1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <asm/unaligned.h>
34 #include "transaction.h"
35 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "ordered-data.h"
41 #include "compression.h"
43 #include "free-space-cache.h"
44 #include "inode-map.h"
50 struct btrfs_iget_args {
51 struct btrfs_key *location;
52 struct btrfs_root *root;
55 struct btrfs_dio_data {
57 u64 unsubmitted_oe_range_start;
58 u64 unsubmitted_oe_range_end;
62 static const struct inode_operations btrfs_dir_inode_operations;
63 static const struct inode_operations btrfs_symlink_inode_operations;
64 static const struct inode_operations btrfs_dir_ro_inode_operations;
65 static const struct inode_operations btrfs_special_inode_operations;
66 static const struct inode_operations btrfs_file_inode_operations;
67 static const struct address_space_operations btrfs_aops;
68 static const struct file_operations btrfs_dir_file_operations;
69 static const struct extent_io_ops btrfs_extent_io_ops;
71 static struct kmem_cache *btrfs_inode_cachep;
72 struct kmem_cache *btrfs_trans_handle_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
77 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
88 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
89 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, u64 delalloc_end,
93 int *page_started, unsigned long *nr_written,
94 int unlock, struct btrfs_dedupe_hash *hash);
95 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
96 u64 orig_start, u64 block_start,
97 u64 block_len, u64 orig_block_len,
98 u64 ram_bytes, int compress_type,
101 static void __endio_write_update_ordered(struct inode *inode,
102 const u64 offset, const u64 bytes,
103 const bool uptodate);
106 * Cleanup all submitted ordered extents in specified range to handle errors
107 * from the fill_dellaloc() callback.
109 * NOTE: caller must ensure that when an error happens, it can not call
110 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
111 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
112 * to be released, which we want to happen only when finishing the ordered
113 * extent (btrfs_finish_ordered_io()). Also note that the caller of
114 * btrfs_run_delalloc_range already does proper cleanup for the first page of
115 * the range, that is, it invokes the callback writepage_end_io_hook() for the
116 * range of the first page.
118 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
122 unsigned long index = offset >> PAGE_SHIFT;
123 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
126 while (index <= end_index) {
127 page = find_get_page(inode->i_mapping, index);
131 ClearPagePrivate2(page);
134 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
135 bytes - PAGE_SIZE, false);
138 static int btrfs_dirty_inode(struct inode *inode);
140 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
141 void btrfs_test_inode_set_ops(struct inode *inode)
143 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
147 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
148 struct inode *inode, struct inode *dir,
149 const struct qstr *qstr)
153 err = btrfs_init_acl(trans, inode, dir);
155 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
160 * this does all the hard work for inserting an inline extent into
161 * the btree. The caller should have done a btrfs_drop_extents so that
162 * no overlapping inline items exist in the btree
164 static int insert_inline_extent(struct btrfs_trans_handle *trans,
165 struct btrfs_path *path, int extent_inserted,
166 struct btrfs_root *root, struct inode *inode,
167 u64 start, size_t size, size_t compressed_size,
169 struct page **compressed_pages)
171 struct extent_buffer *leaf;
172 struct page *page = NULL;
175 struct btrfs_file_extent_item *ei;
177 size_t cur_size = size;
178 unsigned long offset;
180 if (compressed_size && compressed_pages)
181 cur_size = compressed_size;
183 inode_add_bytes(inode, size);
185 if (!extent_inserted) {
186 struct btrfs_key key;
189 key.objectid = btrfs_ino(BTRFS_I(inode));
191 key.type = BTRFS_EXTENT_DATA_KEY;
193 datasize = btrfs_file_extent_calc_inline_size(cur_size);
194 path->leave_spinning = 1;
195 ret = btrfs_insert_empty_item(trans, root, path, &key,
200 leaf = path->nodes[0];
201 ei = btrfs_item_ptr(leaf, path->slots[0],
202 struct btrfs_file_extent_item);
203 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
204 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
205 btrfs_set_file_extent_encryption(leaf, ei, 0);
206 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
207 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
208 ptr = btrfs_file_extent_inline_start(ei);
210 if (compress_type != BTRFS_COMPRESS_NONE) {
213 while (compressed_size > 0) {
214 cpage = compressed_pages[i];
215 cur_size = min_t(unsigned long, compressed_size,
218 kaddr = kmap_atomic(cpage);
219 write_extent_buffer(leaf, kaddr, ptr, cur_size);
220 kunmap_atomic(kaddr);
224 compressed_size -= cur_size;
226 btrfs_set_file_extent_compression(leaf, ei,
229 page = find_get_page(inode->i_mapping,
230 start >> PAGE_SHIFT);
231 btrfs_set_file_extent_compression(leaf, ei, 0);
232 kaddr = kmap_atomic(page);
233 offset = start & (PAGE_SIZE - 1);
234 write_extent_buffer(leaf, kaddr + offset, ptr, size);
235 kunmap_atomic(kaddr);
238 btrfs_mark_buffer_dirty(leaf);
239 btrfs_release_path(path);
242 * we're an inline extent, so nobody can
243 * extend the file past i_size without locking
244 * a page we already have locked.
246 * We must do any isize and inode updates
247 * before we unlock the pages. Otherwise we
248 * could end up racing with unlink.
250 BTRFS_I(inode)->disk_i_size = inode->i_size;
251 ret = btrfs_update_inode(trans, root, inode);
259 * conditionally insert an inline extent into the file. This
260 * does the checks required to make sure the data is small enough
261 * to fit as an inline extent.
263 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
264 u64 end, size_t compressed_size,
266 struct page **compressed_pages)
268 struct btrfs_root *root = BTRFS_I(inode)->root;
269 struct btrfs_fs_info *fs_info = root->fs_info;
270 struct btrfs_trans_handle *trans;
271 u64 isize = i_size_read(inode);
272 u64 actual_end = min(end + 1, isize);
273 u64 inline_len = actual_end - start;
274 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
275 u64 data_len = inline_len;
277 struct btrfs_path *path;
278 int extent_inserted = 0;
279 u32 extent_item_size;
282 data_len = compressed_size;
285 actual_end > fs_info->sectorsize ||
286 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
288 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
290 data_len > fs_info->max_inline) {
294 path = btrfs_alloc_path();
298 trans = btrfs_join_transaction(root);
300 btrfs_free_path(path);
301 return PTR_ERR(trans);
303 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
305 if (compressed_size && compressed_pages)
306 extent_item_size = btrfs_file_extent_calc_inline_size(
309 extent_item_size = btrfs_file_extent_calc_inline_size(
312 ret = __btrfs_drop_extents(trans, root, inode, path,
313 start, aligned_end, NULL,
314 1, 1, extent_item_size, &extent_inserted);
316 btrfs_abort_transaction(trans, ret);
320 if (isize > actual_end)
321 inline_len = min_t(u64, isize, actual_end);
322 ret = insert_inline_extent(trans, path, extent_inserted,
324 inline_len, compressed_size,
325 compress_type, compressed_pages);
326 if (ret && ret != -ENOSPC) {
327 btrfs_abort_transaction(trans, ret);
329 } else if (ret == -ENOSPC) {
334 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
335 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
338 * Don't forget to free the reserved space, as for inlined extent
339 * it won't count as data extent, free them directly here.
340 * And at reserve time, it's always aligned to page size, so
341 * just free one page here.
343 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
344 btrfs_free_path(path);
345 btrfs_end_transaction(trans);
349 struct async_extent {
354 unsigned long nr_pages;
356 struct list_head list;
361 struct btrfs_fs_info *fs_info;
362 struct page *locked_page;
365 unsigned int write_flags;
366 struct list_head extents;
367 struct btrfs_work work;
370 static noinline int add_async_extent(struct async_cow *cow,
371 u64 start, u64 ram_size,
374 unsigned long nr_pages,
377 struct async_extent *async_extent;
379 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
380 BUG_ON(!async_extent); /* -ENOMEM */
381 async_extent->start = start;
382 async_extent->ram_size = ram_size;
383 async_extent->compressed_size = compressed_size;
384 async_extent->pages = pages;
385 async_extent->nr_pages = nr_pages;
386 async_extent->compress_type = compress_type;
387 list_add_tail(&async_extent->list, &cow->extents);
391 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
393 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
396 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
399 if (BTRFS_I(inode)->defrag_compress)
401 /* bad compression ratios */
402 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
404 if (btrfs_test_opt(fs_info, COMPRESS) ||
405 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
406 BTRFS_I(inode)->prop_compress)
407 return btrfs_compress_heuristic(inode, start, end);
411 static inline void inode_should_defrag(struct btrfs_inode *inode,
412 u64 start, u64 end, u64 num_bytes, u64 small_write)
414 /* If this is a small write inside eof, kick off a defrag */
415 if (num_bytes < small_write &&
416 (start > 0 || end + 1 < inode->disk_i_size))
417 btrfs_add_inode_defrag(NULL, inode);
421 * we create compressed extents in two phases. The first
422 * phase compresses a range of pages that have already been
423 * locked (both pages and state bits are locked).
425 * This is done inside an ordered work queue, and the compression
426 * is spread across many cpus. The actual IO submission is step
427 * two, and the ordered work queue takes care of making sure that
428 * happens in the same order things were put onto the queue by
429 * writepages and friends.
431 * If this code finds it can't get good compression, it puts an
432 * entry onto the work queue to write the uncompressed bytes. This
433 * makes sure that both compressed inodes and uncompressed inodes
434 * are written in the same order that the flusher thread sent them
437 static noinline void compress_file_range(struct inode *inode,
438 struct page *locked_page,
440 struct async_cow *async_cow,
443 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
444 u64 blocksize = fs_info->sectorsize;
446 u64 isize = i_size_read(inode);
448 struct page **pages = NULL;
449 unsigned long nr_pages;
450 unsigned long total_compressed = 0;
451 unsigned long total_in = 0;
454 int compress_type = fs_info->compress_type;
457 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
460 actual_end = min_t(u64, isize, end + 1);
463 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
464 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
465 nr_pages = min_t(unsigned long, nr_pages,
466 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
469 * we don't want to send crud past the end of i_size through
470 * compression, that's just a waste of CPU time. So, if the
471 * end of the file is before the start of our current
472 * requested range of bytes, we bail out to the uncompressed
473 * cleanup code that can deal with all of this.
475 * It isn't really the fastest way to fix things, but this is a
476 * very uncommon corner.
478 if (actual_end <= start)
479 goto cleanup_and_bail_uncompressed;
481 total_compressed = actual_end - start;
484 * skip compression for a small file range(<=blocksize) that
485 * isn't an inline extent, since it doesn't save disk space at all.
487 if (total_compressed <= blocksize &&
488 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
489 goto cleanup_and_bail_uncompressed;
491 total_compressed = min_t(unsigned long, total_compressed,
492 BTRFS_MAX_UNCOMPRESSED);
497 * we do compression for mount -o compress and when the
498 * inode has not been flagged as nocompress. This flag can
499 * change at any time if we discover bad compression ratios.
501 if (inode_need_compress(inode, start, end)) {
503 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
505 /* just bail out to the uncompressed code */
510 if (BTRFS_I(inode)->defrag_compress)
511 compress_type = BTRFS_I(inode)->defrag_compress;
512 else if (BTRFS_I(inode)->prop_compress)
513 compress_type = BTRFS_I(inode)->prop_compress;
516 * we need to call clear_page_dirty_for_io on each
517 * page in the range. Otherwise applications with the file
518 * mmap'd can wander in and change the page contents while
519 * we are compressing them.
521 * If the compression fails for any reason, we set the pages
522 * dirty again later on.
524 * Note that the remaining part is redirtied, the start pointer
525 * has moved, the end is the original one.
528 extent_range_clear_dirty_for_io(inode, start, end);
532 /* Compression level is applied here and only here */
533 ret = btrfs_compress_pages(
534 compress_type | (fs_info->compress_level << 4),
535 inode->i_mapping, start,
542 unsigned long offset = total_compressed &
544 struct page *page = pages[nr_pages - 1];
547 /* zero the tail end of the last page, we might be
548 * sending it down to disk
551 kaddr = kmap_atomic(page);
552 memset(kaddr + offset, 0,
554 kunmap_atomic(kaddr);
561 /* lets try to make an inline extent */
562 if (ret || total_in < actual_end) {
563 /* we didn't compress the entire range, try
564 * to make an uncompressed inline extent.
566 ret = cow_file_range_inline(inode, start, end, 0,
567 BTRFS_COMPRESS_NONE, NULL);
569 /* try making a compressed inline extent */
570 ret = cow_file_range_inline(inode, start, end,
572 compress_type, pages);
575 unsigned long clear_flags = EXTENT_DELALLOC |
576 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
577 EXTENT_DO_ACCOUNTING;
578 unsigned long page_error_op;
580 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
583 * inline extent creation worked or returned error,
584 * we don't need to create any more async work items.
585 * Unlock and free up our temp pages.
587 * We use DO_ACCOUNTING here because we need the
588 * delalloc_release_metadata to be done _after_ we drop
589 * our outstanding extent for clearing delalloc for this
592 extent_clear_unlock_delalloc(inode, start, end, end,
605 * we aren't doing an inline extent round the compressed size
606 * up to a block size boundary so the allocator does sane
609 total_compressed = ALIGN(total_compressed, blocksize);
612 * one last check to make sure the compression is really a
613 * win, compare the page count read with the blocks on disk,
614 * compression must free at least one sector size
616 total_in = ALIGN(total_in, PAGE_SIZE);
617 if (total_compressed + blocksize <= total_in) {
621 * The async work queues will take care of doing actual
622 * allocation on disk for these compressed pages, and
623 * will submit them to the elevator.
625 add_async_extent(async_cow, start, total_in,
626 total_compressed, pages, nr_pages,
629 if (start + total_in < end) {
640 * the compression code ran but failed to make things smaller,
641 * free any pages it allocated and our page pointer array
643 for (i = 0; i < nr_pages; i++) {
644 WARN_ON(pages[i]->mapping);
649 total_compressed = 0;
652 /* flag the file so we don't compress in the future */
653 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
654 !(BTRFS_I(inode)->prop_compress)) {
655 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
658 cleanup_and_bail_uncompressed:
660 * No compression, but we still need to write the pages in the file
661 * we've been given so far. redirty the locked page if it corresponds
662 * to our extent and set things up for the async work queue to run
663 * cow_file_range to do the normal delalloc dance.
665 if (page_offset(locked_page) >= start &&
666 page_offset(locked_page) <= end)
667 __set_page_dirty_nobuffers(locked_page);
668 /* unlocked later on in the async handlers */
671 extent_range_redirty_for_io(inode, start, end);
672 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
673 BTRFS_COMPRESS_NONE);
679 for (i = 0; i < nr_pages; i++) {
680 WARN_ON(pages[i]->mapping);
686 static void free_async_extent_pages(struct async_extent *async_extent)
690 if (!async_extent->pages)
693 for (i = 0; i < async_extent->nr_pages; i++) {
694 WARN_ON(async_extent->pages[i]->mapping);
695 put_page(async_extent->pages[i]);
697 kfree(async_extent->pages);
698 async_extent->nr_pages = 0;
699 async_extent->pages = NULL;
703 * phase two of compressed writeback. This is the ordered portion
704 * of the code, which only gets called in the order the work was
705 * queued. We walk all the async extents created by compress_file_range
706 * and send them down to the disk.
708 static noinline void submit_compressed_extents(struct inode *inode,
709 struct async_cow *async_cow)
711 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
712 struct async_extent *async_extent;
714 struct btrfs_key ins;
715 struct extent_map *em;
716 struct btrfs_root *root = BTRFS_I(inode)->root;
717 struct extent_io_tree *io_tree;
721 while (!list_empty(&async_cow->extents)) {
722 async_extent = list_entry(async_cow->extents.next,
723 struct async_extent, list);
724 list_del(&async_extent->list);
726 io_tree = &BTRFS_I(inode)->io_tree;
729 /* did the compression code fall back to uncompressed IO? */
730 if (!async_extent->pages) {
731 int page_started = 0;
732 unsigned long nr_written = 0;
734 lock_extent(io_tree, async_extent->start,
735 async_extent->start +
736 async_extent->ram_size - 1);
738 /* allocate blocks */
739 ret = cow_file_range(inode, async_cow->locked_page,
741 async_extent->start +
742 async_extent->ram_size - 1,
743 async_extent->start +
744 async_extent->ram_size - 1,
745 &page_started, &nr_written, 0,
751 * if page_started, cow_file_range inserted an
752 * inline extent and took care of all the unlocking
753 * and IO for us. Otherwise, we need to submit
754 * all those pages down to the drive.
756 if (!page_started && !ret)
757 extent_write_locked_range(inode,
759 async_extent->start +
760 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 async_cow->write_flags)) {
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 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
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;
946 u64 cur_alloc_size = 0;
947 u64 blocksize = fs_info->sectorsize;
948 struct btrfs_key ins;
949 struct extent_map *em;
951 unsigned long page_ops;
952 bool extent_reserved = false;
955 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
961 num_bytes = ALIGN(end - start + 1, blocksize);
962 num_bytes = max(blocksize, num_bytes);
963 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
965 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
968 /* lets try to make an inline extent */
969 ret = cow_file_range_inline(inode, start, end, 0,
970 BTRFS_COMPRESS_NONE, NULL);
973 * We use DO_ACCOUNTING here because we need the
974 * delalloc_release_metadata to be run _after_ we drop
975 * our outstanding extent for clearing delalloc for this
978 extent_clear_unlock_delalloc(inode, start, end,
980 EXTENT_LOCKED | EXTENT_DELALLOC |
981 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
982 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
983 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
985 *nr_written = *nr_written +
986 (end - start + PAGE_SIZE) / PAGE_SIZE;
989 } else if (ret < 0) {
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 (num_bytes > 0) {
999 cur_alloc_size = 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 */);
1021 free_extent_map(em);
1023 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1024 ram_size, cur_alloc_size, 0);
1026 goto out_drop_extent_cache;
1028 if (root->root_key.objectid ==
1029 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1030 ret = btrfs_reloc_clone_csums(inode, start,
1033 * Only drop cache here, and process as normal.
1035 * We must not allow extent_clear_unlock_delalloc()
1036 * at out_unlock label to free meta of this ordered
1037 * extent, as its meta should be freed by
1038 * btrfs_finish_ordered_io().
1040 * So we must continue until @start is increased to
1041 * skip current ordered extent.
1044 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1045 start + ram_size - 1, 0);
1048 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1050 /* we're not doing compressed IO, don't unlock the first
1051 * page (which the caller expects to stay locked), don't
1052 * clear any dirty bits and don't set any writeback bits
1054 * Do set the Private2 bit so we know this page was properly
1055 * setup for writepage
1057 page_ops = unlock ? PAGE_UNLOCK : 0;
1058 page_ops |= PAGE_SET_PRIVATE2;
1060 extent_clear_unlock_delalloc(inode, start,
1061 start + ram_size - 1,
1062 delalloc_end, locked_page,
1063 EXTENT_LOCKED | EXTENT_DELALLOC,
1065 if (num_bytes < cur_alloc_size)
1068 num_bytes -= cur_alloc_size;
1069 alloc_hint = ins.objectid + ins.offset;
1070 start += cur_alloc_size;
1071 extent_reserved = false;
1074 * btrfs_reloc_clone_csums() error, since start is increased
1075 * extent_clear_unlock_delalloc() at out_unlock label won't
1076 * free metadata of current ordered extent, we're OK to exit.
1084 out_drop_extent_cache:
1085 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1087 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1088 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1090 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1091 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1092 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1095 * If we reserved an extent for our delalloc range (or a subrange) and
1096 * failed to create the respective ordered extent, then it means that
1097 * when we reserved the extent we decremented the extent's size from
1098 * the data space_info's bytes_may_use counter and incremented the
1099 * space_info's bytes_reserved counter by the same amount. We must make
1100 * sure extent_clear_unlock_delalloc() does not try to decrement again
1101 * the data space_info's bytes_may_use counter, therefore we do not pass
1102 * it the flag EXTENT_CLEAR_DATA_RESV.
1104 if (extent_reserved) {
1105 extent_clear_unlock_delalloc(inode, start,
1106 start + cur_alloc_size,
1107 start + cur_alloc_size,
1111 start += cur_alloc_size;
1115 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1117 clear_bits | EXTENT_CLEAR_DATA_RESV,
1123 * work queue call back to started compression on a file and pages
1125 static noinline void async_cow_start(struct btrfs_work *work)
1127 struct async_cow *async_cow;
1129 async_cow = container_of(work, struct async_cow, work);
1131 compress_file_range(async_cow->inode, async_cow->locked_page,
1132 async_cow->start, async_cow->end, async_cow,
1134 if (num_added == 0) {
1135 btrfs_add_delayed_iput(async_cow->inode);
1136 async_cow->inode = NULL;
1141 * work queue call back to submit previously compressed pages
1143 static noinline void async_cow_submit(struct btrfs_work *work)
1145 struct btrfs_fs_info *fs_info;
1146 struct async_cow *async_cow;
1147 unsigned long nr_pages;
1149 async_cow = container_of(work, struct async_cow, work);
1151 fs_info = async_cow->fs_info;
1152 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1155 /* atomic_sub_return implies a barrier */
1156 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1158 cond_wake_up_nomb(&fs_info->async_submit_wait);
1160 if (async_cow->inode)
1161 submit_compressed_extents(async_cow->inode, async_cow);
1164 static noinline void async_cow_free(struct btrfs_work *work)
1166 struct async_cow *async_cow;
1167 async_cow = container_of(work, struct async_cow, work);
1168 if (async_cow->inode)
1169 btrfs_add_delayed_iput(async_cow->inode);
1173 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1174 u64 start, u64 end, int *page_started,
1175 unsigned long *nr_written,
1176 unsigned int write_flags)
1178 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1179 struct async_cow *async_cow;
1180 unsigned long nr_pages;
1183 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1185 while (start < end) {
1186 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1187 BUG_ON(!async_cow); /* -ENOMEM */
1188 async_cow->inode = igrab(inode);
1189 async_cow->fs_info = fs_info;
1190 async_cow->locked_page = locked_page;
1191 async_cow->start = start;
1192 async_cow->write_flags = write_flags;
1194 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1195 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1198 cur_end = min(end, start + SZ_512K - 1);
1200 async_cow->end = cur_end;
1201 INIT_LIST_HEAD(&async_cow->extents);
1203 btrfs_init_work(&async_cow->work,
1204 btrfs_delalloc_helper,
1205 async_cow_start, async_cow_submit,
1208 nr_pages = (cur_end - start + PAGE_SIZE) >>
1210 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1212 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1214 *nr_written += nr_pages;
1215 start = cur_end + 1;
1221 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1222 u64 bytenr, u64 num_bytes)
1225 struct btrfs_ordered_sum *sums;
1228 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1229 bytenr + num_bytes - 1, &list, 0);
1230 if (ret == 0 && list_empty(&list))
1233 while (!list_empty(&list)) {
1234 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1235 list_del(&sums->list);
1244 * when nowcow writeback call back. This checks for snapshots or COW copies
1245 * of the extents that exist in the file, and COWs the file as required.
1247 * If no cow copies or snapshots exist, we write directly to the existing
1250 static noinline int run_delalloc_nocow(struct inode *inode,
1251 struct page *locked_page,
1252 u64 start, u64 end, int *page_started, int force,
1253 unsigned long *nr_written)
1255 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1256 struct btrfs_root *root = BTRFS_I(inode)->root;
1257 struct extent_buffer *leaf;
1258 struct btrfs_path *path;
1259 struct btrfs_file_extent_item *fi;
1260 struct btrfs_key found_key;
1261 struct extent_map *em;
1276 u64 ino = btrfs_ino(BTRFS_I(inode));
1278 path = btrfs_alloc_path();
1280 extent_clear_unlock_delalloc(inode, start, end, end,
1282 EXTENT_LOCKED | EXTENT_DELALLOC |
1283 EXTENT_DO_ACCOUNTING |
1284 EXTENT_DEFRAG, PAGE_UNLOCK |
1286 PAGE_SET_WRITEBACK |
1287 PAGE_END_WRITEBACK);
1291 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1293 cow_start = (u64)-1;
1296 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1300 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1301 leaf = path->nodes[0];
1302 btrfs_item_key_to_cpu(leaf, &found_key,
1303 path->slots[0] - 1);
1304 if (found_key.objectid == ino &&
1305 found_key.type == BTRFS_EXTENT_DATA_KEY)
1310 leaf = path->nodes[0];
1311 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1312 ret = btrfs_next_leaf(root, path);
1314 if (cow_start != (u64)-1)
1315 cur_offset = cow_start;
1320 leaf = path->nodes[0];
1326 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1328 if (found_key.objectid > ino)
1330 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1331 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1335 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1336 found_key.offset > end)
1339 if (found_key.offset > cur_offset) {
1340 extent_end = found_key.offset;
1345 fi = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_file_extent_item);
1347 extent_type = btrfs_file_extent_type(leaf, fi);
1349 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1350 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1351 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1352 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1353 extent_offset = btrfs_file_extent_offset(leaf, fi);
1354 extent_end = found_key.offset +
1355 btrfs_file_extent_num_bytes(leaf, fi);
1357 btrfs_file_extent_disk_num_bytes(leaf, fi);
1358 if (extent_end <= start) {
1362 if (disk_bytenr == 0)
1364 if (btrfs_file_extent_compression(leaf, fi) ||
1365 btrfs_file_extent_encryption(leaf, fi) ||
1366 btrfs_file_extent_other_encoding(leaf, fi))
1369 * Do the same check as in btrfs_cross_ref_exist but
1370 * without the unnecessary search.
1373 btrfs_file_extent_generation(leaf, fi) <=
1374 btrfs_root_last_snapshot(&root->root_item))
1376 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1378 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1380 ret = btrfs_cross_ref_exist(root, ino,
1382 extent_offset, disk_bytenr);
1385 * ret could be -EIO if the above fails to read
1389 if (cow_start != (u64)-1)
1390 cur_offset = cow_start;
1394 WARN_ON_ONCE(nolock);
1397 disk_bytenr += extent_offset;
1398 disk_bytenr += cur_offset - found_key.offset;
1399 num_bytes = min(end + 1, extent_end) - cur_offset;
1401 * if there are pending snapshots for this root,
1402 * we fall into common COW way.
1404 if (!nolock && atomic_read(&root->snapshot_force_cow))
1407 * force cow if csum exists in the range.
1408 * this ensure that csum for a given extent are
1409 * either valid or do not exist.
1411 ret = csum_exist_in_range(fs_info, disk_bytenr,
1415 * ret could be -EIO if the above fails to read
1419 if (cow_start != (u64)-1)
1420 cur_offset = cow_start;
1423 WARN_ON_ONCE(nolock);
1426 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1429 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1430 extent_end = found_key.offset +
1431 btrfs_file_extent_ram_bytes(leaf, fi);
1432 extent_end = ALIGN(extent_end,
1433 fs_info->sectorsize);
1438 if (extent_end <= start) {
1441 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1445 if (cow_start == (u64)-1)
1446 cow_start = cur_offset;
1447 cur_offset = extent_end;
1448 if (cur_offset > end)
1454 btrfs_release_path(path);
1455 if (cow_start != (u64)-1) {
1456 ret = cow_file_range(inode, locked_page,
1457 cow_start, found_key.offset - 1,
1458 end, page_started, nr_written, 1,
1462 btrfs_dec_nocow_writers(fs_info,
1466 cow_start = (u64)-1;
1469 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1470 u64 orig_start = found_key.offset - extent_offset;
1472 em = create_io_em(inode, cur_offset, num_bytes,
1474 disk_bytenr, /* block_start */
1475 num_bytes, /* block_len */
1476 disk_num_bytes, /* orig_block_len */
1477 ram_bytes, BTRFS_COMPRESS_NONE,
1478 BTRFS_ORDERED_PREALLOC);
1481 btrfs_dec_nocow_writers(fs_info,
1486 free_extent_map(em);
1489 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1490 type = BTRFS_ORDERED_PREALLOC;
1492 type = BTRFS_ORDERED_NOCOW;
1495 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1496 num_bytes, num_bytes, type);
1498 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1499 BUG_ON(ret); /* -ENOMEM */
1501 if (root->root_key.objectid ==
1502 BTRFS_DATA_RELOC_TREE_OBJECTID)
1504 * Error handled later, as we must prevent
1505 * extent_clear_unlock_delalloc() in error handler
1506 * from freeing metadata of created ordered extent.
1508 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1511 extent_clear_unlock_delalloc(inode, cur_offset,
1512 cur_offset + num_bytes - 1, end,
1513 locked_page, EXTENT_LOCKED |
1515 EXTENT_CLEAR_DATA_RESV,
1516 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1518 cur_offset = extent_end;
1521 * btrfs_reloc_clone_csums() error, now we're OK to call error
1522 * handler, as metadata for created ordered extent will only
1523 * be freed by btrfs_finish_ordered_io().
1527 if (cur_offset > end)
1530 btrfs_release_path(path);
1532 if (cur_offset <= end && cow_start == (u64)-1)
1533 cow_start = cur_offset;
1535 if (cow_start != (u64)-1) {
1537 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1538 page_started, nr_written, 1, NULL);
1544 if (ret && cur_offset < end)
1545 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1546 locked_page, EXTENT_LOCKED |
1547 EXTENT_DELALLOC | EXTENT_DEFRAG |
1548 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1550 PAGE_SET_WRITEBACK |
1551 PAGE_END_WRITEBACK);
1552 btrfs_free_path(path);
1556 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1559 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1560 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1564 * @defrag_bytes is a hint value, no spinlock held here,
1565 * if is not zero, it means the file is defragging.
1566 * Force cow if given extent needs to be defragged.
1568 if (BTRFS_I(inode)->defrag_bytes &&
1569 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1570 EXTENT_DEFRAG, 0, NULL))
1577 * Function to process delayed allocation (create CoW) for ranges which are
1578 * being touched for the first time.
1580 int btrfs_run_delalloc_range(void *private_data, struct page *locked_page,
1581 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1582 struct writeback_control *wbc)
1584 struct inode *inode = private_data;
1586 int force_cow = need_force_cow(inode, start, end);
1587 unsigned int write_flags = wbc_to_write_flags(wbc);
1589 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1590 ret = run_delalloc_nocow(inode, locked_page, start, end,
1591 page_started, 1, nr_written);
1592 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1593 ret = run_delalloc_nocow(inode, locked_page, start, end,
1594 page_started, 0, nr_written);
1595 } else if (!inode_need_compress(inode, start, end)) {
1596 ret = cow_file_range(inode, locked_page, start, end, end,
1597 page_started, nr_written, 1, NULL);
1599 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1600 &BTRFS_I(inode)->runtime_flags);
1601 ret = cow_file_range_async(inode, locked_page, start, end,
1602 page_started, nr_written,
1606 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1610 void btrfs_split_delalloc_extent(struct inode *inode,
1611 struct extent_state *orig, u64 split)
1615 /* not delalloc, ignore it */
1616 if (!(orig->state & EXTENT_DELALLOC))
1619 size = orig->end - orig->start + 1;
1620 if (size > BTRFS_MAX_EXTENT_SIZE) {
1625 * See the explanation in btrfs_merge_delalloc_extent, the same
1626 * applies here, just in reverse.
1628 new_size = orig->end - split + 1;
1629 num_extents = count_max_extents(new_size);
1630 new_size = split - orig->start;
1631 num_extents += count_max_extents(new_size);
1632 if (count_max_extents(size) >= num_extents)
1636 spin_lock(&BTRFS_I(inode)->lock);
1637 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1638 spin_unlock(&BTRFS_I(inode)->lock);
1642 * Handle merged delayed allocation extents so we can keep track of new extents
1643 * that are just merged onto old extents, such as when we are doing sequential
1644 * writes, so we can properly account for the metadata space we'll need.
1646 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1647 struct extent_state *other)
1649 u64 new_size, old_size;
1652 /* not delalloc, ignore it */
1653 if (!(other->state & EXTENT_DELALLOC))
1656 if (new->start > other->start)
1657 new_size = new->end - other->start + 1;
1659 new_size = other->end - new->start + 1;
1661 /* we're not bigger than the max, unreserve the space and go */
1662 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1663 spin_lock(&BTRFS_I(inode)->lock);
1664 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1665 spin_unlock(&BTRFS_I(inode)->lock);
1670 * We have to add up either side to figure out how many extents were
1671 * accounted for before we merged into one big extent. If the number of
1672 * extents we accounted for is <= the amount we need for the new range
1673 * then we can return, otherwise drop. Think of it like this
1677 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1678 * need 2 outstanding extents, on one side we have 1 and the other side
1679 * we have 1 so they are == and we can return. But in this case
1681 * [MAX_SIZE+4k][MAX_SIZE+4k]
1683 * Each range on their own accounts for 2 extents, but merged together
1684 * they are only 3 extents worth of accounting, so we need to drop in
1687 old_size = other->end - other->start + 1;
1688 num_extents = count_max_extents(old_size);
1689 old_size = new->end - new->start + 1;
1690 num_extents += count_max_extents(old_size);
1691 if (count_max_extents(new_size) >= num_extents)
1694 spin_lock(&BTRFS_I(inode)->lock);
1695 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1696 spin_unlock(&BTRFS_I(inode)->lock);
1699 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1700 struct inode *inode)
1702 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1704 spin_lock(&root->delalloc_lock);
1705 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1706 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1707 &root->delalloc_inodes);
1708 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1709 &BTRFS_I(inode)->runtime_flags);
1710 root->nr_delalloc_inodes++;
1711 if (root->nr_delalloc_inodes == 1) {
1712 spin_lock(&fs_info->delalloc_root_lock);
1713 BUG_ON(!list_empty(&root->delalloc_root));
1714 list_add_tail(&root->delalloc_root,
1715 &fs_info->delalloc_roots);
1716 spin_unlock(&fs_info->delalloc_root_lock);
1719 spin_unlock(&root->delalloc_lock);
1723 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1724 struct btrfs_inode *inode)
1726 struct btrfs_fs_info *fs_info = root->fs_info;
1728 if (!list_empty(&inode->delalloc_inodes)) {
1729 list_del_init(&inode->delalloc_inodes);
1730 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1731 &inode->runtime_flags);
1732 root->nr_delalloc_inodes--;
1733 if (!root->nr_delalloc_inodes) {
1734 ASSERT(list_empty(&root->delalloc_inodes));
1735 spin_lock(&fs_info->delalloc_root_lock);
1736 BUG_ON(list_empty(&root->delalloc_root));
1737 list_del_init(&root->delalloc_root);
1738 spin_unlock(&fs_info->delalloc_root_lock);
1743 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1744 struct btrfs_inode *inode)
1746 spin_lock(&root->delalloc_lock);
1747 __btrfs_del_delalloc_inode(root, inode);
1748 spin_unlock(&root->delalloc_lock);
1752 * Properly track delayed allocation bytes in the inode and to maintain the
1753 * list of inodes that have pending delalloc work to be done.
1755 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
1758 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1760 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1763 * set_bit and clear bit hooks normally require _irqsave/restore
1764 * but in this case, we are only testing for the DELALLOC
1765 * bit, which is only set or cleared with irqs on
1767 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1768 struct btrfs_root *root = BTRFS_I(inode)->root;
1769 u64 len = state->end + 1 - state->start;
1770 u32 num_extents = count_max_extents(len);
1771 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1773 spin_lock(&BTRFS_I(inode)->lock);
1774 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1775 spin_unlock(&BTRFS_I(inode)->lock);
1777 /* For sanity tests */
1778 if (btrfs_is_testing(fs_info))
1781 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1782 fs_info->delalloc_batch);
1783 spin_lock(&BTRFS_I(inode)->lock);
1784 BTRFS_I(inode)->delalloc_bytes += len;
1785 if (*bits & EXTENT_DEFRAG)
1786 BTRFS_I(inode)->defrag_bytes += len;
1787 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1788 &BTRFS_I(inode)->runtime_flags))
1789 btrfs_add_delalloc_inodes(root, inode);
1790 spin_unlock(&BTRFS_I(inode)->lock);
1793 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1794 (*bits & EXTENT_DELALLOC_NEW)) {
1795 spin_lock(&BTRFS_I(inode)->lock);
1796 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1798 spin_unlock(&BTRFS_I(inode)->lock);
1803 * Once a range is no longer delalloc this function ensures that proper
1804 * accounting happens.
1806 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
1807 struct extent_state *state, unsigned *bits)
1809 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
1810 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
1811 u64 len = state->end + 1 - state->start;
1812 u32 num_extents = count_max_extents(len);
1814 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1815 spin_lock(&inode->lock);
1816 inode->defrag_bytes -= len;
1817 spin_unlock(&inode->lock);
1821 * set_bit and clear bit hooks normally require _irqsave/restore
1822 * but in this case, we are only testing for the DELALLOC
1823 * bit, which is only set or cleared with irqs on
1825 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1826 struct btrfs_root *root = inode->root;
1827 bool do_list = !btrfs_is_free_space_inode(inode);
1829 spin_lock(&inode->lock);
1830 btrfs_mod_outstanding_extents(inode, -num_extents);
1831 spin_unlock(&inode->lock);
1834 * We don't reserve metadata space for space cache inodes so we
1835 * don't need to call dellalloc_release_metadata if there is an
1838 if (*bits & EXTENT_CLEAR_META_RESV &&
1839 root != fs_info->tree_root)
1840 btrfs_delalloc_release_metadata(inode, len, false);
1842 /* For sanity tests. */
1843 if (btrfs_is_testing(fs_info))
1846 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1847 do_list && !(state->state & EXTENT_NORESERVE) &&
1848 (*bits & EXTENT_CLEAR_DATA_RESV))
1849 btrfs_free_reserved_data_space_noquota(
1853 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1854 fs_info->delalloc_batch);
1855 spin_lock(&inode->lock);
1856 inode->delalloc_bytes -= len;
1857 if (do_list && inode->delalloc_bytes == 0 &&
1858 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1859 &inode->runtime_flags))
1860 btrfs_del_delalloc_inode(root, inode);
1861 spin_unlock(&inode->lock);
1864 if ((state->state & EXTENT_DELALLOC_NEW) &&
1865 (*bits & EXTENT_DELALLOC_NEW)) {
1866 spin_lock(&inode->lock);
1867 ASSERT(inode->new_delalloc_bytes >= len);
1868 inode->new_delalloc_bytes -= len;
1869 spin_unlock(&inode->lock);
1874 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
1875 * in a chunk's stripe. This function ensures that bios do not span a
1878 * @page - The page we are about to add to the bio
1879 * @size - size we want to add to the bio
1880 * @bio - bio we want to ensure is smaller than a stripe
1881 * @bio_flags - flags of the bio
1883 * return 1 if page cannot be added to the bio
1884 * return 0 if page can be added to the bio
1885 * return error otherwise
1887 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
1888 unsigned long bio_flags)
1890 struct inode *inode = page->mapping->host;
1891 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1892 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1897 if (bio_flags & EXTENT_BIO_COMPRESSED)
1900 length = bio->bi_iter.bi_size;
1901 map_length = length;
1902 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1906 if (map_length < length + size)
1912 * in order to insert checksums into the metadata in large chunks,
1913 * we wait until bio submission time. All the pages in the bio are
1914 * checksummed and sums are attached onto the ordered extent record.
1916 * At IO completion time the cums attached on the ordered extent record
1917 * are inserted into the btree
1919 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1922 struct inode *inode = private_data;
1923 blk_status_t ret = 0;
1925 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1926 BUG_ON(ret); /* -ENOMEM */
1931 * extent_io.c submission hook. This does the right thing for csum calculation
1932 * on write, or reading the csums from the tree before a read.
1934 * Rules about async/sync submit,
1935 * a) read: sync submit
1937 * b) write without checksum: sync submit
1939 * c) write with checksum:
1940 * c-1) if bio is issued by fsync: sync submit
1941 * (sync_writers != 0)
1943 * c-2) if root is reloc root: sync submit
1944 * (only in case of buffered IO)
1946 * c-3) otherwise: async submit
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);
1990 } else if (!skip_sum) {
1991 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1997 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2001 bio->bi_status = ret;
2008 * given a list of ordered sums record them in the inode. This happens
2009 * at IO completion time based on sums calculated at bio submission time.
2011 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2012 struct inode *inode, struct list_head *list)
2014 struct btrfs_ordered_sum *sum;
2017 list_for_each_entry(sum, list, list) {
2018 trans->adding_csums = true;
2019 ret = btrfs_csum_file_blocks(trans,
2020 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2021 trans->adding_csums = false;
2028 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2029 unsigned int extra_bits,
2030 struct extent_state **cached_state, int dedupe)
2032 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2033 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2034 extra_bits, cached_state);
2037 /* see btrfs_writepage_start_hook for details on why this is required */
2038 struct btrfs_writepage_fixup {
2040 struct btrfs_work work;
2043 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2045 struct btrfs_writepage_fixup *fixup;
2046 struct btrfs_ordered_extent *ordered;
2047 struct extent_state *cached_state = NULL;
2048 struct extent_changeset *data_reserved = NULL;
2050 struct inode *inode;
2055 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2059 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2060 ClearPageChecked(page);
2064 inode = page->mapping->host;
2065 page_start = page_offset(page);
2066 page_end = page_offset(page) + PAGE_SIZE - 1;
2068 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2071 /* already ordered? We're done */
2072 if (PagePrivate2(page))
2075 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2078 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2079 page_end, &cached_state);
2081 btrfs_start_ordered_extent(inode, ordered, 1);
2082 btrfs_put_ordered_extent(ordered);
2086 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2089 mapping_set_error(page->mapping, ret);
2090 end_extent_writepage(page, ret, page_start, page_end);
2091 ClearPageChecked(page);
2095 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2098 mapping_set_error(page->mapping, ret);
2099 end_extent_writepage(page, ret, page_start, page_end);
2100 ClearPageChecked(page);
2104 ClearPageChecked(page);
2105 set_page_dirty(page);
2106 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2108 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2114 extent_changeset_free(data_reserved);
2118 * There are a few paths in the higher layers of the kernel that directly
2119 * set the page dirty bit without asking the filesystem if it is a
2120 * good idea. This causes problems because we want to make sure COW
2121 * properly happens and the data=ordered rules are followed.
2123 * In our case any range that doesn't have the ORDERED bit set
2124 * hasn't been properly setup for IO. We kick off an async process
2125 * to fix it up. The async helper will wait for ordered extents, set
2126 * the delalloc bit and make it safe to write the page.
2128 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2130 struct inode *inode = page->mapping->host;
2131 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2132 struct btrfs_writepage_fixup *fixup;
2134 /* this page is properly in the ordered list */
2135 if (TestClearPagePrivate2(page))
2138 if (PageChecked(page))
2141 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2145 SetPageChecked(page);
2147 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2148 btrfs_writepage_fixup_worker, NULL, NULL);
2150 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2154 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2155 struct inode *inode, u64 file_pos,
2156 u64 disk_bytenr, u64 disk_num_bytes,
2157 u64 num_bytes, u64 ram_bytes,
2158 u8 compression, u8 encryption,
2159 u16 other_encoding, int extent_type)
2161 struct btrfs_root *root = BTRFS_I(inode)->root;
2162 struct btrfs_file_extent_item *fi;
2163 struct btrfs_path *path;
2164 struct extent_buffer *leaf;
2165 struct btrfs_key ins;
2167 int extent_inserted = 0;
2170 path = btrfs_alloc_path();
2175 * we may be replacing one extent in the tree with another.
2176 * The new extent is pinned in the extent map, and we don't want
2177 * to drop it from the cache until it is completely in the btree.
2179 * So, tell btrfs_drop_extents to leave this extent in the cache.
2180 * the caller is expected to unpin it and allow it to be merged
2183 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2184 file_pos + num_bytes, NULL, 0,
2185 1, sizeof(*fi), &extent_inserted);
2189 if (!extent_inserted) {
2190 ins.objectid = btrfs_ino(BTRFS_I(inode));
2191 ins.offset = file_pos;
2192 ins.type = BTRFS_EXTENT_DATA_KEY;
2194 path->leave_spinning = 1;
2195 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2200 leaf = path->nodes[0];
2201 fi = btrfs_item_ptr(leaf, path->slots[0],
2202 struct btrfs_file_extent_item);
2203 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2204 btrfs_set_file_extent_type(leaf, fi, extent_type);
2205 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2206 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2207 btrfs_set_file_extent_offset(leaf, fi, 0);
2208 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2209 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2210 btrfs_set_file_extent_compression(leaf, fi, compression);
2211 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2212 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2214 btrfs_mark_buffer_dirty(leaf);
2215 btrfs_release_path(path);
2217 inode_add_bytes(inode, num_bytes);
2219 ins.objectid = disk_bytenr;
2220 ins.offset = disk_num_bytes;
2221 ins.type = BTRFS_EXTENT_ITEM_KEY;
2224 * Release the reserved range from inode dirty range map, as it is
2225 * already moved into delayed_ref_head
2227 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2231 ret = btrfs_alloc_reserved_file_extent(trans, root,
2232 btrfs_ino(BTRFS_I(inode)),
2233 file_pos, qg_released, &ins);
2235 btrfs_free_path(path);
2240 /* snapshot-aware defrag */
2241 struct sa_defrag_extent_backref {
2242 struct rb_node node;
2243 struct old_sa_defrag_extent *old;
2252 struct old_sa_defrag_extent {
2253 struct list_head list;
2254 struct new_sa_defrag_extent *new;
2263 struct new_sa_defrag_extent {
2264 struct rb_root root;
2265 struct list_head head;
2266 struct btrfs_path *path;
2267 struct inode *inode;
2275 static int backref_comp(struct sa_defrag_extent_backref *b1,
2276 struct sa_defrag_extent_backref *b2)
2278 if (b1->root_id < b2->root_id)
2280 else if (b1->root_id > b2->root_id)
2283 if (b1->inum < b2->inum)
2285 else if (b1->inum > b2->inum)
2288 if (b1->file_pos < b2->file_pos)
2290 else if (b1->file_pos > b2->file_pos)
2294 * [------------------------------] ===> (a range of space)
2295 * |<--->| |<---->| =============> (fs/file tree A)
2296 * |<---------------------------->| ===> (fs/file tree B)
2298 * A range of space can refer to two file extents in one tree while
2299 * refer to only one file extent in another tree.
2301 * So we may process a disk offset more than one time(two extents in A)
2302 * and locate at the same extent(one extent in B), then insert two same
2303 * backrefs(both refer to the extent in B).
2308 static void backref_insert(struct rb_root *root,
2309 struct sa_defrag_extent_backref *backref)
2311 struct rb_node **p = &root->rb_node;
2312 struct rb_node *parent = NULL;
2313 struct sa_defrag_extent_backref *entry;
2318 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2320 ret = backref_comp(backref, entry);
2324 p = &(*p)->rb_right;
2327 rb_link_node(&backref->node, parent, p);
2328 rb_insert_color(&backref->node, root);
2332 * Note the backref might has changed, and in this case we just return 0.
2334 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2337 struct btrfs_file_extent_item *extent;
2338 struct old_sa_defrag_extent *old = ctx;
2339 struct new_sa_defrag_extent *new = old->new;
2340 struct btrfs_path *path = new->path;
2341 struct btrfs_key key;
2342 struct btrfs_root *root;
2343 struct sa_defrag_extent_backref *backref;
2344 struct extent_buffer *leaf;
2345 struct inode *inode = new->inode;
2346 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2352 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2353 inum == btrfs_ino(BTRFS_I(inode)))
2356 key.objectid = root_id;
2357 key.type = BTRFS_ROOT_ITEM_KEY;
2358 key.offset = (u64)-1;
2360 root = btrfs_read_fs_root_no_name(fs_info, &key);
2362 if (PTR_ERR(root) == -ENOENT)
2365 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2366 inum, offset, root_id);
2367 return PTR_ERR(root);
2370 key.objectid = inum;
2371 key.type = BTRFS_EXTENT_DATA_KEY;
2372 if (offset > (u64)-1 << 32)
2375 key.offset = offset;
2377 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2378 if (WARN_ON(ret < 0))
2385 leaf = path->nodes[0];
2386 slot = path->slots[0];
2388 if (slot >= btrfs_header_nritems(leaf)) {
2389 ret = btrfs_next_leaf(root, path);
2392 } else if (ret > 0) {
2401 btrfs_item_key_to_cpu(leaf, &key, slot);
2403 if (key.objectid > inum)
2406 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2409 extent = btrfs_item_ptr(leaf, slot,
2410 struct btrfs_file_extent_item);
2412 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2416 * 'offset' refers to the exact key.offset,
2417 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2418 * (key.offset - extent_offset).
2420 if (key.offset != offset)
2423 extent_offset = btrfs_file_extent_offset(leaf, extent);
2424 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2426 if (extent_offset >= old->extent_offset + old->offset +
2427 old->len || extent_offset + num_bytes <=
2428 old->extent_offset + old->offset)
2433 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2439 backref->root_id = root_id;
2440 backref->inum = inum;
2441 backref->file_pos = offset;
2442 backref->num_bytes = num_bytes;
2443 backref->extent_offset = extent_offset;
2444 backref->generation = btrfs_file_extent_generation(leaf, extent);
2446 backref_insert(&new->root, backref);
2449 btrfs_release_path(path);
2454 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2455 struct new_sa_defrag_extent *new)
2457 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2458 struct old_sa_defrag_extent *old, *tmp;
2463 list_for_each_entry_safe(old, tmp, &new->head, list) {
2464 ret = iterate_inodes_from_logical(old->bytenr +
2465 old->extent_offset, fs_info,
2466 path, record_one_backref,
2468 if (ret < 0 && ret != -ENOENT)
2471 /* no backref to be processed for this extent */
2473 list_del(&old->list);
2478 if (list_empty(&new->head))
2484 static int relink_is_mergable(struct extent_buffer *leaf,
2485 struct btrfs_file_extent_item *fi,
2486 struct new_sa_defrag_extent *new)
2488 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2491 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2494 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2497 if (btrfs_file_extent_encryption(leaf, fi) ||
2498 btrfs_file_extent_other_encoding(leaf, fi))
2505 * Note the backref might has changed, and in this case we just return 0.
2507 static noinline int relink_extent_backref(struct btrfs_path *path,
2508 struct sa_defrag_extent_backref *prev,
2509 struct sa_defrag_extent_backref *backref)
2511 struct btrfs_file_extent_item *extent;
2512 struct btrfs_file_extent_item *item;
2513 struct btrfs_ordered_extent *ordered;
2514 struct btrfs_trans_handle *trans;
2515 struct btrfs_root *root;
2516 struct btrfs_key key;
2517 struct extent_buffer *leaf;
2518 struct old_sa_defrag_extent *old = backref->old;
2519 struct new_sa_defrag_extent *new = old->new;
2520 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2521 struct inode *inode;
2522 struct extent_state *cached = NULL;
2531 if (prev && prev->root_id == backref->root_id &&
2532 prev->inum == backref->inum &&
2533 prev->file_pos + prev->num_bytes == backref->file_pos)
2536 /* step 1: get root */
2537 key.objectid = backref->root_id;
2538 key.type = BTRFS_ROOT_ITEM_KEY;
2539 key.offset = (u64)-1;
2541 index = srcu_read_lock(&fs_info->subvol_srcu);
2543 root = btrfs_read_fs_root_no_name(fs_info, &key);
2545 srcu_read_unlock(&fs_info->subvol_srcu, index);
2546 if (PTR_ERR(root) == -ENOENT)
2548 return PTR_ERR(root);
2551 if (btrfs_root_readonly(root)) {
2552 srcu_read_unlock(&fs_info->subvol_srcu, index);
2556 /* step 2: get inode */
2557 key.objectid = backref->inum;
2558 key.type = BTRFS_INODE_ITEM_KEY;
2561 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2562 if (IS_ERR(inode)) {
2563 srcu_read_unlock(&fs_info->subvol_srcu, index);
2567 srcu_read_unlock(&fs_info->subvol_srcu, index);
2569 /* step 3: relink backref */
2570 lock_start = backref->file_pos;
2571 lock_end = backref->file_pos + backref->num_bytes - 1;
2572 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2575 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2577 btrfs_put_ordered_extent(ordered);
2581 trans = btrfs_join_transaction(root);
2582 if (IS_ERR(trans)) {
2583 ret = PTR_ERR(trans);
2587 key.objectid = backref->inum;
2588 key.type = BTRFS_EXTENT_DATA_KEY;
2589 key.offset = backref->file_pos;
2591 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2594 } else if (ret > 0) {
2599 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2600 struct btrfs_file_extent_item);
2602 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2603 backref->generation)
2606 btrfs_release_path(path);
2608 start = backref->file_pos;
2609 if (backref->extent_offset < old->extent_offset + old->offset)
2610 start += old->extent_offset + old->offset -
2611 backref->extent_offset;
2613 len = min(backref->extent_offset + backref->num_bytes,
2614 old->extent_offset + old->offset + old->len);
2615 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2617 ret = btrfs_drop_extents(trans, root, inode, start,
2622 key.objectid = btrfs_ino(BTRFS_I(inode));
2623 key.type = BTRFS_EXTENT_DATA_KEY;
2626 path->leave_spinning = 1;
2628 struct btrfs_file_extent_item *fi;
2630 struct btrfs_key found_key;
2632 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2637 leaf = path->nodes[0];
2638 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2640 fi = btrfs_item_ptr(leaf, path->slots[0],
2641 struct btrfs_file_extent_item);
2642 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2644 if (extent_len + found_key.offset == start &&
2645 relink_is_mergable(leaf, fi, new)) {
2646 btrfs_set_file_extent_num_bytes(leaf, fi,
2648 btrfs_mark_buffer_dirty(leaf);
2649 inode_add_bytes(inode, len);
2655 btrfs_release_path(path);
2660 ret = btrfs_insert_empty_item(trans, root, path, &key,
2663 btrfs_abort_transaction(trans, ret);
2667 leaf = path->nodes[0];
2668 item = btrfs_item_ptr(leaf, path->slots[0],
2669 struct btrfs_file_extent_item);
2670 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2671 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2672 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2673 btrfs_set_file_extent_num_bytes(leaf, item, len);
2674 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2675 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2676 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2677 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2678 btrfs_set_file_extent_encryption(leaf, item, 0);
2679 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2681 btrfs_mark_buffer_dirty(leaf);
2682 inode_add_bytes(inode, len);
2683 btrfs_release_path(path);
2685 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2687 backref->root_id, backref->inum,
2688 new->file_pos); /* start - extent_offset */
2690 btrfs_abort_transaction(trans, ret);
2696 btrfs_release_path(path);
2697 path->leave_spinning = 0;
2698 btrfs_end_transaction(trans);
2700 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2706 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2708 struct old_sa_defrag_extent *old, *tmp;
2713 list_for_each_entry_safe(old, tmp, &new->head, list) {
2719 static void relink_file_extents(struct new_sa_defrag_extent *new)
2721 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2722 struct btrfs_path *path;
2723 struct sa_defrag_extent_backref *backref;
2724 struct sa_defrag_extent_backref *prev = NULL;
2725 struct rb_node *node;
2728 path = btrfs_alloc_path();
2732 if (!record_extent_backrefs(path, new)) {
2733 btrfs_free_path(path);
2736 btrfs_release_path(path);
2739 node = rb_first(&new->root);
2742 rb_erase(node, &new->root);
2744 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2746 ret = relink_extent_backref(path, prev, backref);
2759 btrfs_free_path(path);
2761 free_sa_defrag_extent(new);
2763 atomic_dec(&fs_info->defrag_running);
2764 wake_up(&fs_info->transaction_wait);
2767 static struct new_sa_defrag_extent *
2768 record_old_file_extents(struct inode *inode,
2769 struct btrfs_ordered_extent *ordered)
2771 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2772 struct btrfs_root *root = BTRFS_I(inode)->root;
2773 struct btrfs_path *path;
2774 struct btrfs_key key;
2775 struct old_sa_defrag_extent *old;
2776 struct new_sa_defrag_extent *new;
2779 new = kmalloc(sizeof(*new), GFP_NOFS);
2784 new->file_pos = ordered->file_offset;
2785 new->len = ordered->len;
2786 new->bytenr = ordered->start;
2787 new->disk_len = ordered->disk_len;
2788 new->compress_type = ordered->compress_type;
2789 new->root = RB_ROOT;
2790 INIT_LIST_HEAD(&new->head);
2792 path = btrfs_alloc_path();
2796 key.objectid = btrfs_ino(BTRFS_I(inode));
2797 key.type = BTRFS_EXTENT_DATA_KEY;
2798 key.offset = new->file_pos;
2800 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2803 if (ret > 0 && path->slots[0] > 0)
2806 /* find out all the old extents for the file range */
2808 struct btrfs_file_extent_item *extent;
2809 struct extent_buffer *l;
2818 slot = path->slots[0];
2820 if (slot >= btrfs_header_nritems(l)) {
2821 ret = btrfs_next_leaf(root, path);
2829 btrfs_item_key_to_cpu(l, &key, slot);
2831 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2833 if (key.type != BTRFS_EXTENT_DATA_KEY)
2835 if (key.offset >= new->file_pos + new->len)
2838 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2840 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2841 if (key.offset + num_bytes < new->file_pos)
2844 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2848 extent_offset = btrfs_file_extent_offset(l, extent);
2850 old = kmalloc(sizeof(*old), GFP_NOFS);
2854 offset = max(new->file_pos, key.offset);
2855 end = min(new->file_pos + new->len, key.offset + num_bytes);
2857 old->bytenr = disk_bytenr;
2858 old->extent_offset = extent_offset;
2859 old->offset = offset - key.offset;
2860 old->len = end - offset;
2863 list_add_tail(&old->list, &new->head);
2869 btrfs_free_path(path);
2870 atomic_inc(&fs_info->defrag_running);
2875 btrfs_free_path(path);
2877 free_sa_defrag_extent(new);
2881 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2884 struct btrfs_block_group_cache *cache;
2886 cache = btrfs_lookup_block_group(fs_info, start);
2889 spin_lock(&cache->lock);
2890 cache->delalloc_bytes -= len;
2891 spin_unlock(&cache->lock);
2893 btrfs_put_block_group(cache);
2896 /* as ordered data IO finishes, this gets called so we can finish
2897 * an ordered extent if the range of bytes in the file it covers are
2900 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2902 struct inode *inode = ordered_extent->inode;
2903 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2904 struct btrfs_root *root = BTRFS_I(inode)->root;
2905 struct btrfs_trans_handle *trans = NULL;
2906 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2907 struct extent_state *cached_state = NULL;
2908 struct new_sa_defrag_extent *new = NULL;
2909 int compress_type = 0;
2911 u64 logical_len = ordered_extent->len;
2913 bool truncated = false;
2914 bool range_locked = false;
2915 bool clear_new_delalloc_bytes = false;
2916 bool clear_reserved_extent = true;
2918 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2919 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2920 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2921 clear_new_delalloc_bytes = true;
2923 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2925 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2930 btrfs_free_io_failure_record(BTRFS_I(inode),
2931 ordered_extent->file_offset,
2932 ordered_extent->file_offset +
2933 ordered_extent->len - 1);
2935 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2937 logical_len = ordered_extent->truncated_len;
2938 /* Truncated the entire extent, don't bother adding */
2943 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2944 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2947 * For mwrite(mmap + memset to write) case, we still reserve
2948 * space for NOCOW range.
2949 * As NOCOW won't cause a new delayed ref, just free the space
2951 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2952 ordered_extent->len);
2953 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2955 trans = btrfs_join_transaction_nolock(root);
2957 trans = btrfs_join_transaction(root);
2958 if (IS_ERR(trans)) {
2959 ret = PTR_ERR(trans);
2963 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2964 ret = btrfs_update_inode_fallback(trans, root, inode);
2965 if (ret) /* -ENOMEM or corruption */
2966 btrfs_abort_transaction(trans, ret);
2970 range_locked = true;
2971 lock_extent_bits(io_tree, ordered_extent->file_offset,
2972 ordered_extent->file_offset + ordered_extent->len - 1,
2975 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2976 ordered_extent->file_offset + ordered_extent->len - 1,
2977 EXTENT_DEFRAG, 0, cached_state);
2979 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2980 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2981 /* the inode is shared */
2982 new = record_old_file_extents(inode, ordered_extent);
2984 clear_extent_bit(io_tree, ordered_extent->file_offset,
2985 ordered_extent->file_offset + ordered_extent->len - 1,
2986 EXTENT_DEFRAG, 0, 0, &cached_state);
2990 trans = btrfs_join_transaction_nolock(root);
2992 trans = btrfs_join_transaction(root);
2993 if (IS_ERR(trans)) {
2994 ret = PTR_ERR(trans);
2999 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3001 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3002 compress_type = ordered_extent->compress_type;
3003 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3004 BUG_ON(compress_type);
3005 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3006 ordered_extent->len);
3007 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3008 ordered_extent->file_offset,
3009 ordered_extent->file_offset +
3012 BUG_ON(root == fs_info->tree_root);
3013 ret = insert_reserved_file_extent(trans, inode,
3014 ordered_extent->file_offset,
3015 ordered_extent->start,
3016 ordered_extent->disk_len,
3017 logical_len, logical_len,
3018 compress_type, 0, 0,
3019 BTRFS_FILE_EXTENT_REG);
3021 clear_reserved_extent = false;
3022 btrfs_release_delalloc_bytes(fs_info,
3023 ordered_extent->start,
3024 ordered_extent->disk_len);
3027 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3028 ordered_extent->file_offset, ordered_extent->len,
3031 btrfs_abort_transaction(trans, ret);
3035 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3037 btrfs_abort_transaction(trans, ret);
3041 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3042 ret = btrfs_update_inode_fallback(trans, root, inode);
3043 if (ret) { /* -ENOMEM or corruption */
3044 btrfs_abort_transaction(trans, ret);
3049 if (range_locked || clear_new_delalloc_bytes) {
3050 unsigned int clear_bits = 0;
3053 clear_bits |= EXTENT_LOCKED;
3054 if (clear_new_delalloc_bytes)
3055 clear_bits |= EXTENT_DELALLOC_NEW;
3056 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3057 ordered_extent->file_offset,
3058 ordered_extent->file_offset +
3059 ordered_extent->len - 1,
3061 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3066 btrfs_end_transaction(trans);
3068 if (ret || truncated) {
3072 start = ordered_extent->file_offset + logical_len;
3074 start = ordered_extent->file_offset;
3075 end = ordered_extent->file_offset + ordered_extent->len - 1;
3076 clear_extent_uptodate(io_tree, start, end, NULL);
3078 /* Drop the cache for the part of the extent we didn't write. */
3079 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3082 * If the ordered extent had an IOERR or something else went
3083 * wrong we need to return the space for this ordered extent
3084 * back to the allocator. We only free the extent in the
3085 * truncated case if we didn't write out the extent at all.
3087 * If we made it past insert_reserved_file_extent before we
3088 * errored out then we don't need to do this as the accounting
3089 * has already been done.
3091 if ((ret || !logical_len) &&
3092 clear_reserved_extent &&
3093 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3094 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3095 btrfs_free_reserved_extent(fs_info,
3096 ordered_extent->start,
3097 ordered_extent->disk_len, 1);
3102 * This needs to be done to make sure anybody waiting knows we are done
3103 * updating everything for this ordered extent.
3105 btrfs_remove_ordered_extent(inode, ordered_extent);
3107 /* for snapshot-aware defrag */
3110 free_sa_defrag_extent(new);
3111 atomic_dec(&fs_info->defrag_running);
3113 relink_file_extents(new);
3118 btrfs_put_ordered_extent(ordered_extent);
3119 /* once for the tree */
3120 btrfs_put_ordered_extent(ordered_extent);
3122 /* Try to release some metadata so we don't get an OOM but don't wait */
3123 btrfs_btree_balance_dirty_nodelay(fs_info);
3128 static void finish_ordered_fn(struct btrfs_work *work)
3130 struct btrfs_ordered_extent *ordered_extent;
3131 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3132 btrfs_finish_ordered_io(ordered_extent);
3135 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
3136 u64 end, int uptodate)
3138 struct inode *inode = page->mapping->host;
3139 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3140 struct btrfs_ordered_extent *ordered_extent = NULL;
3141 struct btrfs_workqueue *wq;
3142 btrfs_work_func_t func;
3144 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3146 ClearPagePrivate2(page);
3147 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3148 end - start + 1, uptodate))
3151 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3152 wq = fs_info->endio_freespace_worker;
3153 func = btrfs_freespace_write_helper;
3155 wq = fs_info->endio_write_workers;
3156 func = btrfs_endio_write_helper;
3159 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3161 btrfs_queue_work(wq, &ordered_extent->work);
3164 static int __readpage_endio_check(struct inode *inode,
3165 struct btrfs_io_bio *io_bio,
3166 int icsum, struct page *page,
3167 int pgoff, u64 start, size_t len)
3173 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3175 kaddr = kmap_atomic(page);
3176 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3177 btrfs_csum_final(csum, (u8 *)&csum);
3178 if (csum != csum_expected)
3181 kunmap_atomic(kaddr);
3184 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3185 io_bio->mirror_num);
3186 memset(kaddr + pgoff, 1, len);
3187 flush_dcache_page(page);
3188 kunmap_atomic(kaddr);
3193 * when reads are done, we need to check csums to verify the data is correct
3194 * if there's a match, we allow the bio to finish. If not, the code in
3195 * extent_io.c will try to find good copies for us.
3197 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3198 u64 phy_offset, struct page *page,
3199 u64 start, u64 end, int mirror)
3201 size_t offset = start - page_offset(page);
3202 struct inode *inode = page->mapping->host;
3203 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3204 struct btrfs_root *root = BTRFS_I(inode)->root;
3206 if (PageChecked(page)) {
3207 ClearPageChecked(page);
3211 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3214 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3215 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3216 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3220 phy_offset >>= inode->i_sb->s_blocksize_bits;
3221 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3222 start, (size_t)(end - start + 1));
3226 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3228 * @inode: The inode we want to perform iput on
3230 * This function uses the generic vfs_inode::i_count to track whether we should
3231 * just decrement it (in case it's > 1) or if this is the last iput then link
3232 * the inode to the delayed iput machinery. Delayed iputs are processed at
3233 * transaction commit time/superblock commit/cleaner kthread.
3235 void btrfs_add_delayed_iput(struct inode *inode)
3237 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3238 struct btrfs_inode *binode = BTRFS_I(inode);
3240 if (atomic_add_unless(&inode->i_count, -1, 1))
3243 spin_lock(&fs_info->delayed_iput_lock);
3244 ASSERT(list_empty(&binode->delayed_iput));
3245 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3246 spin_unlock(&fs_info->delayed_iput_lock);
3249 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3252 spin_lock(&fs_info->delayed_iput_lock);
3253 while (!list_empty(&fs_info->delayed_iputs)) {
3254 struct btrfs_inode *inode;
3256 inode = list_first_entry(&fs_info->delayed_iputs,
3257 struct btrfs_inode, delayed_iput);
3258 list_del_init(&inode->delayed_iput);
3259 spin_unlock(&fs_info->delayed_iput_lock);
3260 iput(&inode->vfs_inode);
3261 spin_lock(&fs_info->delayed_iput_lock);
3263 spin_unlock(&fs_info->delayed_iput_lock);
3267 * This creates an orphan entry for the given inode in case something goes wrong
3268 * in the middle of an unlink.
3270 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3271 struct btrfs_inode *inode)
3275 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3276 if (ret && ret != -EEXIST) {
3277 btrfs_abort_transaction(trans, ret);
3285 * We have done the delete so we can go ahead and remove the orphan item for
3286 * this particular inode.
3288 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3289 struct btrfs_inode *inode)
3291 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3295 * this cleans up any orphans that may be left on the list from the last use
3298 int btrfs_orphan_cleanup(struct btrfs_root *root)
3300 struct btrfs_fs_info *fs_info = root->fs_info;
3301 struct btrfs_path *path;
3302 struct extent_buffer *leaf;
3303 struct btrfs_key key, found_key;
3304 struct btrfs_trans_handle *trans;
3305 struct inode *inode;
3306 u64 last_objectid = 0;
3307 int ret = 0, nr_unlink = 0;
3309 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3312 path = btrfs_alloc_path();
3317 path->reada = READA_BACK;
3319 key.objectid = BTRFS_ORPHAN_OBJECTID;
3320 key.type = BTRFS_ORPHAN_ITEM_KEY;
3321 key.offset = (u64)-1;
3324 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3329 * if ret == 0 means we found what we were searching for, which
3330 * is weird, but possible, so only screw with path if we didn't
3331 * find the key and see if we have stuff that matches
3335 if (path->slots[0] == 0)
3340 /* pull out the item */
3341 leaf = path->nodes[0];
3342 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3344 /* make sure the item matches what we want */
3345 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3347 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3350 /* release the path since we're done with it */
3351 btrfs_release_path(path);
3354 * this is where we are basically btrfs_lookup, without the
3355 * crossing root thing. we store the inode number in the
3356 * offset of the orphan item.
3359 if (found_key.offset == last_objectid) {
3361 "Error removing orphan entry, stopping orphan cleanup");
3366 last_objectid = found_key.offset;
3368 found_key.objectid = found_key.offset;
3369 found_key.type = BTRFS_INODE_ITEM_KEY;
3370 found_key.offset = 0;
3371 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3372 ret = PTR_ERR_OR_ZERO(inode);
3373 if (ret && ret != -ENOENT)
3376 if (ret == -ENOENT && root == fs_info->tree_root) {
3377 struct btrfs_root *dead_root;
3378 struct btrfs_fs_info *fs_info = root->fs_info;
3379 int is_dead_root = 0;
3382 * this is an orphan in the tree root. Currently these
3383 * could come from 2 sources:
3384 * a) a snapshot deletion in progress
3385 * b) a free space cache inode
3386 * We need to distinguish those two, as the snapshot
3387 * orphan must not get deleted.
3388 * find_dead_roots already ran before us, so if this
3389 * is a snapshot deletion, we should find the root
3390 * in the dead_roots list
3392 spin_lock(&fs_info->trans_lock);
3393 list_for_each_entry(dead_root, &fs_info->dead_roots,
3395 if (dead_root->root_key.objectid ==
3396 found_key.objectid) {
3401 spin_unlock(&fs_info->trans_lock);
3403 /* prevent this orphan from being found again */
3404 key.offset = found_key.objectid - 1;
3411 * If we have an inode with links, there are a couple of
3412 * possibilities. Old kernels (before v3.12) used to create an
3413 * orphan item for truncate indicating that there were possibly
3414 * extent items past i_size that needed to be deleted. In v3.12,
3415 * truncate was changed to update i_size in sync with the extent
3416 * items, but the (useless) orphan item was still created. Since
3417 * v4.18, we don't create the orphan item for truncate at all.
3419 * So, this item could mean that we need to do a truncate, but
3420 * only if this filesystem was last used on a pre-v3.12 kernel
3421 * and was not cleanly unmounted. The odds of that are quite
3422 * slim, and it's a pain to do the truncate now, so just delete
3425 * It's also possible that this orphan item was supposed to be
3426 * deleted but wasn't. The inode number may have been reused,
3427 * but either way, we can delete the orphan item.
3429 if (ret == -ENOENT || inode->i_nlink) {
3432 trans = btrfs_start_transaction(root, 1);
3433 if (IS_ERR(trans)) {
3434 ret = PTR_ERR(trans);
3437 btrfs_debug(fs_info, "auto deleting %Lu",
3438 found_key.objectid);
3439 ret = btrfs_del_orphan_item(trans, root,
3440 found_key.objectid);
3441 btrfs_end_transaction(trans);
3449 /* this will do delete_inode and everything for us */
3452 /* release the path since we're done with it */
3453 btrfs_release_path(path);
3455 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3457 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3458 trans = btrfs_join_transaction(root);
3460 btrfs_end_transaction(trans);
3464 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3468 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3469 btrfs_free_path(path);
3474 * very simple check to peek ahead in the leaf looking for xattrs. If we
3475 * don't find any xattrs, we know there can't be any acls.
3477 * slot is the slot the inode is in, objectid is the objectid of the inode
3479 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3480 int slot, u64 objectid,
3481 int *first_xattr_slot)
3483 u32 nritems = btrfs_header_nritems(leaf);
3484 struct btrfs_key found_key;
3485 static u64 xattr_access = 0;
3486 static u64 xattr_default = 0;
3489 if (!xattr_access) {
3490 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3491 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3492 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3493 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3497 *first_xattr_slot = -1;
3498 while (slot < nritems) {
3499 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3501 /* we found a different objectid, there must not be acls */
3502 if (found_key.objectid != objectid)
3505 /* we found an xattr, assume we've got an acl */
3506 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3507 if (*first_xattr_slot == -1)
3508 *first_xattr_slot = slot;
3509 if (found_key.offset == xattr_access ||
3510 found_key.offset == xattr_default)
3515 * we found a key greater than an xattr key, there can't
3516 * be any acls later on
3518 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3525 * it goes inode, inode backrefs, xattrs, extents,
3526 * so if there are a ton of hard links to an inode there can
3527 * be a lot of backrefs. Don't waste time searching too hard,
3528 * this is just an optimization
3533 /* we hit the end of the leaf before we found an xattr or
3534 * something larger than an xattr. We have to assume the inode
3537 if (*first_xattr_slot == -1)
3538 *first_xattr_slot = slot;
3543 * read an inode from the btree into the in-memory inode
3545 static int btrfs_read_locked_inode(struct inode *inode,
3546 struct btrfs_path *in_path)
3548 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3549 struct btrfs_path *path = in_path;
3550 struct extent_buffer *leaf;
3551 struct btrfs_inode_item *inode_item;
3552 struct btrfs_root *root = BTRFS_I(inode)->root;
3553 struct btrfs_key location;
3558 bool filled = false;
3559 int first_xattr_slot;
3561 ret = btrfs_fill_inode(inode, &rdev);
3566 path = btrfs_alloc_path();
3571 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3573 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3575 if (path != in_path)
3576 btrfs_free_path(path);
3580 leaf = path->nodes[0];
3585 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3586 struct btrfs_inode_item);
3587 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3588 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3589 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3590 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3591 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3593 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3594 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3596 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3597 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3599 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3600 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3602 BTRFS_I(inode)->i_otime.tv_sec =
3603 btrfs_timespec_sec(leaf, &inode_item->otime);
3604 BTRFS_I(inode)->i_otime.tv_nsec =
3605 btrfs_timespec_nsec(leaf, &inode_item->otime);
3607 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3608 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3609 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3611 inode_set_iversion_queried(inode,
3612 btrfs_inode_sequence(leaf, inode_item));
3613 inode->i_generation = BTRFS_I(inode)->generation;
3615 rdev = btrfs_inode_rdev(leaf, inode_item);
3617 BTRFS_I(inode)->index_cnt = (u64)-1;
3618 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3622 * If we were modified in the current generation and evicted from memory
3623 * and then re-read we need to do a full sync since we don't have any
3624 * idea about which extents were modified before we were evicted from
3627 * This is required for both inode re-read from disk and delayed inode
3628 * in delayed_nodes_tree.
3630 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3631 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3632 &BTRFS_I(inode)->runtime_flags);
3635 * We don't persist the id of the transaction where an unlink operation
3636 * against the inode was last made. So here we assume the inode might
3637 * have been evicted, and therefore the exact value of last_unlink_trans
3638 * lost, and set it to last_trans to avoid metadata inconsistencies
3639 * between the inode and its parent if the inode is fsync'ed and the log
3640 * replayed. For example, in the scenario:
3643 * ln mydir/foo mydir/bar
3646 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3647 * xfs_io -c fsync mydir/foo
3649 * mount fs, triggers fsync log replay
3651 * We must make sure that when we fsync our inode foo we also log its
3652 * parent inode, otherwise after log replay the parent still has the
3653 * dentry with the "bar" name but our inode foo has a link count of 1
3654 * and doesn't have an inode ref with the name "bar" anymore.
3656 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3657 * but it guarantees correctness at the expense of occasional full
3658 * transaction commits on fsync if our inode is a directory, or if our
3659 * inode is not a directory, logging its parent unnecessarily.
3661 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3663 * Similar reasoning for last_link_trans, needs to be set otherwise
3664 * for a case like the following:
3669 * echo 2 > /proc/sys/vm/drop_caches
3673 * Would result in link bar and directory A not existing after the power
3676 BTRFS_I(inode)->last_link_trans = BTRFS_I(inode)->last_trans;
3679 if (inode->i_nlink != 1 ||
3680 path->slots[0] >= btrfs_header_nritems(leaf))
3683 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3684 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3687 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3688 if (location.type == BTRFS_INODE_REF_KEY) {
3689 struct btrfs_inode_ref *ref;
3691 ref = (struct btrfs_inode_ref *)ptr;
3692 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3693 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3694 struct btrfs_inode_extref *extref;
3696 extref = (struct btrfs_inode_extref *)ptr;
3697 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3702 * try to precache a NULL acl entry for files that don't have
3703 * any xattrs or acls
3705 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3706 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3707 if (first_xattr_slot != -1) {
3708 path->slots[0] = first_xattr_slot;
3709 ret = btrfs_load_inode_props(inode, path);
3712 "error loading props for ino %llu (root %llu): %d",
3713 btrfs_ino(BTRFS_I(inode)),
3714 root->root_key.objectid, ret);
3716 if (path != in_path)
3717 btrfs_free_path(path);
3720 cache_no_acl(inode);
3722 switch (inode->i_mode & S_IFMT) {
3724 inode->i_mapping->a_ops = &btrfs_aops;
3725 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3726 inode->i_fop = &btrfs_file_operations;
3727 inode->i_op = &btrfs_file_inode_operations;
3730 inode->i_fop = &btrfs_dir_file_operations;
3731 inode->i_op = &btrfs_dir_inode_operations;
3734 inode->i_op = &btrfs_symlink_inode_operations;
3735 inode_nohighmem(inode);
3736 inode->i_mapping->a_ops = &btrfs_aops;
3739 inode->i_op = &btrfs_special_inode_operations;
3740 init_special_inode(inode, inode->i_mode, rdev);
3744 btrfs_sync_inode_flags_to_i_flags(inode);
3749 * given a leaf and an inode, copy the inode fields into the leaf
3751 static void fill_inode_item(struct btrfs_trans_handle *trans,
3752 struct extent_buffer *leaf,
3753 struct btrfs_inode_item *item,
3754 struct inode *inode)
3756 struct btrfs_map_token token;
3758 btrfs_init_map_token(&token);
3760 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3761 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3762 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3764 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3765 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3767 btrfs_set_token_timespec_sec(leaf, &item->atime,
3768 inode->i_atime.tv_sec, &token);
3769 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3770 inode->i_atime.tv_nsec, &token);
3772 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3773 inode->i_mtime.tv_sec, &token);
3774 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3775 inode->i_mtime.tv_nsec, &token);
3777 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3778 inode->i_ctime.tv_sec, &token);
3779 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3780 inode->i_ctime.tv_nsec, &token);
3782 btrfs_set_token_timespec_sec(leaf, &item->otime,
3783 BTRFS_I(inode)->i_otime.tv_sec, &token);
3784 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3785 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3787 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3789 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3791 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3793 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3794 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3795 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3796 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3800 * copy everything in the in-memory inode into the btree.
3802 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3803 struct btrfs_root *root, struct inode *inode)
3805 struct btrfs_inode_item *inode_item;
3806 struct btrfs_path *path;
3807 struct extent_buffer *leaf;
3810 path = btrfs_alloc_path();
3814 path->leave_spinning = 1;
3815 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3823 leaf = path->nodes[0];
3824 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3825 struct btrfs_inode_item);
3827 fill_inode_item(trans, leaf, inode_item, inode);
3828 btrfs_mark_buffer_dirty(leaf);
3829 btrfs_set_inode_last_trans(trans, inode);
3832 btrfs_free_path(path);
3837 * copy everything in the in-memory inode into the btree.
3839 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3840 struct btrfs_root *root, struct inode *inode)
3842 struct btrfs_fs_info *fs_info = root->fs_info;
3846 * If the inode is a free space inode, we can deadlock during commit
3847 * if we put it into the delayed code.
3849 * The data relocation inode should also be directly updated
3852 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3853 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3854 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3855 btrfs_update_root_times(trans, root);
3857 ret = btrfs_delayed_update_inode(trans, root, inode);
3859 btrfs_set_inode_last_trans(trans, inode);
3863 return btrfs_update_inode_item(trans, root, inode);
3866 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3867 struct btrfs_root *root,
3868 struct inode *inode)
3872 ret = btrfs_update_inode(trans, root, inode);
3874 return btrfs_update_inode_item(trans, root, inode);
3879 * unlink helper that gets used here in inode.c and in the tree logging
3880 * recovery code. It remove a link in a directory with a given name, and
3881 * also drops the back refs in the inode to the directory
3883 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3884 struct btrfs_root *root,
3885 struct btrfs_inode *dir,
3886 struct btrfs_inode *inode,
3887 const char *name, int name_len)
3889 struct btrfs_fs_info *fs_info = root->fs_info;
3890 struct btrfs_path *path;
3892 struct extent_buffer *leaf;
3893 struct btrfs_dir_item *di;
3894 struct btrfs_key key;
3896 u64 ino = btrfs_ino(inode);
3897 u64 dir_ino = btrfs_ino(dir);
3899 path = btrfs_alloc_path();
3905 path->leave_spinning = 1;
3906 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3907 name, name_len, -1);
3908 if (IS_ERR_OR_NULL(di)) {
3909 ret = di ? PTR_ERR(di) : -ENOENT;
3912 leaf = path->nodes[0];
3913 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3914 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3917 btrfs_release_path(path);
3920 * If we don't have dir index, we have to get it by looking up
3921 * the inode ref, since we get the inode ref, remove it directly,
3922 * it is unnecessary to do delayed deletion.
3924 * But if we have dir index, needn't search inode ref to get it.
3925 * Since the inode ref is close to the inode item, it is better
3926 * that we delay to delete it, and just do this deletion when
3927 * we update the inode item.
3929 if (inode->dir_index) {
3930 ret = btrfs_delayed_delete_inode_ref(inode);
3932 index = inode->dir_index;
3937 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3941 "failed to delete reference to %.*s, inode %llu parent %llu",
3942 name_len, name, ino, dir_ino);
3943 btrfs_abort_transaction(trans, ret);
3947 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3949 btrfs_abort_transaction(trans, ret);
3953 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3955 if (ret != 0 && ret != -ENOENT) {
3956 btrfs_abort_transaction(trans, ret);
3960 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3965 btrfs_abort_transaction(trans, ret);
3967 btrfs_free_path(path);
3971 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3972 inode_inc_iversion(&inode->vfs_inode);
3973 inode_inc_iversion(&dir->vfs_inode);
3974 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3975 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3976 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3981 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3982 struct btrfs_root *root,
3983 struct btrfs_inode *dir, struct btrfs_inode *inode,
3984 const char *name, int name_len)
3987 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3989 drop_nlink(&inode->vfs_inode);
3990 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3996 * helper to start transaction for unlink and rmdir.
3998 * unlink and rmdir are special in btrfs, they do not always free space, so
3999 * if we cannot make our reservations the normal way try and see if there is
4000 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4001 * allow the unlink to occur.
4003 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4005 struct btrfs_root *root = BTRFS_I(dir)->root;
4008 * 1 for the possible orphan item
4009 * 1 for the dir item
4010 * 1 for the dir index
4011 * 1 for the inode ref
4014 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4017 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4019 struct btrfs_root *root = BTRFS_I(dir)->root;
4020 struct btrfs_trans_handle *trans;
4021 struct inode *inode = d_inode(dentry);
4024 trans = __unlink_start_trans(dir);
4026 return PTR_ERR(trans);
4028 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4031 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4032 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4033 dentry->d_name.len);
4037 if (inode->i_nlink == 0) {
4038 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4044 btrfs_end_transaction(trans);
4045 btrfs_btree_balance_dirty(root->fs_info);
4049 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4050 struct inode *dir, u64 objectid,
4051 const char *name, int name_len)
4053 struct btrfs_root *root = BTRFS_I(dir)->root;
4054 struct btrfs_path *path;
4055 struct extent_buffer *leaf;
4056 struct btrfs_dir_item *di;
4057 struct btrfs_key key;
4060 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4062 path = btrfs_alloc_path();
4066 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4067 name, name_len, -1);
4068 if (IS_ERR_OR_NULL(di)) {
4069 ret = di ? PTR_ERR(di) : -ENOENT;
4073 leaf = path->nodes[0];
4074 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4075 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4076 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4078 btrfs_abort_transaction(trans, ret);
4081 btrfs_release_path(path);
4083 ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4084 dir_ino, &index, name, name_len);
4086 if (ret != -ENOENT) {
4087 btrfs_abort_transaction(trans, ret);
4090 di = btrfs_search_dir_index_item(root, path, dir_ino,
4092 if (IS_ERR_OR_NULL(di)) {
4097 btrfs_abort_transaction(trans, ret);
4101 leaf = path->nodes[0];
4102 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4105 btrfs_release_path(path);
4107 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4109 btrfs_abort_transaction(trans, ret);
4113 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4114 inode_inc_iversion(dir);
4115 dir->i_mtime = dir->i_ctime = current_time(dir);
4116 ret = btrfs_update_inode_fallback(trans, root, dir);
4118 btrfs_abort_transaction(trans, ret);
4120 btrfs_free_path(path);
4125 * Helper to check if the subvolume references other subvolumes or if it's
4128 static noinline int may_destroy_subvol(struct btrfs_root *root)
4130 struct btrfs_fs_info *fs_info = root->fs_info;
4131 struct btrfs_path *path;
4132 struct btrfs_dir_item *di;
4133 struct btrfs_key key;
4137 path = btrfs_alloc_path();
4141 /* Make sure this root isn't set as the default subvol */
4142 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4143 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4144 dir_id, "default", 7, 0);
4145 if (di && !IS_ERR(di)) {
4146 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4147 if (key.objectid == root->root_key.objectid) {
4150 "deleting default subvolume %llu is not allowed",
4154 btrfs_release_path(path);
4157 key.objectid = root->root_key.objectid;
4158 key.type = BTRFS_ROOT_REF_KEY;
4159 key.offset = (u64)-1;
4161 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4167 if (path->slots[0] > 0) {
4169 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4170 if (key.objectid == root->root_key.objectid &&
4171 key.type == BTRFS_ROOT_REF_KEY)
4175 btrfs_free_path(path);
4179 /* Delete all dentries for inodes belonging to the root */
4180 static void btrfs_prune_dentries(struct btrfs_root *root)
4182 struct btrfs_fs_info *fs_info = root->fs_info;
4183 struct rb_node *node;
4184 struct rb_node *prev;
4185 struct btrfs_inode *entry;
4186 struct inode *inode;
4189 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4190 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4192 spin_lock(&root->inode_lock);
4194 node = root->inode_tree.rb_node;
4198 entry = rb_entry(node, struct btrfs_inode, rb_node);
4200 if (objectid < btrfs_ino(entry))
4201 node = node->rb_left;
4202 else if (objectid > btrfs_ino(entry))
4203 node = node->rb_right;
4209 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4210 if (objectid <= btrfs_ino(entry)) {
4214 prev = rb_next(prev);
4218 entry = rb_entry(node, struct btrfs_inode, rb_node);
4219 objectid = btrfs_ino(entry) + 1;
4220 inode = igrab(&entry->vfs_inode);
4222 spin_unlock(&root->inode_lock);
4223 if (atomic_read(&inode->i_count) > 1)
4224 d_prune_aliases(inode);
4226 * btrfs_drop_inode will have it removed from the inode
4227 * cache when its usage count hits zero.
4231 spin_lock(&root->inode_lock);
4235 if (cond_resched_lock(&root->inode_lock))
4238 node = rb_next(node);
4240 spin_unlock(&root->inode_lock);
4243 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4245 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4246 struct btrfs_root *root = BTRFS_I(dir)->root;
4247 struct inode *inode = d_inode(dentry);
4248 struct btrfs_root *dest = BTRFS_I(inode)->root;
4249 struct btrfs_trans_handle *trans;
4250 struct btrfs_block_rsv block_rsv;
4256 * Don't allow to delete a subvolume with send in progress. This is
4257 * inside the inode lock so the error handling that has to drop the bit
4258 * again is not run concurrently.
4260 spin_lock(&dest->root_item_lock);
4261 if (dest->send_in_progress) {
4262 spin_unlock(&dest->root_item_lock);
4264 "attempt to delete subvolume %llu during send",
4265 dest->root_key.objectid);
4268 root_flags = btrfs_root_flags(&dest->root_item);
4269 btrfs_set_root_flags(&dest->root_item,
4270 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4271 spin_unlock(&dest->root_item_lock);
4273 down_write(&fs_info->subvol_sem);
4275 err = may_destroy_subvol(dest);
4279 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4281 * One for dir inode,
4282 * two for dir entries,
4283 * two for root ref/backref.
4285 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4289 trans = btrfs_start_transaction(root, 0);
4290 if (IS_ERR(trans)) {
4291 err = PTR_ERR(trans);
4294 trans->block_rsv = &block_rsv;
4295 trans->bytes_reserved = block_rsv.size;
4297 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4299 ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4300 dentry->d_name.name, dentry->d_name.len);
4303 btrfs_abort_transaction(trans, ret);
4307 btrfs_record_root_in_trans(trans, dest);
4309 memset(&dest->root_item.drop_progress, 0,
4310 sizeof(dest->root_item.drop_progress));
4311 dest->root_item.drop_level = 0;
4312 btrfs_set_root_refs(&dest->root_item, 0);
4314 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4315 ret = btrfs_insert_orphan_item(trans,
4317 dest->root_key.objectid);
4319 btrfs_abort_transaction(trans, ret);
4325 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4326 BTRFS_UUID_KEY_SUBVOL,
4327 dest->root_key.objectid);
4328 if (ret && ret != -ENOENT) {
4329 btrfs_abort_transaction(trans, ret);
4333 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4334 ret = btrfs_uuid_tree_remove(trans,
4335 dest->root_item.received_uuid,
4336 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4337 dest->root_key.objectid);
4338 if (ret && ret != -ENOENT) {
4339 btrfs_abort_transaction(trans, ret);
4346 trans->block_rsv = NULL;
4347 trans->bytes_reserved = 0;
4348 ret = btrfs_end_transaction(trans);
4351 inode->i_flags |= S_DEAD;
4353 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4355 up_write(&fs_info->subvol_sem);
4357 spin_lock(&dest->root_item_lock);
4358 root_flags = btrfs_root_flags(&dest->root_item);
4359 btrfs_set_root_flags(&dest->root_item,
4360 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4361 spin_unlock(&dest->root_item_lock);
4363 d_invalidate(dentry);
4364 btrfs_prune_dentries(dest);
4365 ASSERT(dest->send_in_progress == 0);
4368 if (dest->ino_cache_inode) {
4369 iput(dest->ino_cache_inode);
4370 dest->ino_cache_inode = NULL;
4377 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4379 struct inode *inode = d_inode(dentry);
4381 struct btrfs_root *root = BTRFS_I(dir)->root;
4382 struct btrfs_trans_handle *trans;
4383 u64 last_unlink_trans;
4385 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4387 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4388 return btrfs_delete_subvolume(dir, dentry);
4390 trans = __unlink_start_trans(dir);
4392 return PTR_ERR(trans);
4394 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4395 err = btrfs_unlink_subvol(trans, dir,
4396 BTRFS_I(inode)->location.objectid,
4397 dentry->d_name.name,
4398 dentry->d_name.len);
4402 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4406 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4408 /* now the directory is empty */
4409 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4410 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4411 dentry->d_name.len);
4413 btrfs_i_size_write(BTRFS_I(inode), 0);
4415 * Propagate the last_unlink_trans value of the deleted dir to
4416 * its parent directory. This is to prevent an unrecoverable
4417 * log tree in the case we do something like this:
4419 * 2) create snapshot under dir foo
4420 * 3) delete the snapshot
4423 * 6) fsync foo or some file inside foo
4425 if (last_unlink_trans >= trans->transid)
4426 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4429 btrfs_end_transaction(trans);
4430 btrfs_btree_balance_dirty(root->fs_info);
4435 static int truncate_space_check(struct btrfs_trans_handle *trans,
4436 struct btrfs_root *root,
4439 struct btrfs_fs_info *fs_info = root->fs_info;
4443 * This is only used to apply pressure to the enospc system, we don't
4444 * intend to use this reservation at all.
4446 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4447 bytes_deleted *= fs_info->nodesize;
4448 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4449 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4451 trace_btrfs_space_reservation(fs_info, "transaction",
4454 trans->bytes_reserved += bytes_deleted;
4461 * Return this if we need to call truncate_block for the last bit of the
4464 #define NEED_TRUNCATE_BLOCK 1
4467 * this can truncate away extent items, csum items and directory items.
4468 * It starts at a high offset and removes keys until it can't find
4469 * any higher than new_size
4471 * csum items that cross the new i_size are truncated to the new size
4474 * min_type is the minimum key type to truncate down to. If set to 0, this
4475 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4477 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4478 struct btrfs_root *root,
4479 struct inode *inode,
4480 u64 new_size, u32 min_type)
4482 struct btrfs_fs_info *fs_info = root->fs_info;
4483 struct btrfs_path *path;
4484 struct extent_buffer *leaf;
4485 struct btrfs_file_extent_item *fi;
4486 struct btrfs_key key;
4487 struct btrfs_key found_key;
4488 u64 extent_start = 0;
4489 u64 extent_num_bytes = 0;
4490 u64 extent_offset = 0;
4492 u64 last_size = new_size;
4493 u32 found_type = (u8)-1;
4496 int pending_del_nr = 0;
4497 int pending_del_slot = 0;
4498 int extent_type = -1;
4500 u64 ino = btrfs_ino(BTRFS_I(inode));
4501 u64 bytes_deleted = 0;
4502 bool be_nice = false;
4503 bool should_throttle = false;
4504 bool should_end = false;
4506 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4509 * for non-free space inodes and ref cows, we want to back off from
4512 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4513 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4516 path = btrfs_alloc_path();
4519 path->reada = READA_BACK;
4522 * We want to drop from the next block forward in case this new size is
4523 * not block aligned since we will be keeping the last block of the
4524 * extent just the way it is.
4526 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4527 root == fs_info->tree_root)
4528 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4529 fs_info->sectorsize),
4533 * This function is also used to drop the items in the log tree before
4534 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4535 * it is used to drop the loged items. So we shouldn't kill the delayed
4538 if (min_type == 0 && root == BTRFS_I(inode)->root)
4539 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4542 key.offset = (u64)-1;
4547 * with a 16K leaf size and 128MB extents, you can actually queue
4548 * up a huge file in a single leaf. Most of the time that
4549 * bytes_deleted is > 0, it will be huge by the time we get here
4551 if (be_nice && bytes_deleted > SZ_32M &&
4552 btrfs_should_end_transaction(trans)) {
4557 path->leave_spinning = 1;
4558 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4564 /* there are no items in the tree for us to truncate, we're
4567 if (path->slots[0] == 0)
4574 leaf = path->nodes[0];
4575 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4576 found_type = found_key.type;
4578 if (found_key.objectid != ino)
4581 if (found_type < min_type)
4584 item_end = found_key.offset;
4585 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4586 fi = btrfs_item_ptr(leaf, path->slots[0],
4587 struct btrfs_file_extent_item);
4588 extent_type = btrfs_file_extent_type(leaf, fi);
4589 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4591 btrfs_file_extent_num_bytes(leaf, fi);
4593 trace_btrfs_truncate_show_fi_regular(
4594 BTRFS_I(inode), leaf, fi,
4596 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4597 item_end += btrfs_file_extent_ram_bytes(leaf,
4600 trace_btrfs_truncate_show_fi_inline(
4601 BTRFS_I(inode), leaf, fi, path->slots[0],
4606 if (found_type > min_type) {
4609 if (item_end < new_size)
4611 if (found_key.offset >= new_size)
4617 /* FIXME, shrink the extent if the ref count is only 1 */
4618 if (found_type != BTRFS_EXTENT_DATA_KEY)
4621 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4623 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4625 u64 orig_num_bytes =
4626 btrfs_file_extent_num_bytes(leaf, fi);
4627 extent_num_bytes = ALIGN(new_size -
4629 fs_info->sectorsize);
4630 btrfs_set_file_extent_num_bytes(leaf, fi,
4632 num_dec = (orig_num_bytes -
4634 if (test_bit(BTRFS_ROOT_REF_COWS,
4637 inode_sub_bytes(inode, num_dec);
4638 btrfs_mark_buffer_dirty(leaf);
4641 btrfs_file_extent_disk_num_bytes(leaf,
4643 extent_offset = found_key.offset -
4644 btrfs_file_extent_offset(leaf, fi);
4646 /* FIXME blocksize != 4096 */
4647 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4648 if (extent_start != 0) {
4650 if (test_bit(BTRFS_ROOT_REF_COWS,
4652 inode_sub_bytes(inode, num_dec);
4655 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4657 * we can't truncate inline items that have had
4661 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4662 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4663 btrfs_file_extent_compression(leaf, fi) == 0) {
4664 u32 size = (u32)(new_size - found_key.offset);
4666 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4667 size = btrfs_file_extent_calc_inline_size(size);
4668 btrfs_truncate_item(root->fs_info, path, size, 1);
4669 } else if (!del_item) {
4671 * We have to bail so the last_size is set to
4672 * just before this extent.
4674 ret = NEED_TRUNCATE_BLOCK;
4678 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4679 inode_sub_bytes(inode, item_end + 1 - new_size);
4683 last_size = found_key.offset;
4685 last_size = new_size;
4687 if (!pending_del_nr) {
4688 /* no pending yet, add ourselves */
4689 pending_del_slot = path->slots[0];
4691 } else if (pending_del_nr &&
4692 path->slots[0] + 1 == pending_del_slot) {
4693 /* hop on the pending chunk */
4695 pending_del_slot = path->slots[0];
4702 should_throttle = false;
4705 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4706 root == fs_info->tree_root)) {
4707 btrfs_set_path_blocking(path);
4708 bytes_deleted += extent_num_bytes;
4709 ret = btrfs_free_extent(trans, root, extent_start,
4710 extent_num_bytes, 0,
4711 btrfs_header_owner(leaf),
4712 ino, extent_offset);
4714 btrfs_abort_transaction(trans, ret);
4717 if (btrfs_should_throttle_delayed_refs(trans))
4718 btrfs_async_run_delayed_refs(fs_info,
4719 trans->delayed_ref_updates * 2,
4722 if (truncate_space_check(trans, root,
4723 extent_num_bytes)) {
4726 if (btrfs_should_throttle_delayed_refs(trans))
4727 should_throttle = true;
4731 if (found_type == BTRFS_INODE_ITEM_KEY)
4734 if (path->slots[0] == 0 ||
4735 path->slots[0] != pending_del_slot ||
4736 should_throttle || should_end) {
4737 if (pending_del_nr) {
4738 ret = btrfs_del_items(trans, root, path,
4742 btrfs_abort_transaction(trans, ret);
4747 btrfs_release_path(path);
4748 if (should_throttle) {
4749 unsigned long updates = trans->delayed_ref_updates;
4751 trans->delayed_ref_updates = 0;
4752 ret = btrfs_run_delayed_refs(trans,
4759 * if we failed to refill our space rsv, bail out
4760 * and let the transaction restart
4772 if (ret >= 0 && pending_del_nr) {
4775 err = btrfs_del_items(trans, root, path, pending_del_slot,
4778 btrfs_abort_transaction(trans, err);
4782 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4783 ASSERT(last_size >= new_size);
4784 if (!ret && last_size > new_size)
4785 last_size = new_size;
4786 btrfs_ordered_update_i_size(inode, last_size, NULL);
4789 btrfs_free_path(path);
4791 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4792 unsigned long updates = trans->delayed_ref_updates;
4796 trans->delayed_ref_updates = 0;
4797 err = btrfs_run_delayed_refs(trans, updates * 2);
4806 * btrfs_truncate_block - read, zero a chunk and write a block
4807 * @inode - inode that we're zeroing
4808 * @from - the offset to start zeroing
4809 * @len - the length to zero, 0 to zero the entire range respective to the
4811 * @front - zero up to the offset instead of from the offset on
4813 * This will find the block for the "from" offset and cow the block and zero the
4814 * part we want to zero. This is used with truncate and hole punching.
4816 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4819 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4820 struct address_space *mapping = inode->i_mapping;
4821 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4822 struct btrfs_ordered_extent *ordered;
4823 struct extent_state *cached_state = NULL;
4824 struct extent_changeset *data_reserved = NULL;
4826 u32 blocksize = fs_info->sectorsize;
4827 pgoff_t index = from >> PAGE_SHIFT;
4828 unsigned offset = from & (blocksize - 1);
4830 gfp_t mask = btrfs_alloc_write_mask(mapping);
4835 if (IS_ALIGNED(offset, blocksize) &&
4836 (!len || IS_ALIGNED(len, blocksize)))
4839 block_start = round_down(from, blocksize);
4840 block_end = block_start + blocksize - 1;
4842 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4843 block_start, blocksize);
4848 page = find_or_create_page(mapping, index, mask);
4850 btrfs_delalloc_release_space(inode, data_reserved,
4851 block_start, blocksize, true);
4852 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4857 if (!PageUptodate(page)) {
4858 ret = btrfs_readpage(NULL, page);
4860 if (page->mapping != mapping) {
4865 if (!PageUptodate(page)) {
4870 wait_on_page_writeback(page);
4872 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4873 set_page_extent_mapped(page);
4875 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4877 unlock_extent_cached(io_tree, block_start, block_end,
4881 btrfs_start_ordered_extent(inode, ordered, 1);
4882 btrfs_put_ordered_extent(ordered);
4886 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4887 EXTENT_DIRTY | EXTENT_DELALLOC |
4888 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4889 0, 0, &cached_state);
4891 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4894 unlock_extent_cached(io_tree, block_start, block_end,
4899 if (offset != blocksize) {
4901 len = blocksize - offset;
4904 memset(kaddr + (block_start - page_offset(page)),
4907 memset(kaddr + (block_start - page_offset(page)) + offset,
4909 flush_dcache_page(page);
4912 ClearPageChecked(page);
4913 set_page_dirty(page);
4914 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4918 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4920 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4924 extent_changeset_free(data_reserved);
4928 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4929 u64 offset, u64 len)
4931 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4932 struct btrfs_trans_handle *trans;
4936 * Still need to make sure the inode looks like it's been updated so
4937 * that any holes get logged if we fsync.
4939 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4940 BTRFS_I(inode)->last_trans = fs_info->generation;
4941 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4942 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4947 * 1 - for the one we're dropping
4948 * 1 - for the one we're adding
4949 * 1 - for updating the inode.
4951 trans = btrfs_start_transaction(root, 3);
4953 return PTR_ERR(trans);
4955 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4957 btrfs_abort_transaction(trans, ret);
4958 btrfs_end_transaction(trans);
4962 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4963 offset, 0, 0, len, 0, len, 0, 0, 0);
4965 btrfs_abort_transaction(trans, ret);
4967 btrfs_update_inode(trans, root, inode);
4968 btrfs_end_transaction(trans);
4973 * This function puts in dummy file extents for the area we're creating a hole
4974 * for. So if we are truncating this file to a larger size we need to insert
4975 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4976 * the range between oldsize and size
4978 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4980 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4981 struct btrfs_root *root = BTRFS_I(inode)->root;
4982 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4983 struct extent_map *em = NULL;
4984 struct extent_state *cached_state = NULL;
4985 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4986 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4987 u64 block_end = ALIGN(size, fs_info->sectorsize);
4994 * If our size started in the middle of a block we need to zero out the
4995 * rest of the block before we expand the i_size, otherwise we could
4996 * expose stale data.
4998 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5002 if (size <= hole_start)
5006 struct btrfs_ordered_extent *ordered;
5008 lock_extent_bits(io_tree, hole_start, block_end - 1,
5010 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5011 block_end - hole_start);
5014 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5016 btrfs_start_ordered_extent(inode, ordered, 1);
5017 btrfs_put_ordered_extent(ordered);
5020 cur_offset = hole_start;
5022 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5023 block_end - cur_offset, 0);
5029 last_byte = min(extent_map_end(em), block_end);
5030 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5031 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5032 struct extent_map *hole_em;
5033 hole_size = last_byte - cur_offset;
5035 err = maybe_insert_hole(root, inode, cur_offset,
5039 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5040 cur_offset + hole_size - 1, 0);
5041 hole_em = alloc_extent_map();
5043 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5044 &BTRFS_I(inode)->runtime_flags);
5047 hole_em->start = cur_offset;
5048 hole_em->len = hole_size;
5049 hole_em->orig_start = cur_offset;
5051 hole_em->block_start = EXTENT_MAP_HOLE;
5052 hole_em->block_len = 0;
5053 hole_em->orig_block_len = 0;
5054 hole_em->ram_bytes = hole_size;
5055 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5056 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5057 hole_em->generation = fs_info->generation;
5060 write_lock(&em_tree->lock);
5061 err = add_extent_mapping(em_tree, hole_em, 1);
5062 write_unlock(&em_tree->lock);
5065 btrfs_drop_extent_cache(BTRFS_I(inode),
5070 free_extent_map(hole_em);
5073 free_extent_map(em);
5075 cur_offset = last_byte;
5076 if (cur_offset >= block_end)
5079 free_extent_map(em);
5080 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5084 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5086 struct btrfs_root *root = BTRFS_I(inode)->root;
5087 struct btrfs_trans_handle *trans;
5088 loff_t oldsize = i_size_read(inode);
5089 loff_t newsize = attr->ia_size;
5090 int mask = attr->ia_valid;
5094 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5095 * special case where we need to update the times despite not having
5096 * these flags set. For all other operations the VFS set these flags
5097 * explicitly if it wants a timestamp update.
5099 if (newsize != oldsize) {
5100 inode_inc_iversion(inode);
5101 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5102 inode->i_ctime = inode->i_mtime =
5103 current_time(inode);
5106 if (newsize > oldsize) {
5108 * Don't do an expanding truncate while snapshotting is ongoing.
5109 * This is to ensure the snapshot captures a fully consistent
5110 * state of this file - if the snapshot captures this expanding
5111 * truncation, it must capture all writes that happened before
5114 btrfs_wait_for_snapshot_creation(root);
5115 ret = btrfs_cont_expand(inode, oldsize, newsize);
5117 btrfs_end_write_no_snapshotting(root);
5121 trans = btrfs_start_transaction(root, 1);
5122 if (IS_ERR(trans)) {
5123 btrfs_end_write_no_snapshotting(root);
5124 return PTR_ERR(trans);
5127 i_size_write(inode, newsize);
5128 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5129 pagecache_isize_extended(inode, oldsize, newsize);
5130 ret = btrfs_update_inode(trans, root, inode);
5131 btrfs_end_write_no_snapshotting(root);
5132 btrfs_end_transaction(trans);
5136 * We're truncating a file that used to have good data down to
5137 * zero. Make sure it gets into the ordered flush list so that
5138 * any new writes get down to disk quickly.
5141 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5142 &BTRFS_I(inode)->runtime_flags);
5144 truncate_setsize(inode, newsize);
5146 /* Disable nonlocked read DIO to avoid the end less truncate */
5147 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5148 inode_dio_wait(inode);
5149 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5151 ret = btrfs_truncate(inode, newsize == oldsize);
5152 if (ret && inode->i_nlink) {
5156 * Truncate failed, so fix up the in-memory size. We
5157 * adjusted disk_i_size down as we removed extents, so
5158 * wait for disk_i_size to be stable and then update the
5159 * in-memory size to match.
5161 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5164 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5171 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5173 struct inode *inode = d_inode(dentry);
5174 struct btrfs_root *root = BTRFS_I(inode)->root;
5177 if (btrfs_root_readonly(root))
5180 err = setattr_prepare(dentry, attr);
5184 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5185 err = btrfs_setsize(inode, attr);
5190 if (attr->ia_valid) {
5191 setattr_copy(inode, attr);
5192 inode_inc_iversion(inode);
5193 err = btrfs_dirty_inode(inode);
5195 if (!err && attr->ia_valid & ATTR_MODE)
5196 err = posix_acl_chmod(inode, inode->i_mode);
5203 * While truncating the inode pages during eviction, we get the VFS calling
5204 * btrfs_invalidatepage() against each page of the inode. This is slow because
5205 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5206 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5207 * extent_state structures over and over, wasting lots of time.
5209 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5210 * those expensive operations on a per page basis and do only the ordered io
5211 * finishing, while we release here the extent_map and extent_state structures,
5212 * without the excessive merging and splitting.
5214 static void evict_inode_truncate_pages(struct inode *inode)
5216 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5217 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5218 struct rb_node *node;
5220 ASSERT(inode->i_state & I_FREEING);
5221 truncate_inode_pages_final(&inode->i_data);
5223 write_lock(&map_tree->lock);
5224 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5225 struct extent_map *em;
5227 node = rb_first_cached(&map_tree->map);
5228 em = rb_entry(node, struct extent_map, rb_node);
5229 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5230 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5231 remove_extent_mapping(map_tree, em);
5232 free_extent_map(em);
5233 if (need_resched()) {
5234 write_unlock(&map_tree->lock);
5236 write_lock(&map_tree->lock);
5239 write_unlock(&map_tree->lock);
5242 * Keep looping until we have no more ranges in the io tree.
5243 * We can have ongoing bios started by readpages (called from readahead)
5244 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5245 * still in progress (unlocked the pages in the bio but did not yet
5246 * unlocked the ranges in the io tree). Therefore this means some
5247 * ranges can still be locked and eviction started because before
5248 * submitting those bios, which are executed by a separate task (work
5249 * queue kthread), inode references (inode->i_count) were not taken
5250 * (which would be dropped in the end io callback of each bio).
5251 * Therefore here we effectively end up waiting for those bios and
5252 * anyone else holding locked ranges without having bumped the inode's
5253 * reference count - if we don't do it, when they access the inode's
5254 * io_tree to unlock a range it may be too late, leading to an
5255 * use-after-free issue.
5257 spin_lock(&io_tree->lock);
5258 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5259 struct extent_state *state;
5260 struct extent_state *cached_state = NULL;
5263 unsigned state_flags;
5265 node = rb_first(&io_tree->state);
5266 state = rb_entry(node, struct extent_state, rb_node);
5267 start = state->start;
5269 state_flags = state->state;
5270 spin_unlock(&io_tree->lock);
5272 lock_extent_bits(io_tree, start, end, &cached_state);
5275 * If still has DELALLOC flag, the extent didn't reach disk,
5276 * and its reserved space won't be freed by delayed_ref.
5277 * So we need to free its reserved space here.
5278 * (Refer to comment in btrfs_invalidatepage, case 2)
5280 * Note, end is the bytenr of last byte, so we need + 1 here.
5282 if (state_flags & EXTENT_DELALLOC)
5283 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5285 clear_extent_bit(io_tree, start, end,
5286 EXTENT_LOCKED | EXTENT_DIRTY |
5287 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5288 EXTENT_DEFRAG, 1, 1, &cached_state);
5291 spin_lock(&io_tree->lock);
5293 spin_unlock(&io_tree->lock);
5296 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5297 struct btrfs_block_rsv *rsv)
5299 struct btrfs_fs_info *fs_info = root->fs_info;
5300 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5304 struct btrfs_trans_handle *trans;
5307 ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5308 BTRFS_RESERVE_FLUSH_LIMIT);
5310 if (ret && ++failures > 2) {
5312 "could not allocate space for a delete; will truncate on mount");
5313 return ERR_PTR(-ENOSPC);
5316 trans = btrfs_join_transaction(root);
5317 if (IS_ERR(trans) || !ret)
5321 * Try to steal from the global reserve if there is space for
5324 if (!btrfs_check_space_for_delayed_refs(trans) &&
5325 !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5328 /* If not, commit and try again. */
5329 ret = btrfs_commit_transaction(trans);
5331 return ERR_PTR(ret);
5335 void btrfs_evict_inode(struct inode *inode)
5337 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5338 struct btrfs_trans_handle *trans;
5339 struct btrfs_root *root = BTRFS_I(inode)->root;
5340 struct btrfs_block_rsv *rsv;
5343 trace_btrfs_inode_evict(inode);
5350 evict_inode_truncate_pages(inode);
5352 if (inode->i_nlink &&
5353 ((btrfs_root_refs(&root->root_item) != 0 &&
5354 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5355 btrfs_is_free_space_inode(BTRFS_I(inode))))
5358 if (is_bad_inode(inode))
5361 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5363 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5366 if (inode->i_nlink > 0) {
5367 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5368 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5372 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5376 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5379 rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5382 btrfs_i_size_write(BTRFS_I(inode), 0);
5385 trans = evict_refill_and_join(root, rsv);
5389 trans->block_rsv = rsv;
5391 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5392 trans->block_rsv = &fs_info->trans_block_rsv;
5393 btrfs_end_transaction(trans);
5394 btrfs_btree_balance_dirty(fs_info);
5395 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5402 * Errors here aren't a big deal, it just means we leave orphan items in
5403 * the tree. They will be cleaned up on the next mount. If the inode
5404 * number gets reused, cleanup deletes the orphan item without doing
5405 * anything, and unlink reuses the existing orphan item.
5407 * If it turns out that we are dropping too many of these, we might want
5408 * to add a mechanism for retrying these after a commit.
5410 trans = evict_refill_and_join(root, rsv);
5411 if (!IS_ERR(trans)) {
5412 trans->block_rsv = rsv;
5413 btrfs_orphan_del(trans, BTRFS_I(inode));
5414 trans->block_rsv = &fs_info->trans_block_rsv;
5415 btrfs_end_transaction(trans);
5418 if (!(root == fs_info->tree_root ||
5419 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5420 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5423 btrfs_free_block_rsv(fs_info, rsv);
5426 * If we didn't successfully delete, the orphan item will still be in
5427 * the tree and we'll retry on the next mount. Again, we might also want
5428 * to retry these periodically in the future.
5430 btrfs_remove_delayed_node(BTRFS_I(inode));
5435 * this returns the key found in the dir entry in the location pointer.
5436 * If no dir entries were found, returns -ENOENT.
5437 * If found a corrupted location in dir entry, returns -EUCLEAN.
5439 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5440 struct btrfs_key *location)
5442 const char *name = dentry->d_name.name;
5443 int namelen = dentry->d_name.len;
5444 struct btrfs_dir_item *di;
5445 struct btrfs_path *path;
5446 struct btrfs_root *root = BTRFS_I(dir)->root;
5449 path = btrfs_alloc_path();
5453 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5455 if (IS_ERR_OR_NULL(di)) {
5456 ret = di ? PTR_ERR(di) : -ENOENT;
5460 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5461 if (location->type != BTRFS_INODE_ITEM_KEY &&
5462 location->type != BTRFS_ROOT_ITEM_KEY) {
5464 btrfs_warn(root->fs_info,
5465 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5466 __func__, name, btrfs_ino(BTRFS_I(dir)),
5467 location->objectid, location->type, location->offset);
5470 btrfs_free_path(path);
5475 * when we hit a tree root in a directory, the btrfs part of the inode
5476 * needs to be changed to reflect the root directory of the tree root. This
5477 * is kind of like crossing a mount point.
5479 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5481 struct dentry *dentry,
5482 struct btrfs_key *location,
5483 struct btrfs_root **sub_root)
5485 struct btrfs_path *path;
5486 struct btrfs_root *new_root;
5487 struct btrfs_root_ref *ref;
5488 struct extent_buffer *leaf;
5489 struct btrfs_key key;
5493 path = btrfs_alloc_path();
5500 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5501 key.type = BTRFS_ROOT_REF_KEY;
5502 key.offset = location->objectid;
5504 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5511 leaf = path->nodes[0];
5512 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5513 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5514 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5517 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5518 (unsigned long)(ref + 1),
5519 dentry->d_name.len);
5523 btrfs_release_path(path);
5525 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5526 if (IS_ERR(new_root)) {
5527 err = PTR_ERR(new_root);
5531 *sub_root = new_root;
5532 location->objectid = btrfs_root_dirid(&new_root->root_item);
5533 location->type = BTRFS_INODE_ITEM_KEY;
5534 location->offset = 0;
5537 btrfs_free_path(path);
5541 static void inode_tree_add(struct inode *inode)
5543 struct btrfs_root *root = BTRFS_I(inode)->root;
5544 struct btrfs_inode *entry;
5546 struct rb_node *parent;
5547 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5548 u64 ino = btrfs_ino(BTRFS_I(inode));
5550 if (inode_unhashed(inode))
5553 spin_lock(&root->inode_lock);
5554 p = &root->inode_tree.rb_node;
5557 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5559 if (ino < btrfs_ino(entry))
5560 p = &parent->rb_left;
5561 else if (ino > btrfs_ino(entry))
5562 p = &parent->rb_right;
5564 WARN_ON(!(entry->vfs_inode.i_state &
5565 (I_WILL_FREE | I_FREEING)));
5566 rb_replace_node(parent, new, &root->inode_tree);
5567 RB_CLEAR_NODE(parent);
5568 spin_unlock(&root->inode_lock);
5572 rb_link_node(new, parent, p);
5573 rb_insert_color(new, &root->inode_tree);
5574 spin_unlock(&root->inode_lock);
5577 static void inode_tree_del(struct inode *inode)
5579 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5580 struct btrfs_root *root = BTRFS_I(inode)->root;
5583 spin_lock(&root->inode_lock);
5584 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5585 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5586 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5587 empty = RB_EMPTY_ROOT(&root->inode_tree);
5589 spin_unlock(&root->inode_lock);
5591 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5592 synchronize_srcu(&fs_info->subvol_srcu);
5593 spin_lock(&root->inode_lock);
5594 empty = RB_EMPTY_ROOT(&root->inode_tree);
5595 spin_unlock(&root->inode_lock);
5597 btrfs_add_dead_root(root);
5602 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5604 struct btrfs_iget_args *args = p;
5605 inode->i_ino = args->location->objectid;
5606 memcpy(&BTRFS_I(inode)->location, args->location,
5607 sizeof(*args->location));
5608 BTRFS_I(inode)->root = args->root;
5612 static int btrfs_find_actor(struct inode *inode, void *opaque)
5614 struct btrfs_iget_args *args = opaque;
5615 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5616 args->root == BTRFS_I(inode)->root;
5619 static struct inode *btrfs_iget_locked(struct super_block *s,
5620 struct btrfs_key *location,
5621 struct btrfs_root *root)
5623 struct inode *inode;
5624 struct btrfs_iget_args args;
5625 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5627 args.location = location;
5630 inode = iget5_locked(s, hashval, btrfs_find_actor,
5631 btrfs_init_locked_inode,
5636 /* Get an inode object given its location and corresponding root.
5637 * Returns in *is_new if the inode was read from disk
5639 struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location,
5640 struct btrfs_root *root, int *new,
5641 struct btrfs_path *path)
5643 struct inode *inode;
5645 inode = btrfs_iget_locked(s, location, root);
5647 return ERR_PTR(-ENOMEM);
5649 if (inode->i_state & I_NEW) {
5652 ret = btrfs_read_locked_inode(inode, path);
5654 inode_tree_add(inode);
5655 unlock_new_inode(inode);
5661 * ret > 0 can come from btrfs_search_slot called by
5662 * btrfs_read_locked_inode, this means the inode item
5667 inode = ERR_PTR(ret);
5674 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5675 struct btrfs_root *root, int *new)
5677 return btrfs_iget_path(s, location, root, new, NULL);
5680 static struct inode *new_simple_dir(struct super_block *s,
5681 struct btrfs_key *key,
5682 struct btrfs_root *root)
5684 struct inode *inode = new_inode(s);
5687 return ERR_PTR(-ENOMEM);
5689 BTRFS_I(inode)->root = root;
5690 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5691 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5693 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5694 inode->i_op = &btrfs_dir_ro_inode_operations;
5695 inode->i_opflags &= ~IOP_XATTR;
5696 inode->i_fop = &simple_dir_operations;
5697 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5698 inode->i_mtime = current_time(inode);
5699 inode->i_atime = inode->i_mtime;
5700 inode->i_ctime = inode->i_mtime;
5701 BTRFS_I(inode)->i_otime = inode->i_mtime;
5706 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5708 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5709 struct inode *inode;
5710 struct btrfs_root *root = BTRFS_I(dir)->root;
5711 struct btrfs_root *sub_root = root;
5712 struct btrfs_key location;
5716 if (dentry->d_name.len > BTRFS_NAME_LEN)
5717 return ERR_PTR(-ENAMETOOLONG);
5719 ret = btrfs_inode_by_name(dir, dentry, &location);
5721 return ERR_PTR(ret);
5723 if (location.type == BTRFS_INODE_ITEM_KEY) {
5724 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5728 index = srcu_read_lock(&fs_info->subvol_srcu);
5729 ret = fixup_tree_root_location(fs_info, dir, dentry,
5730 &location, &sub_root);
5733 inode = ERR_PTR(ret);
5735 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5737 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5739 srcu_read_unlock(&fs_info->subvol_srcu, index);
5741 if (!IS_ERR(inode) && root != sub_root) {
5742 down_read(&fs_info->cleanup_work_sem);
5743 if (!sb_rdonly(inode->i_sb))
5744 ret = btrfs_orphan_cleanup(sub_root);
5745 up_read(&fs_info->cleanup_work_sem);
5748 inode = ERR_PTR(ret);
5755 static int btrfs_dentry_delete(const struct dentry *dentry)
5757 struct btrfs_root *root;
5758 struct inode *inode = d_inode(dentry);
5760 if (!inode && !IS_ROOT(dentry))
5761 inode = d_inode(dentry->d_parent);
5764 root = BTRFS_I(inode)->root;
5765 if (btrfs_root_refs(&root->root_item) == 0)
5768 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5774 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5777 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5779 if (inode == ERR_PTR(-ENOENT))
5781 return d_splice_alias(inode, dentry);
5784 unsigned char btrfs_filetype_table[] = {
5785 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5789 * All this infrastructure exists because dir_emit can fault, and we are holding
5790 * the tree lock when doing readdir. For now just allocate a buffer and copy
5791 * our information into that, and then dir_emit from the buffer. This is
5792 * similar to what NFS does, only we don't keep the buffer around in pagecache
5793 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5794 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5797 static int btrfs_opendir(struct inode *inode, struct file *file)
5799 struct btrfs_file_private *private;
5801 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5804 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5805 if (!private->filldir_buf) {
5809 file->private_data = private;
5820 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5823 struct dir_entry *entry = addr;
5824 char *name = (char *)(entry + 1);
5826 ctx->pos = get_unaligned(&entry->offset);
5827 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5828 get_unaligned(&entry->ino),
5829 get_unaligned(&entry->type)))
5831 addr += sizeof(struct dir_entry) +
5832 get_unaligned(&entry->name_len);
5838 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5840 struct inode *inode = file_inode(file);
5841 struct btrfs_root *root = BTRFS_I(inode)->root;
5842 struct btrfs_file_private *private = file->private_data;
5843 struct btrfs_dir_item *di;
5844 struct btrfs_key key;
5845 struct btrfs_key found_key;
5846 struct btrfs_path *path;
5848 struct list_head ins_list;
5849 struct list_head del_list;
5851 struct extent_buffer *leaf;
5858 struct btrfs_key location;
5860 if (!dir_emit_dots(file, ctx))
5863 path = btrfs_alloc_path();
5867 addr = private->filldir_buf;
5868 path->reada = READA_FORWARD;
5870 INIT_LIST_HEAD(&ins_list);
5871 INIT_LIST_HEAD(&del_list);
5872 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5875 key.type = BTRFS_DIR_INDEX_KEY;
5876 key.offset = ctx->pos;
5877 key.objectid = btrfs_ino(BTRFS_I(inode));
5879 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5884 struct dir_entry *entry;
5886 leaf = path->nodes[0];
5887 slot = path->slots[0];
5888 if (slot >= btrfs_header_nritems(leaf)) {
5889 ret = btrfs_next_leaf(root, path);
5897 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5899 if (found_key.objectid != key.objectid)
5901 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5903 if (found_key.offset < ctx->pos)
5905 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5907 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5908 name_len = btrfs_dir_name_len(leaf, di);
5909 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5911 btrfs_release_path(path);
5912 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5915 addr = private->filldir_buf;
5922 put_unaligned(name_len, &entry->name_len);
5923 name_ptr = (char *)(entry + 1);
5924 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5926 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5928 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5929 put_unaligned(location.objectid, &entry->ino);
5930 put_unaligned(found_key.offset, &entry->offset);
5932 addr += sizeof(struct dir_entry) + name_len;
5933 total_len += sizeof(struct dir_entry) + name_len;
5937 btrfs_release_path(path);
5939 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5943 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5948 * Stop new entries from being returned after we return the last
5951 * New directory entries are assigned a strictly increasing
5952 * offset. This means that new entries created during readdir
5953 * are *guaranteed* to be seen in the future by that readdir.
5954 * This has broken buggy programs which operate on names as
5955 * they're returned by readdir. Until we re-use freed offsets
5956 * we have this hack to stop new entries from being returned
5957 * under the assumption that they'll never reach this huge
5960 * This is being careful not to overflow 32bit loff_t unless the
5961 * last entry requires it because doing so has broken 32bit apps
5964 if (ctx->pos >= INT_MAX)
5965 ctx->pos = LLONG_MAX;
5972 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5973 btrfs_free_path(path);
5978 * This is somewhat expensive, updating the tree every time the
5979 * inode changes. But, it is most likely to find the inode in cache.
5980 * FIXME, needs more benchmarking...there are no reasons other than performance
5981 * to keep or drop this code.
5983 static int btrfs_dirty_inode(struct inode *inode)
5985 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5986 struct btrfs_root *root = BTRFS_I(inode)->root;
5987 struct btrfs_trans_handle *trans;
5990 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5993 trans = btrfs_join_transaction(root);
5995 return PTR_ERR(trans);
5997 ret = btrfs_update_inode(trans, root, inode);
5998 if (ret && ret == -ENOSPC) {
5999 /* whoops, lets try again with the full transaction */
6000 btrfs_end_transaction(trans);
6001 trans = btrfs_start_transaction(root, 1);
6003 return PTR_ERR(trans);
6005 ret = btrfs_update_inode(trans, root, inode);
6007 btrfs_end_transaction(trans);
6008 if (BTRFS_I(inode)->delayed_node)
6009 btrfs_balance_delayed_items(fs_info);
6015 * This is a copy of file_update_time. We need this so we can return error on
6016 * ENOSPC for updating the inode in the case of file write and mmap writes.
6018 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6021 struct btrfs_root *root = BTRFS_I(inode)->root;
6022 bool dirty = flags & ~S_VERSION;
6024 if (btrfs_root_readonly(root))
6027 if (flags & S_VERSION)
6028 dirty |= inode_maybe_inc_iversion(inode, dirty);
6029 if (flags & S_CTIME)
6030 inode->i_ctime = *now;
6031 if (flags & S_MTIME)
6032 inode->i_mtime = *now;
6033 if (flags & S_ATIME)
6034 inode->i_atime = *now;
6035 return dirty ? btrfs_dirty_inode(inode) : 0;
6039 * find the highest existing sequence number in a directory
6040 * and then set the in-memory index_cnt variable to reflect
6041 * free sequence numbers
6043 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6045 struct btrfs_root *root = inode->root;
6046 struct btrfs_key key, found_key;
6047 struct btrfs_path *path;
6048 struct extent_buffer *leaf;
6051 key.objectid = btrfs_ino(inode);
6052 key.type = BTRFS_DIR_INDEX_KEY;
6053 key.offset = (u64)-1;
6055 path = btrfs_alloc_path();
6059 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6062 /* FIXME: we should be able to handle this */
6068 * MAGIC NUMBER EXPLANATION:
6069 * since we search a directory based on f_pos we have to start at 2
6070 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6071 * else has to start at 2
6073 if (path->slots[0] == 0) {
6074 inode->index_cnt = 2;
6080 leaf = path->nodes[0];
6081 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6083 if (found_key.objectid != btrfs_ino(inode) ||
6084 found_key.type != BTRFS_DIR_INDEX_KEY) {
6085 inode->index_cnt = 2;
6089 inode->index_cnt = found_key.offset + 1;
6091 btrfs_free_path(path);
6096 * helper to find a free sequence number in a given directory. This current
6097 * code is very simple, later versions will do smarter things in the btree
6099 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6103 if (dir->index_cnt == (u64)-1) {
6104 ret = btrfs_inode_delayed_dir_index_count(dir);
6106 ret = btrfs_set_inode_index_count(dir);
6112 *index = dir->index_cnt;
6118 static int btrfs_insert_inode_locked(struct inode *inode)
6120 struct btrfs_iget_args args;
6121 args.location = &BTRFS_I(inode)->location;
6122 args.root = BTRFS_I(inode)->root;
6124 return insert_inode_locked4(inode,
6125 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6126 btrfs_find_actor, &args);
6130 * Inherit flags from the parent inode.
6132 * Currently only the compression flags and the cow flags are inherited.
6134 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6141 flags = BTRFS_I(dir)->flags;
6143 if (flags & BTRFS_INODE_NOCOMPRESS) {
6144 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6145 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6146 } else if (flags & BTRFS_INODE_COMPRESS) {
6147 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6148 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6151 if (flags & BTRFS_INODE_NODATACOW) {
6152 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6153 if (S_ISREG(inode->i_mode))
6154 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6157 btrfs_sync_inode_flags_to_i_flags(inode);
6160 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6161 struct btrfs_root *root,
6163 const char *name, int name_len,
6164 u64 ref_objectid, u64 objectid,
6165 umode_t mode, u64 *index)
6167 struct btrfs_fs_info *fs_info = root->fs_info;
6168 struct inode *inode;
6169 struct btrfs_inode_item *inode_item;
6170 struct btrfs_key *location;
6171 struct btrfs_path *path;
6172 struct btrfs_inode_ref *ref;
6173 struct btrfs_key key[2];
6175 int nitems = name ? 2 : 1;
6179 path = btrfs_alloc_path();
6181 return ERR_PTR(-ENOMEM);
6183 inode = new_inode(fs_info->sb);
6185 btrfs_free_path(path);
6186 return ERR_PTR(-ENOMEM);
6190 * O_TMPFILE, set link count to 0, so that after this point,
6191 * we fill in an inode item with the correct link count.
6194 set_nlink(inode, 0);
6197 * we have to initialize this early, so we can reclaim the inode
6198 * number if we fail afterwards in this function.
6200 inode->i_ino = objectid;
6203 trace_btrfs_inode_request(dir);
6205 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6207 btrfs_free_path(path);
6209 return ERR_PTR(ret);
6215 * index_cnt is ignored for everything but a dir,
6216 * btrfs_set_inode_index_count has an explanation for the magic
6219 BTRFS_I(inode)->index_cnt = 2;
6220 BTRFS_I(inode)->dir_index = *index;
6221 BTRFS_I(inode)->root = root;
6222 BTRFS_I(inode)->generation = trans->transid;
6223 inode->i_generation = BTRFS_I(inode)->generation;
6226 * We could have gotten an inode number from somebody who was fsynced
6227 * and then removed in this same transaction, so let's just set full
6228 * sync since it will be a full sync anyway and this will blow away the
6229 * old info in the log.
6231 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6233 key[0].objectid = objectid;
6234 key[0].type = BTRFS_INODE_ITEM_KEY;
6237 sizes[0] = sizeof(struct btrfs_inode_item);
6241 * Start new inodes with an inode_ref. This is slightly more
6242 * efficient for small numbers of hard links since they will
6243 * be packed into one item. Extended refs will kick in if we
6244 * add more hard links than can fit in the ref item.
6246 key[1].objectid = objectid;
6247 key[1].type = BTRFS_INODE_REF_KEY;
6248 key[1].offset = ref_objectid;
6250 sizes[1] = name_len + sizeof(*ref);
6253 location = &BTRFS_I(inode)->location;
6254 location->objectid = objectid;
6255 location->offset = 0;
6256 location->type = BTRFS_INODE_ITEM_KEY;
6258 ret = btrfs_insert_inode_locked(inode);
6264 path->leave_spinning = 1;
6265 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6269 inode_init_owner(inode, dir, mode);
6270 inode_set_bytes(inode, 0);
6272 inode->i_mtime = current_time(inode);
6273 inode->i_atime = inode->i_mtime;
6274 inode->i_ctime = inode->i_mtime;
6275 BTRFS_I(inode)->i_otime = inode->i_mtime;
6277 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6278 struct btrfs_inode_item);
6279 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6280 sizeof(*inode_item));
6281 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6284 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6285 struct btrfs_inode_ref);
6286 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6287 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6288 ptr = (unsigned long)(ref + 1);
6289 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6292 btrfs_mark_buffer_dirty(path->nodes[0]);
6293 btrfs_free_path(path);
6295 btrfs_inherit_iflags(inode, dir);
6297 if (S_ISREG(mode)) {
6298 if (btrfs_test_opt(fs_info, NODATASUM))
6299 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6300 if (btrfs_test_opt(fs_info, NODATACOW))
6301 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6302 BTRFS_INODE_NODATASUM;
6305 inode_tree_add(inode);
6307 trace_btrfs_inode_new(inode);
6308 btrfs_set_inode_last_trans(trans, inode);
6310 btrfs_update_root_times(trans, root);
6312 ret = btrfs_inode_inherit_props(trans, inode, dir);
6315 "error inheriting props for ino %llu (root %llu): %d",
6316 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6321 discard_new_inode(inode);
6324 BTRFS_I(dir)->index_cnt--;
6325 btrfs_free_path(path);
6326 return ERR_PTR(ret);
6329 static inline u8 btrfs_inode_type(struct inode *inode)
6331 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6335 * utility function to add 'inode' into 'parent_inode' with
6336 * a give name and a given sequence number.
6337 * if 'add_backref' is true, also insert a backref from the
6338 * inode to the parent directory.
6340 int btrfs_add_link(struct btrfs_trans_handle *trans,
6341 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6342 const char *name, int name_len, int add_backref, u64 index)
6345 struct btrfs_key key;
6346 struct btrfs_root *root = parent_inode->root;
6347 u64 ino = btrfs_ino(inode);
6348 u64 parent_ino = btrfs_ino(parent_inode);
6350 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6351 memcpy(&key, &inode->root->root_key, sizeof(key));
6354 key.type = BTRFS_INODE_ITEM_KEY;
6358 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6359 ret = btrfs_add_root_ref(trans, key.objectid,
6360 root->root_key.objectid, parent_ino,
6361 index, name, name_len);
6362 } else if (add_backref) {
6363 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6367 /* Nothing to clean up yet */
6371 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6372 btrfs_inode_type(&inode->vfs_inode), index);
6373 if (ret == -EEXIST || ret == -EOVERFLOW)
6376 btrfs_abort_transaction(trans, ret);
6380 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6382 inode_inc_iversion(&parent_inode->vfs_inode);
6383 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6384 current_time(&parent_inode->vfs_inode);
6385 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6387 btrfs_abort_transaction(trans, ret);
6391 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6394 err = btrfs_del_root_ref(trans, key.objectid,
6395 root->root_key.objectid, parent_ino,
6396 &local_index, name, name_len);
6398 } else if (add_backref) {
6402 err = btrfs_del_inode_ref(trans, root, name, name_len,
6403 ino, parent_ino, &local_index);
6408 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6409 struct btrfs_inode *dir, struct dentry *dentry,
6410 struct btrfs_inode *inode, int backref, u64 index)
6412 int err = btrfs_add_link(trans, dir, inode,
6413 dentry->d_name.name, dentry->d_name.len,
6420 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6421 umode_t mode, dev_t rdev)
6423 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6424 struct btrfs_trans_handle *trans;
6425 struct btrfs_root *root = BTRFS_I(dir)->root;
6426 struct inode *inode = NULL;
6432 * 2 for inode item and ref
6434 * 1 for xattr if selinux is on
6436 trans = btrfs_start_transaction(root, 5);
6438 return PTR_ERR(trans);
6440 err = btrfs_find_free_ino(root, &objectid);
6444 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6445 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6447 if (IS_ERR(inode)) {
6448 err = PTR_ERR(inode);
6454 * If the active LSM wants to access the inode during
6455 * d_instantiate it needs these. Smack checks to see
6456 * if the filesystem supports xattrs by looking at the
6459 inode->i_op = &btrfs_special_inode_operations;
6460 init_special_inode(inode, inode->i_mode, rdev);
6462 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6466 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6471 btrfs_update_inode(trans, root, inode);
6472 d_instantiate_new(dentry, inode);
6475 btrfs_end_transaction(trans);
6476 btrfs_btree_balance_dirty(fs_info);
6478 inode_dec_link_count(inode);
6479 discard_new_inode(inode);
6484 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6485 umode_t mode, bool excl)
6487 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6488 struct btrfs_trans_handle *trans;
6489 struct btrfs_root *root = BTRFS_I(dir)->root;
6490 struct inode *inode = NULL;
6496 * 2 for inode item and ref
6498 * 1 for xattr if selinux is on
6500 trans = btrfs_start_transaction(root, 5);
6502 return PTR_ERR(trans);
6504 err = btrfs_find_free_ino(root, &objectid);
6508 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6509 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6511 if (IS_ERR(inode)) {
6512 err = PTR_ERR(inode);
6517 * If the active LSM wants to access the inode during
6518 * d_instantiate it needs these. Smack checks to see
6519 * if the filesystem supports xattrs by looking at the
6522 inode->i_fop = &btrfs_file_operations;
6523 inode->i_op = &btrfs_file_inode_operations;
6524 inode->i_mapping->a_ops = &btrfs_aops;
6526 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6530 err = btrfs_update_inode(trans, root, inode);
6534 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6539 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6540 d_instantiate_new(dentry, inode);
6543 btrfs_end_transaction(trans);
6545 inode_dec_link_count(inode);
6546 discard_new_inode(inode);
6548 btrfs_btree_balance_dirty(fs_info);
6552 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6553 struct dentry *dentry)
6555 struct btrfs_trans_handle *trans = NULL;
6556 struct btrfs_root *root = BTRFS_I(dir)->root;
6557 struct inode *inode = d_inode(old_dentry);
6558 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6563 /* do not allow sys_link's with other subvols of the same device */
6564 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6567 if (inode->i_nlink >= BTRFS_LINK_MAX)
6570 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6575 * 2 items for inode and inode ref
6576 * 2 items for dir items
6577 * 1 item for parent inode
6578 * 1 item for orphan item deletion if O_TMPFILE
6580 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6581 if (IS_ERR(trans)) {
6582 err = PTR_ERR(trans);
6587 /* There are several dir indexes for this inode, clear the cache. */
6588 BTRFS_I(inode)->dir_index = 0ULL;
6590 inode_inc_iversion(inode);
6591 inode->i_ctime = current_time(inode);
6593 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6595 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6601 struct dentry *parent = dentry->d_parent;
6604 err = btrfs_update_inode(trans, root, inode);
6607 if (inode->i_nlink == 1) {
6609 * If new hard link count is 1, it's a file created
6610 * with open(2) O_TMPFILE flag.
6612 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6616 BTRFS_I(inode)->last_link_trans = trans->transid;
6617 d_instantiate(dentry, inode);
6618 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6620 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6621 err = btrfs_commit_transaction(trans);
6628 btrfs_end_transaction(trans);
6630 inode_dec_link_count(inode);
6633 btrfs_btree_balance_dirty(fs_info);
6637 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6639 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6640 struct inode *inode = NULL;
6641 struct btrfs_trans_handle *trans;
6642 struct btrfs_root *root = BTRFS_I(dir)->root;
6648 * 2 items for inode and ref
6649 * 2 items for dir items
6650 * 1 for xattr if selinux is on
6652 trans = btrfs_start_transaction(root, 5);
6654 return PTR_ERR(trans);
6656 err = btrfs_find_free_ino(root, &objectid);
6660 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6661 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6662 S_IFDIR | mode, &index);
6663 if (IS_ERR(inode)) {
6664 err = PTR_ERR(inode);
6669 /* these must be set before we unlock the inode */
6670 inode->i_op = &btrfs_dir_inode_operations;
6671 inode->i_fop = &btrfs_dir_file_operations;
6673 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6677 btrfs_i_size_write(BTRFS_I(inode), 0);
6678 err = btrfs_update_inode(trans, root, inode);
6682 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6683 dentry->d_name.name,
6684 dentry->d_name.len, 0, index);
6688 d_instantiate_new(dentry, inode);
6691 btrfs_end_transaction(trans);
6693 inode_dec_link_count(inode);
6694 discard_new_inode(inode);
6696 btrfs_btree_balance_dirty(fs_info);
6700 static noinline int uncompress_inline(struct btrfs_path *path,
6702 size_t pg_offset, u64 extent_offset,
6703 struct btrfs_file_extent_item *item)
6706 struct extent_buffer *leaf = path->nodes[0];
6709 unsigned long inline_size;
6713 WARN_ON(pg_offset != 0);
6714 compress_type = btrfs_file_extent_compression(leaf, item);
6715 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6716 inline_size = btrfs_file_extent_inline_item_len(leaf,
6717 btrfs_item_nr(path->slots[0]));
6718 tmp = kmalloc(inline_size, GFP_NOFS);
6721 ptr = btrfs_file_extent_inline_start(item);
6723 read_extent_buffer(leaf, tmp, ptr, inline_size);
6725 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6726 ret = btrfs_decompress(compress_type, tmp, page,
6727 extent_offset, inline_size, max_size);
6730 * decompression code contains a memset to fill in any space between the end
6731 * of the uncompressed data and the end of max_size in case the decompressed
6732 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6733 * the end of an inline extent and the beginning of the next block, so we
6734 * cover that region here.
6737 if (max_size + pg_offset < PAGE_SIZE) {
6738 char *map = kmap(page);
6739 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6747 * a bit scary, this does extent mapping from logical file offset to the disk.
6748 * the ugly parts come from merging extents from the disk with the in-ram
6749 * representation. This gets more complex because of the data=ordered code,
6750 * where the in-ram extents might be locked pending data=ordered completion.
6752 * This also copies inline extents directly into the page.
6754 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6756 size_t pg_offset, u64 start, u64 len,
6759 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6762 u64 extent_start = 0;
6764 u64 objectid = btrfs_ino(inode);
6766 struct btrfs_path *path = NULL;
6767 struct btrfs_root *root = inode->root;
6768 struct btrfs_file_extent_item *item;
6769 struct extent_buffer *leaf;
6770 struct btrfs_key found_key;
6771 struct extent_map *em = NULL;
6772 struct extent_map_tree *em_tree = &inode->extent_tree;
6773 struct extent_io_tree *io_tree = &inode->io_tree;
6774 const bool new_inline = !page || create;
6776 read_lock(&em_tree->lock);
6777 em = lookup_extent_mapping(em_tree, start, len);
6779 em->bdev = fs_info->fs_devices->latest_bdev;
6780 read_unlock(&em_tree->lock);
6783 if (em->start > start || em->start + em->len <= start)
6784 free_extent_map(em);
6785 else if (em->block_start == EXTENT_MAP_INLINE && page)
6786 free_extent_map(em);
6790 em = alloc_extent_map();
6795 em->bdev = fs_info->fs_devices->latest_bdev;
6796 em->start = EXTENT_MAP_HOLE;
6797 em->orig_start = EXTENT_MAP_HOLE;
6799 em->block_len = (u64)-1;
6801 path = btrfs_alloc_path();
6807 /* Chances are we'll be called again, so go ahead and do readahead */
6808 path->reada = READA_FORWARD;
6811 * Unless we're going to uncompress the inline extent, no sleep would
6814 path->leave_spinning = 1;
6816 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6823 if (path->slots[0] == 0)
6828 leaf = path->nodes[0];
6829 item = btrfs_item_ptr(leaf, path->slots[0],
6830 struct btrfs_file_extent_item);
6831 /* are we inside the extent that was found? */
6832 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6833 found_type = found_key.type;
6834 if (found_key.objectid != objectid ||
6835 found_type != BTRFS_EXTENT_DATA_KEY) {
6837 * If we backup past the first extent we want to move forward
6838 * and see if there is an extent in front of us, otherwise we'll
6839 * say there is a hole for our whole search range which can
6846 found_type = btrfs_file_extent_type(leaf, item);
6847 extent_start = found_key.offset;
6848 if (found_type == BTRFS_FILE_EXTENT_REG ||
6849 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6850 extent_end = extent_start +
6851 btrfs_file_extent_num_bytes(leaf, item);
6853 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6855 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6858 size = btrfs_file_extent_ram_bytes(leaf, item);
6859 extent_end = ALIGN(extent_start + size,
6860 fs_info->sectorsize);
6862 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6867 if (start >= extent_end) {
6869 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6870 ret = btrfs_next_leaf(root, path);
6877 leaf = path->nodes[0];
6879 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6880 if (found_key.objectid != objectid ||
6881 found_key.type != BTRFS_EXTENT_DATA_KEY)
6883 if (start + len <= found_key.offset)
6885 if (start > found_key.offset)
6888 em->orig_start = start;
6889 em->len = found_key.offset - start;
6893 btrfs_extent_item_to_extent_map(inode, path, item,
6896 if (found_type == BTRFS_FILE_EXTENT_REG ||
6897 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6899 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6903 size_t extent_offset;
6909 size = btrfs_file_extent_ram_bytes(leaf, item);
6910 extent_offset = page_offset(page) + pg_offset - extent_start;
6911 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6912 size - extent_offset);
6913 em->start = extent_start + extent_offset;
6914 em->len = ALIGN(copy_size, fs_info->sectorsize);
6915 em->orig_block_len = em->len;
6916 em->orig_start = em->start;
6917 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6919 btrfs_set_path_blocking(path);
6920 if (!PageUptodate(page)) {
6921 if (btrfs_file_extent_compression(leaf, item) !=
6922 BTRFS_COMPRESS_NONE) {
6923 ret = uncompress_inline(path, page, pg_offset,
6924 extent_offset, item);
6931 read_extent_buffer(leaf, map + pg_offset, ptr,
6933 if (pg_offset + copy_size < PAGE_SIZE) {
6934 memset(map + pg_offset + copy_size, 0,
6935 PAGE_SIZE - pg_offset -
6940 flush_dcache_page(page);
6942 set_extent_uptodate(io_tree, em->start,
6943 extent_map_end(em) - 1, NULL, GFP_NOFS);
6948 em->orig_start = start;
6951 em->block_start = EXTENT_MAP_HOLE;
6953 btrfs_release_path(path);
6954 if (em->start > start || extent_map_end(em) <= start) {
6956 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6957 em->start, em->len, start, len);
6963 write_lock(&em_tree->lock);
6964 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6965 write_unlock(&em_tree->lock);
6967 btrfs_free_path(path);
6969 trace_btrfs_get_extent(root, inode, em);
6972 free_extent_map(em);
6973 return ERR_PTR(err);
6975 BUG_ON(!em); /* Error is always set */
6979 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6981 size_t pg_offset, u64 start, u64 len,
6984 struct extent_map *em;
6985 struct extent_map *hole_em = NULL;
6986 u64 range_start = start;
6992 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6996 * If our em maps to:
6998 * - a pre-alloc extent,
6999 * there might actually be delalloc bytes behind it.
7001 if (em->block_start != EXTENT_MAP_HOLE &&
7002 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7007 /* check to see if we've wrapped (len == -1 or similar) */
7016 /* ok, we didn't find anything, lets look for delalloc */
7017 found = count_range_bits(&inode->io_tree, &range_start,
7018 end, len, EXTENT_DELALLOC, 1);
7019 found_end = range_start + found;
7020 if (found_end < range_start)
7021 found_end = (u64)-1;
7024 * we didn't find anything useful, return
7025 * the original results from get_extent()
7027 if (range_start > end || found_end <= start) {
7033 /* adjust the range_start to make sure it doesn't
7034 * go backwards from the start they passed in
7036 range_start = max(start, range_start);
7037 found = found_end - range_start;
7040 u64 hole_start = start;
7043 em = alloc_extent_map();
7049 * when btrfs_get_extent can't find anything it
7050 * returns one huge hole
7052 * make sure what it found really fits our range, and
7053 * adjust to make sure it is based on the start from
7057 u64 calc_end = extent_map_end(hole_em);
7059 if (calc_end <= start || (hole_em->start > end)) {
7060 free_extent_map(hole_em);
7063 hole_start = max(hole_em->start, start);
7064 hole_len = calc_end - hole_start;
7068 if (hole_em && range_start > hole_start) {
7069 /* our hole starts before our delalloc, so we
7070 * have to return just the parts of the hole
7071 * that go until the delalloc starts
7073 em->len = min(hole_len,
7074 range_start - hole_start);
7075 em->start = hole_start;
7076 em->orig_start = hole_start;
7078 * don't adjust block start at all,
7079 * it is fixed at EXTENT_MAP_HOLE
7081 em->block_start = hole_em->block_start;
7082 em->block_len = hole_len;
7083 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7084 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7086 em->start = range_start;
7088 em->orig_start = range_start;
7089 em->block_start = EXTENT_MAP_DELALLOC;
7090 em->block_len = found;
7097 free_extent_map(hole_em);
7099 free_extent_map(em);
7100 return ERR_PTR(err);
7105 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7108 const u64 orig_start,
7109 const u64 block_start,
7110 const u64 block_len,
7111 const u64 orig_block_len,
7112 const u64 ram_bytes,
7115 struct extent_map *em = NULL;
7118 if (type != BTRFS_ORDERED_NOCOW) {
7119 em = create_io_em(inode, start, len, orig_start,
7120 block_start, block_len, orig_block_len,
7122 BTRFS_COMPRESS_NONE, /* compress_type */
7127 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7128 len, block_len, type);
7131 free_extent_map(em);
7132 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7133 start + len - 1, 0);
7142 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7145 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7146 struct btrfs_root *root = BTRFS_I(inode)->root;
7147 struct extent_map *em;
7148 struct btrfs_key ins;
7152 alloc_hint = get_extent_allocation_hint(inode, start, len);
7153 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7154 0, alloc_hint, &ins, 1, 1);
7156 return ERR_PTR(ret);
7158 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7159 ins.objectid, ins.offset, ins.offset,
7160 ins.offset, BTRFS_ORDERED_REGULAR);
7161 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7163 btrfs_free_reserved_extent(fs_info, ins.objectid,
7170 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7171 * block must be cow'd
7173 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7174 u64 *orig_start, u64 *orig_block_len,
7177 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7178 struct btrfs_path *path;
7180 struct extent_buffer *leaf;
7181 struct btrfs_root *root = BTRFS_I(inode)->root;
7182 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7183 struct btrfs_file_extent_item *fi;
7184 struct btrfs_key key;
7191 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7193 path = btrfs_alloc_path();
7197 ret = btrfs_lookup_file_extent(NULL, root, path,
7198 btrfs_ino(BTRFS_I(inode)), offset, 0);
7202 slot = path->slots[0];
7205 /* can't find the item, must cow */
7212 leaf = path->nodes[0];
7213 btrfs_item_key_to_cpu(leaf, &key, slot);
7214 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7215 key.type != BTRFS_EXTENT_DATA_KEY) {
7216 /* not our file or wrong item type, must cow */
7220 if (key.offset > offset) {
7221 /* Wrong offset, must cow */
7225 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7226 found_type = btrfs_file_extent_type(leaf, fi);
7227 if (found_type != BTRFS_FILE_EXTENT_REG &&
7228 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7229 /* not a regular extent, must cow */
7233 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7236 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7237 if (extent_end <= offset)
7240 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7241 if (disk_bytenr == 0)
7244 if (btrfs_file_extent_compression(leaf, fi) ||
7245 btrfs_file_extent_encryption(leaf, fi) ||
7246 btrfs_file_extent_other_encoding(leaf, fi))
7250 * Do the same check as in btrfs_cross_ref_exist but without the
7251 * unnecessary search.
7253 if (btrfs_file_extent_generation(leaf, fi) <=
7254 btrfs_root_last_snapshot(&root->root_item))
7257 backref_offset = btrfs_file_extent_offset(leaf, fi);
7260 *orig_start = key.offset - backref_offset;
7261 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7262 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7265 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7268 num_bytes = min(offset + *len, extent_end) - offset;
7269 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7272 range_end = round_up(offset + num_bytes,
7273 root->fs_info->sectorsize) - 1;
7274 ret = test_range_bit(io_tree, offset, range_end,
7275 EXTENT_DELALLOC, 0, NULL);
7282 btrfs_release_path(path);
7285 * look for other files referencing this extent, if we
7286 * find any we must cow
7289 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7290 key.offset - backref_offset, disk_bytenr);
7297 * adjust disk_bytenr and num_bytes to cover just the bytes
7298 * in this extent we are about to write. If there
7299 * are any csums in that range we have to cow in order
7300 * to keep the csums correct
7302 disk_bytenr += backref_offset;
7303 disk_bytenr += offset - key.offset;
7304 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7307 * all of the above have passed, it is safe to overwrite this extent
7313 btrfs_free_path(path);
7317 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7318 struct extent_state **cached_state, int writing)
7320 struct btrfs_ordered_extent *ordered;
7324 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7327 * We're concerned with the entire range that we're going to be
7328 * doing DIO to, so we need to make sure there's no ordered
7329 * extents in this range.
7331 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7332 lockend - lockstart + 1);
7335 * We need to make sure there are no buffered pages in this
7336 * range either, we could have raced between the invalidate in
7337 * generic_file_direct_write and locking the extent. The
7338 * invalidate needs to happen so that reads after a write do not
7342 (!writing || !filemap_range_has_page(inode->i_mapping,
7343 lockstart, lockend)))
7346 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7351 * If we are doing a DIO read and the ordered extent we
7352 * found is for a buffered write, we can not wait for it
7353 * to complete and retry, because if we do so we can
7354 * deadlock with concurrent buffered writes on page
7355 * locks. This happens only if our DIO read covers more
7356 * than one extent map, if at this point has already
7357 * created an ordered extent for a previous extent map
7358 * and locked its range in the inode's io tree, and a
7359 * concurrent write against that previous extent map's
7360 * range and this range started (we unlock the ranges
7361 * in the io tree only when the bios complete and
7362 * buffered writes always lock pages before attempting
7363 * to lock range in the io tree).
7366 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7367 btrfs_start_ordered_extent(inode, ordered, 1);
7370 btrfs_put_ordered_extent(ordered);
7373 * We could trigger writeback for this range (and wait
7374 * for it to complete) and then invalidate the pages for
7375 * this range (through invalidate_inode_pages2_range()),
7376 * but that can lead us to a deadlock with a concurrent
7377 * call to readpages() (a buffered read or a defrag call
7378 * triggered a readahead) on a page lock due to an
7379 * ordered dio extent we created before but did not have
7380 * yet a corresponding bio submitted (whence it can not
7381 * complete), which makes readpages() wait for that
7382 * ordered extent to complete while holding a lock on
7397 /* The callers of this must take lock_extent() */
7398 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7399 u64 orig_start, u64 block_start,
7400 u64 block_len, u64 orig_block_len,
7401 u64 ram_bytes, int compress_type,
7404 struct extent_map_tree *em_tree;
7405 struct extent_map *em;
7406 struct btrfs_root *root = BTRFS_I(inode)->root;
7409 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7410 type == BTRFS_ORDERED_COMPRESSED ||
7411 type == BTRFS_ORDERED_NOCOW ||
7412 type == BTRFS_ORDERED_REGULAR);
7414 em_tree = &BTRFS_I(inode)->extent_tree;
7415 em = alloc_extent_map();
7417 return ERR_PTR(-ENOMEM);
7420 em->orig_start = orig_start;
7422 em->block_len = block_len;
7423 em->block_start = block_start;
7424 em->bdev = root->fs_info->fs_devices->latest_bdev;
7425 em->orig_block_len = orig_block_len;
7426 em->ram_bytes = ram_bytes;
7427 em->generation = -1;
7428 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7429 if (type == BTRFS_ORDERED_PREALLOC) {
7430 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7431 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7432 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7433 em->compress_type = compress_type;
7437 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7438 em->start + em->len - 1, 0);
7439 write_lock(&em_tree->lock);
7440 ret = add_extent_mapping(em_tree, em, 1);
7441 write_unlock(&em_tree->lock);
7443 * The caller has taken lock_extent(), who could race with us
7446 } while (ret == -EEXIST);
7449 free_extent_map(em);
7450 return ERR_PTR(ret);
7453 /* em got 2 refs now, callers needs to do free_extent_map once. */
7458 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7459 struct buffer_head *bh_result,
7460 struct inode *inode,
7463 if (em->block_start == EXTENT_MAP_HOLE ||
7464 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7467 len = min(len, em->len - (start - em->start));
7469 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7471 bh_result->b_size = len;
7472 bh_result->b_bdev = em->bdev;
7473 set_buffer_mapped(bh_result);
7478 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7479 struct buffer_head *bh_result,
7480 struct inode *inode,
7481 struct btrfs_dio_data *dio_data,
7484 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7485 struct extent_map *em = *map;
7489 * We don't allocate a new extent in the following cases
7491 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7493 * 2) The extent is marked as PREALLOC. We're good to go here and can
7494 * just use the extent.
7497 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7498 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7499 em->block_start != EXTENT_MAP_HOLE)) {
7501 u64 block_start, orig_start, orig_block_len, ram_bytes;
7503 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7504 type = BTRFS_ORDERED_PREALLOC;
7506 type = BTRFS_ORDERED_NOCOW;
7507 len = min(len, em->len - (start - em->start));
7508 block_start = em->block_start + (start - em->start);
7510 if (can_nocow_extent(inode, start, &len, &orig_start,
7511 &orig_block_len, &ram_bytes) == 1 &&
7512 btrfs_inc_nocow_writers(fs_info, block_start)) {
7513 struct extent_map *em2;
7515 em2 = btrfs_create_dio_extent(inode, start, len,
7516 orig_start, block_start,
7517 len, orig_block_len,
7519 btrfs_dec_nocow_writers(fs_info, block_start);
7520 if (type == BTRFS_ORDERED_PREALLOC) {
7521 free_extent_map(em);
7525 if (em2 && IS_ERR(em2)) {
7530 * For inode marked NODATACOW or extent marked PREALLOC,
7531 * use the existing or preallocated extent, so does not
7532 * need to adjust btrfs_space_info's bytes_may_use.
7534 btrfs_free_reserved_data_space_noquota(inode, start,
7540 /* this will cow the extent */
7541 len = bh_result->b_size;
7542 free_extent_map(em);
7543 *map = em = btrfs_new_extent_direct(inode, start, len);
7549 len = min(len, em->len - (start - em->start));
7552 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7554 bh_result->b_size = len;
7555 bh_result->b_bdev = em->bdev;
7556 set_buffer_mapped(bh_result);
7558 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7559 set_buffer_new(bh_result);
7562 * Need to update the i_size under the extent lock so buffered
7563 * readers will get the updated i_size when we unlock.
7565 if (!dio_data->overwrite && start + len > i_size_read(inode))
7566 i_size_write(inode, start + len);
7568 WARN_ON(dio_data->reserve < len);
7569 dio_data->reserve -= len;
7570 dio_data->unsubmitted_oe_range_end = start + len;
7571 current->journal_info = dio_data;
7576 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7577 struct buffer_head *bh_result, int create)
7579 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7580 struct extent_map *em;
7581 struct extent_state *cached_state = NULL;
7582 struct btrfs_dio_data *dio_data = NULL;
7583 u64 start = iblock << inode->i_blkbits;
7584 u64 lockstart, lockend;
7585 u64 len = bh_result->b_size;
7586 int unlock_bits = EXTENT_LOCKED;
7590 unlock_bits |= EXTENT_DIRTY;
7592 len = min_t(u64, len, fs_info->sectorsize);
7595 lockend = start + len - 1;
7597 if (current->journal_info) {
7599 * Need to pull our outstanding extents and set journal_info to NULL so
7600 * that anything that needs to check if there's a transaction doesn't get
7603 dio_data = current->journal_info;
7604 current->journal_info = NULL;
7608 * If this errors out it's because we couldn't invalidate pagecache for
7609 * this range and we need to fallback to buffered.
7611 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7617 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7624 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7625 * io. INLINE is special, and we could probably kludge it in here, but
7626 * it's still buffered so for safety lets just fall back to the generic
7629 * For COMPRESSED we _have_ to read the entire extent in so we can
7630 * decompress it, so there will be buffering required no matter what we
7631 * do, so go ahead and fallback to buffered.
7633 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7634 * to buffered IO. Don't blame me, this is the price we pay for using
7637 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7638 em->block_start == EXTENT_MAP_INLINE) {
7639 free_extent_map(em);
7645 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7646 dio_data, start, len);
7650 /* clear and unlock the entire range */
7651 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7652 unlock_bits, 1, 0, &cached_state);
7654 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7656 /* Can be negative only if we read from a hole */
7659 free_extent_map(em);
7663 * We need to unlock only the end area that we aren't using.
7664 * The rest is going to be unlocked by the endio routine.
7666 lockstart = start + bh_result->b_size;
7667 if (lockstart < lockend) {
7668 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7669 lockend, unlock_bits, 1, 0,
7672 free_extent_state(cached_state);
7676 free_extent_map(em);
7681 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7682 unlock_bits, 1, 0, &cached_state);
7685 current->journal_info = dio_data;
7689 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7693 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7696 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7698 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7702 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7707 static int btrfs_check_dio_repairable(struct inode *inode,
7708 struct bio *failed_bio,
7709 struct io_failure_record *failrec,
7712 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7715 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7716 if (num_copies == 1) {
7718 * we only have a single copy of the data, so don't bother with
7719 * all the retry and error correction code that follows. no
7720 * matter what the error is, it is very likely to persist.
7722 btrfs_debug(fs_info,
7723 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7724 num_copies, failrec->this_mirror, failed_mirror);
7728 failrec->failed_mirror = failed_mirror;
7729 failrec->this_mirror++;
7730 if (failrec->this_mirror == failed_mirror)
7731 failrec->this_mirror++;
7733 if (failrec->this_mirror > num_copies) {
7734 btrfs_debug(fs_info,
7735 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7736 num_copies, failrec->this_mirror, failed_mirror);
7743 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7744 struct page *page, unsigned int pgoff,
7745 u64 start, u64 end, int failed_mirror,
7746 bio_end_io_t *repair_endio, void *repair_arg)
7748 struct io_failure_record *failrec;
7749 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7750 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7753 unsigned int read_mode = 0;
7756 blk_status_t status;
7757 struct bio_vec bvec;
7759 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7761 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7763 return errno_to_blk_status(ret);
7765 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7768 free_io_failure(failure_tree, io_tree, failrec);
7769 return BLK_STS_IOERR;
7772 segs = bio_segments(failed_bio);
7773 bio_get_first_bvec(failed_bio, &bvec);
7775 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7776 read_mode |= REQ_FAILFAST_DEV;
7778 isector = start - btrfs_io_bio(failed_bio)->logical;
7779 isector >>= inode->i_sb->s_blocksize_bits;
7780 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7781 pgoff, isector, repair_endio, repair_arg);
7782 bio->bi_opf = REQ_OP_READ | read_mode;
7784 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7785 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7786 read_mode, failrec->this_mirror, failrec->in_validation);
7788 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7790 free_io_failure(failure_tree, io_tree, failrec);
7797 struct btrfs_retry_complete {
7798 struct completion done;
7799 struct inode *inode;
7804 static void btrfs_retry_endio_nocsum(struct bio *bio)
7806 struct btrfs_retry_complete *done = bio->bi_private;
7807 struct inode *inode = done->inode;
7808 struct bio_vec *bvec;
7809 struct extent_io_tree *io_tree, *failure_tree;
7815 ASSERT(bio->bi_vcnt == 1);
7816 io_tree = &BTRFS_I(inode)->io_tree;
7817 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7818 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7821 ASSERT(!bio_flagged(bio, BIO_CLONED));
7822 bio_for_each_segment_all(bvec, bio, i)
7823 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7824 io_tree, done->start, bvec->bv_page,
7825 btrfs_ino(BTRFS_I(inode)), 0);
7827 complete(&done->done);
7831 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7832 struct btrfs_io_bio *io_bio)
7834 struct btrfs_fs_info *fs_info;
7835 struct bio_vec bvec;
7836 struct bvec_iter iter;
7837 struct btrfs_retry_complete done;
7843 blk_status_t err = BLK_STS_OK;
7845 fs_info = BTRFS_I(inode)->root->fs_info;
7846 sectorsize = fs_info->sectorsize;
7848 start = io_bio->logical;
7850 io_bio->bio.bi_iter = io_bio->iter;
7852 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7853 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7854 pgoff = bvec.bv_offset;
7856 next_block_or_try_again:
7859 init_completion(&done.done);
7861 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7862 pgoff, start, start + sectorsize - 1,
7864 btrfs_retry_endio_nocsum, &done);
7870 wait_for_completion_io(&done.done);
7872 if (!done.uptodate) {
7873 /* We might have another mirror, so try again */
7874 goto next_block_or_try_again;
7878 start += sectorsize;
7882 pgoff += sectorsize;
7883 ASSERT(pgoff < PAGE_SIZE);
7884 goto next_block_or_try_again;
7891 static void btrfs_retry_endio(struct bio *bio)
7893 struct btrfs_retry_complete *done = bio->bi_private;
7894 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7895 struct extent_io_tree *io_tree, *failure_tree;
7896 struct inode *inode = done->inode;
7897 struct bio_vec *bvec;
7907 ASSERT(bio->bi_vcnt == 1);
7908 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7910 io_tree = &BTRFS_I(inode)->io_tree;
7911 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7913 ASSERT(!bio_flagged(bio, BIO_CLONED));
7914 bio_for_each_segment_all(bvec, bio, i) {
7915 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7916 bvec->bv_offset, done->start,
7919 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7920 failure_tree, io_tree, done->start,
7922 btrfs_ino(BTRFS_I(inode)),
7928 done->uptodate = uptodate;
7930 complete(&done->done);
7934 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7935 struct btrfs_io_bio *io_bio, blk_status_t err)
7937 struct btrfs_fs_info *fs_info;
7938 struct bio_vec bvec;
7939 struct bvec_iter iter;
7940 struct btrfs_retry_complete done;
7947 bool uptodate = (err == 0);
7949 blk_status_t status;
7951 fs_info = BTRFS_I(inode)->root->fs_info;
7952 sectorsize = fs_info->sectorsize;
7955 start = io_bio->logical;
7957 io_bio->bio.bi_iter = io_bio->iter;
7959 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7960 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7962 pgoff = bvec.bv_offset;
7965 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7966 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7967 bvec.bv_page, pgoff, start, sectorsize);
7974 init_completion(&done.done);
7976 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7977 pgoff, start, start + sectorsize - 1,
7978 io_bio->mirror_num, btrfs_retry_endio,
7985 wait_for_completion_io(&done.done);
7987 if (!done.uptodate) {
7988 /* We might have another mirror, so try again */
7992 offset += sectorsize;
7993 start += sectorsize;
7999 pgoff += sectorsize;
8000 ASSERT(pgoff < PAGE_SIZE);
8008 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8009 struct btrfs_io_bio *io_bio, blk_status_t err)
8011 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8015 return __btrfs_correct_data_nocsum(inode, io_bio);
8019 return __btrfs_subio_endio_read(inode, io_bio, err);
8023 static void btrfs_endio_direct_read(struct bio *bio)
8025 struct btrfs_dio_private *dip = bio->bi_private;
8026 struct inode *inode = dip->inode;
8027 struct bio *dio_bio;
8028 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8029 blk_status_t err = bio->bi_status;
8031 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8032 err = btrfs_subio_endio_read(inode, io_bio, err);
8034 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8035 dip->logical_offset + dip->bytes - 1);
8036 dio_bio = dip->dio_bio;
8040 dio_bio->bi_status = err;
8041 dio_end_io(dio_bio);
8042 btrfs_io_bio_free_csum(io_bio);
8046 static void __endio_write_update_ordered(struct inode *inode,
8047 const u64 offset, const u64 bytes,
8048 const bool uptodate)
8050 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8051 struct btrfs_ordered_extent *ordered = NULL;
8052 struct btrfs_workqueue *wq;
8053 btrfs_work_func_t func;
8054 u64 ordered_offset = offset;
8055 u64 ordered_bytes = bytes;
8058 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8059 wq = fs_info->endio_freespace_worker;
8060 func = btrfs_freespace_write_helper;
8062 wq = fs_info->endio_write_workers;
8063 func = btrfs_endio_write_helper;
8066 while (ordered_offset < offset + bytes) {
8067 last_offset = ordered_offset;
8068 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8072 btrfs_init_work(&ordered->work, func,
8075 btrfs_queue_work(wq, &ordered->work);
8078 * If btrfs_dec_test_ordered_pending does not find any ordered
8079 * extent in the range, we can exit.
8081 if (ordered_offset == last_offset)
8084 * Our bio might span multiple ordered extents. In this case
8085 * we keep goin until we have accounted the whole dio.
8087 if (ordered_offset < offset + bytes) {
8088 ordered_bytes = offset + bytes - ordered_offset;
8094 static void btrfs_endio_direct_write(struct bio *bio)
8096 struct btrfs_dio_private *dip = bio->bi_private;
8097 struct bio *dio_bio = dip->dio_bio;
8099 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8100 dip->bytes, !bio->bi_status);
8104 dio_bio->bi_status = bio->bi_status;
8105 dio_end_io(dio_bio);
8109 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8110 struct bio *bio, u64 offset)
8112 struct inode *inode = private_data;
8114 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8115 BUG_ON(ret); /* -ENOMEM */
8119 static void btrfs_end_dio_bio(struct bio *bio)
8121 struct btrfs_dio_private *dip = bio->bi_private;
8122 blk_status_t err = bio->bi_status;
8125 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8126 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8127 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8129 (unsigned long long)bio->bi_iter.bi_sector,
8130 bio->bi_iter.bi_size, err);
8132 if (dip->subio_endio)
8133 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8137 * We want to perceive the errors flag being set before
8138 * decrementing the reference count. We don't need a barrier
8139 * since atomic operations with a return value are fully
8140 * ordered as per atomic_t.txt
8145 /* if there are more bios still pending for this dio, just exit */
8146 if (!atomic_dec_and_test(&dip->pending_bios))
8150 bio_io_error(dip->orig_bio);
8152 dip->dio_bio->bi_status = BLK_STS_OK;
8153 bio_endio(dip->orig_bio);
8159 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8160 struct btrfs_dio_private *dip,
8164 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8165 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8169 * We load all the csum data we need when we submit
8170 * the first bio to reduce the csum tree search and
8173 if (dip->logical_offset == file_offset) {
8174 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8180 if (bio == dip->orig_bio)
8183 file_offset -= dip->logical_offset;
8184 file_offset >>= inode->i_sb->s_blocksize_bits;
8185 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8190 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8191 struct inode *inode, u64 file_offset, int async_submit)
8193 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8194 struct btrfs_dio_private *dip = bio->bi_private;
8195 bool write = bio_op(bio) == REQ_OP_WRITE;
8198 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8200 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8203 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8208 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8211 if (write && async_submit) {
8212 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8214 btrfs_submit_bio_start_direct_io);
8218 * If we aren't doing async submit, calculate the csum of the
8221 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8225 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8231 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8236 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8238 struct inode *inode = dip->inode;
8239 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8241 struct bio *orig_bio = dip->orig_bio;
8242 u64 start_sector = orig_bio->bi_iter.bi_sector;
8243 u64 file_offset = dip->logical_offset;
8245 int async_submit = 0;
8247 int clone_offset = 0;
8250 blk_status_t status;
8252 map_length = orig_bio->bi_iter.bi_size;
8253 submit_len = map_length;
8254 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8255 &map_length, NULL, 0);
8259 if (map_length >= submit_len) {
8261 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8265 /* async crcs make it difficult to collect full stripe writes. */
8266 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8272 ASSERT(map_length <= INT_MAX);
8273 atomic_inc(&dip->pending_bios);
8275 clone_len = min_t(int, submit_len, map_length);
8278 * This will never fail as it's passing GPF_NOFS and
8279 * the allocation is backed by btrfs_bioset.
8281 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8283 bio->bi_private = dip;
8284 bio->bi_end_io = btrfs_end_dio_bio;
8285 btrfs_io_bio(bio)->logical = file_offset;
8287 ASSERT(submit_len >= clone_len);
8288 submit_len -= clone_len;
8289 if (submit_len == 0)
8293 * Increase the count before we submit the bio so we know
8294 * the end IO handler won't happen before we increase the
8295 * count. Otherwise, the dip might get freed before we're
8296 * done setting it up.
8298 atomic_inc(&dip->pending_bios);
8300 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8304 atomic_dec(&dip->pending_bios);
8308 clone_offset += clone_len;
8309 start_sector += clone_len >> 9;
8310 file_offset += clone_len;
8312 map_length = submit_len;
8313 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8314 start_sector << 9, &map_length, NULL, 0);
8317 } while (submit_len > 0);
8320 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8328 * Before atomic variable goto zero, we must make sure dip->errors is
8329 * perceived to be set. This ordering is ensured by the fact that an
8330 * atomic operations with a return value are fully ordered as per
8333 if (atomic_dec_and_test(&dip->pending_bios))
8334 bio_io_error(dip->orig_bio);
8336 /* bio_end_io() will handle error, so we needn't return it */
8340 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8343 struct btrfs_dio_private *dip = NULL;
8344 struct bio *bio = NULL;
8345 struct btrfs_io_bio *io_bio;
8346 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8349 bio = btrfs_bio_clone(dio_bio);
8351 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8357 dip->private = dio_bio->bi_private;
8359 dip->logical_offset = file_offset;
8360 dip->bytes = dio_bio->bi_iter.bi_size;
8361 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8362 bio->bi_private = dip;
8363 dip->orig_bio = bio;
8364 dip->dio_bio = dio_bio;
8365 atomic_set(&dip->pending_bios, 0);
8366 io_bio = btrfs_io_bio(bio);
8367 io_bio->logical = file_offset;
8370 bio->bi_end_io = btrfs_endio_direct_write;
8372 bio->bi_end_io = btrfs_endio_direct_read;
8373 dip->subio_endio = btrfs_subio_endio_read;
8377 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8378 * even if we fail to submit a bio, because in such case we do the
8379 * corresponding error handling below and it must not be done a second
8380 * time by btrfs_direct_IO().
8383 struct btrfs_dio_data *dio_data = current->journal_info;
8385 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8387 dio_data->unsubmitted_oe_range_start =
8388 dio_data->unsubmitted_oe_range_end;
8391 ret = btrfs_submit_direct_hook(dip);
8395 btrfs_io_bio_free_csum(io_bio);
8399 * If we arrived here it means either we failed to submit the dip
8400 * or we either failed to clone the dio_bio or failed to allocate the
8401 * dip. If we cloned the dio_bio and allocated the dip, we can just
8402 * call bio_endio against our io_bio so that we get proper resource
8403 * cleanup if we fail to submit the dip, otherwise, we must do the
8404 * same as btrfs_endio_direct_[write|read] because we can't call these
8405 * callbacks - they require an allocated dip and a clone of dio_bio.
8410 * The end io callbacks free our dip, do the final put on bio
8411 * and all the cleanup and final put for dio_bio (through
8418 __endio_write_update_ordered(inode,
8420 dio_bio->bi_iter.bi_size,
8423 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8424 file_offset + dio_bio->bi_iter.bi_size - 1);
8426 dio_bio->bi_status = BLK_STS_IOERR;
8428 * Releases and cleans up our dio_bio, no need to bio_put()
8429 * nor bio_endio()/bio_io_error() against dio_bio.
8431 dio_end_io(dio_bio);
8438 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8439 const struct iov_iter *iter, loff_t offset)
8443 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8444 ssize_t retval = -EINVAL;
8446 if (offset & blocksize_mask)
8449 if (iov_iter_alignment(iter) & blocksize_mask)
8452 /* If this is a write we don't need to check anymore */
8453 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8456 * Check to make sure we don't have duplicate iov_base's in this
8457 * iovec, if so return EINVAL, otherwise we'll get csum errors
8458 * when reading back.
8460 for (seg = 0; seg < iter->nr_segs; seg++) {
8461 for (i = seg + 1; i < iter->nr_segs; i++) {
8462 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8471 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8473 struct file *file = iocb->ki_filp;
8474 struct inode *inode = file->f_mapping->host;
8475 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8476 struct btrfs_dio_data dio_data = { 0 };
8477 struct extent_changeset *data_reserved = NULL;
8478 loff_t offset = iocb->ki_pos;
8482 bool relock = false;
8485 if (check_direct_IO(fs_info, iter, offset))
8488 inode_dio_begin(inode);
8491 * The generic stuff only does filemap_write_and_wait_range, which
8492 * isn't enough if we've written compressed pages to this area, so
8493 * we need to flush the dirty pages again to make absolutely sure
8494 * that any outstanding dirty pages are on disk.
8496 count = iov_iter_count(iter);
8497 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8498 &BTRFS_I(inode)->runtime_flags))
8499 filemap_fdatawrite_range(inode->i_mapping, offset,
8500 offset + count - 1);
8502 if (iov_iter_rw(iter) == WRITE) {
8504 * If the write DIO is beyond the EOF, we need update
8505 * the isize, but it is protected by i_mutex. So we can
8506 * not unlock the i_mutex at this case.
8508 if (offset + count <= inode->i_size) {
8509 dio_data.overwrite = 1;
8510 inode_unlock(inode);
8512 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8516 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8522 * We need to know how many extents we reserved so that we can
8523 * do the accounting properly if we go over the number we
8524 * originally calculated. Abuse current->journal_info for this.
8526 dio_data.reserve = round_up(count,
8527 fs_info->sectorsize);
8528 dio_data.unsubmitted_oe_range_start = (u64)offset;
8529 dio_data.unsubmitted_oe_range_end = (u64)offset;
8530 current->journal_info = &dio_data;
8531 down_read(&BTRFS_I(inode)->dio_sem);
8532 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8533 &BTRFS_I(inode)->runtime_flags)) {
8534 inode_dio_end(inode);
8535 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8539 ret = __blockdev_direct_IO(iocb, inode,
8540 fs_info->fs_devices->latest_bdev,
8541 iter, btrfs_get_blocks_direct, NULL,
8542 btrfs_submit_direct, flags);
8543 if (iov_iter_rw(iter) == WRITE) {
8544 up_read(&BTRFS_I(inode)->dio_sem);
8545 current->journal_info = NULL;
8546 if (ret < 0 && ret != -EIOCBQUEUED) {
8547 if (dio_data.reserve)
8548 btrfs_delalloc_release_space(inode, data_reserved,
8549 offset, dio_data.reserve, true);
8551 * On error we might have left some ordered extents
8552 * without submitting corresponding bios for them, so
8553 * cleanup them up to avoid other tasks getting them
8554 * and waiting for them to complete forever.
8556 if (dio_data.unsubmitted_oe_range_start <
8557 dio_data.unsubmitted_oe_range_end)
8558 __endio_write_update_ordered(inode,
8559 dio_data.unsubmitted_oe_range_start,
8560 dio_data.unsubmitted_oe_range_end -
8561 dio_data.unsubmitted_oe_range_start,
8563 } else if (ret >= 0 && (size_t)ret < count)
8564 btrfs_delalloc_release_space(inode, data_reserved,
8565 offset, count - (size_t)ret, true);
8566 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8570 inode_dio_end(inode);
8574 extent_changeset_free(data_reserved);
8578 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8580 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8581 __u64 start, __u64 len)
8585 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8589 return extent_fiemap(inode, fieinfo, start, len);
8592 int btrfs_readpage(struct file *file, struct page *page)
8594 struct extent_io_tree *tree;
8595 tree = &BTRFS_I(page->mapping->host)->io_tree;
8596 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8599 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8601 struct inode *inode = page->mapping->host;
8604 if (current->flags & PF_MEMALLOC) {
8605 redirty_page_for_writepage(wbc, page);
8611 * If we are under memory pressure we will call this directly from the
8612 * VM, we need to make sure we have the inode referenced for the ordered
8613 * extent. If not just return like we didn't do anything.
8615 if (!igrab(inode)) {
8616 redirty_page_for_writepage(wbc, page);
8617 return AOP_WRITEPAGE_ACTIVATE;
8619 ret = extent_write_full_page(page, wbc);
8620 btrfs_add_delayed_iput(inode);
8624 static int btrfs_writepages(struct address_space *mapping,
8625 struct writeback_control *wbc)
8627 return extent_writepages(mapping, wbc);
8631 btrfs_readpages(struct file *file, struct address_space *mapping,
8632 struct list_head *pages, unsigned nr_pages)
8634 return extent_readpages(mapping, pages, nr_pages);
8637 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8639 int ret = try_release_extent_mapping(page, gfp_flags);
8641 ClearPagePrivate(page);
8642 set_page_private(page, 0);
8648 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8650 if (PageWriteback(page) || PageDirty(page))
8652 return __btrfs_releasepage(page, gfp_flags);
8655 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8656 unsigned int length)
8658 struct inode *inode = page->mapping->host;
8659 struct extent_io_tree *tree;
8660 struct btrfs_ordered_extent *ordered;
8661 struct extent_state *cached_state = NULL;
8662 u64 page_start = page_offset(page);
8663 u64 page_end = page_start + PAGE_SIZE - 1;
8666 int inode_evicting = inode->i_state & I_FREEING;
8669 * we have the page locked, so new writeback can't start,
8670 * and the dirty bit won't be cleared while we are here.
8672 * Wait for IO on this page so that we can safely clear
8673 * the PagePrivate2 bit and do ordered accounting
8675 wait_on_page_writeback(page);
8677 tree = &BTRFS_I(inode)->io_tree;
8679 btrfs_releasepage(page, GFP_NOFS);
8683 if (!inode_evicting)
8684 lock_extent_bits(tree, page_start, page_end, &cached_state);
8687 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8688 page_end - start + 1);
8690 end = min(page_end, ordered->file_offset + ordered->len - 1);
8692 * IO on this page will never be started, so we need
8693 * to account for any ordered extents now
8695 if (!inode_evicting)
8696 clear_extent_bit(tree, start, end,
8697 EXTENT_DIRTY | EXTENT_DELALLOC |
8698 EXTENT_DELALLOC_NEW |
8699 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8700 EXTENT_DEFRAG, 1, 0, &cached_state);
8702 * whoever cleared the private bit is responsible
8703 * for the finish_ordered_io
8705 if (TestClearPagePrivate2(page)) {
8706 struct btrfs_ordered_inode_tree *tree;
8709 tree = &BTRFS_I(inode)->ordered_tree;
8711 spin_lock_irq(&tree->lock);
8712 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8713 new_len = start - ordered->file_offset;
8714 if (new_len < ordered->truncated_len)
8715 ordered->truncated_len = new_len;
8716 spin_unlock_irq(&tree->lock);
8718 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8720 end - start + 1, 1))
8721 btrfs_finish_ordered_io(ordered);
8723 btrfs_put_ordered_extent(ordered);
8724 if (!inode_evicting) {
8725 cached_state = NULL;
8726 lock_extent_bits(tree, start, end,
8731 if (start < page_end)
8736 * Qgroup reserved space handler
8737 * Page here will be either
8738 * 1) Already written to disk
8739 * In this case, its reserved space is released from data rsv map
8740 * and will be freed by delayed_ref handler finally.
8741 * So even we call qgroup_free_data(), it won't decrease reserved
8743 * 2) Not written to disk
8744 * This means the reserved space should be freed here. However,
8745 * if a truncate invalidates the page (by clearing PageDirty)
8746 * and the page is accounted for while allocating extent
8747 * in btrfs_check_data_free_space() we let delayed_ref to
8748 * free the entire extent.
8750 if (PageDirty(page))
8751 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8752 if (!inode_evicting) {
8753 clear_extent_bit(tree, page_start, page_end,
8754 EXTENT_LOCKED | EXTENT_DIRTY |
8755 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8756 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8759 __btrfs_releasepage(page, GFP_NOFS);
8762 ClearPageChecked(page);
8763 if (PagePrivate(page)) {
8764 ClearPagePrivate(page);
8765 set_page_private(page, 0);
8771 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8772 * called from a page fault handler when a page is first dirtied. Hence we must
8773 * be careful to check for EOF conditions here. We set the page up correctly
8774 * for a written page which means we get ENOSPC checking when writing into
8775 * holes and correct delalloc and unwritten extent mapping on filesystems that
8776 * support these features.
8778 * We are not allowed to take the i_mutex here so we have to play games to
8779 * protect against truncate races as the page could now be beyond EOF. Because
8780 * truncate_setsize() writes the inode size before removing pages, once we have
8781 * the page lock we can determine safely if the page is beyond EOF. If it is not
8782 * beyond EOF, then the page is guaranteed safe against truncation until we
8785 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8787 struct page *page = vmf->page;
8788 struct inode *inode = file_inode(vmf->vma->vm_file);
8789 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8790 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8791 struct btrfs_ordered_extent *ordered;
8792 struct extent_state *cached_state = NULL;
8793 struct extent_changeset *data_reserved = NULL;
8795 unsigned long zero_start;
8805 reserved_space = PAGE_SIZE;
8807 sb_start_pagefault(inode->i_sb);
8808 page_start = page_offset(page);
8809 page_end = page_start + PAGE_SIZE - 1;
8813 * Reserving delalloc space after obtaining the page lock can lead to
8814 * deadlock. For example, if a dirty page is locked by this function
8815 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8816 * dirty page write out, then the btrfs_writepage() function could
8817 * end up waiting indefinitely to get a lock on the page currently
8818 * being processed by btrfs_page_mkwrite() function.
8820 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8823 ret2 = file_update_time(vmf->vma->vm_file);
8827 ret = vmf_error(ret2);
8833 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8836 size = i_size_read(inode);
8838 if ((page->mapping != inode->i_mapping) ||
8839 (page_start >= size)) {
8840 /* page got truncated out from underneath us */
8843 wait_on_page_writeback(page);
8845 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8846 set_page_extent_mapped(page);
8849 * we can't set the delalloc bits if there are pending ordered
8850 * extents. Drop our locks and wait for them to finish
8852 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8855 unlock_extent_cached(io_tree, page_start, page_end,
8858 btrfs_start_ordered_extent(inode, ordered, 1);
8859 btrfs_put_ordered_extent(ordered);
8863 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8864 reserved_space = round_up(size - page_start,
8865 fs_info->sectorsize);
8866 if (reserved_space < PAGE_SIZE) {
8867 end = page_start + reserved_space - 1;
8868 btrfs_delalloc_release_space(inode, data_reserved,
8869 page_start, PAGE_SIZE - reserved_space,
8875 * page_mkwrite gets called when the page is firstly dirtied after it's
8876 * faulted in, but write(2) could also dirty a page and set delalloc
8877 * bits, thus in this case for space account reason, we still need to
8878 * clear any delalloc bits within this page range since we have to
8879 * reserve data&meta space before lock_page() (see above comments).
8881 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8882 EXTENT_DIRTY | EXTENT_DELALLOC |
8883 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8884 0, 0, &cached_state);
8886 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8889 unlock_extent_cached(io_tree, page_start, page_end,
8891 ret = VM_FAULT_SIGBUS;
8896 /* page is wholly or partially inside EOF */
8897 if (page_start + PAGE_SIZE > size)
8898 zero_start = size & ~PAGE_MASK;
8900 zero_start = PAGE_SIZE;
8902 if (zero_start != PAGE_SIZE) {
8904 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8905 flush_dcache_page(page);
8908 ClearPageChecked(page);
8909 set_page_dirty(page);
8910 SetPageUptodate(page);
8912 BTRFS_I(inode)->last_trans = fs_info->generation;
8913 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8914 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8916 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8919 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8920 sb_end_pagefault(inode->i_sb);
8921 extent_changeset_free(data_reserved);
8922 return VM_FAULT_LOCKED;
8928 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8929 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8930 reserved_space, (ret != 0));
8932 sb_end_pagefault(inode->i_sb);
8933 extent_changeset_free(data_reserved);
8937 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8939 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8940 struct btrfs_root *root = BTRFS_I(inode)->root;
8941 struct btrfs_block_rsv *rsv;
8943 struct btrfs_trans_handle *trans;
8944 u64 mask = fs_info->sectorsize - 1;
8945 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8947 if (!skip_writeback) {
8948 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8955 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8956 * things going on here:
8958 * 1) We need to reserve space to update our inode.
8960 * 2) We need to have something to cache all the space that is going to
8961 * be free'd up by the truncate operation, but also have some slack
8962 * space reserved in case it uses space during the truncate (thank you
8963 * very much snapshotting).
8965 * And we need these to be separate. The fact is we can use a lot of
8966 * space doing the truncate, and we have no earthly idea how much space
8967 * we will use, so we need the truncate reservation to be separate so it
8968 * doesn't end up using space reserved for updating the inode. We also
8969 * need to be able to stop the transaction and start a new one, which
8970 * means we need to be able to update the inode several times, and we
8971 * have no idea of knowing how many times that will be, so we can't just
8972 * reserve 1 item for the entirety of the operation, so that has to be
8973 * done separately as well.
8975 * So that leaves us with
8977 * 1) rsv - for the truncate reservation, which we will steal from the
8978 * transaction reservation.
8979 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8980 * updating the inode.
8982 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8985 rsv->size = min_size;
8989 * 1 for the truncate slack space
8990 * 1 for updating the inode.
8992 trans = btrfs_start_transaction(root, 2);
8993 if (IS_ERR(trans)) {
8994 ret = PTR_ERR(trans);
8998 /* Migrate the slack space for the truncate to our reserve */
8999 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9004 * So if we truncate and then write and fsync we normally would just
9005 * write the extents that changed, which is a problem if we need to
9006 * first truncate that entire inode. So set this flag so we write out
9007 * all of the extents in the inode to the sync log so we're completely
9010 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9011 trans->block_rsv = rsv;
9014 ret = btrfs_truncate_inode_items(trans, root, inode,
9016 BTRFS_EXTENT_DATA_KEY);
9017 trans->block_rsv = &fs_info->trans_block_rsv;
9018 if (ret != -ENOSPC && ret != -EAGAIN)
9021 ret = btrfs_update_inode(trans, root, inode);
9025 btrfs_end_transaction(trans);
9026 btrfs_btree_balance_dirty(fs_info);
9028 trans = btrfs_start_transaction(root, 2);
9029 if (IS_ERR(trans)) {
9030 ret = PTR_ERR(trans);
9035 btrfs_block_rsv_release(fs_info, rsv, -1);
9036 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9037 rsv, min_size, false);
9038 BUG_ON(ret); /* shouldn't happen */
9039 trans->block_rsv = rsv;
9043 * We can't call btrfs_truncate_block inside a trans handle as we could
9044 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9045 * we've truncated everything except the last little bit, and can do
9046 * btrfs_truncate_block and then update the disk_i_size.
9048 if (ret == NEED_TRUNCATE_BLOCK) {
9049 btrfs_end_transaction(trans);
9050 btrfs_btree_balance_dirty(fs_info);
9052 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9055 trans = btrfs_start_transaction(root, 1);
9056 if (IS_ERR(trans)) {
9057 ret = PTR_ERR(trans);
9060 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9066 trans->block_rsv = &fs_info->trans_block_rsv;
9067 ret2 = btrfs_update_inode(trans, root, inode);
9071 ret2 = btrfs_end_transaction(trans);
9074 btrfs_btree_balance_dirty(fs_info);
9077 btrfs_free_block_rsv(fs_info, rsv);
9083 * create a new subvolume directory/inode (helper for the ioctl).
9085 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9086 struct btrfs_root *new_root,
9087 struct btrfs_root *parent_root,
9090 struct inode *inode;
9094 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9095 new_dirid, new_dirid,
9096 S_IFDIR | (~current_umask() & S_IRWXUGO),
9099 return PTR_ERR(inode);
9100 inode->i_op = &btrfs_dir_inode_operations;
9101 inode->i_fop = &btrfs_dir_file_operations;
9103 set_nlink(inode, 1);
9104 btrfs_i_size_write(BTRFS_I(inode), 0);
9105 unlock_new_inode(inode);
9107 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9109 btrfs_err(new_root->fs_info,
9110 "error inheriting subvolume %llu properties: %d",
9111 new_root->root_key.objectid, err);
9113 err = btrfs_update_inode(trans, new_root, inode);
9119 struct inode *btrfs_alloc_inode(struct super_block *sb)
9121 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9122 struct btrfs_inode *ei;
9123 struct inode *inode;
9125 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9132 ei->last_sub_trans = 0;
9133 ei->logged_trans = 0;
9134 ei->delalloc_bytes = 0;
9135 ei->new_delalloc_bytes = 0;
9136 ei->defrag_bytes = 0;
9137 ei->disk_i_size = 0;
9140 ei->index_cnt = (u64)-1;
9142 ei->last_unlink_trans = 0;
9143 ei->last_link_trans = 0;
9144 ei->last_log_commit = 0;
9146 spin_lock_init(&ei->lock);
9147 ei->outstanding_extents = 0;
9148 if (sb->s_magic != BTRFS_TEST_MAGIC)
9149 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9150 BTRFS_BLOCK_RSV_DELALLOC);
9151 ei->runtime_flags = 0;
9152 ei->prop_compress = BTRFS_COMPRESS_NONE;
9153 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9155 ei->delayed_node = NULL;
9157 ei->i_otime.tv_sec = 0;
9158 ei->i_otime.tv_nsec = 0;
9160 inode = &ei->vfs_inode;
9161 extent_map_tree_init(&ei->extent_tree);
9162 extent_io_tree_init(&ei->io_tree, inode);
9163 extent_io_tree_init(&ei->io_failure_tree, inode);
9164 ei->io_tree.track_uptodate = 1;
9165 ei->io_failure_tree.track_uptodate = 1;
9166 atomic_set(&ei->sync_writers, 0);
9167 mutex_init(&ei->log_mutex);
9168 mutex_init(&ei->delalloc_mutex);
9169 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9170 INIT_LIST_HEAD(&ei->delalloc_inodes);
9171 INIT_LIST_HEAD(&ei->delayed_iput);
9172 RB_CLEAR_NODE(&ei->rb_node);
9173 init_rwsem(&ei->dio_sem);
9178 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9179 void btrfs_test_destroy_inode(struct inode *inode)
9181 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9182 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9186 static void btrfs_i_callback(struct rcu_head *head)
9188 struct inode *inode = container_of(head, struct inode, i_rcu);
9189 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9192 void btrfs_destroy_inode(struct inode *inode)
9194 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9195 struct btrfs_ordered_extent *ordered;
9196 struct btrfs_root *root = BTRFS_I(inode)->root;
9198 WARN_ON(!hlist_empty(&inode->i_dentry));
9199 WARN_ON(inode->i_data.nrpages);
9200 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9201 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9202 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9203 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9204 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9205 WARN_ON(BTRFS_I(inode)->csum_bytes);
9206 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9209 * This can happen where we create an inode, but somebody else also
9210 * created the same inode and we need to destroy the one we already
9217 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9222 "found ordered extent %llu %llu on inode cleanup",
9223 ordered->file_offset, ordered->len);
9224 btrfs_remove_ordered_extent(inode, ordered);
9225 btrfs_put_ordered_extent(ordered);
9226 btrfs_put_ordered_extent(ordered);
9229 btrfs_qgroup_check_reserved_leak(inode);
9230 inode_tree_del(inode);
9231 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9233 call_rcu(&inode->i_rcu, btrfs_i_callback);
9236 int btrfs_drop_inode(struct inode *inode)
9238 struct btrfs_root *root = BTRFS_I(inode)->root;
9243 /* the snap/subvol tree is on deleting */
9244 if (btrfs_root_refs(&root->root_item) == 0)
9247 return generic_drop_inode(inode);
9250 static void init_once(void *foo)
9252 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9254 inode_init_once(&ei->vfs_inode);
9257 void __cold btrfs_destroy_cachep(void)
9260 * Make sure all delayed rcu free inodes are flushed before we
9264 kmem_cache_destroy(btrfs_inode_cachep);
9265 kmem_cache_destroy(btrfs_trans_handle_cachep);
9266 kmem_cache_destroy(btrfs_path_cachep);
9267 kmem_cache_destroy(btrfs_free_space_cachep);
9270 int __init btrfs_init_cachep(void)
9272 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9273 sizeof(struct btrfs_inode), 0,
9274 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9276 if (!btrfs_inode_cachep)
9279 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9280 sizeof(struct btrfs_trans_handle), 0,
9281 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9282 if (!btrfs_trans_handle_cachep)
9285 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9286 sizeof(struct btrfs_path), 0,
9287 SLAB_MEM_SPREAD, NULL);
9288 if (!btrfs_path_cachep)
9291 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9292 sizeof(struct btrfs_free_space), 0,
9293 SLAB_MEM_SPREAD, NULL);
9294 if (!btrfs_free_space_cachep)
9299 btrfs_destroy_cachep();
9303 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9304 u32 request_mask, unsigned int flags)
9307 struct inode *inode = d_inode(path->dentry);
9308 u32 blocksize = inode->i_sb->s_blocksize;
9309 u32 bi_flags = BTRFS_I(inode)->flags;
9311 stat->result_mask |= STATX_BTIME;
9312 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9313 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9314 if (bi_flags & BTRFS_INODE_APPEND)
9315 stat->attributes |= STATX_ATTR_APPEND;
9316 if (bi_flags & BTRFS_INODE_COMPRESS)
9317 stat->attributes |= STATX_ATTR_COMPRESSED;
9318 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9319 stat->attributes |= STATX_ATTR_IMMUTABLE;
9320 if (bi_flags & BTRFS_INODE_NODUMP)
9321 stat->attributes |= STATX_ATTR_NODUMP;
9323 stat->attributes_mask |= (STATX_ATTR_APPEND |
9324 STATX_ATTR_COMPRESSED |
9325 STATX_ATTR_IMMUTABLE |
9328 generic_fillattr(inode, stat);
9329 stat->dev = BTRFS_I(inode)->root->anon_dev;
9331 spin_lock(&BTRFS_I(inode)->lock);
9332 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9333 spin_unlock(&BTRFS_I(inode)->lock);
9334 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9335 ALIGN(delalloc_bytes, blocksize)) >> 9;
9339 static int btrfs_rename_exchange(struct inode *old_dir,
9340 struct dentry *old_dentry,
9341 struct inode *new_dir,
9342 struct dentry *new_dentry)
9344 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9345 struct btrfs_trans_handle *trans;
9346 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9347 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9348 struct inode *new_inode = new_dentry->d_inode;
9349 struct inode *old_inode = old_dentry->d_inode;
9350 struct timespec64 ctime = current_time(old_inode);
9351 struct dentry *parent;
9352 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9353 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9358 bool root_log_pinned = false;
9359 bool dest_log_pinned = false;
9360 struct btrfs_log_ctx ctx_root;
9361 struct btrfs_log_ctx ctx_dest;
9362 bool sync_log_root = false;
9363 bool sync_log_dest = false;
9364 bool commit_transaction = false;
9366 /* we only allow rename subvolume link between subvolumes */
9367 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9370 btrfs_init_log_ctx(&ctx_root, old_inode);
9371 btrfs_init_log_ctx(&ctx_dest, new_inode);
9373 /* close the race window with snapshot create/destroy ioctl */
9374 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9375 down_read(&fs_info->subvol_sem);
9376 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9377 down_read(&fs_info->subvol_sem);
9380 * We want to reserve the absolute worst case amount of items. So if
9381 * both inodes are subvols and we need to unlink them then that would
9382 * require 4 item modifications, but if they are both normal inodes it
9383 * would require 5 item modifications, so we'll assume their normal
9384 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9385 * should cover the worst case number of items we'll modify.
9387 trans = btrfs_start_transaction(root, 12);
9388 if (IS_ERR(trans)) {
9389 ret = PTR_ERR(trans);
9394 * We need to find a free sequence number both in the source and
9395 * in the destination directory for the exchange.
9397 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9400 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9404 BTRFS_I(old_inode)->dir_index = 0ULL;
9405 BTRFS_I(new_inode)->dir_index = 0ULL;
9407 /* Reference for the source. */
9408 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9409 /* force full log commit if subvolume involved. */
9410 btrfs_set_log_full_commit(fs_info, trans);
9412 btrfs_pin_log_trans(root);
9413 root_log_pinned = true;
9414 ret = btrfs_insert_inode_ref(trans, dest,
9415 new_dentry->d_name.name,
9416 new_dentry->d_name.len,
9418 btrfs_ino(BTRFS_I(new_dir)),
9424 /* And now for the dest. */
9425 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9426 /* force full log commit if subvolume involved. */
9427 btrfs_set_log_full_commit(fs_info, trans);
9429 btrfs_pin_log_trans(dest);
9430 dest_log_pinned = true;
9431 ret = btrfs_insert_inode_ref(trans, root,
9432 old_dentry->d_name.name,
9433 old_dentry->d_name.len,
9435 btrfs_ino(BTRFS_I(old_dir)),
9441 /* Update inode version and ctime/mtime. */
9442 inode_inc_iversion(old_dir);
9443 inode_inc_iversion(new_dir);
9444 inode_inc_iversion(old_inode);
9445 inode_inc_iversion(new_inode);
9446 old_dir->i_ctime = old_dir->i_mtime = ctime;
9447 new_dir->i_ctime = new_dir->i_mtime = ctime;
9448 old_inode->i_ctime = ctime;
9449 new_inode->i_ctime = ctime;
9451 if (old_dentry->d_parent != new_dentry->d_parent) {
9452 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9453 BTRFS_I(old_inode), 1);
9454 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9455 BTRFS_I(new_inode), 1);
9458 /* src is a subvolume */
9459 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9460 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9461 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9462 old_dentry->d_name.name,
9463 old_dentry->d_name.len);
9464 } else { /* src is an inode */
9465 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9466 BTRFS_I(old_dentry->d_inode),
9467 old_dentry->d_name.name,
9468 old_dentry->d_name.len);
9470 ret = btrfs_update_inode(trans, root, old_inode);
9473 btrfs_abort_transaction(trans, ret);
9477 /* dest is a subvolume */
9478 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9479 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9480 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9481 new_dentry->d_name.name,
9482 new_dentry->d_name.len);
9483 } else { /* dest is an inode */
9484 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9485 BTRFS_I(new_dentry->d_inode),
9486 new_dentry->d_name.name,
9487 new_dentry->d_name.len);
9489 ret = btrfs_update_inode(trans, dest, new_inode);
9492 btrfs_abort_transaction(trans, ret);
9496 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9497 new_dentry->d_name.name,
9498 new_dentry->d_name.len, 0, old_idx);
9500 btrfs_abort_transaction(trans, ret);
9504 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9505 old_dentry->d_name.name,
9506 old_dentry->d_name.len, 0, new_idx);
9508 btrfs_abort_transaction(trans, ret);
9512 if (old_inode->i_nlink == 1)
9513 BTRFS_I(old_inode)->dir_index = old_idx;
9514 if (new_inode->i_nlink == 1)
9515 BTRFS_I(new_inode)->dir_index = new_idx;
9517 if (root_log_pinned) {
9518 parent = new_dentry->d_parent;
9519 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9520 BTRFS_I(old_dir), parent,
9522 if (ret == BTRFS_NEED_LOG_SYNC)
9523 sync_log_root = true;
9524 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9525 commit_transaction = true;
9527 btrfs_end_log_trans(root);
9528 root_log_pinned = false;
9530 if (dest_log_pinned) {
9531 if (!commit_transaction) {
9532 parent = old_dentry->d_parent;
9533 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9534 BTRFS_I(new_dir), parent,
9536 if (ret == BTRFS_NEED_LOG_SYNC)
9537 sync_log_dest = true;
9538 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9539 commit_transaction = true;
9542 btrfs_end_log_trans(dest);
9543 dest_log_pinned = false;
9547 * If we have pinned a log and an error happened, we unpin tasks
9548 * trying to sync the log and force them to fallback to a transaction
9549 * commit if the log currently contains any of the inodes involved in
9550 * this rename operation (to ensure we do not persist a log with an
9551 * inconsistent state for any of these inodes or leading to any
9552 * inconsistencies when replayed). If the transaction was aborted, the
9553 * abortion reason is propagated to userspace when attempting to commit
9554 * the transaction. If the log does not contain any of these inodes, we
9555 * allow the tasks to sync it.
9557 if (ret && (root_log_pinned || dest_log_pinned)) {
9558 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9559 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9560 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9562 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9563 btrfs_set_log_full_commit(fs_info, trans);
9565 if (root_log_pinned) {
9566 btrfs_end_log_trans(root);
9567 root_log_pinned = false;
9569 if (dest_log_pinned) {
9570 btrfs_end_log_trans(dest);
9571 dest_log_pinned = false;
9574 if (!ret && sync_log_root && !commit_transaction) {
9575 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9578 commit_transaction = true;
9580 if (!ret && sync_log_dest && !commit_transaction) {
9581 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9584 commit_transaction = true;
9586 if (commit_transaction) {
9587 ret = btrfs_commit_transaction(trans);
9591 ret2 = btrfs_end_transaction(trans);
9592 ret = ret ? ret : ret2;
9595 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9596 up_read(&fs_info->subvol_sem);
9597 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9598 up_read(&fs_info->subvol_sem);
9603 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9604 struct btrfs_root *root,
9606 struct dentry *dentry)
9609 struct inode *inode;
9613 ret = btrfs_find_free_ino(root, &objectid);
9617 inode = btrfs_new_inode(trans, root, dir,
9618 dentry->d_name.name,
9620 btrfs_ino(BTRFS_I(dir)),
9622 S_IFCHR | WHITEOUT_MODE,
9625 if (IS_ERR(inode)) {
9626 ret = PTR_ERR(inode);
9630 inode->i_op = &btrfs_special_inode_operations;
9631 init_special_inode(inode, inode->i_mode,
9634 ret = btrfs_init_inode_security(trans, inode, dir,
9639 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9640 BTRFS_I(inode), 0, index);
9644 ret = btrfs_update_inode(trans, root, inode);
9646 unlock_new_inode(inode);
9648 inode_dec_link_count(inode);
9654 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9655 struct inode *new_dir, struct dentry *new_dentry,
9658 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9659 struct btrfs_trans_handle *trans;
9660 unsigned int trans_num_items;
9661 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9662 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9663 struct inode *new_inode = d_inode(new_dentry);
9664 struct inode *old_inode = d_inode(old_dentry);
9668 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9669 bool log_pinned = false;
9670 struct btrfs_log_ctx ctx;
9671 bool sync_log = false;
9672 bool commit_transaction = false;
9674 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9677 /* we only allow rename subvolume link between subvolumes */
9678 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9681 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9682 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9685 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9686 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9690 /* check for collisions, even if the name isn't there */
9691 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9692 new_dentry->d_name.name,
9693 new_dentry->d_name.len);
9696 if (ret == -EEXIST) {
9698 * eexist without a new_inode */
9699 if (WARN_ON(!new_inode)) {
9703 /* maybe -EOVERFLOW */
9710 * we're using rename to replace one file with another. Start IO on it
9711 * now so we don't add too much work to the end of the transaction
9713 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9714 filemap_flush(old_inode->i_mapping);
9716 /* close the racy window with snapshot create/destroy ioctl */
9717 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9718 down_read(&fs_info->subvol_sem);
9720 * We want to reserve the absolute worst case amount of items. So if
9721 * both inodes are subvols and we need to unlink them then that would
9722 * require 4 item modifications, but if they are both normal inodes it
9723 * would require 5 item modifications, so we'll assume they are normal
9724 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9725 * should cover the worst case number of items we'll modify.
9726 * If our rename has the whiteout flag, we need more 5 units for the
9727 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9728 * when selinux is enabled).
9730 trans_num_items = 11;
9731 if (flags & RENAME_WHITEOUT)
9732 trans_num_items += 5;
9733 trans = btrfs_start_transaction(root, trans_num_items);
9734 if (IS_ERR(trans)) {
9735 ret = PTR_ERR(trans);
9740 btrfs_record_root_in_trans(trans, dest);
9742 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9746 BTRFS_I(old_inode)->dir_index = 0ULL;
9747 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9748 /* force full log commit if subvolume involved. */
9749 btrfs_set_log_full_commit(fs_info, trans);
9751 btrfs_pin_log_trans(root);
9753 ret = btrfs_insert_inode_ref(trans, dest,
9754 new_dentry->d_name.name,
9755 new_dentry->d_name.len,
9757 btrfs_ino(BTRFS_I(new_dir)), index);
9762 inode_inc_iversion(old_dir);
9763 inode_inc_iversion(new_dir);
9764 inode_inc_iversion(old_inode);
9765 old_dir->i_ctime = old_dir->i_mtime =
9766 new_dir->i_ctime = new_dir->i_mtime =
9767 old_inode->i_ctime = current_time(old_dir);
9769 if (old_dentry->d_parent != new_dentry->d_parent)
9770 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9771 BTRFS_I(old_inode), 1);
9773 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9774 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9775 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9776 old_dentry->d_name.name,
9777 old_dentry->d_name.len);
9779 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9780 BTRFS_I(d_inode(old_dentry)),
9781 old_dentry->d_name.name,
9782 old_dentry->d_name.len);
9784 ret = btrfs_update_inode(trans, root, old_inode);
9787 btrfs_abort_transaction(trans, ret);
9792 inode_inc_iversion(new_inode);
9793 new_inode->i_ctime = current_time(new_inode);
9794 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9795 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9796 root_objectid = BTRFS_I(new_inode)->location.objectid;
9797 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9798 new_dentry->d_name.name,
9799 new_dentry->d_name.len);
9800 BUG_ON(new_inode->i_nlink == 0);
9802 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9803 BTRFS_I(d_inode(new_dentry)),
9804 new_dentry->d_name.name,
9805 new_dentry->d_name.len);
9807 if (!ret && new_inode->i_nlink == 0)
9808 ret = btrfs_orphan_add(trans,
9809 BTRFS_I(d_inode(new_dentry)));
9811 btrfs_abort_transaction(trans, ret);
9816 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9817 new_dentry->d_name.name,
9818 new_dentry->d_name.len, 0, index);
9820 btrfs_abort_transaction(trans, ret);
9824 if (old_inode->i_nlink == 1)
9825 BTRFS_I(old_inode)->dir_index = index;
9828 struct dentry *parent = new_dentry->d_parent;
9830 btrfs_init_log_ctx(&ctx, old_inode);
9831 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9832 BTRFS_I(old_dir), parent,
9834 if (ret == BTRFS_NEED_LOG_SYNC)
9836 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9837 commit_transaction = true;
9839 btrfs_end_log_trans(root);
9843 if (flags & RENAME_WHITEOUT) {
9844 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9848 btrfs_abort_transaction(trans, ret);
9854 * If we have pinned the log and an error happened, we unpin tasks
9855 * trying to sync the log and force them to fallback to a transaction
9856 * commit if the log currently contains any of the inodes involved in
9857 * this rename operation (to ensure we do not persist a log with an
9858 * inconsistent state for any of these inodes or leading to any
9859 * inconsistencies when replayed). If the transaction was aborted, the
9860 * abortion reason is propagated to userspace when attempting to commit
9861 * the transaction. If the log does not contain any of these inodes, we
9862 * allow the tasks to sync it.
9864 if (ret && log_pinned) {
9865 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9866 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9867 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9869 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9870 btrfs_set_log_full_commit(fs_info, trans);
9872 btrfs_end_log_trans(root);
9875 if (!ret && sync_log) {
9876 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9878 commit_transaction = true;
9880 if (commit_transaction) {
9881 ret = btrfs_commit_transaction(trans);
9885 ret2 = btrfs_end_transaction(trans);
9886 ret = ret ? ret : ret2;
9889 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9890 up_read(&fs_info->subvol_sem);
9895 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9896 struct inode *new_dir, struct dentry *new_dentry,
9899 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9902 if (flags & RENAME_EXCHANGE)
9903 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9906 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9909 struct btrfs_delalloc_work {
9910 struct inode *inode;
9911 struct completion completion;
9912 struct list_head list;
9913 struct btrfs_work work;
9916 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9918 struct btrfs_delalloc_work *delalloc_work;
9919 struct inode *inode;
9921 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9923 inode = delalloc_work->inode;
9924 filemap_flush(inode->i_mapping);
9925 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9926 &BTRFS_I(inode)->runtime_flags))
9927 filemap_flush(inode->i_mapping);
9930 complete(&delalloc_work->completion);
9933 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9935 struct btrfs_delalloc_work *work;
9937 work = kmalloc(sizeof(*work), GFP_NOFS);
9941 init_completion(&work->completion);
9942 INIT_LIST_HEAD(&work->list);
9943 work->inode = inode;
9944 WARN_ON_ONCE(!inode);
9945 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9946 btrfs_run_delalloc_work, NULL, NULL);
9952 * some fairly slow code that needs optimization. This walks the list
9953 * of all the inodes with pending delalloc and forces them to disk.
9955 static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot)
9957 struct btrfs_inode *binode;
9958 struct inode *inode;
9959 struct btrfs_delalloc_work *work, *next;
9960 struct list_head works;
9961 struct list_head splice;
9964 INIT_LIST_HEAD(&works);
9965 INIT_LIST_HEAD(&splice);
9967 mutex_lock(&root->delalloc_mutex);
9968 spin_lock(&root->delalloc_lock);
9969 list_splice_init(&root->delalloc_inodes, &splice);
9970 while (!list_empty(&splice)) {
9971 binode = list_entry(splice.next, struct btrfs_inode,
9974 list_move_tail(&binode->delalloc_inodes,
9975 &root->delalloc_inodes);
9976 inode = igrab(&binode->vfs_inode);
9978 cond_resched_lock(&root->delalloc_lock);
9981 spin_unlock(&root->delalloc_lock);
9984 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9985 &binode->runtime_flags);
9986 work = btrfs_alloc_delalloc_work(inode);
9992 list_add_tail(&work->list, &works);
9993 btrfs_queue_work(root->fs_info->flush_workers,
9996 if (nr != -1 && ret >= nr)
9999 spin_lock(&root->delalloc_lock);
10001 spin_unlock(&root->delalloc_lock);
10004 list_for_each_entry_safe(work, next, &works, list) {
10005 list_del_init(&work->list);
10006 wait_for_completion(&work->completion);
10010 if (!list_empty(&splice)) {
10011 spin_lock(&root->delalloc_lock);
10012 list_splice_tail(&splice, &root->delalloc_inodes);
10013 spin_unlock(&root->delalloc_lock);
10015 mutex_unlock(&root->delalloc_mutex);
10019 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
10021 struct btrfs_fs_info *fs_info = root->fs_info;
10024 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10027 ret = start_delalloc_inodes(root, -1, true);
10033 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10035 struct btrfs_root *root;
10036 struct list_head splice;
10039 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10042 INIT_LIST_HEAD(&splice);
10044 mutex_lock(&fs_info->delalloc_root_mutex);
10045 spin_lock(&fs_info->delalloc_root_lock);
10046 list_splice_init(&fs_info->delalloc_roots, &splice);
10047 while (!list_empty(&splice) && nr) {
10048 root = list_first_entry(&splice, struct btrfs_root,
10050 root = btrfs_grab_fs_root(root);
10052 list_move_tail(&root->delalloc_root,
10053 &fs_info->delalloc_roots);
10054 spin_unlock(&fs_info->delalloc_root_lock);
10056 ret = start_delalloc_inodes(root, nr, false);
10057 btrfs_put_fs_root(root);
10065 spin_lock(&fs_info->delalloc_root_lock);
10067 spin_unlock(&fs_info->delalloc_root_lock);
10071 if (!list_empty(&splice)) {
10072 spin_lock(&fs_info->delalloc_root_lock);
10073 list_splice_tail(&splice, &fs_info->delalloc_roots);
10074 spin_unlock(&fs_info->delalloc_root_lock);
10076 mutex_unlock(&fs_info->delalloc_root_mutex);
10080 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10081 const char *symname)
10083 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10084 struct btrfs_trans_handle *trans;
10085 struct btrfs_root *root = BTRFS_I(dir)->root;
10086 struct btrfs_path *path;
10087 struct btrfs_key key;
10088 struct inode *inode = NULL;
10095 struct btrfs_file_extent_item *ei;
10096 struct extent_buffer *leaf;
10098 name_len = strlen(symname);
10099 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10100 return -ENAMETOOLONG;
10103 * 2 items for inode item and ref
10104 * 2 items for dir items
10105 * 1 item for updating parent inode item
10106 * 1 item for the inline extent item
10107 * 1 item for xattr if selinux is on
10109 trans = btrfs_start_transaction(root, 7);
10111 return PTR_ERR(trans);
10113 err = btrfs_find_free_ino(root, &objectid);
10117 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10118 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10119 objectid, S_IFLNK|S_IRWXUGO, &index);
10120 if (IS_ERR(inode)) {
10121 err = PTR_ERR(inode);
10127 * If the active LSM wants to access the inode during
10128 * d_instantiate it needs these. Smack checks to see
10129 * if the filesystem supports xattrs by looking at the
10132 inode->i_fop = &btrfs_file_operations;
10133 inode->i_op = &btrfs_file_inode_operations;
10134 inode->i_mapping->a_ops = &btrfs_aops;
10135 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10137 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10141 path = btrfs_alloc_path();
10146 key.objectid = btrfs_ino(BTRFS_I(inode));
10148 key.type = BTRFS_EXTENT_DATA_KEY;
10149 datasize = btrfs_file_extent_calc_inline_size(name_len);
10150 err = btrfs_insert_empty_item(trans, root, path, &key,
10153 btrfs_free_path(path);
10156 leaf = path->nodes[0];
10157 ei = btrfs_item_ptr(leaf, path->slots[0],
10158 struct btrfs_file_extent_item);
10159 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10160 btrfs_set_file_extent_type(leaf, ei,
10161 BTRFS_FILE_EXTENT_INLINE);
10162 btrfs_set_file_extent_encryption(leaf, ei, 0);
10163 btrfs_set_file_extent_compression(leaf, ei, 0);
10164 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10165 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10167 ptr = btrfs_file_extent_inline_start(ei);
10168 write_extent_buffer(leaf, symname, ptr, name_len);
10169 btrfs_mark_buffer_dirty(leaf);
10170 btrfs_free_path(path);
10172 inode->i_op = &btrfs_symlink_inode_operations;
10173 inode_nohighmem(inode);
10174 inode->i_mapping->a_ops = &btrfs_aops;
10175 inode_set_bytes(inode, name_len);
10176 btrfs_i_size_write(BTRFS_I(inode), name_len);
10177 err = btrfs_update_inode(trans, root, inode);
10179 * Last step, add directory indexes for our symlink inode. This is the
10180 * last step to avoid extra cleanup of these indexes if an error happens
10184 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10185 BTRFS_I(inode), 0, index);
10189 d_instantiate_new(dentry, inode);
10192 btrfs_end_transaction(trans);
10193 if (err && inode) {
10194 inode_dec_link_count(inode);
10195 discard_new_inode(inode);
10197 btrfs_btree_balance_dirty(fs_info);
10201 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10202 u64 start, u64 num_bytes, u64 min_size,
10203 loff_t actual_len, u64 *alloc_hint,
10204 struct btrfs_trans_handle *trans)
10206 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10207 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10208 struct extent_map *em;
10209 struct btrfs_root *root = BTRFS_I(inode)->root;
10210 struct btrfs_key ins;
10211 u64 cur_offset = start;
10214 u64 last_alloc = (u64)-1;
10216 bool own_trans = true;
10217 u64 end = start + num_bytes - 1;
10221 while (num_bytes > 0) {
10223 trans = btrfs_start_transaction(root, 3);
10224 if (IS_ERR(trans)) {
10225 ret = PTR_ERR(trans);
10230 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10231 cur_bytes = max(cur_bytes, min_size);
10233 * If we are severely fragmented we could end up with really
10234 * small allocations, so if the allocator is returning small
10235 * chunks lets make its job easier by only searching for those
10238 cur_bytes = min(cur_bytes, last_alloc);
10239 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10240 min_size, 0, *alloc_hint, &ins, 1, 0);
10243 btrfs_end_transaction(trans);
10246 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10248 last_alloc = ins.offset;
10249 ret = insert_reserved_file_extent(trans, inode,
10250 cur_offset, ins.objectid,
10251 ins.offset, ins.offset,
10252 ins.offset, 0, 0, 0,
10253 BTRFS_FILE_EXTENT_PREALLOC);
10255 btrfs_free_reserved_extent(fs_info, ins.objectid,
10257 btrfs_abort_transaction(trans, ret);
10259 btrfs_end_transaction(trans);
10263 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10264 cur_offset + ins.offset -1, 0);
10266 em = alloc_extent_map();
10268 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10269 &BTRFS_I(inode)->runtime_flags);
10273 em->start = cur_offset;
10274 em->orig_start = cur_offset;
10275 em->len = ins.offset;
10276 em->block_start = ins.objectid;
10277 em->block_len = ins.offset;
10278 em->orig_block_len = ins.offset;
10279 em->ram_bytes = ins.offset;
10280 em->bdev = fs_info->fs_devices->latest_bdev;
10281 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10282 em->generation = trans->transid;
10285 write_lock(&em_tree->lock);
10286 ret = add_extent_mapping(em_tree, em, 1);
10287 write_unlock(&em_tree->lock);
10288 if (ret != -EEXIST)
10290 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10291 cur_offset + ins.offset - 1,
10294 free_extent_map(em);
10296 num_bytes -= ins.offset;
10297 cur_offset += ins.offset;
10298 *alloc_hint = ins.objectid + ins.offset;
10300 inode_inc_iversion(inode);
10301 inode->i_ctime = current_time(inode);
10302 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10303 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10304 (actual_len > inode->i_size) &&
10305 (cur_offset > inode->i_size)) {
10306 if (cur_offset > actual_len)
10307 i_size = actual_len;
10309 i_size = cur_offset;
10310 i_size_write(inode, i_size);
10311 btrfs_ordered_update_i_size(inode, i_size, NULL);
10314 ret = btrfs_update_inode(trans, root, inode);
10317 btrfs_abort_transaction(trans, ret);
10319 btrfs_end_transaction(trans);
10324 btrfs_end_transaction(trans);
10326 if (cur_offset < end)
10327 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10328 end - cur_offset + 1);
10332 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10333 u64 start, u64 num_bytes, u64 min_size,
10334 loff_t actual_len, u64 *alloc_hint)
10336 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10337 min_size, actual_len, alloc_hint,
10341 int btrfs_prealloc_file_range_trans(struct inode *inode,
10342 struct btrfs_trans_handle *trans, int mode,
10343 u64 start, u64 num_bytes, u64 min_size,
10344 loff_t actual_len, u64 *alloc_hint)
10346 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10347 min_size, actual_len, alloc_hint, trans);
10350 static int btrfs_set_page_dirty(struct page *page)
10352 return __set_page_dirty_nobuffers(page);
10355 static int btrfs_permission(struct inode *inode, int mask)
10357 struct btrfs_root *root = BTRFS_I(inode)->root;
10358 umode_t mode = inode->i_mode;
10360 if (mask & MAY_WRITE &&
10361 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10362 if (btrfs_root_readonly(root))
10364 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10367 return generic_permission(inode, mask);
10370 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10372 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10373 struct btrfs_trans_handle *trans;
10374 struct btrfs_root *root = BTRFS_I(dir)->root;
10375 struct inode *inode = NULL;
10381 * 5 units required for adding orphan entry
10383 trans = btrfs_start_transaction(root, 5);
10385 return PTR_ERR(trans);
10387 ret = btrfs_find_free_ino(root, &objectid);
10391 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10392 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10393 if (IS_ERR(inode)) {
10394 ret = PTR_ERR(inode);
10399 inode->i_fop = &btrfs_file_operations;
10400 inode->i_op = &btrfs_file_inode_operations;
10402 inode->i_mapping->a_ops = &btrfs_aops;
10403 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10405 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10409 ret = btrfs_update_inode(trans, root, inode);
10412 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10417 * We set number of links to 0 in btrfs_new_inode(), and here we set
10418 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10421 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10423 set_nlink(inode, 1);
10424 d_tmpfile(dentry, inode);
10425 unlock_new_inode(inode);
10426 mark_inode_dirty(inode);
10428 btrfs_end_transaction(trans);
10430 discard_new_inode(inode);
10431 btrfs_btree_balance_dirty(fs_info);
10435 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10437 struct inode *inode = tree->private_data;
10438 unsigned long index = start >> PAGE_SHIFT;
10439 unsigned long end_index = end >> PAGE_SHIFT;
10442 while (index <= end_index) {
10443 page = find_get_page(inode->i_mapping, index);
10444 ASSERT(page); /* Pages should be in the extent_io_tree */
10445 set_page_writeback(page);
10453 * Add an entry indicating a block group or device which is pinned by a
10454 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10455 * negative errno on failure.
10457 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10458 bool is_block_group)
10460 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10461 struct btrfs_swapfile_pin *sp, *entry;
10462 struct rb_node **p;
10463 struct rb_node *parent = NULL;
10465 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10470 sp->is_block_group = is_block_group;
10472 spin_lock(&fs_info->swapfile_pins_lock);
10473 p = &fs_info->swapfile_pins.rb_node;
10476 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10477 if (sp->ptr < entry->ptr ||
10478 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10479 p = &(*p)->rb_left;
10480 } else if (sp->ptr > entry->ptr ||
10481 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10482 p = &(*p)->rb_right;
10484 spin_unlock(&fs_info->swapfile_pins_lock);
10489 rb_link_node(&sp->node, parent, p);
10490 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10491 spin_unlock(&fs_info->swapfile_pins_lock);
10495 /* Free all of the entries pinned by this swapfile. */
10496 static void btrfs_free_swapfile_pins(struct inode *inode)
10498 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10499 struct btrfs_swapfile_pin *sp;
10500 struct rb_node *node, *next;
10502 spin_lock(&fs_info->swapfile_pins_lock);
10503 node = rb_first(&fs_info->swapfile_pins);
10505 next = rb_next(node);
10506 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10507 if (sp->inode == inode) {
10508 rb_erase(&sp->node, &fs_info->swapfile_pins);
10509 if (sp->is_block_group)
10510 btrfs_put_block_group(sp->ptr);
10515 spin_unlock(&fs_info->swapfile_pins_lock);
10518 struct btrfs_swap_info {
10524 unsigned long nr_pages;
10528 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10529 struct btrfs_swap_info *bsi)
10531 unsigned long nr_pages;
10532 u64 first_ppage, first_ppage_reported, next_ppage;
10535 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10536 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10537 PAGE_SIZE) >> PAGE_SHIFT;
10539 if (first_ppage >= next_ppage)
10541 nr_pages = next_ppage - first_ppage;
10543 first_ppage_reported = first_ppage;
10544 if (bsi->start == 0)
10545 first_ppage_reported++;
10546 if (bsi->lowest_ppage > first_ppage_reported)
10547 bsi->lowest_ppage = first_ppage_reported;
10548 if (bsi->highest_ppage < (next_ppage - 1))
10549 bsi->highest_ppage = next_ppage - 1;
10551 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10554 bsi->nr_extents += ret;
10555 bsi->nr_pages += nr_pages;
10559 static void btrfs_swap_deactivate(struct file *file)
10561 struct inode *inode = file_inode(file);
10563 btrfs_free_swapfile_pins(inode);
10564 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10567 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10570 struct inode *inode = file_inode(file);
10571 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10572 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10573 struct extent_state *cached_state = NULL;
10574 struct extent_map *em = NULL;
10575 struct btrfs_device *device = NULL;
10576 struct btrfs_swap_info bsi = {
10577 .lowest_ppage = (sector_t)-1ULL,
10584 * If the swap file was just created, make sure delalloc is done. If the
10585 * file changes again after this, the user is doing something stupid and
10586 * we don't really care.
10588 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10593 * The inode is locked, so these flags won't change after we check them.
10595 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10596 btrfs_warn(fs_info, "swapfile must not be compressed");
10599 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10600 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10603 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10604 btrfs_warn(fs_info, "swapfile must not be checksummed");
10609 * Balance or device remove/replace/resize can move stuff around from
10610 * under us. The EXCL_OP flag makes sure they aren't running/won't run
10611 * concurrently while we are mapping the swap extents, and
10612 * fs_info->swapfile_pins prevents them from running while the swap file
10613 * is active and moving the extents. Note that this also prevents a
10614 * concurrent device add which isn't actually necessary, but it's not
10615 * really worth the trouble to allow it.
10617 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
10618 btrfs_warn(fs_info,
10619 "cannot activate swapfile while exclusive operation is running");
10623 * Snapshots can create extents which require COW even if NODATACOW is
10624 * set. We use this counter to prevent snapshots. We must increment it
10625 * before walking the extents because we don't want a concurrent
10626 * snapshot to run after we've already checked the extents.
10628 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10630 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10632 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10634 while (start < isize) {
10635 u64 logical_block_start, physical_block_start;
10636 struct btrfs_block_group_cache *bg;
10637 u64 len = isize - start;
10639 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
10645 if (em->block_start == EXTENT_MAP_HOLE) {
10646 btrfs_warn(fs_info, "swapfile must not have holes");
10650 if (em->block_start == EXTENT_MAP_INLINE) {
10652 * It's unlikely we'll ever actually find ourselves
10653 * here, as a file small enough to fit inline won't be
10654 * big enough to store more than the swap header, but in
10655 * case something changes in the future, let's catch it
10656 * here rather than later.
10658 btrfs_warn(fs_info, "swapfile must not be inline");
10662 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10663 btrfs_warn(fs_info, "swapfile must not be compressed");
10668 logical_block_start = em->block_start + (start - em->start);
10669 len = min(len, em->len - (start - em->start));
10670 free_extent_map(em);
10673 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL);
10679 btrfs_warn(fs_info,
10680 "swapfile must not be copy-on-write");
10685 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10691 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10692 btrfs_warn(fs_info,
10693 "swapfile must have single data profile");
10698 if (device == NULL) {
10699 device = em->map_lookup->stripes[0].dev;
10700 ret = btrfs_add_swapfile_pin(inode, device, false);
10705 } else if (device != em->map_lookup->stripes[0].dev) {
10706 btrfs_warn(fs_info, "swapfile must be on one device");
10711 physical_block_start = (em->map_lookup->stripes[0].physical +
10712 (logical_block_start - em->start));
10713 len = min(len, em->len - (logical_block_start - em->start));
10714 free_extent_map(em);
10717 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10719 btrfs_warn(fs_info,
10720 "could not find block group containing swapfile");
10725 ret = btrfs_add_swapfile_pin(inode, bg, true);
10727 btrfs_put_block_group(bg);
10734 if (bsi.block_len &&
10735 bsi.block_start + bsi.block_len == physical_block_start) {
10736 bsi.block_len += len;
10738 if (bsi.block_len) {
10739 ret = btrfs_add_swap_extent(sis, &bsi);
10744 bsi.block_start = physical_block_start;
10745 bsi.block_len = len;
10752 ret = btrfs_add_swap_extent(sis, &bsi);
10755 if (!IS_ERR_OR_NULL(em))
10756 free_extent_map(em);
10758 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10761 btrfs_swap_deactivate(file);
10763 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
10769 sis->bdev = device->bdev;
10770 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10771 sis->max = bsi.nr_pages;
10772 sis->pages = bsi.nr_pages - 1;
10773 sis->highest_bit = bsi.nr_pages - 1;
10774 return bsi.nr_extents;
10777 static void btrfs_swap_deactivate(struct file *file)
10781 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10784 return -EOPNOTSUPP;
10788 static const struct inode_operations btrfs_dir_inode_operations = {
10789 .getattr = btrfs_getattr,
10790 .lookup = btrfs_lookup,
10791 .create = btrfs_create,
10792 .unlink = btrfs_unlink,
10793 .link = btrfs_link,
10794 .mkdir = btrfs_mkdir,
10795 .rmdir = btrfs_rmdir,
10796 .rename = btrfs_rename2,
10797 .symlink = btrfs_symlink,
10798 .setattr = btrfs_setattr,
10799 .mknod = btrfs_mknod,
10800 .listxattr = btrfs_listxattr,
10801 .permission = btrfs_permission,
10802 .get_acl = btrfs_get_acl,
10803 .set_acl = btrfs_set_acl,
10804 .update_time = btrfs_update_time,
10805 .tmpfile = btrfs_tmpfile,
10807 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10808 .lookup = btrfs_lookup,
10809 .permission = btrfs_permission,
10810 .update_time = btrfs_update_time,
10813 static const struct file_operations btrfs_dir_file_operations = {
10814 .llseek = generic_file_llseek,
10815 .read = generic_read_dir,
10816 .iterate_shared = btrfs_real_readdir,
10817 .open = btrfs_opendir,
10818 .unlocked_ioctl = btrfs_ioctl,
10819 #ifdef CONFIG_COMPAT
10820 .compat_ioctl = btrfs_compat_ioctl,
10822 .release = btrfs_release_file,
10823 .fsync = btrfs_sync_file,
10826 static const struct extent_io_ops btrfs_extent_io_ops = {
10827 /* mandatory callbacks */
10828 .submit_bio_hook = btrfs_submit_bio_hook,
10829 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10833 * btrfs doesn't support the bmap operation because swapfiles
10834 * use bmap to make a mapping of extents in the file. They assume
10835 * these extents won't change over the life of the file and they
10836 * use the bmap result to do IO directly to the drive.
10838 * the btrfs bmap call would return logical addresses that aren't
10839 * suitable for IO and they also will change frequently as COW
10840 * operations happen. So, swapfile + btrfs == corruption.
10842 * For now we're avoiding this by dropping bmap.
10844 static const struct address_space_operations btrfs_aops = {
10845 .readpage = btrfs_readpage,
10846 .writepage = btrfs_writepage,
10847 .writepages = btrfs_writepages,
10848 .readpages = btrfs_readpages,
10849 .direct_IO = btrfs_direct_IO,
10850 .invalidatepage = btrfs_invalidatepage,
10851 .releasepage = btrfs_releasepage,
10852 .set_page_dirty = btrfs_set_page_dirty,
10853 .error_remove_page = generic_error_remove_page,
10854 .swap_activate = btrfs_swap_activate,
10855 .swap_deactivate = btrfs_swap_deactivate,
10858 static const struct inode_operations btrfs_file_inode_operations = {
10859 .getattr = btrfs_getattr,
10860 .setattr = btrfs_setattr,
10861 .listxattr = btrfs_listxattr,
10862 .permission = btrfs_permission,
10863 .fiemap = btrfs_fiemap,
10864 .get_acl = btrfs_get_acl,
10865 .set_acl = btrfs_set_acl,
10866 .update_time = btrfs_update_time,
10868 static const struct inode_operations btrfs_special_inode_operations = {
10869 .getattr = btrfs_getattr,
10870 .setattr = btrfs_setattr,
10871 .permission = btrfs_permission,
10872 .listxattr = btrfs_listxattr,
10873 .get_acl = btrfs_get_acl,
10874 .set_acl = btrfs_set_acl,
10875 .update_time = btrfs_update_time,
10877 static const struct inode_operations btrfs_symlink_inode_operations = {
10878 .get_link = page_get_link,
10879 .getattr = btrfs_getattr,
10880 .setattr = btrfs_setattr,
10881 .permission = btrfs_permission,
10882 .listxattr = btrfs_listxattr,
10883 .update_time = btrfs_update_time,
10886 const struct dentry_operations btrfs_dentry_operations = {
10887 .d_delete = btrfs_dentry_delete,