * NOTE: caller must ensure that when an error happens, it can not call
126 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128 * to be released, which we want to happen only when finishing the ordered
129 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130 * fill_delalloc() callback already does proper cleanup for the first page of
131 * the range, that is, it invokes the callback writepage_end_io_hook() for the
132 * range of the first page.
134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
138 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
139 bytes - PAGE_SIZE, false);
142 static int btrfs_dirty_inode(struct inode *inode);
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
147 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152 struct inode *inode, struct inode *dir,
153 const struct qstr *qstr)
157 err = btrfs_init_acl(trans, inode, dir);
159 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
164 * this does all the hard work for inserting an inline extent into
165 * the btree. The caller should have done a btrfs_drop_extents so that
166 * no overlapping inline items exist in the btree
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169 struct btrfs_path *path, int extent_inserted,
170 struct btrfs_root *root, struct inode *inode,
171 u64 start, size_t size, size_t compressed_size,
173 struct page **compressed_pages)
175 struct extent_buffer *leaf;
176 struct page *page = NULL;
179 struct btrfs_file_extent_item *ei;
181 size_t cur_size = size;
182 unsigned long offset;
184 if (compressed_size && compressed_pages)
185 cur_size = compressed_size;
187 inode_add_bytes(inode, size);
189 if (!extent_inserted) {
190 struct btrfs_key key;
193 key.objectid = btrfs_ino(BTRFS_I(inode));
195 key.type = BTRFS_EXTENT_DATA_KEY;
197 datasize = btrfs_file_extent_calc_inline_size(cur_size);
198 path->leave_spinning = 1;
199 ret = btrfs_insert_empty_item(trans, root, path, &key,
204 leaf = path->nodes[0];
205 ei = btrfs_item_ptr(leaf, path->slots[0],
206 struct btrfs_file_extent_item);
207 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
208 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
209 btrfs_set_file_extent_encryption(leaf, ei, 0);
210 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
211 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
212 ptr = btrfs_file_extent_inline_start(ei);
214 if (compress_type != BTRFS_COMPRESS_NONE) {
217 while (compressed_size > 0) {
218 cpage = compressed_pages[i];
219 cur_size = min_t(unsigned long, compressed_size,
222 kaddr = kmap_atomic(cpage);
223 write_extent_buffer(leaf, kaddr, ptr, cur_size);
224 kunmap_atomic(kaddr);
228 compressed_size -= cur_size;
230 btrfs_set_file_extent_compression(leaf, ei,
233 page = find_get_page(inode->i_mapping,
234 start >> PAGE_SHIFT);
235 btrfs_set_file_extent_compression(leaf, ei, 0);
236 kaddr = kmap_atomic(page);
237 offset = start & (PAGE_SIZE - 1);
238 write_extent_buffer(leaf, kaddr + offset, ptr, size);
239 kunmap_atomic(kaddr);
242 btrfs_mark_buffer_dirty(leaf);
243 btrfs_release_path(path);
246 * we're an inline extent, so nobody can
247 * extend the file past i_size without locking
248 * a page we already have locked.
250 * We must do any isize and inode updates
251 * before we unlock the pages. Otherwise we
252 * could end up racing with unlink.
254 BTRFS_I(inode)->disk_i_size = inode->i_size;
255 ret = btrfs_update_inode(trans, root, inode);
263 * conditionally insert an inline extent into the file. This
264 * does the checks required to make sure the data is small enough
265 * to fit as an inline extent.
267 static noinline int cow_file_range_inline(struct btrfs_root *root,
268 struct inode *inode, u64 start,
269 u64 end, size_t compressed_size,
271 struct page **compressed_pages)
273 struct btrfs_fs_info *fs_info = root->fs_info;
274 struct btrfs_trans_handle *trans;
275 u64 isize = i_size_read(inode);
276 u64 actual_end = min(end + 1, isize);
277 u64 inline_len = actual_end - start;
278 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
279 u64 data_len = inline_len;
281 struct btrfs_path *path;
282 int extent_inserted = 0;
283 u32 extent_item_size;
286 data_len = compressed_size;
289 actual_end > fs_info->sectorsize ||
290 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
292 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
294 data_len > fs_info->max_inline) {
298 path = btrfs_alloc_path();
302 trans = btrfs_join_transaction(root);
304 btrfs_free_path(path);
305 return PTR_ERR(trans);
307 trans->block_rsv = &fs_info->delalloc_block_rsv;
309 if (compressed_size && compressed_pages)
310 extent_item_size = btrfs_file_extent_calc_inline_size(
313 extent_item_size = btrfs_file_extent_calc_inline_size(
316 ret = __btrfs_drop_extents(trans, root, inode, path,
317 start, aligned_end, NULL,
318 1, 1, extent_item_size, &extent_inserted);
320 btrfs_abort_transaction(trans, ret);
324 if (isize > actual_end)
325 inline_len = min_t(u64, isize, actual_end);
326 ret = insert_inline_extent(trans, path, extent_inserted,
328 inline_len, compressed_size,
329 compress_type, compressed_pages);
330 if (ret && ret != -ENOSPC) {
331 btrfs_abort_transaction(trans, ret);
333 } else if (ret == -ENOSPC) {
338 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
339 btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
340 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
343 * Don't forget to free the reserved space, as for inlined extent
344 * it won't count as data extent, free them directly here.
345 * And at reserve time, it's always aligned to page size, so
346 * just free one page here.
348 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
349 btrfs_free_path(path);
350 btrfs_end_transaction(trans);
354 struct async_extent {
359 unsigned long nr_pages;
361 struct list_head list;
366 struct btrfs_root *root;
367 struct page *locked_page;
370 struct list_head extents;
371 struct btrfs_work work;
374 static noinline int add_async_extent(struct async_cow *cow,
375 u64 start, u64 ram_size,
378 unsigned long nr_pages,
381 struct async_extent *async_extent;
383 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
384 BUG_ON(!async_extent); /* -ENOMEM */
385 async_extent->start = start;
386 async_extent->ram_size = ram_size;
387 async_extent->compressed_size = compressed_size;
388 async_extent->pages = pages;
389 async_extent->nr_pages = nr_pages;
390 async_extent->compress_type = compress_type;
391 list_add_tail(&async_extent->list, &cow->extents);
395 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
400 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
403 if (BTRFS_I(inode)->defrag_compress)
405 /* bad compression ratios */
406 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
408 if (btrfs_test_opt(fs_info, COMPRESS) ||
409 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
410 BTRFS_I(inode)->prop_compress)
411 return btrfs_compress_heuristic(inode, start, end);
415 static inline void inode_should_defrag(struct btrfs_inode *inode,
416 u64 start, u64 end, u64 num_bytes, u64 small_write)
418 /* If this is a small write inside eof, kick off a defrag */
419 if (num_bytes < small_write &&
420 (start > 0 || end + 1 < inode->disk_i_size))
421 btrfs_add_inode_defrag(NULL, inode);
425 * we create compressed extents in two phases. The first
426 * phase compresses a range of pages that have already been
427 * locked (both pages and state bits are locked).
429 * This is done inside an ordered work queue, and the compression
430 * is spread across many cpus. The actual IO submission is step
431 * two, and the ordered work queue takes care of making sure that
432 * happens in the same order things were put onto the queue by
433 * writepages and friends.
435 * If this code finds it can't get good compression, it puts an
436 * entry onto the work queue to write the uncompressed bytes. This
437 * makes sure that both compressed inodes and uncompressed inodes
438 * are written in the same order that the flusher thread sent them
441 static noinline void compress_file_range(struct inode *inode,
442 struct page *locked_page,
444 struct async_cow *async_cow,
447 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
448 struct btrfs_root *root = BTRFS_I(inode)->root;
450 u64 blocksize = fs_info->sectorsize;
452 u64 isize = i_size_read(inode);
454 struct page **pages = NULL;
455 unsigned long nr_pages;
456 unsigned long total_compressed = 0;
457 unsigned long total_in = 0;
460 int compress_type = fs_info->compress_type;
463 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
466 actual_end = min_t(u64, isize, end + 1);
469 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
470 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
471 nr_pages = min_t(unsigned long, nr_pages,
472 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
475 * we don't want to send crud past the end of i_size through
476 * compression, that's just a waste of CPU time. So, if the
477 * end of the file is before the start of our current
478 * requested range of bytes, we bail out to the uncompressed
479 * cleanup code that can deal with all of this.
481 * It isn't really the fastest way to fix things, but this is a
482 * very uncommon corner.
484 if (actual_end <= start)
485 goto cleanup_and_bail_uncompressed;
487 total_compressed = actual_end - start;
490 * skip compression for a small file range(<=blocksize) that
491 * isn't an inline extent, since it doesn't save disk space at all.
493 if (total_compressed <= blocksize &&
494 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
495 goto cleanup_and_bail_uncompressed;
497 total_compressed = min_t(unsigned long, total_compressed,
498 BTRFS_MAX_UNCOMPRESSED);
499 num_bytes = ALIGN(end - start + 1, blocksize);
500 num_bytes = max(blocksize, num_bytes);
505 * we do compression for mount -o compress and when the
506 * inode has not been flagged as nocompress. This flag can
507 * change at any time if we discover bad compression ratios.
509 if (inode_need_compress(inode, start, end)) {
511 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
513 /* just bail out to the uncompressed code */
517 if (BTRFS_I(inode)->defrag_compress)
518 compress_type = BTRFS_I(inode)->defrag_compress;
519 else if (BTRFS_I(inode)->prop_compress)
520 compress_type = BTRFS_I(inode)->prop_compress;
523 * we need to call clear_page_dirty_for_io on each
524 * page in the range. Otherwise applications with the file
525 * mmap'd can wander in and change the page contents while
526 * we are compressing them.
528 * If the compression fails for any reason, we set the pages
529 * dirty again later on.
531 extent_range_clear_dirty_for_io(inode, start, end);
533 ret = btrfs_compress_pages(compress_type,
534 inode->i_mapping, start,
541 unsigned long offset = total_compressed &
543 struct page *page = pages[nr_pages - 1];
546 /* zero the tail end of the last page, we might be
547 * sending it down to disk
550 kaddr = kmap_atomic(page);
551 memset(kaddr + offset, 0,
553 kunmap_atomic(kaddr);
560 /* lets try to make an inline extent */
561 if (ret || total_in < (actual_end - start)) {
562 /* we didn't compress the entire range, try
563 * to make an uncompressed inline extent.
565 ret = cow_file_range_inline(root, inode, start, end,
566 0, BTRFS_COMPRESS_NONE, NULL);
568 /* try making a compressed inline extent */
569 ret = cow_file_range_inline(root, inode, start, end,
571 compress_type, pages);
574 unsigned long clear_flags = EXTENT_DELALLOC |
575 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG;
576 unsigned long page_error_op;
578 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
579 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
582 * inline extent creation worked or returned error,
583 * we don't need to create any more async work items.
584 * Unlock and free up our temp pages.
586 extent_clear_unlock_delalloc(inode, start, end, end,
594 btrfs_free_reserved_data_space_noquota(inode,
603 * we aren't doing an inline extent round the compressed size
604 * up to a block size boundary so the allocator does sane
607 total_compressed = ALIGN(total_compressed, blocksize);
610 * one last check to make sure the compression is really a
611 * win, compare the page count read with the blocks on disk,
612 * compression must free at least one sector size
614 total_in = ALIGN(total_in, PAGE_SIZE);
615 if (total_compressed + blocksize <= total_in) {
616 num_bytes = total_in;
620 * The async work queues will take care of doing actual
621 * allocation on disk for these compressed pages, and
622 * will submit them to the elevator.
624 add_async_extent(async_cow, start, num_bytes,
625 total_compressed, pages, nr_pages,
628 if (start + num_bytes < end) {
639 * the compression code ran but failed to make things smaller,
640 * free any pages it allocated and our page pointer array
642 for (i = 0; i < nr_pages; i++) {
643 WARN_ON(pages[i]->mapping);
648 total_compressed = 0;
651 /* flag the file so we don't compress in the future */
652 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
653 !(BTRFS_I(inode)->prop_compress)) {
654 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
657 cleanup_and_bail_uncompressed:
659 * No compression, but we still need to write the pages in the file
660 * we've been given so far. redirty the locked page if it corresponds
661 * to our extent and set things up for the async work queue to run
662 * cow_file_range to do the normal delalloc dance.
664 if (page_offset(locked_page) >= start &&
665 page_offset(locked_page) <= end)
666 __set_page_dirty_nobuffers(locked_page);
667 /* unlocked later on in the async handlers */
670 extent_range_redirty_for_io(inode, start, end);
671 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
672 BTRFS_COMPRESS_NONE);
678 for (i = 0; i < nr_pages; i++) {
679 WARN_ON(pages[i]->mapping);
685 static void free_async_extent_pages(struct async_extent *async_extent)
689 if (!async_extent->pages)
692 for (i = 0; i < async_extent->nr_pages; i++) {
693 WARN_ON(async_extent->pages[i]->mapping);
694 put_page(async_extent->pages[i]);
696 kfree(async_extent->pages);
697 async_extent->nr_pages = 0;
698 async_extent->pages = NULL;
702 * phase two of compressed writeback. This is the ordered portion
703 * of the code, which only gets called in the order the work was
704 * queued. We walk all the async extents created by compress_file_range
705 * and send them down to the disk.
707 static noinline void submit_compressed_extents(struct inode *inode,
708 struct async_cow *async_cow)
710 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
711 struct async_extent *async_extent;
713 struct btrfs_key ins;
714 struct extent_map *em;
715 struct btrfs_root *root = BTRFS_I(inode)->root;
716 struct extent_io_tree *io_tree;
720 while (!list_empty(&async_cow->extents)) {
721 async_extent = list_entry(async_cow->extents.next,
722 struct async_extent, list);
723 list_del(&async_extent->list);
725 io_tree = &BTRFS_I(inode)->io_tree;
728 /* did the compression code fall back to uncompressed IO? */
729 if (!async_extent->pages) {
730 int page_started = 0;
731 unsigned long nr_written = 0;
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start +
735 async_extent->ram_size - 1);
737 /* allocate blocks */
738 ret = cow_file_range(inode, async_cow->locked_page,
740 async_extent->start +
741 async_extent->ram_size - 1,
742 async_extent->start +
743 async_extent->ram_size - 1,
744 &page_started, &nr_written, 0,
750 * if page_started, cow_file_range inserted an
751 * inline extent and took care of all the unlocking
752 * and IO for us. Otherwise, we need to submit
753 * all those pages down to the drive.
755 if (!page_started && !ret)
756 extent_write_locked_range(io_tree,
757 inode, async_extent->start,
758 async_extent->start +
759 async_extent->ram_size - 1,
763 unlock_page(async_cow->locked_page);
769 lock_extent(io_tree, async_extent->start,
770 async_extent->start + async_extent->ram_size - 1);
772 ret = btrfs_reserve_extent(root, async_extent->ram_size,
773 async_extent->compressed_size,
774 async_extent->compressed_size,
775 0, alloc_hint, &ins, 1, 1);
777 free_async_extent_pages(async_extent);
779 if (ret == -ENOSPC) {
780 unlock_extent(io_tree, async_extent->start,
781 async_extent->start +
782 async_extent->ram_size - 1);
785 * we need to redirty the pages if we decide to
786 * fallback to uncompressed IO, otherwise we
787 * will not submit these pages down to lower
790 extent_range_redirty_for_io(inode,
792 async_extent->start +
793 async_extent->ram_size - 1);
800 * here we're doing allocation and writeback of the
803 em = create_io_em(inode, async_extent->start,
804 async_extent->ram_size, /* len */
805 async_extent->start, /* orig_start */
806 ins.objectid, /* block_start */
807 ins.offset, /* block_len */
808 ins.offset, /* orig_block_len */
809 async_extent->ram_size, /* ram_bytes */
810 async_extent->compress_type,
811 BTRFS_ORDERED_COMPRESSED);
813 /* ret value is not necessary due to void function */
814 goto out_free_reserve;
817 ret = btrfs_add_ordered_extent_compress(inode,
820 async_extent->ram_size,
822 BTRFS_ORDERED_COMPRESSED,
823 async_extent->compress_type);
825 btrfs_drop_extent_cache(BTRFS_I(inode),
827 async_extent->start +
828 async_extent->ram_size - 1, 0);
829 goto out_free_reserve;
831 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
834 * clear dirty, set writeback and unlock the pages.
836 extent_clear_unlock_delalloc(inode, async_extent->start,
837 async_extent->start +
838 async_extent->ram_size - 1,
839 async_extent->start +
840 async_extent->ram_size - 1,
841 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
842 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
844 if (btrfs_submit_compressed_write(inode,
846 async_extent->ram_size,
848 ins.offset, async_extent->pages,
849 async_extent->nr_pages)) {
850 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
851 struct page *p = async_extent->pages[0];
852 const u64 start = async_extent->start;
853 const u64 end = start + async_extent->ram_size - 1;
855 p->mapping = inode->i_mapping;
856 tree->ops->writepage_end_io_hook(p, start, end,
859 extent_clear_unlock_delalloc(inode, start, end, end,
863 free_async_extent_pages(async_extent);
865 alloc_hint = ins.objectid + ins.offset;
871 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
872 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
874 extent_clear_unlock_delalloc(inode, async_extent->start,
875 async_extent->start +
876 async_extent->ram_size - 1,
877 async_extent->start +
878 async_extent->ram_size - 1,
879 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
880 EXTENT_DELALLOC_NEW |
881 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
882 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
883 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
885 free_async_extent_pages(async_extent);
890 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
893 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
894 struct extent_map *em;
897 read_lock(&em_tree->lock);
898 em = search_extent_mapping(em_tree, start, num_bytes);
901 * if block start isn't an actual block number then find the
902 * first block in this inode and use that as a hint. If that
903 * block is also bogus then just don't worry about it.
905 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
907 em = search_extent_mapping(em_tree, 0, 0);
908 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
909 alloc_hint = em->block_start;
913 alloc_hint = em->block_start;
917 read_unlock(&em_tree->lock);
923 * when extent_io.c finds a delayed allocation range in the file,
924 * the call backs end up in this code. The basic idea is to
925 * allocate extents on disk for the range, and create ordered data structs
926 * in ram to track those extents.
928 * locked_page is the page that writepage had locked already. We use
929 * it to make sure we don't do extra locks or unlocks.
931 * *page_started is set to one if we unlock locked_page and do everything
932 * required to start IO on it. It may be clean and already done with
935 static noinline int cow_file_range(struct inode *inode,
936 struct page *locked_page,
937 u64 start, u64 end, u64 delalloc_end,
938 int *page_started, unsigned long *nr_written,
939 int unlock, struct btrfs_dedupe_hash *hash)
941 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
942 struct btrfs_root *root = BTRFS_I(inode)->root;
945 unsigned long ram_size;
947 u64 cur_alloc_size = 0;
948 u64 blocksize = fs_info->sectorsize;
949 struct btrfs_key ins;
950 struct extent_map *em;
952 unsigned long page_ops;
953 bool extent_reserved = false;
956 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
962 num_bytes = ALIGN(end - start + 1, blocksize);
963 num_bytes = max(blocksize, num_bytes);
964 disk_num_bytes = num_bytes;
966 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
969 /* lets try to make an inline extent */
970 ret = cow_file_range_inline(root, inode, start, end, 0,
971 BTRFS_COMPRESS_NONE, NULL);
973 extent_clear_unlock_delalloc(inode, start, end,
975 EXTENT_LOCKED | EXTENT_DELALLOC |
976 EXTENT_DELALLOC_NEW |
977 EXTENT_DEFRAG, PAGE_UNLOCK |
978 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
980 btrfs_free_reserved_data_space_noquota(inode, start,
982 *nr_written = *nr_written +
983 (end - start + PAGE_SIZE) / PAGE_SIZE;
986 } else if (ret < 0) {
991 BUG_ON(disk_num_bytes >
992 btrfs_super_total_bytes(fs_info->super_copy));
994 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
995 btrfs_drop_extent_cache(BTRFS_I(inode), start,
996 start + num_bytes - 1, 0);
998 while (disk_num_bytes > 0) {
999 cur_alloc_size = disk_num_bytes;
1000 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1001 fs_info->sectorsize, 0, alloc_hint,
1005 cur_alloc_size = ins.offset;
1006 extent_reserved = true;
1008 ram_size = ins.offset;
1009 em = create_io_em(inode, start, ins.offset, /* len */
1010 start, /* orig_start */
1011 ins.objectid, /* block_start */
1012 ins.offset, /* block_len */
1013 ins.offset, /* orig_block_len */
1014 ram_size, /* ram_bytes */
1015 BTRFS_COMPRESS_NONE, /* compress_type */
1016 BTRFS_ORDERED_REGULAR /* type */);
1019 free_extent_map(em);
1021 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1022 ram_size, cur_alloc_size, 0);
1024 goto out_drop_extent_cache;
1026 if (root->root_key.objectid ==
1027 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1028 ret = btrfs_reloc_clone_csums(inode, start,
1031 * Only drop cache here, and process as normal.
1033 * We must not allow extent_clear_unlock_delalloc()
1034 * at out_unlock label to free meta of this ordered
1035 * extent, as its meta should be freed by
1036 * btrfs_finish_ordered_io().
1038 * So we must continue until @start is increased to
1039 * skip current ordered extent.
1042 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1043 start + ram_size - 1, 0);
1046 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1048 /* we're not doing compressed IO, don't unlock the first
1049 * page (which the caller expects to stay locked), don't
1050 * clear any dirty bits and don't set any writeback bits
1052 * Do set the Private2 bit so we know this page was properly
1053 * setup for writepage
1055 page_ops = unlock ? PAGE_UNLOCK : 0;
1056 page_ops |= PAGE_SET_PRIVATE2;
1058 extent_clear_unlock_delalloc(inode, start,
1059 start + ram_size - 1,
1060 delalloc_end, locked_page,
1061 EXTENT_LOCKED | EXTENT_DELALLOC,
1063 if (disk_num_bytes < cur_alloc_size)
1066 disk_num_bytes -= cur_alloc_size;
1067 num_bytes -= cur_alloc_size;
1068 alloc_hint = ins.objectid + ins.offset;
1069 start += cur_alloc_size;
1070 extent_reserved = false;
1073 * btrfs_reloc_clone_csums() error, since start is increased
1074 * extent_clear_unlock_delalloc() at out_unlock label won't
1075 * free metadata of current ordered extent, we're OK to exit.
1083 out_drop_extent_cache:
1084 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1086 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1087 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1089 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1090 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1091 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1094 * If we reserved an extent for our delalloc range (or a subrange) and
1095 * failed to create the respective ordered extent, then it means that
1096 * when we reserved the extent we decremented the extent's size from
1097 * the data space_info's bytes_may_use counter and incremented the
1098 * space_info's bytes_reserved counter by the same amount. We must make
1099 * sure extent_clear_unlock_delalloc() does not try to decrement again
1100 * the data space_info's bytes_may_use counter, therefore we do not pass
1101 * it the flag EXTENT_CLEAR_DATA_RESV.
1103 if (extent_reserved) {
1104 extent_clear_unlock_delalloc(inode, start,
1105 start + cur_alloc_size,
1106 start + cur_alloc_size,
1110 start += cur_alloc_size;
1114 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1116 clear_bits | EXTENT_CLEAR_DATA_RESV,
1122 * work queue call back to started compression on a file and pages
1124 static noinline void async_cow_start(struct btrfs_work *work)
1126 struct async_cow *async_cow;
1128 async_cow = container_of(work, struct async_cow, work);
1130 compress_file_range(async_cow->inode, async_cow->locked_page,
1131 async_cow->start, async_cow->end, async_cow,
1133 if (num_added == 0) {
1134 btrfs_add_delayed_iput(async_cow->inode);
1135 async_cow->inode = NULL;
1140 * work queue call back to submit previously compressed pages
1142 static noinline void async_cow_submit(struct btrfs_work *work)
1144 struct btrfs_fs_info *fs_info;
1145 struct async_cow *async_cow;
1146 struct btrfs_root *root;
1147 unsigned long nr_pages;
1149 async_cow = container_of(work, struct async_cow, work);
1151 root = async_cow->root;
1152 fs_info = root->fs_info;
1153 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1157 * atomic_sub_return implies a barrier for waitqueue_active
1159 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1161 waitqueue_active(&fs_info->async_submit_wait))
1162 wake_up(&fs_info->async_submit_wait);
1164 if (async_cow->inode)
1165 submit_compressed_extents(async_cow->inode, async_cow);
1168 static noinline void async_cow_free(struct btrfs_work *work)
1170 struct async_cow *async_cow;
1171 async_cow = container_of(work, struct async_cow, work);
1172 if (async_cow->inode)
1173 btrfs_add_delayed_iput(async_cow->inode);
1177 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1178 u64 start, u64 end, int *page_started,
1179 unsigned long *nr_written)
1181 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1182 struct async_cow *async_cow;
1183 struct btrfs_root *root = BTRFS_I(inode)->root;
1184 unsigned long nr_pages;
1187 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1188 1, 0, NULL, GFP_NOFS);
1189 while (start < end) {
1190 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1191 BUG_ON(!async_cow); /* -ENOMEM */
1192 async_cow->inode = igrab(inode);
1193 async_cow->root = root;
1194 async_cow->locked_page = locked_page;
1195 async_cow->start = start;
1197 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1198 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1201 cur_end = min(end, start + SZ_512K - 1);
1203 async_cow->end = cur_end;
1204 INIT_LIST_HEAD(&async_cow->extents);
1206 btrfs_init_work(&async_cow->work,
1207 btrfs_delalloc_helper,
1208 async_cow_start, async_cow_submit,
1211 nr_pages = (cur_end - start + PAGE_SIZE) >>
1213 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1215 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1217 while (atomic_read(&fs_info->async_submit_draining) &&
1218 atomic_read(&fs_info->async_delalloc_pages)) {
1219 wait_event(fs_info->async_submit_wait,
1220 (atomic_read(&fs_info->async_delalloc_pages) ==
1224 *nr_written += nr_pages;
1225 start = cur_end + 1;
1231 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1232 u64 bytenr, u64 num_bytes)
1235 struct btrfs_ordered_sum *sums;
1238 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1239 bytenr + num_bytes - 1, &list, 0);
1240 if (ret == 0 && list_empty(&list))
1243 while (!list_empty(&list)) {
1244 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1245 list_del(&sums->list);
1252 * when nowcow writeback call back. This checks for snapshots or COW copies
1253 * of the extents that exist in the file, and COWs the file as required.
1255 * If no cow copies or snapshots exist, we write directly to the existing
1258 static noinline int run_delalloc_nocow(struct inode *inode,
1259 struct page *locked_page,
1260 u64 start, u64 end, int *page_started, int force,
1261 unsigned long *nr_written)
1263 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1264 struct btrfs_root *root = BTRFS_I(inode)->root;
1265 struct extent_buffer *leaf;
1266 struct btrfs_path *path;
1267 struct btrfs_file_extent_item *fi;
1268 struct btrfs_key found_key;
1269 struct extent_map *em;
1284 u64 ino = btrfs_ino(BTRFS_I(inode));
1286 path = btrfs_alloc_path();
1288 extent_clear_unlock_delalloc(inode, start, end, end,
1290 EXTENT_LOCKED | EXTENT_DELALLOC |
1291 EXTENT_DO_ACCOUNTING |
1292 EXTENT_DEFRAG, PAGE_UNLOCK |
1294 PAGE_SET_WRITEBACK |
1295 PAGE_END_WRITEBACK);
1299 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1301 cow_start = (u64)-1;
1304 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1308 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1309 leaf = path->nodes[0];
1310 btrfs_item_key_to_cpu(leaf, &found_key,
1311 path->slots[0] - 1);
1312 if (found_key.objectid == ino &&
1313 found_key.type == BTRFS_EXTENT_DATA_KEY)
1318 leaf = path->nodes[0];
1319 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1320 ret = btrfs_next_leaf(root, path);
1325 leaf = path->nodes[0];
1331 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1333 if (found_key.objectid > ino)
1335 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1336 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1340 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1341 found_key.offset > end)
1344 if (found_key.offset > cur_offset) {
1345 extent_end = found_key.offset;
1350 fi = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_file_extent_item);
1352 extent_type = btrfs_file_extent_type(leaf, fi);
1354 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1355 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1356 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1357 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1358 extent_offset = btrfs_file_extent_offset(leaf, fi);
1359 extent_end = found_key.offset +
1360 btrfs_file_extent_num_bytes(leaf, fi);
1362 btrfs_file_extent_disk_num_bytes(leaf, fi);
1363 if (extent_end <= start) {
1367 if (disk_bytenr == 0)
1369 if (btrfs_file_extent_compression(leaf, fi) ||
1370 btrfs_file_extent_encryption(leaf, fi) ||
1371 btrfs_file_extent_other_encoding(leaf, fi))
1373 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1375 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1377 if (btrfs_cross_ref_exist(root, ino,
1379 extent_offset, disk_bytenr))
1381 disk_bytenr += extent_offset;
1382 disk_bytenr += cur_offset - found_key.offset;
1383 num_bytes = min(end + 1, extent_end) - cur_offset;
1385 * if there are pending snapshots for this root,
1386 * we fall into common COW way.
1389 err = btrfs_start_write_no_snapshotting(root);
1394 * force cow if csum exists in the range.
1395 * this ensure that csum for a given extent are
1396 * either valid or do not exist.
1398 if (csum_exist_in_range(fs_info, disk_bytenr,
1401 btrfs_end_write_no_snapshotting(root);
1404 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1406 btrfs_end_write_no_snapshotting(root);
1410 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1411 extent_end = found_key.offset +
1412 btrfs_file_extent_inline_len(leaf,
1413 path->slots[0], fi);
1414 extent_end = ALIGN(extent_end,
1415 fs_info->sectorsize);
1420 if (extent_end <= start) {
1422 if (!nolock && nocow)
1423 btrfs_end_write_no_snapshotting(root);
1425 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1429 if (cow_start == (u64)-1)
1430 cow_start = cur_offset;
1431 cur_offset = extent_end;
1432 if (cur_offset > end)
1438 btrfs_release_path(path);
1439 if (cow_start != (u64)-1) {
1440 ret = cow_file_range(inode, locked_page,
1441 cow_start, found_key.offset - 1,
1442 end, page_started, nr_written, 1,
1445 if (!nolock && nocow)
1446 btrfs_end_write_no_snapshotting(root);
1448 btrfs_dec_nocow_writers(fs_info,
1452 cow_start = (u64)-1;
1455 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1456 u64 orig_start = found_key.offset - extent_offset;
1458 em = create_io_em(inode, cur_offset, num_bytes,
1460 disk_bytenr, /* block_start */
1461 num_bytes, /* block_len */
1462 disk_num_bytes, /* orig_block_len */
1463 ram_bytes, BTRFS_COMPRESS_NONE,
1464 BTRFS_ORDERED_PREALLOC);
1466 if (!nolock && nocow)
1467 btrfs_end_write_no_snapshotting(root);
1469 btrfs_dec_nocow_writers(fs_info,
1474 free_extent_map(em);
1477 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1478 type = BTRFS_ORDERED_PREALLOC;
1480 type = BTRFS_ORDERED_NOCOW;
1483 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1484 num_bytes, num_bytes, type);
1486 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1487 BUG_ON(ret); /* -ENOMEM */
1489 if (root->root_key.objectid ==
1490 BTRFS_DATA_RELOC_TREE_OBJECTID)
1492 * Error handled later, as we must prevent
1493 * extent_clear_unlock_delalloc() in error handler
1494 * from freeing metadata of created ordered extent.
1496 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1499 extent_clear_unlock_delalloc(inode, cur_offset,
1500 cur_offset + num_bytes - 1, end,
1501 locked_page, EXTENT_LOCKED |
1503 EXTENT_CLEAR_DATA_RESV,
1504 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1506 if (!nolock && nocow)
1507 btrfs_end_write_no_snapshotting(root);
1508 cur_offset = extent_end;
1511 * btrfs_reloc_clone_csums() error, now we're OK to call error
1512 * handler, as metadata for created ordered extent will only
1513 * be freed by btrfs_finish_ordered_io().
1517 if (cur_offset > end)
1520 btrfs_release_path(path);
1522 if (cur_offset <= end && cow_start == (u64)-1) {
1523 cow_start = cur_offset;
1527 if (cow_start != (u64)-1) {
1528 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1529 page_started, nr_written, 1, NULL);
1535 if (ret && cur_offset < end)
1536 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1537 locked_page, EXTENT_LOCKED |
1538 EXTENT_DELALLOC | EXTENT_DEFRAG |
1539 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1541 PAGE_SET_WRITEBACK |
1542 PAGE_END_WRITEBACK);
1543 btrfs_free_path(path);
1547 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1550 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1551 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1555 * @defrag_bytes is a hint value, no spinlock held here,
1556 * if is not zero, it means the file is defragging.
1557 * Force cow if given extent needs to be defragged.
1559 if (BTRFS_I(inode)->defrag_bytes &&
1560 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1561 EXTENT_DEFRAG, 0, NULL))
1568 * extent_io.c call back to do delayed allocation processing
1570 static int run_delalloc_range(void *private_data, struct page *locked_page,
1571 u64 start, u64 end, int *page_started,
1572 unsigned long *nr_written)
1574 struct inode *inode = private_data;
1576 int force_cow = need_force_cow(inode, start, end);
1578 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1579 ret = run_delalloc_nocow(inode, locked_page, start, end,
1580 page_started, 1, nr_written);
1581 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1582 ret = run_delalloc_nocow(inode, locked_page, start, end,
1583 page_started, 0, nr_written);
1584 } else if (!inode_need_compress(inode, start, end)) {
1585 ret = cow_file_range(inode, locked_page, start, end, end,
1586 page_started, nr_written, 1, NULL);
1588 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1589 &BTRFS_I(inode)->runtime_flags);
1590 ret = cow_file_range_async(inode, locked_page, start, end,
1591 page_started, nr_written);
1594 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1598 static void btrfs_split_extent_hook(void *private_data,
1599 struct extent_state *orig, u64 split)
1601 struct inode *inode = private_data;
1604 /* not delalloc, ignore it */
1605 if (!(orig->state & EXTENT_DELALLOC))
1608 size = orig->end - orig->start + 1;
1609 if (size > BTRFS_MAX_EXTENT_SIZE) {
1614 * See the explanation in btrfs_merge_extent_hook, the same
1615 * applies here, just in reverse.
1617 new_size = orig->end - split + 1;
1618 num_extents = count_max_extents(new_size);
1619 new_size = split - orig->start;
1620 num_extents += count_max_extents(new_size);
1621 if (count_max_extents(size) >= num_extents)
1625 spin_lock(&BTRFS_I(inode)->lock);
1626 BTRFS_I(inode)->outstanding_extents++;
1627 spin_unlock(&BTRFS_I(inode)->lock);
1631 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1632 * extents so we can keep track of new extents that are just merged onto old
1633 * extents, such as when we are doing sequential writes, so we can properly
1634 * account for the metadata space we'll need.
1636 static void btrfs_merge_extent_hook(void *private_data,
1637 struct extent_state *new,
1638 struct extent_state *other)
1640 struct inode *inode = private_data;
1641 u64 new_size, old_size;
1644 /* not delalloc, ignore it */
1645 if (!(other->state & EXTENT_DELALLOC))
1648 if (new->start > other->start)
1649 new_size = new->end - other->start + 1;
1651 new_size = other->end - new->start + 1;
1653 /* we're not bigger than the max, unreserve the space and go */
1654 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1655 spin_lock(&BTRFS_I(inode)->lock);
1656 BTRFS_I(inode)->outstanding_extents--;
1657 spin_unlock(&BTRFS_I(inode)->lock);
1662 * We have to add up either side to figure out how many extents were
1663 * accounted for before we merged into one big extent. If the number of
1664 * extents we accounted for is <= the amount we need for the new range
1665 * then we can return, otherwise drop. Think of it like this
1669 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1670 * need 2 outstanding extents, on one side we have 1 and the other side
1671 * we have 1 so they are == and we can return. But in this case
1673 * [MAX_SIZE+4k][MAX_SIZE+4k]
1675 * Each range on their own accounts for 2 extents, but merged together
1676 * they are only 3 extents worth of accounting, so we need to drop in
1679 old_size = other->end - other->start + 1;
1680 num_extents = count_max_extents(old_size);
1681 old_size = new->end - new->start + 1;
1682 num_extents += count_max_extents(old_size);
1683 if (count_max_extents(new_size) >= num_extents)
1686 spin_lock(&BTRFS_I(inode)->lock);
1687 BTRFS_I(inode)->outstanding_extents--;
1688 spin_unlock(&BTRFS_I(inode)->lock);
1691 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1692 struct inode *inode)
1694 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1696 spin_lock(&root->delalloc_lock);
1697 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1698 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1699 &root->delalloc_inodes);
1700 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1701 &BTRFS_I(inode)->runtime_flags);
1702 root->nr_delalloc_inodes++;
1703 if (root->nr_delalloc_inodes == 1) {
1704 spin_lock(&fs_info->delalloc_root_lock);
1705 BUG_ON(!list_empty(&root->delalloc_root));
1706 list_add_tail(&root->delalloc_root,
1707 &fs_info->delalloc_roots);
1708 spin_unlock(&fs_info->delalloc_root_lock);
1711 spin_unlock(&root->delalloc_lock);
1714 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1715 struct btrfs_inode *inode)
1717 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1719 spin_lock(&root->delalloc_lock);
1720 if (!list_empty(&inode->delalloc_inodes)) {
1721 list_del_init(&inode->delalloc_inodes);
1722 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1723 &inode->runtime_flags);
1724 root->nr_delalloc_inodes--;
1725 if (!root->nr_delalloc_inodes) {
1726 spin_lock(&fs_info->delalloc_root_lock);
1727 BUG_ON(list_empty(&root->delalloc_root));
1728 list_del_init(&root->delalloc_root);
1729 spin_unlock(&fs_info->delalloc_root_lock);
1732 spin_unlock(&root->delalloc_lock);
1736 * extent_io.c set_bit_hook, used to track delayed allocation
1737 * bytes in this file, and to maintain the list of inodes that
1738 * have pending delalloc work to be done.
1740 static void btrfs_set_bit_hook(void *private_data,
1741 struct extent_state *state, unsigned *bits)
1743 struct inode *inode = private_data;
1745 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1747 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1750 * set_bit and clear bit hooks normally require _irqsave/restore
1751 * but in this case, we are only testing for the DELALLOC
1752 * bit, which is only set or cleared with irqs on
1754 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1755 struct btrfs_root *root = BTRFS_I(inode)->root;
1756 u64 len = state->end + 1 - state->start;
1757 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1759 if (*bits & EXTENT_FIRST_DELALLOC) {
1760 *bits &= ~EXTENT_FIRST_DELALLOC;
1762 spin_lock(&BTRFS_I(inode)->lock);
1763 BTRFS_I(inode)->outstanding_extents++;
1764 spin_unlock(&BTRFS_I(inode)->lock);
1767 /* For sanity tests */
1768 if (btrfs_is_testing(fs_info))
1771 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1772 fs_info->delalloc_batch);
1773 spin_lock(&BTRFS_I(inode)->lock);
1774 BTRFS_I(inode)->delalloc_bytes += len;
1775 if (*bits & EXTENT_DEFRAG)
1776 BTRFS_I(inode)->defrag_bytes += len;
1777 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1778 &BTRFS_I(inode)->runtime_flags))
1779 btrfs_add_delalloc_inodes(root, inode);
1780 spin_unlock(&BTRFS_I(inode)->lock);
1783 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1784 (*bits & EXTENT_DELALLOC_NEW)) {
1785 spin_lock(&BTRFS_I(inode)->lock);
1786 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1788 spin_unlock(&BTRFS_I(inode)->lock);
1793 * extent_io.c clear_bit_hook, see set_bit_hook for why
1795 static void btrfs_clear_bit_hook(void *private_data,
1796 struct extent_state *state,
1799 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1800 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1801 u64 len = state->end + 1 - state->start;
1802 u32 num_extents = count_max_extents(len);
1804 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1805 spin_lock(&inode->lock);
1806 inode->defrag_bytes -= len;
1807 spin_unlock(&inode->lock);
1811 * set_bit and clear bit hooks normally require _irqsave/restore
1812 * but in this case, we are only testing for the DELALLOC
1813 * bit, which is only set or cleared with irqs on
1815 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1816 struct btrfs_root *root = inode->root;
1817 bool do_list = !btrfs_is_free_space_inode(inode);
1819 if (*bits & EXTENT_FIRST_DELALLOC) {
1820 *bits &= ~EXTENT_FIRST_DELALLOC;
1821 } else if (!(*bits & EXTENT_CLEAR_META_RESV)) {
1822 spin_lock(&inode->lock);
1823 inode->outstanding_extents -= num_extents;
1824 spin_unlock(&inode->lock);
1828 * We don't reserve metadata space for space cache inodes so we
1829 * don't need to call dellalloc_release_metadata if there is an
1832 if (*bits & EXTENT_CLEAR_META_RESV &&
1833 root != fs_info->tree_root)
1834 btrfs_delalloc_release_metadata(inode, len);
1836 /* For sanity tests. */
1837 if (btrfs_is_testing(fs_info))
1840 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1841 do_list && !(state->state & EXTENT_NORESERVE) &&
1842 (*bits & EXTENT_CLEAR_DATA_RESV))
1843 btrfs_free_reserved_data_space_noquota(
1847 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1848 fs_info->delalloc_batch);
1849 spin_lock(&inode->lock);
1850 inode->delalloc_bytes -= len;
1851 if (do_list && inode->delalloc_bytes == 0 &&
1852 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1853 &inode->runtime_flags))
1854 btrfs_del_delalloc_inode(root, inode);
1855 spin_unlock(&inode->lock);
1858 if ((state->state & EXTENT_DELALLOC_NEW) &&
1859 (*bits & EXTENT_DELALLOC_NEW)) {
1860 spin_lock(&inode->lock);
1861 ASSERT(inode->new_delalloc_bytes >= len);
1862 inode->new_delalloc_bytes -= len;
1863 spin_unlock(&inode->lock);
1868 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1869 * we don't create bios that span stripes or chunks
1871 * return 1 if page cannot be merged to bio
1872 * return 0 if page can be merged to bio
1873 * return error otherwise
1875 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1876 size_t size, struct bio *bio,
1877 unsigned long bio_flags)
1879 struct inode *inode = page->mapping->host;
1880 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1881 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1886 if (bio_flags & EXTENT_BIO_COMPRESSED)
1889 length = bio->bi_iter.bi_size;
1890 map_length = length;
1891 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1895 if (map_length < length + size)
1901 * in order to insert checksums into the metadata in large chunks,
1902 * we wait until bio submission time. All the pages in the bio are
1903 * checksummed and sums are attached onto the ordered extent record.
1905 * At IO completion time the cums attached on the ordered extent record
1906 * are inserted into the btree
1908 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1909 int mirror_num, unsigned long bio_flags,
1912 struct inode *inode = private_data;
1913 blk_status_t ret = 0;
1915 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1916 BUG_ON(ret); /* -ENOMEM */
1921 * in order to insert checksums into the metadata in large chunks,
1922 * we wait until bio submission time. All the pages in the bio are
1923 * checksummed and sums are attached onto the ordered extent record.
1925 * At IO completion time the cums attached on the ordered extent record
1926 * are inserted into the btree
1928 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1929 int mirror_num, unsigned long bio_flags,
1932 struct inode *inode = private_data;
1933 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1936 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1938 bio->bi_status = ret;
1945 * extent_io.c submission hook. This does the right thing for csum calculation
1946 * on write, or reading the csums from the tree before a read
1948 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1949 int mirror_num, unsigned long bio_flags,
1952 struct inode *inode = private_data;
1953 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1954 struct btrfs_root *root = BTRFS_I(inode)->root;
1955 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1956 blk_status_t ret = 0;
1958 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1960 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1962 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1963 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1965 if (bio_op(bio) != REQ_OP_WRITE) {
1966 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1970 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1971 ret = btrfs_submit_compressed_read(inode, bio,
1975 } else if (!skip_sum) {
1976 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1981 } else if (async && !skip_sum) {
1982 /* csum items have already been cloned */
1983 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1985 /* we're doing a write, do the async checksumming */
1986 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1988 __btrfs_submit_bio_start,
1989 __btrfs_submit_bio_done);
1991 } else if (!skip_sum) {
1992 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1998 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2002 bio->bi_status = ret;
2009 * given a list of ordered sums record them in the inode. This happens
2010 * at IO completion time based on sums calculated at bio submission time.
2012 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2013 struct inode *inode, struct list_head *list)
2015 struct btrfs_ordered_sum *sum;
2017 list_for_each_entry(sum, list, list) {
2018 trans->adding_csums = 1;
2019 btrfs_csum_file_blocks(trans,
2020 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2021 trans->adding_csums = 0;
2026 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2027 struct extent_state **cached_state, int dedupe)
2029 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2030 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2034 /* see btrfs_writepage_start_hook for details on why this is required */
2035 struct btrfs_writepage_fixup {
2037 struct btrfs_work work;
2040 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2042 struct btrfs_writepage_fixup *fixup;
2043 struct btrfs_ordered_extent *ordered;
2044 struct extent_state *cached_state = NULL;
2045 struct extent_changeset *data_reserved = NULL;
2047 struct inode *inode;
2052 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2056 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2057 ClearPageChecked(page);
2061 inode = page->mapping->host;
2062 page_start = page_offset(page);
2063 page_end = page_offset(page) + PAGE_SIZE - 1;
2065 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2068 /* already ordered? We're done */
2069 if (PagePrivate2(page))
2072 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2075 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2076 page_end, &cached_state, GFP_NOFS);
2078 btrfs_start_ordered_extent(inode, ordered, 1);
2079 btrfs_put_ordered_extent(ordered);
2083 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2086 mapping_set_error(page->mapping, ret);
2087 end_extent_writepage(page, ret, page_start, page_end);
2088 ClearPageChecked(page);
2092 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2094 ClearPageChecked(page);
2095 set_page_dirty(page);
2097 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2098 &cached_state, GFP_NOFS);
2103 extent_changeset_free(data_reserved);
2107 * There are a few paths in the higher layers of the kernel that directly
2108 * set the page dirty bit without asking the filesystem if it is a
2109 * good idea. This causes problems because we want to make sure COW
2110 * properly happens and the data=ordered rules are followed.
2112 * In our case any range that doesn't have the ORDERED bit set
2113 * hasn't been properly setup for IO. We kick off an async process
2114 * to fix it up. The async helper will wait for ordered extents, set
2115 * the delalloc bit and make it safe to write the page.
2117 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2119 struct inode *inode = page->mapping->host;
2120 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2121 struct btrfs_writepage_fixup *fixup;
2123 /* this page is properly in the ordered list */
2124 if (TestClearPagePrivate2(page))
2127 if (PageChecked(page))
2130 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2134 SetPageChecked(page);
2136 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2137 btrfs_writepage_fixup_worker, NULL, NULL);
2139 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2143 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2144 struct inode *inode, u64 file_pos,
2145 u64 disk_bytenr, u64 disk_num_bytes,
2146 u64 num_bytes, u64 ram_bytes,
2147 u8 compression, u8 encryption,
2148 u16 other_encoding, int extent_type)
2150 struct btrfs_root *root = BTRFS_I(inode)->root;
2151 struct btrfs_file_extent_item *fi;
2152 struct btrfs_path *path;
2153 struct extent_buffer *leaf;
2154 struct btrfs_key ins;
2156 int extent_inserted = 0;
2159 path = btrfs_alloc_path();
2164 * we may be replacing one extent in the tree with another.
2165 * The new extent is pinned in the extent map, and we don't want
2166 * to drop it from the cache until it is completely in the btree.
2168 * So, tell btrfs_drop_extents to leave this extent in the cache.
2169 * the caller is expected to unpin it and allow it to be merged
2172 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2173 file_pos + num_bytes, NULL, 0,
2174 1, sizeof(*fi), &extent_inserted);
2178 if (!extent_inserted) {
2179 ins.objectid = btrfs_ino(BTRFS_I(inode));
2180 ins.offset = file_pos;
2181 ins.type = BTRFS_EXTENT_DATA_KEY;
2183 path->leave_spinning = 1;
2184 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2189 leaf = path->nodes[0];
2190 fi = btrfs_item_ptr(leaf, path->slots[0],
2191 struct btrfs_file_extent_item);
2192 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2193 btrfs_set_file_extent_type(leaf, fi, extent_type);
2194 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2195 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2196 btrfs_set_file_extent_offset(leaf, fi, 0);
2197 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2198 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2199 btrfs_set_file_extent_compression(leaf, fi, compression);
2200 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2201 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2203 btrfs_mark_buffer_dirty(leaf);
2204 btrfs_release_path(path);
2206 inode_add_bytes(inode, num_bytes);
2208 ins.objectid = disk_bytenr;
2209 ins.offset = disk_num_bytes;
2210 ins.type = BTRFS_EXTENT_ITEM_KEY;
2213 * Release the reserved range from inode dirty range map, as it is
2214 * already moved into delayed_ref_head
2216 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2220 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2221 btrfs_ino(BTRFS_I(inode)), file_pos, qg_released, &ins);
2223 btrfs_free_path(path);
2228 /* snapshot-aware defrag */
2229 struct sa_defrag_extent_backref {
2230 struct rb_node node;
2231 struct old_sa_defrag_extent *old;
2240 struct old_sa_defrag_extent {
2241 struct list_head list;
2242 struct new_sa_defrag_extent *new;
2251 struct new_sa_defrag_extent {
2252 struct rb_root root;
2253 struct list_head head;
2254 struct btrfs_path *path;
2255 struct inode *inode;
2263 static int backref_comp(struct sa_defrag_extent_backref *b1,
2264 struct sa_defrag_extent_backref *b2)
2266 if (b1->root_id < b2->root_id)
2268 else if (b1->root_id > b2->root_id)
2271 if (b1->inum < b2->inum)
2273 else if (b1->inum > b2->inum)
2276 if (b1->file_pos < b2->file_pos)
2278 else if (b1->file_pos > b2->file_pos)
2282 * [------------------------------] ===> (a range of space)
2283 * |<--->| |<---->| =============> (fs/file tree A)
2284 * |<---------------------------->| ===> (fs/file tree B)
2286 * A range of space can refer to two file extents in one tree while
2287 * refer to only one file extent in another tree.
2289 * So we may process a disk offset more than one time(two extents in A)
2290 * and locate at the same extent(one extent in B), then insert two same
2291 * backrefs(both refer to the extent in B).
2296 static void backref_insert(struct rb_root *root,
2297 struct sa_defrag_extent_backref *backref)
2299 struct rb_node **p = &root->rb_node;
2300 struct rb_node *parent = NULL;
2301 struct sa_defrag_extent_backref *entry;
2306 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2308 ret = backref_comp(backref, entry);
2312 p = &(*p)->rb_right;
2315 rb_link_node(&backref->node, parent, p);
2316 rb_insert_color(&backref->node, root);
2320 * Note the backref might has changed, and in this case we just return 0.
2322 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2325 struct btrfs_file_extent_item *extent;
2326 struct old_sa_defrag_extent *old = ctx;
2327 struct new_sa_defrag_extent *new = old->new;
2328 struct btrfs_path *path = new->path;
2329 struct btrfs_key key;
2330 struct btrfs_root *root;
2331 struct sa_defrag_extent_backref *backref;
2332 struct extent_buffer *leaf;
2333 struct inode *inode = new->inode;
2334 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2340 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2341 inum == btrfs_ino(BTRFS_I(inode)))
2344 key.objectid = root_id;
2345 key.type = BTRFS_ROOT_ITEM_KEY;
2346 key.offset = (u64)-1;
2348 root = btrfs_read_fs_root_no_name(fs_info, &key);
2350 if (PTR_ERR(root) == -ENOENT)
2353 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2354 inum, offset, root_id);
2355 return PTR_ERR(root);
2358 key.objectid = inum;
2359 key.type = BTRFS_EXTENT_DATA_KEY;
2360 if (offset > (u64)-1 << 32)
2363 key.offset = offset;
2365 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2366 if (WARN_ON(ret < 0))
2373 leaf = path->nodes[0];
2374 slot = path->slots[0];
2376 if (slot >= btrfs_header_nritems(leaf)) {
2377 ret = btrfs_next_leaf(root, path);
2380 } else if (ret > 0) {
2389 btrfs_item_key_to_cpu(leaf, &key, slot);
2391 if (key.objectid > inum)
2394 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2397 extent = btrfs_item_ptr(leaf, slot,
2398 struct btrfs_file_extent_item);
2400 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2404 * 'offset' refers to the exact key.offset,
2405 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2406 * (key.offset - extent_offset).
2408 if (key.offset != offset)
2411 extent_offset = btrfs_file_extent_offset(leaf, extent);
2412 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2414 if (extent_offset >= old->extent_offset + old->offset +
2415 old->len || extent_offset + num_bytes <=
2416 old->extent_offset + old->offset)
2421 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2427 backref->root_id = root_id;
2428 backref->inum = inum;
2429 backref->file_pos = offset;
2430 backref->num_bytes = num_bytes;
2431 backref->extent_offset = extent_offset;
2432 backref->generation = btrfs_file_extent_generation(leaf, extent);
2434 backref_insert(&new->root, backref);
2437 btrfs_release_path(path);
2442 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2443 struct new_sa_defrag_extent *new)
2445 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2446 struct old_sa_defrag_extent *old, *tmp;
2451 list_for_each_entry_safe(old, tmp, &new->head, list) {
2452 ret = iterate_inodes_from_logical(old->bytenr +
2453 old->extent_offset, fs_info,
2454 path, record_one_backref,
2456 if (ret < 0 && ret != -ENOENT)
2459 /* no backref to be processed for this extent */
2461 list_del(&old->list);
2466 if (list_empty(&new->head))
2472 static int relink_is_mergable(struct extent_buffer *leaf,
2473 struct btrfs_file_extent_item *fi,
2474 struct new_sa_defrag_extent *new)
2476 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2479 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2482 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2485 if (btrfs_file_extent_encryption(leaf, fi) ||
2486 btrfs_file_extent_other_encoding(leaf, fi))
2493 * Note the backref might has changed, and in this case we just return 0.
2495 static noinline int relink_extent_backref(struct btrfs_path *path,
2496 struct sa_defrag_extent_backref *prev,
2497 struct sa_defrag_extent_backref *backref)
2499 struct btrfs_file_extent_item *extent;
2500 struct btrfs_file_extent_item *item;
2501 struct btrfs_ordered_extent *ordered;
2502 struct btrfs_trans_handle *trans;
2503 struct btrfs_root *root;
2504 struct btrfs_key key;
2505 struct extent_buffer *leaf;
2506 struct old_sa_defrag_extent *old = backref->old;
2507 struct new_sa_defrag_extent *new = old->new;
2508 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2509 struct inode *inode;
2510 struct extent_state *cached = NULL;
2519 if (prev && prev->root_id == backref->root_id &&
2520 prev->inum == backref->inum &&
2521 prev->file_pos + prev->num_bytes == backref->file_pos)
2524 /* step 1: get root */
2525 key.objectid = backref->root_id;
2526 key.type = BTRFS_ROOT_ITEM_KEY;
2527 key.offset = (u64)-1;
2529 index = srcu_read_lock(&fs_info->subvol_srcu);
2531 root = btrfs_read_fs_root_no_name(fs_info, &key);
2533 srcu_read_unlock(&fs_info->subvol_srcu, index);
2534 if (PTR_ERR(root) == -ENOENT)
2536 return PTR_ERR(root);
2539 if (btrfs_root_readonly(root)) {
2540 srcu_read_unlock(&fs_info->subvol_srcu, index);
2544 /* step 2: get inode */
2545 key.objectid = backref->inum;
2546 key.type = BTRFS_INODE_ITEM_KEY;
2549 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2550 if (IS_ERR(inode)) {
2551 srcu_read_unlock(&fs_info->subvol_srcu, index);
2555 srcu_read_unlock(&fs_info->subvol_srcu, index);
2557 /* step 3: relink backref */
2558 lock_start = backref->file_pos;
2559 lock_end = backref->file_pos + backref->num_bytes - 1;
2560 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2563 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2565 btrfs_put_ordered_extent(ordered);
2569 trans = btrfs_join_transaction(root);
2570 if (IS_ERR(trans)) {
2571 ret = PTR_ERR(trans);
2575 key.objectid = backref->inum;
2576 key.type = BTRFS_EXTENT_DATA_KEY;
2577 key.offset = backref->file_pos;
2579 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2582 } else if (ret > 0) {
2587 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2588 struct btrfs_file_extent_item);
2590 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2591 backref->generation)
2594 btrfs_release_path(path);
2596 start = backref->file_pos;
2597 if (backref->extent_offset < old->extent_offset + old->offset)
2598 start += old->extent_offset + old->offset -
2599 backref->extent_offset;
2601 len = min(backref->extent_offset + backref->num_bytes,
2602 old->extent_offset + old->offset + old->len);
2603 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2605 ret = btrfs_drop_extents(trans, root, inode, start,
2610 key.objectid = btrfs_ino(BTRFS_I(inode));
2611 key.type = BTRFS_EXTENT_DATA_KEY;
2614 path->leave_spinning = 1;
2616 struct btrfs_file_extent_item *fi;
2618 struct btrfs_key found_key;
2620 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2625 leaf = path->nodes[0];
2626 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2628 fi = btrfs_item_ptr(leaf, path->slots[0],
2629 struct btrfs_file_extent_item);
2630 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2632 if (extent_len + found_key.offset == start &&
2633 relink_is_mergable(leaf, fi, new)) {
2634 btrfs_set_file_extent_num_bytes(leaf, fi,
2636 btrfs_mark_buffer_dirty(leaf);
2637 inode_add_bytes(inode, len);
2643 btrfs_release_path(path);
2648 ret = btrfs_insert_empty_item(trans, root, path, &key,
2651 btrfs_abort_transaction(trans, ret);
2655 leaf = path->nodes[0];
2656 item = btrfs_item_ptr(leaf, path->slots[0],
2657 struct btrfs_file_extent_item);
2658 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2659 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2660 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2661 btrfs_set_file_extent_num_bytes(leaf, item, len);
2662 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2663 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2664 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2665 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2666 btrfs_set_file_extent_encryption(leaf, item, 0);
2667 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2669 btrfs_mark_buffer_dirty(leaf);
2670 inode_add_bytes(inode, len);
2671 btrfs_release_path(path);
2673 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2675 backref->root_id, backref->inum,
2676 new->file_pos); /* start - extent_offset */
2678 btrfs_abort_transaction(trans, ret);
2684 btrfs_release_path(path);
2685 path->leave_spinning = 0;
2686 btrfs_end_transaction(trans);
2688 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2694 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2696 struct old_sa_defrag_extent *old, *tmp;
2701 list_for_each_entry_safe(old, tmp, &new->head, list) {
2707 static void relink_file_extents(struct new_sa_defrag_extent *new)
2709 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2710 struct btrfs_path *path;
2711 struct sa_defrag_extent_backref *backref;
2712 struct sa_defrag_extent_backref *prev = NULL;
2713 struct inode *inode;
2714 struct btrfs_root *root;
2715 struct rb_node *node;
2719 root = BTRFS_I(inode)->root;
2721 path = btrfs_alloc_path();
2725 if (!record_extent_backrefs(path, new)) {
2726 btrfs_free_path(path);
2729 btrfs_release_path(path);
2732 node = rb_first(&new->root);
2735 rb_erase(node, &new->root);
2737 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2739 ret = relink_extent_backref(path, prev, backref);
2752 btrfs_free_path(path);
2754 free_sa_defrag_extent(new);
2756 atomic_dec(&fs_info->defrag_running);
2757 wake_up(&fs_info->transaction_wait);
2760 static struct new_sa_defrag_extent *
2761 record_old_file_extents(struct inode *inode,
2762 struct btrfs_ordered_extent *ordered)
2764 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2765 struct btrfs_root *root = BTRFS_I(inode)->root;
2766 struct btrfs_path *path;
2767 struct btrfs_key key;
2768 struct old_sa_defrag_extent *old;
2769 struct new_sa_defrag_extent *new;
2772 new = kmalloc(sizeof(*new), GFP_NOFS);
2777 new->file_pos = ordered->file_offset;
2778 new->len = ordered->len;
2779 new->bytenr = ordered->start;
2780 new->disk_len = ordered->disk_len;
2781 new->compress_type = ordered->compress_type;
2782 new->root = RB_ROOT;
2783 INIT_LIST_HEAD(&new->head);
2785 path = btrfs_alloc_path();
2789 key.objectid = btrfs_ino(BTRFS_I(inode));
2790 key.type = BTRFS_EXTENT_DATA_KEY;
2791 key.offset = new->file_pos;
2793 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2796 if (ret > 0 && path->slots[0] > 0)
2799 /* find out all the old extents for the file range */
2801 struct btrfs_file_extent_item *extent;
2802 struct extent_buffer *l;
2811 slot = path->slots[0];
2813 if (slot >= btrfs_header_nritems(l)) {
2814 ret = btrfs_next_leaf(root, path);
2822 btrfs_item_key_to_cpu(l, &key, slot);
2824 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2826 if (key.type != BTRFS_EXTENT_DATA_KEY)
2828 if (key.offset >= new->file_pos + new->len)
2831 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2833 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2834 if (key.offset + num_bytes < new->file_pos)
2837 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2841 extent_offset = btrfs_file_extent_offset(l, extent);
2843 old = kmalloc(sizeof(*old), GFP_NOFS);
2847 offset = max(new->file_pos, key.offset);
2848 end = min(new->file_pos + new->len, key.offset + num_bytes);
2850 old->bytenr = disk_bytenr;
2851 old->extent_offset = extent_offset;
2852 old->offset = offset - key.offset;
2853 old->len = end - offset;
2856 list_add_tail(&old->list, &new->head);
2862 btrfs_free_path(path);
2863 atomic_inc(&fs_info->defrag_running);
2868 btrfs_free_path(path);
2870 free_sa_defrag_extent(new);
2874 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2877 struct btrfs_block_group_cache *cache;
2879 cache = btrfs_lookup_block_group(fs_info, start);
2882 spin_lock(&cache->lock);
2883 cache->delalloc_bytes -= len;
2884 spin_unlock(&cache->lock);
2886 btrfs_put_block_group(cache);
2889 /* as ordered data IO finishes, this gets called so we can finish
2890 * an ordered extent if the range of bytes in the file it covers are
2893 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2895 struct inode *inode = ordered_extent->inode;
2896 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2897 struct btrfs_root *root = BTRFS_I(inode)->root;
2898 struct btrfs_trans_handle *trans = NULL;
2899 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2900 struct extent_state *cached_state = NULL;
2901 struct new_sa_defrag_extent *new = NULL;
2902 int compress_type = 0;
2904 u64 logical_len = ordered_extent->len;
2906 bool truncated = false;
2907 bool range_locked = false;
2908 bool clear_new_delalloc_bytes = false;
2910 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2911 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2912 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2913 clear_new_delalloc_bytes = true;
2915 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2917 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2922 btrfs_free_io_failure_record(BTRFS_I(inode),
2923 ordered_extent->file_offset,
2924 ordered_extent->file_offset +
2925 ordered_extent->len - 1);
2927 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2929 logical_len = ordered_extent->truncated_len;
2930 /* Truncated the entire extent, don't bother adding */
2935 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2936 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2939 * For mwrite(mmap + memset to write) case, we still reserve
2940 * space for NOCOW range.
2941 * As NOCOW won't cause a new delayed ref, just free the space
2943 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2944 ordered_extent->len);
2945 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2947 trans = btrfs_join_transaction_nolock(root);
2949 trans = btrfs_join_transaction(root);
2950 if (IS_ERR(trans)) {
2951 ret = PTR_ERR(trans);
2955 trans->block_rsv = &fs_info->delalloc_block_rsv;
2956 ret = btrfs_update_inode_fallback(trans, root, inode);
2957 if (ret) /* -ENOMEM or corruption */
2958 btrfs_abort_transaction(trans, ret);
2962 range_locked = true;
2963 lock_extent_bits(io_tree, ordered_extent->file_offset,
2964 ordered_extent->file_offset + ordered_extent->len - 1,
2967 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2968 ordered_extent->file_offset + ordered_extent->len - 1,
2969 EXTENT_DEFRAG, 0, cached_state);
2971 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2972 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2973 /* the inode is shared */
2974 new = record_old_file_extents(inode, ordered_extent);
2976 clear_extent_bit(io_tree, ordered_extent->file_offset,
2977 ordered_extent->file_offset + ordered_extent->len - 1,
2978 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2982 trans = btrfs_join_transaction_nolock(root);
2984 trans = btrfs_join_transaction(root);
2985 if (IS_ERR(trans)) {
2986 ret = PTR_ERR(trans);
2991 trans->block_rsv = &fs_info->delalloc_block_rsv;
2993 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2994 compress_type = ordered_extent->compress_type;
2995 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2996 BUG_ON(compress_type);
2997 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2998 ordered_extent->file_offset,
2999 ordered_extent->file_offset +
3002 BUG_ON(root == fs_info->tree_root);
3003 ret = insert_reserved_file_extent(trans, inode,
3004 ordered_extent->file_offset,
3005 ordered_extent->start,
3006 ordered_extent->disk_len,
3007 logical_len, logical_len,
3008 compress_type, 0, 0,
3009 BTRFS_FILE_EXTENT_REG);
3011 btrfs_release_delalloc_bytes(fs_info,
3012 ordered_extent->start,
3013 ordered_extent->disk_len);
3015 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3016 ordered_extent->file_offset, ordered_extent->len,
3019 btrfs_abort_transaction(trans, ret);
3023 add_pending_csums(trans, inode, &ordered_extent->list);
3025 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3026 ret = btrfs_update_inode_fallback(trans, root, inode);
3027 if (ret) { /* -ENOMEM or corruption */
3028 btrfs_abort_transaction(trans, ret);
3033 if (range_locked || clear_new_delalloc_bytes) {
3034 unsigned int clear_bits = 0;
3037 clear_bits |= EXTENT_LOCKED;
3038 if (clear_new_delalloc_bytes)
3039 clear_bits |= EXTENT_DELALLOC_NEW;
3040 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3041 ordered_extent->file_offset,
3042 ordered_extent->file_offset +
3043 ordered_extent->len - 1,
3045 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3046 0, &cached_state, GFP_NOFS);
3049 if (root != fs_info->tree_root)
3050 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3051 ordered_extent->len);
3053 btrfs_end_transaction(trans);
3055 if (ret || truncated) {
3059 start = ordered_extent->file_offset + logical_len;
3061 start = ordered_extent->file_offset;
3062 end = ordered_extent->file_offset + ordered_extent->len - 1;
3063 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3065 /* Drop the cache for the part of the extent we didn't write. */
3066 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3069 * If the ordered extent had an IOERR or something else went
3070 * wrong we need to return the space for this ordered extent
3071 * back to the allocator. We only free the extent in the
3072 * truncated case if we didn't write out the extent at all.
3074 if ((ret || !logical_len) &&
3075 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3076 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3077 btrfs_free_reserved_extent(fs_info,
3078 ordered_extent->start,
3079 ordered_extent->disk_len, 1);
3084 * This needs to be done to make sure anybody waiting knows we are done
3085 * updating everything for this ordered extent.
3087 btrfs_remove_ordered_extent(inode, ordered_extent);
3089 /* for snapshot-aware defrag */
3092 free_sa_defrag_extent(new);
3093 atomic_dec(&fs_info->defrag_running);
3095 relink_file_extents(new);
3100 btrfs_put_ordered_extent(ordered_extent);
3101 /* once for the tree */
3102 btrfs_put_ordered_extent(ordered_extent);
3107 static void finish_ordered_fn(struct btrfs_work *work)
3109 struct btrfs_ordered_extent *ordered_extent;
3110 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3111 btrfs_finish_ordered_io(ordered_extent);
3114 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3115 struct extent_state *state, int uptodate)
3117 struct inode *inode = page->mapping->host;
3118 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3119 struct btrfs_ordered_extent *ordered_extent = NULL;
3120 struct btrfs_workqueue *wq;
3121 btrfs_work_func_t func;
3123 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3125 ClearPagePrivate2(page);
3126 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3127 end - start + 1, uptodate))
3130 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3131 wq = fs_info->endio_freespace_worker;
3132 func = btrfs_freespace_write_helper;
3134 wq = fs_info->endio_write_workers;
3135 func = btrfs_endio_write_helper;
3138 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3140 btrfs_queue_work(wq, &ordered_extent->work);
3143 static int __readpage_endio_check(struct inode *inode,
3144 struct btrfs_io_bio *io_bio,
3145 int icsum, struct page *page,
3146 int pgoff, u64 start, size_t len)
3152 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3154 kaddr = kmap_atomic(page);
3155 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3156 btrfs_csum_final(csum, (u8 *)&csum);
3157 if (csum != csum_expected)
3160 kunmap_atomic(kaddr);
3163 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3164 io_bio->mirror_num);
3165 memset(kaddr + pgoff, 1, len);
3166 flush_dcache_page(page);
3167 kunmap_atomic(kaddr);
3172 * when reads are done, we need to check csums to verify the data is correct
3173 * if there's a match, we allow the bio to finish. If not, the code in
3174 * extent_io.c will try to find good copies for us.
3176 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3177 u64 phy_offset, struct page *page,
3178 u64 start, u64 end, int mirror)
3180 size_t offset = start - page_offset(page);
3181 struct inode *inode = page->mapping->host;
3182 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3183 struct btrfs_root *root = BTRFS_I(inode)->root;
3185 if (PageChecked(page)) {
3186 ClearPageChecked(page);
3190 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3193 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3194 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3195 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3199 phy_offset >>= inode->i_sb->s_blocksize_bits;
3200 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3201 start, (size_t)(end - start + 1));
3204 void btrfs_add_delayed_iput(struct inode *inode)
3206 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3207 struct btrfs_inode *binode = BTRFS_I(inode);
3209 if (atomic_add_unless(&inode->i_count, -1, 1))
3212 spin_lock(&fs_info->delayed_iput_lock);
3213 if (binode->delayed_iput_count == 0) {
3214 ASSERT(list_empty(&binode->delayed_iput));
3215 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3217 binode->delayed_iput_count++;
3219 spin_unlock(&fs_info->delayed_iput_lock);
3222 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3225 spin_lock(&fs_info->delayed_iput_lock);
3226 while (!list_empty(&fs_info->delayed_iputs)) {
3227 struct btrfs_inode *inode;
3229 inode = list_first_entry(&fs_info->delayed_iputs,
3230 struct btrfs_inode, delayed_iput);
3231 if (inode->delayed_iput_count) {
3232 inode->delayed_iput_count--;
3233 list_move_tail(&inode->delayed_iput,
3234 &fs_info->delayed_iputs);
3236 list_del_init(&inode->delayed_iput);
3238 spin_unlock(&fs_info->delayed_iput_lock);
3239 iput(&inode->vfs_inode);
3240 spin_lock(&fs_info->delayed_iput_lock);
3242 spin_unlock(&fs_info->delayed_iput_lock);
3246 * This is called in transaction commit time. If there are no orphan
3247 * files in the subvolume, it removes orphan item and frees block_rsv
3250 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3251 struct btrfs_root *root)
3253 struct btrfs_fs_info *fs_info = root->fs_info;
3254 struct btrfs_block_rsv *block_rsv;
3257 if (atomic_read(&root->orphan_inodes) ||
3258 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3261 spin_lock(&root->orphan_lock);
3262 if (atomic_read(&root->orphan_inodes)) {
3263 spin_unlock(&root->orphan_lock);
3267 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3268 spin_unlock(&root->orphan_lock);
3272 block_rsv = root->orphan_block_rsv;
3273 root->orphan_block_rsv = NULL;
3274 spin_unlock(&root->orphan_lock);
3276 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3277 btrfs_root_refs(&root->root_item) > 0) {
3278 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3279 root->root_key.objectid);
3281 btrfs_abort_transaction(trans, ret);
3283 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3288 WARN_ON(block_rsv->size > 0);
3289 btrfs_free_block_rsv(fs_info, block_rsv);
3294 * This creates an orphan entry for the given inode in case something goes
3295 * wrong in the middle of an unlink/truncate.
3297 * NOTE: caller of this function should reserve 5 units of metadata for
3300 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3301 struct btrfs_inode *inode)
3303 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3304 struct btrfs_root *root = inode->root;
3305 struct btrfs_block_rsv *block_rsv = NULL;
3310 if (!root->orphan_block_rsv) {
3311 block_rsv = btrfs_alloc_block_rsv(fs_info,
3312 BTRFS_BLOCK_RSV_TEMP);
3317 spin_lock(&root->orphan_lock);
3318 if (!root->orphan_block_rsv) {
3319 root->orphan_block_rsv = block_rsv;
3320 } else if (block_rsv) {
3321 btrfs_free_block_rsv(fs_info, block_rsv);
3325 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3326 &inode->runtime_flags)) {
3329 * For proper ENOSPC handling, we should do orphan
3330 * cleanup when mounting. But this introduces backward
3331 * compatibility issue.
3333 if (!xchg(&root->orphan_item_inserted, 1))
3339 atomic_inc(&root->orphan_inodes);
3342 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3343 &inode->runtime_flags))
3345 spin_unlock(&root->orphan_lock);
3347 /* grab metadata reservation from transaction handle */
3349 ret = btrfs_orphan_reserve_metadata(trans, inode);
3352 atomic_dec(&root->orphan_inodes);
3353 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3354 &inode->runtime_flags);
3356 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3357 &inode->runtime_flags);
3362 /* insert an orphan item to track this unlinked/truncated file */
3364 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3366 atomic_dec(&root->orphan_inodes);
3368 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3369 &inode->runtime_flags);
3370 btrfs_orphan_release_metadata(inode);
3372 if (ret != -EEXIST) {
3373 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3374 &inode->runtime_flags);
3375 btrfs_abort_transaction(trans, ret);
3382 /* insert an orphan item to track subvolume contains orphan files */
3384 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3385 root->root_key.objectid);
3386 if (ret && ret != -EEXIST) {
3387 btrfs_abort_transaction(trans, ret);
3395 * We have done the truncate/delete so we can go ahead and remove the orphan
3396 * item for this particular inode.
3398 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3399 struct btrfs_inode *inode)
3401 struct btrfs_root *root = inode->root;
3402 int delete_item = 0;
3403 int release_rsv = 0;
3406 spin_lock(&root->orphan_lock);
3407 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3408 &inode->runtime_flags))
3411 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3412 &inode->runtime_flags))
3414 spin_unlock(&root->orphan_lock);
3417 atomic_dec(&root->orphan_inodes);
3419 ret = btrfs_del_orphan_item(trans, root,
3424 btrfs_orphan_release_metadata(inode);
3430 * this cleans up any orphans that may be left on the list from the last use
3433 int btrfs_orphan_cleanup(struct btrfs_root *root)
3435 struct btrfs_fs_info *fs_info = root->fs_info;
3436 struct btrfs_path *path;
3437 struct extent_buffer *leaf;
3438 struct btrfs_key key, found_key;
3439 struct btrfs_trans_handle *trans;
3440 struct inode *inode;
3441 u64 last_objectid = 0;
3442 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3444 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3447 path = btrfs_alloc_path();
3452 path->reada = READA_BACK;
3454 key.objectid = BTRFS_ORPHAN_OBJECTID;
3455 key.type = BTRFS_ORPHAN_ITEM_KEY;
3456 key.offset = (u64)-1;
3459 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3464 * if ret == 0 means we found what we were searching for, which
3465 * is weird, but possible, so only screw with path if we didn't
3466 * find the key and see if we have stuff that matches
3470 if (path->slots[0] == 0)
3475 /* pull out the item */
3476 leaf = path->nodes[0];
3477 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3479 /* make sure the item matches what we want */
3480 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3482 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3485 /* release the path since we're done with it */
3486 btrfs_release_path(path);
3489 * this is where we are basically btrfs_lookup, without the
3490 * crossing root thing. we store the inode number in the
3491 * offset of the orphan item.
3494 if (found_key.offset == last_objectid) {
3496 "Error removing orphan entry, stopping orphan cleanup");
3501 last_objectid = found_key.offset;
3503 found_key.objectid = found_key.offset;
3504 found_key.type = BTRFS_INODE_ITEM_KEY;
3505 found_key.offset = 0;
3506 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3507 ret = PTR_ERR_OR_ZERO(inode);
3508 if (ret && ret != -ENOENT)
3511 if (ret == -ENOENT && root == fs_info->tree_root) {
3512 struct btrfs_root *dead_root;
3513 struct btrfs_fs_info *fs_info = root->fs_info;
3514 int is_dead_root = 0;
3517 * this is an orphan in the tree root. Currently these
3518 * could come from 2 sources:
3519 * a) a snapshot deletion in progress
3520 * b) a free space cache inode
3521 * We need to distinguish those two, as the snapshot
3522 * orphan must not get deleted.
3523 * find_dead_roots already ran before us, so if this
3524 * is a snapshot deletion, we should find the root
3525 * in the dead_roots list
3527 spin_lock(&fs_info->trans_lock);
3528 list_for_each_entry(dead_root, &fs_info->dead_roots,
3530 if (dead_root->root_key.objectid ==
3531 found_key.objectid) {
3536 spin_unlock(&fs_info->trans_lock);
3538 /* prevent this orphan from being found again */
3539 key.offset = found_key.objectid - 1;
3544 * Inode is already gone but the orphan item is still there,
3545 * kill the orphan item.
3547 if (ret == -ENOENT) {
3548 trans = btrfs_start_transaction(root, 1);
3549 if (IS_ERR(trans)) {
3550 ret = PTR_ERR(trans);
3553 btrfs_debug(fs_info, "auto deleting %Lu",
3554 found_key.objectid);
3555 ret = btrfs_del_orphan_item(trans, root,
3556 found_key.objectid);
3557 btrfs_end_transaction(trans);
3564 * add this inode to the orphan list so btrfs_orphan_del does
3565 * the proper thing when we hit it
3567 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3568 &BTRFS_I(inode)->runtime_flags);
3569 atomic_inc(&root->orphan_inodes);
3571 /* if we have links, this was a truncate, lets do that */
3572 if (inode->i_nlink) {
3573 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3579 /* 1 for the orphan item deletion. */
3580 trans = btrfs_start_transaction(root, 1);
3581 if (IS_ERR(trans)) {
3583 ret = PTR_ERR(trans);
3586 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3587 btrfs_end_transaction(trans);
3593 ret = btrfs_truncate(inode);
3595 btrfs_orphan_del(NULL, BTRFS_I(inode));
3600 /* this will do delete_inode and everything for us */
3605 /* release the path since we're done with it */
3606 btrfs_release_path(path);
3608 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3610 if (root->orphan_block_rsv)
3611 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3614 if (root->orphan_block_rsv ||
3615 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3616 trans = btrfs_join_transaction(root);
3618 btrfs_end_transaction(trans);
3622 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3624 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3628 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3629 btrfs_free_path(path);
3634 * very simple check to peek ahead in the leaf looking for xattrs. If we
3635 * don't find any xattrs, we know there can't be any acls.
3637 * slot is the slot the inode is in, objectid is the objectid of the inode
3639 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3640 int slot, u64 objectid,
3641 int *first_xattr_slot)
3643 u32 nritems = btrfs_header_nritems(leaf);
3644 struct btrfs_key found_key;
3645 static u64 xattr_access = 0;
3646 static u64 xattr_default = 0;
3649 if (!xattr_access) {
3650 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3651 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3652 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3653 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3657 *first_xattr_slot = -1;
3658 while (slot < nritems) {
3659 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3661 /* we found a different objectid, there must not be acls */
3662 if (found_key.objectid != objectid)
3665 /* we found an xattr, assume we've got an acl */
3666 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3667 if (*first_xattr_slot == -1)
3668 *first_xattr_slot = slot;
3669 if (found_key.offset == xattr_access ||
3670 found_key.offset == xattr_default)
3675 * we found a key greater than an xattr key, there can't
3676 * be any acls later on
3678 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3685 * it goes inode, inode backrefs, xattrs, extents,
3686 * so if there are a ton of hard links to an inode there can
3687 * be a lot of backrefs. Don't waste time searching too hard,
3688 * this is just an optimization
3693 /* we hit the end of the leaf before we found an xattr or
3694 * something larger than an xattr. We have to assume the inode
3697 if (*first_xattr_slot == -1)
3698 *first_xattr_slot = slot;
3703 * read an inode from the btree into the in-memory inode
3705 static int btrfs_read_locked_inode(struct inode *inode)
3707 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3708 struct btrfs_path *path;
3709 struct extent_buffer *leaf;
3710 struct btrfs_inode_item *inode_item;
3711 struct btrfs_root *root = BTRFS_I(inode)->root;
3712 struct btrfs_key location;
3717 bool filled = false;
3718 int first_xattr_slot;
3720 ret = btrfs_fill_inode(inode, &rdev);
3724 path = btrfs_alloc_path();
3730 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3732 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3739 leaf = path->nodes[0];
3744 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3745 struct btrfs_inode_item);
3746 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3747 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3748 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3749 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3750 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3752 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3753 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3755 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3756 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3758 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3759 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3761 BTRFS_I(inode)->i_otime.tv_sec =
3762 btrfs_timespec_sec(leaf, &inode_item->otime);
3763 BTRFS_I(inode)->i_otime.tv_nsec =
3764 btrfs_timespec_nsec(leaf, &inode_item->otime);
3766 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3767 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3768 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3770 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3771 inode->i_generation = BTRFS_I(inode)->generation;
3773 rdev = btrfs_inode_rdev(leaf, inode_item);
3775 BTRFS_I(inode)->index_cnt = (u64)-1;
3776 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3780 * If we were modified in the current generation and evicted from memory
3781 * and then re-read we need to do a full sync since we don't have any
3782 * idea about which extents were modified before we were evicted from
3785 * This is required for both inode re-read from disk and delayed inode
3786 * in delayed_nodes_tree.
3788 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3789 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3790 &BTRFS_I(inode)->runtime_flags);
3793 * We don't persist the id of the transaction where an unlink operation
3794 * against the inode was last made. So here we assume the inode might
3795 * have been evicted, and therefore the exact value of last_unlink_trans
3796 * lost, and set it to last_trans to avoid metadata inconsistencies
3797 * between the inode and its parent if the inode is fsync'ed and the log
3798 * replayed. For example, in the scenario:
3801 * ln mydir/foo mydir/bar
3804 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3805 * xfs_io -c fsync mydir/foo
3807 * mount fs, triggers fsync log replay
3809 * We must make sure that when we fsync our inode foo we also log its
3810 * parent inode, otherwise after log replay the parent still has the
3811 * dentry with the "bar" name but our inode foo has a link count of 1
3812 * and doesn't have an inode ref with the name "bar" anymore.
3814 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3815 * but it guarantees correctness at the expense of occasional full
3816 * transaction commits on fsync if our inode is a directory, or if our
3817 * inode is not a directory, logging its parent unnecessarily.
3819 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3822 if (inode->i_nlink != 1 ||
3823 path->slots[0] >= btrfs_header_nritems(leaf))
3826 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3827 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3830 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3831 if (location.type == BTRFS_INODE_REF_KEY) {
3832 struct btrfs_inode_ref *ref;
3834 ref = (struct btrfs_inode_ref *)ptr;
3835 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3836 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3837 struct btrfs_inode_extref *extref;
3839 extref = (struct btrfs_inode_extref *)ptr;
3840 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3845 * try to precache a NULL acl entry for files that don't have
3846 * any xattrs or acls
3848 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3849 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3850 if (first_xattr_slot != -1) {
3851 path->slots[0] = first_xattr_slot;
3852 ret = btrfs_load_inode_props(inode, path);
3855 "error loading props for ino %llu (root %llu): %d",
3856 btrfs_ino(BTRFS_I(inode)),
3857 root->root_key.objectid, ret);
3859 btrfs_free_path(path);
3862 cache_no_acl(inode);
3864 switch (inode->i_mode & S_IFMT) {
3866 inode->i_mapping->a_ops = &btrfs_aops;
3867 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3868 inode->i_fop = &btrfs_file_operations;
3869 inode->i_op = &btrfs_file_inode_operations;
3872 inode->i_fop = &btrfs_dir_file_operations;
3873 inode->i_op = &btrfs_dir_inode_operations;
3876 inode->i_op = &btrfs_symlink_inode_operations;
3877 inode_nohighmem(inode);
3878 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3881 inode->i_op = &btrfs_special_inode_operations;
3882 init_special_inode(inode, inode->i_mode, rdev);
3886 btrfs_update_iflags(inode);
3890 btrfs_free_path(path);
3891 make_bad_inode(inode);
3896 * given a leaf and an inode, copy the inode fields into the leaf
3898 static void fill_inode_item(struct btrfs_trans_handle *trans,
3899 struct extent_buffer *leaf,
3900 struct btrfs_inode_item *item,
3901 struct inode *inode)
3903 struct btrfs_map_token token;
3905 btrfs_init_map_token(&token);
3907 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3908 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3909 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3911 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3912 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3914 btrfs_set_token_timespec_sec(leaf, &item->atime,
3915 inode->i_atime.tv_sec, &token);
3916 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3917 inode->i_atime.tv_nsec, &token);
3919 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3920 inode->i_mtime.tv_sec, &token);
3921 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3922 inode->i_mtime.tv_nsec, &token);
3924 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3925 inode->i_ctime.tv_sec, &token);
3926 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3927 inode->i_ctime.tv_nsec, &token);
3929 btrfs_set_token_timespec_sec(leaf, &item->otime,
3930 BTRFS_I(inode)->i_otime.tv_sec, &token);
3931 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3932 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3934 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3936 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3938 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3939 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3940 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3941 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3942 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3946 * copy everything in the in-memory inode into the btree.
3948 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3949 struct btrfs_root *root, struct inode *inode)
3951 struct btrfs_inode_item *inode_item;
3952 struct btrfs_path *path;
3953 struct extent_buffer *leaf;
3956 path = btrfs_alloc_path();
3960 path->leave_spinning = 1;
3961 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3969 leaf = path->nodes[0];
3970 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3971 struct btrfs_inode_item);
3973 fill_inode_item(trans, leaf, inode_item, inode);
3974 btrfs_mark_buffer_dirty(leaf);
3975 btrfs_set_inode_last_trans(trans, inode);
3978 btrfs_free_path(path);
3983 * copy everything in the in-memory inode into the btree.
3985 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3986 struct btrfs_root *root, struct inode *inode)
3988 struct btrfs_fs_info *fs_info = root->fs_info;
3992 * If the inode is a free space inode, we can deadlock during commit
3993 * if we put it into the delayed code.
3995 * The data relocation inode should also be directly updated
3998 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3999 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4000 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4001 btrfs_update_root_times(trans, root);
4003 ret = btrfs_delayed_update_inode(trans, root, inode);
4005 btrfs_set_inode_last_trans(trans, inode);
4009 return btrfs_update_inode_item(trans, root, inode);
4012 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4013 struct btrfs_root *root,
4014 struct inode *inode)
4018 ret = btrfs_update_inode(trans, root, inode);
4020 return btrfs_update_inode_item(trans, root, inode);
4025 * unlink helper that gets used here in inode.c and in the tree logging
4026 * recovery code. It remove a link in a directory with a given name, and
4027 * also drops the back refs in the inode to the directory
4029 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4030 struct btrfs_root *root,
4031 struct btrfs_inode *dir,
4032 struct btrfs_inode *inode,
4033 const char *name, int name_len)
4035 struct btrfs_fs_info *fs_info = root->fs_info;
4036 struct btrfs_path *path;
4038 struct extent_buffer *leaf;
4039 struct btrfs_dir_item *di;
4040 struct btrfs_key key;
4042 u64 ino = btrfs_ino(inode);
4043 u64 dir_ino = btrfs_ino(dir);
4045 path = btrfs_alloc_path();
4051 path->leave_spinning = 1;
4052 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4053 name, name_len, -1);
4062 leaf = path->nodes[0];
4063 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4064 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4067 btrfs_release_path(path);
4070 * If we don't have dir index, we have to get it by looking up
4071 * the inode ref, since we get the inode ref, remove it directly,
4072 * it is unnecessary to do delayed deletion.
4074 * But if we have dir index, needn't search inode ref to get it.
4075 * Since the inode ref is close to the inode item, it is better
4076 * that we delay to delete it, and just do this deletion when
4077 * we update the inode item.
4079 if (inode->dir_index) {
4080 ret = btrfs_delayed_delete_inode_ref(inode);
4082 index = inode->dir_index;
4087 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4091 "failed to delete reference to %.*s, inode %llu parent %llu",
4092 name_len, name, ino, dir_ino);
4093 btrfs_abort_transaction(trans, ret);
4097 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4099 btrfs_abort_transaction(trans, ret);
4103 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4105 if (ret != 0 && ret != -ENOENT) {
4106 btrfs_abort_transaction(trans, ret);
4110 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4115 btrfs_abort_transaction(trans, ret);
4117 btrfs_free_path(path);
4121 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4122 inode_inc_iversion(&inode->vfs_inode);
4123 inode_inc_iversion(&dir->vfs_inode);
4124 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4125 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4126 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4131 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4132 struct btrfs_root *root,
4133 struct btrfs_inode *dir, struct btrfs_inode *inode,
4134 const char *name, int name_len)
4137 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4139 drop_nlink(&inode->vfs_inode);
4140 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4146 * helper to start transaction for unlink and rmdir.
4148 * unlink and rmdir are special in btrfs, they do not always free space, so
4149 * if we cannot make our reservations the normal way try and see if there is
4150 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4151 * allow the unlink to occur.
4153 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4155 struct btrfs_root *root = BTRFS_I(dir)->root;
4158 * 1 for the possible orphan item
4159 * 1 for the dir item
4160 * 1 for the dir index
4161 * 1 for the inode ref
4164 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4167 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4169 struct btrfs_root *root = BTRFS_I(dir)->root;
4170 struct btrfs_trans_handle *trans;
4171 struct inode *inode = d_inode(dentry);
4174 trans = __unlink_start_trans(dir);
4176 return PTR_ERR(trans);
4178 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4181 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4182 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4183 dentry->d_name.len);
4187 if (inode->i_nlink == 0) {
4188 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4194 btrfs_end_transaction(trans);
4195 btrfs_btree_balance_dirty(root->fs_info);
4199 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4200 struct btrfs_root *root,
4201 struct inode *dir, u64 objectid,
4202 const char *name, int name_len)
4204 struct btrfs_fs_info *fs_info = root->fs_info;
4205 struct btrfs_path *path;
4206 struct extent_buffer *leaf;
4207 struct btrfs_dir_item *di;
4208 struct btrfs_key key;
4211 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4213 path = btrfs_alloc_path();
4217 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4218 name, name_len, -1);
4219 if (IS_ERR_OR_NULL(di)) {
4227 leaf = path->nodes[0];
4228 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4229 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4230 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4232 btrfs_abort_transaction(trans, ret);
4235 btrfs_release_path(path);
4237 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4238 root->root_key.objectid, dir_ino,
4239 &index, name, name_len);
4241 if (ret != -ENOENT) {
4242 btrfs_abort_transaction(trans, ret);
4245 di = btrfs_search_dir_index_item(root, path, dir_ino,
4247 if (IS_ERR_OR_NULL(di)) {
4252 btrfs_abort_transaction(trans, ret);
4256 leaf = path->nodes[0];
4257 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4258 btrfs_release_path(path);
4261 btrfs_release_path(path);
4263 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4265 btrfs_abort_transaction(trans, ret);
4269 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4270 inode_inc_iversion(dir);
4271 dir->i_mtime = dir->i_ctime = current_time(dir);
4272 ret = btrfs_update_inode_fallback(trans, root, dir);
4274 btrfs_abort_transaction(trans, ret);
4276 btrfs_free_path(path);
4280 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4282 struct inode *inode = d_inode(dentry);
4284 struct btrfs_root *root = BTRFS_I(dir)->root;
4285 struct btrfs_trans_handle *trans;
4286 u64 last_unlink_trans;
4288 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4290 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4293 trans = __unlink_start_trans(dir);
4295 return PTR_ERR(trans);
4297 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4298 err = btrfs_unlink_subvol(trans, root, dir,
4299 BTRFS_I(inode)->location.objectid,
4300 dentry->d_name.name,
4301 dentry->d_name.len);
4305 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4309 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4311 /* now the directory is empty */
4312 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4313 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4314 dentry->d_name.len);
4316 btrfs_i_size_write(BTRFS_I(inode), 0);
4318 * Propagate the last_unlink_trans value of the deleted dir to
4319 * its parent directory. This is to prevent an unrecoverable
4320 * log tree in the case we do something like this:
4322 * 2) create snapshot under dir foo
4323 * 3) delete the snapshot
4326 * 6) fsync foo or some file inside foo
4328 if (last_unlink_trans >= trans->transid)
4329 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4332 btrfs_end_transaction(trans);
4333 btrfs_btree_balance_dirty(root->fs_info);
4338 static int truncate_space_check(struct btrfs_trans_handle *trans,
4339 struct btrfs_root *root,
4342 struct btrfs_fs_info *fs_info = root->fs_info;
4346 * This is only used to apply pressure to the enospc system, we don't
4347 * intend to use this reservation at all.
4349 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4350 bytes_deleted *= fs_info->nodesize;
4351 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4352 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4354 trace_btrfs_space_reservation(fs_info, "transaction",
4357 trans->bytes_reserved += bytes_deleted;
4363 static int truncate_inline_extent(struct inode *inode,
4364 struct btrfs_path *path,
4365 struct btrfs_key *found_key,
4369 struct extent_buffer *leaf = path->nodes[0];
4370 int slot = path->slots[0];
4371 struct btrfs_file_extent_item *fi;
4372 u32 size = (u32)(new_size - found_key->offset);
4373 struct btrfs_root *root = BTRFS_I(inode)->root;
4375 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4377 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4378 loff_t offset = new_size;
4379 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4382 * Zero out the remaining of the last page of our inline extent,
4383 * instead of directly truncating our inline extent here - that
4384 * would be much more complex (decompressing all the data, then
4385 * compressing the truncated data, which might be bigger than
4386 * the size of the inline extent, resize the extent, etc).
4387 * We release the path because to get the page we might need to
4388 * read the extent item from disk (data not in the page cache).
4390 btrfs_release_path(path);
4391 return btrfs_truncate_block(inode, offset, page_end - offset,
4395 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4396 size = btrfs_file_extent_calc_inline_size(size);
4397 btrfs_truncate_item(root->fs_info, path, size, 1);
4399 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4400 inode_sub_bytes(inode, item_end + 1 - new_size);
4406 * this can truncate away extent items, csum items and directory items.
4407 * It starts at a high offset and removes keys until it can't find
4408 * any higher than new_size
4410 * csum items that cross the new i_size are truncated to the new size
4413 * min_type is the minimum key type to truncate down to. If set to 0, this
4414 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4416 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4417 struct btrfs_root *root,
4418 struct inode *inode,
4419 u64 new_size, u32 min_type)
4421 struct btrfs_fs_info *fs_info = root->fs_info;
4422 struct btrfs_path *path;
4423 struct extent_buffer *leaf;
4424 struct btrfs_file_extent_item *fi;
4425 struct btrfs_key key;
4426 struct btrfs_key found_key;
4427 u64 extent_start = 0;
4428 u64 extent_num_bytes = 0;
4429 u64 extent_offset = 0;
4431 u64 last_size = new_size;
4432 u32 found_type = (u8)-1;
4435 int pending_del_nr = 0;
4436 int pending_del_slot = 0;
4437 int extent_type = -1;
4440 u64 ino = btrfs_ino(BTRFS_I(inode));
4441 u64 bytes_deleted = 0;
4443 bool should_throttle = 0;
4444 bool should_end = 0;
4446 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4449 * for non-free space inodes and ref cows, we want to back off from
4452 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4453 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4456 path = btrfs_alloc_path();
4459 path->reada = READA_BACK;
4462 * We want to drop from the next block forward in case this new size is
4463 * not block aligned since we will be keeping the last block of the
4464 * extent just the way it is.
4466 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4467 root == fs_info->tree_root)
4468 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4469 fs_info->sectorsize),
4473 * This function is also used to drop the items in the log tree before
4474 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4475 * it is used to drop the loged items. So we shouldn't kill the delayed
4478 if (min_type == 0 && root == BTRFS_I(inode)->root)
4479 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4482 key.offset = (u64)-1;
4487 * with a 16K leaf size and 128MB extents, you can actually queue
4488 * up a huge file in a single leaf. Most of the time that
4489 * bytes_deleted is > 0, it will be huge by the time we get here
4491 if (be_nice && bytes_deleted > SZ_32M) {
4492 if (btrfs_should_end_transaction(trans)) {
4499 path->leave_spinning = 1;
4500 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4507 /* there are no items in the tree for us to truncate, we're
4510 if (path->slots[0] == 0)
4517 leaf = path->nodes[0];
4518 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4519 found_type = found_key.type;
4521 if (found_key.objectid != ino)
4524 if (found_type < min_type)
4527 item_end = found_key.offset;
4528 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4529 fi = btrfs_item_ptr(leaf, path->slots[0],
4530 struct btrfs_file_extent_item);
4531 extent_type = btrfs_file_extent_type(leaf, fi);
4532 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4534 btrfs_file_extent_num_bytes(leaf, fi);
4536 trace_btrfs_truncate_show_fi_regular(
4537 BTRFS_I(inode), leaf, fi,
4539 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4540 item_end += btrfs_file_extent_inline_len(leaf,
4541 path->slots[0], fi);
4543 trace_btrfs_truncate_show_fi_inline(
4544 BTRFS_I(inode), leaf, fi, path->slots[0],
4549 if (found_type > min_type) {
4552 if (item_end < new_size)
4554 if (found_key.offset >= new_size)
4560 /* FIXME, shrink the extent if the ref count is only 1 */
4561 if (found_type != BTRFS_EXTENT_DATA_KEY)
4565 last_size = found_key.offset;
4567 last_size = new_size;
4569 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4571 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4573 u64 orig_num_bytes =
4574 btrfs_file_extent_num_bytes(leaf, fi);
4575 extent_num_bytes = ALIGN(new_size -
4577 fs_info->sectorsize);
4578 btrfs_set_file_extent_num_bytes(leaf, fi,
4580 num_dec = (orig_num_bytes -
4582 if (test_bit(BTRFS_ROOT_REF_COWS,
4585 inode_sub_bytes(inode, num_dec);
4586 btrfs_mark_buffer_dirty(leaf);
4589 btrfs_file_extent_disk_num_bytes(leaf,
4591 extent_offset = found_key.offset -
4592 btrfs_file_extent_offset(leaf, fi);
4594 /* FIXME blocksize != 4096 */
4595 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4596 if (extent_start != 0) {
4598 if (test_bit(BTRFS_ROOT_REF_COWS,
4600 inode_sub_bytes(inode, num_dec);
4603 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4605 * we can't truncate inline items that have had
4609 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4610 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4613 * Need to release path in order to truncate a
4614 * compressed extent. So delete any accumulated
4615 * extent items so far.
4617 if (btrfs_file_extent_compression(leaf, fi) !=
4618 BTRFS_COMPRESS_NONE && pending_del_nr) {
4619 err = btrfs_del_items(trans, root, path,
4623 btrfs_abort_transaction(trans,
4630 err = truncate_inline_extent(inode, path,
4635 btrfs_abort_transaction(trans, err);
4638 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4640 inode_sub_bytes(inode, item_end + 1 - new_size);
4645 if (!pending_del_nr) {
4646 /* no pending yet, add ourselves */
4647 pending_del_slot = path->slots[0];
4649 } else if (pending_del_nr &&
4650 path->slots[0] + 1 == pending_del_slot) {
4651 /* hop on the pending chunk */
4653 pending_del_slot = path->slots[0];
4660 should_throttle = 0;
4663 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4664 root == fs_info->tree_root)) {
4665 btrfs_set_path_blocking(path);
4666 bytes_deleted += extent_num_bytes;
4667 ret = btrfs_free_extent(trans, fs_info, extent_start,
4668 extent_num_bytes, 0,
4669 btrfs_header_owner(leaf),
4670 ino, extent_offset);
4672 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4673 btrfs_async_run_delayed_refs(fs_info,
4674 trans->delayed_ref_updates * 2,
4677 if (truncate_space_check(trans, root,
4678 extent_num_bytes)) {
4681 if (btrfs_should_throttle_delayed_refs(trans,
4683 should_throttle = 1;
4687 if (found_type == BTRFS_INODE_ITEM_KEY)
4690 if (path->slots[0] == 0 ||
4691 path->slots[0] != pending_del_slot ||
4692 should_throttle || should_end) {
4693 if (pending_del_nr) {
4694 ret = btrfs_del_items(trans, root, path,
4698 btrfs_abort_transaction(trans, ret);
4703 btrfs_release_path(path);
4704 if (should_throttle) {
4705 unsigned long updates = trans->delayed_ref_updates;
4707 trans->delayed_ref_updates = 0;
4708 ret = btrfs_run_delayed_refs(trans,
4716 * if we failed to refill our space rsv, bail out
4717 * and let the transaction restart
4729 if (pending_del_nr) {
4730 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4733 btrfs_abort_transaction(trans, ret);
4736 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4737 ASSERT(last_size >= new_size);
4738 if (!err && last_size > new_size)
4739 last_size = new_size;
4740 btrfs_ordered_update_i_size(inode, last_size, NULL);
4743 btrfs_free_path(path);
4745 if (be_nice && bytes_deleted > SZ_32M) {
4746 unsigned long updates = trans->delayed_ref_updates;
4748 trans->delayed_ref_updates = 0;
4749 ret = btrfs_run_delayed_refs(trans, fs_info,
4759 * btrfs_truncate_block - read, zero a chunk and write a block
4760 * @inode - inode that we're zeroing
4761 * @from - the offset to start zeroing
4762 * @len - the length to zero, 0 to zero the entire range respective to the
4764 * @front - zero up to the offset instead of from the offset on
4766 * This will find the block for the "from" offset and cow the block and zero the
4767 * part we want to zero. This is used with truncate and hole punching.
4769 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4772 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4773 struct address_space *mapping = inode->i_mapping;
4774 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4775 struct btrfs_ordered_extent *ordered;
4776 struct extent_state *cached_state = NULL;
4777 struct extent_changeset *data_reserved = NULL;
4779 u32 blocksize = fs_info->sectorsize;
4780 pgoff_t index = from >> PAGE_SHIFT;
4781 unsigned offset = from & (blocksize - 1);
4783 gfp_t mask = btrfs_alloc_write_mask(mapping);
4788 if ((offset & (blocksize - 1)) == 0 &&
4789 (!len || ((len & (blocksize - 1)) == 0)))
4792 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4793 round_down(from, blocksize), blocksize);
4798 page = find_or_create_page(mapping, index, mask);
4800 btrfs_delalloc_release_space(inode, data_reserved,
4801 round_down(from, blocksize),
4807 block_start = round_down(from, blocksize);
4808 block_end = block_start + blocksize - 1;
4810 if (!PageUptodate(page)) {
4811 ret = btrfs_readpage(NULL, page);
4813 if (page->mapping != mapping) {
4818 if (!PageUptodate(page)) {
4823 wait_on_page_writeback(page);
4825 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4826 set_page_extent_mapped(page);
4828 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4830 unlock_extent_cached(io_tree, block_start, block_end,
4831 &cached_state, GFP_NOFS);
4834 btrfs_start_ordered_extent(inode, ordered, 1);
4835 btrfs_put_ordered_extent(ordered);
4839 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4840 EXTENT_DIRTY | EXTENT_DELALLOC |
4841 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4842 0, 0, &cached_state, GFP_NOFS);
4844 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4847 unlock_extent_cached(io_tree, block_start, block_end,
4848 &cached_state, GFP_NOFS);
4852 if (offset != blocksize) {
4854 len = blocksize - offset;
4857 memset(kaddr + (block_start - page_offset(page)),
4860 memset(kaddr + (block_start - page_offset(page)) + offset,
4862 flush_dcache_page(page);
4865 ClearPageChecked(page);
4866 set_page_dirty(page);
4867 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4872 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4877 extent_changeset_free(data_reserved);
4881 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4882 u64 offset, u64 len)
4884 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4885 struct btrfs_trans_handle *trans;
4889 * Still need to make sure the inode looks like it's been updated so
4890 * that any holes get logged if we fsync.
4892 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4893 BTRFS_I(inode)->last_trans = fs_info->generation;
4894 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4895 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4900 * 1 - for the one we're dropping
4901 * 1 - for the one we're adding
4902 * 1 - for updating the inode.
4904 trans = btrfs_start_transaction(root, 3);
4906 return PTR_ERR(trans);
4908 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4910 btrfs_abort_transaction(trans, ret);
4911 btrfs_end_transaction(trans);
4915 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4916 offset, 0, 0, len, 0, len, 0, 0, 0);
4918 btrfs_abort_transaction(trans, ret);
4920 btrfs_update_inode(trans, root, inode);
4921 btrfs_end_transaction(trans);
4926 * This function puts in dummy file extents for the area we're creating a hole
4927 * for. So if we are truncating this file to a larger size we need to insert
4928 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4929 * the range between oldsize and size
4931 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4933 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4934 struct btrfs_root *root = BTRFS_I(inode)->root;
4935 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4936 struct extent_map *em = NULL;
4937 struct extent_state *cached_state = NULL;
4938 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4939 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4940 u64 block_end = ALIGN(size, fs_info->sectorsize);
4947 * If our size started in the middle of a block we need to zero out the
4948 * rest of the block before we expand the i_size, otherwise we could
4949 * expose stale data.
4951 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4955 if (size <= hole_start)
4959 struct btrfs_ordered_extent *ordered;
4961 lock_extent_bits(io_tree, hole_start, block_end - 1,
4963 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4964 block_end - hole_start);
4967 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4968 &cached_state, GFP_NOFS);
4969 btrfs_start_ordered_extent(inode, ordered, 1);
4970 btrfs_put_ordered_extent(ordered);
4973 cur_offset = hole_start;
4975 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4976 block_end - cur_offset, 0);
4982 last_byte = min(extent_map_end(em), block_end);
4983 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4984 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4985 struct extent_map *hole_em;
4986 hole_size = last_byte - cur_offset;
4988 err = maybe_insert_hole(root, inode, cur_offset,
4992 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4993 cur_offset + hole_size - 1, 0);
4994 hole_em = alloc_extent_map();
4996 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4997 &BTRFS_I(inode)->runtime_flags);
5000 hole_em->start = cur_offset;
5001 hole_em->len = hole_size;
5002 hole_em->orig_start = cur_offset;
5004 hole_em->block_start = EXTENT_MAP_HOLE;
5005 hole_em->block_len = 0;
5006 hole_em->orig_block_len = 0;
5007 hole_em->ram_bytes = hole_size;
5008 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5009 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5010 hole_em->generation = fs_info->generation;
5013 write_lock(&em_tree->lock);
5014 err = add_extent_mapping(em_tree, hole_em, 1);
5015 write_unlock(&em_tree->lock);
5018 btrfs_drop_extent_cache(BTRFS_I(inode),
5023 free_extent_map(hole_em);
5026 free_extent_map(em);
5028 cur_offset = last_byte;
5029 if (cur_offset >= block_end)
5032 free_extent_map(em);
5033 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5038 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5040 struct btrfs_root *root = BTRFS_I(inode)->root;
5041 struct btrfs_trans_handle *trans;
5042 loff_t oldsize = i_size_read(inode);
5043 loff_t newsize = attr->ia_size;
5044 int mask = attr->ia_valid;
5048 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5049 * special case where we need to update the times despite not having
5050 * these flags set. For all other operations the VFS set these flags
5051 * explicitly if it wants a timestamp update.
5053 if (newsize != oldsize) {
5054 inode_inc_iversion(inode);
5055 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5056 inode->i_ctime = inode->i_mtime =
5057 current_time(inode);
5060 if (newsize > oldsize) {
5062 * Don't do an expanding truncate while snapshotting is ongoing.
5063 * This is to ensure the snapshot captures a fully consistent
5064 * state of this file - if the snapshot captures this expanding
5065 * truncation, it must capture all writes that happened before
5068 btrfs_wait_for_snapshot_creation(root);
5069 ret = btrfs_cont_expand(inode, oldsize, newsize);
5071 btrfs_end_write_no_snapshotting(root);
5075 trans = btrfs_start_transaction(root, 1);
5076 if (IS_ERR(trans)) {
5077 btrfs_end_write_no_snapshotting(root);
5078 return PTR_ERR(trans);
5081 i_size_write(inode, newsize);
5082 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5083 pagecache_isize_extended(inode, oldsize, newsize);
5084 ret = btrfs_update_inode(trans, root, inode);
5085 btrfs_end_write_no_snapshotting(root);
5086 btrfs_end_transaction(trans);
5090 * We're truncating a file that used to have good data down to
5091 * zero. Make sure it gets into the ordered flush list so that
5092 * any new writes get down to disk quickly.
5095 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5096 &BTRFS_I(inode)->runtime_flags);
5099 * 1 for the orphan item we're going to add
5100 * 1 for the orphan item deletion.
5102 trans = btrfs_start_transaction(root, 2);
5104 return PTR_ERR(trans);
5107 * We need to do this in case we fail at _any_ point during the
5108 * actual truncate. Once we do the truncate_setsize we could
5109 * invalidate pages which forces any outstanding ordered io to
5110 * be instantly completed which will give us extents that need
5111 * to be truncated. If we fail to get an orphan inode down we
5112 * could have left over extents that were never meant to live,
5113 * so we need to guarantee from this point on that everything
5114 * will be consistent.
5116 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5117 btrfs_end_transaction(trans);
5121 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5122 truncate_setsize(inode, newsize);
5124 /* Disable nonlocked read DIO to avoid the end less truncate */
5125 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5126 inode_dio_wait(inode);
5127 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5129 ret = btrfs_truncate(inode);
5130 if (ret && inode->i_nlink) {
5133 /* To get a stable disk_i_size */
5134 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5136 btrfs_orphan_del(NULL, BTRFS_I(inode));
5141 * failed to truncate, disk_i_size is only adjusted down
5142 * as we remove extents, so it should represent the true
5143 * size of the inode, so reset the in memory size and
5144 * delete our orphan entry.
5146 trans = btrfs_join_transaction(root);
5147 if (IS_ERR(trans)) {
5148 btrfs_orphan_del(NULL, BTRFS_I(inode));
5151 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5152 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5154 btrfs_abort_transaction(trans, err);
5155 btrfs_end_transaction(trans);
5162 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5164 struct inode *inode = d_inode(dentry);
5165 struct btrfs_root *root = BTRFS_I(inode)->root;
5168 if (btrfs_root_readonly(root))
5171 err = setattr_prepare(dentry, attr);
5175 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5176 err = btrfs_setsize(inode, attr);
5181 if (attr->ia_valid) {
5182 setattr_copy(inode, attr);
5183 inode_inc_iversion(inode);
5184 err = btrfs_dirty_inode(inode);
5186 if (!err && attr->ia_valid & ATTR_MODE)
5187 err = posix_acl_chmod(inode, inode->i_mode);
5194 * While truncating the inode pages during eviction, we get the VFS calling
5195 * btrfs_invalidatepage() against each page of the inode. This is slow because
5196 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5197 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5198 * extent_state structures over and over, wasting lots of time.
5200 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5201 * those expensive operations on a per page basis and do only the ordered io
5202 * finishing, while we release here the extent_map and extent_state structures,
5203 * without the excessive merging and splitting.
5205 static void evict_inode_truncate_pages(struct inode *inode)
5207 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5208 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5209 struct rb_node *node;
5211 ASSERT(inode->i_state & I_FREEING);
5212 truncate_inode_pages_final(&inode->i_data);
5214 write_lock(&map_tree->lock);
5215 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5216 struct extent_map *em;
5218 node = rb_first(&map_tree->map);
5219 em = rb_entry(node, struct extent_map, rb_node);
5220 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5221 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5222 remove_extent_mapping(map_tree, em);
5223 free_extent_map(em);
5224 if (need_resched()) {
5225 write_unlock(&map_tree->lock);
5227 write_lock(&map_tree->lock);
5230 write_unlock(&map_tree->lock);
5233 * Keep looping until we have no more ranges in the io tree.
5234 * We can have ongoing bios started by readpages (called from readahead)
5235 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5236 * still in progress (unlocked the pages in the bio but did not yet
5237 * unlocked the ranges in the io tree). Therefore this means some
5238 * ranges can still be locked and eviction started because before
5239 * submitting those bios, which are executed by a separate task (work
5240 * queue kthread), inode references (inode->i_count) were not taken
5241 * (which would be dropped in the end io callback of each bio).
5242 * Therefore here we effectively end up waiting for those bios and
5243 * anyone else holding locked ranges without having bumped the inode's
5244 * reference count - if we don't do it, when they access the inode's
5245 * io_tree to unlock a range it may be too late, leading to an
5246 * use-after-free issue.
5248 spin_lock(&io_tree->lock);
5249 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5250 struct extent_state *state;
5251 struct extent_state *cached_state = NULL;
5255 node = rb_first(&io_tree->state);
5256 state = rb_entry(node, struct extent_state, rb_node);
5257 start = state->start;
5259 spin_unlock(&io_tree->lock);
5261 lock_extent_bits(io_tree, start, end, &cached_state);
5264 * If still has DELALLOC flag, the extent didn't reach disk,
5265 * and its reserved space won't be freed by delayed_ref.
5266 * So we need to free its reserved space here.
5267 * (Refer to comment in btrfs_invalidatepage, case 2)
5269 * Note, end is the bytenr of last byte, so we need + 1 here.
5271 if (state->state & EXTENT_DELALLOC)
5272 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5274 clear_extent_bit(io_tree, start, end,
5275 EXTENT_LOCKED | EXTENT_DIRTY |
5276 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5277 EXTENT_DEFRAG, 1, 1,
5278 &cached_state, GFP_NOFS);
5281 spin_lock(&io_tree->lock);
5283 spin_unlock(&io_tree->lock);
5286 void btrfs_evict_inode(struct inode *inode)
5288 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5289 struct btrfs_trans_handle *trans;
5290 struct btrfs_root *root = BTRFS_I(inode)->root;
5291 struct btrfs_block_rsv *rsv, *global_rsv;
5292 int steal_from_global = 0;
5296 trace_btrfs_inode_evict(inode);
5299 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5303 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5305 evict_inode_truncate_pages(inode);
5307 if (inode->i_nlink &&
5308 ((btrfs_root_refs(&root->root_item) != 0 &&
5309 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5310 btrfs_is_free_space_inode(BTRFS_I(inode))))
5313 if (is_bad_inode(inode)) {
5314 btrfs_orphan_del(NULL, BTRFS_I(inode));
5317 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5318 if (!special_file(inode->i_mode))
5319 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5321 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5323 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5324 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5325 &BTRFS_I(inode)->runtime_flags));
5329 if (inode->i_nlink > 0) {
5330 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5331 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5335 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5337 btrfs_orphan_del(NULL, BTRFS_I(inode));
5341 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5343 btrfs_orphan_del(NULL, BTRFS_I(inode));
5346 rsv->size = min_size;
5348 global_rsv = &fs_info->global_block_rsv;
5350 btrfs_i_size_write(BTRFS_I(inode), 0);
5353 * This is a bit simpler than btrfs_truncate since we've already
5354 * reserved our space for our orphan item in the unlink, so we just
5355 * need to reserve some slack space in case we add bytes and update
5356 * inode item when doing the truncate.
5359 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5360 BTRFS_RESERVE_FLUSH_LIMIT);
5363 * Try and steal from the global reserve since we will
5364 * likely not use this space anyway, we want to try as
5365 * hard as possible to get this to work.
5368 steal_from_global++;
5370 steal_from_global = 0;
5374 * steal_from_global == 0: we reserved stuff, hooray!
5375 * steal_from_global == 1: we didn't reserve stuff, boo!
5376 * steal_from_global == 2: we've committed, still not a lot of
5377 * room but maybe we'll have room in the global reserve this
5379 * steal_from_global == 3: abandon all hope!
5381 if (steal_from_global > 2) {
5383 "Could not get space for a delete, will truncate on mount %d",
5385 btrfs_orphan_del(NULL, BTRFS_I(inode));
5386 btrfs_free_block_rsv(fs_info, rsv);
5390 trans = btrfs_join_transaction(root);
5391 if (IS_ERR(trans)) {
5392 btrfs_orphan_del(NULL, BTRFS_I(inode));
5393 btrfs_free_block_rsv(fs_info, rsv);
5398 * We can't just steal from the global reserve, we need to make
5399 * sure there is room to do it, if not we need to commit and try
5402 if (steal_from_global) {
5403 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5404 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5411 * Couldn't steal from the global reserve, we have too much
5412 * pending stuff built up, commit the transaction and try it
5416 ret = btrfs_commit_transaction(trans);
5418 btrfs_orphan_del(NULL, BTRFS_I(inode));
5419 btrfs_free_block_rsv(fs_info, rsv);
5424 steal_from_global = 0;
5427 trans->block_rsv = rsv;
5429 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5430 if (ret != -ENOSPC && ret != -EAGAIN)
5433 trans->block_rsv = &fs_info->trans_block_rsv;
5434 btrfs_end_transaction(trans);
5436 btrfs_btree_balance_dirty(fs_info);
5439 btrfs_free_block_rsv(fs_info, rsv);
5442 * Errors here aren't a big deal, it just means we leave orphan items
5443 * in the tree. They will be cleaned up on the next mount.
5446 trans->block_rsv = root->orphan_block_rsv;
5447 btrfs_orphan_del(trans, BTRFS_I(inode));
5449 btrfs_orphan_del(NULL, BTRFS_I(inode));
5452 trans->block_rsv = &fs_info->trans_block_rsv;
5453 if (!(root == fs_info->tree_root ||
5454 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5455 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5457 btrfs_end_transaction(trans);
5458 btrfs_btree_balance_dirty(fs_info);
5460 btrfs_remove_delayed_node(BTRFS_I(inode));
5465 * this returns the key found in the dir entry in the location pointer.
5466 * If no dir entries were found, location->objectid is 0.
5468 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5469 struct btrfs_key *location)
5471 const char *name = dentry->d_name.name;
5472 int namelen = dentry->d_name.len;
5473 struct btrfs_dir_item *di;
5474 struct btrfs_path *path;
5475 struct btrfs_root *root = BTRFS_I(dir)->root;
5478 path = btrfs_alloc_path();
5482 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5487 if (IS_ERR_OR_NULL(di))
5490 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5492 btrfs_free_path(path);
5495 location->objectid = 0;
5500 * when we hit a tree root in a directory, the btrfs part of the inode
5501 * needs to be changed to reflect the root directory of the tree root. This
5502 * is kind of like crossing a mount point.
5504 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5506 struct dentry *dentry,
5507 struct btrfs_key *location,
5508 struct btrfs_root **sub_root)
5510 struct btrfs_path *path;
5511 struct btrfs_root *new_root;
5512 struct btrfs_root_ref *ref;
5513 struct extent_buffer *leaf;
5514 struct btrfs_key key;
5518 path = btrfs_alloc_path();
5525 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5526 key.type = BTRFS_ROOT_REF_KEY;
5527 key.offset = location->objectid;
5529 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5536 leaf = path->nodes[0];
5537 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5538 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5539 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5542 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5543 (unsigned long)(ref + 1),
5544 dentry->d_name.len);
5548 btrfs_release_path(path);
5550 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5551 if (IS_ERR(new_root)) {
5552 err = PTR_ERR(new_root);
5556 *sub_root = new_root;
5557 location->objectid = btrfs_root_dirid(&new_root->root_item);
5558 location->type = BTRFS_INODE_ITEM_KEY;
5559 location->offset = 0;
5562 btrfs_free_path(path);
5566 static void inode_tree_add(struct inode *inode)
5568 struct btrfs_root *root = BTRFS_I(inode)->root;
5569 struct btrfs_inode *entry;
5571 struct rb_node *parent;
5572 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5573 u64 ino = btrfs_ino(BTRFS_I(inode));
5575 if (inode_unhashed(inode))
5578 spin_lock(&root->inode_lock);
5579 p = &root->inode_tree.rb_node;
5582 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5584 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5585 p = &parent->rb_left;
5586 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5587 p = &parent->rb_right;
5589 WARN_ON(!(entry->vfs_inode.i_state &
5590 (I_WILL_FREE | I_FREEING)));
5591 rb_replace_node(parent, new, &root->inode_tree);
5592 RB_CLEAR_NODE(parent);
5593 spin_unlock(&root->inode_lock);
5597 rb_link_node(new, parent, p);
5598 rb_insert_color(new, &root->inode_tree);
5599 spin_unlock(&root->inode_lock);
5602 static void inode_tree_del(struct inode *inode)
5604 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5605 struct btrfs_root *root = BTRFS_I(inode)->root;
5608 spin_lock(&root->inode_lock);
5609 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5610 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5611 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5612 empty = RB_EMPTY_ROOT(&root->inode_tree);
5614 spin_unlock(&root->inode_lock);
5616 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5617 synchronize_srcu(&fs_info->subvol_srcu);
5618 spin_lock(&root->inode_lock);
5619 empty = RB_EMPTY_ROOT(&root->inode_tree);
5620 spin_unlock(&root->inode_lock);
5622 btrfs_add_dead_root(root);
5626 void btrfs_invalidate_inodes(struct btrfs_root *root)
5628 struct btrfs_fs_info *fs_info = root->fs_info;
5629 struct rb_node *node;
5630 struct rb_node *prev;
5631 struct btrfs_inode *entry;
5632 struct inode *inode;
5635 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5636 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5638 spin_lock(&root->inode_lock);
5640 node = root->inode_tree.rb_node;
5644 entry = rb_entry(node, struct btrfs_inode, rb_node);
5646 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5647 node = node->rb_left;
5648 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5649 node = node->rb_right;
5655 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5656 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5660 prev = rb_next(prev);
5664 entry = rb_entry(node, struct btrfs_inode, rb_node);
5665 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5666 inode = igrab(&entry->vfs_inode);
5668 spin_unlock(&root->inode_lock);
5669 if (atomic_read(&inode->i_count) > 1)
5670 d_prune_aliases(inode);
5672 * btrfs_drop_inode will have it removed from
5673 * the inode cache when its usage count
5678 spin_lock(&root->inode_lock);
5682 if (cond_resched_lock(&root->inode_lock))
5685 node = rb_next(node);
5687 spin_unlock(&root->inode_lock);
5690 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5692 struct btrfs_iget_args *args = p;
5693 inode->i_ino = args->location->objectid;
5694 memcpy(&BTRFS_I(inode)->location, args->location,
5695 sizeof(*args->location));
5696 BTRFS_I(inode)->root = args->root;
5700 static int btrfs_find_actor(struct inode *inode, void *opaque)
5702 struct btrfs_iget_args *args = opaque;
5703 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5704 args->root == BTRFS_I(inode)->root;
5707 static struct inode *btrfs_iget_locked(struct super_block *s,
5708 struct btrfs_key *location,
5709 struct btrfs_root *root)
5711 struct inode *inode;
5712 struct btrfs_iget_args args;
5713 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5715 args.location = location;
5718 inode = iget5_locked(s, hashval, btrfs_find_actor,
5719 btrfs_init_locked_inode,
5724 /* Get an inode object given its location and corresponding root.
5725 * Returns in *is_new if the inode was read from disk
5727 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5728 struct btrfs_root *root, int *new)
5730 struct inode *inode;
5732 inode = btrfs_iget_locked(s, location, root);
5734 return ERR_PTR(-ENOMEM);
5736 if (inode->i_state & I_NEW) {
5739 ret = btrfs_read_locked_inode(inode);
5740 if (!is_bad_inode(inode)) {
5741 inode_tree_add(inode);
5742 unlock_new_inode(inode);
5746 unlock_new_inode(inode);
5749 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5756 static struct inode *new_simple_dir(struct super_block *s,
5757 struct btrfs_key *key,
5758 struct btrfs_root *root)
5760 struct inode *inode = new_inode(s);
5763 return ERR_PTR(-ENOMEM);
5765 BTRFS_I(inode)->root = root;
5766 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5767 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5769 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5770 inode->i_op = &btrfs_dir_ro_inode_operations;
5771 inode->i_opflags &= ~IOP_XATTR;
5772 inode->i_fop = &simple_dir_operations;
5773 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5774 inode->i_mtime = current_time(inode);
5775 inode->i_atime = inode->i_mtime;
5776 inode->i_ctime = inode->i_mtime;
5777 BTRFS_I(inode)->i_otime = inode->i_mtime;
5782 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5784 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5785 struct inode *inode;
5786 struct btrfs_root *root = BTRFS_I(dir)->root;
5787 struct btrfs_root *sub_root = root;
5788 struct btrfs_key location;
5792 if (dentry->d_name.len > BTRFS_NAME_LEN)
5793 return ERR_PTR(-ENAMETOOLONG);
5795 ret = btrfs_inode_by_name(dir, dentry, &location);
5797 return ERR_PTR(ret);
5799 if (location.objectid == 0)
5800 return ERR_PTR(-ENOENT);
5802 if (location.type == BTRFS_INODE_ITEM_KEY) {
5803 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5807 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5809 index = srcu_read_lock(&fs_info->subvol_srcu);
5810 ret = fixup_tree_root_location(fs_info, dir, dentry,
5811 &location, &sub_root);
5814 inode = ERR_PTR(ret);
5816 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5818 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5820 srcu_read_unlock(&fs_info->subvol_srcu, index);
5822 if (!IS_ERR(inode) && root != sub_root) {
5823 down_read(&fs_info->cleanup_work_sem);
5824 if (!(inode->i_sb->s_flags & MS_RDONLY))
5825 ret = btrfs_orphan_cleanup(sub_root);
5826 up_read(&fs_info->cleanup_work_sem);
5829 inode = ERR_PTR(ret);
5836 static int btrfs_dentry_delete(const struct dentry *dentry)
5838 struct btrfs_root *root;
5839 struct inode *inode = d_inode(dentry);
5841 if (!inode && !IS_ROOT(dentry))
5842 inode = d_inode(dentry->d_parent);
5845 root = BTRFS_I(inode)->root;
5846 if (btrfs_root_refs(&root->root_item) == 0)
5849 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5855 static void btrfs_dentry_release(struct dentry *dentry)
5857 kfree(dentry->d_fsdata);
5860 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5863 struct inode *inode;
5865 inode = btrfs_lookup_dentry(dir, dentry);
5866 if (IS_ERR(inode)) {
5867 if (PTR_ERR(inode) == -ENOENT)
5870 return ERR_CAST(inode);
5873 return d_splice_alias(inode, dentry);
5876 unsigned char btrfs_filetype_table[] = {
5877 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5881 * All this infrastructure exists because dir_emit can fault, and we are holding
5882 * the tree lock when doing readdir. For now just allocate a buffer and copy
5883 * our information into that, and then dir_emit from the buffer. This is
5884 * similar to what NFS does, only we don't keep the buffer around in pagecache
5885 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5886 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5889 static int btrfs_opendir(struct inode *inode, struct file *file)
5891 struct btrfs_file_private *private;
5893 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5896 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5897 if (!private->filldir_buf) {
5901 file->private_data = private;
5912 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5915 struct dir_entry *entry = addr;
5916 char *name = (char *)(entry + 1);
5918 ctx->pos = entry->offset;
5919 if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5922 addr += sizeof(struct dir_entry) + entry->name_len;
5928 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5930 struct inode *inode = file_inode(file);
5931 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5932 struct btrfs_root *root = BTRFS_I(inode)->root;
5933 struct btrfs_file_private *private = file->private_data;
5934 struct btrfs_dir_item *di;
5935 struct btrfs_key key;
5936 struct btrfs_key found_key;
5937 struct btrfs_path *path;
5939 struct list_head ins_list;
5940 struct list_head del_list;
5942 struct extent_buffer *leaf;
5949 struct btrfs_key location;
5951 if (!dir_emit_dots(file, ctx))
5954 path = btrfs_alloc_path();
5958 addr = private->filldir_buf;
5959 path->reada = READA_FORWARD;
5961 INIT_LIST_HEAD(&ins_list);
5962 INIT_LIST_HEAD(&del_list);
5963 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5966 key.type = BTRFS_DIR_INDEX_KEY;
5967 key.offset = ctx->pos;
5968 key.objectid = btrfs_ino(BTRFS_I(inode));
5970 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5975 struct dir_entry *entry;
5977 leaf = path->nodes[0];
5978 slot = path->slots[0];
5979 if (slot >= btrfs_header_nritems(leaf)) {
5980 ret = btrfs_next_leaf(root, path);
5988 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5990 if (found_key.objectid != key.objectid)
5992 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5994 if (found_key.offset < ctx->pos)
5996 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5998 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5999 if (verify_dir_item(fs_info, leaf, slot, di))
6002 name_len = btrfs_dir_name_len(leaf, di);
6003 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6005 btrfs_release_path(path);
6006 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6009 addr = private->filldir_buf;
6016 entry->name_len = name_len;
6017 name_ptr = (char *)(entry + 1);
6018 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6020 entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
6021 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6022 entry->ino = location.objectid;
6023 entry->offset = found_key.offset;
6025 addr += sizeof(struct dir_entry) + name_len;
6026 total_len += sizeof(struct dir_entry) + name_len;
6030 btrfs_release_path(path);
6032 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6036 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6041 * Stop new entries from being returned after we return the last
6044 * New directory entries are assigned a strictly increasing
6045 * offset. This means that new entries created during readdir
6046 * are *guaranteed* to be seen in the future by that readdir.
6047 * This has broken buggy programs which operate on names as
6048 * they're returned by readdir. Until we re-use freed offsets
6049 * we have this hack to stop new entries from being returned
6050 * under the assumption that they'll never reach this huge
6053 * This is being careful not to overflow 32bit loff_t unless the
6054 * last entry requires it because doing so has broken 32bit apps
6057 if (ctx->pos >= INT_MAX)
6058 ctx->pos = LLONG_MAX;
6065 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6066 btrfs_free_path(path);
6070 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6072 struct btrfs_root *root = BTRFS_I(inode)->root;
6073 struct btrfs_trans_handle *trans;
6075 bool nolock = false;
6077 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6080 if (btrfs_fs_closing(root->fs_info) &&
6081 btrfs_is_free_space_inode(BTRFS_I(inode)))
6084 if (wbc->sync_mode == WB_SYNC_ALL) {
6086 trans = btrfs_join_transaction_nolock(root);
6088 trans = btrfs_join_transaction(root);
6090 return PTR_ERR(trans);
6091 ret = btrfs_commit_transaction(trans);
6097 * This is somewhat expensive, updating the tree every time the
6098 * inode changes. But, it is most likely to find the inode in cache.
6099 * FIXME, needs more benchmarking...there are no reasons other than performance
6100 * to keep or drop this code.
6102 static int btrfs_dirty_inode(struct inode *inode)
6104 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6105 struct btrfs_root *root = BTRFS_I(inode)->root;
6106 struct btrfs_trans_handle *trans;
6109 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6112 trans = btrfs_join_transaction(root);
6114 return PTR_ERR(trans);
6116 ret = btrfs_update_inode(trans, root, inode);
6117 if (ret && ret == -ENOSPC) {
6118 /* whoops, lets try again with the full transaction */
6119 btrfs_end_transaction(trans);
6120 trans = btrfs_start_transaction(root, 1);
6122 return PTR_ERR(trans);
6124 ret = btrfs_update_inode(trans, root, inode);
6126 btrfs_end_transaction(trans);
6127 if (BTRFS_I(inode)->delayed_node)
6128 btrfs_balance_delayed_items(fs_info);
6134 * This is a copy of file_update_time. We need this so we can return error on
6135 * ENOSPC for updating the inode in the case of file write and mmap writes.
6137 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6140 struct btrfs_root *root = BTRFS_I(inode)->root;
6142 if (btrfs_root_readonly(root))
6145 if (flags & S_VERSION)
6146 inode_inc_iversion(inode);
6147 if (flags & S_CTIME)
6148 inode->i_ctime = *now;
6149 if (flags & S_MTIME)
6150 inode->i_mtime = *now;
6151 if (flags & S_ATIME)
6152 inode->i_atime = *now;
6153 return btrfs_dirty_inode(inode);
6157 * find the highest existing sequence number in a directory
6158 * and then set the in-memory index_cnt variable to reflect
6159 * free sequence numbers
6161 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6163 struct btrfs_root *root = inode->root;
6164 struct btrfs_key key, found_key;
6165 struct btrfs_path *path;
6166 struct extent_buffer *leaf;
6169 key.objectid = btrfs_ino(inode);
6170 key.type = BTRFS_DIR_INDEX_KEY;
6171 key.offset = (u64)-1;
6173 path = btrfs_alloc_path();
6177 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6180 /* FIXME: we should be able to handle this */
6186 * MAGIC NUMBER EXPLANATION:
6187 * since we search a directory based on f_pos we have to start at 2
6188 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6189 * else has to start at 2
6191 if (path->slots[0] == 0) {
6192 inode->index_cnt = 2;
6198 leaf = path->nodes[0];
6199 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6201 if (found_key.objectid != btrfs_ino(inode) ||
6202 found_key.type != BTRFS_DIR_INDEX_KEY) {
6203 inode->index_cnt = 2;
6207 inode->index_cnt = found_key.offset + 1;
6209 btrfs_free_path(path);
6214 * helper to find a free sequence number in a given directory. This current
6215 * code is very simple, later versions will do smarter things in the btree
6217 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6221 if (dir->index_cnt == (u64)-1) {
6222 ret = btrfs_inode_delayed_dir_index_count(dir);
6224 ret = btrfs_set_inode_index_count(dir);
6230 *index = dir->index_cnt;
6236 static int btrfs_insert_inode_locked(struct inode *inode)
6238 struct btrfs_iget_args args;
6239 args.location = &BTRFS_I(inode)->location;
6240 args.root = BTRFS_I(inode)->root;
6242 return insert_inode_locked4(inode,
6243 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6244 btrfs_find_actor, &args);
6248 * Inherit flags from the parent inode.
6250 * Currently only the compression flags and the cow flags are inherited.
6252 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6259 flags = BTRFS_I(dir)->flags;
6261 if (flags & BTRFS_INODE_NOCOMPRESS) {
6262 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6263 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6264 } else if (flags & BTRFS_INODE_COMPRESS) {
6265 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6266 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6269 if (flags & BTRFS_INODE_NODATACOW) {
6270 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6271 if (S_ISREG(inode->i_mode))
6272 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6275 btrfs_update_iflags(inode);
6278 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6279 struct btrfs_root *root,
6281 const char *name, int name_len,
6282 u64 ref_objectid, u64 objectid,
6283 umode_t mode, u64 *index)
6285 struct btrfs_fs_info *fs_info = root->fs_info;
6286 struct inode *inode;
6287 struct btrfs_inode_item *inode_item;
6288 struct btrfs_key *location;
6289 struct btrfs_path *path;
6290 struct btrfs_inode_ref *ref;
6291 struct btrfs_key key[2];
6293 int nitems = name ? 2 : 1;
6297 path = btrfs_alloc_path();
6299 return ERR_PTR(-ENOMEM);
6301 inode = new_inode(fs_info->sb);
6303 btrfs_free_path(path);
6304 return ERR_PTR(-ENOMEM);
6308 * O_TMPFILE, set link count to 0, so that after this point,
6309 * we fill in an inode item with the correct link count.
6312 set_nlink(inode, 0);
6315 * we have to initialize this early, so we can reclaim the inode
6316 * number if we fail afterwards in this function.
6318 inode->i_ino = objectid;
6321 trace_btrfs_inode_request(dir);
6323 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6325 btrfs_free_path(path);
6327 return ERR_PTR(ret);
6333 * index_cnt is ignored for everything but a dir,
6334 * btrfs_get_inode_index_count has an explanation for the magic
6337 BTRFS_I(inode)->index_cnt = 2;
6338 BTRFS_I(inode)->dir_index = *index;
6339 BTRFS_I(inode)->root = root;
6340 BTRFS_I(inode)->generation = trans->transid;
6341 inode->i_generation = BTRFS_I(inode)->generation;
6344 * We could have gotten an inode number from somebody who was fsynced
6345 * and then removed in this same transaction, so let's just set full
6346 * sync since it will be a full sync anyway and this will blow away the
6347 * old info in the log.
6349 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6351 key[0].objectid = objectid;
6352 key[0].type = BTRFS_INODE_ITEM_KEY;
6355 sizes[0] = sizeof(struct btrfs_inode_item);
6359 * Start new inodes with an inode_ref. This is slightly more
6360 * efficient for small numbers of hard links since they will
6361 * be packed into one item. Extended refs will kick in if we
6362 * add more hard links than can fit in the ref item.
6364 key[1].objectid = objectid;
6365 key[1].type = BTRFS_INODE_REF_KEY;
6366 key[1].offset = ref_objectid;
6368 sizes[1] = name_len + sizeof(*ref);
6371 location = &BTRFS_I(inode)->location;
6372 location->objectid = objectid;
6373 location->offset = 0;
6374 location->type = BTRFS_INODE_ITEM_KEY;
6376 ret = btrfs_insert_inode_locked(inode);
6380 path->leave_spinning = 1;
6381 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6385 inode_init_owner(inode, dir, mode);
6386 inode_set_bytes(inode, 0);
6388 inode->i_mtime = current_time(inode);
6389 inode->i_atime = inode->i_mtime;
6390 inode->i_ctime = inode->i_mtime;
6391 BTRFS_I(inode)->i_otime = inode->i_mtime;
6393 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6394 struct btrfs_inode_item);
6395 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6396 sizeof(*inode_item));
6397 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6400 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6401 struct btrfs_inode_ref);
6402 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6403 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6404 ptr = (unsigned long)(ref + 1);
6405 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6408 btrfs_mark_buffer_dirty(path->nodes[0]);
6409 btrfs_free_path(path);
6411 btrfs_inherit_iflags(inode, dir);
6413 if (S_ISREG(mode)) {
6414 if (btrfs_test_opt(fs_info, NODATASUM))
6415 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6416 if (btrfs_test_opt(fs_info, NODATACOW))
6417 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6418 BTRFS_INODE_NODATASUM;
6421 inode_tree_add(inode);
6423 trace_btrfs_inode_new(inode);
6424 btrfs_set_inode_last_trans(trans, inode);
6426 btrfs_update_root_times(trans, root);
6428 ret = btrfs_inode_inherit_props(trans, inode, dir);
6431 "error inheriting props for ino %llu (root %llu): %d",
6432 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6437 unlock_new_inode(inode);
6440 BTRFS_I(dir)->index_cnt--;
6441 btrfs_free_path(path);
6443 return ERR_PTR(ret);
6446 static inline u8 btrfs_inode_type(struct inode *inode)
6448 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6452 * utility function to add 'inode' into 'parent_inode' with
6453 * a give name and a given sequence number.
6454 * if 'add_backref' is true, also insert a backref from the
6455 * inode to the parent directory.
6457 int btrfs_add_link(struct btrfs_trans_handle *trans,
6458 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6459 const char *name, int name_len, int add_backref, u64 index)
6461 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6463 struct btrfs_key key;
6464 struct btrfs_root *root = parent_inode->root;
6465 u64 ino = btrfs_ino(inode);
6466 u64 parent_ino = btrfs_ino(parent_inode);
6468 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6469 memcpy(&key, &inode->root->root_key, sizeof(key));
6472 key.type = BTRFS_INODE_ITEM_KEY;
6476 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6477 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6478 root->root_key.objectid, parent_ino,
6479 index, name, name_len);
6480 } else if (add_backref) {
6481 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6485 /* Nothing to clean up yet */
6489 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6491 btrfs_inode_type(&inode->vfs_inode), index);
6492 if (ret == -EEXIST || ret == -EOVERFLOW)
6495 btrfs_abort_transaction(trans, ret);
6499 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6501 inode_inc_iversion(&parent_inode->vfs_inode);
6502 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6503 current_time(&parent_inode->vfs_inode);
6504 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6506 btrfs_abort_transaction(trans, ret);
6510 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6513 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6514 root->root_key.objectid, parent_ino,
6515 &local_index, name, name_len);
6517 } else if (add_backref) {
6521 err = btrfs_del_inode_ref(trans, root, name, name_len,
6522 ino, parent_ino, &local_index);
6527 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6528 struct btrfs_inode *dir, struct dentry *dentry,
6529 struct btrfs_inode *inode, int backref, u64 index)
6531 int err = btrfs_add_link(trans, dir, inode,
6532 dentry->d_name.name, dentry->d_name.len,
6539 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6540 umode_t mode, dev_t rdev)
6542 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6543 struct btrfs_trans_handle *trans;
6544 struct btrfs_root *root = BTRFS_I(dir)->root;
6545 struct inode *inode = NULL;
6552 * 2 for inode item and ref
6554 * 1 for xattr if selinux is on
6556 trans = btrfs_start_transaction(root, 5);
6558 return PTR_ERR(trans);
6560 err = btrfs_find_free_ino(root, &objectid);
6564 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6565 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6567 if (IS_ERR(inode)) {
6568 err = PTR_ERR(inode);
6573 * If the active LSM wants to access the inode during
6574 * d_instantiate it needs these. Smack checks to see
6575 * if the filesystem supports xattrs by looking at the
6578 inode->i_op = &btrfs_special_inode_operations;
6579 init_special_inode(inode, inode->i_mode, rdev);
6581 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6583 goto out_unlock_inode;
6585 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6588 goto out_unlock_inode;
6590 btrfs_update_inode(trans, root, inode);
6591 unlock_new_inode(inode);
6592 d_instantiate(dentry, inode);
6596 btrfs_end_transaction(trans);
6597 btrfs_balance_delayed_items(fs_info);
6598 btrfs_btree_balance_dirty(fs_info);
6600 inode_dec_link_count(inode);
6607 unlock_new_inode(inode);
6612 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6613 umode_t mode, bool excl)
6615 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6616 struct btrfs_trans_handle *trans;
6617 struct btrfs_root *root = BTRFS_I(dir)->root;
6618 struct inode *inode = NULL;
6619 int drop_inode_on_err = 0;
6625 * 2 for inode item and ref
6627 * 1 for xattr if selinux is on
6629 trans = btrfs_start_transaction(root, 5);
6631 return PTR_ERR(trans);
6633 err = btrfs_find_free_ino(root, &objectid);
6637 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6638 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6640 if (IS_ERR(inode)) {
6641 err = PTR_ERR(inode);
6644 drop_inode_on_err = 1;
6646 * If the active LSM wants to access the inode during
6647 * d_instantiate it needs these. Smack checks to see
6648 * if the filesystem supports xattrs by looking at the
6651 inode->i_fop = &btrfs_file_operations;
6652 inode->i_op = &btrfs_file_inode_operations;
6653 inode->i_mapping->a_ops = &btrfs_aops;
6655 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6657 goto out_unlock_inode;
6659 err = btrfs_update_inode(trans, root, inode);
6661 goto out_unlock_inode;
6663 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6666 goto out_unlock_inode;
6668 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6669 unlock_new_inode(inode);
6670 d_instantiate(dentry, inode);
6673 btrfs_end_transaction(trans);
6674 if (err && drop_inode_on_err) {
6675 inode_dec_link_count(inode);
6678 btrfs_balance_delayed_items(fs_info);
6679 btrfs_btree_balance_dirty(fs_info);
6683 unlock_new_inode(inode);
6688 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6689 struct dentry *dentry)
6691 struct btrfs_trans_handle *trans = NULL;
6692 struct btrfs_root *root = BTRFS_I(dir)->root;
6693 struct inode *inode = d_inode(old_dentry);
6694 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6699 /* do not allow sys_link's with other subvols of the same device */
6700 if (root->objectid != BTRFS_I(inode)->root->objectid)
6703 if (inode->i_nlink >= BTRFS_LINK_MAX)
6706 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6711 * 2 items for inode and inode ref
6712 * 2 items for dir items
6713 * 1 item for parent inode
6715 trans = btrfs_start_transaction(root, 5);
6716 if (IS_ERR(trans)) {
6717 err = PTR_ERR(trans);
6722 /* There are several dir indexes for this inode, clear the cache. */
6723 BTRFS_I(inode)->dir_index = 0ULL;
6725 inode_inc_iversion(inode);
6726 inode->i_ctime = current_time(inode);
6728 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6730 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6736 struct dentry *parent = dentry->d_parent;
6737 err = btrfs_update_inode(trans, root, inode);
6740 if (inode->i_nlink == 1) {
6742 * If new hard link count is 1, it's a file created
6743 * with open(2) O_TMPFILE flag.
6745 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6749 d_instantiate(dentry, inode);
6750 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6753 btrfs_balance_delayed_items(fs_info);
6756 btrfs_end_transaction(trans);
6758 inode_dec_link_count(inode);
6761 btrfs_btree_balance_dirty(fs_info);
6765 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6767 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6768 struct inode *inode = NULL;
6769 struct btrfs_trans_handle *trans;
6770 struct btrfs_root *root = BTRFS_I(dir)->root;
6772 int drop_on_err = 0;
6777 * 2 items for inode and ref
6778 * 2 items for dir items
6779 * 1 for xattr if selinux is on
6781 trans = btrfs_start_transaction(root, 5);
6783 return PTR_ERR(trans);
6785 err = btrfs_find_free_ino(root, &objectid);
6789 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6790 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6791 S_IFDIR | mode, &index);
6792 if (IS_ERR(inode)) {
6793 err = PTR_ERR(inode);
6798 /* these must be set before we unlock the inode */
6799 inode->i_op = &btrfs_dir_inode_operations;
6800 inode->i_fop = &btrfs_dir_file_operations;
6802 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6804 goto out_fail_inode;
6806 btrfs_i_size_write(BTRFS_I(inode), 0);
6807 err = btrfs_update_inode(trans, root, inode);
6809 goto out_fail_inode;
6811 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6812 dentry->d_name.name,
6813 dentry->d_name.len, 0, index);
6815 goto out_fail_inode;
6817 d_instantiate(dentry, inode);
6819 * mkdir is special. We're unlocking after we call d_instantiate
6820 * to avoid a race with nfsd calling d_instantiate.
6822 unlock_new_inode(inode);
6826 btrfs_end_transaction(trans);
6828 inode_dec_link_count(inode);
6831 btrfs_balance_delayed_items(fs_info);
6832 btrfs_btree_balance_dirty(fs_info);
6836 unlock_new_inode(inode);
6840 /* Find next extent map of a given extent map, caller needs to ensure locks */
6841 static struct extent_map *next_extent_map(struct extent_map *em)
6843 struct rb_node *next;
6845 next = rb_next(&em->rb_node);
6848 return container_of(next, struct extent_map, rb_node);
6851 static struct extent_map *prev_extent_map(struct extent_map *em)
6853 struct rb_node *prev;
6855 prev = rb_prev(&em->rb_node);
6858 return container_of(prev, struct extent_map, rb_node);
6861 /* helper for btfs_get_extent. Given an existing extent in the tree,
6862 * the existing extent is the nearest extent to map_start,
6863 * and an extent that you want to insert, deal with overlap and insert
6864 * the best fitted new extent into the tree.
6866 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6867 struct extent_map *existing,
6868 struct extent_map *em,
6871 struct extent_map *prev;
6872 struct extent_map *next;
6877 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6879 if (existing->start > map_start) {
6881 prev = prev_extent_map(next);
6884 next = next_extent_map(prev);
6887 start = prev ? extent_map_end(prev) : em->start;
6888 start = max_t(u64, start, em->start);
6889 end = next ? next->start : extent_map_end(em);
6890 end = min_t(u64, end, extent_map_end(em));
6891 start_diff = start - em->start;
6893 em->len = end - start;
6894 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6895 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6896 em->block_start += start_diff;
6897 em->block_len -= start_diff;
6899 return add_extent_mapping(em_tree, em, 0);
6902 static noinline int uncompress_inline(struct btrfs_path *path,
6904 size_t pg_offset, u64 extent_offset,
6905 struct btrfs_file_extent_item *item)
6908 struct extent_buffer *leaf = path->nodes[0];
6911 unsigned long inline_size;
6915 WARN_ON(pg_offset != 0);
6916 compress_type = btrfs_file_extent_compression(leaf, item);
6917 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6918 inline_size = btrfs_file_extent_inline_item_len(leaf,
6919 btrfs_item_nr(path->slots[0]));
6920 tmp = kmalloc(inline_size, GFP_NOFS);
6923 ptr = btrfs_file_extent_inline_start(item);
6925 read_extent_buffer(leaf, tmp, ptr, inline_size);
6927 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6928 ret = btrfs_decompress(compress_type, tmp, page,
6929 extent_offset, inline_size, max_size);
6932 * decompression code contains a memset to fill in any space between the end
6933 * of the uncompressed data and the end of max_size in case the decompressed
6934 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6935 * the end of an inline extent and the beginning of the next block, so we
6936 * cover that region here.
6939 if (max_size + pg_offset < PAGE_SIZE) {
6940 char *map = kmap(page);
6941 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6949 * a bit scary, this does extent mapping from logical file offset to the disk.
6950 * the ugly parts come from merging extents from the disk with the in-ram
6951 * representation. This gets more complex because of the data=ordered code,
6952 * where the in-ram extents might be locked pending data=ordered completion.
6954 * This also copies inline extents directly into the page.
6956 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6958 size_t pg_offset, u64 start, u64 len,
6961 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6964 u64 extent_start = 0;
6966 u64 objectid = btrfs_ino(inode);
6968 struct btrfs_path *path = NULL;
6969 struct btrfs_root *root = inode->root;
6970 struct btrfs_file_extent_item *item;
6971 struct extent_buffer *leaf;
6972 struct btrfs_key found_key;
6973 struct extent_map *em = NULL;
6974 struct extent_map_tree *em_tree = &inode->extent_tree;
6975 struct extent_io_tree *io_tree = &inode->io_tree;
6976 struct btrfs_trans_handle *trans = NULL;
6977 const bool new_inline = !page || create;
6980 read_lock(&em_tree->lock);
6981 em = lookup_extent_mapping(em_tree, start, len);
6983 em->bdev = fs_info->fs_devices->latest_bdev;
6984 read_unlock(&em_tree->lock);
6987 if (em->start > start || em->start + em->len <= start)
6988 free_extent_map(em);
6989 else if (em->block_start == EXTENT_MAP_INLINE && page)
6990 free_extent_map(em);
6994 em = alloc_extent_map();
6999 em->bdev = fs_info->fs_devices->latest_bdev;
7000 em->start = EXTENT_MAP_HOLE;
7001 em->orig_start = EXTENT_MAP_HOLE;
7003 em->block_len = (u64)-1;
7006 path = btrfs_alloc_path();
7012 * Chances are we'll be called again, so go ahead and do
7015 path->reada = READA_FORWARD;
7018 ret = btrfs_lookup_file_extent(trans, root, path,
7019 objectid, start, trans != NULL);
7026 if (path->slots[0] == 0)
7031 leaf = path->nodes[0];
7032 item = btrfs_item_ptr(leaf, path->slots[0],
7033 struct btrfs_file_extent_item);
7034 /* are we inside the extent that was found? */
7035 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7036 found_type = found_key.type;
7037 if (found_key.objectid != objectid ||
7038 found_type != BTRFS_EXTENT_DATA_KEY) {
7040 * If we backup past the first extent we want to move forward
7041 * and see if there is an extent in front of us, otherwise we'll
7042 * say there is a hole for our whole search range which can
7049 found_type = btrfs_file_extent_type(leaf, item);
7050 extent_start = found_key.offset;
7051 if (found_type == BTRFS_FILE_EXTENT_REG ||
7052 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7053 extent_end = extent_start +
7054 btrfs_file_extent_num_bytes(leaf, item);
7056 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7058 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7060 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7061 extent_end = ALIGN(extent_start + size,
7062 fs_info->sectorsize);
7064 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7069 if (start >= extent_end) {
7071 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7072 ret = btrfs_next_leaf(root, path);
7079 leaf = path->nodes[0];
7081 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7082 if (found_key.objectid != objectid ||
7083 found_key.type != BTRFS_EXTENT_DATA_KEY)
7085 if (start + len <= found_key.offset)
7087 if (start > found_key.offset)
7090 em->orig_start = start;
7091 em->len = found_key.offset - start;
7095 btrfs_extent_item_to_extent_map(inode, path, item,
7098 if (found_type == BTRFS_FILE_EXTENT_REG ||
7099 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7101 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7105 size_t extent_offset;
7111 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7112 extent_offset = page_offset(page) + pg_offset - extent_start;
7113 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7114 size - extent_offset);
7115 em->start = extent_start + extent_offset;
7116 em->len = ALIGN(copy_size, fs_info->sectorsize);
7117 em->orig_block_len = em->len;
7118 em->orig_start = em->start;
7119 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7120 if (create == 0 && !PageUptodate(page)) {
7121 if (btrfs_file_extent_compression(leaf, item) !=
7122 BTRFS_COMPRESS_NONE) {
7123 ret = uncompress_inline(path, page, pg_offset,
7124 extent_offset, item);
7131 read_extent_buffer(leaf, map + pg_offset, ptr,
7133 if (pg_offset + copy_size < PAGE_SIZE) {
7134 memset(map + pg_offset + copy_size, 0,
7135 PAGE_SIZE - pg_offset -
7140 flush_dcache_page(page);
7141 } else if (create && PageUptodate(page)) {
7145 free_extent_map(em);
7148 btrfs_release_path(path);
7149 trans = btrfs_join_transaction(root);
7152 return ERR_CAST(trans);
7156 write_extent_buffer(leaf, map + pg_offset, ptr,
7159 btrfs_mark_buffer_dirty(leaf);
7161 set_extent_uptodate(io_tree, em->start,
7162 extent_map_end(em) - 1, NULL, GFP_NOFS);
7167 em->orig_start = start;
7170 em->block_start = EXTENT_MAP_HOLE;
7171 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7173 btrfs_release_path(path);
7174 if (em->start > start || extent_map_end(em) <= start) {
7176 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7177 em->start, em->len, start, len);
7183 write_lock(&em_tree->lock);
7184 ret = add_extent_mapping(em_tree, em, 0);
7185 /* it is possible that someone inserted the extent into the tree
7186 * while we had the lock dropped. It is also possible that
7187 * an overlapping map exists in the tree
7189 if (ret == -EEXIST) {
7190 struct extent_map *existing;
7194 existing = search_extent_mapping(em_tree, start, len);
7196 * existing will always be non-NULL, since there must be
7197 * extent causing the -EEXIST.
7199 if (existing->start == em->start &&
7200 extent_map_end(existing) >= extent_map_end(em) &&
7201 em->block_start == existing->block_start) {
7203 * The existing extent map already encompasses the
7204 * entire extent map we tried to add.
7206 free_extent_map(em);
7210 } else if (start >= extent_map_end(existing) ||
7211 start <= existing->start) {
7213 * The existing extent map is the one nearest to
7214 * the [start, start + len) range which overlaps
7216 err = merge_extent_mapping(em_tree, existing,
7218 free_extent_map(existing);
7220 free_extent_map(em);
7224 free_extent_map(em);
7229 write_unlock(&em_tree->lock);
7232 trace_btrfs_get_extent(root, inode, em);
7234 btrfs_free_path(path);
7236 ret = btrfs_end_transaction(trans);
7241 free_extent_map(em);
7242 return ERR_PTR(err);
7244 BUG_ON(!em); /* Error is always set */
7248 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7250 size_t pg_offset, u64 start, u64 len,
7253 struct extent_map *em;
7254 struct extent_map *hole_em = NULL;
7255 u64 range_start = start;
7261 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7265 * If our em maps to:
7267 * - a pre-alloc extent,
7268 * there might actually be delalloc bytes behind it.
7270 if (em->block_start != EXTENT_MAP_HOLE &&
7271 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7276 /* check to see if we've wrapped (len == -1 or similar) */
7285 /* ok, we didn't find anything, lets look for delalloc */
7286 found = count_range_bits(&inode->io_tree, &range_start,
7287 end, len, EXTENT_DELALLOC, 1);
7288 found_end = range_start + found;
7289 if (found_end < range_start)
7290 found_end = (u64)-1;
7293 * we didn't find anything useful, return
7294 * the original results from get_extent()
7296 if (range_start > end || found_end <= start) {
7302 /* adjust the range_start to make sure it doesn't
7303 * go backwards from the start they passed in
7305 range_start = max(start, range_start);
7306 found = found_end - range_start;
7309 u64 hole_start = start;
7312 em = alloc_extent_map();
7318 * when btrfs_get_extent can't find anything it
7319 * returns one huge hole
7321 * make sure what it found really fits our range, and
7322 * adjust to make sure it is based on the start from
7326 u64 calc_end = extent_map_end(hole_em);
7328 if (calc_end <= start || (hole_em->start > end)) {
7329 free_extent_map(hole_em);
7332 hole_start = max(hole_em->start, start);
7333 hole_len = calc_end - hole_start;
7337 if (hole_em && range_start > hole_start) {
7338 /* our hole starts before our delalloc, so we
7339 * have to return just the parts of the hole
7340 * that go until the delalloc starts
7342 em->len = min(hole_len,
7343 range_start - hole_start);
7344 em->start = hole_start;
7345 em->orig_start = hole_start;
7347 * don't adjust block start at all,
7348 * it is fixed at EXTENT_MAP_HOLE
7350 em->block_start = hole_em->block_start;
7351 em->block_len = hole_len;
7352 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7353 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7355 em->start = range_start;
7357 em->orig_start = range_start;
7358 em->block_start = EXTENT_MAP_DELALLOC;
7359 em->block_len = found;
7361 } else if (hole_em) {
7366 free_extent_map(hole_em);
7368 free_extent_map(em);
7369 return ERR_PTR(err);
7374 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7377 const u64 orig_start,
7378 const u64 block_start,
7379 const u64 block_len,
7380 const u64 orig_block_len,
7381 const u64 ram_bytes,
7384 struct extent_map *em = NULL;
7387 if (type != BTRFS_ORDERED_NOCOW) {
7388 em = create_io_em(inode, start, len, orig_start,
7389 block_start, block_len, orig_block_len,
7391 BTRFS_COMPRESS_NONE, /* compress_type */
7396 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7397 len, block_len, type);
7400 free_extent_map(em);
7401 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7402 start + len - 1, 0);
7411 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7414 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7415 struct btrfs_root *root = BTRFS_I(inode)->root;
7416 struct extent_map *em;
7417 struct btrfs_key ins;
7421 alloc_hint = get_extent_allocation_hint(inode, start, len);
7422 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7423 0, alloc_hint, &ins, 1, 1);
7425 return ERR_PTR(ret);
7427 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7428 ins.objectid, ins.offset, ins.offset,
7429 ins.offset, BTRFS_ORDERED_REGULAR);
7430 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7432 btrfs_free_reserved_extent(fs_info, ins.objectid,
7439 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7440 * block must be cow'd
7442 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7443 u64 *orig_start, u64 *orig_block_len,
7446 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7447 struct btrfs_path *path;
7449 struct extent_buffer *leaf;
7450 struct btrfs_root *root = BTRFS_I(inode)->root;
7451 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7452 struct btrfs_file_extent_item *fi;
7453 struct btrfs_key key;
7460 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7462 path = btrfs_alloc_path();
7466 ret = btrfs_lookup_file_extent(NULL, root, path,
7467 btrfs_ino(BTRFS_I(inode)), offset, 0);
7471 slot = path->slots[0];
7474 /* can't find the item, must cow */
7481 leaf = path->nodes[0];
7482 btrfs_item_key_to_cpu(leaf, &key, slot);
7483 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7484 key.type != BTRFS_EXTENT_DATA_KEY) {
7485 /* not our file or wrong item type, must cow */
7489 if (key.offset > offset) {
7490 /* Wrong offset, must cow */
7494 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7495 found_type = btrfs_file_extent_type(leaf, fi);
7496 if (found_type != BTRFS_FILE_EXTENT_REG &&
7497 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7498 /* not a regular extent, must cow */
7502 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7505 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7506 if (extent_end <= offset)
7509 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7510 if (disk_bytenr == 0)
7513 if (btrfs_file_extent_compression(leaf, fi) ||
7514 btrfs_file_extent_encryption(leaf, fi) ||
7515 btrfs_file_extent_other_encoding(leaf, fi))
7518 backref_offset = btrfs_file_extent_offset(leaf, fi);
7521 *orig_start = key.offset - backref_offset;
7522 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7523 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7526 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7529 num_bytes = min(offset + *len, extent_end) - offset;
7530 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7533 range_end = round_up(offset + num_bytes,
7534 root->fs_info->sectorsize) - 1;
7535 ret = test_range_bit(io_tree, offset, range_end,
7536 EXTENT_DELALLOC, 0, NULL);
7543 btrfs_release_path(path);
7546 * look for other files referencing this extent, if we
7547 * find any we must cow
7550 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7551 key.offset - backref_offset, disk_bytenr);
7558 * adjust disk_bytenr and num_bytes to cover just the bytes
7559 * in this extent we are about to write. If there
7560 * are any csums in that range we have to cow in order
7561 * to keep the csums correct
7563 disk_bytenr += backref_offset;
7564 disk_bytenr += offset - key.offset;
7565 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7568 * all of the above have passed, it is safe to overwrite this extent
7574 btrfs_free_path(path);
7578 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7580 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7582 void **pagep = NULL;
7583 struct page *page = NULL;
7584 unsigned long start_idx;
7585 unsigned long end_idx;
7587 start_idx = start >> PAGE_SHIFT;
7590 * end is the last byte in the last page. end == start is legal
7592 end_idx = end >> PAGE_SHIFT;
7596 /* Most of the code in this while loop is lifted from
7597 * find_get_page. It's been modified to begin searching from a
7598 * page and return just the first page found in that range. If the
7599 * found idx is less than or equal to the end idx then we know that
7600 * a page exists. If no pages are found or if those pages are
7601 * outside of the range then we're fine (yay!) */
7602 while (page == NULL &&
7603 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7604 page = radix_tree_deref_slot(pagep);
7605 if (unlikely(!page))
7608 if (radix_tree_exception(page)) {
7609 if (radix_tree_deref_retry(page)) {
7614 * Otherwise, shmem/tmpfs must be storing a swap entry
7615 * here as an exceptional entry: so return it without
7616 * attempting to raise page count.
7619 break; /* TODO: Is this relevant for this use case? */
7622 if (!page_cache_get_speculative(page)) {
7628 * Has the page moved?
7629 * This is part of the lockless pagecache protocol. See
7630 * include/linux/pagemap.h for details.
7632 if (unlikely(page != *pagep)) {
7639 if (page->index <= end_idx)
7648 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7649 struct extent_state **cached_state, int writing)
7651 struct btrfs_ordered_extent *ordered;
7655 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7658 * We're concerned with the entire range that we're going to be
7659 * doing DIO to, so we need to make sure there's no ordered
7660 * extents in this range.
7662 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7663 lockend - lockstart + 1);
7666 * We need to make sure there are no buffered pages in this
7667 * range either, we could have raced between the invalidate in
7668 * generic_file_direct_write and locking the extent. The
7669 * invalidate needs to happen so that reads after a write do not
7674 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7677 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7678 cached_state, GFP_NOFS);
7682 * If we are doing a DIO read and the ordered extent we
7683 * found is for a buffered write, we can not wait for it
7684 * to complete and retry, because if we do so we can
7685 * deadlock with concurrent buffered writes on page
7686 * locks. This happens only if our DIO read covers more
7687 * than one extent map, if at this point has already
7688 * created an ordered extent for a previous extent map
7689 * and locked its range in the inode's io tree, and a
7690 * concurrent write against that previous extent map's
7691 * range and this range started (we unlock the ranges
7692 * in the io tree only when the bios complete and
7693 * buffered writes always lock pages before attempting
7694 * to lock range in the io tree).
7697 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7698 btrfs_start_ordered_extent(inode, ordered, 1);
7701 btrfs_put_ordered_extent(ordered);
7704 * We could trigger writeback for this range (and wait
7705 * for it to complete) and then invalidate the pages for
7706 * this range (through invalidate_inode_pages2_range()),
7707 * but that can lead us to a deadlock with a concurrent
7708 * call to readpages() (a buffered read or a defrag call
7709 * triggered a readahead) on a page lock due to an
7710 * ordered dio extent we created before but did not have
7711 * yet a corresponding bio submitted (whence it can not
7712 * complete), which makes readpages() wait for that
7713 * ordered extent to complete while holding a lock on
7728 /* The callers of this must take lock_extent() */
7729 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7730 u64 orig_start, u64 block_start,
7731 u64 block_len, u64 orig_block_len,
7732 u64 ram_bytes, int compress_type,
7735 struct extent_map_tree *em_tree;
7736 struct extent_map *em;
7737 struct btrfs_root *root = BTRFS_I(inode)->root;
7740 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7741 type == BTRFS_ORDERED_COMPRESSED ||
7742 type == BTRFS_ORDERED_NOCOW ||
7743 type == BTRFS_ORDERED_REGULAR);
7745 em_tree = &BTRFS_I(inode)->extent_tree;
7746 em = alloc_extent_map();
7748 return ERR_PTR(-ENOMEM);
7751 em->orig_start = orig_start;
7753 em->block_len = block_len;
7754 em->block_start = block_start;
7755 em->bdev = root->fs_info->fs_devices->latest_bdev;
7756 em->orig_block_len = orig_block_len;
7757 em->ram_bytes = ram_bytes;
7758 em->generation = -1;
7759 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7760 if (type == BTRFS_ORDERED_PREALLOC) {
7761 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7762 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7763 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7764 em->compress_type = compress_type;
7768 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7769 em->start + em->len - 1, 0);
7770 write_lock(&em_tree->lock);
7771 ret = add_extent_mapping(em_tree, em, 1);
7772 write_unlock(&em_tree->lock);
7774 * The caller has taken lock_extent(), who could race with us
7777 } while (ret == -EEXIST);
7780 free_extent_map(em);
7781 return ERR_PTR(ret);
7784 /* em got 2 refs now, callers needs to do free_extent_map once. */
7788 static void adjust_dio_outstanding_extents(struct inode *inode,
7789 struct btrfs_dio_data *dio_data,
7792 unsigned num_extents = count_max_extents(len);
7795 * If we have an outstanding_extents count still set then we're
7796 * within our reservation, otherwise we need to adjust our inode
7797 * counter appropriately.
7799 if (dio_data->outstanding_extents >= num_extents) {
7800 dio_data->outstanding_extents -= num_extents;
7803 * If dio write length has been split due to no large enough
7804 * contiguous space, we need to compensate our inode counter
7807 u64 num_needed = num_extents - dio_data->outstanding_extents;
7809 spin_lock(&BTRFS_I(inode)->lock);
7810 BTRFS_I(inode)->outstanding_extents += num_needed;
7811 spin_unlock(&BTRFS_I(inode)->lock);
7815 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7816 struct buffer_head *bh_result, int create)
7818 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7819 struct extent_map *em;
7820 struct extent_state *cached_state = NULL;
7821 struct btrfs_dio_data *dio_data = NULL;
7822 u64 start = iblock << inode->i_blkbits;
7823 u64 lockstart, lockend;
7824 u64 len = bh_result->b_size;
7825 int unlock_bits = EXTENT_LOCKED;
7829 unlock_bits |= EXTENT_DIRTY;
7831 len = min_t(u64, len, fs_info->sectorsize);
7834 lockend = start + len - 1;
7836 if (current->journal_info) {
7838 * Need to pull our outstanding extents and set journal_info to NULL so
7839 * that anything that needs to check if there's a transaction doesn't get
7842 dio_data = current->journal_info;
7843 current->journal_info = NULL;
7847 * If this errors out it's because we couldn't invalidate pagecache for
7848 * this range and we need to fallback to buffered.
7850 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7856 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7863 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7864 * io. INLINE is special, and we could probably kludge it in here, but
7865 * it's still buffered so for safety lets just fall back to the generic
7868 * For COMPRESSED we _have_ to read the entire extent in so we can
7869 * decompress it, so there will be buffering required no matter what we
7870 * do, so go ahead and fallback to buffered.
7872 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7873 * to buffered IO. Don't blame me, this is the price we pay for using
7876 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7877 em->block_start == EXTENT_MAP_INLINE) {
7878 free_extent_map(em);
7883 /* Just a good old fashioned hole, return */
7884 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7885 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7886 free_extent_map(em);
7891 * We don't allocate a new extent in the following cases
7893 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7895 * 2) The extent is marked as PREALLOC. We're good to go here and can
7896 * just use the extent.
7900 len = min(len, em->len - (start - em->start));
7901 lockstart = start + len;
7905 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7906 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7907 em->block_start != EXTENT_MAP_HOLE)) {
7909 u64 block_start, orig_start, orig_block_len, ram_bytes;
7911 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7912 type = BTRFS_ORDERED_PREALLOC;
7914 type = BTRFS_ORDERED_NOCOW;
7915 len = min(len, em->len - (start - em->start));
7916 block_start = em->block_start + (start - em->start);
7918 if (can_nocow_extent(inode, start, &len, &orig_start,
7919 &orig_block_len, &ram_bytes) == 1 &&
7920 btrfs_inc_nocow_writers(fs_info, block_start)) {
7921 struct extent_map *em2;
7923 em2 = btrfs_create_dio_extent(inode, start, len,
7924 orig_start, block_start,
7925 len, orig_block_len,
7927 btrfs_dec_nocow_writers(fs_info, block_start);
7928 if (type == BTRFS_ORDERED_PREALLOC) {
7929 free_extent_map(em);
7932 if (em2 && IS_ERR(em2)) {
7937 * For inode marked NODATACOW or extent marked PREALLOC,
7938 * use the existing or preallocated extent, so does not
7939 * need to adjust btrfs_space_info's bytes_may_use.
7941 btrfs_free_reserved_data_space_noquota(inode,
7948 * this will cow the extent, reset the len in case we changed
7951 len = bh_result->b_size;
7952 free_extent_map(em);
7953 em = btrfs_new_extent_direct(inode, start, len);
7958 len = min(len, em->len - (start - em->start));
7960 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7962 bh_result->b_size = len;
7963 bh_result->b_bdev = em->bdev;
7964 set_buffer_mapped(bh_result);
7966 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7967 set_buffer_new(bh_result);
7970 * Need to update the i_size under the extent lock so buffered
7971 * readers will get the updated i_size when we unlock.
7973 if (!dio_data->overwrite && start + len > i_size_read(inode))
7974 i_size_write(inode, start + len);
7976 adjust_dio_outstanding_extents(inode, dio_data, len);
7977 WARN_ON(dio_data->reserve < len);
7978 dio_data->reserve -= len;
7979 dio_data->unsubmitted_oe_range_end = start + len;
7980 current->journal_info = dio_data;
7984 * In the case of write we need to clear and unlock the entire range,
7985 * in the case of read we need to unlock only the end area that we
7986 * aren't using if there is any left over space.
7988 if (lockstart < lockend) {
7989 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7990 lockend, unlock_bits, 1, 0,
7991 &cached_state, GFP_NOFS);
7993 free_extent_state(cached_state);
7996 free_extent_map(em);
8001 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
8002 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
8005 current->journal_info = dio_data;
8007 * Compensate the delalloc release we do in btrfs_direct_IO() when we
8008 * write less data then expected, so that we don't underflow our inode's
8009 * outstanding extents counter.
8011 if (create && dio_data)
8012 adjust_dio_outstanding_extents(inode, dio_data, len);
8017 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
8021 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8024 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
8028 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
8032 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
8038 static int btrfs_check_dio_repairable(struct inode *inode,
8039 struct bio *failed_bio,
8040 struct io_failure_record *failrec,
8043 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8046 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
8047 if (num_copies == 1) {
8049 * we only have a single copy of the data, so don't bother with
8050 * all the retry and error correction code that follows. no
8051 * matter what the error is, it is very likely to persist.
8053 btrfs_debug(fs_info,
8054 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
8055 num_copies, failrec->this_mirror, failed_mirror);
8059 failrec->failed_mirror = failed_mirror;
8060 failrec->this_mirror++;
8061 if (failrec->this_mirror == failed_mirror)
8062 failrec->this_mirror++;
8064 if (failrec->this_mirror > num_copies) {
8065 btrfs_debug(fs_info,
8066 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
8067 num_copies, failrec->this_mirror, failed_mirror);
8074 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
8075 struct page *page, unsigned int pgoff,
8076 u64 start, u64 end, int failed_mirror,
8077 bio_end_io_t *repair_endio, void *repair_arg)
8079 struct io_failure_record *failrec;
8080 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8081 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8084 unsigned int read_mode = 0;
8087 blk_status_t status;
8089 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
8091 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8093 return errno_to_blk_status(ret);
8095 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8098 free_io_failure(failure_tree, io_tree, failrec);
8099 return BLK_STS_IOERR;
8102 segs = bio_segments(failed_bio);
8104 (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8105 read_mode |= REQ_FAILFAST_DEV;
8107 isector = start - btrfs_io_bio(failed_bio)->logical;
8108 isector >>= inode->i_sb->s_blocksize_bits;
8109 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8110 pgoff, isector, repair_endio, repair_arg);
8111 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8113 btrfs_debug(BTRFS_I(inode)->root->fs_info,
8114 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8115 read_mode, failrec->this_mirror, failrec->in_validation);
8117 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8119 free_io_failure(failure_tree, io_tree, failrec);
8126 struct btrfs_retry_complete {
8127 struct completion done;
8128 struct inode *inode;
8133 static void btrfs_retry_endio_nocsum(struct bio *bio)
8135 struct btrfs_retry_complete *done = bio->bi_private;
8136 struct inode *inode = done->inode;
8137 struct bio_vec *bvec;
8138 struct extent_io_tree *io_tree, *failure_tree;
8144 ASSERT(bio->bi_vcnt == 1);
8145 io_tree = &BTRFS_I(inode)->io_tree;
8146 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8147 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8150 ASSERT(!bio_flagged(bio, BIO_CLONED));
8151 bio_for_each_segment_all(bvec, bio, i)
8152 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8153 io_tree, done->start, bvec->bv_page,
8154 btrfs_ino(BTRFS_I(inode)), 0);
8156 complete(&done->done);
8160 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8161 struct btrfs_io_bio *io_bio)
8163 struct btrfs_fs_info *fs_info;
8164 struct bio_vec bvec;
8165 struct bvec_iter iter;
8166 struct btrfs_retry_complete done;
8172 blk_status_t err = BLK_STS_OK;
8174 fs_info = BTRFS_I(inode)->root->fs_info;
8175 sectorsize = fs_info->sectorsize;
8177 start = io_bio->logical;
8179 io_bio->bio.bi_iter = io_bio->iter;
8181 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8182 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8183 pgoff = bvec.bv_offset;
8185 next_block_or_try_again:
8188 init_completion(&done.done);
8190 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8191 pgoff, start, start + sectorsize - 1,
8193 btrfs_retry_endio_nocsum, &done);
8199 wait_for_completion_io(&done.done);
8201 if (!done.uptodate) {
8202 /* We might have another mirror, so try again */
8203 goto next_block_or_try_again;
8207 start += sectorsize;
8211 pgoff += sectorsize;
8212 ASSERT(pgoff < PAGE_SIZE);
8213 goto next_block_or_try_again;
8220 static void btrfs_retry_endio(struct bio *bio)
8222 struct btrfs_retry_complete *done = bio->bi_private;
8223 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8224 struct extent_io_tree *io_tree, *failure_tree;
8225 struct inode *inode = done->inode;
8226 struct bio_vec *bvec;
8236 ASSERT(bio->bi_vcnt == 1);
8237 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8239 io_tree = &BTRFS_I(inode)->io_tree;
8240 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8242 ASSERT(!bio_flagged(bio, BIO_CLONED));
8243 bio_for_each_segment_all(bvec, bio, i) {
8244 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8245 bvec->bv_offset, done->start,
8248 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8249 failure_tree, io_tree, done->start,
8251 btrfs_ino(BTRFS_I(inode)),
8257 done->uptodate = uptodate;
8259 complete(&done->done);
8263 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8264 struct btrfs_io_bio *io_bio, blk_status_t err)
8266 struct btrfs_fs_info *fs_info;
8267 struct bio_vec bvec;
8268 struct bvec_iter iter;
8269 struct btrfs_retry_complete done;
8276 bool uptodate = (err == 0);
8278 blk_status_t status;
8280 fs_info = BTRFS_I(inode)->root->fs_info;
8281 sectorsize = fs_info->sectorsize;
8284 start = io_bio->logical;
8286 io_bio->bio.bi_iter = io_bio->iter;
8288 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8289 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8291 pgoff = bvec.bv_offset;
8294 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8295 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8296 bvec.bv_page, pgoff, start, sectorsize);
8303 init_completion(&done.done);
8305 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8306 pgoff, start, start + sectorsize - 1,
8307 io_bio->mirror_num, btrfs_retry_endio,
8314 wait_for_completion_io(&done.done);
8316 if (!done.uptodate) {
8317 /* We might have another mirror, so try again */
8321 offset += sectorsize;
8322 start += sectorsize;
8328 pgoff += sectorsize;
8329 ASSERT(pgoff < PAGE_SIZE);
8337 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8338 struct btrfs_io_bio *io_bio, blk_status_t err)
8340 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8344 return __btrfs_correct_data_nocsum(inode, io_bio);
8348 return __btrfs_subio_endio_read(inode, io_bio, err);
8352 static void btrfs_endio_direct_read(struct bio *bio)
8354 struct btrfs_dio_private *dip = bio->bi_private;
8355 struct inode *inode = dip->inode;
8356 struct bio *dio_bio;
8357 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8358 blk_status_t err = bio->bi_status;
8360 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) {
8361 err = btrfs_subio_endio_read(inode, io_bio, err);
8366 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8367 dip->logical_offset + dip->bytes - 1);
8368 dio_bio = dip->dio_bio;
8372 dio_bio->bi_status = bio->bi_status;
8373 dio_end_io(dio_bio);
8376 io_bio->end_io(io_bio, blk_status_to_errno(err));
8380 static void __endio_write_update_ordered(struct inode *inode,
8381 const u64 offset, const u64 bytes,
8382 const bool uptodate)
8384 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8385 struct btrfs_ordered_extent *ordered = NULL;
8386 struct btrfs_workqueue *wq;
8387 btrfs_work_func_t func;
8388 u64 ordered_offset = offset;
8389 u64 ordered_bytes = bytes;
8392 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8393 wq = fs_info->endio_freespace_worker;
8394 func = btrfs_freespace_write_helper;
8396 wq = fs_info->endio_write_workers;
8397 func = btrfs_endio_write_helper;
8401 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8408 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8409 btrfs_queue_work(wq, &ordered->work);
8412 * our bio might span multiple ordered extents. If we haven't
8413 * completed the accounting for the whole dio, go back and try again
8415 if (ordered_offset < offset + bytes) {
8416 ordered_bytes = offset + bytes - ordered_offset;
8422 static void btrfs_endio_direct_write(struct bio *bio)
8424 struct btrfs_dio_private *dip = bio->bi_private;
8425 struct bio *dio_bio = dip->dio_bio;
8427 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8428 dip->bytes, !bio->bi_status);
8432 dio_bio->bi_status = bio->bi_status;
8433 dio_end_io(dio_bio);
8437 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8438 struct bio *bio, int mirror_num,
8439 unsigned long bio_flags, u64 offset)
8441 struct inode *inode = private_data;
8443 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8444 BUG_ON(ret); /* -ENOMEM */
8448 static void btrfs_end_dio_bio(struct bio *bio)
8450 struct btrfs_dio_private *dip = bio->bi_private;
8451 blk_status_t err = bio->bi_status;
8454 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8455 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8456 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8458 (unsigned long long)bio->bi_iter.bi_sector,
8459 bio->bi_iter.bi_size, err);
8461 if (dip->subio_endio)
8462 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8468 * before atomic variable goto zero, we must make sure
8469 * dip->errors is perceived to be set.
8471 smp_mb__before_atomic();
8474 /* if there are more bios still pending for this dio, just exit */
8475 if (!atomic_dec_and_test(&dip->pending_bios))
8479 bio_io_error(dip->orig_bio);
8481 dip->dio_bio->bi_status = 0;
8482 bio_endio(dip->orig_bio);
8488 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8489 struct btrfs_dio_private *dip,
8493 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8494 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8498 * We load all the csum data we need when we submit
8499 * the first bio to reduce the csum tree search and
8502 if (dip->logical_offset == file_offset) {
8503 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8509 if (bio == dip->orig_bio)
8512 file_offset -= dip->logical_offset;
8513 file_offset >>= inode->i_sb->s_blocksize_bits;
8514 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8519 static inline blk_status_t
8520 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8523 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8524 struct btrfs_dio_private *dip = bio->bi_private;
8525 bool write = bio_op(bio) == REQ_OP_WRITE;
8529 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8534 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8539 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8542 if (write && async_submit) {
8543 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8545 __btrfs_submit_bio_start_direct_io,
8546 __btrfs_submit_bio_done);
8550 * If we aren't doing async submit, calculate the csum of the
8553 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8557 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8563 ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8569 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8571 struct inode *inode = dip->inode;
8572 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8574 struct bio *orig_bio = dip->orig_bio;
8575 u64 start_sector = orig_bio->bi_iter.bi_sector;
8576 u64 file_offset = dip->logical_offset;
8578 int async_submit = 0;
8580 int clone_offset = 0;
8583 blk_status_t status;
8585 map_length = orig_bio->bi_iter.bi_size;
8586 submit_len = map_length;
8587 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8588 &map_length, NULL, 0);
8592 if (map_length >= submit_len) {
8594 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8598 /* async crcs make it difficult to collect full stripe writes. */
8599 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8605 ASSERT(map_length <= INT_MAX);
8606 atomic_inc(&dip->pending_bios);
8608 clone_len = min_t(int, submit_len, map_length);
8611 * This will never fail as it's passing GPF_NOFS and
8612 * the allocation is backed by btrfs_bioset.
8614 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8616 bio->bi_private = dip;
8617 bio->bi_end_io = btrfs_end_dio_bio;
8618 btrfs_io_bio(bio)->logical = file_offset;
8620 ASSERT(submit_len >= clone_len);
8621 submit_len -= clone_len;
8622 if (submit_len == 0)
8626 * Increase the count before we submit the bio so we know
8627 * the end IO handler won't happen before we increase the
8628 * count. Otherwise, the dip might get freed before we're
8629 * done setting it up.
8631 atomic_inc(&dip->pending_bios);
8633 status = __btrfs_submit_dio_bio(bio, inode, file_offset,
8637 atomic_dec(&dip->pending_bios);
8641 clone_offset += clone_len;
8642 start_sector += clone_len >> 9;
8643 file_offset += clone_len;
8645 map_length = submit_len;
8646 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8647 start_sector << 9, &map_length, NULL, 0);
8650 } while (submit_len > 0);
8653 status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8661 * before atomic variable goto zero, we must
8662 * make sure dip->errors is perceived to be set.
8664 smp_mb__before_atomic();
8665 if (atomic_dec_and_test(&dip->pending_bios))
8666 bio_io_error(dip->orig_bio);
8668 /* bio_end_io() will handle error, so we needn't return it */
8672 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8675 struct btrfs_dio_private *dip = NULL;
8676 struct bio *bio = NULL;
8677 struct btrfs_io_bio *io_bio;
8678 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8681 bio = btrfs_bio_clone(dio_bio);
8683 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8689 dip->private = dio_bio->bi_private;
8691 dip->logical_offset = file_offset;
8692 dip->bytes = dio_bio->bi_iter.bi_size;
8693 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8694 bio->bi_private = dip;
8695 dip->orig_bio = bio;
8696 dip->dio_bio = dio_bio;
8697 atomic_set(&dip->pending_bios, 0);
8698 io_bio = btrfs_io_bio(bio);
8699 io_bio->logical = file_offset;
8702 bio->bi_end_io = btrfs_endio_direct_write;
8704 bio->bi_end_io = btrfs_endio_direct_read;
8705 dip->subio_endio = btrfs_subio_endio_read;
8709 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8710 * even if we fail to submit a bio, because in such case we do the
8711 * corresponding error handling below and it must not be done a second
8712 * time by btrfs_direct_IO().
8715 struct btrfs_dio_data *dio_data = current->journal_info;
8717 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8719 dio_data->unsubmitted_oe_range_start =
8720 dio_data->unsubmitted_oe_range_end;
8723 ret = btrfs_submit_direct_hook(dip);
8728 io_bio->end_io(io_bio, ret);
8732 * If we arrived here it means either we failed to submit the dip
8733 * or we either failed to clone the dio_bio or failed to allocate the
8734 * dip. If we cloned the dio_bio and allocated the dip, we can just
8735 * call bio_endio against our io_bio so that we get proper resource
8736 * cleanup if we fail to submit the dip, otherwise, we must do the
8737 * same as btrfs_endio_direct_[write|read] because we can't call these
8738 * callbacks - they require an allocated dip and a clone of dio_bio.
8743 * The end io callbacks free our dip, do the final put on bio
8744 * and all the cleanup and final put for dio_bio (through
8751 __endio_write_update_ordered(inode,
8753 dio_bio->bi_iter.bi_size,
8756 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8757 file_offset + dio_bio->bi_iter.bi_size - 1);
8759 dio_bio->bi_status = BLK_STS_IOERR;
8761 * Releases and cleans up our dio_bio, no need to bio_put()
8762 * nor bio_endio()/bio_io_error() against dio_bio.
8764 dio_end_io(dio_bio);
8771 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8773 const struct iov_iter *iter, loff_t offset)
8777 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8778 ssize_t retval = -EINVAL;
8780 if (offset & blocksize_mask)
8783 if (iov_iter_alignment(iter) & blocksize_mask)
8786 /* If this is a write we don't need to check anymore */
8787 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8790 * Check to make sure we don't have duplicate iov_base's in this
8791 * iovec, if so return EINVAL, otherwise we'll get csum errors
8792 * when reading back.
8794 for (seg = 0; seg < iter->nr_segs; seg++) {
8795 for (i = seg + 1; i < iter->nr_segs; i++) {
8796 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8805 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8807 struct file *file = iocb->ki_filp;
8808 struct inode *inode = file->f_mapping->host;
8809 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8810 struct btrfs_dio_data dio_data = { 0 };
8811 struct extent_changeset *data_reserved = NULL;
8812 loff_t offset = iocb->ki_pos;
8816 bool relock = false;
8819 if (check_direct_IO(fs_info, iocb, iter, offset))
8822 inode_dio_begin(inode);
8825 * The generic stuff only does filemap_write_and_wait_range, which
8826 * isn't enough if we've written compressed pages to this area, so
8827 * we need to flush the dirty pages again to make absolutely sure
8828 * that any outstanding dirty pages are on disk.
8830 count = iov_iter_count(iter);
8831 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8832 &BTRFS_I(inode)->runtime_flags))
8833 filemap_fdatawrite_range(inode->i_mapping, offset,
8834 offset + count - 1);
8836 if (iov_iter_rw(iter) == WRITE) {
8838 * If the write DIO is beyond the EOF, we need update
8839 * the isize, but it is protected by i_mutex. So we can
8840 * not unlock the i_mutex at this case.
8842 if (offset + count <= inode->i_size) {
8843 dio_data.overwrite = 1;
8844 inode_unlock(inode);
8846 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8850 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8854 dio_data.outstanding_extents = count_max_extents(count);
8857 * We need to know how many extents we reserved so that we can
8858 * do the accounting properly if we go over the number we
8859 * originally calculated. Abuse current->journal_info for this.
8861 dio_data.reserve = round_up(count,
8862 fs_info->sectorsize);
8863 dio_data.unsubmitted_oe_range_start = (u64)offset;
8864 dio_data.unsubmitted_oe_range_end = (u64)offset;
8865 current->journal_info = &dio_data;
8866 down_read(&BTRFS_I(inode)->dio_sem);
8867 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8868 &BTRFS_I(inode)->runtime_flags)) {
8869 inode_dio_end(inode);
8870 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8874 ret = __blockdev_direct_IO(iocb, inode,
8875 fs_info->fs_devices->latest_bdev,
8876 iter, btrfs_get_blocks_direct, NULL,
8877 btrfs_submit_direct, flags);
8878 if (iov_iter_rw(iter) == WRITE) {
8879 up_read(&BTRFS_I(inode)->dio_sem);
8880 current->journal_info = NULL;
8881 if (ret < 0 && ret != -EIOCBQUEUED) {
8882 if (dio_data.reserve)
8883 btrfs_delalloc_release_space(inode, data_reserved,
8884 offset, dio_data.reserve);
8886 * On error we might have left some ordered extents
8887 * without submitting corresponding bios for them, so
8888 * cleanup them up to avoid other tasks getting them
8889 * and waiting for them to complete forever.
8891 if (dio_data.unsubmitted_oe_range_start <
8892 dio_data.unsubmitted_oe_range_end)
8893 __endio_write_update_ordered(inode,
8894 dio_data.unsubmitted_oe_range_start,
8895 dio_data.unsubmitted_oe_range_end -
8896 dio_data.unsubmitted_oe_range_start,
8898 } else if (ret >= 0 && (size_t)ret < count)
8899 btrfs_delalloc_release_space(inode, data_reserved,
8900 offset, count - (size_t)ret);
8904 inode_dio_end(inode);
8908 extent_changeset_free(data_reserved);
8912 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8914 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8915 __u64 start, __u64 len)
8919 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8923 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8926 int btrfs_readpage(struct file *file, struct page *page)
8928 struct extent_io_tree *tree;
8929 tree = &BTRFS_I(page->mapping->host)->io_tree;
8930 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8933 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8935 struct extent_io_tree *tree;
8936 struct inode *inode = page->mapping->host;
8939 if (current->flags & PF_MEMALLOC) {
8940 redirty_page_for_writepage(wbc, page);
8946 * If we are under memory pressure we will call this directly from the
8947 * VM, we need to make sure we have the inode referenced for the ordered
8948 * extent. If not just return like we didn't do anything.
8950 if (!igrab(inode)) {
8951 redirty_page_for_writepage(wbc, page);
8952 return AOP_WRITEPAGE_ACTIVATE;
8954 tree = &BTRFS_I(page->mapping->host)->io_tree;
8955 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8956 btrfs_add_delayed_iput(inode);
8960 static int btrfs_writepages(struct address_space *mapping,
8961 struct writeback_control *wbc)
8963 struct extent_io_tree *tree;
8965 tree = &BTRFS_I(mapping->host)->io_tree;
8966 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8970 btrfs_readpages(struct file *file, struct address_space *mapping,
8971 struct list_head *pages, unsigned nr_pages)
8973 struct extent_io_tree *tree;
8974 tree = &BTRFS_I(mapping->host)->io_tree;
8975 return extent_readpages(tree, mapping, pages, nr_pages,
8978 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8980 struct extent_io_tree *tree;
8981 struct extent_map_tree *map;
8984 tree = &BTRFS_I(page->mapping->host)->io_tree;
8985 map = &BTRFS_I(page->mapping->host)->extent_tree;
8986 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8988 ClearPagePrivate(page);
8989 set_page_private(page, 0);
8995 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8997 if (PageWriteback(page) || PageDirty(page))
8999 return __btrfs_releasepage(page, gfp_flags);
9002 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
9003 unsigned int length)
9005 struct inode *inode = page->mapping->host;
9006 struct extent_io_tree *tree;
9007 struct btrfs_ordered_extent *ordered;
9008 struct extent_state *cached_state = NULL;
9009 u64 page_start = page_offset(page);
9010 u64 page_end = page_start + PAGE_SIZE - 1;
9013 int inode_evicting = inode->i_state & I_FREEING;
9016 * we have the page locked, so new writeback can't start,
9017 * and the dirty bit won't be cleared while we are here.
9019 * Wait for IO on this page so that we can safely clear
9020 * the PagePrivate2 bit and do ordered accounting
9022 wait_on_page_writeback(page);
9024 tree = &BTRFS_I(inode)->io_tree;
9026 btrfs_releasepage(page, GFP_NOFS);
9030 if (!inode_evicting)
9031 lock_extent_bits(tree, page_start, page_end, &cached_state);
9034 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
9035 page_end - start + 1);
9037 end = min(page_end, ordered->file_offset + ordered->len - 1);
9039 * IO on this page will never be started, so we need
9040 * to account for any ordered extents now
9042 if (!inode_evicting)
9043 clear_extent_bit(tree, start, end,
9044 EXTENT_DIRTY | EXTENT_DELALLOC |
9045 EXTENT_DELALLOC_NEW |
9046 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
9047 EXTENT_DEFRAG, 1, 0, &cached_state,
9050 * whoever cleared the private bit is responsible
9051 * for the finish_ordered_io
9053 if (TestClearPagePrivate2(page)) {
9054 struct btrfs_ordered_inode_tree *tree;
9057 tree = &BTRFS_I(inode)->ordered_tree;
9059 spin_lock_irq(&tree->lock);
9060 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
9061 new_len = start - ordered->file_offset;
9062 if (new_len < ordered->truncated_len)
9063 ordered->truncated_len = new_len;
9064 spin_unlock_irq(&tree->lock);
9066 if (btrfs_dec_test_ordered_pending(inode, &ordered,
9068 end - start + 1, 1))
9069 btrfs_finish_ordered_io(ordered);
9071 btrfs_put_ordered_extent(ordered);
9072 if (!inode_evicting) {
9073 cached_state = NULL;
9074 lock_extent_bits(tree, start, end,
9079 if (start < page_end)
9084 * Qgroup reserved space handler
9085 * Page here will be either
9086 * 1) Already written to disk
9087 * In this case, its reserved space is released from data rsv map
9088 * and will be freed by delayed_ref handler finally.
9089 * So even we call qgroup_free_data(), it won't decrease reserved
9091 * 2) Not written to disk
9092 * This means the reserved space should be freed here. However,
9093 * if a truncate invalidates the page (by clearing PageDirty)
9094 * and the page is accounted for while allocating extent
9095 * in btrfs_check_data_free_space() we let delayed_ref to
9096 * free the entire extent.
9098 if (PageDirty(page))
9099 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9100 if (!inode_evicting) {
9101 clear_extent_bit(tree, page_start, page_end,
9102 EXTENT_LOCKED | EXTENT_DIRTY |
9103 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9104 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9105 &cached_state, GFP_NOFS);
9107 __btrfs_releasepage(page, GFP_NOFS);
9110 ClearPageChecked(page);
9111 if (PagePrivate(page)) {
9112 ClearPagePrivate(page);
9113 set_page_private(page, 0);
9119 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9120 * called from a page fault handler when a page is first dirtied. Hence we must
9121 * be careful to check for EOF conditions here. We set the page up correctly
9122 * for a written page which means we get ENOSPC checking when writing into
9123 * holes and correct delalloc and unwritten extent mapping on filesystems that
9124 * support these features.
9126 * We are not allowed to take the i_mutex here so we have to play games to
9127 * protect against truncate races as the page could now be beyond EOF. Because
9128 * vmtruncate() writes the inode size before removing pages, once we have the
9129 * page lock we can determine safely if the page is beyond EOF. If it is not
9130 * beyond EOF, then the page is guaranteed safe against truncation until we
9133 int btrfs_page_mkwrite(struct vm_fault *vmf)
9135 struct page *page = vmf->page;
9136 struct inode *inode = file_inode(vmf->vma->vm_file);
9137 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9138 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9139 struct btrfs_ordered_extent *ordered;
9140 struct extent_state *cached_state = NULL;
9141 struct extent_changeset *data_reserved = NULL;
9143 unsigned long zero_start;
9152 reserved_space = PAGE_SIZE;
9154 sb_start_pagefault(inode->i_sb);
9155 page_start = page_offset(page);
9156 page_end = page_start + PAGE_SIZE - 1;
9160 * Reserving delalloc space after obtaining the page lock can lead to
9161 * deadlock. For example, if a dirty page is locked by this function
9162 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9163 * dirty page write out, then the btrfs_writepage() function could
9164 * end up waiting indefinitely to get a lock on the page currently
9165 * being processed by btrfs_page_mkwrite() function.
9167 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9170 ret = file_update_time(vmf->vma->vm_file);
9176 else /* -ENOSPC, -EIO, etc */
9177 ret = VM_FAULT_SIGBUS;
9183 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9186 size = i_size_read(inode);
9188 if ((page->mapping != inode->i_mapping) ||
9189 (page_start >= size)) {
9190 /* page got truncated out from underneath us */
9193 wait_on_page_writeback(page);
9195 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9196 set_page_extent_mapped(page);
9199 * we can't set the delalloc bits if there are pending ordered
9200 * extents. Drop our locks and wait for them to finish
9202 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9205 unlock_extent_cached(io_tree, page_start, page_end,
9206 &cached_state, GFP_NOFS);
9208 btrfs_start_ordered_extent(inode, ordered, 1);
9209 btrfs_put_ordered_extent(ordered);
9213 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9214 reserved_space = round_up(size - page_start,
9215 fs_info->sectorsize);
9216 if (reserved_space < PAGE_SIZE) {
9217 end = page_start + reserved_space - 1;
9218 spin_lock(&BTRFS_I(inode)->lock);
9219 BTRFS_I(inode)->outstanding_extents++;
9220 spin_unlock(&BTRFS_I(inode)->lock);
9221 btrfs_delalloc_release_space(inode, data_reserved,
9222 page_start, PAGE_SIZE - reserved_space);
9227 * page_mkwrite gets called when the page is firstly dirtied after it's
9228 * faulted in, but write(2) could also dirty a page and set delalloc
9229 * bits, thus in this case for space account reason, we still need to
9230 * clear any delalloc bits within this page range since we have to
9231 * reserve data&meta space before lock_page() (see above comments).
9233 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9234 EXTENT_DIRTY | EXTENT_DELALLOC |
9235 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9236 0, 0, &cached_state, GFP_NOFS);
9238 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9241 unlock_extent_cached(io_tree, page_start, page_end,
9242 &cached_state, GFP_NOFS);
9243 ret = VM_FAULT_SIGBUS;
9248 /* page is wholly or partially inside EOF */
9249 if (page_start + PAGE_SIZE > size)
9250 zero_start = size & ~PAGE_MASK;
9252 zero_start = PAGE_SIZE;
9254 if (zero_start != PAGE_SIZE) {
9256 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9257 flush_dcache_page(page);
9260 ClearPageChecked(page);
9261 set_page_dirty(page);
9262 SetPageUptodate(page);
9264 BTRFS_I(inode)->last_trans = fs_info->generation;
9265 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9266 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9268 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9272 sb_end_pagefault(inode->i_sb);
9273 extent_changeset_free(data_reserved);
9274 return VM_FAULT_LOCKED;
9278 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9281 sb_end_pagefault(inode->i_sb);
9282 extent_changeset_free(data_reserved);
9286 static int btrfs_truncate(struct inode *inode)
9288 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9289 struct btrfs_root *root = BTRFS_I(inode)->root;
9290 struct btrfs_block_rsv *rsv;
9293 struct btrfs_trans_handle *trans;
9294 u64 mask = fs_info->sectorsize - 1;
9295 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9297 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9303 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9304 * 3 things going on here
9306 * 1) We need to reserve space for our orphan item and the space to
9307 * delete our orphan item. Lord knows we don't want to have a dangling
9308 * orphan item because we didn't reserve space to remove it.
9310 * 2) We need to reserve space to update our inode.
9312 * 3) We need to have something to cache all the space that is going to
9313 * be free'd up by the truncate operation, but also have some slack
9314 * space reserved in case it uses space during the truncate (thank you
9315 * very much snapshotting).
9317 * And we need these to all be separate. The fact is we can use a lot of
9318 * space doing the truncate, and we have no earthly idea how much space
9319 * we will use, so we need the truncate reservation to be separate so it
9320 * doesn't end up using space reserved for updating the inode or
9321 * removing the orphan item. We also need to be able to stop the
9322 * transaction and start a new one, which means we need to be able to
9323 * update the inode several times, and we have no idea of knowing how
9324 * many times that will be, so we can't just reserve 1 item for the
9325 * entirety of the operation, so that has to be done separately as well.
9326 * Then there is the orphan item, which does indeed need to be held on
9327 * to for the whole operation, and we need nobody to touch this reserved
9328 * space except the orphan code.
9330 * So that leaves us with
9332 * 1) root->orphan_block_rsv - for the orphan deletion.
9333 * 2) rsv - for the truncate reservation, which we will steal from the
9334 * transaction reservation.
9335 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9336 * updating the inode.
9338 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9341 rsv->size = min_size;
9345 * 1 for the truncate slack space
9346 * 1 for updating the inode.
9348 trans = btrfs_start_transaction(root, 2);
9349 if (IS_ERR(trans)) {
9350 err = PTR_ERR(trans);
9354 /* Migrate the slack space for the truncate to our reserve */
9355 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9360 * So if we truncate and then write and fsync we normally would just
9361 * write the extents that changed, which is a problem if we need to
9362 * first truncate that entire inode. So set this flag so we write out
9363 * all of the extents in the inode to the sync log so we're completely
9366 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9367 trans->block_rsv = rsv;
9370 ret = btrfs_truncate_inode_items(trans, root, inode,
9372 BTRFS_EXTENT_DATA_KEY);
9373 if (ret != -ENOSPC && ret != -EAGAIN) {
9378 trans->block_rsv = &fs_info->trans_block_rsv;
9379 ret = btrfs_update_inode(trans, root, inode);
9385 btrfs_end_transaction(trans);
9386 btrfs_btree_balance_dirty(fs_info);
9388 trans = btrfs_start_transaction(root, 2);
9389 if (IS_ERR(trans)) {
9390 ret = err = PTR_ERR(trans);
9395 btrfs_block_rsv_release(fs_info, rsv, -1);
9396 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9398 BUG_ON(ret); /* shouldn't happen */
9399 trans->block_rsv = rsv;
9402 if (ret == 0 && inode->i_nlink > 0) {
9403 trans->block_rsv = root->orphan_block_rsv;
9404 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9410 trans->block_rsv = &fs_info->trans_block_rsv;
9411 ret = btrfs_update_inode(trans, root, inode);
9415 ret = btrfs_end_transaction(trans);
9416 btrfs_btree_balance_dirty(fs_info);
9419 btrfs_free_block_rsv(fs_info, rsv);
9428 * create a new subvolume directory/inode (helper for the ioctl).
9430 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9431 struct btrfs_root *new_root,
9432 struct btrfs_root *parent_root,
9435 struct inode *inode;
9439 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9440 new_dirid, new_dirid,
9441 S_IFDIR | (~current_umask() & S_IRWXUGO),
9444 return PTR_ERR(inode);
9445 inode->i_op = &btrfs_dir_inode_operations;
9446 inode->i_fop = &btrfs_dir_file_operations;
9448 set_nlink(inode, 1);
9449 btrfs_i_size_write(BTRFS_I(inode), 0);
9450 unlock_new_inode(inode);
9452 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9454 btrfs_err(new_root->fs_info,
9455 "error inheriting subvolume %llu properties: %d",
9456 new_root->root_key.objectid, err);
9458 err = btrfs_update_inode(trans, new_root, inode);
9464 struct inode *btrfs_alloc_inode(struct super_block *sb)
9466 struct btrfs_inode *ei;
9467 struct inode *inode;
9469 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9476 ei->last_sub_trans = 0;
9477 ei->logged_trans = 0;
9478 ei->delalloc_bytes = 0;
9479 ei->new_delalloc_bytes = 0;
9480 ei->defrag_bytes = 0;
9481 ei->disk_i_size = 0;
9484 ei->index_cnt = (u64)-1;
9486 ei->last_unlink_trans = 0;
9487 ei->last_log_commit = 0;
9488 ei->delayed_iput_count = 0;
9490 spin_lock_init(&ei->lock);
9491 ei->outstanding_extents = 0;
9492 ei->reserved_extents = 0;
9494 ei->runtime_flags = 0;
9495 ei->prop_compress = BTRFS_COMPRESS_NONE;
9496 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9498 ei->delayed_node = NULL;
9500 ei->i_otime.tv_sec = 0;
9501 ei->i_otime.tv_nsec = 0;
9503 inode = &ei->vfs_inode;
9504 extent_map_tree_init(&ei->extent_tree);
9505 extent_io_tree_init(&ei->io_tree, inode);
9506 extent_io_tree_init(&ei->io_failure_tree, inode);
9507 ei->io_tree.track_uptodate = 1;
9508 ei->io_failure_tree.track_uptodate = 1;
9509 atomic_set(&ei->sync_writers, 0);
9510 mutex_init(&ei->log_mutex);
9511 mutex_init(&ei->delalloc_mutex);
9512 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9513 INIT_LIST_HEAD(&ei->delalloc_inodes);
9514 INIT_LIST_HEAD(&ei->delayed_iput);
9515 RB_CLEAR_NODE(&ei->rb_node);
9516 init_rwsem(&ei->dio_sem);
9521 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9522 void btrfs_test_destroy_inode(struct inode *inode)
9524 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9525 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9529 static void btrfs_i_callback(struct rcu_head *head)
9531 struct inode *inode = container_of(head, struct inode, i_rcu);
9532 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9535 void btrfs_destroy_inode(struct inode *inode)
9537 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9538 struct btrfs_ordered_extent *ordered;
9539 struct btrfs_root *root = BTRFS_I(inode)->root;
9541 WARN_ON(!hlist_empty(&inode->i_dentry));
9542 WARN_ON(inode->i_data.nrpages);
9543 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9544 WARN_ON(BTRFS_I(inode)->reserved_extents);
9545 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9546 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9547 WARN_ON(BTRFS_I(inode)->csum_bytes);
9548 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9551 * This can happen where we create an inode, but somebody else also
9552 * created the same inode and we need to destroy the one we already
9558 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9559 &BTRFS_I(inode)->runtime_flags)) {
9560 btrfs_info(fs_info, "inode %llu still on the orphan list",
9561 btrfs_ino(BTRFS_I(inode)));
9562 atomic_dec(&root->orphan_inodes);
9566 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9571 "found ordered extent %llu %llu on inode cleanup",
9572 ordered->file_offset, ordered->len);
9573 btrfs_remove_ordered_extent(inode, ordered);
9574 btrfs_put_ordered_extent(ordered);
9575 btrfs_put_ordered_extent(ordered);
9578 btrfs_qgroup_check_reserved_leak(inode);
9579 inode_tree_del(inode);
9580 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9582 call_rcu(&inode->i_rcu, btrfs_i_callback);
9585 int btrfs_drop_inode(struct inode *inode)
9587 struct btrfs_root *root = BTRFS_I(inode)->root;
9592 /* the snap/subvol tree is on deleting */
9593 if (btrfs_root_refs(&root->root_item) == 0)
9596 return generic_drop_inode(inode);
9599 static void init_once(void *foo)
9601 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9603 inode_init_once(&ei->vfs_inode);
9606 void btrfs_destroy_cachep(void)
9609 * Make sure all delayed rcu free inodes are flushed before we
9613 kmem_cache_destroy(btrfs_inode_cachep);
9614 kmem_cache_destroy(btrfs_trans_handle_cachep);
9615 kmem_cache_destroy(btrfs_path_cachep);
9616 kmem_cache_destroy(btrfs_free_space_cachep);
9619 int btrfs_init_cachep(void)
9621 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9622 sizeof(struct btrfs_inode), 0,
9623 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9625 if (!btrfs_inode_cachep)
9628 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9629 sizeof(struct btrfs_trans_handle), 0,
9630 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9631 if (!btrfs_trans_handle_cachep)
9634 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9635 sizeof(struct btrfs_path), 0,
9636 SLAB_MEM_SPREAD, NULL);
9637 if (!btrfs_path_cachep)
9640 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9641 sizeof(struct btrfs_free_space), 0,
9642 SLAB_MEM_SPREAD, NULL);
9643 if (!btrfs_free_space_cachep)
9648 btrfs_destroy_cachep();
9652 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9653 u32 request_mask, unsigned int flags)
9656 struct inode *inode = d_inode(path->dentry);
9657 u32 blocksize = inode->i_sb->s_blocksize;
9658 u32 bi_flags = BTRFS_I(inode)->flags;
9660 stat->result_mask |= STATX_BTIME;
9661 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9662 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9663 if (bi_flags & BTRFS_INODE_APPEND)
9664 stat->attributes |= STATX_ATTR_APPEND;
9665 if (bi_flags & BTRFS_INODE_COMPRESS)
9666 stat->attributes |= STATX_ATTR_COMPRESSED;
9667 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9668 stat->attributes |= STATX_ATTR_IMMUTABLE;
9669 if (bi_flags & BTRFS_INODE_NODUMP)
9670 stat->attributes |= STATX_ATTR_NODUMP;
9672 stat->attributes_mask |= (STATX_ATTR_APPEND |
9673 STATX_ATTR_COMPRESSED |
9674 STATX_ATTR_IMMUTABLE |
9677 generic_fillattr(inode, stat);
9678 stat->dev = BTRFS_I(inode)->root->anon_dev;
9680 spin_lock(&BTRFS_I(inode)->lock);
9681 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9682 spin_unlock(&BTRFS_I(inode)->lock);
9683 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9684 ALIGN(delalloc_bytes, blocksize)) >> 9;
9688 static int btrfs_rename_exchange(struct inode *old_dir,
9689 struct dentry *old_dentry,
9690 struct inode *new_dir,
9691 struct dentry *new_dentry)
9693 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9694 struct btrfs_trans_handle *trans;
9695 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9696 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9697 struct inode *new_inode = new_dentry->d_inode;
9698 struct inode *old_inode = old_dentry->d_inode;
9699 struct timespec ctime = current_time(old_inode);
9700 struct dentry *parent;
9701 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9702 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9707 bool root_log_pinned = false;
9708 bool dest_log_pinned = false;
9710 /* we only allow rename subvolume link between subvolumes */
9711 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9714 /* close the race window with snapshot create/destroy ioctl */
9715 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9716 down_read(&fs_info->subvol_sem);
9717 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9718 down_read(&fs_info->subvol_sem);
9721 * We want to reserve the absolute worst case amount of items. So if
9722 * both inodes are subvols and we need to unlink them then that would
9723 * require 4 item modifications, but if they are both normal inodes it
9724 * would require 5 item modifications, so we'll assume their normal
9725 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9726 * should cover the worst case number of items we'll modify.
9728 trans = btrfs_start_transaction(root, 12);
9729 if (IS_ERR(trans)) {
9730 ret = PTR_ERR(trans);
9735 * We need to find a free sequence number both in the source and
9736 * in the destination directory for the exchange.
9738 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9741 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9745 BTRFS_I(old_inode)->dir_index = 0ULL;
9746 BTRFS_I(new_inode)->dir_index = 0ULL;
9748 /* Reference for the source. */
9749 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9750 /* force full log commit if subvolume involved. */
9751 btrfs_set_log_full_commit(fs_info, trans);
9753 btrfs_pin_log_trans(root);
9754 root_log_pinned = true;
9755 ret = btrfs_insert_inode_ref(trans, dest,
9756 new_dentry->d_name.name,
9757 new_dentry->d_name.len,
9759 btrfs_ino(BTRFS_I(new_dir)),
9765 /* And now for the dest. */
9766 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9767 /* force full log commit if subvolume involved. */
9768 btrfs_set_log_full_commit(fs_info, trans);
9770 btrfs_pin_log_trans(dest);
9771 dest_log_pinned = true;
9772 ret = btrfs_insert_inode_ref(trans, root,
9773 old_dentry->d_name.name,
9774 old_dentry->d_name.len,
9776 btrfs_ino(BTRFS_I(old_dir)),
9782 /* Update inode version and ctime/mtime. */
9783 inode_inc_iversion(old_dir);
9784 inode_inc_iversion(new_dir);
9785 inode_inc_iversion(old_inode);
9786 inode_inc_iversion(new_inode);
9787 old_dir->i_ctime = old_dir->i_mtime = ctime;
9788 new_dir->i_ctime = new_dir->i_mtime = ctime;
9789 old_inode->i_ctime = ctime;
9790 new_inode->i_ctime = ctime;
9792 if (old_dentry->d_parent != new_dentry->d_parent) {
9793 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9794 BTRFS_I(old_inode), 1);
9795 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9796 BTRFS_I(new_inode), 1);
9799 /* src is a subvolume */
9800 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9801 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9802 ret = btrfs_unlink_subvol(trans, root, old_dir,
9804 old_dentry->d_name.name,
9805 old_dentry->d_name.len);
9806 } else { /* src is an inode */
9807 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9808 BTRFS_I(old_dentry->d_inode),
9809 old_dentry->d_name.name,
9810 old_dentry->d_name.len);
9812 ret = btrfs_update_inode(trans, root, old_inode);
9815 btrfs_abort_transaction(trans, ret);
9819 /* dest is a subvolume */
9820 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9821 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9822 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9824 new_dentry->d_name.name,
9825 new_dentry->d_name.len);
9826 } else { /* dest is an inode */
9827 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9828 BTRFS_I(new_dentry->d_inode),
9829 new_dentry->d_name.name,
9830 new_dentry->d_name.len);
9832 ret = btrfs_update_inode(trans, dest, new_inode);
9835 btrfs_abort_transaction(trans, ret);
9839 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9840 new_dentry->d_name.name,
9841 new_dentry->d_name.len, 0, old_idx);
9843 btrfs_abort_transaction(trans, ret);
9847 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9848 old_dentry->d_name.name,
9849 old_dentry->d_name.len, 0, new_idx);
9851 btrfs_abort_transaction(trans, ret);
9855 if (old_inode->i_nlink == 1)
9856 BTRFS_I(old_inode)->dir_index = old_idx;
9857 if (new_inode->i_nlink == 1)
9858 BTRFS_I(new_inode)->dir_index = new_idx;
9860 if (root_log_pinned) {
9861 parent = new_dentry->d_parent;
9862 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9864 btrfs_end_log_trans(root);
9865 root_log_pinned = false;
9867 if (dest_log_pinned) {
9868 parent = old_dentry->d_parent;
9869 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9871 btrfs_end_log_trans(dest);
9872 dest_log_pinned = false;
9876 * If we have pinned a log and an error happened, we unpin tasks
9877 * trying to sync the log and force them to fallback to a transaction
9878 * commit if the log currently contains any of the inodes involved in
9879 * this rename operation (to ensure we do not persist a log with an
9880 * inconsistent state for any of these inodes or leading to any
9881 * inconsistencies when replayed). If the transaction was aborted, the
9882 * abortion reason is propagated to userspace when attempting to commit
9883 * the transaction. If the log does not contain any of these inodes, we
9884 * allow the tasks to sync it.
9886 if (ret && (root_log_pinned || dest_log_pinned)) {
9887 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9888 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9889 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9891 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9892 btrfs_set_log_full_commit(fs_info, trans);
9894 if (root_log_pinned) {
9895 btrfs_end_log_trans(root);
9896 root_log_pinned = false;
9898 if (dest_log_pinned) {
9899 btrfs_end_log_trans(dest);
9900 dest_log_pinned = false;
9903 ret = btrfs_end_transaction(trans);
9905 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9906 up_read(&fs_info->subvol_sem);
9907 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9908 up_read(&fs_info->subvol_sem);
9913 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9914 struct btrfs_root *root,
9916 struct dentry *dentry)
9919 struct inode *inode;
9923 ret = btrfs_find_free_ino(root, &objectid);
9927 inode = btrfs_new_inode(trans, root, dir,
9928 dentry->d_name.name,
9930 btrfs_ino(BTRFS_I(dir)),
9932 S_IFCHR | WHITEOUT_MODE,
9935 if (IS_ERR(inode)) {
9936 ret = PTR_ERR(inode);
9940 inode->i_op = &btrfs_special_inode_operations;
9941 init_special_inode(inode, inode->i_mode,
9944 ret = btrfs_init_inode_security(trans, inode, dir,
9949 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9950 BTRFS_I(inode), 0, index);
9954 ret = btrfs_update_inode(trans, root, inode);
9956 unlock_new_inode(inode);
9958 inode_dec_link_count(inode);
9964 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9965 struct inode *new_dir, struct dentry *new_dentry,
9968 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9969 struct btrfs_trans_handle *trans;
9970 unsigned int trans_num_items;
9971 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9972 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9973 struct inode *new_inode = d_inode(new_dentry);
9974 struct inode *old_inode = d_inode(old_dentry);
9978 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9979 bool log_pinned = false;
9981 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9984 /* we only allow rename subvolume link between subvolumes */
9985 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9988 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9989 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9992 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9993 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9997 /* check for collisions, even if the name isn't there */
9998 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9999 new_dentry->d_name.name,
10000 new_dentry->d_name.len);
10003 if (ret == -EEXIST) {
10004 /* we shouldn't get
10005 * eexist without a new_inode */
10006 if (WARN_ON(!new_inode)) {
10010 /* maybe -EOVERFLOW */
10017 * we're using rename to replace one file with another. Start IO on it
10018 * now so we don't add too much work to the end of the transaction
10020 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
10021 filemap_flush(old_inode->i_mapping);
10023 /* close the racy window with snapshot create/destroy ioctl */
10024 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10025 down_read(&fs_info->subvol_sem);
10027 * We want to reserve the absolute worst case amount of items. So if
10028 * both inodes are subvols and we need to unlink them then that would
10029 * require 4 item modifications, but if they are both normal inodes it
10030 * would require 5 item modifications, so we'll assume they are normal
10031 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
10032 * should cover the worst case number of items we'll modify.
10033 * If our rename has the whiteout flag, we need more 5 units for the
10034 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
10035 * when selinux is enabled).
10037 trans_num_items = 11;
10038 if (flags & RENAME_WHITEOUT)
10039 trans_num_items += 5;
10040 trans = btrfs_start_transaction(root, trans_num_items);
10041 if (IS_ERR(trans)) {
10042 ret = PTR_ERR(trans);
10047 btrfs_record_root_in_trans(trans, dest);
10049 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
10053 BTRFS_I(old_inode)->dir_index = 0ULL;
10054 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10055 /* force full log commit if subvolume involved. */
10056 btrfs_set_log_full_commit(fs_info, trans);
10058 btrfs_pin_log_trans(root);
10060 ret = btrfs_insert_inode_ref(trans, dest,
10061 new_dentry->d_name.name,
10062 new_dentry->d_name.len,
10064 btrfs_ino(BTRFS_I(new_dir)), index);
10069 inode_inc_iversion(old_dir);
10070 inode_inc_iversion(new_dir);
10071 inode_inc_iversion(old_inode);
10072 old_dir->i_ctime = old_dir->i_mtime =
10073 new_dir->i_ctime = new_dir->i_mtime =
10074 old_inode->i_ctime = current_time(old_dir);
10076 if (old_dentry->d_parent != new_dentry->d_parent)
10077 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
10078 BTRFS_I(old_inode), 1);
10080 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10081 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
10082 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
10083 old_dentry->d_name.name,
10084 old_dentry->d_name.len);
10086 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10087 BTRFS_I(d_inode(old_dentry)),
10088 old_dentry->d_name.name,
10089 old_dentry->d_name.len);
10091 ret = btrfs_update_inode(trans, root, old_inode);
10094 btrfs_abort_transaction(trans, ret);
10099 inode_inc_iversion(new_inode);
10100 new_inode->i_ctime = current_time(new_inode);
10101 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10102 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10103 root_objectid = BTRFS_I(new_inode)->location.objectid;
10104 ret = btrfs_unlink_subvol(trans, dest, new_dir,
10106 new_dentry->d_name.name,
10107 new_dentry->d_name.len);
10108 BUG_ON(new_inode->i_nlink == 0);
10110 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10111 BTRFS_I(d_inode(new_dentry)),
10112 new_dentry->d_name.name,
10113 new_dentry->d_name.len);
10115 if (!ret && new_inode->i_nlink == 0)
10116 ret = btrfs_orphan_add(trans,
10117 BTRFS_I(d_inode(new_dentry)));
10119 btrfs_abort_transaction(trans, ret);
10124 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10125 new_dentry->d_name.name,
10126 new_dentry->d_name.len, 0, index);
10128 btrfs_abort_transaction(trans, ret);
10132 if (old_inode->i_nlink == 1)
10133 BTRFS_I(old_inode)->dir_index = index;
10136 struct dentry *parent = new_dentry->d_parent;
10138 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10140 btrfs_end_log_trans(root);
10141 log_pinned = false;
10144 if (flags & RENAME_WHITEOUT) {
10145 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10149 btrfs_abort_transaction(trans, ret);
10155 * If we have pinned the log and an error happened, we unpin tasks
10156 * trying to sync the log and force them to fallback to a transaction
10157 * commit if the log currently contains any of the inodes involved in
10158 * this rename operation (to ensure we do not persist a log with an
10159 * inconsistent state for any of these inodes or leading to any
10160 * inconsistencies when replayed). If the transaction was aborted, the
10161 * abortion reason is propagated to userspace when attempting to commit
10162 * the transaction. If the log does not contain any of these inodes, we
10163 * allow the tasks to sync it.
10165 if (ret && log_pinned) {
10166 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10167 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10168 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10170 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10171 btrfs_set_log_full_commit(fs_info, trans);
10173 btrfs_end_log_trans(root);
10174 log_pinned = false;
10176 btrfs_end_transaction(trans);
10178 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10179 up_read(&fs_info->subvol_sem);
10184 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10185 struct inode *new_dir, struct dentry *new_dentry,
10186 unsigned int flags)
10188 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10191 if (flags & RENAME_EXCHANGE)
10192 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10195 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10198 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10200 struct btrfs_delalloc_work *delalloc_work;
10201 struct inode *inode;
10203 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10205 inode = delalloc_work->inode;
10206 filemap_flush(inode->i_mapping);
10207 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10208 &BTRFS_I(inode)->runtime_flags))
10209 filemap_flush(inode->i_mapping);
10211 if (delalloc_work->delay_iput)
10212 btrfs_add_delayed_iput(inode);
10215 complete(&delalloc_work->completion);
10218 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10221 struct btrfs_delalloc_work *work;
10223 work = kmalloc(sizeof(*work), GFP_NOFS);
10227 init_completion(&work->completion);
10228 INIT_LIST_HEAD(&work->list);
10229 work->inode = inode;
10230 work->delay_iput = delay_iput;
10231 WARN_ON_ONCE(!inode);
10232 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10233 btrfs_run_delalloc_work, NULL, NULL);
10238 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10240 wait_for_completion(&work->completion);
10245 * some fairly slow code that needs optimization. This walks the list
10246 * of all the inodes with pending delalloc and forces them to disk.
10248 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10251 struct btrfs_inode *binode;
10252 struct inode *inode;
10253 struct btrfs_delalloc_work *work, *next;
10254 struct list_head works;
10255 struct list_head splice;
10258 INIT_LIST_HEAD(&works);
10259 INIT_LIST_HEAD(&splice);
10261 mutex_lock(&root->delalloc_mutex);
10262 spin_lock(&root->delalloc_lock);
10263 list_splice_init(&root->delalloc_inodes, &splice);
10264 while (!list_empty(&splice)) {
10265 binode = list_entry(splice.next, struct btrfs_inode,
10268 list_move_tail(&binode->delalloc_inodes,
10269 &root->delalloc_inodes);
10270 inode = igrab(&binode->vfs_inode);
10272 cond_resched_lock(&root->delalloc_lock);
10275 spin_unlock(&root->delalloc_lock);
10277 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10280 btrfs_add_delayed_iput(inode);
10286 list_add_tail(&work->list, &works);
10287 btrfs_queue_work(root->fs_info->flush_workers,
10290 if (nr != -1 && ret >= nr)
10293 spin_lock(&root->delalloc_lock);
10295 spin_unlock(&root->delalloc_lock);
10298 list_for_each_entry_safe(work, next, &works, list) {
10299 list_del_init(&work->list);
10300 btrfs_wait_and_free_delalloc_work(work);
10303 if (!list_empty_careful(&splice)) {
10304 spin_lock(&root->delalloc_lock);
10305 list_splice_tail(&splice, &root->delalloc_inodes);
10306 spin_unlock(&root->delalloc_lock);
10308 mutex_unlock(&root->delalloc_mutex);
10312 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10314 struct btrfs_fs_info *fs_info = root->fs_info;
10317 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10320 ret = __start_delalloc_inodes(root, delay_iput, -1);
10324 * the filemap_flush will queue IO into the worker threads, but
10325 * we have to make sure the IO is actually started and that
10326 * ordered extents get created before we return
10328 atomic_inc(&fs_info->async_submit_draining);
10329 while (atomic_read(&fs_info->nr_async_submits) ||
10330 atomic_read(&fs_info->async_delalloc_pages)) {
10331 wait_event(fs_info->async_submit_wait,
10332 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10333 atomic_read(&fs_info->async_delalloc_pages) == 0));
10335 atomic_dec(&fs_info->async_submit_draining);
10339 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10342 struct btrfs_root *root;
10343 struct list_head splice;
10346 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10349 INIT_LIST_HEAD(&splice);
10351 mutex_lock(&fs_info->delalloc_root_mutex);
10352 spin_lock(&fs_info->delalloc_root_lock);
10353 list_splice_init(&fs_info->delalloc_roots, &splice);
10354 while (!list_empty(&splice) && nr) {
10355 root = list_first_entry(&splice, struct btrfs_root,
10357 root = btrfs_grab_fs_root(root);
10359 list_move_tail(&root->delalloc_root,
10360 &fs_info->delalloc_roots);
10361 spin_unlock(&fs_info->delalloc_root_lock);
10363 ret = __start_delalloc_inodes(root, delay_iput, nr);
10364 btrfs_put_fs_root(root);
10372 spin_lock(&fs_info->delalloc_root_lock);
10374 spin_unlock(&fs_info->delalloc_root_lock);
10377 atomic_inc(&fs_info->async_submit_draining);
10378 while (atomic_read(&fs_info->nr_async_submits) ||
10379 atomic_read(&fs_info->async_delalloc_pages)) {
10380 wait_event(fs_info->async_submit_wait,
10381 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10382 atomic_read(&fs_info->async_delalloc_pages) == 0));
10384 atomic_dec(&fs_info->async_submit_draining);
10386 if (!list_empty_careful(&splice)) {
10387 spin_lock(&fs_info->delalloc_root_lock);
10388 list_splice_tail(&splice, &fs_info->delalloc_roots);
10389 spin_unlock(&fs_info->delalloc_root_lock);
10391 mutex_unlock(&fs_info->delalloc_root_mutex);
10395 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10396 const char *symname)
10398 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10399 struct btrfs_trans_handle *trans;
10400 struct btrfs_root *root = BTRFS_I(dir)->root;
10401 struct btrfs_path *path;
10402 struct btrfs_key key;
10403 struct inode *inode = NULL;
10405 int drop_inode = 0;
10411 struct btrfs_file_extent_item *ei;
10412 struct extent_buffer *leaf;
10414 name_len = strlen(symname);
10415 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10416 return -ENAMETOOLONG;
10419 * 2 items for inode item and ref
10420 * 2 items for dir items
10421 * 1 item for updating parent inode item
10422 * 1 item for the inline extent item
10423 * 1 item for xattr if selinux is on
10425 trans = btrfs_start_transaction(root, 7);
10427 return PTR_ERR(trans);
10429 err = btrfs_find_free_ino(root, &objectid);
10433 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10434 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10435 objectid, S_IFLNK|S_IRWXUGO, &index);
10436 if (IS_ERR(inode)) {
10437 err = PTR_ERR(inode);
10442 * If the active LSM wants to access the inode during
10443 * d_instantiate it needs these. Smack checks to see
10444 * if the filesystem supports xattrs by looking at the
10447 inode->i_fop = &btrfs_file_operations;
10448 inode->i_op = &btrfs_file_inode_operations;
10449 inode->i_mapping->a_ops = &btrfs_aops;
10450 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10452 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10454 goto out_unlock_inode;
10456 path = btrfs_alloc_path();
10459 goto out_unlock_inode;
10461 key.objectid = btrfs_ino(BTRFS_I(inode));
10463 key.type = BTRFS_EXTENT_DATA_KEY;
10464 datasize = btrfs_file_extent_calc_inline_size(name_len);
10465 err = btrfs_insert_empty_item(trans, root, path, &key,
10468 btrfs_free_path(path);
10469 goto out_unlock_inode;
10471 leaf = path->nodes[0];
10472 ei = btrfs_item_ptr(leaf, path->slots[0],
10473 struct btrfs_file_extent_item);
10474 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10475 btrfs_set_file_extent_type(leaf, ei,
10476 BTRFS_FILE_EXTENT_INLINE);
10477 btrfs_set_file_extent_encryption(leaf, ei, 0);
10478 btrfs_set_file_extent_compression(leaf, ei, 0);
10479 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10480 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10482 ptr = btrfs_file_extent_inline_start(ei);
10483 write_extent_buffer(leaf, symname, ptr, name_len);
10484 btrfs_mark_buffer_dirty(leaf);
10485 btrfs_free_path(path);
10487 inode->i_op = &btrfs_symlink_inode_operations;
10488 inode_nohighmem(inode);
10489 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10490 inode_set_bytes(inode, name_len);
10491 btrfs_i_size_write(BTRFS_I(inode), name_len);
10492 err = btrfs_update_inode(trans, root, inode);
10494 * Last step, add directory indexes for our symlink inode. This is the
10495 * last step to avoid extra cleanup of these indexes if an error happens
10499 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10500 BTRFS_I(inode), 0, index);
10503 goto out_unlock_inode;
10506 unlock_new_inode(inode);
10507 d_instantiate(dentry, inode);
10510 btrfs_end_transaction(trans);
10512 inode_dec_link_count(inode);
10515 btrfs_btree_balance_dirty(fs_info);
10520 unlock_new_inode(inode);
10524 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10525 u64 start, u64 num_bytes, u64 min_size,
10526 loff_t actual_len, u64 *alloc_hint,
10527 struct btrfs_trans_handle *trans)
10529 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10530 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10531 struct extent_map *em;
10532 struct btrfs_root *root = BTRFS_I(inode)->root;
10533 struct btrfs_key ins;
10534 u64 cur_offset = start;
10537 u64 last_alloc = (u64)-1;
10539 bool own_trans = true;
10540 u64 end = start + num_bytes - 1;
10544 while (num_bytes > 0) {
10546 trans = btrfs_start_transaction(root, 3);
10547 if (IS_ERR(trans)) {
10548 ret = PTR_ERR(trans);
10553 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10554 cur_bytes = max(cur_bytes, min_size);
10556 * If we are severely fragmented we could end up with really
10557 * small allocations, so if the allocator is returning small
10558 * chunks lets make its job easier by only searching for those
10561 cur_bytes = min(cur_bytes, last_alloc);
10562 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10563 min_size, 0, *alloc_hint, &ins, 1, 0);
10566 btrfs_end_transaction(trans);
10569 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10571 last_alloc = ins.offset;
10572 ret = insert_reserved_file_extent(trans, inode,
10573 cur_offset, ins.objectid,
10574 ins.offset, ins.offset,
10575 ins.offset, 0, 0, 0,
10576 BTRFS_FILE_EXTENT_PREALLOC);
10578 btrfs_free_reserved_extent(fs_info, ins.objectid,
10580 btrfs_abort_transaction(trans, ret);
10582 btrfs_end_transaction(trans);
10586 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10587 cur_offset + ins.offset -1, 0);
10589 em = alloc_extent_map();
10591 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10592 &BTRFS_I(inode)->runtime_flags);
10596 em->start = cur_offset;
10597 em->orig_start = cur_offset;
10598 em->len = ins.offset;
10599 em->block_start = ins.objectid;
10600 em->block_len = ins.offset;
10601 em->orig_block_len = ins.offset;
10602 em->ram_bytes = ins.offset;
10603 em->bdev = fs_info->fs_devices->latest_bdev;
10604 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10605 em->generation = trans->transid;
10608 write_lock(&em_tree->lock);
10609 ret = add_extent_mapping(em_tree, em, 1);
10610 write_unlock(&em_tree->lock);
10611 if (ret != -EEXIST)
10613 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10614 cur_offset + ins.offset - 1,
10617 free_extent_map(em);
10619 num_bytes -= ins.offset;
10620 cur_offset += ins.offset;
10621 *alloc_hint = ins.objectid + ins.offset;
10623 inode_inc_iversion(inode);
10624 inode->i_ctime = current_time(inode);
10625 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10626 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10627 (actual_len > inode->i_size) &&
10628 (cur_offset > inode->i_size)) {
10629 if (cur_offset > actual_len)
10630 i_size = actual_len;
10632 i_size = cur_offset;
10633 i_size_write(inode, i_size);
10634 btrfs_ordered_update_i_size(inode, i_size, NULL);
10637 ret = btrfs_update_inode(trans, root, inode);
10640 btrfs_abort_transaction(trans, ret);
10642 btrfs_end_transaction(trans);
10647 btrfs_end_transaction(trans);
10649 if (cur_offset < end)
10650 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10651 end - cur_offset + 1);
10655 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10656 u64 start, u64 num_bytes, u64 min_size,
10657 loff_t actual_len, u64 *alloc_hint)
10659 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10660 min_size, actual_len, alloc_hint,
10664 int btrfs_prealloc_file_range_trans(struct inode *inode,
10665 struct btrfs_trans_handle *trans, int mode,
10666 u64 start, u64 num_bytes, u64 min_size,
10667 loff_t actual_len, u64 *alloc_hint)
10669 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10670 min_size, actual_len, alloc_hint, trans);
10673 static int btrfs_set_page_dirty(struct page *page)
10675 return __set_page_dirty_nobuffers(page);
10678 static int btrfs_permission(struct inode *inode, int mask)
10680 struct btrfs_root *root = BTRFS_I(inode)->root;
10681 umode_t mode = inode->i_mode;
10683 if (mask & MAY_WRITE &&
10684 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10685 if (btrfs_root_readonly(root))
10687 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10690 return generic_permission(inode, mask);
10693 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10695 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10696 struct btrfs_trans_handle *trans;
10697 struct btrfs_root *root = BTRFS_I(dir)->root;
10698 struct inode *inode = NULL;
10704 * 5 units required for adding orphan entry
10706 trans = btrfs_start_transaction(root, 5);
10708 return PTR_ERR(trans);
10710 ret = btrfs_find_free_ino(root, &objectid);
10714 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10715 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10716 if (IS_ERR(inode)) {
10717 ret = PTR_ERR(inode);
10722 inode->i_fop = &btrfs_file_operations;
10723 inode->i_op = &btrfs_file_inode_operations;
10725 inode->i_mapping->a_ops = &btrfs_aops;
10726 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10728 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10732 ret = btrfs_update_inode(trans, root, inode);
10735 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10740 * We set number of links to 0 in btrfs_new_inode(), and here we set
10741 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10744 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10746 set_nlink(inode, 1);
10747 unlock_new_inode(inode);
10748 d_tmpfile(dentry, inode);
10749 mark_inode_dirty(inode);
10752 btrfs_end_transaction(trans);
10755 btrfs_balance_delayed_items(fs_info);
10756 btrfs_btree_balance_dirty(fs_info);
10760 unlock_new_inode(inode);
10765 __attribute__((const))
10766 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10771 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10773 struct inode *inode = private_data;
10774 return btrfs_sb(inode->i_sb);
10777 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10778 u64 start, u64 end)
10780 struct inode *inode = private_data;
10783 isize = i_size_read(inode);
10784 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10785 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10786 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10787 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10791 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10793 struct inode *inode = private_data;
10794 unsigned long index = start >> PAGE_SHIFT;
10795 unsigned long end_index = end >> PAGE_SHIFT;
10798 while (index <= end_index) {
10799 page = find_get_page(inode->i_mapping, index);
10800 ASSERT(page); /* Pages should be in the extent_io_tree */
10801 set_page_writeback(page);
10807 static const struct inode_operations btrfs_dir_inode_operations = {
10808 .getattr = btrfs_getattr,
10809 .lookup = btrfs_lookup,
10810 .create = btrfs_create,
10811 .unlink = btrfs_unlink,
10812 .link = btrfs_link,
10813 .mkdir = btrfs_mkdir,
10814 .rmdir = btrfs_rmdir,
10815 .rename = btrfs_rename2,
10816 .symlink = btrfs_symlink,
10817 .setattr = btrfs_setattr,
10818 .mknod = btrfs_mknod,
10819 .listxattr = btrfs_listxattr,
10820 .permission = btrfs_permission,
10821 .get_acl = btrfs_get_acl,
10822 .set_acl = btrfs_set_acl,
10823 .update_time = btrfs_update_time,
10824 .tmpfile = btrfs_tmpfile,
10826 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10827 .lookup = btrfs_lookup,
10828 .permission = btrfs_permission,
10829 .update_time = btrfs_update_time,
10832 static const struct file_operations btrfs_dir_file_operations = {
10833 .llseek = generic_file_llseek,
10834 .read = generic_read_dir,
10835 .iterate_shared = btrfs_real_readdir,
10836 .open = btrfs_opendir,
10837 .unlocked_ioctl = btrfs_ioctl,
10838 #ifdef CONFIG_COMPAT
10839 .compat_ioctl = btrfs_compat_ioctl,
10841 .release = btrfs_release_file,
10842 .fsync = btrfs_sync_file,
10845 static const struct extent_io_ops btrfs_extent_io_ops = {
10846 /* mandatory callbacks */
10847 .submit_bio_hook = btrfs_submit_bio_hook,
10848 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10849 .merge_bio_hook = btrfs_merge_bio_hook,
10850 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10851 .tree_fs_info = iotree_fs_info,
10852 .set_range_writeback = btrfs_set_range_writeback,
10854 /* optional callbacks */
10855 .fill_delalloc = run_delalloc_range,
10856 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10857 .writepage_start_hook = btrfs_writepage_start_hook,
10858 .set_bit_hook = btrfs_set_bit_hook,
10859 .clear_bit_hook = btrfs_clear_bit_hook,
10860 .merge_extent_hook = btrfs_merge_extent_hook,
10861 .split_extent_hook = btrfs_split_extent_hook,
10862 .check_extent_io_range = btrfs_check_extent_io_range,
10866 * btrfs doesn't support the bmap operation because swapfiles
10867 * use bmap to make a mapping of extents in the file. They assume
10868 * these extents won't change over the life of the file and they
10869 * use the bmap result to do IO directly to the drive.
10871 * the btrfs bmap call would return logical addresses that aren't
10872 * suitable for IO and they also will change frequently as COW
10873 * operations happen. So, swapfile + btrfs == corruption.
10875 * For now we're avoiding this by dropping bmap.
10877 static const struct address_space_operations btrfs_aops = {
10878 .readpage = btrfs_readpage,
10879 .writepage = btrfs_writepage,
10880 .writepages = btrfs_writepages,
10881 .readpages = btrfs_readpages,
10882 .direct_IO = btrfs_direct_IO,
10883 .invalidatepage = btrfs_invalidatepage,
10884 .releasepage = btrfs_releasepage,
10885 .set_page_dirty = btrfs_set_page_dirty,
10886 .error_remove_page = generic_error_remove_page,
10889 static const struct address_space_operations btrfs_symlink_aops = {
10890 .readpage = btrfs_readpage,
10891 .writepage = btrfs_writepage,
10892 .invalidatepage = btrfs_invalidatepage,
10893 .releasepage = btrfs_releasepage,
10896 static const struct inode_operations btrfs_file_inode_operations = {
10897 .getattr = btrfs_getattr,
10898 .setattr = btrfs_setattr,
10899 .listxattr = btrfs_listxattr,
10900 .permission = btrfs_permission,
10901 .fiemap = btrfs_fiemap,
10902 .get_acl = btrfs_get_acl,
10903 .set_acl = btrfs_set_acl,
10904 .update_time = btrfs_update_time,
10906 static const struct inode_operations btrfs_special_inode_operations = {
10907 .getattr = btrfs_getattr,
10908 .setattr = btrfs_setattr,
10909 .permission = btrfs_permission,
10910 .listxattr = btrfs_listxattr,
10911 .get_acl = btrfs_get_acl,
10912 .set_acl = btrfs_set_acl,
10913 .update_time = btrfs_update_time,
10915 static const struct inode_operations btrfs_symlink_inode_operations = {
10916 .get_link = page_get_link,
10917 .getattr = btrfs_getattr,
10918 .setattr = btrfs_setattr,
10919 .permission = btrfs_permission,
10920 .listxattr = btrfs_listxattr,
10921 .update_time = btrfs_update_time,
10924 const struct dentry_operations btrfs_dentry_operations = {
10925 .d_delete = btrfs_dentry_delete,
10926 .d_release = btrfs_dentry_release,