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 <asm/unaligned.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "print-tree.h"
36 #include "ordered-data.h"
40 #include "compression.h"
42 #include "free-space-cache.h"
43 #include "inode-map.h"
49 struct btrfs_iget_args {
50 struct btrfs_key *location;
51 struct btrfs_root *root;
54 struct btrfs_dio_data {
56 u64 unsubmitted_oe_range_start;
57 u64 unsubmitted_oe_range_end;
61 static const struct inode_operations btrfs_dir_inode_operations;
62 static const struct inode_operations btrfs_symlink_inode_operations;
63 static const struct inode_operations btrfs_dir_ro_inode_operations;
64 static const struct inode_operations btrfs_special_inode_operations;
65 static const struct inode_operations btrfs_file_inode_operations;
66 static const struct address_space_operations btrfs_aops;
67 static const struct address_space_operations btrfs_symlink_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 the
114 * fill_delalloc() callback 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_root *root;
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 */
509 if (BTRFS_I(inode)->defrag_compress)
510 compress_type = BTRFS_I(inode)->defrag_compress;
511 else if (BTRFS_I(inode)->prop_compress)
512 compress_type = BTRFS_I(inode)->prop_compress;
515 * we need to call clear_page_dirty_for_io on each
516 * page in the range. Otherwise applications with the file
517 * mmap'd can wander in and change the page contents while
518 * we are compressing them.
520 * If the compression fails for any reason, we set the pages
521 * dirty again later on.
523 * Note that the remaining part is redirtied, the start pointer
524 * has moved, the end is the original one.
527 extent_range_clear_dirty_for_io(inode, start, end);
531 /* Compression level is applied here and only here */
532 ret = btrfs_compress_pages(
533 compress_type | (fs_info->compress_level << 4),
534 inode->i_mapping, start,
541 unsigned long offset = total_compressed &
543 struct page *page = pages[nr_pages - 1];
546 /* zero the tail end of the last page, we might be
547 * sending it down to disk
550 kaddr = kmap_atomic(page);
551 memset(kaddr + offset, 0,
553 kunmap_atomic(kaddr);
560 /* lets try to make an inline extent */
561 if (ret || total_in < actual_end) {
562 /* we didn't compress the entire range, try
563 * to make an uncompressed inline extent.
565 ret = cow_file_range_inline(inode, start, end, 0,
566 BTRFS_COMPRESS_NONE, NULL);
568 /* try making a compressed inline extent */
569 ret = cow_file_range_inline(inode, start, end,
571 compress_type, pages);
574 unsigned long clear_flags = EXTENT_DELALLOC |
575 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
576 EXTENT_DO_ACCOUNTING;
577 unsigned long page_error_op;
579 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
582 * inline extent creation worked or returned error,
583 * we don't need to create any more async work items.
584 * Unlock and free up our temp pages.
586 * We use DO_ACCOUNTING here because we need the
587 * delalloc_release_metadata to be done _after_ we drop
588 * our outstanding extent for clearing delalloc for this
591 extent_clear_unlock_delalloc(inode, start, end, end,
604 * we aren't doing an inline extent round the compressed size
605 * up to a block size boundary so the allocator does sane
608 total_compressed = ALIGN(total_compressed, blocksize);
611 * one last check to make sure the compression is really a
612 * win, compare the page count read with the blocks on disk,
613 * compression must free at least one sector size
615 total_in = ALIGN(total_in, PAGE_SIZE);
616 if (total_compressed + blocksize <= total_in) {
620 * The async work queues will take care of doing actual
621 * allocation on disk for these compressed pages, and
622 * will submit them to the elevator.
624 add_async_extent(async_cow, start, total_in,
625 total_compressed, pages, nr_pages,
628 if (start + total_in < end) {
639 * the compression code ran but failed to make things smaller,
640 * free any pages it allocated and our page pointer array
642 for (i = 0; i < nr_pages; i++) {
643 WARN_ON(pages[i]->mapping);
648 total_compressed = 0;
651 /* flag the file so we don't compress in the future */
652 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
653 !(BTRFS_I(inode)->prop_compress)) {
654 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
657 cleanup_and_bail_uncompressed:
659 * No compression, but we still need to write the pages in the file
660 * we've been given so far. redirty the locked page if it corresponds
661 * to our extent and set things up for the async work queue to run
662 * cow_file_range to do the normal delalloc dance.
664 if (page_offset(locked_page) >= start &&
665 page_offset(locked_page) <= end)
666 __set_page_dirty_nobuffers(locked_page);
667 /* unlocked later on in the async handlers */
670 extent_range_redirty_for_io(inode, start, end);
671 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
672 BTRFS_COMPRESS_NONE);
678 for (i = 0; i < nr_pages; i++) {
679 WARN_ON(pages[i]->mapping);
685 static void free_async_extent_pages(struct async_extent *async_extent)
689 if (!async_extent->pages)
692 for (i = 0; i < async_extent->nr_pages; i++) {
693 WARN_ON(async_extent->pages[i]->mapping);
694 put_page(async_extent->pages[i]);
696 kfree(async_extent->pages);
697 async_extent->nr_pages = 0;
698 async_extent->pages = NULL;
702 * phase two of compressed writeback. This is the ordered portion
703 * of the code, which only gets called in the order the work was
704 * queued. We walk all the async extents created by compress_file_range
705 * and send them down to the disk.
707 static noinline void submit_compressed_extents(struct inode *inode,
708 struct async_cow *async_cow)
710 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
711 struct async_extent *async_extent;
713 struct btrfs_key ins;
714 struct extent_map *em;
715 struct btrfs_root *root = BTRFS_I(inode)->root;
716 struct extent_io_tree *io_tree;
720 while (!list_empty(&async_cow->extents)) {
721 async_extent = list_entry(async_cow->extents.next,
722 struct async_extent, list);
723 list_del(&async_extent->list);
725 io_tree = &BTRFS_I(inode)->io_tree;
728 /* did the compression code fall back to uncompressed IO? */
729 if (!async_extent->pages) {
730 int page_started = 0;
731 unsigned long nr_written = 0;
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start +
735 async_extent->ram_size - 1);
737 /* allocate blocks */
738 ret = cow_file_range(inode, async_cow->locked_page,
740 async_extent->start +
741 async_extent->ram_size - 1,
742 async_extent->start +
743 async_extent->ram_size - 1,
744 &page_started, &nr_written, 0,
750 * if page_started, cow_file_range inserted an
751 * inline extent and took care of all the unlocking
752 * and IO for us. Otherwise, we need to submit
753 * all those pages down to the drive.
755 if (!page_started && !ret)
756 extent_write_locked_range(inode,
758 async_extent->start +
759 async_extent->ram_size - 1,
762 unlock_page(async_cow->locked_page);
768 lock_extent(io_tree, async_extent->start,
769 async_extent->start + async_extent->ram_size - 1);
771 ret = btrfs_reserve_extent(root, async_extent->ram_size,
772 async_extent->compressed_size,
773 async_extent->compressed_size,
774 0, alloc_hint, &ins, 1, 1);
776 free_async_extent_pages(async_extent);
778 if (ret == -ENOSPC) {
779 unlock_extent(io_tree, async_extent->start,
780 async_extent->start +
781 async_extent->ram_size - 1);
784 * we need to redirty the pages if we decide to
785 * fallback to uncompressed IO, otherwise we
786 * will not submit these pages down to lower
789 extent_range_redirty_for_io(inode,
791 async_extent->start +
792 async_extent->ram_size - 1);
799 * here we're doing allocation and writeback of the
802 em = create_io_em(inode, async_extent->start,
803 async_extent->ram_size, /* len */
804 async_extent->start, /* orig_start */
805 ins.objectid, /* block_start */
806 ins.offset, /* block_len */
807 ins.offset, /* orig_block_len */
808 async_extent->ram_size, /* ram_bytes */
809 async_extent->compress_type,
810 BTRFS_ORDERED_COMPRESSED);
812 /* ret value is not necessary due to void function */
813 goto out_free_reserve;
816 ret = btrfs_add_ordered_extent_compress(inode,
819 async_extent->ram_size,
821 BTRFS_ORDERED_COMPRESSED,
822 async_extent->compress_type);
824 btrfs_drop_extent_cache(BTRFS_I(inode),
826 async_extent->start +
827 async_extent->ram_size - 1, 0);
828 goto out_free_reserve;
830 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
833 * clear dirty, set writeback and unlock the pages.
835 extent_clear_unlock_delalloc(inode, async_extent->start,
836 async_extent->start +
837 async_extent->ram_size - 1,
838 async_extent->start +
839 async_extent->ram_size - 1,
840 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
841 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
843 if (btrfs_submit_compressed_write(inode,
845 async_extent->ram_size,
847 ins.offset, async_extent->pages,
848 async_extent->nr_pages,
849 async_cow->write_flags)) {
850 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
851 struct page *p = async_extent->pages[0];
852 const u64 start = async_extent->start;
853 const u64 end = start + async_extent->ram_size - 1;
855 p->mapping = inode->i_mapping;
856 tree->ops->writepage_end_io_hook(p, start, end,
859 extent_clear_unlock_delalloc(inode, start, end, end,
863 free_async_extent_pages(async_extent);
865 alloc_hint = ins.objectid + ins.offset;
871 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
872 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
874 extent_clear_unlock_delalloc(inode, async_extent->start,
875 async_extent->start +
876 async_extent->ram_size - 1,
877 async_extent->start +
878 async_extent->ram_size - 1,
879 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
880 EXTENT_DELALLOC_NEW |
881 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
882 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
883 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
885 free_async_extent_pages(async_extent);
890 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
893 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
894 struct extent_map *em;
897 read_lock(&em_tree->lock);
898 em = search_extent_mapping(em_tree, start, num_bytes);
901 * if block start isn't an actual block number then find the
902 * first block in this inode and use that as a hint. If that
903 * block is also bogus then just don't worry about it.
905 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
907 em = search_extent_mapping(em_tree, 0, 0);
908 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
909 alloc_hint = em->block_start;
913 alloc_hint = em->block_start;
917 read_unlock(&em_tree->lock);
923 * when extent_io.c finds a delayed allocation range in the file,
924 * the call backs end up in this code. The basic idea is to
925 * allocate extents on disk for the range, and create ordered data structs
926 * in ram to track those extents.
928 * locked_page is the page that writepage had locked already. We use
929 * it to make sure we don't do extra locks or unlocks.
931 * *page_started is set to one if we unlock locked_page and do everything
932 * required to start IO on it. It may be clean and already done with
935 static noinline int cow_file_range(struct inode *inode,
936 struct page *locked_page,
937 u64 start, u64 end, u64 delalloc_end,
938 int *page_started, unsigned long *nr_written,
939 int unlock, struct btrfs_dedupe_hash *hash)
941 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
942 struct btrfs_root *root = BTRFS_I(inode)->root;
945 unsigned long ram_size;
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 struct btrfs_root *root;
1148 unsigned long nr_pages;
1150 async_cow = container_of(work, struct async_cow, work);
1152 root = async_cow->root;
1153 fs_info = root->fs_info;
1154 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1157 /* atomic_sub_return implies a barrier */
1158 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1160 cond_wake_up_nomb(&fs_info->async_submit_wait);
1162 if (async_cow->inode)
1163 submit_compressed_extents(async_cow->inode, async_cow);
1166 static noinline void async_cow_free(struct btrfs_work *work)
1168 struct async_cow *async_cow;
1169 async_cow = container_of(work, struct async_cow, work);
1170 if (async_cow->inode)
1171 btrfs_add_delayed_iput(async_cow->inode);
1175 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1176 u64 start, u64 end, int *page_started,
1177 unsigned long *nr_written,
1178 unsigned int write_flags)
1180 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1181 struct async_cow *async_cow;
1182 struct btrfs_root *root = BTRFS_I(inode)->root;
1183 unsigned long nr_pages;
1186 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1188 while (start < end) {
1189 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1190 BUG_ON(!async_cow); /* -ENOMEM */
1191 async_cow->inode = igrab(inode);
1192 async_cow->root = root;
1193 async_cow->locked_page = locked_page;
1194 async_cow->start = start;
1195 async_cow->write_flags = write_flags;
1197 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1198 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1201 cur_end = min(end, start + SZ_512K - 1);
1203 async_cow->end = cur_end;
1204 INIT_LIST_HEAD(&async_cow->extents);
1206 btrfs_init_work(&async_cow->work,
1207 btrfs_delalloc_helper,
1208 async_cow_start, async_cow_submit,
1211 nr_pages = (cur_end - start + PAGE_SIZE) >>
1213 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1215 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1217 *nr_written += nr_pages;
1218 start = cur_end + 1;
1224 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1225 u64 bytenr, u64 num_bytes)
1228 struct btrfs_ordered_sum *sums;
1231 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1232 bytenr + num_bytes - 1, &list, 0);
1233 if (ret == 0 && list_empty(&list))
1236 while (!list_empty(&list)) {
1237 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1238 list_del(&sums->list);
1247 * when nowcow writeback call back. This checks for snapshots or COW copies
1248 * of the extents that exist in the file, and COWs the file as required.
1250 * If no cow copies or snapshots exist, we write directly to the existing
1253 static noinline int run_delalloc_nocow(struct inode *inode,
1254 struct page *locked_page,
1255 u64 start, u64 end, int *page_started, int force,
1256 unsigned long *nr_written)
1258 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1259 struct btrfs_root *root = BTRFS_I(inode)->root;
1260 struct extent_buffer *leaf;
1261 struct btrfs_path *path;
1262 struct btrfs_file_extent_item *fi;
1263 struct btrfs_key found_key;
1264 struct extent_map *em;
1279 u64 ino = btrfs_ino(BTRFS_I(inode));
1281 path = btrfs_alloc_path();
1283 extent_clear_unlock_delalloc(inode, start, end, end,
1285 EXTENT_LOCKED | EXTENT_DELALLOC |
1286 EXTENT_DO_ACCOUNTING |
1287 EXTENT_DEFRAG, PAGE_UNLOCK |
1289 PAGE_SET_WRITEBACK |
1290 PAGE_END_WRITEBACK);
1294 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1296 cow_start = (u64)-1;
1299 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1303 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1304 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &found_key,
1306 path->slots[0] - 1);
1307 if (found_key.objectid == ino &&
1308 found_key.type == BTRFS_EXTENT_DATA_KEY)
1313 leaf = path->nodes[0];
1314 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1315 ret = btrfs_next_leaf(root, path);
1317 if (cow_start != (u64)-1)
1318 cur_offset = cow_start;
1323 leaf = path->nodes[0];
1329 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1331 if (found_key.objectid > ino)
1333 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1334 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1338 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1339 found_key.offset > end)
1342 if (found_key.offset > cur_offset) {
1343 extent_end = found_key.offset;
1348 fi = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_file_extent_item);
1350 extent_type = btrfs_file_extent_type(leaf, fi);
1352 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1353 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1354 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1355 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1356 extent_offset = btrfs_file_extent_offset(leaf, fi);
1357 extent_end = found_key.offset +
1358 btrfs_file_extent_num_bytes(leaf, fi);
1360 btrfs_file_extent_disk_num_bytes(leaf, fi);
1361 if (extent_end <= start) {
1365 if (disk_bytenr == 0)
1367 if (btrfs_file_extent_compression(leaf, fi) ||
1368 btrfs_file_extent_encryption(leaf, fi) ||
1369 btrfs_file_extent_other_encoding(leaf, fi))
1372 * Do the same check as in btrfs_cross_ref_exist but
1373 * without the unnecessary search.
1375 if (btrfs_file_extent_generation(leaf, fi) <=
1376 btrfs_root_last_snapshot(&root->root_item))
1378 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1380 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1382 ret = btrfs_cross_ref_exist(root, ino,
1384 extent_offset, disk_bytenr);
1387 * ret could be -EIO if the above fails to read
1391 if (cow_start != (u64)-1)
1392 cur_offset = cow_start;
1396 WARN_ON_ONCE(nolock);
1399 disk_bytenr += extent_offset;
1400 disk_bytenr += cur_offset - found_key.offset;
1401 num_bytes = min(end + 1, extent_end) - cur_offset;
1403 * if there are pending snapshots for this root,
1404 * we fall into common COW way.
1406 if (!nolock && atomic_read(&root->snapshot_force_cow))
1409 * force cow if csum exists in the range.
1410 * this ensure that csum for a given extent are
1411 * either valid or do not exist.
1413 ret = csum_exist_in_range(fs_info, disk_bytenr,
1417 * ret could be -EIO if the above fails to read
1421 if (cow_start != (u64)-1)
1422 cur_offset = cow_start;
1425 WARN_ON_ONCE(nolock);
1428 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1431 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1432 extent_end = found_key.offset +
1433 btrfs_file_extent_ram_bytes(leaf, fi);
1434 extent_end = ALIGN(extent_end,
1435 fs_info->sectorsize);
1440 if (extent_end <= start) {
1443 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1447 if (cow_start == (u64)-1)
1448 cow_start = cur_offset;
1449 cur_offset = extent_end;
1450 if (cur_offset > end)
1456 btrfs_release_path(path);
1457 if (cow_start != (u64)-1) {
1458 ret = cow_file_range(inode, locked_page,
1459 cow_start, found_key.offset - 1,
1460 end, page_started, nr_written, 1,
1464 btrfs_dec_nocow_writers(fs_info,
1468 cow_start = (u64)-1;
1471 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1472 u64 orig_start = found_key.offset - extent_offset;
1474 em = create_io_em(inode, cur_offset, num_bytes,
1476 disk_bytenr, /* block_start */
1477 num_bytes, /* block_len */
1478 disk_num_bytes, /* orig_block_len */
1479 ram_bytes, BTRFS_COMPRESS_NONE,
1480 BTRFS_ORDERED_PREALLOC);
1483 btrfs_dec_nocow_writers(fs_info,
1488 free_extent_map(em);
1491 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1492 type = BTRFS_ORDERED_PREALLOC;
1494 type = BTRFS_ORDERED_NOCOW;
1497 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1498 num_bytes, num_bytes, type);
1500 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1501 BUG_ON(ret); /* -ENOMEM */
1503 if (root->root_key.objectid ==
1504 BTRFS_DATA_RELOC_TREE_OBJECTID)
1506 * Error handled later, as we must prevent
1507 * extent_clear_unlock_delalloc() in error handler
1508 * from freeing metadata of created ordered extent.
1510 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1513 extent_clear_unlock_delalloc(inode, cur_offset,
1514 cur_offset + num_bytes - 1, end,
1515 locked_page, EXTENT_LOCKED |
1517 EXTENT_CLEAR_DATA_RESV,
1518 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1520 cur_offset = extent_end;
1523 * btrfs_reloc_clone_csums() error, now we're OK to call error
1524 * handler, as metadata for created ordered extent will only
1525 * be freed by btrfs_finish_ordered_io().
1529 if (cur_offset > end)
1532 btrfs_release_path(path);
1534 if (cur_offset <= end && cow_start == (u64)-1) {
1535 cow_start = cur_offset;
1539 if (cow_start != (u64)-1) {
1540 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1541 page_started, nr_written, 1, NULL);
1547 if (ret && cur_offset < end)
1548 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1549 locked_page, EXTENT_LOCKED |
1550 EXTENT_DELALLOC | EXTENT_DEFRAG |
1551 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1553 PAGE_SET_WRITEBACK |
1554 PAGE_END_WRITEBACK);
1555 btrfs_free_path(path);
1559 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1562 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1563 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1567 * @defrag_bytes is a hint value, no spinlock held here,
1568 * if is not zero, it means the file is defragging.
1569 * Force cow if given extent needs to be defragged.
1571 if (BTRFS_I(inode)->defrag_bytes &&
1572 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1573 EXTENT_DEFRAG, 0, NULL))
1580 * extent_io.c call back to do delayed allocation processing
1582 static int run_delalloc_range(void *private_data, struct page *locked_page,
1583 u64 start, u64 end, int *page_started,
1584 unsigned long *nr_written,
1585 struct writeback_control *wbc)
1587 struct inode *inode = private_data;
1589 int force_cow = need_force_cow(inode, start, end);
1590 unsigned int write_flags = wbc_to_write_flags(wbc);
1592 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1593 ret = run_delalloc_nocow(inode, locked_page, start, end,
1594 page_started, 1, nr_written);
1595 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1596 ret = run_delalloc_nocow(inode, locked_page, start, end,
1597 page_started, 0, nr_written);
1598 } else if (!inode_need_compress(inode, start, end)) {
1599 ret = cow_file_range(inode, locked_page, start, end, end,
1600 page_started, nr_written, 1, NULL);
1602 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1603 &BTRFS_I(inode)->runtime_flags);
1604 ret = cow_file_range_async(inode, locked_page, start, end,
1605 page_started, nr_written,
1609 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1613 static void btrfs_split_extent_hook(void *private_data,
1614 struct extent_state *orig, u64 split)
1616 struct inode *inode = private_data;
1619 /* not delalloc, ignore it */
1620 if (!(orig->state & EXTENT_DELALLOC))
1623 size = orig->end - orig->start + 1;
1624 if (size > BTRFS_MAX_EXTENT_SIZE) {
1629 * See the explanation in btrfs_merge_extent_hook, the same
1630 * applies here, just in reverse.
1632 new_size = orig->end - split + 1;
1633 num_extents = count_max_extents(new_size);
1634 new_size = split - orig->start;
1635 num_extents += count_max_extents(new_size);
1636 if (count_max_extents(size) >= num_extents)
1640 spin_lock(&BTRFS_I(inode)->lock);
1641 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1642 spin_unlock(&BTRFS_I(inode)->lock);
1646 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1647 * extents so we can keep track of new extents that are just merged onto old
1648 * extents, such as when we are doing sequential writes, so we can properly
1649 * account for the metadata space we'll need.
1651 static void btrfs_merge_extent_hook(void *private_data,
1652 struct extent_state *new,
1653 struct extent_state *other)
1655 struct inode *inode = private_data;
1656 u64 new_size, old_size;
1659 /* not delalloc, ignore it */
1660 if (!(other->state & EXTENT_DELALLOC))
1663 if (new->start > other->start)
1664 new_size = new->end - other->start + 1;
1666 new_size = other->end - new->start + 1;
1668 /* we're not bigger than the max, unreserve the space and go */
1669 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1670 spin_lock(&BTRFS_I(inode)->lock);
1671 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1672 spin_unlock(&BTRFS_I(inode)->lock);
1677 * We have to add up either side to figure out how many extents were
1678 * accounted for before we merged into one big extent. If the number of
1679 * extents we accounted for is <= the amount we need for the new range
1680 * then we can return, otherwise drop. Think of it like this
1684 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1685 * need 2 outstanding extents, on one side we have 1 and the other side
1686 * we have 1 so they are == and we can return. But in this case
1688 * [MAX_SIZE+4k][MAX_SIZE+4k]
1690 * Each range on their own accounts for 2 extents, but merged together
1691 * they are only 3 extents worth of accounting, so we need to drop in
1694 old_size = other->end - other->start + 1;
1695 num_extents = count_max_extents(old_size);
1696 old_size = new->end - new->start + 1;
1697 num_extents += count_max_extents(old_size);
1698 if (count_max_extents(new_size) >= num_extents)
1701 spin_lock(&BTRFS_I(inode)->lock);
1702 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1703 spin_unlock(&BTRFS_I(inode)->lock);
1706 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1707 struct inode *inode)
1709 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1711 spin_lock(&root->delalloc_lock);
1712 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1713 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1714 &root->delalloc_inodes);
1715 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1716 &BTRFS_I(inode)->runtime_flags);
1717 root->nr_delalloc_inodes++;
1718 if (root->nr_delalloc_inodes == 1) {
1719 spin_lock(&fs_info->delalloc_root_lock);
1720 BUG_ON(!list_empty(&root->delalloc_root));
1721 list_add_tail(&root->delalloc_root,
1722 &fs_info->delalloc_roots);
1723 spin_unlock(&fs_info->delalloc_root_lock);
1726 spin_unlock(&root->delalloc_lock);
1730 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1731 struct btrfs_inode *inode)
1733 struct btrfs_fs_info *fs_info = root->fs_info;
1735 if (!list_empty(&inode->delalloc_inodes)) {
1736 list_del_init(&inode->delalloc_inodes);
1737 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1738 &inode->runtime_flags);
1739 root->nr_delalloc_inodes--;
1740 if (!root->nr_delalloc_inodes) {
1741 ASSERT(list_empty(&root->delalloc_inodes));
1742 spin_lock(&fs_info->delalloc_root_lock);
1743 BUG_ON(list_empty(&root->delalloc_root));
1744 list_del_init(&root->delalloc_root);
1745 spin_unlock(&fs_info->delalloc_root_lock);
1750 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1751 struct btrfs_inode *inode)
1753 spin_lock(&root->delalloc_lock);
1754 __btrfs_del_delalloc_inode(root, inode);
1755 spin_unlock(&root->delalloc_lock);
1759 * extent_io.c set_bit_hook, used to track delayed allocation
1760 * bytes in this file, and to maintain the list of inodes that
1761 * have pending delalloc work to be done.
1763 static void btrfs_set_bit_hook(void *private_data,
1764 struct extent_state *state, unsigned *bits)
1766 struct inode *inode = private_data;
1768 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1770 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1773 * set_bit and clear bit hooks normally require _irqsave/restore
1774 * but in this case, we are only testing for the DELALLOC
1775 * bit, which is only set or cleared with irqs on
1777 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1778 struct btrfs_root *root = BTRFS_I(inode)->root;
1779 u64 len = state->end + 1 - state->start;
1780 u32 num_extents = count_max_extents(len);
1781 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1783 spin_lock(&BTRFS_I(inode)->lock);
1784 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1785 spin_unlock(&BTRFS_I(inode)->lock);
1787 /* For sanity tests */
1788 if (btrfs_is_testing(fs_info))
1791 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1792 fs_info->delalloc_batch);
1793 spin_lock(&BTRFS_I(inode)->lock);
1794 BTRFS_I(inode)->delalloc_bytes += len;
1795 if (*bits & EXTENT_DEFRAG)
1796 BTRFS_I(inode)->defrag_bytes += len;
1797 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1798 &BTRFS_I(inode)->runtime_flags))
1799 btrfs_add_delalloc_inodes(root, inode);
1800 spin_unlock(&BTRFS_I(inode)->lock);
1803 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1804 (*bits & EXTENT_DELALLOC_NEW)) {
1805 spin_lock(&BTRFS_I(inode)->lock);
1806 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1808 spin_unlock(&BTRFS_I(inode)->lock);
1813 * extent_io.c clear_bit_hook, see set_bit_hook for why
1815 static void btrfs_clear_bit_hook(void *private_data,
1816 struct extent_state *state,
1819 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1821 u64 len = state->end + 1 - state->start;
1822 u32 num_extents = count_max_extents(len);
1824 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1825 spin_lock(&inode->lock);
1826 inode->defrag_bytes -= len;
1827 spin_unlock(&inode->lock);
1831 * set_bit and clear bit hooks normally require _irqsave/restore
1832 * but in this case, we are only testing for the DELALLOC
1833 * bit, which is only set or cleared with irqs on
1835 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1836 struct btrfs_root *root = inode->root;
1837 bool do_list = !btrfs_is_free_space_inode(inode);
1839 spin_lock(&inode->lock);
1840 btrfs_mod_outstanding_extents(inode, -num_extents);
1841 spin_unlock(&inode->lock);
1844 * We don't reserve metadata space for space cache inodes so we
1845 * don't need to call dellalloc_release_metadata if there is an
1848 if (*bits & EXTENT_CLEAR_META_RESV &&
1849 root != fs_info->tree_root)
1850 btrfs_delalloc_release_metadata(inode, len, false);
1852 /* For sanity tests. */
1853 if (btrfs_is_testing(fs_info))
1856 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1857 do_list && !(state->state & EXTENT_NORESERVE) &&
1858 (*bits & EXTENT_CLEAR_DATA_RESV))
1859 btrfs_free_reserved_data_space_noquota(
1863 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1864 fs_info->delalloc_batch);
1865 spin_lock(&inode->lock);
1866 inode->delalloc_bytes -= len;
1867 if (do_list && inode->delalloc_bytes == 0 &&
1868 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1869 &inode->runtime_flags))
1870 btrfs_del_delalloc_inode(root, inode);
1871 spin_unlock(&inode->lock);
1874 if ((state->state & EXTENT_DELALLOC_NEW) &&
1875 (*bits & EXTENT_DELALLOC_NEW)) {
1876 spin_lock(&inode->lock);
1877 ASSERT(inode->new_delalloc_bytes >= len);
1878 inode->new_delalloc_bytes -= len;
1879 spin_unlock(&inode->lock);
1884 * Merge bio hook, this must check the chunk tree to make sure we don't create
1885 * bios that span stripes or chunks
1887 * return 1 if page cannot be merged to bio
1888 * return 0 if page can be merged to bio
1889 * return error otherwise
1891 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1892 size_t size, struct bio *bio,
1893 unsigned long bio_flags)
1895 struct inode *inode = page->mapping->host;
1896 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1897 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1902 if (bio_flags & EXTENT_BIO_COMPRESSED)
1905 length = bio->bi_iter.bi_size;
1906 map_length = length;
1907 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1911 if (map_length < length + size)
1917 * in order to insert checksums into the metadata in large chunks,
1918 * we wait until bio submission time. All the pages in the bio are
1919 * checksummed and sums are attached onto the ordered extent record.
1921 * At IO completion time the cums attached on the ordered extent record
1922 * are inserted into the btree
1924 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1927 struct inode *inode = private_data;
1928 blk_status_t ret = 0;
1930 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1931 BUG_ON(ret); /* -ENOMEM */
1936 * in order to insert checksums into the metadata in large chunks,
1937 * we wait until bio submission time. All the pages in the bio are
1938 * checksummed and sums are attached onto the ordered extent record.
1940 * At IO completion time the cums attached on the ordered extent record
1941 * are inserted into the btree
1943 blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1946 struct inode *inode = private_data;
1947 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1950 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1952 bio->bi_status = ret;
1959 * extent_io.c submission hook. This does the right thing for csum calculation
1960 * on write, or reading the csums from the tree before a read.
1962 * Rules about async/sync submit,
1963 * a) read: sync submit
1965 * b) write without checksum: sync submit
1967 * c) write with checksum:
1968 * c-1) if bio is issued by fsync: sync submit
1969 * (sync_writers != 0)
1971 * c-2) if root is reloc root: sync submit
1972 * (only in case of buffered IO)
1974 * c-3) otherwise: async submit
1976 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1977 int mirror_num, unsigned long bio_flags,
1980 struct inode *inode = private_data;
1981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1982 struct btrfs_root *root = BTRFS_I(inode)->root;
1983 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1984 blk_status_t ret = 0;
1986 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1988 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1990 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1991 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1993 if (bio_op(bio) != REQ_OP_WRITE) {
1994 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1998 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1999 ret = btrfs_submit_compressed_read(inode, bio,
2003 } else if (!skip_sum) {
2004 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2009 } else if (async && !skip_sum) {
2010 /* csum items have already been cloned */
2011 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2013 /* we're doing a write, do the async checksumming */
2014 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2016 btrfs_submit_bio_start);
2018 } else if (!skip_sum) {
2019 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2025 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2029 bio->bi_status = ret;
2036 * given a list of ordered sums record them in the inode. This happens
2037 * at IO completion time based on sums calculated at bio submission time.
2039 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2040 struct inode *inode, struct list_head *list)
2042 struct btrfs_ordered_sum *sum;
2045 list_for_each_entry(sum, list, list) {
2046 trans->adding_csums = true;
2047 ret = btrfs_csum_file_blocks(trans,
2048 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2049 trans->adding_csums = false;
2056 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2057 unsigned int extra_bits,
2058 struct extent_state **cached_state, int dedupe)
2060 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2061 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2062 extra_bits, cached_state);
2065 /* see btrfs_writepage_start_hook for details on why this is required */
2066 struct btrfs_writepage_fixup {
2068 struct btrfs_work work;
2071 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2073 struct btrfs_writepage_fixup *fixup;
2074 struct btrfs_ordered_extent *ordered;
2075 struct extent_state *cached_state = NULL;
2076 struct extent_changeset *data_reserved = NULL;
2078 struct inode *inode;
2083 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2087 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2088 ClearPageChecked(page);
2092 inode = page->mapping->host;
2093 page_start = page_offset(page);
2094 page_end = page_offset(page) + PAGE_SIZE - 1;
2096 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2099 /* already ordered? We're done */
2100 if (PagePrivate2(page))
2103 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2106 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2107 page_end, &cached_state);
2109 btrfs_start_ordered_extent(inode, ordered, 1);
2110 btrfs_put_ordered_extent(ordered);
2114 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2117 mapping_set_error(page->mapping, ret);
2118 end_extent_writepage(page, ret, page_start, page_end);
2119 ClearPageChecked(page);
2123 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2126 mapping_set_error(page->mapping, ret);
2127 end_extent_writepage(page, ret, page_start, page_end);
2128 ClearPageChecked(page);
2132 ClearPageChecked(page);
2133 set_page_dirty(page);
2134 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2136 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2142 extent_changeset_free(data_reserved);
2146 * There are a few paths in the higher layers of the kernel that directly
2147 * set the page dirty bit without asking the filesystem if it is a
2148 * good idea. This causes problems because we want to make sure COW
2149 * properly happens and the data=ordered rules are followed.
2151 * In our case any range that doesn't have the ORDERED bit set
2152 * hasn't been properly setup for IO. We kick off an async process
2153 * to fix it up. The async helper will wait for ordered extents, set
2154 * the delalloc bit and make it safe to write the page.
2156 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2158 struct inode *inode = page->mapping->host;
2159 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2160 struct btrfs_writepage_fixup *fixup;
2162 /* this page is properly in the ordered list */
2163 if (TestClearPagePrivate2(page))
2166 if (PageChecked(page))
2169 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2173 SetPageChecked(page);
2175 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2176 btrfs_writepage_fixup_worker, NULL, NULL);
2178 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2182 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2183 struct inode *inode, u64 file_pos,
2184 u64 disk_bytenr, u64 disk_num_bytes,
2185 u64 num_bytes, u64 ram_bytes,
2186 u8 compression, u8 encryption,
2187 u16 other_encoding, int extent_type)
2189 struct btrfs_root *root = BTRFS_I(inode)->root;
2190 struct btrfs_file_extent_item *fi;
2191 struct btrfs_path *path;
2192 struct extent_buffer *leaf;
2193 struct btrfs_key ins;
2195 int extent_inserted = 0;
2198 path = btrfs_alloc_path();
2203 * we may be replacing one extent in the tree with another.
2204 * The new extent is pinned in the extent map, and we don't want
2205 * to drop it from the cache until it is completely in the btree.
2207 * So, tell btrfs_drop_extents to leave this extent in the cache.
2208 * the caller is expected to unpin it and allow it to be merged
2211 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2212 file_pos + num_bytes, NULL, 0,
2213 1, sizeof(*fi), &extent_inserted);
2217 if (!extent_inserted) {
2218 ins.objectid = btrfs_ino(BTRFS_I(inode));
2219 ins.offset = file_pos;
2220 ins.type = BTRFS_EXTENT_DATA_KEY;
2222 path->leave_spinning = 1;
2223 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2228 leaf = path->nodes[0];
2229 fi = btrfs_item_ptr(leaf, path->slots[0],
2230 struct btrfs_file_extent_item);
2231 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2232 btrfs_set_file_extent_type(leaf, fi, extent_type);
2233 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2234 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2235 btrfs_set_file_extent_offset(leaf, fi, 0);
2236 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2237 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2238 btrfs_set_file_extent_compression(leaf, fi, compression);
2239 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2240 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2242 btrfs_mark_buffer_dirty(leaf);
2243 btrfs_release_path(path);
2245 inode_add_bytes(inode, num_bytes);
2247 ins.objectid = disk_bytenr;
2248 ins.offset = disk_num_bytes;
2249 ins.type = BTRFS_EXTENT_ITEM_KEY;
2252 * Release the reserved range from inode dirty range map, as it is
2253 * already moved into delayed_ref_head
2255 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2259 ret = btrfs_alloc_reserved_file_extent(trans, root,
2260 btrfs_ino(BTRFS_I(inode)),
2261 file_pos, qg_released, &ins);
2263 btrfs_free_path(path);
2268 /* snapshot-aware defrag */
2269 struct sa_defrag_extent_backref {
2270 struct rb_node node;
2271 struct old_sa_defrag_extent *old;
2280 struct old_sa_defrag_extent {
2281 struct list_head list;
2282 struct new_sa_defrag_extent *new;
2291 struct new_sa_defrag_extent {
2292 struct rb_root root;
2293 struct list_head head;
2294 struct btrfs_path *path;
2295 struct inode *inode;
2303 static int backref_comp(struct sa_defrag_extent_backref *b1,
2304 struct sa_defrag_extent_backref *b2)
2306 if (b1->root_id < b2->root_id)
2308 else if (b1->root_id > b2->root_id)
2311 if (b1->inum < b2->inum)
2313 else if (b1->inum > b2->inum)
2316 if (b1->file_pos < b2->file_pos)
2318 else if (b1->file_pos > b2->file_pos)
2322 * [------------------------------] ===> (a range of space)
2323 * |<--->| |<---->| =============> (fs/file tree A)
2324 * |<---------------------------->| ===> (fs/file tree B)
2326 * A range of space can refer to two file extents in one tree while
2327 * refer to only one file extent in another tree.
2329 * So we may process a disk offset more than one time(two extents in A)
2330 * and locate at the same extent(one extent in B), then insert two same
2331 * backrefs(both refer to the extent in B).
2336 static void backref_insert(struct rb_root *root,
2337 struct sa_defrag_extent_backref *backref)
2339 struct rb_node **p = &root->rb_node;
2340 struct rb_node *parent = NULL;
2341 struct sa_defrag_extent_backref *entry;
2346 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2348 ret = backref_comp(backref, entry);
2352 p = &(*p)->rb_right;
2355 rb_link_node(&backref->node, parent, p);
2356 rb_insert_color(&backref->node, root);
2360 * Note the backref might has changed, and in this case we just return 0.
2362 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2365 struct btrfs_file_extent_item *extent;
2366 struct old_sa_defrag_extent *old = ctx;
2367 struct new_sa_defrag_extent *new = old->new;
2368 struct btrfs_path *path = new->path;
2369 struct btrfs_key key;
2370 struct btrfs_root *root;
2371 struct sa_defrag_extent_backref *backref;
2372 struct extent_buffer *leaf;
2373 struct inode *inode = new->inode;
2374 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2380 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2381 inum == btrfs_ino(BTRFS_I(inode)))
2384 key.objectid = root_id;
2385 key.type = BTRFS_ROOT_ITEM_KEY;
2386 key.offset = (u64)-1;
2388 root = btrfs_read_fs_root_no_name(fs_info, &key);
2390 if (PTR_ERR(root) == -ENOENT)
2393 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2394 inum, offset, root_id);
2395 return PTR_ERR(root);
2398 key.objectid = inum;
2399 key.type = BTRFS_EXTENT_DATA_KEY;
2400 if (offset > (u64)-1 << 32)
2403 key.offset = offset;
2405 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2406 if (WARN_ON(ret < 0))
2413 leaf = path->nodes[0];
2414 slot = path->slots[0];
2416 if (slot >= btrfs_header_nritems(leaf)) {
2417 ret = btrfs_next_leaf(root, path);
2420 } else if (ret > 0) {
2429 btrfs_item_key_to_cpu(leaf, &key, slot);
2431 if (key.objectid > inum)
2434 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2437 extent = btrfs_item_ptr(leaf, slot,
2438 struct btrfs_file_extent_item);
2440 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2444 * 'offset' refers to the exact key.offset,
2445 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2446 * (key.offset - extent_offset).
2448 if (key.offset != offset)
2451 extent_offset = btrfs_file_extent_offset(leaf, extent);
2452 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2454 if (extent_offset >= old->extent_offset + old->offset +
2455 old->len || extent_offset + num_bytes <=
2456 old->extent_offset + old->offset)
2461 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2467 backref->root_id = root_id;
2468 backref->inum = inum;
2469 backref->file_pos = offset;
2470 backref->num_bytes = num_bytes;
2471 backref->extent_offset = extent_offset;
2472 backref->generation = btrfs_file_extent_generation(leaf, extent);
2474 backref_insert(&new->root, backref);
2477 btrfs_release_path(path);
2482 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2483 struct new_sa_defrag_extent *new)
2485 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2486 struct old_sa_defrag_extent *old, *tmp;
2491 list_for_each_entry_safe(old, tmp, &new->head, list) {
2492 ret = iterate_inodes_from_logical(old->bytenr +
2493 old->extent_offset, fs_info,
2494 path, record_one_backref,
2496 if (ret < 0 && ret != -ENOENT)
2499 /* no backref to be processed for this extent */
2501 list_del(&old->list);
2506 if (list_empty(&new->head))
2512 static int relink_is_mergable(struct extent_buffer *leaf,
2513 struct btrfs_file_extent_item *fi,
2514 struct new_sa_defrag_extent *new)
2516 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2519 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2522 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2525 if (btrfs_file_extent_encryption(leaf, fi) ||
2526 btrfs_file_extent_other_encoding(leaf, fi))
2533 * Note the backref might has changed, and in this case we just return 0.
2535 static noinline int relink_extent_backref(struct btrfs_path *path,
2536 struct sa_defrag_extent_backref *prev,
2537 struct sa_defrag_extent_backref *backref)
2539 struct btrfs_file_extent_item *extent;
2540 struct btrfs_file_extent_item *item;
2541 struct btrfs_ordered_extent *ordered;
2542 struct btrfs_trans_handle *trans;
2543 struct btrfs_root *root;
2544 struct btrfs_key key;
2545 struct extent_buffer *leaf;
2546 struct old_sa_defrag_extent *old = backref->old;
2547 struct new_sa_defrag_extent *new = old->new;
2548 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2549 struct inode *inode;
2550 struct extent_state *cached = NULL;
2559 if (prev && prev->root_id == backref->root_id &&
2560 prev->inum == backref->inum &&
2561 prev->file_pos + prev->num_bytes == backref->file_pos)
2564 /* step 1: get root */
2565 key.objectid = backref->root_id;
2566 key.type = BTRFS_ROOT_ITEM_KEY;
2567 key.offset = (u64)-1;
2569 index = srcu_read_lock(&fs_info->subvol_srcu);
2571 root = btrfs_read_fs_root_no_name(fs_info, &key);
2573 srcu_read_unlock(&fs_info->subvol_srcu, index);
2574 if (PTR_ERR(root) == -ENOENT)
2576 return PTR_ERR(root);
2579 if (btrfs_root_readonly(root)) {
2580 srcu_read_unlock(&fs_info->subvol_srcu, index);
2584 /* step 2: get inode */
2585 key.objectid = backref->inum;
2586 key.type = BTRFS_INODE_ITEM_KEY;
2589 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2590 if (IS_ERR(inode)) {
2591 srcu_read_unlock(&fs_info->subvol_srcu, index);
2595 srcu_read_unlock(&fs_info->subvol_srcu, index);
2597 /* step 3: relink backref */
2598 lock_start = backref->file_pos;
2599 lock_end = backref->file_pos + backref->num_bytes - 1;
2600 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2603 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2605 btrfs_put_ordered_extent(ordered);
2609 trans = btrfs_join_transaction(root);
2610 if (IS_ERR(trans)) {
2611 ret = PTR_ERR(trans);
2615 key.objectid = backref->inum;
2616 key.type = BTRFS_EXTENT_DATA_KEY;
2617 key.offset = backref->file_pos;
2619 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2622 } else if (ret > 0) {
2627 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2628 struct btrfs_file_extent_item);
2630 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2631 backref->generation)
2634 btrfs_release_path(path);
2636 start = backref->file_pos;
2637 if (backref->extent_offset < old->extent_offset + old->offset)
2638 start += old->extent_offset + old->offset -
2639 backref->extent_offset;
2641 len = min(backref->extent_offset + backref->num_bytes,
2642 old->extent_offset + old->offset + old->len);
2643 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2645 ret = btrfs_drop_extents(trans, root, inode, start,
2650 key.objectid = btrfs_ino(BTRFS_I(inode));
2651 key.type = BTRFS_EXTENT_DATA_KEY;
2654 path->leave_spinning = 1;
2656 struct btrfs_file_extent_item *fi;
2658 struct btrfs_key found_key;
2660 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2665 leaf = path->nodes[0];
2666 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2668 fi = btrfs_item_ptr(leaf, path->slots[0],
2669 struct btrfs_file_extent_item);
2670 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2672 if (extent_len + found_key.offset == start &&
2673 relink_is_mergable(leaf, fi, new)) {
2674 btrfs_set_file_extent_num_bytes(leaf, fi,
2676 btrfs_mark_buffer_dirty(leaf);
2677 inode_add_bytes(inode, len);
2683 btrfs_release_path(path);
2688 ret = btrfs_insert_empty_item(trans, root, path, &key,
2691 btrfs_abort_transaction(trans, ret);
2695 leaf = path->nodes[0];
2696 item = btrfs_item_ptr(leaf, path->slots[0],
2697 struct btrfs_file_extent_item);
2698 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2699 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2700 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2701 btrfs_set_file_extent_num_bytes(leaf, item, len);
2702 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2703 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2704 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2705 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2706 btrfs_set_file_extent_encryption(leaf, item, 0);
2707 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2709 btrfs_mark_buffer_dirty(leaf);
2710 inode_add_bytes(inode, len);
2711 btrfs_release_path(path);
2713 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2715 backref->root_id, backref->inum,
2716 new->file_pos); /* start - extent_offset */
2718 btrfs_abort_transaction(trans, ret);
2724 btrfs_release_path(path);
2725 path->leave_spinning = 0;
2726 btrfs_end_transaction(trans);
2728 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2734 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2736 struct old_sa_defrag_extent *old, *tmp;
2741 list_for_each_entry_safe(old, tmp, &new->head, list) {
2747 static void relink_file_extents(struct new_sa_defrag_extent *new)
2749 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2750 struct btrfs_path *path;
2751 struct sa_defrag_extent_backref *backref;
2752 struct sa_defrag_extent_backref *prev = NULL;
2753 struct rb_node *node;
2756 path = btrfs_alloc_path();
2760 if (!record_extent_backrefs(path, new)) {
2761 btrfs_free_path(path);
2764 btrfs_release_path(path);
2767 node = rb_first(&new->root);
2770 rb_erase(node, &new->root);
2772 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2774 ret = relink_extent_backref(path, prev, backref);
2787 btrfs_free_path(path);
2789 free_sa_defrag_extent(new);
2791 atomic_dec(&fs_info->defrag_running);
2792 wake_up(&fs_info->transaction_wait);
2795 static struct new_sa_defrag_extent *
2796 record_old_file_extents(struct inode *inode,
2797 struct btrfs_ordered_extent *ordered)
2799 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2800 struct btrfs_root *root = BTRFS_I(inode)->root;
2801 struct btrfs_path *path;
2802 struct btrfs_key key;
2803 struct old_sa_defrag_extent *old;
2804 struct new_sa_defrag_extent *new;
2807 new = kmalloc(sizeof(*new), GFP_NOFS);
2812 new->file_pos = ordered->file_offset;
2813 new->len = ordered->len;
2814 new->bytenr = ordered->start;
2815 new->disk_len = ordered->disk_len;
2816 new->compress_type = ordered->compress_type;
2817 new->root = RB_ROOT;
2818 INIT_LIST_HEAD(&new->head);
2820 path = btrfs_alloc_path();
2824 key.objectid = btrfs_ino(BTRFS_I(inode));
2825 key.type = BTRFS_EXTENT_DATA_KEY;
2826 key.offset = new->file_pos;
2828 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2831 if (ret > 0 && path->slots[0] > 0)
2834 /* find out all the old extents for the file range */
2836 struct btrfs_file_extent_item *extent;
2837 struct extent_buffer *l;
2846 slot = path->slots[0];
2848 if (slot >= btrfs_header_nritems(l)) {
2849 ret = btrfs_next_leaf(root, path);
2857 btrfs_item_key_to_cpu(l, &key, slot);
2859 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2861 if (key.type != BTRFS_EXTENT_DATA_KEY)
2863 if (key.offset >= new->file_pos + new->len)
2866 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2868 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2869 if (key.offset + num_bytes < new->file_pos)
2872 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2876 extent_offset = btrfs_file_extent_offset(l, extent);
2878 old = kmalloc(sizeof(*old), GFP_NOFS);
2882 offset = max(new->file_pos, key.offset);
2883 end = min(new->file_pos + new->len, key.offset + num_bytes);
2885 old->bytenr = disk_bytenr;
2886 old->extent_offset = extent_offset;
2887 old->offset = offset - key.offset;
2888 old->len = end - offset;
2891 list_add_tail(&old->list, &new->head);
2897 btrfs_free_path(path);
2898 atomic_inc(&fs_info->defrag_running);
2903 btrfs_free_path(path);
2905 free_sa_defrag_extent(new);
2909 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2912 struct btrfs_block_group_cache *cache;
2914 cache = btrfs_lookup_block_group(fs_info, start);
2917 spin_lock(&cache->lock);
2918 cache->delalloc_bytes -= len;
2919 spin_unlock(&cache->lock);
2921 btrfs_put_block_group(cache);
2924 /* as ordered data IO finishes, this gets called so we can finish
2925 * an ordered extent if the range of bytes in the file it covers are
2928 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2930 struct inode *inode = ordered_extent->inode;
2931 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2932 struct btrfs_root *root = BTRFS_I(inode)->root;
2933 struct btrfs_trans_handle *trans = NULL;
2934 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2935 struct extent_state *cached_state = NULL;
2936 struct new_sa_defrag_extent *new = NULL;
2937 int compress_type = 0;
2939 u64 logical_len = ordered_extent->len;
2941 bool truncated = false;
2942 bool range_locked = false;
2943 bool clear_new_delalloc_bytes = false;
2945 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2946 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2947 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2948 clear_new_delalloc_bytes = true;
2950 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2952 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2957 btrfs_free_io_failure_record(BTRFS_I(inode),
2958 ordered_extent->file_offset,
2959 ordered_extent->file_offset +
2960 ordered_extent->len - 1);
2962 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2964 logical_len = ordered_extent->truncated_len;
2965 /* Truncated the entire extent, don't bother adding */
2970 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2971 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2974 * For mwrite(mmap + memset to write) case, we still reserve
2975 * space for NOCOW range.
2976 * As NOCOW won't cause a new delayed ref, just free the space
2978 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2979 ordered_extent->len);
2980 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2982 trans = btrfs_join_transaction_nolock(root);
2984 trans = btrfs_join_transaction(root);
2985 if (IS_ERR(trans)) {
2986 ret = PTR_ERR(trans);
2990 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2991 ret = btrfs_update_inode_fallback(trans, root, inode);
2992 if (ret) /* -ENOMEM or corruption */
2993 btrfs_abort_transaction(trans, ret);
2997 range_locked = true;
2998 lock_extent_bits(io_tree, ordered_extent->file_offset,
2999 ordered_extent->file_offset + ordered_extent->len - 1,
3002 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3003 ordered_extent->file_offset + ordered_extent->len - 1,
3004 EXTENT_DEFRAG, 0, cached_state);
3006 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3007 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3008 /* the inode is shared */
3009 new = record_old_file_extents(inode, ordered_extent);
3011 clear_extent_bit(io_tree, ordered_extent->file_offset,
3012 ordered_extent->file_offset + ordered_extent->len - 1,
3013 EXTENT_DEFRAG, 0, 0, &cached_state);
3017 trans = btrfs_join_transaction_nolock(root);
3019 trans = btrfs_join_transaction(root);
3020 if (IS_ERR(trans)) {
3021 ret = PTR_ERR(trans);
3026 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3028 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3029 compress_type = ordered_extent->compress_type;
3030 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3031 BUG_ON(compress_type);
3032 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3033 ordered_extent->len);
3034 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3035 ordered_extent->file_offset,
3036 ordered_extent->file_offset +
3039 BUG_ON(root == fs_info->tree_root);
3040 ret = insert_reserved_file_extent(trans, inode,
3041 ordered_extent->file_offset,
3042 ordered_extent->start,
3043 ordered_extent->disk_len,
3044 logical_len, logical_len,
3045 compress_type, 0, 0,
3046 BTRFS_FILE_EXTENT_REG);
3048 btrfs_release_delalloc_bytes(fs_info,
3049 ordered_extent->start,
3050 ordered_extent->disk_len);
3052 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3053 ordered_extent->file_offset, ordered_extent->len,
3056 btrfs_abort_transaction(trans, ret);
3060 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3062 btrfs_abort_transaction(trans, ret);
3066 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3067 ret = btrfs_update_inode_fallback(trans, root, inode);
3068 if (ret) { /* -ENOMEM or corruption */
3069 btrfs_abort_transaction(trans, ret);
3074 if (range_locked || clear_new_delalloc_bytes) {
3075 unsigned int clear_bits = 0;
3078 clear_bits |= EXTENT_LOCKED;
3079 if (clear_new_delalloc_bytes)
3080 clear_bits |= EXTENT_DELALLOC_NEW;
3081 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3082 ordered_extent->file_offset,
3083 ordered_extent->file_offset +
3084 ordered_extent->len - 1,
3086 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3091 btrfs_end_transaction(trans);
3093 if (ret || truncated) {
3097 start = ordered_extent->file_offset + logical_len;
3099 start = ordered_extent->file_offset;
3100 end = ordered_extent->file_offset + ordered_extent->len - 1;
3101 clear_extent_uptodate(io_tree, start, end, NULL);
3103 /* Drop the cache for the part of the extent we didn't write. */
3104 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3107 * If the ordered extent had an IOERR or something else went
3108 * wrong we need to return the space for this ordered extent
3109 * back to the allocator. We only free the extent in the
3110 * truncated case if we didn't write out the extent at all.
3112 if ((ret || !logical_len) &&
3113 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3114 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3115 btrfs_free_reserved_extent(fs_info,
3116 ordered_extent->start,
3117 ordered_extent->disk_len, 1);
3122 * This needs to be done to make sure anybody waiting knows we are done
3123 * updating everything for this ordered extent.
3125 btrfs_remove_ordered_extent(inode, ordered_extent);
3127 /* for snapshot-aware defrag */
3130 free_sa_defrag_extent(new);
3131 atomic_dec(&fs_info->defrag_running);
3133 relink_file_extents(new);
3138 btrfs_put_ordered_extent(ordered_extent);
3139 /* once for the tree */
3140 btrfs_put_ordered_extent(ordered_extent);
3142 /* Try to release some metadata so we don't get an OOM but don't wait */
3143 btrfs_btree_balance_dirty_nodelay(fs_info);
3148 static void finish_ordered_fn(struct btrfs_work *work)
3150 struct btrfs_ordered_extent *ordered_extent;
3151 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3152 btrfs_finish_ordered_io(ordered_extent);
3155 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3156 struct extent_state *state, int uptodate)
3158 struct inode *inode = page->mapping->host;
3159 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3160 struct btrfs_ordered_extent *ordered_extent = NULL;
3161 struct btrfs_workqueue *wq;
3162 btrfs_work_func_t func;
3164 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3166 ClearPagePrivate2(page);
3167 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3168 end - start + 1, uptodate))
3171 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3172 wq = fs_info->endio_freespace_worker;
3173 func = btrfs_freespace_write_helper;
3175 wq = fs_info->endio_write_workers;
3176 func = btrfs_endio_write_helper;
3179 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3181 btrfs_queue_work(wq, &ordered_extent->work);
3184 static int __readpage_endio_check(struct inode *inode,
3185 struct btrfs_io_bio *io_bio,
3186 int icsum, struct page *page,
3187 int pgoff, u64 start, size_t len)
3193 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3195 kaddr = kmap_atomic(page);
3196 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3197 btrfs_csum_final(csum, (u8 *)&csum);
3198 if (csum != csum_expected)
3201 kunmap_atomic(kaddr);
3204 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3205 io_bio->mirror_num);
3206 memset(kaddr + pgoff, 1, len);
3207 flush_dcache_page(page);
3208 kunmap_atomic(kaddr);
3213 * when reads are done, we need to check csums to verify the data is correct
3214 * if there's a match, we allow the bio to finish. If not, the code in
3215 * extent_io.c will try to find good copies for us.
3217 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3218 u64 phy_offset, struct page *page,
3219 u64 start, u64 end, int mirror)
3221 size_t offset = start - page_offset(page);
3222 struct inode *inode = page->mapping->host;
3223 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3224 struct btrfs_root *root = BTRFS_I(inode)->root;
3226 if (PageChecked(page)) {
3227 ClearPageChecked(page);
3231 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3234 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3235 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3236 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3240 phy_offset >>= inode->i_sb->s_blocksize_bits;
3241 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3242 start, (size_t)(end - start + 1));
3246 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3248 * @inode: The inode we want to perform iput on
3250 * This function uses the generic vfs_inode::i_count to track whether we should
3251 * just decrement it (in case it's > 1) or if this is the last iput then link
3252 * the inode to the delayed iput machinery. Delayed iputs are processed at
3253 * transaction commit time/superblock commit/cleaner kthread.
3255 void btrfs_add_delayed_iput(struct inode *inode)
3257 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3258 struct btrfs_inode *binode = BTRFS_I(inode);
3260 if (atomic_add_unless(&inode->i_count, -1, 1))
3263 spin_lock(&fs_info->delayed_iput_lock);
3264 ASSERT(list_empty(&binode->delayed_iput));
3265 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3266 spin_unlock(&fs_info->delayed_iput_lock);
3269 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3272 spin_lock(&fs_info->delayed_iput_lock);
3273 while (!list_empty(&fs_info->delayed_iputs)) {
3274 struct btrfs_inode *inode;
3276 inode = list_first_entry(&fs_info->delayed_iputs,
3277 struct btrfs_inode, delayed_iput);
3278 list_del_init(&inode->delayed_iput);
3279 spin_unlock(&fs_info->delayed_iput_lock);
3280 iput(&inode->vfs_inode);
3281 spin_lock(&fs_info->delayed_iput_lock);
3283 spin_unlock(&fs_info->delayed_iput_lock);
3287 * This creates an orphan entry for the given inode in case something goes wrong
3288 * in the middle of an unlink.
3290 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3291 struct btrfs_inode *inode)
3295 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3296 if (ret && ret != -EEXIST) {
3297 btrfs_abort_transaction(trans, ret);
3305 * We have done the delete so we can go ahead and remove the orphan item for
3306 * this particular inode.
3308 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3309 struct btrfs_inode *inode)
3311 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3315 * this cleans up any orphans that may be left on the list from the last use
3318 int btrfs_orphan_cleanup(struct btrfs_root *root)
3320 struct btrfs_fs_info *fs_info = root->fs_info;
3321 struct btrfs_path *path;
3322 struct extent_buffer *leaf;
3323 struct btrfs_key key, found_key;
3324 struct btrfs_trans_handle *trans;
3325 struct inode *inode;
3326 u64 last_objectid = 0;
3327 int ret = 0, nr_unlink = 0;
3329 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3332 path = btrfs_alloc_path();
3337 path->reada = READA_BACK;
3339 key.objectid = BTRFS_ORPHAN_OBJECTID;
3340 key.type = BTRFS_ORPHAN_ITEM_KEY;
3341 key.offset = (u64)-1;
3344 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3349 * if ret == 0 means we found what we were searching for, which
3350 * is weird, but possible, so only screw with path if we didn't
3351 * find the key and see if we have stuff that matches
3355 if (path->slots[0] == 0)
3360 /* pull out the item */
3361 leaf = path->nodes[0];
3362 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3364 /* make sure the item matches what we want */
3365 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3367 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3370 /* release the path since we're done with it */
3371 btrfs_release_path(path);
3374 * this is where we are basically btrfs_lookup, without the
3375 * crossing root thing. we store the inode number in the
3376 * offset of the orphan item.
3379 if (found_key.offset == last_objectid) {
3381 "Error removing orphan entry, stopping orphan cleanup");
3386 last_objectid = found_key.offset;
3388 found_key.objectid = found_key.offset;
3389 found_key.type = BTRFS_INODE_ITEM_KEY;
3390 found_key.offset = 0;
3391 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3392 ret = PTR_ERR_OR_ZERO(inode);
3393 if (ret && ret != -ENOENT)
3396 if (ret == -ENOENT && root == fs_info->tree_root) {
3397 struct btrfs_root *dead_root;
3398 struct btrfs_fs_info *fs_info = root->fs_info;
3399 int is_dead_root = 0;
3402 * this is an orphan in the tree root. Currently these
3403 * could come from 2 sources:
3404 * a) a snapshot deletion in progress
3405 * b) a free space cache inode
3406 * We need to distinguish those two, as the snapshot
3407 * orphan must not get deleted.
3408 * find_dead_roots already ran before us, so if this
3409 * is a snapshot deletion, we should find the root
3410 * in the dead_roots list
3412 spin_lock(&fs_info->trans_lock);
3413 list_for_each_entry(dead_root, &fs_info->dead_roots,
3415 if (dead_root->root_key.objectid ==
3416 found_key.objectid) {
3421 spin_unlock(&fs_info->trans_lock);
3423 /* prevent this orphan from being found again */
3424 key.offset = found_key.objectid - 1;
3431 * If we have an inode with links, there are a couple of
3432 * possibilities. Old kernels (before v3.12) used to create an
3433 * orphan item for truncate indicating that there were possibly
3434 * extent items past i_size that needed to be deleted. In v3.12,
3435 * truncate was changed to update i_size in sync with the extent
3436 * items, but the (useless) orphan item was still created. Since
3437 * v4.18, we don't create the orphan item for truncate at all.
3439 * So, this item could mean that we need to do a truncate, but
3440 * only if this filesystem was last used on a pre-v3.12 kernel
3441 * and was not cleanly unmounted. The odds of that are quite
3442 * slim, and it's a pain to do the truncate now, so just delete
3445 * It's also possible that this orphan item was supposed to be
3446 * deleted but wasn't. The inode number may have been reused,
3447 * but either way, we can delete the orphan item.
3449 if (ret == -ENOENT || inode->i_nlink) {
3452 trans = btrfs_start_transaction(root, 1);
3453 if (IS_ERR(trans)) {
3454 ret = PTR_ERR(trans);
3457 btrfs_debug(fs_info, "auto deleting %Lu",
3458 found_key.objectid);
3459 ret = btrfs_del_orphan_item(trans, root,
3460 found_key.objectid);
3461 btrfs_end_transaction(trans);
3469 /* this will do delete_inode and everything for us */
3472 /* release the path since we're done with it */
3473 btrfs_release_path(path);
3475 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3477 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3478 trans = btrfs_join_transaction(root);
3480 btrfs_end_transaction(trans);
3484 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3488 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3489 btrfs_free_path(path);
3494 * very simple check to peek ahead in the leaf looking for xattrs. If we
3495 * don't find any xattrs, we know there can't be any acls.
3497 * slot is the slot the inode is in, objectid is the objectid of the inode
3499 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3500 int slot, u64 objectid,
3501 int *first_xattr_slot)
3503 u32 nritems = btrfs_header_nritems(leaf);
3504 struct btrfs_key found_key;
3505 static u64 xattr_access = 0;
3506 static u64 xattr_default = 0;
3509 if (!xattr_access) {
3510 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3511 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3512 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3513 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3517 *first_xattr_slot = -1;
3518 while (slot < nritems) {
3519 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3521 /* we found a different objectid, there must not be acls */
3522 if (found_key.objectid != objectid)
3525 /* we found an xattr, assume we've got an acl */
3526 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3527 if (*first_xattr_slot == -1)
3528 *first_xattr_slot = slot;
3529 if (found_key.offset == xattr_access ||
3530 found_key.offset == xattr_default)
3535 * we found a key greater than an xattr key, there can't
3536 * be any acls later on
3538 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3545 * it goes inode, inode backrefs, xattrs, extents,
3546 * so if there are a ton of hard links to an inode there can
3547 * be a lot of backrefs. Don't waste time searching too hard,
3548 * this is just an optimization
3553 /* we hit the end of the leaf before we found an xattr or
3554 * something larger than an xattr. We have to assume the inode
3557 if (*first_xattr_slot == -1)
3558 *first_xattr_slot = slot;
3563 * read an inode from the btree into the in-memory inode
3565 static int btrfs_read_locked_inode(struct inode *inode)
3567 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3568 struct btrfs_path *path;
3569 struct extent_buffer *leaf;
3570 struct btrfs_inode_item *inode_item;
3571 struct btrfs_root *root = BTRFS_I(inode)->root;
3572 struct btrfs_key location;
3577 bool filled = false;
3578 int first_xattr_slot;
3580 ret = btrfs_fill_inode(inode, &rdev);
3584 path = btrfs_alloc_path();
3588 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3590 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3592 btrfs_free_path(path);
3596 leaf = path->nodes[0];
3601 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3602 struct btrfs_inode_item);
3603 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3604 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3605 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3606 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3607 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3609 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3610 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3612 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3613 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3615 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3616 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3618 BTRFS_I(inode)->i_otime.tv_sec =
3619 btrfs_timespec_sec(leaf, &inode_item->otime);
3620 BTRFS_I(inode)->i_otime.tv_nsec =
3621 btrfs_timespec_nsec(leaf, &inode_item->otime);
3623 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3624 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3625 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3627 inode_set_iversion_queried(inode,
3628 btrfs_inode_sequence(leaf, inode_item));
3629 inode->i_generation = BTRFS_I(inode)->generation;
3631 rdev = btrfs_inode_rdev(leaf, inode_item);
3633 BTRFS_I(inode)->index_cnt = (u64)-1;
3634 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3638 * If we were modified in the current generation and evicted from memory
3639 * and then re-read we need to do a full sync since we don't have any
3640 * idea about which extents were modified before we were evicted from
3643 * This is required for both inode re-read from disk and delayed inode
3644 * in delayed_nodes_tree.
3646 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3647 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3648 &BTRFS_I(inode)->runtime_flags);
3651 * We don't persist the id of the transaction where an unlink operation
3652 * against the inode was last made. So here we assume the inode might
3653 * have been evicted, and therefore the exact value of last_unlink_trans
3654 * lost, and set it to last_trans to avoid metadata inconsistencies
3655 * between the inode and its parent if the inode is fsync'ed and the log
3656 * replayed. For example, in the scenario:
3659 * ln mydir/foo mydir/bar
3662 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3663 * xfs_io -c fsync mydir/foo
3665 * mount fs, triggers fsync log replay
3667 * We must make sure that when we fsync our inode foo we also log its
3668 * parent inode, otherwise after log replay the parent still has the
3669 * dentry with the "bar" name but our inode foo has a link count of 1
3670 * and doesn't have an inode ref with the name "bar" anymore.
3672 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3673 * but it guarantees correctness at the expense of occasional full
3674 * transaction commits on fsync if our inode is a directory, or if our
3675 * inode is not a directory, logging its parent unnecessarily.
3677 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3680 if (inode->i_nlink != 1 ||
3681 path->slots[0] >= btrfs_header_nritems(leaf))
3684 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3685 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3688 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3689 if (location.type == BTRFS_INODE_REF_KEY) {
3690 struct btrfs_inode_ref *ref;
3692 ref = (struct btrfs_inode_ref *)ptr;
3693 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3694 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3695 struct btrfs_inode_extref *extref;
3697 extref = (struct btrfs_inode_extref *)ptr;
3698 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3703 * try to precache a NULL acl entry for files that don't have
3704 * any xattrs or acls
3706 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3707 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3708 if (first_xattr_slot != -1) {
3709 path->slots[0] = first_xattr_slot;
3710 ret = btrfs_load_inode_props(inode, path);
3713 "error loading props for ino %llu (root %llu): %d",
3714 btrfs_ino(BTRFS_I(inode)),
3715 root->root_key.objectid, ret);
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_symlink_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, fs_info))
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,
4728 should_throttle = true;
4732 if (found_type == BTRFS_INODE_ITEM_KEY)
4735 if (path->slots[0] == 0 ||
4736 path->slots[0] != pending_del_slot ||
4737 should_throttle || should_end) {
4738 if (pending_del_nr) {
4739 ret = btrfs_del_items(trans, root, path,
4743 btrfs_abort_transaction(trans, ret);
4748 btrfs_release_path(path);
4749 if (should_throttle) {
4750 unsigned long updates = trans->delayed_ref_updates;
4752 trans->delayed_ref_updates = 0;
4753 ret = btrfs_run_delayed_refs(trans,
4760 * if we failed to refill our space rsv, bail out
4761 * and let the transaction restart
4773 if (ret >= 0 && pending_del_nr) {
4776 err = btrfs_del_items(trans, root, path, pending_del_slot,
4779 btrfs_abort_transaction(trans, err);
4783 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4784 ASSERT(last_size >= new_size);
4785 if (!ret && last_size > new_size)
4786 last_size = new_size;
4787 btrfs_ordered_update_i_size(inode, last_size, NULL);
4790 btrfs_free_path(path);
4792 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4793 unsigned long updates = trans->delayed_ref_updates;
4797 trans->delayed_ref_updates = 0;
4798 err = btrfs_run_delayed_refs(trans, updates * 2);
4807 * btrfs_truncate_block - read, zero a chunk and write a block
4808 * @inode - inode that we're zeroing
4809 * @from - the offset to start zeroing
4810 * @len - the length to zero, 0 to zero the entire range respective to the
4812 * @front - zero up to the offset instead of from the offset on
4814 * This will find the block for the "from" offset and cow the block and zero the
4815 * part we want to zero. This is used with truncate and hole punching.
4817 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4821 struct address_space *mapping = inode->i_mapping;
4822 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4823 struct btrfs_ordered_extent *ordered;
4824 struct extent_state *cached_state = NULL;
4825 struct extent_changeset *data_reserved = NULL;
4827 u32 blocksize = fs_info->sectorsize;
4828 pgoff_t index = from >> PAGE_SHIFT;
4829 unsigned offset = from & (blocksize - 1);
4831 gfp_t mask = btrfs_alloc_write_mask(mapping);
4836 if (IS_ALIGNED(offset, blocksize) &&
4837 (!len || IS_ALIGNED(len, blocksize)))
4840 block_start = round_down(from, blocksize);
4841 block_end = block_start + blocksize - 1;
4843 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4844 block_start, blocksize);
4849 page = find_or_create_page(mapping, index, mask);
4851 btrfs_delalloc_release_space(inode, data_reserved,
4852 block_start, blocksize, true);
4853 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4858 if (!PageUptodate(page)) {
4859 ret = btrfs_readpage(NULL, page);
4861 if (page->mapping != mapping) {
4866 if (!PageUptodate(page)) {
4871 wait_on_page_writeback(page);
4873 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4874 set_page_extent_mapped(page);
4876 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4878 unlock_extent_cached(io_tree, block_start, block_end,
4882 btrfs_start_ordered_extent(inode, ordered, 1);
4883 btrfs_put_ordered_extent(ordered);
4887 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4888 EXTENT_DIRTY | EXTENT_DELALLOC |
4889 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4890 0, 0, &cached_state);
4892 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4895 unlock_extent_cached(io_tree, block_start, block_end,
4900 if (offset != blocksize) {
4902 len = blocksize - offset;
4905 memset(kaddr + (block_start - page_offset(page)),
4908 memset(kaddr + (block_start - page_offset(page)) + offset,
4910 flush_dcache_page(page);
4913 ClearPageChecked(page);
4914 set_page_dirty(page);
4915 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4919 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4921 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4925 extent_changeset_free(data_reserved);
4929 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4930 u64 offset, u64 len)
4932 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4933 struct btrfs_trans_handle *trans;
4937 * Still need to make sure the inode looks like it's been updated so
4938 * that any holes get logged if we fsync.
4940 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4941 BTRFS_I(inode)->last_trans = fs_info->generation;
4942 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4943 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4948 * 1 - for the one we're dropping
4949 * 1 - for the one we're adding
4950 * 1 - for updating the inode.
4952 trans = btrfs_start_transaction(root, 3);
4954 return PTR_ERR(trans);
4956 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4958 btrfs_abort_transaction(trans, ret);
4959 btrfs_end_transaction(trans);
4963 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4964 offset, 0, 0, len, 0, len, 0, 0, 0);
4966 btrfs_abort_transaction(trans, ret);
4968 btrfs_update_inode(trans, root, inode);
4969 btrfs_end_transaction(trans);
4974 * This function puts in dummy file extents for the area we're creating a hole
4975 * for. So if we are truncating this file to a larger size we need to insert
4976 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4977 * the range between oldsize and size
4979 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4982 struct btrfs_root *root = BTRFS_I(inode)->root;
4983 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4984 struct extent_map *em = NULL;
4985 struct extent_state *cached_state = NULL;
4986 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4987 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4988 u64 block_end = ALIGN(size, fs_info->sectorsize);
4995 * If our size started in the middle of a block we need to zero out the
4996 * rest of the block before we expand the i_size, otherwise we could
4997 * expose stale data.
4999 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5003 if (size <= hole_start)
5007 struct btrfs_ordered_extent *ordered;
5009 lock_extent_bits(io_tree, hole_start, block_end - 1,
5011 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5012 block_end - hole_start);
5015 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5017 btrfs_start_ordered_extent(inode, ordered, 1);
5018 btrfs_put_ordered_extent(ordered);
5021 cur_offset = hole_start;
5023 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5024 block_end - cur_offset, 0);
5030 last_byte = min(extent_map_end(em), block_end);
5031 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5032 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5033 struct extent_map *hole_em;
5034 hole_size = last_byte - cur_offset;
5036 err = maybe_insert_hole(root, inode, cur_offset,
5040 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5041 cur_offset + hole_size - 1, 0);
5042 hole_em = alloc_extent_map();
5044 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5045 &BTRFS_I(inode)->runtime_flags);
5048 hole_em->start = cur_offset;
5049 hole_em->len = hole_size;
5050 hole_em->orig_start = cur_offset;
5052 hole_em->block_start = EXTENT_MAP_HOLE;
5053 hole_em->block_len = 0;
5054 hole_em->orig_block_len = 0;
5055 hole_em->ram_bytes = hole_size;
5056 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5057 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5058 hole_em->generation = fs_info->generation;
5061 write_lock(&em_tree->lock);
5062 err = add_extent_mapping(em_tree, hole_em, 1);
5063 write_unlock(&em_tree->lock);
5066 btrfs_drop_extent_cache(BTRFS_I(inode),
5071 free_extent_map(hole_em);
5074 free_extent_map(em);
5076 cur_offset = last_byte;
5077 if (cur_offset >= block_end)
5080 free_extent_map(em);
5081 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5085 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5087 struct btrfs_root *root = BTRFS_I(inode)->root;
5088 struct btrfs_trans_handle *trans;
5089 loff_t oldsize = i_size_read(inode);
5090 loff_t newsize = attr->ia_size;
5091 int mask = attr->ia_valid;
5095 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5096 * special case where we need to update the times despite not having
5097 * these flags set. For all other operations the VFS set these flags
5098 * explicitly if it wants a timestamp update.
5100 if (newsize != oldsize) {
5101 inode_inc_iversion(inode);
5102 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5103 inode->i_ctime = inode->i_mtime =
5104 current_time(inode);
5107 if (newsize > oldsize) {
5109 * Don't do an expanding truncate while snapshotting is ongoing.
5110 * This is to ensure the snapshot captures a fully consistent
5111 * state of this file - if the snapshot captures this expanding
5112 * truncation, it must capture all writes that happened before
5115 btrfs_wait_for_snapshot_creation(root);
5116 ret = btrfs_cont_expand(inode, oldsize, newsize);
5118 btrfs_end_write_no_snapshotting(root);
5122 trans = btrfs_start_transaction(root, 1);
5123 if (IS_ERR(trans)) {
5124 btrfs_end_write_no_snapshotting(root);
5125 return PTR_ERR(trans);
5128 i_size_write(inode, newsize);
5129 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5130 pagecache_isize_extended(inode, oldsize, newsize);
5131 ret = btrfs_update_inode(trans, root, inode);
5132 btrfs_end_write_no_snapshotting(root);
5133 btrfs_end_transaction(trans);
5137 * We're truncating a file that used to have good data down to
5138 * zero. Make sure it gets into the ordered flush list so that
5139 * any new writes get down to disk quickly.
5142 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5143 &BTRFS_I(inode)->runtime_flags);
5145 truncate_setsize(inode, newsize);
5147 /* Disable nonlocked read DIO to avoid the end less truncate */
5148 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5149 inode_dio_wait(inode);
5150 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5152 ret = btrfs_truncate(inode, newsize == oldsize);
5153 if (ret && inode->i_nlink) {
5157 * Truncate failed, so fix up the in-memory size. We
5158 * adjusted disk_i_size down as we removed extents, so
5159 * wait for disk_i_size to be stable and then update the
5160 * in-memory size to match.
5162 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5165 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5172 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5174 struct inode *inode = d_inode(dentry);
5175 struct btrfs_root *root = BTRFS_I(inode)->root;
5178 if (btrfs_root_readonly(root))
5181 err = setattr_prepare(dentry, attr);
5185 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5186 err = btrfs_setsize(inode, attr);
5191 if (attr->ia_valid) {
5192 setattr_copy(inode, attr);
5193 inode_inc_iversion(inode);
5194 err = btrfs_dirty_inode(inode);
5196 if (!err && attr->ia_valid & ATTR_MODE)
5197 err = posix_acl_chmod(inode, inode->i_mode);
5204 * While truncating the inode pages during eviction, we get the VFS calling
5205 * btrfs_invalidatepage() against each page of the inode. This is slow because
5206 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5207 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5208 * extent_state structures over and over, wasting lots of time.
5210 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5211 * those expensive operations on a per page basis and do only the ordered io
5212 * finishing, while we release here the extent_map and extent_state structures,
5213 * without the excessive merging and splitting.
5215 static void evict_inode_truncate_pages(struct inode *inode)
5217 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5218 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5219 struct rb_node *node;
5221 ASSERT(inode->i_state & I_FREEING);
5222 truncate_inode_pages_final(&inode->i_data);
5224 write_lock(&map_tree->lock);
5225 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5226 struct extent_map *em;
5228 node = rb_first(&map_tree->map);
5229 em = rb_entry(node, struct extent_map, rb_node);
5230 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5231 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5232 remove_extent_mapping(map_tree, em);
5233 free_extent_map(em);
5234 if (need_resched()) {
5235 write_unlock(&map_tree->lock);
5237 write_lock(&map_tree->lock);
5240 write_unlock(&map_tree->lock);
5243 * Keep looping until we have no more ranges in the io tree.
5244 * We can have ongoing bios started by readpages (called from readahead)
5245 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5246 * still in progress (unlocked the pages in the bio but did not yet
5247 * unlocked the ranges in the io tree). Therefore this means some
5248 * ranges can still be locked and eviction started because before
5249 * submitting those bios, which are executed by a separate task (work
5250 * queue kthread), inode references (inode->i_count) were not taken
5251 * (which would be dropped in the end io callback of each bio).
5252 * Therefore here we effectively end up waiting for those bios and
5253 * anyone else holding locked ranges without having bumped the inode's
5254 * reference count - if we don't do it, when they access the inode's
5255 * io_tree to unlock a range it may be too late, leading to an
5256 * use-after-free issue.
5258 spin_lock(&io_tree->lock);
5259 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5260 struct extent_state *state;
5261 struct extent_state *cached_state = NULL;
5265 node = rb_first(&io_tree->state);
5266 state = rb_entry(node, struct extent_state, rb_node);
5267 start = state->start;
5269 spin_unlock(&io_tree->lock);
5271 lock_extent_bits(io_tree, start, end, &cached_state);
5274 * If still has DELALLOC flag, the extent didn't reach disk,
5275 * and its reserved space won't be freed by delayed_ref.
5276 * So we need to free its reserved space here.
5277 * (Refer to comment in btrfs_invalidatepage, case 2)
5279 * Note, end is the bytenr of last byte, so we need + 1 here.
5281 if (state->state & EXTENT_DELALLOC)
5282 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5284 clear_extent_bit(io_tree, start, end,
5285 EXTENT_LOCKED | EXTENT_DIRTY |
5286 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5287 EXTENT_DEFRAG, 1, 1, &cached_state);
5290 spin_lock(&io_tree->lock);
5292 spin_unlock(&io_tree->lock);
5295 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5296 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, min_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, fs_info) &&
5325 !btrfs_block_rsv_migrate(global_rsv, rsv, min_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;
5344 trace_btrfs_inode_evict(inode);
5351 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5353 evict_inode_truncate_pages(inode);
5355 if (inode->i_nlink &&
5356 ((btrfs_root_refs(&root->root_item) != 0 &&
5357 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5358 btrfs_is_free_space_inode(BTRFS_I(inode))))
5361 if (is_bad_inode(inode))
5364 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5366 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5369 if (inode->i_nlink > 0) {
5370 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5371 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5375 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5379 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5382 rsv->size = min_size;
5385 btrfs_i_size_write(BTRFS_I(inode), 0);
5388 trans = evict_refill_and_join(root, rsv, min_size);
5392 trans->block_rsv = rsv;
5394 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5395 trans->block_rsv = &fs_info->trans_block_rsv;
5396 btrfs_end_transaction(trans);
5397 btrfs_btree_balance_dirty(fs_info);
5398 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5405 * Errors here aren't a big deal, it just means we leave orphan items in
5406 * the tree. They will be cleaned up on the next mount. If the inode
5407 * number gets reused, cleanup deletes the orphan item without doing
5408 * anything, and unlink reuses the existing orphan item.
5410 * If it turns out that we are dropping too many of these, we might want
5411 * to add a mechanism for retrying these after a commit.
5413 trans = evict_refill_and_join(root, rsv, min_size);
5414 if (!IS_ERR(trans)) {
5415 trans->block_rsv = rsv;
5416 btrfs_orphan_del(trans, BTRFS_I(inode));
5417 trans->block_rsv = &fs_info->trans_block_rsv;
5418 btrfs_end_transaction(trans);
5421 if (!(root == fs_info->tree_root ||
5422 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5423 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5426 btrfs_free_block_rsv(fs_info, rsv);
5429 * If we didn't successfully delete, the orphan item will still be in
5430 * the tree and we'll retry on the next mount. Again, we might also want
5431 * to retry these periodically in the future.
5433 btrfs_remove_delayed_node(BTRFS_I(inode));
5438 * this returns the key found in the dir entry in the location pointer.
5439 * If no dir entries were found, returns -ENOENT.
5440 * If found a corrupted location in dir entry, returns -EUCLEAN.
5442 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5443 struct btrfs_key *location)
5445 const char *name = dentry->d_name.name;
5446 int namelen = dentry->d_name.len;
5447 struct btrfs_dir_item *di;
5448 struct btrfs_path *path;
5449 struct btrfs_root *root = BTRFS_I(dir)->root;
5452 path = btrfs_alloc_path();
5456 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5458 if (IS_ERR_OR_NULL(di)) {
5459 ret = di ? PTR_ERR(di) : -ENOENT;
5463 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5464 if (location->type != BTRFS_INODE_ITEM_KEY &&
5465 location->type != BTRFS_ROOT_ITEM_KEY) {
5467 btrfs_warn(root->fs_info,
5468 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5469 __func__, name, btrfs_ino(BTRFS_I(dir)),
5470 location->objectid, location->type, location->offset);
5473 btrfs_free_path(path);
5478 * when we hit a tree root in a directory, the btrfs part of the inode
5479 * needs to be changed to reflect the root directory of the tree root. This
5480 * is kind of like crossing a mount point.
5482 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5484 struct dentry *dentry,
5485 struct btrfs_key *location,
5486 struct btrfs_root **sub_root)
5488 struct btrfs_path *path;
5489 struct btrfs_root *new_root;
5490 struct btrfs_root_ref *ref;
5491 struct extent_buffer *leaf;
5492 struct btrfs_key key;
5496 path = btrfs_alloc_path();
5503 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5504 key.type = BTRFS_ROOT_REF_KEY;
5505 key.offset = location->objectid;
5507 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5514 leaf = path->nodes[0];
5515 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5516 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5517 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5520 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5521 (unsigned long)(ref + 1),
5522 dentry->d_name.len);
5526 btrfs_release_path(path);
5528 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5529 if (IS_ERR(new_root)) {
5530 err = PTR_ERR(new_root);
5534 *sub_root = new_root;
5535 location->objectid = btrfs_root_dirid(&new_root->root_item);
5536 location->type = BTRFS_INODE_ITEM_KEY;
5537 location->offset = 0;
5540 btrfs_free_path(path);
5544 static void inode_tree_add(struct inode *inode)
5546 struct btrfs_root *root = BTRFS_I(inode)->root;
5547 struct btrfs_inode *entry;
5549 struct rb_node *parent;
5550 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5551 u64 ino = btrfs_ino(BTRFS_I(inode));
5553 if (inode_unhashed(inode))
5556 spin_lock(&root->inode_lock);
5557 p = &root->inode_tree.rb_node;
5560 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5562 if (ino < btrfs_ino(entry))
5563 p = &parent->rb_left;
5564 else if (ino > btrfs_ino(entry))
5565 p = &parent->rb_right;
5567 WARN_ON(!(entry->vfs_inode.i_state &
5568 (I_WILL_FREE | I_FREEING)));
5569 rb_replace_node(parent, new, &root->inode_tree);
5570 RB_CLEAR_NODE(parent);
5571 spin_unlock(&root->inode_lock);
5575 rb_link_node(new, parent, p);
5576 rb_insert_color(new, &root->inode_tree);
5577 spin_unlock(&root->inode_lock);
5580 static void inode_tree_del(struct inode *inode)
5582 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5583 struct btrfs_root *root = BTRFS_I(inode)->root;
5586 spin_lock(&root->inode_lock);
5587 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5588 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5589 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5590 empty = RB_EMPTY_ROOT(&root->inode_tree);
5592 spin_unlock(&root->inode_lock);
5594 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5595 synchronize_srcu(&fs_info->subvol_srcu);
5596 spin_lock(&root->inode_lock);
5597 empty = RB_EMPTY_ROOT(&root->inode_tree);
5598 spin_unlock(&root->inode_lock);
5600 btrfs_add_dead_root(root);
5605 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5607 struct btrfs_iget_args *args = p;
5608 inode->i_ino = args->location->objectid;
5609 memcpy(&BTRFS_I(inode)->location, args->location,
5610 sizeof(*args->location));
5611 BTRFS_I(inode)->root = args->root;
5615 static int btrfs_find_actor(struct inode *inode, void *opaque)
5617 struct btrfs_iget_args *args = opaque;
5618 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5619 args->root == BTRFS_I(inode)->root;
5622 static struct inode *btrfs_iget_locked(struct super_block *s,
5623 struct btrfs_key *location,
5624 struct btrfs_root *root)
5626 struct inode *inode;
5627 struct btrfs_iget_args args;
5628 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5630 args.location = location;
5633 inode = iget5_locked(s, hashval, btrfs_find_actor,
5634 btrfs_init_locked_inode,
5639 /* Get an inode object given its location and corresponding root.
5640 * Returns in *is_new if the inode was read from disk
5642 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5643 struct btrfs_root *root, int *new)
5645 struct inode *inode;
5647 inode = btrfs_iget_locked(s, location, root);
5649 return ERR_PTR(-ENOMEM);
5651 if (inode->i_state & I_NEW) {
5654 ret = btrfs_read_locked_inode(inode);
5656 inode_tree_add(inode);
5657 unlock_new_inode(inode);
5663 * ret > 0 can come from btrfs_search_slot called by
5664 * btrfs_read_locked_inode, this means the inode item
5669 inode = ERR_PTR(ret);
5676 static struct inode *new_simple_dir(struct super_block *s,
5677 struct btrfs_key *key,
5678 struct btrfs_root *root)
5680 struct inode *inode = new_inode(s);
5683 return ERR_PTR(-ENOMEM);
5685 BTRFS_I(inode)->root = root;
5686 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5687 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5689 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5690 inode->i_op = &btrfs_dir_ro_inode_operations;
5691 inode->i_opflags &= ~IOP_XATTR;
5692 inode->i_fop = &simple_dir_operations;
5693 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5694 inode->i_mtime = current_time(inode);
5695 inode->i_atime = inode->i_mtime;
5696 inode->i_ctime = inode->i_mtime;
5697 BTRFS_I(inode)->i_otime = inode->i_mtime;
5702 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5704 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5705 struct inode *inode;
5706 struct btrfs_root *root = BTRFS_I(dir)->root;
5707 struct btrfs_root *sub_root = root;
5708 struct btrfs_key location;
5712 if (dentry->d_name.len > BTRFS_NAME_LEN)
5713 return ERR_PTR(-ENAMETOOLONG);
5715 ret = btrfs_inode_by_name(dir, dentry, &location);
5717 return ERR_PTR(ret);
5719 if (location.type == BTRFS_INODE_ITEM_KEY) {
5720 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5724 index = srcu_read_lock(&fs_info->subvol_srcu);
5725 ret = fixup_tree_root_location(fs_info, dir, dentry,
5726 &location, &sub_root);
5729 inode = ERR_PTR(ret);
5731 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5733 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5735 srcu_read_unlock(&fs_info->subvol_srcu, index);
5737 if (!IS_ERR(inode) && root != sub_root) {
5738 down_read(&fs_info->cleanup_work_sem);
5739 if (!sb_rdonly(inode->i_sb))
5740 ret = btrfs_orphan_cleanup(sub_root);
5741 up_read(&fs_info->cleanup_work_sem);
5744 inode = ERR_PTR(ret);
5751 static int btrfs_dentry_delete(const struct dentry *dentry)
5753 struct btrfs_root *root;
5754 struct inode *inode = d_inode(dentry);
5756 if (!inode && !IS_ROOT(dentry))
5757 inode = d_inode(dentry->d_parent);
5760 root = BTRFS_I(inode)->root;
5761 if (btrfs_root_refs(&root->root_item) == 0)
5764 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5770 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5773 struct inode *inode;
5775 inode = btrfs_lookup_dentry(dir, dentry);
5776 if (IS_ERR(inode)) {
5777 if (PTR_ERR(inode) == -ENOENT)
5780 return ERR_CAST(inode);
5783 return d_splice_alias(inode, dentry);
5786 unsigned char btrfs_filetype_table[] = {
5787 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5791 * All this infrastructure exists because dir_emit can fault, and we are holding
5792 * the tree lock when doing readdir. For now just allocate a buffer and copy
5793 * our information into that, and then dir_emit from the buffer. This is
5794 * similar to what NFS does, only we don't keep the buffer around in pagecache
5795 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5796 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5799 static int btrfs_opendir(struct inode *inode, struct file *file)
5801 struct btrfs_file_private *private;
5803 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5806 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5807 if (!private->filldir_buf) {
5811 file->private_data = private;
5822 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5825 struct dir_entry *entry = addr;
5826 char *name = (char *)(entry + 1);
5828 ctx->pos = get_unaligned(&entry->offset);
5829 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5830 get_unaligned(&entry->ino),
5831 get_unaligned(&entry->type)))
5833 addr += sizeof(struct dir_entry) +
5834 get_unaligned(&entry->name_len);
5840 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5842 struct inode *inode = file_inode(file);
5843 struct btrfs_root *root = BTRFS_I(inode)->root;
5844 struct btrfs_file_private *private = file->private_data;
5845 struct btrfs_dir_item *di;
5846 struct btrfs_key key;
5847 struct btrfs_key found_key;
5848 struct btrfs_path *path;
5850 struct list_head ins_list;
5851 struct list_head del_list;
5853 struct extent_buffer *leaf;
5860 struct btrfs_key location;
5862 if (!dir_emit_dots(file, ctx))
5865 path = btrfs_alloc_path();
5869 addr = private->filldir_buf;
5870 path->reada = READA_FORWARD;
5872 INIT_LIST_HEAD(&ins_list);
5873 INIT_LIST_HEAD(&del_list);
5874 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5877 key.type = BTRFS_DIR_INDEX_KEY;
5878 key.offset = ctx->pos;
5879 key.objectid = btrfs_ino(BTRFS_I(inode));
5881 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5886 struct dir_entry *entry;
5888 leaf = path->nodes[0];
5889 slot = path->slots[0];
5890 if (slot >= btrfs_header_nritems(leaf)) {
5891 ret = btrfs_next_leaf(root, path);
5899 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5901 if (found_key.objectid != key.objectid)
5903 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5905 if (found_key.offset < ctx->pos)
5907 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5909 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5910 name_len = btrfs_dir_name_len(leaf, di);
5911 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5913 btrfs_release_path(path);
5914 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5917 addr = private->filldir_buf;
5924 put_unaligned(name_len, &entry->name_len);
5925 name_ptr = (char *)(entry + 1);
5926 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5928 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5930 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5931 put_unaligned(location.objectid, &entry->ino);
5932 put_unaligned(found_key.offset, &entry->offset);
5934 addr += sizeof(struct dir_entry) + name_len;
5935 total_len += sizeof(struct dir_entry) + name_len;
5939 btrfs_release_path(path);
5941 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5945 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5950 * Stop new entries from being returned after we return the last
5953 * New directory entries are assigned a strictly increasing
5954 * offset. This means that new entries created during readdir
5955 * are *guaranteed* to be seen in the future by that readdir.
5956 * This has broken buggy programs which operate on names as
5957 * they're returned by readdir. Until we re-use freed offsets
5958 * we have this hack to stop new entries from being returned
5959 * under the assumption that they'll never reach this huge
5962 * This is being careful not to overflow 32bit loff_t unless the
5963 * last entry requires it because doing so has broken 32bit apps
5966 if (ctx->pos >= INT_MAX)
5967 ctx->pos = LLONG_MAX;
5974 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5975 btrfs_free_path(path);
5980 * This is somewhat expensive, updating the tree every time the
5981 * inode changes. But, it is most likely to find the inode in cache.
5982 * FIXME, needs more benchmarking...there are no reasons other than performance
5983 * to keep or drop this code.
5985 static int btrfs_dirty_inode(struct inode *inode)
5987 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5988 struct btrfs_root *root = BTRFS_I(inode)->root;
5989 struct btrfs_trans_handle *trans;
5992 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5995 trans = btrfs_join_transaction(root);
5997 return PTR_ERR(trans);
5999 ret = btrfs_update_inode(trans, root, inode);
6000 if (ret && ret == -ENOSPC) {
6001 /* whoops, lets try again with the full transaction */
6002 btrfs_end_transaction(trans);
6003 trans = btrfs_start_transaction(root, 1);
6005 return PTR_ERR(trans);
6007 ret = btrfs_update_inode(trans, root, inode);
6009 btrfs_end_transaction(trans);
6010 if (BTRFS_I(inode)->delayed_node)
6011 btrfs_balance_delayed_items(fs_info);
6017 * This is a copy of file_update_time. We need this so we can return error on
6018 * ENOSPC for updating the inode in the case of file write and mmap writes.
6020 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6023 struct btrfs_root *root = BTRFS_I(inode)->root;
6024 bool dirty = flags & ~S_VERSION;
6026 if (btrfs_root_readonly(root))
6029 if (flags & S_VERSION)
6030 dirty |= inode_maybe_inc_iversion(inode, dirty);
6031 if (flags & S_CTIME)
6032 inode->i_ctime = *now;
6033 if (flags & S_MTIME)
6034 inode->i_mtime = *now;
6035 if (flags & S_ATIME)
6036 inode->i_atime = *now;
6037 return dirty ? btrfs_dirty_inode(inode) : 0;
6041 * find the highest existing sequence number in a directory
6042 * and then set the in-memory index_cnt variable to reflect
6043 * free sequence numbers
6045 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6047 struct btrfs_root *root = inode->root;
6048 struct btrfs_key key, found_key;
6049 struct btrfs_path *path;
6050 struct extent_buffer *leaf;
6053 key.objectid = btrfs_ino(inode);
6054 key.type = BTRFS_DIR_INDEX_KEY;
6055 key.offset = (u64)-1;
6057 path = btrfs_alloc_path();
6061 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6064 /* FIXME: we should be able to handle this */
6070 * MAGIC NUMBER EXPLANATION:
6071 * since we search a directory based on f_pos we have to start at 2
6072 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6073 * else has to start at 2
6075 if (path->slots[0] == 0) {
6076 inode->index_cnt = 2;
6082 leaf = path->nodes[0];
6083 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6085 if (found_key.objectid != btrfs_ino(inode) ||
6086 found_key.type != BTRFS_DIR_INDEX_KEY) {
6087 inode->index_cnt = 2;
6091 inode->index_cnt = found_key.offset + 1;
6093 btrfs_free_path(path);
6098 * helper to find a free sequence number in a given directory. This current
6099 * code is very simple, later versions will do smarter things in the btree
6101 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6105 if (dir->index_cnt == (u64)-1) {
6106 ret = btrfs_inode_delayed_dir_index_count(dir);
6108 ret = btrfs_set_inode_index_count(dir);
6114 *index = dir->index_cnt;
6120 static int btrfs_insert_inode_locked(struct inode *inode)
6122 struct btrfs_iget_args args;
6123 args.location = &BTRFS_I(inode)->location;
6124 args.root = BTRFS_I(inode)->root;
6126 return insert_inode_locked4(inode,
6127 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6128 btrfs_find_actor, &args);
6132 * Inherit flags from the parent inode.
6134 * Currently only the compression flags and the cow flags are inherited.
6136 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6143 flags = BTRFS_I(dir)->flags;
6145 if (flags & BTRFS_INODE_NOCOMPRESS) {
6146 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6147 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6148 } else if (flags & BTRFS_INODE_COMPRESS) {
6149 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6150 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6153 if (flags & BTRFS_INODE_NODATACOW) {
6154 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6155 if (S_ISREG(inode->i_mode))
6156 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6159 btrfs_sync_inode_flags_to_i_flags(inode);
6162 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6163 struct btrfs_root *root,
6165 const char *name, int name_len,
6166 u64 ref_objectid, u64 objectid,
6167 umode_t mode, u64 *index)
6169 struct btrfs_fs_info *fs_info = root->fs_info;
6170 struct inode *inode;
6171 struct btrfs_inode_item *inode_item;
6172 struct btrfs_key *location;
6173 struct btrfs_path *path;
6174 struct btrfs_inode_ref *ref;
6175 struct btrfs_key key[2];
6177 int nitems = name ? 2 : 1;
6181 path = btrfs_alloc_path();
6183 return ERR_PTR(-ENOMEM);
6185 inode = new_inode(fs_info->sb);
6187 btrfs_free_path(path);
6188 return ERR_PTR(-ENOMEM);
6192 * O_TMPFILE, set link count to 0, so that after this point,
6193 * we fill in an inode item with the correct link count.
6196 set_nlink(inode, 0);
6199 * we have to initialize this early, so we can reclaim the inode
6200 * number if we fail afterwards in this function.
6202 inode->i_ino = objectid;
6205 trace_btrfs_inode_request(dir);
6207 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6209 btrfs_free_path(path);
6211 return ERR_PTR(ret);
6217 * index_cnt is ignored for everything but a dir,
6218 * btrfs_set_inode_index_count has an explanation for the magic
6221 BTRFS_I(inode)->index_cnt = 2;
6222 BTRFS_I(inode)->dir_index = *index;
6223 BTRFS_I(inode)->root = root;
6224 BTRFS_I(inode)->generation = trans->transid;
6225 inode->i_generation = BTRFS_I(inode)->generation;
6228 * We could have gotten an inode number from somebody who was fsynced
6229 * and then removed in this same transaction, so let's just set full
6230 * sync since it will be a full sync anyway and this will blow away the
6231 * old info in the log.
6233 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6235 key[0].objectid = objectid;
6236 key[0].type = BTRFS_INODE_ITEM_KEY;
6239 sizes[0] = sizeof(struct btrfs_inode_item);
6243 * Start new inodes with an inode_ref. This is slightly more
6244 * efficient for small numbers of hard links since they will
6245 * be packed into one item. Extended refs will kick in if we
6246 * add more hard links than can fit in the ref item.
6248 key[1].objectid = objectid;
6249 key[1].type = BTRFS_INODE_REF_KEY;
6250 key[1].offset = ref_objectid;
6252 sizes[1] = name_len + sizeof(*ref);
6255 location = &BTRFS_I(inode)->location;
6256 location->objectid = objectid;
6257 location->offset = 0;
6258 location->type = BTRFS_INODE_ITEM_KEY;
6260 ret = btrfs_insert_inode_locked(inode);
6266 path->leave_spinning = 1;
6267 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6271 inode_init_owner(inode, dir, mode);
6272 inode_set_bytes(inode, 0);
6274 inode->i_mtime = current_time(inode);
6275 inode->i_atime = inode->i_mtime;
6276 inode->i_ctime = inode->i_mtime;
6277 BTRFS_I(inode)->i_otime = inode->i_mtime;
6279 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6280 struct btrfs_inode_item);
6281 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6282 sizeof(*inode_item));
6283 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6286 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6287 struct btrfs_inode_ref);
6288 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6289 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6290 ptr = (unsigned long)(ref + 1);
6291 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6294 btrfs_mark_buffer_dirty(path->nodes[0]);
6295 btrfs_free_path(path);
6297 btrfs_inherit_iflags(inode, dir);
6299 if (S_ISREG(mode)) {
6300 if (btrfs_test_opt(fs_info, NODATASUM))
6301 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6302 if (btrfs_test_opt(fs_info, NODATACOW))
6303 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6304 BTRFS_INODE_NODATASUM;
6307 inode_tree_add(inode);
6309 trace_btrfs_inode_new(inode);
6310 btrfs_set_inode_last_trans(trans, inode);
6312 btrfs_update_root_times(trans, root);
6314 ret = btrfs_inode_inherit_props(trans, inode, dir);
6317 "error inheriting props for ino %llu (root %llu): %d",
6318 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6323 discard_new_inode(inode);
6326 BTRFS_I(dir)->index_cnt--;
6327 btrfs_free_path(path);
6328 return ERR_PTR(ret);
6331 static inline u8 btrfs_inode_type(struct inode *inode)
6333 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6337 * utility function to add 'inode' into 'parent_inode' with
6338 * a give name and a given sequence number.
6339 * if 'add_backref' is true, also insert a backref from the
6340 * inode to the parent directory.
6342 int btrfs_add_link(struct btrfs_trans_handle *trans,
6343 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6344 const char *name, int name_len, int add_backref, u64 index)
6347 struct btrfs_key key;
6348 struct btrfs_root *root = parent_inode->root;
6349 u64 ino = btrfs_ino(inode);
6350 u64 parent_ino = btrfs_ino(parent_inode);
6352 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6353 memcpy(&key, &inode->root->root_key, sizeof(key));
6356 key.type = BTRFS_INODE_ITEM_KEY;
6360 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6361 ret = btrfs_add_root_ref(trans, key.objectid,
6362 root->root_key.objectid, parent_ino,
6363 index, name, name_len);
6364 } else if (add_backref) {
6365 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6369 /* Nothing to clean up yet */
6373 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6374 btrfs_inode_type(&inode->vfs_inode), index);
6375 if (ret == -EEXIST || ret == -EOVERFLOW)
6378 btrfs_abort_transaction(trans, ret);
6382 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6384 inode_inc_iversion(&parent_inode->vfs_inode);
6385 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6386 current_time(&parent_inode->vfs_inode);
6387 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6389 btrfs_abort_transaction(trans, ret);
6393 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6396 err = btrfs_del_root_ref(trans, key.objectid,
6397 root->root_key.objectid, parent_ino,
6398 &local_index, name, name_len);
6400 } else if (add_backref) {
6404 err = btrfs_del_inode_ref(trans, root, name, name_len,
6405 ino, parent_ino, &local_index);
6410 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6411 struct btrfs_inode *dir, struct dentry *dentry,
6412 struct btrfs_inode *inode, int backref, u64 index)
6414 int err = btrfs_add_link(trans, dir, inode,
6415 dentry->d_name.name, dentry->d_name.len,
6422 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6423 umode_t mode, dev_t rdev)
6425 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6426 struct btrfs_trans_handle *trans;
6427 struct btrfs_root *root = BTRFS_I(dir)->root;
6428 struct inode *inode = NULL;
6434 * 2 for inode item and ref
6436 * 1 for xattr if selinux is on
6438 trans = btrfs_start_transaction(root, 5);
6440 return PTR_ERR(trans);
6442 err = btrfs_find_free_ino(root, &objectid);
6446 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6447 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6449 if (IS_ERR(inode)) {
6450 err = PTR_ERR(inode);
6456 * If the active LSM wants to access the inode during
6457 * d_instantiate it needs these. Smack checks to see
6458 * if the filesystem supports xattrs by looking at the
6461 inode->i_op = &btrfs_special_inode_operations;
6462 init_special_inode(inode, inode->i_mode, rdev);
6464 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6468 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6473 btrfs_update_inode(trans, root, inode);
6474 d_instantiate_new(dentry, inode);
6477 btrfs_end_transaction(trans);
6478 btrfs_btree_balance_dirty(fs_info);
6480 inode_dec_link_count(inode);
6481 discard_new_inode(inode);
6486 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6487 umode_t mode, bool excl)
6489 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6490 struct btrfs_trans_handle *trans;
6491 struct btrfs_root *root = BTRFS_I(dir)->root;
6492 struct inode *inode = NULL;
6498 * 2 for inode item and ref
6500 * 1 for xattr if selinux is on
6502 trans = btrfs_start_transaction(root, 5);
6504 return PTR_ERR(trans);
6506 err = btrfs_find_free_ino(root, &objectid);
6510 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6511 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6513 if (IS_ERR(inode)) {
6514 err = PTR_ERR(inode);
6519 * If the active LSM wants to access the inode during
6520 * d_instantiate it needs these. Smack checks to see
6521 * if the filesystem supports xattrs by looking at the
6524 inode->i_fop = &btrfs_file_operations;
6525 inode->i_op = &btrfs_file_inode_operations;
6526 inode->i_mapping->a_ops = &btrfs_aops;
6528 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6532 err = btrfs_update_inode(trans, root, inode);
6536 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6541 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6542 d_instantiate_new(dentry, inode);
6545 btrfs_end_transaction(trans);
6547 inode_dec_link_count(inode);
6548 discard_new_inode(inode);
6550 btrfs_btree_balance_dirty(fs_info);
6554 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6555 struct dentry *dentry)
6557 struct btrfs_trans_handle *trans = NULL;
6558 struct btrfs_root *root = BTRFS_I(dir)->root;
6559 struct inode *inode = d_inode(old_dentry);
6560 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6565 /* do not allow sys_link's with other subvols of the same device */
6566 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6569 if (inode->i_nlink >= BTRFS_LINK_MAX)
6572 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6577 * 2 items for inode and inode ref
6578 * 2 items for dir items
6579 * 1 item for parent inode
6580 * 1 item for orphan item deletion if O_TMPFILE
6582 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6583 if (IS_ERR(trans)) {
6584 err = PTR_ERR(trans);
6589 /* There are several dir indexes for this inode, clear the cache. */
6590 BTRFS_I(inode)->dir_index = 0ULL;
6592 inode_inc_iversion(inode);
6593 inode->i_ctime = current_time(inode);
6595 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6597 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6603 struct dentry *parent = dentry->d_parent;
6606 err = btrfs_update_inode(trans, root, inode);
6609 if (inode->i_nlink == 1) {
6611 * If new hard link count is 1, it's a file created
6612 * with open(2) O_TMPFILE flag.
6614 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6618 d_instantiate(dentry, inode);
6619 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6621 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6622 err = btrfs_commit_transaction(trans);
6629 btrfs_end_transaction(trans);
6631 inode_dec_link_count(inode);
6634 btrfs_btree_balance_dirty(fs_info);
6638 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6640 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6641 struct inode *inode = NULL;
6642 struct btrfs_trans_handle *trans;
6643 struct btrfs_root *root = BTRFS_I(dir)->root;
6645 int drop_on_err = 0;
6650 * 2 items for inode and ref
6651 * 2 items for dir items
6652 * 1 for xattr if selinux is on
6654 trans = btrfs_start_transaction(root, 5);
6656 return PTR_ERR(trans);
6658 err = btrfs_find_free_ino(root, &objectid);
6662 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6663 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6664 S_IFDIR | mode, &index);
6665 if (IS_ERR(inode)) {
6666 err = PTR_ERR(inode);
6672 /* these must be set before we unlock the inode */
6673 inode->i_op = &btrfs_dir_inode_operations;
6674 inode->i_fop = &btrfs_dir_file_operations;
6676 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6680 btrfs_i_size_write(BTRFS_I(inode), 0);
6681 err = btrfs_update_inode(trans, root, inode);
6685 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6686 dentry->d_name.name,
6687 dentry->d_name.len, 0, index);
6691 d_instantiate_new(dentry, inode);
6695 btrfs_end_transaction(trans);
6697 inode_dec_link_count(inode);
6698 discard_new_inode(inode);
6700 btrfs_btree_balance_dirty(fs_info);
6704 static noinline int uncompress_inline(struct btrfs_path *path,
6706 size_t pg_offset, u64 extent_offset,
6707 struct btrfs_file_extent_item *item)
6710 struct extent_buffer *leaf = path->nodes[0];
6713 unsigned long inline_size;
6717 WARN_ON(pg_offset != 0);
6718 compress_type = btrfs_file_extent_compression(leaf, item);
6719 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6720 inline_size = btrfs_file_extent_inline_item_len(leaf,
6721 btrfs_item_nr(path->slots[0]));
6722 tmp = kmalloc(inline_size, GFP_NOFS);
6725 ptr = btrfs_file_extent_inline_start(item);
6727 read_extent_buffer(leaf, tmp, ptr, inline_size);
6729 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6730 ret = btrfs_decompress(compress_type, tmp, page,
6731 extent_offset, inline_size, max_size);
6734 * decompression code contains a memset to fill in any space between the end
6735 * of the uncompressed data and the end of max_size in case the decompressed
6736 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6737 * the end of an inline extent and the beginning of the next block, so we
6738 * cover that region here.
6741 if (max_size + pg_offset < PAGE_SIZE) {
6742 char *map = kmap(page);
6743 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6751 * a bit scary, this does extent mapping from logical file offset to the disk.
6752 * the ugly parts come from merging extents from the disk with the in-ram
6753 * representation. This gets more complex because of the data=ordered code,
6754 * where the in-ram extents might be locked pending data=ordered completion.
6756 * This also copies inline extents directly into the page.
6758 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6760 size_t pg_offset, u64 start, u64 len,
6763 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6766 u64 extent_start = 0;
6768 u64 objectid = btrfs_ino(inode);
6770 struct btrfs_path *path = NULL;
6771 struct btrfs_root *root = inode->root;
6772 struct btrfs_file_extent_item *item;
6773 struct extent_buffer *leaf;
6774 struct btrfs_key found_key;
6775 struct extent_map *em = NULL;
6776 struct extent_map_tree *em_tree = &inode->extent_tree;
6777 struct extent_io_tree *io_tree = &inode->io_tree;
6778 const bool new_inline = !page || create;
6780 read_lock(&em_tree->lock);
6781 em = lookup_extent_mapping(em_tree, start, len);
6783 em->bdev = fs_info->fs_devices->latest_bdev;
6784 read_unlock(&em_tree->lock);
6787 if (em->start > start || em->start + em->len <= start)
6788 free_extent_map(em);
6789 else if (em->block_start == EXTENT_MAP_INLINE && page)
6790 free_extent_map(em);
6794 em = alloc_extent_map();
6799 em->bdev = fs_info->fs_devices->latest_bdev;
6800 em->start = EXTENT_MAP_HOLE;
6801 em->orig_start = EXTENT_MAP_HOLE;
6803 em->block_len = (u64)-1;
6805 path = btrfs_alloc_path();
6811 /* Chances are we'll be called again, so go ahead and do readahead */
6812 path->reada = READA_FORWARD;
6815 * Unless we're going to uncompress the inline extent, no sleep would
6818 path->leave_spinning = 1;
6820 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6827 if (path->slots[0] == 0)
6832 leaf = path->nodes[0];
6833 item = btrfs_item_ptr(leaf, path->slots[0],
6834 struct btrfs_file_extent_item);
6835 /* are we inside the extent that was found? */
6836 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6837 found_type = found_key.type;
6838 if (found_key.objectid != objectid ||
6839 found_type != BTRFS_EXTENT_DATA_KEY) {
6841 * If we backup past the first extent we want to move forward
6842 * and see if there is an extent in front of us, otherwise we'll
6843 * say there is a hole for our whole search range which can
6850 found_type = btrfs_file_extent_type(leaf, item);
6851 extent_start = found_key.offset;
6852 if (found_type == BTRFS_FILE_EXTENT_REG ||
6853 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6854 extent_end = extent_start +
6855 btrfs_file_extent_num_bytes(leaf, item);
6857 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6859 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6862 size = btrfs_file_extent_ram_bytes(leaf, item);
6863 extent_end = ALIGN(extent_start + size,
6864 fs_info->sectorsize);
6866 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6871 if (start >= extent_end) {
6873 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6874 ret = btrfs_next_leaf(root, path);
6881 leaf = path->nodes[0];
6883 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6884 if (found_key.objectid != objectid ||
6885 found_key.type != BTRFS_EXTENT_DATA_KEY)
6887 if (start + len <= found_key.offset)
6889 if (start > found_key.offset)
6892 em->orig_start = start;
6893 em->len = found_key.offset - start;
6897 btrfs_extent_item_to_extent_map(inode, path, item,
6900 if (found_type == BTRFS_FILE_EXTENT_REG ||
6901 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6903 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6907 size_t extent_offset;
6913 size = btrfs_file_extent_ram_bytes(leaf, item);
6914 extent_offset = page_offset(page) + pg_offset - extent_start;
6915 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6916 size - extent_offset);
6917 em->start = extent_start + extent_offset;
6918 em->len = ALIGN(copy_size, fs_info->sectorsize);
6919 em->orig_block_len = em->len;
6920 em->orig_start = em->start;
6921 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6923 btrfs_set_path_blocking(path);
6924 if (!PageUptodate(page)) {
6925 if (btrfs_file_extent_compression(leaf, item) !=
6926 BTRFS_COMPRESS_NONE) {
6927 ret = uncompress_inline(path, page, pg_offset,
6928 extent_offset, item);
6935 read_extent_buffer(leaf, map + pg_offset, ptr,
6937 if (pg_offset + copy_size < PAGE_SIZE) {
6938 memset(map + pg_offset + copy_size, 0,
6939 PAGE_SIZE - pg_offset -
6944 flush_dcache_page(page);
6946 set_extent_uptodate(io_tree, em->start,
6947 extent_map_end(em) - 1, NULL, GFP_NOFS);
6952 em->orig_start = start;
6955 em->block_start = EXTENT_MAP_HOLE;
6957 btrfs_release_path(path);
6958 if (em->start > start || extent_map_end(em) <= start) {
6960 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6961 em->start, em->len, start, len);
6967 write_lock(&em_tree->lock);
6968 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6969 write_unlock(&em_tree->lock);
6971 btrfs_free_path(path);
6973 trace_btrfs_get_extent(root, inode, em);
6976 free_extent_map(em);
6977 return ERR_PTR(err);
6979 BUG_ON(!em); /* Error is always set */
6983 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6985 size_t pg_offset, u64 start, u64 len,
6988 struct extent_map *em;
6989 struct extent_map *hole_em = NULL;
6990 u64 range_start = start;
6996 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7000 * If our em maps to:
7002 * - a pre-alloc extent,
7003 * there might actually be delalloc bytes behind it.
7005 if (em->block_start != EXTENT_MAP_HOLE &&
7006 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7011 /* check to see if we've wrapped (len == -1 or similar) */
7020 /* ok, we didn't find anything, lets look for delalloc */
7021 found = count_range_bits(&inode->io_tree, &range_start,
7022 end, len, EXTENT_DELALLOC, 1);
7023 found_end = range_start + found;
7024 if (found_end < range_start)
7025 found_end = (u64)-1;
7028 * we didn't find anything useful, return
7029 * the original results from get_extent()
7031 if (range_start > end || found_end <= start) {
7037 /* adjust the range_start to make sure it doesn't
7038 * go backwards from the start they passed in
7040 range_start = max(start, range_start);
7041 found = found_end - range_start;
7044 u64 hole_start = start;
7047 em = alloc_extent_map();
7053 * when btrfs_get_extent can't find anything it
7054 * returns one huge hole
7056 * make sure what it found really fits our range, and
7057 * adjust to make sure it is based on the start from
7061 u64 calc_end = extent_map_end(hole_em);
7063 if (calc_end <= start || (hole_em->start > end)) {
7064 free_extent_map(hole_em);
7067 hole_start = max(hole_em->start, start);
7068 hole_len = calc_end - hole_start;
7072 if (hole_em && range_start > hole_start) {
7073 /* our hole starts before our delalloc, so we
7074 * have to return just the parts of the hole
7075 * that go until the delalloc starts
7077 em->len = min(hole_len,
7078 range_start - hole_start);
7079 em->start = hole_start;
7080 em->orig_start = hole_start;
7082 * don't adjust block start at all,
7083 * it is fixed at EXTENT_MAP_HOLE
7085 em->block_start = hole_em->block_start;
7086 em->block_len = hole_len;
7087 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7088 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7090 em->start = range_start;
7092 em->orig_start = range_start;
7093 em->block_start = EXTENT_MAP_DELALLOC;
7094 em->block_len = found;
7101 free_extent_map(hole_em);
7103 free_extent_map(em);
7104 return ERR_PTR(err);
7109 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7112 const u64 orig_start,
7113 const u64 block_start,
7114 const u64 block_len,
7115 const u64 orig_block_len,
7116 const u64 ram_bytes,
7119 struct extent_map *em = NULL;
7122 if (type != BTRFS_ORDERED_NOCOW) {
7123 em = create_io_em(inode, start, len, orig_start,
7124 block_start, block_len, orig_block_len,
7126 BTRFS_COMPRESS_NONE, /* compress_type */
7131 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7132 len, block_len, type);
7135 free_extent_map(em);
7136 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7137 start + len - 1, 0);
7146 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7149 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7150 struct btrfs_root *root = BTRFS_I(inode)->root;
7151 struct extent_map *em;
7152 struct btrfs_key ins;
7156 alloc_hint = get_extent_allocation_hint(inode, start, len);
7157 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7158 0, alloc_hint, &ins, 1, 1);
7160 return ERR_PTR(ret);
7162 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7163 ins.objectid, ins.offset, ins.offset,
7164 ins.offset, BTRFS_ORDERED_REGULAR);
7165 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7167 btrfs_free_reserved_extent(fs_info, ins.objectid,
7174 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7175 * block must be cow'd
7177 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7178 u64 *orig_start, u64 *orig_block_len,
7181 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7182 struct btrfs_path *path;
7184 struct extent_buffer *leaf;
7185 struct btrfs_root *root = BTRFS_I(inode)->root;
7186 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7187 struct btrfs_file_extent_item *fi;
7188 struct btrfs_key key;
7195 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7197 path = btrfs_alloc_path();
7201 ret = btrfs_lookup_file_extent(NULL, root, path,
7202 btrfs_ino(BTRFS_I(inode)), offset, 0);
7206 slot = path->slots[0];
7209 /* can't find the item, must cow */
7216 leaf = path->nodes[0];
7217 btrfs_item_key_to_cpu(leaf, &key, slot);
7218 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7219 key.type != BTRFS_EXTENT_DATA_KEY) {
7220 /* not our file or wrong item type, must cow */
7224 if (key.offset > offset) {
7225 /* Wrong offset, must cow */
7229 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7230 found_type = btrfs_file_extent_type(leaf, fi);
7231 if (found_type != BTRFS_FILE_EXTENT_REG &&
7232 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7233 /* not a regular extent, must cow */
7237 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7240 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7241 if (extent_end <= offset)
7244 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7245 if (disk_bytenr == 0)
7248 if (btrfs_file_extent_compression(leaf, fi) ||
7249 btrfs_file_extent_encryption(leaf, fi) ||
7250 btrfs_file_extent_other_encoding(leaf, fi))
7254 * Do the same check as in btrfs_cross_ref_exist but without the
7255 * unnecessary search.
7257 if (btrfs_file_extent_generation(leaf, fi) <=
7258 btrfs_root_last_snapshot(&root->root_item))
7261 backref_offset = btrfs_file_extent_offset(leaf, fi);
7264 *orig_start = key.offset - backref_offset;
7265 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7266 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7269 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7272 num_bytes = min(offset + *len, extent_end) - offset;
7273 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7276 range_end = round_up(offset + num_bytes,
7277 root->fs_info->sectorsize) - 1;
7278 ret = test_range_bit(io_tree, offset, range_end,
7279 EXTENT_DELALLOC, 0, NULL);
7286 btrfs_release_path(path);
7289 * look for other files referencing this extent, if we
7290 * find any we must cow
7293 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7294 key.offset - backref_offset, disk_bytenr);
7301 * adjust disk_bytenr and num_bytes to cover just the bytes
7302 * in this extent we are about to write. If there
7303 * are any csums in that range we have to cow in order
7304 * to keep the csums correct
7306 disk_bytenr += backref_offset;
7307 disk_bytenr += offset - key.offset;
7308 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7311 * all of the above have passed, it is safe to overwrite this extent
7317 btrfs_free_path(path);
7321 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7322 struct extent_state **cached_state, int writing)
7324 struct btrfs_ordered_extent *ordered;
7328 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7331 * We're concerned with the entire range that we're going to be
7332 * doing DIO to, so we need to make sure there's no ordered
7333 * extents in this range.
7335 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7336 lockend - lockstart + 1);
7339 * We need to make sure there are no buffered pages in this
7340 * range either, we could have raced between the invalidate in
7341 * generic_file_direct_write and locking the extent. The
7342 * invalidate needs to happen so that reads after a write do not
7346 (!writing || !filemap_range_has_page(inode->i_mapping,
7347 lockstart, lockend)))
7350 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7355 * If we are doing a DIO read and the ordered extent we
7356 * found is for a buffered write, we can not wait for it
7357 * to complete and retry, because if we do so we can
7358 * deadlock with concurrent buffered writes on page
7359 * locks. This happens only if our DIO read covers more
7360 * than one extent map, if at this point has already
7361 * created an ordered extent for a previous extent map
7362 * and locked its range in the inode's io tree, and a
7363 * concurrent write against that previous extent map's
7364 * range and this range started (we unlock the ranges
7365 * in the io tree only when the bios complete and
7366 * buffered writes always lock pages before attempting
7367 * to lock range in the io tree).
7370 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7371 btrfs_start_ordered_extent(inode, ordered, 1);
7374 btrfs_put_ordered_extent(ordered);
7377 * We could trigger writeback for this range (and wait
7378 * for it to complete) and then invalidate the pages for
7379 * this range (through invalidate_inode_pages2_range()),
7380 * but that can lead us to a deadlock with a concurrent
7381 * call to readpages() (a buffered read or a defrag call
7382 * triggered a readahead) on a page lock due to an
7383 * ordered dio extent we created before but did not have
7384 * yet a corresponding bio submitted (whence it can not
7385 * complete), which makes readpages() wait for that
7386 * ordered extent to complete while holding a lock on
7401 /* The callers of this must take lock_extent() */
7402 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7403 u64 orig_start, u64 block_start,
7404 u64 block_len, u64 orig_block_len,
7405 u64 ram_bytes, int compress_type,
7408 struct extent_map_tree *em_tree;
7409 struct extent_map *em;
7410 struct btrfs_root *root = BTRFS_I(inode)->root;
7413 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7414 type == BTRFS_ORDERED_COMPRESSED ||
7415 type == BTRFS_ORDERED_NOCOW ||
7416 type == BTRFS_ORDERED_REGULAR);
7418 em_tree = &BTRFS_I(inode)->extent_tree;
7419 em = alloc_extent_map();
7421 return ERR_PTR(-ENOMEM);
7424 em->orig_start = orig_start;
7426 em->block_len = block_len;
7427 em->block_start = block_start;
7428 em->bdev = root->fs_info->fs_devices->latest_bdev;
7429 em->orig_block_len = orig_block_len;
7430 em->ram_bytes = ram_bytes;
7431 em->generation = -1;
7432 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7433 if (type == BTRFS_ORDERED_PREALLOC) {
7434 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7435 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7436 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7437 em->compress_type = compress_type;
7441 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7442 em->start + em->len - 1, 0);
7443 write_lock(&em_tree->lock);
7444 ret = add_extent_mapping(em_tree, em, 1);
7445 write_unlock(&em_tree->lock);
7447 * The caller has taken lock_extent(), who could race with us
7450 } while (ret == -EEXIST);
7453 free_extent_map(em);
7454 return ERR_PTR(ret);
7457 /* em got 2 refs now, callers needs to do free_extent_map once. */
7462 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7463 struct buffer_head *bh_result,
7464 struct inode *inode,
7467 if (em->block_start == EXTENT_MAP_HOLE ||
7468 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7471 len = min(len, em->len - (start - em->start));
7473 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7475 bh_result->b_size = len;
7476 bh_result->b_bdev = em->bdev;
7477 set_buffer_mapped(bh_result);
7482 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7483 struct buffer_head *bh_result,
7484 struct inode *inode,
7485 struct btrfs_dio_data *dio_data,
7488 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7489 struct extent_map *em = *map;
7493 * We don't allocate a new extent in the following cases
7495 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7497 * 2) The extent is marked as PREALLOC. We're good to go here and can
7498 * just use the extent.
7501 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7502 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7503 em->block_start != EXTENT_MAP_HOLE)) {
7505 u64 block_start, orig_start, orig_block_len, ram_bytes;
7507 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7508 type = BTRFS_ORDERED_PREALLOC;
7510 type = BTRFS_ORDERED_NOCOW;
7511 len = min(len, em->len - (start - em->start));
7512 block_start = em->block_start + (start - em->start);
7514 if (can_nocow_extent(inode, start, &len, &orig_start,
7515 &orig_block_len, &ram_bytes) == 1 &&
7516 btrfs_inc_nocow_writers(fs_info, block_start)) {
7517 struct extent_map *em2;
7519 em2 = btrfs_create_dio_extent(inode, start, len,
7520 orig_start, block_start,
7521 len, orig_block_len,
7523 btrfs_dec_nocow_writers(fs_info, block_start);
7524 if (type == BTRFS_ORDERED_PREALLOC) {
7525 free_extent_map(em);
7529 if (em2 && IS_ERR(em2)) {
7534 * For inode marked NODATACOW or extent marked PREALLOC,
7535 * use the existing or preallocated extent, so does not
7536 * need to adjust btrfs_space_info's bytes_may_use.
7538 btrfs_free_reserved_data_space_noquota(inode, start,
7544 /* this will cow the extent */
7545 len = bh_result->b_size;
7546 free_extent_map(em);
7547 *map = em = btrfs_new_extent_direct(inode, start, len);
7553 len = min(len, em->len - (start - em->start));
7556 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7558 bh_result->b_size = len;
7559 bh_result->b_bdev = em->bdev;
7560 set_buffer_mapped(bh_result);
7562 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7563 set_buffer_new(bh_result);
7566 * Need to update the i_size under the extent lock so buffered
7567 * readers will get the updated i_size when we unlock.
7569 if (!dio_data->overwrite && start + len > i_size_read(inode))
7570 i_size_write(inode, start + len);
7572 WARN_ON(dio_data->reserve < len);
7573 dio_data->reserve -= len;
7574 dio_data->unsubmitted_oe_range_end = start + len;
7575 current->journal_info = dio_data;
7580 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7581 struct buffer_head *bh_result, int create)
7583 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7584 struct extent_map *em;
7585 struct extent_state *cached_state = NULL;
7586 struct btrfs_dio_data *dio_data = NULL;
7587 u64 start = iblock << inode->i_blkbits;
7588 u64 lockstart, lockend;
7589 u64 len = bh_result->b_size;
7590 int unlock_bits = EXTENT_LOCKED;
7594 unlock_bits |= EXTENT_DIRTY;
7596 len = min_t(u64, len, fs_info->sectorsize);
7599 lockend = start + len - 1;
7601 if (current->journal_info) {
7603 * Need to pull our outstanding extents and set journal_info to NULL so
7604 * that anything that needs to check if there's a transaction doesn't get
7607 dio_data = current->journal_info;
7608 current->journal_info = NULL;
7612 * If this errors out it's because we couldn't invalidate pagecache for
7613 * this range and we need to fallback to buffered.
7615 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7621 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7628 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7629 * io. INLINE is special, and we could probably kludge it in here, but
7630 * it's still buffered so for safety lets just fall back to the generic
7633 * For COMPRESSED we _have_ to read the entire extent in so we can
7634 * decompress it, so there will be buffering required no matter what we
7635 * do, so go ahead and fallback to buffered.
7637 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7638 * to buffered IO. Don't blame me, this is the price we pay for using
7641 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7642 em->block_start == EXTENT_MAP_INLINE) {
7643 free_extent_map(em);
7649 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7650 dio_data, start, len);
7654 /* clear and unlock the entire range */
7655 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7656 unlock_bits, 1, 0, &cached_state);
7658 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7660 /* Can be negative only if we read from a hole */
7663 free_extent_map(em);
7667 * We need to unlock only the end area that we aren't using.
7668 * The rest is going to be unlocked by the endio routine.
7670 lockstart = start + bh_result->b_size;
7671 if (lockstart < lockend) {
7672 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7673 lockend, unlock_bits, 1, 0,
7676 free_extent_state(cached_state);
7680 free_extent_map(em);
7685 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7686 unlock_bits, 1, 0, &cached_state);
7689 current->journal_info = dio_data;
7693 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7697 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7700 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7702 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7706 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7711 static int btrfs_check_dio_repairable(struct inode *inode,
7712 struct bio *failed_bio,
7713 struct io_failure_record *failrec,
7716 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7719 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7720 if (num_copies == 1) {
7722 * we only have a single copy of the data, so don't bother with
7723 * all the retry and error correction code that follows. no
7724 * matter what the error is, it is very likely to persist.
7726 btrfs_debug(fs_info,
7727 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7728 num_copies, failrec->this_mirror, failed_mirror);
7732 failrec->failed_mirror = failed_mirror;
7733 failrec->this_mirror++;
7734 if (failrec->this_mirror == failed_mirror)
7735 failrec->this_mirror++;
7737 if (failrec->this_mirror > num_copies) {
7738 btrfs_debug(fs_info,
7739 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7740 num_copies, failrec->this_mirror, failed_mirror);
7747 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7748 struct page *page, unsigned int pgoff,
7749 u64 start, u64 end, int failed_mirror,
7750 bio_end_io_t *repair_endio, void *repair_arg)
7752 struct io_failure_record *failrec;
7753 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7754 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7757 unsigned int read_mode = 0;
7760 blk_status_t status;
7761 struct bio_vec bvec;
7763 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7765 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7767 return errno_to_blk_status(ret);
7769 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7772 free_io_failure(failure_tree, io_tree, failrec);
7773 return BLK_STS_IOERR;
7776 segs = bio_segments(failed_bio);
7777 bio_get_first_bvec(failed_bio, &bvec);
7779 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7780 read_mode |= REQ_FAILFAST_DEV;
7782 isector = start - btrfs_io_bio(failed_bio)->logical;
7783 isector >>= inode->i_sb->s_blocksize_bits;
7784 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7785 pgoff, isector, repair_endio, repair_arg);
7786 bio->bi_opf = REQ_OP_READ | read_mode;
7788 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7789 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7790 read_mode, failrec->this_mirror, failrec->in_validation);
7792 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7794 free_io_failure(failure_tree, io_tree, failrec);
7801 struct btrfs_retry_complete {
7802 struct completion done;
7803 struct inode *inode;
7808 static void btrfs_retry_endio_nocsum(struct bio *bio)
7810 struct btrfs_retry_complete *done = bio->bi_private;
7811 struct inode *inode = done->inode;
7812 struct bio_vec *bvec;
7813 struct extent_io_tree *io_tree, *failure_tree;
7819 ASSERT(bio->bi_vcnt == 1);
7820 io_tree = &BTRFS_I(inode)->io_tree;
7821 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7822 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7825 ASSERT(!bio_flagged(bio, BIO_CLONED));
7826 bio_for_each_segment_all(bvec, bio, i)
7827 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7828 io_tree, done->start, bvec->bv_page,
7829 btrfs_ino(BTRFS_I(inode)), 0);
7831 complete(&done->done);
7835 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7836 struct btrfs_io_bio *io_bio)
7838 struct btrfs_fs_info *fs_info;
7839 struct bio_vec bvec;
7840 struct bvec_iter iter;
7841 struct btrfs_retry_complete done;
7847 blk_status_t err = BLK_STS_OK;
7849 fs_info = BTRFS_I(inode)->root->fs_info;
7850 sectorsize = fs_info->sectorsize;
7852 start = io_bio->logical;
7854 io_bio->bio.bi_iter = io_bio->iter;
7856 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7857 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7858 pgoff = bvec.bv_offset;
7860 next_block_or_try_again:
7863 init_completion(&done.done);
7865 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7866 pgoff, start, start + sectorsize - 1,
7868 btrfs_retry_endio_nocsum, &done);
7874 wait_for_completion_io(&done.done);
7876 if (!done.uptodate) {
7877 /* We might have another mirror, so try again */
7878 goto next_block_or_try_again;
7882 start += sectorsize;
7886 pgoff += sectorsize;
7887 ASSERT(pgoff < PAGE_SIZE);
7888 goto next_block_or_try_again;
7895 static void btrfs_retry_endio(struct bio *bio)
7897 struct btrfs_retry_complete *done = bio->bi_private;
7898 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7899 struct extent_io_tree *io_tree, *failure_tree;
7900 struct inode *inode = done->inode;
7901 struct bio_vec *bvec;
7911 ASSERT(bio->bi_vcnt == 1);
7912 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7914 io_tree = &BTRFS_I(inode)->io_tree;
7915 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7917 ASSERT(!bio_flagged(bio, BIO_CLONED));
7918 bio_for_each_segment_all(bvec, bio, i) {
7919 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7920 bvec->bv_offset, done->start,
7923 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7924 failure_tree, io_tree, done->start,
7926 btrfs_ino(BTRFS_I(inode)),
7932 done->uptodate = uptodate;
7934 complete(&done->done);
7938 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7939 struct btrfs_io_bio *io_bio, blk_status_t err)
7941 struct btrfs_fs_info *fs_info;
7942 struct bio_vec bvec;
7943 struct bvec_iter iter;
7944 struct btrfs_retry_complete done;
7951 bool uptodate = (err == 0);
7953 blk_status_t status;
7955 fs_info = BTRFS_I(inode)->root->fs_info;
7956 sectorsize = fs_info->sectorsize;
7959 start = io_bio->logical;
7961 io_bio->bio.bi_iter = io_bio->iter;
7963 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7964 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7966 pgoff = bvec.bv_offset;
7969 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7970 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7971 bvec.bv_page, pgoff, start, sectorsize);
7978 init_completion(&done.done);
7980 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7981 pgoff, start, start + sectorsize - 1,
7982 io_bio->mirror_num, btrfs_retry_endio,
7989 wait_for_completion_io(&done.done);
7991 if (!done.uptodate) {
7992 /* We might have another mirror, so try again */
7996 offset += sectorsize;
7997 start += sectorsize;
8003 pgoff += sectorsize;
8004 ASSERT(pgoff < PAGE_SIZE);
8012 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8013 struct btrfs_io_bio *io_bio, blk_status_t err)
8015 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8019 return __btrfs_correct_data_nocsum(inode, io_bio);
8023 return __btrfs_subio_endio_read(inode, io_bio, err);
8027 static void btrfs_endio_direct_read(struct bio *bio)
8029 struct btrfs_dio_private *dip = bio->bi_private;
8030 struct inode *inode = dip->inode;
8031 struct bio *dio_bio;
8032 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8033 blk_status_t err = bio->bi_status;
8035 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8036 err = btrfs_subio_endio_read(inode, io_bio, err);
8038 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8039 dip->logical_offset + dip->bytes - 1);
8040 dio_bio = dip->dio_bio;
8044 dio_bio->bi_status = err;
8045 dio_end_io(dio_bio);
8048 io_bio->end_io(io_bio, blk_status_to_errno(err));
8052 static void __endio_write_update_ordered(struct inode *inode,
8053 const u64 offset, const u64 bytes,
8054 const bool uptodate)
8056 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8057 struct btrfs_ordered_extent *ordered = NULL;
8058 struct btrfs_workqueue *wq;
8059 btrfs_work_func_t func;
8060 u64 ordered_offset = offset;
8061 u64 ordered_bytes = bytes;
8064 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8065 wq = fs_info->endio_freespace_worker;
8066 func = btrfs_freespace_write_helper;
8068 wq = fs_info->endio_write_workers;
8069 func = btrfs_endio_write_helper;
8072 while (ordered_offset < offset + bytes) {
8073 last_offset = ordered_offset;
8074 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8078 btrfs_init_work(&ordered->work, func,
8081 btrfs_queue_work(wq, &ordered->work);
8084 * If btrfs_dec_test_ordered_pending does not find any ordered
8085 * extent in the range, we can exit.
8087 if (ordered_offset == last_offset)
8090 * Our bio might span multiple ordered extents. In this case
8091 * we keep goin until we have accounted the whole dio.
8093 if (ordered_offset < offset + bytes) {
8094 ordered_bytes = offset + bytes - ordered_offset;
8100 static void btrfs_endio_direct_write(struct bio *bio)
8102 struct btrfs_dio_private *dip = bio->bi_private;
8103 struct bio *dio_bio = dip->dio_bio;
8105 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8106 dip->bytes, !bio->bi_status);
8110 dio_bio->bi_status = bio->bi_status;
8111 dio_end_io(dio_bio);
8115 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8116 struct bio *bio, u64 offset)
8118 struct inode *inode = private_data;
8120 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8121 BUG_ON(ret); /* -ENOMEM */
8125 static void btrfs_end_dio_bio(struct bio *bio)
8127 struct btrfs_dio_private *dip = bio->bi_private;
8128 blk_status_t err = bio->bi_status;
8131 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8132 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8133 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8135 (unsigned long long)bio->bi_iter.bi_sector,
8136 bio->bi_iter.bi_size, err);
8138 if (dip->subio_endio)
8139 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8143 * We want to perceive the errors flag being set before
8144 * decrementing the reference count. We don't need a barrier
8145 * since atomic operations with a return value are fully
8146 * ordered as per atomic_t.txt
8151 /* if there are more bios still pending for this dio, just exit */
8152 if (!atomic_dec_and_test(&dip->pending_bios))
8156 bio_io_error(dip->orig_bio);
8158 dip->dio_bio->bi_status = BLK_STS_OK;
8159 bio_endio(dip->orig_bio);
8165 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8166 struct btrfs_dio_private *dip,
8170 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8171 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8175 * We load all the csum data we need when we submit
8176 * the first bio to reduce the csum tree search and
8179 if (dip->logical_offset == file_offset) {
8180 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8186 if (bio == dip->orig_bio)
8189 file_offset -= dip->logical_offset;
8190 file_offset >>= inode->i_sb->s_blocksize_bits;
8191 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8196 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8197 struct inode *inode, u64 file_offset, int async_submit)
8199 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8200 struct btrfs_dio_private *dip = bio->bi_private;
8201 bool write = bio_op(bio) == REQ_OP_WRITE;
8204 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8206 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8209 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8214 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8217 if (write && async_submit) {
8218 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8220 btrfs_submit_bio_start_direct_io);
8224 * If we aren't doing async submit, calculate the csum of the
8227 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8231 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8237 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8242 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8244 struct inode *inode = dip->inode;
8245 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8247 struct bio *orig_bio = dip->orig_bio;
8248 u64 start_sector = orig_bio->bi_iter.bi_sector;
8249 u64 file_offset = dip->logical_offset;
8251 int async_submit = 0;
8253 int clone_offset = 0;
8256 blk_status_t status;
8258 map_length = orig_bio->bi_iter.bi_size;
8259 submit_len = map_length;
8260 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8261 &map_length, NULL, 0);
8265 if (map_length >= submit_len) {
8267 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8271 /* async crcs make it difficult to collect full stripe writes. */
8272 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8278 ASSERT(map_length <= INT_MAX);
8279 atomic_inc(&dip->pending_bios);
8281 clone_len = min_t(int, submit_len, map_length);
8284 * This will never fail as it's passing GPF_NOFS and
8285 * the allocation is backed by btrfs_bioset.
8287 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8289 bio->bi_private = dip;
8290 bio->bi_end_io = btrfs_end_dio_bio;
8291 btrfs_io_bio(bio)->logical = file_offset;
8293 ASSERT(submit_len >= clone_len);
8294 submit_len -= clone_len;
8295 if (submit_len == 0)
8299 * Increase the count before we submit the bio so we know
8300 * the end IO handler won't happen before we increase the
8301 * count. Otherwise, the dip might get freed before we're
8302 * done setting it up.
8304 atomic_inc(&dip->pending_bios);
8306 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8310 atomic_dec(&dip->pending_bios);
8314 clone_offset += clone_len;
8315 start_sector += clone_len >> 9;
8316 file_offset += clone_len;
8318 map_length = submit_len;
8319 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8320 start_sector << 9, &map_length, NULL, 0);
8323 } while (submit_len > 0);
8326 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8334 * Before atomic variable goto zero, we must make sure dip->errors is
8335 * perceived to be set. This ordering is ensured by the fact that an
8336 * atomic operations with a return value are fully ordered as per
8339 if (atomic_dec_and_test(&dip->pending_bios))
8340 bio_io_error(dip->orig_bio);
8342 /* bio_end_io() will handle error, so we needn't return it */
8346 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8349 struct btrfs_dio_private *dip = NULL;
8350 struct bio *bio = NULL;
8351 struct btrfs_io_bio *io_bio;
8352 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8355 bio = btrfs_bio_clone(dio_bio);
8357 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8363 dip->private = dio_bio->bi_private;
8365 dip->logical_offset = file_offset;
8366 dip->bytes = dio_bio->bi_iter.bi_size;
8367 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8368 bio->bi_private = dip;
8369 dip->orig_bio = bio;
8370 dip->dio_bio = dio_bio;
8371 atomic_set(&dip->pending_bios, 0);
8372 io_bio = btrfs_io_bio(bio);
8373 io_bio->logical = file_offset;
8376 bio->bi_end_io = btrfs_endio_direct_write;
8378 bio->bi_end_io = btrfs_endio_direct_read;
8379 dip->subio_endio = btrfs_subio_endio_read;
8383 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8384 * even if we fail to submit a bio, because in such case we do the
8385 * corresponding error handling below and it must not be done a second
8386 * time by btrfs_direct_IO().
8389 struct btrfs_dio_data *dio_data = current->journal_info;
8391 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8393 dio_data->unsubmitted_oe_range_start =
8394 dio_data->unsubmitted_oe_range_end;
8397 ret = btrfs_submit_direct_hook(dip);
8402 io_bio->end_io(io_bio, ret);
8406 * If we arrived here it means either we failed to submit the dip
8407 * or we either failed to clone the dio_bio or failed to allocate the
8408 * dip. If we cloned the dio_bio and allocated the dip, we can just
8409 * call bio_endio against our io_bio so that we get proper resource
8410 * cleanup if we fail to submit the dip, otherwise, we must do the
8411 * same as btrfs_endio_direct_[write|read] because we can't call these
8412 * callbacks - they require an allocated dip and a clone of dio_bio.
8417 * The end io callbacks free our dip, do the final put on bio
8418 * and all the cleanup and final put for dio_bio (through
8425 __endio_write_update_ordered(inode,
8427 dio_bio->bi_iter.bi_size,
8430 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8431 file_offset + dio_bio->bi_iter.bi_size - 1);
8433 dio_bio->bi_status = BLK_STS_IOERR;
8435 * Releases and cleans up our dio_bio, no need to bio_put()
8436 * nor bio_endio()/bio_io_error() against dio_bio.
8438 dio_end_io(dio_bio);
8445 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8446 const struct iov_iter *iter, loff_t offset)
8450 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8451 ssize_t retval = -EINVAL;
8453 if (offset & blocksize_mask)
8456 if (iov_iter_alignment(iter) & blocksize_mask)
8459 /* If this is a write we don't need to check anymore */
8460 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8463 * Check to make sure we don't have duplicate iov_base's in this
8464 * iovec, if so return EINVAL, otherwise we'll get csum errors
8465 * when reading back.
8467 for (seg = 0; seg < iter->nr_segs; seg++) {
8468 for (i = seg + 1; i < iter->nr_segs; i++) {
8469 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8478 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8480 struct file *file = iocb->ki_filp;
8481 struct inode *inode = file->f_mapping->host;
8482 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8483 struct btrfs_dio_data dio_data = { 0 };
8484 struct extent_changeset *data_reserved = NULL;
8485 loff_t offset = iocb->ki_pos;
8489 bool relock = false;
8492 if (check_direct_IO(fs_info, iter, offset))
8495 inode_dio_begin(inode);
8498 * The generic stuff only does filemap_write_and_wait_range, which
8499 * isn't enough if we've written compressed pages to this area, so
8500 * we need to flush the dirty pages again to make absolutely sure
8501 * that any outstanding dirty pages are on disk.
8503 count = iov_iter_count(iter);
8504 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8505 &BTRFS_I(inode)->runtime_flags))
8506 filemap_fdatawrite_range(inode->i_mapping, offset,
8507 offset + count - 1);
8509 if (iov_iter_rw(iter) == WRITE) {
8511 * If the write DIO is beyond the EOF, we need update
8512 * the isize, but it is protected by i_mutex. So we can
8513 * not unlock the i_mutex at this case.
8515 if (offset + count <= inode->i_size) {
8516 dio_data.overwrite = 1;
8517 inode_unlock(inode);
8519 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8523 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8529 * We need to know how many extents we reserved so that we can
8530 * do the accounting properly if we go over the number we
8531 * originally calculated. Abuse current->journal_info for this.
8533 dio_data.reserve = round_up(count,
8534 fs_info->sectorsize);
8535 dio_data.unsubmitted_oe_range_start = (u64)offset;
8536 dio_data.unsubmitted_oe_range_end = (u64)offset;
8537 current->journal_info = &dio_data;
8538 down_read(&BTRFS_I(inode)->dio_sem);
8539 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8540 &BTRFS_I(inode)->runtime_flags)) {
8541 inode_dio_end(inode);
8542 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8546 ret = __blockdev_direct_IO(iocb, inode,
8547 fs_info->fs_devices->latest_bdev,
8548 iter, btrfs_get_blocks_direct, NULL,
8549 btrfs_submit_direct, flags);
8550 if (iov_iter_rw(iter) == WRITE) {
8551 up_read(&BTRFS_I(inode)->dio_sem);
8552 current->journal_info = NULL;
8553 if (ret < 0 && ret != -EIOCBQUEUED) {
8554 if (dio_data.reserve)
8555 btrfs_delalloc_release_space(inode, data_reserved,
8556 offset, dio_data.reserve, true);
8558 * On error we might have left some ordered extents
8559 * without submitting corresponding bios for them, so
8560 * cleanup them up to avoid other tasks getting them
8561 * and waiting for them to complete forever.
8563 if (dio_data.unsubmitted_oe_range_start <
8564 dio_data.unsubmitted_oe_range_end)
8565 __endio_write_update_ordered(inode,
8566 dio_data.unsubmitted_oe_range_start,
8567 dio_data.unsubmitted_oe_range_end -
8568 dio_data.unsubmitted_oe_range_start,
8570 } else if (ret >= 0 && (size_t)ret < count)
8571 btrfs_delalloc_release_space(inode, data_reserved,
8572 offset, count - (size_t)ret, true);
8573 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8577 inode_dio_end(inode);
8581 extent_changeset_free(data_reserved);
8585 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8587 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8588 __u64 start, __u64 len)
8592 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8596 return extent_fiemap(inode, fieinfo, start, len);
8599 int btrfs_readpage(struct file *file, struct page *page)
8601 struct extent_io_tree *tree;
8602 tree = &BTRFS_I(page->mapping->host)->io_tree;
8603 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8606 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8608 struct inode *inode = page->mapping->host;
8611 if (current->flags & PF_MEMALLOC) {
8612 redirty_page_for_writepage(wbc, page);
8618 * If we are under memory pressure we will call this directly from the
8619 * VM, we need to make sure we have the inode referenced for the ordered
8620 * extent. If not just return like we didn't do anything.
8622 if (!igrab(inode)) {
8623 redirty_page_for_writepage(wbc, page);
8624 return AOP_WRITEPAGE_ACTIVATE;
8626 ret = extent_write_full_page(page, wbc);
8627 btrfs_add_delayed_iput(inode);
8631 static int btrfs_writepages(struct address_space *mapping,
8632 struct writeback_control *wbc)
8634 return extent_writepages(mapping, wbc);
8638 btrfs_readpages(struct file *file, struct address_space *mapping,
8639 struct list_head *pages, unsigned nr_pages)
8641 return extent_readpages(mapping, pages, nr_pages);
8644 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8646 int ret = try_release_extent_mapping(page, gfp_flags);
8648 ClearPagePrivate(page);
8649 set_page_private(page, 0);
8655 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8657 if (PageWriteback(page) || PageDirty(page))
8659 return __btrfs_releasepage(page, gfp_flags);
8662 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8663 unsigned int length)
8665 struct inode *inode = page->mapping->host;
8666 struct extent_io_tree *tree;
8667 struct btrfs_ordered_extent *ordered;
8668 struct extent_state *cached_state = NULL;
8669 u64 page_start = page_offset(page);
8670 u64 page_end = page_start + PAGE_SIZE - 1;
8673 int inode_evicting = inode->i_state & I_FREEING;
8676 * we have the page locked, so new writeback can't start,
8677 * and the dirty bit won't be cleared while we are here.
8679 * Wait for IO on this page so that we can safely clear
8680 * the PagePrivate2 bit and do ordered accounting
8682 wait_on_page_writeback(page);
8684 tree = &BTRFS_I(inode)->io_tree;
8686 btrfs_releasepage(page, GFP_NOFS);
8690 if (!inode_evicting)
8691 lock_extent_bits(tree, page_start, page_end, &cached_state);
8694 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8695 page_end - start + 1);
8697 end = min(page_end, ordered->file_offset + ordered->len - 1);
8699 * IO on this page will never be started, so we need
8700 * to account for any ordered extents now
8702 if (!inode_evicting)
8703 clear_extent_bit(tree, start, end,
8704 EXTENT_DIRTY | EXTENT_DELALLOC |
8705 EXTENT_DELALLOC_NEW |
8706 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8707 EXTENT_DEFRAG, 1, 0, &cached_state);
8709 * whoever cleared the private bit is responsible
8710 * for the finish_ordered_io
8712 if (TestClearPagePrivate2(page)) {
8713 struct btrfs_ordered_inode_tree *tree;
8716 tree = &BTRFS_I(inode)->ordered_tree;
8718 spin_lock_irq(&tree->lock);
8719 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8720 new_len = start - ordered->file_offset;
8721 if (new_len < ordered->truncated_len)
8722 ordered->truncated_len = new_len;
8723 spin_unlock_irq(&tree->lock);
8725 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8727 end - start + 1, 1))
8728 btrfs_finish_ordered_io(ordered);
8730 btrfs_put_ordered_extent(ordered);
8731 if (!inode_evicting) {
8732 cached_state = NULL;
8733 lock_extent_bits(tree, start, end,
8738 if (start < page_end)
8743 * Qgroup reserved space handler
8744 * Page here will be either
8745 * 1) Already written to disk
8746 * In this case, its reserved space is released from data rsv map
8747 * and will be freed by delayed_ref handler finally.
8748 * So even we call qgroup_free_data(), it won't decrease reserved
8750 * 2) Not written to disk
8751 * This means the reserved space should be freed here. However,
8752 * if a truncate invalidates the page (by clearing PageDirty)
8753 * and the page is accounted for while allocating extent
8754 * in btrfs_check_data_free_space() we let delayed_ref to
8755 * free the entire extent.
8757 if (PageDirty(page))
8758 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8759 if (!inode_evicting) {
8760 clear_extent_bit(tree, page_start, page_end,
8761 EXTENT_LOCKED | EXTENT_DIRTY |
8762 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8763 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8766 __btrfs_releasepage(page, GFP_NOFS);
8769 ClearPageChecked(page);
8770 if (PagePrivate(page)) {
8771 ClearPagePrivate(page);
8772 set_page_private(page, 0);
8778 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8779 * called from a page fault handler when a page is first dirtied. Hence we must
8780 * be careful to check for EOF conditions here. We set the page up correctly
8781 * for a written page which means we get ENOSPC checking when writing into
8782 * holes and correct delalloc and unwritten extent mapping on filesystems that
8783 * support these features.
8785 * We are not allowed to take the i_mutex here so we have to play games to
8786 * protect against truncate races as the page could now be beyond EOF. Because
8787 * truncate_setsize() writes the inode size before removing pages, once we have
8788 * the page lock we can determine safely if the page is beyond EOF. If it is not
8789 * beyond EOF, then the page is guaranteed safe against truncation until we
8792 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8794 struct page *page = vmf->page;
8795 struct inode *inode = file_inode(vmf->vma->vm_file);
8796 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8797 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8798 struct btrfs_ordered_extent *ordered;
8799 struct extent_state *cached_state = NULL;
8800 struct extent_changeset *data_reserved = NULL;
8802 unsigned long zero_start;
8812 reserved_space = PAGE_SIZE;
8814 sb_start_pagefault(inode->i_sb);
8815 page_start = page_offset(page);
8816 page_end = page_start + PAGE_SIZE - 1;
8820 * Reserving delalloc space after obtaining the page lock can lead to
8821 * deadlock. For example, if a dirty page is locked by this function
8822 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8823 * dirty page write out, then the btrfs_writepage() function could
8824 * end up waiting indefinitely to get a lock on the page currently
8825 * being processed by btrfs_page_mkwrite() function.
8827 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8830 ret2 = file_update_time(vmf->vma->vm_file);
8834 ret = vmf_error(ret2);
8840 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8843 size = i_size_read(inode);
8845 if ((page->mapping != inode->i_mapping) ||
8846 (page_start >= size)) {
8847 /* page got truncated out from underneath us */
8850 wait_on_page_writeback(page);
8852 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8853 set_page_extent_mapped(page);
8856 * we can't set the delalloc bits if there are pending ordered
8857 * extents. Drop our locks and wait for them to finish
8859 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8862 unlock_extent_cached(io_tree, page_start, page_end,
8865 btrfs_start_ordered_extent(inode, ordered, 1);
8866 btrfs_put_ordered_extent(ordered);
8870 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8871 reserved_space = round_up(size - page_start,
8872 fs_info->sectorsize);
8873 if (reserved_space < PAGE_SIZE) {
8874 end = page_start + reserved_space - 1;
8875 btrfs_delalloc_release_space(inode, data_reserved,
8876 page_start, PAGE_SIZE - reserved_space,
8882 * page_mkwrite gets called when the page is firstly dirtied after it's
8883 * faulted in, but write(2) could also dirty a page and set delalloc
8884 * bits, thus in this case for space account reason, we still need to
8885 * clear any delalloc bits within this page range since we have to
8886 * reserve data&meta space before lock_page() (see above comments).
8888 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8889 EXTENT_DIRTY | EXTENT_DELALLOC |
8890 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8891 0, 0, &cached_state);
8893 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8896 unlock_extent_cached(io_tree, page_start, page_end,
8898 ret = VM_FAULT_SIGBUS;
8903 /* page is wholly or partially inside EOF */
8904 if (page_start + PAGE_SIZE > size)
8905 zero_start = size & ~PAGE_MASK;
8907 zero_start = PAGE_SIZE;
8909 if (zero_start != PAGE_SIZE) {
8911 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8912 flush_dcache_page(page);
8915 ClearPageChecked(page);
8916 set_page_dirty(page);
8917 SetPageUptodate(page);
8919 BTRFS_I(inode)->last_trans = fs_info->generation;
8920 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8921 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8923 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8926 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8927 sb_end_pagefault(inode->i_sb);
8928 extent_changeset_free(data_reserved);
8929 return VM_FAULT_LOCKED;
8935 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8936 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8937 reserved_space, (ret != 0));
8939 sb_end_pagefault(inode->i_sb);
8940 extent_changeset_free(data_reserved);
8944 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8946 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8947 struct btrfs_root *root = BTRFS_I(inode)->root;
8948 struct btrfs_block_rsv *rsv;
8950 struct btrfs_trans_handle *trans;
8951 u64 mask = fs_info->sectorsize - 1;
8952 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8954 if (!skip_writeback) {
8955 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8962 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8963 * things going on here:
8965 * 1) We need to reserve space to update our inode.
8967 * 2) We need to have something to cache all the space that is going to
8968 * be free'd up by the truncate operation, but also have some slack
8969 * space reserved in case it uses space during the truncate (thank you
8970 * very much snapshotting).
8972 * And we need these to be separate. The fact is we can use a lot of
8973 * space doing the truncate, and we have no earthly idea how much space
8974 * we will use, so we need the truncate reservation to be separate so it
8975 * doesn't end up using space reserved for updating the inode. We also
8976 * need to be able to stop the transaction and start a new one, which
8977 * means we need to be able to update the inode several times, and we
8978 * have no idea of knowing how many times that will be, so we can't just
8979 * reserve 1 item for the entirety of the operation, so that has to be
8980 * done separately as well.
8982 * So that leaves us with
8984 * 1) rsv - for the truncate reservation, which we will steal from the
8985 * transaction reservation.
8986 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8987 * updating the inode.
8989 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8992 rsv->size = min_size;
8996 * 1 for the truncate slack space
8997 * 1 for updating the inode.
8999 trans = btrfs_start_transaction(root, 2);
9000 if (IS_ERR(trans)) {
9001 ret = PTR_ERR(trans);
9005 /* Migrate the slack space for the truncate to our reserve */
9006 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9011 * So if we truncate and then write and fsync we normally would just
9012 * write the extents that changed, which is a problem if we need to
9013 * first truncate that entire inode. So set this flag so we write out
9014 * all of the extents in the inode to the sync log so we're completely
9017 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9018 trans->block_rsv = rsv;
9021 ret = btrfs_truncate_inode_items(trans, root, inode,
9023 BTRFS_EXTENT_DATA_KEY);
9024 trans->block_rsv = &fs_info->trans_block_rsv;
9025 if (ret != -ENOSPC && ret != -EAGAIN)
9028 ret = btrfs_update_inode(trans, root, inode);
9032 btrfs_end_transaction(trans);
9033 btrfs_btree_balance_dirty(fs_info);
9035 trans = btrfs_start_transaction(root, 2);
9036 if (IS_ERR(trans)) {
9037 ret = PTR_ERR(trans);
9042 btrfs_block_rsv_release(fs_info, rsv, -1);
9043 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9044 rsv, min_size, false);
9045 BUG_ON(ret); /* shouldn't happen */
9046 trans->block_rsv = rsv;
9050 * We can't call btrfs_truncate_block inside a trans handle as we could
9051 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9052 * we've truncated everything except the last little bit, and can do
9053 * btrfs_truncate_block and then update the disk_i_size.
9055 if (ret == NEED_TRUNCATE_BLOCK) {
9056 btrfs_end_transaction(trans);
9057 btrfs_btree_balance_dirty(fs_info);
9059 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9062 trans = btrfs_start_transaction(root, 1);
9063 if (IS_ERR(trans)) {
9064 ret = PTR_ERR(trans);
9067 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9073 trans->block_rsv = &fs_info->trans_block_rsv;
9074 ret2 = btrfs_update_inode(trans, root, inode);
9078 ret2 = btrfs_end_transaction(trans);
9081 btrfs_btree_balance_dirty(fs_info);
9084 btrfs_free_block_rsv(fs_info, rsv);
9090 * create a new subvolume directory/inode (helper for the ioctl).
9092 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9093 struct btrfs_root *new_root,
9094 struct btrfs_root *parent_root,
9097 struct inode *inode;
9101 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9102 new_dirid, new_dirid,
9103 S_IFDIR | (~current_umask() & S_IRWXUGO),
9106 return PTR_ERR(inode);
9107 inode->i_op = &btrfs_dir_inode_operations;
9108 inode->i_fop = &btrfs_dir_file_operations;
9110 set_nlink(inode, 1);
9111 btrfs_i_size_write(BTRFS_I(inode), 0);
9112 unlock_new_inode(inode);
9114 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9116 btrfs_err(new_root->fs_info,
9117 "error inheriting subvolume %llu properties: %d",
9118 new_root->root_key.objectid, err);
9120 err = btrfs_update_inode(trans, new_root, inode);
9126 struct inode *btrfs_alloc_inode(struct super_block *sb)
9128 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9129 struct btrfs_inode *ei;
9130 struct inode *inode;
9132 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9139 ei->last_sub_trans = 0;
9140 ei->logged_trans = 0;
9141 ei->delalloc_bytes = 0;
9142 ei->new_delalloc_bytes = 0;
9143 ei->defrag_bytes = 0;
9144 ei->disk_i_size = 0;
9147 ei->index_cnt = (u64)-1;
9149 ei->last_unlink_trans = 0;
9150 ei->last_log_commit = 0;
9152 spin_lock_init(&ei->lock);
9153 ei->outstanding_extents = 0;
9154 if (sb->s_magic != BTRFS_TEST_MAGIC)
9155 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9156 BTRFS_BLOCK_RSV_DELALLOC);
9157 ei->runtime_flags = 0;
9158 ei->prop_compress = BTRFS_COMPRESS_NONE;
9159 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9161 ei->delayed_node = NULL;
9163 ei->i_otime.tv_sec = 0;
9164 ei->i_otime.tv_nsec = 0;
9166 inode = &ei->vfs_inode;
9167 extent_map_tree_init(&ei->extent_tree);
9168 extent_io_tree_init(&ei->io_tree, inode);
9169 extent_io_tree_init(&ei->io_failure_tree, inode);
9170 ei->io_tree.track_uptodate = 1;
9171 ei->io_failure_tree.track_uptodate = 1;
9172 atomic_set(&ei->sync_writers, 0);
9173 mutex_init(&ei->log_mutex);
9174 mutex_init(&ei->delalloc_mutex);
9175 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9176 INIT_LIST_HEAD(&ei->delalloc_inodes);
9177 INIT_LIST_HEAD(&ei->delayed_iput);
9178 RB_CLEAR_NODE(&ei->rb_node);
9179 init_rwsem(&ei->dio_sem);
9184 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9185 void btrfs_test_destroy_inode(struct inode *inode)
9187 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9188 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9192 static void btrfs_i_callback(struct rcu_head *head)
9194 struct inode *inode = container_of(head, struct inode, i_rcu);
9195 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9198 void btrfs_destroy_inode(struct inode *inode)
9200 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9201 struct btrfs_ordered_extent *ordered;
9202 struct btrfs_root *root = BTRFS_I(inode)->root;
9204 WARN_ON(!hlist_empty(&inode->i_dentry));
9205 WARN_ON(inode->i_data.nrpages);
9206 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9207 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9208 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9209 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9210 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9211 WARN_ON(BTRFS_I(inode)->csum_bytes);
9212 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9215 * This can happen where we create an inode, but somebody else also
9216 * created the same inode and we need to destroy the one we already
9223 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9228 "found ordered extent %llu %llu on inode cleanup",
9229 ordered->file_offset, ordered->len);
9230 btrfs_remove_ordered_extent(inode, ordered);
9231 btrfs_put_ordered_extent(ordered);
9232 btrfs_put_ordered_extent(ordered);
9235 btrfs_qgroup_check_reserved_leak(inode);
9236 inode_tree_del(inode);
9237 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9239 call_rcu(&inode->i_rcu, btrfs_i_callback);
9242 int btrfs_drop_inode(struct inode *inode)
9244 struct btrfs_root *root = BTRFS_I(inode)->root;
9249 /* the snap/subvol tree is on deleting */
9250 if (btrfs_root_refs(&root->root_item) == 0)
9253 return generic_drop_inode(inode);
9256 static void init_once(void *foo)
9258 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9260 inode_init_once(&ei->vfs_inode);
9263 void __cold btrfs_destroy_cachep(void)
9266 * Make sure all delayed rcu free inodes are flushed before we
9270 kmem_cache_destroy(btrfs_inode_cachep);
9271 kmem_cache_destroy(btrfs_trans_handle_cachep);
9272 kmem_cache_destroy(btrfs_path_cachep);
9273 kmem_cache_destroy(btrfs_free_space_cachep);
9276 int __init btrfs_init_cachep(void)
9278 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9279 sizeof(struct btrfs_inode), 0,
9280 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9282 if (!btrfs_inode_cachep)
9285 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9286 sizeof(struct btrfs_trans_handle), 0,
9287 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9288 if (!btrfs_trans_handle_cachep)
9291 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9292 sizeof(struct btrfs_path), 0,
9293 SLAB_MEM_SPREAD, NULL);
9294 if (!btrfs_path_cachep)
9297 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9298 sizeof(struct btrfs_free_space), 0,
9299 SLAB_MEM_SPREAD, NULL);
9300 if (!btrfs_free_space_cachep)
9305 btrfs_destroy_cachep();
9309 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9310 u32 request_mask, unsigned int flags)
9313 struct inode *inode = d_inode(path->dentry);
9314 u32 blocksize = inode->i_sb->s_blocksize;
9315 u32 bi_flags = BTRFS_I(inode)->flags;
9317 stat->result_mask |= STATX_BTIME;
9318 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9319 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9320 if (bi_flags & BTRFS_INODE_APPEND)
9321 stat->attributes |= STATX_ATTR_APPEND;
9322 if (bi_flags & BTRFS_INODE_COMPRESS)
9323 stat->attributes |= STATX_ATTR_COMPRESSED;
9324 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9325 stat->attributes |= STATX_ATTR_IMMUTABLE;
9326 if (bi_flags & BTRFS_INODE_NODUMP)
9327 stat->attributes |= STATX_ATTR_NODUMP;
9329 stat->attributes_mask |= (STATX_ATTR_APPEND |
9330 STATX_ATTR_COMPRESSED |
9331 STATX_ATTR_IMMUTABLE |
9334 generic_fillattr(inode, stat);
9335 stat->dev = BTRFS_I(inode)->root->anon_dev;
9337 spin_lock(&BTRFS_I(inode)->lock);
9338 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9339 spin_unlock(&BTRFS_I(inode)->lock);
9340 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9341 ALIGN(delalloc_bytes, blocksize)) >> 9;
9345 static int btrfs_rename_exchange(struct inode *old_dir,
9346 struct dentry *old_dentry,
9347 struct inode *new_dir,
9348 struct dentry *new_dentry)
9350 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9351 struct btrfs_trans_handle *trans;
9352 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9353 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9354 struct inode *new_inode = new_dentry->d_inode;
9355 struct inode *old_inode = old_dentry->d_inode;
9356 struct timespec64 ctime = current_time(old_inode);
9357 struct dentry *parent;
9358 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9359 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9364 bool root_log_pinned = false;
9365 bool dest_log_pinned = false;
9366 struct btrfs_log_ctx ctx_root;
9367 struct btrfs_log_ctx ctx_dest;
9368 bool sync_log_root = false;
9369 bool sync_log_dest = false;
9370 bool commit_transaction = false;
9372 /* we only allow rename subvolume link between subvolumes */
9373 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9376 btrfs_init_log_ctx(&ctx_root, old_inode);
9377 btrfs_init_log_ctx(&ctx_dest, new_inode);
9379 /* close the race window with snapshot create/destroy ioctl */
9380 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9381 down_read(&fs_info->subvol_sem);
9382 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9383 down_read(&fs_info->subvol_sem);
9386 * We want to reserve the absolute worst case amount of items. So if
9387 * both inodes are subvols and we need to unlink them then that would
9388 * require 4 item modifications, but if they are both normal inodes it
9389 * would require 5 item modifications, so we'll assume their normal
9390 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9391 * should cover the worst case number of items we'll modify.
9393 trans = btrfs_start_transaction(root, 12);
9394 if (IS_ERR(trans)) {
9395 ret = PTR_ERR(trans);
9400 * We need to find a free sequence number both in the source and
9401 * in the destination directory for the exchange.
9403 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9406 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9410 BTRFS_I(old_inode)->dir_index = 0ULL;
9411 BTRFS_I(new_inode)->dir_index = 0ULL;
9413 /* Reference for the source. */
9414 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9415 /* force full log commit if subvolume involved. */
9416 btrfs_set_log_full_commit(fs_info, trans);
9418 btrfs_pin_log_trans(root);
9419 root_log_pinned = true;
9420 ret = btrfs_insert_inode_ref(trans, dest,
9421 new_dentry->d_name.name,
9422 new_dentry->d_name.len,
9424 btrfs_ino(BTRFS_I(new_dir)),
9430 /* And now for the dest. */
9431 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9432 /* force full log commit if subvolume involved. */
9433 btrfs_set_log_full_commit(fs_info, trans);
9435 btrfs_pin_log_trans(dest);
9436 dest_log_pinned = true;
9437 ret = btrfs_insert_inode_ref(trans, root,
9438 old_dentry->d_name.name,
9439 old_dentry->d_name.len,
9441 btrfs_ino(BTRFS_I(old_dir)),
9447 /* Update inode version and ctime/mtime. */
9448 inode_inc_iversion(old_dir);
9449 inode_inc_iversion(new_dir);
9450 inode_inc_iversion(old_inode);
9451 inode_inc_iversion(new_inode);
9452 old_dir->i_ctime = old_dir->i_mtime = ctime;
9453 new_dir->i_ctime = new_dir->i_mtime = ctime;
9454 old_inode->i_ctime = ctime;
9455 new_inode->i_ctime = ctime;
9457 if (old_dentry->d_parent != new_dentry->d_parent) {
9458 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9459 BTRFS_I(old_inode), 1);
9460 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9461 BTRFS_I(new_inode), 1);
9464 /* src is a subvolume */
9465 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9466 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9467 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9468 old_dentry->d_name.name,
9469 old_dentry->d_name.len);
9470 } else { /* src is an inode */
9471 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9472 BTRFS_I(old_dentry->d_inode),
9473 old_dentry->d_name.name,
9474 old_dentry->d_name.len);
9476 ret = btrfs_update_inode(trans, root, old_inode);
9479 btrfs_abort_transaction(trans, ret);
9483 /* dest is a subvolume */
9484 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9485 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9486 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9487 new_dentry->d_name.name,
9488 new_dentry->d_name.len);
9489 } else { /* dest is an inode */
9490 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9491 BTRFS_I(new_dentry->d_inode),
9492 new_dentry->d_name.name,
9493 new_dentry->d_name.len);
9495 ret = btrfs_update_inode(trans, dest, new_inode);
9498 btrfs_abort_transaction(trans, ret);
9502 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9503 new_dentry->d_name.name,
9504 new_dentry->d_name.len, 0, old_idx);
9506 btrfs_abort_transaction(trans, ret);
9510 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9511 old_dentry->d_name.name,
9512 old_dentry->d_name.len, 0, new_idx);
9514 btrfs_abort_transaction(trans, ret);
9518 if (old_inode->i_nlink == 1)
9519 BTRFS_I(old_inode)->dir_index = old_idx;
9520 if (new_inode->i_nlink == 1)
9521 BTRFS_I(new_inode)->dir_index = new_idx;
9523 if (root_log_pinned) {
9524 parent = new_dentry->d_parent;
9525 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9526 BTRFS_I(old_dir), parent,
9528 if (ret == BTRFS_NEED_LOG_SYNC)
9529 sync_log_root = true;
9530 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9531 commit_transaction = true;
9533 btrfs_end_log_trans(root);
9534 root_log_pinned = false;
9536 if (dest_log_pinned) {
9537 if (!commit_transaction) {
9538 parent = old_dentry->d_parent;
9539 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9540 BTRFS_I(new_dir), parent,
9542 if (ret == BTRFS_NEED_LOG_SYNC)
9543 sync_log_dest = true;
9544 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9545 commit_transaction = true;
9548 btrfs_end_log_trans(dest);
9549 dest_log_pinned = false;
9553 * If we have pinned a log and an error happened, we unpin tasks
9554 * trying to sync the log and force them to fallback to a transaction
9555 * commit if the log currently contains any of the inodes involved in
9556 * this rename operation (to ensure we do not persist a log with an
9557 * inconsistent state for any of these inodes or leading to any
9558 * inconsistencies when replayed). If the transaction was aborted, the
9559 * abortion reason is propagated to userspace when attempting to commit
9560 * the transaction. If the log does not contain any of these inodes, we
9561 * allow the tasks to sync it.
9563 if (ret && (root_log_pinned || dest_log_pinned)) {
9564 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9565 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9566 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9568 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9569 btrfs_set_log_full_commit(fs_info, trans);
9571 if (root_log_pinned) {
9572 btrfs_end_log_trans(root);
9573 root_log_pinned = false;
9575 if (dest_log_pinned) {
9576 btrfs_end_log_trans(dest);
9577 dest_log_pinned = false;
9580 if (!ret && sync_log_root && !commit_transaction) {
9581 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9584 commit_transaction = true;
9586 if (!ret && sync_log_dest && !commit_transaction) {
9587 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9590 commit_transaction = true;
9592 if (commit_transaction) {
9593 ret = btrfs_commit_transaction(trans);
9597 ret2 = btrfs_end_transaction(trans);
9598 ret = ret ? ret : ret2;
9601 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9602 up_read(&fs_info->subvol_sem);
9603 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9604 up_read(&fs_info->subvol_sem);
9609 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9610 struct btrfs_root *root,
9612 struct dentry *dentry)
9615 struct inode *inode;
9619 ret = btrfs_find_free_ino(root, &objectid);
9623 inode = btrfs_new_inode(trans, root, dir,
9624 dentry->d_name.name,
9626 btrfs_ino(BTRFS_I(dir)),
9628 S_IFCHR | WHITEOUT_MODE,
9631 if (IS_ERR(inode)) {
9632 ret = PTR_ERR(inode);
9636 inode->i_op = &btrfs_special_inode_operations;
9637 init_special_inode(inode, inode->i_mode,
9640 ret = btrfs_init_inode_security(trans, inode, dir,
9645 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9646 BTRFS_I(inode), 0, index);
9650 ret = btrfs_update_inode(trans, root, inode);
9652 unlock_new_inode(inode);
9654 inode_dec_link_count(inode);
9660 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9661 struct inode *new_dir, struct dentry *new_dentry,
9664 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9665 struct btrfs_trans_handle *trans;
9666 unsigned int trans_num_items;
9667 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9668 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9669 struct inode *new_inode = d_inode(new_dentry);
9670 struct inode *old_inode = d_inode(old_dentry);
9674 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9675 bool log_pinned = false;
9676 struct btrfs_log_ctx ctx;
9677 bool sync_log = false;
9678 bool commit_transaction = false;
9680 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9683 /* we only allow rename subvolume link between subvolumes */
9684 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9687 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9688 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9691 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9692 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9696 /* check for collisions, even if the name isn't there */
9697 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9698 new_dentry->d_name.name,
9699 new_dentry->d_name.len);
9702 if (ret == -EEXIST) {
9704 * eexist without a new_inode */
9705 if (WARN_ON(!new_inode)) {
9709 /* maybe -EOVERFLOW */
9716 * we're using rename to replace one file with another. Start IO on it
9717 * now so we don't add too much work to the end of the transaction
9719 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9720 filemap_flush(old_inode->i_mapping);
9722 /* close the racy window with snapshot create/destroy ioctl */
9723 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9724 down_read(&fs_info->subvol_sem);
9726 * We want to reserve the absolute worst case amount of items. So if
9727 * both inodes are subvols and we need to unlink them then that would
9728 * require 4 item modifications, but if they are both normal inodes it
9729 * would require 5 item modifications, so we'll assume they are normal
9730 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9731 * should cover the worst case number of items we'll modify.
9732 * If our rename has the whiteout flag, we need more 5 units for the
9733 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9734 * when selinux is enabled).
9736 trans_num_items = 11;
9737 if (flags & RENAME_WHITEOUT)
9738 trans_num_items += 5;
9739 trans = btrfs_start_transaction(root, trans_num_items);
9740 if (IS_ERR(trans)) {
9741 ret = PTR_ERR(trans);
9746 btrfs_record_root_in_trans(trans, dest);
9748 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9752 BTRFS_I(old_inode)->dir_index = 0ULL;
9753 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9754 /* force full log commit if subvolume involved. */
9755 btrfs_set_log_full_commit(fs_info, trans);
9757 btrfs_pin_log_trans(root);
9759 ret = btrfs_insert_inode_ref(trans, dest,
9760 new_dentry->d_name.name,
9761 new_dentry->d_name.len,
9763 btrfs_ino(BTRFS_I(new_dir)), index);
9768 inode_inc_iversion(old_dir);
9769 inode_inc_iversion(new_dir);
9770 inode_inc_iversion(old_inode);
9771 old_dir->i_ctime = old_dir->i_mtime =
9772 new_dir->i_ctime = new_dir->i_mtime =
9773 old_inode->i_ctime = current_time(old_dir);
9775 if (old_dentry->d_parent != new_dentry->d_parent)
9776 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9777 BTRFS_I(old_inode), 1);
9779 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9780 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9781 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9782 old_dentry->d_name.name,
9783 old_dentry->d_name.len);
9785 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9786 BTRFS_I(d_inode(old_dentry)),
9787 old_dentry->d_name.name,
9788 old_dentry->d_name.len);
9790 ret = btrfs_update_inode(trans, root, old_inode);
9793 btrfs_abort_transaction(trans, ret);
9798 inode_inc_iversion(new_inode);
9799 new_inode->i_ctime = current_time(new_inode);
9800 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9801 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9802 root_objectid = BTRFS_I(new_inode)->location.objectid;
9803 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9804 new_dentry->d_name.name,
9805 new_dentry->d_name.len);
9806 BUG_ON(new_inode->i_nlink == 0);
9808 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9809 BTRFS_I(d_inode(new_dentry)),
9810 new_dentry->d_name.name,
9811 new_dentry->d_name.len);
9813 if (!ret && new_inode->i_nlink == 0)
9814 ret = btrfs_orphan_add(trans,
9815 BTRFS_I(d_inode(new_dentry)));
9817 btrfs_abort_transaction(trans, ret);
9822 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9823 new_dentry->d_name.name,
9824 new_dentry->d_name.len, 0, index);
9826 btrfs_abort_transaction(trans, ret);
9830 if (old_inode->i_nlink == 1)
9831 BTRFS_I(old_inode)->dir_index = index;
9834 struct dentry *parent = new_dentry->d_parent;
9836 btrfs_init_log_ctx(&ctx, old_inode);
9837 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9838 BTRFS_I(old_dir), parent,
9840 if (ret == BTRFS_NEED_LOG_SYNC)
9842 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9843 commit_transaction = true;
9845 btrfs_end_log_trans(root);
9849 if (flags & RENAME_WHITEOUT) {
9850 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9854 btrfs_abort_transaction(trans, ret);
9860 * If we have pinned the log and an error happened, we unpin tasks
9861 * trying to sync the log and force them to fallback to a transaction
9862 * commit if the log currently contains any of the inodes involved in
9863 * this rename operation (to ensure we do not persist a log with an
9864 * inconsistent state for any of these inodes or leading to any
9865 * inconsistencies when replayed). If the transaction was aborted, the
9866 * abortion reason is propagated to userspace when attempting to commit
9867 * the transaction. If the log does not contain any of these inodes, we
9868 * allow the tasks to sync it.
9870 if (ret && log_pinned) {
9871 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9872 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9873 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9875 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9876 btrfs_set_log_full_commit(fs_info, trans);
9878 btrfs_end_log_trans(root);
9881 if (!ret && sync_log) {
9882 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9884 commit_transaction = true;
9886 if (commit_transaction) {
9887 ret = btrfs_commit_transaction(trans);
9891 ret2 = btrfs_end_transaction(trans);
9892 ret = ret ? ret : ret2;
9895 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9896 up_read(&fs_info->subvol_sem);
9901 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9902 struct inode *new_dir, struct dentry *new_dentry,
9905 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9908 if (flags & RENAME_EXCHANGE)
9909 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9912 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9915 struct btrfs_delalloc_work {
9916 struct inode *inode;
9917 struct completion completion;
9918 struct list_head list;
9919 struct btrfs_work work;
9922 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9924 struct btrfs_delalloc_work *delalloc_work;
9925 struct inode *inode;
9927 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9929 inode = delalloc_work->inode;
9930 filemap_flush(inode->i_mapping);
9931 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9932 &BTRFS_I(inode)->runtime_flags))
9933 filemap_flush(inode->i_mapping);
9936 complete(&delalloc_work->completion);
9939 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9941 struct btrfs_delalloc_work *work;
9943 work = kmalloc(sizeof(*work), GFP_NOFS);
9947 init_completion(&work->completion);
9948 INIT_LIST_HEAD(&work->list);
9949 work->inode = inode;
9950 WARN_ON_ONCE(!inode);
9951 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9952 btrfs_run_delalloc_work, NULL, NULL);
9958 * some fairly slow code that needs optimization. This walks the list
9959 * of all the inodes with pending delalloc and forces them to disk.
9961 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9963 struct btrfs_inode *binode;
9964 struct inode *inode;
9965 struct btrfs_delalloc_work *work, *next;
9966 struct list_head works;
9967 struct list_head splice;
9970 INIT_LIST_HEAD(&works);
9971 INIT_LIST_HEAD(&splice);
9973 mutex_lock(&root->delalloc_mutex);
9974 spin_lock(&root->delalloc_lock);
9975 list_splice_init(&root->delalloc_inodes, &splice);
9976 while (!list_empty(&splice)) {
9977 binode = list_entry(splice.next, struct btrfs_inode,
9980 list_move_tail(&binode->delalloc_inodes,
9981 &root->delalloc_inodes);
9982 inode = igrab(&binode->vfs_inode);
9984 cond_resched_lock(&root->delalloc_lock);
9987 spin_unlock(&root->delalloc_lock);
9989 work = btrfs_alloc_delalloc_work(inode);
9995 list_add_tail(&work->list, &works);
9996 btrfs_queue_work(root->fs_info->flush_workers,
9999 if (nr != -1 && ret >= nr)
10002 spin_lock(&root->delalloc_lock);
10004 spin_unlock(&root->delalloc_lock);
10007 list_for_each_entry_safe(work, next, &works, list) {
10008 list_del_init(&work->list);
10009 wait_for_completion(&work->completion);
10013 if (!list_empty(&splice)) {
10014 spin_lock(&root->delalloc_lock);
10015 list_splice_tail(&splice, &root->delalloc_inodes);
10016 spin_unlock(&root->delalloc_lock);
10018 mutex_unlock(&root->delalloc_mutex);
10022 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10024 struct btrfs_fs_info *fs_info = root->fs_info;
10027 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10030 ret = start_delalloc_inodes(root, -1);
10036 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10038 struct btrfs_root *root;
10039 struct list_head splice;
10042 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10045 INIT_LIST_HEAD(&splice);
10047 mutex_lock(&fs_info->delalloc_root_mutex);
10048 spin_lock(&fs_info->delalloc_root_lock);
10049 list_splice_init(&fs_info->delalloc_roots, &splice);
10050 while (!list_empty(&splice) && nr) {
10051 root = list_first_entry(&splice, struct btrfs_root,
10053 root = btrfs_grab_fs_root(root);
10055 list_move_tail(&root->delalloc_root,
10056 &fs_info->delalloc_roots);
10057 spin_unlock(&fs_info->delalloc_root_lock);
10059 ret = start_delalloc_inodes(root, nr);
10060 btrfs_put_fs_root(root);
10068 spin_lock(&fs_info->delalloc_root_lock);
10070 spin_unlock(&fs_info->delalloc_root_lock);
10074 if (!list_empty(&splice)) {
10075 spin_lock(&fs_info->delalloc_root_lock);
10076 list_splice_tail(&splice, &fs_info->delalloc_roots);
10077 spin_unlock(&fs_info->delalloc_root_lock);
10079 mutex_unlock(&fs_info->delalloc_root_mutex);
10083 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10084 const char *symname)
10086 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10087 struct btrfs_trans_handle *trans;
10088 struct btrfs_root *root = BTRFS_I(dir)->root;
10089 struct btrfs_path *path;
10090 struct btrfs_key key;
10091 struct inode *inode = NULL;
10098 struct btrfs_file_extent_item *ei;
10099 struct extent_buffer *leaf;
10101 name_len = strlen(symname);
10102 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10103 return -ENAMETOOLONG;
10106 * 2 items for inode item and ref
10107 * 2 items for dir items
10108 * 1 item for updating parent inode item
10109 * 1 item for the inline extent item
10110 * 1 item for xattr if selinux is on
10112 trans = btrfs_start_transaction(root, 7);
10114 return PTR_ERR(trans);
10116 err = btrfs_find_free_ino(root, &objectid);
10120 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10121 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10122 objectid, S_IFLNK|S_IRWXUGO, &index);
10123 if (IS_ERR(inode)) {
10124 err = PTR_ERR(inode);
10130 * If the active LSM wants to access the inode during
10131 * d_instantiate it needs these. Smack checks to see
10132 * if the filesystem supports xattrs by looking at the
10135 inode->i_fop = &btrfs_file_operations;
10136 inode->i_op = &btrfs_file_inode_operations;
10137 inode->i_mapping->a_ops = &btrfs_aops;
10138 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10140 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10144 path = btrfs_alloc_path();
10149 key.objectid = btrfs_ino(BTRFS_I(inode));
10151 key.type = BTRFS_EXTENT_DATA_KEY;
10152 datasize = btrfs_file_extent_calc_inline_size(name_len);
10153 err = btrfs_insert_empty_item(trans, root, path, &key,
10156 btrfs_free_path(path);
10159 leaf = path->nodes[0];
10160 ei = btrfs_item_ptr(leaf, path->slots[0],
10161 struct btrfs_file_extent_item);
10162 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10163 btrfs_set_file_extent_type(leaf, ei,
10164 BTRFS_FILE_EXTENT_INLINE);
10165 btrfs_set_file_extent_encryption(leaf, ei, 0);
10166 btrfs_set_file_extent_compression(leaf, ei, 0);
10167 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10168 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10170 ptr = btrfs_file_extent_inline_start(ei);
10171 write_extent_buffer(leaf, symname, ptr, name_len);
10172 btrfs_mark_buffer_dirty(leaf);
10173 btrfs_free_path(path);
10175 inode->i_op = &btrfs_symlink_inode_operations;
10176 inode_nohighmem(inode);
10177 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10178 inode_set_bytes(inode, name_len);
10179 btrfs_i_size_write(BTRFS_I(inode), name_len);
10180 err = btrfs_update_inode(trans, root, inode);
10182 * Last step, add directory indexes for our symlink inode. This is the
10183 * last step to avoid extra cleanup of these indexes if an error happens
10187 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10188 BTRFS_I(inode), 0, index);
10192 d_instantiate_new(dentry, inode);
10195 btrfs_end_transaction(trans);
10196 if (err && inode) {
10197 inode_dec_link_count(inode);
10198 discard_new_inode(inode);
10200 btrfs_btree_balance_dirty(fs_info);
10204 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10205 u64 start, u64 num_bytes, u64 min_size,
10206 loff_t actual_len, u64 *alloc_hint,
10207 struct btrfs_trans_handle *trans)
10209 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10210 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10211 struct extent_map *em;
10212 struct btrfs_root *root = BTRFS_I(inode)->root;
10213 struct btrfs_key ins;
10214 u64 cur_offset = start;
10217 u64 last_alloc = (u64)-1;
10219 bool own_trans = true;
10220 u64 end = start + num_bytes - 1;
10224 while (num_bytes > 0) {
10226 trans = btrfs_start_transaction(root, 3);
10227 if (IS_ERR(trans)) {
10228 ret = PTR_ERR(trans);
10233 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10234 cur_bytes = max(cur_bytes, min_size);
10236 * If we are severely fragmented we could end up with really
10237 * small allocations, so if the allocator is returning small
10238 * chunks lets make its job easier by only searching for those
10241 cur_bytes = min(cur_bytes, last_alloc);
10242 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10243 min_size, 0, *alloc_hint, &ins, 1, 0);
10246 btrfs_end_transaction(trans);
10249 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10251 last_alloc = ins.offset;
10252 ret = insert_reserved_file_extent(trans, inode,
10253 cur_offset, ins.objectid,
10254 ins.offset, ins.offset,
10255 ins.offset, 0, 0, 0,
10256 BTRFS_FILE_EXTENT_PREALLOC);
10258 btrfs_free_reserved_extent(fs_info, ins.objectid,
10260 btrfs_abort_transaction(trans, ret);
10262 btrfs_end_transaction(trans);
10266 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10267 cur_offset + ins.offset -1, 0);
10269 em = alloc_extent_map();
10271 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10272 &BTRFS_I(inode)->runtime_flags);
10276 em->start = cur_offset;
10277 em->orig_start = cur_offset;
10278 em->len = ins.offset;
10279 em->block_start = ins.objectid;
10280 em->block_len = ins.offset;
10281 em->orig_block_len = ins.offset;
10282 em->ram_bytes = ins.offset;
10283 em->bdev = fs_info->fs_devices->latest_bdev;
10284 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10285 em->generation = trans->transid;
10288 write_lock(&em_tree->lock);
10289 ret = add_extent_mapping(em_tree, em, 1);
10290 write_unlock(&em_tree->lock);
10291 if (ret != -EEXIST)
10293 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10294 cur_offset + ins.offset - 1,
10297 free_extent_map(em);
10299 num_bytes -= ins.offset;
10300 cur_offset += ins.offset;
10301 *alloc_hint = ins.objectid + ins.offset;
10303 inode_inc_iversion(inode);
10304 inode->i_ctime = current_time(inode);
10305 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10306 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10307 (actual_len > inode->i_size) &&
10308 (cur_offset > inode->i_size)) {
10309 if (cur_offset > actual_len)
10310 i_size = actual_len;
10312 i_size = cur_offset;
10313 i_size_write(inode, i_size);
10314 btrfs_ordered_update_i_size(inode, i_size, NULL);
10317 ret = btrfs_update_inode(trans, root, inode);
10320 btrfs_abort_transaction(trans, ret);
10322 btrfs_end_transaction(trans);
10327 btrfs_end_transaction(trans);
10329 if (cur_offset < end)
10330 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10331 end - cur_offset + 1);
10335 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10336 u64 start, u64 num_bytes, u64 min_size,
10337 loff_t actual_len, u64 *alloc_hint)
10339 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10340 min_size, actual_len, alloc_hint,
10344 int btrfs_prealloc_file_range_trans(struct inode *inode,
10345 struct btrfs_trans_handle *trans, int mode,
10346 u64 start, u64 num_bytes, u64 min_size,
10347 loff_t actual_len, u64 *alloc_hint)
10349 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10350 min_size, actual_len, alloc_hint, trans);
10353 static int btrfs_set_page_dirty(struct page *page)
10355 return __set_page_dirty_nobuffers(page);
10358 static int btrfs_permission(struct inode *inode, int mask)
10360 struct btrfs_root *root = BTRFS_I(inode)->root;
10361 umode_t mode = inode->i_mode;
10363 if (mask & MAY_WRITE &&
10364 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10365 if (btrfs_root_readonly(root))
10367 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10370 return generic_permission(inode, mask);
10373 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10375 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10376 struct btrfs_trans_handle *trans;
10377 struct btrfs_root *root = BTRFS_I(dir)->root;
10378 struct inode *inode = NULL;
10384 * 5 units required for adding orphan entry
10386 trans = btrfs_start_transaction(root, 5);
10388 return PTR_ERR(trans);
10390 ret = btrfs_find_free_ino(root, &objectid);
10394 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10395 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10396 if (IS_ERR(inode)) {
10397 ret = PTR_ERR(inode);
10402 inode->i_fop = &btrfs_file_operations;
10403 inode->i_op = &btrfs_file_inode_operations;
10405 inode->i_mapping->a_ops = &btrfs_aops;
10406 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10408 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10412 ret = btrfs_update_inode(trans, root, inode);
10415 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10420 * We set number of links to 0 in btrfs_new_inode(), and here we set
10421 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10424 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10426 set_nlink(inode, 1);
10427 d_tmpfile(dentry, inode);
10428 unlock_new_inode(inode);
10429 mark_inode_dirty(inode);
10431 btrfs_end_transaction(trans);
10433 discard_new_inode(inode);
10434 btrfs_btree_balance_dirty(fs_info);
10438 __attribute__((const))
10439 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10444 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10445 u64 start, u64 end)
10447 struct inode *inode = private_data;
10450 isize = i_size_read(inode);
10451 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10452 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10453 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10454 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10458 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10460 struct inode *inode = tree->private_data;
10461 unsigned long index = start >> PAGE_SHIFT;
10462 unsigned long end_index = end >> PAGE_SHIFT;
10465 while (index <= end_index) {
10466 page = find_get_page(inode->i_mapping, index);
10467 ASSERT(page); /* Pages should be in the extent_io_tree */
10468 set_page_writeback(page);
10474 static const struct inode_operations btrfs_dir_inode_operations = {
10475 .getattr = btrfs_getattr,
10476 .lookup = btrfs_lookup,
10477 .create = btrfs_create,
10478 .unlink = btrfs_unlink,
10479 .link = btrfs_link,
10480 .mkdir = btrfs_mkdir,
10481 .rmdir = btrfs_rmdir,
10482 .rename = btrfs_rename2,
10483 .symlink = btrfs_symlink,
10484 .setattr = btrfs_setattr,
10485 .mknod = btrfs_mknod,
10486 .listxattr = btrfs_listxattr,
10487 .permission = btrfs_permission,
10488 .get_acl = btrfs_get_acl,
10489 .set_acl = btrfs_set_acl,
10490 .update_time = btrfs_update_time,
10491 .tmpfile = btrfs_tmpfile,
10493 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10494 .lookup = btrfs_lookup,
10495 .permission = btrfs_permission,
10496 .update_time = btrfs_update_time,
10499 static const struct file_operations btrfs_dir_file_operations = {
10500 .llseek = generic_file_llseek,
10501 .read = generic_read_dir,
10502 .iterate_shared = btrfs_real_readdir,
10503 .open = btrfs_opendir,
10504 .unlocked_ioctl = btrfs_ioctl,
10505 #ifdef CONFIG_COMPAT
10506 .compat_ioctl = btrfs_compat_ioctl,
10508 .release = btrfs_release_file,
10509 .fsync = btrfs_sync_file,
10512 static const struct extent_io_ops btrfs_extent_io_ops = {
10513 /* mandatory callbacks */
10514 .submit_bio_hook = btrfs_submit_bio_hook,
10515 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10516 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10518 /* optional callbacks */
10519 .fill_delalloc = run_delalloc_range,
10520 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10521 .writepage_start_hook = btrfs_writepage_start_hook,
10522 .set_bit_hook = btrfs_set_bit_hook,
10523 .clear_bit_hook = btrfs_clear_bit_hook,
10524 .merge_extent_hook = btrfs_merge_extent_hook,
10525 .split_extent_hook = btrfs_split_extent_hook,
10526 .check_extent_io_range = btrfs_check_extent_io_range,
10530 * btrfs doesn't support the bmap operation because swapfiles
10531 * use bmap to make a mapping of extents in the file. They assume
10532 * these extents won't change over the life of the file and they
10533 * use the bmap result to do IO directly to the drive.
10535 * the btrfs bmap call would return logical addresses that aren't
10536 * suitable for IO and they also will change frequently as COW
10537 * operations happen. So, swapfile + btrfs == corruption.
10539 * For now we're avoiding this by dropping bmap.
10541 static const struct address_space_operations btrfs_aops = {
10542 .readpage = btrfs_readpage,
10543 .writepage = btrfs_writepage,
10544 .writepages = btrfs_writepages,
10545 .readpages = btrfs_readpages,
10546 .direct_IO = btrfs_direct_IO,
10547 .invalidatepage = btrfs_invalidatepage,
10548 .releasepage = btrfs_releasepage,
10549 .set_page_dirty = btrfs_set_page_dirty,
10550 .error_remove_page = generic_error_remove_page,
10553 static const struct address_space_operations btrfs_symlink_aops = {
10554 .readpage = btrfs_readpage,
10555 .writepage = btrfs_writepage,
10556 .invalidatepage = btrfs_invalidatepage,
10557 .releasepage = btrfs_releasepage,
10560 static const struct inode_operations btrfs_file_inode_operations = {
10561 .getattr = btrfs_getattr,
10562 .setattr = btrfs_setattr,
10563 .listxattr = btrfs_listxattr,
10564 .permission = btrfs_permission,
10565 .fiemap = btrfs_fiemap,
10566 .get_acl = btrfs_get_acl,
10567 .set_acl = btrfs_set_acl,
10568 .update_time = btrfs_update_time,
10570 static const struct inode_operations btrfs_special_inode_operations = {
10571 .getattr = btrfs_getattr,
10572 .setattr = btrfs_setattr,
10573 .permission = btrfs_permission,
10574 .listxattr = btrfs_listxattr,
10575 .get_acl = btrfs_get_acl,
10576 .set_acl = btrfs_set_acl,
10577 .update_time = btrfs_update_time,
10579 static const struct inode_operations btrfs_symlink_inode_operations = {
10580 .get_link = page_get_link,
10581 .getattr = btrfs_getattr,
10582 .setattr = btrfs_setattr,
10583 .permission = btrfs_permission,
10584 .listxattr = btrfs_listxattr,
10585 .update_time = btrfs_update_time,
10588 const struct dentry_operations btrfs_dentry_operations = {
10589 .d_delete = btrfs_dentry_delete,