1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <linux/migrate.h>
32 #include <linux/sched/mm.h>
33 #include <linux/iomap.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "ordered-data.h"
45 #include "compression.h"
47 #include "free-space-cache.h"
48 #include "inode-map.h"
51 #include "delalloc-space.h"
52 #include "block-group.h"
53 #include "space-info.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 struct btrfs_dio_data {
64 struct extent_changeset *data_reserved;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_special_inode_operations;
71 static const struct inode_operations btrfs_file_inode_operations;
72 static const struct address_space_operations btrfs_aops;
73 static const struct file_operations btrfs_dir_file_operations;
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
79 struct kmem_cache *btrfs_free_space_bitmap_cachep;
81 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
82 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
83 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
84 static noinline int cow_file_range(struct btrfs_inode *inode,
85 struct page *locked_page,
86 u64 start, u64 end, int *page_started,
87 unsigned long *nr_written, int unlock);
88 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
89 u64 len, u64 orig_start, u64 block_start,
90 u64 block_len, u64 orig_block_len,
91 u64 ram_bytes, int compress_type,
94 static void __endio_write_update_ordered(struct btrfs_inode *inode,
95 const u64 offset, const u64 bytes,
99 * Cleanup all submitted ordered extents in specified range to handle errors
100 * from the btrfs_run_delalloc_range() callback.
102 * NOTE: caller must ensure that when an error happens, it can not call
103 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
104 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
105 * to be released, which we want to happen only when finishing the ordered
106 * extent (btrfs_finish_ordered_io()).
108 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
109 struct page *locked_page,
110 u64 offset, u64 bytes)
112 unsigned long index = offset >> PAGE_SHIFT;
113 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
114 u64 page_start = page_offset(locked_page);
115 u64 page_end = page_start + PAGE_SIZE - 1;
119 while (index <= end_index) {
120 page = find_get_page(inode->vfs_inode.i_mapping, index);
124 ClearPagePrivate2(page);
129 * In case this page belongs to the delalloc range being instantiated
130 * then skip it, since the first page of a range is going to be
131 * properly cleaned up by the caller of run_delalloc_range
133 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
138 return __endio_write_update_ordered(inode, offset, bytes, false);
141 static int btrfs_dirty_inode(struct inode *inode);
143 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
144 struct inode *inode, struct inode *dir,
145 const struct qstr *qstr)
149 err = btrfs_init_acl(trans, inode, dir);
151 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
156 * this does all the hard work for inserting an inline extent into
157 * the btree. The caller should have done a btrfs_drop_extents so that
158 * no overlapping inline items exist in the btree
160 static int insert_inline_extent(struct btrfs_trans_handle *trans,
161 struct btrfs_path *path, int extent_inserted,
162 struct btrfs_root *root, struct inode *inode,
163 u64 start, size_t size, size_t compressed_size,
165 struct page **compressed_pages)
167 struct extent_buffer *leaf;
168 struct page *page = NULL;
171 struct btrfs_file_extent_item *ei;
173 size_t cur_size = size;
174 unsigned long offset;
176 ASSERT((compressed_size > 0 && compressed_pages) ||
177 (compressed_size == 0 && !compressed_pages));
179 if (compressed_size && compressed_pages)
180 cur_size = compressed_size;
182 inode_add_bytes(inode, size);
184 if (!extent_inserted) {
185 struct btrfs_key key;
188 key.objectid = btrfs_ino(BTRFS_I(inode));
190 key.type = BTRFS_EXTENT_DATA_KEY;
192 datasize = btrfs_file_extent_calc_inline_size(cur_size);
193 path->leave_spinning = 1;
194 ret = btrfs_insert_empty_item(trans, root, path, &key,
199 leaf = path->nodes[0];
200 ei = btrfs_item_ptr(leaf, path->slots[0],
201 struct btrfs_file_extent_item);
202 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
203 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
204 btrfs_set_file_extent_encryption(leaf, ei, 0);
205 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
206 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
207 ptr = btrfs_file_extent_inline_start(ei);
209 if (compress_type != BTRFS_COMPRESS_NONE) {
212 while (compressed_size > 0) {
213 cpage = compressed_pages[i];
214 cur_size = min_t(unsigned long, compressed_size,
217 kaddr = kmap_atomic(cpage);
218 write_extent_buffer(leaf, kaddr, ptr, cur_size);
219 kunmap_atomic(kaddr);
223 compressed_size -= cur_size;
225 btrfs_set_file_extent_compression(leaf, ei,
228 page = find_get_page(inode->i_mapping,
229 start >> PAGE_SHIFT);
230 btrfs_set_file_extent_compression(leaf, ei, 0);
231 kaddr = kmap_atomic(page);
232 offset = offset_in_page(start);
233 write_extent_buffer(leaf, kaddr + offset, ptr, size);
234 kunmap_atomic(kaddr);
237 btrfs_mark_buffer_dirty(leaf);
238 btrfs_release_path(path);
241 * We align size to sectorsize for inline extents just for simplicity
244 size = ALIGN(size, root->fs_info->sectorsize);
245 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
250 * we're an inline extent, so nobody can
251 * extend the file past i_size without locking
252 * a page we already have locked.
254 * We must do any isize and inode updates
255 * before we unlock the pages. Otherwise we
256 * could end up racing with unlink.
258 BTRFS_I(inode)->disk_i_size = inode->i_size;
259 ret = btrfs_update_inode(trans, root, inode);
267 * conditionally insert an inline extent into the file. This
268 * does the checks required to make sure the data is small enough
269 * to fit as an inline extent.
271 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
272 u64 end, size_t compressed_size,
274 struct page **compressed_pages)
276 struct btrfs_root *root = inode->root;
277 struct btrfs_fs_info *fs_info = root->fs_info;
278 struct btrfs_trans_handle *trans;
279 u64 isize = i_size_read(&inode->vfs_inode);
280 u64 actual_end = min(end + 1, isize);
281 u64 inline_len = actual_end - start;
282 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
283 u64 data_len = inline_len;
285 struct btrfs_path *path;
286 int extent_inserted = 0;
287 u32 extent_item_size;
290 data_len = compressed_size;
293 actual_end > fs_info->sectorsize ||
294 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
296 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
298 data_len > fs_info->max_inline) {
302 path = btrfs_alloc_path();
306 trans = btrfs_join_transaction(root);
308 btrfs_free_path(path);
309 return PTR_ERR(trans);
311 trans->block_rsv = &inode->block_rsv;
313 if (compressed_size && compressed_pages)
314 extent_item_size = btrfs_file_extent_calc_inline_size(
317 extent_item_size = btrfs_file_extent_calc_inline_size(
320 ret = __btrfs_drop_extents(trans, root, inode, path, start, aligned_end,
321 NULL, 1, 1, extent_item_size,
324 btrfs_abort_transaction(trans, ret);
328 if (isize > actual_end)
329 inline_len = min_t(u64, isize, actual_end);
330 ret = insert_inline_extent(trans, path, extent_inserted,
331 root, &inode->vfs_inode, start,
332 inline_len, compressed_size,
333 compress_type, compressed_pages);
334 if (ret && ret != -ENOSPC) {
335 btrfs_abort_transaction(trans, ret);
337 } else if (ret == -ENOSPC) {
342 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
343 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
346 * Don't forget to free the reserved space, as for inlined extent
347 * it won't count as data extent, free them directly here.
348 * And at reserve time, it's always aligned to page size, so
349 * just free one page here.
351 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
352 btrfs_free_path(path);
353 btrfs_end_transaction(trans);
357 struct async_extent {
362 unsigned long nr_pages;
364 struct list_head list;
369 struct page *locked_page;
372 unsigned int write_flags;
373 struct list_head extents;
374 struct cgroup_subsys_state *blkcg_css;
375 struct btrfs_work work;
380 /* Number of chunks in flight; must be first in the structure */
382 struct async_chunk chunks[];
385 static noinline int add_async_extent(struct async_chunk *cow,
386 u64 start, u64 ram_size,
389 unsigned long nr_pages,
392 struct async_extent *async_extent;
394 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
395 BUG_ON(!async_extent); /* -ENOMEM */
396 async_extent->start = start;
397 async_extent->ram_size = ram_size;
398 async_extent->compressed_size = compressed_size;
399 async_extent->pages = pages;
400 async_extent->nr_pages = nr_pages;
401 async_extent->compress_type = compress_type;
402 list_add_tail(&async_extent->list, &cow->extents);
407 * Check if the inode has flags compatible with compression
409 static inline bool inode_can_compress(struct btrfs_inode *inode)
411 if (inode->flags & BTRFS_INODE_NODATACOW ||
412 inode->flags & BTRFS_INODE_NODATASUM)
418 * Check if the inode needs to be submitted to compression, based on mount
419 * options, defragmentation, properties or heuristics.
421 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
424 struct btrfs_fs_info *fs_info = inode->root->fs_info;
426 if (!inode_can_compress(inode)) {
427 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
428 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
433 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
436 if (inode->defrag_compress)
438 /* bad compression ratios */
439 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
441 if (btrfs_test_opt(fs_info, COMPRESS) ||
442 inode->flags & BTRFS_INODE_COMPRESS ||
443 inode->prop_compress)
444 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
448 static inline void inode_should_defrag(struct btrfs_inode *inode,
449 u64 start, u64 end, u64 num_bytes, u64 small_write)
451 /* If this is a small write inside eof, kick off a defrag */
452 if (num_bytes < small_write &&
453 (start > 0 || end + 1 < inode->disk_i_size))
454 btrfs_add_inode_defrag(NULL, inode);
458 * we create compressed extents in two phases. The first
459 * phase compresses a range of pages that have already been
460 * locked (both pages and state bits are locked).
462 * This is done inside an ordered work queue, and the compression
463 * is spread across many cpus. The actual IO submission is step
464 * two, and the ordered work queue takes care of making sure that
465 * happens in the same order things were put onto the queue by
466 * writepages and friends.
468 * If this code finds it can't get good compression, it puts an
469 * entry onto the work queue to write the uncompressed bytes. This
470 * makes sure that both compressed inodes and uncompressed inodes
471 * are written in the same order that the flusher thread sent them
474 static noinline int compress_file_range(struct async_chunk *async_chunk)
476 struct inode *inode = async_chunk->inode;
477 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
478 u64 blocksize = fs_info->sectorsize;
479 u64 start = async_chunk->start;
480 u64 end = async_chunk->end;
484 struct page **pages = NULL;
485 unsigned long nr_pages;
486 unsigned long total_compressed = 0;
487 unsigned long total_in = 0;
490 int compress_type = fs_info->compress_type;
491 int compressed_extents = 0;
494 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
498 * We need to save i_size before now because it could change in between
499 * us evaluating the size and assigning it. This is because we lock and
500 * unlock the page in truncate and fallocate, and then modify the i_size
503 * The barriers are to emulate READ_ONCE, remove that once i_size_read
507 i_size = i_size_read(inode);
509 actual_end = min_t(u64, i_size, end + 1);
512 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
513 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
514 nr_pages = min_t(unsigned long, nr_pages,
515 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
518 * we don't want to send crud past the end of i_size through
519 * compression, that's just a waste of CPU time. So, if the
520 * end of the file is before the start of our current
521 * requested range of bytes, we bail out to the uncompressed
522 * cleanup code that can deal with all of this.
524 * It isn't really the fastest way to fix things, but this is a
525 * very uncommon corner.
527 if (actual_end <= start)
528 goto cleanup_and_bail_uncompressed;
530 total_compressed = actual_end - start;
533 * skip compression for a small file range(<=blocksize) that
534 * isn't an inline extent, since it doesn't save disk space at all.
536 if (total_compressed <= blocksize &&
537 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
538 goto cleanup_and_bail_uncompressed;
540 total_compressed = min_t(unsigned long, total_compressed,
541 BTRFS_MAX_UNCOMPRESSED);
546 * we do compression for mount -o compress and when the
547 * inode has not been flagged as nocompress. This flag can
548 * change at any time if we discover bad compression ratios.
550 if (inode_need_compress(BTRFS_I(inode), start, end)) {
552 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
554 /* just bail out to the uncompressed code */
559 if (BTRFS_I(inode)->defrag_compress)
560 compress_type = BTRFS_I(inode)->defrag_compress;
561 else if (BTRFS_I(inode)->prop_compress)
562 compress_type = BTRFS_I(inode)->prop_compress;
565 * we need to call clear_page_dirty_for_io on each
566 * page in the range. Otherwise applications with the file
567 * mmap'd can wander in and change the page contents while
568 * we are compressing them.
570 * If the compression fails for any reason, we set the pages
571 * dirty again later on.
573 * Note that the remaining part is redirtied, the start pointer
574 * has moved, the end is the original one.
577 extent_range_clear_dirty_for_io(inode, start, end);
581 /* Compression level is applied here and only here */
582 ret = btrfs_compress_pages(
583 compress_type | (fs_info->compress_level << 4),
584 inode->i_mapping, start,
591 unsigned long offset = offset_in_page(total_compressed);
592 struct page *page = pages[nr_pages - 1];
595 /* zero the tail end of the last page, we might be
596 * sending it down to disk
599 kaddr = kmap_atomic(page);
600 memset(kaddr + offset, 0,
602 kunmap_atomic(kaddr);
609 /* lets try to make an inline extent */
610 if (ret || total_in < actual_end) {
611 /* we didn't compress the entire range, try
612 * to make an uncompressed inline extent.
614 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
615 0, BTRFS_COMPRESS_NONE,
618 /* try making a compressed inline extent */
619 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
621 compress_type, pages);
624 unsigned long clear_flags = EXTENT_DELALLOC |
625 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
626 EXTENT_DO_ACCOUNTING;
627 unsigned long page_error_op;
629 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
632 * inline extent creation worked or returned error,
633 * we don't need to create any more async work items.
634 * Unlock and free up our temp pages.
636 * We use DO_ACCOUNTING here because we need the
637 * delalloc_release_metadata to be done _after_ we drop
638 * our outstanding extent for clearing delalloc for this
641 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
651 * Ensure we only free the compressed pages if we have
652 * them allocated, as we can still reach here with
653 * inode_need_compress() == false.
656 for (i = 0; i < nr_pages; i++) {
657 WARN_ON(pages[i]->mapping);
668 * we aren't doing an inline extent round the compressed size
669 * up to a block size boundary so the allocator does sane
672 total_compressed = ALIGN(total_compressed, blocksize);
675 * one last check to make sure the compression is really a
676 * win, compare the page count read with the blocks on disk,
677 * compression must free at least one sector size
679 total_in = ALIGN(total_in, PAGE_SIZE);
680 if (total_compressed + blocksize <= total_in) {
681 compressed_extents++;
684 * The async work queues will take care of doing actual
685 * allocation on disk for these compressed pages, and
686 * will submit them to the elevator.
688 add_async_extent(async_chunk, start, total_in,
689 total_compressed, pages, nr_pages,
692 if (start + total_in < end) {
698 return compressed_extents;
703 * the compression code ran but failed to make things smaller,
704 * free any pages it allocated and our page pointer array
706 for (i = 0; i < nr_pages; i++) {
707 WARN_ON(pages[i]->mapping);
712 total_compressed = 0;
715 /* flag the file so we don't compress in the future */
716 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
717 !(BTRFS_I(inode)->prop_compress)) {
718 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
721 cleanup_and_bail_uncompressed:
723 * No compression, but we still need to write the pages in the file
724 * we've been given so far. redirty the locked page if it corresponds
725 * to our extent and set things up for the async work queue to run
726 * cow_file_range to do the normal delalloc dance.
728 if (async_chunk->locked_page &&
729 (page_offset(async_chunk->locked_page) >= start &&
730 page_offset(async_chunk->locked_page)) <= end) {
731 __set_page_dirty_nobuffers(async_chunk->locked_page);
732 /* unlocked later on in the async handlers */
736 extent_range_redirty_for_io(inode, start, end);
737 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
738 BTRFS_COMPRESS_NONE);
739 compressed_extents++;
741 return compressed_extents;
744 static void free_async_extent_pages(struct async_extent *async_extent)
748 if (!async_extent->pages)
751 for (i = 0; i < async_extent->nr_pages; i++) {
752 WARN_ON(async_extent->pages[i]->mapping);
753 put_page(async_extent->pages[i]);
755 kfree(async_extent->pages);
756 async_extent->nr_pages = 0;
757 async_extent->pages = NULL;
761 * phase two of compressed writeback. This is the ordered portion
762 * of the code, which only gets called in the order the work was
763 * queued. We walk all the async extents created by compress_file_range
764 * and send them down to the disk.
766 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
768 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
769 struct btrfs_fs_info *fs_info = inode->root->fs_info;
770 struct async_extent *async_extent;
772 struct btrfs_key ins;
773 struct extent_map *em;
774 struct btrfs_root *root = inode->root;
775 struct extent_io_tree *io_tree = &inode->io_tree;
779 while (!list_empty(&async_chunk->extents)) {
780 async_extent = list_entry(async_chunk->extents.next,
781 struct async_extent, list);
782 list_del(&async_extent->list);
785 lock_extent(io_tree, async_extent->start,
786 async_extent->start + async_extent->ram_size - 1);
787 /* did the compression code fall back to uncompressed IO? */
788 if (!async_extent->pages) {
789 int page_started = 0;
790 unsigned long nr_written = 0;
792 /* allocate blocks */
793 ret = cow_file_range(inode, async_chunk->locked_page,
795 async_extent->start +
796 async_extent->ram_size - 1,
797 &page_started, &nr_written, 0);
802 * if page_started, cow_file_range inserted an
803 * inline extent and took care of all the unlocking
804 * and IO for us. Otherwise, we need to submit
805 * all those pages down to the drive.
807 if (!page_started && !ret)
808 extent_write_locked_range(&inode->vfs_inode,
810 async_extent->start +
811 async_extent->ram_size - 1,
813 else if (ret && async_chunk->locked_page)
814 unlock_page(async_chunk->locked_page);
820 ret = btrfs_reserve_extent(root, async_extent->ram_size,
821 async_extent->compressed_size,
822 async_extent->compressed_size,
823 0, alloc_hint, &ins, 1, 1);
825 free_async_extent_pages(async_extent);
827 if (ret == -ENOSPC) {
828 unlock_extent(io_tree, async_extent->start,
829 async_extent->start +
830 async_extent->ram_size - 1);
833 * we need to redirty the pages if we decide to
834 * fallback to uncompressed IO, otherwise we
835 * will not submit these pages down to lower
838 extent_range_redirty_for_io(&inode->vfs_inode,
840 async_extent->start +
841 async_extent->ram_size - 1);
848 * here we're doing allocation and writeback of the
851 em = create_io_em(inode, async_extent->start,
852 async_extent->ram_size, /* len */
853 async_extent->start, /* orig_start */
854 ins.objectid, /* block_start */
855 ins.offset, /* block_len */
856 ins.offset, /* orig_block_len */
857 async_extent->ram_size, /* ram_bytes */
858 async_extent->compress_type,
859 BTRFS_ORDERED_COMPRESSED);
861 /* ret value is not necessary due to void function */
862 goto out_free_reserve;
865 ret = btrfs_add_ordered_extent_compress(inode,
868 async_extent->ram_size,
870 BTRFS_ORDERED_COMPRESSED,
871 async_extent->compress_type);
873 btrfs_drop_extent_cache(inode, async_extent->start,
874 async_extent->start +
875 async_extent->ram_size - 1, 0);
876 goto out_free_reserve;
878 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
881 * clear dirty, set writeback and unlock the pages.
883 extent_clear_unlock_delalloc(inode, async_extent->start,
884 async_extent->start +
885 async_extent->ram_size - 1,
886 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
887 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
889 if (btrfs_submit_compressed_write(inode, async_extent->start,
890 async_extent->ram_size,
892 ins.offset, async_extent->pages,
893 async_extent->nr_pages,
894 async_chunk->write_flags,
895 async_chunk->blkcg_css)) {
896 struct page *p = async_extent->pages[0];
897 const u64 start = async_extent->start;
898 const u64 end = start + async_extent->ram_size - 1;
900 p->mapping = inode->vfs_inode.i_mapping;
901 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
904 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
907 free_async_extent_pages(async_extent);
909 alloc_hint = ins.objectid + ins.offset;
915 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
916 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
918 extent_clear_unlock_delalloc(inode, async_extent->start,
919 async_extent->start +
920 async_extent->ram_size - 1,
921 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
922 EXTENT_DELALLOC_NEW |
923 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
924 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
925 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
927 free_async_extent_pages(async_extent);
932 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
935 struct extent_map_tree *em_tree = &inode->extent_tree;
936 struct extent_map *em;
939 read_lock(&em_tree->lock);
940 em = search_extent_mapping(em_tree, start, num_bytes);
943 * if block start isn't an actual block number then find the
944 * first block in this inode and use that as a hint. If that
945 * block is also bogus then just don't worry about it.
947 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
949 em = search_extent_mapping(em_tree, 0, 0);
950 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
951 alloc_hint = em->block_start;
955 alloc_hint = em->block_start;
959 read_unlock(&em_tree->lock);
965 * when extent_io.c finds a delayed allocation range in the file,
966 * the call backs end up in this code. The basic idea is to
967 * allocate extents on disk for the range, and create ordered data structs
968 * in ram to track those extents.
970 * locked_page is the page that writepage had locked already. We use
971 * it to make sure we don't do extra locks or unlocks.
973 * *page_started is set to one if we unlock locked_page and do everything
974 * required to start IO on it. It may be clean and already done with
977 static noinline int cow_file_range(struct btrfs_inode *inode,
978 struct page *locked_page,
979 u64 start, u64 end, int *page_started,
980 unsigned long *nr_written, int unlock)
982 struct btrfs_root *root = inode->root;
983 struct btrfs_fs_info *fs_info = root->fs_info;
986 unsigned long ram_size;
987 u64 cur_alloc_size = 0;
989 u64 blocksize = fs_info->sectorsize;
990 struct btrfs_key ins;
991 struct extent_map *em;
993 unsigned long page_ops;
994 bool extent_reserved = false;
997 if (btrfs_is_free_space_inode(inode)) {
1003 num_bytes = ALIGN(end - start + 1, blocksize);
1004 num_bytes = max(blocksize, num_bytes);
1005 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1007 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1010 /* lets try to make an inline extent */
1011 ret = cow_file_range_inline(inode, start, end, 0,
1012 BTRFS_COMPRESS_NONE, NULL);
1015 * We use DO_ACCOUNTING here because we need the
1016 * delalloc_release_metadata to be run _after_ we drop
1017 * our outstanding extent for clearing delalloc for this
1020 extent_clear_unlock_delalloc(inode, start, end, NULL,
1021 EXTENT_LOCKED | EXTENT_DELALLOC |
1022 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1023 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1024 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1025 PAGE_END_WRITEBACK);
1026 *nr_written = *nr_written +
1027 (end - start + PAGE_SIZE) / PAGE_SIZE;
1030 } else if (ret < 0) {
1035 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1036 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1039 * Relocation relies on the relocated extents to have exactly the same
1040 * size as the original extents. Normally writeback for relocation data
1041 * extents follows a NOCOW path because relocation preallocates the
1042 * extents. However, due to an operation such as scrub turning a block
1043 * group to RO mode, it may fallback to COW mode, so we must make sure
1044 * an extent allocated during COW has exactly the requested size and can
1045 * not be split into smaller extents, otherwise relocation breaks and
1046 * fails during the stage where it updates the bytenr of file extent
1049 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1050 min_alloc_size = num_bytes;
1052 min_alloc_size = fs_info->sectorsize;
1054 while (num_bytes > 0) {
1055 cur_alloc_size = num_bytes;
1056 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1057 min_alloc_size, 0, alloc_hint,
1061 cur_alloc_size = ins.offset;
1062 extent_reserved = true;
1064 ram_size = ins.offset;
1065 em = create_io_em(inode, start, ins.offset, /* len */
1066 start, /* orig_start */
1067 ins.objectid, /* block_start */
1068 ins.offset, /* block_len */
1069 ins.offset, /* orig_block_len */
1070 ram_size, /* ram_bytes */
1071 BTRFS_COMPRESS_NONE, /* compress_type */
1072 BTRFS_ORDERED_REGULAR /* type */);
1077 free_extent_map(em);
1079 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1080 ram_size, cur_alloc_size, 0);
1082 goto out_drop_extent_cache;
1084 if (root->root_key.objectid ==
1085 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1086 ret = btrfs_reloc_clone_csums(inode, start,
1089 * Only drop cache here, and process as normal.
1091 * We must not allow extent_clear_unlock_delalloc()
1092 * at out_unlock label to free meta of this ordered
1093 * extent, as its meta should be freed by
1094 * btrfs_finish_ordered_io().
1096 * So we must continue until @start is increased to
1097 * skip current ordered extent.
1100 btrfs_drop_extent_cache(inode, start,
1101 start + ram_size - 1, 0);
1104 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1106 /* we're not doing compressed IO, don't unlock the first
1107 * page (which the caller expects to stay locked), don't
1108 * clear any dirty bits and don't set any writeback bits
1110 * Do set the Private2 bit so we know this page was properly
1111 * setup for writepage
1113 page_ops = unlock ? PAGE_UNLOCK : 0;
1114 page_ops |= PAGE_SET_PRIVATE2;
1116 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1118 EXTENT_LOCKED | EXTENT_DELALLOC,
1120 if (num_bytes < cur_alloc_size)
1123 num_bytes -= cur_alloc_size;
1124 alloc_hint = ins.objectid + ins.offset;
1125 start += cur_alloc_size;
1126 extent_reserved = false;
1129 * btrfs_reloc_clone_csums() error, since start is increased
1130 * extent_clear_unlock_delalloc() at out_unlock label won't
1131 * free metadata of current ordered extent, we're OK to exit.
1139 out_drop_extent_cache:
1140 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1142 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1143 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1145 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1146 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1147 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1150 * If we reserved an extent for our delalloc range (or a subrange) and
1151 * failed to create the respective ordered extent, then it means that
1152 * when we reserved the extent we decremented the extent's size from
1153 * the data space_info's bytes_may_use counter and incremented the
1154 * space_info's bytes_reserved counter by the same amount. We must make
1155 * sure extent_clear_unlock_delalloc() does not try to decrement again
1156 * the data space_info's bytes_may_use counter, therefore we do not pass
1157 * it the flag EXTENT_CLEAR_DATA_RESV.
1159 if (extent_reserved) {
1160 extent_clear_unlock_delalloc(inode, start,
1161 start + cur_alloc_size - 1,
1165 start += cur_alloc_size;
1169 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1170 clear_bits | EXTENT_CLEAR_DATA_RESV,
1176 * work queue call back to started compression on a file and pages
1178 static noinline void async_cow_start(struct btrfs_work *work)
1180 struct async_chunk *async_chunk;
1181 int compressed_extents;
1183 async_chunk = container_of(work, struct async_chunk, work);
1185 compressed_extents = compress_file_range(async_chunk);
1186 if (compressed_extents == 0) {
1187 btrfs_add_delayed_iput(async_chunk->inode);
1188 async_chunk->inode = NULL;
1193 * work queue call back to submit previously compressed pages
1195 static noinline void async_cow_submit(struct btrfs_work *work)
1197 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1199 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1200 unsigned long nr_pages;
1202 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1205 /* atomic_sub_return implies a barrier */
1206 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1208 cond_wake_up_nomb(&fs_info->async_submit_wait);
1211 * ->inode could be NULL if async_chunk_start has failed to compress,
1212 * in which case we don't have anything to submit, yet we need to
1213 * always adjust ->async_delalloc_pages as its paired with the init
1214 * happening in cow_file_range_async
1216 if (async_chunk->inode)
1217 submit_compressed_extents(async_chunk);
1220 static noinline void async_cow_free(struct btrfs_work *work)
1222 struct async_chunk *async_chunk;
1224 async_chunk = container_of(work, struct async_chunk, work);
1225 if (async_chunk->inode)
1226 btrfs_add_delayed_iput(async_chunk->inode);
1227 if (async_chunk->blkcg_css)
1228 css_put(async_chunk->blkcg_css);
1230 * Since the pointer to 'pending' is at the beginning of the array of
1231 * async_chunk's, freeing it ensures the whole array has been freed.
1233 if (atomic_dec_and_test(async_chunk->pending))
1234 kvfree(async_chunk->pending);
1237 static int cow_file_range_async(struct btrfs_inode *inode,
1238 struct writeback_control *wbc,
1239 struct page *locked_page,
1240 u64 start, u64 end, int *page_started,
1241 unsigned long *nr_written)
1243 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1244 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1245 struct async_cow *ctx;
1246 struct async_chunk *async_chunk;
1247 unsigned long nr_pages;
1249 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1251 bool should_compress;
1253 const unsigned int write_flags = wbc_to_write_flags(wbc);
1255 unlock_extent(&inode->io_tree, start, end);
1257 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1258 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1260 should_compress = false;
1262 should_compress = true;
1265 nofs_flag = memalloc_nofs_save();
1266 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1267 memalloc_nofs_restore(nofs_flag);
1270 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1271 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1272 EXTENT_DO_ACCOUNTING;
1273 unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1274 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
1277 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1278 clear_bits, page_ops);
1282 async_chunk = ctx->chunks;
1283 atomic_set(&ctx->num_chunks, num_chunks);
1285 for (i = 0; i < num_chunks; i++) {
1286 if (should_compress)
1287 cur_end = min(end, start + SZ_512K - 1);
1292 * igrab is called higher up in the call chain, take only the
1293 * lightweight reference for the callback lifetime
1295 ihold(&inode->vfs_inode);
1296 async_chunk[i].pending = &ctx->num_chunks;
1297 async_chunk[i].inode = &inode->vfs_inode;
1298 async_chunk[i].start = start;
1299 async_chunk[i].end = cur_end;
1300 async_chunk[i].write_flags = write_flags;
1301 INIT_LIST_HEAD(&async_chunk[i].extents);
1304 * The locked_page comes all the way from writepage and its
1305 * the original page we were actually given. As we spread
1306 * this large delalloc region across multiple async_chunk
1307 * structs, only the first struct needs a pointer to locked_page
1309 * This way we don't need racey decisions about who is supposed
1314 * Depending on the compressibility, the pages might or
1315 * might not go through async. We want all of them to
1316 * be accounted against wbc once. Let's do it here
1317 * before the paths diverge. wbc accounting is used
1318 * only for foreign writeback detection and doesn't
1319 * need full accuracy. Just account the whole thing
1320 * against the first page.
1322 wbc_account_cgroup_owner(wbc, locked_page,
1324 async_chunk[i].locked_page = locked_page;
1327 async_chunk[i].locked_page = NULL;
1330 if (blkcg_css != blkcg_root_css) {
1332 async_chunk[i].blkcg_css = blkcg_css;
1334 async_chunk[i].blkcg_css = NULL;
1337 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1338 async_cow_submit, async_cow_free);
1340 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1341 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1343 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1345 *nr_written += nr_pages;
1346 start = cur_end + 1;
1352 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1353 u64 bytenr, u64 num_bytes)
1356 struct btrfs_ordered_sum *sums;
1359 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1360 bytenr + num_bytes - 1, &list, 0);
1361 if (ret == 0 && list_empty(&list))
1364 while (!list_empty(&list)) {
1365 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1366 list_del(&sums->list);
1374 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1375 const u64 start, const u64 end,
1376 int *page_started, unsigned long *nr_written)
1378 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1379 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1380 BTRFS_DATA_RELOC_TREE_OBJECTID);
1381 const u64 range_bytes = end + 1 - start;
1382 struct extent_io_tree *io_tree = &inode->io_tree;
1383 u64 range_start = start;
1387 * If EXTENT_NORESERVE is set it means that when the buffered write was
1388 * made we had not enough available data space and therefore we did not
1389 * reserve data space for it, since we though we could do NOCOW for the
1390 * respective file range (either there is prealloc extent or the inode
1391 * has the NOCOW bit set).
1393 * However when we need to fallback to COW mode (because for example the
1394 * block group for the corresponding extent was turned to RO mode by a
1395 * scrub or relocation) we need to do the following:
1397 * 1) We increment the bytes_may_use counter of the data space info.
1398 * If COW succeeds, it allocates a new data extent and after doing
1399 * that it decrements the space info's bytes_may_use counter and
1400 * increments its bytes_reserved counter by the same amount (we do
1401 * this at btrfs_add_reserved_bytes()). So we need to increment the
1402 * bytes_may_use counter to compensate (when space is reserved at
1403 * buffered write time, the bytes_may_use counter is incremented);
1405 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1406 * that if the COW path fails for any reason, it decrements (through
1407 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1408 * data space info, which we incremented in the step above.
1410 * If we need to fallback to cow and the inode corresponds to a free
1411 * space cache inode or an inode of the data relocation tree, we must
1412 * also increment bytes_may_use of the data space_info for the same
1413 * reason. Space caches and relocated data extents always get a prealloc
1414 * extent for them, however scrub or balance may have set the block
1415 * group that contains that extent to RO mode and therefore force COW
1416 * when starting writeback.
1418 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1419 EXTENT_NORESERVE, 0);
1420 if (count > 0 || is_space_ino || is_reloc_ino) {
1422 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1423 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1425 if (is_space_ino || is_reloc_ino)
1426 bytes = range_bytes;
1428 spin_lock(&sinfo->lock);
1429 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1430 spin_unlock(&sinfo->lock);
1433 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1437 return cow_file_range(inode, locked_page, start, end, page_started,
1442 * when nowcow writeback call back. This checks for snapshots or COW copies
1443 * of the extents that exist in the file, and COWs the file as required.
1445 * If no cow copies or snapshots exist, we write directly to the existing
1448 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1449 struct page *locked_page,
1450 const u64 start, const u64 end,
1451 int *page_started, int force,
1452 unsigned long *nr_written)
1454 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1455 struct btrfs_root *root = inode->root;
1456 struct btrfs_path *path;
1457 u64 cow_start = (u64)-1;
1458 u64 cur_offset = start;
1460 bool check_prev = true;
1461 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1462 u64 ino = btrfs_ino(inode);
1464 u64 disk_bytenr = 0;
1466 path = btrfs_alloc_path();
1468 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1469 EXTENT_LOCKED | EXTENT_DELALLOC |
1470 EXTENT_DO_ACCOUNTING |
1471 EXTENT_DEFRAG, PAGE_UNLOCK |
1473 PAGE_SET_WRITEBACK |
1474 PAGE_END_WRITEBACK);
1479 struct btrfs_key found_key;
1480 struct btrfs_file_extent_item *fi;
1481 struct extent_buffer *leaf;
1491 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1497 * If there is no extent for our range when doing the initial
1498 * search, then go back to the previous slot as it will be the
1499 * one containing the search offset
1501 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1502 leaf = path->nodes[0];
1503 btrfs_item_key_to_cpu(leaf, &found_key,
1504 path->slots[0] - 1);
1505 if (found_key.objectid == ino &&
1506 found_key.type == BTRFS_EXTENT_DATA_KEY)
1511 /* Go to next leaf if we have exhausted the current one */
1512 leaf = path->nodes[0];
1513 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1514 ret = btrfs_next_leaf(root, path);
1516 if (cow_start != (u64)-1)
1517 cur_offset = cow_start;
1522 leaf = path->nodes[0];
1525 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1527 /* Didn't find anything for our INO */
1528 if (found_key.objectid > ino)
1531 * Keep searching until we find an EXTENT_ITEM or there are no
1532 * more extents for this inode
1534 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1535 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1540 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1541 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1542 found_key.offset > end)
1546 * If the found extent starts after requested offset, then
1547 * adjust extent_end to be right before this extent begins
1549 if (found_key.offset > cur_offset) {
1550 extent_end = found_key.offset;
1556 * Found extent which begins before our range and potentially
1559 fi = btrfs_item_ptr(leaf, path->slots[0],
1560 struct btrfs_file_extent_item);
1561 extent_type = btrfs_file_extent_type(leaf, fi);
1563 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1564 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1565 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1566 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1567 extent_offset = btrfs_file_extent_offset(leaf, fi);
1568 extent_end = found_key.offset +
1569 btrfs_file_extent_num_bytes(leaf, fi);
1571 btrfs_file_extent_disk_num_bytes(leaf, fi);
1573 * If the extent we got ends before our current offset,
1574 * skip to the next extent.
1576 if (extent_end <= cur_offset) {
1581 if (disk_bytenr == 0)
1583 /* Skip compressed/encrypted/encoded extents */
1584 if (btrfs_file_extent_compression(leaf, fi) ||
1585 btrfs_file_extent_encryption(leaf, fi) ||
1586 btrfs_file_extent_other_encoding(leaf, fi))
1589 * If extent is created before the last volume's snapshot
1590 * this implies the extent is shared, hence we can't do
1591 * nocow. This is the same check as in
1592 * btrfs_cross_ref_exist but without calling
1593 * btrfs_search_slot.
1595 if (!freespace_inode &&
1596 btrfs_file_extent_generation(leaf, fi) <=
1597 btrfs_root_last_snapshot(&root->root_item))
1599 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1601 /* If extent is RO, we must COW it */
1602 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1604 ret = btrfs_cross_ref_exist(root, ino,
1606 extent_offset, disk_bytenr, false);
1609 * ret could be -EIO if the above fails to read
1613 if (cow_start != (u64)-1)
1614 cur_offset = cow_start;
1618 WARN_ON_ONCE(freespace_inode);
1621 disk_bytenr += extent_offset;
1622 disk_bytenr += cur_offset - found_key.offset;
1623 num_bytes = min(end + 1, extent_end) - cur_offset;
1625 * If there are pending snapshots for this root, we
1626 * fall into common COW way
1628 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1631 * force cow if csum exists in the range.
1632 * this ensure that csum for a given extent are
1633 * either valid or do not exist.
1635 ret = csum_exist_in_range(fs_info, disk_bytenr,
1639 * ret could be -EIO if the above fails to read
1643 if (cow_start != (u64)-1)
1644 cur_offset = cow_start;
1647 WARN_ON_ONCE(freespace_inode);
1650 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1653 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1654 extent_end = found_key.offset + ram_bytes;
1655 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1656 /* Skip extents outside of our requested range */
1657 if (extent_end <= start) {
1662 /* If this triggers then we have a memory corruption */
1667 * If nocow is false then record the beginning of the range
1668 * that needs to be COWed
1671 if (cow_start == (u64)-1)
1672 cow_start = cur_offset;
1673 cur_offset = extent_end;
1674 if (cur_offset > end)
1680 btrfs_release_path(path);
1683 * COW range from cow_start to found_key.offset - 1. As the key
1684 * will contain the beginning of the first extent that can be
1685 * NOCOW, following one which needs to be COW'ed
1687 if (cow_start != (u64)-1) {
1688 ret = fallback_to_cow(inode, locked_page,
1689 cow_start, found_key.offset - 1,
1690 page_started, nr_written);
1693 cow_start = (u64)-1;
1696 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1697 u64 orig_start = found_key.offset - extent_offset;
1698 struct extent_map *em;
1700 em = create_io_em(inode, cur_offset, num_bytes,
1702 disk_bytenr, /* block_start */
1703 num_bytes, /* block_len */
1704 disk_num_bytes, /* orig_block_len */
1705 ram_bytes, BTRFS_COMPRESS_NONE,
1706 BTRFS_ORDERED_PREALLOC);
1711 free_extent_map(em);
1712 ret = btrfs_add_ordered_extent(inode, cur_offset,
1713 disk_bytenr, num_bytes,
1715 BTRFS_ORDERED_PREALLOC);
1717 btrfs_drop_extent_cache(inode, cur_offset,
1718 cur_offset + num_bytes - 1,
1723 ret = btrfs_add_ordered_extent(inode, cur_offset,
1724 disk_bytenr, num_bytes,
1726 BTRFS_ORDERED_NOCOW);
1732 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1735 if (root->root_key.objectid ==
1736 BTRFS_DATA_RELOC_TREE_OBJECTID)
1738 * Error handled later, as we must prevent
1739 * extent_clear_unlock_delalloc() in error handler
1740 * from freeing metadata of created ordered extent.
1742 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1745 extent_clear_unlock_delalloc(inode, cur_offset,
1746 cur_offset + num_bytes - 1,
1747 locked_page, EXTENT_LOCKED |
1749 EXTENT_CLEAR_DATA_RESV,
1750 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1752 cur_offset = extent_end;
1755 * btrfs_reloc_clone_csums() error, now we're OK to call error
1756 * handler, as metadata for created ordered extent will only
1757 * be freed by btrfs_finish_ordered_io().
1761 if (cur_offset > end)
1764 btrfs_release_path(path);
1766 if (cur_offset <= end && cow_start == (u64)-1)
1767 cow_start = cur_offset;
1769 if (cow_start != (u64)-1) {
1771 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1772 page_started, nr_written);
1779 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1781 if (ret && cur_offset < end)
1782 extent_clear_unlock_delalloc(inode, cur_offset, end,
1783 locked_page, EXTENT_LOCKED |
1784 EXTENT_DELALLOC | EXTENT_DEFRAG |
1785 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1787 PAGE_SET_WRITEBACK |
1788 PAGE_END_WRITEBACK);
1789 btrfs_free_path(path);
1793 static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end)
1796 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1797 !(inode->flags & BTRFS_INODE_PREALLOC))
1801 * @defrag_bytes is a hint value, no spinlock held here,
1802 * if is not zero, it means the file is defragging.
1803 * Force cow if given extent needs to be defragged.
1805 if (inode->defrag_bytes &&
1806 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL))
1813 * Function to process delayed allocation (create CoW) for ranges which are
1814 * being touched for the first time.
1816 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1817 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1818 struct writeback_control *wbc)
1821 int force_cow = need_force_cow(inode, start, end);
1823 if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1824 ret = run_delalloc_nocow(inode, locked_page, start, end,
1825 page_started, 1, nr_written);
1826 } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1827 ret = run_delalloc_nocow(inode, locked_page, start, end,
1828 page_started, 0, nr_written);
1829 } else if (!inode_can_compress(inode) ||
1830 !inode_need_compress(inode, start, end)) {
1831 ret = cow_file_range(inode, locked_page, start, end,
1832 page_started, nr_written, 1);
1834 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1835 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1836 page_started, nr_written);
1839 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1844 void btrfs_split_delalloc_extent(struct inode *inode,
1845 struct extent_state *orig, u64 split)
1849 /* not delalloc, ignore it */
1850 if (!(orig->state & EXTENT_DELALLOC))
1853 size = orig->end - orig->start + 1;
1854 if (size > BTRFS_MAX_EXTENT_SIZE) {
1859 * See the explanation in btrfs_merge_delalloc_extent, the same
1860 * applies here, just in reverse.
1862 new_size = orig->end - split + 1;
1863 num_extents = count_max_extents(new_size);
1864 new_size = split - orig->start;
1865 num_extents += count_max_extents(new_size);
1866 if (count_max_extents(size) >= num_extents)
1870 spin_lock(&BTRFS_I(inode)->lock);
1871 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1872 spin_unlock(&BTRFS_I(inode)->lock);
1876 * Handle merged delayed allocation extents so we can keep track of new extents
1877 * that are just merged onto old extents, such as when we are doing sequential
1878 * writes, so we can properly account for the metadata space we'll need.
1880 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1881 struct extent_state *other)
1883 u64 new_size, old_size;
1886 /* not delalloc, ignore it */
1887 if (!(other->state & EXTENT_DELALLOC))
1890 if (new->start > other->start)
1891 new_size = new->end - other->start + 1;
1893 new_size = other->end - new->start + 1;
1895 /* we're not bigger than the max, unreserve the space and go */
1896 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1897 spin_lock(&BTRFS_I(inode)->lock);
1898 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1899 spin_unlock(&BTRFS_I(inode)->lock);
1904 * We have to add up either side to figure out how many extents were
1905 * accounted for before we merged into one big extent. If the number of
1906 * extents we accounted for is <= the amount we need for the new range
1907 * then we can return, otherwise drop. Think of it like this
1911 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1912 * need 2 outstanding extents, on one side we have 1 and the other side
1913 * we have 1 so they are == and we can return. But in this case
1915 * [MAX_SIZE+4k][MAX_SIZE+4k]
1917 * Each range on their own accounts for 2 extents, but merged together
1918 * they are only 3 extents worth of accounting, so we need to drop in
1921 old_size = other->end - other->start + 1;
1922 num_extents = count_max_extents(old_size);
1923 old_size = new->end - new->start + 1;
1924 num_extents += count_max_extents(old_size);
1925 if (count_max_extents(new_size) >= num_extents)
1928 spin_lock(&BTRFS_I(inode)->lock);
1929 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1930 spin_unlock(&BTRFS_I(inode)->lock);
1933 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1934 struct inode *inode)
1936 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1938 spin_lock(&root->delalloc_lock);
1939 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1940 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1941 &root->delalloc_inodes);
1942 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1943 &BTRFS_I(inode)->runtime_flags);
1944 root->nr_delalloc_inodes++;
1945 if (root->nr_delalloc_inodes == 1) {
1946 spin_lock(&fs_info->delalloc_root_lock);
1947 BUG_ON(!list_empty(&root->delalloc_root));
1948 list_add_tail(&root->delalloc_root,
1949 &fs_info->delalloc_roots);
1950 spin_unlock(&fs_info->delalloc_root_lock);
1953 spin_unlock(&root->delalloc_lock);
1957 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1958 struct btrfs_inode *inode)
1960 struct btrfs_fs_info *fs_info = root->fs_info;
1962 if (!list_empty(&inode->delalloc_inodes)) {
1963 list_del_init(&inode->delalloc_inodes);
1964 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1965 &inode->runtime_flags);
1966 root->nr_delalloc_inodes--;
1967 if (!root->nr_delalloc_inodes) {
1968 ASSERT(list_empty(&root->delalloc_inodes));
1969 spin_lock(&fs_info->delalloc_root_lock);
1970 BUG_ON(list_empty(&root->delalloc_root));
1971 list_del_init(&root->delalloc_root);
1972 spin_unlock(&fs_info->delalloc_root_lock);
1977 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1978 struct btrfs_inode *inode)
1980 spin_lock(&root->delalloc_lock);
1981 __btrfs_del_delalloc_inode(root, inode);
1982 spin_unlock(&root->delalloc_lock);
1986 * Properly track delayed allocation bytes in the inode and to maintain the
1987 * list of inodes that have pending delalloc work to be done.
1989 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
1992 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1994 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1997 * set_bit and clear bit hooks normally require _irqsave/restore
1998 * but in this case, we are only testing for the DELALLOC
1999 * bit, which is only set or cleared with irqs on
2001 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2002 struct btrfs_root *root = BTRFS_I(inode)->root;
2003 u64 len = state->end + 1 - state->start;
2004 u32 num_extents = count_max_extents(len);
2005 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2007 spin_lock(&BTRFS_I(inode)->lock);
2008 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2009 spin_unlock(&BTRFS_I(inode)->lock);
2011 /* For sanity tests */
2012 if (btrfs_is_testing(fs_info))
2015 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2016 fs_info->delalloc_batch);
2017 spin_lock(&BTRFS_I(inode)->lock);
2018 BTRFS_I(inode)->delalloc_bytes += len;
2019 if (*bits & EXTENT_DEFRAG)
2020 BTRFS_I(inode)->defrag_bytes += len;
2021 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2022 &BTRFS_I(inode)->runtime_flags))
2023 btrfs_add_delalloc_inodes(root, inode);
2024 spin_unlock(&BTRFS_I(inode)->lock);
2027 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2028 (*bits & EXTENT_DELALLOC_NEW)) {
2029 spin_lock(&BTRFS_I(inode)->lock);
2030 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2032 spin_unlock(&BTRFS_I(inode)->lock);
2037 * Once a range is no longer delalloc this function ensures that proper
2038 * accounting happens.
2040 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2041 struct extent_state *state, unsigned *bits)
2043 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2044 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2045 u64 len = state->end + 1 - state->start;
2046 u32 num_extents = count_max_extents(len);
2048 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2049 spin_lock(&inode->lock);
2050 inode->defrag_bytes -= len;
2051 spin_unlock(&inode->lock);
2055 * set_bit and clear bit hooks normally require _irqsave/restore
2056 * but in this case, we are only testing for the DELALLOC
2057 * bit, which is only set or cleared with irqs on
2059 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2060 struct btrfs_root *root = inode->root;
2061 bool do_list = !btrfs_is_free_space_inode(inode);
2063 spin_lock(&inode->lock);
2064 btrfs_mod_outstanding_extents(inode, -num_extents);
2065 spin_unlock(&inode->lock);
2068 * We don't reserve metadata space for space cache inodes so we
2069 * don't need to call delalloc_release_metadata if there is an
2072 if (*bits & EXTENT_CLEAR_META_RESV &&
2073 root != fs_info->tree_root)
2074 btrfs_delalloc_release_metadata(inode, len, false);
2076 /* For sanity tests. */
2077 if (btrfs_is_testing(fs_info))
2080 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2081 do_list && !(state->state & EXTENT_NORESERVE) &&
2082 (*bits & EXTENT_CLEAR_DATA_RESV))
2083 btrfs_free_reserved_data_space_noquota(fs_info, len);
2085 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2086 fs_info->delalloc_batch);
2087 spin_lock(&inode->lock);
2088 inode->delalloc_bytes -= len;
2089 if (do_list && inode->delalloc_bytes == 0 &&
2090 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2091 &inode->runtime_flags))
2092 btrfs_del_delalloc_inode(root, inode);
2093 spin_unlock(&inode->lock);
2096 if ((state->state & EXTENT_DELALLOC_NEW) &&
2097 (*bits & EXTENT_DELALLOC_NEW)) {
2098 spin_lock(&inode->lock);
2099 ASSERT(inode->new_delalloc_bytes >= len);
2100 inode->new_delalloc_bytes -= len;
2101 spin_unlock(&inode->lock);
2106 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2107 * in a chunk's stripe. This function ensures that bios do not span a
2110 * @page - The page we are about to add to the bio
2111 * @size - size we want to add to the bio
2112 * @bio - bio we want to ensure is smaller than a stripe
2113 * @bio_flags - flags of the bio
2115 * return 1 if page cannot be added to the bio
2116 * return 0 if page can be added to the bio
2117 * return error otherwise
2119 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2120 unsigned long bio_flags)
2122 struct inode *inode = page->mapping->host;
2123 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2124 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
2128 struct btrfs_io_geometry geom;
2130 if (bio_flags & EXTENT_BIO_COMPRESSED)
2133 length = bio->bi_iter.bi_size;
2134 map_length = length;
2135 ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length,
2140 if (geom.len < length + size)
2146 * in order to insert checksums into the metadata in large chunks,
2147 * we wait until bio submission time. All the pages in the bio are
2148 * checksummed and sums are attached onto the ordered extent record.
2150 * At IO completion time the cums attached on the ordered extent record
2151 * are inserted into the btree
2153 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
2156 struct inode *inode = private_data;
2158 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2162 * extent_io.c submission hook. This does the right thing for csum calculation
2163 * on write, or reading the csums from the tree before a read.
2165 * Rules about async/sync submit,
2166 * a) read: sync submit
2168 * b) write without checksum: sync submit
2170 * c) write with checksum:
2171 * c-1) if bio is issued by fsync: sync submit
2172 * (sync_writers != 0)
2174 * c-2) if root is reloc root: sync submit
2175 * (only in case of buffered IO)
2177 * c-3) otherwise: async submit
2179 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2180 int mirror_num, unsigned long bio_flags)
2183 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2184 struct btrfs_root *root = BTRFS_I(inode)->root;
2185 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2186 blk_status_t ret = 0;
2188 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2190 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2192 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2193 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2195 if (bio_op(bio) != REQ_OP_WRITE) {
2196 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2200 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2201 ret = btrfs_submit_compressed_read(inode, bio,
2205 } else if (!skip_sum) {
2206 ret = btrfs_lookup_bio_sums(inode, bio, (u64)-1, NULL);
2211 } else if (async && !skip_sum) {
2212 /* csum items have already been cloned */
2213 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2215 /* we're doing a write, do the async checksumming */
2216 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2217 0, inode, btrfs_submit_bio_start);
2219 } else if (!skip_sum) {
2220 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2226 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2230 bio->bi_status = ret;
2237 * given a list of ordered sums record them in the inode. This happens
2238 * at IO completion time based on sums calculated at bio submission time.
2240 static int add_pending_csums(struct btrfs_trans_handle *trans,
2241 struct list_head *list)
2243 struct btrfs_ordered_sum *sum;
2246 list_for_each_entry(sum, list, list) {
2247 trans->adding_csums = true;
2248 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2249 trans->adding_csums = false;
2256 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2259 struct extent_state **cached_state)
2261 u64 search_start = start;
2262 const u64 end = start + len - 1;
2264 while (search_start < end) {
2265 const u64 search_len = end - search_start + 1;
2266 struct extent_map *em;
2270 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2274 if (em->block_start != EXTENT_MAP_HOLE)
2278 if (em->start < search_start)
2279 em_len -= search_start - em->start;
2280 if (em_len > search_len)
2281 em_len = search_len;
2283 ret = set_extent_bit(&inode->io_tree, search_start,
2284 search_start + em_len - 1,
2285 EXTENT_DELALLOC_NEW,
2286 NULL, cached_state, GFP_NOFS);
2288 search_start = extent_map_end(em);
2289 free_extent_map(em);
2296 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2297 unsigned int extra_bits,
2298 struct extent_state **cached_state)
2300 WARN_ON(PAGE_ALIGNED(end));
2302 if (start >= i_size_read(&inode->vfs_inode) &&
2303 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2305 * There can't be any extents following eof in this case so just
2306 * set the delalloc new bit for the range directly.
2308 extra_bits |= EXTENT_DELALLOC_NEW;
2312 ret = btrfs_find_new_delalloc_bytes(inode, start,
2319 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2323 /* see btrfs_writepage_start_hook for details on why this is required */
2324 struct btrfs_writepage_fixup {
2326 struct inode *inode;
2327 struct btrfs_work work;
2330 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2332 struct btrfs_writepage_fixup *fixup;
2333 struct btrfs_ordered_extent *ordered;
2334 struct extent_state *cached_state = NULL;
2335 struct extent_changeset *data_reserved = NULL;
2337 struct btrfs_inode *inode;
2341 bool free_delalloc_space = true;
2343 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2345 inode = BTRFS_I(fixup->inode);
2346 page_start = page_offset(page);
2347 page_end = page_offset(page) + PAGE_SIZE - 1;
2350 * This is similar to page_mkwrite, we need to reserve the space before
2351 * we take the page lock.
2353 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2359 * Before we queued this fixup, we took a reference on the page.
2360 * page->mapping may go NULL, but it shouldn't be moved to a different
2363 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2365 * Unfortunately this is a little tricky, either
2367 * 1) We got here and our page had already been dealt with and
2368 * we reserved our space, thus ret == 0, so we need to just
2369 * drop our space reservation and bail. This can happen the
2370 * first time we come into the fixup worker, or could happen
2371 * while waiting for the ordered extent.
2372 * 2) Our page was already dealt with, but we happened to get an
2373 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2374 * this case we obviously don't have anything to release, but
2375 * because the page was already dealt with we don't want to
2376 * mark the page with an error, so make sure we're resetting
2377 * ret to 0. This is why we have this check _before_ the ret
2378 * check, because we do not want to have a surprise ENOSPC
2379 * when the page was already properly dealt with.
2382 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2383 btrfs_delalloc_release_space(inode, data_reserved,
2384 page_start, PAGE_SIZE,
2392 * We can't mess with the page state unless it is locked, so now that
2393 * it is locked bail if we failed to make our space reservation.
2398 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2400 /* already ordered? We're done */
2401 if (PagePrivate2(page))
2404 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2406 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2409 btrfs_start_ordered_extent(ordered, 1);
2410 btrfs_put_ordered_extent(ordered);
2414 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2420 * Everything went as planned, we're now the owner of a dirty page with
2421 * delayed allocation bits set and space reserved for our COW
2424 * The page was dirty when we started, nothing should have cleaned it.
2426 BUG_ON(!PageDirty(page));
2427 free_delalloc_space = false;
2429 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2430 if (free_delalloc_space)
2431 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2433 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2438 * We hit ENOSPC or other errors. Update the mapping and page
2439 * to reflect the errors and clean the page.
2441 mapping_set_error(page->mapping, ret);
2442 end_extent_writepage(page, ret, page_start, page_end);
2443 clear_page_dirty_for_io(page);
2446 ClearPageChecked(page);
2450 extent_changeset_free(data_reserved);
2452 * As a precaution, do a delayed iput in case it would be the last iput
2453 * that could need flushing space. Recursing back to fixup worker would
2456 btrfs_add_delayed_iput(&inode->vfs_inode);
2460 * There are a few paths in the higher layers of the kernel that directly
2461 * set the page dirty bit without asking the filesystem if it is a
2462 * good idea. This causes problems because we want to make sure COW
2463 * properly happens and the data=ordered rules are followed.
2465 * In our case any range that doesn't have the ORDERED bit set
2466 * hasn't been properly setup for IO. We kick off an async process
2467 * to fix it up. The async helper will wait for ordered extents, set
2468 * the delalloc bit and make it safe to write the page.
2470 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2472 struct inode *inode = page->mapping->host;
2473 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2474 struct btrfs_writepage_fixup *fixup;
2476 /* this page is properly in the ordered list */
2477 if (TestClearPagePrivate2(page))
2481 * PageChecked is set below when we create a fixup worker for this page,
2482 * don't try to create another one if we're already PageChecked()
2484 * The extent_io writepage code will redirty the page if we send back
2487 if (PageChecked(page))
2490 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2495 * We are already holding a reference to this inode from
2496 * write_cache_pages. We need to hold it because the space reservation
2497 * takes place outside of the page lock, and we can't trust
2498 * page->mapping outside of the page lock.
2501 SetPageChecked(page);
2503 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2505 fixup->inode = inode;
2506 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2511 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2512 struct btrfs_inode *inode, u64 file_pos,
2513 struct btrfs_file_extent_item *stack_fi,
2514 u64 qgroup_reserved)
2516 struct btrfs_root *root = inode->root;
2517 struct btrfs_path *path;
2518 struct extent_buffer *leaf;
2519 struct btrfs_key ins;
2520 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2521 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2522 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2523 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2524 int extent_inserted = 0;
2527 path = btrfs_alloc_path();
2532 * we may be replacing one extent in the tree with another.
2533 * The new extent is pinned in the extent map, and we don't want
2534 * to drop it from the cache until it is completely in the btree.
2536 * So, tell btrfs_drop_extents to leave this extent in the cache.
2537 * the caller is expected to unpin it and allow it to be merged
2540 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2541 file_pos + num_bytes, NULL, 0,
2542 1, sizeof(*stack_fi), &extent_inserted);
2546 if (!extent_inserted) {
2547 ins.objectid = btrfs_ino(inode);
2548 ins.offset = file_pos;
2549 ins.type = BTRFS_EXTENT_DATA_KEY;
2551 path->leave_spinning = 1;
2552 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2557 leaf = path->nodes[0];
2558 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2559 write_extent_buffer(leaf, stack_fi,
2560 btrfs_item_ptr_offset(leaf, path->slots[0]),
2561 sizeof(struct btrfs_file_extent_item));
2563 btrfs_mark_buffer_dirty(leaf);
2564 btrfs_release_path(path);
2566 inode_add_bytes(&inode->vfs_inode, num_bytes);
2568 ins.objectid = disk_bytenr;
2569 ins.offset = disk_num_bytes;
2570 ins.type = BTRFS_EXTENT_ITEM_KEY;
2572 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2576 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2577 file_pos, qgroup_reserved, &ins);
2579 btrfs_free_path(path);
2584 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2587 struct btrfs_block_group *cache;
2589 cache = btrfs_lookup_block_group(fs_info, start);
2592 spin_lock(&cache->lock);
2593 cache->delalloc_bytes -= len;
2594 spin_unlock(&cache->lock);
2596 btrfs_put_block_group(cache);
2599 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2600 struct btrfs_ordered_extent *oe)
2602 struct btrfs_file_extent_item stack_fi;
2605 memset(&stack_fi, 0, sizeof(stack_fi));
2606 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2607 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2608 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2609 oe->disk_num_bytes);
2610 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2611 logical_len = oe->truncated_len;
2613 logical_len = oe->num_bytes;
2614 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2615 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2616 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2617 /* Encryption and other encoding is reserved and all 0 */
2619 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2620 oe->file_offset, &stack_fi,
2625 * As ordered data IO finishes, this gets called so we can finish
2626 * an ordered extent if the range of bytes in the file it covers are
2629 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2631 struct inode *inode = ordered_extent->inode;
2632 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2633 struct btrfs_root *root = BTRFS_I(inode)->root;
2634 struct btrfs_trans_handle *trans = NULL;
2635 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2636 struct extent_state *cached_state = NULL;
2638 int compress_type = 0;
2640 u64 logical_len = ordered_extent->num_bytes;
2641 bool freespace_inode;
2642 bool truncated = false;
2643 bool range_locked = false;
2644 bool clear_new_delalloc_bytes = false;
2645 bool clear_reserved_extent = true;
2646 unsigned int clear_bits;
2648 start = ordered_extent->file_offset;
2649 end = start + ordered_extent->num_bytes - 1;
2651 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2652 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2653 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2654 clear_new_delalloc_bytes = true;
2656 freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
2658 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2663 btrfs_free_io_failure_record(BTRFS_I(inode), start, end);
2665 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2667 logical_len = ordered_extent->truncated_len;
2668 /* Truncated the entire extent, don't bother adding */
2673 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2674 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2676 btrfs_inode_safe_disk_i_size_write(inode, 0);
2677 if (freespace_inode)
2678 trans = btrfs_join_transaction_spacecache(root);
2680 trans = btrfs_join_transaction(root);
2681 if (IS_ERR(trans)) {
2682 ret = PTR_ERR(trans);
2686 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2687 ret = btrfs_update_inode_fallback(trans, root, inode);
2688 if (ret) /* -ENOMEM or corruption */
2689 btrfs_abort_transaction(trans, ret);
2693 range_locked = true;
2694 lock_extent_bits(io_tree, start, end, &cached_state);
2696 if (freespace_inode)
2697 trans = btrfs_join_transaction_spacecache(root);
2699 trans = btrfs_join_transaction(root);
2700 if (IS_ERR(trans)) {
2701 ret = PTR_ERR(trans);
2706 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2708 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2709 compress_type = ordered_extent->compress_type;
2710 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2711 BUG_ON(compress_type);
2712 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2713 ordered_extent->file_offset,
2714 ordered_extent->file_offset +
2717 BUG_ON(root == fs_info->tree_root);
2718 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2720 clear_reserved_extent = false;
2721 btrfs_release_delalloc_bytes(fs_info,
2722 ordered_extent->disk_bytenr,
2723 ordered_extent->disk_num_bytes);
2726 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2727 ordered_extent->file_offset,
2728 ordered_extent->num_bytes, trans->transid);
2730 btrfs_abort_transaction(trans, ret);
2734 ret = add_pending_csums(trans, &ordered_extent->list);
2736 btrfs_abort_transaction(trans, ret);
2740 btrfs_inode_safe_disk_i_size_write(inode, 0);
2741 ret = btrfs_update_inode_fallback(trans, root, inode);
2742 if (ret) { /* -ENOMEM or corruption */
2743 btrfs_abort_transaction(trans, ret);
2748 clear_bits = EXTENT_DEFRAG;
2750 clear_bits |= EXTENT_LOCKED;
2751 if (clear_new_delalloc_bytes)
2752 clear_bits |= EXTENT_DELALLOC_NEW;
2753 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits,
2754 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2758 btrfs_end_transaction(trans);
2760 if (ret || truncated) {
2761 u64 unwritten_start = start;
2764 unwritten_start += logical_len;
2765 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2767 /* Drop the cache for the part of the extent we didn't write. */
2768 btrfs_drop_extent_cache(BTRFS_I(inode), unwritten_start, end, 0);
2771 * If the ordered extent had an IOERR or something else went
2772 * wrong we need to return the space for this ordered extent
2773 * back to the allocator. We only free the extent in the
2774 * truncated case if we didn't write out the extent at all.
2776 * If we made it past insert_reserved_file_extent before we
2777 * errored out then we don't need to do this as the accounting
2778 * has already been done.
2780 if ((ret || !logical_len) &&
2781 clear_reserved_extent &&
2782 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2783 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2785 * Discard the range before returning it back to the
2788 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2789 btrfs_discard_extent(fs_info,
2790 ordered_extent->disk_bytenr,
2791 ordered_extent->disk_num_bytes,
2793 btrfs_free_reserved_extent(fs_info,
2794 ordered_extent->disk_bytenr,
2795 ordered_extent->disk_num_bytes, 1);
2800 * This needs to be done to make sure anybody waiting knows we are done
2801 * updating everything for this ordered extent.
2803 btrfs_remove_ordered_extent(BTRFS_I(inode), ordered_extent);
2806 btrfs_put_ordered_extent(ordered_extent);
2807 /* once for the tree */
2808 btrfs_put_ordered_extent(ordered_extent);
2813 static void finish_ordered_fn(struct btrfs_work *work)
2815 struct btrfs_ordered_extent *ordered_extent;
2816 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2817 btrfs_finish_ordered_io(ordered_extent);
2820 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2821 u64 end, int uptodate)
2823 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2824 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2825 struct btrfs_ordered_extent *ordered_extent = NULL;
2826 struct btrfs_workqueue *wq;
2828 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2830 ClearPagePrivate2(page);
2831 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2832 end - start + 1, uptodate))
2835 if (btrfs_is_free_space_inode(inode))
2836 wq = fs_info->endio_freespace_worker;
2838 wq = fs_info->endio_write_workers;
2840 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2841 btrfs_queue_work(wq, &ordered_extent->work);
2844 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
2845 int icsum, struct page *page, int pgoff, u64 start,
2848 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2849 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2851 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2853 u8 csum[BTRFS_CSUM_SIZE];
2855 csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size;
2857 kaddr = kmap_atomic(page);
2858 shash->tfm = fs_info->csum_shash;
2860 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
2862 if (memcmp(csum, csum_expected, csum_size))
2865 kunmap_atomic(kaddr);
2868 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
2869 io_bio->mirror_num);
2871 btrfs_dev_stat_inc_and_print(io_bio->device,
2872 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2873 memset(kaddr + pgoff, 1, len);
2874 flush_dcache_page(page);
2875 kunmap_atomic(kaddr);
2880 * when reads are done, we need to check csums to verify the data is correct
2881 * if there's a match, we allow the bio to finish. If not, the code in
2882 * extent_io.c will try to find good copies for us.
2884 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u64 phy_offset,
2885 struct page *page, u64 start, u64 end, int mirror)
2887 size_t offset = start - page_offset(page);
2888 struct inode *inode = page->mapping->host;
2889 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2890 struct btrfs_root *root = BTRFS_I(inode)->root;
2892 if (PageChecked(page)) {
2893 ClearPageChecked(page);
2897 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2900 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2901 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2902 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
2906 phy_offset >>= inode->i_sb->s_blocksize_bits;
2907 return check_data_csum(inode, io_bio, phy_offset, page, offset, start,
2908 (size_t)(end - start + 1));
2912 * btrfs_add_delayed_iput - perform a delayed iput on @inode
2914 * @inode: The inode we want to perform iput on
2916 * This function uses the generic vfs_inode::i_count to track whether we should
2917 * just decrement it (in case it's > 1) or if this is the last iput then link
2918 * the inode to the delayed iput machinery. Delayed iputs are processed at
2919 * transaction commit time/superblock commit/cleaner kthread.
2921 void btrfs_add_delayed_iput(struct inode *inode)
2923 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2924 struct btrfs_inode *binode = BTRFS_I(inode);
2926 if (atomic_add_unless(&inode->i_count, -1, 1))
2929 atomic_inc(&fs_info->nr_delayed_iputs);
2930 spin_lock(&fs_info->delayed_iput_lock);
2931 ASSERT(list_empty(&binode->delayed_iput));
2932 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
2933 spin_unlock(&fs_info->delayed_iput_lock);
2934 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
2935 wake_up_process(fs_info->cleaner_kthread);
2938 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
2939 struct btrfs_inode *inode)
2941 list_del_init(&inode->delayed_iput);
2942 spin_unlock(&fs_info->delayed_iput_lock);
2943 iput(&inode->vfs_inode);
2944 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
2945 wake_up(&fs_info->delayed_iputs_wait);
2946 spin_lock(&fs_info->delayed_iput_lock);
2949 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
2950 struct btrfs_inode *inode)
2952 if (!list_empty(&inode->delayed_iput)) {
2953 spin_lock(&fs_info->delayed_iput_lock);
2954 if (!list_empty(&inode->delayed_iput))
2955 run_delayed_iput_locked(fs_info, inode);
2956 spin_unlock(&fs_info->delayed_iput_lock);
2960 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
2963 spin_lock(&fs_info->delayed_iput_lock);
2964 while (!list_empty(&fs_info->delayed_iputs)) {
2965 struct btrfs_inode *inode;
2967 inode = list_first_entry(&fs_info->delayed_iputs,
2968 struct btrfs_inode, delayed_iput);
2969 run_delayed_iput_locked(fs_info, inode);
2971 spin_unlock(&fs_info->delayed_iput_lock);
2975 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
2976 * @fs_info - the fs_info for this fs
2977 * @return - EINTR if we were killed, 0 if nothing's pending
2979 * This will wait on any delayed iputs that are currently running with KILLABLE
2980 * set. Once they are all done running we will return, unless we are killed in
2981 * which case we return EINTR. This helps in user operations like fallocate etc
2982 * that might get blocked on the iputs.
2984 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
2986 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
2987 atomic_read(&fs_info->nr_delayed_iputs) == 0);
2994 * This creates an orphan entry for the given inode in case something goes wrong
2995 * in the middle of an unlink.
2997 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
2998 struct btrfs_inode *inode)
3002 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3003 if (ret && ret != -EEXIST) {
3004 btrfs_abort_transaction(trans, ret);
3012 * We have done the delete so we can go ahead and remove the orphan item for
3013 * this particular inode.
3015 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3016 struct btrfs_inode *inode)
3018 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3022 * this cleans up any orphans that may be left on the list from the last use
3025 int btrfs_orphan_cleanup(struct btrfs_root *root)
3027 struct btrfs_fs_info *fs_info = root->fs_info;
3028 struct btrfs_path *path;
3029 struct extent_buffer *leaf;
3030 struct btrfs_key key, found_key;
3031 struct btrfs_trans_handle *trans;
3032 struct inode *inode;
3033 u64 last_objectid = 0;
3034 int ret = 0, nr_unlink = 0;
3036 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3039 path = btrfs_alloc_path();
3044 path->reada = READA_BACK;
3046 key.objectid = BTRFS_ORPHAN_OBJECTID;
3047 key.type = BTRFS_ORPHAN_ITEM_KEY;
3048 key.offset = (u64)-1;
3051 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3056 * if ret == 0 means we found what we were searching for, which
3057 * is weird, but possible, so only screw with path if we didn't
3058 * find the key and see if we have stuff that matches
3062 if (path->slots[0] == 0)
3067 /* pull out the item */
3068 leaf = path->nodes[0];
3069 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3071 /* make sure the item matches what we want */
3072 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3074 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3077 /* release the path since we're done with it */
3078 btrfs_release_path(path);
3081 * this is where we are basically btrfs_lookup, without the
3082 * crossing root thing. we store the inode number in the
3083 * offset of the orphan item.
3086 if (found_key.offset == last_objectid) {
3088 "Error removing orphan entry, stopping orphan cleanup");
3093 last_objectid = found_key.offset;
3095 found_key.objectid = found_key.offset;
3096 found_key.type = BTRFS_INODE_ITEM_KEY;
3097 found_key.offset = 0;
3098 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3099 ret = PTR_ERR_OR_ZERO(inode);
3100 if (ret && ret != -ENOENT)
3103 if (ret == -ENOENT && root == fs_info->tree_root) {
3104 struct btrfs_root *dead_root;
3105 int is_dead_root = 0;
3108 * this is an orphan in the tree root. Currently these
3109 * could come from 2 sources:
3110 * a) a snapshot deletion in progress
3111 * b) a free space cache inode
3112 * We need to distinguish those two, as the snapshot
3113 * orphan must not get deleted.
3114 * find_dead_roots already ran before us, so if this
3115 * is a snapshot deletion, we should find the root
3116 * in the fs_roots radix tree.
3119 spin_lock(&fs_info->fs_roots_radix_lock);
3120 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3121 (unsigned long)found_key.objectid);
3122 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3124 spin_unlock(&fs_info->fs_roots_radix_lock);
3127 /* prevent this orphan from being found again */
3128 key.offset = found_key.objectid - 1;
3135 * If we have an inode with links, there are a couple of
3136 * possibilities. Old kernels (before v3.12) used to create an
3137 * orphan item for truncate indicating that there were possibly
3138 * extent items past i_size that needed to be deleted. In v3.12,
3139 * truncate was changed to update i_size in sync with the extent
3140 * items, but the (useless) orphan item was still created. Since
3141 * v4.18, we don't create the orphan item for truncate at all.
3143 * So, this item could mean that we need to do a truncate, but
3144 * only if this filesystem was last used on a pre-v3.12 kernel
3145 * and was not cleanly unmounted. The odds of that are quite
3146 * slim, and it's a pain to do the truncate now, so just delete
3149 * It's also possible that this orphan item was supposed to be
3150 * deleted but wasn't. The inode number may have been reused,
3151 * but either way, we can delete the orphan item.
3153 if (ret == -ENOENT || inode->i_nlink) {
3156 trans = btrfs_start_transaction(root, 1);
3157 if (IS_ERR(trans)) {
3158 ret = PTR_ERR(trans);
3161 btrfs_debug(fs_info, "auto deleting %Lu",
3162 found_key.objectid);
3163 ret = btrfs_del_orphan_item(trans, root,
3164 found_key.objectid);
3165 btrfs_end_transaction(trans);
3173 /* this will do delete_inode and everything for us */
3176 /* release the path since we're done with it */
3177 btrfs_release_path(path);
3179 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3181 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3182 trans = btrfs_join_transaction(root);
3184 btrfs_end_transaction(trans);
3188 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3192 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3193 btrfs_free_path(path);
3198 * very simple check to peek ahead in the leaf looking for xattrs. If we
3199 * don't find any xattrs, we know there can't be any acls.
3201 * slot is the slot the inode is in, objectid is the objectid of the inode
3203 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3204 int slot, u64 objectid,
3205 int *first_xattr_slot)
3207 u32 nritems = btrfs_header_nritems(leaf);
3208 struct btrfs_key found_key;
3209 static u64 xattr_access = 0;
3210 static u64 xattr_default = 0;
3213 if (!xattr_access) {
3214 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3215 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3216 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3217 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3221 *first_xattr_slot = -1;
3222 while (slot < nritems) {
3223 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3225 /* we found a different objectid, there must not be acls */
3226 if (found_key.objectid != objectid)
3229 /* we found an xattr, assume we've got an acl */
3230 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3231 if (*first_xattr_slot == -1)
3232 *first_xattr_slot = slot;
3233 if (found_key.offset == xattr_access ||
3234 found_key.offset == xattr_default)
3239 * we found a key greater than an xattr key, there can't
3240 * be any acls later on
3242 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3249 * it goes inode, inode backrefs, xattrs, extents,
3250 * so if there are a ton of hard links to an inode there can
3251 * be a lot of backrefs. Don't waste time searching too hard,
3252 * this is just an optimization
3257 /* we hit the end of the leaf before we found an xattr or
3258 * something larger than an xattr. We have to assume the inode
3261 if (*first_xattr_slot == -1)
3262 *first_xattr_slot = slot;
3267 * read an inode from the btree into the in-memory inode
3269 static int btrfs_read_locked_inode(struct inode *inode,
3270 struct btrfs_path *in_path)
3272 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3273 struct btrfs_path *path = in_path;
3274 struct extent_buffer *leaf;
3275 struct btrfs_inode_item *inode_item;
3276 struct btrfs_root *root = BTRFS_I(inode)->root;
3277 struct btrfs_key location;
3282 bool filled = false;
3283 int first_xattr_slot;
3285 ret = btrfs_fill_inode(inode, &rdev);
3290 path = btrfs_alloc_path();
3295 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3297 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3299 if (path != in_path)
3300 btrfs_free_path(path);
3304 leaf = path->nodes[0];
3309 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3310 struct btrfs_inode_item);
3311 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3312 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3313 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3314 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3315 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3316 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3317 round_up(i_size_read(inode), fs_info->sectorsize));
3319 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3320 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3322 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3323 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3325 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3326 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3328 BTRFS_I(inode)->i_otime.tv_sec =
3329 btrfs_timespec_sec(leaf, &inode_item->otime);
3330 BTRFS_I(inode)->i_otime.tv_nsec =
3331 btrfs_timespec_nsec(leaf, &inode_item->otime);
3333 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3334 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3335 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3337 inode_set_iversion_queried(inode,
3338 btrfs_inode_sequence(leaf, inode_item));
3339 inode->i_generation = BTRFS_I(inode)->generation;
3341 rdev = btrfs_inode_rdev(leaf, inode_item);
3343 BTRFS_I(inode)->index_cnt = (u64)-1;
3344 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3348 * If we were modified in the current generation and evicted from memory
3349 * and then re-read we need to do a full sync since we don't have any
3350 * idea about which extents were modified before we were evicted from
3353 * This is required for both inode re-read from disk and delayed inode
3354 * in delayed_nodes_tree.
3356 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3357 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3358 &BTRFS_I(inode)->runtime_flags);
3361 * We don't persist the id of the transaction where an unlink operation
3362 * against the inode was last made. So here we assume the inode might
3363 * have been evicted, and therefore the exact value of last_unlink_trans
3364 * lost, and set it to last_trans to avoid metadata inconsistencies
3365 * between the inode and its parent if the inode is fsync'ed and the log
3366 * replayed. For example, in the scenario:
3369 * ln mydir/foo mydir/bar
3372 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3373 * xfs_io -c fsync mydir/foo
3375 * mount fs, triggers fsync log replay
3377 * We must make sure that when we fsync our inode foo we also log its
3378 * parent inode, otherwise after log replay the parent still has the
3379 * dentry with the "bar" name but our inode foo has a link count of 1
3380 * and doesn't have an inode ref with the name "bar" anymore.
3382 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3383 * but it guarantees correctness at the expense of occasional full
3384 * transaction commits on fsync if our inode is a directory, or if our
3385 * inode is not a directory, logging its parent unnecessarily.
3387 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3390 * Same logic as for last_unlink_trans. We don't persist the generation
3391 * of the last transaction where this inode was used for a reflink
3392 * operation, so after eviction and reloading the inode we must be
3393 * pessimistic and assume the last transaction that modified the inode.
3395 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3398 if (inode->i_nlink != 1 ||
3399 path->slots[0] >= btrfs_header_nritems(leaf))
3402 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3403 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3406 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3407 if (location.type == BTRFS_INODE_REF_KEY) {
3408 struct btrfs_inode_ref *ref;
3410 ref = (struct btrfs_inode_ref *)ptr;
3411 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3412 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3413 struct btrfs_inode_extref *extref;
3415 extref = (struct btrfs_inode_extref *)ptr;
3416 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3421 * try to precache a NULL acl entry for files that don't have
3422 * any xattrs or acls
3424 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3425 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3426 if (first_xattr_slot != -1) {
3427 path->slots[0] = first_xattr_slot;
3428 ret = btrfs_load_inode_props(inode, path);
3431 "error loading props for ino %llu (root %llu): %d",
3432 btrfs_ino(BTRFS_I(inode)),
3433 root->root_key.objectid, ret);
3435 if (path != in_path)
3436 btrfs_free_path(path);
3439 cache_no_acl(inode);
3441 switch (inode->i_mode & S_IFMT) {
3443 inode->i_mapping->a_ops = &btrfs_aops;
3444 inode->i_fop = &btrfs_file_operations;
3445 inode->i_op = &btrfs_file_inode_operations;
3448 inode->i_fop = &btrfs_dir_file_operations;
3449 inode->i_op = &btrfs_dir_inode_operations;
3452 inode->i_op = &btrfs_symlink_inode_operations;
3453 inode_nohighmem(inode);
3454 inode->i_mapping->a_ops = &btrfs_aops;
3457 inode->i_op = &btrfs_special_inode_operations;
3458 init_special_inode(inode, inode->i_mode, rdev);
3462 btrfs_sync_inode_flags_to_i_flags(inode);
3467 * given a leaf and an inode, copy the inode fields into the leaf
3469 static void fill_inode_item(struct btrfs_trans_handle *trans,
3470 struct extent_buffer *leaf,
3471 struct btrfs_inode_item *item,
3472 struct inode *inode)
3474 struct btrfs_map_token token;
3476 btrfs_init_map_token(&token, leaf);
3478 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3479 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3480 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3481 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3482 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3484 btrfs_set_token_timespec_sec(&token, &item->atime,
3485 inode->i_atime.tv_sec);
3486 btrfs_set_token_timespec_nsec(&token, &item->atime,
3487 inode->i_atime.tv_nsec);
3489 btrfs_set_token_timespec_sec(&token, &item->mtime,
3490 inode->i_mtime.tv_sec);
3491 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3492 inode->i_mtime.tv_nsec);
3494 btrfs_set_token_timespec_sec(&token, &item->ctime,
3495 inode->i_ctime.tv_sec);
3496 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3497 inode->i_ctime.tv_nsec);
3499 btrfs_set_token_timespec_sec(&token, &item->otime,
3500 BTRFS_I(inode)->i_otime.tv_sec);
3501 btrfs_set_token_timespec_nsec(&token, &item->otime,
3502 BTRFS_I(inode)->i_otime.tv_nsec);
3504 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3505 btrfs_set_token_inode_generation(&token, item,
3506 BTRFS_I(inode)->generation);
3507 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3508 btrfs_set_token_inode_transid(&token, item, trans->transid);
3509 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3510 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3511 btrfs_set_token_inode_block_group(&token, item, 0);
3515 * copy everything in the in-memory inode into the btree.
3517 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3518 struct btrfs_root *root, struct inode *inode)
3520 struct btrfs_inode_item *inode_item;
3521 struct btrfs_path *path;
3522 struct extent_buffer *leaf;
3525 path = btrfs_alloc_path();
3529 path->leave_spinning = 1;
3530 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3538 leaf = path->nodes[0];
3539 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3540 struct btrfs_inode_item);
3542 fill_inode_item(trans, leaf, inode_item, inode);
3543 btrfs_mark_buffer_dirty(leaf);
3544 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3547 btrfs_free_path(path);
3552 * copy everything in the in-memory inode into the btree.
3554 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3555 struct btrfs_root *root, struct inode *inode)
3557 struct btrfs_fs_info *fs_info = root->fs_info;
3561 * If the inode is a free space inode, we can deadlock during commit
3562 * if we put it into the delayed code.
3564 * The data relocation inode should also be directly updated
3567 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3568 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3569 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3570 btrfs_update_root_times(trans, root);
3572 ret = btrfs_delayed_update_inode(trans, root, inode);
3574 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3578 return btrfs_update_inode_item(trans, root, inode);
3581 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3582 struct btrfs_root *root,
3583 struct inode *inode)
3587 ret = btrfs_update_inode(trans, root, inode);
3589 return btrfs_update_inode_item(trans, root, inode);
3594 * unlink helper that gets used here in inode.c and in the tree logging
3595 * recovery code. It remove a link in a directory with a given name, and
3596 * also drops the back refs in the inode to the directory
3598 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3599 struct btrfs_root *root,
3600 struct btrfs_inode *dir,
3601 struct btrfs_inode *inode,
3602 const char *name, int name_len)
3604 struct btrfs_fs_info *fs_info = root->fs_info;
3605 struct btrfs_path *path;
3607 struct btrfs_dir_item *di;
3609 u64 ino = btrfs_ino(inode);
3610 u64 dir_ino = btrfs_ino(dir);
3612 path = btrfs_alloc_path();
3618 path->leave_spinning = 1;
3619 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3620 name, name_len, -1);
3621 if (IS_ERR_OR_NULL(di)) {
3622 ret = di ? PTR_ERR(di) : -ENOENT;
3625 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3628 btrfs_release_path(path);
3631 * If we don't have dir index, we have to get it by looking up
3632 * the inode ref, since we get the inode ref, remove it directly,
3633 * it is unnecessary to do delayed deletion.
3635 * But if we have dir index, needn't search inode ref to get it.
3636 * Since the inode ref is close to the inode item, it is better
3637 * that we delay to delete it, and just do this deletion when
3638 * we update the inode item.
3640 if (inode->dir_index) {
3641 ret = btrfs_delayed_delete_inode_ref(inode);
3643 index = inode->dir_index;
3648 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3652 "failed to delete reference to %.*s, inode %llu parent %llu",
3653 name_len, name, ino, dir_ino);
3654 btrfs_abort_transaction(trans, ret);
3658 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3660 btrfs_abort_transaction(trans, ret);
3664 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3666 if (ret != 0 && ret != -ENOENT) {
3667 btrfs_abort_transaction(trans, ret);
3671 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3676 btrfs_abort_transaction(trans, ret);
3679 * If we have a pending delayed iput we could end up with the final iput
3680 * being run in btrfs-cleaner context. If we have enough of these built
3681 * up we can end up burning a lot of time in btrfs-cleaner without any
3682 * way to throttle the unlinks. Since we're currently holding a ref on
3683 * the inode we can run the delayed iput here without any issues as the
3684 * final iput won't be done until after we drop the ref we're currently
3687 btrfs_run_delayed_iput(fs_info, inode);
3689 btrfs_free_path(path);
3693 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3694 inode_inc_iversion(&inode->vfs_inode);
3695 inode_inc_iversion(&dir->vfs_inode);
3696 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3697 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3698 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3703 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3704 struct btrfs_root *root,
3705 struct btrfs_inode *dir, struct btrfs_inode *inode,
3706 const char *name, int name_len)
3709 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3711 drop_nlink(&inode->vfs_inode);
3712 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3718 * helper to start transaction for unlink and rmdir.
3720 * unlink and rmdir are special in btrfs, they do not always free space, so
3721 * if we cannot make our reservations the normal way try and see if there is
3722 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3723 * allow the unlink to occur.
3725 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3727 struct btrfs_root *root = BTRFS_I(dir)->root;
3730 * 1 for the possible orphan item
3731 * 1 for the dir item
3732 * 1 for the dir index
3733 * 1 for the inode ref
3736 return btrfs_start_transaction_fallback_global_rsv(root, 5);
3739 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3741 struct btrfs_root *root = BTRFS_I(dir)->root;
3742 struct btrfs_trans_handle *trans;
3743 struct inode *inode = d_inode(dentry);
3746 trans = __unlink_start_trans(dir);
3748 return PTR_ERR(trans);
3750 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
3753 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3754 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3755 dentry->d_name.len);
3759 if (inode->i_nlink == 0) {
3760 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3766 btrfs_end_transaction(trans);
3767 btrfs_btree_balance_dirty(root->fs_info);
3771 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3772 struct inode *dir, struct dentry *dentry)
3774 struct btrfs_root *root = BTRFS_I(dir)->root;
3775 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
3776 struct btrfs_path *path;
3777 struct extent_buffer *leaf;
3778 struct btrfs_dir_item *di;
3779 struct btrfs_key key;
3780 const char *name = dentry->d_name.name;
3781 int name_len = dentry->d_name.len;
3785 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3787 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
3788 objectid = inode->root->root_key.objectid;
3789 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3790 objectid = inode->location.objectid;
3796 path = btrfs_alloc_path();
3800 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3801 name, name_len, -1);
3802 if (IS_ERR_OR_NULL(di)) {
3803 ret = di ? PTR_ERR(di) : -ENOENT;
3807 leaf = path->nodes[0];
3808 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3809 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3810 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3812 btrfs_abort_transaction(trans, ret);
3815 btrfs_release_path(path);
3818 * This is a placeholder inode for a subvolume we didn't have a
3819 * reference to at the time of the snapshot creation. In the meantime
3820 * we could have renamed the real subvol link into our snapshot, so
3821 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
3822 * Instead simply lookup the dir_index_item for this entry so we can
3823 * remove it. Otherwise we know we have a ref to the root and we can
3824 * call btrfs_del_root_ref, and it _shouldn't_ fail.
3826 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3827 di = btrfs_search_dir_index_item(root, path, dir_ino,
3829 if (IS_ERR_OR_NULL(di)) {
3834 btrfs_abort_transaction(trans, ret);
3838 leaf = path->nodes[0];
3839 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3841 btrfs_release_path(path);
3843 ret = btrfs_del_root_ref(trans, objectid,
3844 root->root_key.objectid, dir_ino,
3845 &index, name, name_len);
3847 btrfs_abort_transaction(trans, ret);
3852 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
3854 btrfs_abort_transaction(trans, ret);
3858 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
3859 inode_inc_iversion(dir);
3860 dir->i_mtime = dir->i_ctime = current_time(dir);
3861 ret = btrfs_update_inode_fallback(trans, root, dir);
3863 btrfs_abort_transaction(trans, ret);
3865 btrfs_free_path(path);
3870 * Helper to check if the subvolume references other subvolumes or if it's
3873 static noinline int may_destroy_subvol(struct btrfs_root *root)
3875 struct btrfs_fs_info *fs_info = root->fs_info;
3876 struct btrfs_path *path;
3877 struct btrfs_dir_item *di;
3878 struct btrfs_key key;
3882 path = btrfs_alloc_path();
3886 /* Make sure this root isn't set as the default subvol */
3887 dir_id = btrfs_super_root_dir(fs_info->super_copy);
3888 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
3889 dir_id, "default", 7, 0);
3890 if (di && !IS_ERR(di)) {
3891 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
3892 if (key.objectid == root->root_key.objectid) {
3895 "deleting default subvolume %llu is not allowed",
3899 btrfs_release_path(path);
3902 key.objectid = root->root_key.objectid;
3903 key.type = BTRFS_ROOT_REF_KEY;
3904 key.offset = (u64)-1;
3906 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3912 if (path->slots[0] > 0) {
3914 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3915 if (key.objectid == root->root_key.objectid &&
3916 key.type == BTRFS_ROOT_REF_KEY)
3920 btrfs_free_path(path);
3924 /* Delete all dentries for inodes belonging to the root */
3925 static void btrfs_prune_dentries(struct btrfs_root *root)
3927 struct btrfs_fs_info *fs_info = root->fs_info;
3928 struct rb_node *node;
3929 struct rb_node *prev;
3930 struct btrfs_inode *entry;
3931 struct inode *inode;
3934 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3935 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3937 spin_lock(&root->inode_lock);
3939 node = root->inode_tree.rb_node;
3943 entry = rb_entry(node, struct btrfs_inode, rb_node);
3945 if (objectid < btrfs_ino(entry))
3946 node = node->rb_left;
3947 else if (objectid > btrfs_ino(entry))
3948 node = node->rb_right;
3954 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3955 if (objectid <= btrfs_ino(entry)) {
3959 prev = rb_next(prev);
3963 entry = rb_entry(node, struct btrfs_inode, rb_node);
3964 objectid = btrfs_ino(entry) + 1;
3965 inode = igrab(&entry->vfs_inode);
3967 spin_unlock(&root->inode_lock);
3968 if (atomic_read(&inode->i_count) > 1)
3969 d_prune_aliases(inode);
3971 * btrfs_drop_inode will have it removed from the inode
3972 * cache when its usage count hits zero.
3976 spin_lock(&root->inode_lock);
3980 if (cond_resched_lock(&root->inode_lock))
3983 node = rb_next(node);
3985 spin_unlock(&root->inode_lock);
3988 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
3990 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
3991 struct btrfs_root *root = BTRFS_I(dir)->root;
3992 struct inode *inode = d_inode(dentry);
3993 struct btrfs_root *dest = BTRFS_I(inode)->root;
3994 struct btrfs_trans_handle *trans;
3995 struct btrfs_block_rsv block_rsv;
4001 * Don't allow to delete a subvolume with send in progress. This is
4002 * inside the inode lock so the error handling that has to drop the bit
4003 * again is not run concurrently.
4005 spin_lock(&dest->root_item_lock);
4006 if (dest->send_in_progress) {
4007 spin_unlock(&dest->root_item_lock);
4009 "attempt to delete subvolume %llu during send",
4010 dest->root_key.objectid);
4013 root_flags = btrfs_root_flags(&dest->root_item);
4014 btrfs_set_root_flags(&dest->root_item,
4015 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4016 spin_unlock(&dest->root_item_lock);
4018 down_write(&fs_info->subvol_sem);
4020 err = may_destroy_subvol(dest);
4024 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4026 * One for dir inode,
4027 * two for dir entries,
4028 * two for root ref/backref.
4030 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4034 trans = btrfs_start_transaction(root, 0);
4035 if (IS_ERR(trans)) {
4036 err = PTR_ERR(trans);
4039 trans->block_rsv = &block_rsv;
4040 trans->bytes_reserved = block_rsv.size;
4042 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4044 ret = btrfs_unlink_subvol(trans, dir, dentry);
4047 btrfs_abort_transaction(trans, ret);
4051 btrfs_record_root_in_trans(trans, dest);
4053 memset(&dest->root_item.drop_progress, 0,
4054 sizeof(dest->root_item.drop_progress));
4055 dest->root_item.drop_level = 0;
4056 btrfs_set_root_refs(&dest->root_item, 0);
4058 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4059 ret = btrfs_insert_orphan_item(trans,
4061 dest->root_key.objectid);
4063 btrfs_abort_transaction(trans, ret);
4069 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4070 BTRFS_UUID_KEY_SUBVOL,
4071 dest->root_key.objectid);
4072 if (ret && ret != -ENOENT) {
4073 btrfs_abort_transaction(trans, ret);
4077 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4078 ret = btrfs_uuid_tree_remove(trans,
4079 dest->root_item.received_uuid,
4080 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4081 dest->root_key.objectid);
4082 if (ret && ret != -ENOENT) {
4083 btrfs_abort_transaction(trans, ret);
4089 free_anon_bdev(dest->anon_dev);
4092 trans->block_rsv = NULL;
4093 trans->bytes_reserved = 0;
4094 ret = btrfs_end_transaction(trans);
4097 inode->i_flags |= S_DEAD;
4099 btrfs_subvolume_release_metadata(root, &block_rsv);
4101 up_write(&fs_info->subvol_sem);
4103 spin_lock(&dest->root_item_lock);
4104 root_flags = btrfs_root_flags(&dest->root_item);
4105 btrfs_set_root_flags(&dest->root_item,
4106 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4107 spin_unlock(&dest->root_item_lock);
4109 d_invalidate(dentry);
4110 btrfs_prune_dentries(dest);
4111 ASSERT(dest->send_in_progress == 0);
4114 if (dest->ino_cache_inode) {
4115 iput(dest->ino_cache_inode);
4116 dest->ino_cache_inode = NULL;
4123 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4125 struct inode *inode = d_inode(dentry);
4127 struct btrfs_root *root = BTRFS_I(dir)->root;
4128 struct btrfs_trans_handle *trans;
4129 u64 last_unlink_trans;
4131 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4133 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4134 return btrfs_delete_subvolume(dir, dentry);
4136 trans = __unlink_start_trans(dir);
4138 return PTR_ERR(trans);
4140 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4141 err = btrfs_unlink_subvol(trans, dir, dentry);
4145 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4149 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4151 /* now the directory is empty */
4152 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4153 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4154 dentry->d_name.len);
4156 btrfs_i_size_write(BTRFS_I(inode), 0);
4158 * Propagate the last_unlink_trans value of the deleted dir to
4159 * its parent directory. This is to prevent an unrecoverable
4160 * log tree in the case we do something like this:
4162 * 2) create snapshot under dir foo
4163 * 3) delete the snapshot
4166 * 6) fsync foo or some file inside foo
4168 if (last_unlink_trans >= trans->transid)
4169 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4172 btrfs_end_transaction(trans);
4173 btrfs_btree_balance_dirty(root->fs_info);
4179 * Return this if we need to call truncate_block for the last bit of the
4182 #define NEED_TRUNCATE_BLOCK 1
4185 * this can truncate away extent items, csum items and directory items.
4186 * It starts at a high offset and removes keys until it can't find
4187 * any higher than new_size
4189 * csum items that cross the new i_size are truncated to the new size
4192 * min_type is the minimum key type to truncate down to. If set to 0, this
4193 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4195 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4196 struct btrfs_root *root,
4197 struct inode *inode,
4198 u64 new_size, u32 min_type)
4200 struct btrfs_fs_info *fs_info = root->fs_info;
4201 struct btrfs_path *path;
4202 struct extent_buffer *leaf;
4203 struct btrfs_file_extent_item *fi;
4204 struct btrfs_key key;
4205 struct btrfs_key found_key;
4206 u64 extent_start = 0;
4207 u64 extent_num_bytes = 0;
4208 u64 extent_offset = 0;
4210 u64 last_size = new_size;
4211 u32 found_type = (u8)-1;
4214 int pending_del_nr = 0;
4215 int pending_del_slot = 0;
4216 int extent_type = -1;
4218 u64 ino = btrfs_ino(BTRFS_I(inode));
4219 u64 bytes_deleted = 0;
4220 bool be_nice = false;
4221 bool should_throttle = false;
4222 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4223 struct extent_state *cached_state = NULL;
4225 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4228 * For non-free space inodes and non-shareable roots, we want to back
4229 * off from time to time. This means all inodes in subvolume roots,
4230 * reloc roots, and data reloc roots.
4232 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4233 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4236 path = btrfs_alloc_path();
4239 path->reada = READA_BACK;
4241 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4242 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
4246 * We want to drop from the next block forward in case this
4247 * new size is not block aligned since we will be keeping the
4248 * last block of the extent just the way it is.
4250 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4251 fs_info->sectorsize),
4256 * This function is also used to drop the items in the log tree before
4257 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4258 * it is used to drop the logged items. So we shouldn't kill the delayed
4261 if (min_type == 0 && root == BTRFS_I(inode)->root)
4262 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4265 key.offset = (u64)-1;
4270 * with a 16K leaf size and 128MB extents, you can actually queue
4271 * up a huge file in a single leaf. Most of the time that
4272 * bytes_deleted is > 0, it will be huge by the time we get here
4274 if (be_nice && bytes_deleted > SZ_32M &&
4275 btrfs_should_end_transaction(trans)) {
4280 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4286 /* there are no items in the tree for us to truncate, we're
4289 if (path->slots[0] == 0)
4295 u64 clear_start = 0, clear_len = 0;
4298 leaf = path->nodes[0];
4299 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4300 found_type = found_key.type;
4302 if (found_key.objectid != ino)
4305 if (found_type < min_type)
4308 item_end = found_key.offset;
4309 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4310 fi = btrfs_item_ptr(leaf, path->slots[0],
4311 struct btrfs_file_extent_item);
4312 extent_type = btrfs_file_extent_type(leaf, fi);
4313 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4315 btrfs_file_extent_num_bytes(leaf, fi);
4317 trace_btrfs_truncate_show_fi_regular(
4318 BTRFS_I(inode), leaf, fi,
4320 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4321 item_end += btrfs_file_extent_ram_bytes(leaf,
4324 trace_btrfs_truncate_show_fi_inline(
4325 BTRFS_I(inode), leaf, fi, path->slots[0],
4330 if (found_type > min_type) {
4333 if (item_end < new_size)
4335 if (found_key.offset >= new_size)
4341 /* FIXME, shrink the extent if the ref count is only 1 */
4342 if (found_type != BTRFS_EXTENT_DATA_KEY)
4345 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4348 clear_start = found_key.offset;
4349 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4351 u64 orig_num_bytes =
4352 btrfs_file_extent_num_bytes(leaf, fi);
4353 extent_num_bytes = ALIGN(new_size -
4355 fs_info->sectorsize);
4356 clear_start = ALIGN(new_size, fs_info->sectorsize);
4357 btrfs_set_file_extent_num_bytes(leaf, fi,
4359 num_dec = (orig_num_bytes -
4361 if (test_bit(BTRFS_ROOT_SHAREABLE,
4364 inode_sub_bytes(inode, num_dec);
4365 btrfs_mark_buffer_dirty(leaf);
4368 btrfs_file_extent_disk_num_bytes(leaf,
4370 extent_offset = found_key.offset -
4371 btrfs_file_extent_offset(leaf, fi);
4373 /* FIXME blocksize != 4096 */
4374 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4375 if (extent_start != 0) {
4377 if (test_bit(BTRFS_ROOT_SHAREABLE,
4379 inode_sub_bytes(inode, num_dec);
4382 clear_len = num_dec;
4383 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4385 * we can't truncate inline items that have had
4389 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4390 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4391 btrfs_file_extent_compression(leaf, fi) == 0) {
4392 u32 size = (u32)(new_size - found_key.offset);
4394 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4395 size = btrfs_file_extent_calc_inline_size(size);
4396 btrfs_truncate_item(path, size, 1);
4397 } else if (!del_item) {
4399 * We have to bail so the last_size is set to
4400 * just before this extent.
4402 ret = NEED_TRUNCATE_BLOCK;
4406 * Inline extents are special, we just treat
4407 * them as a full sector worth in the file
4408 * extent tree just for simplicity sake.
4410 clear_len = fs_info->sectorsize;
4413 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4414 inode_sub_bytes(inode, item_end + 1 - new_size);
4418 * We use btrfs_truncate_inode_items() to clean up log trees for
4419 * multiple fsyncs, and in this case we don't want to clear the
4420 * file extent range because it's just the log.
4422 if (root == BTRFS_I(inode)->root) {
4423 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
4424 clear_start, clear_len);
4426 btrfs_abort_transaction(trans, ret);
4432 last_size = found_key.offset;
4434 last_size = new_size;
4436 if (!pending_del_nr) {
4437 /* no pending yet, add ourselves */
4438 pending_del_slot = path->slots[0];
4440 } else if (pending_del_nr &&
4441 path->slots[0] + 1 == pending_del_slot) {
4442 /* hop on the pending chunk */
4444 pending_del_slot = path->slots[0];
4451 should_throttle = false;
4454 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4455 struct btrfs_ref ref = { 0 };
4457 bytes_deleted += extent_num_bytes;
4459 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4460 extent_start, extent_num_bytes, 0);
4461 ref.real_root = root->root_key.objectid;
4462 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4463 ino, extent_offset);
4464 ret = btrfs_free_extent(trans, &ref);
4466 btrfs_abort_transaction(trans, ret);
4470 if (btrfs_should_throttle_delayed_refs(trans))
4471 should_throttle = true;
4475 if (found_type == BTRFS_INODE_ITEM_KEY)
4478 if (path->slots[0] == 0 ||
4479 path->slots[0] != pending_del_slot ||
4481 if (pending_del_nr) {
4482 ret = btrfs_del_items(trans, root, path,
4486 btrfs_abort_transaction(trans, ret);
4491 btrfs_release_path(path);
4494 * We can generate a lot of delayed refs, so we need to
4495 * throttle every once and a while and make sure we're
4496 * adding enough space to keep up with the work we are
4497 * generating. Since we hold a transaction here we
4498 * can't flush, and we don't want to FLUSH_LIMIT because
4499 * we could have generated too many delayed refs to
4500 * actually allocate, so just bail if we're short and
4501 * let the normal reservation dance happen higher up.
4503 if (should_throttle) {
4504 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4505 BTRFS_RESERVE_NO_FLUSH);
4517 if (ret >= 0 && pending_del_nr) {
4520 err = btrfs_del_items(trans, root, path, pending_del_slot,
4523 btrfs_abort_transaction(trans, err);
4527 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4528 ASSERT(last_size >= new_size);
4529 if (!ret && last_size > new_size)
4530 last_size = new_size;
4531 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4532 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
4533 (u64)-1, &cached_state);
4536 btrfs_free_path(path);
4541 * btrfs_truncate_block - read, zero a chunk and write a block
4542 * @inode - inode that we're zeroing
4543 * @from - the offset to start zeroing
4544 * @len - the length to zero, 0 to zero the entire range respective to the
4546 * @front - zero up to the offset instead of from the offset on
4548 * This will find the block for the "from" offset and cow the block and zero the
4549 * part we want to zero. This is used with truncate and hole punching.
4551 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4554 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4555 struct address_space *mapping = inode->i_mapping;
4556 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4557 struct btrfs_ordered_extent *ordered;
4558 struct extent_state *cached_state = NULL;
4559 struct extent_changeset *data_reserved = NULL;
4561 bool only_release_metadata = false;
4562 u32 blocksize = fs_info->sectorsize;
4563 pgoff_t index = from >> PAGE_SHIFT;
4564 unsigned offset = from & (blocksize - 1);
4566 gfp_t mask = btrfs_alloc_write_mask(mapping);
4567 size_t write_bytes = blocksize;
4572 if (IS_ALIGNED(offset, blocksize) &&
4573 (!len || IS_ALIGNED(len, blocksize)))
4576 block_start = round_down(from, blocksize);
4577 block_end = block_start + blocksize - 1;
4579 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved,
4580 block_start, blocksize);
4582 if (btrfs_check_nocow_lock(BTRFS_I(inode), block_start,
4583 &write_bytes) > 0) {
4584 /* For nocow case, no need to reserve data space */
4585 only_release_metadata = true;
4590 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), blocksize);
4592 if (!only_release_metadata)
4593 btrfs_free_reserved_data_space(BTRFS_I(inode),
4594 data_reserved, block_start, blocksize);
4598 page = find_or_create_page(mapping, index, mask);
4600 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4601 block_start, blocksize, true);
4602 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4607 if (!PageUptodate(page)) {
4608 ret = btrfs_readpage(NULL, page);
4610 if (page->mapping != mapping) {
4615 if (!PageUptodate(page)) {
4620 wait_on_page_writeback(page);
4622 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4623 set_page_extent_mapped(page);
4625 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), block_start);
4627 unlock_extent_cached(io_tree, block_start, block_end,
4631 btrfs_start_ordered_extent(ordered, 1);
4632 btrfs_put_ordered_extent(ordered);
4636 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4637 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4638 0, 0, &cached_state);
4640 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), block_start, block_end, 0,
4643 unlock_extent_cached(io_tree, block_start, block_end,
4648 if (offset != blocksize) {
4650 len = blocksize - offset;
4653 memset(kaddr + (block_start - page_offset(page)),
4656 memset(kaddr + (block_start - page_offset(page)) + offset,
4658 flush_dcache_page(page);
4661 ClearPageChecked(page);
4662 set_page_dirty(page);
4663 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4665 if (only_release_metadata)
4666 set_extent_bit(&BTRFS_I(inode)->io_tree, block_start,
4667 block_end, EXTENT_NORESERVE, NULL, NULL,
4672 if (only_release_metadata)
4673 btrfs_delalloc_release_metadata(BTRFS_I(inode),
4676 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4677 block_start, blocksize, true);
4679 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4683 if (only_release_metadata)
4684 btrfs_check_nocow_unlock(BTRFS_I(inode));
4685 extent_changeset_free(data_reserved);
4689 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4690 u64 offset, u64 len)
4692 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4693 struct btrfs_trans_handle *trans;
4697 * Still need to make sure the inode looks like it's been updated so
4698 * that any holes get logged if we fsync.
4700 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4701 BTRFS_I(inode)->last_trans = fs_info->generation;
4702 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4703 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4708 * 1 - for the one we're dropping
4709 * 1 - for the one we're adding
4710 * 1 - for updating the inode.
4712 trans = btrfs_start_transaction(root, 3);
4714 return PTR_ERR(trans);
4716 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4718 btrfs_abort_transaction(trans, ret);
4719 btrfs_end_transaction(trans);
4723 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4724 offset, 0, 0, len, 0, len, 0, 0, 0);
4726 btrfs_abort_transaction(trans, ret);
4728 btrfs_update_inode(trans, root, inode);
4729 btrfs_end_transaction(trans);
4734 * This function puts in dummy file extents for the area we're creating a hole
4735 * for. So if we are truncating this file to a larger size we need to insert
4736 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4737 * the range between oldsize and size
4739 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4741 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4742 struct btrfs_root *root = BTRFS_I(inode)->root;
4743 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4744 struct extent_map *em = NULL;
4745 struct extent_state *cached_state = NULL;
4746 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4747 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4748 u64 block_end = ALIGN(size, fs_info->sectorsize);
4755 * If our size started in the middle of a block we need to zero out the
4756 * rest of the block before we expand the i_size, otherwise we could
4757 * expose stale data.
4759 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4763 if (size <= hole_start)
4766 btrfs_lock_and_flush_ordered_range(BTRFS_I(inode), hole_start,
4767 block_end - 1, &cached_state);
4768 cur_offset = hole_start;
4770 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4771 block_end - cur_offset);
4777 last_byte = min(extent_map_end(em), block_end);
4778 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4779 hole_size = last_byte - cur_offset;
4781 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4782 struct extent_map *hole_em;
4784 err = maybe_insert_hole(root, inode, cur_offset,
4789 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4790 cur_offset, hole_size);
4794 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4795 cur_offset + hole_size - 1, 0);
4796 hole_em = alloc_extent_map();
4798 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4799 &BTRFS_I(inode)->runtime_flags);
4802 hole_em->start = cur_offset;
4803 hole_em->len = hole_size;
4804 hole_em->orig_start = cur_offset;
4806 hole_em->block_start = EXTENT_MAP_HOLE;
4807 hole_em->block_len = 0;
4808 hole_em->orig_block_len = 0;
4809 hole_em->ram_bytes = hole_size;
4810 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4811 hole_em->generation = fs_info->generation;
4814 write_lock(&em_tree->lock);
4815 err = add_extent_mapping(em_tree, hole_em, 1);
4816 write_unlock(&em_tree->lock);
4819 btrfs_drop_extent_cache(BTRFS_I(inode),
4824 free_extent_map(hole_em);
4826 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4827 cur_offset, hole_size);
4832 free_extent_map(em);
4834 cur_offset = last_byte;
4835 if (cur_offset >= block_end)
4838 free_extent_map(em);
4839 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4843 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4845 struct btrfs_root *root = BTRFS_I(inode)->root;
4846 struct btrfs_trans_handle *trans;
4847 loff_t oldsize = i_size_read(inode);
4848 loff_t newsize = attr->ia_size;
4849 int mask = attr->ia_valid;
4853 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4854 * special case where we need to update the times despite not having
4855 * these flags set. For all other operations the VFS set these flags
4856 * explicitly if it wants a timestamp update.
4858 if (newsize != oldsize) {
4859 inode_inc_iversion(inode);
4860 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4861 inode->i_ctime = inode->i_mtime =
4862 current_time(inode);
4865 if (newsize > oldsize) {
4867 * Don't do an expanding truncate while snapshotting is ongoing.
4868 * This is to ensure the snapshot captures a fully consistent
4869 * state of this file - if the snapshot captures this expanding
4870 * truncation, it must capture all writes that happened before
4873 btrfs_drew_write_lock(&root->snapshot_lock);
4874 ret = btrfs_cont_expand(inode, oldsize, newsize);
4876 btrfs_drew_write_unlock(&root->snapshot_lock);
4880 trans = btrfs_start_transaction(root, 1);
4881 if (IS_ERR(trans)) {
4882 btrfs_drew_write_unlock(&root->snapshot_lock);
4883 return PTR_ERR(trans);
4886 i_size_write(inode, newsize);
4887 btrfs_inode_safe_disk_i_size_write(inode, 0);
4888 pagecache_isize_extended(inode, oldsize, newsize);
4889 ret = btrfs_update_inode(trans, root, inode);
4890 btrfs_drew_write_unlock(&root->snapshot_lock);
4891 btrfs_end_transaction(trans);
4895 * We're truncating a file that used to have good data down to
4896 * zero. Make sure any new writes to the file get on disk
4900 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
4901 &BTRFS_I(inode)->runtime_flags);
4903 truncate_setsize(inode, newsize);
4905 inode_dio_wait(inode);
4907 ret = btrfs_truncate(inode, newsize == oldsize);
4908 if (ret && inode->i_nlink) {
4912 * Truncate failed, so fix up the in-memory size. We
4913 * adjusted disk_i_size down as we removed extents, so
4914 * wait for disk_i_size to be stable and then update the
4915 * in-memory size to match.
4917 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
4920 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4927 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4929 struct inode *inode = d_inode(dentry);
4930 struct btrfs_root *root = BTRFS_I(inode)->root;
4933 if (btrfs_root_readonly(root))
4936 err = setattr_prepare(dentry, attr);
4940 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4941 err = btrfs_setsize(inode, attr);
4946 if (attr->ia_valid) {
4947 setattr_copy(inode, attr);
4948 inode_inc_iversion(inode);
4949 err = btrfs_dirty_inode(inode);
4951 if (!err && attr->ia_valid & ATTR_MODE)
4952 err = posix_acl_chmod(inode, inode->i_mode);
4959 * While truncating the inode pages during eviction, we get the VFS calling
4960 * btrfs_invalidatepage() against each page of the inode. This is slow because
4961 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4962 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4963 * extent_state structures over and over, wasting lots of time.
4965 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4966 * those expensive operations on a per page basis and do only the ordered io
4967 * finishing, while we release here the extent_map and extent_state structures,
4968 * without the excessive merging and splitting.
4970 static void evict_inode_truncate_pages(struct inode *inode)
4972 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4973 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4974 struct rb_node *node;
4976 ASSERT(inode->i_state & I_FREEING);
4977 truncate_inode_pages_final(&inode->i_data);
4979 write_lock(&map_tree->lock);
4980 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
4981 struct extent_map *em;
4983 node = rb_first_cached(&map_tree->map);
4984 em = rb_entry(node, struct extent_map, rb_node);
4985 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4986 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4987 remove_extent_mapping(map_tree, em);
4988 free_extent_map(em);
4989 if (need_resched()) {
4990 write_unlock(&map_tree->lock);
4992 write_lock(&map_tree->lock);
4995 write_unlock(&map_tree->lock);
4998 * Keep looping until we have no more ranges in the io tree.
4999 * We can have ongoing bios started by readahead that have
5000 * their endio callback (extent_io.c:end_bio_extent_readpage)
5001 * still in progress (unlocked the pages in the bio but did not yet
5002 * unlocked the ranges in the io tree). Therefore this means some
5003 * ranges can still be locked and eviction started because before
5004 * submitting those bios, which are executed by a separate task (work
5005 * queue kthread), inode references (inode->i_count) were not taken
5006 * (which would be dropped in the end io callback of each bio).
5007 * Therefore here we effectively end up waiting for those bios and
5008 * anyone else holding locked ranges without having bumped the inode's
5009 * reference count - if we don't do it, when they access the inode's
5010 * io_tree to unlock a range it may be too late, leading to an
5011 * use-after-free issue.
5013 spin_lock(&io_tree->lock);
5014 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5015 struct extent_state *state;
5016 struct extent_state *cached_state = NULL;
5019 unsigned state_flags;
5021 node = rb_first(&io_tree->state);
5022 state = rb_entry(node, struct extent_state, rb_node);
5023 start = state->start;
5025 state_flags = state->state;
5026 spin_unlock(&io_tree->lock);
5028 lock_extent_bits(io_tree, start, end, &cached_state);
5031 * If still has DELALLOC flag, the extent didn't reach disk,
5032 * and its reserved space won't be freed by delayed_ref.
5033 * So we need to free its reserved space here.
5034 * (Refer to comment in btrfs_invalidatepage, case 2)
5036 * Note, end is the bytenr of last byte, so we need + 1 here.
5038 if (state_flags & EXTENT_DELALLOC)
5039 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5042 clear_extent_bit(io_tree, start, end,
5043 EXTENT_LOCKED | EXTENT_DELALLOC |
5044 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5048 spin_lock(&io_tree->lock);
5050 spin_unlock(&io_tree->lock);
5053 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5054 struct btrfs_block_rsv *rsv)
5056 struct btrfs_fs_info *fs_info = root->fs_info;
5057 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5058 struct btrfs_trans_handle *trans;
5059 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5063 * Eviction should be taking place at some place safe because of our
5064 * delayed iputs. However the normal flushing code will run delayed
5065 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5067 * We reserve the delayed_refs_extra here again because we can't use
5068 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5069 * above. We reserve our extra bit here because we generate a ton of
5070 * delayed refs activity by truncating.
5072 * If we cannot make our reservation we'll attempt to steal from the
5073 * global reserve, because we really want to be able to free up space.
5075 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5076 BTRFS_RESERVE_FLUSH_EVICT);
5079 * Try to steal from the global reserve if there is space for
5082 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5083 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5085 "could not allocate space for delete; will truncate on mount");
5086 return ERR_PTR(-ENOSPC);
5088 delayed_refs_extra = 0;
5091 trans = btrfs_join_transaction(root);
5095 if (delayed_refs_extra) {
5096 trans->block_rsv = &fs_info->trans_block_rsv;
5097 trans->bytes_reserved = delayed_refs_extra;
5098 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5099 delayed_refs_extra, 1);
5104 void btrfs_evict_inode(struct inode *inode)
5106 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5107 struct btrfs_trans_handle *trans;
5108 struct btrfs_root *root = BTRFS_I(inode)->root;
5109 struct btrfs_block_rsv *rsv;
5112 trace_btrfs_inode_evict(inode);
5119 evict_inode_truncate_pages(inode);
5121 if (inode->i_nlink &&
5122 ((btrfs_root_refs(&root->root_item) != 0 &&
5123 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5124 btrfs_is_free_space_inode(BTRFS_I(inode))))
5127 if (is_bad_inode(inode))
5130 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5132 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5135 if (inode->i_nlink > 0) {
5136 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5137 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5141 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5145 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5148 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5151 btrfs_i_size_write(BTRFS_I(inode), 0);
5154 trans = evict_refill_and_join(root, rsv);
5158 trans->block_rsv = rsv;
5160 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5161 trans->block_rsv = &fs_info->trans_block_rsv;
5162 btrfs_end_transaction(trans);
5163 btrfs_btree_balance_dirty(fs_info);
5164 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5171 * Errors here aren't a big deal, it just means we leave orphan items in
5172 * the tree. They will be cleaned up on the next mount. If the inode
5173 * number gets reused, cleanup deletes the orphan item without doing
5174 * anything, and unlink reuses the existing orphan item.
5176 * If it turns out that we are dropping too many of these, we might want
5177 * to add a mechanism for retrying these after a commit.
5179 trans = evict_refill_and_join(root, rsv);
5180 if (!IS_ERR(trans)) {
5181 trans->block_rsv = rsv;
5182 btrfs_orphan_del(trans, BTRFS_I(inode));
5183 trans->block_rsv = &fs_info->trans_block_rsv;
5184 btrfs_end_transaction(trans);
5187 if (!(root == fs_info->tree_root ||
5188 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5189 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5192 btrfs_free_block_rsv(fs_info, rsv);
5195 * If we didn't successfully delete, the orphan item will still be in
5196 * the tree and we'll retry on the next mount. Again, we might also want
5197 * to retry these periodically in the future.
5199 btrfs_remove_delayed_node(BTRFS_I(inode));
5204 * Return the key found in the dir entry in the location pointer, fill @type
5205 * with BTRFS_FT_*, and return 0.
5207 * If no dir entries were found, returns -ENOENT.
5208 * If found a corrupted location in dir entry, returns -EUCLEAN.
5210 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5211 struct btrfs_key *location, u8 *type)
5213 const char *name = dentry->d_name.name;
5214 int namelen = dentry->d_name.len;
5215 struct btrfs_dir_item *di;
5216 struct btrfs_path *path;
5217 struct btrfs_root *root = BTRFS_I(dir)->root;
5220 path = btrfs_alloc_path();
5224 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5226 if (IS_ERR_OR_NULL(di)) {
5227 ret = di ? PTR_ERR(di) : -ENOENT;
5231 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5232 if (location->type != BTRFS_INODE_ITEM_KEY &&
5233 location->type != BTRFS_ROOT_ITEM_KEY) {
5235 btrfs_warn(root->fs_info,
5236 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5237 __func__, name, btrfs_ino(BTRFS_I(dir)),
5238 location->objectid, location->type, location->offset);
5241 *type = btrfs_dir_type(path->nodes[0], di);
5243 btrfs_free_path(path);
5248 * when we hit a tree root in a directory, the btrfs part of the inode
5249 * needs to be changed to reflect the root directory of the tree root. This
5250 * is kind of like crossing a mount point.
5252 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5254 struct dentry *dentry,
5255 struct btrfs_key *location,
5256 struct btrfs_root **sub_root)
5258 struct btrfs_path *path;
5259 struct btrfs_root *new_root;
5260 struct btrfs_root_ref *ref;
5261 struct extent_buffer *leaf;
5262 struct btrfs_key key;
5266 path = btrfs_alloc_path();
5273 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5274 key.type = BTRFS_ROOT_REF_KEY;
5275 key.offset = location->objectid;
5277 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5284 leaf = path->nodes[0];
5285 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5286 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5287 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5290 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5291 (unsigned long)(ref + 1),
5292 dentry->d_name.len);
5296 btrfs_release_path(path);
5298 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5299 if (IS_ERR(new_root)) {
5300 err = PTR_ERR(new_root);
5304 *sub_root = new_root;
5305 location->objectid = btrfs_root_dirid(&new_root->root_item);
5306 location->type = BTRFS_INODE_ITEM_KEY;
5307 location->offset = 0;
5310 btrfs_free_path(path);
5314 static void inode_tree_add(struct inode *inode)
5316 struct btrfs_root *root = BTRFS_I(inode)->root;
5317 struct btrfs_inode *entry;
5319 struct rb_node *parent;
5320 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5321 u64 ino = btrfs_ino(BTRFS_I(inode));
5323 if (inode_unhashed(inode))
5326 spin_lock(&root->inode_lock);
5327 p = &root->inode_tree.rb_node;
5330 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5332 if (ino < btrfs_ino(entry))
5333 p = &parent->rb_left;
5334 else if (ino > btrfs_ino(entry))
5335 p = &parent->rb_right;
5337 WARN_ON(!(entry->vfs_inode.i_state &
5338 (I_WILL_FREE | I_FREEING)));
5339 rb_replace_node(parent, new, &root->inode_tree);
5340 RB_CLEAR_NODE(parent);
5341 spin_unlock(&root->inode_lock);
5345 rb_link_node(new, parent, p);
5346 rb_insert_color(new, &root->inode_tree);
5347 spin_unlock(&root->inode_lock);
5350 static void inode_tree_del(struct btrfs_inode *inode)
5352 struct btrfs_root *root = inode->root;
5355 spin_lock(&root->inode_lock);
5356 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5357 rb_erase(&inode->rb_node, &root->inode_tree);
5358 RB_CLEAR_NODE(&inode->rb_node);
5359 empty = RB_EMPTY_ROOT(&root->inode_tree);
5361 spin_unlock(&root->inode_lock);
5363 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5364 spin_lock(&root->inode_lock);
5365 empty = RB_EMPTY_ROOT(&root->inode_tree);
5366 spin_unlock(&root->inode_lock);
5368 btrfs_add_dead_root(root);
5373 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5375 struct btrfs_iget_args *args = p;
5377 inode->i_ino = args->ino;
5378 BTRFS_I(inode)->location.objectid = args->ino;
5379 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5380 BTRFS_I(inode)->location.offset = 0;
5381 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5382 BUG_ON(args->root && !BTRFS_I(inode)->root);
5386 static int btrfs_find_actor(struct inode *inode, void *opaque)
5388 struct btrfs_iget_args *args = opaque;
5390 return args->ino == BTRFS_I(inode)->location.objectid &&
5391 args->root == BTRFS_I(inode)->root;
5394 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5395 struct btrfs_root *root)
5397 struct inode *inode;
5398 struct btrfs_iget_args args;
5399 unsigned long hashval = btrfs_inode_hash(ino, root);
5404 inode = iget5_locked(s, hashval, btrfs_find_actor,
5405 btrfs_init_locked_inode,
5411 * Get an inode object given its inode number and corresponding root.
5412 * Path can be preallocated to prevent recursing back to iget through
5413 * allocator. NULL is also valid but may require an additional allocation
5416 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5417 struct btrfs_root *root, struct btrfs_path *path)
5419 struct inode *inode;
5421 inode = btrfs_iget_locked(s, ino, root);
5423 return ERR_PTR(-ENOMEM);
5425 if (inode->i_state & I_NEW) {
5428 ret = btrfs_read_locked_inode(inode, path);
5430 inode_tree_add(inode);
5431 unlock_new_inode(inode);
5435 * ret > 0 can come from btrfs_search_slot called by
5436 * btrfs_read_locked_inode, this means the inode item
5441 inode = ERR_PTR(ret);
5448 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5450 return btrfs_iget_path(s, ino, root, NULL);
5453 static struct inode *new_simple_dir(struct super_block *s,
5454 struct btrfs_key *key,
5455 struct btrfs_root *root)
5457 struct inode *inode = new_inode(s);
5460 return ERR_PTR(-ENOMEM);
5462 BTRFS_I(inode)->root = btrfs_grab_root(root);
5463 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5464 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5466 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5468 * We only need lookup, the rest is read-only and there's no inode
5469 * associated with the dentry
5471 inode->i_op = &simple_dir_inode_operations;
5472 inode->i_opflags &= ~IOP_XATTR;
5473 inode->i_fop = &simple_dir_operations;
5474 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5475 inode->i_mtime = current_time(inode);
5476 inode->i_atime = inode->i_mtime;
5477 inode->i_ctime = inode->i_mtime;
5478 BTRFS_I(inode)->i_otime = inode->i_mtime;
5483 static inline u8 btrfs_inode_type(struct inode *inode)
5486 * Compile-time asserts that generic FT_* types still match
5489 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5490 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5491 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5492 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5493 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5494 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5495 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5496 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5498 return fs_umode_to_ftype(inode->i_mode);
5501 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5503 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5504 struct inode *inode;
5505 struct btrfs_root *root = BTRFS_I(dir)->root;
5506 struct btrfs_root *sub_root = root;
5507 struct btrfs_key location;
5511 if (dentry->d_name.len > BTRFS_NAME_LEN)
5512 return ERR_PTR(-ENAMETOOLONG);
5514 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5516 return ERR_PTR(ret);
5518 if (location.type == BTRFS_INODE_ITEM_KEY) {
5519 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5523 /* Do extra check against inode mode with di_type */
5524 if (btrfs_inode_type(inode) != di_type) {
5526 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5527 inode->i_mode, btrfs_inode_type(inode),
5530 return ERR_PTR(-EUCLEAN);
5535 ret = fixup_tree_root_location(fs_info, dir, dentry,
5536 &location, &sub_root);
5539 inode = ERR_PTR(ret);
5541 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5543 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5545 if (root != sub_root)
5546 btrfs_put_root(sub_root);
5548 if (!IS_ERR(inode) && root != sub_root) {
5549 down_read(&fs_info->cleanup_work_sem);
5550 if (!sb_rdonly(inode->i_sb))
5551 ret = btrfs_orphan_cleanup(sub_root);
5552 up_read(&fs_info->cleanup_work_sem);
5555 inode = ERR_PTR(ret);
5562 static int btrfs_dentry_delete(const struct dentry *dentry)
5564 struct btrfs_root *root;
5565 struct inode *inode = d_inode(dentry);
5567 if (!inode && !IS_ROOT(dentry))
5568 inode = d_inode(dentry->d_parent);
5571 root = BTRFS_I(inode)->root;
5572 if (btrfs_root_refs(&root->root_item) == 0)
5575 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5581 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5584 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5586 if (inode == ERR_PTR(-ENOENT))
5588 return d_splice_alias(inode, dentry);
5592 * All this infrastructure exists because dir_emit can fault, and we are holding
5593 * the tree lock when doing readdir. For now just allocate a buffer and copy
5594 * our information into that, and then dir_emit from the buffer. This is
5595 * similar to what NFS does, only we don't keep the buffer around in pagecache
5596 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5597 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5600 static int btrfs_opendir(struct inode *inode, struct file *file)
5602 struct btrfs_file_private *private;
5604 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5607 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5608 if (!private->filldir_buf) {
5612 file->private_data = private;
5623 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5626 struct dir_entry *entry = addr;
5627 char *name = (char *)(entry + 1);
5629 ctx->pos = get_unaligned(&entry->offset);
5630 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5631 get_unaligned(&entry->ino),
5632 get_unaligned(&entry->type)))
5634 addr += sizeof(struct dir_entry) +
5635 get_unaligned(&entry->name_len);
5641 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5643 struct inode *inode = file_inode(file);
5644 struct btrfs_root *root = BTRFS_I(inode)->root;
5645 struct btrfs_file_private *private = file->private_data;
5646 struct btrfs_dir_item *di;
5647 struct btrfs_key key;
5648 struct btrfs_key found_key;
5649 struct btrfs_path *path;
5651 struct list_head ins_list;
5652 struct list_head del_list;
5654 struct extent_buffer *leaf;
5661 struct btrfs_key location;
5663 if (!dir_emit_dots(file, ctx))
5666 path = btrfs_alloc_path();
5670 addr = private->filldir_buf;
5671 path->reada = READA_FORWARD;
5673 INIT_LIST_HEAD(&ins_list);
5674 INIT_LIST_HEAD(&del_list);
5675 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5678 key.type = BTRFS_DIR_INDEX_KEY;
5679 key.offset = ctx->pos;
5680 key.objectid = btrfs_ino(BTRFS_I(inode));
5682 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5687 struct dir_entry *entry;
5689 leaf = path->nodes[0];
5690 slot = path->slots[0];
5691 if (slot >= btrfs_header_nritems(leaf)) {
5692 ret = btrfs_next_leaf(root, path);
5700 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5702 if (found_key.objectid != key.objectid)
5704 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5706 if (found_key.offset < ctx->pos)
5708 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5710 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5711 name_len = btrfs_dir_name_len(leaf, di);
5712 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5714 btrfs_release_path(path);
5715 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5718 addr = private->filldir_buf;
5725 put_unaligned(name_len, &entry->name_len);
5726 name_ptr = (char *)(entry + 1);
5727 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5729 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5731 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5732 put_unaligned(location.objectid, &entry->ino);
5733 put_unaligned(found_key.offset, &entry->offset);
5735 addr += sizeof(struct dir_entry) + name_len;
5736 total_len += sizeof(struct dir_entry) + name_len;
5740 btrfs_release_path(path);
5742 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5746 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5751 * Stop new entries from being returned after we return the last
5754 * New directory entries are assigned a strictly increasing
5755 * offset. This means that new entries created during readdir
5756 * are *guaranteed* to be seen in the future by that readdir.
5757 * This has broken buggy programs which operate on names as
5758 * they're returned by readdir. Until we re-use freed offsets
5759 * we have this hack to stop new entries from being returned
5760 * under the assumption that they'll never reach this huge
5763 * This is being careful not to overflow 32bit loff_t unless the
5764 * last entry requires it because doing so has broken 32bit apps
5767 if (ctx->pos >= INT_MAX)
5768 ctx->pos = LLONG_MAX;
5775 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5776 btrfs_free_path(path);
5781 * This is somewhat expensive, updating the tree every time the
5782 * inode changes. But, it is most likely to find the inode in cache.
5783 * FIXME, needs more benchmarking...there are no reasons other than performance
5784 * to keep or drop this code.
5786 static int btrfs_dirty_inode(struct inode *inode)
5788 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5789 struct btrfs_root *root = BTRFS_I(inode)->root;
5790 struct btrfs_trans_handle *trans;
5793 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5796 trans = btrfs_join_transaction(root);
5798 return PTR_ERR(trans);
5800 ret = btrfs_update_inode(trans, root, inode);
5801 if (ret && ret == -ENOSPC) {
5802 /* whoops, lets try again with the full transaction */
5803 btrfs_end_transaction(trans);
5804 trans = btrfs_start_transaction(root, 1);
5806 return PTR_ERR(trans);
5808 ret = btrfs_update_inode(trans, root, inode);
5810 btrfs_end_transaction(trans);
5811 if (BTRFS_I(inode)->delayed_node)
5812 btrfs_balance_delayed_items(fs_info);
5818 * This is a copy of file_update_time. We need this so we can return error on
5819 * ENOSPC for updating the inode in the case of file write and mmap writes.
5821 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5824 struct btrfs_root *root = BTRFS_I(inode)->root;
5825 bool dirty = flags & ~S_VERSION;
5827 if (btrfs_root_readonly(root))
5830 if (flags & S_VERSION)
5831 dirty |= inode_maybe_inc_iversion(inode, dirty);
5832 if (flags & S_CTIME)
5833 inode->i_ctime = *now;
5834 if (flags & S_MTIME)
5835 inode->i_mtime = *now;
5836 if (flags & S_ATIME)
5837 inode->i_atime = *now;
5838 return dirty ? btrfs_dirty_inode(inode) : 0;
5842 * find the highest existing sequence number in a directory
5843 * and then set the in-memory index_cnt variable to reflect
5844 * free sequence numbers
5846 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5848 struct btrfs_root *root = inode->root;
5849 struct btrfs_key key, found_key;
5850 struct btrfs_path *path;
5851 struct extent_buffer *leaf;
5854 key.objectid = btrfs_ino(inode);
5855 key.type = BTRFS_DIR_INDEX_KEY;
5856 key.offset = (u64)-1;
5858 path = btrfs_alloc_path();
5862 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5865 /* FIXME: we should be able to handle this */
5871 * MAGIC NUMBER EXPLANATION:
5872 * since we search a directory based on f_pos we have to start at 2
5873 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5874 * else has to start at 2
5876 if (path->slots[0] == 0) {
5877 inode->index_cnt = 2;
5883 leaf = path->nodes[0];
5884 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5886 if (found_key.objectid != btrfs_ino(inode) ||
5887 found_key.type != BTRFS_DIR_INDEX_KEY) {
5888 inode->index_cnt = 2;
5892 inode->index_cnt = found_key.offset + 1;
5894 btrfs_free_path(path);
5899 * helper to find a free sequence number in a given directory. This current
5900 * code is very simple, later versions will do smarter things in the btree
5902 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
5906 if (dir->index_cnt == (u64)-1) {
5907 ret = btrfs_inode_delayed_dir_index_count(dir);
5909 ret = btrfs_set_inode_index_count(dir);
5915 *index = dir->index_cnt;
5921 static int btrfs_insert_inode_locked(struct inode *inode)
5923 struct btrfs_iget_args args;
5925 args.ino = BTRFS_I(inode)->location.objectid;
5926 args.root = BTRFS_I(inode)->root;
5928 return insert_inode_locked4(inode,
5929 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5930 btrfs_find_actor, &args);
5934 * Inherit flags from the parent inode.
5936 * Currently only the compression flags and the cow flags are inherited.
5938 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
5945 flags = BTRFS_I(dir)->flags;
5947 if (flags & BTRFS_INODE_NOCOMPRESS) {
5948 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
5949 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
5950 } else if (flags & BTRFS_INODE_COMPRESS) {
5951 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
5952 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
5955 if (flags & BTRFS_INODE_NODATACOW) {
5956 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
5957 if (S_ISREG(inode->i_mode))
5958 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5961 btrfs_sync_inode_flags_to_i_flags(inode);
5964 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5965 struct btrfs_root *root,
5967 const char *name, int name_len,
5968 u64 ref_objectid, u64 objectid,
5969 umode_t mode, u64 *index)
5971 struct btrfs_fs_info *fs_info = root->fs_info;
5972 struct inode *inode;
5973 struct btrfs_inode_item *inode_item;
5974 struct btrfs_key *location;
5975 struct btrfs_path *path;
5976 struct btrfs_inode_ref *ref;
5977 struct btrfs_key key[2];
5979 int nitems = name ? 2 : 1;
5981 unsigned int nofs_flag;
5984 path = btrfs_alloc_path();
5986 return ERR_PTR(-ENOMEM);
5988 nofs_flag = memalloc_nofs_save();
5989 inode = new_inode(fs_info->sb);
5990 memalloc_nofs_restore(nofs_flag);
5992 btrfs_free_path(path);
5993 return ERR_PTR(-ENOMEM);
5997 * O_TMPFILE, set link count to 0, so that after this point,
5998 * we fill in an inode item with the correct link count.
6001 set_nlink(inode, 0);
6004 * we have to initialize this early, so we can reclaim the inode
6005 * number if we fail afterwards in this function.
6007 inode->i_ino = objectid;
6010 trace_btrfs_inode_request(dir);
6012 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6014 btrfs_free_path(path);
6016 return ERR_PTR(ret);
6022 * index_cnt is ignored for everything but a dir,
6023 * btrfs_set_inode_index_count has an explanation for the magic
6026 BTRFS_I(inode)->index_cnt = 2;
6027 BTRFS_I(inode)->dir_index = *index;
6028 BTRFS_I(inode)->root = btrfs_grab_root(root);
6029 BTRFS_I(inode)->generation = trans->transid;
6030 inode->i_generation = BTRFS_I(inode)->generation;
6033 * We could have gotten an inode number from somebody who was fsynced
6034 * and then removed in this same transaction, so let's just set full
6035 * sync since it will be a full sync anyway and this will blow away the
6036 * old info in the log.
6038 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6040 key[0].objectid = objectid;
6041 key[0].type = BTRFS_INODE_ITEM_KEY;
6044 sizes[0] = sizeof(struct btrfs_inode_item);
6048 * Start new inodes with an inode_ref. This is slightly more
6049 * efficient for small numbers of hard links since they will
6050 * be packed into one item. Extended refs will kick in if we
6051 * add more hard links than can fit in the ref item.
6053 key[1].objectid = objectid;
6054 key[1].type = BTRFS_INODE_REF_KEY;
6055 key[1].offset = ref_objectid;
6057 sizes[1] = name_len + sizeof(*ref);
6060 location = &BTRFS_I(inode)->location;
6061 location->objectid = objectid;
6062 location->offset = 0;
6063 location->type = BTRFS_INODE_ITEM_KEY;
6065 ret = btrfs_insert_inode_locked(inode);
6071 path->leave_spinning = 1;
6072 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6076 inode_init_owner(inode, dir, mode);
6077 inode_set_bytes(inode, 0);
6079 inode->i_mtime = current_time(inode);
6080 inode->i_atime = inode->i_mtime;
6081 inode->i_ctime = inode->i_mtime;
6082 BTRFS_I(inode)->i_otime = inode->i_mtime;
6084 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6085 struct btrfs_inode_item);
6086 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6087 sizeof(*inode_item));
6088 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6091 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6092 struct btrfs_inode_ref);
6093 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6094 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6095 ptr = (unsigned long)(ref + 1);
6096 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6099 btrfs_mark_buffer_dirty(path->nodes[0]);
6100 btrfs_free_path(path);
6102 btrfs_inherit_iflags(inode, dir);
6104 if (S_ISREG(mode)) {
6105 if (btrfs_test_opt(fs_info, NODATASUM))
6106 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6107 if (btrfs_test_opt(fs_info, NODATACOW))
6108 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6109 BTRFS_INODE_NODATASUM;
6112 inode_tree_add(inode);
6114 trace_btrfs_inode_new(inode);
6115 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6117 btrfs_update_root_times(trans, root);
6119 ret = btrfs_inode_inherit_props(trans, inode, dir);
6122 "error inheriting props for ino %llu (root %llu): %d",
6123 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6128 discard_new_inode(inode);
6131 BTRFS_I(dir)->index_cnt--;
6132 btrfs_free_path(path);
6133 return ERR_PTR(ret);
6137 * utility function to add 'inode' into 'parent_inode' with
6138 * a give name and a given sequence number.
6139 * if 'add_backref' is true, also insert a backref from the
6140 * inode to the parent directory.
6142 int btrfs_add_link(struct btrfs_trans_handle *trans,
6143 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6144 const char *name, int name_len, int add_backref, u64 index)
6147 struct btrfs_key key;
6148 struct btrfs_root *root = parent_inode->root;
6149 u64 ino = btrfs_ino(inode);
6150 u64 parent_ino = btrfs_ino(parent_inode);
6152 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6153 memcpy(&key, &inode->root->root_key, sizeof(key));
6156 key.type = BTRFS_INODE_ITEM_KEY;
6160 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6161 ret = btrfs_add_root_ref(trans, key.objectid,
6162 root->root_key.objectid, parent_ino,
6163 index, name, name_len);
6164 } else if (add_backref) {
6165 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6169 /* Nothing to clean up yet */
6173 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6174 btrfs_inode_type(&inode->vfs_inode), index);
6175 if (ret == -EEXIST || ret == -EOVERFLOW)
6178 btrfs_abort_transaction(trans, ret);
6182 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6184 inode_inc_iversion(&parent_inode->vfs_inode);
6186 * If we are replaying a log tree, we do not want to update the mtime
6187 * and ctime of the parent directory with the current time, since the
6188 * log replay procedure is responsible for setting them to their correct
6189 * values (the ones it had when the fsync was done).
6191 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6192 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6194 parent_inode->vfs_inode.i_mtime = now;
6195 parent_inode->vfs_inode.i_ctime = now;
6197 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6199 btrfs_abort_transaction(trans, ret);
6203 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6206 err = btrfs_del_root_ref(trans, key.objectid,
6207 root->root_key.objectid, parent_ino,
6208 &local_index, name, name_len);
6210 btrfs_abort_transaction(trans, err);
6211 } else if (add_backref) {
6215 err = btrfs_del_inode_ref(trans, root, name, name_len,
6216 ino, parent_ino, &local_index);
6218 btrfs_abort_transaction(trans, err);
6221 /* Return the original error code */
6225 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6226 struct btrfs_inode *dir, struct dentry *dentry,
6227 struct btrfs_inode *inode, int backref, u64 index)
6229 int err = btrfs_add_link(trans, dir, inode,
6230 dentry->d_name.name, dentry->d_name.len,
6237 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6238 umode_t mode, dev_t rdev)
6240 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6241 struct btrfs_trans_handle *trans;
6242 struct btrfs_root *root = BTRFS_I(dir)->root;
6243 struct inode *inode = NULL;
6249 * 2 for inode item and ref
6251 * 1 for xattr if selinux is on
6253 trans = btrfs_start_transaction(root, 5);
6255 return PTR_ERR(trans);
6257 err = btrfs_find_free_ino(root, &objectid);
6261 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6262 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6264 if (IS_ERR(inode)) {
6265 err = PTR_ERR(inode);
6271 * If the active LSM wants to access the inode during
6272 * d_instantiate it needs these. Smack checks to see
6273 * if the filesystem supports xattrs by looking at the
6276 inode->i_op = &btrfs_special_inode_operations;
6277 init_special_inode(inode, inode->i_mode, rdev);
6279 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6283 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6288 btrfs_update_inode(trans, root, inode);
6289 d_instantiate_new(dentry, inode);
6292 btrfs_end_transaction(trans);
6293 btrfs_btree_balance_dirty(fs_info);
6295 inode_dec_link_count(inode);
6296 discard_new_inode(inode);
6301 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6302 umode_t mode, bool excl)
6304 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6305 struct btrfs_trans_handle *trans;
6306 struct btrfs_root *root = BTRFS_I(dir)->root;
6307 struct inode *inode = NULL;
6313 * 2 for inode item and ref
6315 * 1 for xattr if selinux is on
6317 trans = btrfs_start_transaction(root, 5);
6319 return PTR_ERR(trans);
6321 err = btrfs_find_free_ino(root, &objectid);
6325 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6326 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6328 if (IS_ERR(inode)) {
6329 err = PTR_ERR(inode);
6334 * If the active LSM wants to access the inode during
6335 * d_instantiate it needs these. Smack checks to see
6336 * if the filesystem supports xattrs by looking at the
6339 inode->i_fop = &btrfs_file_operations;
6340 inode->i_op = &btrfs_file_inode_operations;
6341 inode->i_mapping->a_ops = &btrfs_aops;
6343 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6347 err = btrfs_update_inode(trans, root, inode);
6351 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6356 d_instantiate_new(dentry, inode);
6359 btrfs_end_transaction(trans);
6361 inode_dec_link_count(inode);
6362 discard_new_inode(inode);
6364 btrfs_btree_balance_dirty(fs_info);
6368 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6369 struct dentry *dentry)
6371 struct btrfs_trans_handle *trans = NULL;
6372 struct btrfs_root *root = BTRFS_I(dir)->root;
6373 struct inode *inode = d_inode(old_dentry);
6374 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6379 /* do not allow sys_link's with other subvols of the same device */
6380 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6383 if (inode->i_nlink >= BTRFS_LINK_MAX)
6386 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6391 * 2 items for inode and inode ref
6392 * 2 items for dir items
6393 * 1 item for parent inode
6394 * 1 item for orphan item deletion if O_TMPFILE
6396 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6397 if (IS_ERR(trans)) {
6398 err = PTR_ERR(trans);
6403 /* There are several dir indexes for this inode, clear the cache. */
6404 BTRFS_I(inode)->dir_index = 0ULL;
6406 inode_inc_iversion(inode);
6407 inode->i_ctime = current_time(inode);
6409 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6411 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6417 struct dentry *parent = dentry->d_parent;
6419 err = btrfs_update_inode(trans, root, inode);
6422 if (inode->i_nlink == 1) {
6424 * If new hard link count is 1, it's a file created
6425 * with open(2) O_TMPFILE flag.
6427 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6431 d_instantiate(dentry, inode);
6432 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6437 btrfs_end_transaction(trans);
6439 inode_dec_link_count(inode);
6442 btrfs_btree_balance_dirty(fs_info);
6446 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6448 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6449 struct inode *inode = NULL;
6450 struct btrfs_trans_handle *trans;
6451 struct btrfs_root *root = BTRFS_I(dir)->root;
6457 * 2 items for inode and ref
6458 * 2 items for dir items
6459 * 1 for xattr if selinux is on
6461 trans = btrfs_start_transaction(root, 5);
6463 return PTR_ERR(trans);
6465 err = btrfs_find_free_ino(root, &objectid);
6469 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6470 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6471 S_IFDIR | mode, &index);
6472 if (IS_ERR(inode)) {
6473 err = PTR_ERR(inode);
6478 /* these must be set before we unlock the inode */
6479 inode->i_op = &btrfs_dir_inode_operations;
6480 inode->i_fop = &btrfs_dir_file_operations;
6482 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6486 btrfs_i_size_write(BTRFS_I(inode), 0);
6487 err = btrfs_update_inode(trans, root, inode);
6491 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6492 dentry->d_name.name,
6493 dentry->d_name.len, 0, index);
6497 d_instantiate_new(dentry, inode);
6500 btrfs_end_transaction(trans);
6502 inode_dec_link_count(inode);
6503 discard_new_inode(inode);
6505 btrfs_btree_balance_dirty(fs_info);
6509 static noinline int uncompress_inline(struct btrfs_path *path,
6511 size_t pg_offset, u64 extent_offset,
6512 struct btrfs_file_extent_item *item)
6515 struct extent_buffer *leaf = path->nodes[0];
6518 unsigned long inline_size;
6522 WARN_ON(pg_offset != 0);
6523 compress_type = btrfs_file_extent_compression(leaf, item);
6524 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6525 inline_size = btrfs_file_extent_inline_item_len(leaf,
6526 btrfs_item_nr(path->slots[0]));
6527 tmp = kmalloc(inline_size, GFP_NOFS);
6530 ptr = btrfs_file_extent_inline_start(item);
6532 read_extent_buffer(leaf, tmp, ptr, inline_size);
6534 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6535 ret = btrfs_decompress(compress_type, tmp, page,
6536 extent_offset, inline_size, max_size);
6539 * decompression code contains a memset to fill in any space between the end
6540 * of the uncompressed data and the end of max_size in case the decompressed
6541 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6542 * the end of an inline extent and the beginning of the next block, so we
6543 * cover that region here.
6546 if (max_size + pg_offset < PAGE_SIZE) {
6547 char *map = kmap(page);
6548 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6556 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6557 * @inode: file to search in
6558 * @page: page to read extent data into if the extent is inline
6559 * @pg_offset: offset into @page to copy to
6560 * @start: file offset
6561 * @len: length of range starting at @start
6563 * This returns the first &struct extent_map which overlaps with the given
6564 * range, reading it from the B-tree and caching it if necessary. Note that
6565 * there may be more extents which overlap the given range after the returned
6568 * If @page is not NULL and the extent is inline, this also reads the extent
6569 * data directly into the page and marks the extent up to date in the io_tree.
6571 * Return: ERR_PTR on error, non-NULL extent_map on success.
6573 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6574 struct page *page, size_t pg_offset,
6577 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6579 u64 extent_start = 0;
6581 u64 objectid = btrfs_ino(inode);
6582 int extent_type = -1;
6583 struct btrfs_path *path = NULL;
6584 struct btrfs_root *root = inode->root;
6585 struct btrfs_file_extent_item *item;
6586 struct extent_buffer *leaf;
6587 struct btrfs_key found_key;
6588 struct extent_map *em = NULL;
6589 struct extent_map_tree *em_tree = &inode->extent_tree;
6590 struct extent_io_tree *io_tree = &inode->io_tree;
6592 read_lock(&em_tree->lock);
6593 em = lookup_extent_mapping(em_tree, start, len);
6594 read_unlock(&em_tree->lock);
6597 if (em->start > start || em->start + em->len <= start)
6598 free_extent_map(em);
6599 else if (em->block_start == EXTENT_MAP_INLINE && page)
6600 free_extent_map(em);
6604 em = alloc_extent_map();
6609 em->start = EXTENT_MAP_HOLE;
6610 em->orig_start = EXTENT_MAP_HOLE;
6612 em->block_len = (u64)-1;
6614 path = btrfs_alloc_path();
6620 /* Chances are we'll be called again, so go ahead and do readahead */
6621 path->reada = READA_FORWARD;
6624 * Unless we're going to uncompress the inline extent, no sleep would
6627 path->leave_spinning = 1;
6629 path->recurse = btrfs_is_free_space_inode(inode);
6631 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6634 } else if (ret > 0) {
6635 if (path->slots[0] == 0)
6641 leaf = path->nodes[0];
6642 item = btrfs_item_ptr(leaf, path->slots[0],
6643 struct btrfs_file_extent_item);
6644 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6645 if (found_key.objectid != objectid ||
6646 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6648 * If we backup past the first extent we want to move forward
6649 * and see if there is an extent in front of us, otherwise we'll
6650 * say there is a hole for our whole search range which can
6657 extent_type = btrfs_file_extent_type(leaf, item);
6658 extent_start = found_key.offset;
6659 extent_end = btrfs_file_extent_end(path);
6660 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6661 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6662 /* Only regular file could have regular/prealloc extent */
6663 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6666 "regular/prealloc extent found for non-regular inode %llu",
6670 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6672 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6673 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6678 if (start >= extent_end) {
6680 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6681 ret = btrfs_next_leaf(root, path);
6687 leaf = path->nodes[0];
6689 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6690 if (found_key.objectid != objectid ||
6691 found_key.type != BTRFS_EXTENT_DATA_KEY)
6693 if (start + len <= found_key.offset)
6695 if (start > found_key.offset)
6698 /* New extent overlaps with existing one */
6700 em->orig_start = start;
6701 em->len = found_key.offset - start;
6702 em->block_start = EXTENT_MAP_HOLE;
6706 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6708 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6709 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6711 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6715 size_t extent_offset;
6721 size = btrfs_file_extent_ram_bytes(leaf, item);
6722 extent_offset = page_offset(page) + pg_offset - extent_start;
6723 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6724 size - extent_offset);
6725 em->start = extent_start + extent_offset;
6726 em->len = ALIGN(copy_size, fs_info->sectorsize);
6727 em->orig_block_len = em->len;
6728 em->orig_start = em->start;
6729 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6731 btrfs_set_path_blocking(path);
6732 if (!PageUptodate(page)) {
6733 if (btrfs_file_extent_compression(leaf, item) !=
6734 BTRFS_COMPRESS_NONE) {
6735 ret = uncompress_inline(path, page, pg_offset,
6736 extent_offset, item);
6741 read_extent_buffer(leaf, map + pg_offset, ptr,
6743 if (pg_offset + copy_size < PAGE_SIZE) {
6744 memset(map + pg_offset + copy_size, 0,
6745 PAGE_SIZE - pg_offset -
6750 flush_dcache_page(page);
6752 set_extent_uptodate(io_tree, em->start,
6753 extent_map_end(em) - 1, NULL, GFP_NOFS);
6758 em->orig_start = start;
6760 em->block_start = EXTENT_MAP_HOLE;
6763 btrfs_release_path(path);
6764 if (em->start > start || extent_map_end(em) <= start) {
6766 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6767 em->start, em->len, start, len);
6772 write_lock(&em_tree->lock);
6773 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6774 write_unlock(&em_tree->lock);
6776 btrfs_free_path(path);
6778 trace_btrfs_get_extent(root, inode, em);
6781 free_extent_map(em);
6782 return ERR_PTR(ret);
6787 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6790 struct extent_map *em;
6791 struct extent_map *hole_em = NULL;
6792 u64 delalloc_start = start;
6798 em = btrfs_get_extent(inode, NULL, 0, start, len);
6802 * If our em maps to:
6804 * - a pre-alloc extent,
6805 * there might actually be delalloc bytes behind it.
6807 if (em->block_start != EXTENT_MAP_HOLE &&
6808 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6813 /* check to see if we've wrapped (len == -1 or similar) */
6822 /* ok, we didn't find anything, lets look for delalloc */
6823 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6824 end, len, EXTENT_DELALLOC, 1);
6825 delalloc_end = delalloc_start + delalloc_len;
6826 if (delalloc_end < delalloc_start)
6827 delalloc_end = (u64)-1;
6830 * We didn't find anything useful, return the original results from
6833 if (delalloc_start > end || delalloc_end <= start) {
6840 * Adjust the delalloc_start to make sure it doesn't go backwards from
6841 * the start they passed in
6843 delalloc_start = max(start, delalloc_start);
6844 delalloc_len = delalloc_end - delalloc_start;
6846 if (delalloc_len > 0) {
6849 const u64 hole_end = extent_map_end(hole_em);
6851 em = alloc_extent_map();
6859 * When btrfs_get_extent can't find anything it returns one
6862 * Make sure what it found really fits our range, and adjust to
6863 * make sure it is based on the start from the caller
6865 if (hole_end <= start || hole_em->start > end) {
6866 free_extent_map(hole_em);
6869 hole_start = max(hole_em->start, start);
6870 hole_len = hole_end - hole_start;
6873 if (hole_em && delalloc_start > hole_start) {
6875 * Our hole starts before our delalloc, so we have to
6876 * return just the parts of the hole that go until the
6879 em->len = min(hole_len, delalloc_start - hole_start);
6880 em->start = hole_start;
6881 em->orig_start = hole_start;
6883 * Don't adjust block start at all, it is fixed at
6886 em->block_start = hole_em->block_start;
6887 em->block_len = hole_len;
6888 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6889 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6892 * Hole is out of passed range or it starts after
6895 em->start = delalloc_start;
6896 em->len = delalloc_len;
6897 em->orig_start = delalloc_start;
6898 em->block_start = EXTENT_MAP_DELALLOC;
6899 em->block_len = delalloc_len;
6906 free_extent_map(hole_em);
6908 free_extent_map(em);
6909 return ERR_PTR(err);
6914 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6917 const u64 orig_start,
6918 const u64 block_start,
6919 const u64 block_len,
6920 const u64 orig_block_len,
6921 const u64 ram_bytes,
6924 struct extent_map *em = NULL;
6927 if (type != BTRFS_ORDERED_NOCOW) {
6928 em = create_io_em(inode, start, len, orig_start, block_start,
6929 block_len, orig_block_len, ram_bytes,
6930 BTRFS_COMPRESS_NONE, /* compress_type */
6935 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
6939 free_extent_map(em);
6940 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
6949 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6952 struct btrfs_root *root = inode->root;
6953 struct btrfs_fs_info *fs_info = root->fs_info;
6954 struct extent_map *em;
6955 struct btrfs_key ins;
6959 alloc_hint = get_extent_allocation_hint(inode, start, len);
6960 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6961 0, alloc_hint, &ins, 1, 1);
6963 return ERR_PTR(ret);
6965 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
6966 ins.objectid, ins.offset, ins.offset,
6967 ins.offset, BTRFS_ORDERED_REGULAR);
6968 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6970 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
6977 * Check if we can do nocow write into the range [@offset, @offset + @len)
6979 * @offset: File offset
6980 * @len: The length to write, will be updated to the nocow writeable
6982 * @orig_start: (optional) Return the original file offset of the file extent
6983 * @orig_len: (optional) Return the original on-disk length of the file extent
6984 * @ram_bytes: (optional) Return the ram_bytes of the file extent
6985 * @strict: if true, omit optimizations that might force us into unnecessary
6986 * cow. e.g., don't trust generation number.
6988 * This function will flush ordered extents in the range to ensure proper
6989 * nocow checks for (nowait == false) case.
6992 * >0 and update @len if we can do nocow write
6993 * 0 if we can't do nocow write
6994 * <0 if error happened
6996 * NOTE: This only checks the file extents, caller is responsible to wait for
6997 * any ordered extents.
6999 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7000 u64 *orig_start, u64 *orig_block_len,
7001 u64 *ram_bytes, bool strict)
7003 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7004 struct btrfs_path *path;
7006 struct extent_buffer *leaf;
7007 struct btrfs_root *root = BTRFS_I(inode)->root;
7008 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7009 struct btrfs_file_extent_item *fi;
7010 struct btrfs_key key;
7017 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7019 path = btrfs_alloc_path();
7023 ret = btrfs_lookup_file_extent(NULL, root, path,
7024 btrfs_ino(BTRFS_I(inode)), offset, 0);
7028 slot = path->slots[0];
7031 /* can't find the item, must cow */
7038 leaf = path->nodes[0];
7039 btrfs_item_key_to_cpu(leaf, &key, slot);
7040 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7041 key.type != BTRFS_EXTENT_DATA_KEY) {
7042 /* not our file or wrong item type, must cow */
7046 if (key.offset > offset) {
7047 /* Wrong offset, must cow */
7051 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7052 found_type = btrfs_file_extent_type(leaf, fi);
7053 if (found_type != BTRFS_FILE_EXTENT_REG &&
7054 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7055 /* not a regular extent, must cow */
7059 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7062 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7063 if (extent_end <= offset)
7066 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7067 if (disk_bytenr == 0)
7070 if (btrfs_file_extent_compression(leaf, fi) ||
7071 btrfs_file_extent_encryption(leaf, fi) ||
7072 btrfs_file_extent_other_encoding(leaf, fi))
7076 * Do the same check as in btrfs_cross_ref_exist but without the
7077 * unnecessary search.
7080 (btrfs_file_extent_generation(leaf, fi) <=
7081 btrfs_root_last_snapshot(&root->root_item)))
7084 backref_offset = btrfs_file_extent_offset(leaf, fi);
7087 *orig_start = key.offset - backref_offset;
7088 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7089 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7092 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7095 num_bytes = min(offset + *len, extent_end) - offset;
7096 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7099 range_end = round_up(offset + num_bytes,
7100 root->fs_info->sectorsize) - 1;
7101 ret = test_range_bit(io_tree, offset, range_end,
7102 EXTENT_DELALLOC, 0, NULL);
7109 btrfs_release_path(path);
7112 * look for other files referencing this extent, if we
7113 * find any we must cow
7116 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7117 key.offset - backref_offset, disk_bytenr,
7125 * adjust disk_bytenr and num_bytes to cover just the bytes
7126 * in this extent we are about to write. If there
7127 * are any csums in that range we have to cow in order
7128 * to keep the csums correct
7130 disk_bytenr += backref_offset;
7131 disk_bytenr += offset - key.offset;
7132 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7135 * all of the above have passed, it is safe to overwrite this extent
7141 btrfs_free_path(path);
7145 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7146 struct extent_state **cached_state, bool writing)
7148 struct btrfs_ordered_extent *ordered;
7152 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7155 * We're concerned with the entire range that we're going to be
7156 * doing DIO to, so we need to make sure there's no ordered
7157 * extents in this range.
7159 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7160 lockend - lockstart + 1);
7163 * We need to make sure there are no buffered pages in this
7164 * range either, we could have raced between the invalidate in
7165 * generic_file_direct_write and locking the extent. The
7166 * invalidate needs to happen so that reads after a write do not
7170 (!writing || !filemap_range_has_page(inode->i_mapping,
7171 lockstart, lockend)))
7174 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7179 * If we are doing a DIO read and the ordered extent we
7180 * found is for a buffered write, we can not wait for it
7181 * to complete and retry, because if we do so we can
7182 * deadlock with concurrent buffered writes on page
7183 * locks. This happens only if our DIO read covers more
7184 * than one extent map, if at this point has already
7185 * created an ordered extent for a previous extent map
7186 * and locked its range in the inode's io tree, and a
7187 * concurrent write against that previous extent map's
7188 * range and this range started (we unlock the ranges
7189 * in the io tree only when the bios complete and
7190 * buffered writes always lock pages before attempting
7191 * to lock range in the io tree).
7194 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7195 btrfs_start_ordered_extent(ordered, 1);
7198 btrfs_put_ordered_extent(ordered);
7201 * We could trigger writeback for this range (and wait
7202 * for it to complete) and then invalidate the pages for
7203 * this range (through invalidate_inode_pages2_range()),
7204 * but that can lead us to a deadlock with a concurrent
7205 * call to readahead (a buffered read or a defrag call
7206 * triggered a readahead) on a page lock due to an
7207 * ordered dio extent we created before but did not have
7208 * yet a corresponding bio submitted (whence it can not
7209 * complete), which makes readahead wait for that
7210 * ordered extent to complete while holding a lock on
7225 /* The callers of this must take lock_extent() */
7226 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7227 u64 len, u64 orig_start, u64 block_start,
7228 u64 block_len, u64 orig_block_len,
7229 u64 ram_bytes, int compress_type,
7232 struct extent_map_tree *em_tree;
7233 struct extent_map *em;
7236 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7237 type == BTRFS_ORDERED_COMPRESSED ||
7238 type == BTRFS_ORDERED_NOCOW ||
7239 type == BTRFS_ORDERED_REGULAR);
7241 em_tree = &inode->extent_tree;
7242 em = alloc_extent_map();
7244 return ERR_PTR(-ENOMEM);
7247 em->orig_start = orig_start;
7249 em->block_len = block_len;
7250 em->block_start = block_start;
7251 em->orig_block_len = orig_block_len;
7252 em->ram_bytes = ram_bytes;
7253 em->generation = -1;
7254 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7255 if (type == BTRFS_ORDERED_PREALLOC) {
7256 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7257 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7258 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7259 em->compress_type = compress_type;
7263 btrfs_drop_extent_cache(inode, em->start,
7264 em->start + em->len - 1, 0);
7265 write_lock(&em_tree->lock);
7266 ret = add_extent_mapping(em_tree, em, 1);
7267 write_unlock(&em_tree->lock);
7269 * The caller has taken lock_extent(), who could race with us
7272 } while (ret == -EEXIST);
7275 free_extent_map(em);
7276 return ERR_PTR(ret);
7279 /* em got 2 refs now, callers needs to do free_extent_map once. */
7284 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7285 struct inode *inode,
7286 struct btrfs_dio_data *dio_data,
7289 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7290 struct extent_map *em = *map;
7294 * We don't allocate a new extent in the following cases
7296 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7298 * 2) The extent is marked as PREALLOC. We're good to go here and can
7299 * just use the extent.
7302 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7303 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7304 em->block_start != EXTENT_MAP_HOLE)) {
7306 u64 block_start, orig_start, orig_block_len, ram_bytes;
7308 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7309 type = BTRFS_ORDERED_PREALLOC;
7311 type = BTRFS_ORDERED_NOCOW;
7312 len = min(len, em->len - (start - em->start));
7313 block_start = em->block_start + (start - em->start);
7315 if (can_nocow_extent(inode, start, &len, &orig_start,
7316 &orig_block_len, &ram_bytes, false) == 1 &&
7317 btrfs_inc_nocow_writers(fs_info, block_start)) {
7318 struct extent_map *em2;
7320 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7321 orig_start, block_start,
7322 len, orig_block_len,
7324 btrfs_dec_nocow_writers(fs_info, block_start);
7325 if (type == BTRFS_ORDERED_PREALLOC) {
7326 free_extent_map(em);
7330 if (em2 && IS_ERR(em2)) {
7335 * For inode marked NODATACOW or extent marked PREALLOC,
7336 * use the existing or preallocated extent, so does not
7337 * need to adjust btrfs_space_info's bytes_may_use.
7339 btrfs_free_reserved_data_space_noquota(fs_info, len);
7344 /* this will cow the extent */
7345 free_extent_map(em);
7346 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7352 len = min(len, em->len - (start - em->start));
7356 * Need to update the i_size under the extent lock so buffered
7357 * readers will get the updated i_size when we unlock.
7359 if (start + len > i_size_read(inode))
7360 i_size_write(inode, start + len);
7362 dio_data->reserve -= len;
7367 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7368 loff_t length, unsigned int flags, struct iomap *iomap,
7369 struct iomap *srcmap)
7371 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7372 struct extent_map *em;
7373 struct extent_state *cached_state = NULL;
7374 struct btrfs_dio_data *dio_data = NULL;
7375 u64 lockstart, lockend;
7376 const bool write = !!(flags & IOMAP_WRITE);
7379 bool unlock_extents = false;
7380 bool sync = (current->journal_info == BTRFS_DIO_SYNC_STUB);
7383 * We used current->journal_info here to see if we were sync, but
7384 * there's a lot of tests in the enospc machinery to not do flushing if
7385 * we have a journal_info set, so we need to clear this out and re-set
7388 ASSERT(current->journal_info == NULL ||
7389 current->journal_info == BTRFS_DIO_SYNC_STUB);
7390 current->journal_info = NULL;
7393 len = min_t(u64, len, fs_info->sectorsize);
7396 lockend = start + len - 1;
7399 * The generic stuff only does filemap_write_and_wait_range, which
7400 * isn't enough if we've written compressed pages to this area, so we
7401 * need to flush the dirty pages again to make absolutely sure that any
7402 * outstanding dirty pages are on disk.
7404 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7405 &BTRFS_I(inode)->runtime_flags)) {
7406 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7407 start + length - 1);
7412 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7416 dio_data->sync = sync;
7417 dio_data->length = length;
7419 dio_data->reserve = round_up(length, fs_info->sectorsize);
7420 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7421 &dio_data->data_reserved,
7422 start, dio_data->reserve);
7424 extent_changeset_free(dio_data->data_reserved);
7429 iomap->private = dio_data;
7433 * If this errors out it's because we couldn't invalidate pagecache for
7434 * this range and we need to fallback to buffered.
7436 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7441 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7448 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7449 * io. INLINE is special, and we could probably kludge it in here, but
7450 * it's still buffered so for safety lets just fall back to the generic
7453 * For COMPRESSED we _have_ to read the entire extent in so we can
7454 * decompress it, so there will be buffering required no matter what we
7455 * do, so go ahead and fallback to buffered.
7457 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7458 * to buffered IO. Don't blame me, this is the price we pay for using
7461 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7462 em->block_start == EXTENT_MAP_INLINE) {
7463 free_extent_map(em);
7468 len = min(len, em->len - (start - em->start));
7470 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7474 unlock_extents = true;
7475 /* Recalc len in case the new em is smaller than requested */
7476 len = min(len, em->len - (start - em->start));
7479 * We need to unlock only the end area that we aren't using.
7480 * The rest is going to be unlocked by the endio routine.
7482 lockstart = start + len;
7483 if (lockstart < lockend)
7484 unlock_extents = true;
7488 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7489 lockstart, lockend, &cached_state);
7491 free_extent_state(cached_state);
7494 * Translate extent map information to iomap.
7495 * We trim the extents (and move the addr) even though iomap code does
7496 * that, since we have locked only the parts we are performing I/O in.
7498 if ((em->block_start == EXTENT_MAP_HOLE) ||
7499 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7500 iomap->addr = IOMAP_NULL_ADDR;
7501 iomap->type = IOMAP_HOLE;
7503 iomap->addr = em->block_start + (start - em->start);
7504 iomap->type = IOMAP_MAPPED;
7506 iomap->offset = start;
7507 iomap->bdev = fs_info->fs_devices->latest_bdev;
7508 iomap->length = len;
7510 free_extent_map(em);
7515 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7519 btrfs_delalloc_release_space(BTRFS_I(inode),
7520 dio_data->data_reserved, start,
7521 dio_data->reserve, true);
7522 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7523 extent_changeset_free(dio_data->data_reserved);
7529 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7530 ssize_t written, unsigned int flags, struct iomap *iomap)
7533 struct btrfs_dio_data *dio_data = iomap->private;
7534 size_t submitted = dio_data->submitted;
7535 const bool write = !!(flags & IOMAP_WRITE);
7537 if (!write && (iomap->type == IOMAP_HOLE)) {
7538 /* If reading from a hole, unlock and return */
7539 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7543 if (submitted < length) {
7545 length -= submitted;
7547 __endio_write_update_ordered(BTRFS_I(inode), pos,
7550 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7556 if (dio_data->reserve)
7557 btrfs_delalloc_release_space(BTRFS_I(inode),
7558 dio_data->data_reserved, pos,
7559 dio_data->reserve, true);
7560 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7561 extent_changeset_free(dio_data->data_reserved);
7565 * We're all done, we can re-set the current->journal_info now safely
7568 if (dio_data->sync) {
7569 ASSERT(current->journal_info == NULL);
7570 current->journal_info = BTRFS_DIO_SYNC_STUB;
7573 iomap->private = NULL;
7578 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7581 * This implies a barrier so that stores to dio_bio->bi_status before
7582 * this and loads of dio_bio->bi_status after this are fully ordered.
7584 if (!refcount_dec_and_test(&dip->refs))
7587 if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
7588 __endio_write_update_ordered(BTRFS_I(dip->inode),
7589 dip->logical_offset,
7591 !dip->dio_bio->bi_status);
7593 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7594 dip->logical_offset,
7595 dip->logical_offset + dip->bytes - 1);
7598 bio_endio(dip->dio_bio);
7602 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7604 unsigned long bio_flags)
7606 struct btrfs_dio_private *dip = bio->bi_private;
7607 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7610 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7612 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7616 refcount_inc(&dip->refs);
7617 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7619 refcount_dec(&dip->refs);
7623 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7624 struct btrfs_io_bio *io_bio,
7625 const bool uptodate)
7627 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7628 const u32 sectorsize = fs_info->sectorsize;
7629 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7630 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7631 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7632 struct bio_vec bvec;
7633 struct bvec_iter iter;
7634 u64 start = io_bio->logical;
7636 blk_status_t err = BLK_STS_OK;
7638 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7639 unsigned int i, nr_sectors, pgoff;
7641 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7642 pgoff = bvec.bv_offset;
7643 for (i = 0; i < nr_sectors; i++) {
7644 ASSERT(pgoff < PAGE_SIZE);
7646 (!csum || !check_data_csum(inode, io_bio, icsum,
7647 bvec.bv_page, pgoff,
7648 start, sectorsize))) {
7649 clean_io_failure(fs_info, failure_tree, io_tree,
7650 start, bvec.bv_page,
7651 btrfs_ino(BTRFS_I(inode)),
7654 blk_status_t status;
7656 status = btrfs_submit_read_repair(inode,
7658 start - io_bio->logical,
7659 bvec.bv_page, pgoff,
7661 start + sectorsize - 1,
7663 submit_dio_repair_bio);
7667 start += sectorsize;
7669 pgoff += sectorsize;
7675 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7676 const u64 offset, const u64 bytes,
7677 const bool uptodate)
7679 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7680 struct btrfs_ordered_extent *ordered = NULL;
7681 struct btrfs_workqueue *wq;
7682 u64 ordered_offset = offset;
7683 u64 ordered_bytes = bytes;
7686 if (btrfs_is_free_space_inode(inode))
7687 wq = fs_info->endio_freespace_worker;
7689 wq = fs_info->endio_write_workers;
7691 while (ordered_offset < offset + bytes) {
7692 last_offset = ordered_offset;
7693 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7697 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7699 btrfs_queue_work(wq, &ordered->work);
7702 * If btrfs_dec_test_ordered_pending does not find any ordered
7703 * extent in the range, we can exit.
7705 if (ordered_offset == last_offset)
7708 * Our bio might span multiple ordered extents. In this case
7709 * we keep going until we have accounted the whole dio.
7711 if (ordered_offset < offset + bytes) {
7712 ordered_bytes = offset + bytes - ordered_offset;
7718 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
7719 struct bio *bio, u64 offset)
7721 struct inode *inode = private_data;
7723 return btrfs_csum_one_bio(BTRFS_I(inode), bio, offset, 1);
7726 static void btrfs_end_dio_bio(struct bio *bio)
7728 struct btrfs_dio_private *dip = bio->bi_private;
7729 blk_status_t err = bio->bi_status;
7732 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7733 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7734 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7736 (unsigned long long)bio->bi_iter.bi_sector,
7737 bio->bi_iter.bi_size, err);
7739 if (bio_op(bio) == REQ_OP_READ) {
7740 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7745 dip->dio_bio->bi_status = err;
7748 btrfs_dio_private_put(dip);
7751 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7752 struct inode *inode, u64 file_offset, int async_submit)
7754 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7755 struct btrfs_dio_private *dip = bio->bi_private;
7756 bool write = bio_op(bio) == REQ_OP_WRITE;
7759 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7761 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7764 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7769 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7772 if (write && async_submit) {
7773 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
7775 btrfs_submit_bio_start_direct_io);
7779 * If we aren't doing async submit, calculate the csum of the
7782 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
7788 csum_offset = file_offset - dip->logical_offset;
7789 csum_offset >>= inode->i_sb->s_blocksize_bits;
7790 csum_offset *= btrfs_super_csum_size(fs_info->super_copy);
7791 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
7794 ret = btrfs_map_bio(fs_info, bio, 0);
7800 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7801 * or ordered extents whether or not we submit any bios.
7803 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7804 struct inode *inode,
7807 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7808 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7810 struct btrfs_dio_private *dip;
7812 dip_size = sizeof(*dip);
7813 if (!write && csum) {
7814 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7815 const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
7818 nblocks = dio_bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
7819 dip_size += csum_size * nblocks;
7822 dip = kzalloc(dip_size, GFP_NOFS);
7827 dip->logical_offset = file_offset;
7828 dip->bytes = dio_bio->bi_iter.bi_size;
7829 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7830 dip->dio_bio = dio_bio;
7831 refcount_set(&dip->refs, 1);
7835 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
7836 struct bio *dio_bio, loff_t file_offset)
7838 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7839 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7840 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7841 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7842 BTRFS_BLOCK_GROUP_RAID56_MASK);
7843 struct btrfs_dio_private *dip;
7846 int async_submit = 0;
7848 int clone_offset = 0;
7851 blk_status_t status;
7852 struct btrfs_io_geometry geom;
7853 struct btrfs_dio_data *dio_data = iomap->private;
7855 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7858 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7859 file_offset + dio_bio->bi_iter.bi_size - 1);
7861 dio_bio->bi_status = BLK_STS_RESOURCE;
7863 return BLK_QC_T_NONE;
7866 if (!write && csum) {
7868 * Load the csums up front to reduce csum tree searches and
7869 * contention when submitting bios.
7871 status = btrfs_lookup_bio_sums(inode, dio_bio, file_offset,
7873 if (status != BLK_STS_OK)
7877 start_sector = dio_bio->bi_iter.bi_sector;
7878 submit_len = dio_bio->bi_iter.bi_size;
7881 ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio),
7882 start_sector << 9, submit_len,
7885 status = errno_to_blk_status(ret);
7888 ASSERT(geom.len <= INT_MAX);
7890 clone_len = min_t(int, submit_len, geom.len);
7893 * This will never fail as it's passing GPF_NOFS and
7894 * the allocation is backed by btrfs_bioset.
7896 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7897 bio->bi_private = dip;
7898 bio->bi_end_io = btrfs_end_dio_bio;
7899 btrfs_io_bio(bio)->logical = file_offset;
7901 ASSERT(submit_len >= clone_len);
7902 submit_len -= clone_len;
7905 * Increase the count before we submit the bio so we know
7906 * the end IO handler won't happen before we increase the
7907 * count. Otherwise, the dip might get freed before we're
7908 * done setting it up.
7910 * We transfer the initial reference to the last bio, so we
7911 * don't need to increment the reference count for the last one.
7913 if (submit_len > 0) {
7914 refcount_inc(&dip->refs);
7916 * If we are submitting more than one bio, submit them
7917 * all asynchronously. The exception is RAID 5 or 6, as
7918 * asynchronous checksums make it difficult to collect
7919 * full stripe writes.
7925 status = btrfs_submit_dio_bio(bio, inode, file_offset,
7930 refcount_dec(&dip->refs);
7934 dio_data->submitted += clone_len;
7935 clone_offset += clone_len;
7936 start_sector += clone_len >> 9;
7937 file_offset += clone_len;
7938 } while (submit_len > 0);
7939 return BLK_QC_T_NONE;
7942 dip->dio_bio->bi_status = status;
7943 btrfs_dio_private_put(dip);
7944 return BLK_QC_T_NONE;
7947 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
7948 const struct iov_iter *iter, loff_t offset)
7952 unsigned int blocksize_mask = fs_info->sectorsize - 1;
7953 ssize_t retval = -EINVAL;
7955 if (offset & blocksize_mask)
7958 if (iov_iter_alignment(iter) & blocksize_mask)
7961 /* If this is a write we don't need to check anymore */
7962 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
7965 * Check to make sure we don't have duplicate iov_base's in this
7966 * iovec, if so return EINVAL, otherwise we'll get csum errors
7967 * when reading back.
7969 for (seg = 0; seg < iter->nr_segs; seg++) {
7970 for (i = seg + 1; i < iter->nr_segs; i++) {
7971 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7980 static inline int btrfs_maybe_fsync_end_io(struct kiocb *iocb, ssize_t size,
7981 int error, unsigned flags)
7984 * Now if we're still in the context of our submitter we know we can't
7985 * safely run generic_write_sync(), so clear our flag here so that the
7986 * caller knows to follow up with a sync.
7988 if (current->journal_info == BTRFS_DIO_SYNC_STUB) {
7989 current->journal_info = NULL;
7997 iocb->ki_flags |= IOCB_DSYNC;
7998 return generic_write_sync(iocb, size);
8004 static const struct iomap_ops btrfs_dio_iomap_ops = {
8005 .iomap_begin = btrfs_dio_iomap_begin,
8006 .iomap_end = btrfs_dio_iomap_end,
8009 static const struct iomap_dio_ops btrfs_dio_ops = {
8010 .submit_io = btrfs_submit_direct,
8013 static const struct iomap_dio_ops btrfs_sync_dops = {
8014 .submit_io = btrfs_submit_direct,
8015 .end_io = btrfs_maybe_fsync_end_io,
8018 ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8020 struct file *file = iocb->ki_filp;
8021 struct inode *inode = file->f_mapping->host;
8022 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8023 struct extent_changeset *data_reserved = NULL;
8024 loff_t offset = iocb->ki_pos;
8026 bool relock = false;
8029 if (check_direct_IO(fs_info, iter, offset))
8032 count = iov_iter_count(iter);
8033 if (iov_iter_rw(iter) == WRITE) {
8035 * If the write DIO is beyond the EOF, we need update
8036 * the isize, but it is protected by i_mutex. So we can
8037 * not unlock the i_mutex at this case.
8039 if (offset + count <= inode->i_size) {
8040 inode_unlock(inode);
8043 down_read(&BTRFS_I(inode)->dio_sem);
8047 * We have are actually a sync iocb, so we need our fancy endio to know
8048 * if we need to sync.
8050 if (current->journal_info)
8051 ret = iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops,
8052 &btrfs_sync_dops, is_sync_kiocb(iocb));
8054 ret = iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops,
8055 &btrfs_dio_ops, is_sync_kiocb(iocb));
8057 if (ret == -ENOTBLK)
8060 if (iov_iter_rw(iter) == WRITE)
8061 up_read(&BTRFS_I(inode)->dio_sem);
8066 extent_changeset_free(data_reserved);
8070 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8075 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8079 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8082 int btrfs_readpage(struct file *file, struct page *page)
8084 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8085 u64 start = page_offset(page);
8086 u64 end = start + PAGE_SIZE - 1;
8087 unsigned long bio_flags = 0;
8088 struct bio *bio = NULL;
8091 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8093 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8095 ret = submit_one_bio(bio, 0, bio_flags);
8099 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8101 struct inode *inode = page->mapping->host;
8104 if (current->flags & PF_MEMALLOC) {
8105 redirty_page_for_writepage(wbc, page);
8111 * If we are under memory pressure we will call this directly from the
8112 * VM, we need to make sure we have the inode referenced for the ordered
8113 * extent. If not just return like we didn't do anything.
8115 if (!igrab(inode)) {
8116 redirty_page_for_writepage(wbc, page);
8117 return AOP_WRITEPAGE_ACTIVATE;
8119 ret = extent_write_full_page(page, wbc);
8120 btrfs_add_delayed_iput(inode);
8124 static int btrfs_writepages(struct address_space *mapping,
8125 struct writeback_control *wbc)
8127 return extent_writepages(mapping, wbc);
8130 static void btrfs_readahead(struct readahead_control *rac)
8132 extent_readahead(rac);
8135 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8137 int ret = try_release_extent_mapping(page, gfp_flags);
8139 detach_page_private(page);
8143 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8145 if (PageWriteback(page) || PageDirty(page))
8147 return __btrfs_releasepage(page, gfp_flags);
8150 #ifdef CONFIG_MIGRATION
8151 static int btrfs_migratepage(struct address_space *mapping,
8152 struct page *newpage, struct page *page,
8153 enum migrate_mode mode)
8157 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8158 if (ret != MIGRATEPAGE_SUCCESS)
8161 if (page_has_private(page))
8162 attach_page_private(newpage, detach_page_private(page));
8164 if (PagePrivate2(page)) {
8165 ClearPagePrivate2(page);
8166 SetPagePrivate2(newpage);
8169 if (mode != MIGRATE_SYNC_NO_COPY)
8170 migrate_page_copy(newpage, page);
8172 migrate_page_states(newpage, page);
8173 return MIGRATEPAGE_SUCCESS;
8177 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8178 unsigned int length)
8180 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8181 struct extent_io_tree *tree = &inode->io_tree;
8182 struct btrfs_ordered_extent *ordered;
8183 struct extent_state *cached_state = NULL;
8184 u64 page_start = page_offset(page);
8185 u64 page_end = page_start + PAGE_SIZE - 1;
8188 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8191 * we have the page locked, so new writeback can't start,
8192 * and the dirty bit won't be cleared while we are here.
8194 * Wait for IO on this page so that we can safely clear
8195 * the PagePrivate2 bit and do ordered accounting
8197 wait_on_page_writeback(page);
8200 btrfs_releasepage(page, GFP_NOFS);
8204 if (!inode_evicting)
8205 lock_extent_bits(tree, page_start, page_end, &cached_state);
8208 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8211 ordered->file_offset + ordered->num_bytes - 1);
8213 * IO on this page will never be started, so we need
8214 * to account for any ordered extents now
8216 if (!inode_evicting)
8217 clear_extent_bit(tree, start, end,
8218 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8219 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8220 EXTENT_DEFRAG, 1, 0, &cached_state);
8222 * whoever cleared the private bit is responsible
8223 * for the finish_ordered_io
8225 if (TestClearPagePrivate2(page)) {
8226 struct btrfs_ordered_inode_tree *tree;
8229 tree = &inode->ordered_tree;
8231 spin_lock_irq(&tree->lock);
8232 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8233 new_len = start - ordered->file_offset;
8234 if (new_len < ordered->truncated_len)
8235 ordered->truncated_len = new_len;
8236 spin_unlock_irq(&tree->lock);
8238 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8240 end - start + 1, 1))
8241 btrfs_finish_ordered_io(ordered);
8243 btrfs_put_ordered_extent(ordered);
8244 if (!inode_evicting) {
8245 cached_state = NULL;
8246 lock_extent_bits(tree, start, end,
8251 if (start < page_end)
8256 * Qgroup reserved space handler
8257 * Page here will be either
8258 * 1) Already written to disk or ordered extent already submitted
8259 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8260 * Qgroup will be handled by its qgroup_record then.
8261 * btrfs_qgroup_free_data() call will do nothing here.
8263 * 2) Not written to disk yet
8264 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8265 * bit of its io_tree, and free the qgroup reserved data space.
8266 * Since the IO will never happen for this page.
8268 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8269 if (!inode_evicting) {
8270 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8271 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8272 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8275 __btrfs_releasepage(page, GFP_NOFS);
8278 ClearPageChecked(page);
8279 detach_page_private(page);
8283 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8284 * called from a page fault handler when a page is first dirtied. Hence we must
8285 * be careful to check for EOF conditions here. We set the page up correctly
8286 * for a written page which means we get ENOSPC checking when writing into
8287 * holes and correct delalloc and unwritten extent mapping on filesystems that
8288 * support these features.
8290 * We are not allowed to take the i_mutex here so we have to play games to
8291 * protect against truncate races as the page could now be beyond EOF. Because
8292 * truncate_setsize() writes the inode size before removing pages, once we have
8293 * the page lock we can determine safely if the page is beyond EOF. If it is not
8294 * beyond EOF, then the page is guaranteed safe against truncation until we
8297 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8299 struct page *page = vmf->page;
8300 struct inode *inode = file_inode(vmf->vma->vm_file);
8301 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8302 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8303 struct btrfs_ordered_extent *ordered;
8304 struct extent_state *cached_state = NULL;
8305 struct extent_changeset *data_reserved = NULL;
8307 unsigned long zero_start;
8317 reserved_space = PAGE_SIZE;
8319 sb_start_pagefault(inode->i_sb);
8320 page_start = page_offset(page);
8321 page_end = page_start + PAGE_SIZE - 1;
8325 * Reserving delalloc space after obtaining the page lock can lead to
8326 * deadlock. For example, if a dirty page is locked by this function
8327 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8328 * dirty page write out, then the btrfs_writepage() function could
8329 * end up waiting indefinitely to get a lock on the page currently
8330 * being processed by btrfs_page_mkwrite() function.
8332 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8333 page_start, reserved_space);
8335 ret2 = file_update_time(vmf->vma->vm_file);
8339 ret = vmf_error(ret2);
8345 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8348 size = i_size_read(inode);
8350 if ((page->mapping != inode->i_mapping) ||
8351 (page_start >= size)) {
8352 /* page got truncated out from underneath us */
8355 wait_on_page_writeback(page);
8357 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8358 set_page_extent_mapped(page);
8361 * we can't set the delalloc bits if there are pending ordered
8362 * extents. Drop our locks and wait for them to finish
8364 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8367 unlock_extent_cached(io_tree, page_start, page_end,
8370 btrfs_start_ordered_extent(ordered, 1);
8371 btrfs_put_ordered_extent(ordered);
8375 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8376 reserved_space = round_up(size - page_start,
8377 fs_info->sectorsize);
8378 if (reserved_space < PAGE_SIZE) {
8379 end = page_start + reserved_space - 1;
8380 btrfs_delalloc_release_space(BTRFS_I(inode),
8381 data_reserved, page_start,
8382 PAGE_SIZE - reserved_space, true);
8387 * page_mkwrite gets called when the page is firstly dirtied after it's
8388 * faulted in, but write(2) could also dirty a page and set delalloc
8389 * bits, thus in this case for space account reason, we still need to
8390 * clear any delalloc bits within this page range since we have to
8391 * reserve data&meta space before lock_page() (see above comments).
8393 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8394 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8395 EXTENT_DEFRAG, 0, 0, &cached_state);
8397 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8400 unlock_extent_cached(io_tree, page_start, page_end,
8402 ret = VM_FAULT_SIGBUS;
8406 /* page is wholly or partially inside EOF */
8407 if (page_start + PAGE_SIZE > size)
8408 zero_start = offset_in_page(size);
8410 zero_start = PAGE_SIZE;
8412 if (zero_start != PAGE_SIZE) {
8414 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8415 flush_dcache_page(page);
8418 ClearPageChecked(page);
8419 set_page_dirty(page);
8420 SetPageUptodate(page);
8422 BTRFS_I(inode)->last_trans = fs_info->generation;
8423 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8424 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8426 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8428 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8429 sb_end_pagefault(inode->i_sb);
8430 extent_changeset_free(data_reserved);
8431 return VM_FAULT_LOCKED;
8436 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8437 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8438 reserved_space, (ret != 0));
8440 sb_end_pagefault(inode->i_sb);
8441 extent_changeset_free(data_reserved);
8445 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8447 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8448 struct btrfs_root *root = BTRFS_I(inode)->root;
8449 struct btrfs_block_rsv *rsv;
8451 struct btrfs_trans_handle *trans;
8452 u64 mask = fs_info->sectorsize - 1;
8453 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8455 if (!skip_writeback) {
8456 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8463 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8464 * things going on here:
8466 * 1) We need to reserve space to update our inode.
8468 * 2) We need to have something to cache all the space that is going to
8469 * be free'd up by the truncate operation, but also have some slack
8470 * space reserved in case it uses space during the truncate (thank you
8471 * very much snapshotting).
8473 * And we need these to be separate. The fact is we can use a lot of
8474 * space doing the truncate, and we have no earthly idea how much space
8475 * we will use, so we need the truncate reservation to be separate so it
8476 * doesn't end up using space reserved for updating the inode. We also
8477 * need to be able to stop the transaction and start a new one, which
8478 * means we need to be able to update the inode several times, and we
8479 * have no idea of knowing how many times that will be, so we can't just
8480 * reserve 1 item for the entirety of the operation, so that has to be
8481 * done separately as well.
8483 * So that leaves us with
8485 * 1) rsv - for the truncate reservation, which we will steal from the
8486 * transaction reservation.
8487 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8488 * updating the inode.
8490 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8493 rsv->size = min_size;
8497 * 1 for the truncate slack space
8498 * 1 for updating the inode.
8500 trans = btrfs_start_transaction(root, 2);
8501 if (IS_ERR(trans)) {
8502 ret = PTR_ERR(trans);
8506 /* Migrate the slack space for the truncate to our reserve */
8507 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8512 * So if we truncate and then write and fsync we normally would just
8513 * write the extents that changed, which is a problem if we need to
8514 * first truncate that entire inode. So set this flag so we write out
8515 * all of the extents in the inode to the sync log so we're completely
8518 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8519 trans->block_rsv = rsv;
8522 ret = btrfs_truncate_inode_items(trans, root, inode,
8524 BTRFS_EXTENT_DATA_KEY);
8525 trans->block_rsv = &fs_info->trans_block_rsv;
8526 if (ret != -ENOSPC && ret != -EAGAIN)
8529 ret = btrfs_update_inode(trans, root, inode);
8533 btrfs_end_transaction(trans);
8534 btrfs_btree_balance_dirty(fs_info);
8536 trans = btrfs_start_transaction(root, 2);
8537 if (IS_ERR(trans)) {
8538 ret = PTR_ERR(trans);
8543 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8544 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8545 rsv, min_size, false);
8546 BUG_ON(ret); /* shouldn't happen */
8547 trans->block_rsv = rsv;
8551 * We can't call btrfs_truncate_block inside a trans handle as we could
8552 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8553 * we've truncated everything except the last little bit, and can do
8554 * btrfs_truncate_block and then update the disk_i_size.
8556 if (ret == NEED_TRUNCATE_BLOCK) {
8557 btrfs_end_transaction(trans);
8558 btrfs_btree_balance_dirty(fs_info);
8560 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
8563 trans = btrfs_start_transaction(root, 1);
8564 if (IS_ERR(trans)) {
8565 ret = PTR_ERR(trans);
8568 btrfs_inode_safe_disk_i_size_write(inode, 0);
8574 trans->block_rsv = &fs_info->trans_block_rsv;
8575 ret2 = btrfs_update_inode(trans, root, inode);
8579 ret2 = btrfs_end_transaction(trans);
8582 btrfs_btree_balance_dirty(fs_info);
8585 btrfs_free_block_rsv(fs_info, rsv);
8591 * create a new subvolume directory/inode (helper for the ioctl).
8593 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8594 struct btrfs_root *new_root,
8595 struct btrfs_root *parent_root,
8598 struct inode *inode;
8602 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8603 new_dirid, new_dirid,
8604 S_IFDIR | (~current_umask() & S_IRWXUGO),
8607 return PTR_ERR(inode);
8608 inode->i_op = &btrfs_dir_inode_operations;
8609 inode->i_fop = &btrfs_dir_file_operations;
8611 set_nlink(inode, 1);
8612 btrfs_i_size_write(BTRFS_I(inode), 0);
8613 unlock_new_inode(inode);
8615 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8617 btrfs_err(new_root->fs_info,
8618 "error inheriting subvolume %llu properties: %d",
8619 new_root->root_key.objectid, err);
8621 err = btrfs_update_inode(trans, new_root, inode);
8627 struct inode *btrfs_alloc_inode(struct super_block *sb)
8629 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8630 struct btrfs_inode *ei;
8631 struct inode *inode;
8633 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8640 ei->last_sub_trans = 0;
8641 ei->logged_trans = 0;
8642 ei->delalloc_bytes = 0;
8643 ei->new_delalloc_bytes = 0;
8644 ei->defrag_bytes = 0;
8645 ei->disk_i_size = 0;
8648 ei->index_cnt = (u64)-1;
8650 ei->last_unlink_trans = 0;
8651 ei->last_reflink_trans = 0;
8652 ei->last_log_commit = 0;
8654 spin_lock_init(&ei->lock);
8655 ei->outstanding_extents = 0;
8656 if (sb->s_magic != BTRFS_TEST_MAGIC)
8657 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8658 BTRFS_BLOCK_RSV_DELALLOC);
8659 ei->runtime_flags = 0;
8660 ei->prop_compress = BTRFS_COMPRESS_NONE;
8661 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8663 ei->delayed_node = NULL;
8665 ei->i_otime.tv_sec = 0;
8666 ei->i_otime.tv_nsec = 0;
8668 inode = &ei->vfs_inode;
8669 extent_map_tree_init(&ei->extent_tree);
8670 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8671 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8672 IO_TREE_INODE_IO_FAILURE, inode);
8673 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8674 IO_TREE_INODE_FILE_EXTENT, inode);
8675 ei->io_tree.track_uptodate = true;
8676 ei->io_failure_tree.track_uptodate = true;
8677 atomic_set(&ei->sync_writers, 0);
8678 mutex_init(&ei->log_mutex);
8679 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8680 INIT_LIST_HEAD(&ei->delalloc_inodes);
8681 INIT_LIST_HEAD(&ei->delayed_iput);
8682 RB_CLEAR_NODE(&ei->rb_node);
8683 init_rwsem(&ei->dio_sem);
8688 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8689 void btrfs_test_destroy_inode(struct inode *inode)
8691 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8692 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8696 void btrfs_free_inode(struct inode *inode)
8698 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8701 void btrfs_destroy_inode(struct inode *vfs_inode)
8703 struct btrfs_ordered_extent *ordered;
8704 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8705 struct btrfs_root *root = inode->root;
8707 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8708 WARN_ON(vfs_inode->i_data.nrpages);
8709 WARN_ON(inode->block_rsv.reserved);
8710 WARN_ON(inode->block_rsv.size);
8711 WARN_ON(inode->outstanding_extents);
8712 WARN_ON(inode->delalloc_bytes);
8713 WARN_ON(inode->new_delalloc_bytes);
8714 WARN_ON(inode->csum_bytes);
8715 WARN_ON(inode->defrag_bytes);
8718 * This can happen where we create an inode, but somebody else also
8719 * created the same inode and we need to destroy the one we already
8726 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8730 btrfs_err(root->fs_info,
8731 "found ordered extent %llu %llu on inode cleanup",
8732 ordered->file_offset, ordered->num_bytes);
8733 btrfs_remove_ordered_extent(inode, ordered);
8734 btrfs_put_ordered_extent(ordered);
8735 btrfs_put_ordered_extent(ordered);
8738 btrfs_qgroup_check_reserved_leak(inode);
8739 inode_tree_del(inode);
8740 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8741 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8742 btrfs_put_root(inode->root);
8745 int btrfs_drop_inode(struct inode *inode)
8747 struct btrfs_root *root = BTRFS_I(inode)->root;
8752 /* the snap/subvol tree is on deleting */
8753 if (btrfs_root_refs(&root->root_item) == 0)
8756 return generic_drop_inode(inode);
8759 static void init_once(void *foo)
8761 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8763 inode_init_once(&ei->vfs_inode);
8766 void __cold btrfs_destroy_cachep(void)
8769 * Make sure all delayed rcu free inodes are flushed before we
8773 kmem_cache_destroy(btrfs_inode_cachep);
8774 kmem_cache_destroy(btrfs_trans_handle_cachep);
8775 kmem_cache_destroy(btrfs_path_cachep);
8776 kmem_cache_destroy(btrfs_free_space_cachep);
8777 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8780 int __init btrfs_init_cachep(void)
8782 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8783 sizeof(struct btrfs_inode), 0,
8784 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8786 if (!btrfs_inode_cachep)
8789 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8790 sizeof(struct btrfs_trans_handle), 0,
8791 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8792 if (!btrfs_trans_handle_cachep)
8795 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8796 sizeof(struct btrfs_path), 0,
8797 SLAB_MEM_SPREAD, NULL);
8798 if (!btrfs_path_cachep)
8801 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8802 sizeof(struct btrfs_free_space), 0,
8803 SLAB_MEM_SPREAD, NULL);
8804 if (!btrfs_free_space_cachep)
8807 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8808 PAGE_SIZE, PAGE_SIZE,
8809 SLAB_RED_ZONE, NULL);
8810 if (!btrfs_free_space_bitmap_cachep)
8815 btrfs_destroy_cachep();
8819 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8820 u32 request_mask, unsigned int flags)
8823 struct inode *inode = d_inode(path->dentry);
8824 u32 blocksize = inode->i_sb->s_blocksize;
8825 u32 bi_flags = BTRFS_I(inode)->flags;
8827 stat->result_mask |= STATX_BTIME;
8828 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8829 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8830 if (bi_flags & BTRFS_INODE_APPEND)
8831 stat->attributes |= STATX_ATTR_APPEND;
8832 if (bi_flags & BTRFS_INODE_COMPRESS)
8833 stat->attributes |= STATX_ATTR_COMPRESSED;
8834 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8835 stat->attributes |= STATX_ATTR_IMMUTABLE;
8836 if (bi_flags & BTRFS_INODE_NODUMP)
8837 stat->attributes |= STATX_ATTR_NODUMP;
8839 stat->attributes_mask |= (STATX_ATTR_APPEND |
8840 STATX_ATTR_COMPRESSED |
8841 STATX_ATTR_IMMUTABLE |
8844 generic_fillattr(inode, stat);
8845 stat->dev = BTRFS_I(inode)->root->anon_dev;
8847 spin_lock(&BTRFS_I(inode)->lock);
8848 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8849 spin_unlock(&BTRFS_I(inode)->lock);
8850 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8851 ALIGN(delalloc_bytes, blocksize)) >> 9;
8855 static int btrfs_rename_exchange(struct inode *old_dir,
8856 struct dentry *old_dentry,
8857 struct inode *new_dir,
8858 struct dentry *new_dentry)
8860 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8861 struct btrfs_trans_handle *trans;
8862 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8863 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8864 struct inode *new_inode = new_dentry->d_inode;
8865 struct inode *old_inode = old_dentry->d_inode;
8866 struct timespec64 ctime = current_time(old_inode);
8867 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8868 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8873 bool root_log_pinned = false;
8874 bool dest_log_pinned = false;
8876 /* we only allow rename subvolume link between subvolumes */
8877 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8880 /* close the race window with snapshot create/destroy ioctl */
8881 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8882 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8883 down_read(&fs_info->subvol_sem);
8886 * We want to reserve the absolute worst case amount of items. So if
8887 * both inodes are subvols and we need to unlink them then that would
8888 * require 4 item modifications, but if they are both normal inodes it
8889 * would require 5 item modifications, so we'll assume their normal
8890 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8891 * should cover the worst case number of items we'll modify.
8893 trans = btrfs_start_transaction(root, 12);
8894 if (IS_ERR(trans)) {
8895 ret = PTR_ERR(trans);
8900 btrfs_record_root_in_trans(trans, dest);
8903 * We need to find a free sequence number both in the source and
8904 * in the destination directory for the exchange.
8906 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8909 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8913 BTRFS_I(old_inode)->dir_index = 0ULL;
8914 BTRFS_I(new_inode)->dir_index = 0ULL;
8916 /* Reference for the source. */
8917 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8918 /* force full log commit if subvolume involved. */
8919 btrfs_set_log_full_commit(trans);
8921 btrfs_pin_log_trans(root);
8922 root_log_pinned = true;
8923 ret = btrfs_insert_inode_ref(trans, dest,
8924 new_dentry->d_name.name,
8925 new_dentry->d_name.len,
8927 btrfs_ino(BTRFS_I(new_dir)),
8933 /* And now for the dest. */
8934 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8935 /* force full log commit if subvolume involved. */
8936 btrfs_set_log_full_commit(trans);
8938 btrfs_pin_log_trans(dest);
8939 dest_log_pinned = true;
8940 ret = btrfs_insert_inode_ref(trans, root,
8941 old_dentry->d_name.name,
8942 old_dentry->d_name.len,
8944 btrfs_ino(BTRFS_I(old_dir)),
8950 /* Update inode version and ctime/mtime. */
8951 inode_inc_iversion(old_dir);
8952 inode_inc_iversion(new_dir);
8953 inode_inc_iversion(old_inode);
8954 inode_inc_iversion(new_inode);
8955 old_dir->i_ctime = old_dir->i_mtime = ctime;
8956 new_dir->i_ctime = new_dir->i_mtime = ctime;
8957 old_inode->i_ctime = ctime;
8958 new_inode->i_ctime = ctime;
8960 if (old_dentry->d_parent != new_dentry->d_parent) {
8961 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8962 BTRFS_I(old_inode), 1);
8963 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8964 BTRFS_I(new_inode), 1);
8967 /* src is a subvolume */
8968 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8969 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
8970 } else { /* src is an inode */
8971 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
8972 BTRFS_I(old_dentry->d_inode),
8973 old_dentry->d_name.name,
8974 old_dentry->d_name.len);
8976 ret = btrfs_update_inode(trans, root, old_inode);
8979 btrfs_abort_transaction(trans, ret);
8983 /* dest is a subvolume */
8984 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8985 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
8986 } else { /* dest is an inode */
8987 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
8988 BTRFS_I(new_dentry->d_inode),
8989 new_dentry->d_name.name,
8990 new_dentry->d_name.len);
8992 ret = btrfs_update_inode(trans, dest, new_inode);
8995 btrfs_abort_transaction(trans, ret);
8999 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9000 new_dentry->d_name.name,
9001 new_dentry->d_name.len, 0, old_idx);
9003 btrfs_abort_transaction(trans, ret);
9007 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9008 old_dentry->d_name.name,
9009 old_dentry->d_name.len, 0, new_idx);
9011 btrfs_abort_transaction(trans, ret);
9015 if (old_inode->i_nlink == 1)
9016 BTRFS_I(old_inode)->dir_index = old_idx;
9017 if (new_inode->i_nlink == 1)
9018 BTRFS_I(new_inode)->dir_index = new_idx;
9020 if (root_log_pinned) {
9021 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9022 new_dentry->d_parent);
9023 btrfs_end_log_trans(root);
9024 root_log_pinned = false;
9026 if (dest_log_pinned) {
9027 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9028 old_dentry->d_parent);
9029 btrfs_end_log_trans(dest);
9030 dest_log_pinned = false;
9034 * If we have pinned a log and an error happened, we unpin tasks
9035 * trying to sync the log and force them to fallback to a transaction
9036 * commit if the log currently contains any of the inodes involved in
9037 * this rename operation (to ensure we do not persist a log with an
9038 * inconsistent state for any of these inodes or leading to any
9039 * inconsistencies when replayed). If the transaction was aborted, the
9040 * abortion reason is propagated to userspace when attempting to commit
9041 * the transaction. If the log does not contain any of these inodes, we
9042 * allow the tasks to sync it.
9044 if (ret && (root_log_pinned || dest_log_pinned)) {
9045 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9046 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9047 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9049 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9050 btrfs_set_log_full_commit(trans);
9052 if (root_log_pinned) {
9053 btrfs_end_log_trans(root);
9054 root_log_pinned = false;
9056 if (dest_log_pinned) {
9057 btrfs_end_log_trans(dest);
9058 dest_log_pinned = false;
9061 ret2 = btrfs_end_transaction(trans);
9062 ret = ret ? ret : ret2;
9064 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9065 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9066 up_read(&fs_info->subvol_sem);
9071 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9072 struct btrfs_root *root,
9074 struct dentry *dentry)
9077 struct inode *inode;
9081 ret = btrfs_find_free_ino(root, &objectid);
9085 inode = btrfs_new_inode(trans, root, dir,
9086 dentry->d_name.name,
9088 btrfs_ino(BTRFS_I(dir)),
9090 S_IFCHR | WHITEOUT_MODE,
9093 if (IS_ERR(inode)) {
9094 ret = PTR_ERR(inode);
9098 inode->i_op = &btrfs_special_inode_operations;
9099 init_special_inode(inode, inode->i_mode,
9102 ret = btrfs_init_inode_security(trans, inode, dir,
9107 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9108 BTRFS_I(inode), 0, index);
9112 ret = btrfs_update_inode(trans, root, inode);
9114 unlock_new_inode(inode);
9116 inode_dec_link_count(inode);
9122 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9123 struct inode *new_dir, struct dentry *new_dentry,
9126 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9127 struct btrfs_trans_handle *trans;
9128 unsigned int trans_num_items;
9129 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9130 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9131 struct inode *new_inode = d_inode(new_dentry);
9132 struct inode *old_inode = d_inode(old_dentry);
9136 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9137 bool log_pinned = false;
9139 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9142 /* we only allow rename subvolume link between subvolumes */
9143 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9146 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9147 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9150 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9151 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9155 /* check for collisions, even if the name isn't there */
9156 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9157 new_dentry->d_name.name,
9158 new_dentry->d_name.len);
9161 if (ret == -EEXIST) {
9163 * eexist without a new_inode */
9164 if (WARN_ON(!new_inode)) {
9168 /* maybe -EOVERFLOW */
9175 * we're using rename to replace one file with another. Start IO on it
9176 * now so we don't add too much work to the end of the transaction
9178 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9179 filemap_flush(old_inode->i_mapping);
9181 /* close the racy window with snapshot create/destroy ioctl */
9182 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9183 down_read(&fs_info->subvol_sem);
9185 * We want to reserve the absolute worst case amount of items. So if
9186 * both inodes are subvols and we need to unlink them then that would
9187 * require 4 item modifications, but if they are both normal inodes it
9188 * would require 5 item modifications, so we'll assume they are normal
9189 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9190 * should cover the worst case number of items we'll modify.
9191 * If our rename has the whiteout flag, we need more 5 units for the
9192 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9193 * when selinux is enabled).
9195 trans_num_items = 11;
9196 if (flags & RENAME_WHITEOUT)
9197 trans_num_items += 5;
9198 trans = btrfs_start_transaction(root, trans_num_items);
9199 if (IS_ERR(trans)) {
9200 ret = PTR_ERR(trans);
9205 btrfs_record_root_in_trans(trans, dest);
9207 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9211 BTRFS_I(old_inode)->dir_index = 0ULL;
9212 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9213 /* force full log commit if subvolume involved. */
9214 btrfs_set_log_full_commit(trans);
9216 btrfs_pin_log_trans(root);
9218 ret = btrfs_insert_inode_ref(trans, dest,
9219 new_dentry->d_name.name,
9220 new_dentry->d_name.len,
9222 btrfs_ino(BTRFS_I(new_dir)), index);
9227 inode_inc_iversion(old_dir);
9228 inode_inc_iversion(new_dir);
9229 inode_inc_iversion(old_inode);
9230 old_dir->i_ctime = old_dir->i_mtime =
9231 new_dir->i_ctime = new_dir->i_mtime =
9232 old_inode->i_ctime = current_time(old_dir);
9234 if (old_dentry->d_parent != new_dentry->d_parent)
9235 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9236 BTRFS_I(old_inode), 1);
9238 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9239 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9241 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9242 BTRFS_I(d_inode(old_dentry)),
9243 old_dentry->d_name.name,
9244 old_dentry->d_name.len);
9246 ret = btrfs_update_inode(trans, root, old_inode);
9249 btrfs_abort_transaction(trans, ret);
9254 inode_inc_iversion(new_inode);
9255 new_inode->i_ctime = current_time(new_inode);
9256 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9257 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9258 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9259 BUG_ON(new_inode->i_nlink == 0);
9261 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9262 BTRFS_I(d_inode(new_dentry)),
9263 new_dentry->d_name.name,
9264 new_dentry->d_name.len);
9266 if (!ret && new_inode->i_nlink == 0)
9267 ret = btrfs_orphan_add(trans,
9268 BTRFS_I(d_inode(new_dentry)));
9270 btrfs_abort_transaction(trans, ret);
9275 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9276 new_dentry->d_name.name,
9277 new_dentry->d_name.len, 0, index);
9279 btrfs_abort_transaction(trans, ret);
9283 if (old_inode->i_nlink == 1)
9284 BTRFS_I(old_inode)->dir_index = index;
9287 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9288 new_dentry->d_parent);
9289 btrfs_end_log_trans(root);
9293 if (flags & RENAME_WHITEOUT) {
9294 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9298 btrfs_abort_transaction(trans, ret);
9304 * If we have pinned the log and an error happened, we unpin tasks
9305 * trying to sync the log and force them to fallback to a transaction
9306 * commit if the log currently contains any of the inodes involved in
9307 * this rename operation (to ensure we do not persist a log with an
9308 * inconsistent state for any of these inodes or leading to any
9309 * inconsistencies when replayed). If the transaction was aborted, the
9310 * abortion reason is propagated to userspace when attempting to commit
9311 * the transaction. If the log does not contain any of these inodes, we
9312 * allow the tasks to sync it.
9314 if (ret && log_pinned) {
9315 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9316 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9317 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9319 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9320 btrfs_set_log_full_commit(trans);
9322 btrfs_end_log_trans(root);
9325 ret2 = btrfs_end_transaction(trans);
9326 ret = ret ? ret : ret2;
9328 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9329 up_read(&fs_info->subvol_sem);
9334 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9335 struct inode *new_dir, struct dentry *new_dentry,
9338 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9341 if (flags & RENAME_EXCHANGE)
9342 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9345 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9348 struct btrfs_delalloc_work {
9349 struct inode *inode;
9350 struct completion completion;
9351 struct list_head list;
9352 struct btrfs_work work;
9355 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9357 struct btrfs_delalloc_work *delalloc_work;
9358 struct inode *inode;
9360 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9362 inode = delalloc_work->inode;
9363 filemap_flush(inode->i_mapping);
9364 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9365 &BTRFS_I(inode)->runtime_flags))
9366 filemap_flush(inode->i_mapping);
9369 complete(&delalloc_work->completion);
9372 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9374 struct btrfs_delalloc_work *work;
9376 work = kmalloc(sizeof(*work), GFP_NOFS);
9380 init_completion(&work->completion);
9381 INIT_LIST_HEAD(&work->list);
9382 work->inode = inode;
9383 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9389 * some fairly slow code that needs optimization. This walks the list
9390 * of all the inodes with pending delalloc and forces them to disk.
9392 static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot)
9394 struct btrfs_inode *binode;
9395 struct inode *inode;
9396 struct btrfs_delalloc_work *work, *next;
9397 struct list_head works;
9398 struct list_head splice;
9401 INIT_LIST_HEAD(&works);
9402 INIT_LIST_HEAD(&splice);
9404 mutex_lock(&root->delalloc_mutex);
9405 spin_lock(&root->delalloc_lock);
9406 list_splice_init(&root->delalloc_inodes, &splice);
9407 while (!list_empty(&splice)) {
9408 binode = list_entry(splice.next, struct btrfs_inode,
9411 list_move_tail(&binode->delalloc_inodes,
9412 &root->delalloc_inodes);
9413 inode = igrab(&binode->vfs_inode);
9415 cond_resched_lock(&root->delalloc_lock);
9418 spin_unlock(&root->delalloc_lock);
9421 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9422 &binode->runtime_flags);
9423 work = btrfs_alloc_delalloc_work(inode);
9429 list_add_tail(&work->list, &works);
9430 btrfs_queue_work(root->fs_info->flush_workers,
9432 if (*nr != U64_MAX) {
9438 spin_lock(&root->delalloc_lock);
9440 spin_unlock(&root->delalloc_lock);
9443 list_for_each_entry_safe(work, next, &works, list) {
9444 list_del_init(&work->list);
9445 wait_for_completion(&work->completion);
9449 if (!list_empty(&splice)) {
9450 spin_lock(&root->delalloc_lock);
9451 list_splice_tail(&splice, &root->delalloc_inodes);
9452 spin_unlock(&root->delalloc_lock);
9454 mutex_unlock(&root->delalloc_mutex);
9458 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9460 struct btrfs_fs_info *fs_info = root->fs_info;
9463 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9466 return start_delalloc_inodes(root, &nr, true);
9469 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr)
9471 struct btrfs_root *root;
9472 struct list_head splice;
9475 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9478 INIT_LIST_HEAD(&splice);
9480 mutex_lock(&fs_info->delalloc_root_mutex);
9481 spin_lock(&fs_info->delalloc_root_lock);
9482 list_splice_init(&fs_info->delalloc_roots, &splice);
9483 while (!list_empty(&splice) && nr) {
9484 root = list_first_entry(&splice, struct btrfs_root,
9486 root = btrfs_grab_root(root);
9488 list_move_tail(&root->delalloc_root,
9489 &fs_info->delalloc_roots);
9490 spin_unlock(&fs_info->delalloc_root_lock);
9492 ret = start_delalloc_inodes(root, &nr, false);
9493 btrfs_put_root(root);
9496 spin_lock(&fs_info->delalloc_root_lock);
9498 spin_unlock(&fs_info->delalloc_root_lock);
9502 if (!list_empty(&splice)) {
9503 spin_lock(&fs_info->delalloc_root_lock);
9504 list_splice_tail(&splice, &fs_info->delalloc_roots);
9505 spin_unlock(&fs_info->delalloc_root_lock);
9507 mutex_unlock(&fs_info->delalloc_root_mutex);
9511 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9512 const char *symname)
9514 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9515 struct btrfs_trans_handle *trans;
9516 struct btrfs_root *root = BTRFS_I(dir)->root;
9517 struct btrfs_path *path;
9518 struct btrfs_key key;
9519 struct inode *inode = NULL;
9526 struct btrfs_file_extent_item *ei;
9527 struct extent_buffer *leaf;
9529 name_len = strlen(symname);
9530 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9531 return -ENAMETOOLONG;
9534 * 2 items for inode item and ref
9535 * 2 items for dir items
9536 * 1 item for updating parent inode item
9537 * 1 item for the inline extent item
9538 * 1 item for xattr if selinux is on
9540 trans = btrfs_start_transaction(root, 7);
9542 return PTR_ERR(trans);
9544 err = btrfs_find_free_ino(root, &objectid);
9548 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9549 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9550 objectid, S_IFLNK|S_IRWXUGO, &index);
9551 if (IS_ERR(inode)) {
9552 err = PTR_ERR(inode);
9558 * If the active LSM wants to access the inode during
9559 * d_instantiate it needs these. Smack checks to see
9560 * if the filesystem supports xattrs by looking at the
9563 inode->i_fop = &btrfs_file_operations;
9564 inode->i_op = &btrfs_file_inode_operations;
9565 inode->i_mapping->a_ops = &btrfs_aops;
9567 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9571 path = btrfs_alloc_path();
9576 key.objectid = btrfs_ino(BTRFS_I(inode));
9578 key.type = BTRFS_EXTENT_DATA_KEY;
9579 datasize = btrfs_file_extent_calc_inline_size(name_len);
9580 err = btrfs_insert_empty_item(trans, root, path, &key,
9583 btrfs_free_path(path);
9586 leaf = path->nodes[0];
9587 ei = btrfs_item_ptr(leaf, path->slots[0],
9588 struct btrfs_file_extent_item);
9589 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9590 btrfs_set_file_extent_type(leaf, ei,
9591 BTRFS_FILE_EXTENT_INLINE);
9592 btrfs_set_file_extent_encryption(leaf, ei, 0);
9593 btrfs_set_file_extent_compression(leaf, ei, 0);
9594 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9595 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9597 ptr = btrfs_file_extent_inline_start(ei);
9598 write_extent_buffer(leaf, symname, ptr, name_len);
9599 btrfs_mark_buffer_dirty(leaf);
9600 btrfs_free_path(path);
9602 inode->i_op = &btrfs_symlink_inode_operations;
9603 inode_nohighmem(inode);
9604 inode_set_bytes(inode, name_len);
9605 btrfs_i_size_write(BTRFS_I(inode), name_len);
9606 err = btrfs_update_inode(trans, root, inode);
9608 * Last step, add directory indexes for our symlink inode. This is the
9609 * last step to avoid extra cleanup of these indexes if an error happens
9613 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9614 BTRFS_I(inode), 0, index);
9618 d_instantiate_new(dentry, inode);
9621 btrfs_end_transaction(trans);
9623 inode_dec_link_count(inode);
9624 discard_new_inode(inode);
9626 btrfs_btree_balance_dirty(fs_info);
9630 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9631 struct btrfs_trans_handle *trans_in,
9632 struct inode *inode, struct btrfs_key *ins,
9635 struct btrfs_file_extent_item stack_fi;
9636 struct btrfs_replace_extent_info extent_info;
9637 struct btrfs_trans_handle *trans = trans_in;
9638 struct btrfs_path *path;
9639 u64 start = ins->objectid;
9640 u64 len = ins->offset;
9643 memset(&stack_fi, 0, sizeof(stack_fi));
9645 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9646 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9647 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9648 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9649 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9650 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9651 /* Encryption and other encoding is reserved and all 0 */
9653 ret = btrfs_qgroup_release_data(BTRFS_I(inode), file_offset, len);
9655 return ERR_PTR(ret);
9658 ret = insert_reserved_file_extent(trans, BTRFS_I(inode),
9659 file_offset, &stack_fi, ret);
9661 return ERR_PTR(ret);
9665 extent_info.disk_offset = start;
9666 extent_info.disk_len = len;
9667 extent_info.data_offset = 0;
9668 extent_info.data_len = len;
9669 extent_info.file_offset = file_offset;
9670 extent_info.extent_buf = (char *)&stack_fi;
9671 extent_info.is_new_extent = true;
9672 extent_info.qgroup_reserved = ret;
9673 extent_info.insertions = 0;
9675 path = btrfs_alloc_path();
9677 return ERR_PTR(-ENOMEM);
9679 ret = btrfs_replace_file_extents(inode, path, file_offset,
9680 file_offset + len - 1, &extent_info,
9682 btrfs_free_path(path);
9684 return ERR_PTR(ret);
9689 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9690 u64 start, u64 num_bytes, u64 min_size,
9691 loff_t actual_len, u64 *alloc_hint,
9692 struct btrfs_trans_handle *trans)
9694 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9695 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9696 struct extent_map *em;
9697 struct btrfs_root *root = BTRFS_I(inode)->root;
9698 struct btrfs_key ins;
9699 u64 cur_offset = start;
9700 u64 clear_offset = start;
9703 u64 last_alloc = (u64)-1;
9705 bool own_trans = true;
9706 u64 end = start + num_bytes - 1;
9710 while (num_bytes > 0) {
9711 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9712 cur_bytes = max(cur_bytes, min_size);
9714 * If we are severely fragmented we could end up with really
9715 * small allocations, so if the allocator is returning small
9716 * chunks lets make its job easier by only searching for those
9719 cur_bytes = min(cur_bytes, last_alloc);
9720 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9721 min_size, 0, *alloc_hint, &ins, 1, 0);
9726 * We've reserved this space, and thus converted it from
9727 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9728 * from here on out we will only need to clear our reservation
9729 * for the remaining unreserved area, so advance our
9730 * clear_offset by our extent size.
9732 clear_offset += ins.offset;
9734 last_alloc = ins.offset;
9735 trans = insert_prealloc_file_extent(trans, inode, &ins, cur_offset);
9737 * Now that we inserted the prealloc extent we can finally
9738 * decrement the number of reservations in the block group.
9739 * If we did it before, we could race with relocation and have
9740 * relocation miss the reserved extent, making it fail later.
9742 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9743 if (IS_ERR(trans)) {
9744 ret = PTR_ERR(trans);
9745 btrfs_free_reserved_extent(fs_info, ins.objectid,
9750 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9751 cur_offset + ins.offset -1, 0);
9753 em = alloc_extent_map();
9755 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9756 &BTRFS_I(inode)->runtime_flags);
9760 em->start = cur_offset;
9761 em->orig_start = cur_offset;
9762 em->len = ins.offset;
9763 em->block_start = ins.objectid;
9764 em->block_len = ins.offset;
9765 em->orig_block_len = ins.offset;
9766 em->ram_bytes = ins.offset;
9767 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9768 em->generation = trans->transid;
9771 write_lock(&em_tree->lock);
9772 ret = add_extent_mapping(em_tree, em, 1);
9773 write_unlock(&em_tree->lock);
9776 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9777 cur_offset + ins.offset - 1,
9780 free_extent_map(em);
9782 num_bytes -= ins.offset;
9783 cur_offset += ins.offset;
9784 *alloc_hint = ins.objectid + ins.offset;
9786 inode_inc_iversion(inode);
9787 inode->i_ctime = current_time(inode);
9788 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9789 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9790 (actual_len > inode->i_size) &&
9791 (cur_offset > inode->i_size)) {
9792 if (cur_offset > actual_len)
9793 i_size = actual_len;
9795 i_size = cur_offset;
9796 i_size_write(inode, i_size);
9797 btrfs_inode_safe_disk_i_size_write(inode, 0);
9800 ret = btrfs_update_inode(trans, root, inode);
9803 btrfs_abort_transaction(trans, ret);
9805 btrfs_end_transaction(trans);
9810 btrfs_end_transaction(trans);
9814 if (clear_offset < end)
9815 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9816 end - clear_offset + 1);
9820 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9821 u64 start, u64 num_bytes, u64 min_size,
9822 loff_t actual_len, u64 *alloc_hint)
9824 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9825 min_size, actual_len, alloc_hint,
9829 int btrfs_prealloc_file_range_trans(struct inode *inode,
9830 struct btrfs_trans_handle *trans, int mode,
9831 u64 start, u64 num_bytes, u64 min_size,
9832 loff_t actual_len, u64 *alloc_hint)
9834 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9835 min_size, actual_len, alloc_hint, trans);
9838 static int btrfs_set_page_dirty(struct page *page)
9840 return __set_page_dirty_nobuffers(page);
9843 static int btrfs_permission(struct inode *inode, int mask)
9845 struct btrfs_root *root = BTRFS_I(inode)->root;
9846 umode_t mode = inode->i_mode;
9848 if (mask & MAY_WRITE &&
9849 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9850 if (btrfs_root_readonly(root))
9852 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9855 return generic_permission(inode, mask);
9858 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9860 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9861 struct btrfs_trans_handle *trans;
9862 struct btrfs_root *root = BTRFS_I(dir)->root;
9863 struct inode *inode = NULL;
9869 * 5 units required for adding orphan entry
9871 trans = btrfs_start_transaction(root, 5);
9873 return PTR_ERR(trans);
9875 ret = btrfs_find_free_ino(root, &objectid);
9879 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9880 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9881 if (IS_ERR(inode)) {
9882 ret = PTR_ERR(inode);
9887 inode->i_fop = &btrfs_file_operations;
9888 inode->i_op = &btrfs_file_inode_operations;
9890 inode->i_mapping->a_ops = &btrfs_aops;
9892 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9896 ret = btrfs_update_inode(trans, root, inode);
9899 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9904 * We set number of links to 0 in btrfs_new_inode(), and here we set
9905 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9908 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9910 set_nlink(inode, 1);
9911 d_tmpfile(dentry, inode);
9912 unlock_new_inode(inode);
9913 mark_inode_dirty(inode);
9915 btrfs_end_transaction(trans);
9917 discard_new_inode(inode);
9918 btrfs_btree_balance_dirty(fs_info);
9922 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9924 struct inode *inode = tree->private_data;
9925 unsigned long index = start >> PAGE_SHIFT;
9926 unsigned long end_index = end >> PAGE_SHIFT;
9929 while (index <= end_index) {
9930 page = find_get_page(inode->i_mapping, index);
9931 ASSERT(page); /* Pages should be in the extent_io_tree */
9932 set_page_writeback(page);
9940 * Add an entry indicating a block group or device which is pinned by a
9941 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9942 * negative errno on failure.
9944 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
9945 bool is_block_group)
9947 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9948 struct btrfs_swapfile_pin *sp, *entry;
9950 struct rb_node *parent = NULL;
9952 sp = kmalloc(sizeof(*sp), GFP_NOFS);
9957 sp->is_block_group = is_block_group;
9959 spin_lock(&fs_info->swapfile_pins_lock);
9960 p = &fs_info->swapfile_pins.rb_node;
9963 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
9964 if (sp->ptr < entry->ptr ||
9965 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
9967 } else if (sp->ptr > entry->ptr ||
9968 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
9969 p = &(*p)->rb_right;
9971 spin_unlock(&fs_info->swapfile_pins_lock);
9976 rb_link_node(&sp->node, parent, p);
9977 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
9978 spin_unlock(&fs_info->swapfile_pins_lock);
9982 /* Free all of the entries pinned by this swapfile. */
9983 static void btrfs_free_swapfile_pins(struct inode *inode)
9985 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9986 struct btrfs_swapfile_pin *sp;
9987 struct rb_node *node, *next;
9989 spin_lock(&fs_info->swapfile_pins_lock);
9990 node = rb_first(&fs_info->swapfile_pins);
9992 next = rb_next(node);
9993 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
9994 if (sp->inode == inode) {
9995 rb_erase(&sp->node, &fs_info->swapfile_pins);
9996 if (sp->is_block_group)
9997 btrfs_put_block_group(sp->ptr);
10002 spin_unlock(&fs_info->swapfile_pins_lock);
10005 struct btrfs_swap_info {
10011 unsigned long nr_pages;
10015 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10016 struct btrfs_swap_info *bsi)
10018 unsigned long nr_pages;
10019 u64 first_ppage, first_ppage_reported, next_ppage;
10022 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10023 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10024 PAGE_SIZE) >> PAGE_SHIFT;
10026 if (first_ppage >= next_ppage)
10028 nr_pages = next_ppage - first_ppage;
10030 first_ppage_reported = first_ppage;
10031 if (bsi->start == 0)
10032 first_ppage_reported++;
10033 if (bsi->lowest_ppage > first_ppage_reported)
10034 bsi->lowest_ppage = first_ppage_reported;
10035 if (bsi->highest_ppage < (next_ppage - 1))
10036 bsi->highest_ppage = next_ppage - 1;
10038 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10041 bsi->nr_extents += ret;
10042 bsi->nr_pages += nr_pages;
10046 static void btrfs_swap_deactivate(struct file *file)
10048 struct inode *inode = file_inode(file);
10050 btrfs_free_swapfile_pins(inode);
10051 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10054 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10057 struct inode *inode = file_inode(file);
10058 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10059 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10060 struct extent_state *cached_state = NULL;
10061 struct extent_map *em = NULL;
10062 struct btrfs_device *device = NULL;
10063 struct btrfs_swap_info bsi = {
10064 .lowest_ppage = (sector_t)-1ULL,
10071 * If the swap file was just created, make sure delalloc is done. If the
10072 * file changes again after this, the user is doing something stupid and
10073 * we don't really care.
10075 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10080 * The inode is locked, so these flags won't change after we check them.
10082 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10083 btrfs_warn(fs_info, "swapfile must not be compressed");
10086 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10087 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10090 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10091 btrfs_warn(fs_info, "swapfile must not be checksummed");
10096 * Balance or device remove/replace/resize can move stuff around from
10097 * under us. The exclop protection makes sure they aren't running/won't
10098 * run concurrently while we are mapping the swap extents, and
10099 * fs_info->swapfile_pins prevents them from running while the swap
10100 * file is active and moving the extents. Note that this also prevents
10101 * a concurrent device add which isn't actually necessary, but it's not
10102 * really worth the trouble to allow it.
10104 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10105 btrfs_warn(fs_info,
10106 "cannot activate swapfile while exclusive operation is running");
10110 * Snapshots can create extents which require COW even if NODATACOW is
10111 * set. We use this counter to prevent snapshots. We must increment it
10112 * before walking the extents because we don't want a concurrent
10113 * snapshot to run after we've already checked the extents.
10115 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10117 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10119 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10121 while (start < isize) {
10122 u64 logical_block_start, physical_block_start;
10123 struct btrfs_block_group *bg;
10124 u64 len = isize - start;
10126 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10132 if (em->block_start == EXTENT_MAP_HOLE) {
10133 btrfs_warn(fs_info, "swapfile must not have holes");
10137 if (em->block_start == EXTENT_MAP_INLINE) {
10139 * It's unlikely we'll ever actually find ourselves
10140 * here, as a file small enough to fit inline won't be
10141 * big enough to store more than the swap header, but in
10142 * case something changes in the future, let's catch it
10143 * here rather than later.
10145 btrfs_warn(fs_info, "swapfile must not be inline");
10149 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10150 btrfs_warn(fs_info, "swapfile must not be compressed");
10155 logical_block_start = em->block_start + (start - em->start);
10156 len = min(len, em->len - (start - em->start));
10157 free_extent_map(em);
10160 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10166 btrfs_warn(fs_info,
10167 "swapfile must not be copy-on-write");
10172 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10178 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10179 btrfs_warn(fs_info,
10180 "swapfile must have single data profile");
10185 if (device == NULL) {
10186 device = em->map_lookup->stripes[0].dev;
10187 ret = btrfs_add_swapfile_pin(inode, device, false);
10192 } else if (device != em->map_lookup->stripes[0].dev) {
10193 btrfs_warn(fs_info, "swapfile must be on one device");
10198 physical_block_start = (em->map_lookup->stripes[0].physical +
10199 (logical_block_start - em->start));
10200 len = min(len, em->len - (logical_block_start - em->start));
10201 free_extent_map(em);
10204 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10206 btrfs_warn(fs_info,
10207 "could not find block group containing swapfile");
10212 ret = btrfs_add_swapfile_pin(inode, bg, true);
10214 btrfs_put_block_group(bg);
10221 if (bsi.block_len &&
10222 bsi.block_start + bsi.block_len == physical_block_start) {
10223 bsi.block_len += len;
10225 if (bsi.block_len) {
10226 ret = btrfs_add_swap_extent(sis, &bsi);
10231 bsi.block_start = physical_block_start;
10232 bsi.block_len = len;
10239 ret = btrfs_add_swap_extent(sis, &bsi);
10242 if (!IS_ERR_OR_NULL(em))
10243 free_extent_map(em);
10245 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10248 btrfs_swap_deactivate(file);
10250 btrfs_exclop_finish(fs_info);
10256 sis->bdev = device->bdev;
10257 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10258 sis->max = bsi.nr_pages;
10259 sis->pages = bsi.nr_pages - 1;
10260 sis->highest_bit = bsi.nr_pages - 1;
10261 return bsi.nr_extents;
10264 static void btrfs_swap_deactivate(struct file *file)
10268 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10271 return -EOPNOTSUPP;
10275 static const struct inode_operations btrfs_dir_inode_operations = {
10276 .getattr = btrfs_getattr,
10277 .lookup = btrfs_lookup,
10278 .create = btrfs_create,
10279 .unlink = btrfs_unlink,
10280 .link = btrfs_link,
10281 .mkdir = btrfs_mkdir,
10282 .rmdir = btrfs_rmdir,
10283 .rename = btrfs_rename2,
10284 .symlink = btrfs_symlink,
10285 .setattr = btrfs_setattr,
10286 .mknod = btrfs_mknod,
10287 .listxattr = btrfs_listxattr,
10288 .permission = btrfs_permission,
10289 .get_acl = btrfs_get_acl,
10290 .set_acl = btrfs_set_acl,
10291 .update_time = btrfs_update_time,
10292 .tmpfile = btrfs_tmpfile,
10295 static const struct file_operations btrfs_dir_file_operations = {
10296 .llseek = generic_file_llseek,
10297 .read = generic_read_dir,
10298 .iterate_shared = btrfs_real_readdir,
10299 .open = btrfs_opendir,
10300 .unlocked_ioctl = btrfs_ioctl,
10301 #ifdef CONFIG_COMPAT
10302 .compat_ioctl = btrfs_compat_ioctl,
10304 .release = btrfs_release_file,
10305 .fsync = btrfs_sync_file,
10309 * btrfs doesn't support the bmap operation because swapfiles
10310 * use bmap to make a mapping of extents in the file. They assume
10311 * these extents won't change over the life of the file and they
10312 * use the bmap result to do IO directly to the drive.
10314 * the btrfs bmap call would return logical addresses that aren't
10315 * suitable for IO and they also will change frequently as COW
10316 * operations happen. So, swapfile + btrfs == corruption.
10318 * For now we're avoiding this by dropping bmap.
10320 static const struct address_space_operations btrfs_aops = {
10321 .readpage = btrfs_readpage,
10322 .writepage = btrfs_writepage,
10323 .writepages = btrfs_writepages,
10324 .readahead = btrfs_readahead,
10325 .direct_IO = noop_direct_IO,
10326 .invalidatepage = btrfs_invalidatepage,
10327 .releasepage = btrfs_releasepage,
10328 #ifdef CONFIG_MIGRATION
10329 .migratepage = btrfs_migratepage,
10331 .set_page_dirty = btrfs_set_page_dirty,
10332 .error_remove_page = generic_error_remove_page,
10333 .swap_activate = btrfs_swap_activate,
10334 .swap_deactivate = btrfs_swap_deactivate,
10337 static const struct inode_operations btrfs_file_inode_operations = {
10338 .getattr = btrfs_getattr,
10339 .setattr = btrfs_setattr,
10340 .listxattr = btrfs_listxattr,
10341 .permission = btrfs_permission,
10342 .fiemap = btrfs_fiemap,
10343 .get_acl = btrfs_get_acl,
10344 .set_acl = btrfs_set_acl,
10345 .update_time = btrfs_update_time,
10347 static const struct inode_operations btrfs_special_inode_operations = {
10348 .getattr = btrfs_getattr,
10349 .setattr = btrfs_setattr,
10350 .permission = btrfs_permission,
10351 .listxattr = btrfs_listxattr,
10352 .get_acl = btrfs_get_acl,
10353 .set_acl = btrfs_set_acl,
10354 .update_time = btrfs_update_time,
10356 static const struct inode_operations btrfs_symlink_inode_operations = {
10357 .get_link = page_get_link,
10358 .getattr = btrfs_getattr,
10359 .setattr = btrfs_setattr,
10360 .permission = btrfs_permission,
10361 .listxattr = btrfs_listxattr,
10362 .update_time = btrfs_update_time,
10365 const struct dentry_operations btrfs_dentry_operations = {
10366 .d_delete = btrfs_dentry_delete,