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 wait_on_bit_io(&inode->root->fs_info->flags,
1648 BTRFS_FS_NEED_ZONE_FINISH,
1649 TASK_UNINTERRUPTIBLE);
1653 if (!locked_page_done) {
1654 __set_page_dirty_nobuffers(locked_page);
1655 account_page_redirty(locked_page);
1657 locked_page_done = true;
1658 extent_write_locked_range(&inode->vfs_inode, start, done_offset);
1660 start = done_offset + 1;
1668 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1669 u64 bytenr, u64 num_bytes)
1671 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1672 struct btrfs_ordered_sum *sums;
1676 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1677 bytenr + num_bytes - 1, &list, 0);
1678 if (ret == 0 && list_empty(&list))
1681 while (!list_empty(&list)) {
1682 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1683 list_del(&sums->list);
1691 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1692 const u64 start, const u64 end,
1693 int *page_started, unsigned long *nr_written)
1695 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1696 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1697 const u64 range_bytes = end + 1 - start;
1698 struct extent_io_tree *io_tree = &inode->io_tree;
1699 u64 range_start = start;
1703 * If EXTENT_NORESERVE is set it means that when the buffered write was
1704 * made we had not enough available data space and therefore we did not
1705 * reserve data space for it, since we though we could do NOCOW for the
1706 * respective file range (either there is prealloc extent or the inode
1707 * has the NOCOW bit set).
1709 * However when we need to fallback to COW mode (because for example the
1710 * block group for the corresponding extent was turned to RO mode by a
1711 * scrub or relocation) we need to do the following:
1713 * 1) We increment the bytes_may_use counter of the data space info.
1714 * If COW succeeds, it allocates a new data extent and after doing
1715 * that it decrements the space info's bytes_may_use counter and
1716 * increments its bytes_reserved counter by the same amount (we do
1717 * this at btrfs_add_reserved_bytes()). So we need to increment the
1718 * bytes_may_use counter to compensate (when space is reserved at
1719 * buffered write time, the bytes_may_use counter is incremented);
1721 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1722 * that if the COW path fails for any reason, it decrements (through
1723 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1724 * data space info, which we incremented in the step above.
1726 * If we need to fallback to cow and the inode corresponds to a free
1727 * space cache inode or an inode of the data relocation tree, we must
1728 * also increment bytes_may_use of the data space_info for the same
1729 * reason. Space caches and relocated data extents always get a prealloc
1730 * extent for them, however scrub or balance may have set the block
1731 * group that contains that extent to RO mode and therefore force COW
1732 * when starting writeback.
1734 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1735 EXTENT_NORESERVE, 0);
1736 if (count > 0 || is_space_ino || is_reloc_ino) {
1738 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1739 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1741 if (is_space_ino || is_reloc_ino)
1742 bytes = range_bytes;
1744 spin_lock(&sinfo->lock);
1745 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1746 spin_unlock(&sinfo->lock);
1749 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1753 return cow_file_range(inode, locked_page, start, end, page_started,
1754 nr_written, 1, NULL);
1757 struct can_nocow_file_extent_args {
1760 /* Start file offset of the range we want to NOCOW. */
1762 /* End file offset (inclusive) of the range we want to NOCOW. */
1764 bool writeback_path;
1767 * Free the path passed to can_nocow_file_extent() once it's not needed
1772 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1777 /* Number of bytes that can be written to in NOCOW mode. */
1782 * Check if we can NOCOW the file extent that the path points to.
1783 * This function may return with the path released, so the caller should check
1784 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1786 * Returns: < 0 on error
1787 * 0 if we can not NOCOW
1790 static int can_nocow_file_extent(struct btrfs_path *path,
1791 struct btrfs_key *key,
1792 struct btrfs_inode *inode,
1793 struct can_nocow_file_extent_args *args)
1795 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1796 struct extent_buffer *leaf = path->nodes[0];
1797 struct btrfs_root *root = inode->root;
1798 struct btrfs_file_extent_item *fi;
1804 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1805 extent_type = btrfs_file_extent_type(leaf, fi);
1807 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1810 /* Can't access these fields unless we know it's not an inline extent. */
1811 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1812 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1813 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1815 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1816 extent_type == BTRFS_FILE_EXTENT_REG)
1820 * If the extent was created before the generation where the last snapshot
1821 * for its subvolume was created, then this implies the extent is shared,
1822 * hence we must COW.
1824 if (!args->strict &&
1825 btrfs_file_extent_generation(leaf, fi) <=
1826 btrfs_root_last_snapshot(&root->root_item))
1829 /* An explicit hole, must COW. */
1830 if (args->disk_bytenr == 0)
1833 /* Compressed/encrypted/encoded extents must be COWed. */
1834 if (btrfs_file_extent_compression(leaf, fi) ||
1835 btrfs_file_extent_encryption(leaf, fi) ||
1836 btrfs_file_extent_other_encoding(leaf, fi))
1839 extent_end = btrfs_file_extent_end(path);
1842 * The following checks can be expensive, as they need to take other
1843 * locks and do btree or rbtree searches, so release the path to avoid
1844 * blocking other tasks for too long.
1846 btrfs_release_path(path);
1848 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1849 key->offset - args->extent_offset,
1850 args->disk_bytenr, false, path);
1851 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1855 if (args->free_path) {
1857 * We don't need the path anymore, plus through the
1858 * csum_exist_in_range() call below we will end up allocating
1859 * another path. So free the path to avoid unnecessary extra
1862 btrfs_free_path(path);
1866 /* If there are pending snapshots for this root, we must COW. */
1867 if (args->writeback_path && !is_freespace_inode &&
1868 atomic_read(&root->snapshot_force_cow))
1871 args->disk_bytenr += args->extent_offset;
1872 args->disk_bytenr += args->start - key->offset;
1873 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1876 * Force COW if csums exist in the range. This ensures that csums for a
1877 * given extent are either valid or do not exist.
1879 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes);
1880 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1886 if (args->free_path && path)
1887 btrfs_free_path(path);
1889 return ret < 0 ? ret : can_nocow;
1893 * when nowcow writeback call back. This checks for snapshots or COW copies
1894 * of the extents that exist in the file, and COWs the file as required.
1896 * If no cow copies or snapshots exist, we write directly to the existing
1899 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1900 struct page *locked_page,
1901 const u64 start, const u64 end,
1903 unsigned long *nr_written)
1905 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1906 struct btrfs_root *root = inode->root;
1907 struct btrfs_path *path;
1908 u64 cow_start = (u64)-1;
1909 u64 cur_offset = start;
1911 bool check_prev = true;
1912 u64 ino = btrfs_ino(inode);
1913 struct btrfs_block_group *bg;
1915 struct can_nocow_file_extent_args nocow_args = { 0 };
1917 path = btrfs_alloc_path();
1919 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1920 EXTENT_LOCKED | EXTENT_DELALLOC |
1921 EXTENT_DO_ACCOUNTING |
1922 EXTENT_DEFRAG, PAGE_UNLOCK |
1923 PAGE_START_WRITEBACK |
1924 PAGE_END_WRITEBACK);
1928 nocow_args.end = end;
1929 nocow_args.writeback_path = true;
1932 struct btrfs_key found_key;
1933 struct btrfs_file_extent_item *fi;
1934 struct extent_buffer *leaf;
1942 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1948 * If there is no extent for our range when doing the initial
1949 * search, then go back to the previous slot as it will be the
1950 * one containing the search offset
1952 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1953 leaf = path->nodes[0];
1954 btrfs_item_key_to_cpu(leaf, &found_key,
1955 path->slots[0] - 1);
1956 if (found_key.objectid == ino &&
1957 found_key.type == BTRFS_EXTENT_DATA_KEY)
1962 /* Go to next leaf if we have exhausted the current one */
1963 leaf = path->nodes[0];
1964 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1965 ret = btrfs_next_leaf(root, path);
1967 if (cow_start != (u64)-1)
1968 cur_offset = cow_start;
1973 leaf = path->nodes[0];
1976 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1978 /* Didn't find anything for our INO */
1979 if (found_key.objectid > ino)
1982 * Keep searching until we find an EXTENT_ITEM or there are no
1983 * more extents for this inode
1985 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1986 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1991 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1992 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1993 found_key.offset > end)
1997 * If the found extent starts after requested offset, then
1998 * adjust extent_end to be right before this extent begins
2000 if (found_key.offset > cur_offset) {
2001 extent_end = found_key.offset;
2007 * Found extent which begins before our range and potentially
2010 fi = btrfs_item_ptr(leaf, path->slots[0],
2011 struct btrfs_file_extent_item);
2012 extent_type = btrfs_file_extent_type(leaf, fi);
2013 /* If this is triggered then we have a memory corruption. */
2014 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2015 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2019 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2020 extent_end = btrfs_file_extent_end(path);
2023 * If the extent we got ends before our current offset, skip to
2026 if (extent_end <= cur_offset) {
2031 nocow_args.start = cur_offset;
2032 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2034 if (cow_start != (u64)-1)
2035 cur_offset = cow_start;
2037 } else if (ret == 0) {
2042 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2047 * If nocow is false then record the beginning of the range
2048 * that needs to be COWed
2051 if (cow_start == (u64)-1)
2052 cow_start = cur_offset;
2053 cur_offset = extent_end;
2054 if (cur_offset > end)
2056 if (!path->nodes[0])
2063 * COW range from cow_start to found_key.offset - 1. As the key
2064 * will contain the beginning of the first extent that can be
2065 * NOCOW, following one which needs to be COW'ed
2067 if (cow_start != (u64)-1) {
2068 ret = fallback_to_cow(inode, locked_page,
2069 cow_start, found_key.offset - 1,
2070 page_started, nr_written);
2073 cow_start = (u64)-1;
2076 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2078 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2079 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2080 struct extent_map *em;
2082 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2084 nocow_args.disk_bytenr, /* block_start */
2085 nocow_args.num_bytes, /* block_len */
2086 nocow_args.disk_num_bytes, /* orig_block_len */
2087 ram_bytes, BTRFS_COMPRESS_NONE,
2088 BTRFS_ORDERED_PREALLOC);
2093 free_extent_map(em);
2094 ret = btrfs_add_ordered_extent(inode,
2095 cur_offset, nocow_args.num_bytes,
2096 nocow_args.num_bytes,
2097 nocow_args.disk_bytenr,
2098 nocow_args.num_bytes, 0,
2099 1 << BTRFS_ORDERED_PREALLOC,
2100 BTRFS_COMPRESS_NONE);
2102 btrfs_drop_extent_cache(inode, cur_offset,
2107 ret = btrfs_add_ordered_extent(inode, cur_offset,
2108 nocow_args.num_bytes,
2109 nocow_args.num_bytes,
2110 nocow_args.disk_bytenr,
2111 nocow_args.num_bytes,
2113 1 << BTRFS_ORDERED_NOCOW,
2114 BTRFS_COMPRESS_NONE);
2120 btrfs_dec_nocow_writers(bg);
2124 if (btrfs_is_data_reloc_root(root))
2126 * Error handled later, as we must prevent
2127 * extent_clear_unlock_delalloc() in error handler
2128 * from freeing metadata of created ordered extent.
2130 ret = btrfs_reloc_clone_csums(inode, cur_offset,
2131 nocow_args.num_bytes);
2133 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2134 locked_page, EXTENT_LOCKED |
2136 EXTENT_CLEAR_DATA_RESV,
2137 PAGE_UNLOCK | PAGE_SET_ORDERED);
2139 cur_offset = extent_end;
2142 * btrfs_reloc_clone_csums() error, now we're OK to call error
2143 * handler, as metadata for created ordered extent will only
2144 * be freed by btrfs_finish_ordered_io().
2148 if (cur_offset > end)
2151 btrfs_release_path(path);
2153 if (cur_offset <= end && cow_start == (u64)-1)
2154 cow_start = cur_offset;
2156 if (cow_start != (u64)-1) {
2158 ret = fallback_to_cow(inode, locked_page, cow_start, end,
2159 page_started, nr_written);
2166 btrfs_dec_nocow_writers(bg);
2168 if (ret && cur_offset < end)
2169 extent_clear_unlock_delalloc(inode, cur_offset, end,
2170 locked_page, EXTENT_LOCKED |
2171 EXTENT_DELALLOC | EXTENT_DEFRAG |
2172 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2173 PAGE_START_WRITEBACK |
2174 PAGE_END_WRITEBACK);
2175 btrfs_free_path(path);
2179 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2181 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2182 if (inode->defrag_bytes &&
2183 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2192 * Function to process delayed allocation (create CoW) for ranges which are
2193 * being touched for the first time.
2195 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2196 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2197 struct writeback_control *wbc)
2200 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2203 * The range must cover part of the @locked_page, or the returned
2204 * @page_started can confuse the caller.
2206 ASSERT(!(end <= page_offset(locked_page) ||
2207 start >= page_offset(locked_page) + PAGE_SIZE));
2209 if (should_nocow(inode, start, end)) {
2211 * Normally on a zoned device we're only doing COW writes, but
2212 * in case of relocation on a zoned filesystem we have taken
2213 * precaution, that we're only writing sequentially. It's safe
2214 * to use run_delalloc_nocow() here, like for regular
2215 * preallocated inodes.
2217 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2218 ret = run_delalloc_nocow(inode, locked_page, start, end,
2219 page_started, nr_written);
2220 } else if (!btrfs_inode_can_compress(inode) ||
2221 !inode_need_compress(inode, start, end)) {
2223 ret = run_delalloc_zoned(inode, locked_page, start, end,
2224 page_started, nr_written);
2226 ret = cow_file_range(inode, locked_page, start, end,
2227 page_started, nr_written, 1, NULL);
2229 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2230 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2231 page_started, nr_written);
2235 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2240 void btrfs_split_delalloc_extent(struct inode *inode,
2241 struct extent_state *orig, u64 split)
2243 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2246 /* not delalloc, ignore it */
2247 if (!(orig->state & EXTENT_DELALLOC))
2250 size = orig->end - orig->start + 1;
2251 if (size > fs_info->max_extent_size) {
2256 * See the explanation in btrfs_merge_delalloc_extent, the same
2257 * applies here, just in reverse.
2259 new_size = orig->end - split + 1;
2260 num_extents = count_max_extents(fs_info, new_size);
2261 new_size = split - orig->start;
2262 num_extents += count_max_extents(fs_info, new_size);
2263 if (count_max_extents(fs_info, size) >= num_extents)
2267 spin_lock(&BTRFS_I(inode)->lock);
2268 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2269 spin_unlock(&BTRFS_I(inode)->lock);
2273 * Handle merged delayed allocation extents so we can keep track of new extents
2274 * that are just merged onto old extents, such as when we are doing sequential
2275 * writes, so we can properly account for the metadata space we'll need.
2277 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2278 struct extent_state *other)
2280 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2281 u64 new_size, old_size;
2284 /* not delalloc, ignore it */
2285 if (!(other->state & EXTENT_DELALLOC))
2288 if (new->start > other->start)
2289 new_size = new->end - other->start + 1;
2291 new_size = other->end - new->start + 1;
2293 /* we're not bigger than the max, unreserve the space and go */
2294 if (new_size <= fs_info->max_extent_size) {
2295 spin_lock(&BTRFS_I(inode)->lock);
2296 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2297 spin_unlock(&BTRFS_I(inode)->lock);
2302 * We have to add up either side to figure out how many extents were
2303 * accounted for before we merged into one big extent. If the number of
2304 * extents we accounted for is <= the amount we need for the new range
2305 * then we can return, otherwise drop. Think of it like this
2309 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2310 * need 2 outstanding extents, on one side we have 1 and the other side
2311 * we have 1 so they are == and we can return. But in this case
2313 * [MAX_SIZE+4k][MAX_SIZE+4k]
2315 * Each range on their own accounts for 2 extents, but merged together
2316 * they are only 3 extents worth of accounting, so we need to drop in
2319 old_size = other->end - other->start + 1;
2320 num_extents = count_max_extents(fs_info, old_size);
2321 old_size = new->end - new->start + 1;
2322 num_extents += count_max_extents(fs_info, old_size);
2323 if (count_max_extents(fs_info, new_size) >= num_extents)
2326 spin_lock(&BTRFS_I(inode)->lock);
2327 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2328 spin_unlock(&BTRFS_I(inode)->lock);
2331 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2332 struct inode *inode)
2334 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2336 spin_lock(&root->delalloc_lock);
2337 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2338 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2339 &root->delalloc_inodes);
2340 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2341 &BTRFS_I(inode)->runtime_flags);
2342 root->nr_delalloc_inodes++;
2343 if (root->nr_delalloc_inodes == 1) {
2344 spin_lock(&fs_info->delalloc_root_lock);
2345 BUG_ON(!list_empty(&root->delalloc_root));
2346 list_add_tail(&root->delalloc_root,
2347 &fs_info->delalloc_roots);
2348 spin_unlock(&fs_info->delalloc_root_lock);
2351 spin_unlock(&root->delalloc_lock);
2355 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2356 struct btrfs_inode *inode)
2358 struct btrfs_fs_info *fs_info = root->fs_info;
2360 if (!list_empty(&inode->delalloc_inodes)) {
2361 list_del_init(&inode->delalloc_inodes);
2362 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2363 &inode->runtime_flags);
2364 root->nr_delalloc_inodes--;
2365 if (!root->nr_delalloc_inodes) {
2366 ASSERT(list_empty(&root->delalloc_inodes));
2367 spin_lock(&fs_info->delalloc_root_lock);
2368 BUG_ON(list_empty(&root->delalloc_root));
2369 list_del_init(&root->delalloc_root);
2370 spin_unlock(&fs_info->delalloc_root_lock);
2375 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2376 struct btrfs_inode *inode)
2378 spin_lock(&root->delalloc_lock);
2379 __btrfs_del_delalloc_inode(root, inode);
2380 spin_unlock(&root->delalloc_lock);
2384 * Properly track delayed allocation bytes in the inode and to maintain the
2385 * list of inodes that have pending delalloc work to be done.
2387 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2390 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2392 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2395 * set_bit and clear bit hooks normally require _irqsave/restore
2396 * but in this case, we are only testing for the DELALLOC
2397 * bit, which is only set or cleared with irqs on
2399 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2400 struct btrfs_root *root = BTRFS_I(inode)->root;
2401 u64 len = state->end + 1 - state->start;
2402 u32 num_extents = count_max_extents(fs_info, len);
2403 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2405 spin_lock(&BTRFS_I(inode)->lock);
2406 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2407 spin_unlock(&BTRFS_I(inode)->lock);
2409 /* For sanity tests */
2410 if (btrfs_is_testing(fs_info))
2413 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2414 fs_info->delalloc_batch);
2415 spin_lock(&BTRFS_I(inode)->lock);
2416 BTRFS_I(inode)->delalloc_bytes += len;
2417 if (bits & EXTENT_DEFRAG)
2418 BTRFS_I(inode)->defrag_bytes += len;
2419 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2420 &BTRFS_I(inode)->runtime_flags))
2421 btrfs_add_delalloc_inodes(root, inode);
2422 spin_unlock(&BTRFS_I(inode)->lock);
2425 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2426 (bits & EXTENT_DELALLOC_NEW)) {
2427 spin_lock(&BTRFS_I(inode)->lock);
2428 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2430 spin_unlock(&BTRFS_I(inode)->lock);
2435 * Once a range is no longer delalloc this function ensures that proper
2436 * accounting happens.
2438 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2439 struct extent_state *state, u32 bits)
2441 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2442 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2443 u64 len = state->end + 1 - state->start;
2444 u32 num_extents = count_max_extents(fs_info, len);
2446 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2447 spin_lock(&inode->lock);
2448 inode->defrag_bytes -= len;
2449 spin_unlock(&inode->lock);
2453 * set_bit and clear bit hooks normally require _irqsave/restore
2454 * but in this case, we are only testing for the DELALLOC
2455 * bit, which is only set or cleared with irqs on
2457 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2458 struct btrfs_root *root = inode->root;
2459 bool do_list = !btrfs_is_free_space_inode(inode);
2461 spin_lock(&inode->lock);
2462 btrfs_mod_outstanding_extents(inode, -num_extents);
2463 spin_unlock(&inode->lock);
2466 * We don't reserve metadata space for space cache inodes so we
2467 * don't need to call delalloc_release_metadata if there is an
2470 if (bits & EXTENT_CLEAR_META_RESV &&
2471 root != fs_info->tree_root)
2472 btrfs_delalloc_release_metadata(inode, len, false);
2474 /* For sanity tests. */
2475 if (btrfs_is_testing(fs_info))
2478 if (!btrfs_is_data_reloc_root(root) &&
2479 do_list && !(state->state & EXTENT_NORESERVE) &&
2480 (bits & EXTENT_CLEAR_DATA_RESV))
2481 btrfs_free_reserved_data_space_noquota(fs_info, len);
2483 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2484 fs_info->delalloc_batch);
2485 spin_lock(&inode->lock);
2486 inode->delalloc_bytes -= len;
2487 if (do_list && inode->delalloc_bytes == 0 &&
2488 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2489 &inode->runtime_flags))
2490 btrfs_del_delalloc_inode(root, inode);
2491 spin_unlock(&inode->lock);
2494 if ((state->state & EXTENT_DELALLOC_NEW) &&
2495 (bits & EXTENT_DELALLOC_NEW)) {
2496 spin_lock(&inode->lock);
2497 ASSERT(inode->new_delalloc_bytes >= len);
2498 inode->new_delalloc_bytes -= len;
2499 if (bits & EXTENT_ADD_INODE_BYTES)
2500 inode_add_bytes(&inode->vfs_inode, len);
2501 spin_unlock(&inode->lock);
2506 * in order to insert checksums into the metadata in large chunks,
2507 * we wait until bio submission time. All the pages in the bio are
2508 * checksummed and sums are attached onto the ordered extent record.
2510 * At IO completion time the cums attached on the ordered extent record
2511 * are inserted into the btree
2513 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2514 u64 dio_file_offset)
2516 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2520 * Split an extent_map at [start, start + len]
2522 * This function is intended to be used only for extract_ordered_extent().
2524 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2527 struct extent_map_tree *em_tree = &inode->extent_tree;
2528 struct extent_map *em;
2529 struct extent_map *split_pre = NULL;
2530 struct extent_map *split_mid = NULL;
2531 struct extent_map *split_post = NULL;
2533 unsigned long flags;
2536 if (pre == 0 && post == 0)
2539 split_pre = alloc_extent_map();
2541 split_mid = alloc_extent_map();
2543 split_post = alloc_extent_map();
2544 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2549 ASSERT(pre + post < len);
2551 lock_extent(&inode->io_tree, start, start + len - 1);
2552 write_lock(&em_tree->lock);
2553 em = lookup_extent_mapping(em_tree, start, len);
2559 ASSERT(em->len == len);
2560 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2561 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2562 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2563 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2564 ASSERT(!list_empty(&em->list));
2567 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2569 /* First, replace the em with a new extent_map starting from * em->start */
2570 split_pre->start = em->start;
2571 split_pre->len = (pre ? pre : em->len - post);
2572 split_pre->orig_start = split_pre->start;
2573 split_pre->block_start = em->block_start;
2574 split_pre->block_len = split_pre->len;
2575 split_pre->orig_block_len = split_pre->block_len;
2576 split_pre->ram_bytes = split_pre->len;
2577 split_pre->flags = flags;
2578 split_pre->compress_type = em->compress_type;
2579 split_pre->generation = em->generation;
2581 replace_extent_mapping(em_tree, em, split_pre, 1);
2584 * Now we only have an extent_map at:
2585 * [em->start, em->start + pre] if pre != 0
2586 * [em->start, em->start + em->len - post] if pre == 0
2590 /* Insert the middle extent_map */
2591 split_mid->start = em->start + pre;
2592 split_mid->len = em->len - pre - post;
2593 split_mid->orig_start = split_mid->start;
2594 split_mid->block_start = em->block_start + pre;
2595 split_mid->block_len = split_mid->len;
2596 split_mid->orig_block_len = split_mid->block_len;
2597 split_mid->ram_bytes = split_mid->len;
2598 split_mid->flags = flags;
2599 split_mid->compress_type = em->compress_type;
2600 split_mid->generation = em->generation;
2601 add_extent_mapping(em_tree, split_mid, 1);
2605 split_post->start = em->start + em->len - post;
2606 split_post->len = post;
2607 split_post->orig_start = split_post->start;
2608 split_post->block_start = em->block_start + em->len - post;
2609 split_post->block_len = split_post->len;
2610 split_post->orig_block_len = split_post->block_len;
2611 split_post->ram_bytes = split_post->len;
2612 split_post->flags = flags;
2613 split_post->compress_type = em->compress_type;
2614 split_post->generation = em->generation;
2615 add_extent_mapping(em_tree, split_post, 1);
2619 free_extent_map(em);
2620 /* Once for the tree */
2621 free_extent_map(em);
2624 write_unlock(&em_tree->lock);
2625 unlock_extent(&inode->io_tree, start, start + len - 1);
2627 free_extent_map(split_pre);
2628 free_extent_map(split_mid);
2629 free_extent_map(split_post);
2634 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2635 struct bio *bio, loff_t file_offset)
2637 struct btrfs_ordered_extent *ordered;
2638 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2640 u64 len = bio->bi_iter.bi_size;
2641 u64 end = start + len;
2646 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2647 if (WARN_ON_ONCE(!ordered))
2648 return BLK_STS_IOERR;
2650 /* No need to split */
2651 if (ordered->disk_num_bytes == len)
2654 /* We cannot split once end_bio'd ordered extent */
2655 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2660 /* We cannot split a compressed ordered extent */
2661 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2666 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2667 /* bio must be in one ordered extent */
2668 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2673 /* Checksum list should be empty */
2674 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2679 file_len = ordered->num_bytes;
2680 pre = start - ordered->disk_bytenr;
2681 post = ordered_end - end;
2683 ret = btrfs_split_ordered_extent(ordered, pre, post);
2686 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2689 btrfs_put_ordered_extent(ordered);
2691 return errno_to_blk_status(ret);
2694 void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio, int mirror_num)
2696 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2697 struct btrfs_inode *bi = BTRFS_I(inode);
2700 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2701 ret = extract_ordered_extent(bi, bio,
2702 page_offset(bio_first_bvec_all(bio)->bv_page));
2708 * If we need to checksum, and the I/O is not issued by fsync and
2709 * friends, that is ->sync_writers != 0, defer the submission to a
2710 * workqueue to parallelize it.
2712 * Csum items for reloc roots have already been cloned at this point,
2713 * so they are handled as part of the no-checksum case.
2715 if (!(bi->flags & BTRFS_INODE_NODATASUM) &&
2716 !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state) &&
2717 !btrfs_is_data_reloc_root(bi->root)) {
2718 if (!atomic_read(&bi->sync_writers) &&
2719 btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
2720 btrfs_submit_bio_start))
2723 ret = btrfs_csum_one_bio(bi, bio, (u64)-1, false);
2727 btrfs_submit_bio(fs_info, bio, mirror_num);
2731 bio->bi_status = ret;
2736 void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
2737 int mirror_num, enum btrfs_compression_type compress_type)
2739 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2742 if (compress_type != BTRFS_COMPRESS_NONE) {
2744 * btrfs_submit_compressed_read will handle completing the bio
2745 * if there were any errors, so just return here.
2747 btrfs_submit_compressed_read(inode, bio, mirror_num);
2751 /* Save the original iter for read repair */
2752 btrfs_bio(bio)->iter = bio->bi_iter;
2755 * Lookup bio sums does extra checks around whether we need to csum or
2756 * not, which is why we ignore skip_sum here.
2758 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2760 bio->bi_status = ret;
2765 btrfs_submit_bio(fs_info, bio, mirror_num);
2769 * given a list of ordered sums record them in the inode. This happens
2770 * at IO completion time based on sums calculated at bio submission time.
2772 static int add_pending_csums(struct btrfs_trans_handle *trans,
2773 struct list_head *list)
2775 struct btrfs_ordered_sum *sum;
2776 struct btrfs_root *csum_root = NULL;
2779 list_for_each_entry(sum, list, list) {
2780 trans->adding_csums = true;
2782 csum_root = btrfs_csum_root(trans->fs_info,
2784 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2785 trans->adding_csums = false;
2792 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2795 struct extent_state **cached_state)
2797 u64 search_start = start;
2798 const u64 end = start + len - 1;
2800 while (search_start < end) {
2801 const u64 search_len = end - search_start + 1;
2802 struct extent_map *em;
2806 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2810 if (em->block_start != EXTENT_MAP_HOLE)
2814 if (em->start < search_start)
2815 em_len -= search_start - em->start;
2816 if (em_len > search_len)
2817 em_len = search_len;
2819 ret = set_extent_bit(&inode->io_tree, search_start,
2820 search_start + em_len - 1,
2821 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2824 search_start = extent_map_end(em);
2825 free_extent_map(em);
2832 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2833 unsigned int extra_bits,
2834 struct extent_state **cached_state)
2836 WARN_ON(PAGE_ALIGNED(end));
2838 if (start >= i_size_read(&inode->vfs_inode) &&
2839 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2841 * There can't be any extents following eof in this case so just
2842 * set the delalloc new bit for the range directly.
2844 extra_bits |= EXTENT_DELALLOC_NEW;
2848 ret = btrfs_find_new_delalloc_bytes(inode, start,
2855 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2859 /* see btrfs_writepage_start_hook for details on why this is required */
2860 struct btrfs_writepage_fixup {
2862 struct inode *inode;
2863 struct btrfs_work work;
2866 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2868 struct btrfs_writepage_fixup *fixup;
2869 struct btrfs_ordered_extent *ordered;
2870 struct extent_state *cached_state = NULL;
2871 struct extent_changeset *data_reserved = NULL;
2873 struct btrfs_inode *inode;
2877 bool free_delalloc_space = true;
2879 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2881 inode = BTRFS_I(fixup->inode);
2882 page_start = page_offset(page);
2883 page_end = page_offset(page) + PAGE_SIZE - 1;
2886 * This is similar to page_mkwrite, we need to reserve the space before
2887 * we take the page lock.
2889 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2895 * Before we queued this fixup, we took a reference on the page.
2896 * page->mapping may go NULL, but it shouldn't be moved to a different
2899 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2901 * Unfortunately this is a little tricky, either
2903 * 1) We got here and our page had already been dealt with and
2904 * we reserved our space, thus ret == 0, so we need to just
2905 * drop our space reservation and bail. This can happen the
2906 * first time we come into the fixup worker, or could happen
2907 * while waiting for the ordered extent.
2908 * 2) Our page was already dealt with, but we happened to get an
2909 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2910 * this case we obviously don't have anything to release, but
2911 * because the page was already dealt with we don't want to
2912 * mark the page with an error, so make sure we're resetting
2913 * ret to 0. This is why we have this check _before_ the ret
2914 * check, because we do not want to have a surprise ENOSPC
2915 * when the page was already properly dealt with.
2918 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2919 btrfs_delalloc_release_space(inode, data_reserved,
2920 page_start, PAGE_SIZE,
2928 * We can't mess with the page state unless it is locked, so now that
2929 * it is locked bail if we failed to make our space reservation.
2934 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2936 /* already ordered? We're done */
2937 if (PageOrdered(page))
2940 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2942 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2945 btrfs_start_ordered_extent(ordered, 1);
2946 btrfs_put_ordered_extent(ordered);
2950 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2956 * Everything went as planned, we're now the owner of a dirty page with
2957 * delayed allocation bits set and space reserved for our COW
2960 * The page was dirty when we started, nothing should have cleaned it.
2962 BUG_ON(!PageDirty(page));
2963 free_delalloc_space = false;
2965 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2966 if (free_delalloc_space)
2967 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2969 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2974 * We hit ENOSPC or other errors. Update the mapping and page
2975 * to reflect the errors and clean the page.
2977 mapping_set_error(page->mapping, ret);
2978 end_extent_writepage(page, ret, page_start, page_end);
2979 clear_page_dirty_for_io(page);
2982 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2986 extent_changeset_free(data_reserved);
2988 * As a precaution, do a delayed iput in case it would be the last iput
2989 * that could need flushing space. Recursing back to fixup worker would
2992 btrfs_add_delayed_iput(&inode->vfs_inode);
2996 * There are a few paths in the higher layers of the kernel that directly
2997 * set the page dirty bit without asking the filesystem if it is a
2998 * good idea. This causes problems because we want to make sure COW
2999 * properly happens and the data=ordered rules are followed.
3001 * In our case any range that doesn't have the ORDERED bit set
3002 * hasn't been properly setup for IO. We kick off an async process
3003 * to fix it up. The async helper will wait for ordered extents, set
3004 * the delalloc bit and make it safe to write the page.
3006 int btrfs_writepage_cow_fixup(struct page *page)
3008 struct inode *inode = page->mapping->host;
3009 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3010 struct btrfs_writepage_fixup *fixup;
3012 /* This page has ordered extent covering it already */
3013 if (PageOrdered(page))
3017 * PageChecked is set below when we create a fixup worker for this page,
3018 * don't try to create another one if we're already PageChecked()
3020 * The extent_io writepage code will redirty the page if we send back
3023 if (PageChecked(page))
3026 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
3031 * We are already holding a reference to this inode from
3032 * write_cache_pages. We need to hold it because the space reservation
3033 * takes place outside of the page lock, and we can't trust
3034 * page->mapping outside of the page lock.
3037 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3039 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3041 fixup->inode = inode;
3042 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3047 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3048 struct btrfs_inode *inode, u64 file_pos,
3049 struct btrfs_file_extent_item *stack_fi,
3050 const bool update_inode_bytes,
3051 u64 qgroup_reserved)
3053 struct btrfs_root *root = inode->root;
3054 const u64 sectorsize = root->fs_info->sectorsize;
3055 struct btrfs_path *path;
3056 struct extent_buffer *leaf;
3057 struct btrfs_key ins;
3058 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3059 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3060 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3061 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3062 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3063 struct btrfs_drop_extents_args drop_args = { 0 };
3066 path = btrfs_alloc_path();
3071 * we may be replacing one extent in the tree with another.
3072 * The new extent is pinned in the extent map, and we don't want
3073 * to drop it from the cache until it is completely in the btree.
3075 * So, tell btrfs_drop_extents to leave this extent in the cache.
3076 * the caller is expected to unpin it and allow it to be merged
3079 drop_args.path = path;
3080 drop_args.start = file_pos;
3081 drop_args.end = file_pos + num_bytes;
3082 drop_args.replace_extent = true;
3083 drop_args.extent_item_size = sizeof(*stack_fi);
3084 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3088 if (!drop_args.extent_inserted) {
3089 ins.objectid = btrfs_ino(inode);
3090 ins.offset = file_pos;
3091 ins.type = BTRFS_EXTENT_DATA_KEY;
3093 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3098 leaf = path->nodes[0];
3099 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3100 write_extent_buffer(leaf, stack_fi,
3101 btrfs_item_ptr_offset(leaf, path->slots[0]),
3102 sizeof(struct btrfs_file_extent_item));
3104 btrfs_mark_buffer_dirty(leaf);
3105 btrfs_release_path(path);
3108 * If we dropped an inline extent here, we know the range where it is
3109 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3110 * number of bytes only for that range containing the inline extent.
3111 * The remaining of the range will be processed when clearning the
3112 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3114 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3115 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3117 inline_size = drop_args.bytes_found - inline_size;
3118 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3119 drop_args.bytes_found -= inline_size;
3120 num_bytes -= sectorsize;
3123 if (update_inode_bytes)
3124 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3126 ins.objectid = disk_bytenr;
3127 ins.offset = disk_num_bytes;
3128 ins.type = BTRFS_EXTENT_ITEM_KEY;
3130 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3134 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3136 qgroup_reserved, &ins);
3138 btrfs_free_path(path);
3143 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3146 struct btrfs_block_group *cache;
3148 cache = btrfs_lookup_block_group(fs_info, start);
3151 spin_lock(&cache->lock);
3152 cache->delalloc_bytes -= len;
3153 spin_unlock(&cache->lock);
3155 btrfs_put_block_group(cache);
3158 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3159 struct btrfs_ordered_extent *oe)
3161 struct btrfs_file_extent_item stack_fi;
3162 bool update_inode_bytes;
3163 u64 num_bytes = oe->num_bytes;
3164 u64 ram_bytes = oe->ram_bytes;
3166 memset(&stack_fi, 0, sizeof(stack_fi));
3167 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3168 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3169 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3170 oe->disk_num_bytes);
3171 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3172 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3173 num_bytes = oe->truncated_len;
3174 ram_bytes = num_bytes;
3176 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3177 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3178 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3179 /* Encryption and other encoding is reserved and all 0 */
3182 * For delalloc, when completing an ordered extent we update the inode's
3183 * bytes when clearing the range in the inode's io tree, so pass false
3184 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3185 * except if the ordered extent was truncated.
3187 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3188 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3189 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3191 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3192 oe->file_offset, &stack_fi,
3193 update_inode_bytes, oe->qgroup_rsv);
3197 * As ordered data IO finishes, this gets called so we can finish
3198 * an ordered extent if the range of bytes in the file it covers are
3201 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3203 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3204 struct btrfs_root *root = inode->root;
3205 struct btrfs_fs_info *fs_info = root->fs_info;
3206 struct btrfs_trans_handle *trans = NULL;
3207 struct extent_io_tree *io_tree = &inode->io_tree;
3208 struct extent_state *cached_state = NULL;
3210 int compress_type = 0;
3212 u64 logical_len = ordered_extent->num_bytes;
3213 bool freespace_inode;
3214 bool truncated = false;
3215 bool clear_reserved_extent = true;
3216 unsigned int clear_bits = EXTENT_DEFRAG;
3218 start = ordered_extent->file_offset;
3219 end = start + ordered_extent->num_bytes - 1;
3221 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3222 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3223 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3224 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3225 clear_bits |= EXTENT_DELALLOC_NEW;
3227 freespace_inode = btrfs_is_free_space_inode(inode);
3229 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3234 /* A valid bdev implies a write on a sequential zone */
3235 if (ordered_extent->bdev) {
3236 btrfs_rewrite_logical_zoned(ordered_extent);
3237 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3238 ordered_extent->disk_num_bytes);
3241 btrfs_free_io_failure_record(inode, start, end);
3243 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3245 logical_len = ordered_extent->truncated_len;
3246 /* Truncated the entire extent, don't bother adding */
3251 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3252 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3254 btrfs_inode_safe_disk_i_size_write(inode, 0);
3255 if (freespace_inode)
3256 trans = btrfs_join_transaction_spacecache(root);
3258 trans = btrfs_join_transaction(root);
3259 if (IS_ERR(trans)) {
3260 ret = PTR_ERR(trans);
3264 trans->block_rsv = &inode->block_rsv;
3265 ret = btrfs_update_inode_fallback(trans, root, inode);
3266 if (ret) /* -ENOMEM or corruption */
3267 btrfs_abort_transaction(trans, ret);
3271 clear_bits |= EXTENT_LOCKED;
3272 lock_extent_bits(io_tree, start, end, &cached_state);
3274 if (freespace_inode)
3275 trans = btrfs_join_transaction_spacecache(root);
3277 trans = btrfs_join_transaction(root);
3278 if (IS_ERR(trans)) {
3279 ret = PTR_ERR(trans);
3284 trans->block_rsv = &inode->block_rsv;
3286 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3287 compress_type = ordered_extent->compress_type;
3288 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3289 BUG_ON(compress_type);
3290 ret = btrfs_mark_extent_written(trans, inode,
3291 ordered_extent->file_offset,
3292 ordered_extent->file_offset +
3294 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3295 ordered_extent->disk_num_bytes);
3297 BUG_ON(root == fs_info->tree_root);
3298 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3300 clear_reserved_extent = false;
3301 btrfs_release_delalloc_bytes(fs_info,
3302 ordered_extent->disk_bytenr,
3303 ordered_extent->disk_num_bytes);
3306 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3307 ordered_extent->num_bytes, trans->transid);
3309 btrfs_abort_transaction(trans, ret);
3313 ret = add_pending_csums(trans, &ordered_extent->list);
3315 btrfs_abort_transaction(trans, ret);
3320 * If this is a new delalloc range, clear its new delalloc flag to
3321 * update the inode's number of bytes. This needs to be done first
3322 * before updating the inode item.
3324 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3325 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3326 clear_extent_bit(&inode->io_tree, start, end,
3327 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3328 0, 0, &cached_state);
3330 btrfs_inode_safe_disk_i_size_write(inode, 0);
3331 ret = btrfs_update_inode_fallback(trans, root, inode);
3332 if (ret) { /* -ENOMEM or corruption */
3333 btrfs_abort_transaction(trans, ret);
3338 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3339 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3343 btrfs_end_transaction(trans);
3345 if (ret || truncated) {
3346 u64 unwritten_start = start;
3349 * If we failed to finish this ordered extent for any reason we
3350 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3351 * extent, and mark the inode with the error if it wasn't
3352 * already set. Any error during writeback would have already
3353 * set the mapping error, so we need to set it if we're the ones
3354 * marking this ordered extent as failed.
3356 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3357 &ordered_extent->flags))
3358 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3361 unwritten_start += logical_len;
3362 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3364 /* Drop the cache for the part of the extent we didn't write. */
3365 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3368 * If the ordered extent had an IOERR or something else went
3369 * wrong we need to return the space for this ordered extent
3370 * back to the allocator. We only free the extent in the
3371 * truncated case if we didn't write out the extent at all.
3373 * If we made it past insert_reserved_file_extent before we
3374 * errored out then we don't need to do this as the accounting
3375 * has already been done.
3377 if ((ret || !logical_len) &&
3378 clear_reserved_extent &&
3379 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3380 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3382 * Discard the range before returning it back to the
3385 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3386 btrfs_discard_extent(fs_info,
3387 ordered_extent->disk_bytenr,
3388 ordered_extent->disk_num_bytes,
3390 btrfs_free_reserved_extent(fs_info,
3391 ordered_extent->disk_bytenr,
3392 ordered_extent->disk_num_bytes, 1);
3397 * This needs to be done to make sure anybody waiting knows we are done
3398 * updating everything for this ordered extent.
3400 btrfs_remove_ordered_extent(inode, ordered_extent);
3403 btrfs_put_ordered_extent(ordered_extent);
3404 /* once for the tree */
3405 btrfs_put_ordered_extent(ordered_extent);
3410 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3411 struct page *page, u64 start,
3412 u64 end, bool uptodate)
3414 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3416 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate);
3420 * Verify the checksum for a single sector without any extra action that depend
3421 * on the type of I/O.
3423 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3424 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3426 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3429 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3431 shash->tfm = fs_info->csum_shash;
3433 kaddr = kmap_local_page(page) + pgoff;
3434 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3435 kunmap_local(kaddr);
3437 if (memcmp(csum, csum_expected, fs_info->csum_size))
3443 * check_data_csum - verify checksum of one sector of uncompressed data
3445 * @bbio: btrfs_bio which contains the csum
3446 * @bio_offset: offset to the beginning of the bio (in bytes)
3447 * @page: page where is the data to be verified
3448 * @pgoff: offset inside the page
3450 * The length of such check is always one sector size.
3452 * When csum mismatch is detected, we will also report the error and fill the
3453 * corrupted range with zero. (Thus it needs the extra parameters)
3455 int btrfs_check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3456 u32 bio_offset, struct page *page, u32 pgoff)
3458 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3459 u32 len = fs_info->sectorsize;
3461 u8 csum[BTRFS_CSUM_SIZE];
3463 ASSERT(pgoff + len <= PAGE_SIZE);
3465 csum_expected = btrfs_csum_ptr(fs_info, bbio->csum, bio_offset);
3467 if (btrfs_check_sector_csum(fs_info, page, pgoff, csum, csum_expected))
3472 btrfs_print_data_csum_error(BTRFS_I(inode),
3473 bbio->file_offset + bio_offset,
3474 csum, csum_expected, bbio->mirror_num);
3476 btrfs_dev_stat_inc_and_print(bbio->device,
3477 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3478 memzero_page(page, pgoff, len);
3483 * When reads are done, we need to check csums to verify the data is correct.
3484 * if there's a match, we allow the bio to finish. If not, the code in
3485 * extent_io.c will try to find good copies for us.
3487 * @bio_offset: offset to the beginning of the bio (in bytes)
3488 * @start: file offset of the range start
3489 * @end: file offset of the range end (inclusive)
3491 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3494 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3495 u32 bio_offset, struct page *page,
3498 struct inode *inode = page->mapping->host;
3499 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3500 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3501 struct btrfs_root *root = BTRFS_I(inode)->root;
3502 const u32 sectorsize = root->fs_info->sectorsize;
3504 unsigned int result = 0;
3507 * This only happens for NODATASUM or compressed read.
3508 * Normally this should be covered by above check for compressed read
3509 * or the next check for NODATASUM. Just do a quicker exit here.
3511 if (bbio->csum == NULL)
3514 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3517 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3520 ASSERT(page_offset(page) <= start &&
3521 end <= page_offset(page) + PAGE_SIZE - 1);
3522 for (pg_off = offset_in_page(start);
3523 pg_off < offset_in_page(end);
3524 pg_off += sectorsize, bio_offset += sectorsize) {
3525 u64 file_offset = pg_off + page_offset(page);
3528 if (btrfs_is_data_reloc_root(root) &&
3529 test_range_bit(io_tree, file_offset,
3530 file_offset + sectorsize - 1,
3531 EXTENT_NODATASUM, 1, NULL)) {
3532 /* Skip the range without csum for data reloc inode */
3533 clear_extent_bits(io_tree, file_offset,
3534 file_offset + sectorsize - 1,
3538 ret = btrfs_check_data_csum(inode, bbio, bio_offset, page, pg_off);
3540 const int nr_bit = (pg_off - offset_in_page(start)) >>
3541 root->fs_info->sectorsize_bits;
3543 result |= (1U << nr_bit);
3550 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3552 * @inode: The inode we want to perform iput on
3554 * This function uses the generic vfs_inode::i_count to track whether we should
3555 * just decrement it (in case it's > 1) or if this is the last iput then link
3556 * the inode to the delayed iput machinery. Delayed iputs are processed at
3557 * transaction commit time/superblock commit/cleaner kthread.
3559 void btrfs_add_delayed_iput(struct inode *inode)
3561 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3562 struct btrfs_inode *binode = BTRFS_I(inode);
3564 if (atomic_add_unless(&inode->i_count, -1, 1))
3567 atomic_inc(&fs_info->nr_delayed_iputs);
3568 spin_lock(&fs_info->delayed_iput_lock);
3569 ASSERT(list_empty(&binode->delayed_iput));
3570 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3571 spin_unlock(&fs_info->delayed_iput_lock);
3572 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3573 wake_up_process(fs_info->cleaner_kthread);
3576 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3577 struct btrfs_inode *inode)
3579 list_del_init(&inode->delayed_iput);
3580 spin_unlock(&fs_info->delayed_iput_lock);
3581 iput(&inode->vfs_inode);
3582 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3583 wake_up(&fs_info->delayed_iputs_wait);
3584 spin_lock(&fs_info->delayed_iput_lock);
3587 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3588 struct btrfs_inode *inode)
3590 if (!list_empty(&inode->delayed_iput)) {
3591 spin_lock(&fs_info->delayed_iput_lock);
3592 if (!list_empty(&inode->delayed_iput))
3593 run_delayed_iput_locked(fs_info, inode);
3594 spin_unlock(&fs_info->delayed_iput_lock);
3598 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3601 spin_lock(&fs_info->delayed_iput_lock);
3602 while (!list_empty(&fs_info->delayed_iputs)) {
3603 struct btrfs_inode *inode;
3605 inode = list_first_entry(&fs_info->delayed_iputs,
3606 struct btrfs_inode, delayed_iput);
3607 run_delayed_iput_locked(fs_info, inode);
3608 cond_resched_lock(&fs_info->delayed_iput_lock);
3610 spin_unlock(&fs_info->delayed_iput_lock);
3614 * Wait for flushing all delayed iputs
3616 * @fs_info: the filesystem
3618 * This will wait on any delayed iputs that are currently running with KILLABLE
3619 * set. Once they are all done running we will return, unless we are killed in
3620 * which case we return EINTR. This helps in user operations like fallocate etc
3621 * that might get blocked on the iputs.
3623 * Return EINTR if we were killed, 0 if nothing's pending
3625 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3627 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3628 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3635 * This creates an orphan entry for the given inode in case something goes wrong
3636 * in the middle of an unlink.
3638 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3639 struct btrfs_inode *inode)
3643 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3644 if (ret && ret != -EEXIST) {
3645 btrfs_abort_transaction(trans, ret);
3653 * We have done the delete so we can go ahead and remove the orphan item for
3654 * this particular inode.
3656 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3657 struct btrfs_inode *inode)
3659 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3663 * this cleans up any orphans that may be left on the list from the last use
3666 int btrfs_orphan_cleanup(struct btrfs_root *root)
3668 struct btrfs_fs_info *fs_info = root->fs_info;
3669 struct btrfs_path *path;
3670 struct extent_buffer *leaf;
3671 struct btrfs_key key, found_key;
3672 struct btrfs_trans_handle *trans;
3673 struct inode *inode;
3674 u64 last_objectid = 0;
3675 int ret = 0, nr_unlink = 0;
3677 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3680 path = btrfs_alloc_path();
3685 path->reada = READA_BACK;
3687 key.objectid = BTRFS_ORPHAN_OBJECTID;
3688 key.type = BTRFS_ORPHAN_ITEM_KEY;
3689 key.offset = (u64)-1;
3692 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3697 * if ret == 0 means we found what we were searching for, which
3698 * is weird, but possible, so only screw with path if we didn't
3699 * find the key and see if we have stuff that matches
3703 if (path->slots[0] == 0)
3708 /* pull out the item */
3709 leaf = path->nodes[0];
3710 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3712 /* make sure the item matches what we want */
3713 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3715 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3718 /* release the path since we're done with it */
3719 btrfs_release_path(path);
3722 * this is where we are basically btrfs_lookup, without the
3723 * crossing root thing. we store the inode number in the
3724 * offset of the orphan item.
3727 if (found_key.offset == last_objectid) {
3729 "Error removing orphan entry, stopping orphan cleanup");
3734 last_objectid = found_key.offset;
3736 found_key.objectid = found_key.offset;
3737 found_key.type = BTRFS_INODE_ITEM_KEY;
3738 found_key.offset = 0;
3739 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3740 ret = PTR_ERR_OR_ZERO(inode);
3741 if (ret && ret != -ENOENT)
3744 if (ret == -ENOENT && root == fs_info->tree_root) {
3745 struct btrfs_root *dead_root;
3746 int is_dead_root = 0;
3749 * This is an orphan in the tree root. Currently these
3750 * could come from 2 sources:
3751 * a) a root (snapshot/subvolume) deletion in progress
3752 * b) a free space cache inode
3753 * We need to distinguish those two, as the orphan item
3754 * for a root must not get deleted before the deletion
3755 * of the snapshot/subvolume's tree completes.
3757 * btrfs_find_orphan_roots() ran before us, which has
3758 * found all deleted roots and loaded them into
3759 * fs_info->fs_roots_radix. So here we can find if an
3760 * orphan item corresponds to a deleted root by looking
3761 * up the root from that radix tree.
3764 spin_lock(&fs_info->fs_roots_radix_lock);
3765 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3766 (unsigned long)found_key.objectid);
3767 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3769 spin_unlock(&fs_info->fs_roots_radix_lock);
3772 /* prevent this orphan from being found again */
3773 key.offset = found_key.objectid - 1;
3780 * If we have an inode with links, there are a couple of
3783 * 1. We were halfway through creating fsverity metadata for the
3784 * file. In that case, the orphan item represents incomplete
3785 * fsverity metadata which must be cleaned up with
3786 * btrfs_drop_verity_items and deleting the orphan item.
3788 * 2. Old kernels (before v3.12) used to create an
3789 * orphan item for truncate indicating that there were possibly
3790 * extent items past i_size that needed to be deleted. In v3.12,
3791 * truncate was changed to update i_size in sync with the extent
3792 * items, but the (useless) orphan item was still created. Since
3793 * v4.18, we don't create the orphan item for truncate at all.
3795 * So, this item could mean that we need to do a truncate, but
3796 * only if this filesystem was last used on a pre-v3.12 kernel
3797 * and was not cleanly unmounted. The odds of that are quite
3798 * slim, and it's a pain to do the truncate now, so just delete
3801 * It's also possible that this orphan item was supposed to be
3802 * deleted but wasn't. The inode number may have been reused,
3803 * but either way, we can delete the orphan item.
3805 if (ret == -ENOENT || inode->i_nlink) {
3807 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3812 trans = btrfs_start_transaction(root, 1);
3813 if (IS_ERR(trans)) {
3814 ret = PTR_ERR(trans);
3817 btrfs_debug(fs_info, "auto deleting %Lu",
3818 found_key.objectid);
3819 ret = btrfs_del_orphan_item(trans, root,
3820 found_key.objectid);
3821 btrfs_end_transaction(trans);
3829 /* this will do delete_inode and everything for us */
3832 /* release the path since we're done with it */
3833 btrfs_release_path(path);
3835 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3836 trans = btrfs_join_transaction(root);
3838 btrfs_end_transaction(trans);
3842 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3846 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3847 btrfs_free_path(path);
3852 * very simple check to peek ahead in the leaf looking for xattrs. If we
3853 * don't find any xattrs, we know there can't be any acls.
3855 * slot is the slot the inode is in, objectid is the objectid of the inode
3857 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3858 int slot, u64 objectid,
3859 int *first_xattr_slot)
3861 u32 nritems = btrfs_header_nritems(leaf);
3862 struct btrfs_key found_key;
3863 static u64 xattr_access = 0;
3864 static u64 xattr_default = 0;
3867 if (!xattr_access) {
3868 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3869 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3870 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3871 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3875 *first_xattr_slot = -1;
3876 while (slot < nritems) {
3877 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3879 /* we found a different objectid, there must not be acls */
3880 if (found_key.objectid != objectid)
3883 /* we found an xattr, assume we've got an acl */
3884 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3885 if (*first_xattr_slot == -1)
3886 *first_xattr_slot = slot;
3887 if (found_key.offset == xattr_access ||
3888 found_key.offset == xattr_default)
3893 * we found a key greater than an xattr key, there can't
3894 * be any acls later on
3896 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3903 * it goes inode, inode backrefs, xattrs, extents,
3904 * so if there are a ton of hard links to an inode there can
3905 * be a lot of backrefs. Don't waste time searching too hard,
3906 * this is just an optimization
3911 /* we hit the end of the leaf before we found an xattr or
3912 * something larger than an xattr. We have to assume the inode
3915 if (*first_xattr_slot == -1)
3916 *first_xattr_slot = slot;
3921 * read an inode from the btree into the in-memory inode
3923 static int btrfs_read_locked_inode(struct inode *inode,
3924 struct btrfs_path *in_path)
3926 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3927 struct btrfs_path *path = in_path;
3928 struct extent_buffer *leaf;
3929 struct btrfs_inode_item *inode_item;
3930 struct btrfs_root *root = BTRFS_I(inode)->root;
3931 struct btrfs_key location;
3936 bool filled = false;
3937 int first_xattr_slot;
3939 ret = btrfs_fill_inode(inode, &rdev);
3944 path = btrfs_alloc_path();
3949 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3951 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3953 if (path != in_path)
3954 btrfs_free_path(path);
3958 leaf = path->nodes[0];
3963 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3964 struct btrfs_inode_item);
3965 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3966 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3967 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3968 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3969 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3970 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3971 round_up(i_size_read(inode), fs_info->sectorsize));
3973 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3974 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3976 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3977 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3979 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3980 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3982 BTRFS_I(inode)->i_otime.tv_sec =
3983 btrfs_timespec_sec(leaf, &inode_item->otime);
3984 BTRFS_I(inode)->i_otime.tv_nsec =
3985 btrfs_timespec_nsec(leaf, &inode_item->otime);
3987 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3988 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3989 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3991 inode_set_iversion_queried(inode,
3992 btrfs_inode_sequence(leaf, inode_item));
3993 inode->i_generation = BTRFS_I(inode)->generation;
3995 rdev = btrfs_inode_rdev(leaf, inode_item);
3997 BTRFS_I(inode)->index_cnt = (u64)-1;
3998 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3999 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
4003 * If we were modified in the current generation and evicted from memory
4004 * and then re-read we need to do a full sync since we don't have any
4005 * idea about which extents were modified before we were evicted from
4008 * This is required for both inode re-read from disk and delayed inode
4009 * in delayed_nodes_tree.
4011 if (BTRFS_I(inode)->last_trans == fs_info->generation)
4012 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4013 &BTRFS_I(inode)->runtime_flags);
4016 * We don't persist the id of the transaction where an unlink operation
4017 * against the inode was last made. So here we assume the inode might
4018 * have been evicted, and therefore the exact value of last_unlink_trans
4019 * lost, and set it to last_trans to avoid metadata inconsistencies
4020 * between the inode and its parent if the inode is fsync'ed and the log
4021 * replayed. For example, in the scenario:
4024 * ln mydir/foo mydir/bar
4027 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
4028 * xfs_io -c fsync mydir/foo
4030 * mount fs, triggers fsync log replay
4032 * We must make sure that when we fsync our inode foo we also log its
4033 * parent inode, otherwise after log replay the parent still has the
4034 * dentry with the "bar" name but our inode foo has a link count of 1
4035 * and doesn't have an inode ref with the name "bar" anymore.
4037 * Setting last_unlink_trans to last_trans is a pessimistic approach,
4038 * but it guarantees correctness at the expense of occasional full
4039 * transaction commits on fsync if our inode is a directory, or if our
4040 * inode is not a directory, logging its parent unnecessarily.
4042 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
4045 * Same logic as for last_unlink_trans. We don't persist the generation
4046 * of the last transaction where this inode was used for a reflink
4047 * operation, so after eviction and reloading the inode we must be
4048 * pessimistic and assume the last transaction that modified the inode.
4050 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
4053 if (inode->i_nlink != 1 ||
4054 path->slots[0] >= btrfs_header_nritems(leaf))
4057 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
4058 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
4061 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4062 if (location.type == BTRFS_INODE_REF_KEY) {
4063 struct btrfs_inode_ref *ref;
4065 ref = (struct btrfs_inode_ref *)ptr;
4066 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
4067 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
4068 struct btrfs_inode_extref *extref;
4070 extref = (struct btrfs_inode_extref *)ptr;
4071 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4076 * try to precache a NULL acl entry for files that don't have
4077 * any xattrs or acls
4079 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4080 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4081 if (first_xattr_slot != -1) {
4082 path->slots[0] = first_xattr_slot;
4083 ret = btrfs_load_inode_props(inode, path);
4086 "error loading props for ino %llu (root %llu): %d",
4087 btrfs_ino(BTRFS_I(inode)),
4088 root->root_key.objectid, ret);
4090 if (path != in_path)
4091 btrfs_free_path(path);
4094 cache_no_acl(inode);
4096 switch (inode->i_mode & S_IFMT) {
4098 inode->i_mapping->a_ops = &btrfs_aops;
4099 inode->i_fop = &btrfs_file_operations;
4100 inode->i_op = &btrfs_file_inode_operations;
4103 inode->i_fop = &btrfs_dir_file_operations;
4104 inode->i_op = &btrfs_dir_inode_operations;
4107 inode->i_op = &btrfs_symlink_inode_operations;
4108 inode_nohighmem(inode);
4109 inode->i_mapping->a_ops = &btrfs_aops;
4112 inode->i_op = &btrfs_special_inode_operations;
4113 init_special_inode(inode, inode->i_mode, rdev);
4117 btrfs_sync_inode_flags_to_i_flags(inode);
4122 * given a leaf and an inode, copy the inode fields into the leaf
4124 static void fill_inode_item(struct btrfs_trans_handle *trans,
4125 struct extent_buffer *leaf,
4126 struct btrfs_inode_item *item,
4127 struct inode *inode)
4129 struct btrfs_map_token token;
4132 btrfs_init_map_token(&token, leaf);
4134 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4135 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4136 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4137 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4138 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4140 btrfs_set_token_timespec_sec(&token, &item->atime,
4141 inode->i_atime.tv_sec);
4142 btrfs_set_token_timespec_nsec(&token, &item->atime,
4143 inode->i_atime.tv_nsec);
4145 btrfs_set_token_timespec_sec(&token, &item->mtime,
4146 inode->i_mtime.tv_sec);
4147 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4148 inode->i_mtime.tv_nsec);
4150 btrfs_set_token_timespec_sec(&token, &item->ctime,
4151 inode->i_ctime.tv_sec);
4152 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4153 inode->i_ctime.tv_nsec);
4155 btrfs_set_token_timespec_sec(&token, &item->otime,
4156 BTRFS_I(inode)->i_otime.tv_sec);
4157 btrfs_set_token_timespec_nsec(&token, &item->otime,
4158 BTRFS_I(inode)->i_otime.tv_nsec);
4160 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4161 btrfs_set_token_inode_generation(&token, item,
4162 BTRFS_I(inode)->generation);
4163 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4164 btrfs_set_token_inode_transid(&token, item, trans->transid);
4165 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4166 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4167 BTRFS_I(inode)->ro_flags);
4168 btrfs_set_token_inode_flags(&token, item, flags);
4169 btrfs_set_token_inode_block_group(&token, item, 0);
4173 * copy everything in the in-memory inode into the btree.
4175 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4176 struct btrfs_root *root,
4177 struct btrfs_inode *inode)
4179 struct btrfs_inode_item *inode_item;
4180 struct btrfs_path *path;
4181 struct extent_buffer *leaf;
4184 path = btrfs_alloc_path();
4188 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4195 leaf = path->nodes[0];
4196 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4197 struct btrfs_inode_item);
4199 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4200 btrfs_mark_buffer_dirty(leaf);
4201 btrfs_set_inode_last_trans(trans, inode);
4204 btrfs_free_path(path);
4209 * copy everything in the in-memory inode into the btree.
4211 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4212 struct btrfs_root *root,
4213 struct btrfs_inode *inode)
4215 struct btrfs_fs_info *fs_info = root->fs_info;
4219 * If the inode is a free space inode, we can deadlock during commit
4220 * if we put it into the delayed code.
4222 * The data relocation inode should also be directly updated
4225 if (!btrfs_is_free_space_inode(inode)
4226 && !btrfs_is_data_reloc_root(root)
4227 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4228 btrfs_update_root_times(trans, root);
4230 ret = btrfs_delayed_update_inode(trans, root, inode);
4232 btrfs_set_inode_last_trans(trans, inode);
4236 return btrfs_update_inode_item(trans, root, inode);
4239 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4240 struct btrfs_root *root, struct btrfs_inode *inode)
4244 ret = btrfs_update_inode(trans, root, inode);
4246 return btrfs_update_inode_item(trans, root, inode);
4251 * unlink helper that gets used here in inode.c and in the tree logging
4252 * recovery code. It remove a link in a directory with a given name, and
4253 * also drops the back refs in the inode to the directory
4255 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4256 struct btrfs_inode *dir,
4257 struct btrfs_inode *inode,
4258 const char *name, int name_len,
4259 struct btrfs_rename_ctx *rename_ctx)
4261 struct btrfs_root *root = dir->root;
4262 struct btrfs_fs_info *fs_info = root->fs_info;
4263 struct btrfs_path *path;
4265 struct btrfs_dir_item *di;
4267 u64 ino = btrfs_ino(inode);
4268 u64 dir_ino = btrfs_ino(dir);
4270 path = btrfs_alloc_path();
4276 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4277 name, name_len, -1);
4278 if (IS_ERR_OR_NULL(di)) {
4279 ret = di ? PTR_ERR(di) : -ENOENT;
4282 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4285 btrfs_release_path(path);
4288 * If we don't have dir index, we have to get it by looking up
4289 * the inode ref, since we get the inode ref, remove it directly,
4290 * it is unnecessary to do delayed deletion.
4292 * But if we have dir index, needn't search inode ref to get it.
4293 * Since the inode ref is close to the inode item, it is better
4294 * that we delay to delete it, and just do this deletion when
4295 * we update the inode item.
4297 if (inode->dir_index) {
4298 ret = btrfs_delayed_delete_inode_ref(inode);
4300 index = inode->dir_index;
4305 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4309 "failed to delete reference to %.*s, inode %llu parent %llu",
4310 name_len, name, ino, dir_ino);
4311 btrfs_abort_transaction(trans, ret);
4316 rename_ctx->index = index;
4318 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4320 btrfs_abort_transaction(trans, ret);
4325 * If we are in a rename context, we don't need to update anything in the
4326 * log. That will be done later during the rename by btrfs_log_new_name().
4327 * Besides that, doing it here would only cause extra unnecessary btree
4328 * operations on the log tree, increasing latency for applications.
4331 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4333 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4338 * If we have a pending delayed iput we could end up with the final iput
4339 * being run in btrfs-cleaner context. If we have enough of these built
4340 * up we can end up burning a lot of time in btrfs-cleaner without any
4341 * way to throttle the unlinks. Since we're currently holding a ref on
4342 * the inode we can run the delayed iput here without any issues as the
4343 * final iput won't be done until after we drop the ref we're currently
4346 btrfs_run_delayed_iput(fs_info, inode);
4348 btrfs_free_path(path);
4352 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4353 inode_inc_iversion(&inode->vfs_inode);
4354 inode_inc_iversion(&dir->vfs_inode);
4355 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4356 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4357 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4358 ret = btrfs_update_inode(trans, root, dir);
4363 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4364 struct btrfs_inode *dir, struct btrfs_inode *inode,
4365 const char *name, int name_len)
4368 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4370 drop_nlink(&inode->vfs_inode);
4371 ret = btrfs_update_inode(trans, inode->root, inode);
4377 * helper to start transaction for unlink and rmdir.
4379 * unlink and rmdir are special in btrfs, they do not always free space, so
4380 * if we cannot make our reservations the normal way try and see if there is
4381 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4382 * allow the unlink to occur.
4384 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4386 struct btrfs_root *root = BTRFS_I(dir)->root;
4389 * 1 for the possible orphan item
4390 * 1 for the dir item
4391 * 1 for the dir index
4392 * 1 for the inode ref
4394 * 1 for the parent inode
4396 return btrfs_start_transaction_fallback_global_rsv(root, 6);
4399 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4401 struct btrfs_trans_handle *trans;
4402 struct inode *inode = d_inode(dentry);
4405 trans = __unlink_start_trans(dir);
4407 return PTR_ERR(trans);
4409 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4412 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4413 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4414 dentry->d_name.len);
4418 if (inode->i_nlink == 0) {
4419 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4425 btrfs_end_transaction(trans);
4426 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4430 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4431 struct inode *dir, struct dentry *dentry)
4433 struct btrfs_root *root = BTRFS_I(dir)->root;
4434 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4435 struct btrfs_path *path;
4436 struct extent_buffer *leaf;
4437 struct btrfs_dir_item *di;
4438 struct btrfs_key key;
4439 const char *name = dentry->d_name.name;
4440 int name_len = dentry->d_name.len;
4444 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4446 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4447 objectid = inode->root->root_key.objectid;
4448 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4449 objectid = inode->location.objectid;
4455 path = btrfs_alloc_path();
4459 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4460 name, name_len, -1);
4461 if (IS_ERR_OR_NULL(di)) {
4462 ret = di ? PTR_ERR(di) : -ENOENT;
4466 leaf = path->nodes[0];
4467 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4468 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4469 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4471 btrfs_abort_transaction(trans, ret);
4474 btrfs_release_path(path);
4477 * This is a placeholder inode for a subvolume we didn't have a
4478 * reference to at the time of the snapshot creation. In the meantime
4479 * we could have renamed the real subvol link into our snapshot, so
4480 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4481 * Instead simply lookup the dir_index_item for this entry so we can
4482 * remove it. Otherwise we know we have a ref to the root and we can
4483 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4485 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4486 di = btrfs_search_dir_index_item(root, path, dir_ino,
4488 if (IS_ERR_OR_NULL(di)) {
4493 btrfs_abort_transaction(trans, ret);
4497 leaf = path->nodes[0];
4498 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4500 btrfs_release_path(path);
4502 ret = btrfs_del_root_ref(trans, objectid,
4503 root->root_key.objectid, dir_ino,
4504 &index, name, name_len);
4506 btrfs_abort_transaction(trans, ret);
4511 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4513 btrfs_abort_transaction(trans, ret);
4517 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4518 inode_inc_iversion(dir);
4519 dir->i_mtime = current_time(dir);
4520 dir->i_ctime = dir->i_mtime;
4521 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4523 btrfs_abort_transaction(trans, ret);
4525 btrfs_free_path(path);
4530 * Helper to check if the subvolume references other subvolumes or if it's
4533 static noinline int may_destroy_subvol(struct btrfs_root *root)
4535 struct btrfs_fs_info *fs_info = root->fs_info;
4536 struct btrfs_path *path;
4537 struct btrfs_dir_item *di;
4538 struct btrfs_key key;
4542 path = btrfs_alloc_path();
4546 /* Make sure this root isn't set as the default subvol */
4547 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4548 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4549 dir_id, "default", 7, 0);
4550 if (di && !IS_ERR(di)) {
4551 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4552 if (key.objectid == root->root_key.objectid) {
4555 "deleting default subvolume %llu is not allowed",
4559 btrfs_release_path(path);
4562 key.objectid = root->root_key.objectid;
4563 key.type = BTRFS_ROOT_REF_KEY;
4564 key.offset = (u64)-1;
4566 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4572 if (path->slots[0] > 0) {
4574 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4575 if (key.objectid == root->root_key.objectid &&
4576 key.type == BTRFS_ROOT_REF_KEY)
4580 btrfs_free_path(path);
4584 /* Delete all dentries for inodes belonging to the root */
4585 static void btrfs_prune_dentries(struct btrfs_root *root)
4587 struct btrfs_fs_info *fs_info = root->fs_info;
4588 struct rb_node *node;
4589 struct rb_node *prev;
4590 struct btrfs_inode *entry;
4591 struct inode *inode;
4594 if (!BTRFS_FS_ERROR(fs_info))
4595 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4597 spin_lock(&root->inode_lock);
4599 node = root->inode_tree.rb_node;
4603 entry = rb_entry(node, struct btrfs_inode, rb_node);
4605 if (objectid < btrfs_ino(entry))
4606 node = node->rb_left;
4607 else if (objectid > btrfs_ino(entry))
4608 node = node->rb_right;
4614 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4615 if (objectid <= btrfs_ino(entry)) {
4619 prev = rb_next(prev);
4623 entry = rb_entry(node, struct btrfs_inode, rb_node);
4624 objectid = btrfs_ino(entry) + 1;
4625 inode = igrab(&entry->vfs_inode);
4627 spin_unlock(&root->inode_lock);
4628 if (atomic_read(&inode->i_count) > 1)
4629 d_prune_aliases(inode);
4631 * btrfs_drop_inode will have it removed from the inode
4632 * cache when its usage count hits zero.
4636 spin_lock(&root->inode_lock);
4640 if (cond_resched_lock(&root->inode_lock))
4643 node = rb_next(node);
4645 spin_unlock(&root->inode_lock);
4648 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4650 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4651 struct btrfs_root *root = BTRFS_I(dir)->root;
4652 struct inode *inode = d_inode(dentry);
4653 struct btrfs_root *dest = BTRFS_I(inode)->root;
4654 struct btrfs_trans_handle *trans;
4655 struct btrfs_block_rsv block_rsv;
4660 * Don't allow to delete a subvolume with send in progress. This is
4661 * inside the inode lock so the error handling that has to drop the bit
4662 * again is not run concurrently.
4664 spin_lock(&dest->root_item_lock);
4665 if (dest->send_in_progress) {
4666 spin_unlock(&dest->root_item_lock);
4668 "attempt to delete subvolume %llu during send",
4669 dest->root_key.objectid);
4672 if (atomic_read(&dest->nr_swapfiles)) {
4673 spin_unlock(&dest->root_item_lock);
4675 "attempt to delete subvolume %llu with active swapfile",
4676 root->root_key.objectid);
4679 root_flags = btrfs_root_flags(&dest->root_item);
4680 btrfs_set_root_flags(&dest->root_item,
4681 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4682 spin_unlock(&dest->root_item_lock);
4684 down_write(&fs_info->subvol_sem);
4686 ret = may_destroy_subvol(dest);
4690 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4692 * One for dir inode,
4693 * two for dir entries,
4694 * two for root ref/backref.
4696 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4700 trans = btrfs_start_transaction(root, 0);
4701 if (IS_ERR(trans)) {
4702 ret = PTR_ERR(trans);
4705 trans->block_rsv = &block_rsv;
4706 trans->bytes_reserved = block_rsv.size;
4708 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4710 ret = btrfs_unlink_subvol(trans, dir, dentry);
4712 btrfs_abort_transaction(trans, ret);
4716 ret = btrfs_record_root_in_trans(trans, dest);
4718 btrfs_abort_transaction(trans, ret);
4722 memset(&dest->root_item.drop_progress, 0,
4723 sizeof(dest->root_item.drop_progress));
4724 btrfs_set_root_drop_level(&dest->root_item, 0);
4725 btrfs_set_root_refs(&dest->root_item, 0);
4727 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4728 ret = btrfs_insert_orphan_item(trans,
4730 dest->root_key.objectid);
4732 btrfs_abort_transaction(trans, ret);
4737 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4738 BTRFS_UUID_KEY_SUBVOL,
4739 dest->root_key.objectid);
4740 if (ret && ret != -ENOENT) {
4741 btrfs_abort_transaction(trans, ret);
4744 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4745 ret = btrfs_uuid_tree_remove(trans,
4746 dest->root_item.received_uuid,
4747 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4748 dest->root_key.objectid);
4749 if (ret && ret != -ENOENT) {
4750 btrfs_abort_transaction(trans, ret);
4755 free_anon_bdev(dest->anon_dev);
4758 trans->block_rsv = NULL;
4759 trans->bytes_reserved = 0;
4760 ret = btrfs_end_transaction(trans);
4761 inode->i_flags |= S_DEAD;
4763 btrfs_subvolume_release_metadata(root, &block_rsv);
4765 up_write(&fs_info->subvol_sem);
4767 spin_lock(&dest->root_item_lock);
4768 root_flags = btrfs_root_flags(&dest->root_item);
4769 btrfs_set_root_flags(&dest->root_item,
4770 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4771 spin_unlock(&dest->root_item_lock);
4773 d_invalidate(dentry);
4774 btrfs_prune_dentries(dest);
4775 ASSERT(dest->send_in_progress == 0);
4781 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4783 struct inode *inode = d_inode(dentry);
4784 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4786 struct btrfs_trans_handle *trans;
4787 u64 last_unlink_trans;
4789 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4791 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4792 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4794 "extent tree v2 doesn't support snapshot deletion yet");
4797 return btrfs_delete_subvolume(dir, dentry);
4800 trans = __unlink_start_trans(dir);
4802 return PTR_ERR(trans);
4804 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4805 err = btrfs_unlink_subvol(trans, dir, dentry);
4809 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4813 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4815 /* now the directory is empty */
4816 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4817 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4818 dentry->d_name.len);
4820 btrfs_i_size_write(BTRFS_I(inode), 0);
4822 * Propagate the last_unlink_trans value of the deleted dir to
4823 * its parent directory. This is to prevent an unrecoverable
4824 * log tree in the case we do something like this:
4826 * 2) create snapshot under dir foo
4827 * 3) delete the snapshot
4830 * 6) fsync foo or some file inside foo
4832 if (last_unlink_trans >= trans->transid)
4833 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4836 btrfs_end_transaction(trans);
4837 btrfs_btree_balance_dirty(fs_info);
4843 * btrfs_truncate_block - read, zero a chunk and write a block
4844 * @inode - inode that we're zeroing
4845 * @from - the offset to start zeroing
4846 * @len - the length to zero, 0 to zero the entire range respective to the
4848 * @front - zero up to the offset instead of from the offset on
4850 * This will find the block for the "from" offset and cow the block and zero the
4851 * part we want to zero. This is used with truncate and hole punching.
4853 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4856 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4857 struct address_space *mapping = inode->vfs_inode.i_mapping;
4858 struct extent_io_tree *io_tree = &inode->io_tree;
4859 struct btrfs_ordered_extent *ordered;
4860 struct extent_state *cached_state = NULL;
4861 struct extent_changeset *data_reserved = NULL;
4862 bool only_release_metadata = false;
4863 u32 blocksize = fs_info->sectorsize;
4864 pgoff_t index = from >> PAGE_SHIFT;
4865 unsigned offset = from & (blocksize - 1);
4867 gfp_t mask = btrfs_alloc_write_mask(mapping);
4868 size_t write_bytes = blocksize;
4873 if (IS_ALIGNED(offset, blocksize) &&
4874 (!len || IS_ALIGNED(len, blocksize)))
4877 block_start = round_down(from, blocksize);
4878 block_end = block_start + blocksize - 1;
4880 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4883 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4884 /* For nocow case, no need to reserve data space */
4885 only_release_metadata = true;
4890 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4892 if (!only_release_metadata)
4893 btrfs_free_reserved_data_space(inode, data_reserved,
4894 block_start, blocksize);
4898 page = find_or_create_page(mapping, index, mask);
4900 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4902 btrfs_delalloc_release_extents(inode, blocksize);
4906 ret = set_page_extent_mapped(page);
4910 if (!PageUptodate(page)) {
4911 ret = btrfs_read_folio(NULL, page_folio(page));
4913 if (page->mapping != mapping) {
4918 if (!PageUptodate(page)) {
4923 wait_on_page_writeback(page);
4925 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4927 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4929 unlock_extent_cached(io_tree, block_start, block_end,
4933 btrfs_start_ordered_extent(ordered, 1);
4934 btrfs_put_ordered_extent(ordered);
4938 clear_extent_bit(&inode->io_tree, block_start, block_end,
4939 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4940 0, 0, &cached_state);
4942 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4945 unlock_extent_cached(io_tree, block_start, block_end,
4950 if (offset != blocksize) {
4952 len = blocksize - offset;
4954 memzero_page(page, (block_start - page_offset(page)),
4957 memzero_page(page, (block_start - page_offset(page)) + offset,
4960 btrfs_page_clear_checked(fs_info, page, block_start,
4961 block_end + 1 - block_start);
4962 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4963 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4965 if (only_release_metadata)
4966 set_extent_bit(&inode->io_tree, block_start, block_end,
4967 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4971 if (only_release_metadata)
4972 btrfs_delalloc_release_metadata(inode, blocksize, true);
4974 btrfs_delalloc_release_space(inode, data_reserved,
4975 block_start, blocksize, true);
4977 btrfs_delalloc_release_extents(inode, blocksize);
4981 if (only_release_metadata)
4982 btrfs_check_nocow_unlock(inode);
4983 extent_changeset_free(data_reserved);
4987 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4988 u64 offset, u64 len)
4990 struct btrfs_fs_info *fs_info = root->fs_info;
4991 struct btrfs_trans_handle *trans;
4992 struct btrfs_drop_extents_args drop_args = { 0 };
4996 * If NO_HOLES is enabled, we don't need to do anything.
4997 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4998 * or btrfs_update_inode() will be called, which guarantee that the next
4999 * fsync will know this inode was changed and needs to be logged.
5001 if (btrfs_fs_incompat(fs_info, NO_HOLES))
5005 * 1 - for the one we're dropping
5006 * 1 - for the one we're adding
5007 * 1 - for updating the inode.
5009 trans = btrfs_start_transaction(root, 3);
5011 return PTR_ERR(trans);
5013 drop_args.start = offset;
5014 drop_args.end = offset + len;
5015 drop_args.drop_cache = true;
5017 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5019 btrfs_abort_transaction(trans, ret);
5020 btrfs_end_transaction(trans);
5024 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5025 offset, 0, 0, len, 0, len, 0, 0, 0);
5027 btrfs_abort_transaction(trans, ret);
5029 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5030 btrfs_update_inode(trans, root, inode);
5032 btrfs_end_transaction(trans);
5037 * This function puts in dummy file extents for the area we're creating a hole
5038 * for. So if we are truncating this file to a larger size we need to insert
5039 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5040 * the range between oldsize and size
5042 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5044 struct btrfs_root *root = inode->root;
5045 struct btrfs_fs_info *fs_info = root->fs_info;
5046 struct extent_io_tree *io_tree = &inode->io_tree;
5047 struct extent_map *em = NULL;
5048 struct extent_state *cached_state = NULL;
5049 struct extent_map_tree *em_tree = &inode->extent_tree;
5050 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5051 u64 block_end = ALIGN(size, fs_info->sectorsize);
5058 * If our size started in the middle of a block we need to zero out the
5059 * rest of the block before we expand the i_size, otherwise we could
5060 * expose stale data.
5062 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5066 if (size <= hole_start)
5069 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5071 cur_offset = hole_start;
5073 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5074 block_end - cur_offset);
5080 last_byte = min(extent_map_end(em), block_end);
5081 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5082 hole_size = last_byte - cur_offset;
5084 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5085 struct extent_map *hole_em;
5087 err = maybe_insert_hole(root, inode, cur_offset,
5092 err = btrfs_inode_set_file_extent_range(inode,
5093 cur_offset, hole_size);
5097 btrfs_drop_extent_cache(inode, cur_offset,
5098 cur_offset + hole_size - 1, 0);
5099 hole_em = alloc_extent_map();
5101 btrfs_set_inode_full_sync(inode);
5104 hole_em->start = cur_offset;
5105 hole_em->len = hole_size;
5106 hole_em->orig_start = cur_offset;
5108 hole_em->block_start = EXTENT_MAP_HOLE;
5109 hole_em->block_len = 0;
5110 hole_em->orig_block_len = 0;
5111 hole_em->ram_bytes = hole_size;
5112 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5113 hole_em->generation = fs_info->generation;
5116 write_lock(&em_tree->lock);
5117 err = add_extent_mapping(em_tree, hole_em, 1);
5118 write_unlock(&em_tree->lock);
5121 btrfs_drop_extent_cache(inode, cur_offset,
5125 free_extent_map(hole_em);
5127 err = btrfs_inode_set_file_extent_range(inode,
5128 cur_offset, hole_size);
5133 free_extent_map(em);
5135 cur_offset = last_byte;
5136 if (cur_offset >= block_end)
5139 free_extent_map(em);
5140 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5144 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5146 struct btrfs_root *root = BTRFS_I(inode)->root;
5147 struct btrfs_trans_handle *trans;
5148 loff_t oldsize = i_size_read(inode);
5149 loff_t newsize = attr->ia_size;
5150 int mask = attr->ia_valid;
5154 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5155 * special case where we need to update the times despite not having
5156 * these flags set. For all other operations the VFS set these flags
5157 * explicitly if it wants a timestamp update.
5159 if (newsize != oldsize) {
5160 inode_inc_iversion(inode);
5161 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5162 inode->i_mtime = current_time(inode);
5163 inode->i_ctime = inode->i_mtime;
5167 if (newsize > oldsize) {
5169 * Don't do an expanding truncate while snapshotting is ongoing.
5170 * This is to ensure the snapshot captures a fully consistent
5171 * state of this file - if the snapshot captures this expanding
5172 * truncation, it must capture all writes that happened before
5175 btrfs_drew_write_lock(&root->snapshot_lock);
5176 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5178 btrfs_drew_write_unlock(&root->snapshot_lock);
5182 trans = btrfs_start_transaction(root, 1);
5183 if (IS_ERR(trans)) {
5184 btrfs_drew_write_unlock(&root->snapshot_lock);
5185 return PTR_ERR(trans);
5188 i_size_write(inode, newsize);
5189 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5190 pagecache_isize_extended(inode, oldsize, newsize);
5191 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5192 btrfs_drew_write_unlock(&root->snapshot_lock);
5193 btrfs_end_transaction(trans);
5195 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5197 if (btrfs_is_zoned(fs_info)) {
5198 ret = btrfs_wait_ordered_range(inode,
5199 ALIGN(newsize, fs_info->sectorsize),
5206 * We're truncating a file that used to have good data down to
5207 * zero. Make sure any new writes to the file get on disk
5211 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5212 &BTRFS_I(inode)->runtime_flags);
5214 truncate_setsize(inode, newsize);
5216 inode_dio_wait(inode);
5218 ret = btrfs_truncate(inode, newsize == oldsize);
5219 if (ret && inode->i_nlink) {
5223 * Truncate failed, so fix up the in-memory size. We
5224 * adjusted disk_i_size down as we removed extents, so
5225 * wait for disk_i_size to be stable and then update the
5226 * in-memory size to match.
5228 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5231 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5238 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5241 struct inode *inode = d_inode(dentry);
5242 struct btrfs_root *root = BTRFS_I(inode)->root;
5245 if (btrfs_root_readonly(root))
5248 err = setattr_prepare(mnt_userns, dentry, attr);
5252 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5253 err = btrfs_setsize(inode, attr);
5258 if (attr->ia_valid) {
5259 setattr_copy(mnt_userns, inode, attr);
5260 inode_inc_iversion(inode);
5261 err = btrfs_dirty_inode(inode);
5263 if (!err && attr->ia_valid & ATTR_MODE)
5264 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5271 * While truncating the inode pages during eviction, we get the VFS
5272 * calling btrfs_invalidate_folio() against each folio of the inode. This
5273 * is slow because the calls to btrfs_invalidate_folio() result in a
5274 * huge amount of calls to lock_extent_bits() and clear_extent_bit(),
5275 * which keep merging and splitting extent_state structures over and over,
5276 * wasting lots of time.
5278 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5279 * skip all those expensive operations on a per folio basis and do only
5280 * the ordered io finishing, while we release here the extent_map and
5281 * extent_state structures, without the excessive merging and splitting.
5283 static void evict_inode_truncate_pages(struct inode *inode)
5285 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5286 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5287 struct rb_node *node;
5289 ASSERT(inode->i_state & I_FREEING);
5290 truncate_inode_pages_final(&inode->i_data);
5292 write_lock(&map_tree->lock);
5293 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5294 struct extent_map *em;
5296 node = rb_first_cached(&map_tree->map);
5297 em = rb_entry(node, struct extent_map, rb_node);
5298 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5299 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5300 remove_extent_mapping(map_tree, em);
5301 free_extent_map(em);
5302 if (need_resched()) {
5303 write_unlock(&map_tree->lock);
5305 write_lock(&map_tree->lock);
5308 write_unlock(&map_tree->lock);
5311 * Keep looping until we have no more ranges in the io tree.
5312 * We can have ongoing bios started by readahead that have
5313 * their endio callback (extent_io.c:end_bio_extent_readpage)
5314 * still in progress (unlocked the pages in the bio but did not yet
5315 * unlocked the ranges in the io tree). Therefore this means some
5316 * ranges can still be locked and eviction started because before
5317 * submitting those bios, which are executed by a separate task (work
5318 * queue kthread), inode references (inode->i_count) were not taken
5319 * (which would be dropped in the end io callback of each bio).
5320 * Therefore here we effectively end up waiting for those bios and
5321 * anyone else holding locked ranges without having bumped the inode's
5322 * reference count - if we don't do it, when they access the inode's
5323 * io_tree to unlock a range it may be too late, leading to an
5324 * use-after-free issue.
5326 spin_lock(&io_tree->lock);
5327 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5328 struct extent_state *state;
5329 struct extent_state *cached_state = NULL;
5332 unsigned state_flags;
5334 node = rb_first(&io_tree->state);
5335 state = rb_entry(node, struct extent_state, rb_node);
5336 start = state->start;
5338 state_flags = state->state;
5339 spin_unlock(&io_tree->lock);
5341 lock_extent_bits(io_tree, start, end, &cached_state);
5344 * If still has DELALLOC flag, the extent didn't reach disk,
5345 * and its reserved space won't be freed by delayed_ref.
5346 * So we need to free its reserved space here.
5347 * (Refer to comment in btrfs_invalidate_folio, case 2)
5349 * Note, end is the bytenr of last byte, so we need + 1 here.
5351 if (state_flags & EXTENT_DELALLOC)
5352 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5355 clear_extent_bit(io_tree, start, end,
5356 EXTENT_LOCKED | EXTENT_DELALLOC |
5357 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5361 spin_lock(&io_tree->lock);
5363 spin_unlock(&io_tree->lock);
5366 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5367 struct btrfs_block_rsv *rsv)
5369 struct btrfs_fs_info *fs_info = root->fs_info;
5370 struct btrfs_trans_handle *trans;
5371 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5375 * Eviction should be taking place at some place safe because of our
5376 * delayed iputs. However the normal flushing code will run delayed
5377 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5379 * We reserve the delayed_refs_extra here again because we can't use
5380 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5381 * above. We reserve our extra bit here because we generate a ton of
5382 * delayed refs activity by truncating.
5384 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5385 * if we fail to make this reservation we can re-try without the
5386 * delayed_refs_extra so we can make some forward progress.
5388 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5389 BTRFS_RESERVE_FLUSH_EVICT);
5391 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5392 BTRFS_RESERVE_FLUSH_EVICT);
5395 "could not allocate space for delete; will truncate on mount");
5396 return ERR_PTR(-ENOSPC);
5398 delayed_refs_extra = 0;
5401 trans = btrfs_join_transaction(root);
5405 if (delayed_refs_extra) {
5406 trans->block_rsv = &fs_info->trans_block_rsv;
5407 trans->bytes_reserved = delayed_refs_extra;
5408 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5409 delayed_refs_extra, 1);
5414 void btrfs_evict_inode(struct inode *inode)
5416 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5417 struct btrfs_trans_handle *trans;
5418 struct btrfs_root *root = BTRFS_I(inode)->root;
5419 struct btrfs_block_rsv *rsv;
5422 trace_btrfs_inode_evict(inode);
5425 fsverity_cleanup_inode(inode);
5430 evict_inode_truncate_pages(inode);
5432 if (inode->i_nlink &&
5433 ((btrfs_root_refs(&root->root_item) != 0 &&
5434 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5435 btrfs_is_free_space_inode(BTRFS_I(inode))))
5438 if (is_bad_inode(inode))
5441 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5443 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5446 if (inode->i_nlink > 0) {
5447 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5448 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5453 * This makes sure the inode item in tree is uptodate and the space for
5454 * the inode update is released.
5456 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5461 * This drops any pending insert or delete operations we have for this
5462 * inode. We could have a delayed dir index deletion queued up, but
5463 * we're removing the inode completely so that'll be taken care of in
5466 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5468 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5471 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5472 rsv->failfast = true;
5474 btrfs_i_size_write(BTRFS_I(inode), 0);
5477 struct btrfs_truncate_control control = {
5478 .inode = BTRFS_I(inode),
5479 .ino = btrfs_ino(BTRFS_I(inode)),
5484 trans = evict_refill_and_join(root, rsv);
5488 trans->block_rsv = rsv;
5490 ret = btrfs_truncate_inode_items(trans, root, &control);
5491 trans->block_rsv = &fs_info->trans_block_rsv;
5492 btrfs_end_transaction(trans);
5493 btrfs_btree_balance_dirty(fs_info);
5494 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5501 * Errors here aren't a big deal, it just means we leave orphan items in
5502 * the tree. They will be cleaned up on the next mount. If the inode
5503 * number gets reused, cleanup deletes the orphan item without doing
5504 * anything, and unlink reuses the existing orphan item.
5506 * If it turns out that we are dropping too many of these, we might want
5507 * to add a mechanism for retrying these after a commit.
5509 trans = evict_refill_and_join(root, rsv);
5510 if (!IS_ERR(trans)) {
5511 trans->block_rsv = rsv;
5512 btrfs_orphan_del(trans, BTRFS_I(inode));
5513 trans->block_rsv = &fs_info->trans_block_rsv;
5514 btrfs_end_transaction(trans);
5518 btrfs_free_block_rsv(fs_info, rsv);
5521 * If we didn't successfully delete, the orphan item will still be in
5522 * the tree and we'll retry on the next mount. Again, we might also want
5523 * to retry these periodically in the future.
5525 btrfs_remove_delayed_node(BTRFS_I(inode));
5526 fsverity_cleanup_inode(inode);
5531 * Return the key found in the dir entry in the location pointer, fill @type
5532 * with BTRFS_FT_*, and return 0.
5534 * If no dir entries were found, returns -ENOENT.
5535 * If found a corrupted location in dir entry, returns -EUCLEAN.
5537 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5538 struct btrfs_key *location, u8 *type)
5540 const char *name = dentry->d_name.name;
5541 int namelen = dentry->d_name.len;
5542 struct btrfs_dir_item *di;
5543 struct btrfs_path *path;
5544 struct btrfs_root *root = BTRFS_I(dir)->root;
5547 path = btrfs_alloc_path();
5551 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5553 if (IS_ERR_OR_NULL(di)) {
5554 ret = di ? PTR_ERR(di) : -ENOENT;
5558 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5559 if (location->type != BTRFS_INODE_ITEM_KEY &&
5560 location->type != BTRFS_ROOT_ITEM_KEY) {
5562 btrfs_warn(root->fs_info,
5563 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5564 __func__, name, btrfs_ino(BTRFS_I(dir)),
5565 location->objectid, location->type, location->offset);
5568 *type = btrfs_dir_type(path->nodes[0], di);
5570 btrfs_free_path(path);
5575 * when we hit a tree root in a directory, the btrfs part of the inode
5576 * needs to be changed to reflect the root directory of the tree root. This
5577 * is kind of like crossing a mount point.
5579 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5581 struct dentry *dentry,
5582 struct btrfs_key *location,
5583 struct btrfs_root **sub_root)
5585 struct btrfs_path *path;
5586 struct btrfs_root *new_root;
5587 struct btrfs_root_ref *ref;
5588 struct extent_buffer *leaf;
5589 struct btrfs_key key;
5593 path = btrfs_alloc_path();
5600 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5601 key.type = BTRFS_ROOT_REF_KEY;
5602 key.offset = location->objectid;
5604 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5611 leaf = path->nodes[0];
5612 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5613 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5614 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5617 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5618 (unsigned long)(ref + 1),
5619 dentry->d_name.len);
5623 btrfs_release_path(path);
5625 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5626 if (IS_ERR(new_root)) {
5627 err = PTR_ERR(new_root);
5631 *sub_root = new_root;
5632 location->objectid = btrfs_root_dirid(&new_root->root_item);
5633 location->type = BTRFS_INODE_ITEM_KEY;
5634 location->offset = 0;
5637 btrfs_free_path(path);
5641 static void inode_tree_add(struct inode *inode)
5643 struct btrfs_root *root = BTRFS_I(inode)->root;
5644 struct btrfs_inode *entry;
5646 struct rb_node *parent;
5647 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5648 u64 ino = btrfs_ino(BTRFS_I(inode));
5650 if (inode_unhashed(inode))
5653 spin_lock(&root->inode_lock);
5654 p = &root->inode_tree.rb_node;
5657 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5659 if (ino < btrfs_ino(entry))
5660 p = &parent->rb_left;
5661 else if (ino > btrfs_ino(entry))
5662 p = &parent->rb_right;
5664 WARN_ON(!(entry->vfs_inode.i_state &
5665 (I_WILL_FREE | I_FREEING)));
5666 rb_replace_node(parent, new, &root->inode_tree);
5667 RB_CLEAR_NODE(parent);
5668 spin_unlock(&root->inode_lock);
5672 rb_link_node(new, parent, p);
5673 rb_insert_color(new, &root->inode_tree);
5674 spin_unlock(&root->inode_lock);
5677 static void inode_tree_del(struct btrfs_inode *inode)
5679 struct btrfs_root *root = inode->root;
5682 spin_lock(&root->inode_lock);
5683 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5684 rb_erase(&inode->rb_node, &root->inode_tree);
5685 RB_CLEAR_NODE(&inode->rb_node);
5686 empty = RB_EMPTY_ROOT(&root->inode_tree);
5688 spin_unlock(&root->inode_lock);
5690 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5691 spin_lock(&root->inode_lock);
5692 empty = RB_EMPTY_ROOT(&root->inode_tree);
5693 spin_unlock(&root->inode_lock);
5695 btrfs_add_dead_root(root);
5700 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5702 struct btrfs_iget_args *args = p;
5704 inode->i_ino = args->ino;
5705 BTRFS_I(inode)->location.objectid = args->ino;
5706 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5707 BTRFS_I(inode)->location.offset = 0;
5708 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5709 BUG_ON(args->root && !BTRFS_I(inode)->root);
5713 static int btrfs_find_actor(struct inode *inode, void *opaque)
5715 struct btrfs_iget_args *args = opaque;
5717 return args->ino == BTRFS_I(inode)->location.objectid &&
5718 args->root == BTRFS_I(inode)->root;
5721 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5722 struct btrfs_root *root)
5724 struct inode *inode;
5725 struct btrfs_iget_args args;
5726 unsigned long hashval = btrfs_inode_hash(ino, root);
5731 inode = iget5_locked(s, hashval, btrfs_find_actor,
5732 btrfs_init_locked_inode,
5738 * Get an inode object given its inode number and corresponding root.
5739 * Path can be preallocated to prevent recursing back to iget through
5740 * allocator. NULL is also valid but may require an additional allocation
5743 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5744 struct btrfs_root *root, struct btrfs_path *path)
5746 struct inode *inode;
5748 inode = btrfs_iget_locked(s, ino, root);
5750 return ERR_PTR(-ENOMEM);
5752 if (inode->i_state & I_NEW) {
5755 ret = btrfs_read_locked_inode(inode, path);
5757 inode_tree_add(inode);
5758 unlock_new_inode(inode);
5762 * ret > 0 can come from btrfs_search_slot called by
5763 * btrfs_read_locked_inode, this means the inode item
5768 inode = ERR_PTR(ret);
5775 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5777 return btrfs_iget_path(s, ino, root, NULL);
5780 static struct inode *new_simple_dir(struct super_block *s,
5781 struct btrfs_key *key,
5782 struct btrfs_root *root)
5784 struct inode *inode = new_inode(s);
5787 return ERR_PTR(-ENOMEM);
5789 BTRFS_I(inode)->root = btrfs_grab_root(root);
5790 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5791 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5793 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5795 * We only need lookup, the rest is read-only and there's no inode
5796 * associated with the dentry
5798 inode->i_op = &simple_dir_inode_operations;
5799 inode->i_opflags &= ~IOP_XATTR;
5800 inode->i_fop = &simple_dir_operations;
5801 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5802 inode->i_mtime = current_time(inode);
5803 inode->i_atime = inode->i_mtime;
5804 inode->i_ctime = inode->i_mtime;
5805 BTRFS_I(inode)->i_otime = inode->i_mtime;
5810 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5811 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5812 static_assert(BTRFS_FT_DIR == FT_DIR);
5813 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5814 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5815 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5816 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5817 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5819 static inline u8 btrfs_inode_type(struct inode *inode)
5821 return fs_umode_to_ftype(inode->i_mode);
5824 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5826 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5827 struct inode *inode;
5828 struct btrfs_root *root = BTRFS_I(dir)->root;
5829 struct btrfs_root *sub_root = root;
5830 struct btrfs_key location;
5834 if (dentry->d_name.len > BTRFS_NAME_LEN)
5835 return ERR_PTR(-ENAMETOOLONG);
5837 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5839 return ERR_PTR(ret);
5841 if (location.type == BTRFS_INODE_ITEM_KEY) {
5842 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5846 /* Do extra check against inode mode with di_type */
5847 if (btrfs_inode_type(inode) != di_type) {
5849 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5850 inode->i_mode, btrfs_inode_type(inode),
5853 return ERR_PTR(-EUCLEAN);
5858 ret = fixup_tree_root_location(fs_info, dir, dentry,
5859 &location, &sub_root);
5862 inode = ERR_PTR(ret);
5864 inode = new_simple_dir(dir->i_sb, &location, root);
5866 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5867 btrfs_put_root(sub_root);
5872 down_read(&fs_info->cleanup_work_sem);
5873 if (!sb_rdonly(inode->i_sb))
5874 ret = btrfs_orphan_cleanup(sub_root);
5875 up_read(&fs_info->cleanup_work_sem);
5878 inode = ERR_PTR(ret);
5885 static int btrfs_dentry_delete(const struct dentry *dentry)
5887 struct btrfs_root *root;
5888 struct inode *inode = d_inode(dentry);
5890 if (!inode && !IS_ROOT(dentry))
5891 inode = d_inode(dentry->d_parent);
5894 root = BTRFS_I(inode)->root;
5895 if (btrfs_root_refs(&root->root_item) == 0)
5898 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5904 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5907 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5909 if (inode == ERR_PTR(-ENOENT))
5911 return d_splice_alias(inode, dentry);
5915 * All this infrastructure exists because dir_emit can fault, and we are holding
5916 * the tree lock when doing readdir. For now just allocate a buffer and copy
5917 * our information into that, and then dir_emit from the buffer. This is
5918 * similar to what NFS does, only we don't keep the buffer around in pagecache
5919 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5920 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5923 static int btrfs_opendir(struct inode *inode, struct file *file)
5925 struct btrfs_file_private *private;
5927 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5930 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5931 if (!private->filldir_buf) {
5935 file->private_data = private;
5946 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5949 struct dir_entry *entry = addr;
5950 char *name = (char *)(entry + 1);
5952 ctx->pos = get_unaligned(&entry->offset);
5953 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5954 get_unaligned(&entry->ino),
5955 get_unaligned(&entry->type)))
5957 addr += sizeof(struct dir_entry) +
5958 get_unaligned(&entry->name_len);
5964 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5966 struct inode *inode = file_inode(file);
5967 struct btrfs_root *root = BTRFS_I(inode)->root;
5968 struct btrfs_file_private *private = file->private_data;
5969 struct btrfs_dir_item *di;
5970 struct btrfs_key key;
5971 struct btrfs_key found_key;
5972 struct btrfs_path *path;
5974 struct list_head ins_list;
5975 struct list_head del_list;
5982 struct btrfs_key location;
5984 if (!dir_emit_dots(file, ctx))
5987 path = btrfs_alloc_path();
5991 addr = private->filldir_buf;
5992 path->reada = READA_FORWARD;
5994 INIT_LIST_HEAD(&ins_list);
5995 INIT_LIST_HEAD(&del_list);
5996 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5999 key.type = BTRFS_DIR_INDEX_KEY;
6000 key.offset = ctx->pos;
6001 key.objectid = btrfs_ino(BTRFS_I(inode));
6003 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
6004 struct dir_entry *entry;
6005 struct extent_buffer *leaf = path->nodes[0];
6007 if (found_key.objectid != key.objectid)
6009 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6011 if (found_key.offset < ctx->pos)
6013 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6015 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
6016 name_len = btrfs_dir_name_len(leaf, di);
6017 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6019 btrfs_release_path(path);
6020 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6023 addr = private->filldir_buf;
6030 put_unaligned(name_len, &entry->name_len);
6031 name_ptr = (char *)(entry + 1);
6032 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6034 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6036 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6037 put_unaligned(location.objectid, &entry->ino);
6038 put_unaligned(found_key.offset, &entry->offset);
6040 addr += sizeof(struct dir_entry) + name_len;
6041 total_len += sizeof(struct dir_entry) + name_len;
6043 /* Catch error encountered during iteration */
6047 btrfs_release_path(path);
6049 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6053 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6058 * Stop new entries from being returned after we return the last
6061 * New directory entries are assigned a strictly increasing
6062 * offset. This means that new entries created during readdir
6063 * are *guaranteed* to be seen in the future by that readdir.
6064 * This has broken buggy programs which operate on names as
6065 * they're returned by readdir. Until we re-use freed offsets
6066 * we have this hack to stop new entries from being returned
6067 * under the assumption that they'll never reach this huge
6070 * This is being careful not to overflow 32bit loff_t unless the
6071 * last entry requires it because doing so has broken 32bit apps
6074 if (ctx->pos >= INT_MAX)
6075 ctx->pos = LLONG_MAX;
6082 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6083 btrfs_free_path(path);
6088 * This is somewhat expensive, updating the tree every time the
6089 * inode changes. But, it is most likely to find the inode in cache.
6090 * FIXME, needs more benchmarking...there are no reasons other than performance
6091 * to keep or drop this code.
6093 static int btrfs_dirty_inode(struct inode *inode)
6095 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6096 struct btrfs_root *root = BTRFS_I(inode)->root;
6097 struct btrfs_trans_handle *trans;
6100 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6103 trans = btrfs_join_transaction(root);
6105 return PTR_ERR(trans);
6107 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6108 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6109 /* whoops, lets try again with the full transaction */
6110 btrfs_end_transaction(trans);
6111 trans = btrfs_start_transaction(root, 1);
6113 return PTR_ERR(trans);
6115 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6117 btrfs_end_transaction(trans);
6118 if (BTRFS_I(inode)->delayed_node)
6119 btrfs_balance_delayed_items(fs_info);
6125 * This is a copy of file_update_time. We need this so we can return error on
6126 * ENOSPC for updating the inode in the case of file write and mmap writes.
6128 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6131 struct btrfs_root *root = BTRFS_I(inode)->root;
6132 bool dirty = flags & ~S_VERSION;
6134 if (btrfs_root_readonly(root))
6137 if (flags & S_VERSION)
6138 dirty |= inode_maybe_inc_iversion(inode, dirty);
6139 if (flags & S_CTIME)
6140 inode->i_ctime = *now;
6141 if (flags & S_MTIME)
6142 inode->i_mtime = *now;
6143 if (flags & S_ATIME)
6144 inode->i_atime = *now;
6145 return dirty ? btrfs_dirty_inode(inode) : 0;
6149 * find the highest existing sequence number in a directory
6150 * and then set the in-memory index_cnt variable to reflect
6151 * free sequence numbers
6153 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6155 struct btrfs_root *root = inode->root;
6156 struct btrfs_key key, found_key;
6157 struct btrfs_path *path;
6158 struct extent_buffer *leaf;
6161 key.objectid = btrfs_ino(inode);
6162 key.type = BTRFS_DIR_INDEX_KEY;
6163 key.offset = (u64)-1;
6165 path = btrfs_alloc_path();
6169 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6172 /* FIXME: we should be able to handle this */
6177 if (path->slots[0] == 0) {
6178 inode->index_cnt = BTRFS_DIR_START_INDEX;
6184 leaf = path->nodes[0];
6185 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6187 if (found_key.objectid != btrfs_ino(inode) ||
6188 found_key.type != BTRFS_DIR_INDEX_KEY) {
6189 inode->index_cnt = BTRFS_DIR_START_INDEX;
6193 inode->index_cnt = found_key.offset + 1;
6195 btrfs_free_path(path);
6200 * helper to find a free sequence number in a given directory. This current
6201 * code is very simple, later versions will do smarter things in the btree
6203 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6207 if (dir->index_cnt == (u64)-1) {
6208 ret = btrfs_inode_delayed_dir_index_count(dir);
6210 ret = btrfs_set_inode_index_count(dir);
6216 *index = dir->index_cnt;
6222 static int btrfs_insert_inode_locked(struct inode *inode)
6224 struct btrfs_iget_args args;
6226 args.ino = BTRFS_I(inode)->location.objectid;
6227 args.root = BTRFS_I(inode)->root;
6229 return insert_inode_locked4(inode,
6230 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6231 btrfs_find_actor, &args);
6234 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6235 unsigned int *trans_num_items)
6237 struct inode *dir = args->dir;
6238 struct inode *inode = args->inode;
6241 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6245 /* 1 to add inode item */
6246 *trans_num_items = 1;
6247 /* 1 to add compression property */
6248 if (BTRFS_I(dir)->prop_compress)
6249 (*trans_num_items)++;
6250 /* 1 to add default ACL xattr */
6251 if (args->default_acl)
6252 (*trans_num_items)++;
6253 /* 1 to add access ACL xattr */
6255 (*trans_num_items)++;
6256 #ifdef CONFIG_SECURITY
6257 /* 1 to add LSM xattr */
6258 if (dir->i_security)
6259 (*trans_num_items)++;
6262 /* 1 to add orphan item */
6263 (*trans_num_items)++;
6267 * 1 to add dir index
6268 * 1 to update parent inode item
6270 * No need for 1 unit for the inode ref item because it is
6271 * inserted in a batch together with the inode item at
6272 * btrfs_create_new_inode().
6274 *trans_num_items += 3;
6279 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6281 posix_acl_release(args->acl);
6282 posix_acl_release(args->default_acl);
6286 * Inherit flags from the parent inode.
6288 * Currently only the compression flags and the cow flags are inherited.
6290 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6294 flags = BTRFS_I(dir)->flags;
6296 if (flags & BTRFS_INODE_NOCOMPRESS) {
6297 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6298 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6299 } else if (flags & BTRFS_INODE_COMPRESS) {
6300 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6301 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6304 if (flags & BTRFS_INODE_NODATACOW) {
6305 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6306 if (S_ISREG(inode->i_mode))
6307 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6310 btrfs_sync_inode_flags_to_i_flags(inode);
6313 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6314 struct btrfs_new_inode_args *args)
6316 struct inode *dir = args->dir;
6317 struct inode *inode = args->inode;
6318 const char *name = args->orphan ? NULL : args->dentry->d_name.name;
6319 int name_len = args->orphan ? 0 : args->dentry->d_name.len;
6320 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6321 struct btrfs_root *root;
6322 struct btrfs_inode_item *inode_item;
6323 struct btrfs_key *location;
6324 struct btrfs_path *path;
6326 struct btrfs_inode_ref *ref;
6327 struct btrfs_key key[2];
6329 struct btrfs_item_batch batch;
6333 path = btrfs_alloc_path();
6338 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6339 root = BTRFS_I(inode)->root;
6341 ret = btrfs_get_free_objectid(root, &objectid);
6344 inode->i_ino = objectid;
6348 * O_TMPFILE, set link count to 0, so that after this point, we
6349 * fill in an inode item with the correct link count.
6351 set_nlink(inode, 0);
6353 trace_btrfs_inode_request(dir);
6355 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6359 /* index_cnt is ignored for everything but a dir. */
6360 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6361 BTRFS_I(inode)->generation = trans->transid;
6362 inode->i_generation = BTRFS_I(inode)->generation;
6365 * Subvolumes don't inherit flags from their parent directory.
6366 * Originally this was probably by accident, but we probably can't
6367 * change it now without compatibility issues.
6370 btrfs_inherit_iflags(inode, dir);
6372 if (S_ISREG(inode->i_mode)) {
6373 if (btrfs_test_opt(fs_info, NODATASUM))
6374 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6375 if (btrfs_test_opt(fs_info, NODATACOW))
6376 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6377 BTRFS_INODE_NODATASUM;
6380 location = &BTRFS_I(inode)->location;
6381 location->objectid = objectid;
6382 location->offset = 0;
6383 location->type = BTRFS_INODE_ITEM_KEY;
6385 ret = btrfs_insert_inode_locked(inode);
6388 BTRFS_I(dir)->index_cnt--;
6393 * We could have gotten an inode number from somebody who was fsynced
6394 * and then removed in this same transaction, so let's just set full
6395 * sync since it will be a full sync anyway and this will blow away the
6396 * old info in the log.
6398 btrfs_set_inode_full_sync(BTRFS_I(inode));
6400 key[0].objectid = objectid;
6401 key[0].type = BTRFS_INODE_ITEM_KEY;
6404 sizes[0] = sizeof(struct btrfs_inode_item);
6406 if (!args->orphan) {
6408 * Start new inodes with an inode_ref. This is slightly more
6409 * efficient for small numbers of hard links since they will
6410 * be packed into one item. Extended refs will kick in if we
6411 * add more hard links than can fit in the ref item.
6413 key[1].objectid = objectid;
6414 key[1].type = BTRFS_INODE_REF_KEY;
6416 key[1].offset = objectid;
6417 sizes[1] = 2 + sizeof(*ref);
6419 key[1].offset = btrfs_ino(BTRFS_I(dir));
6420 sizes[1] = name_len + sizeof(*ref);
6424 batch.keys = &key[0];
6425 batch.data_sizes = &sizes[0];
6426 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6427 batch.nr = args->orphan ? 1 : 2;
6428 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6430 btrfs_abort_transaction(trans, ret);
6434 inode->i_mtime = current_time(inode);
6435 inode->i_atime = inode->i_mtime;
6436 inode->i_ctime = inode->i_mtime;
6437 BTRFS_I(inode)->i_otime = inode->i_mtime;
6440 * We're going to fill the inode item now, so at this point the inode
6441 * must be fully initialized.
6444 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6445 struct btrfs_inode_item);
6446 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6447 sizeof(*inode_item));
6448 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6450 if (!args->orphan) {
6451 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6452 struct btrfs_inode_ref);
6453 ptr = (unsigned long)(ref + 1);
6455 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6456 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6457 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6459 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6460 btrfs_set_inode_ref_index(path->nodes[0], ref,
6461 BTRFS_I(inode)->dir_index);
6462 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6466 btrfs_mark_buffer_dirty(path->nodes[0]);
6468 * We don't need the path anymore, plus inheriting properties, adding
6469 * ACLs, security xattrs, orphan item or adding the link, will result in
6470 * allocating yet another path. So just free our path.
6472 btrfs_free_path(path);
6476 struct inode *parent;
6479 * Subvolumes inherit properties from their parent subvolume,
6480 * not the directory they were created in.
6482 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6483 BTRFS_I(dir)->root);
6484 if (IS_ERR(parent)) {
6485 ret = PTR_ERR(parent);
6487 ret = btrfs_inode_inherit_props(trans, inode, parent);
6491 ret = btrfs_inode_inherit_props(trans, inode, dir);
6495 "error inheriting props for ino %llu (root %llu): %d",
6496 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6501 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6504 if (!args->subvol) {
6505 ret = btrfs_init_inode_security(trans, args);
6507 btrfs_abort_transaction(trans, ret);
6512 inode_tree_add(inode);
6514 trace_btrfs_inode_new(inode);
6515 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6517 btrfs_update_root_times(trans, root);
6520 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6522 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6523 name_len, 0, BTRFS_I(inode)->dir_index);
6526 btrfs_abort_transaction(trans, ret);
6534 * discard_new_inode() calls iput(), but the caller owns the reference
6538 discard_new_inode(inode);
6540 btrfs_free_path(path);
6545 * utility function to add 'inode' into 'parent_inode' with
6546 * a give name and a given sequence number.
6547 * if 'add_backref' is true, also insert a backref from the
6548 * inode to the parent directory.
6550 int btrfs_add_link(struct btrfs_trans_handle *trans,
6551 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6552 const char *name, int name_len, int add_backref, u64 index)
6555 struct btrfs_key key;
6556 struct btrfs_root *root = parent_inode->root;
6557 u64 ino = btrfs_ino(inode);
6558 u64 parent_ino = btrfs_ino(parent_inode);
6560 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6561 memcpy(&key, &inode->root->root_key, sizeof(key));
6564 key.type = BTRFS_INODE_ITEM_KEY;
6568 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6569 ret = btrfs_add_root_ref(trans, key.objectid,
6570 root->root_key.objectid, parent_ino,
6571 index, name, name_len);
6572 } else if (add_backref) {
6573 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6577 /* Nothing to clean up yet */
6581 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6582 btrfs_inode_type(&inode->vfs_inode), index);
6583 if (ret == -EEXIST || ret == -EOVERFLOW)
6586 btrfs_abort_transaction(trans, ret);
6590 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6592 inode_inc_iversion(&parent_inode->vfs_inode);
6594 * If we are replaying a log tree, we do not want to update the mtime
6595 * and ctime of the parent directory with the current time, since the
6596 * log replay procedure is responsible for setting them to their correct
6597 * values (the ones it had when the fsync was done).
6599 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6600 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6602 parent_inode->vfs_inode.i_mtime = now;
6603 parent_inode->vfs_inode.i_ctime = now;
6605 ret = btrfs_update_inode(trans, root, parent_inode);
6607 btrfs_abort_transaction(trans, ret);
6611 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6614 err = btrfs_del_root_ref(trans, key.objectid,
6615 root->root_key.objectid, parent_ino,
6616 &local_index, name, name_len);
6618 btrfs_abort_transaction(trans, err);
6619 } else if (add_backref) {
6623 err = btrfs_del_inode_ref(trans, root, name, name_len,
6624 ino, parent_ino, &local_index);
6626 btrfs_abort_transaction(trans, err);
6629 /* Return the original error code */
6633 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6634 struct inode *inode)
6636 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6637 struct btrfs_root *root = BTRFS_I(dir)->root;
6638 struct btrfs_new_inode_args new_inode_args = {
6643 unsigned int trans_num_items;
6644 struct btrfs_trans_handle *trans;
6647 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6651 trans = btrfs_start_transaction(root, trans_num_items);
6652 if (IS_ERR(trans)) {
6653 err = PTR_ERR(trans);
6654 goto out_new_inode_args;
6657 err = btrfs_create_new_inode(trans, &new_inode_args);
6659 d_instantiate_new(dentry, inode);
6661 btrfs_end_transaction(trans);
6662 btrfs_btree_balance_dirty(fs_info);
6664 btrfs_new_inode_args_destroy(&new_inode_args);
6671 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6672 struct dentry *dentry, umode_t mode, dev_t rdev)
6674 struct inode *inode;
6676 inode = new_inode(dir->i_sb);
6679 inode_init_owner(mnt_userns, inode, dir, mode);
6680 inode->i_op = &btrfs_special_inode_operations;
6681 init_special_inode(inode, inode->i_mode, rdev);
6682 return btrfs_create_common(dir, dentry, inode);
6685 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6686 struct dentry *dentry, umode_t mode, bool excl)
6688 struct inode *inode;
6690 inode = new_inode(dir->i_sb);
6693 inode_init_owner(mnt_userns, inode, dir, mode);
6694 inode->i_fop = &btrfs_file_operations;
6695 inode->i_op = &btrfs_file_inode_operations;
6696 inode->i_mapping->a_ops = &btrfs_aops;
6697 return btrfs_create_common(dir, dentry, inode);
6700 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6701 struct dentry *dentry)
6703 struct btrfs_trans_handle *trans = NULL;
6704 struct btrfs_root *root = BTRFS_I(dir)->root;
6705 struct inode *inode = d_inode(old_dentry);
6706 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6711 /* do not allow sys_link's with other subvols of the same device */
6712 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6715 if (inode->i_nlink >= BTRFS_LINK_MAX)
6718 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6723 * 2 items for inode and inode ref
6724 * 2 items for dir items
6725 * 1 item for parent inode
6726 * 1 item for orphan item deletion if O_TMPFILE
6728 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6729 if (IS_ERR(trans)) {
6730 err = PTR_ERR(trans);
6735 /* There are several dir indexes for this inode, clear the cache. */
6736 BTRFS_I(inode)->dir_index = 0ULL;
6738 inode_inc_iversion(inode);
6739 inode->i_ctime = current_time(inode);
6741 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6743 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6744 dentry->d_name.name, dentry->d_name.len, 1, index);
6749 struct dentry *parent = dentry->d_parent;
6751 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6754 if (inode->i_nlink == 1) {
6756 * If new hard link count is 1, it's a file created
6757 * with open(2) O_TMPFILE flag.
6759 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6763 d_instantiate(dentry, inode);
6764 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6769 btrfs_end_transaction(trans);
6771 inode_dec_link_count(inode);
6774 btrfs_btree_balance_dirty(fs_info);
6778 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6779 struct dentry *dentry, umode_t mode)
6781 struct inode *inode;
6783 inode = new_inode(dir->i_sb);
6786 inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode);
6787 inode->i_op = &btrfs_dir_inode_operations;
6788 inode->i_fop = &btrfs_dir_file_operations;
6789 return btrfs_create_common(dir, dentry, inode);
6792 static noinline int uncompress_inline(struct btrfs_path *path,
6794 size_t pg_offset, u64 extent_offset,
6795 struct btrfs_file_extent_item *item)
6798 struct extent_buffer *leaf = path->nodes[0];
6801 unsigned long inline_size;
6805 WARN_ON(pg_offset != 0);
6806 compress_type = btrfs_file_extent_compression(leaf, item);
6807 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6808 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6809 tmp = kmalloc(inline_size, GFP_NOFS);
6812 ptr = btrfs_file_extent_inline_start(item);
6814 read_extent_buffer(leaf, tmp, ptr, inline_size);
6816 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6817 ret = btrfs_decompress(compress_type, tmp, page,
6818 extent_offset, inline_size, max_size);
6821 * decompression code contains a memset to fill in any space between the end
6822 * of the uncompressed data and the end of max_size in case the decompressed
6823 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6824 * the end of an inline extent and the beginning of the next block, so we
6825 * cover that region here.
6828 if (max_size + pg_offset < PAGE_SIZE)
6829 memzero_page(page, pg_offset + max_size,
6830 PAGE_SIZE - max_size - pg_offset);
6836 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6837 * @inode: file to search in
6838 * @page: page to read extent data into if the extent is inline
6839 * @pg_offset: offset into @page to copy to
6840 * @start: file offset
6841 * @len: length of range starting at @start
6843 * This returns the first &struct extent_map which overlaps with the given
6844 * range, reading it from the B-tree and caching it if necessary. Note that
6845 * there may be more extents which overlap the given range after the returned
6848 * If @page is not NULL and the extent is inline, this also reads the extent
6849 * data directly into the page and marks the extent up to date in the io_tree.
6851 * Return: ERR_PTR on error, non-NULL extent_map on success.
6853 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6854 struct page *page, size_t pg_offset,
6857 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6859 u64 extent_start = 0;
6861 u64 objectid = btrfs_ino(inode);
6862 int extent_type = -1;
6863 struct btrfs_path *path = NULL;
6864 struct btrfs_root *root = inode->root;
6865 struct btrfs_file_extent_item *item;
6866 struct extent_buffer *leaf;
6867 struct btrfs_key found_key;
6868 struct extent_map *em = NULL;
6869 struct extent_map_tree *em_tree = &inode->extent_tree;
6870 struct extent_io_tree *io_tree = &inode->io_tree;
6872 read_lock(&em_tree->lock);
6873 em = lookup_extent_mapping(em_tree, start, len);
6874 read_unlock(&em_tree->lock);
6877 if (em->start > start || em->start + em->len <= start)
6878 free_extent_map(em);
6879 else if (em->block_start == EXTENT_MAP_INLINE && page)
6880 free_extent_map(em);
6884 em = alloc_extent_map();
6889 em->start = EXTENT_MAP_HOLE;
6890 em->orig_start = EXTENT_MAP_HOLE;
6892 em->block_len = (u64)-1;
6894 path = btrfs_alloc_path();
6900 /* Chances are we'll be called again, so go ahead and do readahead */
6901 path->reada = READA_FORWARD;
6904 * The same explanation in load_free_space_cache applies here as well,
6905 * we only read when we're loading the free space cache, and at that
6906 * point the commit_root has everything we need.
6908 if (btrfs_is_free_space_inode(inode)) {
6909 path->search_commit_root = 1;
6910 path->skip_locking = 1;
6913 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6916 } else if (ret > 0) {
6917 if (path->slots[0] == 0)
6923 leaf = path->nodes[0];
6924 item = btrfs_item_ptr(leaf, path->slots[0],
6925 struct btrfs_file_extent_item);
6926 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6927 if (found_key.objectid != objectid ||
6928 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6930 * If we backup past the first extent we want to move forward
6931 * and see if there is an extent in front of us, otherwise we'll
6932 * say there is a hole for our whole search range which can
6939 extent_type = btrfs_file_extent_type(leaf, item);
6940 extent_start = found_key.offset;
6941 extent_end = btrfs_file_extent_end(path);
6942 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6943 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6944 /* Only regular file could have regular/prealloc extent */
6945 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6948 "regular/prealloc extent found for non-regular inode %llu",
6952 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6954 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6955 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6960 if (start >= extent_end) {
6962 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6963 ret = btrfs_next_leaf(root, path);
6969 leaf = path->nodes[0];
6971 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6972 if (found_key.objectid != objectid ||
6973 found_key.type != BTRFS_EXTENT_DATA_KEY)
6975 if (start + len <= found_key.offset)
6977 if (start > found_key.offset)
6980 /* New extent overlaps with existing one */
6982 em->orig_start = start;
6983 em->len = found_key.offset - start;
6984 em->block_start = EXTENT_MAP_HOLE;
6988 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6990 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6991 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6993 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6997 size_t extent_offset;
7003 size = btrfs_file_extent_ram_bytes(leaf, item);
7004 extent_offset = page_offset(page) + pg_offset - extent_start;
7005 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7006 size - extent_offset);
7007 em->start = extent_start + extent_offset;
7008 em->len = ALIGN(copy_size, fs_info->sectorsize);
7009 em->orig_block_len = em->len;
7010 em->orig_start = em->start;
7011 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7013 if (!PageUptodate(page)) {
7014 if (btrfs_file_extent_compression(leaf, item) !=
7015 BTRFS_COMPRESS_NONE) {
7016 ret = uncompress_inline(path, page, pg_offset,
7017 extent_offset, item);
7021 map = kmap_local_page(page);
7022 read_extent_buffer(leaf, map + pg_offset, ptr,
7024 if (pg_offset + copy_size < PAGE_SIZE) {
7025 memset(map + pg_offset + copy_size, 0,
7026 PAGE_SIZE - pg_offset -
7031 flush_dcache_page(page);
7033 set_extent_uptodate(io_tree, em->start,
7034 extent_map_end(em) - 1, NULL, GFP_NOFS);
7039 em->orig_start = start;
7041 em->block_start = EXTENT_MAP_HOLE;
7044 btrfs_release_path(path);
7045 if (em->start > start || extent_map_end(em) <= start) {
7047 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7048 em->start, em->len, start, len);
7053 write_lock(&em_tree->lock);
7054 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7055 write_unlock(&em_tree->lock);
7057 btrfs_free_path(path);
7059 trace_btrfs_get_extent(root, inode, em);
7062 free_extent_map(em);
7063 return ERR_PTR(ret);
7068 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7071 struct extent_map *em;
7072 struct extent_map *hole_em = NULL;
7073 u64 delalloc_start = start;
7079 em = btrfs_get_extent(inode, NULL, 0, start, len);
7083 * If our em maps to:
7085 * - a pre-alloc extent,
7086 * there might actually be delalloc bytes behind it.
7088 if (em->block_start != EXTENT_MAP_HOLE &&
7089 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7094 /* check to see if we've wrapped (len == -1 or similar) */
7103 /* ok, we didn't find anything, lets look for delalloc */
7104 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7105 end, len, EXTENT_DELALLOC, 1);
7106 delalloc_end = delalloc_start + delalloc_len;
7107 if (delalloc_end < delalloc_start)
7108 delalloc_end = (u64)-1;
7111 * We didn't find anything useful, return the original results from
7114 if (delalloc_start > end || delalloc_end <= start) {
7121 * Adjust the delalloc_start to make sure it doesn't go backwards from
7122 * the start they passed in
7124 delalloc_start = max(start, delalloc_start);
7125 delalloc_len = delalloc_end - delalloc_start;
7127 if (delalloc_len > 0) {
7130 const u64 hole_end = extent_map_end(hole_em);
7132 em = alloc_extent_map();
7140 * When btrfs_get_extent can't find anything it returns one
7143 * Make sure what it found really fits our range, and adjust to
7144 * make sure it is based on the start from the caller
7146 if (hole_end <= start || hole_em->start > end) {
7147 free_extent_map(hole_em);
7150 hole_start = max(hole_em->start, start);
7151 hole_len = hole_end - hole_start;
7154 if (hole_em && delalloc_start > hole_start) {
7156 * Our hole starts before our delalloc, so we have to
7157 * return just the parts of the hole that go until the
7160 em->len = min(hole_len, delalloc_start - hole_start);
7161 em->start = hole_start;
7162 em->orig_start = hole_start;
7164 * Don't adjust block start at all, it is fixed at
7167 em->block_start = hole_em->block_start;
7168 em->block_len = hole_len;
7169 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7170 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7173 * Hole is out of passed range or it starts after
7176 em->start = delalloc_start;
7177 em->len = delalloc_len;
7178 em->orig_start = delalloc_start;
7179 em->block_start = EXTENT_MAP_DELALLOC;
7180 em->block_len = delalloc_len;
7187 free_extent_map(hole_em);
7189 free_extent_map(em);
7190 return ERR_PTR(err);
7195 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7198 const u64 orig_start,
7199 const u64 block_start,
7200 const u64 block_len,
7201 const u64 orig_block_len,
7202 const u64 ram_bytes,
7205 struct extent_map *em = NULL;
7208 if (type != BTRFS_ORDERED_NOCOW) {
7209 em = create_io_em(inode, start, len, orig_start, block_start,
7210 block_len, orig_block_len, ram_bytes,
7211 BTRFS_COMPRESS_NONE, /* compress_type */
7216 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7219 (1 << BTRFS_ORDERED_DIRECT),
7220 BTRFS_COMPRESS_NONE);
7223 free_extent_map(em);
7224 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7233 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7236 struct btrfs_root *root = inode->root;
7237 struct btrfs_fs_info *fs_info = root->fs_info;
7238 struct extent_map *em;
7239 struct btrfs_key ins;
7243 alloc_hint = get_extent_allocation_hint(inode, start, len);
7244 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7245 0, alloc_hint, &ins, 1, 1);
7247 return ERR_PTR(ret);
7249 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7250 ins.objectid, ins.offset, ins.offset,
7251 ins.offset, BTRFS_ORDERED_REGULAR);
7252 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7254 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7260 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7262 struct btrfs_block_group *block_group;
7263 bool readonly = false;
7265 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7266 if (!block_group || block_group->ro)
7269 btrfs_put_block_group(block_group);
7274 * Check if we can do nocow write into the range [@offset, @offset + @len)
7276 * @offset: File offset
7277 * @len: The length to write, will be updated to the nocow writeable
7279 * @orig_start: (optional) Return the original file offset of the file extent
7280 * @orig_len: (optional) Return the original on-disk length of the file extent
7281 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7282 * @strict: if true, omit optimizations that might force us into unnecessary
7283 * cow. e.g., don't trust generation number.
7286 * >0 and update @len if we can do nocow write
7287 * 0 if we can't do nocow write
7288 * <0 if error happened
7290 * NOTE: This only checks the file extents, caller is responsible to wait for
7291 * any ordered extents.
7293 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7294 u64 *orig_start, u64 *orig_block_len,
7295 u64 *ram_bytes, bool strict)
7297 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7298 struct can_nocow_file_extent_args nocow_args = { 0 };
7299 struct btrfs_path *path;
7301 struct extent_buffer *leaf;
7302 struct btrfs_root *root = BTRFS_I(inode)->root;
7303 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7304 struct btrfs_file_extent_item *fi;
7305 struct btrfs_key key;
7308 path = btrfs_alloc_path();
7312 ret = btrfs_lookup_file_extent(NULL, root, path,
7313 btrfs_ino(BTRFS_I(inode)), offset, 0);
7318 if (path->slots[0] == 0) {
7319 /* can't find the item, must cow */
7326 leaf = path->nodes[0];
7327 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7328 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7329 key.type != BTRFS_EXTENT_DATA_KEY) {
7330 /* not our file or wrong item type, must cow */
7334 if (key.offset > offset) {
7335 /* Wrong offset, must cow */
7339 if (btrfs_file_extent_end(path) <= offset)
7342 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7343 found_type = btrfs_file_extent_type(leaf, fi);
7345 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7347 nocow_args.start = offset;
7348 nocow_args.end = offset + *len - 1;
7349 nocow_args.strict = strict;
7350 nocow_args.free_path = true;
7352 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7353 /* can_nocow_file_extent() has freed the path. */
7357 /* Treat errors as not being able to NOCOW. */
7363 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7366 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7367 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7370 range_end = round_up(offset + nocow_args.num_bytes,
7371 root->fs_info->sectorsize) - 1;
7372 ret = test_range_bit(io_tree, offset, range_end,
7373 EXTENT_DELALLOC, 0, NULL);
7381 *orig_start = key.offset - nocow_args.extent_offset;
7383 *orig_block_len = nocow_args.disk_num_bytes;
7385 *len = nocow_args.num_bytes;
7388 btrfs_free_path(path);
7392 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7393 struct extent_state **cached_state,
7394 unsigned int iomap_flags)
7396 const bool writing = (iomap_flags & IOMAP_WRITE);
7397 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7398 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7399 struct btrfs_ordered_extent *ordered;
7404 if (!try_lock_extent(io_tree, lockstart, lockend))
7407 lock_extent_bits(io_tree, lockstart, lockend, cached_state);
7410 * We're concerned with the entire range that we're going to be
7411 * doing DIO to, so we need to make sure there's no ordered
7412 * extents in this range.
7414 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7415 lockend - lockstart + 1);
7418 * We need to make sure there are no buffered pages in this
7419 * range either, we could have raced between the invalidate in
7420 * generic_file_direct_write and locking the extent. The
7421 * invalidate needs to happen so that reads after a write do not
7425 (!writing || !filemap_range_has_page(inode->i_mapping,
7426 lockstart, lockend)))
7429 unlock_extent_cached(io_tree, lockstart, lockend, cached_state);
7433 btrfs_put_ordered_extent(ordered);
7438 * If we are doing a DIO read and the ordered extent we
7439 * found is for a buffered write, we can not wait for it
7440 * to complete and retry, because if we do so we can
7441 * deadlock with concurrent buffered writes on page
7442 * locks. This happens only if our DIO read covers more
7443 * than one extent map, if at this point has already
7444 * created an ordered extent for a previous extent map
7445 * and locked its range in the inode's io tree, and a
7446 * concurrent write against that previous extent map's
7447 * range and this range started (we unlock the ranges
7448 * in the io tree only when the bios complete and
7449 * buffered writes always lock pages before attempting
7450 * to lock range in the io tree).
7453 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7454 btrfs_start_ordered_extent(ordered, 1);
7456 ret = nowait ? -EAGAIN : -ENOTBLK;
7457 btrfs_put_ordered_extent(ordered);
7460 * We could trigger writeback for this range (and wait
7461 * for it to complete) and then invalidate the pages for
7462 * this range (through invalidate_inode_pages2_range()),
7463 * but that can lead us to a deadlock with a concurrent
7464 * call to readahead (a buffered read or a defrag call
7465 * triggered a readahead) on a page lock due to an
7466 * ordered dio extent we created before but did not have
7467 * yet a corresponding bio submitted (whence it can not
7468 * complete), which makes readahead wait for that
7469 * ordered extent to complete while holding a lock on
7472 ret = nowait ? -EAGAIN : -ENOTBLK;
7484 /* The callers of this must take lock_extent() */
7485 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7486 u64 len, u64 orig_start, u64 block_start,
7487 u64 block_len, u64 orig_block_len,
7488 u64 ram_bytes, int compress_type,
7491 struct extent_map_tree *em_tree;
7492 struct extent_map *em;
7495 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7496 type == BTRFS_ORDERED_COMPRESSED ||
7497 type == BTRFS_ORDERED_NOCOW ||
7498 type == BTRFS_ORDERED_REGULAR);
7500 em_tree = &inode->extent_tree;
7501 em = alloc_extent_map();
7503 return ERR_PTR(-ENOMEM);
7506 em->orig_start = orig_start;
7508 em->block_len = block_len;
7509 em->block_start = block_start;
7510 em->orig_block_len = orig_block_len;
7511 em->ram_bytes = ram_bytes;
7512 em->generation = -1;
7513 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7514 if (type == BTRFS_ORDERED_PREALLOC) {
7515 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7516 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7517 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7518 em->compress_type = compress_type;
7522 btrfs_drop_extent_cache(inode, em->start,
7523 em->start + em->len - 1, 0);
7524 write_lock(&em_tree->lock);
7525 ret = add_extent_mapping(em_tree, em, 1);
7526 write_unlock(&em_tree->lock);
7528 * The caller has taken lock_extent(), who could race with us
7531 } while (ret == -EEXIST);
7534 free_extent_map(em);
7535 return ERR_PTR(ret);
7538 /* em got 2 refs now, callers needs to do free_extent_map once. */
7543 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7544 struct inode *inode,
7545 struct btrfs_dio_data *dio_data,
7547 unsigned int iomap_flags)
7549 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7550 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7551 struct extent_map *em = *map;
7553 u64 block_start, orig_start, orig_block_len, ram_bytes;
7554 struct btrfs_block_group *bg;
7555 bool can_nocow = false;
7556 bool space_reserved = false;
7561 * We don't allocate a new extent in the following cases
7563 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7565 * 2) The extent is marked as PREALLOC. We're good to go here and can
7566 * just use the extent.
7569 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7570 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7571 em->block_start != EXTENT_MAP_HOLE)) {
7572 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7573 type = BTRFS_ORDERED_PREALLOC;
7575 type = BTRFS_ORDERED_NOCOW;
7576 len = min(len, em->len - (start - em->start));
7577 block_start = em->block_start + (start - em->start);
7579 if (can_nocow_extent(inode, start, &len, &orig_start,
7580 &orig_block_len, &ram_bytes, false) == 1) {
7581 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7589 struct extent_map *em2;
7591 /* We can NOCOW, so only need to reserve metadata space. */
7592 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7595 /* Our caller expects us to free the input extent map. */
7596 free_extent_map(em);
7598 btrfs_dec_nocow_writers(bg);
7599 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7603 space_reserved = true;
7605 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7606 orig_start, block_start,
7607 len, orig_block_len,
7609 btrfs_dec_nocow_writers(bg);
7610 if (type == BTRFS_ORDERED_PREALLOC) {
7611 free_extent_map(em);
7621 dio_data->nocow_done = true;
7623 /* Our caller expects us to free the input extent map. */
7624 free_extent_map(em);
7631 * If we could not allocate data space before locking the file
7632 * range and we can't do a NOCOW write, then we have to fail.
7634 if (!dio_data->data_space_reserved)
7638 * We have to COW and we have already reserved data space before,
7639 * so now we reserve only metadata.
7641 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7645 space_reserved = true;
7647 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7653 len = min(len, em->len - (start - em->start));
7655 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7656 prev_len - len, true);
7660 * We have created our ordered extent, so we can now release our reservation
7661 * for an outstanding extent.
7663 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7666 * Need to update the i_size under the extent lock so buffered
7667 * readers will get the updated i_size when we unlock.
7669 if (start + len > i_size_read(inode))
7670 i_size_write(inode, start + len);
7672 if (ret && space_reserved) {
7673 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7674 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7679 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7680 loff_t length, unsigned int flags, struct iomap *iomap,
7681 struct iomap *srcmap)
7683 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7684 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7685 struct extent_map *em;
7686 struct extent_state *cached_state = NULL;
7687 struct btrfs_dio_data *dio_data = iter->private;
7688 u64 lockstart, lockend;
7689 const bool write = !!(flags & IOMAP_WRITE);
7692 const u64 data_alloc_len = length;
7693 bool unlock_extents = false;
7696 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7697 * we're NOWAIT we may submit a bio for a partial range and return
7698 * EIOCBQUEUED, which would result in an errant short read.
7700 * The best way to handle this would be to allow for partial completions
7701 * of iocb's, so we could submit the partial bio, return and fault in
7702 * the rest of the pages, and then submit the io for the rest of the
7703 * range. However we don't have that currently, so simply return
7704 * -EAGAIN at this point so that the normal path is used.
7706 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7710 * Cap the size of reads to that usually seen in buffered I/O as we need
7711 * to allocate a contiguous array for the checksums.
7714 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7717 lockend = start + len - 1;
7720 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7721 * enough if we've written compressed pages to this area, so we need to
7722 * flush the dirty pages again to make absolutely sure that any
7723 * outstanding dirty pages are on disk - the first flush only starts
7724 * compression on the data, while keeping the pages locked, so by the
7725 * time the second flush returns we know bios for the compressed pages
7726 * were submitted and finished, and the pages no longer under writeback.
7728 * If we have a NOWAIT request and we have any pages in the range that
7729 * are locked, likely due to compression still in progress, we don't want
7730 * to block on page locks. We also don't want to block on pages marked as
7731 * dirty or under writeback (same as for the non-compression case).
7732 * iomap_dio_rw() did the same check, but after that and before we got
7733 * here, mmap'ed writes may have happened or buffered reads started
7734 * (readpage() and readahead(), which lock pages), as we haven't locked
7735 * the file range yet.
7737 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7738 &BTRFS_I(inode)->runtime_flags)) {
7739 if (flags & IOMAP_NOWAIT) {
7740 if (filemap_range_needs_writeback(inode->i_mapping,
7741 lockstart, lockend))
7744 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7745 start + length - 1);
7751 memset(dio_data, 0, sizeof(*dio_data));
7754 * We always try to allocate data space and must do it before locking
7755 * the file range, to avoid deadlocks with concurrent writes to the same
7756 * range if the range has several extents and the writes don't expand the
7757 * current i_size (the inode lock is taken in shared mode). If we fail to
7758 * allocate data space here we continue and later, after locking the
7759 * file range, we fail with ENOSPC only if we figure out we can not do a
7762 if (write && !(flags & IOMAP_NOWAIT)) {
7763 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7764 &dio_data->data_reserved,
7765 start, data_alloc_len);
7767 dio_data->data_space_reserved = true;
7768 else if (ret && !(BTRFS_I(inode)->flags &
7769 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7774 * If this errors out it's because we couldn't invalidate pagecache for
7775 * this range and we need to fallback to buffered IO, or we are doing a
7776 * NOWAIT read/write and we need to block.
7778 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7782 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7789 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7790 * io. INLINE is special, and we could probably kludge it in here, but
7791 * it's still buffered so for safety lets just fall back to the generic
7794 * For COMPRESSED we _have_ to read the entire extent in so we can
7795 * decompress it, so there will be buffering required no matter what we
7796 * do, so go ahead and fallback to buffered.
7798 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7799 * to buffered IO. Don't blame me, this is the price we pay for using
7802 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7803 em->block_start == EXTENT_MAP_INLINE) {
7804 free_extent_map(em);
7806 * If we are in a NOWAIT context, return -EAGAIN in order to
7807 * fallback to buffered IO. This is not only because we can
7808 * block with buffered IO (no support for NOWAIT semantics at
7809 * the moment) but also to avoid returning short reads to user
7810 * space - this happens if we were able to read some data from
7811 * previous non-compressed extents and then when we fallback to
7812 * buffered IO, at btrfs_file_read_iter() by calling
7813 * filemap_read(), we fail to fault in pages for the read buffer,
7814 * in which case filemap_read() returns a short read (the number
7815 * of bytes previously read is > 0, so it does not return -EFAULT).
7817 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7821 len = min(len, em->len - (start - em->start));
7824 * If we have a NOWAIT request and the range contains multiple extents
7825 * (or a mix of extents and holes), then we return -EAGAIN to make the
7826 * caller fallback to a context where it can do a blocking (without
7827 * NOWAIT) request. This way we avoid doing partial IO and returning
7828 * success to the caller, which is not optimal for writes and for reads
7829 * it can result in unexpected behaviour for an application.
7831 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7832 * iomap_dio_rw(), we can end up returning less data then what the caller
7833 * asked for, resulting in an unexpected, and incorrect, short read.
7834 * That is, the caller asked to read N bytes and we return less than that,
7835 * which is wrong unless we are crossing EOF. This happens if we get a
7836 * page fault error when trying to fault in pages for the buffer that is
7837 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7838 * have previously submitted bios for other extents in the range, in
7839 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7840 * those bios have completed by the time we get the page fault error,
7841 * which we return back to our caller - we should only return EIOCBQUEUED
7842 * after we have submitted bios for all the extents in the range.
7844 if ((flags & IOMAP_NOWAIT) && len < length) {
7845 free_extent_map(em);
7851 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7855 unlock_extents = true;
7856 /* Recalc len in case the new em is smaller than requested */
7857 len = min(len, em->len - (start - em->start));
7858 if (dio_data->data_space_reserved) {
7860 u64 release_len = 0;
7862 if (dio_data->nocow_done) {
7863 release_offset = start;
7864 release_len = data_alloc_len;
7865 } else if (len < data_alloc_len) {
7866 release_offset = start + len;
7867 release_len = data_alloc_len - len;
7870 if (release_len > 0)
7871 btrfs_free_reserved_data_space(BTRFS_I(inode),
7872 dio_data->data_reserved,
7878 * We need to unlock only the end area that we aren't using.
7879 * The rest is going to be unlocked by the endio routine.
7881 lockstart = start + len;
7882 if (lockstart < lockend)
7883 unlock_extents = true;
7887 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7888 lockstart, lockend, &cached_state);
7890 free_extent_state(cached_state);
7893 * Translate extent map information to iomap.
7894 * We trim the extents (and move the addr) even though iomap code does
7895 * that, since we have locked only the parts we are performing I/O in.
7897 if ((em->block_start == EXTENT_MAP_HOLE) ||
7898 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7899 iomap->addr = IOMAP_NULL_ADDR;
7900 iomap->type = IOMAP_HOLE;
7902 iomap->addr = em->block_start + (start - em->start);
7903 iomap->type = IOMAP_MAPPED;
7905 iomap->offset = start;
7906 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7907 iomap->length = len;
7909 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7910 iomap->flags |= IOMAP_F_ZONE_APPEND;
7912 free_extent_map(em);
7917 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7920 if (dio_data->data_space_reserved) {
7921 btrfs_free_reserved_data_space(BTRFS_I(inode),
7922 dio_data->data_reserved,
7923 start, data_alloc_len);
7924 extent_changeset_free(dio_data->data_reserved);
7930 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7931 ssize_t written, unsigned int flags, struct iomap *iomap)
7933 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7934 struct btrfs_dio_data *dio_data = iter->private;
7935 size_t submitted = dio_data->submitted;
7936 const bool write = !!(flags & IOMAP_WRITE);
7939 if (!write && (iomap->type == IOMAP_HOLE)) {
7940 /* If reading from a hole, unlock and return */
7941 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7945 if (submitted < length) {
7947 length -= submitted;
7949 btrfs_mark_ordered_io_finished(BTRFS_I(inode), NULL,
7950 pos, length, false);
7952 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7958 extent_changeset_free(dio_data->data_reserved);
7962 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7965 * This implies a barrier so that stores to dio_bio->bi_status before
7966 * this and loads of dio_bio->bi_status after this are fully ordered.
7968 if (!refcount_dec_and_test(&dip->refs))
7971 if (btrfs_op(&dip->bio) == BTRFS_MAP_WRITE) {
7972 btrfs_mark_ordered_io_finished(BTRFS_I(dip->inode), NULL,
7973 dip->file_offset, dip->bytes,
7974 !dip->bio.bi_status);
7976 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7978 dip->file_offset + dip->bytes - 1);
7982 bio_endio(&dip->bio);
7985 static void submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7987 enum btrfs_compression_type compress_type)
7989 struct btrfs_dio_private *dip = bio->bi_private;
7990 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7992 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7994 refcount_inc(&dip->refs);
7995 btrfs_submit_bio(fs_info, bio, mirror_num);
7998 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7999 struct btrfs_bio *bbio,
8000 const bool uptodate)
8002 struct inode *inode = dip->inode;
8003 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
8004 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8005 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8006 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8007 blk_status_t err = BLK_STS_OK;
8008 struct bvec_iter iter;
8012 btrfs_bio_for_each_sector(fs_info, bv, bbio, iter, offset) {
8013 u64 start = bbio->file_offset + offset;
8016 (!csum || !btrfs_check_data_csum(inode, bbio, offset, bv.bv_page,
8018 clean_io_failure(fs_info, failure_tree, io_tree, start,
8019 bv.bv_page, btrfs_ino(BTRFS_I(inode)),
8024 ret = btrfs_repair_one_sector(inode, bbio, offset,
8025 bv.bv_page, bv.bv_offset,
8026 submit_dio_repair_bio);
8028 err = errno_to_blk_status(ret);
8035 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8037 u64 dio_file_offset)
8039 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
8042 static void btrfs_end_dio_bio(struct bio *bio)
8044 struct btrfs_dio_private *dip = bio->bi_private;
8045 struct btrfs_bio *bbio = btrfs_bio(bio);
8046 blk_status_t err = bio->bi_status;
8049 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8050 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8051 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8052 bio->bi_opf, bio->bi_iter.bi_sector,
8053 bio->bi_iter.bi_size, err);
8055 if (bio_op(bio) == REQ_OP_READ)
8056 err = btrfs_check_read_dio_bio(dip, bbio, !err);
8059 dip->bio.bi_status = err;
8061 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
8064 btrfs_dio_private_put(dip);
8067 static void btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8068 u64 file_offset, int async_submit)
8070 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8071 struct btrfs_dio_private *dip = bio->bi_private;
8074 /* Save the original iter for read repair */
8075 if (btrfs_op(bio) == BTRFS_MAP_READ)
8076 btrfs_bio(bio)->iter = bio->bi_iter;
8078 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8081 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
8082 /* Check btrfs_submit_data_write_bio() for async submit rules */
8083 if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers) &&
8084 btrfs_wq_submit_bio(inode, bio, 0, file_offset,
8085 btrfs_submit_bio_start_direct_io))
8089 * If we aren't doing async submit, calculate the csum of the
8092 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
8094 bio->bi_status = ret;
8099 btrfs_bio(bio)->csum = btrfs_csum_ptr(fs_info, dip->csums,
8100 file_offset - dip->file_offset);
8103 btrfs_submit_bio(fs_info, bio, 0);
8106 static void btrfs_submit_direct(const struct iomap_iter *iter,
8107 struct bio *dio_bio, loff_t file_offset)
8109 struct btrfs_dio_private *dip =
8110 container_of(dio_bio, struct btrfs_dio_private, bio);
8111 struct inode *inode = iter->inode;
8112 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8113 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8114 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8115 BTRFS_BLOCK_GROUP_RAID56_MASK);
8118 int async_submit = 0;
8120 u64 clone_offset = 0;
8124 blk_status_t status;
8125 struct btrfs_io_geometry geom;
8126 struct btrfs_dio_data *dio_data = iter->private;
8127 struct extent_map *em = NULL;
8130 dip->file_offset = file_offset;
8131 dip->bytes = dio_bio->bi_iter.bi_size;
8132 refcount_set(&dip->refs, 1);
8135 if (!write && !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
8136 unsigned int nr_sectors =
8137 (dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits);
8140 * Load the csums up front to reduce csum tree searches and
8141 * contention when submitting bios.
8143 status = BLK_STS_RESOURCE;
8144 dip->csums = kcalloc(nr_sectors, fs_info->csum_size, GFP_NOFS);
8148 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8149 if (status != BLK_STS_OK)
8153 start_sector = dio_bio->bi_iter.bi_sector;
8154 submit_len = dio_bio->bi_iter.bi_size;
8157 logical = start_sector << 9;
8158 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8160 status = errno_to_blk_status(PTR_ERR(em));
8164 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8167 status = errno_to_blk_status(ret);
8171 clone_len = min(submit_len, geom.len);
8172 ASSERT(clone_len <= UINT_MAX);
8175 * This will never fail as it's passing GPF_NOFS and
8176 * the allocation is backed by btrfs_bioset.
8178 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8179 bio->bi_private = dip;
8180 bio->bi_end_io = btrfs_end_dio_bio;
8181 btrfs_bio(bio)->file_offset = file_offset;
8183 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8184 status = extract_ordered_extent(BTRFS_I(inode), bio,
8192 ASSERT(submit_len >= clone_len);
8193 submit_len -= clone_len;
8196 * Increase the count before we submit the bio so we know
8197 * the end IO handler won't happen before we increase the
8198 * count. Otherwise, the dip might get freed before we're
8199 * done setting it up.
8201 * We transfer the initial reference to the last bio, so we
8202 * don't need to increment the reference count for the last one.
8204 if (submit_len > 0) {
8205 refcount_inc(&dip->refs);
8207 * If we are submitting more than one bio, submit them
8208 * all asynchronously. The exception is RAID 5 or 6, as
8209 * asynchronous checksums make it difficult to collect
8210 * full stripe writes.
8216 btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8218 dio_data->submitted += clone_len;
8219 clone_offset += clone_len;
8220 start_sector += clone_len >> 9;
8221 file_offset += clone_len;
8223 free_extent_map(em);
8224 } while (submit_len > 0);
8228 free_extent_map(em);
8230 dio_bio->bi_status = status;
8231 btrfs_dio_private_put(dip);
8234 static const struct iomap_ops btrfs_dio_iomap_ops = {
8235 .iomap_begin = btrfs_dio_iomap_begin,
8236 .iomap_end = btrfs_dio_iomap_end,
8239 static const struct iomap_dio_ops btrfs_dio_ops = {
8240 .submit_io = btrfs_submit_direct,
8241 .bio_set = &btrfs_dio_bioset,
8244 ssize_t btrfs_dio_rw(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
8246 struct btrfs_dio_data data;
8248 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
8249 IOMAP_DIO_PARTIAL | IOMAP_DIO_NOSYNC,
8250 &data, done_before);
8253 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8258 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8262 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8265 static int btrfs_writepages(struct address_space *mapping,
8266 struct writeback_control *wbc)
8268 return extent_writepages(mapping, wbc);
8271 static void btrfs_readahead(struct readahead_control *rac)
8273 extent_readahead(rac);
8277 * For release_folio() and invalidate_folio() we have a race window where
8278 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8279 * If we continue to release/invalidate the page, we could cause use-after-free
8280 * for subpage spinlock. So this function is to spin and wait for subpage
8283 static void wait_subpage_spinlock(struct page *page)
8285 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8286 struct btrfs_subpage *subpage;
8288 if (!btrfs_is_subpage(fs_info, page))
8291 ASSERT(PagePrivate(page) && page->private);
8292 subpage = (struct btrfs_subpage *)page->private;
8295 * This may look insane as we just acquire the spinlock and release it,
8296 * without doing anything. But we just want to make sure no one is
8297 * still holding the subpage spinlock.
8298 * And since the page is not dirty nor writeback, and we have page
8299 * locked, the only possible way to hold a spinlock is from the endio
8300 * function to clear page writeback.
8302 * Here we just acquire the spinlock so that all existing callers
8303 * should exit and we're safe to release/invalidate the page.
8305 spin_lock_irq(&subpage->lock);
8306 spin_unlock_irq(&subpage->lock);
8309 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8311 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8314 wait_subpage_spinlock(&folio->page);
8315 clear_page_extent_mapped(&folio->page);
8320 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8322 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8324 return __btrfs_release_folio(folio, gfp_flags);
8327 #ifdef CONFIG_MIGRATION
8328 static int btrfs_migrate_folio(struct address_space *mapping,
8329 struct folio *dst, struct folio *src,
8330 enum migrate_mode mode)
8332 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8334 if (ret != MIGRATEPAGE_SUCCESS)
8337 if (folio_test_ordered(src)) {
8338 folio_clear_ordered(src);
8339 folio_set_ordered(dst);
8342 return MIGRATEPAGE_SUCCESS;
8345 #define btrfs_migrate_folio NULL
8348 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8351 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8352 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8353 struct extent_io_tree *tree = &inode->io_tree;
8354 struct extent_state *cached_state = NULL;
8355 u64 page_start = folio_pos(folio);
8356 u64 page_end = page_start + folio_size(folio) - 1;
8358 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8361 * We have folio locked so no new ordered extent can be created on this
8362 * page, nor bio can be submitted for this folio.
8364 * But already submitted bio can still be finished on this folio.
8365 * Furthermore, endio function won't skip folio which has Ordered
8366 * (Private2) already cleared, so it's possible for endio and
8367 * invalidate_folio to do the same ordered extent accounting twice
8370 * So here we wait for any submitted bios to finish, so that we won't
8371 * do double ordered extent accounting on the same folio.
8373 folio_wait_writeback(folio);
8374 wait_subpage_spinlock(&folio->page);
8377 * For subpage case, we have call sites like
8378 * btrfs_punch_hole_lock_range() which passes range not aligned to
8380 * If the range doesn't cover the full folio, we don't need to and
8381 * shouldn't clear page extent mapped, as folio->private can still
8382 * record subpage dirty bits for other part of the range.
8384 * For cases that invalidate the full folio even the range doesn't
8385 * cover the full folio, like invalidating the last folio, we're
8386 * still safe to wait for ordered extent to finish.
8388 if (!(offset == 0 && length == folio_size(folio))) {
8389 btrfs_release_folio(folio, GFP_NOFS);
8393 if (!inode_evicting)
8394 lock_extent_bits(tree, page_start, page_end, &cached_state);
8397 while (cur < page_end) {
8398 struct btrfs_ordered_extent *ordered;
8403 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8404 page_end + 1 - cur);
8406 range_end = page_end;
8408 * No ordered extent covering this range, we are safe
8409 * to delete all extent states in the range.
8411 delete_states = true;
8414 if (ordered->file_offset > cur) {
8416 * There is a range between [cur, oe->file_offset) not
8417 * covered by any ordered extent.
8418 * We are safe to delete all extent states, and handle
8419 * the ordered extent in the next iteration.
8421 range_end = ordered->file_offset - 1;
8422 delete_states = true;
8426 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8428 ASSERT(range_end + 1 - cur < U32_MAX);
8429 range_len = range_end + 1 - cur;
8430 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8432 * If Ordered (Private2) is cleared, it means endio has
8433 * already been executed for the range.
8434 * We can't delete the extent states as
8435 * btrfs_finish_ordered_io() may still use some of them.
8437 delete_states = false;
8440 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8443 * IO on this page will never be started, so we need to account
8444 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8445 * here, must leave that up for the ordered extent completion.
8447 * This will also unlock the range for incoming
8448 * btrfs_finish_ordered_io().
8450 if (!inode_evicting)
8451 clear_extent_bit(tree, cur, range_end,
8453 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8454 EXTENT_DEFRAG, 1, 0, &cached_state);
8456 spin_lock_irq(&inode->ordered_tree.lock);
8457 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8458 ordered->truncated_len = min(ordered->truncated_len,
8459 cur - ordered->file_offset);
8460 spin_unlock_irq(&inode->ordered_tree.lock);
8462 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8463 cur, range_end + 1 - cur)) {
8464 btrfs_finish_ordered_io(ordered);
8466 * The ordered extent has finished, now we're again
8467 * safe to delete all extent states of the range.
8469 delete_states = true;
8472 * btrfs_finish_ordered_io() will get executed by endio
8473 * of other pages, thus we can't delete extent states
8476 delete_states = false;
8480 btrfs_put_ordered_extent(ordered);
8482 * Qgroup reserved space handler
8483 * Sector(s) here will be either:
8485 * 1) Already written to disk or bio already finished
8486 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8487 * Qgroup will be handled by its qgroup_record then.
8488 * btrfs_qgroup_free_data() call will do nothing here.
8490 * 2) Not written to disk yet
8491 * Then btrfs_qgroup_free_data() call will clear the
8492 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8493 * reserved data space.
8494 * Since the IO will never happen for this page.
8496 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8497 if (!inode_evicting) {
8498 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8499 EXTENT_DELALLOC | EXTENT_UPTODATE |
8500 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8501 delete_states, &cached_state);
8503 cur = range_end + 1;
8506 * We have iterated through all ordered extents of the page, the page
8507 * should not have Ordered (Private2) anymore, or the above iteration
8508 * did something wrong.
8510 ASSERT(!folio_test_ordered(folio));
8511 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8512 if (!inode_evicting)
8513 __btrfs_release_folio(folio, GFP_NOFS);
8514 clear_page_extent_mapped(&folio->page);
8518 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8519 * called from a page fault handler when a page is first dirtied. Hence we must
8520 * be careful to check for EOF conditions here. We set the page up correctly
8521 * for a written page which means we get ENOSPC checking when writing into
8522 * holes and correct delalloc and unwritten extent mapping on filesystems that
8523 * support these features.
8525 * We are not allowed to take the i_mutex here so we have to play games to
8526 * protect against truncate races as the page could now be beyond EOF. Because
8527 * truncate_setsize() writes the inode size before removing pages, once we have
8528 * the page lock we can determine safely if the page is beyond EOF. If it is not
8529 * beyond EOF, then the page is guaranteed safe against truncation until we
8532 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8534 struct page *page = vmf->page;
8535 struct inode *inode = file_inode(vmf->vma->vm_file);
8536 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8537 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8538 struct btrfs_ordered_extent *ordered;
8539 struct extent_state *cached_state = NULL;
8540 struct extent_changeset *data_reserved = NULL;
8541 unsigned long zero_start;
8551 reserved_space = PAGE_SIZE;
8553 sb_start_pagefault(inode->i_sb);
8554 page_start = page_offset(page);
8555 page_end = page_start + PAGE_SIZE - 1;
8559 * Reserving delalloc space after obtaining the page lock can lead to
8560 * deadlock. For example, if a dirty page is locked by this function
8561 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8562 * dirty page write out, then the btrfs_writepages() function could
8563 * end up waiting indefinitely to get a lock on the page currently
8564 * being processed by btrfs_page_mkwrite() function.
8566 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8567 page_start, reserved_space);
8569 ret2 = file_update_time(vmf->vma->vm_file);
8573 ret = vmf_error(ret2);
8579 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8581 down_read(&BTRFS_I(inode)->i_mmap_lock);
8583 size = i_size_read(inode);
8585 if ((page->mapping != inode->i_mapping) ||
8586 (page_start >= size)) {
8587 /* page got truncated out from underneath us */
8590 wait_on_page_writeback(page);
8592 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8593 ret2 = set_page_extent_mapped(page);
8595 ret = vmf_error(ret2);
8596 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8601 * we can't set the delalloc bits if there are pending ordered
8602 * extents. Drop our locks and wait for them to finish
8604 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8607 unlock_extent_cached(io_tree, page_start, page_end,
8610 up_read(&BTRFS_I(inode)->i_mmap_lock);
8611 btrfs_start_ordered_extent(ordered, 1);
8612 btrfs_put_ordered_extent(ordered);
8616 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8617 reserved_space = round_up(size - page_start,
8618 fs_info->sectorsize);
8619 if (reserved_space < PAGE_SIZE) {
8620 end = page_start + reserved_space - 1;
8621 btrfs_delalloc_release_space(BTRFS_I(inode),
8622 data_reserved, page_start,
8623 PAGE_SIZE - reserved_space, true);
8628 * page_mkwrite gets called when the page is firstly dirtied after it's
8629 * faulted in, but write(2) could also dirty a page and set delalloc
8630 * bits, thus in this case for space account reason, we still need to
8631 * clear any delalloc bits within this page range since we have to
8632 * reserve data&meta space before lock_page() (see above comments).
8634 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8635 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8636 EXTENT_DEFRAG, 0, 0, &cached_state);
8638 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8641 unlock_extent_cached(io_tree, page_start, page_end,
8643 ret = VM_FAULT_SIGBUS;
8647 /* page is wholly or partially inside EOF */
8648 if (page_start + PAGE_SIZE > size)
8649 zero_start = offset_in_page(size);
8651 zero_start = PAGE_SIZE;
8653 if (zero_start != PAGE_SIZE)
8654 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8656 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8657 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8658 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8660 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8662 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8663 up_read(&BTRFS_I(inode)->i_mmap_lock);
8665 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8666 sb_end_pagefault(inode->i_sb);
8667 extent_changeset_free(data_reserved);
8668 return VM_FAULT_LOCKED;
8672 up_read(&BTRFS_I(inode)->i_mmap_lock);
8674 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8675 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8676 reserved_space, (ret != 0));
8678 sb_end_pagefault(inode->i_sb);
8679 extent_changeset_free(data_reserved);
8683 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8685 struct btrfs_truncate_control control = {
8686 .inode = BTRFS_I(inode),
8687 .ino = btrfs_ino(BTRFS_I(inode)),
8688 .min_type = BTRFS_EXTENT_DATA_KEY,
8689 .clear_extent_range = true,
8691 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8692 struct btrfs_root *root = BTRFS_I(inode)->root;
8693 struct btrfs_block_rsv *rsv;
8695 struct btrfs_trans_handle *trans;
8696 u64 mask = fs_info->sectorsize - 1;
8697 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8699 if (!skip_writeback) {
8700 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8707 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8708 * things going on here:
8710 * 1) We need to reserve space to update our inode.
8712 * 2) We need to have something to cache all the space that is going to
8713 * be free'd up by the truncate operation, but also have some slack
8714 * space reserved in case it uses space during the truncate (thank you
8715 * very much snapshotting).
8717 * And we need these to be separate. The fact is we can use a lot of
8718 * space doing the truncate, and we have no earthly idea how much space
8719 * we will use, so we need the truncate reservation to be separate so it
8720 * doesn't end up using space reserved for updating the inode. We also
8721 * need to be able to stop the transaction and start a new one, which
8722 * means we need to be able to update the inode several times, and we
8723 * have no idea of knowing how many times that will be, so we can't just
8724 * reserve 1 item for the entirety of the operation, so that has to be
8725 * done separately as well.
8727 * So that leaves us with
8729 * 1) rsv - for the truncate reservation, which we will steal from the
8730 * transaction reservation.
8731 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8732 * updating the inode.
8734 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8737 rsv->size = min_size;
8738 rsv->failfast = true;
8741 * 1 for the truncate slack space
8742 * 1 for updating the inode.
8744 trans = btrfs_start_transaction(root, 2);
8745 if (IS_ERR(trans)) {
8746 ret = PTR_ERR(trans);
8750 /* Migrate the slack space for the truncate to our reserve */
8751 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8755 trans->block_rsv = rsv;
8758 struct extent_state *cached_state = NULL;
8759 const u64 new_size = inode->i_size;
8760 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8762 control.new_size = new_size;
8763 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8766 * We want to drop from the next block forward in case this new
8767 * size is not block aligned since we will be keeping the last
8768 * block of the extent just the way it is.
8770 btrfs_drop_extent_cache(BTRFS_I(inode),
8771 ALIGN(new_size, fs_info->sectorsize),
8774 ret = btrfs_truncate_inode_items(trans, root, &control);
8776 inode_sub_bytes(inode, control.sub_bytes);
8777 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8779 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8780 (u64)-1, &cached_state);
8782 trans->block_rsv = &fs_info->trans_block_rsv;
8783 if (ret != -ENOSPC && ret != -EAGAIN)
8786 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8790 btrfs_end_transaction(trans);
8791 btrfs_btree_balance_dirty(fs_info);
8793 trans = btrfs_start_transaction(root, 2);
8794 if (IS_ERR(trans)) {
8795 ret = PTR_ERR(trans);
8800 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8801 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8802 rsv, min_size, false);
8803 BUG_ON(ret); /* shouldn't happen */
8804 trans->block_rsv = rsv;
8808 * We can't call btrfs_truncate_block inside a trans handle as we could
8809 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8810 * know we've truncated everything except the last little bit, and can
8811 * do btrfs_truncate_block and then update the disk_i_size.
8813 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8814 btrfs_end_transaction(trans);
8815 btrfs_btree_balance_dirty(fs_info);
8817 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8820 trans = btrfs_start_transaction(root, 1);
8821 if (IS_ERR(trans)) {
8822 ret = PTR_ERR(trans);
8825 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8831 trans->block_rsv = &fs_info->trans_block_rsv;
8832 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8836 ret2 = btrfs_end_transaction(trans);
8839 btrfs_btree_balance_dirty(fs_info);
8842 btrfs_free_block_rsv(fs_info, rsv);
8844 * So if we truncate and then write and fsync we normally would just
8845 * write the extents that changed, which is a problem if we need to
8846 * first truncate that entire inode. So set this flag so we write out
8847 * all of the extents in the inode to the sync log so we're completely
8850 * If no extents were dropped or trimmed we don't need to force the next
8851 * fsync to truncate all the inode's items from the log and re-log them
8852 * all. This means the truncate operation did not change the file size,
8853 * or changed it to a smaller size but there was only an implicit hole
8854 * between the old i_size and the new i_size, and there were no prealloc
8855 * extents beyond i_size to drop.
8857 if (control.extents_found > 0)
8858 btrfs_set_inode_full_sync(BTRFS_I(inode));
8863 struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns,
8866 struct inode *inode;
8868 inode = new_inode(dir->i_sb);
8871 * Subvolumes don't inherit the sgid bit or the parent's gid if
8872 * the parent's sgid bit is set. This is probably a bug.
8874 inode_init_owner(mnt_userns, inode, NULL,
8875 S_IFDIR | (~current_umask() & S_IRWXUGO));
8876 inode->i_op = &btrfs_dir_inode_operations;
8877 inode->i_fop = &btrfs_dir_file_operations;
8882 struct inode *btrfs_alloc_inode(struct super_block *sb)
8884 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8885 struct btrfs_inode *ei;
8886 struct inode *inode;
8888 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8895 ei->last_sub_trans = 0;
8896 ei->logged_trans = 0;
8897 ei->delalloc_bytes = 0;
8898 ei->new_delalloc_bytes = 0;
8899 ei->defrag_bytes = 0;
8900 ei->disk_i_size = 0;
8904 ei->index_cnt = (u64)-1;
8906 ei->last_unlink_trans = 0;
8907 ei->last_reflink_trans = 0;
8908 ei->last_log_commit = 0;
8910 spin_lock_init(&ei->lock);
8911 ei->outstanding_extents = 0;
8912 if (sb->s_magic != BTRFS_TEST_MAGIC)
8913 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8914 BTRFS_BLOCK_RSV_DELALLOC);
8915 ei->runtime_flags = 0;
8916 ei->prop_compress = BTRFS_COMPRESS_NONE;
8917 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8919 ei->delayed_node = NULL;
8921 ei->i_otime.tv_sec = 0;
8922 ei->i_otime.tv_nsec = 0;
8924 inode = &ei->vfs_inode;
8925 extent_map_tree_init(&ei->extent_tree);
8926 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8927 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8928 IO_TREE_INODE_IO_FAILURE, inode);
8929 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8930 IO_TREE_INODE_FILE_EXTENT, inode);
8931 ei->io_tree.track_uptodate = true;
8932 ei->io_failure_tree.track_uptodate = true;
8933 atomic_set(&ei->sync_writers, 0);
8934 mutex_init(&ei->log_mutex);
8935 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8936 INIT_LIST_HEAD(&ei->delalloc_inodes);
8937 INIT_LIST_HEAD(&ei->delayed_iput);
8938 RB_CLEAR_NODE(&ei->rb_node);
8939 init_rwsem(&ei->i_mmap_lock);
8944 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8945 void btrfs_test_destroy_inode(struct inode *inode)
8947 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8948 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8952 void btrfs_free_inode(struct inode *inode)
8954 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8957 void btrfs_destroy_inode(struct inode *vfs_inode)
8959 struct btrfs_ordered_extent *ordered;
8960 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8961 struct btrfs_root *root = inode->root;
8963 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8964 WARN_ON(vfs_inode->i_data.nrpages);
8965 WARN_ON(inode->block_rsv.reserved);
8966 WARN_ON(inode->block_rsv.size);
8967 WARN_ON(inode->outstanding_extents);
8968 if (!S_ISDIR(vfs_inode->i_mode)) {
8969 WARN_ON(inode->delalloc_bytes);
8970 WARN_ON(inode->new_delalloc_bytes);
8972 WARN_ON(inode->csum_bytes);
8973 WARN_ON(inode->defrag_bytes);
8976 * This can happen where we create an inode, but somebody else also
8977 * created the same inode and we need to destroy the one we already
8984 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8988 btrfs_err(root->fs_info,
8989 "found ordered extent %llu %llu on inode cleanup",
8990 ordered->file_offset, ordered->num_bytes);
8991 btrfs_remove_ordered_extent(inode, ordered);
8992 btrfs_put_ordered_extent(ordered);
8993 btrfs_put_ordered_extent(ordered);
8996 btrfs_qgroup_check_reserved_leak(inode);
8997 inode_tree_del(inode);
8998 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8999 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
9000 btrfs_put_root(inode->root);
9003 int btrfs_drop_inode(struct inode *inode)
9005 struct btrfs_root *root = BTRFS_I(inode)->root;
9010 /* the snap/subvol tree is on deleting */
9011 if (btrfs_root_refs(&root->root_item) == 0)
9014 return generic_drop_inode(inode);
9017 static void init_once(void *foo)
9019 struct btrfs_inode *ei = foo;
9021 inode_init_once(&ei->vfs_inode);
9024 void __cold btrfs_destroy_cachep(void)
9027 * Make sure all delayed rcu free inodes are flushed before we
9031 bioset_exit(&btrfs_dio_bioset);
9032 kmem_cache_destroy(btrfs_inode_cachep);
9033 kmem_cache_destroy(btrfs_trans_handle_cachep);
9034 kmem_cache_destroy(btrfs_path_cachep);
9035 kmem_cache_destroy(btrfs_free_space_cachep);
9036 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9039 int __init btrfs_init_cachep(void)
9041 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9042 sizeof(struct btrfs_inode), 0,
9043 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9045 if (!btrfs_inode_cachep)
9048 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9049 sizeof(struct btrfs_trans_handle), 0,
9050 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9051 if (!btrfs_trans_handle_cachep)
9054 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9055 sizeof(struct btrfs_path), 0,
9056 SLAB_MEM_SPREAD, NULL);
9057 if (!btrfs_path_cachep)
9060 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9061 sizeof(struct btrfs_free_space), 0,
9062 SLAB_MEM_SPREAD, NULL);
9063 if (!btrfs_free_space_cachep)
9066 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9067 PAGE_SIZE, PAGE_SIZE,
9068 SLAB_MEM_SPREAD, NULL);
9069 if (!btrfs_free_space_bitmap_cachep)
9072 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
9073 offsetof(struct btrfs_dio_private, bio),
9079 btrfs_destroy_cachep();
9083 static int btrfs_getattr(struct user_namespace *mnt_userns,
9084 const struct path *path, struct kstat *stat,
9085 u32 request_mask, unsigned int flags)
9089 struct inode *inode = d_inode(path->dentry);
9090 u32 blocksize = inode->i_sb->s_blocksize;
9091 u32 bi_flags = BTRFS_I(inode)->flags;
9092 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9094 stat->result_mask |= STATX_BTIME;
9095 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9096 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9097 if (bi_flags & BTRFS_INODE_APPEND)
9098 stat->attributes |= STATX_ATTR_APPEND;
9099 if (bi_flags & BTRFS_INODE_COMPRESS)
9100 stat->attributes |= STATX_ATTR_COMPRESSED;
9101 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9102 stat->attributes |= STATX_ATTR_IMMUTABLE;
9103 if (bi_flags & BTRFS_INODE_NODUMP)
9104 stat->attributes |= STATX_ATTR_NODUMP;
9105 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9106 stat->attributes |= STATX_ATTR_VERITY;
9108 stat->attributes_mask |= (STATX_ATTR_APPEND |
9109 STATX_ATTR_COMPRESSED |
9110 STATX_ATTR_IMMUTABLE |
9113 generic_fillattr(mnt_userns, inode, stat);
9114 stat->dev = BTRFS_I(inode)->root->anon_dev;
9116 spin_lock(&BTRFS_I(inode)->lock);
9117 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9118 inode_bytes = inode_get_bytes(inode);
9119 spin_unlock(&BTRFS_I(inode)->lock);
9120 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9121 ALIGN(delalloc_bytes, blocksize)) >> 9;
9125 static int btrfs_rename_exchange(struct inode *old_dir,
9126 struct dentry *old_dentry,
9127 struct inode *new_dir,
9128 struct dentry *new_dentry)
9130 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9131 struct btrfs_trans_handle *trans;
9132 unsigned int trans_num_items;
9133 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9134 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9135 struct inode *new_inode = new_dentry->d_inode;
9136 struct inode *old_inode = old_dentry->d_inode;
9137 struct timespec64 ctime = current_time(old_inode);
9138 struct btrfs_rename_ctx old_rename_ctx;
9139 struct btrfs_rename_ctx new_rename_ctx;
9140 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9141 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9146 bool need_abort = false;
9149 * For non-subvolumes allow exchange only within one subvolume, in the
9150 * same inode namespace. Two subvolumes (represented as directory) can
9151 * be exchanged as they're a logical link and have a fixed inode number.
9154 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9155 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9158 /* close the race window with snapshot create/destroy ioctl */
9159 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9160 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9161 down_read(&fs_info->subvol_sem);
9165 * 1 to remove old dir item
9166 * 1 to remove old dir index
9167 * 1 to add new dir item
9168 * 1 to add new dir index
9169 * 1 to update parent inode
9171 * If the parents are the same, we only need to account for one
9173 trans_num_items = (old_dir == new_dir ? 9 : 10);
9174 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9176 * 1 to remove old root ref
9177 * 1 to remove old root backref
9178 * 1 to add new root ref
9179 * 1 to add new root backref
9181 trans_num_items += 4;
9184 * 1 to update inode item
9185 * 1 to remove old inode ref
9186 * 1 to add new inode ref
9188 trans_num_items += 3;
9190 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9191 trans_num_items += 4;
9193 trans_num_items += 3;
9194 trans = btrfs_start_transaction(root, trans_num_items);
9195 if (IS_ERR(trans)) {
9196 ret = PTR_ERR(trans);
9201 ret = btrfs_record_root_in_trans(trans, dest);
9207 * We need to find a free sequence number both in the source and
9208 * in the destination directory for the exchange.
9210 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9213 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9217 BTRFS_I(old_inode)->dir_index = 0ULL;
9218 BTRFS_I(new_inode)->dir_index = 0ULL;
9220 /* Reference for the source. */
9221 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9222 /* force full log commit if subvolume involved. */
9223 btrfs_set_log_full_commit(trans);
9225 ret = btrfs_insert_inode_ref(trans, dest,
9226 new_dentry->d_name.name,
9227 new_dentry->d_name.len,
9229 btrfs_ino(BTRFS_I(new_dir)),
9236 /* And now for the dest. */
9237 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9238 /* force full log commit if subvolume involved. */
9239 btrfs_set_log_full_commit(trans);
9241 ret = btrfs_insert_inode_ref(trans, root,
9242 old_dentry->d_name.name,
9243 old_dentry->d_name.len,
9245 btrfs_ino(BTRFS_I(old_dir)),
9249 btrfs_abort_transaction(trans, ret);
9254 /* Update inode version and ctime/mtime. */
9255 inode_inc_iversion(old_dir);
9256 inode_inc_iversion(new_dir);
9257 inode_inc_iversion(old_inode);
9258 inode_inc_iversion(new_inode);
9259 old_dir->i_mtime = ctime;
9260 old_dir->i_ctime = ctime;
9261 new_dir->i_mtime = ctime;
9262 new_dir->i_ctime = ctime;
9263 old_inode->i_ctime = ctime;
9264 new_inode->i_ctime = ctime;
9266 if (old_dentry->d_parent != new_dentry->d_parent) {
9267 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9268 BTRFS_I(old_inode), 1);
9269 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9270 BTRFS_I(new_inode), 1);
9273 /* src is a subvolume */
9274 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9275 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9276 } else { /* src is an inode */
9277 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9278 BTRFS_I(old_dentry->d_inode),
9279 old_dentry->d_name.name,
9280 old_dentry->d_name.len,
9283 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9286 btrfs_abort_transaction(trans, ret);
9290 /* dest is a subvolume */
9291 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9292 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9293 } else { /* dest is an inode */
9294 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9295 BTRFS_I(new_dentry->d_inode),
9296 new_dentry->d_name.name,
9297 new_dentry->d_name.len,
9300 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9303 btrfs_abort_transaction(trans, ret);
9307 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9308 new_dentry->d_name.name,
9309 new_dentry->d_name.len, 0, old_idx);
9311 btrfs_abort_transaction(trans, ret);
9315 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9316 old_dentry->d_name.name,
9317 old_dentry->d_name.len, 0, new_idx);
9319 btrfs_abort_transaction(trans, ret);
9323 if (old_inode->i_nlink == 1)
9324 BTRFS_I(old_inode)->dir_index = old_idx;
9325 if (new_inode->i_nlink == 1)
9326 BTRFS_I(new_inode)->dir_index = new_idx;
9329 * Now pin the logs of the roots. We do it to ensure that no other task
9330 * can sync the logs while we are in progress with the rename, because
9331 * that could result in an inconsistency in case any of the inodes that
9332 * are part of this rename operation were logged before.
9334 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9335 btrfs_pin_log_trans(root);
9336 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9337 btrfs_pin_log_trans(dest);
9339 /* Do the log updates for all inodes. */
9340 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9341 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9342 old_rename_ctx.index, new_dentry->d_parent);
9343 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9344 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9345 new_rename_ctx.index, old_dentry->d_parent);
9347 /* Now unpin the logs. */
9348 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9349 btrfs_end_log_trans(root);
9350 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9351 btrfs_end_log_trans(dest);
9353 ret2 = btrfs_end_transaction(trans);
9354 ret = ret ? ret : ret2;
9356 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9357 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9358 up_read(&fs_info->subvol_sem);
9363 static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns,
9366 struct inode *inode;
9368 inode = new_inode(dir->i_sb);
9370 inode_init_owner(mnt_userns, inode, dir,
9371 S_IFCHR | WHITEOUT_MODE);
9372 inode->i_op = &btrfs_special_inode_operations;
9373 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9378 static int btrfs_rename(struct user_namespace *mnt_userns,
9379 struct inode *old_dir, struct dentry *old_dentry,
9380 struct inode *new_dir, struct dentry *new_dentry,
9383 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9384 struct btrfs_new_inode_args whiteout_args = {
9386 .dentry = old_dentry,
9388 struct btrfs_trans_handle *trans;
9389 unsigned int trans_num_items;
9390 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9391 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9392 struct inode *new_inode = d_inode(new_dentry);
9393 struct inode *old_inode = d_inode(old_dentry);
9394 struct btrfs_rename_ctx rename_ctx;
9398 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9400 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9403 /* we only allow rename subvolume link between subvolumes */
9404 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9407 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9408 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9411 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9412 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9416 /* check for collisions, even if the name isn't there */
9417 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9418 new_dentry->d_name.name,
9419 new_dentry->d_name.len);
9422 if (ret == -EEXIST) {
9424 * eexist without a new_inode */
9425 if (WARN_ON(!new_inode)) {
9429 /* maybe -EOVERFLOW */
9436 * we're using rename to replace one file with another. Start IO on it
9437 * now so we don't add too much work to the end of the transaction
9439 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9440 filemap_flush(old_inode->i_mapping);
9442 if (flags & RENAME_WHITEOUT) {
9443 whiteout_args.inode = new_whiteout_inode(mnt_userns, old_dir);
9444 if (!whiteout_args.inode)
9446 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9448 goto out_whiteout_inode;
9450 /* 1 to update the old parent inode. */
9451 trans_num_items = 1;
9454 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9455 /* Close the race window with snapshot create/destroy ioctl */
9456 down_read(&fs_info->subvol_sem);
9458 * 1 to remove old root ref
9459 * 1 to remove old root backref
9460 * 1 to add new root ref
9461 * 1 to add new root backref
9463 trans_num_items += 4;
9467 * 1 to remove old inode ref
9468 * 1 to add new inode ref
9470 trans_num_items += 3;
9473 * 1 to remove old dir item
9474 * 1 to remove old dir index
9475 * 1 to add new dir item
9476 * 1 to add new dir index
9478 trans_num_items += 4;
9479 /* 1 to update new parent inode if it's not the same as the old parent */
9480 if (new_dir != old_dir)
9485 * 1 to remove inode ref
9486 * 1 to remove dir item
9487 * 1 to remove dir index
9488 * 1 to possibly add orphan item
9490 trans_num_items += 5;
9492 trans = btrfs_start_transaction(root, trans_num_items);
9493 if (IS_ERR(trans)) {
9494 ret = PTR_ERR(trans);
9499 ret = btrfs_record_root_in_trans(trans, dest);
9504 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9508 BTRFS_I(old_inode)->dir_index = 0ULL;
9509 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9510 /* force full log commit if subvolume involved. */
9511 btrfs_set_log_full_commit(trans);
9513 ret = btrfs_insert_inode_ref(trans, dest,
9514 new_dentry->d_name.name,
9515 new_dentry->d_name.len,
9517 btrfs_ino(BTRFS_I(new_dir)), index);
9522 inode_inc_iversion(old_dir);
9523 inode_inc_iversion(new_dir);
9524 inode_inc_iversion(old_inode);
9525 old_dir->i_mtime = current_time(old_dir);
9526 old_dir->i_ctime = old_dir->i_mtime;
9527 new_dir->i_mtime = old_dir->i_mtime;
9528 new_dir->i_ctime = old_dir->i_mtime;
9529 old_inode->i_ctime = old_dir->i_mtime;
9531 if (old_dentry->d_parent != new_dentry->d_parent)
9532 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9533 BTRFS_I(old_inode), 1);
9535 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9536 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9538 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9539 BTRFS_I(d_inode(old_dentry)),
9540 old_dentry->d_name.name,
9541 old_dentry->d_name.len,
9544 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9547 btrfs_abort_transaction(trans, ret);
9552 inode_inc_iversion(new_inode);
9553 new_inode->i_ctime = current_time(new_inode);
9554 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9555 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9556 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9557 BUG_ON(new_inode->i_nlink == 0);
9559 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9560 BTRFS_I(d_inode(new_dentry)),
9561 new_dentry->d_name.name,
9562 new_dentry->d_name.len);
9564 if (!ret && new_inode->i_nlink == 0)
9565 ret = btrfs_orphan_add(trans,
9566 BTRFS_I(d_inode(new_dentry)));
9568 btrfs_abort_transaction(trans, ret);
9573 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9574 new_dentry->d_name.name,
9575 new_dentry->d_name.len, 0, index);
9577 btrfs_abort_transaction(trans, ret);
9581 if (old_inode->i_nlink == 1)
9582 BTRFS_I(old_inode)->dir_index = index;
9584 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9585 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9586 rename_ctx.index, new_dentry->d_parent);
9588 if (flags & RENAME_WHITEOUT) {
9589 ret = btrfs_create_new_inode(trans, &whiteout_args);
9591 btrfs_abort_transaction(trans, ret);
9594 unlock_new_inode(whiteout_args.inode);
9595 iput(whiteout_args.inode);
9596 whiteout_args.inode = NULL;
9600 ret2 = btrfs_end_transaction(trans);
9601 ret = ret ? ret : ret2;
9603 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9604 up_read(&fs_info->subvol_sem);
9605 if (flags & RENAME_WHITEOUT)
9606 btrfs_new_inode_args_destroy(&whiteout_args);
9608 if (flags & RENAME_WHITEOUT)
9609 iput(whiteout_args.inode);
9613 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9614 struct dentry *old_dentry, struct inode *new_dir,
9615 struct dentry *new_dentry, unsigned int flags)
9619 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9622 if (flags & RENAME_EXCHANGE)
9623 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9626 ret = btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9629 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9634 struct btrfs_delalloc_work {
9635 struct inode *inode;
9636 struct completion completion;
9637 struct list_head list;
9638 struct btrfs_work work;
9641 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9643 struct btrfs_delalloc_work *delalloc_work;
9644 struct inode *inode;
9646 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9648 inode = delalloc_work->inode;
9649 filemap_flush(inode->i_mapping);
9650 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9651 &BTRFS_I(inode)->runtime_flags))
9652 filemap_flush(inode->i_mapping);
9655 complete(&delalloc_work->completion);
9658 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9660 struct btrfs_delalloc_work *work;
9662 work = kmalloc(sizeof(*work), GFP_NOFS);
9666 init_completion(&work->completion);
9667 INIT_LIST_HEAD(&work->list);
9668 work->inode = inode;
9669 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9675 * some fairly slow code that needs optimization. This walks the list
9676 * of all the inodes with pending delalloc and forces them to disk.
9678 static int start_delalloc_inodes(struct btrfs_root *root,
9679 struct writeback_control *wbc, bool snapshot,
9680 bool in_reclaim_context)
9682 struct btrfs_inode *binode;
9683 struct inode *inode;
9684 struct btrfs_delalloc_work *work, *next;
9685 struct list_head works;
9686 struct list_head splice;
9688 bool full_flush = wbc->nr_to_write == LONG_MAX;
9690 INIT_LIST_HEAD(&works);
9691 INIT_LIST_HEAD(&splice);
9693 mutex_lock(&root->delalloc_mutex);
9694 spin_lock(&root->delalloc_lock);
9695 list_splice_init(&root->delalloc_inodes, &splice);
9696 while (!list_empty(&splice)) {
9697 binode = list_entry(splice.next, struct btrfs_inode,
9700 list_move_tail(&binode->delalloc_inodes,
9701 &root->delalloc_inodes);
9703 if (in_reclaim_context &&
9704 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9707 inode = igrab(&binode->vfs_inode);
9709 cond_resched_lock(&root->delalloc_lock);
9712 spin_unlock(&root->delalloc_lock);
9715 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9716 &binode->runtime_flags);
9718 work = btrfs_alloc_delalloc_work(inode);
9724 list_add_tail(&work->list, &works);
9725 btrfs_queue_work(root->fs_info->flush_workers,
9728 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9729 btrfs_add_delayed_iput(inode);
9730 if (ret || wbc->nr_to_write <= 0)
9734 spin_lock(&root->delalloc_lock);
9736 spin_unlock(&root->delalloc_lock);
9739 list_for_each_entry_safe(work, next, &works, list) {
9740 list_del_init(&work->list);
9741 wait_for_completion(&work->completion);
9745 if (!list_empty(&splice)) {
9746 spin_lock(&root->delalloc_lock);
9747 list_splice_tail(&splice, &root->delalloc_inodes);
9748 spin_unlock(&root->delalloc_lock);
9750 mutex_unlock(&root->delalloc_mutex);
9754 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9756 struct writeback_control wbc = {
9757 .nr_to_write = LONG_MAX,
9758 .sync_mode = WB_SYNC_NONE,
9760 .range_end = LLONG_MAX,
9762 struct btrfs_fs_info *fs_info = root->fs_info;
9764 if (BTRFS_FS_ERROR(fs_info))
9767 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9770 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9771 bool in_reclaim_context)
9773 struct writeback_control wbc = {
9775 .sync_mode = WB_SYNC_NONE,
9777 .range_end = LLONG_MAX,
9779 struct btrfs_root *root;
9780 struct list_head splice;
9783 if (BTRFS_FS_ERROR(fs_info))
9786 INIT_LIST_HEAD(&splice);
9788 mutex_lock(&fs_info->delalloc_root_mutex);
9789 spin_lock(&fs_info->delalloc_root_lock);
9790 list_splice_init(&fs_info->delalloc_roots, &splice);
9791 while (!list_empty(&splice)) {
9793 * Reset nr_to_write here so we know that we're doing a full
9797 wbc.nr_to_write = LONG_MAX;
9799 root = list_first_entry(&splice, struct btrfs_root,
9801 root = btrfs_grab_root(root);
9803 list_move_tail(&root->delalloc_root,
9804 &fs_info->delalloc_roots);
9805 spin_unlock(&fs_info->delalloc_root_lock);
9807 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9808 btrfs_put_root(root);
9809 if (ret < 0 || wbc.nr_to_write <= 0)
9811 spin_lock(&fs_info->delalloc_root_lock);
9813 spin_unlock(&fs_info->delalloc_root_lock);
9817 if (!list_empty(&splice)) {
9818 spin_lock(&fs_info->delalloc_root_lock);
9819 list_splice_tail(&splice, &fs_info->delalloc_roots);
9820 spin_unlock(&fs_info->delalloc_root_lock);
9822 mutex_unlock(&fs_info->delalloc_root_mutex);
9826 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9827 struct dentry *dentry, const char *symname)
9829 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9830 struct btrfs_trans_handle *trans;
9831 struct btrfs_root *root = BTRFS_I(dir)->root;
9832 struct btrfs_path *path;
9833 struct btrfs_key key;
9834 struct inode *inode;
9835 struct btrfs_new_inode_args new_inode_args = {
9839 unsigned int trans_num_items;
9844 struct btrfs_file_extent_item *ei;
9845 struct extent_buffer *leaf;
9847 name_len = strlen(symname);
9848 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9849 return -ENAMETOOLONG;
9851 inode = new_inode(dir->i_sb);
9854 inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO);
9855 inode->i_op = &btrfs_symlink_inode_operations;
9856 inode_nohighmem(inode);
9857 inode->i_mapping->a_ops = &btrfs_aops;
9858 btrfs_i_size_write(BTRFS_I(inode), name_len);
9859 inode_set_bytes(inode, name_len);
9861 new_inode_args.inode = inode;
9862 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9865 /* 1 additional item for the inline extent */
9868 trans = btrfs_start_transaction(root, trans_num_items);
9869 if (IS_ERR(trans)) {
9870 err = PTR_ERR(trans);
9871 goto out_new_inode_args;
9874 err = btrfs_create_new_inode(trans, &new_inode_args);
9878 path = btrfs_alloc_path();
9881 btrfs_abort_transaction(trans, err);
9882 discard_new_inode(inode);
9886 key.objectid = btrfs_ino(BTRFS_I(inode));
9888 key.type = BTRFS_EXTENT_DATA_KEY;
9889 datasize = btrfs_file_extent_calc_inline_size(name_len);
9890 err = btrfs_insert_empty_item(trans, root, path, &key,
9893 btrfs_abort_transaction(trans, err);
9894 btrfs_free_path(path);
9895 discard_new_inode(inode);
9899 leaf = path->nodes[0];
9900 ei = btrfs_item_ptr(leaf, path->slots[0],
9901 struct btrfs_file_extent_item);
9902 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9903 btrfs_set_file_extent_type(leaf, ei,
9904 BTRFS_FILE_EXTENT_INLINE);
9905 btrfs_set_file_extent_encryption(leaf, ei, 0);
9906 btrfs_set_file_extent_compression(leaf, ei, 0);
9907 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9908 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9910 ptr = btrfs_file_extent_inline_start(ei);
9911 write_extent_buffer(leaf, symname, ptr, name_len);
9912 btrfs_mark_buffer_dirty(leaf);
9913 btrfs_free_path(path);
9915 d_instantiate_new(dentry, inode);
9918 btrfs_end_transaction(trans);
9919 btrfs_btree_balance_dirty(fs_info);
9921 btrfs_new_inode_args_destroy(&new_inode_args);
9928 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9929 struct btrfs_trans_handle *trans_in,
9930 struct btrfs_inode *inode,
9931 struct btrfs_key *ins,
9934 struct btrfs_file_extent_item stack_fi;
9935 struct btrfs_replace_extent_info extent_info;
9936 struct btrfs_trans_handle *trans = trans_in;
9937 struct btrfs_path *path;
9938 u64 start = ins->objectid;
9939 u64 len = ins->offset;
9940 int qgroup_released;
9943 memset(&stack_fi, 0, sizeof(stack_fi));
9945 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9946 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9947 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9948 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9949 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9950 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9951 /* Encryption and other encoding is reserved and all 0 */
9953 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9954 if (qgroup_released < 0)
9955 return ERR_PTR(qgroup_released);
9958 ret = insert_reserved_file_extent(trans, inode,
9959 file_offset, &stack_fi,
9960 true, qgroup_released);
9966 extent_info.disk_offset = start;
9967 extent_info.disk_len = len;
9968 extent_info.data_offset = 0;
9969 extent_info.data_len = len;
9970 extent_info.file_offset = file_offset;
9971 extent_info.extent_buf = (char *)&stack_fi;
9972 extent_info.is_new_extent = true;
9973 extent_info.update_times = true;
9974 extent_info.qgroup_reserved = qgroup_released;
9975 extent_info.insertions = 0;
9977 path = btrfs_alloc_path();
9983 ret = btrfs_replace_file_extents(inode, path, file_offset,
9984 file_offset + len - 1, &extent_info,
9986 btrfs_free_path(path);
9993 * We have released qgroup data range at the beginning of the function,
9994 * and normally qgroup_released bytes will be freed when committing
9996 * But if we error out early, we have to free what we have released
9997 * or we leak qgroup data reservation.
9999 btrfs_qgroup_free_refroot(inode->root->fs_info,
10000 inode->root->root_key.objectid, qgroup_released,
10001 BTRFS_QGROUP_RSV_DATA);
10002 return ERR_PTR(ret);
10005 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10006 u64 start, u64 num_bytes, u64 min_size,
10007 loff_t actual_len, u64 *alloc_hint,
10008 struct btrfs_trans_handle *trans)
10010 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10011 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10012 struct extent_map *em;
10013 struct btrfs_root *root = BTRFS_I(inode)->root;
10014 struct btrfs_key ins;
10015 u64 cur_offset = start;
10016 u64 clear_offset = start;
10019 u64 last_alloc = (u64)-1;
10021 bool own_trans = true;
10022 u64 end = start + num_bytes - 1;
10026 while (num_bytes > 0) {
10027 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10028 cur_bytes = max(cur_bytes, min_size);
10030 * If we are severely fragmented we could end up with really
10031 * small allocations, so if the allocator is returning small
10032 * chunks lets make its job easier by only searching for those
10035 cur_bytes = min(cur_bytes, last_alloc);
10036 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10037 min_size, 0, *alloc_hint, &ins, 1, 0);
10042 * We've reserved this space, and thus converted it from
10043 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10044 * from here on out we will only need to clear our reservation
10045 * for the remaining unreserved area, so advance our
10046 * clear_offset by our extent size.
10048 clear_offset += ins.offset;
10050 last_alloc = ins.offset;
10051 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10054 * Now that we inserted the prealloc extent we can finally
10055 * decrement the number of reservations in the block group.
10056 * If we did it before, we could race with relocation and have
10057 * relocation miss the reserved extent, making it fail later.
10059 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10060 if (IS_ERR(trans)) {
10061 ret = PTR_ERR(trans);
10062 btrfs_free_reserved_extent(fs_info, ins.objectid,
10067 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10068 cur_offset + ins.offset -1, 0);
10070 em = alloc_extent_map();
10072 btrfs_set_inode_full_sync(BTRFS_I(inode));
10076 em->start = cur_offset;
10077 em->orig_start = cur_offset;
10078 em->len = ins.offset;
10079 em->block_start = ins.objectid;
10080 em->block_len = ins.offset;
10081 em->orig_block_len = ins.offset;
10082 em->ram_bytes = ins.offset;
10083 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10084 em->generation = trans->transid;
10087 write_lock(&em_tree->lock);
10088 ret = add_extent_mapping(em_tree, em, 1);
10089 write_unlock(&em_tree->lock);
10090 if (ret != -EEXIST)
10092 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10093 cur_offset + ins.offset - 1,
10096 free_extent_map(em);
10098 num_bytes -= ins.offset;
10099 cur_offset += ins.offset;
10100 *alloc_hint = ins.objectid + ins.offset;
10102 inode_inc_iversion(inode);
10103 inode->i_ctime = current_time(inode);
10104 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10105 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10106 (actual_len > inode->i_size) &&
10107 (cur_offset > inode->i_size)) {
10108 if (cur_offset > actual_len)
10109 i_size = actual_len;
10111 i_size = cur_offset;
10112 i_size_write(inode, i_size);
10113 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10116 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10119 btrfs_abort_transaction(trans, ret);
10121 btrfs_end_transaction(trans);
10126 btrfs_end_transaction(trans);
10130 if (clear_offset < end)
10131 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10132 end - clear_offset + 1);
10136 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10137 u64 start, u64 num_bytes, u64 min_size,
10138 loff_t actual_len, u64 *alloc_hint)
10140 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10141 min_size, actual_len, alloc_hint,
10145 int btrfs_prealloc_file_range_trans(struct inode *inode,
10146 struct btrfs_trans_handle *trans, int mode,
10147 u64 start, u64 num_bytes, u64 min_size,
10148 loff_t actual_len, u64 *alloc_hint)
10150 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10151 min_size, actual_len, alloc_hint, trans);
10154 static int btrfs_permission(struct user_namespace *mnt_userns,
10155 struct inode *inode, int mask)
10157 struct btrfs_root *root = BTRFS_I(inode)->root;
10158 umode_t mode = inode->i_mode;
10160 if (mask & MAY_WRITE &&
10161 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10162 if (btrfs_root_readonly(root))
10164 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10167 return generic_permission(mnt_userns, inode, mask);
10170 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10171 struct dentry *dentry, umode_t mode)
10173 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10174 struct btrfs_trans_handle *trans;
10175 struct btrfs_root *root = BTRFS_I(dir)->root;
10176 struct inode *inode;
10177 struct btrfs_new_inode_args new_inode_args = {
10182 unsigned int trans_num_items;
10185 inode = new_inode(dir->i_sb);
10188 inode_init_owner(mnt_userns, inode, dir, mode);
10189 inode->i_fop = &btrfs_file_operations;
10190 inode->i_op = &btrfs_file_inode_operations;
10191 inode->i_mapping->a_ops = &btrfs_aops;
10193 new_inode_args.inode = inode;
10194 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
10198 trans = btrfs_start_transaction(root, trans_num_items);
10199 if (IS_ERR(trans)) {
10200 ret = PTR_ERR(trans);
10201 goto out_new_inode_args;
10204 ret = btrfs_create_new_inode(trans, &new_inode_args);
10207 * We set number of links to 0 in btrfs_create_new_inode(), and here we
10208 * set it to 1 because d_tmpfile() will issue a warning if the count is
10211 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10213 set_nlink(inode, 1);
10216 d_tmpfile(dentry, inode);
10217 unlock_new_inode(inode);
10218 mark_inode_dirty(inode);
10221 btrfs_end_transaction(trans);
10222 btrfs_btree_balance_dirty(fs_info);
10223 out_new_inode_args:
10224 btrfs_new_inode_args_destroy(&new_inode_args);
10231 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10233 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10234 unsigned long index = start >> PAGE_SHIFT;
10235 unsigned long end_index = end >> PAGE_SHIFT;
10239 ASSERT(end + 1 - start <= U32_MAX);
10240 len = end + 1 - start;
10241 while (index <= end_index) {
10242 page = find_get_page(inode->vfs_inode.i_mapping, index);
10243 ASSERT(page); /* Pages should be in the extent_io_tree */
10245 btrfs_page_set_writeback(fs_info, page, start, len);
10251 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
10254 switch (compress_type) {
10255 case BTRFS_COMPRESS_NONE:
10256 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10257 case BTRFS_COMPRESS_ZLIB:
10258 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10259 case BTRFS_COMPRESS_LZO:
10261 * The LZO format depends on the sector size. 64K is the maximum
10262 * sector size that we support.
10264 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10266 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10267 (fs_info->sectorsize_bits - 12);
10268 case BTRFS_COMPRESS_ZSTD:
10269 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10275 static ssize_t btrfs_encoded_read_inline(
10276 struct kiocb *iocb,
10277 struct iov_iter *iter, u64 start,
10279 struct extent_state **cached_state,
10280 u64 extent_start, size_t count,
10281 struct btrfs_ioctl_encoded_io_args *encoded,
10284 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10285 struct btrfs_root *root = inode->root;
10286 struct btrfs_fs_info *fs_info = root->fs_info;
10287 struct extent_io_tree *io_tree = &inode->io_tree;
10288 struct btrfs_path *path;
10289 struct extent_buffer *leaf;
10290 struct btrfs_file_extent_item *item;
10296 path = btrfs_alloc_path();
10301 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10305 /* The extent item disappeared? */
10310 leaf = path->nodes[0];
10311 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10313 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10314 ptr = btrfs_file_extent_inline_start(item);
10316 encoded->len = min_t(u64, extent_start + ram_bytes,
10317 inode->vfs_inode.i_size) - iocb->ki_pos;
10318 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10319 btrfs_file_extent_compression(leaf, item));
10322 encoded->compression = ret;
10323 if (encoded->compression) {
10324 size_t inline_size;
10326 inline_size = btrfs_file_extent_inline_item_len(leaf,
10328 if (inline_size > count) {
10332 count = inline_size;
10333 encoded->unencoded_len = ram_bytes;
10334 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10336 count = min_t(u64, count, encoded->len);
10337 encoded->len = count;
10338 encoded->unencoded_len = count;
10339 ptr += iocb->ki_pos - extent_start;
10342 tmp = kmalloc(count, GFP_NOFS);
10347 read_extent_buffer(leaf, tmp, ptr, count);
10348 btrfs_release_path(path);
10349 unlock_extent_cached(io_tree, start, lockend, cached_state);
10350 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10353 ret = copy_to_iter(tmp, count, iter);
10358 btrfs_free_path(path);
10362 struct btrfs_encoded_read_private {
10363 struct btrfs_inode *inode;
10365 wait_queue_head_t wait;
10367 blk_status_t status;
10371 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10372 struct bio *bio, int mirror_num)
10374 struct btrfs_encoded_read_private *priv = bio->bi_private;
10375 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10378 if (!priv->skip_csum) {
10379 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10384 atomic_inc(&priv->pending);
10385 btrfs_submit_bio(fs_info, bio, mirror_num);
10389 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10391 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10392 struct btrfs_encoded_read_private *priv = bbio->bio.bi_private;
10393 struct btrfs_inode *inode = priv->inode;
10394 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10395 u32 sectorsize = fs_info->sectorsize;
10396 struct bio_vec *bvec;
10397 struct bvec_iter_all iter_all;
10398 u32 bio_offset = 0;
10400 if (priv->skip_csum || !uptodate)
10401 return bbio->bio.bi_status;
10403 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10404 unsigned int i, nr_sectors, pgoff;
10406 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10407 pgoff = bvec->bv_offset;
10408 for (i = 0; i < nr_sectors; i++) {
10409 ASSERT(pgoff < PAGE_SIZE);
10410 if (btrfs_check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10411 bvec->bv_page, pgoff))
10412 return BLK_STS_IOERR;
10413 bio_offset += sectorsize;
10414 pgoff += sectorsize;
10420 static void btrfs_encoded_read_endio(struct bio *bio)
10422 struct btrfs_encoded_read_private *priv = bio->bi_private;
10423 struct btrfs_bio *bbio = btrfs_bio(bio);
10424 blk_status_t status;
10426 status = btrfs_encoded_read_verify_csum(bbio);
10429 * The memory barrier implied by the atomic_dec_return() here
10430 * pairs with the memory barrier implied by the
10431 * atomic_dec_return() or io_wait_event() in
10432 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10433 * write is observed before the load of status in
10434 * btrfs_encoded_read_regular_fill_pages().
10436 WRITE_ONCE(priv->status, status);
10438 if (!atomic_dec_return(&priv->pending))
10439 wake_up(&priv->wait);
10440 btrfs_bio_free_csum(bbio);
10444 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10445 u64 file_offset, u64 disk_bytenr,
10446 u64 disk_io_size, struct page **pages)
10448 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10449 struct btrfs_encoded_read_private priv = {
10451 .file_offset = file_offset,
10452 .pending = ATOMIC_INIT(1),
10453 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10455 unsigned long i = 0;
10459 init_waitqueue_head(&priv.wait);
10461 * Submit bios for the extent, splitting due to bio or stripe limits as
10464 while (cur < disk_io_size) {
10465 struct extent_map *em;
10466 struct btrfs_io_geometry geom;
10467 struct bio *bio = NULL;
10470 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10471 disk_io_size - cur);
10475 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10476 disk_bytenr + cur, &geom);
10477 free_extent_map(em);
10480 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10483 remaining = min(geom.len, disk_io_size - cur);
10484 while (bio || remaining) {
10485 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10488 bio = btrfs_bio_alloc(BIO_MAX_VECS);
10489 bio->bi_iter.bi_sector =
10490 (disk_bytenr + cur) >> SECTOR_SHIFT;
10491 bio->bi_end_io = btrfs_encoded_read_endio;
10492 bio->bi_private = &priv;
10493 bio->bi_opf = REQ_OP_READ;
10497 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10498 blk_status_t status;
10500 status = submit_encoded_read_bio(inode, bio, 0);
10502 WRITE_ONCE(priv.status, status);
10512 remaining -= bytes;
10517 if (atomic_dec_return(&priv.pending))
10518 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10519 /* See btrfs_encoded_read_endio() for ordering. */
10520 return blk_status_to_errno(READ_ONCE(priv.status));
10523 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10524 struct iov_iter *iter,
10525 u64 start, u64 lockend,
10526 struct extent_state **cached_state,
10527 u64 disk_bytenr, u64 disk_io_size,
10528 size_t count, bool compressed,
10531 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10532 struct extent_io_tree *io_tree = &inode->io_tree;
10533 struct page **pages;
10534 unsigned long nr_pages, i;
10536 size_t page_offset;
10539 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10540 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10543 ret = btrfs_alloc_page_array(nr_pages, pages);
10549 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10550 disk_io_size, pages);
10554 unlock_extent_cached(io_tree, start, lockend, cached_state);
10555 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10562 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10563 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10566 while (cur < count) {
10567 size_t bytes = min_t(size_t, count - cur,
10568 PAGE_SIZE - page_offset);
10570 if (copy_page_to_iter(pages[i], page_offset, bytes,
10581 for (i = 0; i < nr_pages; i++) {
10583 __free_page(pages[i]);
10589 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10590 struct btrfs_ioctl_encoded_io_args *encoded)
10592 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10593 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10594 struct extent_io_tree *io_tree = &inode->io_tree;
10596 size_t count = iov_iter_count(iter);
10597 u64 start, lockend, disk_bytenr, disk_io_size;
10598 struct extent_state *cached_state = NULL;
10599 struct extent_map *em;
10600 bool unlocked = false;
10602 file_accessed(iocb->ki_filp);
10604 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10606 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10607 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10610 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10612 * We don't know how long the extent containing iocb->ki_pos is, but if
10613 * it's compressed we know that it won't be longer than this.
10615 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10618 struct btrfs_ordered_extent *ordered;
10620 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10621 lockend - start + 1);
10623 goto out_unlock_inode;
10624 lock_extent_bits(io_tree, start, lockend, &cached_state);
10625 ordered = btrfs_lookup_ordered_range(inode, start,
10626 lockend - start + 1);
10629 btrfs_put_ordered_extent(ordered);
10630 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10634 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10637 goto out_unlock_extent;
10640 if (em->block_start == EXTENT_MAP_INLINE) {
10641 u64 extent_start = em->start;
10644 * For inline extents we get everything we need out of the
10647 free_extent_map(em);
10649 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10650 &cached_state, extent_start,
10651 count, encoded, &unlocked);
10656 * We only want to return up to EOF even if the extent extends beyond
10659 encoded->len = min_t(u64, extent_map_end(em),
10660 inode->vfs_inode.i_size) - iocb->ki_pos;
10661 if (em->block_start == EXTENT_MAP_HOLE ||
10662 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10663 disk_bytenr = EXTENT_MAP_HOLE;
10664 count = min_t(u64, count, encoded->len);
10665 encoded->len = count;
10666 encoded->unencoded_len = count;
10667 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10668 disk_bytenr = em->block_start;
10670 * Bail if the buffer isn't large enough to return the whole
10671 * compressed extent.
10673 if (em->block_len > count) {
10677 disk_io_size = em->block_len;
10678 count = em->block_len;
10679 encoded->unencoded_len = em->ram_bytes;
10680 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10681 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10682 em->compress_type);
10685 encoded->compression = ret;
10687 disk_bytenr = em->block_start + (start - em->start);
10688 if (encoded->len > count)
10689 encoded->len = count;
10691 * Don't read beyond what we locked. This also limits the page
10692 * allocations that we'll do.
10694 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10695 count = start + disk_io_size - iocb->ki_pos;
10696 encoded->len = count;
10697 encoded->unencoded_len = count;
10698 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10700 free_extent_map(em);
10703 if (disk_bytenr == EXTENT_MAP_HOLE) {
10704 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10705 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10707 ret = iov_iter_zero(count, iter);
10711 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10712 &cached_state, disk_bytenr,
10713 disk_io_size, count,
10714 encoded->compression,
10720 iocb->ki_pos += encoded->len;
10722 free_extent_map(em);
10725 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10728 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10732 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10733 const struct btrfs_ioctl_encoded_io_args *encoded)
10735 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10736 struct btrfs_root *root = inode->root;
10737 struct btrfs_fs_info *fs_info = root->fs_info;
10738 struct extent_io_tree *io_tree = &inode->io_tree;
10739 struct extent_changeset *data_reserved = NULL;
10740 struct extent_state *cached_state = NULL;
10744 u64 num_bytes, ram_bytes, disk_num_bytes;
10745 unsigned long nr_pages, i;
10746 struct page **pages;
10747 struct btrfs_key ins;
10748 bool extent_reserved = false;
10749 struct extent_map *em;
10752 switch (encoded->compression) {
10753 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10754 compression = BTRFS_COMPRESS_ZLIB;
10756 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10757 compression = BTRFS_COMPRESS_ZSTD;
10759 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10760 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10761 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10762 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10763 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10764 /* The sector size must match for LZO. */
10765 if (encoded->compression -
10766 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10767 fs_info->sectorsize_bits)
10769 compression = BTRFS_COMPRESS_LZO;
10774 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10777 orig_count = iov_iter_count(from);
10779 /* The extent size must be sane. */
10780 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10781 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10785 * The compressed data must be smaller than the decompressed data.
10787 * It's of course possible for data to compress to larger or the same
10788 * size, but the buffered I/O path falls back to no compression for such
10789 * data, and we don't want to break any assumptions by creating these
10792 * Note that this is less strict than the current check we have that the
10793 * compressed data must be at least one sector smaller than the
10794 * decompressed data. We only want to enforce the weaker requirement
10795 * from old kernels that it is at least one byte smaller.
10797 if (orig_count >= encoded->unencoded_len)
10800 /* The extent must start on a sector boundary. */
10801 start = iocb->ki_pos;
10802 if (!IS_ALIGNED(start, fs_info->sectorsize))
10806 * The extent must end on a sector boundary. However, we allow a write
10807 * which ends at or extends i_size to have an unaligned length; we round
10808 * up the extent size and set i_size to the unaligned end.
10810 if (start + encoded->len < inode->vfs_inode.i_size &&
10811 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10814 /* Finally, the offset in the unencoded data must be sector-aligned. */
10815 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10818 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10819 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10820 end = start + num_bytes - 1;
10823 * If the extent cannot be inline, the compressed data on disk must be
10824 * sector-aligned. For convenience, we extend it with zeroes if it
10827 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10828 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10829 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10832 for (i = 0; i < nr_pages; i++) {
10833 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10836 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10841 kaddr = kmap_local_page(pages[i]);
10842 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10843 kunmap_local(kaddr);
10847 if (bytes < PAGE_SIZE)
10848 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10849 kunmap_local(kaddr);
10853 struct btrfs_ordered_extent *ordered;
10855 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10858 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10859 start >> PAGE_SHIFT,
10860 end >> PAGE_SHIFT);
10863 lock_extent_bits(io_tree, start, end, &cached_state);
10864 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10866 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10869 btrfs_put_ordered_extent(ordered);
10870 unlock_extent_cached(io_tree, start, end, &cached_state);
10875 * We don't use the higher-level delalloc space functions because our
10876 * num_bytes and disk_num_bytes are different.
10878 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10881 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10883 goto out_free_data_space;
10884 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10887 goto out_qgroup_free_data;
10889 /* Try an inline extent first. */
10890 if (start == 0 && encoded->unencoded_len == encoded->len &&
10891 encoded->unencoded_offset == 0) {
10892 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10893 compression, pages, true);
10897 goto out_delalloc_release;
10901 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10902 disk_num_bytes, 0, 0, &ins, 1, 1);
10904 goto out_delalloc_release;
10905 extent_reserved = true;
10907 em = create_io_em(inode, start, num_bytes,
10908 start - encoded->unencoded_offset, ins.objectid,
10909 ins.offset, ins.offset, ram_bytes, compression,
10910 BTRFS_ORDERED_COMPRESSED);
10913 goto out_free_reserved;
10915 free_extent_map(em);
10917 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10918 ins.objectid, ins.offset,
10919 encoded->unencoded_offset,
10920 (1 << BTRFS_ORDERED_ENCODED) |
10921 (1 << BTRFS_ORDERED_COMPRESSED),
10924 btrfs_drop_extent_cache(inode, start, end, 0);
10925 goto out_free_reserved;
10927 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10929 if (start + encoded->len > inode->vfs_inode.i_size)
10930 i_size_write(&inode->vfs_inode, start + encoded->len);
10932 unlock_extent_cached(io_tree, start, end, &cached_state);
10934 btrfs_delalloc_release_extents(inode, num_bytes);
10936 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10937 ins.offset, pages, nr_pages, 0, NULL,
10939 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10947 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10948 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10949 out_delalloc_release:
10950 btrfs_delalloc_release_extents(inode, num_bytes);
10951 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10952 out_qgroup_free_data:
10954 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10955 out_free_data_space:
10957 * If btrfs_reserve_extent() succeeded, then we already decremented
10960 if (!extent_reserved)
10961 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10963 unlock_extent_cached(io_tree, start, end, &cached_state);
10965 for (i = 0; i < nr_pages; i++) {
10967 __free_page(pages[i]);
10972 iocb->ki_pos += encoded->len;
10978 * Add an entry indicating a block group or device which is pinned by a
10979 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10980 * negative errno on failure.
10982 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10983 bool is_block_group)
10985 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10986 struct btrfs_swapfile_pin *sp, *entry;
10987 struct rb_node **p;
10988 struct rb_node *parent = NULL;
10990 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10995 sp->is_block_group = is_block_group;
10996 sp->bg_extent_count = 1;
10998 spin_lock(&fs_info->swapfile_pins_lock);
10999 p = &fs_info->swapfile_pins.rb_node;
11002 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
11003 if (sp->ptr < entry->ptr ||
11004 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
11005 p = &(*p)->rb_left;
11006 } else if (sp->ptr > entry->ptr ||
11007 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
11008 p = &(*p)->rb_right;
11010 if (is_block_group)
11011 entry->bg_extent_count++;
11012 spin_unlock(&fs_info->swapfile_pins_lock);
11017 rb_link_node(&sp->node, parent, p);
11018 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
11019 spin_unlock(&fs_info->swapfile_pins_lock);
11023 /* Free all of the entries pinned by this swapfile. */
11024 static void btrfs_free_swapfile_pins(struct inode *inode)
11026 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
11027 struct btrfs_swapfile_pin *sp;
11028 struct rb_node *node, *next;
11030 spin_lock(&fs_info->swapfile_pins_lock);
11031 node = rb_first(&fs_info->swapfile_pins);
11033 next = rb_next(node);
11034 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
11035 if (sp->inode == inode) {
11036 rb_erase(&sp->node, &fs_info->swapfile_pins);
11037 if (sp->is_block_group) {
11038 btrfs_dec_block_group_swap_extents(sp->ptr,
11039 sp->bg_extent_count);
11040 btrfs_put_block_group(sp->ptr);
11046 spin_unlock(&fs_info->swapfile_pins_lock);
11049 struct btrfs_swap_info {
11055 unsigned long nr_pages;
11059 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
11060 struct btrfs_swap_info *bsi)
11062 unsigned long nr_pages;
11063 unsigned long max_pages;
11064 u64 first_ppage, first_ppage_reported, next_ppage;
11068 * Our swapfile may have had its size extended after the swap header was
11069 * written. In that case activating the swapfile should not go beyond
11070 * the max size set in the swap header.
11072 if (bsi->nr_pages >= sis->max)
11075 max_pages = sis->max - bsi->nr_pages;
11076 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
11077 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
11078 PAGE_SIZE) >> PAGE_SHIFT;
11080 if (first_ppage >= next_ppage)
11082 nr_pages = next_ppage - first_ppage;
11083 nr_pages = min(nr_pages, max_pages);
11085 first_ppage_reported = first_ppage;
11086 if (bsi->start == 0)
11087 first_ppage_reported++;
11088 if (bsi->lowest_ppage > first_ppage_reported)
11089 bsi->lowest_ppage = first_ppage_reported;
11090 if (bsi->highest_ppage < (next_ppage - 1))
11091 bsi->highest_ppage = next_ppage - 1;
11093 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
11096 bsi->nr_extents += ret;
11097 bsi->nr_pages += nr_pages;
11101 static void btrfs_swap_deactivate(struct file *file)
11103 struct inode *inode = file_inode(file);
11105 btrfs_free_swapfile_pins(inode);
11106 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
11109 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11112 struct inode *inode = file_inode(file);
11113 struct btrfs_root *root = BTRFS_I(inode)->root;
11114 struct btrfs_fs_info *fs_info = root->fs_info;
11115 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
11116 struct extent_state *cached_state = NULL;
11117 struct extent_map *em = NULL;
11118 struct btrfs_device *device = NULL;
11119 struct btrfs_swap_info bsi = {
11120 .lowest_ppage = (sector_t)-1ULL,
11127 * If the swap file was just created, make sure delalloc is done. If the
11128 * file changes again after this, the user is doing something stupid and
11129 * we don't really care.
11131 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
11136 * The inode is locked, so these flags won't change after we check them.
11138 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
11139 btrfs_warn(fs_info, "swapfile must not be compressed");
11142 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
11143 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
11146 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
11147 btrfs_warn(fs_info, "swapfile must not be checksummed");
11152 * Balance or device remove/replace/resize can move stuff around from
11153 * under us. The exclop protection makes sure they aren't running/won't
11154 * run concurrently while we are mapping the swap extents, and
11155 * fs_info->swapfile_pins prevents them from running while the swap
11156 * file is active and moving the extents. Note that this also prevents
11157 * a concurrent device add which isn't actually necessary, but it's not
11158 * really worth the trouble to allow it.
11160 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
11161 btrfs_warn(fs_info,
11162 "cannot activate swapfile while exclusive operation is running");
11167 * Prevent snapshot creation while we are activating the swap file.
11168 * We do not want to race with snapshot creation. If snapshot creation
11169 * already started before we bumped nr_swapfiles from 0 to 1 and
11170 * completes before the first write into the swap file after it is
11171 * activated, than that write would fallback to COW.
11173 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11174 btrfs_exclop_finish(fs_info);
11175 btrfs_warn(fs_info,
11176 "cannot activate swapfile because snapshot creation is in progress");
11180 * Snapshots can create extents which require COW even if NODATACOW is
11181 * set. We use this counter to prevent snapshots. We must increment it
11182 * before walking the extents because we don't want a concurrent
11183 * snapshot to run after we've already checked the extents.
11185 * It is possible that subvolume is marked for deletion but still not
11186 * removed yet. To prevent this race, we check the root status before
11187 * activating the swapfile.
11189 spin_lock(&root->root_item_lock);
11190 if (btrfs_root_dead(root)) {
11191 spin_unlock(&root->root_item_lock);
11193 btrfs_exclop_finish(fs_info);
11194 btrfs_warn(fs_info,
11195 "cannot activate swapfile because subvolume %llu is being deleted",
11196 root->root_key.objectid);
11199 atomic_inc(&root->nr_swapfiles);
11200 spin_unlock(&root->root_item_lock);
11202 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11204 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
11206 while (start < isize) {
11207 u64 logical_block_start, physical_block_start;
11208 struct btrfs_block_group *bg;
11209 u64 len = isize - start;
11211 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11217 if (em->block_start == EXTENT_MAP_HOLE) {
11218 btrfs_warn(fs_info, "swapfile must not have holes");
11222 if (em->block_start == EXTENT_MAP_INLINE) {
11224 * It's unlikely we'll ever actually find ourselves
11225 * here, as a file small enough to fit inline won't be
11226 * big enough to store more than the swap header, but in
11227 * case something changes in the future, let's catch it
11228 * here rather than later.
11230 btrfs_warn(fs_info, "swapfile must not be inline");
11234 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11235 btrfs_warn(fs_info, "swapfile must not be compressed");
11240 logical_block_start = em->block_start + (start - em->start);
11241 len = min(len, em->len - (start - em->start));
11242 free_extent_map(em);
11245 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
11251 btrfs_warn(fs_info,
11252 "swapfile must not be copy-on-write");
11257 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11263 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11264 btrfs_warn(fs_info,
11265 "swapfile must have single data profile");
11270 if (device == NULL) {
11271 device = em->map_lookup->stripes[0].dev;
11272 ret = btrfs_add_swapfile_pin(inode, device, false);
11277 } else if (device != em->map_lookup->stripes[0].dev) {
11278 btrfs_warn(fs_info, "swapfile must be on one device");
11283 physical_block_start = (em->map_lookup->stripes[0].physical +
11284 (logical_block_start - em->start));
11285 len = min(len, em->len - (logical_block_start - em->start));
11286 free_extent_map(em);
11289 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11291 btrfs_warn(fs_info,
11292 "could not find block group containing swapfile");
11297 if (!btrfs_inc_block_group_swap_extents(bg)) {
11298 btrfs_warn(fs_info,
11299 "block group for swapfile at %llu is read-only%s",
11301 atomic_read(&fs_info->scrubs_running) ?
11302 " (scrub running)" : "");
11303 btrfs_put_block_group(bg);
11308 ret = btrfs_add_swapfile_pin(inode, bg, true);
11310 btrfs_put_block_group(bg);
11317 if (bsi.block_len &&
11318 bsi.block_start + bsi.block_len == physical_block_start) {
11319 bsi.block_len += len;
11321 if (bsi.block_len) {
11322 ret = btrfs_add_swap_extent(sis, &bsi);
11327 bsi.block_start = physical_block_start;
11328 bsi.block_len = len;
11335 ret = btrfs_add_swap_extent(sis, &bsi);
11338 if (!IS_ERR_OR_NULL(em))
11339 free_extent_map(em);
11341 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
11344 btrfs_swap_deactivate(file);
11346 btrfs_drew_write_unlock(&root->snapshot_lock);
11348 btrfs_exclop_finish(fs_info);
11354 sis->bdev = device->bdev;
11355 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11356 sis->max = bsi.nr_pages;
11357 sis->pages = bsi.nr_pages - 1;
11358 sis->highest_bit = bsi.nr_pages - 1;
11359 return bsi.nr_extents;
11362 static void btrfs_swap_deactivate(struct file *file)
11366 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11369 return -EOPNOTSUPP;
11374 * Update the number of bytes used in the VFS' inode. When we replace extents in
11375 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11376 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11377 * always get a correct value.
11379 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11380 const u64 add_bytes,
11381 const u64 del_bytes)
11383 if (add_bytes == del_bytes)
11386 spin_lock(&inode->lock);
11388 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11390 inode_add_bytes(&inode->vfs_inode, add_bytes);
11391 spin_unlock(&inode->lock);
11395 * Verify that there are no ordered extents for a given file range.
11397 * @inode: The target inode.
11398 * @start: Start offset of the file range, should be sector size aligned.
11399 * @end: End offset (inclusive) of the file range, its value +1 should be
11400 * sector size aligned.
11402 * This should typically be used for cases where we locked an inode's VFS lock in
11403 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11404 * we have flushed all delalloc in the range, we have waited for all ordered
11405 * extents in the range to complete and finally we have locked the file range in
11406 * the inode's io_tree.
11408 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11410 struct btrfs_root *root = inode->root;
11411 struct btrfs_ordered_extent *ordered;
11413 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11416 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11418 btrfs_err(root->fs_info,
11419 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11420 start, end, btrfs_ino(inode), root->root_key.objectid,
11421 ordered->file_offset,
11422 ordered->file_offset + ordered->num_bytes - 1);
11423 btrfs_put_ordered_extent(ordered);
11426 ASSERT(ordered == NULL);
11429 static const struct inode_operations btrfs_dir_inode_operations = {
11430 .getattr = btrfs_getattr,
11431 .lookup = btrfs_lookup,
11432 .create = btrfs_create,
11433 .unlink = btrfs_unlink,
11434 .link = btrfs_link,
11435 .mkdir = btrfs_mkdir,
11436 .rmdir = btrfs_rmdir,
11437 .rename = btrfs_rename2,
11438 .symlink = btrfs_symlink,
11439 .setattr = btrfs_setattr,
11440 .mknod = btrfs_mknod,
11441 .listxattr = btrfs_listxattr,
11442 .permission = btrfs_permission,
11443 .get_acl = btrfs_get_acl,
11444 .set_acl = btrfs_set_acl,
11445 .update_time = btrfs_update_time,
11446 .tmpfile = btrfs_tmpfile,
11447 .fileattr_get = btrfs_fileattr_get,
11448 .fileattr_set = btrfs_fileattr_set,
11451 static const struct file_operations btrfs_dir_file_operations = {
11452 .llseek = generic_file_llseek,
11453 .read = generic_read_dir,
11454 .iterate_shared = btrfs_real_readdir,
11455 .open = btrfs_opendir,
11456 .unlocked_ioctl = btrfs_ioctl,
11457 #ifdef CONFIG_COMPAT
11458 .compat_ioctl = btrfs_compat_ioctl,
11460 .release = btrfs_release_file,
11461 .fsync = btrfs_sync_file,
11465 * btrfs doesn't support the bmap operation because swapfiles
11466 * use bmap to make a mapping of extents in the file. They assume
11467 * these extents won't change over the life of the file and they
11468 * use the bmap result to do IO directly to the drive.
11470 * the btrfs bmap call would return logical addresses that aren't
11471 * suitable for IO and they also will change frequently as COW
11472 * operations happen. So, swapfile + btrfs == corruption.
11474 * For now we're avoiding this by dropping bmap.
11476 static const struct address_space_operations btrfs_aops = {
11477 .read_folio = btrfs_read_folio,
11478 .writepages = btrfs_writepages,
11479 .readahead = btrfs_readahead,
11480 .direct_IO = noop_direct_IO,
11481 .invalidate_folio = btrfs_invalidate_folio,
11482 .release_folio = btrfs_release_folio,
11483 .migrate_folio = btrfs_migrate_folio,
11484 .dirty_folio = filemap_dirty_folio,
11485 .error_remove_page = generic_error_remove_page,
11486 .swap_activate = btrfs_swap_activate,
11487 .swap_deactivate = btrfs_swap_deactivate,
11490 static const struct inode_operations btrfs_file_inode_operations = {
11491 .getattr = btrfs_getattr,
11492 .setattr = btrfs_setattr,
11493 .listxattr = btrfs_listxattr,
11494 .permission = btrfs_permission,
11495 .fiemap = btrfs_fiemap,
11496 .get_acl = btrfs_get_acl,
11497 .set_acl = btrfs_set_acl,
11498 .update_time = btrfs_update_time,
11499 .fileattr_get = btrfs_fileattr_get,
11500 .fileattr_set = btrfs_fileattr_set,
11502 static const struct inode_operations btrfs_special_inode_operations = {
11503 .getattr = btrfs_getattr,
11504 .setattr = btrfs_setattr,
11505 .permission = btrfs_permission,
11506 .listxattr = btrfs_listxattr,
11507 .get_acl = btrfs_get_acl,
11508 .set_acl = btrfs_set_acl,
11509 .update_time = btrfs_update_time,
11511 static const struct inode_operations btrfs_symlink_inode_operations = {
11512 .get_link = page_get_link,
11513 .getattr = btrfs_getattr,
11514 .setattr = btrfs_setattr,
11515 .permission = btrfs_permission,
11516 .listxattr = btrfs_listxattr,
11517 .update_time = btrfs_update_time,
11520 const struct dentry_operations btrfs_dentry_operations = {
11521 .d_delete = btrfs_dentry_delete,