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/blk-cgroup.h>
10 #include <linux/file.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/xattr.h>
21 #include <linux/posix_acl.h>
22 #include <linux/falloc.h>
23 #include <linux/slab.h>
24 #include <linux/ratelimit.h>
25 #include <linux/btrfs.h>
26 #include <linux/blkdev.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/uio.h>
29 #include <linux/magic.h>
30 #include <linux/iversion.h>
31 #include <linux/swap.h>
32 #include <linux/migrate.h>
33 #include <linux/sched/mm.h>
34 #include <linux/iomap.h>
35 #include <asm/unaligned.h>
36 #include <linux/fsverity.h>
40 #include "transaction.h"
41 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "ordered-data.h"
47 #include "compression.h"
49 #include "free-space-cache.h"
52 #include "delalloc-space.h"
53 #include "block-group.h"
54 #include "space-info.h"
57 #include "inode-item.h"
59 struct btrfs_iget_args {
61 struct btrfs_root *root;
64 struct btrfs_dio_data {
66 struct extent_changeset *data_reserved;
67 bool data_space_reserved;
71 struct btrfs_dio_private {
75 * Since DIO can use anonymous page, we cannot use page_offset() to
76 * grab the file offset, thus need a dedicated member for file offset.
79 /* Used for bio::bi_size */
83 * References to this structure. There is one reference per in-flight
84 * bio plus one while we're still setting up.
88 /* Array of checksums */
91 /* This must be last */
95 static struct bio_set btrfs_dio_bioset;
97 struct btrfs_rename_ctx {
98 /* Output field. Stores the index number of the old directory entry. */
102 static const struct inode_operations btrfs_dir_inode_operations;
103 static const struct inode_operations btrfs_symlink_inode_operations;
104 static const struct inode_operations btrfs_special_inode_operations;
105 static const struct inode_operations btrfs_file_inode_operations;
106 static const struct address_space_operations btrfs_aops;
107 static const struct file_operations btrfs_dir_file_operations;
109 static struct kmem_cache *btrfs_inode_cachep;
110 struct kmem_cache *btrfs_trans_handle_cachep;
111 struct kmem_cache *btrfs_path_cachep;
112 struct kmem_cache *btrfs_free_space_cachep;
113 struct kmem_cache *btrfs_free_space_bitmap_cachep;
115 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
116 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
117 static noinline int cow_file_range(struct btrfs_inode *inode,
118 struct page *locked_page,
119 u64 start, u64 end, int *page_started,
120 unsigned long *nr_written, int unlock,
122 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
123 u64 len, u64 orig_start, u64 block_start,
124 u64 block_len, u64 orig_block_len,
125 u64 ram_bytes, int compress_type,
129 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
131 * ilock_flags can have the following bit set:
133 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
134 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
136 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
138 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
140 if (ilock_flags & BTRFS_ILOCK_SHARED) {
141 if (ilock_flags & BTRFS_ILOCK_TRY) {
142 if (!inode_trylock_shared(inode))
147 inode_lock_shared(inode);
149 if (ilock_flags & BTRFS_ILOCK_TRY) {
150 if (!inode_trylock(inode))
157 if (ilock_flags & BTRFS_ILOCK_MMAP)
158 down_write(&BTRFS_I(inode)->i_mmap_lock);
163 * btrfs_inode_unlock - unock inode i_rwsem
165 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
166 * to decide whether the lock acquired is shared or exclusive.
168 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
170 if (ilock_flags & BTRFS_ILOCK_MMAP)
171 up_write(&BTRFS_I(inode)->i_mmap_lock);
172 if (ilock_flags & BTRFS_ILOCK_SHARED)
173 inode_unlock_shared(inode);
179 * Cleanup all submitted ordered extents in specified range to handle errors
180 * from the btrfs_run_delalloc_range() callback.
182 * NOTE: caller must ensure that when an error happens, it can not call
183 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
184 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
185 * to be released, which we want to happen only when finishing the ordered
186 * extent (btrfs_finish_ordered_io()).
188 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
189 struct page *locked_page,
190 u64 offset, u64 bytes)
192 unsigned long index = offset >> PAGE_SHIFT;
193 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
194 u64 page_start, page_end;
198 page_start = page_offset(locked_page);
199 page_end = page_start + PAGE_SIZE - 1;
202 while (index <= end_index) {
204 * For locked page, we will call end_extent_writepage() on it
205 * in run_delalloc_range() for the error handling. That
206 * end_extent_writepage() function will call
207 * btrfs_mark_ordered_io_finished() to clear page Ordered and
208 * run the ordered extent accounting.
210 * Here we can't just clear the Ordered bit, or
211 * btrfs_mark_ordered_io_finished() would skip the accounting
212 * for the page range, and the ordered extent will never finish.
214 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
218 page = find_get_page(inode->vfs_inode.i_mapping, index);
224 * Here we just clear all Ordered bits for every page in the
225 * range, then btrfs_mark_ordered_io_finished() will handle
226 * the ordered extent accounting for the range.
228 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
234 /* The locked page covers the full range, nothing needs to be done */
235 if (bytes + offset <= page_start + PAGE_SIZE)
238 * In case this page belongs to the delalloc range being
239 * instantiated then skip it, since the first page of a range is
240 * going to be properly cleaned up by the caller of
243 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
244 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
245 offset = page_offset(locked_page) + PAGE_SIZE;
249 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
252 static int btrfs_dirty_inode(struct inode *inode);
254 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
255 struct btrfs_new_inode_args *args)
259 if (args->default_acl) {
260 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
266 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
270 if (!args->default_acl && !args->acl)
271 cache_no_acl(args->inode);
272 return btrfs_xattr_security_init(trans, args->inode, args->dir,
273 &args->dentry->d_name);
277 * this does all the hard work for inserting an inline extent into
278 * the btree. The caller should have done a btrfs_drop_extents so that
279 * no overlapping inline items exist in the btree
281 static int insert_inline_extent(struct btrfs_trans_handle *trans,
282 struct btrfs_path *path,
283 struct btrfs_inode *inode, bool extent_inserted,
284 size_t size, size_t compressed_size,
286 struct page **compressed_pages,
289 struct btrfs_root *root = inode->root;
290 struct extent_buffer *leaf;
291 struct page *page = NULL;
294 struct btrfs_file_extent_item *ei;
296 size_t cur_size = size;
299 ASSERT((compressed_size > 0 && compressed_pages) ||
300 (compressed_size == 0 && !compressed_pages));
302 if (compressed_size && compressed_pages)
303 cur_size = compressed_size;
305 if (!extent_inserted) {
306 struct btrfs_key key;
309 key.objectid = btrfs_ino(inode);
311 key.type = BTRFS_EXTENT_DATA_KEY;
313 datasize = btrfs_file_extent_calc_inline_size(cur_size);
314 ret = btrfs_insert_empty_item(trans, root, path, &key,
319 leaf = path->nodes[0];
320 ei = btrfs_item_ptr(leaf, path->slots[0],
321 struct btrfs_file_extent_item);
322 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
323 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
324 btrfs_set_file_extent_encryption(leaf, ei, 0);
325 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
326 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
327 ptr = btrfs_file_extent_inline_start(ei);
329 if (compress_type != BTRFS_COMPRESS_NONE) {
332 while (compressed_size > 0) {
333 cpage = compressed_pages[i];
334 cur_size = min_t(unsigned long, compressed_size,
337 kaddr = kmap_local_page(cpage);
338 write_extent_buffer(leaf, kaddr, ptr, cur_size);
343 compressed_size -= cur_size;
345 btrfs_set_file_extent_compression(leaf, ei,
348 page = find_get_page(inode->vfs_inode.i_mapping, 0);
349 btrfs_set_file_extent_compression(leaf, ei, 0);
350 kaddr = kmap_local_page(page);
351 write_extent_buffer(leaf, kaddr, ptr, size);
355 btrfs_mark_buffer_dirty(leaf);
356 btrfs_release_path(path);
359 * We align size to sectorsize for inline extents just for simplicity
362 ret = btrfs_inode_set_file_extent_range(inode, 0,
363 ALIGN(size, root->fs_info->sectorsize));
368 * We're an inline extent, so nobody can extend the file past i_size
369 * without locking a page we already have locked.
371 * We must do any i_size and inode updates before we unlock the pages.
372 * Otherwise we could end up racing with unlink.
374 i_size = i_size_read(&inode->vfs_inode);
375 if (update_i_size && size > i_size) {
376 i_size_write(&inode->vfs_inode, size);
379 inode->disk_i_size = i_size;
387 * conditionally insert an inline extent into the file. This
388 * does the checks required to make sure the data is small enough
389 * to fit as an inline extent.
391 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
392 size_t compressed_size,
394 struct page **compressed_pages,
397 struct btrfs_drop_extents_args drop_args = { 0 };
398 struct btrfs_root *root = inode->root;
399 struct btrfs_fs_info *fs_info = root->fs_info;
400 struct btrfs_trans_handle *trans;
401 u64 data_len = (compressed_size ?: size);
403 struct btrfs_path *path;
406 * We can create an inline extent if it ends at or beyond the current
407 * i_size, is no larger than a sector (decompressed), and the (possibly
408 * compressed) data fits in a leaf and the configured maximum inline
411 if (size < i_size_read(&inode->vfs_inode) ||
412 size > fs_info->sectorsize ||
413 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
414 data_len > fs_info->max_inline)
417 path = btrfs_alloc_path();
421 trans = btrfs_join_transaction(root);
423 btrfs_free_path(path);
424 return PTR_ERR(trans);
426 trans->block_rsv = &inode->block_rsv;
428 drop_args.path = path;
430 drop_args.end = fs_info->sectorsize;
431 drop_args.drop_cache = true;
432 drop_args.replace_extent = true;
433 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
434 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
436 btrfs_abort_transaction(trans, ret);
440 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
441 size, compressed_size, compress_type,
442 compressed_pages, update_i_size);
443 if (ret && ret != -ENOSPC) {
444 btrfs_abort_transaction(trans, ret);
446 } else if (ret == -ENOSPC) {
451 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
452 ret = btrfs_update_inode(trans, root, inode);
453 if (ret && ret != -ENOSPC) {
454 btrfs_abort_transaction(trans, ret);
456 } else if (ret == -ENOSPC) {
461 btrfs_set_inode_full_sync(inode);
464 * Don't forget to free the reserved space, as for inlined extent
465 * it won't count as data extent, free them directly here.
466 * And at reserve time, it's always aligned to page size, so
467 * just free one page here.
469 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
470 btrfs_free_path(path);
471 btrfs_end_transaction(trans);
475 struct async_extent {
480 unsigned long nr_pages;
482 struct list_head list;
487 struct page *locked_page;
490 blk_opf_t write_flags;
491 struct list_head extents;
492 struct cgroup_subsys_state *blkcg_css;
493 struct btrfs_work work;
494 struct async_cow *async_cow;
499 struct async_chunk chunks[];
502 static noinline int add_async_extent(struct async_chunk *cow,
503 u64 start, u64 ram_size,
506 unsigned long nr_pages,
509 struct async_extent *async_extent;
511 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
512 BUG_ON(!async_extent); /* -ENOMEM */
513 async_extent->start = start;
514 async_extent->ram_size = ram_size;
515 async_extent->compressed_size = compressed_size;
516 async_extent->pages = pages;
517 async_extent->nr_pages = nr_pages;
518 async_extent->compress_type = compress_type;
519 list_add_tail(&async_extent->list, &cow->extents);
524 * Check if the inode needs to be submitted to compression, based on mount
525 * options, defragmentation, properties or heuristics.
527 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
530 struct btrfs_fs_info *fs_info = inode->root->fs_info;
532 if (!btrfs_inode_can_compress(inode)) {
533 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
534 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
539 * Special check for subpage.
541 * We lock the full page then run each delalloc range in the page, thus
542 * for the following case, we will hit some subpage specific corner case:
545 * | |///////| |///////|
548 * In above case, both range A and range B will try to unlock the full
549 * page [0, 64K), causing the one finished later will have page
550 * unlocked already, triggering various page lock requirement BUG_ON()s.
552 * So here we add an artificial limit that subpage compression can only
553 * if the range is fully page aligned.
555 * In theory we only need to ensure the first page is fully covered, but
556 * the tailing partial page will be locked until the full compression
557 * finishes, delaying the write of other range.
559 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
560 * first to prevent any submitted async extent to unlock the full page.
561 * By this, we can ensure for subpage case that only the last async_cow
562 * will unlock the full page.
564 if (fs_info->sectorsize < PAGE_SIZE) {
565 if (!PAGE_ALIGNED(start) ||
566 !PAGE_ALIGNED(end + 1))
571 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
574 if (inode->defrag_compress)
576 /* bad compression ratios */
577 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
579 if (btrfs_test_opt(fs_info, COMPRESS) ||
580 inode->flags & BTRFS_INODE_COMPRESS ||
581 inode->prop_compress)
582 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
586 static inline void inode_should_defrag(struct btrfs_inode *inode,
587 u64 start, u64 end, u64 num_bytes, u32 small_write)
589 /* If this is a small write inside eof, kick off a defrag */
590 if (num_bytes < small_write &&
591 (start > 0 || end + 1 < inode->disk_i_size))
592 btrfs_add_inode_defrag(NULL, inode, small_write);
596 * we create compressed extents in two phases. The first
597 * phase compresses a range of pages that have already been
598 * locked (both pages and state bits are locked).
600 * This is done inside an ordered work queue, and the compression
601 * is spread across many cpus. The actual IO submission is step
602 * two, and the ordered work queue takes care of making sure that
603 * happens in the same order things were put onto the queue by
604 * writepages and friends.
606 * If this code finds it can't get good compression, it puts an
607 * entry onto the work queue to write the uncompressed bytes. This
608 * makes sure that both compressed inodes and uncompressed inodes
609 * are written in the same order that the flusher thread sent them
612 static noinline int compress_file_range(struct async_chunk *async_chunk)
614 struct inode *inode = async_chunk->inode;
615 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
616 u64 blocksize = fs_info->sectorsize;
617 u64 start = async_chunk->start;
618 u64 end = async_chunk->end;
622 struct page **pages = NULL;
623 unsigned long nr_pages;
624 unsigned long total_compressed = 0;
625 unsigned long total_in = 0;
628 int compress_type = fs_info->compress_type;
629 int compressed_extents = 0;
632 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
636 * We need to save i_size before now because it could change in between
637 * us evaluating the size and assigning it. This is because we lock and
638 * unlock the page in truncate and fallocate, and then modify the i_size
641 * The barriers are to emulate READ_ONCE, remove that once i_size_read
645 i_size = i_size_read(inode);
647 actual_end = min_t(u64, i_size, end + 1);
650 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
651 nr_pages = min_t(unsigned long, nr_pages,
652 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
655 * we don't want to send crud past the end of i_size through
656 * compression, that's just a waste of CPU time. So, if the
657 * end of the file is before the start of our current
658 * requested range of bytes, we bail out to the uncompressed
659 * cleanup code that can deal with all of this.
661 * It isn't really the fastest way to fix things, but this is a
662 * very uncommon corner.
664 if (actual_end <= start)
665 goto cleanup_and_bail_uncompressed;
667 total_compressed = actual_end - start;
670 * Skip compression for a small file range(<=blocksize) that
671 * isn't an inline extent, since it doesn't save disk space at all.
673 if (total_compressed <= blocksize &&
674 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
675 goto cleanup_and_bail_uncompressed;
678 * For subpage case, we require full page alignment for the sector
680 * Thus we must also check against @actual_end, not just @end.
682 if (blocksize < PAGE_SIZE) {
683 if (!PAGE_ALIGNED(start) ||
684 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
685 goto cleanup_and_bail_uncompressed;
688 total_compressed = min_t(unsigned long, total_compressed,
689 BTRFS_MAX_UNCOMPRESSED);
694 * we do compression for mount -o compress and when the
695 * inode has not been flagged as nocompress. This flag can
696 * change at any time if we discover bad compression ratios.
698 if (inode_need_compress(BTRFS_I(inode), start, end)) {
700 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
702 /* just bail out to the uncompressed code */
707 if (BTRFS_I(inode)->defrag_compress)
708 compress_type = BTRFS_I(inode)->defrag_compress;
709 else if (BTRFS_I(inode)->prop_compress)
710 compress_type = BTRFS_I(inode)->prop_compress;
713 * we need to call clear_page_dirty_for_io on each
714 * page in the range. Otherwise applications with the file
715 * mmap'd can wander in and change the page contents while
716 * we are compressing them.
718 * If the compression fails for any reason, we set the pages
719 * dirty again later on.
721 * Note that the remaining part is redirtied, the start pointer
722 * has moved, the end is the original one.
725 extent_range_clear_dirty_for_io(inode, start, end);
729 /* Compression level is applied here and only here */
730 ret = btrfs_compress_pages(
731 compress_type | (fs_info->compress_level << 4),
732 inode->i_mapping, start,
739 unsigned long offset = offset_in_page(total_compressed);
740 struct page *page = pages[nr_pages - 1];
742 /* zero the tail end of the last page, we might be
743 * sending it down to disk
746 memzero_page(page, offset, PAGE_SIZE - offset);
752 * Check cow_file_range() for why we don't even try to create inline
753 * extent for subpage case.
755 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
756 /* lets try to make an inline extent */
757 if (ret || total_in < actual_end) {
758 /* we didn't compress the entire range, try
759 * to make an uncompressed inline extent.
761 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
762 0, BTRFS_COMPRESS_NONE,
765 /* try making a compressed inline extent */
766 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
768 compress_type, pages,
772 unsigned long clear_flags = EXTENT_DELALLOC |
773 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
774 EXTENT_DO_ACCOUNTING;
775 unsigned long page_error_op;
777 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
780 * inline extent creation worked or returned error,
781 * we don't need to create any more async work items.
782 * Unlock and free up our temp pages.
784 * We use DO_ACCOUNTING here because we need the
785 * delalloc_release_metadata to be done _after_ we drop
786 * our outstanding extent for clearing delalloc for this
789 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
793 PAGE_START_WRITEBACK |
798 * Ensure we only free the compressed pages if we have
799 * them allocated, as we can still reach here with
800 * inode_need_compress() == false.
803 for (i = 0; i < nr_pages; i++) {
804 WARN_ON(pages[i]->mapping);
815 * we aren't doing an inline extent round the compressed size
816 * up to a block size boundary so the allocator does sane
819 total_compressed = ALIGN(total_compressed, blocksize);
822 * one last check to make sure the compression is really a
823 * win, compare the page count read with the blocks on disk,
824 * compression must free at least one sector size
826 total_in = round_up(total_in, fs_info->sectorsize);
827 if (total_compressed + blocksize <= total_in) {
828 compressed_extents++;
831 * The async work queues will take care of doing actual
832 * allocation on disk for these compressed pages, and
833 * will submit them to the elevator.
835 add_async_extent(async_chunk, start, total_in,
836 total_compressed, pages, nr_pages,
839 if (start + total_in < end) {
845 return compressed_extents;
850 * the compression code ran but failed to make things smaller,
851 * free any pages it allocated and our page pointer array
853 for (i = 0; i < nr_pages; i++) {
854 WARN_ON(pages[i]->mapping);
859 total_compressed = 0;
862 /* flag the file so we don't compress in the future */
863 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
864 !(BTRFS_I(inode)->prop_compress)) {
865 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
868 cleanup_and_bail_uncompressed:
870 * No compression, but we still need to write the pages in the file
871 * we've been given so far. redirty the locked page if it corresponds
872 * to our extent and set things up for the async work queue to run
873 * cow_file_range to do the normal delalloc dance.
875 if (async_chunk->locked_page &&
876 (page_offset(async_chunk->locked_page) >= start &&
877 page_offset(async_chunk->locked_page)) <= end) {
878 __set_page_dirty_nobuffers(async_chunk->locked_page);
879 /* unlocked later on in the async handlers */
883 extent_range_redirty_for_io(inode, start, end);
884 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
885 BTRFS_COMPRESS_NONE);
886 compressed_extents++;
888 return compressed_extents;
891 static void free_async_extent_pages(struct async_extent *async_extent)
895 if (!async_extent->pages)
898 for (i = 0; i < async_extent->nr_pages; i++) {
899 WARN_ON(async_extent->pages[i]->mapping);
900 put_page(async_extent->pages[i]);
902 kfree(async_extent->pages);
903 async_extent->nr_pages = 0;
904 async_extent->pages = NULL;
907 static int submit_uncompressed_range(struct btrfs_inode *inode,
908 struct async_extent *async_extent,
909 struct page *locked_page)
911 u64 start = async_extent->start;
912 u64 end = async_extent->start + async_extent->ram_size - 1;
913 unsigned long nr_written = 0;
914 int page_started = 0;
918 * Call cow_file_range() to run the delalloc range directly, since we
919 * won't go to NOCOW or async path again.
921 * Also we call cow_file_range() with @unlock_page == 0, so that we
922 * can directly submit them without interruption.
924 ret = cow_file_range(inode, locked_page, start, end, &page_started,
925 &nr_written, 0, NULL);
926 /* Inline extent inserted, page gets unlocked and everything is done */
932 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
934 const u64 page_start = page_offset(locked_page);
935 const u64 page_end = page_start + PAGE_SIZE - 1;
937 btrfs_page_set_error(inode->root->fs_info, locked_page,
938 page_start, PAGE_SIZE);
939 set_page_writeback(locked_page);
940 end_page_writeback(locked_page);
941 end_extent_writepage(locked_page, ret, page_start, page_end);
942 unlock_page(locked_page);
947 ret = extent_write_locked_range(&inode->vfs_inode, start, end);
948 /* All pages will be unlocked, including @locked_page */
954 static int submit_one_async_extent(struct btrfs_inode *inode,
955 struct async_chunk *async_chunk,
956 struct async_extent *async_extent,
959 struct extent_io_tree *io_tree = &inode->io_tree;
960 struct btrfs_root *root = inode->root;
961 struct btrfs_fs_info *fs_info = root->fs_info;
962 struct btrfs_key ins;
963 struct page *locked_page = NULL;
964 struct extent_map *em;
966 u64 start = async_extent->start;
967 u64 end = async_extent->start + async_extent->ram_size - 1;
970 * If async_chunk->locked_page is in the async_extent range, we need to
973 if (async_chunk->locked_page) {
974 u64 locked_page_start = page_offset(async_chunk->locked_page);
975 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
977 if (!(start >= locked_page_end || end <= locked_page_start))
978 locked_page = async_chunk->locked_page;
980 lock_extent(io_tree, start, end);
982 /* We have fall back to uncompressed write */
983 if (!async_extent->pages)
984 return submit_uncompressed_range(inode, async_extent, locked_page);
986 ret = btrfs_reserve_extent(root, async_extent->ram_size,
987 async_extent->compressed_size,
988 async_extent->compressed_size,
989 0, *alloc_hint, &ins, 1, 1);
991 free_async_extent_pages(async_extent);
993 * Here we used to try again by going back to non-compressed
994 * path for ENOSPC. But we can't reserve space even for
995 * compressed size, how could it work for uncompressed size
996 * which requires larger size? So here we directly go error
1002 /* Here we're doing allocation and writeback of the compressed pages */
1003 em = create_io_em(inode, start,
1004 async_extent->ram_size, /* len */
1005 start, /* orig_start */
1006 ins.objectid, /* block_start */
1007 ins.offset, /* block_len */
1008 ins.offset, /* orig_block_len */
1009 async_extent->ram_size, /* ram_bytes */
1010 async_extent->compress_type,
1011 BTRFS_ORDERED_COMPRESSED);
1014 goto out_free_reserve;
1016 free_extent_map(em);
1018 ret = btrfs_add_ordered_extent(inode, start, /* file_offset */
1019 async_extent->ram_size, /* num_bytes */
1020 async_extent->ram_size, /* ram_bytes */
1021 ins.objectid, /* disk_bytenr */
1022 ins.offset, /* disk_num_bytes */
1024 1 << BTRFS_ORDERED_COMPRESSED,
1025 async_extent->compress_type);
1027 btrfs_drop_extent_cache(inode, start, end, 0);
1028 goto out_free_reserve;
1030 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1032 /* Clear dirty, set writeback and unlock the pages. */
1033 extent_clear_unlock_delalloc(inode, start, end,
1034 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1035 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1036 if (btrfs_submit_compressed_write(inode, start, /* file_offset */
1037 async_extent->ram_size, /* num_bytes */
1038 ins.objectid, /* disk_bytenr */
1039 ins.offset, /* compressed_len */
1040 async_extent->pages, /* compressed_pages */
1041 async_extent->nr_pages,
1042 async_chunk->write_flags,
1043 async_chunk->blkcg_css, true)) {
1044 const u64 start = async_extent->start;
1045 const u64 end = start + async_extent->ram_size - 1;
1047 btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0);
1049 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
1050 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1051 free_async_extent_pages(async_extent);
1053 *alloc_hint = ins.objectid + ins.offset;
1054 kfree(async_extent);
1058 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1059 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1061 extent_clear_unlock_delalloc(inode, start, end,
1062 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1063 EXTENT_DELALLOC_NEW |
1064 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1065 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1066 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1067 free_async_extent_pages(async_extent);
1068 kfree(async_extent);
1073 * Phase two of compressed writeback. This is the ordered portion of the code,
1074 * which only gets called in the order the work was queued. We walk all the
1075 * async extents created by compress_file_range and send them down to the disk.
1077 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1079 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
1080 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1081 struct async_extent *async_extent;
1085 while (!list_empty(&async_chunk->extents)) {
1089 async_extent = list_entry(async_chunk->extents.next,
1090 struct async_extent, list);
1091 list_del(&async_extent->list);
1092 extent_start = async_extent->start;
1093 ram_size = async_extent->ram_size;
1095 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1097 btrfs_debug(fs_info,
1098 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1099 inode->root->root_key.objectid,
1100 btrfs_ino(inode), extent_start, ram_size, ret);
1104 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1107 struct extent_map_tree *em_tree = &inode->extent_tree;
1108 struct extent_map *em;
1111 read_lock(&em_tree->lock);
1112 em = search_extent_mapping(em_tree, start, num_bytes);
1115 * if block start isn't an actual block number then find the
1116 * first block in this inode and use that as a hint. If that
1117 * block is also bogus then just don't worry about it.
1119 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1120 free_extent_map(em);
1121 em = search_extent_mapping(em_tree, 0, 0);
1122 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1123 alloc_hint = em->block_start;
1125 free_extent_map(em);
1127 alloc_hint = em->block_start;
1128 free_extent_map(em);
1131 read_unlock(&em_tree->lock);
1137 * when extent_io.c finds a delayed allocation range in the file,
1138 * the call backs end up in this code. The basic idea is to
1139 * allocate extents on disk for the range, and create ordered data structs
1140 * in ram to track those extents.
1142 * locked_page is the page that writepage had locked already. We use
1143 * it to make sure we don't do extra locks or unlocks.
1145 * *page_started is set to one if we unlock locked_page and do everything
1146 * required to start IO on it. It may be clean and already done with
1147 * IO when we return.
1149 * When unlock == 1, we unlock the pages in successfully allocated regions.
1150 * When unlock == 0, we leave them locked for writing them out.
1152 * However, we unlock all the pages except @locked_page in case of failure.
1154 * In summary, page locking state will be as follow:
1156 * - page_started == 1 (return value)
1157 * - All the pages are unlocked. IO is started.
1158 * - Note that this can happen only on success
1160 * - All the pages except @locked_page are unlocked in any case
1162 * - On success, all the pages are locked for writing out them
1163 * - On failure, all the pages except @locked_page are unlocked
1165 * When a failure happens in the second or later iteration of the
1166 * while-loop, the ordered extents created in previous iterations are kept
1167 * intact. So, the caller must clean them up by calling
1168 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1171 static noinline int cow_file_range(struct btrfs_inode *inode,
1172 struct page *locked_page,
1173 u64 start, u64 end, int *page_started,
1174 unsigned long *nr_written, int unlock,
1177 struct btrfs_root *root = inode->root;
1178 struct btrfs_fs_info *fs_info = root->fs_info;
1180 u64 orig_start = start;
1182 unsigned long ram_size;
1183 u64 cur_alloc_size = 0;
1185 u64 blocksize = fs_info->sectorsize;
1186 struct btrfs_key ins;
1187 struct extent_map *em;
1188 unsigned clear_bits;
1189 unsigned long page_ops;
1190 bool extent_reserved = false;
1193 if (btrfs_is_free_space_inode(inode)) {
1198 num_bytes = ALIGN(end - start + 1, blocksize);
1199 num_bytes = max(blocksize, num_bytes);
1200 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1202 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1205 * Due to the page size limit, for subpage we can only trigger the
1206 * writeback for the dirty sectors of page, that means data writeback
1207 * is doing more writeback than what we want.
1209 * This is especially unexpected for some call sites like fallocate,
1210 * where we only increase i_size after everything is done.
1211 * This means we can trigger inline extent even if we didn't want to.
1212 * So here we skip inline extent creation completely.
1214 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1215 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1218 /* lets try to make an inline extent */
1219 ret = cow_file_range_inline(inode, actual_end, 0,
1220 BTRFS_COMPRESS_NONE, NULL, false);
1223 * We use DO_ACCOUNTING here because we need the
1224 * delalloc_release_metadata to be run _after_ we drop
1225 * our outstanding extent for clearing delalloc for this
1228 extent_clear_unlock_delalloc(inode, start, end,
1230 EXTENT_LOCKED | EXTENT_DELALLOC |
1231 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1232 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1233 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1234 *nr_written = *nr_written +
1235 (end - start + PAGE_SIZE) / PAGE_SIZE;
1238 * locked_page is locked by the caller of
1239 * writepage_delalloc(), not locked by
1240 * __process_pages_contig().
1242 * We can't let __process_pages_contig() to unlock it,
1243 * as it doesn't have any subpage::writers recorded.
1245 * Here we manually unlock the page, since the caller
1246 * can't use page_started to determine if it's an
1247 * inline extent or a compressed extent.
1249 unlock_page(locked_page);
1251 } else if (ret < 0) {
1256 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1257 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1260 * Relocation relies on the relocated extents to have exactly the same
1261 * size as the original extents. Normally writeback for relocation data
1262 * extents follows a NOCOW path because relocation preallocates the
1263 * extents. However, due to an operation such as scrub turning a block
1264 * group to RO mode, it may fallback to COW mode, so we must make sure
1265 * an extent allocated during COW has exactly the requested size and can
1266 * not be split into smaller extents, otherwise relocation breaks and
1267 * fails during the stage where it updates the bytenr of file extent
1270 if (btrfs_is_data_reloc_root(root))
1271 min_alloc_size = num_bytes;
1273 min_alloc_size = fs_info->sectorsize;
1275 while (num_bytes > 0) {
1276 cur_alloc_size = num_bytes;
1277 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1278 min_alloc_size, 0, alloc_hint,
1282 cur_alloc_size = ins.offset;
1283 extent_reserved = true;
1285 ram_size = ins.offset;
1286 em = create_io_em(inode, start, ins.offset, /* len */
1287 start, /* orig_start */
1288 ins.objectid, /* block_start */
1289 ins.offset, /* block_len */
1290 ins.offset, /* orig_block_len */
1291 ram_size, /* ram_bytes */
1292 BTRFS_COMPRESS_NONE, /* compress_type */
1293 BTRFS_ORDERED_REGULAR /* type */);
1298 free_extent_map(em);
1300 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1301 ins.objectid, cur_alloc_size, 0,
1302 1 << BTRFS_ORDERED_REGULAR,
1303 BTRFS_COMPRESS_NONE);
1305 goto out_drop_extent_cache;
1307 if (btrfs_is_data_reloc_root(root)) {
1308 ret = btrfs_reloc_clone_csums(inode, start,
1311 * Only drop cache here, and process as normal.
1313 * We must not allow extent_clear_unlock_delalloc()
1314 * at out_unlock label to free meta of this ordered
1315 * extent, as its meta should be freed by
1316 * btrfs_finish_ordered_io().
1318 * So we must continue until @start is increased to
1319 * skip current ordered extent.
1322 btrfs_drop_extent_cache(inode, start,
1323 start + ram_size - 1, 0);
1326 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1329 * We're not doing compressed IO, don't unlock the first page
1330 * (which the caller expects to stay locked), don't clear any
1331 * dirty bits and don't set any writeback bits
1333 * Do set the Ordered (Private2) bit so we know this page was
1334 * properly setup for writepage.
1336 page_ops = unlock ? PAGE_UNLOCK : 0;
1337 page_ops |= PAGE_SET_ORDERED;
1339 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1341 EXTENT_LOCKED | EXTENT_DELALLOC,
1343 if (num_bytes < cur_alloc_size)
1346 num_bytes -= cur_alloc_size;
1347 alloc_hint = ins.objectid + ins.offset;
1348 start += cur_alloc_size;
1349 extent_reserved = false;
1352 * btrfs_reloc_clone_csums() error, since start is increased
1353 * extent_clear_unlock_delalloc() at out_unlock label won't
1354 * free metadata of current ordered extent, we're OK to exit.
1362 out_drop_extent_cache:
1363 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1365 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1366 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1369 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1370 * caller to write out the successfully allocated region and retry.
1372 if (done_offset && ret == -EAGAIN) {
1373 if (orig_start < start)
1374 *done_offset = start - 1;
1376 *done_offset = start;
1378 } else if (ret == -EAGAIN) {
1379 /* Convert to -ENOSPC since the caller cannot retry. */
1384 * Now, we have three regions to clean up:
1386 * |-------(1)----|---(2)---|-------------(3)----------|
1387 * `- orig_start `- start `- start + cur_alloc_size `- end
1389 * We process each region below.
1392 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1393 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1394 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1397 * For the range (1). We have already instantiated the ordered extents
1398 * for this region. They are cleaned up by
1399 * btrfs_cleanup_ordered_extents() in e.g,
1400 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1401 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1402 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1405 * However, in case of unlock == 0, we still need to unlock the pages
1406 * (except @locked_page) to ensure all the pages are unlocked.
1408 if (!unlock && orig_start < start) {
1410 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1411 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1412 locked_page, 0, page_ops);
1416 * For the range (2). If we reserved an extent for our delalloc range
1417 * (or a subrange) and failed to create the respective ordered extent,
1418 * then it means that when we reserved the extent we decremented the
1419 * extent's size from the data space_info's bytes_may_use counter and
1420 * incremented the space_info's bytes_reserved counter by the same
1421 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1422 * to decrement again the data space_info's bytes_may_use counter,
1423 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1425 if (extent_reserved) {
1426 extent_clear_unlock_delalloc(inode, start,
1427 start + cur_alloc_size - 1,
1431 start += cur_alloc_size;
1437 * For the range (3). We never touched the region. In addition to the
1438 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1439 * space_info's bytes_may_use counter, reserved in
1440 * btrfs_check_data_free_space().
1442 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1443 clear_bits | EXTENT_CLEAR_DATA_RESV,
1449 * work queue call back to started compression on a file and pages
1451 static noinline void async_cow_start(struct btrfs_work *work)
1453 struct async_chunk *async_chunk;
1454 int compressed_extents;
1456 async_chunk = container_of(work, struct async_chunk, work);
1458 compressed_extents = compress_file_range(async_chunk);
1459 if (compressed_extents == 0) {
1460 btrfs_add_delayed_iput(async_chunk->inode);
1461 async_chunk->inode = NULL;
1466 * work queue call back to submit previously compressed pages
1468 static noinline void async_cow_submit(struct btrfs_work *work)
1470 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1472 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1473 unsigned long nr_pages;
1475 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1479 * ->inode could be NULL if async_chunk_start has failed to compress,
1480 * in which case we don't have anything to submit, yet we need to
1481 * always adjust ->async_delalloc_pages as its paired with the init
1482 * happening in cow_file_range_async
1484 if (async_chunk->inode)
1485 submit_compressed_extents(async_chunk);
1487 /* atomic_sub_return implies a barrier */
1488 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1490 cond_wake_up_nomb(&fs_info->async_submit_wait);
1493 static noinline void async_cow_free(struct btrfs_work *work)
1495 struct async_chunk *async_chunk;
1496 struct async_cow *async_cow;
1498 async_chunk = container_of(work, struct async_chunk, work);
1499 if (async_chunk->inode)
1500 btrfs_add_delayed_iput(async_chunk->inode);
1501 if (async_chunk->blkcg_css)
1502 css_put(async_chunk->blkcg_css);
1504 async_cow = async_chunk->async_cow;
1505 if (atomic_dec_and_test(&async_cow->num_chunks))
1509 static int cow_file_range_async(struct btrfs_inode *inode,
1510 struct writeback_control *wbc,
1511 struct page *locked_page,
1512 u64 start, u64 end, int *page_started,
1513 unsigned long *nr_written)
1515 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1516 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1517 struct async_cow *ctx;
1518 struct async_chunk *async_chunk;
1519 unsigned long nr_pages;
1521 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1523 bool should_compress;
1525 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1527 unlock_extent(&inode->io_tree, start, end);
1529 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1530 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1532 should_compress = false;
1534 should_compress = true;
1537 nofs_flag = memalloc_nofs_save();
1538 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1539 memalloc_nofs_restore(nofs_flag);
1542 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1543 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1544 EXTENT_DO_ACCOUNTING;
1545 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1546 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1548 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1549 clear_bits, page_ops);
1553 async_chunk = ctx->chunks;
1554 atomic_set(&ctx->num_chunks, num_chunks);
1556 for (i = 0; i < num_chunks; i++) {
1557 if (should_compress)
1558 cur_end = min(end, start + SZ_512K - 1);
1563 * igrab is called higher up in the call chain, take only the
1564 * lightweight reference for the callback lifetime
1566 ihold(&inode->vfs_inode);
1567 async_chunk[i].async_cow = ctx;
1568 async_chunk[i].inode = &inode->vfs_inode;
1569 async_chunk[i].start = start;
1570 async_chunk[i].end = cur_end;
1571 async_chunk[i].write_flags = write_flags;
1572 INIT_LIST_HEAD(&async_chunk[i].extents);
1575 * The locked_page comes all the way from writepage and its
1576 * the original page we were actually given. As we spread
1577 * this large delalloc region across multiple async_chunk
1578 * structs, only the first struct needs a pointer to locked_page
1580 * This way we don't need racey decisions about who is supposed
1585 * Depending on the compressibility, the pages might or
1586 * might not go through async. We want all of them to
1587 * be accounted against wbc once. Let's do it here
1588 * before the paths diverge. wbc accounting is used
1589 * only for foreign writeback detection and doesn't
1590 * need full accuracy. Just account the whole thing
1591 * against the first page.
1593 wbc_account_cgroup_owner(wbc, locked_page,
1595 async_chunk[i].locked_page = locked_page;
1598 async_chunk[i].locked_page = NULL;
1601 if (blkcg_css != blkcg_root_css) {
1603 async_chunk[i].blkcg_css = blkcg_css;
1605 async_chunk[i].blkcg_css = NULL;
1608 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1609 async_cow_submit, async_cow_free);
1611 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1612 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1614 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1616 *nr_written += nr_pages;
1617 start = cur_end + 1;
1623 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1624 struct page *locked_page, u64 start,
1625 u64 end, int *page_started,
1626 unsigned long *nr_written)
1628 u64 done_offset = end;
1630 bool locked_page_done = false;
1632 while (start <= end) {
1633 ret = cow_file_range(inode, locked_page, start, end, page_started,
1634 nr_written, 0, &done_offset);
1635 if (ret && ret != -EAGAIN)
1638 if (*page_started) {
1646 if (done_offset == start) {
1647 struct btrfs_fs_info *info = inode->root->fs_info;
1649 wait_var_event(&info->zone_finish_wait,
1650 !test_bit(BTRFS_FS_NEED_ZONE_FINISH, &info->flags));
1654 if (!locked_page_done) {
1655 __set_page_dirty_nobuffers(locked_page);
1656 account_page_redirty(locked_page);
1658 locked_page_done = true;
1659 extent_write_locked_range(&inode->vfs_inode, start, done_offset);
1661 start = done_offset + 1;
1669 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1670 u64 bytenr, u64 num_bytes)
1672 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1673 struct btrfs_ordered_sum *sums;
1677 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1678 bytenr + num_bytes - 1, &list, 0);
1679 if (ret == 0 && list_empty(&list))
1682 while (!list_empty(&list)) {
1683 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1684 list_del(&sums->list);
1692 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1693 const u64 start, const u64 end,
1694 int *page_started, unsigned long *nr_written)
1696 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1697 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1698 const u64 range_bytes = end + 1 - start;
1699 struct extent_io_tree *io_tree = &inode->io_tree;
1700 u64 range_start = start;
1704 * If EXTENT_NORESERVE is set it means that when the buffered write was
1705 * made we had not enough available data space and therefore we did not
1706 * reserve data space for it, since we though we could do NOCOW for the
1707 * respective file range (either there is prealloc extent or the inode
1708 * has the NOCOW bit set).
1710 * However when we need to fallback to COW mode (because for example the
1711 * block group for the corresponding extent was turned to RO mode by a
1712 * scrub or relocation) we need to do the following:
1714 * 1) We increment the bytes_may_use counter of the data space info.
1715 * If COW succeeds, it allocates a new data extent and after doing
1716 * that it decrements the space info's bytes_may_use counter and
1717 * increments its bytes_reserved counter by the same amount (we do
1718 * this at btrfs_add_reserved_bytes()). So we need to increment the
1719 * bytes_may_use counter to compensate (when space is reserved at
1720 * buffered write time, the bytes_may_use counter is incremented);
1722 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1723 * that if the COW path fails for any reason, it decrements (through
1724 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1725 * data space info, which we incremented in the step above.
1727 * If we need to fallback to cow and the inode corresponds to a free
1728 * space cache inode or an inode of the data relocation tree, we must
1729 * also increment bytes_may_use of the data space_info for the same
1730 * reason. Space caches and relocated data extents always get a prealloc
1731 * extent for them, however scrub or balance may have set the block
1732 * group that contains that extent to RO mode and therefore force COW
1733 * when starting writeback.
1735 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1736 EXTENT_NORESERVE, 0);
1737 if (count > 0 || is_space_ino || is_reloc_ino) {
1739 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1740 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1742 if (is_space_ino || is_reloc_ino)
1743 bytes = range_bytes;
1745 spin_lock(&sinfo->lock);
1746 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1747 spin_unlock(&sinfo->lock);
1750 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1754 return cow_file_range(inode, locked_page, start, end, page_started,
1755 nr_written, 1, NULL);
1758 struct can_nocow_file_extent_args {
1761 /* Start file offset of the range we want to NOCOW. */
1763 /* End file offset (inclusive) of the range we want to NOCOW. */
1765 bool writeback_path;
1768 * Free the path passed to can_nocow_file_extent() once it's not needed
1773 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1778 /* Number of bytes that can be written to in NOCOW mode. */
1783 * Check if we can NOCOW the file extent that the path points to.
1784 * This function may return with the path released, so the caller should check
1785 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1787 * Returns: < 0 on error
1788 * 0 if we can not NOCOW
1791 static int can_nocow_file_extent(struct btrfs_path *path,
1792 struct btrfs_key *key,
1793 struct btrfs_inode *inode,
1794 struct can_nocow_file_extent_args *args)
1796 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1797 struct extent_buffer *leaf = path->nodes[0];
1798 struct btrfs_root *root = inode->root;
1799 struct btrfs_file_extent_item *fi;
1805 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1806 extent_type = btrfs_file_extent_type(leaf, fi);
1808 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1811 /* Can't access these fields unless we know it's not an inline extent. */
1812 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1813 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1814 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1816 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1817 extent_type == BTRFS_FILE_EXTENT_REG)
1821 * If the extent was created before the generation where the last snapshot
1822 * for its subvolume was created, then this implies the extent is shared,
1823 * hence we must COW.
1825 if (!args->strict &&
1826 btrfs_file_extent_generation(leaf, fi) <=
1827 btrfs_root_last_snapshot(&root->root_item))
1830 /* An explicit hole, must COW. */
1831 if (args->disk_bytenr == 0)
1834 /* Compressed/encrypted/encoded extents must be COWed. */
1835 if (btrfs_file_extent_compression(leaf, fi) ||
1836 btrfs_file_extent_encryption(leaf, fi) ||
1837 btrfs_file_extent_other_encoding(leaf, fi))
1840 extent_end = btrfs_file_extent_end(path);
1843 * The following checks can be expensive, as they need to take other
1844 * locks and do btree or rbtree searches, so release the path to avoid
1845 * blocking other tasks for too long.
1847 btrfs_release_path(path);
1849 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1850 key->offset - args->extent_offset,
1851 args->disk_bytenr, false, path);
1852 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1856 if (args->free_path) {
1858 * We don't need the path anymore, plus through the
1859 * csum_exist_in_range() call below we will end up allocating
1860 * another path. So free the path to avoid unnecessary extra
1863 btrfs_free_path(path);
1867 /* If there are pending snapshots for this root, we must COW. */
1868 if (args->writeback_path && !is_freespace_inode &&
1869 atomic_read(&root->snapshot_force_cow))
1872 args->disk_bytenr += args->extent_offset;
1873 args->disk_bytenr += args->start - key->offset;
1874 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1877 * Force COW if csums exist in the range. This ensures that csums for a
1878 * given extent are either valid or do not exist.
1880 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes);
1881 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1887 if (args->free_path && path)
1888 btrfs_free_path(path);
1890 return ret < 0 ? ret : can_nocow;
1894 * when nowcow writeback call back. This checks for snapshots or COW copies
1895 * of the extents that exist in the file, and COWs the file as required.
1897 * If no cow copies or snapshots exist, we write directly to the existing
1900 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1901 struct page *locked_page,
1902 const u64 start, const u64 end,
1904 unsigned long *nr_written)
1906 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1907 struct btrfs_root *root = inode->root;
1908 struct btrfs_path *path;
1909 u64 cow_start = (u64)-1;
1910 u64 cur_offset = start;
1912 bool check_prev = true;
1913 u64 ino = btrfs_ino(inode);
1914 struct btrfs_block_group *bg;
1916 struct can_nocow_file_extent_args nocow_args = { 0 };
1918 path = btrfs_alloc_path();
1920 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1921 EXTENT_LOCKED | EXTENT_DELALLOC |
1922 EXTENT_DO_ACCOUNTING |
1923 EXTENT_DEFRAG, PAGE_UNLOCK |
1924 PAGE_START_WRITEBACK |
1925 PAGE_END_WRITEBACK);
1929 nocow_args.end = end;
1930 nocow_args.writeback_path = true;
1933 struct btrfs_key found_key;
1934 struct btrfs_file_extent_item *fi;
1935 struct extent_buffer *leaf;
1943 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1949 * If there is no extent for our range when doing the initial
1950 * search, then go back to the previous slot as it will be the
1951 * one containing the search offset
1953 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1954 leaf = path->nodes[0];
1955 btrfs_item_key_to_cpu(leaf, &found_key,
1956 path->slots[0] - 1);
1957 if (found_key.objectid == ino &&
1958 found_key.type == BTRFS_EXTENT_DATA_KEY)
1963 /* Go to next leaf if we have exhausted the current one */
1964 leaf = path->nodes[0];
1965 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1966 ret = btrfs_next_leaf(root, path);
1968 if (cow_start != (u64)-1)
1969 cur_offset = cow_start;
1974 leaf = path->nodes[0];
1977 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1979 /* Didn't find anything for our INO */
1980 if (found_key.objectid > ino)
1983 * Keep searching until we find an EXTENT_ITEM or there are no
1984 * more extents for this inode
1986 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1987 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1992 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1993 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1994 found_key.offset > end)
1998 * If the found extent starts after requested offset, then
1999 * adjust extent_end to be right before this extent begins
2001 if (found_key.offset > cur_offset) {
2002 extent_end = found_key.offset;
2008 * Found extent which begins before our range and potentially
2011 fi = btrfs_item_ptr(leaf, path->slots[0],
2012 struct btrfs_file_extent_item);
2013 extent_type = btrfs_file_extent_type(leaf, fi);
2014 /* If this is triggered then we have a memory corruption. */
2015 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2016 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2020 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2021 extent_end = btrfs_file_extent_end(path);
2024 * If the extent we got ends before our current offset, skip to
2027 if (extent_end <= cur_offset) {
2032 nocow_args.start = cur_offset;
2033 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2035 if (cow_start != (u64)-1)
2036 cur_offset = cow_start;
2038 } else if (ret == 0) {
2043 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2048 * If nocow is false then record the beginning of the range
2049 * that needs to be COWed
2052 if (cow_start == (u64)-1)
2053 cow_start = cur_offset;
2054 cur_offset = extent_end;
2055 if (cur_offset > end)
2057 if (!path->nodes[0])
2064 * COW range from cow_start to found_key.offset - 1. As the key
2065 * will contain the beginning of the first extent that can be
2066 * NOCOW, following one which needs to be COW'ed
2068 if (cow_start != (u64)-1) {
2069 ret = fallback_to_cow(inode, locked_page,
2070 cow_start, found_key.offset - 1,
2071 page_started, nr_written);
2074 cow_start = (u64)-1;
2077 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2079 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2080 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2081 struct extent_map *em;
2083 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2085 nocow_args.disk_bytenr, /* block_start */
2086 nocow_args.num_bytes, /* block_len */
2087 nocow_args.disk_num_bytes, /* orig_block_len */
2088 ram_bytes, BTRFS_COMPRESS_NONE,
2089 BTRFS_ORDERED_PREALLOC);
2094 free_extent_map(em);
2095 ret = btrfs_add_ordered_extent(inode,
2096 cur_offset, nocow_args.num_bytes,
2097 nocow_args.num_bytes,
2098 nocow_args.disk_bytenr,
2099 nocow_args.num_bytes, 0,
2100 1 << BTRFS_ORDERED_PREALLOC,
2101 BTRFS_COMPRESS_NONE);
2103 btrfs_drop_extent_cache(inode, cur_offset,
2108 ret = btrfs_add_ordered_extent(inode, cur_offset,
2109 nocow_args.num_bytes,
2110 nocow_args.num_bytes,
2111 nocow_args.disk_bytenr,
2112 nocow_args.num_bytes,
2114 1 << BTRFS_ORDERED_NOCOW,
2115 BTRFS_COMPRESS_NONE);
2121 btrfs_dec_nocow_writers(bg);
2125 if (btrfs_is_data_reloc_root(root))
2127 * Error handled later, as we must prevent
2128 * extent_clear_unlock_delalloc() in error handler
2129 * from freeing metadata of created ordered extent.
2131 ret = btrfs_reloc_clone_csums(inode, cur_offset,
2132 nocow_args.num_bytes);
2134 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2135 locked_page, EXTENT_LOCKED |
2137 EXTENT_CLEAR_DATA_RESV,
2138 PAGE_UNLOCK | PAGE_SET_ORDERED);
2140 cur_offset = extent_end;
2143 * btrfs_reloc_clone_csums() error, now we're OK to call error
2144 * handler, as metadata for created ordered extent will only
2145 * be freed by btrfs_finish_ordered_io().
2149 if (cur_offset > end)
2152 btrfs_release_path(path);
2154 if (cur_offset <= end && cow_start == (u64)-1)
2155 cow_start = cur_offset;
2157 if (cow_start != (u64)-1) {
2159 ret = fallback_to_cow(inode, locked_page, cow_start, end,
2160 page_started, nr_written);
2167 btrfs_dec_nocow_writers(bg);
2169 if (ret && cur_offset < end)
2170 extent_clear_unlock_delalloc(inode, cur_offset, end,
2171 locked_page, EXTENT_LOCKED |
2172 EXTENT_DELALLOC | EXTENT_DEFRAG |
2173 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2174 PAGE_START_WRITEBACK |
2175 PAGE_END_WRITEBACK);
2176 btrfs_free_path(path);
2180 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2182 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2183 if (inode->defrag_bytes &&
2184 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2193 * Function to process delayed allocation (create CoW) for ranges which are
2194 * being touched for the first time.
2196 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2197 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2198 struct writeback_control *wbc)
2201 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2204 * The range must cover part of the @locked_page, or the returned
2205 * @page_started can confuse the caller.
2207 ASSERT(!(end <= page_offset(locked_page) ||
2208 start >= page_offset(locked_page) + PAGE_SIZE));
2210 if (should_nocow(inode, start, end)) {
2212 * Normally on a zoned device we're only doing COW writes, but
2213 * in case of relocation on a zoned filesystem we have taken
2214 * precaution, that we're only writing sequentially. It's safe
2215 * to use run_delalloc_nocow() here, like for regular
2216 * preallocated inodes.
2218 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2219 ret = run_delalloc_nocow(inode, locked_page, start, end,
2220 page_started, nr_written);
2221 } else if (!btrfs_inode_can_compress(inode) ||
2222 !inode_need_compress(inode, start, end)) {
2224 ret = run_delalloc_zoned(inode, locked_page, start, end,
2225 page_started, nr_written);
2227 ret = cow_file_range(inode, locked_page, start, end,
2228 page_started, nr_written, 1, NULL);
2230 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2231 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2232 page_started, nr_written);
2236 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2241 void btrfs_split_delalloc_extent(struct inode *inode,
2242 struct extent_state *orig, u64 split)
2244 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2247 /* not delalloc, ignore it */
2248 if (!(orig->state & EXTENT_DELALLOC))
2251 size = orig->end - orig->start + 1;
2252 if (size > fs_info->max_extent_size) {
2257 * See the explanation in btrfs_merge_delalloc_extent, the same
2258 * applies here, just in reverse.
2260 new_size = orig->end - split + 1;
2261 num_extents = count_max_extents(fs_info, new_size);
2262 new_size = split - orig->start;
2263 num_extents += count_max_extents(fs_info, new_size);
2264 if (count_max_extents(fs_info, size) >= num_extents)
2268 spin_lock(&BTRFS_I(inode)->lock);
2269 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2270 spin_unlock(&BTRFS_I(inode)->lock);
2274 * Handle merged delayed allocation extents so we can keep track of new extents
2275 * that are just merged onto old extents, such as when we are doing sequential
2276 * writes, so we can properly account for the metadata space we'll need.
2278 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2279 struct extent_state *other)
2281 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2282 u64 new_size, old_size;
2285 /* not delalloc, ignore it */
2286 if (!(other->state & EXTENT_DELALLOC))
2289 if (new->start > other->start)
2290 new_size = new->end - other->start + 1;
2292 new_size = other->end - new->start + 1;
2294 /* we're not bigger than the max, unreserve the space and go */
2295 if (new_size <= fs_info->max_extent_size) {
2296 spin_lock(&BTRFS_I(inode)->lock);
2297 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2298 spin_unlock(&BTRFS_I(inode)->lock);
2303 * We have to add up either side to figure out how many extents were
2304 * accounted for before we merged into one big extent. If the number of
2305 * extents we accounted for is <= the amount we need for the new range
2306 * then we can return, otherwise drop. Think of it like this
2310 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2311 * need 2 outstanding extents, on one side we have 1 and the other side
2312 * we have 1 so they are == and we can return. But in this case
2314 * [MAX_SIZE+4k][MAX_SIZE+4k]
2316 * Each range on their own accounts for 2 extents, but merged together
2317 * they are only 3 extents worth of accounting, so we need to drop in
2320 old_size = other->end - other->start + 1;
2321 num_extents = count_max_extents(fs_info, old_size);
2322 old_size = new->end - new->start + 1;
2323 num_extents += count_max_extents(fs_info, old_size);
2324 if (count_max_extents(fs_info, new_size) >= num_extents)
2327 spin_lock(&BTRFS_I(inode)->lock);
2328 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2329 spin_unlock(&BTRFS_I(inode)->lock);
2332 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2333 struct inode *inode)
2335 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2337 spin_lock(&root->delalloc_lock);
2338 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2339 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2340 &root->delalloc_inodes);
2341 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2342 &BTRFS_I(inode)->runtime_flags);
2343 root->nr_delalloc_inodes++;
2344 if (root->nr_delalloc_inodes == 1) {
2345 spin_lock(&fs_info->delalloc_root_lock);
2346 BUG_ON(!list_empty(&root->delalloc_root));
2347 list_add_tail(&root->delalloc_root,
2348 &fs_info->delalloc_roots);
2349 spin_unlock(&fs_info->delalloc_root_lock);
2352 spin_unlock(&root->delalloc_lock);
2356 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2357 struct btrfs_inode *inode)
2359 struct btrfs_fs_info *fs_info = root->fs_info;
2361 if (!list_empty(&inode->delalloc_inodes)) {
2362 list_del_init(&inode->delalloc_inodes);
2363 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2364 &inode->runtime_flags);
2365 root->nr_delalloc_inodes--;
2366 if (!root->nr_delalloc_inodes) {
2367 ASSERT(list_empty(&root->delalloc_inodes));
2368 spin_lock(&fs_info->delalloc_root_lock);
2369 BUG_ON(list_empty(&root->delalloc_root));
2370 list_del_init(&root->delalloc_root);
2371 spin_unlock(&fs_info->delalloc_root_lock);
2376 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2377 struct btrfs_inode *inode)
2379 spin_lock(&root->delalloc_lock);
2380 __btrfs_del_delalloc_inode(root, inode);
2381 spin_unlock(&root->delalloc_lock);
2385 * Properly track delayed allocation bytes in the inode and to maintain the
2386 * list of inodes that have pending delalloc work to be done.
2388 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2391 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2393 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2396 * set_bit and clear bit hooks normally require _irqsave/restore
2397 * but in this case, we are only testing for the DELALLOC
2398 * bit, which is only set or cleared with irqs on
2400 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2401 struct btrfs_root *root = BTRFS_I(inode)->root;
2402 u64 len = state->end + 1 - state->start;
2403 u32 num_extents = count_max_extents(fs_info, len);
2404 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2406 spin_lock(&BTRFS_I(inode)->lock);
2407 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2408 spin_unlock(&BTRFS_I(inode)->lock);
2410 /* For sanity tests */
2411 if (btrfs_is_testing(fs_info))
2414 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2415 fs_info->delalloc_batch);
2416 spin_lock(&BTRFS_I(inode)->lock);
2417 BTRFS_I(inode)->delalloc_bytes += len;
2418 if (bits & EXTENT_DEFRAG)
2419 BTRFS_I(inode)->defrag_bytes += len;
2420 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2421 &BTRFS_I(inode)->runtime_flags))
2422 btrfs_add_delalloc_inodes(root, inode);
2423 spin_unlock(&BTRFS_I(inode)->lock);
2426 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2427 (bits & EXTENT_DELALLOC_NEW)) {
2428 spin_lock(&BTRFS_I(inode)->lock);
2429 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2431 spin_unlock(&BTRFS_I(inode)->lock);
2436 * Once a range is no longer delalloc this function ensures that proper
2437 * accounting happens.
2439 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2440 struct extent_state *state, u32 bits)
2442 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2443 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2444 u64 len = state->end + 1 - state->start;
2445 u32 num_extents = count_max_extents(fs_info, len);
2447 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2448 spin_lock(&inode->lock);
2449 inode->defrag_bytes -= len;
2450 spin_unlock(&inode->lock);
2454 * set_bit and clear bit hooks normally require _irqsave/restore
2455 * but in this case, we are only testing for the DELALLOC
2456 * bit, which is only set or cleared with irqs on
2458 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2459 struct btrfs_root *root = inode->root;
2460 bool do_list = !btrfs_is_free_space_inode(inode);
2462 spin_lock(&inode->lock);
2463 btrfs_mod_outstanding_extents(inode, -num_extents);
2464 spin_unlock(&inode->lock);
2467 * We don't reserve metadata space for space cache inodes so we
2468 * don't need to call delalloc_release_metadata if there is an
2471 if (bits & EXTENT_CLEAR_META_RESV &&
2472 root != fs_info->tree_root)
2473 btrfs_delalloc_release_metadata(inode, len, false);
2475 /* For sanity tests. */
2476 if (btrfs_is_testing(fs_info))
2479 if (!btrfs_is_data_reloc_root(root) &&
2480 do_list && !(state->state & EXTENT_NORESERVE) &&
2481 (bits & EXTENT_CLEAR_DATA_RESV))
2482 btrfs_free_reserved_data_space_noquota(fs_info, len);
2484 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2485 fs_info->delalloc_batch);
2486 spin_lock(&inode->lock);
2487 inode->delalloc_bytes -= len;
2488 if (do_list && inode->delalloc_bytes == 0 &&
2489 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2490 &inode->runtime_flags))
2491 btrfs_del_delalloc_inode(root, inode);
2492 spin_unlock(&inode->lock);
2495 if ((state->state & EXTENT_DELALLOC_NEW) &&
2496 (bits & EXTENT_DELALLOC_NEW)) {
2497 spin_lock(&inode->lock);
2498 ASSERT(inode->new_delalloc_bytes >= len);
2499 inode->new_delalloc_bytes -= len;
2500 if (bits & EXTENT_ADD_INODE_BYTES)
2501 inode_add_bytes(&inode->vfs_inode, len);
2502 spin_unlock(&inode->lock);
2507 * in order to insert checksums into the metadata in large chunks,
2508 * we wait until bio submission time. All the pages in the bio are
2509 * checksummed and sums are attached onto the ordered extent record.
2511 * At IO completion time the cums attached on the ordered extent record
2512 * are inserted into the btree
2514 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2515 u64 dio_file_offset)
2517 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2521 * Split an extent_map at [start, start + len]
2523 * This function is intended to be used only for extract_ordered_extent().
2525 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2528 struct extent_map_tree *em_tree = &inode->extent_tree;
2529 struct extent_map *em;
2530 struct extent_map *split_pre = NULL;
2531 struct extent_map *split_mid = NULL;
2532 struct extent_map *split_post = NULL;
2534 unsigned long flags;
2537 if (pre == 0 && post == 0)
2540 split_pre = alloc_extent_map();
2542 split_mid = alloc_extent_map();
2544 split_post = alloc_extent_map();
2545 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2550 ASSERT(pre + post < len);
2552 lock_extent(&inode->io_tree, start, start + len - 1);
2553 write_lock(&em_tree->lock);
2554 em = lookup_extent_mapping(em_tree, start, len);
2560 ASSERT(em->len == len);
2561 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2562 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2563 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2564 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2565 ASSERT(!list_empty(&em->list));
2568 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2570 /* First, replace the em with a new extent_map starting from * em->start */
2571 split_pre->start = em->start;
2572 split_pre->len = (pre ? pre : em->len - post);
2573 split_pre->orig_start = split_pre->start;
2574 split_pre->block_start = em->block_start;
2575 split_pre->block_len = split_pre->len;
2576 split_pre->orig_block_len = split_pre->block_len;
2577 split_pre->ram_bytes = split_pre->len;
2578 split_pre->flags = flags;
2579 split_pre->compress_type = em->compress_type;
2580 split_pre->generation = em->generation;
2582 replace_extent_mapping(em_tree, em, split_pre, 1);
2585 * Now we only have an extent_map at:
2586 * [em->start, em->start + pre] if pre != 0
2587 * [em->start, em->start + em->len - post] if pre == 0
2591 /* Insert the middle extent_map */
2592 split_mid->start = em->start + pre;
2593 split_mid->len = em->len - pre - post;
2594 split_mid->orig_start = split_mid->start;
2595 split_mid->block_start = em->block_start + pre;
2596 split_mid->block_len = split_mid->len;
2597 split_mid->orig_block_len = split_mid->block_len;
2598 split_mid->ram_bytes = split_mid->len;
2599 split_mid->flags = flags;
2600 split_mid->compress_type = em->compress_type;
2601 split_mid->generation = em->generation;
2602 add_extent_mapping(em_tree, split_mid, 1);
2606 split_post->start = em->start + em->len - post;
2607 split_post->len = post;
2608 split_post->orig_start = split_post->start;
2609 split_post->block_start = em->block_start + em->len - post;
2610 split_post->block_len = split_post->len;
2611 split_post->orig_block_len = split_post->block_len;
2612 split_post->ram_bytes = split_post->len;
2613 split_post->flags = flags;
2614 split_post->compress_type = em->compress_type;
2615 split_post->generation = em->generation;
2616 add_extent_mapping(em_tree, split_post, 1);
2620 free_extent_map(em);
2621 /* Once for the tree */
2622 free_extent_map(em);
2625 write_unlock(&em_tree->lock);
2626 unlock_extent(&inode->io_tree, start, start + len - 1);
2628 free_extent_map(split_pre);
2629 free_extent_map(split_mid);
2630 free_extent_map(split_post);
2635 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2636 struct bio *bio, loff_t file_offset)
2638 struct btrfs_ordered_extent *ordered;
2639 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2641 u64 len = bio->bi_iter.bi_size;
2642 u64 end = start + len;
2647 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2648 if (WARN_ON_ONCE(!ordered))
2649 return BLK_STS_IOERR;
2651 /* No need to split */
2652 if (ordered->disk_num_bytes == len)
2655 /* We cannot split once end_bio'd ordered extent */
2656 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2661 /* We cannot split a compressed ordered extent */
2662 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2667 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2668 /* bio must be in one ordered extent */
2669 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2674 /* Checksum list should be empty */
2675 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2680 file_len = ordered->num_bytes;
2681 pre = start - ordered->disk_bytenr;
2682 post = ordered_end - end;
2684 ret = btrfs_split_ordered_extent(ordered, pre, post);
2687 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2690 btrfs_put_ordered_extent(ordered);
2692 return errno_to_blk_status(ret);
2695 void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio, int mirror_num)
2697 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2698 struct btrfs_inode *bi = BTRFS_I(inode);
2701 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2702 ret = extract_ordered_extent(bi, bio,
2703 page_offset(bio_first_bvec_all(bio)->bv_page));
2709 * If we need to checksum, and the I/O is not issued by fsync and
2710 * friends, that is ->sync_writers != 0, defer the submission to a
2711 * workqueue to parallelize it.
2713 * Csum items for reloc roots have already been cloned at this point,
2714 * so they are handled as part of the no-checksum case.
2716 if (!(bi->flags & BTRFS_INODE_NODATASUM) &&
2717 !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state) &&
2718 !btrfs_is_data_reloc_root(bi->root)) {
2719 if (!atomic_read(&bi->sync_writers) &&
2720 btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
2721 btrfs_submit_bio_start))
2724 ret = btrfs_csum_one_bio(bi, bio, (u64)-1, false);
2728 btrfs_submit_bio(fs_info, bio, mirror_num);
2732 bio->bi_status = ret;
2737 void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
2738 int mirror_num, enum btrfs_compression_type compress_type)
2740 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2743 if (compress_type != BTRFS_COMPRESS_NONE) {
2745 * btrfs_submit_compressed_read will handle completing the bio
2746 * if there were any errors, so just return here.
2748 btrfs_submit_compressed_read(inode, bio, mirror_num);
2752 /* Save the original iter for read repair */
2753 btrfs_bio(bio)->iter = bio->bi_iter;
2756 * Lookup bio sums does extra checks around whether we need to csum or
2757 * not, which is why we ignore skip_sum here.
2759 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2761 bio->bi_status = ret;
2766 btrfs_submit_bio(fs_info, bio, mirror_num);
2770 * given a list of ordered sums record them in the inode. This happens
2771 * at IO completion time based on sums calculated at bio submission time.
2773 static int add_pending_csums(struct btrfs_trans_handle *trans,
2774 struct list_head *list)
2776 struct btrfs_ordered_sum *sum;
2777 struct btrfs_root *csum_root = NULL;
2780 list_for_each_entry(sum, list, list) {
2781 trans->adding_csums = true;
2783 csum_root = btrfs_csum_root(trans->fs_info,
2785 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2786 trans->adding_csums = false;
2793 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2796 struct extent_state **cached_state)
2798 u64 search_start = start;
2799 const u64 end = start + len - 1;
2801 while (search_start < end) {
2802 const u64 search_len = end - search_start + 1;
2803 struct extent_map *em;
2807 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2811 if (em->block_start != EXTENT_MAP_HOLE)
2815 if (em->start < search_start)
2816 em_len -= search_start - em->start;
2817 if (em_len > search_len)
2818 em_len = search_len;
2820 ret = set_extent_bit(&inode->io_tree, search_start,
2821 search_start + em_len - 1,
2822 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2825 search_start = extent_map_end(em);
2826 free_extent_map(em);
2833 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2834 unsigned int extra_bits,
2835 struct extent_state **cached_state)
2837 WARN_ON(PAGE_ALIGNED(end));
2839 if (start >= i_size_read(&inode->vfs_inode) &&
2840 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2842 * There can't be any extents following eof in this case so just
2843 * set the delalloc new bit for the range directly.
2845 extra_bits |= EXTENT_DELALLOC_NEW;
2849 ret = btrfs_find_new_delalloc_bytes(inode, start,
2856 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2860 /* see btrfs_writepage_start_hook for details on why this is required */
2861 struct btrfs_writepage_fixup {
2863 struct inode *inode;
2864 struct btrfs_work work;
2867 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2869 struct btrfs_writepage_fixup *fixup;
2870 struct btrfs_ordered_extent *ordered;
2871 struct extent_state *cached_state = NULL;
2872 struct extent_changeset *data_reserved = NULL;
2874 struct btrfs_inode *inode;
2878 bool free_delalloc_space = true;
2880 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2882 inode = BTRFS_I(fixup->inode);
2883 page_start = page_offset(page);
2884 page_end = page_offset(page) + PAGE_SIZE - 1;
2887 * This is similar to page_mkwrite, we need to reserve the space before
2888 * we take the page lock.
2890 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2896 * Before we queued this fixup, we took a reference on the page.
2897 * page->mapping may go NULL, but it shouldn't be moved to a different
2900 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2902 * Unfortunately this is a little tricky, either
2904 * 1) We got here and our page had already been dealt with and
2905 * we reserved our space, thus ret == 0, so we need to just
2906 * drop our space reservation and bail. This can happen the
2907 * first time we come into the fixup worker, or could happen
2908 * while waiting for the ordered extent.
2909 * 2) Our page was already dealt with, but we happened to get an
2910 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2911 * this case we obviously don't have anything to release, but
2912 * because the page was already dealt with we don't want to
2913 * mark the page with an error, so make sure we're resetting
2914 * ret to 0. This is why we have this check _before_ the ret
2915 * check, because we do not want to have a surprise ENOSPC
2916 * when the page was already properly dealt with.
2919 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2920 btrfs_delalloc_release_space(inode, data_reserved,
2921 page_start, PAGE_SIZE,
2929 * We can't mess with the page state unless it is locked, so now that
2930 * it is locked bail if we failed to make our space reservation.
2935 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2937 /* already ordered? We're done */
2938 if (PageOrdered(page))
2941 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2943 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2946 btrfs_start_ordered_extent(ordered, 1);
2947 btrfs_put_ordered_extent(ordered);
2951 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2957 * Everything went as planned, we're now the owner of a dirty page with
2958 * delayed allocation bits set and space reserved for our COW
2961 * The page was dirty when we started, nothing should have cleaned it.
2963 BUG_ON(!PageDirty(page));
2964 free_delalloc_space = false;
2966 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2967 if (free_delalloc_space)
2968 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2970 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2975 * We hit ENOSPC or other errors. Update the mapping and page
2976 * to reflect the errors and clean the page.
2978 mapping_set_error(page->mapping, ret);
2979 end_extent_writepage(page, ret, page_start, page_end);
2980 clear_page_dirty_for_io(page);
2983 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2987 extent_changeset_free(data_reserved);
2989 * As a precaution, do a delayed iput in case it would be the last iput
2990 * that could need flushing space. Recursing back to fixup worker would
2993 btrfs_add_delayed_iput(&inode->vfs_inode);
2997 * There are a few paths in the higher layers of the kernel that directly
2998 * set the page dirty bit without asking the filesystem if it is a
2999 * good idea. This causes problems because we want to make sure COW
3000 * properly happens and the data=ordered rules are followed.
3002 * In our case any range that doesn't have the ORDERED bit set
3003 * hasn't been properly setup for IO. We kick off an async process
3004 * to fix it up. The async helper will wait for ordered extents, set
3005 * the delalloc bit and make it safe to write the page.
3007 int btrfs_writepage_cow_fixup(struct page *page)
3009 struct inode *inode = page->mapping->host;
3010 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3011 struct btrfs_writepage_fixup *fixup;
3013 /* This page has ordered extent covering it already */
3014 if (PageOrdered(page))
3018 * PageChecked is set below when we create a fixup worker for this page,
3019 * don't try to create another one if we're already PageChecked()
3021 * The extent_io writepage code will redirty the page if we send back
3024 if (PageChecked(page))
3027 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
3032 * We are already holding a reference to this inode from
3033 * write_cache_pages. We need to hold it because the space reservation
3034 * takes place outside of the page lock, and we can't trust
3035 * page->mapping outside of the page lock.
3038 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3040 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3042 fixup->inode = inode;
3043 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3048 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3049 struct btrfs_inode *inode, u64 file_pos,
3050 struct btrfs_file_extent_item *stack_fi,
3051 const bool update_inode_bytes,
3052 u64 qgroup_reserved)
3054 struct btrfs_root *root = inode->root;
3055 const u64 sectorsize = root->fs_info->sectorsize;
3056 struct btrfs_path *path;
3057 struct extent_buffer *leaf;
3058 struct btrfs_key ins;
3059 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3060 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3061 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3062 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3063 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3064 struct btrfs_drop_extents_args drop_args = { 0 };
3067 path = btrfs_alloc_path();
3072 * we may be replacing one extent in the tree with another.
3073 * The new extent is pinned in the extent map, and we don't want
3074 * to drop it from the cache until it is completely in the btree.
3076 * So, tell btrfs_drop_extents to leave this extent in the cache.
3077 * the caller is expected to unpin it and allow it to be merged
3080 drop_args.path = path;
3081 drop_args.start = file_pos;
3082 drop_args.end = file_pos + num_bytes;
3083 drop_args.replace_extent = true;
3084 drop_args.extent_item_size = sizeof(*stack_fi);
3085 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3089 if (!drop_args.extent_inserted) {
3090 ins.objectid = btrfs_ino(inode);
3091 ins.offset = file_pos;
3092 ins.type = BTRFS_EXTENT_DATA_KEY;
3094 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3099 leaf = path->nodes[0];
3100 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3101 write_extent_buffer(leaf, stack_fi,
3102 btrfs_item_ptr_offset(leaf, path->slots[0]),
3103 sizeof(struct btrfs_file_extent_item));
3105 btrfs_mark_buffer_dirty(leaf);
3106 btrfs_release_path(path);
3109 * If we dropped an inline extent here, we know the range where it is
3110 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3111 * number of bytes only for that range containing the inline extent.
3112 * The remaining of the range will be processed when clearning the
3113 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3115 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3116 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3118 inline_size = drop_args.bytes_found - inline_size;
3119 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3120 drop_args.bytes_found -= inline_size;
3121 num_bytes -= sectorsize;
3124 if (update_inode_bytes)
3125 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3127 ins.objectid = disk_bytenr;
3128 ins.offset = disk_num_bytes;
3129 ins.type = BTRFS_EXTENT_ITEM_KEY;
3131 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3135 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3137 qgroup_reserved, &ins);
3139 btrfs_free_path(path);
3144 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3147 struct btrfs_block_group *cache;
3149 cache = btrfs_lookup_block_group(fs_info, start);
3152 spin_lock(&cache->lock);
3153 cache->delalloc_bytes -= len;
3154 spin_unlock(&cache->lock);
3156 btrfs_put_block_group(cache);
3159 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3160 struct btrfs_ordered_extent *oe)
3162 struct btrfs_file_extent_item stack_fi;
3163 bool update_inode_bytes;
3164 u64 num_bytes = oe->num_bytes;
3165 u64 ram_bytes = oe->ram_bytes;
3167 memset(&stack_fi, 0, sizeof(stack_fi));
3168 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3169 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3170 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3171 oe->disk_num_bytes);
3172 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3173 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3174 num_bytes = oe->truncated_len;
3175 ram_bytes = num_bytes;
3177 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3178 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3179 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3180 /* Encryption and other encoding is reserved and all 0 */
3183 * For delalloc, when completing an ordered extent we update the inode's
3184 * bytes when clearing the range in the inode's io tree, so pass false
3185 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3186 * except if the ordered extent was truncated.
3188 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3189 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3190 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3192 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3193 oe->file_offset, &stack_fi,
3194 update_inode_bytes, oe->qgroup_rsv);
3198 * As ordered data IO finishes, this gets called so we can finish
3199 * an ordered extent if the range of bytes in the file it covers are
3202 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3204 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3205 struct btrfs_root *root = inode->root;
3206 struct btrfs_fs_info *fs_info = root->fs_info;
3207 struct btrfs_trans_handle *trans = NULL;
3208 struct extent_io_tree *io_tree = &inode->io_tree;
3209 struct extent_state *cached_state = NULL;
3211 int compress_type = 0;
3213 u64 logical_len = ordered_extent->num_bytes;
3214 bool freespace_inode;
3215 bool truncated = false;
3216 bool clear_reserved_extent = true;
3217 unsigned int clear_bits = EXTENT_DEFRAG;
3219 start = ordered_extent->file_offset;
3220 end = start + ordered_extent->num_bytes - 1;
3222 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3223 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3224 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3225 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3226 clear_bits |= EXTENT_DELALLOC_NEW;
3228 freespace_inode = btrfs_is_free_space_inode(inode);
3230 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3235 /* A valid bdev implies a write on a sequential zone */
3236 if (ordered_extent->bdev) {
3237 btrfs_rewrite_logical_zoned(ordered_extent);
3238 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3239 ordered_extent->disk_num_bytes);
3242 btrfs_free_io_failure_record(inode, start, end);
3244 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3246 logical_len = ordered_extent->truncated_len;
3247 /* Truncated the entire extent, don't bother adding */
3252 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3253 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3255 btrfs_inode_safe_disk_i_size_write(inode, 0);
3256 if (freespace_inode)
3257 trans = btrfs_join_transaction_spacecache(root);
3259 trans = btrfs_join_transaction(root);
3260 if (IS_ERR(trans)) {
3261 ret = PTR_ERR(trans);
3265 trans->block_rsv = &inode->block_rsv;
3266 ret = btrfs_update_inode_fallback(trans, root, inode);
3267 if (ret) /* -ENOMEM or corruption */
3268 btrfs_abort_transaction(trans, ret);
3272 clear_bits |= EXTENT_LOCKED;
3273 lock_extent_bits(io_tree, start, end, &cached_state);
3275 if (freespace_inode)
3276 trans = btrfs_join_transaction_spacecache(root);
3278 trans = btrfs_join_transaction(root);
3279 if (IS_ERR(trans)) {
3280 ret = PTR_ERR(trans);
3285 trans->block_rsv = &inode->block_rsv;
3287 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3288 compress_type = ordered_extent->compress_type;
3289 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3290 BUG_ON(compress_type);
3291 ret = btrfs_mark_extent_written(trans, inode,
3292 ordered_extent->file_offset,
3293 ordered_extent->file_offset +
3295 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3296 ordered_extent->disk_num_bytes);
3298 BUG_ON(root == fs_info->tree_root);
3299 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3301 clear_reserved_extent = false;
3302 btrfs_release_delalloc_bytes(fs_info,
3303 ordered_extent->disk_bytenr,
3304 ordered_extent->disk_num_bytes);
3307 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3308 ordered_extent->num_bytes, trans->transid);
3310 btrfs_abort_transaction(trans, ret);
3314 ret = add_pending_csums(trans, &ordered_extent->list);
3316 btrfs_abort_transaction(trans, ret);
3321 * If this is a new delalloc range, clear its new delalloc flag to
3322 * update the inode's number of bytes. This needs to be done first
3323 * before updating the inode item.
3325 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3326 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3327 clear_extent_bit(&inode->io_tree, start, end,
3328 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3329 0, 0, &cached_state);
3331 btrfs_inode_safe_disk_i_size_write(inode, 0);
3332 ret = btrfs_update_inode_fallback(trans, root, inode);
3333 if (ret) { /* -ENOMEM or corruption */
3334 btrfs_abort_transaction(trans, ret);
3339 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3340 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3344 btrfs_end_transaction(trans);
3346 if (ret || truncated) {
3347 u64 unwritten_start = start;
3350 * If we failed to finish this ordered extent for any reason we
3351 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3352 * extent, and mark the inode with the error if it wasn't
3353 * already set. Any error during writeback would have already
3354 * set the mapping error, so we need to set it if we're the ones
3355 * marking this ordered extent as failed.
3357 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3358 &ordered_extent->flags))
3359 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3362 unwritten_start += logical_len;
3363 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3365 /* Drop the cache for the part of the extent we didn't write. */
3366 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3369 * If the ordered extent had an IOERR or something else went
3370 * wrong we need to return the space for this ordered extent
3371 * back to the allocator. We only free the extent in the
3372 * truncated case if we didn't write out the extent at all.
3374 * If we made it past insert_reserved_file_extent before we
3375 * errored out then we don't need to do this as the accounting
3376 * has already been done.
3378 if ((ret || !logical_len) &&
3379 clear_reserved_extent &&
3380 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3381 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3383 * Discard the range before returning it back to the
3386 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3387 btrfs_discard_extent(fs_info,
3388 ordered_extent->disk_bytenr,
3389 ordered_extent->disk_num_bytes,
3391 btrfs_free_reserved_extent(fs_info,
3392 ordered_extent->disk_bytenr,
3393 ordered_extent->disk_num_bytes, 1);
3398 * This needs to be done to make sure anybody waiting knows we are done
3399 * updating everything for this ordered extent.
3401 btrfs_remove_ordered_extent(inode, ordered_extent);
3404 btrfs_put_ordered_extent(ordered_extent);
3405 /* once for the tree */
3406 btrfs_put_ordered_extent(ordered_extent);
3411 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3412 struct page *page, u64 start,
3413 u64 end, bool uptodate)
3415 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3417 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate);
3421 * Verify the checksum for a single sector without any extra action that depend
3422 * on the type of I/O.
3424 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3425 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3427 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3430 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3432 shash->tfm = fs_info->csum_shash;
3434 kaddr = kmap_local_page(page) + pgoff;
3435 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3436 kunmap_local(kaddr);
3438 if (memcmp(csum, csum_expected, fs_info->csum_size))
3444 * check_data_csum - verify checksum of one sector of uncompressed data
3446 * @bbio: btrfs_bio which contains the csum
3447 * @bio_offset: offset to the beginning of the bio (in bytes)
3448 * @page: page where is the data to be verified
3449 * @pgoff: offset inside the page
3451 * The length of such check is always one sector size.
3453 * When csum mismatch is detected, we will also report the error and fill the
3454 * corrupted range with zero. (Thus it needs the extra parameters)
3456 int btrfs_check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3457 u32 bio_offset, struct page *page, u32 pgoff)
3459 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3460 u32 len = fs_info->sectorsize;
3462 u8 csum[BTRFS_CSUM_SIZE];
3464 ASSERT(pgoff + len <= PAGE_SIZE);
3466 csum_expected = btrfs_csum_ptr(fs_info, bbio->csum, bio_offset);
3468 if (btrfs_check_sector_csum(fs_info, page, pgoff, csum, csum_expected))
3473 btrfs_print_data_csum_error(BTRFS_I(inode),
3474 bbio->file_offset + bio_offset,
3475 csum, csum_expected, bbio->mirror_num);
3477 btrfs_dev_stat_inc_and_print(bbio->device,
3478 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3479 memzero_page(page, pgoff, len);
3484 * When reads are done, we need to check csums to verify the data is correct.
3485 * if there's a match, we allow the bio to finish. If not, the code in
3486 * extent_io.c will try to find good copies for us.
3488 * @bio_offset: offset to the beginning of the bio (in bytes)
3489 * @start: file offset of the range start
3490 * @end: file offset of the range end (inclusive)
3492 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3495 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3496 u32 bio_offset, struct page *page,
3499 struct inode *inode = page->mapping->host;
3500 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3501 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3502 struct btrfs_root *root = BTRFS_I(inode)->root;
3503 const u32 sectorsize = root->fs_info->sectorsize;
3505 unsigned int result = 0;
3508 * This only happens for NODATASUM or compressed read.
3509 * Normally this should be covered by above check for compressed read
3510 * or the next check for NODATASUM. Just do a quicker exit here.
3512 if (bbio->csum == NULL)
3515 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3518 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3521 ASSERT(page_offset(page) <= start &&
3522 end <= page_offset(page) + PAGE_SIZE - 1);
3523 for (pg_off = offset_in_page(start);
3524 pg_off < offset_in_page(end);
3525 pg_off += sectorsize, bio_offset += sectorsize) {
3526 u64 file_offset = pg_off + page_offset(page);
3529 if (btrfs_is_data_reloc_root(root) &&
3530 test_range_bit(io_tree, file_offset,
3531 file_offset + sectorsize - 1,
3532 EXTENT_NODATASUM, 1, NULL)) {
3533 /* Skip the range without csum for data reloc inode */
3534 clear_extent_bits(io_tree, file_offset,
3535 file_offset + sectorsize - 1,
3539 ret = btrfs_check_data_csum(inode, bbio, bio_offset, page, pg_off);
3541 const int nr_bit = (pg_off - offset_in_page(start)) >>
3542 root->fs_info->sectorsize_bits;
3544 result |= (1U << nr_bit);
3551 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3553 * @inode: The inode we want to perform iput on
3555 * This function uses the generic vfs_inode::i_count to track whether we should
3556 * just decrement it (in case it's > 1) or if this is the last iput then link
3557 * the inode to the delayed iput machinery. Delayed iputs are processed at
3558 * transaction commit time/superblock commit/cleaner kthread.
3560 void btrfs_add_delayed_iput(struct inode *inode)
3562 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3563 struct btrfs_inode *binode = BTRFS_I(inode);
3565 if (atomic_add_unless(&inode->i_count, -1, 1))
3568 atomic_inc(&fs_info->nr_delayed_iputs);
3569 spin_lock(&fs_info->delayed_iput_lock);
3570 ASSERT(list_empty(&binode->delayed_iput));
3571 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3572 spin_unlock(&fs_info->delayed_iput_lock);
3573 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3574 wake_up_process(fs_info->cleaner_kthread);
3577 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3578 struct btrfs_inode *inode)
3580 list_del_init(&inode->delayed_iput);
3581 spin_unlock(&fs_info->delayed_iput_lock);
3582 iput(&inode->vfs_inode);
3583 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3584 wake_up(&fs_info->delayed_iputs_wait);
3585 spin_lock(&fs_info->delayed_iput_lock);
3588 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3589 struct btrfs_inode *inode)
3591 if (!list_empty(&inode->delayed_iput)) {
3592 spin_lock(&fs_info->delayed_iput_lock);
3593 if (!list_empty(&inode->delayed_iput))
3594 run_delayed_iput_locked(fs_info, inode);
3595 spin_unlock(&fs_info->delayed_iput_lock);
3599 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3602 spin_lock(&fs_info->delayed_iput_lock);
3603 while (!list_empty(&fs_info->delayed_iputs)) {
3604 struct btrfs_inode *inode;
3606 inode = list_first_entry(&fs_info->delayed_iputs,
3607 struct btrfs_inode, delayed_iput);
3608 run_delayed_iput_locked(fs_info, inode);
3609 cond_resched_lock(&fs_info->delayed_iput_lock);
3611 spin_unlock(&fs_info->delayed_iput_lock);
3615 * Wait for flushing all delayed iputs
3617 * @fs_info: the filesystem
3619 * This will wait on any delayed iputs that are currently running with KILLABLE
3620 * set. Once they are all done running we will return, unless we are killed in
3621 * which case we return EINTR. This helps in user operations like fallocate etc
3622 * that might get blocked on the iputs.
3624 * Return EINTR if we were killed, 0 if nothing's pending
3626 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3628 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3629 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3636 * This creates an orphan entry for the given inode in case something goes wrong
3637 * in the middle of an unlink.
3639 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3640 struct btrfs_inode *inode)
3644 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3645 if (ret && ret != -EEXIST) {
3646 btrfs_abort_transaction(trans, ret);
3654 * We have done the delete so we can go ahead and remove the orphan item for
3655 * this particular inode.
3657 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3658 struct btrfs_inode *inode)
3660 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3664 * this cleans up any orphans that may be left on the list from the last use
3667 int btrfs_orphan_cleanup(struct btrfs_root *root)
3669 struct btrfs_fs_info *fs_info = root->fs_info;
3670 struct btrfs_path *path;
3671 struct extent_buffer *leaf;
3672 struct btrfs_key key, found_key;
3673 struct btrfs_trans_handle *trans;
3674 struct inode *inode;
3675 u64 last_objectid = 0;
3676 int ret = 0, nr_unlink = 0;
3678 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3681 path = btrfs_alloc_path();
3686 path->reada = READA_BACK;
3688 key.objectid = BTRFS_ORPHAN_OBJECTID;
3689 key.type = BTRFS_ORPHAN_ITEM_KEY;
3690 key.offset = (u64)-1;
3693 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3698 * if ret == 0 means we found what we were searching for, which
3699 * is weird, but possible, so only screw with path if we didn't
3700 * find the key and see if we have stuff that matches
3704 if (path->slots[0] == 0)
3709 /* pull out the item */
3710 leaf = path->nodes[0];
3711 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3713 /* make sure the item matches what we want */
3714 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3716 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3719 /* release the path since we're done with it */
3720 btrfs_release_path(path);
3723 * this is where we are basically btrfs_lookup, without the
3724 * crossing root thing. we store the inode number in the
3725 * offset of the orphan item.
3728 if (found_key.offset == last_objectid) {
3730 "Error removing orphan entry, stopping orphan cleanup");
3735 last_objectid = found_key.offset;
3737 found_key.objectid = found_key.offset;
3738 found_key.type = BTRFS_INODE_ITEM_KEY;
3739 found_key.offset = 0;
3740 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3741 ret = PTR_ERR_OR_ZERO(inode);
3742 if (ret && ret != -ENOENT)
3745 if (ret == -ENOENT && root == fs_info->tree_root) {
3746 struct btrfs_root *dead_root;
3747 int is_dead_root = 0;
3750 * This is an orphan in the tree root. Currently these
3751 * could come from 2 sources:
3752 * a) a root (snapshot/subvolume) deletion in progress
3753 * b) a free space cache inode
3754 * We need to distinguish those two, as the orphan item
3755 * for a root must not get deleted before the deletion
3756 * of the snapshot/subvolume's tree completes.
3758 * btrfs_find_orphan_roots() ran before us, which has
3759 * found all deleted roots and loaded them into
3760 * fs_info->fs_roots_radix. So here we can find if an
3761 * orphan item corresponds to a deleted root by looking
3762 * up the root from that radix tree.
3765 spin_lock(&fs_info->fs_roots_radix_lock);
3766 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3767 (unsigned long)found_key.objectid);
3768 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3770 spin_unlock(&fs_info->fs_roots_radix_lock);
3773 /* prevent this orphan from being found again */
3774 key.offset = found_key.objectid - 1;
3781 * If we have an inode with links, there are a couple of
3784 * 1. We were halfway through creating fsverity metadata for the
3785 * file. In that case, the orphan item represents incomplete
3786 * fsverity metadata which must be cleaned up with
3787 * btrfs_drop_verity_items and deleting the orphan item.
3789 * 2. Old kernels (before v3.12) used to create an
3790 * orphan item for truncate indicating that there were possibly
3791 * extent items past i_size that needed to be deleted. In v3.12,
3792 * truncate was changed to update i_size in sync with the extent
3793 * items, but the (useless) orphan item was still created. Since
3794 * v4.18, we don't create the orphan item for truncate at all.
3796 * So, this item could mean that we need to do a truncate, but
3797 * only if this filesystem was last used on a pre-v3.12 kernel
3798 * and was not cleanly unmounted. The odds of that are quite
3799 * slim, and it's a pain to do the truncate now, so just delete
3802 * It's also possible that this orphan item was supposed to be
3803 * deleted but wasn't. The inode number may have been reused,
3804 * but either way, we can delete the orphan item.
3806 if (ret == -ENOENT || inode->i_nlink) {
3808 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3813 trans = btrfs_start_transaction(root, 1);
3814 if (IS_ERR(trans)) {
3815 ret = PTR_ERR(trans);
3818 btrfs_debug(fs_info, "auto deleting %Lu",
3819 found_key.objectid);
3820 ret = btrfs_del_orphan_item(trans, root,
3821 found_key.objectid);
3822 btrfs_end_transaction(trans);
3830 /* this will do delete_inode and everything for us */
3833 /* release the path since we're done with it */
3834 btrfs_release_path(path);
3836 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3837 trans = btrfs_join_transaction(root);
3839 btrfs_end_transaction(trans);
3843 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3847 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3848 btrfs_free_path(path);
3853 * very simple check to peek ahead in the leaf looking for xattrs. If we
3854 * don't find any xattrs, we know there can't be any acls.
3856 * slot is the slot the inode is in, objectid is the objectid of the inode
3858 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3859 int slot, u64 objectid,
3860 int *first_xattr_slot)
3862 u32 nritems = btrfs_header_nritems(leaf);
3863 struct btrfs_key found_key;
3864 static u64 xattr_access = 0;
3865 static u64 xattr_default = 0;
3868 if (!xattr_access) {
3869 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3870 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3871 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3872 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3876 *first_xattr_slot = -1;
3877 while (slot < nritems) {
3878 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3880 /* we found a different objectid, there must not be acls */
3881 if (found_key.objectid != objectid)
3884 /* we found an xattr, assume we've got an acl */
3885 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3886 if (*first_xattr_slot == -1)
3887 *first_xattr_slot = slot;
3888 if (found_key.offset == xattr_access ||
3889 found_key.offset == xattr_default)
3894 * we found a key greater than an xattr key, there can't
3895 * be any acls later on
3897 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3904 * it goes inode, inode backrefs, xattrs, extents,
3905 * so if there are a ton of hard links to an inode there can
3906 * be a lot of backrefs. Don't waste time searching too hard,
3907 * this is just an optimization
3912 /* we hit the end of the leaf before we found an xattr or
3913 * something larger than an xattr. We have to assume the inode
3916 if (*first_xattr_slot == -1)
3917 *first_xattr_slot = slot;
3922 * read an inode from the btree into the in-memory inode
3924 static int btrfs_read_locked_inode(struct inode *inode,
3925 struct btrfs_path *in_path)
3927 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3928 struct btrfs_path *path = in_path;
3929 struct extent_buffer *leaf;
3930 struct btrfs_inode_item *inode_item;
3931 struct btrfs_root *root = BTRFS_I(inode)->root;
3932 struct btrfs_key location;
3937 bool filled = false;
3938 int first_xattr_slot;
3940 ret = btrfs_fill_inode(inode, &rdev);
3945 path = btrfs_alloc_path();
3950 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3952 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3954 if (path != in_path)
3955 btrfs_free_path(path);
3959 leaf = path->nodes[0];
3964 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3965 struct btrfs_inode_item);
3966 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3967 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3968 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3969 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3970 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3971 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3972 round_up(i_size_read(inode), fs_info->sectorsize));
3974 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3975 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3977 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3978 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3980 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3981 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3983 BTRFS_I(inode)->i_otime.tv_sec =
3984 btrfs_timespec_sec(leaf, &inode_item->otime);
3985 BTRFS_I(inode)->i_otime.tv_nsec =
3986 btrfs_timespec_nsec(leaf, &inode_item->otime);
3988 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3989 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3990 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3992 inode_set_iversion_queried(inode,
3993 btrfs_inode_sequence(leaf, inode_item));
3994 inode->i_generation = BTRFS_I(inode)->generation;
3996 rdev = btrfs_inode_rdev(leaf, inode_item);
3998 BTRFS_I(inode)->index_cnt = (u64)-1;
3999 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
4000 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
4004 * If we were modified in the current generation and evicted from memory
4005 * and then re-read we need to do a full sync since we don't have any
4006 * idea about which extents were modified before we were evicted from
4009 * This is required for both inode re-read from disk and delayed inode
4010 * in delayed_nodes_tree.
4012 if (BTRFS_I(inode)->last_trans == fs_info->generation)
4013 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4014 &BTRFS_I(inode)->runtime_flags);
4017 * We don't persist the id of the transaction where an unlink operation
4018 * against the inode was last made. So here we assume the inode might
4019 * have been evicted, and therefore the exact value of last_unlink_trans
4020 * lost, and set it to last_trans to avoid metadata inconsistencies
4021 * between the inode and its parent if the inode is fsync'ed and the log
4022 * replayed. For example, in the scenario:
4025 * ln mydir/foo mydir/bar
4028 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
4029 * xfs_io -c fsync mydir/foo
4031 * mount fs, triggers fsync log replay
4033 * We must make sure that when we fsync our inode foo we also log its
4034 * parent inode, otherwise after log replay the parent still has the
4035 * dentry with the "bar" name but our inode foo has a link count of 1
4036 * and doesn't have an inode ref with the name "bar" anymore.
4038 * Setting last_unlink_trans to last_trans is a pessimistic approach,
4039 * but it guarantees correctness at the expense of occasional full
4040 * transaction commits on fsync if our inode is a directory, or if our
4041 * inode is not a directory, logging its parent unnecessarily.
4043 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
4046 * Same logic as for last_unlink_trans. We don't persist the generation
4047 * of the last transaction where this inode was used for a reflink
4048 * operation, so after eviction and reloading the inode we must be
4049 * pessimistic and assume the last transaction that modified the inode.
4051 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
4054 if (inode->i_nlink != 1 ||
4055 path->slots[0] >= btrfs_header_nritems(leaf))
4058 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
4059 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
4062 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4063 if (location.type == BTRFS_INODE_REF_KEY) {
4064 struct btrfs_inode_ref *ref;
4066 ref = (struct btrfs_inode_ref *)ptr;
4067 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
4068 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
4069 struct btrfs_inode_extref *extref;
4071 extref = (struct btrfs_inode_extref *)ptr;
4072 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4077 * try to precache a NULL acl entry for files that don't have
4078 * any xattrs or acls
4080 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4081 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4082 if (first_xattr_slot != -1) {
4083 path->slots[0] = first_xattr_slot;
4084 ret = btrfs_load_inode_props(inode, path);
4087 "error loading props for ino %llu (root %llu): %d",
4088 btrfs_ino(BTRFS_I(inode)),
4089 root->root_key.objectid, ret);
4091 if (path != in_path)
4092 btrfs_free_path(path);
4095 cache_no_acl(inode);
4097 switch (inode->i_mode & S_IFMT) {
4099 inode->i_mapping->a_ops = &btrfs_aops;
4100 inode->i_fop = &btrfs_file_operations;
4101 inode->i_op = &btrfs_file_inode_operations;
4104 inode->i_fop = &btrfs_dir_file_operations;
4105 inode->i_op = &btrfs_dir_inode_operations;
4108 inode->i_op = &btrfs_symlink_inode_operations;
4109 inode_nohighmem(inode);
4110 inode->i_mapping->a_ops = &btrfs_aops;
4113 inode->i_op = &btrfs_special_inode_operations;
4114 init_special_inode(inode, inode->i_mode, rdev);
4118 btrfs_sync_inode_flags_to_i_flags(inode);
4123 * given a leaf and an inode, copy the inode fields into the leaf
4125 static void fill_inode_item(struct btrfs_trans_handle *trans,
4126 struct extent_buffer *leaf,
4127 struct btrfs_inode_item *item,
4128 struct inode *inode)
4130 struct btrfs_map_token token;
4133 btrfs_init_map_token(&token, leaf);
4135 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4136 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4137 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4138 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4139 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4141 btrfs_set_token_timespec_sec(&token, &item->atime,
4142 inode->i_atime.tv_sec);
4143 btrfs_set_token_timespec_nsec(&token, &item->atime,
4144 inode->i_atime.tv_nsec);
4146 btrfs_set_token_timespec_sec(&token, &item->mtime,
4147 inode->i_mtime.tv_sec);
4148 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4149 inode->i_mtime.tv_nsec);
4151 btrfs_set_token_timespec_sec(&token, &item->ctime,
4152 inode->i_ctime.tv_sec);
4153 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4154 inode->i_ctime.tv_nsec);
4156 btrfs_set_token_timespec_sec(&token, &item->otime,
4157 BTRFS_I(inode)->i_otime.tv_sec);
4158 btrfs_set_token_timespec_nsec(&token, &item->otime,
4159 BTRFS_I(inode)->i_otime.tv_nsec);
4161 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4162 btrfs_set_token_inode_generation(&token, item,
4163 BTRFS_I(inode)->generation);
4164 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4165 btrfs_set_token_inode_transid(&token, item, trans->transid);
4166 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4167 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4168 BTRFS_I(inode)->ro_flags);
4169 btrfs_set_token_inode_flags(&token, item, flags);
4170 btrfs_set_token_inode_block_group(&token, item, 0);
4174 * copy everything in the in-memory inode into the btree.
4176 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4177 struct btrfs_root *root,
4178 struct btrfs_inode *inode)
4180 struct btrfs_inode_item *inode_item;
4181 struct btrfs_path *path;
4182 struct extent_buffer *leaf;
4185 path = btrfs_alloc_path();
4189 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4196 leaf = path->nodes[0];
4197 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4198 struct btrfs_inode_item);
4200 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4201 btrfs_mark_buffer_dirty(leaf);
4202 btrfs_set_inode_last_trans(trans, inode);
4205 btrfs_free_path(path);
4210 * copy everything in the in-memory inode into the btree.
4212 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4213 struct btrfs_root *root,
4214 struct btrfs_inode *inode)
4216 struct btrfs_fs_info *fs_info = root->fs_info;
4220 * If the inode is a free space inode, we can deadlock during commit
4221 * if we put it into the delayed code.
4223 * The data relocation inode should also be directly updated
4226 if (!btrfs_is_free_space_inode(inode)
4227 && !btrfs_is_data_reloc_root(root)
4228 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4229 btrfs_update_root_times(trans, root);
4231 ret = btrfs_delayed_update_inode(trans, root, inode);
4233 btrfs_set_inode_last_trans(trans, inode);
4237 return btrfs_update_inode_item(trans, root, inode);
4240 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4241 struct btrfs_root *root, struct btrfs_inode *inode)
4245 ret = btrfs_update_inode(trans, root, inode);
4247 return btrfs_update_inode_item(trans, root, inode);
4252 * unlink helper that gets used here in inode.c and in the tree logging
4253 * recovery code. It remove a link in a directory with a given name, and
4254 * also drops the back refs in the inode to the directory
4256 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4257 struct btrfs_inode *dir,
4258 struct btrfs_inode *inode,
4259 const char *name, int name_len,
4260 struct btrfs_rename_ctx *rename_ctx)
4262 struct btrfs_root *root = dir->root;
4263 struct btrfs_fs_info *fs_info = root->fs_info;
4264 struct btrfs_path *path;
4266 struct btrfs_dir_item *di;
4268 u64 ino = btrfs_ino(inode);
4269 u64 dir_ino = btrfs_ino(dir);
4271 path = btrfs_alloc_path();
4277 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4278 name, name_len, -1);
4279 if (IS_ERR_OR_NULL(di)) {
4280 ret = di ? PTR_ERR(di) : -ENOENT;
4283 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4286 btrfs_release_path(path);
4289 * If we don't have dir index, we have to get it by looking up
4290 * the inode ref, since we get the inode ref, remove it directly,
4291 * it is unnecessary to do delayed deletion.
4293 * But if we have dir index, needn't search inode ref to get it.
4294 * Since the inode ref is close to the inode item, it is better
4295 * that we delay to delete it, and just do this deletion when
4296 * we update the inode item.
4298 if (inode->dir_index) {
4299 ret = btrfs_delayed_delete_inode_ref(inode);
4301 index = inode->dir_index;
4306 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4310 "failed to delete reference to %.*s, inode %llu parent %llu",
4311 name_len, name, ino, dir_ino);
4312 btrfs_abort_transaction(trans, ret);
4317 rename_ctx->index = index;
4319 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4321 btrfs_abort_transaction(trans, ret);
4326 * If we are in a rename context, we don't need to update anything in the
4327 * log. That will be done later during the rename by btrfs_log_new_name().
4328 * Besides that, doing it here would only cause extra unnecessary btree
4329 * operations on the log tree, increasing latency for applications.
4332 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4334 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4339 * If we have a pending delayed iput we could end up with the final iput
4340 * being run in btrfs-cleaner context. If we have enough of these built
4341 * up we can end up burning a lot of time in btrfs-cleaner without any
4342 * way to throttle the unlinks. Since we're currently holding a ref on
4343 * the inode we can run the delayed iput here without any issues as the
4344 * final iput won't be done until after we drop the ref we're currently
4347 btrfs_run_delayed_iput(fs_info, inode);
4349 btrfs_free_path(path);
4353 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4354 inode_inc_iversion(&inode->vfs_inode);
4355 inode_inc_iversion(&dir->vfs_inode);
4356 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4357 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4358 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4359 ret = btrfs_update_inode(trans, root, dir);
4364 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4365 struct btrfs_inode *dir, struct btrfs_inode *inode,
4366 const char *name, int name_len)
4369 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4371 drop_nlink(&inode->vfs_inode);
4372 ret = btrfs_update_inode(trans, inode->root, inode);
4378 * helper to start transaction for unlink and rmdir.
4380 * unlink and rmdir are special in btrfs, they do not always free space, so
4381 * if we cannot make our reservations the normal way try and see if there is
4382 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4383 * allow the unlink to occur.
4385 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4387 struct btrfs_root *root = BTRFS_I(dir)->root;
4390 * 1 for the possible orphan item
4391 * 1 for the dir item
4392 * 1 for the dir index
4393 * 1 for the inode ref
4395 * 1 for the parent inode
4397 return btrfs_start_transaction_fallback_global_rsv(root, 6);
4400 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4402 struct btrfs_trans_handle *trans;
4403 struct inode *inode = d_inode(dentry);
4406 trans = __unlink_start_trans(dir);
4408 return PTR_ERR(trans);
4410 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4413 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4414 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4415 dentry->d_name.len);
4419 if (inode->i_nlink == 0) {
4420 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4426 btrfs_end_transaction(trans);
4427 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4431 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4432 struct inode *dir, struct dentry *dentry)
4434 struct btrfs_root *root = BTRFS_I(dir)->root;
4435 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4436 struct btrfs_path *path;
4437 struct extent_buffer *leaf;
4438 struct btrfs_dir_item *di;
4439 struct btrfs_key key;
4440 const char *name = dentry->d_name.name;
4441 int name_len = dentry->d_name.len;
4445 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4447 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4448 objectid = inode->root->root_key.objectid;
4449 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4450 objectid = inode->location.objectid;
4456 path = btrfs_alloc_path();
4460 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4461 name, name_len, -1);
4462 if (IS_ERR_OR_NULL(di)) {
4463 ret = di ? PTR_ERR(di) : -ENOENT;
4467 leaf = path->nodes[0];
4468 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4469 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4470 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4472 btrfs_abort_transaction(trans, ret);
4475 btrfs_release_path(path);
4478 * This is a placeholder inode for a subvolume we didn't have a
4479 * reference to at the time of the snapshot creation. In the meantime
4480 * we could have renamed the real subvol link into our snapshot, so
4481 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4482 * Instead simply lookup the dir_index_item for this entry so we can
4483 * remove it. Otherwise we know we have a ref to the root and we can
4484 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4486 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4487 di = btrfs_search_dir_index_item(root, path, dir_ino,
4489 if (IS_ERR_OR_NULL(di)) {
4494 btrfs_abort_transaction(trans, ret);
4498 leaf = path->nodes[0];
4499 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4501 btrfs_release_path(path);
4503 ret = btrfs_del_root_ref(trans, objectid,
4504 root->root_key.objectid, dir_ino,
4505 &index, name, name_len);
4507 btrfs_abort_transaction(trans, ret);
4512 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4514 btrfs_abort_transaction(trans, ret);
4518 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4519 inode_inc_iversion(dir);
4520 dir->i_mtime = current_time(dir);
4521 dir->i_ctime = dir->i_mtime;
4522 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4524 btrfs_abort_transaction(trans, ret);
4526 btrfs_free_path(path);
4531 * Helper to check if the subvolume references other subvolumes or if it's
4534 static noinline int may_destroy_subvol(struct btrfs_root *root)
4536 struct btrfs_fs_info *fs_info = root->fs_info;
4537 struct btrfs_path *path;
4538 struct btrfs_dir_item *di;
4539 struct btrfs_key key;
4543 path = btrfs_alloc_path();
4547 /* Make sure this root isn't set as the default subvol */
4548 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4549 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4550 dir_id, "default", 7, 0);
4551 if (di && !IS_ERR(di)) {
4552 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4553 if (key.objectid == root->root_key.objectid) {
4556 "deleting default subvolume %llu is not allowed",
4560 btrfs_release_path(path);
4563 key.objectid = root->root_key.objectid;
4564 key.type = BTRFS_ROOT_REF_KEY;
4565 key.offset = (u64)-1;
4567 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4573 if (path->slots[0] > 0) {
4575 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4576 if (key.objectid == root->root_key.objectid &&
4577 key.type == BTRFS_ROOT_REF_KEY)
4581 btrfs_free_path(path);
4585 /* Delete all dentries for inodes belonging to the root */
4586 static void btrfs_prune_dentries(struct btrfs_root *root)
4588 struct btrfs_fs_info *fs_info = root->fs_info;
4589 struct rb_node *node;
4590 struct rb_node *prev;
4591 struct btrfs_inode *entry;
4592 struct inode *inode;
4595 if (!BTRFS_FS_ERROR(fs_info))
4596 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4598 spin_lock(&root->inode_lock);
4600 node = root->inode_tree.rb_node;
4604 entry = rb_entry(node, struct btrfs_inode, rb_node);
4606 if (objectid < btrfs_ino(entry))
4607 node = node->rb_left;
4608 else if (objectid > btrfs_ino(entry))
4609 node = node->rb_right;
4615 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4616 if (objectid <= btrfs_ino(entry)) {
4620 prev = rb_next(prev);
4624 entry = rb_entry(node, struct btrfs_inode, rb_node);
4625 objectid = btrfs_ino(entry) + 1;
4626 inode = igrab(&entry->vfs_inode);
4628 spin_unlock(&root->inode_lock);
4629 if (atomic_read(&inode->i_count) > 1)
4630 d_prune_aliases(inode);
4632 * btrfs_drop_inode will have it removed from the inode
4633 * cache when its usage count hits zero.
4637 spin_lock(&root->inode_lock);
4641 if (cond_resched_lock(&root->inode_lock))
4644 node = rb_next(node);
4646 spin_unlock(&root->inode_lock);
4649 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4651 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4652 struct btrfs_root *root = BTRFS_I(dir)->root;
4653 struct inode *inode = d_inode(dentry);
4654 struct btrfs_root *dest = BTRFS_I(inode)->root;
4655 struct btrfs_trans_handle *trans;
4656 struct btrfs_block_rsv block_rsv;
4661 * Don't allow to delete a subvolume with send in progress. This is
4662 * inside the inode lock so the error handling that has to drop the bit
4663 * again is not run concurrently.
4665 spin_lock(&dest->root_item_lock);
4666 if (dest->send_in_progress) {
4667 spin_unlock(&dest->root_item_lock);
4669 "attempt to delete subvolume %llu during send",
4670 dest->root_key.objectid);
4673 if (atomic_read(&dest->nr_swapfiles)) {
4674 spin_unlock(&dest->root_item_lock);
4676 "attempt to delete subvolume %llu with active swapfile",
4677 root->root_key.objectid);
4680 root_flags = btrfs_root_flags(&dest->root_item);
4681 btrfs_set_root_flags(&dest->root_item,
4682 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4683 spin_unlock(&dest->root_item_lock);
4685 down_write(&fs_info->subvol_sem);
4687 ret = may_destroy_subvol(dest);
4691 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4693 * One for dir inode,
4694 * two for dir entries,
4695 * two for root ref/backref.
4697 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4701 trans = btrfs_start_transaction(root, 0);
4702 if (IS_ERR(trans)) {
4703 ret = PTR_ERR(trans);
4706 trans->block_rsv = &block_rsv;
4707 trans->bytes_reserved = block_rsv.size;
4709 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4711 ret = btrfs_unlink_subvol(trans, dir, dentry);
4713 btrfs_abort_transaction(trans, ret);
4717 ret = btrfs_record_root_in_trans(trans, dest);
4719 btrfs_abort_transaction(trans, ret);
4723 memset(&dest->root_item.drop_progress, 0,
4724 sizeof(dest->root_item.drop_progress));
4725 btrfs_set_root_drop_level(&dest->root_item, 0);
4726 btrfs_set_root_refs(&dest->root_item, 0);
4728 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4729 ret = btrfs_insert_orphan_item(trans,
4731 dest->root_key.objectid);
4733 btrfs_abort_transaction(trans, ret);
4738 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4739 BTRFS_UUID_KEY_SUBVOL,
4740 dest->root_key.objectid);
4741 if (ret && ret != -ENOENT) {
4742 btrfs_abort_transaction(trans, ret);
4745 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4746 ret = btrfs_uuid_tree_remove(trans,
4747 dest->root_item.received_uuid,
4748 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4749 dest->root_key.objectid);
4750 if (ret && ret != -ENOENT) {
4751 btrfs_abort_transaction(trans, ret);
4756 free_anon_bdev(dest->anon_dev);
4759 trans->block_rsv = NULL;
4760 trans->bytes_reserved = 0;
4761 ret = btrfs_end_transaction(trans);
4762 inode->i_flags |= S_DEAD;
4764 btrfs_subvolume_release_metadata(root, &block_rsv);
4766 up_write(&fs_info->subvol_sem);
4768 spin_lock(&dest->root_item_lock);
4769 root_flags = btrfs_root_flags(&dest->root_item);
4770 btrfs_set_root_flags(&dest->root_item,
4771 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4772 spin_unlock(&dest->root_item_lock);
4774 d_invalidate(dentry);
4775 btrfs_prune_dentries(dest);
4776 ASSERT(dest->send_in_progress == 0);
4782 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4784 struct inode *inode = d_inode(dentry);
4785 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4787 struct btrfs_trans_handle *trans;
4788 u64 last_unlink_trans;
4790 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4792 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4793 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4795 "extent tree v2 doesn't support snapshot deletion yet");
4798 return btrfs_delete_subvolume(dir, dentry);
4801 trans = __unlink_start_trans(dir);
4803 return PTR_ERR(trans);
4805 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4806 err = btrfs_unlink_subvol(trans, dir, dentry);
4810 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4814 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4816 /* now the directory is empty */
4817 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4818 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4819 dentry->d_name.len);
4821 btrfs_i_size_write(BTRFS_I(inode), 0);
4823 * Propagate the last_unlink_trans value of the deleted dir to
4824 * its parent directory. This is to prevent an unrecoverable
4825 * log tree in the case we do something like this:
4827 * 2) create snapshot under dir foo
4828 * 3) delete the snapshot
4831 * 6) fsync foo or some file inside foo
4833 if (last_unlink_trans >= trans->transid)
4834 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4837 btrfs_end_transaction(trans);
4838 btrfs_btree_balance_dirty(fs_info);
4844 * btrfs_truncate_block - read, zero a chunk and write a block
4845 * @inode - inode that we're zeroing
4846 * @from - the offset to start zeroing
4847 * @len - the length to zero, 0 to zero the entire range respective to the
4849 * @front - zero up to the offset instead of from the offset on
4851 * This will find the block for the "from" offset and cow the block and zero the
4852 * part we want to zero. This is used with truncate and hole punching.
4854 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4857 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4858 struct address_space *mapping = inode->vfs_inode.i_mapping;
4859 struct extent_io_tree *io_tree = &inode->io_tree;
4860 struct btrfs_ordered_extent *ordered;
4861 struct extent_state *cached_state = NULL;
4862 struct extent_changeset *data_reserved = NULL;
4863 bool only_release_metadata = false;
4864 u32 blocksize = fs_info->sectorsize;
4865 pgoff_t index = from >> PAGE_SHIFT;
4866 unsigned offset = from & (blocksize - 1);
4868 gfp_t mask = btrfs_alloc_write_mask(mapping);
4869 size_t write_bytes = blocksize;
4874 if (IS_ALIGNED(offset, blocksize) &&
4875 (!len || IS_ALIGNED(len, blocksize)))
4878 block_start = round_down(from, blocksize);
4879 block_end = block_start + blocksize - 1;
4881 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4884 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4885 /* For nocow case, no need to reserve data space */
4886 only_release_metadata = true;
4891 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4893 if (!only_release_metadata)
4894 btrfs_free_reserved_data_space(inode, data_reserved,
4895 block_start, blocksize);
4899 page = find_or_create_page(mapping, index, mask);
4901 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4903 btrfs_delalloc_release_extents(inode, blocksize);
4907 ret = set_page_extent_mapped(page);
4911 if (!PageUptodate(page)) {
4912 ret = btrfs_read_folio(NULL, page_folio(page));
4914 if (page->mapping != mapping) {
4919 if (!PageUptodate(page)) {
4924 wait_on_page_writeback(page);
4926 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4928 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4930 unlock_extent_cached(io_tree, block_start, block_end,
4934 btrfs_start_ordered_extent(ordered, 1);
4935 btrfs_put_ordered_extent(ordered);
4939 clear_extent_bit(&inode->io_tree, block_start, block_end,
4940 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4941 0, 0, &cached_state);
4943 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4946 unlock_extent_cached(io_tree, block_start, block_end,
4951 if (offset != blocksize) {
4953 len = blocksize - offset;
4955 memzero_page(page, (block_start - page_offset(page)),
4958 memzero_page(page, (block_start - page_offset(page)) + offset,
4961 btrfs_page_clear_checked(fs_info, page, block_start,
4962 block_end + 1 - block_start);
4963 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4964 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4966 if (only_release_metadata)
4967 set_extent_bit(&inode->io_tree, block_start, block_end,
4968 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4972 if (only_release_metadata)
4973 btrfs_delalloc_release_metadata(inode, blocksize, true);
4975 btrfs_delalloc_release_space(inode, data_reserved,
4976 block_start, blocksize, true);
4978 btrfs_delalloc_release_extents(inode, blocksize);
4982 if (only_release_metadata)
4983 btrfs_check_nocow_unlock(inode);
4984 extent_changeset_free(data_reserved);
4988 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4989 u64 offset, u64 len)
4991 struct btrfs_fs_info *fs_info = root->fs_info;
4992 struct btrfs_trans_handle *trans;
4993 struct btrfs_drop_extents_args drop_args = { 0 };
4997 * If NO_HOLES is enabled, we don't need to do anything.
4998 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4999 * or btrfs_update_inode() will be called, which guarantee that the next
5000 * fsync will know this inode was changed and needs to be logged.
5002 if (btrfs_fs_incompat(fs_info, NO_HOLES))
5006 * 1 - for the one we're dropping
5007 * 1 - for the one we're adding
5008 * 1 - for updating the inode.
5010 trans = btrfs_start_transaction(root, 3);
5012 return PTR_ERR(trans);
5014 drop_args.start = offset;
5015 drop_args.end = offset + len;
5016 drop_args.drop_cache = true;
5018 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5020 btrfs_abort_transaction(trans, ret);
5021 btrfs_end_transaction(trans);
5025 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5026 offset, 0, 0, len, 0, len, 0, 0, 0);
5028 btrfs_abort_transaction(trans, ret);
5030 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5031 btrfs_update_inode(trans, root, inode);
5033 btrfs_end_transaction(trans);
5038 * This function puts in dummy file extents for the area we're creating a hole
5039 * for. So if we are truncating this file to a larger size we need to insert
5040 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5041 * the range between oldsize and size
5043 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5045 struct btrfs_root *root = inode->root;
5046 struct btrfs_fs_info *fs_info = root->fs_info;
5047 struct extent_io_tree *io_tree = &inode->io_tree;
5048 struct extent_map *em = NULL;
5049 struct extent_state *cached_state = NULL;
5050 struct extent_map_tree *em_tree = &inode->extent_tree;
5051 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5052 u64 block_end = ALIGN(size, fs_info->sectorsize);
5059 * If our size started in the middle of a block we need to zero out the
5060 * rest of the block before we expand the i_size, otherwise we could
5061 * expose stale data.
5063 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5067 if (size <= hole_start)
5070 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5072 cur_offset = hole_start;
5074 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5075 block_end - cur_offset);
5081 last_byte = min(extent_map_end(em), block_end);
5082 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5083 hole_size = last_byte - cur_offset;
5085 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5086 struct extent_map *hole_em;
5088 err = maybe_insert_hole(root, inode, cur_offset,
5093 err = btrfs_inode_set_file_extent_range(inode,
5094 cur_offset, hole_size);
5098 btrfs_drop_extent_cache(inode, cur_offset,
5099 cur_offset + hole_size - 1, 0);
5100 hole_em = alloc_extent_map();
5102 btrfs_set_inode_full_sync(inode);
5105 hole_em->start = cur_offset;
5106 hole_em->len = hole_size;
5107 hole_em->orig_start = cur_offset;
5109 hole_em->block_start = EXTENT_MAP_HOLE;
5110 hole_em->block_len = 0;
5111 hole_em->orig_block_len = 0;
5112 hole_em->ram_bytes = hole_size;
5113 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5114 hole_em->generation = fs_info->generation;
5117 write_lock(&em_tree->lock);
5118 err = add_extent_mapping(em_tree, hole_em, 1);
5119 write_unlock(&em_tree->lock);
5122 btrfs_drop_extent_cache(inode, cur_offset,
5126 free_extent_map(hole_em);
5128 err = btrfs_inode_set_file_extent_range(inode,
5129 cur_offset, hole_size);
5134 free_extent_map(em);
5136 cur_offset = last_byte;
5137 if (cur_offset >= block_end)
5140 free_extent_map(em);
5141 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5145 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5147 struct btrfs_root *root = BTRFS_I(inode)->root;
5148 struct btrfs_trans_handle *trans;
5149 loff_t oldsize = i_size_read(inode);
5150 loff_t newsize = attr->ia_size;
5151 int mask = attr->ia_valid;
5155 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5156 * special case where we need to update the times despite not having
5157 * these flags set. For all other operations the VFS set these flags
5158 * explicitly if it wants a timestamp update.
5160 if (newsize != oldsize) {
5161 inode_inc_iversion(inode);
5162 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5163 inode->i_mtime = current_time(inode);
5164 inode->i_ctime = inode->i_mtime;
5168 if (newsize > oldsize) {
5170 * Don't do an expanding truncate while snapshotting is ongoing.
5171 * This is to ensure the snapshot captures a fully consistent
5172 * state of this file - if the snapshot captures this expanding
5173 * truncation, it must capture all writes that happened before
5176 btrfs_drew_write_lock(&root->snapshot_lock);
5177 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5179 btrfs_drew_write_unlock(&root->snapshot_lock);
5183 trans = btrfs_start_transaction(root, 1);
5184 if (IS_ERR(trans)) {
5185 btrfs_drew_write_unlock(&root->snapshot_lock);
5186 return PTR_ERR(trans);
5189 i_size_write(inode, newsize);
5190 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5191 pagecache_isize_extended(inode, oldsize, newsize);
5192 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5193 btrfs_drew_write_unlock(&root->snapshot_lock);
5194 btrfs_end_transaction(trans);
5196 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5198 if (btrfs_is_zoned(fs_info)) {
5199 ret = btrfs_wait_ordered_range(inode,
5200 ALIGN(newsize, fs_info->sectorsize),
5207 * We're truncating a file that used to have good data down to
5208 * zero. Make sure any new writes to the file get on disk
5212 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5213 &BTRFS_I(inode)->runtime_flags);
5215 truncate_setsize(inode, newsize);
5217 inode_dio_wait(inode);
5219 ret = btrfs_truncate(inode, newsize == oldsize);
5220 if (ret && inode->i_nlink) {
5224 * Truncate failed, so fix up the in-memory size. We
5225 * adjusted disk_i_size down as we removed extents, so
5226 * wait for disk_i_size to be stable and then update the
5227 * in-memory size to match.
5229 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5232 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5239 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5242 struct inode *inode = d_inode(dentry);
5243 struct btrfs_root *root = BTRFS_I(inode)->root;
5246 if (btrfs_root_readonly(root))
5249 err = setattr_prepare(mnt_userns, dentry, attr);
5253 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5254 err = btrfs_setsize(inode, attr);
5259 if (attr->ia_valid) {
5260 setattr_copy(mnt_userns, inode, attr);
5261 inode_inc_iversion(inode);
5262 err = btrfs_dirty_inode(inode);
5264 if (!err && attr->ia_valid & ATTR_MODE)
5265 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5272 * While truncating the inode pages during eviction, we get the VFS
5273 * calling btrfs_invalidate_folio() against each folio of the inode. This
5274 * is slow because the calls to btrfs_invalidate_folio() result in a
5275 * huge amount of calls to lock_extent_bits() and clear_extent_bit(),
5276 * which keep merging and splitting extent_state structures over and over,
5277 * wasting lots of time.
5279 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5280 * skip all those expensive operations on a per folio basis and do only
5281 * the ordered io finishing, while we release here the extent_map and
5282 * extent_state structures, without the excessive merging and splitting.
5284 static void evict_inode_truncate_pages(struct inode *inode)
5286 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5287 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5288 struct rb_node *node;
5290 ASSERT(inode->i_state & I_FREEING);
5291 truncate_inode_pages_final(&inode->i_data);
5293 write_lock(&map_tree->lock);
5294 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5295 struct extent_map *em;
5297 node = rb_first_cached(&map_tree->map);
5298 em = rb_entry(node, struct extent_map, rb_node);
5299 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5300 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5301 remove_extent_mapping(map_tree, em);
5302 free_extent_map(em);
5303 if (need_resched()) {
5304 write_unlock(&map_tree->lock);
5306 write_lock(&map_tree->lock);
5309 write_unlock(&map_tree->lock);
5312 * Keep looping until we have no more ranges in the io tree.
5313 * We can have ongoing bios started by readahead that have
5314 * their endio callback (extent_io.c:end_bio_extent_readpage)
5315 * still in progress (unlocked the pages in the bio but did not yet
5316 * unlocked the ranges in the io tree). Therefore this means some
5317 * ranges can still be locked and eviction started because before
5318 * submitting those bios, which are executed by a separate task (work
5319 * queue kthread), inode references (inode->i_count) were not taken
5320 * (which would be dropped in the end io callback of each bio).
5321 * Therefore here we effectively end up waiting for those bios and
5322 * anyone else holding locked ranges without having bumped the inode's
5323 * reference count - if we don't do it, when they access the inode's
5324 * io_tree to unlock a range it may be too late, leading to an
5325 * use-after-free issue.
5327 spin_lock(&io_tree->lock);
5328 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5329 struct extent_state *state;
5330 struct extent_state *cached_state = NULL;
5333 unsigned state_flags;
5335 node = rb_first(&io_tree->state);
5336 state = rb_entry(node, struct extent_state, rb_node);
5337 start = state->start;
5339 state_flags = state->state;
5340 spin_unlock(&io_tree->lock);
5342 lock_extent_bits(io_tree, start, end, &cached_state);
5345 * If still has DELALLOC flag, the extent didn't reach disk,
5346 * and its reserved space won't be freed by delayed_ref.
5347 * So we need to free its reserved space here.
5348 * (Refer to comment in btrfs_invalidate_folio, case 2)
5350 * Note, end is the bytenr of last byte, so we need + 1 here.
5352 if (state_flags & EXTENT_DELALLOC)
5353 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5356 clear_extent_bit(io_tree, start, end,
5357 EXTENT_LOCKED | EXTENT_DELALLOC |
5358 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5362 spin_lock(&io_tree->lock);
5364 spin_unlock(&io_tree->lock);
5367 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5368 struct btrfs_block_rsv *rsv)
5370 struct btrfs_fs_info *fs_info = root->fs_info;
5371 struct btrfs_trans_handle *trans;
5372 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5376 * Eviction should be taking place at some place safe because of our
5377 * delayed iputs. However the normal flushing code will run delayed
5378 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5380 * We reserve the delayed_refs_extra here again because we can't use
5381 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5382 * above. We reserve our extra bit here because we generate a ton of
5383 * delayed refs activity by truncating.
5385 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5386 * if we fail to make this reservation we can re-try without the
5387 * delayed_refs_extra so we can make some forward progress.
5389 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5390 BTRFS_RESERVE_FLUSH_EVICT);
5392 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5393 BTRFS_RESERVE_FLUSH_EVICT);
5396 "could not allocate space for delete; will truncate on mount");
5397 return ERR_PTR(-ENOSPC);
5399 delayed_refs_extra = 0;
5402 trans = btrfs_join_transaction(root);
5406 if (delayed_refs_extra) {
5407 trans->block_rsv = &fs_info->trans_block_rsv;
5408 trans->bytes_reserved = delayed_refs_extra;
5409 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5410 delayed_refs_extra, 1);
5415 void btrfs_evict_inode(struct inode *inode)
5417 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5418 struct btrfs_trans_handle *trans;
5419 struct btrfs_root *root = BTRFS_I(inode)->root;
5420 struct btrfs_block_rsv *rsv;
5423 trace_btrfs_inode_evict(inode);
5426 fsverity_cleanup_inode(inode);
5431 evict_inode_truncate_pages(inode);
5433 if (inode->i_nlink &&
5434 ((btrfs_root_refs(&root->root_item) != 0 &&
5435 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5436 btrfs_is_free_space_inode(BTRFS_I(inode))))
5439 if (is_bad_inode(inode))
5442 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5444 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5447 if (inode->i_nlink > 0) {
5448 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5449 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5454 * This makes sure the inode item in tree is uptodate and the space for
5455 * the inode update is released.
5457 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5462 * This drops any pending insert or delete operations we have for this
5463 * inode. We could have a delayed dir index deletion queued up, but
5464 * we're removing the inode completely so that'll be taken care of in
5467 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5469 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5472 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5473 rsv->failfast = true;
5475 btrfs_i_size_write(BTRFS_I(inode), 0);
5478 struct btrfs_truncate_control control = {
5479 .inode = BTRFS_I(inode),
5480 .ino = btrfs_ino(BTRFS_I(inode)),
5485 trans = evict_refill_and_join(root, rsv);
5489 trans->block_rsv = rsv;
5491 ret = btrfs_truncate_inode_items(trans, root, &control);
5492 trans->block_rsv = &fs_info->trans_block_rsv;
5493 btrfs_end_transaction(trans);
5494 btrfs_btree_balance_dirty(fs_info);
5495 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5502 * Errors here aren't a big deal, it just means we leave orphan items in
5503 * the tree. They will be cleaned up on the next mount. If the inode
5504 * number gets reused, cleanup deletes the orphan item without doing
5505 * anything, and unlink reuses the existing orphan item.
5507 * If it turns out that we are dropping too many of these, we might want
5508 * to add a mechanism for retrying these after a commit.
5510 trans = evict_refill_and_join(root, rsv);
5511 if (!IS_ERR(trans)) {
5512 trans->block_rsv = rsv;
5513 btrfs_orphan_del(trans, BTRFS_I(inode));
5514 trans->block_rsv = &fs_info->trans_block_rsv;
5515 btrfs_end_transaction(trans);
5519 btrfs_free_block_rsv(fs_info, rsv);
5522 * If we didn't successfully delete, the orphan item will still be in
5523 * the tree and we'll retry on the next mount. Again, we might also want
5524 * to retry these periodically in the future.
5526 btrfs_remove_delayed_node(BTRFS_I(inode));
5527 fsverity_cleanup_inode(inode);
5532 * Return the key found in the dir entry in the location pointer, fill @type
5533 * with BTRFS_FT_*, and return 0.
5535 * If no dir entries were found, returns -ENOENT.
5536 * If found a corrupted location in dir entry, returns -EUCLEAN.
5538 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5539 struct btrfs_key *location, u8 *type)
5541 const char *name = dentry->d_name.name;
5542 int namelen = dentry->d_name.len;
5543 struct btrfs_dir_item *di;
5544 struct btrfs_path *path;
5545 struct btrfs_root *root = BTRFS_I(dir)->root;
5548 path = btrfs_alloc_path();
5552 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5554 if (IS_ERR_OR_NULL(di)) {
5555 ret = di ? PTR_ERR(di) : -ENOENT;
5559 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5560 if (location->type != BTRFS_INODE_ITEM_KEY &&
5561 location->type != BTRFS_ROOT_ITEM_KEY) {
5563 btrfs_warn(root->fs_info,
5564 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5565 __func__, name, btrfs_ino(BTRFS_I(dir)),
5566 location->objectid, location->type, location->offset);
5569 *type = btrfs_dir_type(path->nodes[0], di);
5571 btrfs_free_path(path);
5576 * when we hit a tree root in a directory, the btrfs part of the inode
5577 * needs to be changed to reflect the root directory of the tree root. This
5578 * is kind of like crossing a mount point.
5580 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5582 struct dentry *dentry,
5583 struct btrfs_key *location,
5584 struct btrfs_root **sub_root)
5586 struct btrfs_path *path;
5587 struct btrfs_root *new_root;
5588 struct btrfs_root_ref *ref;
5589 struct extent_buffer *leaf;
5590 struct btrfs_key key;
5594 path = btrfs_alloc_path();
5601 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5602 key.type = BTRFS_ROOT_REF_KEY;
5603 key.offset = location->objectid;
5605 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5612 leaf = path->nodes[0];
5613 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5614 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5615 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5618 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5619 (unsigned long)(ref + 1),
5620 dentry->d_name.len);
5624 btrfs_release_path(path);
5626 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5627 if (IS_ERR(new_root)) {
5628 err = PTR_ERR(new_root);
5632 *sub_root = new_root;
5633 location->objectid = btrfs_root_dirid(&new_root->root_item);
5634 location->type = BTRFS_INODE_ITEM_KEY;
5635 location->offset = 0;
5638 btrfs_free_path(path);
5642 static void inode_tree_add(struct inode *inode)
5644 struct btrfs_root *root = BTRFS_I(inode)->root;
5645 struct btrfs_inode *entry;
5647 struct rb_node *parent;
5648 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5649 u64 ino = btrfs_ino(BTRFS_I(inode));
5651 if (inode_unhashed(inode))
5654 spin_lock(&root->inode_lock);
5655 p = &root->inode_tree.rb_node;
5658 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5660 if (ino < btrfs_ino(entry))
5661 p = &parent->rb_left;
5662 else if (ino > btrfs_ino(entry))
5663 p = &parent->rb_right;
5665 WARN_ON(!(entry->vfs_inode.i_state &
5666 (I_WILL_FREE | I_FREEING)));
5667 rb_replace_node(parent, new, &root->inode_tree);
5668 RB_CLEAR_NODE(parent);
5669 spin_unlock(&root->inode_lock);
5673 rb_link_node(new, parent, p);
5674 rb_insert_color(new, &root->inode_tree);
5675 spin_unlock(&root->inode_lock);
5678 static void inode_tree_del(struct btrfs_inode *inode)
5680 struct btrfs_root *root = inode->root;
5683 spin_lock(&root->inode_lock);
5684 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5685 rb_erase(&inode->rb_node, &root->inode_tree);
5686 RB_CLEAR_NODE(&inode->rb_node);
5687 empty = RB_EMPTY_ROOT(&root->inode_tree);
5689 spin_unlock(&root->inode_lock);
5691 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5692 spin_lock(&root->inode_lock);
5693 empty = RB_EMPTY_ROOT(&root->inode_tree);
5694 spin_unlock(&root->inode_lock);
5696 btrfs_add_dead_root(root);
5701 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5703 struct btrfs_iget_args *args = p;
5705 inode->i_ino = args->ino;
5706 BTRFS_I(inode)->location.objectid = args->ino;
5707 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5708 BTRFS_I(inode)->location.offset = 0;
5709 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5710 BUG_ON(args->root && !BTRFS_I(inode)->root);
5714 static int btrfs_find_actor(struct inode *inode, void *opaque)
5716 struct btrfs_iget_args *args = opaque;
5718 return args->ino == BTRFS_I(inode)->location.objectid &&
5719 args->root == BTRFS_I(inode)->root;
5722 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5723 struct btrfs_root *root)
5725 struct inode *inode;
5726 struct btrfs_iget_args args;
5727 unsigned long hashval = btrfs_inode_hash(ino, root);
5732 inode = iget5_locked(s, hashval, btrfs_find_actor,
5733 btrfs_init_locked_inode,
5739 * Get an inode object given its inode number and corresponding root.
5740 * Path can be preallocated to prevent recursing back to iget through
5741 * allocator. NULL is also valid but may require an additional allocation
5744 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5745 struct btrfs_root *root, struct btrfs_path *path)
5747 struct inode *inode;
5749 inode = btrfs_iget_locked(s, ino, root);
5751 return ERR_PTR(-ENOMEM);
5753 if (inode->i_state & I_NEW) {
5756 ret = btrfs_read_locked_inode(inode, path);
5758 inode_tree_add(inode);
5759 unlock_new_inode(inode);
5763 * ret > 0 can come from btrfs_search_slot called by
5764 * btrfs_read_locked_inode, this means the inode item
5769 inode = ERR_PTR(ret);
5776 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5778 return btrfs_iget_path(s, ino, root, NULL);
5781 static struct inode *new_simple_dir(struct super_block *s,
5782 struct btrfs_key *key,
5783 struct btrfs_root *root)
5785 struct inode *inode = new_inode(s);
5788 return ERR_PTR(-ENOMEM);
5790 BTRFS_I(inode)->root = btrfs_grab_root(root);
5791 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5792 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5794 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5796 * We only need lookup, the rest is read-only and there's no inode
5797 * associated with the dentry
5799 inode->i_op = &simple_dir_inode_operations;
5800 inode->i_opflags &= ~IOP_XATTR;
5801 inode->i_fop = &simple_dir_operations;
5802 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5803 inode->i_mtime = current_time(inode);
5804 inode->i_atime = inode->i_mtime;
5805 inode->i_ctime = inode->i_mtime;
5806 BTRFS_I(inode)->i_otime = inode->i_mtime;
5811 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5812 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5813 static_assert(BTRFS_FT_DIR == FT_DIR);
5814 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5815 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5816 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5817 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5818 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5820 static inline u8 btrfs_inode_type(struct inode *inode)
5822 return fs_umode_to_ftype(inode->i_mode);
5825 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5827 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5828 struct inode *inode;
5829 struct btrfs_root *root = BTRFS_I(dir)->root;
5830 struct btrfs_root *sub_root = root;
5831 struct btrfs_key location;
5835 if (dentry->d_name.len > BTRFS_NAME_LEN)
5836 return ERR_PTR(-ENAMETOOLONG);
5838 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5840 return ERR_PTR(ret);
5842 if (location.type == BTRFS_INODE_ITEM_KEY) {
5843 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5847 /* Do extra check against inode mode with di_type */
5848 if (btrfs_inode_type(inode) != di_type) {
5850 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5851 inode->i_mode, btrfs_inode_type(inode),
5854 return ERR_PTR(-EUCLEAN);
5859 ret = fixup_tree_root_location(fs_info, dir, dentry,
5860 &location, &sub_root);
5863 inode = ERR_PTR(ret);
5865 inode = new_simple_dir(dir->i_sb, &location, root);
5867 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5868 btrfs_put_root(sub_root);
5873 down_read(&fs_info->cleanup_work_sem);
5874 if (!sb_rdonly(inode->i_sb))
5875 ret = btrfs_orphan_cleanup(sub_root);
5876 up_read(&fs_info->cleanup_work_sem);
5879 inode = ERR_PTR(ret);
5886 static int btrfs_dentry_delete(const struct dentry *dentry)
5888 struct btrfs_root *root;
5889 struct inode *inode = d_inode(dentry);
5891 if (!inode && !IS_ROOT(dentry))
5892 inode = d_inode(dentry->d_parent);
5895 root = BTRFS_I(inode)->root;
5896 if (btrfs_root_refs(&root->root_item) == 0)
5899 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5905 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5908 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5910 if (inode == ERR_PTR(-ENOENT))
5912 return d_splice_alias(inode, dentry);
5916 * All this infrastructure exists because dir_emit can fault, and we are holding
5917 * the tree lock when doing readdir. For now just allocate a buffer and copy
5918 * our information into that, and then dir_emit from the buffer. This is
5919 * similar to what NFS does, only we don't keep the buffer around in pagecache
5920 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5921 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5924 static int btrfs_opendir(struct inode *inode, struct file *file)
5926 struct btrfs_file_private *private;
5928 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5931 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5932 if (!private->filldir_buf) {
5936 file->private_data = private;
5947 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5950 struct dir_entry *entry = addr;
5951 char *name = (char *)(entry + 1);
5953 ctx->pos = get_unaligned(&entry->offset);
5954 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5955 get_unaligned(&entry->ino),
5956 get_unaligned(&entry->type)))
5958 addr += sizeof(struct dir_entry) +
5959 get_unaligned(&entry->name_len);
5965 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5967 struct inode *inode = file_inode(file);
5968 struct btrfs_root *root = BTRFS_I(inode)->root;
5969 struct btrfs_file_private *private = file->private_data;
5970 struct btrfs_dir_item *di;
5971 struct btrfs_key key;
5972 struct btrfs_key found_key;
5973 struct btrfs_path *path;
5975 struct list_head ins_list;
5976 struct list_head del_list;
5983 struct btrfs_key location;
5985 if (!dir_emit_dots(file, ctx))
5988 path = btrfs_alloc_path();
5992 addr = private->filldir_buf;
5993 path->reada = READA_FORWARD;
5995 INIT_LIST_HEAD(&ins_list);
5996 INIT_LIST_HEAD(&del_list);
5997 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
6000 key.type = BTRFS_DIR_INDEX_KEY;
6001 key.offset = ctx->pos;
6002 key.objectid = btrfs_ino(BTRFS_I(inode));
6004 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
6005 struct dir_entry *entry;
6006 struct extent_buffer *leaf = path->nodes[0];
6008 if (found_key.objectid != key.objectid)
6010 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6012 if (found_key.offset < ctx->pos)
6014 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6016 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
6017 name_len = btrfs_dir_name_len(leaf, di);
6018 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6020 btrfs_release_path(path);
6021 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6024 addr = private->filldir_buf;
6031 put_unaligned(name_len, &entry->name_len);
6032 name_ptr = (char *)(entry + 1);
6033 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6035 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6037 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6038 put_unaligned(location.objectid, &entry->ino);
6039 put_unaligned(found_key.offset, &entry->offset);
6041 addr += sizeof(struct dir_entry) + name_len;
6042 total_len += sizeof(struct dir_entry) + name_len;
6044 /* Catch error encountered during iteration */
6048 btrfs_release_path(path);
6050 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6054 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6059 * Stop new entries from being returned after we return the last
6062 * New directory entries are assigned a strictly increasing
6063 * offset. This means that new entries created during readdir
6064 * are *guaranteed* to be seen in the future by that readdir.
6065 * This has broken buggy programs which operate on names as
6066 * they're returned by readdir. Until we re-use freed offsets
6067 * we have this hack to stop new entries from being returned
6068 * under the assumption that they'll never reach this huge
6071 * This is being careful not to overflow 32bit loff_t unless the
6072 * last entry requires it because doing so has broken 32bit apps
6075 if (ctx->pos >= INT_MAX)
6076 ctx->pos = LLONG_MAX;
6083 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6084 btrfs_free_path(path);
6089 * This is somewhat expensive, updating the tree every time the
6090 * inode changes. But, it is most likely to find the inode in cache.
6091 * FIXME, needs more benchmarking...there are no reasons other than performance
6092 * to keep or drop this code.
6094 static int btrfs_dirty_inode(struct inode *inode)
6096 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6097 struct btrfs_root *root = BTRFS_I(inode)->root;
6098 struct btrfs_trans_handle *trans;
6101 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6104 trans = btrfs_join_transaction(root);
6106 return PTR_ERR(trans);
6108 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6109 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6110 /* whoops, lets try again with the full transaction */
6111 btrfs_end_transaction(trans);
6112 trans = btrfs_start_transaction(root, 1);
6114 return PTR_ERR(trans);
6116 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6118 btrfs_end_transaction(trans);
6119 if (BTRFS_I(inode)->delayed_node)
6120 btrfs_balance_delayed_items(fs_info);
6126 * This is a copy of file_update_time. We need this so we can return error on
6127 * ENOSPC for updating the inode in the case of file write and mmap writes.
6129 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6132 struct btrfs_root *root = BTRFS_I(inode)->root;
6133 bool dirty = flags & ~S_VERSION;
6135 if (btrfs_root_readonly(root))
6138 if (flags & S_VERSION)
6139 dirty |= inode_maybe_inc_iversion(inode, dirty);
6140 if (flags & S_CTIME)
6141 inode->i_ctime = *now;
6142 if (flags & S_MTIME)
6143 inode->i_mtime = *now;
6144 if (flags & S_ATIME)
6145 inode->i_atime = *now;
6146 return dirty ? btrfs_dirty_inode(inode) : 0;
6150 * find the highest existing sequence number in a directory
6151 * and then set the in-memory index_cnt variable to reflect
6152 * free sequence numbers
6154 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6156 struct btrfs_root *root = inode->root;
6157 struct btrfs_key key, found_key;
6158 struct btrfs_path *path;
6159 struct extent_buffer *leaf;
6162 key.objectid = btrfs_ino(inode);
6163 key.type = BTRFS_DIR_INDEX_KEY;
6164 key.offset = (u64)-1;
6166 path = btrfs_alloc_path();
6170 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6173 /* FIXME: we should be able to handle this */
6178 if (path->slots[0] == 0) {
6179 inode->index_cnt = BTRFS_DIR_START_INDEX;
6185 leaf = path->nodes[0];
6186 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6188 if (found_key.objectid != btrfs_ino(inode) ||
6189 found_key.type != BTRFS_DIR_INDEX_KEY) {
6190 inode->index_cnt = BTRFS_DIR_START_INDEX;
6194 inode->index_cnt = found_key.offset + 1;
6196 btrfs_free_path(path);
6201 * helper to find a free sequence number in a given directory. This current
6202 * code is very simple, later versions will do smarter things in the btree
6204 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6208 if (dir->index_cnt == (u64)-1) {
6209 ret = btrfs_inode_delayed_dir_index_count(dir);
6211 ret = btrfs_set_inode_index_count(dir);
6217 *index = dir->index_cnt;
6223 static int btrfs_insert_inode_locked(struct inode *inode)
6225 struct btrfs_iget_args args;
6227 args.ino = BTRFS_I(inode)->location.objectid;
6228 args.root = BTRFS_I(inode)->root;
6230 return insert_inode_locked4(inode,
6231 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6232 btrfs_find_actor, &args);
6235 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6236 unsigned int *trans_num_items)
6238 struct inode *dir = args->dir;
6239 struct inode *inode = args->inode;
6242 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6246 /* 1 to add inode item */
6247 *trans_num_items = 1;
6248 /* 1 to add compression property */
6249 if (BTRFS_I(dir)->prop_compress)
6250 (*trans_num_items)++;
6251 /* 1 to add default ACL xattr */
6252 if (args->default_acl)
6253 (*trans_num_items)++;
6254 /* 1 to add access ACL xattr */
6256 (*trans_num_items)++;
6257 #ifdef CONFIG_SECURITY
6258 /* 1 to add LSM xattr */
6259 if (dir->i_security)
6260 (*trans_num_items)++;
6263 /* 1 to add orphan item */
6264 (*trans_num_items)++;
6268 * 1 to add dir index
6269 * 1 to update parent inode item
6271 * No need for 1 unit for the inode ref item because it is
6272 * inserted in a batch together with the inode item at
6273 * btrfs_create_new_inode().
6275 *trans_num_items += 3;
6280 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6282 posix_acl_release(args->acl);
6283 posix_acl_release(args->default_acl);
6287 * Inherit flags from the parent inode.
6289 * Currently only the compression flags and the cow flags are inherited.
6291 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6295 flags = BTRFS_I(dir)->flags;
6297 if (flags & BTRFS_INODE_NOCOMPRESS) {
6298 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6299 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6300 } else if (flags & BTRFS_INODE_COMPRESS) {
6301 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6302 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6305 if (flags & BTRFS_INODE_NODATACOW) {
6306 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6307 if (S_ISREG(inode->i_mode))
6308 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6311 btrfs_sync_inode_flags_to_i_flags(inode);
6314 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6315 struct btrfs_new_inode_args *args)
6317 struct inode *dir = args->dir;
6318 struct inode *inode = args->inode;
6319 const char *name = args->orphan ? NULL : args->dentry->d_name.name;
6320 int name_len = args->orphan ? 0 : args->dentry->d_name.len;
6321 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6322 struct btrfs_root *root;
6323 struct btrfs_inode_item *inode_item;
6324 struct btrfs_key *location;
6325 struct btrfs_path *path;
6327 struct btrfs_inode_ref *ref;
6328 struct btrfs_key key[2];
6330 struct btrfs_item_batch batch;
6334 path = btrfs_alloc_path();
6339 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6340 root = BTRFS_I(inode)->root;
6342 ret = btrfs_get_free_objectid(root, &objectid);
6345 inode->i_ino = objectid;
6349 * O_TMPFILE, set link count to 0, so that after this point, we
6350 * fill in an inode item with the correct link count.
6352 set_nlink(inode, 0);
6354 trace_btrfs_inode_request(dir);
6356 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6360 /* index_cnt is ignored for everything but a dir. */
6361 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6362 BTRFS_I(inode)->generation = trans->transid;
6363 inode->i_generation = BTRFS_I(inode)->generation;
6366 * Subvolumes don't inherit flags from their parent directory.
6367 * Originally this was probably by accident, but we probably can't
6368 * change it now without compatibility issues.
6371 btrfs_inherit_iflags(inode, dir);
6373 if (S_ISREG(inode->i_mode)) {
6374 if (btrfs_test_opt(fs_info, NODATASUM))
6375 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6376 if (btrfs_test_opt(fs_info, NODATACOW))
6377 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6378 BTRFS_INODE_NODATASUM;
6381 location = &BTRFS_I(inode)->location;
6382 location->objectid = objectid;
6383 location->offset = 0;
6384 location->type = BTRFS_INODE_ITEM_KEY;
6386 ret = btrfs_insert_inode_locked(inode);
6389 BTRFS_I(dir)->index_cnt--;
6394 * We could have gotten an inode number from somebody who was fsynced
6395 * and then removed in this same transaction, so let's just set full
6396 * sync since it will be a full sync anyway and this will blow away the
6397 * old info in the log.
6399 btrfs_set_inode_full_sync(BTRFS_I(inode));
6401 key[0].objectid = objectid;
6402 key[0].type = BTRFS_INODE_ITEM_KEY;
6405 sizes[0] = sizeof(struct btrfs_inode_item);
6407 if (!args->orphan) {
6409 * Start new inodes with an inode_ref. This is slightly more
6410 * efficient for small numbers of hard links since they will
6411 * be packed into one item. Extended refs will kick in if we
6412 * add more hard links than can fit in the ref item.
6414 key[1].objectid = objectid;
6415 key[1].type = BTRFS_INODE_REF_KEY;
6417 key[1].offset = objectid;
6418 sizes[1] = 2 + sizeof(*ref);
6420 key[1].offset = btrfs_ino(BTRFS_I(dir));
6421 sizes[1] = name_len + sizeof(*ref);
6425 batch.keys = &key[0];
6426 batch.data_sizes = &sizes[0];
6427 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6428 batch.nr = args->orphan ? 1 : 2;
6429 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6431 btrfs_abort_transaction(trans, ret);
6435 inode->i_mtime = current_time(inode);
6436 inode->i_atime = inode->i_mtime;
6437 inode->i_ctime = inode->i_mtime;
6438 BTRFS_I(inode)->i_otime = inode->i_mtime;
6441 * We're going to fill the inode item now, so at this point the inode
6442 * must be fully initialized.
6445 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6446 struct btrfs_inode_item);
6447 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6448 sizeof(*inode_item));
6449 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6451 if (!args->orphan) {
6452 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6453 struct btrfs_inode_ref);
6454 ptr = (unsigned long)(ref + 1);
6456 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6457 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6458 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6460 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6461 btrfs_set_inode_ref_index(path->nodes[0], ref,
6462 BTRFS_I(inode)->dir_index);
6463 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6467 btrfs_mark_buffer_dirty(path->nodes[0]);
6469 * We don't need the path anymore, plus inheriting properties, adding
6470 * ACLs, security xattrs, orphan item or adding the link, will result in
6471 * allocating yet another path. So just free our path.
6473 btrfs_free_path(path);
6477 struct inode *parent;
6480 * Subvolumes inherit properties from their parent subvolume,
6481 * not the directory they were created in.
6483 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6484 BTRFS_I(dir)->root);
6485 if (IS_ERR(parent)) {
6486 ret = PTR_ERR(parent);
6488 ret = btrfs_inode_inherit_props(trans, inode, parent);
6492 ret = btrfs_inode_inherit_props(trans, inode, dir);
6496 "error inheriting props for ino %llu (root %llu): %d",
6497 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6502 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6505 if (!args->subvol) {
6506 ret = btrfs_init_inode_security(trans, args);
6508 btrfs_abort_transaction(trans, ret);
6513 inode_tree_add(inode);
6515 trace_btrfs_inode_new(inode);
6516 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6518 btrfs_update_root_times(trans, root);
6521 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6523 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6524 name_len, 0, BTRFS_I(inode)->dir_index);
6527 btrfs_abort_transaction(trans, ret);
6535 * discard_new_inode() calls iput(), but the caller owns the reference
6539 discard_new_inode(inode);
6541 btrfs_free_path(path);
6546 * utility function to add 'inode' into 'parent_inode' with
6547 * a give name and a given sequence number.
6548 * if 'add_backref' is true, also insert a backref from the
6549 * inode to the parent directory.
6551 int btrfs_add_link(struct btrfs_trans_handle *trans,
6552 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6553 const char *name, int name_len, int add_backref, u64 index)
6556 struct btrfs_key key;
6557 struct btrfs_root *root = parent_inode->root;
6558 u64 ino = btrfs_ino(inode);
6559 u64 parent_ino = btrfs_ino(parent_inode);
6561 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6562 memcpy(&key, &inode->root->root_key, sizeof(key));
6565 key.type = BTRFS_INODE_ITEM_KEY;
6569 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6570 ret = btrfs_add_root_ref(trans, key.objectid,
6571 root->root_key.objectid, parent_ino,
6572 index, name, name_len);
6573 } else if (add_backref) {
6574 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6578 /* Nothing to clean up yet */
6582 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6583 btrfs_inode_type(&inode->vfs_inode), index);
6584 if (ret == -EEXIST || ret == -EOVERFLOW)
6587 btrfs_abort_transaction(trans, ret);
6591 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6593 inode_inc_iversion(&parent_inode->vfs_inode);
6595 * If we are replaying a log tree, we do not want to update the mtime
6596 * and ctime of the parent directory with the current time, since the
6597 * log replay procedure is responsible for setting them to their correct
6598 * values (the ones it had when the fsync was done).
6600 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6601 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6603 parent_inode->vfs_inode.i_mtime = now;
6604 parent_inode->vfs_inode.i_ctime = now;
6606 ret = btrfs_update_inode(trans, root, parent_inode);
6608 btrfs_abort_transaction(trans, ret);
6612 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6615 err = btrfs_del_root_ref(trans, key.objectid,
6616 root->root_key.objectid, parent_ino,
6617 &local_index, name, name_len);
6619 btrfs_abort_transaction(trans, err);
6620 } else if (add_backref) {
6624 err = btrfs_del_inode_ref(trans, root, name, name_len,
6625 ino, parent_ino, &local_index);
6627 btrfs_abort_transaction(trans, err);
6630 /* Return the original error code */
6634 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6635 struct inode *inode)
6637 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6638 struct btrfs_root *root = BTRFS_I(dir)->root;
6639 struct btrfs_new_inode_args new_inode_args = {
6644 unsigned int trans_num_items;
6645 struct btrfs_trans_handle *trans;
6648 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6652 trans = btrfs_start_transaction(root, trans_num_items);
6653 if (IS_ERR(trans)) {
6654 err = PTR_ERR(trans);
6655 goto out_new_inode_args;
6658 err = btrfs_create_new_inode(trans, &new_inode_args);
6660 d_instantiate_new(dentry, inode);
6662 btrfs_end_transaction(trans);
6663 btrfs_btree_balance_dirty(fs_info);
6665 btrfs_new_inode_args_destroy(&new_inode_args);
6672 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6673 struct dentry *dentry, umode_t mode, dev_t rdev)
6675 struct inode *inode;
6677 inode = new_inode(dir->i_sb);
6680 inode_init_owner(mnt_userns, inode, dir, mode);
6681 inode->i_op = &btrfs_special_inode_operations;
6682 init_special_inode(inode, inode->i_mode, rdev);
6683 return btrfs_create_common(dir, dentry, inode);
6686 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6687 struct dentry *dentry, umode_t mode, bool excl)
6689 struct inode *inode;
6691 inode = new_inode(dir->i_sb);
6694 inode_init_owner(mnt_userns, inode, dir, mode);
6695 inode->i_fop = &btrfs_file_operations;
6696 inode->i_op = &btrfs_file_inode_operations;
6697 inode->i_mapping->a_ops = &btrfs_aops;
6698 return btrfs_create_common(dir, dentry, inode);
6701 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6702 struct dentry *dentry)
6704 struct btrfs_trans_handle *trans = NULL;
6705 struct btrfs_root *root = BTRFS_I(dir)->root;
6706 struct inode *inode = d_inode(old_dentry);
6707 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6712 /* do not allow sys_link's with other subvols of the same device */
6713 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6716 if (inode->i_nlink >= BTRFS_LINK_MAX)
6719 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6724 * 2 items for inode and inode ref
6725 * 2 items for dir items
6726 * 1 item for parent inode
6727 * 1 item for orphan item deletion if O_TMPFILE
6729 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6730 if (IS_ERR(trans)) {
6731 err = PTR_ERR(trans);
6736 /* There are several dir indexes for this inode, clear the cache. */
6737 BTRFS_I(inode)->dir_index = 0ULL;
6739 inode_inc_iversion(inode);
6740 inode->i_ctime = current_time(inode);
6742 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6744 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6745 dentry->d_name.name, dentry->d_name.len, 1, index);
6750 struct dentry *parent = dentry->d_parent;
6752 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6755 if (inode->i_nlink == 1) {
6757 * If new hard link count is 1, it's a file created
6758 * with open(2) O_TMPFILE flag.
6760 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6764 d_instantiate(dentry, inode);
6765 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6770 btrfs_end_transaction(trans);
6772 inode_dec_link_count(inode);
6775 btrfs_btree_balance_dirty(fs_info);
6779 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6780 struct dentry *dentry, umode_t mode)
6782 struct inode *inode;
6784 inode = new_inode(dir->i_sb);
6787 inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode);
6788 inode->i_op = &btrfs_dir_inode_operations;
6789 inode->i_fop = &btrfs_dir_file_operations;
6790 return btrfs_create_common(dir, dentry, inode);
6793 static noinline int uncompress_inline(struct btrfs_path *path,
6795 size_t pg_offset, u64 extent_offset,
6796 struct btrfs_file_extent_item *item)
6799 struct extent_buffer *leaf = path->nodes[0];
6802 unsigned long inline_size;
6806 WARN_ON(pg_offset != 0);
6807 compress_type = btrfs_file_extent_compression(leaf, item);
6808 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6809 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6810 tmp = kmalloc(inline_size, GFP_NOFS);
6813 ptr = btrfs_file_extent_inline_start(item);
6815 read_extent_buffer(leaf, tmp, ptr, inline_size);
6817 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6818 ret = btrfs_decompress(compress_type, tmp, page,
6819 extent_offset, inline_size, max_size);
6822 * decompression code contains a memset to fill in any space between the end
6823 * of the uncompressed data and the end of max_size in case the decompressed
6824 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6825 * the end of an inline extent and the beginning of the next block, so we
6826 * cover that region here.
6829 if (max_size + pg_offset < PAGE_SIZE)
6830 memzero_page(page, pg_offset + max_size,
6831 PAGE_SIZE - max_size - pg_offset);
6837 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6838 * @inode: file to search in
6839 * @page: page to read extent data into if the extent is inline
6840 * @pg_offset: offset into @page to copy to
6841 * @start: file offset
6842 * @len: length of range starting at @start
6844 * This returns the first &struct extent_map which overlaps with the given
6845 * range, reading it from the B-tree and caching it if necessary. Note that
6846 * there may be more extents which overlap the given range after the returned
6849 * If @page is not NULL and the extent is inline, this also reads the extent
6850 * data directly into the page and marks the extent up to date in the io_tree.
6852 * Return: ERR_PTR on error, non-NULL extent_map on success.
6854 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6855 struct page *page, size_t pg_offset,
6858 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6860 u64 extent_start = 0;
6862 u64 objectid = btrfs_ino(inode);
6863 int extent_type = -1;
6864 struct btrfs_path *path = NULL;
6865 struct btrfs_root *root = inode->root;
6866 struct btrfs_file_extent_item *item;
6867 struct extent_buffer *leaf;
6868 struct btrfs_key found_key;
6869 struct extent_map *em = NULL;
6870 struct extent_map_tree *em_tree = &inode->extent_tree;
6871 struct extent_io_tree *io_tree = &inode->io_tree;
6873 read_lock(&em_tree->lock);
6874 em = lookup_extent_mapping(em_tree, start, len);
6875 read_unlock(&em_tree->lock);
6878 if (em->start > start || em->start + em->len <= start)
6879 free_extent_map(em);
6880 else if (em->block_start == EXTENT_MAP_INLINE && page)
6881 free_extent_map(em);
6885 em = alloc_extent_map();
6890 em->start = EXTENT_MAP_HOLE;
6891 em->orig_start = EXTENT_MAP_HOLE;
6893 em->block_len = (u64)-1;
6895 path = btrfs_alloc_path();
6901 /* Chances are we'll be called again, so go ahead and do readahead */
6902 path->reada = READA_FORWARD;
6905 * The same explanation in load_free_space_cache applies here as well,
6906 * we only read when we're loading the free space cache, and at that
6907 * point the commit_root has everything we need.
6909 if (btrfs_is_free_space_inode(inode)) {
6910 path->search_commit_root = 1;
6911 path->skip_locking = 1;
6914 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6917 } else if (ret > 0) {
6918 if (path->slots[0] == 0)
6924 leaf = path->nodes[0];
6925 item = btrfs_item_ptr(leaf, path->slots[0],
6926 struct btrfs_file_extent_item);
6927 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6928 if (found_key.objectid != objectid ||
6929 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6931 * If we backup past the first extent we want to move forward
6932 * and see if there is an extent in front of us, otherwise we'll
6933 * say there is a hole for our whole search range which can
6940 extent_type = btrfs_file_extent_type(leaf, item);
6941 extent_start = found_key.offset;
6942 extent_end = btrfs_file_extent_end(path);
6943 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6944 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6945 /* Only regular file could have regular/prealloc extent */
6946 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6949 "regular/prealloc extent found for non-regular inode %llu",
6953 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6955 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6956 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6961 if (start >= extent_end) {
6963 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6964 ret = btrfs_next_leaf(root, path);
6970 leaf = path->nodes[0];
6972 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6973 if (found_key.objectid != objectid ||
6974 found_key.type != BTRFS_EXTENT_DATA_KEY)
6976 if (start + len <= found_key.offset)
6978 if (start > found_key.offset)
6981 /* New extent overlaps with existing one */
6983 em->orig_start = start;
6984 em->len = found_key.offset - start;
6985 em->block_start = EXTENT_MAP_HOLE;
6989 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6991 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6992 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6994 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6998 size_t extent_offset;
7004 size = btrfs_file_extent_ram_bytes(leaf, item);
7005 extent_offset = page_offset(page) + pg_offset - extent_start;
7006 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7007 size - extent_offset);
7008 em->start = extent_start + extent_offset;
7009 em->len = ALIGN(copy_size, fs_info->sectorsize);
7010 em->orig_block_len = em->len;
7011 em->orig_start = em->start;
7012 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7014 if (!PageUptodate(page)) {
7015 if (btrfs_file_extent_compression(leaf, item) !=
7016 BTRFS_COMPRESS_NONE) {
7017 ret = uncompress_inline(path, page, pg_offset,
7018 extent_offset, item);
7022 map = kmap_local_page(page);
7023 read_extent_buffer(leaf, map + pg_offset, ptr,
7025 if (pg_offset + copy_size < PAGE_SIZE) {
7026 memset(map + pg_offset + copy_size, 0,
7027 PAGE_SIZE - pg_offset -
7032 flush_dcache_page(page);
7034 set_extent_uptodate(io_tree, em->start,
7035 extent_map_end(em) - 1, NULL, GFP_NOFS);
7040 em->orig_start = start;
7042 em->block_start = EXTENT_MAP_HOLE;
7045 btrfs_release_path(path);
7046 if (em->start > start || extent_map_end(em) <= start) {
7048 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7049 em->start, em->len, start, len);
7054 write_lock(&em_tree->lock);
7055 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7056 write_unlock(&em_tree->lock);
7058 btrfs_free_path(path);
7060 trace_btrfs_get_extent(root, inode, em);
7063 free_extent_map(em);
7064 return ERR_PTR(ret);
7069 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7072 struct extent_map *em;
7073 struct extent_map *hole_em = NULL;
7074 u64 delalloc_start = start;
7080 em = btrfs_get_extent(inode, NULL, 0, start, len);
7084 * If our em maps to:
7086 * - a pre-alloc extent,
7087 * there might actually be delalloc bytes behind it.
7089 if (em->block_start != EXTENT_MAP_HOLE &&
7090 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7095 /* check to see if we've wrapped (len == -1 or similar) */
7104 /* ok, we didn't find anything, lets look for delalloc */
7105 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7106 end, len, EXTENT_DELALLOC, 1);
7107 delalloc_end = delalloc_start + delalloc_len;
7108 if (delalloc_end < delalloc_start)
7109 delalloc_end = (u64)-1;
7112 * We didn't find anything useful, return the original results from
7115 if (delalloc_start > end || delalloc_end <= start) {
7122 * Adjust the delalloc_start to make sure it doesn't go backwards from
7123 * the start they passed in
7125 delalloc_start = max(start, delalloc_start);
7126 delalloc_len = delalloc_end - delalloc_start;
7128 if (delalloc_len > 0) {
7131 const u64 hole_end = extent_map_end(hole_em);
7133 em = alloc_extent_map();
7141 * When btrfs_get_extent can't find anything it returns one
7144 * Make sure what it found really fits our range, and adjust to
7145 * make sure it is based on the start from the caller
7147 if (hole_end <= start || hole_em->start > end) {
7148 free_extent_map(hole_em);
7151 hole_start = max(hole_em->start, start);
7152 hole_len = hole_end - hole_start;
7155 if (hole_em && delalloc_start > hole_start) {
7157 * Our hole starts before our delalloc, so we have to
7158 * return just the parts of the hole that go until the
7161 em->len = min(hole_len, delalloc_start - hole_start);
7162 em->start = hole_start;
7163 em->orig_start = hole_start;
7165 * Don't adjust block start at all, it is fixed at
7168 em->block_start = hole_em->block_start;
7169 em->block_len = hole_len;
7170 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7171 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7174 * Hole is out of passed range or it starts after
7177 em->start = delalloc_start;
7178 em->len = delalloc_len;
7179 em->orig_start = delalloc_start;
7180 em->block_start = EXTENT_MAP_DELALLOC;
7181 em->block_len = delalloc_len;
7188 free_extent_map(hole_em);
7190 free_extent_map(em);
7191 return ERR_PTR(err);
7196 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7199 const u64 orig_start,
7200 const u64 block_start,
7201 const u64 block_len,
7202 const u64 orig_block_len,
7203 const u64 ram_bytes,
7206 struct extent_map *em = NULL;
7209 if (type != BTRFS_ORDERED_NOCOW) {
7210 em = create_io_em(inode, start, len, orig_start, block_start,
7211 block_len, orig_block_len, ram_bytes,
7212 BTRFS_COMPRESS_NONE, /* compress_type */
7217 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7220 (1 << BTRFS_ORDERED_DIRECT),
7221 BTRFS_COMPRESS_NONE);
7224 free_extent_map(em);
7225 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7234 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7237 struct btrfs_root *root = inode->root;
7238 struct btrfs_fs_info *fs_info = root->fs_info;
7239 struct extent_map *em;
7240 struct btrfs_key ins;
7244 alloc_hint = get_extent_allocation_hint(inode, start, len);
7245 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7246 0, alloc_hint, &ins, 1, 1);
7248 return ERR_PTR(ret);
7250 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7251 ins.objectid, ins.offset, ins.offset,
7252 ins.offset, BTRFS_ORDERED_REGULAR);
7253 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7255 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7261 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7263 struct btrfs_block_group *block_group;
7264 bool readonly = false;
7266 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7267 if (!block_group || block_group->ro)
7270 btrfs_put_block_group(block_group);
7275 * Check if we can do nocow write into the range [@offset, @offset + @len)
7277 * @offset: File offset
7278 * @len: The length to write, will be updated to the nocow writeable
7280 * @orig_start: (optional) Return the original file offset of the file extent
7281 * @orig_len: (optional) Return the original on-disk length of the file extent
7282 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7283 * @strict: if true, omit optimizations that might force us into unnecessary
7284 * cow. e.g., don't trust generation number.
7287 * >0 and update @len if we can do nocow write
7288 * 0 if we can't do nocow write
7289 * <0 if error happened
7291 * NOTE: This only checks the file extents, caller is responsible to wait for
7292 * any ordered extents.
7294 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7295 u64 *orig_start, u64 *orig_block_len,
7296 u64 *ram_bytes, bool strict)
7298 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7299 struct can_nocow_file_extent_args nocow_args = { 0 };
7300 struct btrfs_path *path;
7302 struct extent_buffer *leaf;
7303 struct btrfs_root *root = BTRFS_I(inode)->root;
7304 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7305 struct btrfs_file_extent_item *fi;
7306 struct btrfs_key key;
7309 path = btrfs_alloc_path();
7313 ret = btrfs_lookup_file_extent(NULL, root, path,
7314 btrfs_ino(BTRFS_I(inode)), offset, 0);
7319 if (path->slots[0] == 0) {
7320 /* can't find the item, must cow */
7327 leaf = path->nodes[0];
7328 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7329 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7330 key.type != BTRFS_EXTENT_DATA_KEY) {
7331 /* not our file or wrong item type, must cow */
7335 if (key.offset > offset) {
7336 /* Wrong offset, must cow */
7340 if (btrfs_file_extent_end(path) <= offset)
7343 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7344 found_type = btrfs_file_extent_type(leaf, fi);
7346 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7348 nocow_args.start = offset;
7349 nocow_args.end = offset + *len - 1;
7350 nocow_args.strict = strict;
7351 nocow_args.free_path = true;
7353 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7354 /* can_nocow_file_extent() has freed the path. */
7358 /* Treat errors as not being able to NOCOW. */
7364 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7367 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7368 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7371 range_end = round_up(offset + nocow_args.num_bytes,
7372 root->fs_info->sectorsize) - 1;
7373 ret = test_range_bit(io_tree, offset, range_end,
7374 EXTENT_DELALLOC, 0, NULL);
7382 *orig_start = key.offset - nocow_args.extent_offset;
7384 *orig_block_len = nocow_args.disk_num_bytes;
7386 *len = nocow_args.num_bytes;
7389 btrfs_free_path(path);
7393 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7394 struct extent_state **cached_state,
7395 unsigned int iomap_flags)
7397 const bool writing = (iomap_flags & IOMAP_WRITE);
7398 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7399 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7400 struct btrfs_ordered_extent *ordered;
7405 if (!try_lock_extent(io_tree, lockstart, lockend))
7408 lock_extent_bits(io_tree, lockstart, lockend, cached_state);
7411 * We're concerned with the entire range that we're going to be
7412 * doing DIO to, so we need to make sure there's no ordered
7413 * extents in this range.
7415 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7416 lockend - lockstart + 1);
7419 * We need to make sure there are no buffered pages in this
7420 * range either, we could have raced between the invalidate in
7421 * generic_file_direct_write and locking the extent. The
7422 * invalidate needs to happen so that reads after a write do not
7426 (!writing || !filemap_range_has_page(inode->i_mapping,
7427 lockstart, lockend)))
7430 unlock_extent_cached(io_tree, lockstart, lockend, cached_state);
7434 btrfs_put_ordered_extent(ordered);
7439 * If we are doing a DIO read and the ordered extent we
7440 * found is for a buffered write, we can not wait for it
7441 * to complete and retry, because if we do so we can
7442 * deadlock with concurrent buffered writes on page
7443 * locks. This happens only if our DIO read covers more
7444 * than one extent map, if at this point has already
7445 * created an ordered extent for a previous extent map
7446 * and locked its range in the inode's io tree, and a
7447 * concurrent write against that previous extent map's
7448 * range and this range started (we unlock the ranges
7449 * in the io tree only when the bios complete and
7450 * buffered writes always lock pages before attempting
7451 * to lock range in the io tree).
7454 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7455 btrfs_start_ordered_extent(ordered, 1);
7457 ret = nowait ? -EAGAIN : -ENOTBLK;
7458 btrfs_put_ordered_extent(ordered);
7461 * We could trigger writeback for this range (and wait
7462 * for it to complete) and then invalidate the pages for
7463 * this range (through invalidate_inode_pages2_range()),
7464 * but that can lead us to a deadlock with a concurrent
7465 * call to readahead (a buffered read or a defrag call
7466 * triggered a readahead) on a page lock due to an
7467 * ordered dio extent we created before but did not have
7468 * yet a corresponding bio submitted (whence it can not
7469 * complete), which makes readahead wait for that
7470 * ordered extent to complete while holding a lock on
7473 ret = nowait ? -EAGAIN : -ENOTBLK;
7485 /* The callers of this must take lock_extent() */
7486 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7487 u64 len, u64 orig_start, u64 block_start,
7488 u64 block_len, u64 orig_block_len,
7489 u64 ram_bytes, int compress_type,
7492 struct extent_map_tree *em_tree;
7493 struct extent_map *em;
7496 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7497 type == BTRFS_ORDERED_COMPRESSED ||
7498 type == BTRFS_ORDERED_NOCOW ||
7499 type == BTRFS_ORDERED_REGULAR);
7501 em_tree = &inode->extent_tree;
7502 em = alloc_extent_map();
7504 return ERR_PTR(-ENOMEM);
7507 em->orig_start = orig_start;
7509 em->block_len = block_len;
7510 em->block_start = block_start;
7511 em->orig_block_len = orig_block_len;
7512 em->ram_bytes = ram_bytes;
7513 em->generation = -1;
7514 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7515 if (type == BTRFS_ORDERED_PREALLOC) {
7516 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7517 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7518 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7519 em->compress_type = compress_type;
7523 btrfs_drop_extent_cache(inode, em->start,
7524 em->start + em->len - 1, 0);
7525 write_lock(&em_tree->lock);
7526 ret = add_extent_mapping(em_tree, em, 1);
7527 write_unlock(&em_tree->lock);
7529 * The caller has taken lock_extent(), who could race with us
7532 } while (ret == -EEXIST);
7535 free_extent_map(em);
7536 return ERR_PTR(ret);
7539 /* em got 2 refs now, callers needs to do free_extent_map once. */
7544 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7545 struct inode *inode,
7546 struct btrfs_dio_data *dio_data,
7548 unsigned int iomap_flags)
7550 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7551 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7552 struct extent_map *em = *map;
7554 u64 block_start, orig_start, orig_block_len, ram_bytes;
7555 struct btrfs_block_group *bg;
7556 bool can_nocow = false;
7557 bool space_reserved = false;
7562 * We don't allocate a new extent in the following cases
7564 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7566 * 2) The extent is marked as PREALLOC. We're good to go here and can
7567 * just use the extent.
7570 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7571 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7572 em->block_start != EXTENT_MAP_HOLE)) {
7573 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7574 type = BTRFS_ORDERED_PREALLOC;
7576 type = BTRFS_ORDERED_NOCOW;
7577 len = min(len, em->len - (start - em->start));
7578 block_start = em->block_start + (start - em->start);
7580 if (can_nocow_extent(inode, start, &len, &orig_start,
7581 &orig_block_len, &ram_bytes, false) == 1) {
7582 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7590 struct extent_map *em2;
7592 /* We can NOCOW, so only need to reserve metadata space. */
7593 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7596 /* Our caller expects us to free the input extent map. */
7597 free_extent_map(em);
7599 btrfs_dec_nocow_writers(bg);
7600 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7604 space_reserved = true;
7606 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7607 orig_start, block_start,
7608 len, orig_block_len,
7610 btrfs_dec_nocow_writers(bg);
7611 if (type == BTRFS_ORDERED_PREALLOC) {
7612 free_extent_map(em);
7622 dio_data->nocow_done = true;
7624 /* Our caller expects us to free the input extent map. */
7625 free_extent_map(em);
7632 * If we could not allocate data space before locking the file
7633 * range and we can't do a NOCOW write, then we have to fail.
7635 if (!dio_data->data_space_reserved)
7639 * We have to COW and we have already reserved data space before,
7640 * so now we reserve only metadata.
7642 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7646 space_reserved = true;
7648 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7654 len = min(len, em->len - (start - em->start));
7656 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7657 prev_len - len, true);
7661 * We have created our ordered extent, so we can now release our reservation
7662 * for an outstanding extent.
7664 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7667 * Need to update the i_size under the extent lock so buffered
7668 * readers will get the updated i_size when we unlock.
7670 if (start + len > i_size_read(inode))
7671 i_size_write(inode, start + len);
7673 if (ret && space_reserved) {
7674 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7675 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7680 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7681 loff_t length, unsigned int flags, struct iomap *iomap,
7682 struct iomap *srcmap)
7684 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7685 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7686 struct extent_map *em;
7687 struct extent_state *cached_state = NULL;
7688 struct btrfs_dio_data *dio_data = iter->private;
7689 u64 lockstart, lockend;
7690 const bool write = !!(flags & IOMAP_WRITE);
7693 const u64 data_alloc_len = length;
7694 bool unlock_extents = false;
7697 * Cap the size of reads to that usually seen in buffered I/O as we need
7698 * to allocate a contiguous array for the checksums.
7701 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7704 lockend = start + len - 1;
7707 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7708 * enough if we've written compressed pages to this area, so we need to
7709 * flush the dirty pages again to make absolutely sure that any
7710 * outstanding dirty pages are on disk - the first flush only starts
7711 * compression on the data, while keeping the pages locked, so by the
7712 * time the second flush returns we know bios for the compressed pages
7713 * were submitted and finished, and the pages no longer under writeback.
7715 * If we have a NOWAIT request and we have any pages in the range that
7716 * are locked, likely due to compression still in progress, we don't want
7717 * to block on page locks. We also don't want to block on pages marked as
7718 * dirty or under writeback (same as for the non-compression case).
7719 * iomap_dio_rw() did the same check, but after that and before we got
7720 * here, mmap'ed writes may have happened or buffered reads started
7721 * (readpage() and readahead(), which lock pages), as we haven't locked
7722 * the file range yet.
7724 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7725 &BTRFS_I(inode)->runtime_flags)) {
7726 if (flags & IOMAP_NOWAIT) {
7727 if (filemap_range_needs_writeback(inode->i_mapping,
7728 lockstart, lockend))
7731 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7732 start + length - 1);
7738 memset(dio_data, 0, sizeof(*dio_data));
7741 * We always try to allocate data space and must do it before locking
7742 * the file range, to avoid deadlocks with concurrent writes to the same
7743 * range if the range has several extents and the writes don't expand the
7744 * current i_size (the inode lock is taken in shared mode). If we fail to
7745 * allocate data space here we continue and later, after locking the
7746 * file range, we fail with ENOSPC only if we figure out we can not do a
7749 if (write && !(flags & IOMAP_NOWAIT)) {
7750 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7751 &dio_data->data_reserved,
7752 start, data_alloc_len);
7754 dio_data->data_space_reserved = true;
7755 else if (ret && !(BTRFS_I(inode)->flags &
7756 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7761 * If this errors out it's because we couldn't invalidate pagecache for
7762 * this range and we need to fallback to buffered IO, or we are doing a
7763 * NOWAIT read/write and we need to block.
7765 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7769 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7776 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7777 * io. INLINE is special, and we could probably kludge it in here, but
7778 * it's still buffered so for safety lets just fall back to the generic
7781 * For COMPRESSED we _have_ to read the entire extent in so we can
7782 * decompress it, so there will be buffering required no matter what we
7783 * do, so go ahead and fallback to buffered.
7785 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7786 * to buffered IO. Don't blame me, this is the price we pay for using
7789 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7790 em->block_start == EXTENT_MAP_INLINE) {
7791 free_extent_map(em);
7793 * If we are in a NOWAIT context, return -EAGAIN in order to
7794 * fallback to buffered IO. This is not only because we can
7795 * block with buffered IO (no support for NOWAIT semantics at
7796 * the moment) but also to avoid returning short reads to user
7797 * space - this happens if we were able to read some data from
7798 * previous non-compressed extents and then when we fallback to
7799 * buffered IO, at btrfs_file_read_iter() by calling
7800 * filemap_read(), we fail to fault in pages for the read buffer,
7801 * in which case filemap_read() returns a short read (the number
7802 * of bytes previously read is > 0, so it does not return -EFAULT).
7804 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7808 len = min(len, em->len - (start - em->start));
7811 * If we have a NOWAIT request and the range contains multiple extents
7812 * (or a mix of extents and holes), then we return -EAGAIN to make the
7813 * caller fallback to a context where it can do a blocking (without
7814 * NOWAIT) request. This way we avoid doing partial IO and returning
7815 * success to the caller, which is not optimal for writes and for reads
7816 * it can result in unexpected behaviour for an application.
7818 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7819 * iomap_dio_rw(), we can end up returning less data then what the caller
7820 * asked for, resulting in an unexpected, and incorrect, short read.
7821 * That is, the caller asked to read N bytes and we return less than that,
7822 * which is wrong unless we are crossing EOF. This happens if we get a
7823 * page fault error when trying to fault in pages for the buffer that is
7824 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7825 * have previously submitted bios for other extents in the range, in
7826 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7827 * those bios have completed by the time we get the page fault error,
7828 * which we return back to our caller - we should only return EIOCBQUEUED
7829 * after we have submitted bios for all the extents in the range.
7831 if ((flags & IOMAP_NOWAIT) && len < length) {
7832 free_extent_map(em);
7838 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7842 unlock_extents = true;
7843 /* Recalc len in case the new em is smaller than requested */
7844 len = min(len, em->len - (start - em->start));
7845 if (dio_data->data_space_reserved) {
7847 u64 release_len = 0;
7849 if (dio_data->nocow_done) {
7850 release_offset = start;
7851 release_len = data_alloc_len;
7852 } else if (len < data_alloc_len) {
7853 release_offset = start + len;
7854 release_len = data_alloc_len - len;
7857 if (release_len > 0)
7858 btrfs_free_reserved_data_space(BTRFS_I(inode),
7859 dio_data->data_reserved,
7865 * We need to unlock only the end area that we aren't using.
7866 * The rest is going to be unlocked by the endio routine.
7868 lockstart = start + len;
7869 if (lockstart < lockend)
7870 unlock_extents = true;
7874 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7875 lockstart, lockend, &cached_state);
7877 free_extent_state(cached_state);
7880 * Translate extent map information to iomap.
7881 * We trim the extents (and move the addr) even though iomap code does
7882 * that, since we have locked only the parts we are performing I/O in.
7884 if ((em->block_start == EXTENT_MAP_HOLE) ||
7885 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7886 iomap->addr = IOMAP_NULL_ADDR;
7887 iomap->type = IOMAP_HOLE;
7889 iomap->addr = em->block_start + (start - em->start);
7890 iomap->type = IOMAP_MAPPED;
7892 iomap->offset = start;
7893 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7894 iomap->length = len;
7896 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7897 iomap->flags |= IOMAP_F_ZONE_APPEND;
7899 free_extent_map(em);
7904 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7907 if (dio_data->data_space_reserved) {
7908 btrfs_free_reserved_data_space(BTRFS_I(inode),
7909 dio_data->data_reserved,
7910 start, data_alloc_len);
7911 extent_changeset_free(dio_data->data_reserved);
7917 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7918 ssize_t written, unsigned int flags, struct iomap *iomap)
7920 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7921 struct btrfs_dio_data *dio_data = iter->private;
7922 size_t submitted = dio_data->submitted;
7923 const bool write = !!(flags & IOMAP_WRITE);
7926 if (!write && (iomap->type == IOMAP_HOLE)) {
7927 /* If reading from a hole, unlock and return */
7928 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7932 if (submitted < length) {
7934 length -= submitted;
7936 btrfs_mark_ordered_io_finished(BTRFS_I(inode), NULL,
7937 pos, length, false);
7939 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7945 extent_changeset_free(dio_data->data_reserved);
7949 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7952 * This implies a barrier so that stores to dio_bio->bi_status before
7953 * this and loads of dio_bio->bi_status after this are fully ordered.
7955 if (!refcount_dec_and_test(&dip->refs))
7958 if (btrfs_op(&dip->bio) == BTRFS_MAP_WRITE) {
7959 btrfs_mark_ordered_io_finished(BTRFS_I(dip->inode), NULL,
7960 dip->file_offset, dip->bytes,
7961 !dip->bio.bi_status);
7963 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7965 dip->file_offset + dip->bytes - 1);
7969 bio_endio(&dip->bio);
7972 static void submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7974 enum btrfs_compression_type compress_type)
7976 struct btrfs_dio_private *dip = bio->bi_private;
7977 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7979 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7981 refcount_inc(&dip->refs);
7982 btrfs_submit_bio(fs_info, bio, mirror_num);
7985 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7986 struct btrfs_bio *bbio,
7987 const bool uptodate)
7989 struct inode *inode = dip->inode;
7990 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7991 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7992 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7993 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7994 blk_status_t err = BLK_STS_OK;
7995 struct bvec_iter iter;
7999 btrfs_bio_for_each_sector(fs_info, bv, bbio, iter, offset) {
8000 u64 start = bbio->file_offset + offset;
8003 (!csum || !btrfs_check_data_csum(inode, bbio, offset, bv.bv_page,
8005 clean_io_failure(fs_info, failure_tree, io_tree, start,
8006 bv.bv_page, btrfs_ino(BTRFS_I(inode)),
8011 ret = btrfs_repair_one_sector(inode, bbio, offset,
8012 bv.bv_page, bv.bv_offset,
8013 submit_dio_repair_bio);
8015 err = errno_to_blk_status(ret);
8022 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8024 u64 dio_file_offset)
8026 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
8029 static void btrfs_end_dio_bio(struct bio *bio)
8031 struct btrfs_dio_private *dip = bio->bi_private;
8032 struct btrfs_bio *bbio = btrfs_bio(bio);
8033 blk_status_t err = bio->bi_status;
8036 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8037 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8038 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8039 bio->bi_opf, bio->bi_iter.bi_sector,
8040 bio->bi_iter.bi_size, err);
8042 if (bio_op(bio) == REQ_OP_READ)
8043 err = btrfs_check_read_dio_bio(dip, bbio, !err);
8046 dip->bio.bi_status = err;
8048 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
8051 btrfs_dio_private_put(dip);
8054 static void btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8055 u64 file_offset, int async_submit)
8057 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8058 struct btrfs_dio_private *dip = bio->bi_private;
8061 /* Save the original iter for read repair */
8062 if (btrfs_op(bio) == BTRFS_MAP_READ)
8063 btrfs_bio(bio)->iter = bio->bi_iter;
8065 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8068 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
8069 /* Check btrfs_submit_data_write_bio() for async submit rules */
8070 if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers) &&
8071 btrfs_wq_submit_bio(inode, bio, 0, file_offset,
8072 btrfs_submit_bio_start_direct_io))
8076 * If we aren't doing async submit, calculate the csum of the
8079 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
8081 bio->bi_status = ret;
8086 btrfs_bio(bio)->csum = btrfs_csum_ptr(fs_info, dip->csums,
8087 file_offset - dip->file_offset);
8090 btrfs_submit_bio(fs_info, bio, 0);
8093 static void btrfs_submit_direct(const struct iomap_iter *iter,
8094 struct bio *dio_bio, loff_t file_offset)
8096 struct btrfs_dio_private *dip =
8097 container_of(dio_bio, struct btrfs_dio_private, bio);
8098 struct inode *inode = iter->inode;
8099 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8100 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8101 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8102 BTRFS_BLOCK_GROUP_RAID56_MASK);
8105 int async_submit = 0;
8107 u64 clone_offset = 0;
8111 blk_status_t status;
8112 struct btrfs_io_geometry geom;
8113 struct btrfs_dio_data *dio_data = iter->private;
8114 struct extent_map *em = NULL;
8117 dip->file_offset = file_offset;
8118 dip->bytes = dio_bio->bi_iter.bi_size;
8119 refcount_set(&dip->refs, 1);
8122 if (!write && !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
8123 unsigned int nr_sectors =
8124 (dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits);
8127 * Load the csums up front to reduce csum tree searches and
8128 * contention when submitting bios.
8130 status = BLK_STS_RESOURCE;
8131 dip->csums = kcalloc(nr_sectors, fs_info->csum_size, GFP_NOFS);
8135 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8136 if (status != BLK_STS_OK)
8140 start_sector = dio_bio->bi_iter.bi_sector;
8141 submit_len = dio_bio->bi_iter.bi_size;
8144 logical = start_sector << 9;
8145 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8147 status = errno_to_blk_status(PTR_ERR(em));
8151 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8154 status = errno_to_blk_status(ret);
8158 clone_len = min(submit_len, geom.len);
8159 ASSERT(clone_len <= UINT_MAX);
8162 * This will never fail as it's passing GPF_NOFS and
8163 * the allocation is backed by btrfs_bioset.
8165 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8166 bio->bi_private = dip;
8167 bio->bi_end_io = btrfs_end_dio_bio;
8168 btrfs_bio(bio)->file_offset = file_offset;
8170 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8171 status = extract_ordered_extent(BTRFS_I(inode), bio,
8179 ASSERT(submit_len >= clone_len);
8180 submit_len -= clone_len;
8183 * Increase the count before we submit the bio so we know
8184 * the end IO handler won't happen before we increase the
8185 * count. Otherwise, the dip might get freed before we're
8186 * done setting it up.
8188 * We transfer the initial reference to the last bio, so we
8189 * don't need to increment the reference count for the last one.
8191 if (submit_len > 0) {
8192 refcount_inc(&dip->refs);
8194 * If we are submitting more than one bio, submit them
8195 * all asynchronously. The exception is RAID 5 or 6, as
8196 * asynchronous checksums make it difficult to collect
8197 * full stripe writes.
8203 btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8205 dio_data->submitted += clone_len;
8206 clone_offset += clone_len;
8207 start_sector += clone_len >> 9;
8208 file_offset += clone_len;
8210 free_extent_map(em);
8211 } while (submit_len > 0);
8215 free_extent_map(em);
8217 dio_bio->bi_status = status;
8218 btrfs_dio_private_put(dip);
8221 static const struct iomap_ops btrfs_dio_iomap_ops = {
8222 .iomap_begin = btrfs_dio_iomap_begin,
8223 .iomap_end = btrfs_dio_iomap_end,
8226 static const struct iomap_dio_ops btrfs_dio_ops = {
8227 .submit_io = btrfs_submit_direct,
8228 .bio_set = &btrfs_dio_bioset,
8231 ssize_t btrfs_dio_rw(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
8233 struct btrfs_dio_data data;
8235 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
8236 IOMAP_DIO_PARTIAL | IOMAP_DIO_NOSYNC,
8237 &data, done_before);
8240 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8245 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8249 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8252 static int btrfs_writepages(struct address_space *mapping,
8253 struct writeback_control *wbc)
8255 return extent_writepages(mapping, wbc);
8258 static void btrfs_readahead(struct readahead_control *rac)
8260 extent_readahead(rac);
8264 * For release_folio() and invalidate_folio() we have a race window where
8265 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8266 * If we continue to release/invalidate the page, we could cause use-after-free
8267 * for subpage spinlock. So this function is to spin and wait for subpage
8270 static void wait_subpage_spinlock(struct page *page)
8272 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8273 struct btrfs_subpage *subpage;
8275 if (!btrfs_is_subpage(fs_info, page))
8278 ASSERT(PagePrivate(page) && page->private);
8279 subpage = (struct btrfs_subpage *)page->private;
8282 * This may look insane as we just acquire the spinlock and release it,
8283 * without doing anything. But we just want to make sure no one is
8284 * still holding the subpage spinlock.
8285 * And since the page is not dirty nor writeback, and we have page
8286 * locked, the only possible way to hold a spinlock is from the endio
8287 * function to clear page writeback.
8289 * Here we just acquire the spinlock so that all existing callers
8290 * should exit and we're safe to release/invalidate the page.
8292 spin_lock_irq(&subpage->lock);
8293 spin_unlock_irq(&subpage->lock);
8296 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8298 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8301 wait_subpage_spinlock(&folio->page);
8302 clear_page_extent_mapped(&folio->page);
8307 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8309 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8311 return __btrfs_release_folio(folio, gfp_flags);
8314 #ifdef CONFIG_MIGRATION
8315 static int btrfs_migrate_folio(struct address_space *mapping,
8316 struct folio *dst, struct folio *src,
8317 enum migrate_mode mode)
8319 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8321 if (ret != MIGRATEPAGE_SUCCESS)
8324 if (folio_test_ordered(src)) {
8325 folio_clear_ordered(src);
8326 folio_set_ordered(dst);
8329 return MIGRATEPAGE_SUCCESS;
8332 #define btrfs_migrate_folio NULL
8335 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8338 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8339 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8340 struct extent_io_tree *tree = &inode->io_tree;
8341 struct extent_state *cached_state = NULL;
8342 u64 page_start = folio_pos(folio);
8343 u64 page_end = page_start + folio_size(folio) - 1;
8345 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8348 * We have folio locked so no new ordered extent can be created on this
8349 * page, nor bio can be submitted for this folio.
8351 * But already submitted bio can still be finished on this folio.
8352 * Furthermore, endio function won't skip folio which has Ordered
8353 * (Private2) already cleared, so it's possible for endio and
8354 * invalidate_folio to do the same ordered extent accounting twice
8357 * So here we wait for any submitted bios to finish, so that we won't
8358 * do double ordered extent accounting on the same folio.
8360 folio_wait_writeback(folio);
8361 wait_subpage_spinlock(&folio->page);
8364 * For subpage case, we have call sites like
8365 * btrfs_punch_hole_lock_range() which passes range not aligned to
8367 * If the range doesn't cover the full folio, we don't need to and
8368 * shouldn't clear page extent mapped, as folio->private can still
8369 * record subpage dirty bits for other part of the range.
8371 * For cases that invalidate the full folio even the range doesn't
8372 * cover the full folio, like invalidating the last folio, we're
8373 * still safe to wait for ordered extent to finish.
8375 if (!(offset == 0 && length == folio_size(folio))) {
8376 btrfs_release_folio(folio, GFP_NOFS);
8380 if (!inode_evicting)
8381 lock_extent_bits(tree, page_start, page_end, &cached_state);
8384 while (cur < page_end) {
8385 struct btrfs_ordered_extent *ordered;
8390 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8391 page_end + 1 - cur);
8393 range_end = page_end;
8395 * No ordered extent covering this range, we are safe
8396 * to delete all extent states in the range.
8398 delete_states = true;
8401 if (ordered->file_offset > cur) {
8403 * There is a range between [cur, oe->file_offset) not
8404 * covered by any ordered extent.
8405 * We are safe to delete all extent states, and handle
8406 * the ordered extent in the next iteration.
8408 range_end = ordered->file_offset - 1;
8409 delete_states = true;
8413 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8415 ASSERT(range_end + 1 - cur < U32_MAX);
8416 range_len = range_end + 1 - cur;
8417 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8419 * If Ordered (Private2) is cleared, it means endio has
8420 * already been executed for the range.
8421 * We can't delete the extent states as
8422 * btrfs_finish_ordered_io() may still use some of them.
8424 delete_states = false;
8427 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8430 * IO on this page will never be started, so we need to account
8431 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8432 * here, must leave that up for the ordered extent completion.
8434 * This will also unlock the range for incoming
8435 * btrfs_finish_ordered_io().
8437 if (!inode_evicting)
8438 clear_extent_bit(tree, cur, range_end,
8440 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8441 EXTENT_DEFRAG, 1, 0, &cached_state);
8443 spin_lock_irq(&inode->ordered_tree.lock);
8444 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8445 ordered->truncated_len = min(ordered->truncated_len,
8446 cur - ordered->file_offset);
8447 spin_unlock_irq(&inode->ordered_tree.lock);
8449 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8450 cur, range_end + 1 - cur)) {
8451 btrfs_finish_ordered_io(ordered);
8453 * The ordered extent has finished, now we're again
8454 * safe to delete all extent states of the range.
8456 delete_states = true;
8459 * btrfs_finish_ordered_io() will get executed by endio
8460 * of other pages, thus we can't delete extent states
8463 delete_states = false;
8467 btrfs_put_ordered_extent(ordered);
8469 * Qgroup reserved space handler
8470 * Sector(s) here will be either:
8472 * 1) Already written to disk or bio already finished
8473 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8474 * Qgroup will be handled by its qgroup_record then.
8475 * btrfs_qgroup_free_data() call will do nothing here.
8477 * 2) Not written to disk yet
8478 * Then btrfs_qgroup_free_data() call will clear the
8479 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8480 * reserved data space.
8481 * Since the IO will never happen for this page.
8483 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8484 if (!inode_evicting) {
8485 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8486 EXTENT_DELALLOC | EXTENT_UPTODATE |
8487 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8488 delete_states, &cached_state);
8490 cur = range_end + 1;
8493 * We have iterated through all ordered extents of the page, the page
8494 * should not have Ordered (Private2) anymore, or the above iteration
8495 * did something wrong.
8497 ASSERT(!folio_test_ordered(folio));
8498 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8499 if (!inode_evicting)
8500 __btrfs_release_folio(folio, GFP_NOFS);
8501 clear_page_extent_mapped(&folio->page);
8505 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8506 * called from a page fault handler when a page is first dirtied. Hence we must
8507 * be careful to check for EOF conditions here. We set the page up correctly
8508 * for a written page which means we get ENOSPC checking when writing into
8509 * holes and correct delalloc and unwritten extent mapping on filesystems that
8510 * support these features.
8512 * We are not allowed to take the i_mutex here so we have to play games to
8513 * protect against truncate races as the page could now be beyond EOF. Because
8514 * truncate_setsize() writes the inode size before removing pages, once we have
8515 * the page lock we can determine safely if the page is beyond EOF. If it is not
8516 * beyond EOF, then the page is guaranteed safe against truncation until we
8519 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8521 struct page *page = vmf->page;
8522 struct inode *inode = file_inode(vmf->vma->vm_file);
8523 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8524 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8525 struct btrfs_ordered_extent *ordered;
8526 struct extent_state *cached_state = NULL;
8527 struct extent_changeset *data_reserved = NULL;
8528 unsigned long zero_start;
8538 reserved_space = PAGE_SIZE;
8540 sb_start_pagefault(inode->i_sb);
8541 page_start = page_offset(page);
8542 page_end = page_start + PAGE_SIZE - 1;
8546 * Reserving delalloc space after obtaining the page lock can lead to
8547 * deadlock. For example, if a dirty page is locked by this function
8548 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8549 * dirty page write out, then the btrfs_writepages() function could
8550 * end up waiting indefinitely to get a lock on the page currently
8551 * being processed by btrfs_page_mkwrite() function.
8553 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8554 page_start, reserved_space);
8556 ret2 = file_update_time(vmf->vma->vm_file);
8560 ret = vmf_error(ret2);
8566 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8568 down_read(&BTRFS_I(inode)->i_mmap_lock);
8570 size = i_size_read(inode);
8572 if ((page->mapping != inode->i_mapping) ||
8573 (page_start >= size)) {
8574 /* page got truncated out from underneath us */
8577 wait_on_page_writeback(page);
8579 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8580 ret2 = set_page_extent_mapped(page);
8582 ret = vmf_error(ret2);
8583 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8588 * we can't set the delalloc bits if there are pending ordered
8589 * extents. Drop our locks and wait for them to finish
8591 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8594 unlock_extent_cached(io_tree, page_start, page_end,
8597 up_read(&BTRFS_I(inode)->i_mmap_lock);
8598 btrfs_start_ordered_extent(ordered, 1);
8599 btrfs_put_ordered_extent(ordered);
8603 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8604 reserved_space = round_up(size - page_start,
8605 fs_info->sectorsize);
8606 if (reserved_space < PAGE_SIZE) {
8607 end = page_start + reserved_space - 1;
8608 btrfs_delalloc_release_space(BTRFS_I(inode),
8609 data_reserved, page_start,
8610 PAGE_SIZE - reserved_space, true);
8615 * page_mkwrite gets called when the page is firstly dirtied after it's
8616 * faulted in, but write(2) could also dirty a page and set delalloc
8617 * bits, thus in this case for space account reason, we still need to
8618 * clear any delalloc bits within this page range since we have to
8619 * reserve data&meta space before lock_page() (see above comments).
8621 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8622 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8623 EXTENT_DEFRAG, 0, 0, &cached_state);
8625 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8628 unlock_extent_cached(io_tree, page_start, page_end,
8630 ret = VM_FAULT_SIGBUS;
8634 /* page is wholly or partially inside EOF */
8635 if (page_start + PAGE_SIZE > size)
8636 zero_start = offset_in_page(size);
8638 zero_start = PAGE_SIZE;
8640 if (zero_start != PAGE_SIZE)
8641 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8643 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8644 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8645 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8647 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8649 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8650 up_read(&BTRFS_I(inode)->i_mmap_lock);
8652 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8653 sb_end_pagefault(inode->i_sb);
8654 extent_changeset_free(data_reserved);
8655 return VM_FAULT_LOCKED;
8659 up_read(&BTRFS_I(inode)->i_mmap_lock);
8661 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8662 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8663 reserved_space, (ret != 0));
8665 sb_end_pagefault(inode->i_sb);
8666 extent_changeset_free(data_reserved);
8670 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8672 struct btrfs_truncate_control control = {
8673 .inode = BTRFS_I(inode),
8674 .ino = btrfs_ino(BTRFS_I(inode)),
8675 .min_type = BTRFS_EXTENT_DATA_KEY,
8676 .clear_extent_range = true,
8678 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8679 struct btrfs_root *root = BTRFS_I(inode)->root;
8680 struct btrfs_block_rsv *rsv;
8682 struct btrfs_trans_handle *trans;
8683 u64 mask = fs_info->sectorsize - 1;
8684 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8686 if (!skip_writeback) {
8687 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8694 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8695 * things going on here:
8697 * 1) We need to reserve space to update our inode.
8699 * 2) We need to have something to cache all the space that is going to
8700 * be free'd up by the truncate operation, but also have some slack
8701 * space reserved in case it uses space during the truncate (thank you
8702 * very much snapshotting).
8704 * And we need these to be separate. The fact is we can use a lot of
8705 * space doing the truncate, and we have no earthly idea how much space
8706 * we will use, so we need the truncate reservation to be separate so it
8707 * doesn't end up using space reserved for updating the inode. We also
8708 * need to be able to stop the transaction and start a new one, which
8709 * means we need to be able to update the inode several times, and we
8710 * have no idea of knowing how many times that will be, so we can't just
8711 * reserve 1 item for the entirety of the operation, so that has to be
8712 * done separately as well.
8714 * So that leaves us with
8716 * 1) rsv - for the truncate reservation, which we will steal from the
8717 * transaction reservation.
8718 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8719 * updating the inode.
8721 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8724 rsv->size = min_size;
8725 rsv->failfast = true;
8728 * 1 for the truncate slack space
8729 * 1 for updating the inode.
8731 trans = btrfs_start_transaction(root, 2);
8732 if (IS_ERR(trans)) {
8733 ret = PTR_ERR(trans);
8737 /* Migrate the slack space for the truncate to our reserve */
8738 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8742 trans->block_rsv = rsv;
8745 struct extent_state *cached_state = NULL;
8746 const u64 new_size = inode->i_size;
8747 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8749 control.new_size = new_size;
8750 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8753 * We want to drop from the next block forward in case this new
8754 * size is not block aligned since we will be keeping the last
8755 * block of the extent just the way it is.
8757 btrfs_drop_extent_cache(BTRFS_I(inode),
8758 ALIGN(new_size, fs_info->sectorsize),
8761 ret = btrfs_truncate_inode_items(trans, root, &control);
8763 inode_sub_bytes(inode, control.sub_bytes);
8764 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8766 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8767 (u64)-1, &cached_state);
8769 trans->block_rsv = &fs_info->trans_block_rsv;
8770 if (ret != -ENOSPC && ret != -EAGAIN)
8773 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8777 btrfs_end_transaction(trans);
8778 btrfs_btree_balance_dirty(fs_info);
8780 trans = btrfs_start_transaction(root, 2);
8781 if (IS_ERR(trans)) {
8782 ret = PTR_ERR(trans);
8787 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8788 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8789 rsv, min_size, false);
8790 BUG_ON(ret); /* shouldn't happen */
8791 trans->block_rsv = rsv;
8795 * We can't call btrfs_truncate_block inside a trans handle as we could
8796 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8797 * know we've truncated everything except the last little bit, and can
8798 * do btrfs_truncate_block and then update the disk_i_size.
8800 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8801 btrfs_end_transaction(trans);
8802 btrfs_btree_balance_dirty(fs_info);
8804 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8807 trans = btrfs_start_transaction(root, 1);
8808 if (IS_ERR(trans)) {
8809 ret = PTR_ERR(trans);
8812 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8818 trans->block_rsv = &fs_info->trans_block_rsv;
8819 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8823 ret2 = btrfs_end_transaction(trans);
8826 btrfs_btree_balance_dirty(fs_info);
8829 btrfs_free_block_rsv(fs_info, rsv);
8831 * So if we truncate and then write and fsync we normally would just
8832 * write the extents that changed, which is a problem if we need to
8833 * first truncate that entire inode. So set this flag so we write out
8834 * all of the extents in the inode to the sync log so we're completely
8837 * If no extents were dropped or trimmed we don't need to force the next
8838 * fsync to truncate all the inode's items from the log and re-log them
8839 * all. This means the truncate operation did not change the file size,
8840 * or changed it to a smaller size but there was only an implicit hole
8841 * between the old i_size and the new i_size, and there were no prealloc
8842 * extents beyond i_size to drop.
8844 if (control.extents_found > 0)
8845 btrfs_set_inode_full_sync(BTRFS_I(inode));
8850 struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns,
8853 struct inode *inode;
8855 inode = new_inode(dir->i_sb);
8858 * Subvolumes don't inherit the sgid bit or the parent's gid if
8859 * the parent's sgid bit is set. This is probably a bug.
8861 inode_init_owner(mnt_userns, inode, NULL,
8862 S_IFDIR | (~current_umask() & S_IRWXUGO));
8863 inode->i_op = &btrfs_dir_inode_operations;
8864 inode->i_fop = &btrfs_dir_file_operations;
8869 struct inode *btrfs_alloc_inode(struct super_block *sb)
8871 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8872 struct btrfs_inode *ei;
8873 struct inode *inode;
8875 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8882 ei->last_sub_trans = 0;
8883 ei->logged_trans = 0;
8884 ei->delalloc_bytes = 0;
8885 ei->new_delalloc_bytes = 0;
8886 ei->defrag_bytes = 0;
8887 ei->disk_i_size = 0;
8891 ei->index_cnt = (u64)-1;
8893 ei->last_unlink_trans = 0;
8894 ei->last_reflink_trans = 0;
8895 ei->last_log_commit = 0;
8897 spin_lock_init(&ei->lock);
8898 ei->outstanding_extents = 0;
8899 if (sb->s_magic != BTRFS_TEST_MAGIC)
8900 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8901 BTRFS_BLOCK_RSV_DELALLOC);
8902 ei->runtime_flags = 0;
8903 ei->prop_compress = BTRFS_COMPRESS_NONE;
8904 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8906 ei->delayed_node = NULL;
8908 ei->i_otime.tv_sec = 0;
8909 ei->i_otime.tv_nsec = 0;
8911 inode = &ei->vfs_inode;
8912 extent_map_tree_init(&ei->extent_tree);
8913 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8914 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8915 IO_TREE_INODE_IO_FAILURE, inode);
8916 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8917 IO_TREE_INODE_FILE_EXTENT, inode);
8918 ei->io_tree.track_uptodate = true;
8919 ei->io_failure_tree.track_uptodate = true;
8920 atomic_set(&ei->sync_writers, 0);
8921 mutex_init(&ei->log_mutex);
8922 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8923 INIT_LIST_HEAD(&ei->delalloc_inodes);
8924 INIT_LIST_HEAD(&ei->delayed_iput);
8925 RB_CLEAR_NODE(&ei->rb_node);
8926 init_rwsem(&ei->i_mmap_lock);
8931 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8932 void btrfs_test_destroy_inode(struct inode *inode)
8934 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8935 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8939 void btrfs_free_inode(struct inode *inode)
8941 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8944 void btrfs_destroy_inode(struct inode *vfs_inode)
8946 struct btrfs_ordered_extent *ordered;
8947 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8948 struct btrfs_root *root = inode->root;
8950 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8951 WARN_ON(vfs_inode->i_data.nrpages);
8952 WARN_ON(inode->block_rsv.reserved);
8953 WARN_ON(inode->block_rsv.size);
8954 WARN_ON(inode->outstanding_extents);
8955 if (!S_ISDIR(vfs_inode->i_mode)) {
8956 WARN_ON(inode->delalloc_bytes);
8957 WARN_ON(inode->new_delalloc_bytes);
8959 WARN_ON(inode->csum_bytes);
8960 WARN_ON(inode->defrag_bytes);
8963 * This can happen where we create an inode, but somebody else also
8964 * created the same inode and we need to destroy the one we already
8971 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8975 btrfs_err(root->fs_info,
8976 "found ordered extent %llu %llu on inode cleanup",
8977 ordered->file_offset, ordered->num_bytes);
8978 btrfs_remove_ordered_extent(inode, ordered);
8979 btrfs_put_ordered_extent(ordered);
8980 btrfs_put_ordered_extent(ordered);
8983 btrfs_qgroup_check_reserved_leak(inode);
8984 inode_tree_del(inode);
8985 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8986 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8987 btrfs_put_root(inode->root);
8990 int btrfs_drop_inode(struct inode *inode)
8992 struct btrfs_root *root = BTRFS_I(inode)->root;
8997 /* the snap/subvol tree is on deleting */
8998 if (btrfs_root_refs(&root->root_item) == 0)
9001 return generic_drop_inode(inode);
9004 static void init_once(void *foo)
9006 struct btrfs_inode *ei = foo;
9008 inode_init_once(&ei->vfs_inode);
9011 void __cold btrfs_destroy_cachep(void)
9014 * Make sure all delayed rcu free inodes are flushed before we
9018 bioset_exit(&btrfs_dio_bioset);
9019 kmem_cache_destroy(btrfs_inode_cachep);
9020 kmem_cache_destroy(btrfs_trans_handle_cachep);
9021 kmem_cache_destroy(btrfs_path_cachep);
9022 kmem_cache_destroy(btrfs_free_space_cachep);
9023 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9026 int __init btrfs_init_cachep(void)
9028 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9029 sizeof(struct btrfs_inode), 0,
9030 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9032 if (!btrfs_inode_cachep)
9035 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9036 sizeof(struct btrfs_trans_handle), 0,
9037 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9038 if (!btrfs_trans_handle_cachep)
9041 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9042 sizeof(struct btrfs_path), 0,
9043 SLAB_MEM_SPREAD, NULL);
9044 if (!btrfs_path_cachep)
9047 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9048 sizeof(struct btrfs_free_space), 0,
9049 SLAB_MEM_SPREAD, NULL);
9050 if (!btrfs_free_space_cachep)
9053 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9054 PAGE_SIZE, PAGE_SIZE,
9055 SLAB_MEM_SPREAD, NULL);
9056 if (!btrfs_free_space_bitmap_cachep)
9059 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
9060 offsetof(struct btrfs_dio_private, bio),
9066 btrfs_destroy_cachep();
9070 static int btrfs_getattr(struct user_namespace *mnt_userns,
9071 const struct path *path, struct kstat *stat,
9072 u32 request_mask, unsigned int flags)
9076 struct inode *inode = d_inode(path->dentry);
9077 u32 blocksize = inode->i_sb->s_blocksize;
9078 u32 bi_flags = BTRFS_I(inode)->flags;
9079 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9081 stat->result_mask |= STATX_BTIME;
9082 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9083 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9084 if (bi_flags & BTRFS_INODE_APPEND)
9085 stat->attributes |= STATX_ATTR_APPEND;
9086 if (bi_flags & BTRFS_INODE_COMPRESS)
9087 stat->attributes |= STATX_ATTR_COMPRESSED;
9088 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9089 stat->attributes |= STATX_ATTR_IMMUTABLE;
9090 if (bi_flags & BTRFS_INODE_NODUMP)
9091 stat->attributes |= STATX_ATTR_NODUMP;
9092 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9093 stat->attributes |= STATX_ATTR_VERITY;
9095 stat->attributes_mask |= (STATX_ATTR_APPEND |
9096 STATX_ATTR_COMPRESSED |
9097 STATX_ATTR_IMMUTABLE |
9100 generic_fillattr(mnt_userns, inode, stat);
9101 stat->dev = BTRFS_I(inode)->root->anon_dev;
9103 spin_lock(&BTRFS_I(inode)->lock);
9104 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9105 inode_bytes = inode_get_bytes(inode);
9106 spin_unlock(&BTRFS_I(inode)->lock);
9107 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9108 ALIGN(delalloc_bytes, blocksize)) >> 9;
9112 static int btrfs_rename_exchange(struct inode *old_dir,
9113 struct dentry *old_dentry,
9114 struct inode *new_dir,
9115 struct dentry *new_dentry)
9117 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9118 struct btrfs_trans_handle *trans;
9119 unsigned int trans_num_items;
9120 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9121 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9122 struct inode *new_inode = new_dentry->d_inode;
9123 struct inode *old_inode = old_dentry->d_inode;
9124 struct timespec64 ctime = current_time(old_inode);
9125 struct btrfs_rename_ctx old_rename_ctx;
9126 struct btrfs_rename_ctx new_rename_ctx;
9127 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9128 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9133 bool need_abort = false;
9136 * For non-subvolumes allow exchange only within one subvolume, in the
9137 * same inode namespace. Two subvolumes (represented as directory) can
9138 * be exchanged as they're a logical link and have a fixed inode number.
9141 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9142 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9145 /* close the race window with snapshot create/destroy ioctl */
9146 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9147 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9148 down_read(&fs_info->subvol_sem);
9152 * 1 to remove old dir item
9153 * 1 to remove old dir index
9154 * 1 to add new dir item
9155 * 1 to add new dir index
9156 * 1 to update parent inode
9158 * If the parents are the same, we only need to account for one
9160 trans_num_items = (old_dir == new_dir ? 9 : 10);
9161 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9163 * 1 to remove old root ref
9164 * 1 to remove old root backref
9165 * 1 to add new root ref
9166 * 1 to add new root backref
9168 trans_num_items += 4;
9171 * 1 to update inode item
9172 * 1 to remove old inode ref
9173 * 1 to add new inode ref
9175 trans_num_items += 3;
9177 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9178 trans_num_items += 4;
9180 trans_num_items += 3;
9181 trans = btrfs_start_transaction(root, trans_num_items);
9182 if (IS_ERR(trans)) {
9183 ret = PTR_ERR(trans);
9188 ret = btrfs_record_root_in_trans(trans, dest);
9194 * We need to find a free sequence number both in the source and
9195 * in the destination directory for the exchange.
9197 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9200 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9204 BTRFS_I(old_inode)->dir_index = 0ULL;
9205 BTRFS_I(new_inode)->dir_index = 0ULL;
9207 /* Reference for the source. */
9208 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9209 /* force full log commit if subvolume involved. */
9210 btrfs_set_log_full_commit(trans);
9212 ret = btrfs_insert_inode_ref(trans, dest,
9213 new_dentry->d_name.name,
9214 new_dentry->d_name.len,
9216 btrfs_ino(BTRFS_I(new_dir)),
9223 /* And now for the dest. */
9224 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9225 /* force full log commit if subvolume involved. */
9226 btrfs_set_log_full_commit(trans);
9228 ret = btrfs_insert_inode_ref(trans, root,
9229 old_dentry->d_name.name,
9230 old_dentry->d_name.len,
9232 btrfs_ino(BTRFS_I(old_dir)),
9236 btrfs_abort_transaction(trans, ret);
9241 /* Update inode version and ctime/mtime. */
9242 inode_inc_iversion(old_dir);
9243 inode_inc_iversion(new_dir);
9244 inode_inc_iversion(old_inode);
9245 inode_inc_iversion(new_inode);
9246 old_dir->i_mtime = ctime;
9247 old_dir->i_ctime = ctime;
9248 new_dir->i_mtime = ctime;
9249 new_dir->i_ctime = ctime;
9250 old_inode->i_ctime = ctime;
9251 new_inode->i_ctime = ctime;
9253 if (old_dentry->d_parent != new_dentry->d_parent) {
9254 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9255 BTRFS_I(old_inode), 1);
9256 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9257 BTRFS_I(new_inode), 1);
9260 /* src is a subvolume */
9261 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9262 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9263 } else { /* src is an inode */
9264 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9265 BTRFS_I(old_dentry->d_inode),
9266 old_dentry->d_name.name,
9267 old_dentry->d_name.len,
9270 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9273 btrfs_abort_transaction(trans, ret);
9277 /* dest is a subvolume */
9278 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9279 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9280 } else { /* dest is an inode */
9281 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9282 BTRFS_I(new_dentry->d_inode),
9283 new_dentry->d_name.name,
9284 new_dentry->d_name.len,
9287 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9290 btrfs_abort_transaction(trans, ret);
9294 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9295 new_dentry->d_name.name,
9296 new_dentry->d_name.len, 0, old_idx);
9298 btrfs_abort_transaction(trans, ret);
9302 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9303 old_dentry->d_name.name,
9304 old_dentry->d_name.len, 0, new_idx);
9306 btrfs_abort_transaction(trans, ret);
9310 if (old_inode->i_nlink == 1)
9311 BTRFS_I(old_inode)->dir_index = old_idx;
9312 if (new_inode->i_nlink == 1)
9313 BTRFS_I(new_inode)->dir_index = new_idx;
9316 * Now pin the logs of the roots. We do it to ensure that no other task
9317 * can sync the logs while we are in progress with the rename, because
9318 * that could result in an inconsistency in case any of the inodes that
9319 * are part of this rename operation were logged before.
9321 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9322 btrfs_pin_log_trans(root);
9323 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9324 btrfs_pin_log_trans(dest);
9326 /* Do the log updates for all inodes. */
9327 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9328 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9329 old_rename_ctx.index, new_dentry->d_parent);
9330 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9331 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9332 new_rename_ctx.index, old_dentry->d_parent);
9334 /* Now unpin the logs. */
9335 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9336 btrfs_end_log_trans(root);
9337 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9338 btrfs_end_log_trans(dest);
9340 ret2 = btrfs_end_transaction(trans);
9341 ret = ret ? ret : ret2;
9343 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9344 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9345 up_read(&fs_info->subvol_sem);
9350 static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns,
9353 struct inode *inode;
9355 inode = new_inode(dir->i_sb);
9357 inode_init_owner(mnt_userns, inode, dir,
9358 S_IFCHR | WHITEOUT_MODE);
9359 inode->i_op = &btrfs_special_inode_operations;
9360 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9365 static int btrfs_rename(struct user_namespace *mnt_userns,
9366 struct inode *old_dir, struct dentry *old_dentry,
9367 struct inode *new_dir, struct dentry *new_dentry,
9370 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9371 struct btrfs_new_inode_args whiteout_args = {
9373 .dentry = old_dentry,
9375 struct btrfs_trans_handle *trans;
9376 unsigned int trans_num_items;
9377 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9378 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9379 struct inode *new_inode = d_inode(new_dentry);
9380 struct inode *old_inode = d_inode(old_dentry);
9381 struct btrfs_rename_ctx rename_ctx;
9385 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9387 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9390 /* we only allow rename subvolume link between subvolumes */
9391 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9394 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9395 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9398 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9399 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9403 /* check for collisions, even if the name isn't there */
9404 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9405 new_dentry->d_name.name,
9406 new_dentry->d_name.len);
9409 if (ret == -EEXIST) {
9411 * eexist without a new_inode */
9412 if (WARN_ON(!new_inode)) {
9416 /* maybe -EOVERFLOW */
9423 * we're using rename to replace one file with another. Start IO on it
9424 * now so we don't add too much work to the end of the transaction
9426 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9427 filemap_flush(old_inode->i_mapping);
9429 if (flags & RENAME_WHITEOUT) {
9430 whiteout_args.inode = new_whiteout_inode(mnt_userns, old_dir);
9431 if (!whiteout_args.inode)
9433 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9435 goto out_whiteout_inode;
9437 /* 1 to update the old parent inode. */
9438 trans_num_items = 1;
9441 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9442 /* Close the race window with snapshot create/destroy ioctl */
9443 down_read(&fs_info->subvol_sem);
9445 * 1 to remove old root ref
9446 * 1 to remove old root backref
9447 * 1 to add new root ref
9448 * 1 to add new root backref
9450 trans_num_items += 4;
9454 * 1 to remove old inode ref
9455 * 1 to add new inode ref
9457 trans_num_items += 3;
9460 * 1 to remove old dir item
9461 * 1 to remove old dir index
9462 * 1 to add new dir item
9463 * 1 to add new dir index
9465 trans_num_items += 4;
9466 /* 1 to update new parent inode if it's not the same as the old parent */
9467 if (new_dir != old_dir)
9472 * 1 to remove inode ref
9473 * 1 to remove dir item
9474 * 1 to remove dir index
9475 * 1 to possibly add orphan item
9477 trans_num_items += 5;
9479 trans = btrfs_start_transaction(root, trans_num_items);
9480 if (IS_ERR(trans)) {
9481 ret = PTR_ERR(trans);
9486 ret = btrfs_record_root_in_trans(trans, dest);
9491 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9495 BTRFS_I(old_inode)->dir_index = 0ULL;
9496 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9497 /* force full log commit if subvolume involved. */
9498 btrfs_set_log_full_commit(trans);
9500 ret = btrfs_insert_inode_ref(trans, dest,
9501 new_dentry->d_name.name,
9502 new_dentry->d_name.len,
9504 btrfs_ino(BTRFS_I(new_dir)), index);
9509 inode_inc_iversion(old_dir);
9510 inode_inc_iversion(new_dir);
9511 inode_inc_iversion(old_inode);
9512 old_dir->i_mtime = current_time(old_dir);
9513 old_dir->i_ctime = old_dir->i_mtime;
9514 new_dir->i_mtime = old_dir->i_mtime;
9515 new_dir->i_ctime = old_dir->i_mtime;
9516 old_inode->i_ctime = old_dir->i_mtime;
9518 if (old_dentry->d_parent != new_dentry->d_parent)
9519 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9520 BTRFS_I(old_inode), 1);
9522 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9523 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9525 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9526 BTRFS_I(d_inode(old_dentry)),
9527 old_dentry->d_name.name,
9528 old_dentry->d_name.len,
9531 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9534 btrfs_abort_transaction(trans, ret);
9539 inode_inc_iversion(new_inode);
9540 new_inode->i_ctime = current_time(new_inode);
9541 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9542 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9543 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9544 BUG_ON(new_inode->i_nlink == 0);
9546 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9547 BTRFS_I(d_inode(new_dentry)),
9548 new_dentry->d_name.name,
9549 new_dentry->d_name.len);
9551 if (!ret && new_inode->i_nlink == 0)
9552 ret = btrfs_orphan_add(trans,
9553 BTRFS_I(d_inode(new_dentry)));
9555 btrfs_abort_transaction(trans, ret);
9560 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9561 new_dentry->d_name.name,
9562 new_dentry->d_name.len, 0, index);
9564 btrfs_abort_transaction(trans, ret);
9568 if (old_inode->i_nlink == 1)
9569 BTRFS_I(old_inode)->dir_index = index;
9571 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9572 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9573 rename_ctx.index, new_dentry->d_parent);
9575 if (flags & RENAME_WHITEOUT) {
9576 ret = btrfs_create_new_inode(trans, &whiteout_args);
9578 btrfs_abort_transaction(trans, ret);
9581 unlock_new_inode(whiteout_args.inode);
9582 iput(whiteout_args.inode);
9583 whiteout_args.inode = NULL;
9587 ret2 = btrfs_end_transaction(trans);
9588 ret = ret ? ret : ret2;
9590 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9591 up_read(&fs_info->subvol_sem);
9592 if (flags & RENAME_WHITEOUT)
9593 btrfs_new_inode_args_destroy(&whiteout_args);
9595 if (flags & RENAME_WHITEOUT)
9596 iput(whiteout_args.inode);
9600 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9601 struct dentry *old_dentry, struct inode *new_dir,
9602 struct dentry *new_dentry, unsigned int flags)
9606 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9609 if (flags & RENAME_EXCHANGE)
9610 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9613 ret = btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9616 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9621 struct btrfs_delalloc_work {
9622 struct inode *inode;
9623 struct completion completion;
9624 struct list_head list;
9625 struct btrfs_work work;
9628 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9630 struct btrfs_delalloc_work *delalloc_work;
9631 struct inode *inode;
9633 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9635 inode = delalloc_work->inode;
9636 filemap_flush(inode->i_mapping);
9637 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9638 &BTRFS_I(inode)->runtime_flags))
9639 filemap_flush(inode->i_mapping);
9642 complete(&delalloc_work->completion);
9645 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9647 struct btrfs_delalloc_work *work;
9649 work = kmalloc(sizeof(*work), GFP_NOFS);
9653 init_completion(&work->completion);
9654 INIT_LIST_HEAD(&work->list);
9655 work->inode = inode;
9656 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9662 * some fairly slow code that needs optimization. This walks the list
9663 * of all the inodes with pending delalloc and forces them to disk.
9665 static int start_delalloc_inodes(struct btrfs_root *root,
9666 struct writeback_control *wbc, bool snapshot,
9667 bool in_reclaim_context)
9669 struct btrfs_inode *binode;
9670 struct inode *inode;
9671 struct btrfs_delalloc_work *work, *next;
9672 struct list_head works;
9673 struct list_head splice;
9675 bool full_flush = wbc->nr_to_write == LONG_MAX;
9677 INIT_LIST_HEAD(&works);
9678 INIT_LIST_HEAD(&splice);
9680 mutex_lock(&root->delalloc_mutex);
9681 spin_lock(&root->delalloc_lock);
9682 list_splice_init(&root->delalloc_inodes, &splice);
9683 while (!list_empty(&splice)) {
9684 binode = list_entry(splice.next, struct btrfs_inode,
9687 list_move_tail(&binode->delalloc_inodes,
9688 &root->delalloc_inodes);
9690 if (in_reclaim_context &&
9691 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9694 inode = igrab(&binode->vfs_inode);
9696 cond_resched_lock(&root->delalloc_lock);
9699 spin_unlock(&root->delalloc_lock);
9702 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9703 &binode->runtime_flags);
9705 work = btrfs_alloc_delalloc_work(inode);
9711 list_add_tail(&work->list, &works);
9712 btrfs_queue_work(root->fs_info->flush_workers,
9715 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9716 btrfs_add_delayed_iput(inode);
9717 if (ret || wbc->nr_to_write <= 0)
9721 spin_lock(&root->delalloc_lock);
9723 spin_unlock(&root->delalloc_lock);
9726 list_for_each_entry_safe(work, next, &works, list) {
9727 list_del_init(&work->list);
9728 wait_for_completion(&work->completion);
9732 if (!list_empty(&splice)) {
9733 spin_lock(&root->delalloc_lock);
9734 list_splice_tail(&splice, &root->delalloc_inodes);
9735 spin_unlock(&root->delalloc_lock);
9737 mutex_unlock(&root->delalloc_mutex);
9741 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9743 struct writeback_control wbc = {
9744 .nr_to_write = LONG_MAX,
9745 .sync_mode = WB_SYNC_NONE,
9747 .range_end = LLONG_MAX,
9749 struct btrfs_fs_info *fs_info = root->fs_info;
9751 if (BTRFS_FS_ERROR(fs_info))
9754 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9757 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9758 bool in_reclaim_context)
9760 struct writeback_control wbc = {
9762 .sync_mode = WB_SYNC_NONE,
9764 .range_end = LLONG_MAX,
9766 struct btrfs_root *root;
9767 struct list_head splice;
9770 if (BTRFS_FS_ERROR(fs_info))
9773 INIT_LIST_HEAD(&splice);
9775 mutex_lock(&fs_info->delalloc_root_mutex);
9776 spin_lock(&fs_info->delalloc_root_lock);
9777 list_splice_init(&fs_info->delalloc_roots, &splice);
9778 while (!list_empty(&splice)) {
9780 * Reset nr_to_write here so we know that we're doing a full
9784 wbc.nr_to_write = LONG_MAX;
9786 root = list_first_entry(&splice, struct btrfs_root,
9788 root = btrfs_grab_root(root);
9790 list_move_tail(&root->delalloc_root,
9791 &fs_info->delalloc_roots);
9792 spin_unlock(&fs_info->delalloc_root_lock);
9794 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9795 btrfs_put_root(root);
9796 if (ret < 0 || wbc.nr_to_write <= 0)
9798 spin_lock(&fs_info->delalloc_root_lock);
9800 spin_unlock(&fs_info->delalloc_root_lock);
9804 if (!list_empty(&splice)) {
9805 spin_lock(&fs_info->delalloc_root_lock);
9806 list_splice_tail(&splice, &fs_info->delalloc_roots);
9807 spin_unlock(&fs_info->delalloc_root_lock);
9809 mutex_unlock(&fs_info->delalloc_root_mutex);
9813 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9814 struct dentry *dentry, const char *symname)
9816 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9817 struct btrfs_trans_handle *trans;
9818 struct btrfs_root *root = BTRFS_I(dir)->root;
9819 struct btrfs_path *path;
9820 struct btrfs_key key;
9821 struct inode *inode;
9822 struct btrfs_new_inode_args new_inode_args = {
9826 unsigned int trans_num_items;
9831 struct btrfs_file_extent_item *ei;
9832 struct extent_buffer *leaf;
9834 name_len = strlen(symname);
9835 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9836 return -ENAMETOOLONG;
9838 inode = new_inode(dir->i_sb);
9841 inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO);
9842 inode->i_op = &btrfs_symlink_inode_operations;
9843 inode_nohighmem(inode);
9844 inode->i_mapping->a_ops = &btrfs_aops;
9845 btrfs_i_size_write(BTRFS_I(inode), name_len);
9846 inode_set_bytes(inode, name_len);
9848 new_inode_args.inode = inode;
9849 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9852 /* 1 additional item for the inline extent */
9855 trans = btrfs_start_transaction(root, trans_num_items);
9856 if (IS_ERR(trans)) {
9857 err = PTR_ERR(trans);
9858 goto out_new_inode_args;
9861 err = btrfs_create_new_inode(trans, &new_inode_args);
9865 path = btrfs_alloc_path();
9868 btrfs_abort_transaction(trans, err);
9869 discard_new_inode(inode);
9873 key.objectid = btrfs_ino(BTRFS_I(inode));
9875 key.type = BTRFS_EXTENT_DATA_KEY;
9876 datasize = btrfs_file_extent_calc_inline_size(name_len);
9877 err = btrfs_insert_empty_item(trans, root, path, &key,
9880 btrfs_abort_transaction(trans, err);
9881 btrfs_free_path(path);
9882 discard_new_inode(inode);
9886 leaf = path->nodes[0];
9887 ei = btrfs_item_ptr(leaf, path->slots[0],
9888 struct btrfs_file_extent_item);
9889 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9890 btrfs_set_file_extent_type(leaf, ei,
9891 BTRFS_FILE_EXTENT_INLINE);
9892 btrfs_set_file_extent_encryption(leaf, ei, 0);
9893 btrfs_set_file_extent_compression(leaf, ei, 0);
9894 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9895 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9897 ptr = btrfs_file_extent_inline_start(ei);
9898 write_extent_buffer(leaf, symname, ptr, name_len);
9899 btrfs_mark_buffer_dirty(leaf);
9900 btrfs_free_path(path);
9902 d_instantiate_new(dentry, inode);
9905 btrfs_end_transaction(trans);
9906 btrfs_btree_balance_dirty(fs_info);
9908 btrfs_new_inode_args_destroy(&new_inode_args);
9915 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9916 struct btrfs_trans_handle *trans_in,
9917 struct btrfs_inode *inode,
9918 struct btrfs_key *ins,
9921 struct btrfs_file_extent_item stack_fi;
9922 struct btrfs_replace_extent_info extent_info;
9923 struct btrfs_trans_handle *trans = trans_in;
9924 struct btrfs_path *path;
9925 u64 start = ins->objectid;
9926 u64 len = ins->offset;
9927 int qgroup_released;
9930 memset(&stack_fi, 0, sizeof(stack_fi));
9932 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9933 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9934 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9935 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9936 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9937 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9938 /* Encryption and other encoding is reserved and all 0 */
9940 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9941 if (qgroup_released < 0)
9942 return ERR_PTR(qgroup_released);
9945 ret = insert_reserved_file_extent(trans, inode,
9946 file_offset, &stack_fi,
9947 true, qgroup_released);
9953 extent_info.disk_offset = start;
9954 extent_info.disk_len = len;
9955 extent_info.data_offset = 0;
9956 extent_info.data_len = len;
9957 extent_info.file_offset = file_offset;
9958 extent_info.extent_buf = (char *)&stack_fi;
9959 extent_info.is_new_extent = true;
9960 extent_info.update_times = true;
9961 extent_info.qgroup_reserved = qgroup_released;
9962 extent_info.insertions = 0;
9964 path = btrfs_alloc_path();
9970 ret = btrfs_replace_file_extents(inode, path, file_offset,
9971 file_offset + len - 1, &extent_info,
9973 btrfs_free_path(path);
9980 * We have released qgroup data range at the beginning of the function,
9981 * and normally qgroup_released bytes will be freed when committing
9983 * But if we error out early, we have to free what we have released
9984 * or we leak qgroup data reservation.
9986 btrfs_qgroup_free_refroot(inode->root->fs_info,
9987 inode->root->root_key.objectid, qgroup_released,
9988 BTRFS_QGROUP_RSV_DATA);
9989 return ERR_PTR(ret);
9992 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9993 u64 start, u64 num_bytes, u64 min_size,
9994 loff_t actual_len, u64 *alloc_hint,
9995 struct btrfs_trans_handle *trans)
9997 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9998 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9999 struct extent_map *em;
10000 struct btrfs_root *root = BTRFS_I(inode)->root;
10001 struct btrfs_key ins;
10002 u64 cur_offset = start;
10003 u64 clear_offset = start;
10006 u64 last_alloc = (u64)-1;
10008 bool own_trans = true;
10009 u64 end = start + num_bytes - 1;
10013 while (num_bytes > 0) {
10014 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10015 cur_bytes = max(cur_bytes, min_size);
10017 * If we are severely fragmented we could end up with really
10018 * small allocations, so if the allocator is returning small
10019 * chunks lets make its job easier by only searching for those
10022 cur_bytes = min(cur_bytes, last_alloc);
10023 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10024 min_size, 0, *alloc_hint, &ins, 1, 0);
10029 * We've reserved this space, and thus converted it from
10030 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10031 * from here on out we will only need to clear our reservation
10032 * for the remaining unreserved area, so advance our
10033 * clear_offset by our extent size.
10035 clear_offset += ins.offset;
10037 last_alloc = ins.offset;
10038 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10041 * Now that we inserted the prealloc extent we can finally
10042 * decrement the number of reservations in the block group.
10043 * If we did it before, we could race with relocation and have
10044 * relocation miss the reserved extent, making it fail later.
10046 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10047 if (IS_ERR(trans)) {
10048 ret = PTR_ERR(trans);
10049 btrfs_free_reserved_extent(fs_info, ins.objectid,
10054 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10055 cur_offset + ins.offset -1, 0);
10057 em = alloc_extent_map();
10059 btrfs_set_inode_full_sync(BTRFS_I(inode));
10063 em->start = cur_offset;
10064 em->orig_start = cur_offset;
10065 em->len = ins.offset;
10066 em->block_start = ins.objectid;
10067 em->block_len = ins.offset;
10068 em->orig_block_len = ins.offset;
10069 em->ram_bytes = ins.offset;
10070 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10071 em->generation = trans->transid;
10074 write_lock(&em_tree->lock);
10075 ret = add_extent_mapping(em_tree, em, 1);
10076 write_unlock(&em_tree->lock);
10077 if (ret != -EEXIST)
10079 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10080 cur_offset + ins.offset - 1,
10083 free_extent_map(em);
10085 num_bytes -= ins.offset;
10086 cur_offset += ins.offset;
10087 *alloc_hint = ins.objectid + ins.offset;
10089 inode_inc_iversion(inode);
10090 inode->i_ctime = current_time(inode);
10091 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10092 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10093 (actual_len > inode->i_size) &&
10094 (cur_offset > inode->i_size)) {
10095 if (cur_offset > actual_len)
10096 i_size = actual_len;
10098 i_size = cur_offset;
10099 i_size_write(inode, i_size);
10100 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10103 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10106 btrfs_abort_transaction(trans, ret);
10108 btrfs_end_transaction(trans);
10113 btrfs_end_transaction(trans);
10117 if (clear_offset < end)
10118 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10119 end - clear_offset + 1);
10123 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10124 u64 start, u64 num_bytes, u64 min_size,
10125 loff_t actual_len, u64 *alloc_hint)
10127 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10128 min_size, actual_len, alloc_hint,
10132 int btrfs_prealloc_file_range_trans(struct inode *inode,
10133 struct btrfs_trans_handle *trans, int mode,
10134 u64 start, u64 num_bytes, u64 min_size,
10135 loff_t actual_len, u64 *alloc_hint)
10137 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10138 min_size, actual_len, alloc_hint, trans);
10141 static int btrfs_permission(struct user_namespace *mnt_userns,
10142 struct inode *inode, int mask)
10144 struct btrfs_root *root = BTRFS_I(inode)->root;
10145 umode_t mode = inode->i_mode;
10147 if (mask & MAY_WRITE &&
10148 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10149 if (btrfs_root_readonly(root))
10151 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10154 return generic_permission(mnt_userns, inode, mask);
10157 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10158 struct dentry *dentry, umode_t mode)
10160 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10161 struct btrfs_trans_handle *trans;
10162 struct btrfs_root *root = BTRFS_I(dir)->root;
10163 struct inode *inode;
10164 struct btrfs_new_inode_args new_inode_args = {
10169 unsigned int trans_num_items;
10172 inode = new_inode(dir->i_sb);
10175 inode_init_owner(mnt_userns, inode, dir, mode);
10176 inode->i_fop = &btrfs_file_operations;
10177 inode->i_op = &btrfs_file_inode_operations;
10178 inode->i_mapping->a_ops = &btrfs_aops;
10180 new_inode_args.inode = inode;
10181 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
10185 trans = btrfs_start_transaction(root, trans_num_items);
10186 if (IS_ERR(trans)) {
10187 ret = PTR_ERR(trans);
10188 goto out_new_inode_args;
10191 ret = btrfs_create_new_inode(trans, &new_inode_args);
10194 * We set number of links to 0 in btrfs_create_new_inode(), and here we
10195 * set it to 1 because d_tmpfile() will issue a warning if the count is
10198 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10200 set_nlink(inode, 1);
10203 d_tmpfile(dentry, inode);
10204 unlock_new_inode(inode);
10205 mark_inode_dirty(inode);
10208 btrfs_end_transaction(trans);
10209 btrfs_btree_balance_dirty(fs_info);
10210 out_new_inode_args:
10211 btrfs_new_inode_args_destroy(&new_inode_args);
10218 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10220 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10221 unsigned long index = start >> PAGE_SHIFT;
10222 unsigned long end_index = end >> PAGE_SHIFT;
10226 ASSERT(end + 1 - start <= U32_MAX);
10227 len = end + 1 - start;
10228 while (index <= end_index) {
10229 page = find_get_page(inode->vfs_inode.i_mapping, index);
10230 ASSERT(page); /* Pages should be in the extent_io_tree */
10232 btrfs_page_set_writeback(fs_info, page, start, len);
10238 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
10241 switch (compress_type) {
10242 case BTRFS_COMPRESS_NONE:
10243 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10244 case BTRFS_COMPRESS_ZLIB:
10245 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10246 case BTRFS_COMPRESS_LZO:
10248 * The LZO format depends on the sector size. 64K is the maximum
10249 * sector size that we support.
10251 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10253 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10254 (fs_info->sectorsize_bits - 12);
10255 case BTRFS_COMPRESS_ZSTD:
10256 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10262 static ssize_t btrfs_encoded_read_inline(
10263 struct kiocb *iocb,
10264 struct iov_iter *iter, u64 start,
10266 struct extent_state **cached_state,
10267 u64 extent_start, size_t count,
10268 struct btrfs_ioctl_encoded_io_args *encoded,
10271 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10272 struct btrfs_root *root = inode->root;
10273 struct btrfs_fs_info *fs_info = root->fs_info;
10274 struct extent_io_tree *io_tree = &inode->io_tree;
10275 struct btrfs_path *path;
10276 struct extent_buffer *leaf;
10277 struct btrfs_file_extent_item *item;
10283 path = btrfs_alloc_path();
10288 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10292 /* The extent item disappeared? */
10297 leaf = path->nodes[0];
10298 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10300 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10301 ptr = btrfs_file_extent_inline_start(item);
10303 encoded->len = min_t(u64, extent_start + ram_bytes,
10304 inode->vfs_inode.i_size) - iocb->ki_pos;
10305 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10306 btrfs_file_extent_compression(leaf, item));
10309 encoded->compression = ret;
10310 if (encoded->compression) {
10311 size_t inline_size;
10313 inline_size = btrfs_file_extent_inline_item_len(leaf,
10315 if (inline_size > count) {
10319 count = inline_size;
10320 encoded->unencoded_len = ram_bytes;
10321 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10323 count = min_t(u64, count, encoded->len);
10324 encoded->len = count;
10325 encoded->unencoded_len = count;
10326 ptr += iocb->ki_pos - extent_start;
10329 tmp = kmalloc(count, GFP_NOFS);
10334 read_extent_buffer(leaf, tmp, ptr, count);
10335 btrfs_release_path(path);
10336 unlock_extent_cached(io_tree, start, lockend, cached_state);
10337 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10340 ret = copy_to_iter(tmp, count, iter);
10345 btrfs_free_path(path);
10349 struct btrfs_encoded_read_private {
10350 struct btrfs_inode *inode;
10352 wait_queue_head_t wait;
10354 blk_status_t status;
10358 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10359 struct bio *bio, int mirror_num)
10361 struct btrfs_encoded_read_private *priv = bio->bi_private;
10362 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10365 if (!priv->skip_csum) {
10366 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10371 atomic_inc(&priv->pending);
10372 btrfs_submit_bio(fs_info, bio, mirror_num);
10376 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10378 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10379 struct btrfs_encoded_read_private *priv = bbio->bio.bi_private;
10380 struct btrfs_inode *inode = priv->inode;
10381 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10382 u32 sectorsize = fs_info->sectorsize;
10383 struct bio_vec *bvec;
10384 struct bvec_iter_all iter_all;
10385 u32 bio_offset = 0;
10387 if (priv->skip_csum || !uptodate)
10388 return bbio->bio.bi_status;
10390 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10391 unsigned int i, nr_sectors, pgoff;
10393 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10394 pgoff = bvec->bv_offset;
10395 for (i = 0; i < nr_sectors; i++) {
10396 ASSERT(pgoff < PAGE_SIZE);
10397 if (btrfs_check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10398 bvec->bv_page, pgoff))
10399 return BLK_STS_IOERR;
10400 bio_offset += sectorsize;
10401 pgoff += sectorsize;
10407 static void btrfs_encoded_read_endio(struct bio *bio)
10409 struct btrfs_encoded_read_private *priv = bio->bi_private;
10410 struct btrfs_bio *bbio = btrfs_bio(bio);
10411 blk_status_t status;
10413 status = btrfs_encoded_read_verify_csum(bbio);
10416 * The memory barrier implied by the atomic_dec_return() here
10417 * pairs with the memory barrier implied by the
10418 * atomic_dec_return() or io_wait_event() in
10419 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10420 * write is observed before the load of status in
10421 * btrfs_encoded_read_regular_fill_pages().
10423 WRITE_ONCE(priv->status, status);
10425 if (!atomic_dec_return(&priv->pending))
10426 wake_up(&priv->wait);
10427 btrfs_bio_free_csum(bbio);
10431 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10432 u64 file_offset, u64 disk_bytenr,
10433 u64 disk_io_size, struct page **pages)
10435 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10436 struct btrfs_encoded_read_private priv = {
10438 .file_offset = file_offset,
10439 .pending = ATOMIC_INIT(1),
10440 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10442 unsigned long i = 0;
10446 init_waitqueue_head(&priv.wait);
10448 * Submit bios for the extent, splitting due to bio or stripe limits as
10451 while (cur < disk_io_size) {
10452 struct extent_map *em;
10453 struct btrfs_io_geometry geom;
10454 struct bio *bio = NULL;
10457 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10458 disk_io_size - cur);
10462 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10463 disk_bytenr + cur, &geom);
10464 free_extent_map(em);
10467 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10470 remaining = min(geom.len, disk_io_size - cur);
10471 while (bio || remaining) {
10472 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10475 bio = btrfs_bio_alloc(BIO_MAX_VECS);
10476 bio->bi_iter.bi_sector =
10477 (disk_bytenr + cur) >> SECTOR_SHIFT;
10478 bio->bi_end_io = btrfs_encoded_read_endio;
10479 bio->bi_private = &priv;
10480 bio->bi_opf = REQ_OP_READ;
10484 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10485 blk_status_t status;
10487 status = submit_encoded_read_bio(inode, bio, 0);
10489 WRITE_ONCE(priv.status, status);
10499 remaining -= bytes;
10504 if (atomic_dec_return(&priv.pending))
10505 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10506 /* See btrfs_encoded_read_endio() for ordering. */
10507 return blk_status_to_errno(READ_ONCE(priv.status));
10510 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10511 struct iov_iter *iter,
10512 u64 start, u64 lockend,
10513 struct extent_state **cached_state,
10514 u64 disk_bytenr, u64 disk_io_size,
10515 size_t count, bool compressed,
10518 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10519 struct extent_io_tree *io_tree = &inode->io_tree;
10520 struct page **pages;
10521 unsigned long nr_pages, i;
10523 size_t page_offset;
10526 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10527 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10530 ret = btrfs_alloc_page_array(nr_pages, pages);
10536 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10537 disk_io_size, pages);
10541 unlock_extent_cached(io_tree, start, lockend, cached_state);
10542 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10549 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10550 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10553 while (cur < count) {
10554 size_t bytes = min_t(size_t, count - cur,
10555 PAGE_SIZE - page_offset);
10557 if (copy_page_to_iter(pages[i], page_offset, bytes,
10568 for (i = 0; i < nr_pages; i++) {
10570 __free_page(pages[i]);
10576 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10577 struct btrfs_ioctl_encoded_io_args *encoded)
10579 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10580 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10581 struct extent_io_tree *io_tree = &inode->io_tree;
10583 size_t count = iov_iter_count(iter);
10584 u64 start, lockend, disk_bytenr, disk_io_size;
10585 struct extent_state *cached_state = NULL;
10586 struct extent_map *em;
10587 bool unlocked = false;
10589 file_accessed(iocb->ki_filp);
10591 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10593 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10594 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10597 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10599 * We don't know how long the extent containing iocb->ki_pos is, but if
10600 * it's compressed we know that it won't be longer than this.
10602 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10605 struct btrfs_ordered_extent *ordered;
10607 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10608 lockend - start + 1);
10610 goto out_unlock_inode;
10611 lock_extent_bits(io_tree, start, lockend, &cached_state);
10612 ordered = btrfs_lookup_ordered_range(inode, start,
10613 lockend - start + 1);
10616 btrfs_put_ordered_extent(ordered);
10617 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10621 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10624 goto out_unlock_extent;
10627 if (em->block_start == EXTENT_MAP_INLINE) {
10628 u64 extent_start = em->start;
10631 * For inline extents we get everything we need out of the
10634 free_extent_map(em);
10636 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10637 &cached_state, extent_start,
10638 count, encoded, &unlocked);
10643 * We only want to return up to EOF even if the extent extends beyond
10646 encoded->len = min_t(u64, extent_map_end(em),
10647 inode->vfs_inode.i_size) - iocb->ki_pos;
10648 if (em->block_start == EXTENT_MAP_HOLE ||
10649 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10650 disk_bytenr = EXTENT_MAP_HOLE;
10651 count = min_t(u64, count, encoded->len);
10652 encoded->len = count;
10653 encoded->unencoded_len = count;
10654 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10655 disk_bytenr = em->block_start;
10657 * Bail if the buffer isn't large enough to return the whole
10658 * compressed extent.
10660 if (em->block_len > count) {
10664 disk_io_size = em->block_len;
10665 count = em->block_len;
10666 encoded->unencoded_len = em->ram_bytes;
10667 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10668 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10669 em->compress_type);
10672 encoded->compression = ret;
10674 disk_bytenr = em->block_start + (start - em->start);
10675 if (encoded->len > count)
10676 encoded->len = count;
10678 * Don't read beyond what we locked. This also limits the page
10679 * allocations that we'll do.
10681 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10682 count = start + disk_io_size - iocb->ki_pos;
10683 encoded->len = count;
10684 encoded->unencoded_len = count;
10685 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10687 free_extent_map(em);
10690 if (disk_bytenr == EXTENT_MAP_HOLE) {
10691 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10692 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10694 ret = iov_iter_zero(count, iter);
10698 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10699 &cached_state, disk_bytenr,
10700 disk_io_size, count,
10701 encoded->compression,
10707 iocb->ki_pos += encoded->len;
10709 free_extent_map(em);
10712 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10715 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10719 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10720 const struct btrfs_ioctl_encoded_io_args *encoded)
10722 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10723 struct btrfs_root *root = inode->root;
10724 struct btrfs_fs_info *fs_info = root->fs_info;
10725 struct extent_io_tree *io_tree = &inode->io_tree;
10726 struct extent_changeset *data_reserved = NULL;
10727 struct extent_state *cached_state = NULL;
10731 u64 num_bytes, ram_bytes, disk_num_bytes;
10732 unsigned long nr_pages, i;
10733 struct page **pages;
10734 struct btrfs_key ins;
10735 bool extent_reserved = false;
10736 struct extent_map *em;
10739 switch (encoded->compression) {
10740 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10741 compression = BTRFS_COMPRESS_ZLIB;
10743 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10744 compression = BTRFS_COMPRESS_ZSTD;
10746 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10747 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10748 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10749 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10750 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10751 /* The sector size must match for LZO. */
10752 if (encoded->compression -
10753 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10754 fs_info->sectorsize_bits)
10756 compression = BTRFS_COMPRESS_LZO;
10761 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10764 orig_count = iov_iter_count(from);
10766 /* The extent size must be sane. */
10767 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10768 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10772 * The compressed data must be smaller than the decompressed data.
10774 * It's of course possible for data to compress to larger or the same
10775 * size, but the buffered I/O path falls back to no compression for such
10776 * data, and we don't want to break any assumptions by creating these
10779 * Note that this is less strict than the current check we have that the
10780 * compressed data must be at least one sector smaller than the
10781 * decompressed data. We only want to enforce the weaker requirement
10782 * from old kernels that it is at least one byte smaller.
10784 if (orig_count >= encoded->unencoded_len)
10787 /* The extent must start on a sector boundary. */
10788 start = iocb->ki_pos;
10789 if (!IS_ALIGNED(start, fs_info->sectorsize))
10793 * The extent must end on a sector boundary. However, we allow a write
10794 * which ends at or extends i_size to have an unaligned length; we round
10795 * up the extent size and set i_size to the unaligned end.
10797 if (start + encoded->len < inode->vfs_inode.i_size &&
10798 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10801 /* Finally, the offset in the unencoded data must be sector-aligned. */
10802 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10805 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10806 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10807 end = start + num_bytes - 1;
10810 * If the extent cannot be inline, the compressed data on disk must be
10811 * sector-aligned. For convenience, we extend it with zeroes if it
10814 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10815 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10816 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10819 for (i = 0; i < nr_pages; i++) {
10820 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10823 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10828 kaddr = kmap_local_page(pages[i]);
10829 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10830 kunmap_local(kaddr);
10834 if (bytes < PAGE_SIZE)
10835 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10836 kunmap_local(kaddr);
10840 struct btrfs_ordered_extent *ordered;
10842 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10845 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10846 start >> PAGE_SHIFT,
10847 end >> PAGE_SHIFT);
10850 lock_extent_bits(io_tree, start, end, &cached_state);
10851 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10853 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10856 btrfs_put_ordered_extent(ordered);
10857 unlock_extent_cached(io_tree, start, end, &cached_state);
10862 * We don't use the higher-level delalloc space functions because our
10863 * num_bytes and disk_num_bytes are different.
10865 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10868 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10870 goto out_free_data_space;
10871 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10874 goto out_qgroup_free_data;
10876 /* Try an inline extent first. */
10877 if (start == 0 && encoded->unencoded_len == encoded->len &&
10878 encoded->unencoded_offset == 0) {
10879 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10880 compression, pages, true);
10884 goto out_delalloc_release;
10888 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10889 disk_num_bytes, 0, 0, &ins, 1, 1);
10891 goto out_delalloc_release;
10892 extent_reserved = true;
10894 em = create_io_em(inode, start, num_bytes,
10895 start - encoded->unencoded_offset, ins.objectid,
10896 ins.offset, ins.offset, ram_bytes, compression,
10897 BTRFS_ORDERED_COMPRESSED);
10900 goto out_free_reserved;
10902 free_extent_map(em);
10904 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10905 ins.objectid, ins.offset,
10906 encoded->unencoded_offset,
10907 (1 << BTRFS_ORDERED_ENCODED) |
10908 (1 << BTRFS_ORDERED_COMPRESSED),
10911 btrfs_drop_extent_cache(inode, start, end, 0);
10912 goto out_free_reserved;
10914 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10916 if (start + encoded->len > inode->vfs_inode.i_size)
10917 i_size_write(&inode->vfs_inode, start + encoded->len);
10919 unlock_extent_cached(io_tree, start, end, &cached_state);
10921 btrfs_delalloc_release_extents(inode, num_bytes);
10923 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10924 ins.offset, pages, nr_pages, 0, NULL,
10926 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10934 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10935 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10936 out_delalloc_release:
10937 btrfs_delalloc_release_extents(inode, num_bytes);
10938 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10939 out_qgroup_free_data:
10941 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10942 out_free_data_space:
10944 * If btrfs_reserve_extent() succeeded, then we already decremented
10947 if (!extent_reserved)
10948 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10950 unlock_extent_cached(io_tree, start, end, &cached_state);
10952 for (i = 0; i < nr_pages; i++) {
10954 __free_page(pages[i]);
10959 iocb->ki_pos += encoded->len;
10965 * Add an entry indicating a block group or device which is pinned by a
10966 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10967 * negative errno on failure.
10969 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10970 bool is_block_group)
10972 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10973 struct btrfs_swapfile_pin *sp, *entry;
10974 struct rb_node **p;
10975 struct rb_node *parent = NULL;
10977 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10982 sp->is_block_group = is_block_group;
10983 sp->bg_extent_count = 1;
10985 spin_lock(&fs_info->swapfile_pins_lock);
10986 p = &fs_info->swapfile_pins.rb_node;
10989 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10990 if (sp->ptr < entry->ptr ||
10991 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10992 p = &(*p)->rb_left;
10993 } else if (sp->ptr > entry->ptr ||
10994 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10995 p = &(*p)->rb_right;
10997 if (is_block_group)
10998 entry->bg_extent_count++;
10999 spin_unlock(&fs_info->swapfile_pins_lock);
11004 rb_link_node(&sp->node, parent, p);
11005 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
11006 spin_unlock(&fs_info->swapfile_pins_lock);
11010 /* Free all of the entries pinned by this swapfile. */
11011 static void btrfs_free_swapfile_pins(struct inode *inode)
11013 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
11014 struct btrfs_swapfile_pin *sp;
11015 struct rb_node *node, *next;
11017 spin_lock(&fs_info->swapfile_pins_lock);
11018 node = rb_first(&fs_info->swapfile_pins);
11020 next = rb_next(node);
11021 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
11022 if (sp->inode == inode) {
11023 rb_erase(&sp->node, &fs_info->swapfile_pins);
11024 if (sp->is_block_group) {
11025 btrfs_dec_block_group_swap_extents(sp->ptr,
11026 sp->bg_extent_count);
11027 btrfs_put_block_group(sp->ptr);
11033 spin_unlock(&fs_info->swapfile_pins_lock);
11036 struct btrfs_swap_info {
11042 unsigned long nr_pages;
11046 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
11047 struct btrfs_swap_info *bsi)
11049 unsigned long nr_pages;
11050 unsigned long max_pages;
11051 u64 first_ppage, first_ppage_reported, next_ppage;
11055 * Our swapfile may have had its size extended after the swap header was
11056 * written. In that case activating the swapfile should not go beyond
11057 * the max size set in the swap header.
11059 if (bsi->nr_pages >= sis->max)
11062 max_pages = sis->max - bsi->nr_pages;
11063 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
11064 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
11065 PAGE_SIZE) >> PAGE_SHIFT;
11067 if (first_ppage >= next_ppage)
11069 nr_pages = next_ppage - first_ppage;
11070 nr_pages = min(nr_pages, max_pages);
11072 first_ppage_reported = first_ppage;
11073 if (bsi->start == 0)
11074 first_ppage_reported++;
11075 if (bsi->lowest_ppage > first_ppage_reported)
11076 bsi->lowest_ppage = first_ppage_reported;
11077 if (bsi->highest_ppage < (next_ppage - 1))
11078 bsi->highest_ppage = next_ppage - 1;
11080 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
11083 bsi->nr_extents += ret;
11084 bsi->nr_pages += nr_pages;
11088 static void btrfs_swap_deactivate(struct file *file)
11090 struct inode *inode = file_inode(file);
11092 btrfs_free_swapfile_pins(inode);
11093 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
11096 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11099 struct inode *inode = file_inode(file);
11100 struct btrfs_root *root = BTRFS_I(inode)->root;
11101 struct btrfs_fs_info *fs_info = root->fs_info;
11102 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
11103 struct extent_state *cached_state = NULL;
11104 struct extent_map *em = NULL;
11105 struct btrfs_device *device = NULL;
11106 struct btrfs_swap_info bsi = {
11107 .lowest_ppage = (sector_t)-1ULL,
11114 * If the swap file was just created, make sure delalloc is done. If the
11115 * file changes again after this, the user is doing something stupid and
11116 * we don't really care.
11118 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
11123 * The inode is locked, so these flags won't change after we check them.
11125 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
11126 btrfs_warn(fs_info, "swapfile must not be compressed");
11129 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
11130 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
11133 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
11134 btrfs_warn(fs_info, "swapfile must not be checksummed");
11139 * Balance or device remove/replace/resize can move stuff around from
11140 * under us. The exclop protection makes sure they aren't running/won't
11141 * run concurrently while we are mapping the swap extents, and
11142 * fs_info->swapfile_pins prevents them from running while the swap
11143 * file is active and moving the extents. Note that this also prevents
11144 * a concurrent device add which isn't actually necessary, but it's not
11145 * really worth the trouble to allow it.
11147 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
11148 btrfs_warn(fs_info,
11149 "cannot activate swapfile while exclusive operation is running");
11154 * Prevent snapshot creation while we are activating the swap file.
11155 * We do not want to race with snapshot creation. If snapshot creation
11156 * already started before we bumped nr_swapfiles from 0 to 1 and
11157 * completes before the first write into the swap file after it is
11158 * activated, than that write would fallback to COW.
11160 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11161 btrfs_exclop_finish(fs_info);
11162 btrfs_warn(fs_info,
11163 "cannot activate swapfile because snapshot creation is in progress");
11167 * Snapshots can create extents which require COW even if NODATACOW is
11168 * set. We use this counter to prevent snapshots. We must increment it
11169 * before walking the extents because we don't want a concurrent
11170 * snapshot to run after we've already checked the extents.
11172 * It is possible that subvolume is marked for deletion but still not
11173 * removed yet. To prevent this race, we check the root status before
11174 * activating the swapfile.
11176 spin_lock(&root->root_item_lock);
11177 if (btrfs_root_dead(root)) {
11178 spin_unlock(&root->root_item_lock);
11180 btrfs_exclop_finish(fs_info);
11181 btrfs_warn(fs_info,
11182 "cannot activate swapfile because subvolume %llu is being deleted",
11183 root->root_key.objectid);
11186 atomic_inc(&root->nr_swapfiles);
11187 spin_unlock(&root->root_item_lock);
11189 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11191 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
11193 while (start < isize) {
11194 u64 logical_block_start, physical_block_start;
11195 struct btrfs_block_group *bg;
11196 u64 len = isize - start;
11198 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11204 if (em->block_start == EXTENT_MAP_HOLE) {
11205 btrfs_warn(fs_info, "swapfile must not have holes");
11209 if (em->block_start == EXTENT_MAP_INLINE) {
11211 * It's unlikely we'll ever actually find ourselves
11212 * here, as a file small enough to fit inline won't be
11213 * big enough to store more than the swap header, but in
11214 * case something changes in the future, let's catch it
11215 * here rather than later.
11217 btrfs_warn(fs_info, "swapfile must not be inline");
11221 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11222 btrfs_warn(fs_info, "swapfile must not be compressed");
11227 logical_block_start = em->block_start + (start - em->start);
11228 len = min(len, em->len - (start - em->start));
11229 free_extent_map(em);
11232 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
11238 btrfs_warn(fs_info,
11239 "swapfile must not be copy-on-write");
11244 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11250 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11251 btrfs_warn(fs_info,
11252 "swapfile must have single data profile");
11257 if (device == NULL) {
11258 device = em->map_lookup->stripes[0].dev;
11259 ret = btrfs_add_swapfile_pin(inode, device, false);
11264 } else if (device != em->map_lookup->stripes[0].dev) {
11265 btrfs_warn(fs_info, "swapfile must be on one device");
11270 physical_block_start = (em->map_lookup->stripes[0].physical +
11271 (logical_block_start - em->start));
11272 len = min(len, em->len - (logical_block_start - em->start));
11273 free_extent_map(em);
11276 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11278 btrfs_warn(fs_info,
11279 "could not find block group containing swapfile");
11284 if (!btrfs_inc_block_group_swap_extents(bg)) {
11285 btrfs_warn(fs_info,
11286 "block group for swapfile at %llu is read-only%s",
11288 atomic_read(&fs_info->scrubs_running) ?
11289 " (scrub running)" : "");
11290 btrfs_put_block_group(bg);
11295 ret = btrfs_add_swapfile_pin(inode, bg, true);
11297 btrfs_put_block_group(bg);
11304 if (bsi.block_len &&
11305 bsi.block_start + bsi.block_len == physical_block_start) {
11306 bsi.block_len += len;
11308 if (bsi.block_len) {
11309 ret = btrfs_add_swap_extent(sis, &bsi);
11314 bsi.block_start = physical_block_start;
11315 bsi.block_len = len;
11322 ret = btrfs_add_swap_extent(sis, &bsi);
11325 if (!IS_ERR_OR_NULL(em))
11326 free_extent_map(em);
11328 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
11331 btrfs_swap_deactivate(file);
11333 btrfs_drew_write_unlock(&root->snapshot_lock);
11335 btrfs_exclop_finish(fs_info);
11341 sis->bdev = device->bdev;
11342 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11343 sis->max = bsi.nr_pages;
11344 sis->pages = bsi.nr_pages - 1;
11345 sis->highest_bit = bsi.nr_pages - 1;
11346 return bsi.nr_extents;
11349 static void btrfs_swap_deactivate(struct file *file)
11353 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11356 return -EOPNOTSUPP;
11361 * Update the number of bytes used in the VFS' inode. When we replace extents in
11362 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11363 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11364 * always get a correct value.
11366 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11367 const u64 add_bytes,
11368 const u64 del_bytes)
11370 if (add_bytes == del_bytes)
11373 spin_lock(&inode->lock);
11375 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11377 inode_add_bytes(&inode->vfs_inode, add_bytes);
11378 spin_unlock(&inode->lock);
11382 * Verify that there are no ordered extents for a given file range.
11384 * @inode: The target inode.
11385 * @start: Start offset of the file range, should be sector size aligned.
11386 * @end: End offset (inclusive) of the file range, its value +1 should be
11387 * sector size aligned.
11389 * This should typically be used for cases where we locked an inode's VFS lock in
11390 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11391 * we have flushed all delalloc in the range, we have waited for all ordered
11392 * extents in the range to complete and finally we have locked the file range in
11393 * the inode's io_tree.
11395 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11397 struct btrfs_root *root = inode->root;
11398 struct btrfs_ordered_extent *ordered;
11400 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11403 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11405 btrfs_err(root->fs_info,
11406 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11407 start, end, btrfs_ino(inode), root->root_key.objectid,
11408 ordered->file_offset,
11409 ordered->file_offset + ordered->num_bytes - 1);
11410 btrfs_put_ordered_extent(ordered);
11413 ASSERT(ordered == NULL);
11416 static const struct inode_operations btrfs_dir_inode_operations = {
11417 .getattr = btrfs_getattr,
11418 .lookup = btrfs_lookup,
11419 .create = btrfs_create,
11420 .unlink = btrfs_unlink,
11421 .link = btrfs_link,
11422 .mkdir = btrfs_mkdir,
11423 .rmdir = btrfs_rmdir,
11424 .rename = btrfs_rename2,
11425 .symlink = btrfs_symlink,
11426 .setattr = btrfs_setattr,
11427 .mknod = btrfs_mknod,
11428 .listxattr = btrfs_listxattr,
11429 .permission = btrfs_permission,
11430 .get_acl = btrfs_get_acl,
11431 .set_acl = btrfs_set_acl,
11432 .update_time = btrfs_update_time,
11433 .tmpfile = btrfs_tmpfile,
11434 .fileattr_get = btrfs_fileattr_get,
11435 .fileattr_set = btrfs_fileattr_set,
11438 static const struct file_operations btrfs_dir_file_operations = {
11439 .llseek = generic_file_llseek,
11440 .read = generic_read_dir,
11441 .iterate_shared = btrfs_real_readdir,
11442 .open = btrfs_opendir,
11443 .unlocked_ioctl = btrfs_ioctl,
11444 #ifdef CONFIG_COMPAT
11445 .compat_ioctl = btrfs_compat_ioctl,
11447 .release = btrfs_release_file,
11448 .fsync = btrfs_sync_file,
11452 * btrfs doesn't support the bmap operation because swapfiles
11453 * use bmap to make a mapping of extents in the file. They assume
11454 * these extents won't change over the life of the file and they
11455 * use the bmap result to do IO directly to the drive.
11457 * the btrfs bmap call would return logical addresses that aren't
11458 * suitable for IO and they also will change frequently as COW
11459 * operations happen. So, swapfile + btrfs == corruption.
11461 * For now we're avoiding this by dropping bmap.
11463 static const struct address_space_operations btrfs_aops = {
11464 .read_folio = btrfs_read_folio,
11465 .writepages = btrfs_writepages,
11466 .readahead = btrfs_readahead,
11467 .direct_IO = noop_direct_IO,
11468 .invalidate_folio = btrfs_invalidate_folio,
11469 .release_folio = btrfs_release_folio,
11470 .migrate_folio = btrfs_migrate_folio,
11471 .dirty_folio = filemap_dirty_folio,
11472 .error_remove_page = generic_error_remove_page,
11473 .swap_activate = btrfs_swap_activate,
11474 .swap_deactivate = btrfs_swap_deactivate,
11477 static const struct inode_operations btrfs_file_inode_operations = {
11478 .getattr = btrfs_getattr,
11479 .setattr = btrfs_setattr,
11480 .listxattr = btrfs_listxattr,
11481 .permission = btrfs_permission,
11482 .fiemap = btrfs_fiemap,
11483 .get_acl = btrfs_get_acl,
11484 .set_acl = btrfs_set_acl,
11485 .update_time = btrfs_update_time,
11486 .fileattr_get = btrfs_fileattr_get,
11487 .fileattr_set = btrfs_fileattr_set,
11489 static const struct inode_operations btrfs_special_inode_operations = {
11490 .getattr = btrfs_getattr,
11491 .setattr = btrfs_setattr,
11492 .permission = btrfs_permission,
11493 .listxattr = btrfs_listxattr,
11494 .get_acl = btrfs_get_acl,
11495 .set_acl = btrfs_set_acl,
11496 .update_time = btrfs_update_time,
11498 static const struct inode_operations btrfs_symlink_inode_operations = {
11499 .get_link = page_get_link,
11500 .getattr = btrfs_getattr,
11501 .setattr = btrfs_setattr,
11502 .permission = btrfs_permission,
11503 .listxattr = btrfs_listxattr,
11504 .update_time = btrfs_update_time,
11507 const struct dentry_operations btrfs_dentry_operations = {
11508 .d_delete = btrfs_dentry_delete,