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, NULL);
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, NULL);
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_map_range(inode, start, end, false);
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);
1259 * Relocation relies on the relocated extents to have exactly the same
1260 * size as the original extents. Normally writeback for relocation data
1261 * extents follows a NOCOW path because relocation preallocates the
1262 * extents. However, due to an operation such as scrub turning a block
1263 * group to RO mode, it may fallback to COW mode, so we must make sure
1264 * an extent allocated during COW has exactly the requested size and can
1265 * not be split into smaller extents, otherwise relocation breaks and
1266 * fails during the stage where it updates the bytenr of file extent
1269 if (btrfs_is_data_reloc_root(root))
1270 min_alloc_size = num_bytes;
1272 min_alloc_size = fs_info->sectorsize;
1274 while (num_bytes > 0) {
1275 cur_alloc_size = num_bytes;
1276 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1277 min_alloc_size, 0, alloc_hint,
1281 cur_alloc_size = ins.offset;
1282 extent_reserved = true;
1284 ram_size = ins.offset;
1285 em = create_io_em(inode, start, ins.offset, /* len */
1286 start, /* orig_start */
1287 ins.objectid, /* block_start */
1288 ins.offset, /* block_len */
1289 ins.offset, /* orig_block_len */
1290 ram_size, /* ram_bytes */
1291 BTRFS_COMPRESS_NONE, /* compress_type */
1292 BTRFS_ORDERED_REGULAR /* type */);
1297 free_extent_map(em);
1299 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1300 ins.objectid, cur_alloc_size, 0,
1301 1 << BTRFS_ORDERED_REGULAR,
1302 BTRFS_COMPRESS_NONE);
1304 goto out_drop_extent_cache;
1306 if (btrfs_is_data_reloc_root(root)) {
1307 ret = btrfs_reloc_clone_csums(inode, start,
1310 * Only drop cache here, and process as normal.
1312 * We must not allow extent_clear_unlock_delalloc()
1313 * at out_unlock label to free meta of this ordered
1314 * extent, as its meta should be freed by
1315 * btrfs_finish_ordered_io().
1317 * So we must continue until @start is increased to
1318 * skip current ordered extent.
1321 btrfs_drop_extent_map_range(inode, start,
1322 start + ram_size - 1,
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_map_range(inode, start, start + ram_size - 1, false);
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;
1435 * For the range (3). We never touched the region. In addition to the
1436 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1437 * space_info's bytes_may_use counter, reserved in
1438 * btrfs_check_data_free_space().
1441 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1442 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1443 clear_bits, page_ops);
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, NULL);
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, bool nowait)
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,
1679 if (ret == 0 && list_empty(&list))
1682 while (!list_empty(&list)) {
1683 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1684 list_del(&sums->list);
1692 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1693 const u64 start, const u64 end,
1694 int *page_started, unsigned long *nr_written)
1696 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1697 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1698 const u64 range_bytes = end + 1 - start;
1699 struct extent_io_tree *io_tree = &inode->io_tree;
1700 u64 range_start = start;
1704 * If EXTENT_NORESERVE is set it means that when the buffered write was
1705 * made we had not enough available data space and therefore we did not
1706 * reserve data space for it, since we though we could do NOCOW for the
1707 * respective file range (either there is prealloc extent or the inode
1708 * has the NOCOW bit set).
1710 * However when we need to fallback to COW mode (because for example the
1711 * block group for the corresponding extent was turned to RO mode by a
1712 * scrub or relocation) we need to do the following:
1714 * 1) We increment the bytes_may_use counter of the data space info.
1715 * If COW succeeds, it allocates a new data extent and after doing
1716 * that it decrements the space info's bytes_may_use counter and
1717 * increments its bytes_reserved counter by the same amount (we do
1718 * this at btrfs_add_reserved_bytes()). So we need to increment the
1719 * bytes_may_use counter to compensate (when space is reserved at
1720 * buffered write time, the bytes_may_use counter is incremented);
1722 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1723 * that if the COW path fails for any reason, it decrements (through
1724 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1725 * data space info, which we incremented in the step above.
1727 * If we need to fallback to cow and the inode corresponds to a free
1728 * space cache inode or an inode of the data relocation tree, we must
1729 * also increment bytes_may_use of the data space_info for the same
1730 * reason. Space caches and relocated data extents always get a prealloc
1731 * extent for them, however scrub or balance may have set the block
1732 * group that contains that extent to RO mode and therefore force COW
1733 * when starting writeback.
1735 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1736 EXTENT_NORESERVE, 0);
1737 if (count > 0 || is_space_ino || is_reloc_ino) {
1739 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1740 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1742 if (is_space_ino || is_reloc_ino)
1743 bytes = range_bytes;
1745 spin_lock(&sinfo->lock);
1746 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1747 spin_unlock(&sinfo->lock);
1750 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1754 return cow_file_range(inode, locked_page, start, end, page_started,
1755 nr_written, 1, NULL);
1758 struct can_nocow_file_extent_args {
1761 /* Start file offset of the range we want to NOCOW. */
1763 /* End file offset (inclusive) of the range we want to NOCOW. */
1765 bool writeback_path;
1768 * Free the path passed to can_nocow_file_extent() once it's not needed
1773 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1778 /* Number of bytes that can be written to in NOCOW mode. */
1783 * Check if we can NOCOW the file extent that the path points to.
1784 * This function may return with the path released, so the caller should check
1785 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1787 * Returns: < 0 on error
1788 * 0 if we can not NOCOW
1791 static int can_nocow_file_extent(struct btrfs_path *path,
1792 struct btrfs_key *key,
1793 struct btrfs_inode *inode,
1794 struct can_nocow_file_extent_args *args)
1796 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1797 struct extent_buffer *leaf = path->nodes[0];
1798 struct btrfs_root *root = inode->root;
1799 struct btrfs_file_extent_item *fi;
1804 bool nowait = path->nowait;
1806 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1807 extent_type = btrfs_file_extent_type(leaf, fi);
1809 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1812 /* Can't access these fields unless we know it's not an inline extent. */
1813 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1814 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1815 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1817 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1818 extent_type == BTRFS_FILE_EXTENT_REG)
1822 * If the extent was created before the generation where the last snapshot
1823 * for its subvolume was created, then this implies the extent is shared,
1824 * hence we must COW.
1826 if (!args->strict &&
1827 btrfs_file_extent_generation(leaf, fi) <=
1828 btrfs_root_last_snapshot(&root->root_item))
1831 /* An explicit hole, must COW. */
1832 if (args->disk_bytenr == 0)
1835 /* Compressed/encrypted/encoded extents must be COWed. */
1836 if (btrfs_file_extent_compression(leaf, fi) ||
1837 btrfs_file_extent_encryption(leaf, fi) ||
1838 btrfs_file_extent_other_encoding(leaf, fi))
1841 extent_end = btrfs_file_extent_end(path);
1844 * The following checks can be expensive, as they need to take other
1845 * locks and do btree or rbtree searches, so release the path to avoid
1846 * blocking other tasks for too long.
1848 btrfs_release_path(path);
1850 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1851 key->offset - args->extent_offset,
1852 args->disk_bytenr, args->strict, path);
1853 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1857 if (args->free_path) {
1859 * We don't need the path anymore, plus through the
1860 * csum_exist_in_range() call below we will end up allocating
1861 * another path. So free the path to avoid unnecessary extra
1864 btrfs_free_path(path);
1868 /* If there are pending snapshots for this root, we must COW. */
1869 if (args->writeback_path && !is_freespace_inode &&
1870 atomic_read(&root->snapshot_force_cow))
1873 args->disk_bytenr += args->extent_offset;
1874 args->disk_bytenr += args->start - key->offset;
1875 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1878 * Force COW if csums exist in the range. This ensures that csums for a
1879 * given extent are either valid or do not exist.
1881 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1883 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1889 if (args->free_path && path)
1890 btrfs_free_path(path);
1892 return ret < 0 ? ret : can_nocow;
1896 * when nowcow writeback call back. This checks for snapshots or COW copies
1897 * of the extents that exist in the file, and COWs the file as required.
1899 * If no cow copies or snapshots exist, we write directly to the existing
1902 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1903 struct page *locked_page,
1904 const u64 start, const u64 end,
1906 unsigned long *nr_written)
1908 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1909 struct btrfs_root *root = inode->root;
1910 struct btrfs_path *path;
1911 u64 cow_start = (u64)-1;
1912 u64 cur_offset = start;
1914 bool check_prev = true;
1915 u64 ino = btrfs_ino(inode);
1916 struct btrfs_block_group *bg;
1918 struct can_nocow_file_extent_args nocow_args = { 0 };
1920 path = btrfs_alloc_path();
1922 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1923 EXTENT_LOCKED | EXTENT_DELALLOC |
1924 EXTENT_DO_ACCOUNTING |
1925 EXTENT_DEFRAG, PAGE_UNLOCK |
1926 PAGE_START_WRITEBACK |
1927 PAGE_END_WRITEBACK);
1931 nocow_args.end = end;
1932 nocow_args.writeback_path = true;
1935 struct btrfs_key found_key;
1936 struct btrfs_file_extent_item *fi;
1937 struct extent_buffer *leaf;
1945 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1951 * If there is no extent for our range when doing the initial
1952 * search, then go back to the previous slot as it will be the
1953 * one containing the search offset
1955 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1956 leaf = path->nodes[0];
1957 btrfs_item_key_to_cpu(leaf, &found_key,
1958 path->slots[0] - 1);
1959 if (found_key.objectid == ino &&
1960 found_key.type == BTRFS_EXTENT_DATA_KEY)
1965 /* Go to next leaf if we have exhausted the current one */
1966 leaf = path->nodes[0];
1967 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1968 ret = btrfs_next_leaf(root, path);
1970 if (cow_start != (u64)-1)
1971 cur_offset = cow_start;
1976 leaf = path->nodes[0];
1979 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1981 /* Didn't find anything for our INO */
1982 if (found_key.objectid > ino)
1985 * Keep searching until we find an EXTENT_ITEM or there are no
1986 * more extents for this inode
1988 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1989 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1994 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1995 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1996 found_key.offset > end)
2000 * If the found extent starts after requested offset, then
2001 * adjust extent_end to be right before this extent begins
2003 if (found_key.offset > cur_offset) {
2004 extent_end = found_key.offset;
2010 * Found extent which begins before our range and potentially
2013 fi = btrfs_item_ptr(leaf, path->slots[0],
2014 struct btrfs_file_extent_item);
2015 extent_type = btrfs_file_extent_type(leaf, fi);
2016 /* If this is triggered then we have a memory corruption. */
2017 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2018 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2022 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2023 extent_end = btrfs_file_extent_end(path);
2026 * If the extent we got ends before our current offset, skip to
2029 if (extent_end <= cur_offset) {
2034 nocow_args.start = cur_offset;
2035 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2037 if (cow_start != (u64)-1)
2038 cur_offset = cow_start;
2040 } else if (ret == 0) {
2045 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2050 * If nocow is false then record the beginning of the range
2051 * that needs to be COWed
2054 if (cow_start == (u64)-1)
2055 cow_start = cur_offset;
2056 cur_offset = extent_end;
2057 if (cur_offset > end)
2059 if (!path->nodes[0])
2066 * COW range from cow_start to found_key.offset - 1. As the key
2067 * will contain the beginning of the first extent that can be
2068 * NOCOW, following one which needs to be COW'ed
2070 if (cow_start != (u64)-1) {
2071 ret = fallback_to_cow(inode, locked_page,
2072 cow_start, found_key.offset - 1,
2073 page_started, nr_written);
2076 cow_start = (u64)-1;
2079 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2081 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2082 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2083 struct extent_map *em;
2085 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2087 nocow_args.disk_bytenr, /* block_start */
2088 nocow_args.num_bytes, /* block_len */
2089 nocow_args.disk_num_bytes, /* orig_block_len */
2090 ram_bytes, BTRFS_COMPRESS_NONE,
2091 BTRFS_ORDERED_PREALLOC);
2096 free_extent_map(em);
2097 ret = btrfs_add_ordered_extent(inode,
2098 cur_offset, nocow_args.num_bytes,
2099 nocow_args.num_bytes,
2100 nocow_args.disk_bytenr,
2101 nocow_args.num_bytes, 0,
2102 1 << BTRFS_ORDERED_PREALLOC,
2103 BTRFS_COMPRESS_NONE);
2105 btrfs_drop_extent_map_range(inode, cur_offset,
2110 ret = btrfs_add_ordered_extent(inode, cur_offset,
2111 nocow_args.num_bytes,
2112 nocow_args.num_bytes,
2113 nocow_args.disk_bytenr,
2114 nocow_args.num_bytes,
2116 1 << BTRFS_ORDERED_NOCOW,
2117 BTRFS_COMPRESS_NONE);
2123 btrfs_dec_nocow_writers(bg);
2127 if (btrfs_is_data_reloc_root(root))
2129 * Error handled later, as we must prevent
2130 * extent_clear_unlock_delalloc() in error handler
2131 * from freeing metadata of created ordered extent.
2133 ret = btrfs_reloc_clone_csums(inode, cur_offset,
2134 nocow_args.num_bytes);
2136 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2137 locked_page, EXTENT_LOCKED |
2139 EXTENT_CLEAR_DATA_RESV,
2140 PAGE_UNLOCK | PAGE_SET_ORDERED);
2142 cur_offset = extent_end;
2145 * btrfs_reloc_clone_csums() error, now we're OK to call error
2146 * handler, as metadata for created ordered extent will only
2147 * be freed by btrfs_finish_ordered_io().
2151 if (cur_offset > end)
2154 btrfs_release_path(path);
2156 if (cur_offset <= end && cow_start == (u64)-1)
2157 cow_start = cur_offset;
2159 if (cow_start != (u64)-1) {
2161 ret = fallback_to_cow(inode, locked_page, cow_start, end,
2162 page_started, nr_written);
2169 btrfs_dec_nocow_writers(bg);
2171 if (ret && cur_offset < end)
2172 extent_clear_unlock_delalloc(inode, cur_offset, end,
2173 locked_page, EXTENT_LOCKED |
2174 EXTENT_DELALLOC | EXTENT_DEFRAG |
2175 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2176 PAGE_START_WRITEBACK |
2177 PAGE_END_WRITEBACK);
2178 btrfs_free_path(path);
2182 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2184 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2185 if (inode->defrag_bytes &&
2186 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2195 * Function to process delayed allocation (create CoW) for ranges which are
2196 * being touched for the first time.
2198 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2199 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2200 struct writeback_control *wbc)
2203 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2206 * The range must cover part of the @locked_page, or the returned
2207 * @page_started can confuse the caller.
2209 ASSERT(!(end <= page_offset(locked_page) ||
2210 start >= page_offset(locked_page) + PAGE_SIZE));
2212 if (should_nocow(inode, start, end)) {
2214 * Normally on a zoned device we're only doing COW writes, but
2215 * in case of relocation on a zoned filesystem we have taken
2216 * precaution, that we're only writing sequentially. It's safe
2217 * to use run_delalloc_nocow() here, like for regular
2218 * preallocated inodes.
2220 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2221 ret = run_delalloc_nocow(inode, locked_page, start, end,
2222 page_started, nr_written);
2223 } else if (!btrfs_inode_can_compress(inode) ||
2224 !inode_need_compress(inode, start, end)) {
2226 ret = run_delalloc_zoned(inode, locked_page, start, end,
2227 page_started, nr_written);
2229 ret = cow_file_range(inode, locked_page, start, end,
2230 page_started, nr_written, 1, NULL);
2232 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2233 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2234 page_started, nr_written);
2238 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2243 void btrfs_split_delalloc_extent(struct inode *inode,
2244 struct extent_state *orig, u64 split)
2246 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2249 /* not delalloc, ignore it */
2250 if (!(orig->state & EXTENT_DELALLOC))
2253 size = orig->end - orig->start + 1;
2254 if (size > fs_info->max_extent_size) {
2259 * See the explanation in btrfs_merge_delalloc_extent, the same
2260 * applies here, just in reverse.
2262 new_size = orig->end - split + 1;
2263 num_extents = count_max_extents(fs_info, new_size);
2264 new_size = split - orig->start;
2265 num_extents += count_max_extents(fs_info, new_size);
2266 if (count_max_extents(fs_info, size) >= num_extents)
2270 spin_lock(&BTRFS_I(inode)->lock);
2271 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2272 spin_unlock(&BTRFS_I(inode)->lock);
2276 * Handle merged delayed allocation extents so we can keep track of new extents
2277 * that are just merged onto old extents, such as when we are doing sequential
2278 * writes, so we can properly account for the metadata space we'll need.
2280 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2281 struct extent_state *other)
2283 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2284 u64 new_size, old_size;
2287 /* not delalloc, ignore it */
2288 if (!(other->state & EXTENT_DELALLOC))
2291 if (new->start > other->start)
2292 new_size = new->end - other->start + 1;
2294 new_size = other->end - new->start + 1;
2296 /* we're not bigger than the max, unreserve the space and go */
2297 if (new_size <= fs_info->max_extent_size) {
2298 spin_lock(&BTRFS_I(inode)->lock);
2299 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2300 spin_unlock(&BTRFS_I(inode)->lock);
2305 * We have to add up either side to figure out how many extents were
2306 * accounted for before we merged into one big extent. If the number of
2307 * extents we accounted for is <= the amount we need for the new range
2308 * then we can return, otherwise drop. Think of it like this
2312 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2313 * need 2 outstanding extents, on one side we have 1 and the other side
2314 * we have 1 so they are == and we can return. But in this case
2316 * [MAX_SIZE+4k][MAX_SIZE+4k]
2318 * Each range on their own accounts for 2 extents, but merged together
2319 * they are only 3 extents worth of accounting, so we need to drop in
2322 old_size = other->end - other->start + 1;
2323 num_extents = count_max_extents(fs_info, old_size);
2324 old_size = new->end - new->start + 1;
2325 num_extents += count_max_extents(fs_info, old_size);
2326 if (count_max_extents(fs_info, new_size) >= num_extents)
2329 spin_lock(&BTRFS_I(inode)->lock);
2330 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2331 spin_unlock(&BTRFS_I(inode)->lock);
2334 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2335 struct inode *inode)
2337 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2339 spin_lock(&root->delalloc_lock);
2340 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2341 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2342 &root->delalloc_inodes);
2343 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2344 &BTRFS_I(inode)->runtime_flags);
2345 root->nr_delalloc_inodes++;
2346 if (root->nr_delalloc_inodes == 1) {
2347 spin_lock(&fs_info->delalloc_root_lock);
2348 BUG_ON(!list_empty(&root->delalloc_root));
2349 list_add_tail(&root->delalloc_root,
2350 &fs_info->delalloc_roots);
2351 spin_unlock(&fs_info->delalloc_root_lock);
2354 spin_unlock(&root->delalloc_lock);
2358 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2359 struct btrfs_inode *inode)
2361 struct btrfs_fs_info *fs_info = root->fs_info;
2363 if (!list_empty(&inode->delalloc_inodes)) {
2364 list_del_init(&inode->delalloc_inodes);
2365 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2366 &inode->runtime_flags);
2367 root->nr_delalloc_inodes--;
2368 if (!root->nr_delalloc_inodes) {
2369 ASSERT(list_empty(&root->delalloc_inodes));
2370 spin_lock(&fs_info->delalloc_root_lock);
2371 BUG_ON(list_empty(&root->delalloc_root));
2372 list_del_init(&root->delalloc_root);
2373 spin_unlock(&fs_info->delalloc_root_lock);
2378 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2379 struct btrfs_inode *inode)
2381 spin_lock(&root->delalloc_lock);
2382 __btrfs_del_delalloc_inode(root, inode);
2383 spin_unlock(&root->delalloc_lock);
2387 * Properly track delayed allocation bytes in the inode and to maintain the
2388 * list of inodes that have pending delalloc work to be done.
2390 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2393 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2395 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2398 * set_bit and clear bit hooks normally require _irqsave/restore
2399 * but in this case, we are only testing for the DELALLOC
2400 * bit, which is only set or cleared with irqs on
2402 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2403 struct btrfs_root *root = BTRFS_I(inode)->root;
2404 u64 len = state->end + 1 - state->start;
2405 u32 num_extents = count_max_extents(fs_info, len);
2406 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2408 spin_lock(&BTRFS_I(inode)->lock);
2409 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2410 spin_unlock(&BTRFS_I(inode)->lock);
2412 /* For sanity tests */
2413 if (btrfs_is_testing(fs_info))
2416 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2417 fs_info->delalloc_batch);
2418 spin_lock(&BTRFS_I(inode)->lock);
2419 BTRFS_I(inode)->delalloc_bytes += len;
2420 if (bits & EXTENT_DEFRAG)
2421 BTRFS_I(inode)->defrag_bytes += len;
2422 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2423 &BTRFS_I(inode)->runtime_flags))
2424 btrfs_add_delalloc_inodes(root, inode);
2425 spin_unlock(&BTRFS_I(inode)->lock);
2428 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2429 (bits & EXTENT_DELALLOC_NEW)) {
2430 spin_lock(&BTRFS_I(inode)->lock);
2431 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2433 spin_unlock(&BTRFS_I(inode)->lock);
2438 * Once a range is no longer delalloc this function ensures that proper
2439 * accounting happens.
2441 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2442 struct extent_state *state, u32 bits)
2444 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2445 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2446 u64 len = state->end + 1 - state->start;
2447 u32 num_extents = count_max_extents(fs_info, len);
2449 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2450 spin_lock(&inode->lock);
2451 inode->defrag_bytes -= len;
2452 spin_unlock(&inode->lock);
2456 * set_bit and clear bit hooks normally require _irqsave/restore
2457 * but in this case, we are only testing for the DELALLOC
2458 * bit, which is only set or cleared with irqs on
2460 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2461 struct btrfs_root *root = inode->root;
2462 bool do_list = !btrfs_is_free_space_inode(inode);
2464 spin_lock(&inode->lock);
2465 btrfs_mod_outstanding_extents(inode, -num_extents);
2466 spin_unlock(&inode->lock);
2469 * We don't reserve metadata space for space cache inodes so we
2470 * don't need to call delalloc_release_metadata if there is an
2473 if (bits & EXTENT_CLEAR_META_RESV &&
2474 root != fs_info->tree_root)
2475 btrfs_delalloc_release_metadata(inode, len, false);
2477 /* For sanity tests. */
2478 if (btrfs_is_testing(fs_info))
2481 if (!btrfs_is_data_reloc_root(root) &&
2482 do_list && !(state->state & EXTENT_NORESERVE) &&
2483 (bits & EXTENT_CLEAR_DATA_RESV))
2484 btrfs_free_reserved_data_space_noquota(fs_info, len);
2486 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2487 fs_info->delalloc_batch);
2488 spin_lock(&inode->lock);
2489 inode->delalloc_bytes -= len;
2490 if (do_list && inode->delalloc_bytes == 0 &&
2491 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2492 &inode->runtime_flags))
2493 btrfs_del_delalloc_inode(root, inode);
2494 spin_unlock(&inode->lock);
2497 if ((state->state & EXTENT_DELALLOC_NEW) &&
2498 (bits & EXTENT_DELALLOC_NEW)) {
2499 spin_lock(&inode->lock);
2500 ASSERT(inode->new_delalloc_bytes >= len);
2501 inode->new_delalloc_bytes -= len;
2502 if (bits & EXTENT_ADD_INODE_BYTES)
2503 inode_add_bytes(&inode->vfs_inode, len);
2504 spin_unlock(&inode->lock);
2509 * in order to insert checksums into the metadata in large chunks,
2510 * we wait until bio submission time. All the pages in the bio are
2511 * checksummed and sums are attached onto the ordered extent record.
2513 * At IO completion time the cums attached on the ordered extent record
2514 * are inserted into the btree
2516 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2517 u64 dio_file_offset)
2519 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2523 * Split an extent_map at [start, start + len]
2525 * This function is intended to be used only for extract_ordered_extent().
2527 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2530 struct extent_map_tree *em_tree = &inode->extent_tree;
2531 struct extent_map *em;
2532 struct extent_map *split_pre = NULL;
2533 struct extent_map *split_mid = NULL;
2534 struct extent_map *split_post = NULL;
2536 unsigned long flags;
2539 if (pre == 0 && post == 0)
2542 split_pre = alloc_extent_map();
2544 split_mid = alloc_extent_map();
2546 split_post = alloc_extent_map();
2547 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2552 ASSERT(pre + post < len);
2554 lock_extent(&inode->io_tree, start, start + len - 1, NULL);
2555 write_lock(&em_tree->lock);
2556 em = lookup_extent_mapping(em_tree, start, len);
2562 ASSERT(em->len == len);
2563 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2564 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2565 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2566 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2567 ASSERT(!list_empty(&em->list));
2570 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2572 /* First, replace the em with a new extent_map starting from * em->start */
2573 split_pre->start = em->start;
2574 split_pre->len = (pre ? pre : em->len - post);
2575 split_pre->orig_start = split_pre->start;
2576 split_pre->block_start = em->block_start;
2577 split_pre->block_len = split_pre->len;
2578 split_pre->orig_block_len = split_pre->block_len;
2579 split_pre->ram_bytes = split_pre->len;
2580 split_pre->flags = flags;
2581 split_pre->compress_type = em->compress_type;
2582 split_pre->generation = em->generation;
2584 replace_extent_mapping(em_tree, em, split_pre, 1);
2587 * Now we only have an extent_map at:
2588 * [em->start, em->start + pre] if pre != 0
2589 * [em->start, em->start + em->len - post] if pre == 0
2593 /* Insert the middle extent_map */
2594 split_mid->start = em->start + pre;
2595 split_mid->len = em->len - pre - post;
2596 split_mid->orig_start = split_mid->start;
2597 split_mid->block_start = em->block_start + pre;
2598 split_mid->block_len = split_mid->len;
2599 split_mid->orig_block_len = split_mid->block_len;
2600 split_mid->ram_bytes = split_mid->len;
2601 split_mid->flags = flags;
2602 split_mid->compress_type = em->compress_type;
2603 split_mid->generation = em->generation;
2604 add_extent_mapping(em_tree, split_mid, 1);
2608 split_post->start = em->start + em->len - post;
2609 split_post->len = post;
2610 split_post->orig_start = split_post->start;
2611 split_post->block_start = em->block_start + em->len - post;
2612 split_post->block_len = split_post->len;
2613 split_post->orig_block_len = split_post->block_len;
2614 split_post->ram_bytes = split_post->len;
2615 split_post->flags = flags;
2616 split_post->compress_type = em->compress_type;
2617 split_post->generation = em->generation;
2618 add_extent_mapping(em_tree, split_post, 1);
2622 free_extent_map(em);
2623 /* Once for the tree */
2624 free_extent_map(em);
2627 write_unlock(&em_tree->lock);
2628 unlock_extent(&inode->io_tree, start, start + len - 1, NULL);
2630 free_extent_map(split_pre);
2631 free_extent_map(split_mid);
2632 free_extent_map(split_post);
2637 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2638 struct bio *bio, loff_t file_offset)
2640 struct btrfs_ordered_extent *ordered;
2641 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2643 u64 len = bio->bi_iter.bi_size;
2644 u64 end = start + len;
2649 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2650 if (WARN_ON_ONCE(!ordered))
2651 return BLK_STS_IOERR;
2653 /* No need to split */
2654 if (ordered->disk_num_bytes == len)
2657 /* We cannot split once end_bio'd ordered extent */
2658 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2663 /* We cannot split a compressed ordered extent */
2664 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2669 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2670 /* bio must be in one ordered extent */
2671 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2676 /* Checksum list should be empty */
2677 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2682 file_len = ordered->num_bytes;
2683 pre = start - ordered->disk_bytenr;
2684 post = ordered_end - end;
2686 ret = btrfs_split_ordered_extent(ordered, pre, post);
2689 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2692 btrfs_put_ordered_extent(ordered);
2694 return errno_to_blk_status(ret);
2697 void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio, int mirror_num)
2699 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2700 struct btrfs_inode *bi = BTRFS_I(inode);
2703 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2704 ret = extract_ordered_extent(bi, bio,
2705 page_offset(bio_first_bvec_all(bio)->bv_page));
2707 btrfs_bio_end_io(btrfs_bio(bio), ret);
2713 * If we need to checksum, and the I/O is not issued by fsync and
2714 * friends, that is ->sync_writers != 0, defer the submission to a
2715 * workqueue to parallelize it.
2717 * Csum items for reloc roots have already been cloned at this point,
2718 * so they are handled as part of the no-checksum case.
2720 if (!(bi->flags & BTRFS_INODE_NODATASUM) &&
2721 !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state) &&
2722 !btrfs_is_data_reloc_root(bi->root)) {
2723 if (!atomic_read(&bi->sync_writers) &&
2724 btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
2725 btrfs_submit_bio_start))
2728 ret = btrfs_csum_one_bio(bi, bio, (u64)-1, false);
2730 btrfs_bio_end_io(btrfs_bio(bio), ret);
2734 btrfs_submit_bio(fs_info, bio, mirror_num);
2737 void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
2738 int mirror_num, enum btrfs_compression_type compress_type)
2740 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2743 if (compress_type != BTRFS_COMPRESS_NONE) {
2745 * btrfs_submit_compressed_read will handle completing the bio
2746 * if there were any errors, so just return here.
2748 btrfs_submit_compressed_read(inode, bio, mirror_num);
2752 /* Save the original iter for read repair */
2753 btrfs_bio(bio)->iter = bio->bi_iter;
2756 * Lookup bio sums does extra checks around whether we need to csum or
2757 * not, which is why we ignore skip_sum here.
2759 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2761 btrfs_bio_end_io(btrfs_bio(bio), 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, 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(&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(&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(&inode->io_tree, page_start, page_end, &cached_state);
2973 * We hit ENOSPC or other errors. Update the mapping and page
2974 * to reflect the errors and clean the page.
2976 mapping_set_error(page->mapping, ret);
2977 end_extent_writepage(page, ret, page_start, page_end);
2978 clear_page_dirty_for_io(page);
2981 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2985 extent_changeset_free(data_reserved);
2987 * As a precaution, do a delayed iput in case it would be the last iput
2988 * that could need flushing space. Recursing back to fixup worker would
2991 btrfs_add_delayed_iput(&inode->vfs_inode);
2995 * There are a few paths in the higher layers of the kernel that directly
2996 * set the page dirty bit without asking the filesystem if it is a
2997 * good idea. This causes problems because we want to make sure COW
2998 * properly happens and the data=ordered rules are followed.
3000 * In our case any range that doesn't have the ORDERED bit set
3001 * hasn't been properly setup for IO. We kick off an async process
3002 * to fix it up. The async helper will wait for ordered extents, set
3003 * the delalloc bit and make it safe to write the page.
3005 int btrfs_writepage_cow_fixup(struct page *page)
3007 struct inode *inode = page->mapping->host;
3008 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3009 struct btrfs_writepage_fixup *fixup;
3011 /* This page has ordered extent covering it already */
3012 if (PageOrdered(page))
3016 * PageChecked is set below when we create a fixup worker for this page,
3017 * don't try to create another one if we're already PageChecked()
3019 * The extent_io writepage code will redirty the page if we send back
3022 if (PageChecked(page))
3025 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
3030 * We are already holding a reference to this inode from
3031 * write_cache_pages. We need to hold it because the space reservation
3032 * takes place outside of the page lock, and we can't trust
3033 * page->mapping outside of the page lock.
3036 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3038 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3040 fixup->inode = inode;
3041 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3046 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3047 struct btrfs_inode *inode, u64 file_pos,
3048 struct btrfs_file_extent_item *stack_fi,
3049 const bool update_inode_bytes,
3050 u64 qgroup_reserved)
3052 struct btrfs_root *root = inode->root;
3053 const u64 sectorsize = root->fs_info->sectorsize;
3054 struct btrfs_path *path;
3055 struct extent_buffer *leaf;
3056 struct btrfs_key ins;
3057 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3058 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3059 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3060 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3061 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3062 struct btrfs_drop_extents_args drop_args = { 0 };
3065 path = btrfs_alloc_path();
3070 * we may be replacing one extent in the tree with another.
3071 * The new extent is pinned in the extent map, and we don't want
3072 * to drop it from the cache until it is completely in the btree.
3074 * So, tell btrfs_drop_extents to leave this extent in the cache.
3075 * the caller is expected to unpin it and allow it to be merged
3078 drop_args.path = path;
3079 drop_args.start = file_pos;
3080 drop_args.end = file_pos + num_bytes;
3081 drop_args.replace_extent = true;
3082 drop_args.extent_item_size = sizeof(*stack_fi);
3083 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3087 if (!drop_args.extent_inserted) {
3088 ins.objectid = btrfs_ino(inode);
3089 ins.offset = file_pos;
3090 ins.type = BTRFS_EXTENT_DATA_KEY;
3092 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3097 leaf = path->nodes[0];
3098 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3099 write_extent_buffer(leaf, stack_fi,
3100 btrfs_item_ptr_offset(leaf, path->slots[0]),
3101 sizeof(struct btrfs_file_extent_item));
3103 btrfs_mark_buffer_dirty(leaf);
3104 btrfs_release_path(path);
3107 * If we dropped an inline extent here, we know the range where it is
3108 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3109 * number of bytes only for that range containing the inline extent.
3110 * The remaining of the range will be processed when clearning the
3111 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3113 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3114 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3116 inline_size = drop_args.bytes_found - inline_size;
3117 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3118 drop_args.bytes_found -= inline_size;
3119 num_bytes -= sectorsize;
3122 if (update_inode_bytes)
3123 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3125 ins.objectid = disk_bytenr;
3126 ins.offset = disk_num_bytes;
3127 ins.type = BTRFS_EXTENT_ITEM_KEY;
3129 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3133 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3135 qgroup_reserved, &ins);
3137 btrfs_free_path(path);
3142 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3145 struct btrfs_block_group *cache;
3147 cache = btrfs_lookup_block_group(fs_info, start);
3150 spin_lock(&cache->lock);
3151 cache->delalloc_bytes -= len;
3152 spin_unlock(&cache->lock);
3154 btrfs_put_block_group(cache);
3157 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3158 struct btrfs_ordered_extent *oe)
3160 struct btrfs_file_extent_item stack_fi;
3161 bool update_inode_bytes;
3162 u64 num_bytes = oe->num_bytes;
3163 u64 ram_bytes = oe->ram_bytes;
3165 memset(&stack_fi, 0, sizeof(stack_fi));
3166 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3167 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3168 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3169 oe->disk_num_bytes);
3170 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3171 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3172 num_bytes = oe->truncated_len;
3173 ram_bytes = num_bytes;
3175 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3176 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3177 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3178 /* Encryption and other encoding is reserved and all 0 */
3181 * For delalloc, when completing an ordered extent we update the inode's
3182 * bytes when clearing the range in the inode's io tree, so pass false
3183 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3184 * except if the ordered extent was truncated.
3186 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3187 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3188 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3190 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3191 oe->file_offset, &stack_fi,
3192 update_inode_bytes, oe->qgroup_rsv);
3196 * As ordered data IO finishes, this gets called so we can finish
3197 * an ordered extent if the range of bytes in the file it covers are
3200 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3202 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3203 struct btrfs_root *root = inode->root;
3204 struct btrfs_fs_info *fs_info = root->fs_info;
3205 struct btrfs_trans_handle *trans = NULL;
3206 struct extent_io_tree *io_tree = &inode->io_tree;
3207 struct extent_state *cached_state = NULL;
3209 int compress_type = 0;
3211 u64 logical_len = ordered_extent->num_bytes;
3212 bool freespace_inode;
3213 bool truncated = false;
3214 bool clear_reserved_extent = true;
3215 unsigned int clear_bits = EXTENT_DEFRAG;
3217 start = ordered_extent->file_offset;
3218 end = start + ordered_extent->num_bytes - 1;
3220 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3221 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3222 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3223 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3224 clear_bits |= EXTENT_DELALLOC_NEW;
3226 freespace_inode = btrfs_is_free_space_inode(inode);
3227 if (!freespace_inode)
3228 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3230 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3235 /* A valid bdev implies a write on a sequential zone */
3236 if (ordered_extent->bdev) {
3237 btrfs_rewrite_logical_zoned(ordered_extent);
3238 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3239 ordered_extent->disk_num_bytes);
3240 } else if (btrfs_is_data_reloc_root(inode->root)) {
3241 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3242 ordered_extent->disk_num_bytes);
3245 btrfs_free_io_failure_record(inode, start, end);
3247 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3249 logical_len = ordered_extent->truncated_len;
3250 /* Truncated the entire extent, don't bother adding */
3255 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3256 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3258 btrfs_inode_safe_disk_i_size_write(inode, 0);
3259 if (freespace_inode)
3260 trans = btrfs_join_transaction_spacecache(root);
3262 trans = btrfs_join_transaction(root);
3263 if (IS_ERR(trans)) {
3264 ret = PTR_ERR(trans);
3268 trans->block_rsv = &inode->block_rsv;
3269 ret = btrfs_update_inode_fallback(trans, root, inode);
3270 if (ret) /* -ENOMEM or corruption */
3271 btrfs_abort_transaction(trans, ret);
3275 clear_bits |= EXTENT_LOCKED;
3276 lock_extent(io_tree, start, end, &cached_state);
3278 if (freespace_inode)
3279 trans = btrfs_join_transaction_spacecache(root);
3281 trans = btrfs_join_transaction(root);
3282 if (IS_ERR(trans)) {
3283 ret = PTR_ERR(trans);
3288 trans->block_rsv = &inode->block_rsv;
3290 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3291 compress_type = ordered_extent->compress_type;
3292 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3293 BUG_ON(compress_type);
3294 ret = btrfs_mark_extent_written(trans, inode,
3295 ordered_extent->file_offset,
3296 ordered_extent->file_offset +
3298 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3299 ordered_extent->disk_num_bytes);
3301 BUG_ON(root == fs_info->tree_root);
3302 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3304 clear_reserved_extent = false;
3305 btrfs_release_delalloc_bytes(fs_info,
3306 ordered_extent->disk_bytenr,
3307 ordered_extent->disk_num_bytes);
3310 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3311 ordered_extent->num_bytes, trans->transid);
3313 btrfs_abort_transaction(trans, ret);
3317 ret = add_pending_csums(trans, &ordered_extent->list);
3319 btrfs_abort_transaction(trans, ret);
3324 * If this is a new delalloc range, clear its new delalloc flag to
3325 * update the inode's number of bytes. This needs to be done first
3326 * before updating the inode item.
3328 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3329 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3330 clear_extent_bit(&inode->io_tree, start, end,
3331 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3334 btrfs_inode_safe_disk_i_size_write(inode, 0);
3335 ret = btrfs_update_inode_fallback(trans, root, inode);
3336 if (ret) { /* -ENOMEM or corruption */
3337 btrfs_abort_transaction(trans, ret);
3342 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3346 btrfs_end_transaction(trans);
3348 if (ret || truncated) {
3349 u64 unwritten_start = start;
3352 * If we failed to finish this ordered extent for any reason we
3353 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3354 * extent, and mark the inode with the error if it wasn't
3355 * already set. Any error during writeback would have already
3356 * set the mapping error, so we need to set it if we're the ones
3357 * marking this ordered extent as failed.
3359 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3360 &ordered_extent->flags))
3361 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3364 unwritten_start += logical_len;
3365 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3368 * Drop extent maps for the part of the extent we didn't write.
3370 * We have an exception here for the free_space_inode, this is
3371 * because when we do btrfs_get_extent() on the free space inode
3372 * we will search the commit root. If this is a new block group
3373 * we won't find anything, and we will trip over the assert in
3374 * writepage where we do ASSERT(em->block_start !=
3377 * Theoretically we could also skip this for any NOCOW extent as
3378 * we don't mess with the extent map tree in the NOCOW case, but
3379 * for now simply skip this if we are the free space inode.
3381 if (!btrfs_is_free_space_inode(inode))
3382 btrfs_drop_extent_map_range(inode, unwritten_start,
3386 * If the ordered extent had an IOERR or something else went
3387 * wrong we need to return the space for this ordered extent
3388 * back to the allocator. We only free the extent in the
3389 * truncated case if we didn't write out the extent at all.
3391 * If we made it past insert_reserved_file_extent before we
3392 * errored out then we don't need to do this as the accounting
3393 * has already been done.
3395 if ((ret || !logical_len) &&
3396 clear_reserved_extent &&
3397 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3398 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3400 * Discard the range before returning it back to the
3403 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3404 btrfs_discard_extent(fs_info,
3405 ordered_extent->disk_bytenr,
3406 ordered_extent->disk_num_bytes,
3408 btrfs_free_reserved_extent(fs_info,
3409 ordered_extent->disk_bytenr,
3410 ordered_extent->disk_num_bytes, 1);
3412 * Actually free the qgroup rsv which was released when
3413 * the ordered extent was created.
3415 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3416 ordered_extent->qgroup_rsv,
3417 BTRFS_QGROUP_RSV_DATA);
3422 * This needs to be done to make sure anybody waiting knows we are done
3423 * updating everything for this ordered extent.
3425 btrfs_remove_ordered_extent(inode, ordered_extent);
3428 btrfs_put_ordered_extent(ordered_extent);
3429 /* once for the tree */
3430 btrfs_put_ordered_extent(ordered_extent);
3435 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3436 struct page *page, u64 start,
3437 u64 end, bool uptodate)
3439 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3441 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate);
3445 * Verify the checksum for a single sector without any extra action that depend
3446 * on the type of I/O.
3448 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3449 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3451 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3454 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3456 shash->tfm = fs_info->csum_shash;
3458 kaddr = kmap_local_page(page) + pgoff;
3459 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3460 kunmap_local(kaddr);
3462 if (memcmp(csum, csum_expected, fs_info->csum_size))
3467 static u8 *btrfs_csum_ptr(const struct btrfs_fs_info *fs_info, u8 *csums, u64 offset)
3469 u64 offset_in_sectors = offset >> fs_info->sectorsize_bits;
3471 return csums + offset_in_sectors * fs_info->csum_size;
3475 * check_data_csum - verify checksum of one sector of uncompressed data
3477 * @bbio: btrfs_bio which contains the csum
3478 * @bio_offset: offset to the beginning of the bio (in bytes)
3479 * @page: page where is the data to be verified
3480 * @pgoff: offset inside the page
3482 * The length of such check is always one sector size.
3484 * When csum mismatch is detected, we will also report the error and fill the
3485 * corrupted range with zero. (Thus it needs the extra parameters)
3487 int btrfs_check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3488 u32 bio_offset, struct page *page, u32 pgoff)
3490 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3491 u32 len = fs_info->sectorsize;
3493 u8 csum[BTRFS_CSUM_SIZE];
3495 ASSERT(pgoff + len <= PAGE_SIZE);
3497 csum_expected = btrfs_csum_ptr(fs_info, bbio->csum, bio_offset);
3499 if (btrfs_check_sector_csum(fs_info, page, pgoff, csum, csum_expected))
3504 btrfs_print_data_csum_error(BTRFS_I(inode),
3505 bbio->file_offset + bio_offset,
3506 csum, csum_expected, bbio->mirror_num);
3508 btrfs_dev_stat_inc_and_print(bbio->device,
3509 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3510 memzero_page(page, pgoff, len);
3515 * When reads are done, we need to check csums to verify the data is correct.
3516 * if there's a match, we allow the bio to finish. If not, the code in
3517 * extent_io.c will try to find good copies for us.
3519 * @bio_offset: offset to the beginning of the bio (in bytes)
3520 * @start: file offset of the range start
3521 * @end: file offset of the range end (inclusive)
3523 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3526 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3527 u32 bio_offset, struct page *page,
3530 struct inode *inode = page->mapping->host;
3531 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3532 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3533 struct btrfs_root *root = BTRFS_I(inode)->root;
3534 const u32 sectorsize = root->fs_info->sectorsize;
3536 unsigned int result = 0;
3539 * This only happens for NODATASUM or compressed read.
3540 * Normally this should be covered by above check for compressed read
3541 * or the next check for NODATASUM. Just do a quicker exit here.
3543 if (bbio->csum == NULL)
3546 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3549 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3552 ASSERT(page_offset(page) <= start &&
3553 end <= page_offset(page) + PAGE_SIZE - 1);
3554 for (pg_off = offset_in_page(start);
3555 pg_off < offset_in_page(end);
3556 pg_off += sectorsize, bio_offset += sectorsize) {
3557 u64 file_offset = pg_off + page_offset(page);
3560 if (btrfs_is_data_reloc_root(root) &&
3561 test_range_bit(io_tree, file_offset,
3562 file_offset + sectorsize - 1,
3563 EXTENT_NODATASUM, 1, NULL)) {
3564 /* Skip the range without csum for data reloc inode */
3565 clear_extent_bits(io_tree, file_offset,
3566 file_offset + sectorsize - 1,
3570 ret = btrfs_check_data_csum(inode, bbio, bio_offset, page, pg_off);
3572 const int nr_bit = (pg_off - offset_in_page(start)) >>
3573 root->fs_info->sectorsize_bits;
3575 result |= (1U << nr_bit);
3582 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3584 * @inode: The inode we want to perform iput on
3586 * This function uses the generic vfs_inode::i_count to track whether we should
3587 * just decrement it (in case it's > 1) or if this is the last iput then link
3588 * the inode to the delayed iput machinery. Delayed iputs are processed at
3589 * transaction commit time/superblock commit/cleaner kthread.
3591 void btrfs_add_delayed_iput(struct inode *inode)
3593 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3594 struct btrfs_inode *binode = BTRFS_I(inode);
3596 if (atomic_add_unless(&inode->i_count, -1, 1))
3599 atomic_inc(&fs_info->nr_delayed_iputs);
3600 spin_lock(&fs_info->delayed_iput_lock);
3601 ASSERT(list_empty(&binode->delayed_iput));
3602 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3603 spin_unlock(&fs_info->delayed_iput_lock);
3604 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3605 wake_up_process(fs_info->cleaner_kthread);
3608 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3609 struct btrfs_inode *inode)
3611 list_del_init(&inode->delayed_iput);
3612 spin_unlock(&fs_info->delayed_iput_lock);
3613 iput(&inode->vfs_inode);
3614 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3615 wake_up(&fs_info->delayed_iputs_wait);
3616 spin_lock(&fs_info->delayed_iput_lock);
3619 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3620 struct btrfs_inode *inode)
3622 if (!list_empty(&inode->delayed_iput)) {
3623 spin_lock(&fs_info->delayed_iput_lock);
3624 if (!list_empty(&inode->delayed_iput))
3625 run_delayed_iput_locked(fs_info, inode);
3626 spin_unlock(&fs_info->delayed_iput_lock);
3630 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3633 spin_lock(&fs_info->delayed_iput_lock);
3634 while (!list_empty(&fs_info->delayed_iputs)) {
3635 struct btrfs_inode *inode;
3637 inode = list_first_entry(&fs_info->delayed_iputs,
3638 struct btrfs_inode, delayed_iput);
3639 run_delayed_iput_locked(fs_info, inode);
3640 cond_resched_lock(&fs_info->delayed_iput_lock);
3642 spin_unlock(&fs_info->delayed_iput_lock);
3646 * Wait for flushing all delayed iputs
3648 * @fs_info: the filesystem
3650 * This will wait on any delayed iputs that are currently running with KILLABLE
3651 * set. Once they are all done running we will return, unless we are killed in
3652 * which case we return EINTR. This helps in user operations like fallocate etc
3653 * that might get blocked on the iputs.
3655 * Return EINTR if we were killed, 0 if nothing's pending
3657 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3659 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3660 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3667 * This creates an orphan entry for the given inode in case something goes wrong
3668 * in the middle of an unlink.
3670 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3671 struct btrfs_inode *inode)
3675 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3676 if (ret && ret != -EEXIST) {
3677 btrfs_abort_transaction(trans, ret);
3685 * We have done the delete so we can go ahead and remove the orphan item for
3686 * this particular inode.
3688 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3689 struct btrfs_inode *inode)
3691 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3695 * this cleans up any orphans that may be left on the list from the last use
3698 int btrfs_orphan_cleanup(struct btrfs_root *root)
3700 struct btrfs_fs_info *fs_info = root->fs_info;
3701 struct btrfs_path *path;
3702 struct extent_buffer *leaf;
3703 struct btrfs_key key, found_key;
3704 struct btrfs_trans_handle *trans;
3705 struct inode *inode;
3706 u64 last_objectid = 0;
3707 int ret = 0, nr_unlink = 0;
3709 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3712 path = btrfs_alloc_path();
3717 path->reada = READA_BACK;
3719 key.objectid = BTRFS_ORPHAN_OBJECTID;
3720 key.type = BTRFS_ORPHAN_ITEM_KEY;
3721 key.offset = (u64)-1;
3724 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3729 * if ret == 0 means we found what we were searching for, which
3730 * is weird, but possible, so only screw with path if we didn't
3731 * find the key and see if we have stuff that matches
3735 if (path->slots[0] == 0)
3740 /* pull out the item */
3741 leaf = path->nodes[0];
3742 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3744 /* make sure the item matches what we want */
3745 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3747 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3750 /* release the path since we're done with it */
3751 btrfs_release_path(path);
3754 * this is where we are basically btrfs_lookup, without the
3755 * crossing root thing. we store the inode number in the
3756 * offset of the orphan item.
3759 if (found_key.offset == last_objectid) {
3761 "Error removing orphan entry, stopping orphan cleanup");
3766 last_objectid = found_key.offset;
3768 found_key.objectid = found_key.offset;
3769 found_key.type = BTRFS_INODE_ITEM_KEY;
3770 found_key.offset = 0;
3771 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3772 ret = PTR_ERR_OR_ZERO(inode);
3773 if (ret && ret != -ENOENT)
3776 if (ret == -ENOENT && root == fs_info->tree_root) {
3777 struct btrfs_root *dead_root;
3778 int is_dead_root = 0;
3781 * This is an orphan in the tree root. Currently these
3782 * could come from 2 sources:
3783 * a) a root (snapshot/subvolume) deletion in progress
3784 * b) a free space cache inode
3785 * We need to distinguish those two, as the orphan item
3786 * for a root must not get deleted before the deletion
3787 * of the snapshot/subvolume's tree completes.
3789 * btrfs_find_orphan_roots() ran before us, which has
3790 * found all deleted roots and loaded them into
3791 * fs_info->fs_roots_radix. So here we can find if an
3792 * orphan item corresponds to a deleted root by looking
3793 * up the root from that radix tree.
3796 spin_lock(&fs_info->fs_roots_radix_lock);
3797 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3798 (unsigned long)found_key.objectid);
3799 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3801 spin_unlock(&fs_info->fs_roots_radix_lock);
3804 /* prevent this orphan from being found again */
3805 key.offset = found_key.objectid - 1;
3812 * If we have an inode with links, there are a couple of
3815 * 1. We were halfway through creating fsverity metadata for the
3816 * file. In that case, the orphan item represents incomplete
3817 * fsverity metadata which must be cleaned up with
3818 * btrfs_drop_verity_items and deleting the orphan item.
3820 * 2. Old kernels (before v3.12) used to create an
3821 * orphan item for truncate indicating that there were possibly
3822 * extent items past i_size that needed to be deleted. In v3.12,
3823 * truncate was changed to update i_size in sync with the extent
3824 * items, but the (useless) orphan item was still created. Since
3825 * v4.18, we don't create the orphan item for truncate at all.
3827 * So, this item could mean that we need to do a truncate, but
3828 * only if this filesystem was last used on a pre-v3.12 kernel
3829 * and was not cleanly unmounted. The odds of that are quite
3830 * slim, and it's a pain to do the truncate now, so just delete
3833 * It's also possible that this orphan item was supposed to be
3834 * deleted but wasn't. The inode number may have been reused,
3835 * but either way, we can delete the orphan item.
3837 if (ret == -ENOENT || inode->i_nlink) {
3839 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3844 trans = btrfs_start_transaction(root, 1);
3845 if (IS_ERR(trans)) {
3846 ret = PTR_ERR(trans);
3849 btrfs_debug(fs_info, "auto deleting %Lu",
3850 found_key.objectid);
3851 ret = btrfs_del_orphan_item(trans, root,
3852 found_key.objectid);
3853 btrfs_end_transaction(trans);
3861 /* this will do delete_inode and everything for us */
3864 /* release the path since we're done with it */
3865 btrfs_release_path(path);
3867 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3868 trans = btrfs_join_transaction(root);
3870 btrfs_end_transaction(trans);
3874 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3878 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3879 btrfs_free_path(path);
3884 * very simple check to peek ahead in the leaf looking for xattrs. If we
3885 * don't find any xattrs, we know there can't be any acls.
3887 * slot is the slot the inode is in, objectid is the objectid of the inode
3889 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3890 int slot, u64 objectid,
3891 int *first_xattr_slot)
3893 u32 nritems = btrfs_header_nritems(leaf);
3894 struct btrfs_key found_key;
3895 static u64 xattr_access = 0;
3896 static u64 xattr_default = 0;
3899 if (!xattr_access) {
3900 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3901 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3902 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3903 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3907 *first_xattr_slot = -1;
3908 while (slot < nritems) {
3909 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3911 /* we found a different objectid, there must not be acls */
3912 if (found_key.objectid != objectid)
3915 /* we found an xattr, assume we've got an acl */
3916 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3917 if (*first_xattr_slot == -1)
3918 *first_xattr_slot = slot;
3919 if (found_key.offset == xattr_access ||
3920 found_key.offset == xattr_default)
3925 * we found a key greater than an xattr key, there can't
3926 * be any acls later on
3928 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3935 * it goes inode, inode backrefs, xattrs, extents,
3936 * so if there are a ton of hard links to an inode there can
3937 * be a lot of backrefs. Don't waste time searching too hard,
3938 * this is just an optimization
3943 /* we hit the end of the leaf before we found an xattr or
3944 * something larger than an xattr. We have to assume the inode
3947 if (*first_xattr_slot == -1)
3948 *first_xattr_slot = slot;
3953 * read an inode from the btree into the in-memory inode
3955 static int btrfs_read_locked_inode(struct inode *inode,
3956 struct btrfs_path *in_path)
3958 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3959 struct btrfs_path *path = in_path;
3960 struct extent_buffer *leaf;
3961 struct btrfs_inode_item *inode_item;
3962 struct btrfs_root *root = BTRFS_I(inode)->root;
3963 struct btrfs_key location;
3968 bool filled = false;
3969 int first_xattr_slot;
3971 ret = btrfs_fill_inode(inode, &rdev);
3976 path = btrfs_alloc_path();
3981 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3983 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3985 if (path != in_path)
3986 btrfs_free_path(path);
3990 leaf = path->nodes[0];
3995 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3996 struct btrfs_inode_item);
3997 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3998 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3999 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
4000 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
4001 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
4002 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
4003 round_up(i_size_read(inode), fs_info->sectorsize));
4005 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
4006 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
4008 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
4009 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
4011 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
4012 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
4014 BTRFS_I(inode)->i_otime.tv_sec =
4015 btrfs_timespec_sec(leaf, &inode_item->otime);
4016 BTRFS_I(inode)->i_otime.tv_nsec =
4017 btrfs_timespec_nsec(leaf, &inode_item->otime);
4019 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
4020 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
4021 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
4023 inode_set_iversion_queried(inode,
4024 btrfs_inode_sequence(leaf, inode_item));
4025 inode->i_generation = BTRFS_I(inode)->generation;
4027 rdev = btrfs_inode_rdev(leaf, inode_item);
4029 BTRFS_I(inode)->index_cnt = (u64)-1;
4030 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
4031 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
4035 * If we were modified in the current generation and evicted from memory
4036 * and then re-read we need to do a full sync since we don't have any
4037 * idea about which extents were modified before we were evicted from
4040 * This is required for both inode re-read from disk and delayed inode
4041 * in delayed_nodes_tree.
4043 if (BTRFS_I(inode)->last_trans == fs_info->generation)
4044 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4045 &BTRFS_I(inode)->runtime_flags);
4048 * We don't persist the id of the transaction where an unlink operation
4049 * against the inode was last made. So here we assume the inode might
4050 * have been evicted, and therefore the exact value of last_unlink_trans
4051 * lost, and set it to last_trans to avoid metadata inconsistencies
4052 * between the inode and its parent if the inode is fsync'ed and the log
4053 * replayed. For example, in the scenario:
4056 * ln mydir/foo mydir/bar
4059 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
4060 * xfs_io -c fsync mydir/foo
4062 * mount fs, triggers fsync log replay
4064 * We must make sure that when we fsync our inode foo we also log its
4065 * parent inode, otherwise after log replay the parent still has the
4066 * dentry with the "bar" name but our inode foo has a link count of 1
4067 * and doesn't have an inode ref with the name "bar" anymore.
4069 * Setting last_unlink_trans to last_trans is a pessimistic approach,
4070 * but it guarantees correctness at the expense of occasional full
4071 * transaction commits on fsync if our inode is a directory, or if our
4072 * inode is not a directory, logging its parent unnecessarily.
4074 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
4077 * Same logic as for last_unlink_trans. We don't persist the generation
4078 * of the last transaction where this inode was used for a reflink
4079 * operation, so after eviction and reloading the inode we must be
4080 * pessimistic and assume the last transaction that modified the inode.
4082 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
4085 if (inode->i_nlink != 1 ||
4086 path->slots[0] >= btrfs_header_nritems(leaf))
4089 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
4090 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
4093 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4094 if (location.type == BTRFS_INODE_REF_KEY) {
4095 struct btrfs_inode_ref *ref;
4097 ref = (struct btrfs_inode_ref *)ptr;
4098 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
4099 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
4100 struct btrfs_inode_extref *extref;
4102 extref = (struct btrfs_inode_extref *)ptr;
4103 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4108 * try to precache a NULL acl entry for files that don't have
4109 * any xattrs or acls
4111 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4112 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4113 if (first_xattr_slot != -1) {
4114 path->slots[0] = first_xattr_slot;
4115 ret = btrfs_load_inode_props(inode, path);
4118 "error loading props for ino %llu (root %llu): %d",
4119 btrfs_ino(BTRFS_I(inode)),
4120 root->root_key.objectid, ret);
4122 if (path != in_path)
4123 btrfs_free_path(path);
4126 cache_no_acl(inode);
4128 switch (inode->i_mode & S_IFMT) {
4130 inode->i_mapping->a_ops = &btrfs_aops;
4131 inode->i_fop = &btrfs_file_operations;
4132 inode->i_op = &btrfs_file_inode_operations;
4135 inode->i_fop = &btrfs_dir_file_operations;
4136 inode->i_op = &btrfs_dir_inode_operations;
4139 inode->i_op = &btrfs_symlink_inode_operations;
4140 inode_nohighmem(inode);
4141 inode->i_mapping->a_ops = &btrfs_aops;
4144 inode->i_op = &btrfs_special_inode_operations;
4145 init_special_inode(inode, inode->i_mode, rdev);
4149 btrfs_sync_inode_flags_to_i_flags(inode);
4154 * given a leaf and an inode, copy the inode fields into the leaf
4156 static void fill_inode_item(struct btrfs_trans_handle *trans,
4157 struct extent_buffer *leaf,
4158 struct btrfs_inode_item *item,
4159 struct inode *inode)
4161 struct btrfs_map_token token;
4164 btrfs_init_map_token(&token, leaf);
4166 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4167 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4168 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4169 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4170 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4172 btrfs_set_token_timespec_sec(&token, &item->atime,
4173 inode->i_atime.tv_sec);
4174 btrfs_set_token_timespec_nsec(&token, &item->atime,
4175 inode->i_atime.tv_nsec);
4177 btrfs_set_token_timespec_sec(&token, &item->mtime,
4178 inode->i_mtime.tv_sec);
4179 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4180 inode->i_mtime.tv_nsec);
4182 btrfs_set_token_timespec_sec(&token, &item->ctime,
4183 inode->i_ctime.tv_sec);
4184 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4185 inode->i_ctime.tv_nsec);
4187 btrfs_set_token_timespec_sec(&token, &item->otime,
4188 BTRFS_I(inode)->i_otime.tv_sec);
4189 btrfs_set_token_timespec_nsec(&token, &item->otime,
4190 BTRFS_I(inode)->i_otime.tv_nsec);
4192 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4193 btrfs_set_token_inode_generation(&token, item,
4194 BTRFS_I(inode)->generation);
4195 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4196 btrfs_set_token_inode_transid(&token, item, trans->transid);
4197 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4198 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4199 BTRFS_I(inode)->ro_flags);
4200 btrfs_set_token_inode_flags(&token, item, flags);
4201 btrfs_set_token_inode_block_group(&token, item, 0);
4205 * copy everything in the in-memory inode into the btree.
4207 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root,
4209 struct btrfs_inode *inode)
4211 struct btrfs_inode_item *inode_item;
4212 struct btrfs_path *path;
4213 struct extent_buffer *leaf;
4216 path = btrfs_alloc_path();
4220 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4227 leaf = path->nodes[0];
4228 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4229 struct btrfs_inode_item);
4231 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4232 btrfs_mark_buffer_dirty(leaf);
4233 btrfs_set_inode_last_trans(trans, inode);
4236 btrfs_free_path(path);
4241 * copy everything in the in-memory inode into the btree.
4243 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4244 struct btrfs_root *root,
4245 struct btrfs_inode *inode)
4247 struct btrfs_fs_info *fs_info = root->fs_info;
4251 * If the inode is a free space inode, we can deadlock during commit
4252 * if we put it into the delayed code.
4254 * The data relocation inode should also be directly updated
4257 if (!btrfs_is_free_space_inode(inode)
4258 && !btrfs_is_data_reloc_root(root)
4259 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4260 btrfs_update_root_times(trans, root);
4262 ret = btrfs_delayed_update_inode(trans, root, inode);
4264 btrfs_set_inode_last_trans(trans, inode);
4268 return btrfs_update_inode_item(trans, root, inode);
4271 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4272 struct btrfs_root *root, struct btrfs_inode *inode)
4276 ret = btrfs_update_inode(trans, root, inode);
4278 return btrfs_update_inode_item(trans, root, inode);
4283 * unlink helper that gets used here in inode.c and in the tree logging
4284 * recovery code. It remove a link in a directory with a given name, and
4285 * also drops the back refs in the inode to the directory
4287 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4288 struct btrfs_inode *dir,
4289 struct btrfs_inode *inode,
4290 const struct fscrypt_str *name,
4291 struct btrfs_rename_ctx *rename_ctx)
4293 struct btrfs_root *root = dir->root;
4294 struct btrfs_fs_info *fs_info = root->fs_info;
4295 struct btrfs_path *path;
4297 struct btrfs_dir_item *di;
4299 u64 ino = btrfs_ino(inode);
4300 u64 dir_ino = btrfs_ino(dir);
4302 path = btrfs_alloc_path();
4308 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4309 if (IS_ERR_OR_NULL(di)) {
4310 ret = di ? PTR_ERR(di) : -ENOENT;
4313 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4316 btrfs_release_path(path);
4319 * If we don't have dir index, we have to get it by looking up
4320 * the inode ref, since we get the inode ref, remove it directly,
4321 * it is unnecessary to do delayed deletion.
4323 * But if we have dir index, needn't search inode ref to get it.
4324 * Since the inode ref is close to the inode item, it is better
4325 * that we delay to delete it, and just do this deletion when
4326 * we update the inode item.
4328 if (inode->dir_index) {
4329 ret = btrfs_delayed_delete_inode_ref(inode);
4331 index = inode->dir_index;
4336 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4339 "failed to delete reference to %.*s, inode %llu parent %llu",
4340 name->len, name->name, ino, dir_ino);
4341 btrfs_abort_transaction(trans, ret);
4346 rename_ctx->index = index;
4348 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4350 btrfs_abort_transaction(trans, ret);
4355 * If we are in a rename context, we don't need to update anything in the
4356 * log. That will be done later during the rename by btrfs_log_new_name().
4357 * Besides that, doing it here would only cause extra unnecessary btree
4358 * operations on the log tree, increasing latency for applications.
4361 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4362 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4366 * If we have a pending delayed iput we could end up with the final iput
4367 * being run in btrfs-cleaner context. If we have enough of these built
4368 * up we can end up burning a lot of time in btrfs-cleaner without any
4369 * way to throttle the unlinks. Since we're currently holding a ref on
4370 * the inode we can run the delayed iput here without any issues as the
4371 * final iput won't be done until after we drop the ref we're currently
4374 btrfs_run_delayed_iput(fs_info, inode);
4376 btrfs_free_path(path);
4380 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4381 inode_inc_iversion(&inode->vfs_inode);
4382 inode_inc_iversion(&dir->vfs_inode);
4383 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4384 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4385 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4386 ret = btrfs_update_inode(trans, root, dir);
4391 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4392 struct btrfs_inode *dir, struct btrfs_inode *inode,
4393 const struct fscrypt_str *name)
4397 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4399 drop_nlink(&inode->vfs_inode);
4400 ret = btrfs_update_inode(trans, inode->root, inode);
4406 * helper to start transaction for unlink and rmdir.
4408 * unlink and rmdir are special in btrfs, they do not always free space, so
4409 * if we cannot make our reservations the normal way try and see if there is
4410 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4411 * allow the unlink to occur.
4413 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4415 struct btrfs_root *root = BTRFS_I(dir)->root;
4418 * 1 for the possible orphan item
4419 * 1 for the dir item
4420 * 1 for the dir index
4421 * 1 for the inode ref
4423 * 1 for the parent inode
4425 return btrfs_start_transaction_fallback_global_rsv(root, 6);
4428 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4430 struct btrfs_trans_handle *trans;
4431 struct inode *inode = d_inode(dentry);
4433 struct fscrypt_name fname;
4435 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4439 /* This needs to handle no-key deletions later on */
4441 trans = __unlink_start_trans(dir);
4442 if (IS_ERR(trans)) {
4443 ret = PTR_ERR(trans);
4447 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4450 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4455 if (inode->i_nlink == 0) {
4456 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4462 btrfs_end_transaction(trans);
4463 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4465 fscrypt_free_filename(&fname);
4469 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4470 struct inode *dir, struct dentry *dentry)
4472 struct btrfs_root *root = BTRFS_I(dir)->root;
4473 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4474 struct btrfs_path *path;
4475 struct extent_buffer *leaf;
4476 struct btrfs_dir_item *di;
4477 struct btrfs_key key;
4481 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4482 struct fscrypt_name fname;
4484 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4488 /* This needs to handle no-key deletions later on */
4490 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4491 objectid = inode->root->root_key.objectid;
4492 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4493 objectid = inode->location.objectid;
4496 fscrypt_free_filename(&fname);
4500 path = btrfs_alloc_path();
4506 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4507 &fname.disk_name, -1);
4508 if (IS_ERR_OR_NULL(di)) {
4509 ret = di ? PTR_ERR(di) : -ENOENT;
4513 leaf = path->nodes[0];
4514 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4515 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4516 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4518 btrfs_abort_transaction(trans, ret);
4521 btrfs_release_path(path);
4524 * This is a placeholder inode for a subvolume we didn't have a
4525 * reference to at the time of the snapshot creation. In the meantime
4526 * we could have renamed the real subvol link into our snapshot, so
4527 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4528 * Instead simply lookup the dir_index_item for this entry so we can
4529 * remove it. Otherwise we know we have a ref to the root and we can
4530 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4532 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4533 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4534 if (IS_ERR_OR_NULL(di)) {
4539 btrfs_abort_transaction(trans, ret);
4543 leaf = path->nodes[0];
4544 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4546 btrfs_release_path(path);
4548 ret = btrfs_del_root_ref(trans, objectid,
4549 root->root_key.objectid, dir_ino,
4550 &index, &fname.disk_name);
4552 btrfs_abort_transaction(trans, ret);
4557 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4559 btrfs_abort_transaction(trans, ret);
4563 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - fname.disk_name.len * 2);
4564 inode_inc_iversion(dir);
4565 dir->i_mtime = current_time(dir);
4566 dir->i_ctime = dir->i_mtime;
4567 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4569 btrfs_abort_transaction(trans, ret);
4571 btrfs_free_path(path);
4572 fscrypt_free_filename(&fname);
4577 * Helper to check if the subvolume references other subvolumes or if it's
4580 static noinline int may_destroy_subvol(struct btrfs_root *root)
4582 struct btrfs_fs_info *fs_info = root->fs_info;
4583 struct btrfs_path *path;
4584 struct btrfs_dir_item *di;
4585 struct btrfs_key key;
4586 struct fscrypt_str name = FSTR_INIT("default", 7);
4590 path = btrfs_alloc_path();
4594 /* Make sure this root isn't set as the default subvol */
4595 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4596 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4598 if (di && !IS_ERR(di)) {
4599 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4600 if (key.objectid == root->root_key.objectid) {
4603 "deleting default subvolume %llu is not allowed",
4607 btrfs_release_path(path);
4610 key.objectid = root->root_key.objectid;
4611 key.type = BTRFS_ROOT_REF_KEY;
4612 key.offset = (u64)-1;
4614 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4620 if (path->slots[0] > 0) {
4622 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4623 if (key.objectid == root->root_key.objectid &&
4624 key.type == BTRFS_ROOT_REF_KEY)
4628 btrfs_free_path(path);
4632 /* Delete all dentries for inodes belonging to the root */
4633 static void btrfs_prune_dentries(struct btrfs_root *root)
4635 struct btrfs_fs_info *fs_info = root->fs_info;
4636 struct rb_node *node;
4637 struct rb_node *prev;
4638 struct btrfs_inode *entry;
4639 struct inode *inode;
4642 if (!BTRFS_FS_ERROR(fs_info))
4643 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4645 spin_lock(&root->inode_lock);
4647 node = root->inode_tree.rb_node;
4651 entry = rb_entry(node, struct btrfs_inode, rb_node);
4653 if (objectid < btrfs_ino(entry))
4654 node = node->rb_left;
4655 else if (objectid > btrfs_ino(entry))
4656 node = node->rb_right;
4662 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4663 if (objectid <= btrfs_ino(entry)) {
4667 prev = rb_next(prev);
4671 entry = rb_entry(node, struct btrfs_inode, rb_node);
4672 objectid = btrfs_ino(entry) + 1;
4673 inode = igrab(&entry->vfs_inode);
4675 spin_unlock(&root->inode_lock);
4676 if (atomic_read(&inode->i_count) > 1)
4677 d_prune_aliases(inode);
4679 * btrfs_drop_inode will have it removed from the inode
4680 * cache when its usage count hits zero.
4684 spin_lock(&root->inode_lock);
4688 if (cond_resched_lock(&root->inode_lock))
4691 node = rb_next(node);
4693 spin_unlock(&root->inode_lock);
4696 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4698 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4699 struct btrfs_root *root = BTRFS_I(dir)->root;
4700 struct inode *inode = d_inode(dentry);
4701 struct btrfs_root *dest = BTRFS_I(inode)->root;
4702 struct btrfs_trans_handle *trans;
4703 struct btrfs_block_rsv block_rsv;
4707 down_write(&fs_info->subvol_sem);
4710 * Don't allow to delete a subvolume with send in progress. This is
4711 * inside the inode lock so the error handling that has to drop the bit
4712 * again is not run concurrently.
4714 spin_lock(&dest->root_item_lock);
4715 if (dest->send_in_progress) {
4716 spin_unlock(&dest->root_item_lock);
4718 "attempt to delete subvolume %llu during send",
4719 dest->root_key.objectid);
4723 if (atomic_read(&dest->nr_swapfiles)) {
4724 spin_unlock(&dest->root_item_lock);
4726 "attempt to delete subvolume %llu with active swapfile",
4727 root->root_key.objectid);
4731 root_flags = btrfs_root_flags(&dest->root_item);
4732 btrfs_set_root_flags(&dest->root_item,
4733 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4734 spin_unlock(&dest->root_item_lock);
4736 ret = may_destroy_subvol(dest);
4740 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4742 * One for dir inode,
4743 * two for dir entries,
4744 * two for root ref/backref.
4746 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4750 trans = btrfs_start_transaction(root, 0);
4751 if (IS_ERR(trans)) {
4752 ret = PTR_ERR(trans);
4755 trans->block_rsv = &block_rsv;
4756 trans->bytes_reserved = block_rsv.size;
4758 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4760 ret = btrfs_unlink_subvol(trans, dir, dentry);
4762 btrfs_abort_transaction(trans, ret);
4766 ret = btrfs_record_root_in_trans(trans, dest);
4768 btrfs_abort_transaction(trans, ret);
4772 memset(&dest->root_item.drop_progress, 0,
4773 sizeof(dest->root_item.drop_progress));
4774 btrfs_set_root_drop_level(&dest->root_item, 0);
4775 btrfs_set_root_refs(&dest->root_item, 0);
4777 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4778 ret = btrfs_insert_orphan_item(trans,
4780 dest->root_key.objectid);
4782 btrfs_abort_transaction(trans, ret);
4787 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4788 BTRFS_UUID_KEY_SUBVOL,
4789 dest->root_key.objectid);
4790 if (ret && ret != -ENOENT) {
4791 btrfs_abort_transaction(trans, ret);
4794 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4795 ret = btrfs_uuid_tree_remove(trans,
4796 dest->root_item.received_uuid,
4797 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4798 dest->root_key.objectid);
4799 if (ret && ret != -ENOENT) {
4800 btrfs_abort_transaction(trans, ret);
4805 free_anon_bdev(dest->anon_dev);
4808 trans->block_rsv = NULL;
4809 trans->bytes_reserved = 0;
4810 ret = btrfs_end_transaction(trans);
4811 inode->i_flags |= S_DEAD;
4813 btrfs_subvolume_release_metadata(root, &block_rsv);
4816 spin_lock(&dest->root_item_lock);
4817 root_flags = btrfs_root_flags(&dest->root_item);
4818 btrfs_set_root_flags(&dest->root_item,
4819 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4820 spin_unlock(&dest->root_item_lock);
4823 up_write(&fs_info->subvol_sem);
4825 d_invalidate(dentry);
4826 btrfs_prune_dentries(dest);
4827 ASSERT(dest->send_in_progress == 0);
4833 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4835 struct inode *inode = d_inode(dentry);
4836 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4838 struct btrfs_trans_handle *trans;
4839 u64 last_unlink_trans;
4840 struct fscrypt_name fname;
4842 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4844 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4845 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4847 "extent tree v2 doesn't support snapshot deletion yet");
4850 return btrfs_delete_subvolume(dir, dentry);
4853 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4857 /* This needs to handle no-key deletions later on */
4859 trans = __unlink_start_trans(dir);
4860 if (IS_ERR(trans)) {
4861 err = PTR_ERR(trans);
4865 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4866 err = btrfs_unlink_subvol(trans, dir, dentry);
4870 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4874 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4876 /* now the directory is empty */
4877 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4880 btrfs_i_size_write(BTRFS_I(inode), 0);
4882 * Propagate the last_unlink_trans value of the deleted dir to
4883 * its parent directory. This is to prevent an unrecoverable
4884 * log tree in the case we do something like this:
4886 * 2) create snapshot under dir foo
4887 * 3) delete the snapshot
4890 * 6) fsync foo or some file inside foo
4892 if (last_unlink_trans >= trans->transid)
4893 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4896 btrfs_end_transaction(trans);
4898 btrfs_btree_balance_dirty(fs_info);
4899 fscrypt_free_filename(&fname);
4905 * btrfs_truncate_block - read, zero a chunk and write a block
4906 * @inode - inode that we're zeroing
4907 * @from - the offset to start zeroing
4908 * @len - the length to zero, 0 to zero the entire range respective to the
4910 * @front - zero up to the offset instead of from the offset on
4912 * This will find the block for the "from" offset and cow the block and zero the
4913 * part we want to zero. This is used with truncate and hole punching.
4915 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4918 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4919 struct address_space *mapping = inode->vfs_inode.i_mapping;
4920 struct extent_io_tree *io_tree = &inode->io_tree;
4921 struct btrfs_ordered_extent *ordered;
4922 struct extent_state *cached_state = NULL;
4923 struct extent_changeset *data_reserved = NULL;
4924 bool only_release_metadata = false;
4925 u32 blocksize = fs_info->sectorsize;
4926 pgoff_t index = from >> PAGE_SHIFT;
4927 unsigned offset = from & (blocksize - 1);
4929 gfp_t mask = btrfs_alloc_write_mask(mapping);
4930 size_t write_bytes = blocksize;
4935 if (IS_ALIGNED(offset, blocksize) &&
4936 (!len || IS_ALIGNED(len, blocksize)))
4939 block_start = round_down(from, blocksize);
4940 block_end = block_start + blocksize - 1;
4942 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4945 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4946 /* For nocow case, no need to reserve data space */
4947 only_release_metadata = true;
4952 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4954 if (!only_release_metadata)
4955 btrfs_free_reserved_data_space(inode, data_reserved,
4956 block_start, blocksize);
4960 page = find_or_create_page(mapping, index, mask);
4962 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4964 btrfs_delalloc_release_extents(inode, blocksize);
4969 if (!PageUptodate(page)) {
4970 ret = btrfs_read_folio(NULL, page_folio(page));
4972 if (page->mapping != mapping) {
4977 if (!PageUptodate(page)) {
4984 * We unlock the page after the io is completed and then re-lock it
4985 * above. release_folio() could have come in between that and cleared
4986 * PagePrivate(), but left the page in the mapping. Set the page mapped
4987 * here to make sure it's properly set for the subpage stuff.
4989 ret = set_page_extent_mapped(page);
4993 wait_on_page_writeback(page);
4995 lock_extent(io_tree, block_start, block_end, &cached_state);
4997 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4999 unlock_extent(io_tree, block_start, block_end, &cached_state);
5002 btrfs_start_ordered_extent(ordered, 1);
5003 btrfs_put_ordered_extent(ordered);
5007 clear_extent_bit(&inode->io_tree, block_start, block_end,
5008 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
5011 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
5014 unlock_extent(io_tree, block_start, block_end, &cached_state);
5018 if (offset != blocksize) {
5020 len = blocksize - offset;
5022 memzero_page(page, (block_start - page_offset(page)),
5025 memzero_page(page, (block_start - page_offset(page)) + offset,
5028 btrfs_page_clear_checked(fs_info, page, block_start,
5029 block_end + 1 - block_start);
5030 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
5031 unlock_extent(io_tree, block_start, block_end, &cached_state);
5033 if (only_release_metadata)
5034 set_extent_bit(&inode->io_tree, block_start, block_end,
5035 EXTENT_NORESERVE, NULL, GFP_NOFS);
5039 if (only_release_metadata)
5040 btrfs_delalloc_release_metadata(inode, blocksize, true);
5042 btrfs_delalloc_release_space(inode, data_reserved,
5043 block_start, blocksize, true);
5045 btrfs_delalloc_release_extents(inode, blocksize);
5049 if (only_release_metadata)
5050 btrfs_check_nocow_unlock(inode);
5051 extent_changeset_free(data_reserved);
5055 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
5056 u64 offset, u64 len)
5058 struct btrfs_fs_info *fs_info = root->fs_info;
5059 struct btrfs_trans_handle *trans;
5060 struct btrfs_drop_extents_args drop_args = { 0 };
5064 * If NO_HOLES is enabled, we don't need to do anything.
5065 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
5066 * or btrfs_update_inode() will be called, which guarantee that the next
5067 * fsync will know this inode was changed and needs to be logged.
5069 if (btrfs_fs_incompat(fs_info, NO_HOLES))
5073 * 1 - for the one we're dropping
5074 * 1 - for the one we're adding
5075 * 1 - for updating the inode.
5077 trans = btrfs_start_transaction(root, 3);
5079 return PTR_ERR(trans);
5081 drop_args.start = offset;
5082 drop_args.end = offset + len;
5083 drop_args.drop_cache = true;
5085 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5087 btrfs_abort_transaction(trans, ret);
5088 btrfs_end_transaction(trans);
5092 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
5094 btrfs_abort_transaction(trans, ret);
5096 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5097 btrfs_update_inode(trans, root, inode);
5099 btrfs_end_transaction(trans);
5104 * This function puts in dummy file extents for the area we're creating a hole
5105 * for. So if we are truncating this file to a larger size we need to insert
5106 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5107 * the range between oldsize and size
5109 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5111 struct btrfs_root *root = inode->root;
5112 struct btrfs_fs_info *fs_info = root->fs_info;
5113 struct extent_io_tree *io_tree = &inode->io_tree;
5114 struct extent_map *em = NULL;
5115 struct extent_state *cached_state = NULL;
5116 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5117 u64 block_end = ALIGN(size, fs_info->sectorsize);
5124 * If our size started in the middle of a block we need to zero out the
5125 * rest of the block before we expand the i_size, otherwise we could
5126 * expose stale data.
5128 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5132 if (size <= hole_start)
5135 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5137 cur_offset = hole_start;
5139 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5140 block_end - cur_offset);
5146 last_byte = min(extent_map_end(em), block_end);
5147 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5148 hole_size = last_byte - cur_offset;
5150 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5151 struct extent_map *hole_em;
5153 err = maybe_insert_hole(root, inode, cur_offset,
5158 err = btrfs_inode_set_file_extent_range(inode,
5159 cur_offset, hole_size);
5163 hole_em = alloc_extent_map();
5165 btrfs_drop_extent_map_range(inode, cur_offset,
5166 cur_offset + hole_size - 1,
5168 btrfs_set_inode_full_sync(inode);
5171 hole_em->start = cur_offset;
5172 hole_em->len = hole_size;
5173 hole_em->orig_start = cur_offset;
5175 hole_em->block_start = EXTENT_MAP_HOLE;
5176 hole_em->block_len = 0;
5177 hole_em->orig_block_len = 0;
5178 hole_em->ram_bytes = hole_size;
5179 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5180 hole_em->generation = fs_info->generation;
5182 err = btrfs_replace_extent_map_range(inode, hole_em, true);
5183 free_extent_map(hole_em);
5185 err = btrfs_inode_set_file_extent_range(inode,
5186 cur_offset, hole_size);
5191 free_extent_map(em);
5193 cur_offset = last_byte;
5194 if (cur_offset >= block_end)
5197 free_extent_map(em);
5198 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5202 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5204 struct btrfs_root *root = BTRFS_I(inode)->root;
5205 struct btrfs_trans_handle *trans;
5206 loff_t oldsize = i_size_read(inode);
5207 loff_t newsize = attr->ia_size;
5208 int mask = attr->ia_valid;
5212 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5213 * special case where we need to update the times despite not having
5214 * these flags set. For all other operations the VFS set these flags
5215 * explicitly if it wants a timestamp update.
5217 if (newsize != oldsize) {
5218 inode_inc_iversion(inode);
5219 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5220 inode->i_mtime = current_time(inode);
5221 inode->i_ctime = inode->i_mtime;
5225 if (newsize > oldsize) {
5227 * Don't do an expanding truncate while snapshotting is ongoing.
5228 * This is to ensure the snapshot captures a fully consistent
5229 * state of this file - if the snapshot captures this expanding
5230 * truncation, it must capture all writes that happened before
5233 btrfs_drew_write_lock(&root->snapshot_lock);
5234 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5236 btrfs_drew_write_unlock(&root->snapshot_lock);
5240 trans = btrfs_start_transaction(root, 1);
5241 if (IS_ERR(trans)) {
5242 btrfs_drew_write_unlock(&root->snapshot_lock);
5243 return PTR_ERR(trans);
5246 i_size_write(inode, newsize);
5247 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5248 pagecache_isize_extended(inode, oldsize, newsize);
5249 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5250 btrfs_drew_write_unlock(&root->snapshot_lock);
5251 btrfs_end_transaction(trans);
5253 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5255 if (btrfs_is_zoned(fs_info)) {
5256 ret = btrfs_wait_ordered_range(inode,
5257 ALIGN(newsize, fs_info->sectorsize),
5264 * We're truncating a file that used to have good data down to
5265 * zero. Make sure any new writes to the file get on disk
5269 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5270 &BTRFS_I(inode)->runtime_flags);
5272 truncate_setsize(inode, newsize);
5274 inode_dio_wait(inode);
5276 ret = btrfs_truncate(inode, newsize == oldsize);
5277 if (ret && inode->i_nlink) {
5281 * Truncate failed, so fix up the in-memory size. We
5282 * adjusted disk_i_size down as we removed extents, so
5283 * wait for disk_i_size to be stable and then update the
5284 * in-memory size to match.
5286 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5289 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5296 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5299 struct inode *inode = d_inode(dentry);
5300 struct btrfs_root *root = BTRFS_I(inode)->root;
5303 if (btrfs_root_readonly(root))
5306 err = setattr_prepare(mnt_userns, dentry, attr);
5310 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5311 err = btrfs_setsize(inode, attr);
5316 if (attr->ia_valid) {
5317 setattr_copy(mnt_userns, inode, attr);
5318 inode_inc_iversion(inode);
5319 err = btrfs_dirty_inode(inode);
5321 if (!err && attr->ia_valid & ATTR_MODE)
5322 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5329 * While truncating the inode pages during eviction, we get the VFS
5330 * calling btrfs_invalidate_folio() against each folio of the inode. This
5331 * is slow because the calls to btrfs_invalidate_folio() result in a
5332 * huge amount of calls to lock_extent() and clear_extent_bit(),
5333 * which keep merging and splitting extent_state structures over and over,
5334 * wasting lots of time.
5336 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5337 * skip all those expensive operations on a per folio basis and do only
5338 * the ordered io finishing, while we release here the extent_map and
5339 * extent_state structures, without the excessive merging and splitting.
5341 static void evict_inode_truncate_pages(struct inode *inode)
5343 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5344 struct rb_node *node;
5346 ASSERT(inode->i_state & I_FREEING);
5347 truncate_inode_pages_final(&inode->i_data);
5349 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5352 * Keep looping until we have no more ranges in the io tree.
5353 * We can have ongoing bios started by readahead that have
5354 * their endio callback (extent_io.c:end_bio_extent_readpage)
5355 * still in progress (unlocked the pages in the bio but did not yet
5356 * unlocked the ranges in the io tree). Therefore this means some
5357 * ranges can still be locked and eviction started because before
5358 * submitting those bios, which are executed by a separate task (work
5359 * queue kthread), inode references (inode->i_count) were not taken
5360 * (which would be dropped in the end io callback of each bio).
5361 * Therefore here we effectively end up waiting for those bios and
5362 * anyone else holding locked ranges without having bumped the inode's
5363 * reference count - if we don't do it, when they access the inode's
5364 * io_tree to unlock a range it may be too late, leading to an
5365 * use-after-free issue.
5367 spin_lock(&io_tree->lock);
5368 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5369 struct extent_state *state;
5370 struct extent_state *cached_state = NULL;
5373 unsigned state_flags;
5375 node = rb_first(&io_tree->state);
5376 state = rb_entry(node, struct extent_state, rb_node);
5377 start = state->start;
5379 state_flags = state->state;
5380 spin_unlock(&io_tree->lock);
5382 lock_extent(io_tree, start, end, &cached_state);
5385 * If still has DELALLOC flag, the extent didn't reach disk,
5386 * and its reserved space won't be freed by delayed_ref.
5387 * So we need to free its reserved space here.
5388 * (Refer to comment in btrfs_invalidate_folio, case 2)
5390 * Note, end is the bytenr of last byte, so we need + 1 here.
5392 if (state_flags & EXTENT_DELALLOC)
5393 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5394 end - start + 1, NULL);
5396 clear_extent_bit(io_tree, start, end,
5397 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5401 spin_lock(&io_tree->lock);
5403 spin_unlock(&io_tree->lock);
5406 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5407 struct btrfs_block_rsv *rsv)
5409 struct btrfs_fs_info *fs_info = root->fs_info;
5410 struct btrfs_trans_handle *trans;
5411 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5415 * Eviction should be taking place at some place safe because of our
5416 * delayed iputs. However the normal flushing code will run delayed
5417 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5419 * We reserve the delayed_refs_extra here again because we can't use
5420 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5421 * above. We reserve our extra bit here because we generate a ton of
5422 * delayed refs activity by truncating.
5424 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5425 * if we fail to make this reservation we can re-try without the
5426 * delayed_refs_extra so we can make some forward progress.
5428 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5429 BTRFS_RESERVE_FLUSH_EVICT);
5431 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5432 BTRFS_RESERVE_FLUSH_EVICT);
5435 "could not allocate space for delete; will truncate on mount");
5436 return ERR_PTR(-ENOSPC);
5438 delayed_refs_extra = 0;
5441 trans = btrfs_join_transaction(root);
5445 if (delayed_refs_extra) {
5446 trans->block_rsv = &fs_info->trans_block_rsv;
5447 trans->bytes_reserved = delayed_refs_extra;
5448 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5449 delayed_refs_extra, 1);
5454 void btrfs_evict_inode(struct inode *inode)
5456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5457 struct btrfs_trans_handle *trans;
5458 struct btrfs_root *root = BTRFS_I(inode)->root;
5459 struct btrfs_block_rsv *rsv;
5462 trace_btrfs_inode_evict(inode);
5465 fsverity_cleanup_inode(inode);
5470 evict_inode_truncate_pages(inode);
5472 if (inode->i_nlink &&
5473 ((btrfs_root_refs(&root->root_item) != 0 &&
5474 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5475 btrfs_is_free_space_inode(BTRFS_I(inode))))
5478 if (is_bad_inode(inode))
5481 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5483 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5486 if (inode->i_nlink > 0) {
5487 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5488 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5493 * This makes sure the inode item in tree is uptodate and the space for
5494 * the inode update is released.
5496 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5501 * This drops any pending insert or delete operations we have for this
5502 * inode. We could have a delayed dir index deletion queued up, but
5503 * we're removing the inode completely so that'll be taken care of in
5506 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5508 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5511 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5512 rsv->failfast = true;
5514 btrfs_i_size_write(BTRFS_I(inode), 0);
5517 struct btrfs_truncate_control control = {
5518 .inode = BTRFS_I(inode),
5519 .ino = btrfs_ino(BTRFS_I(inode)),
5524 trans = evict_refill_and_join(root, rsv);
5528 trans->block_rsv = rsv;
5530 ret = btrfs_truncate_inode_items(trans, root, &control);
5531 trans->block_rsv = &fs_info->trans_block_rsv;
5532 btrfs_end_transaction(trans);
5533 btrfs_btree_balance_dirty(fs_info);
5534 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5541 * Errors here aren't a big deal, it just means we leave orphan items in
5542 * the tree. They will be cleaned up on the next mount. If the inode
5543 * number gets reused, cleanup deletes the orphan item without doing
5544 * anything, and unlink reuses the existing orphan item.
5546 * If it turns out that we are dropping too many of these, we might want
5547 * to add a mechanism for retrying these after a commit.
5549 trans = evict_refill_and_join(root, rsv);
5550 if (!IS_ERR(trans)) {
5551 trans->block_rsv = rsv;
5552 btrfs_orphan_del(trans, BTRFS_I(inode));
5553 trans->block_rsv = &fs_info->trans_block_rsv;
5554 btrfs_end_transaction(trans);
5558 btrfs_free_block_rsv(fs_info, rsv);
5561 * If we didn't successfully delete, the orphan item will still be in
5562 * the tree and we'll retry on the next mount. Again, we might also want
5563 * to retry these periodically in the future.
5565 btrfs_remove_delayed_node(BTRFS_I(inode));
5566 fsverity_cleanup_inode(inode);
5571 * Return the key found in the dir entry in the location pointer, fill @type
5572 * with BTRFS_FT_*, and return 0.
5574 * If no dir entries were found, returns -ENOENT.
5575 * If found a corrupted location in dir entry, returns -EUCLEAN.
5577 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5578 struct btrfs_key *location, u8 *type)
5580 struct btrfs_dir_item *di;
5581 struct btrfs_path *path;
5582 struct btrfs_root *root = BTRFS_I(dir)->root;
5584 struct fscrypt_name fname;
5586 path = btrfs_alloc_path();
5590 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
5594 /* This needs to handle no-key deletions later on */
5596 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5597 &fname.disk_name, 0);
5598 if (IS_ERR_OR_NULL(di)) {
5599 ret = di ? PTR_ERR(di) : -ENOENT;
5603 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5604 if (location->type != BTRFS_INODE_ITEM_KEY &&
5605 location->type != BTRFS_ROOT_ITEM_KEY) {
5607 btrfs_warn(root->fs_info,
5608 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5609 __func__, fname.disk_name.name, btrfs_ino(BTRFS_I(dir)),
5610 location->objectid, location->type, location->offset);
5613 *type = btrfs_dir_type(path->nodes[0], di);
5615 fscrypt_free_filename(&fname);
5616 btrfs_free_path(path);
5621 * when we hit a tree root in a directory, the btrfs part of the inode
5622 * needs to be changed to reflect the root directory of the tree root. This
5623 * is kind of like crossing a mount point.
5625 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5627 struct dentry *dentry,
5628 struct btrfs_key *location,
5629 struct btrfs_root **sub_root)
5631 struct btrfs_path *path;
5632 struct btrfs_root *new_root;
5633 struct btrfs_root_ref *ref;
5634 struct extent_buffer *leaf;
5635 struct btrfs_key key;
5638 struct fscrypt_name fname;
5640 ret = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
5644 path = btrfs_alloc_path();
5651 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5652 key.type = BTRFS_ROOT_REF_KEY;
5653 key.offset = location->objectid;
5655 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5662 leaf = path->nodes[0];
5663 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5664 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5665 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5668 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5669 (unsigned long)(ref + 1), fname.disk_name.len);
5673 btrfs_release_path(path);
5675 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5676 if (IS_ERR(new_root)) {
5677 err = PTR_ERR(new_root);
5681 *sub_root = new_root;
5682 location->objectid = btrfs_root_dirid(&new_root->root_item);
5683 location->type = BTRFS_INODE_ITEM_KEY;
5684 location->offset = 0;
5687 btrfs_free_path(path);
5688 fscrypt_free_filename(&fname);
5692 static void inode_tree_add(struct inode *inode)
5694 struct btrfs_root *root = BTRFS_I(inode)->root;
5695 struct btrfs_inode *entry;
5697 struct rb_node *parent;
5698 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5699 u64 ino = btrfs_ino(BTRFS_I(inode));
5701 if (inode_unhashed(inode))
5704 spin_lock(&root->inode_lock);
5705 p = &root->inode_tree.rb_node;
5708 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5710 if (ino < btrfs_ino(entry))
5711 p = &parent->rb_left;
5712 else if (ino > btrfs_ino(entry))
5713 p = &parent->rb_right;
5715 WARN_ON(!(entry->vfs_inode.i_state &
5716 (I_WILL_FREE | I_FREEING)));
5717 rb_replace_node(parent, new, &root->inode_tree);
5718 RB_CLEAR_NODE(parent);
5719 spin_unlock(&root->inode_lock);
5723 rb_link_node(new, parent, p);
5724 rb_insert_color(new, &root->inode_tree);
5725 spin_unlock(&root->inode_lock);
5728 static void inode_tree_del(struct btrfs_inode *inode)
5730 struct btrfs_root *root = inode->root;
5733 spin_lock(&root->inode_lock);
5734 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5735 rb_erase(&inode->rb_node, &root->inode_tree);
5736 RB_CLEAR_NODE(&inode->rb_node);
5737 empty = RB_EMPTY_ROOT(&root->inode_tree);
5739 spin_unlock(&root->inode_lock);
5741 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5742 spin_lock(&root->inode_lock);
5743 empty = RB_EMPTY_ROOT(&root->inode_tree);
5744 spin_unlock(&root->inode_lock);
5746 btrfs_add_dead_root(root);
5751 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5753 struct btrfs_iget_args *args = p;
5755 inode->i_ino = args->ino;
5756 BTRFS_I(inode)->location.objectid = args->ino;
5757 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5758 BTRFS_I(inode)->location.offset = 0;
5759 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5760 BUG_ON(args->root && !BTRFS_I(inode)->root);
5762 if (args->root && args->root == args->root->fs_info->tree_root &&
5763 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5764 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5765 &BTRFS_I(inode)->runtime_flags);
5769 static int btrfs_find_actor(struct inode *inode, void *opaque)
5771 struct btrfs_iget_args *args = opaque;
5773 return args->ino == BTRFS_I(inode)->location.objectid &&
5774 args->root == BTRFS_I(inode)->root;
5777 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5778 struct btrfs_root *root)
5780 struct inode *inode;
5781 struct btrfs_iget_args args;
5782 unsigned long hashval = btrfs_inode_hash(ino, root);
5787 inode = iget5_locked(s, hashval, btrfs_find_actor,
5788 btrfs_init_locked_inode,
5794 * Get an inode object given its inode number and corresponding root.
5795 * Path can be preallocated to prevent recursing back to iget through
5796 * allocator. NULL is also valid but may require an additional allocation
5799 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5800 struct btrfs_root *root, struct btrfs_path *path)
5802 struct inode *inode;
5804 inode = btrfs_iget_locked(s, ino, root);
5806 return ERR_PTR(-ENOMEM);
5808 if (inode->i_state & I_NEW) {
5811 ret = btrfs_read_locked_inode(inode, path);
5813 inode_tree_add(inode);
5814 unlock_new_inode(inode);
5818 * ret > 0 can come from btrfs_search_slot called by
5819 * btrfs_read_locked_inode, this means the inode item
5824 inode = ERR_PTR(ret);
5831 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5833 return btrfs_iget_path(s, ino, root, NULL);
5836 static struct inode *new_simple_dir(struct super_block *s,
5837 struct btrfs_key *key,
5838 struct btrfs_root *root)
5840 struct inode *inode = new_inode(s);
5843 return ERR_PTR(-ENOMEM);
5845 BTRFS_I(inode)->root = btrfs_grab_root(root);
5846 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5847 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5849 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5851 * We only need lookup, the rest is read-only and there's no inode
5852 * associated with the dentry
5854 inode->i_op = &simple_dir_inode_operations;
5855 inode->i_opflags &= ~IOP_XATTR;
5856 inode->i_fop = &simple_dir_operations;
5857 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5858 inode->i_mtime = current_time(inode);
5859 inode->i_atime = inode->i_mtime;
5860 inode->i_ctime = inode->i_mtime;
5861 BTRFS_I(inode)->i_otime = inode->i_mtime;
5866 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5867 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5868 static_assert(BTRFS_FT_DIR == FT_DIR);
5869 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5870 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5871 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5872 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5873 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5875 static inline u8 btrfs_inode_type(struct inode *inode)
5877 return fs_umode_to_ftype(inode->i_mode);
5880 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5882 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5883 struct inode *inode;
5884 struct btrfs_root *root = BTRFS_I(dir)->root;
5885 struct btrfs_root *sub_root = root;
5886 struct btrfs_key location;
5890 if (dentry->d_name.len > BTRFS_NAME_LEN)
5891 return ERR_PTR(-ENAMETOOLONG);
5893 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5895 return ERR_PTR(ret);
5897 if (location.type == BTRFS_INODE_ITEM_KEY) {
5898 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5902 /* Do extra check against inode mode with di_type */
5903 if (btrfs_inode_type(inode) != di_type) {
5905 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5906 inode->i_mode, btrfs_inode_type(inode),
5909 return ERR_PTR(-EUCLEAN);
5914 ret = fixup_tree_root_location(fs_info, dir, dentry,
5915 &location, &sub_root);
5918 inode = ERR_PTR(ret);
5920 inode = new_simple_dir(dir->i_sb, &location, root);
5922 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5923 btrfs_put_root(sub_root);
5928 down_read(&fs_info->cleanup_work_sem);
5929 if (!sb_rdonly(inode->i_sb))
5930 ret = btrfs_orphan_cleanup(sub_root);
5931 up_read(&fs_info->cleanup_work_sem);
5934 inode = ERR_PTR(ret);
5941 static int btrfs_dentry_delete(const struct dentry *dentry)
5943 struct btrfs_root *root;
5944 struct inode *inode = d_inode(dentry);
5946 if (!inode && !IS_ROOT(dentry))
5947 inode = d_inode(dentry->d_parent);
5950 root = BTRFS_I(inode)->root;
5951 if (btrfs_root_refs(&root->root_item) == 0)
5954 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5960 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5963 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5965 if (inode == ERR_PTR(-ENOENT))
5967 return d_splice_alias(inode, dentry);
5971 * Find the highest existing sequence number in a directory and then set the
5972 * in-memory index_cnt variable to the first free sequence number.
5974 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5976 struct btrfs_root *root = inode->root;
5977 struct btrfs_key key, found_key;
5978 struct btrfs_path *path;
5979 struct extent_buffer *leaf;
5982 key.objectid = btrfs_ino(inode);
5983 key.type = BTRFS_DIR_INDEX_KEY;
5984 key.offset = (u64)-1;
5986 path = btrfs_alloc_path();
5990 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5993 /* FIXME: we should be able to handle this */
5998 if (path->slots[0] == 0) {
5999 inode->index_cnt = BTRFS_DIR_START_INDEX;
6005 leaf = path->nodes[0];
6006 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6008 if (found_key.objectid != btrfs_ino(inode) ||
6009 found_key.type != BTRFS_DIR_INDEX_KEY) {
6010 inode->index_cnt = BTRFS_DIR_START_INDEX;
6014 inode->index_cnt = found_key.offset + 1;
6016 btrfs_free_path(path);
6020 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
6024 btrfs_inode_lock(&dir->vfs_inode, 0);
6025 if (dir->index_cnt == (u64)-1) {
6026 ret = btrfs_inode_delayed_dir_index_count(dir);
6028 ret = btrfs_set_inode_index_count(dir);
6034 /* index_cnt is the index number of next new entry, so decrement it. */
6035 *index = dir->index_cnt - 1;
6037 btrfs_inode_unlock(&dir->vfs_inode, 0);
6043 * All this infrastructure exists because dir_emit can fault, and we are holding
6044 * the tree lock when doing readdir. For now just allocate a buffer and copy
6045 * our information into that, and then dir_emit from the buffer. This is
6046 * similar to what NFS does, only we don't keep the buffer around in pagecache
6047 * because I'm afraid I'll mess that up. Long term we need to make filldir do
6048 * copy_to_user_inatomic so we don't have to worry about page faulting under the
6051 static int btrfs_opendir(struct inode *inode, struct file *file)
6053 struct btrfs_file_private *private;
6057 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
6061 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
6064 private->last_index = last_index;
6065 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
6066 if (!private->filldir_buf) {
6070 file->private_data = private;
6074 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
6076 struct btrfs_file_private *private = file->private_data;
6079 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
6080 &private->last_index);
6084 return generic_file_llseek(file, offset, whence);
6094 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
6097 struct dir_entry *entry = addr;
6098 char *name = (char *)(entry + 1);
6100 ctx->pos = get_unaligned(&entry->offset);
6101 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
6102 get_unaligned(&entry->ino),
6103 get_unaligned(&entry->type)))
6105 addr += sizeof(struct dir_entry) +
6106 get_unaligned(&entry->name_len);
6112 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
6114 struct inode *inode = file_inode(file);
6115 struct btrfs_root *root = BTRFS_I(inode)->root;
6116 struct btrfs_file_private *private = file->private_data;
6117 struct btrfs_dir_item *di;
6118 struct btrfs_key key;
6119 struct btrfs_key found_key;
6120 struct btrfs_path *path;
6122 struct list_head ins_list;
6123 struct list_head del_list;
6130 struct btrfs_key location;
6132 if (!dir_emit_dots(file, ctx))
6135 path = btrfs_alloc_path();
6139 addr = private->filldir_buf;
6140 path->reada = READA_FORWARD;
6142 INIT_LIST_HEAD(&ins_list);
6143 INIT_LIST_HEAD(&del_list);
6144 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
6145 &ins_list, &del_list);
6148 key.type = BTRFS_DIR_INDEX_KEY;
6149 key.offset = ctx->pos;
6150 key.objectid = btrfs_ino(BTRFS_I(inode));
6152 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
6153 struct dir_entry *entry;
6154 struct extent_buffer *leaf = path->nodes[0];
6156 if (found_key.objectid != key.objectid)
6158 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6160 if (found_key.offset < ctx->pos)
6162 if (found_key.offset > private->last_index)
6164 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6166 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
6167 name_len = btrfs_dir_name_len(leaf, di);
6168 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6170 btrfs_release_path(path);
6171 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6174 addr = private->filldir_buf;
6181 put_unaligned(name_len, &entry->name_len);
6182 name_ptr = (char *)(entry + 1);
6183 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6185 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6187 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6188 put_unaligned(location.objectid, &entry->ino);
6189 put_unaligned(found_key.offset, &entry->offset);
6191 addr += sizeof(struct dir_entry) + name_len;
6192 total_len += sizeof(struct dir_entry) + name_len;
6194 /* Catch error encountered during iteration */
6198 btrfs_release_path(path);
6200 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6204 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6209 * Stop new entries from being returned after we return the last
6212 * New directory entries are assigned a strictly increasing
6213 * offset. This means that new entries created during readdir
6214 * are *guaranteed* to be seen in the future by that readdir.
6215 * This has broken buggy programs which operate on names as
6216 * they're returned by readdir. Until we re-use freed offsets
6217 * we have this hack to stop new entries from being returned
6218 * under the assumption that they'll never reach this huge
6221 * This is being careful not to overflow 32bit loff_t unless the
6222 * last entry requires it because doing so has broken 32bit apps
6225 if (ctx->pos >= INT_MAX)
6226 ctx->pos = LLONG_MAX;
6233 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6234 btrfs_free_path(path);
6239 * This is somewhat expensive, updating the tree every time the
6240 * inode changes. But, it is most likely to find the inode in cache.
6241 * FIXME, needs more benchmarking...there are no reasons other than performance
6242 * to keep or drop this code.
6244 static int btrfs_dirty_inode(struct inode *inode)
6246 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6247 struct btrfs_root *root = BTRFS_I(inode)->root;
6248 struct btrfs_trans_handle *trans;
6251 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6254 trans = btrfs_join_transaction(root);
6256 return PTR_ERR(trans);
6258 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6259 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6260 /* whoops, lets try again with the full transaction */
6261 btrfs_end_transaction(trans);
6262 trans = btrfs_start_transaction(root, 1);
6264 return PTR_ERR(trans);
6266 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6268 btrfs_end_transaction(trans);
6269 if (BTRFS_I(inode)->delayed_node)
6270 btrfs_balance_delayed_items(fs_info);
6276 * This is a copy of file_update_time. We need this so we can return error on
6277 * ENOSPC for updating the inode in the case of file write and mmap writes.
6279 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6282 struct btrfs_root *root = BTRFS_I(inode)->root;
6283 bool dirty = flags & ~S_VERSION;
6285 if (btrfs_root_readonly(root))
6288 if (flags & S_VERSION)
6289 dirty |= inode_maybe_inc_iversion(inode, dirty);
6290 if (flags & S_CTIME)
6291 inode->i_ctime = *now;
6292 if (flags & S_MTIME)
6293 inode->i_mtime = *now;
6294 if (flags & S_ATIME)
6295 inode->i_atime = *now;
6296 return dirty ? btrfs_dirty_inode(inode) : 0;
6300 * helper to find a free sequence number in a given directory. This current
6301 * code is very simple, later versions will do smarter things in the btree
6303 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6307 if (dir->index_cnt == (u64)-1) {
6308 ret = btrfs_inode_delayed_dir_index_count(dir);
6310 ret = btrfs_set_inode_index_count(dir);
6316 *index = dir->index_cnt;
6322 static int btrfs_insert_inode_locked(struct inode *inode)
6324 struct btrfs_iget_args args;
6326 args.ino = BTRFS_I(inode)->location.objectid;
6327 args.root = BTRFS_I(inode)->root;
6329 return insert_inode_locked4(inode,
6330 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6331 btrfs_find_actor, &args);
6334 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6335 unsigned int *trans_num_items)
6337 struct inode *dir = args->dir;
6338 struct inode *inode = args->inode;
6341 if (!args->orphan) {
6342 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6348 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6350 fscrypt_free_filename(&args->fname);
6354 /* 1 to add inode item */
6355 *trans_num_items = 1;
6356 /* 1 to add compression property */
6357 if (BTRFS_I(dir)->prop_compress)
6358 (*trans_num_items)++;
6359 /* 1 to add default ACL xattr */
6360 if (args->default_acl)
6361 (*trans_num_items)++;
6362 /* 1 to add access ACL xattr */
6364 (*trans_num_items)++;
6365 #ifdef CONFIG_SECURITY
6366 /* 1 to add LSM xattr */
6367 if (dir->i_security)
6368 (*trans_num_items)++;
6371 /* 1 to add orphan item */
6372 (*trans_num_items)++;
6376 * 1 to add dir index
6377 * 1 to update parent inode item
6379 * No need for 1 unit for the inode ref item because it is
6380 * inserted in a batch together with the inode item at
6381 * btrfs_create_new_inode().
6383 *trans_num_items += 3;
6388 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6390 posix_acl_release(args->acl);
6391 posix_acl_release(args->default_acl);
6392 fscrypt_free_filename(&args->fname);
6396 * Inherit flags from the parent inode.
6398 * Currently only the compression flags and the cow flags are inherited.
6400 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6404 flags = BTRFS_I(dir)->flags;
6406 if (flags & BTRFS_INODE_NOCOMPRESS) {
6407 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6408 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6409 } else if (flags & BTRFS_INODE_COMPRESS) {
6410 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6411 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6414 if (flags & BTRFS_INODE_NODATACOW) {
6415 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6416 if (S_ISREG(inode->i_mode))
6417 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6420 btrfs_sync_inode_flags_to_i_flags(inode);
6423 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6424 struct btrfs_new_inode_args *args)
6426 struct inode *dir = args->dir;
6427 struct inode *inode = args->inode;
6428 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6429 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6430 struct btrfs_root *root;
6431 struct btrfs_inode_item *inode_item;
6432 struct btrfs_key *location;
6433 struct btrfs_path *path;
6435 struct btrfs_inode_ref *ref;
6436 struct btrfs_key key[2];
6438 struct btrfs_item_batch batch;
6442 path = btrfs_alloc_path();
6447 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6448 root = BTRFS_I(inode)->root;
6450 ret = btrfs_get_free_objectid(root, &objectid);
6453 inode->i_ino = objectid;
6457 * O_TMPFILE, set link count to 0, so that after this point, we
6458 * fill in an inode item with the correct link count.
6460 set_nlink(inode, 0);
6462 trace_btrfs_inode_request(dir);
6464 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6468 /* index_cnt is ignored for everything but a dir. */
6469 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6470 BTRFS_I(inode)->generation = trans->transid;
6471 inode->i_generation = BTRFS_I(inode)->generation;
6474 * Subvolumes don't inherit flags from their parent directory.
6475 * Originally this was probably by accident, but we probably can't
6476 * change it now without compatibility issues.
6479 btrfs_inherit_iflags(inode, dir);
6481 if (S_ISREG(inode->i_mode)) {
6482 if (btrfs_test_opt(fs_info, NODATASUM))
6483 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6484 if (btrfs_test_opt(fs_info, NODATACOW))
6485 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6486 BTRFS_INODE_NODATASUM;
6489 location = &BTRFS_I(inode)->location;
6490 location->objectid = objectid;
6491 location->offset = 0;
6492 location->type = BTRFS_INODE_ITEM_KEY;
6494 ret = btrfs_insert_inode_locked(inode);
6497 BTRFS_I(dir)->index_cnt--;
6502 * We could have gotten an inode number from somebody who was fsynced
6503 * and then removed in this same transaction, so let's just set full
6504 * sync since it will be a full sync anyway and this will blow away the
6505 * old info in the log.
6507 btrfs_set_inode_full_sync(BTRFS_I(inode));
6509 key[0].objectid = objectid;
6510 key[0].type = BTRFS_INODE_ITEM_KEY;
6513 sizes[0] = sizeof(struct btrfs_inode_item);
6515 if (!args->orphan) {
6517 * Start new inodes with an inode_ref. This is slightly more
6518 * efficient for small numbers of hard links since they will
6519 * be packed into one item. Extended refs will kick in if we
6520 * add more hard links than can fit in the ref item.
6522 key[1].objectid = objectid;
6523 key[1].type = BTRFS_INODE_REF_KEY;
6525 key[1].offset = objectid;
6526 sizes[1] = 2 + sizeof(*ref);
6528 key[1].offset = btrfs_ino(BTRFS_I(dir));
6529 sizes[1] = name->len + sizeof(*ref);
6533 batch.keys = &key[0];
6534 batch.data_sizes = &sizes[0];
6535 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6536 batch.nr = args->orphan ? 1 : 2;
6537 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6539 btrfs_abort_transaction(trans, ret);
6543 inode->i_mtime = current_time(inode);
6544 inode->i_atime = inode->i_mtime;
6545 inode->i_ctime = inode->i_mtime;
6546 BTRFS_I(inode)->i_otime = inode->i_mtime;
6549 * We're going to fill the inode item now, so at this point the inode
6550 * must be fully initialized.
6553 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6554 struct btrfs_inode_item);
6555 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6556 sizeof(*inode_item));
6557 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6559 if (!args->orphan) {
6560 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6561 struct btrfs_inode_ref);
6562 ptr = (unsigned long)(ref + 1);
6564 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6565 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6566 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6568 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6570 btrfs_set_inode_ref_index(path->nodes[0], ref,
6571 BTRFS_I(inode)->dir_index);
6572 write_extent_buffer(path->nodes[0], name->name, ptr,
6577 btrfs_mark_buffer_dirty(path->nodes[0]);
6579 * We don't need the path anymore, plus inheriting properties, adding
6580 * ACLs, security xattrs, orphan item or adding the link, will result in
6581 * allocating yet another path. So just free our path.
6583 btrfs_free_path(path);
6587 struct inode *parent;
6590 * Subvolumes inherit properties from their parent subvolume,
6591 * not the directory they were created in.
6593 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6594 BTRFS_I(dir)->root);
6595 if (IS_ERR(parent)) {
6596 ret = PTR_ERR(parent);
6598 ret = btrfs_inode_inherit_props(trans, inode, parent);
6602 ret = btrfs_inode_inherit_props(trans, inode, dir);
6606 "error inheriting props for ino %llu (root %llu): %d",
6607 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6612 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6615 if (!args->subvol) {
6616 ret = btrfs_init_inode_security(trans, args);
6618 btrfs_abort_transaction(trans, ret);
6623 inode_tree_add(inode);
6625 trace_btrfs_inode_new(inode);
6626 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6628 btrfs_update_root_times(trans, root);
6631 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6633 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6634 0, BTRFS_I(inode)->dir_index);
6637 btrfs_abort_transaction(trans, ret);
6645 * discard_new_inode() calls iput(), but the caller owns the reference
6649 discard_new_inode(inode);
6651 btrfs_free_path(path);
6656 * utility function to add 'inode' into 'parent_inode' with
6657 * a give name and a given sequence number.
6658 * if 'add_backref' is true, also insert a backref from the
6659 * inode to the parent directory.
6661 int btrfs_add_link(struct btrfs_trans_handle *trans,
6662 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6663 const struct fscrypt_str *name, int add_backref, u64 index)
6666 struct btrfs_key key;
6667 struct btrfs_root *root = parent_inode->root;
6668 u64 ino = btrfs_ino(inode);
6669 u64 parent_ino = btrfs_ino(parent_inode);
6671 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6672 memcpy(&key, &inode->root->root_key, sizeof(key));
6675 key.type = BTRFS_INODE_ITEM_KEY;
6679 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6680 ret = btrfs_add_root_ref(trans, key.objectid,
6681 root->root_key.objectid, parent_ino,
6683 } else if (add_backref) {
6684 ret = btrfs_insert_inode_ref(trans, root, name,
6685 ino, parent_ino, index);
6688 /* Nothing to clean up yet */
6692 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6693 btrfs_inode_type(&inode->vfs_inode), index);
6694 if (ret == -EEXIST || ret == -EOVERFLOW)
6697 btrfs_abort_transaction(trans, ret);
6701 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6703 inode_inc_iversion(&parent_inode->vfs_inode);
6705 * If we are replaying a log tree, we do not want to update the mtime
6706 * and ctime of the parent directory with the current time, since the
6707 * log replay procedure is responsible for setting them to their correct
6708 * values (the ones it had when the fsync was done).
6710 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6711 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6713 parent_inode->vfs_inode.i_mtime = now;
6714 parent_inode->vfs_inode.i_ctime = now;
6716 ret = btrfs_update_inode(trans, root, parent_inode);
6718 btrfs_abort_transaction(trans, ret);
6722 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6725 err = btrfs_del_root_ref(trans, key.objectid,
6726 root->root_key.objectid, parent_ino,
6727 &local_index, name);
6729 btrfs_abort_transaction(trans, err);
6730 } else if (add_backref) {
6734 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6737 btrfs_abort_transaction(trans, err);
6740 /* Return the original error code */
6744 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6745 struct inode *inode)
6747 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6748 struct btrfs_root *root = BTRFS_I(dir)->root;
6749 struct btrfs_new_inode_args new_inode_args = {
6754 unsigned int trans_num_items;
6755 struct btrfs_trans_handle *trans;
6758 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6762 trans = btrfs_start_transaction(root, trans_num_items);
6763 if (IS_ERR(trans)) {
6764 err = PTR_ERR(trans);
6765 goto out_new_inode_args;
6768 err = btrfs_create_new_inode(trans, &new_inode_args);
6770 d_instantiate_new(dentry, inode);
6772 btrfs_end_transaction(trans);
6773 btrfs_btree_balance_dirty(fs_info);
6775 btrfs_new_inode_args_destroy(&new_inode_args);
6782 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6783 struct dentry *dentry, umode_t mode, dev_t rdev)
6785 struct inode *inode;
6787 inode = new_inode(dir->i_sb);
6790 inode_init_owner(mnt_userns, inode, dir, mode);
6791 inode->i_op = &btrfs_special_inode_operations;
6792 init_special_inode(inode, inode->i_mode, rdev);
6793 return btrfs_create_common(dir, dentry, inode);
6796 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6797 struct dentry *dentry, umode_t mode, bool excl)
6799 struct inode *inode;
6801 inode = new_inode(dir->i_sb);
6804 inode_init_owner(mnt_userns, inode, dir, mode);
6805 inode->i_fop = &btrfs_file_operations;
6806 inode->i_op = &btrfs_file_inode_operations;
6807 inode->i_mapping->a_ops = &btrfs_aops;
6808 return btrfs_create_common(dir, dentry, inode);
6811 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6812 struct dentry *dentry)
6814 struct btrfs_trans_handle *trans = NULL;
6815 struct btrfs_root *root = BTRFS_I(dir)->root;
6816 struct inode *inode = d_inode(old_dentry);
6817 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6818 struct fscrypt_name fname;
6823 /* do not allow sys_link's with other subvols of the same device */
6824 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6827 if (inode->i_nlink >= BTRFS_LINK_MAX)
6830 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6834 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6839 * 2 items for inode and inode ref
6840 * 2 items for dir items
6841 * 1 item for parent inode
6842 * 1 item for orphan item deletion if O_TMPFILE
6844 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6845 if (IS_ERR(trans)) {
6846 err = PTR_ERR(trans);
6851 /* There are several dir indexes for this inode, clear the cache. */
6852 BTRFS_I(inode)->dir_index = 0ULL;
6854 inode_inc_iversion(inode);
6855 inode->i_ctime = current_time(inode);
6857 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6859 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6860 &fname.disk_name, 1, index);
6865 struct dentry *parent = dentry->d_parent;
6867 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6870 if (inode->i_nlink == 1) {
6872 * If new hard link count is 1, it's a file created
6873 * with open(2) O_TMPFILE flag.
6875 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6879 d_instantiate(dentry, inode);
6880 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6884 fscrypt_free_filename(&fname);
6886 btrfs_end_transaction(trans);
6888 inode_dec_link_count(inode);
6891 btrfs_btree_balance_dirty(fs_info);
6895 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6896 struct dentry *dentry, umode_t mode)
6898 struct inode *inode;
6900 inode = new_inode(dir->i_sb);
6903 inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode);
6904 inode->i_op = &btrfs_dir_inode_operations;
6905 inode->i_fop = &btrfs_dir_file_operations;
6906 return btrfs_create_common(dir, dentry, inode);
6909 static noinline int uncompress_inline(struct btrfs_path *path,
6911 size_t pg_offset, u64 extent_offset,
6912 struct btrfs_file_extent_item *item)
6915 struct extent_buffer *leaf = path->nodes[0];
6918 unsigned long inline_size;
6922 WARN_ON(pg_offset != 0);
6923 compress_type = btrfs_file_extent_compression(leaf, item);
6924 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6925 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6926 tmp = kmalloc(inline_size, GFP_NOFS);
6929 ptr = btrfs_file_extent_inline_start(item);
6931 read_extent_buffer(leaf, tmp, ptr, inline_size);
6933 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6934 ret = btrfs_decompress(compress_type, tmp, page,
6935 extent_offset, inline_size, max_size);
6938 * decompression code contains a memset to fill in any space between the end
6939 * of the uncompressed data and the end of max_size in case the decompressed
6940 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6941 * the end of an inline extent and the beginning of the next block, so we
6942 * cover that region here.
6945 if (max_size + pg_offset < PAGE_SIZE)
6946 memzero_page(page, pg_offset + max_size,
6947 PAGE_SIZE - max_size - pg_offset);
6953 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6954 * @inode: file to search in
6955 * @page: page to read extent data into if the extent is inline
6956 * @pg_offset: offset into @page to copy to
6957 * @start: file offset
6958 * @len: length of range starting at @start
6960 * This returns the first &struct extent_map which overlaps with the given
6961 * range, reading it from the B-tree and caching it if necessary. Note that
6962 * there may be more extents which overlap the given range after the returned
6965 * If @page is not NULL and the extent is inline, this also reads the extent
6966 * data directly into the page and marks the extent up to date in the io_tree.
6968 * Return: ERR_PTR on error, non-NULL extent_map on success.
6970 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6971 struct page *page, size_t pg_offset,
6974 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6976 u64 extent_start = 0;
6978 u64 objectid = btrfs_ino(inode);
6979 int extent_type = -1;
6980 struct btrfs_path *path = NULL;
6981 struct btrfs_root *root = inode->root;
6982 struct btrfs_file_extent_item *item;
6983 struct extent_buffer *leaf;
6984 struct btrfs_key found_key;
6985 struct extent_map *em = NULL;
6986 struct extent_map_tree *em_tree = &inode->extent_tree;
6988 read_lock(&em_tree->lock);
6989 em = lookup_extent_mapping(em_tree, start, len);
6990 read_unlock(&em_tree->lock);
6993 if (em->start > start || em->start + em->len <= start)
6994 free_extent_map(em);
6995 else if (em->block_start == EXTENT_MAP_INLINE && page)
6996 free_extent_map(em);
7000 em = alloc_extent_map();
7005 em->start = EXTENT_MAP_HOLE;
7006 em->orig_start = EXTENT_MAP_HOLE;
7008 em->block_len = (u64)-1;
7010 path = btrfs_alloc_path();
7016 /* Chances are we'll be called again, so go ahead and do readahead */
7017 path->reada = READA_FORWARD;
7020 * The same explanation in load_free_space_cache applies here as well,
7021 * we only read when we're loading the free space cache, and at that
7022 * point the commit_root has everything we need.
7024 if (btrfs_is_free_space_inode(inode)) {
7025 path->search_commit_root = 1;
7026 path->skip_locking = 1;
7029 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
7032 } else if (ret > 0) {
7033 if (path->slots[0] == 0)
7039 leaf = path->nodes[0];
7040 item = btrfs_item_ptr(leaf, path->slots[0],
7041 struct btrfs_file_extent_item);
7042 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7043 if (found_key.objectid != objectid ||
7044 found_key.type != BTRFS_EXTENT_DATA_KEY) {
7046 * If we backup past the first extent we want to move forward
7047 * and see if there is an extent in front of us, otherwise we'll
7048 * say there is a hole for our whole search range which can
7055 extent_type = btrfs_file_extent_type(leaf, item);
7056 extent_start = found_key.offset;
7057 extent_end = btrfs_file_extent_end(path);
7058 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7059 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7060 /* Only regular file could have regular/prealloc extent */
7061 if (!S_ISREG(inode->vfs_inode.i_mode)) {
7064 "regular/prealloc extent found for non-regular inode %llu",
7068 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7070 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7071 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7076 if (start >= extent_end) {
7078 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7079 ret = btrfs_next_leaf(root, path);
7085 leaf = path->nodes[0];
7087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7088 if (found_key.objectid != objectid ||
7089 found_key.type != BTRFS_EXTENT_DATA_KEY)
7091 if (start + len <= found_key.offset)
7093 if (start > found_key.offset)
7096 /* New extent overlaps with existing one */
7098 em->orig_start = start;
7099 em->len = found_key.offset - start;
7100 em->block_start = EXTENT_MAP_HOLE;
7104 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
7106 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7107 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7109 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7113 size_t extent_offset;
7119 size = btrfs_file_extent_ram_bytes(leaf, item);
7120 extent_offset = page_offset(page) + pg_offset - extent_start;
7121 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7122 size - extent_offset);
7123 em->start = extent_start + extent_offset;
7124 em->len = ALIGN(copy_size, fs_info->sectorsize);
7125 em->orig_block_len = em->len;
7126 em->orig_start = em->start;
7127 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7129 if (!PageUptodate(page)) {
7130 if (btrfs_file_extent_compression(leaf, item) !=
7131 BTRFS_COMPRESS_NONE) {
7132 ret = uncompress_inline(path, page, pg_offset,
7133 extent_offset, item);
7137 map = kmap_local_page(page);
7138 read_extent_buffer(leaf, map + pg_offset, ptr,
7140 if (pg_offset + copy_size < PAGE_SIZE) {
7141 memset(map + pg_offset + copy_size, 0,
7142 PAGE_SIZE - pg_offset -
7147 flush_dcache_page(page);
7153 em->orig_start = start;
7155 em->block_start = EXTENT_MAP_HOLE;
7158 btrfs_release_path(path);
7159 if (em->start > start || extent_map_end(em) <= start) {
7161 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7162 em->start, em->len, start, len);
7167 write_lock(&em_tree->lock);
7168 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7169 write_unlock(&em_tree->lock);
7171 btrfs_free_path(path);
7173 trace_btrfs_get_extent(root, inode, em);
7176 free_extent_map(em);
7177 return ERR_PTR(ret);
7182 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7185 const u64 orig_start,
7186 const u64 block_start,
7187 const u64 block_len,
7188 const u64 orig_block_len,
7189 const u64 ram_bytes,
7192 struct extent_map *em = NULL;
7195 if (type != BTRFS_ORDERED_NOCOW) {
7196 em = create_io_em(inode, start, len, orig_start, block_start,
7197 block_len, orig_block_len, ram_bytes,
7198 BTRFS_COMPRESS_NONE, /* compress_type */
7203 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7206 (1 << BTRFS_ORDERED_DIRECT),
7207 BTRFS_COMPRESS_NONE);
7210 free_extent_map(em);
7211 btrfs_drop_extent_map_range(inode, start,
7212 start + len - 1, false);
7221 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7224 struct btrfs_root *root = inode->root;
7225 struct btrfs_fs_info *fs_info = root->fs_info;
7226 struct extent_map *em;
7227 struct btrfs_key ins;
7231 alloc_hint = get_extent_allocation_hint(inode, start, len);
7233 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7234 0, alloc_hint, &ins, 1, 1);
7235 if (ret == -EAGAIN) {
7236 ASSERT(btrfs_is_zoned(fs_info));
7237 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
7238 TASK_UNINTERRUPTIBLE);
7242 return ERR_PTR(ret);
7244 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7245 ins.objectid, ins.offset, ins.offset,
7246 ins.offset, BTRFS_ORDERED_REGULAR);
7247 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7249 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7255 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7257 struct btrfs_block_group *block_group;
7258 bool readonly = false;
7260 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7261 if (!block_group || block_group->ro)
7264 btrfs_put_block_group(block_group);
7269 * Check if we can do nocow write into the range [@offset, @offset + @len)
7271 * @offset: File offset
7272 * @len: The length to write, will be updated to the nocow writeable
7274 * @orig_start: (optional) Return the original file offset of the file extent
7275 * @orig_len: (optional) Return the original on-disk length of the file extent
7276 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7277 * @strict: if true, omit optimizations that might force us into unnecessary
7278 * cow. e.g., don't trust generation number.
7281 * >0 and update @len if we can do nocow write
7282 * 0 if we can't do nocow write
7283 * <0 if error happened
7285 * NOTE: This only checks the file extents, caller is responsible to wait for
7286 * any ordered extents.
7288 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7289 u64 *orig_start, u64 *orig_block_len,
7290 u64 *ram_bytes, bool nowait, bool strict)
7292 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7293 struct can_nocow_file_extent_args nocow_args = { 0 };
7294 struct btrfs_path *path;
7296 struct extent_buffer *leaf;
7297 struct btrfs_root *root = BTRFS_I(inode)->root;
7298 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7299 struct btrfs_file_extent_item *fi;
7300 struct btrfs_key key;
7303 path = btrfs_alloc_path();
7306 path->nowait = nowait;
7308 ret = btrfs_lookup_file_extent(NULL, root, path,
7309 btrfs_ino(BTRFS_I(inode)), offset, 0);
7314 if (path->slots[0] == 0) {
7315 /* can't find the item, must cow */
7322 leaf = path->nodes[0];
7323 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7324 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7325 key.type != BTRFS_EXTENT_DATA_KEY) {
7326 /* not our file or wrong item type, must cow */
7330 if (key.offset > offset) {
7331 /* Wrong offset, must cow */
7335 if (btrfs_file_extent_end(path) <= offset)
7338 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7339 found_type = btrfs_file_extent_type(leaf, fi);
7341 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7343 nocow_args.start = offset;
7344 nocow_args.end = offset + *len - 1;
7345 nocow_args.strict = strict;
7346 nocow_args.free_path = true;
7348 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7349 /* can_nocow_file_extent() has freed the path. */
7353 /* Treat errors as not being able to NOCOW. */
7359 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7362 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7363 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7366 range_end = round_up(offset + nocow_args.num_bytes,
7367 root->fs_info->sectorsize) - 1;
7368 ret = test_range_bit(io_tree, offset, range_end,
7369 EXTENT_DELALLOC, 0, NULL);
7377 *orig_start = key.offset - nocow_args.extent_offset;
7379 *orig_block_len = nocow_args.disk_num_bytes;
7381 *len = nocow_args.num_bytes;
7384 btrfs_free_path(path);
7388 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7389 struct extent_state **cached_state,
7390 unsigned int iomap_flags)
7392 const bool writing = (iomap_flags & IOMAP_WRITE);
7393 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7394 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7395 struct btrfs_ordered_extent *ordered;
7400 if (!try_lock_extent(io_tree, lockstart, lockend))
7403 lock_extent(io_tree, lockstart, lockend, cached_state);
7406 * We're concerned with the entire range that we're going to be
7407 * doing DIO to, so we need to make sure there's no ordered
7408 * extents in this range.
7410 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7411 lockend - lockstart + 1);
7414 * We need to make sure there are no buffered pages in this
7415 * range either, we could have raced between the invalidate in
7416 * generic_file_direct_write and locking the extent. The
7417 * invalidate needs to happen so that reads after a write do not
7421 (!writing || !filemap_range_has_page(inode->i_mapping,
7422 lockstart, lockend)))
7425 unlock_extent(io_tree, lockstart, lockend, cached_state);
7429 btrfs_put_ordered_extent(ordered);
7434 * If we are doing a DIO read and the ordered extent we
7435 * found is for a buffered write, we can not wait for it
7436 * to complete and retry, because if we do so we can
7437 * deadlock with concurrent buffered writes on page
7438 * locks. This happens only if our DIO read covers more
7439 * than one extent map, if at this point has already
7440 * created an ordered extent for a previous extent map
7441 * and locked its range in the inode's io tree, and a
7442 * concurrent write against that previous extent map's
7443 * range and this range started (we unlock the ranges
7444 * in the io tree only when the bios complete and
7445 * buffered writes always lock pages before attempting
7446 * to lock range in the io tree).
7449 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7450 btrfs_start_ordered_extent(ordered, 1);
7452 ret = nowait ? -EAGAIN : -ENOTBLK;
7453 btrfs_put_ordered_extent(ordered);
7456 * We could trigger writeback for this range (and wait
7457 * for it to complete) and then invalidate the pages for
7458 * this range (through invalidate_inode_pages2_range()),
7459 * but that can lead us to a deadlock with a concurrent
7460 * call to readahead (a buffered read or a defrag call
7461 * triggered a readahead) on a page lock due to an
7462 * ordered dio extent we created before but did not have
7463 * yet a corresponding bio submitted (whence it can not
7464 * complete), which makes readahead wait for that
7465 * ordered extent to complete while holding a lock on
7468 ret = nowait ? -EAGAIN : -ENOTBLK;
7480 /* The callers of this must take lock_extent() */
7481 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7482 u64 len, u64 orig_start, u64 block_start,
7483 u64 block_len, u64 orig_block_len,
7484 u64 ram_bytes, int compress_type,
7487 struct extent_map *em;
7490 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7491 type == BTRFS_ORDERED_COMPRESSED ||
7492 type == BTRFS_ORDERED_NOCOW ||
7493 type == BTRFS_ORDERED_REGULAR);
7495 em = alloc_extent_map();
7497 return ERR_PTR(-ENOMEM);
7500 em->orig_start = orig_start;
7502 em->block_len = block_len;
7503 em->block_start = block_start;
7504 em->orig_block_len = orig_block_len;
7505 em->ram_bytes = ram_bytes;
7506 em->generation = -1;
7507 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7508 if (type == BTRFS_ORDERED_PREALLOC) {
7509 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7510 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7511 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7512 em->compress_type = compress_type;
7515 ret = btrfs_replace_extent_map_range(inode, em, true);
7517 free_extent_map(em);
7518 return ERR_PTR(ret);
7521 /* em got 2 refs now, callers needs to do free_extent_map once. */
7526 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7527 struct inode *inode,
7528 struct btrfs_dio_data *dio_data,
7529 u64 start, u64 *lenp,
7530 unsigned int iomap_flags)
7532 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7533 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7534 struct extent_map *em = *map;
7536 u64 block_start, orig_start, orig_block_len, ram_bytes;
7537 struct btrfs_block_group *bg;
7538 bool can_nocow = false;
7539 bool space_reserved = false;
7545 * We don't allocate a new extent in the following cases
7547 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7549 * 2) The extent is marked as PREALLOC. We're good to go here and can
7550 * just use the extent.
7553 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7554 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7555 em->block_start != EXTENT_MAP_HOLE)) {
7556 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7557 type = BTRFS_ORDERED_PREALLOC;
7559 type = BTRFS_ORDERED_NOCOW;
7560 len = min(len, em->len - (start - em->start));
7561 block_start = em->block_start + (start - em->start);
7563 if (can_nocow_extent(inode, start, &len, &orig_start,
7564 &orig_block_len, &ram_bytes, false, false) == 1) {
7565 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7573 struct extent_map *em2;
7575 /* We can NOCOW, so only need to reserve metadata space. */
7576 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7579 /* Our caller expects us to free the input extent map. */
7580 free_extent_map(em);
7582 btrfs_dec_nocow_writers(bg);
7583 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7587 space_reserved = true;
7589 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7590 orig_start, block_start,
7591 len, orig_block_len,
7593 btrfs_dec_nocow_writers(bg);
7594 if (type == BTRFS_ORDERED_PREALLOC) {
7595 free_extent_map(em);
7605 dio_data->nocow_done = true;
7607 /* Our caller expects us to free the input extent map. */
7608 free_extent_map(em);
7617 * If we could not allocate data space before locking the file
7618 * range and we can't do a NOCOW write, then we have to fail.
7620 if (!dio_data->data_space_reserved) {
7626 * We have to COW and we have already reserved data space before,
7627 * so now we reserve only metadata.
7629 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7633 space_reserved = true;
7635 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7641 len = min(len, em->len - (start - em->start));
7643 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7644 prev_len - len, true);
7648 * We have created our ordered extent, so we can now release our reservation
7649 * for an outstanding extent.
7651 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7654 * Need to update the i_size under the extent lock so buffered
7655 * readers will get the updated i_size when we unlock.
7657 if (start + len > i_size_read(inode))
7658 i_size_write(inode, start + len);
7660 if (ret && space_reserved) {
7661 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7662 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7668 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7669 loff_t length, unsigned int flags, struct iomap *iomap,
7670 struct iomap *srcmap)
7672 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7673 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7674 struct extent_map *em;
7675 struct extent_state *cached_state = NULL;
7676 struct btrfs_dio_data *dio_data = iter->private;
7677 u64 lockstart, lockend;
7678 const bool write = !!(flags & IOMAP_WRITE);
7681 const u64 data_alloc_len = length;
7682 bool unlock_extents = false;
7685 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7686 * we're NOWAIT we may submit a bio for a partial range and return
7687 * EIOCBQUEUED, which would result in an errant short read.
7689 * The best way to handle this would be to allow for partial completions
7690 * of iocb's, so we could submit the partial bio, return and fault in
7691 * the rest of the pages, and then submit the io for the rest of the
7692 * range. However we don't have that currently, so simply return
7693 * -EAGAIN at this point so that the normal path is used.
7695 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7699 * Cap the size of reads to that usually seen in buffered I/O as we need
7700 * to allocate a contiguous array for the checksums.
7703 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7706 lockend = start + len - 1;
7709 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7710 * enough if we've written compressed pages to this area, so we need to
7711 * flush the dirty pages again to make absolutely sure that any
7712 * outstanding dirty pages are on disk - the first flush only starts
7713 * compression on the data, while keeping the pages locked, so by the
7714 * time the second flush returns we know bios for the compressed pages
7715 * were submitted and finished, and the pages no longer under writeback.
7717 * If we have a NOWAIT request and we have any pages in the range that
7718 * are locked, likely due to compression still in progress, we don't want
7719 * to block on page locks. We also don't want to block on pages marked as
7720 * dirty or under writeback (same as for the non-compression case).
7721 * iomap_dio_rw() did the same check, but after that and before we got
7722 * here, mmap'ed writes may have happened or buffered reads started
7723 * (readpage() and readahead(), which lock pages), as we haven't locked
7724 * the file range yet.
7726 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7727 &BTRFS_I(inode)->runtime_flags)) {
7728 if (flags & IOMAP_NOWAIT) {
7729 if (filemap_range_needs_writeback(inode->i_mapping,
7730 lockstart, lockend))
7733 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7734 start + length - 1);
7740 memset(dio_data, 0, sizeof(*dio_data));
7743 * We always try to allocate data space and must do it before locking
7744 * the file range, to avoid deadlocks with concurrent writes to the same
7745 * range if the range has several extents and the writes don't expand the
7746 * current i_size (the inode lock is taken in shared mode). If we fail to
7747 * allocate data space here we continue and later, after locking the
7748 * file range, we fail with ENOSPC only if we figure out we can not do a
7751 if (write && !(flags & IOMAP_NOWAIT)) {
7752 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7753 &dio_data->data_reserved,
7754 start, data_alloc_len, false);
7756 dio_data->data_space_reserved = true;
7757 else if (ret && !(BTRFS_I(inode)->flags &
7758 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7763 * If this errors out it's because we couldn't invalidate pagecache for
7764 * this range and we need to fallback to buffered IO, or we are doing a
7765 * NOWAIT read/write and we need to block.
7767 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7771 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7778 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7779 * io. INLINE is special, and we could probably kludge it in here, but
7780 * it's still buffered so for safety lets just fall back to the generic
7783 * For COMPRESSED we _have_ to read the entire extent in so we can
7784 * decompress it, so there will be buffering required no matter what we
7785 * do, so go ahead and fallback to buffered.
7787 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7788 * to buffered IO. Don't blame me, this is the price we pay for using
7791 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7792 em->block_start == EXTENT_MAP_INLINE) {
7793 free_extent_map(em);
7795 * If we are in a NOWAIT context, return -EAGAIN in order to
7796 * fallback to buffered IO. This is not only because we can
7797 * block with buffered IO (no support for NOWAIT semantics at
7798 * the moment) but also to avoid returning short reads to user
7799 * space - this happens if we were able to read some data from
7800 * previous non-compressed extents and then when we fallback to
7801 * buffered IO, at btrfs_file_read_iter() by calling
7802 * filemap_read(), we fail to fault in pages for the read buffer,
7803 * in which case filemap_read() returns a short read (the number
7804 * of bytes previously read is > 0, so it does not return -EFAULT).
7806 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7810 len = min(len, em->len - (start - em->start));
7813 * If we have a NOWAIT request and the range contains multiple extents
7814 * (or a mix of extents and holes), then we return -EAGAIN to make the
7815 * caller fallback to a context where it can do a blocking (without
7816 * NOWAIT) request. This way we avoid doing partial IO and returning
7817 * success to the caller, which is not optimal for writes and for reads
7818 * it can result in unexpected behaviour for an application.
7820 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7821 * iomap_dio_rw(), we can end up returning less data then what the caller
7822 * asked for, resulting in an unexpected, and incorrect, short read.
7823 * That is, the caller asked to read N bytes and we return less than that,
7824 * which is wrong unless we are crossing EOF. This happens if we get a
7825 * page fault error when trying to fault in pages for the buffer that is
7826 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7827 * have previously submitted bios for other extents in the range, in
7828 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7829 * those bios have completed by the time we get the page fault error,
7830 * which we return back to our caller - we should only return EIOCBQUEUED
7831 * after we have submitted bios for all the extents in the range.
7833 if ((flags & IOMAP_NOWAIT) && len < length) {
7834 free_extent_map(em);
7840 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7841 start, &len, flags);
7844 unlock_extents = true;
7845 /* Recalc len in case the new em is smaller than requested */
7846 len = min(len, em->len - (start - em->start));
7847 if (dio_data->data_space_reserved) {
7849 u64 release_len = 0;
7851 if (dio_data->nocow_done) {
7852 release_offset = start;
7853 release_len = data_alloc_len;
7854 } else if (len < data_alloc_len) {
7855 release_offset = start + len;
7856 release_len = data_alloc_len - len;
7859 if (release_len > 0)
7860 btrfs_free_reserved_data_space(BTRFS_I(inode),
7861 dio_data->data_reserved,
7867 * We need to unlock only the end area that we aren't using.
7868 * The rest is going to be unlocked by the endio routine.
7870 lockstart = start + len;
7871 if (lockstart < lockend)
7872 unlock_extents = true;
7876 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7879 free_extent_state(cached_state);
7882 * Translate extent map information to iomap.
7883 * We trim the extents (and move the addr) even though iomap code does
7884 * that, since we have locked only the parts we are performing I/O in.
7886 if ((em->block_start == EXTENT_MAP_HOLE) ||
7887 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7888 iomap->addr = IOMAP_NULL_ADDR;
7889 iomap->type = IOMAP_HOLE;
7891 iomap->addr = em->block_start + (start - em->start);
7892 iomap->type = IOMAP_MAPPED;
7894 iomap->offset = start;
7895 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7896 iomap->length = len;
7898 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7899 iomap->flags |= IOMAP_F_ZONE_APPEND;
7901 free_extent_map(em);
7906 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7909 if (dio_data->data_space_reserved) {
7910 btrfs_free_reserved_data_space(BTRFS_I(inode),
7911 dio_data->data_reserved,
7912 start, data_alloc_len);
7913 extent_changeset_free(dio_data->data_reserved);
7919 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7920 ssize_t written, unsigned int flags, struct iomap *iomap)
7922 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7923 struct btrfs_dio_data *dio_data = iter->private;
7924 size_t submitted = dio_data->submitted;
7925 const bool write = !!(flags & IOMAP_WRITE);
7928 if (!write && (iomap->type == IOMAP_HOLE)) {
7929 /* If reading from a hole, unlock and return */
7930 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7935 if (submitted < length) {
7937 length -= submitted;
7939 btrfs_mark_ordered_io_finished(BTRFS_I(inode), NULL,
7940 pos, length, false);
7942 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7943 pos + length - 1, NULL);
7948 extent_changeset_free(dio_data->data_reserved);
7952 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7955 * This implies a barrier so that stores to dio_bio->bi_status before
7956 * this and loads of dio_bio->bi_status after this are fully ordered.
7958 if (!refcount_dec_and_test(&dip->refs))
7961 if (btrfs_op(&dip->bio) == BTRFS_MAP_WRITE) {
7962 btrfs_mark_ordered_io_finished(BTRFS_I(dip->inode), NULL,
7963 dip->file_offset, dip->bytes,
7964 !dip->bio.bi_status);
7966 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7968 dip->file_offset + dip->bytes - 1, NULL);
7972 bio_endio(&dip->bio);
7975 static void submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7977 enum btrfs_compression_type compress_type)
7979 struct btrfs_dio_private *dip = btrfs_bio(bio)->private;
7980 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7982 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7984 refcount_inc(&dip->refs);
7985 btrfs_submit_bio(fs_info, bio, mirror_num);
7988 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7989 struct btrfs_bio *bbio,
7990 const bool uptodate)
7992 struct inode *inode = dip->inode;
7993 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7994 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7995 blk_status_t err = BLK_STS_OK;
7996 struct bvec_iter iter;
8000 btrfs_bio_for_each_sector(fs_info, bv, bbio, iter, offset) {
8001 u64 start = bbio->file_offset + offset;
8004 (!csum || !btrfs_check_data_csum(inode, bbio, offset, bv.bv_page,
8006 btrfs_clean_io_failure(BTRFS_I(inode), start,
8007 bv.bv_page, bv.bv_offset);
8011 ret = btrfs_repair_one_sector(inode, bbio, offset,
8012 bv.bv_page, bv.bv_offset,
8013 submit_dio_repair_bio);
8015 err = errno_to_blk_status(ret);
8022 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8024 u64 dio_file_offset)
8026 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
8029 static void btrfs_end_dio_bio(struct btrfs_bio *bbio)
8031 struct btrfs_dio_private *dip = bbio->private;
8032 struct bio *bio = &bbio->bio;
8033 blk_status_t err = bio->bi_status;
8036 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8037 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8038 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8039 bio->bi_opf, bio->bi_iter.bi_sector,
8040 bio->bi_iter.bi_size, err);
8042 if (bio_op(bio) == REQ_OP_READ)
8043 err = btrfs_check_read_dio_bio(dip, bbio, !err);
8046 dip->bio.bi_status = err;
8048 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
8051 btrfs_dio_private_put(dip);
8054 static void btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8055 u64 file_offset, int async_submit)
8057 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8058 struct btrfs_dio_private *dip = btrfs_bio(bio)->private;
8061 /* Save the original iter for read repair */
8062 if (btrfs_op(bio) == BTRFS_MAP_READ)
8063 btrfs_bio(bio)->iter = bio->bi_iter;
8065 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8068 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
8069 /* Check btrfs_submit_data_write_bio() for async submit rules */
8070 if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers) &&
8071 btrfs_wq_submit_bio(inode, bio, 0, file_offset,
8072 btrfs_submit_bio_start_direct_io))
8076 * If we aren't doing async submit, calculate the csum of the
8079 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
8081 btrfs_bio_end_io(btrfs_bio(bio), ret);
8085 btrfs_bio(bio)->csum = btrfs_csum_ptr(fs_info, dip->csums,
8086 file_offset - dip->file_offset);
8089 btrfs_submit_bio(fs_info, bio, 0);
8092 static void btrfs_submit_direct(const struct iomap_iter *iter,
8093 struct bio *dio_bio, loff_t file_offset)
8095 struct btrfs_dio_private *dip =
8096 container_of(dio_bio, struct btrfs_dio_private, bio);
8097 struct inode *inode = iter->inode;
8098 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8099 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8100 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8101 BTRFS_BLOCK_GROUP_RAID56_MASK);
8104 int async_submit = 0;
8106 u64 clone_offset = 0;
8110 blk_status_t status;
8111 struct btrfs_io_geometry geom;
8112 struct btrfs_dio_data *dio_data = iter->private;
8113 struct extent_map *em = NULL;
8116 dip->file_offset = file_offset;
8117 dip->bytes = dio_bio->bi_iter.bi_size;
8118 refcount_set(&dip->refs, 1);
8121 if (!write && !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
8122 unsigned int nr_sectors =
8123 (dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits);
8126 * Load the csums up front to reduce csum tree searches and
8127 * contention when submitting bios.
8129 status = BLK_STS_RESOURCE;
8130 dip->csums = kcalloc(nr_sectors, fs_info->csum_size, GFP_NOFS);
8134 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8135 if (status != BLK_STS_OK)
8139 start_sector = dio_bio->bi_iter.bi_sector;
8140 submit_len = dio_bio->bi_iter.bi_size;
8143 logical = start_sector << 9;
8144 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8146 status = errno_to_blk_status(PTR_ERR(em));
8150 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8153 status = errno_to_blk_status(ret);
8157 clone_len = min(submit_len, geom.len);
8158 ASSERT(clone_len <= UINT_MAX);
8161 * This will never fail as it's passing GPF_NOFS and
8162 * the allocation is backed by btrfs_bioset.
8164 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len,
8165 btrfs_end_dio_bio, dip);
8166 btrfs_bio(bio)->file_offset = file_offset;
8168 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8169 status = extract_ordered_extent(BTRFS_I(inode), bio,
8177 ASSERT(submit_len >= clone_len);
8178 submit_len -= clone_len;
8181 * Increase the count before we submit the bio so we know
8182 * the end IO handler won't happen before we increase the
8183 * count. Otherwise, the dip might get freed before we're
8184 * done setting it up.
8186 * We transfer the initial reference to the last bio, so we
8187 * don't need to increment the reference count for the last one.
8189 if (submit_len > 0) {
8190 refcount_inc(&dip->refs);
8192 * If we are submitting more than one bio, submit them
8193 * all asynchronously. The exception is RAID 5 or 6, as
8194 * asynchronous checksums make it difficult to collect
8195 * full stripe writes.
8201 btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8203 dio_data->submitted += clone_len;
8204 clone_offset += clone_len;
8205 start_sector += clone_len >> 9;
8206 file_offset += clone_len;
8208 free_extent_map(em);
8209 } while (submit_len > 0);
8213 free_extent_map(em);
8215 dio_bio->bi_status = status;
8216 btrfs_dio_private_put(dip);
8219 static const struct iomap_ops btrfs_dio_iomap_ops = {
8220 .iomap_begin = btrfs_dio_iomap_begin,
8221 .iomap_end = btrfs_dio_iomap_end,
8224 static const struct iomap_dio_ops btrfs_dio_ops = {
8225 .submit_io = btrfs_submit_direct,
8226 .bio_set = &btrfs_dio_bioset,
8229 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
8231 struct btrfs_dio_data data;
8233 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
8234 IOMAP_DIO_PARTIAL, &data, done_before);
8237 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
8240 struct btrfs_dio_data data;
8242 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
8243 IOMAP_DIO_PARTIAL, &data, done_before);
8246 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8251 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8256 * fiemap_prep() called filemap_write_and_wait() for the whole possible
8257 * file range (0 to LLONG_MAX), but that is not enough if we have
8258 * compression enabled. The first filemap_fdatawrite_range() only kicks
8259 * in the compression of data (in an async thread) and will return
8260 * before the compression is done and writeback is started. A second
8261 * filemap_fdatawrite_range() is needed to wait for the compression to
8262 * complete and writeback to start. We also need to wait for ordered
8263 * extents to complete, because our fiemap implementation uses mainly
8264 * file extent items to list the extents, searching for extent maps
8265 * only for file ranges with holes or prealloc extents to figure out
8266 * if we have delalloc in those ranges.
8268 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
8269 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
8274 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8277 static int btrfs_writepages(struct address_space *mapping,
8278 struct writeback_control *wbc)
8280 return extent_writepages(mapping, wbc);
8283 static void btrfs_readahead(struct readahead_control *rac)
8285 extent_readahead(rac);
8289 * For release_folio() and invalidate_folio() we have a race window where
8290 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8291 * If we continue to release/invalidate the page, we could cause use-after-free
8292 * for subpage spinlock. So this function is to spin and wait for subpage
8295 static void wait_subpage_spinlock(struct page *page)
8297 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8298 struct btrfs_subpage *subpage;
8300 if (!btrfs_is_subpage(fs_info, page))
8303 ASSERT(PagePrivate(page) && page->private);
8304 subpage = (struct btrfs_subpage *)page->private;
8307 * This may look insane as we just acquire the spinlock and release it,
8308 * without doing anything. But we just want to make sure no one is
8309 * still holding the subpage spinlock.
8310 * And since the page is not dirty nor writeback, and we have page
8311 * locked, the only possible way to hold a spinlock is from the endio
8312 * function to clear page writeback.
8314 * Here we just acquire the spinlock so that all existing callers
8315 * should exit and we're safe to release/invalidate the page.
8317 spin_lock_irq(&subpage->lock);
8318 spin_unlock_irq(&subpage->lock);
8321 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8323 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8326 wait_subpage_spinlock(&folio->page);
8327 clear_page_extent_mapped(&folio->page);
8332 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8334 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8336 return __btrfs_release_folio(folio, gfp_flags);
8339 #ifdef CONFIG_MIGRATION
8340 static int btrfs_migrate_folio(struct address_space *mapping,
8341 struct folio *dst, struct folio *src,
8342 enum migrate_mode mode)
8344 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8346 if (ret != MIGRATEPAGE_SUCCESS)
8349 if (folio_test_ordered(src)) {
8350 folio_clear_ordered(src);
8351 folio_set_ordered(dst);
8354 return MIGRATEPAGE_SUCCESS;
8357 #define btrfs_migrate_folio NULL
8360 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8363 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8364 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8365 struct extent_io_tree *tree = &inode->io_tree;
8366 struct extent_state *cached_state = NULL;
8367 u64 page_start = folio_pos(folio);
8368 u64 page_end = page_start + folio_size(folio) - 1;
8370 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8373 * We have folio locked so no new ordered extent can be created on this
8374 * page, nor bio can be submitted for this folio.
8376 * But already submitted bio can still be finished on this folio.
8377 * Furthermore, endio function won't skip folio which has Ordered
8378 * (Private2) already cleared, so it's possible for endio and
8379 * invalidate_folio to do the same ordered extent accounting twice
8382 * So here we wait for any submitted bios to finish, so that we won't
8383 * do double ordered extent accounting on the same folio.
8385 folio_wait_writeback(folio);
8386 wait_subpage_spinlock(&folio->page);
8389 * For subpage case, we have call sites like
8390 * btrfs_punch_hole_lock_range() which passes range not aligned to
8392 * If the range doesn't cover the full folio, we don't need to and
8393 * shouldn't clear page extent mapped, as folio->private can still
8394 * record subpage dirty bits for other part of the range.
8396 * For cases that invalidate the full folio even the range doesn't
8397 * cover the full folio, like invalidating the last folio, we're
8398 * still safe to wait for ordered extent to finish.
8400 if (!(offset == 0 && length == folio_size(folio))) {
8401 btrfs_release_folio(folio, GFP_NOFS);
8405 if (!inode_evicting)
8406 lock_extent(tree, page_start, page_end, &cached_state);
8409 while (cur < page_end) {
8410 struct btrfs_ordered_extent *ordered;
8413 u32 extra_flags = 0;
8415 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8416 page_end + 1 - cur);
8418 range_end = page_end;
8420 * No ordered extent covering this range, we are safe
8421 * to delete all extent states in the range.
8423 extra_flags = EXTENT_CLEAR_ALL_BITS;
8426 if (ordered->file_offset > cur) {
8428 * There is a range between [cur, oe->file_offset) not
8429 * covered by any ordered extent.
8430 * We are safe to delete all extent states, and handle
8431 * the ordered extent in the next iteration.
8433 range_end = ordered->file_offset - 1;
8434 extra_flags = EXTENT_CLEAR_ALL_BITS;
8438 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8440 ASSERT(range_end + 1 - cur < U32_MAX);
8441 range_len = range_end + 1 - cur;
8442 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8444 * If Ordered (Private2) is cleared, it means endio has
8445 * already been executed for the range.
8446 * We can't delete the extent states as
8447 * btrfs_finish_ordered_io() may still use some of them.
8451 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8454 * IO on this page will never be started, so we need to account
8455 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8456 * here, must leave that up for the ordered extent completion.
8458 * This will also unlock the range for incoming
8459 * btrfs_finish_ordered_io().
8461 if (!inode_evicting)
8462 clear_extent_bit(tree, cur, range_end,
8464 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8465 EXTENT_DEFRAG, &cached_state);
8467 spin_lock_irq(&inode->ordered_tree.lock);
8468 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8469 ordered->truncated_len = min(ordered->truncated_len,
8470 cur - ordered->file_offset);
8471 spin_unlock_irq(&inode->ordered_tree.lock);
8474 * If the ordered extent has finished, we're safe to delete all
8475 * the extent states of the range, otherwise
8476 * btrfs_finish_ordered_io() will get executed by endio for
8477 * other pages, so we can't delete extent states.
8479 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8480 cur, range_end + 1 - cur)) {
8481 btrfs_finish_ordered_io(ordered);
8483 * The ordered extent has finished, now we're again
8484 * safe to delete all extent states of the range.
8486 extra_flags = EXTENT_CLEAR_ALL_BITS;
8490 btrfs_put_ordered_extent(ordered);
8492 * Qgroup reserved space handler
8493 * Sector(s) here will be either:
8495 * 1) Already written to disk or bio already finished
8496 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8497 * Qgroup will be handled by its qgroup_record then.
8498 * btrfs_qgroup_free_data() call will do nothing here.
8500 * 2) Not written to disk yet
8501 * Then btrfs_qgroup_free_data() call will clear the
8502 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8503 * reserved data space.
8504 * Since the IO will never happen for this page.
8506 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8507 if (!inode_evicting) {
8508 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8509 EXTENT_DELALLOC | EXTENT_UPTODATE |
8510 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8511 extra_flags, &cached_state);
8513 cur = range_end + 1;
8516 * We have iterated through all ordered extents of the page, the page
8517 * should not have Ordered (Private2) anymore, or the above iteration
8518 * did something wrong.
8520 ASSERT(!folio_test_ordered(folio));
8521 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8522 if (!inode_evicting)
8523 __btrfs_release_folio(folio, GFP_NOFS);
8524 clear_page_extent_mapped(&folio->page);
8528 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8529 * called from a page fault handler when a page is first dirtied. Hence we must
8530 * be careful to check for EOF conditions here. We set the page up correctly
8531 * for a written page which means we get ENOSPC checking when writing into
8532 * holes and correct delalloc and unwritten extent mapping on filesystems that
8533 * support these features.
8535 * We are not allowed to take the i_mutex here so we have to play games to
8536 * protect against truncate races as the page could now be beyond EOF. Because
8537 * truncate_setsize() writes the inode size before removing pages, once we have
8538 * the page lock we can determine safely if the page is beyond EOF. If it is not
8539 * beyond EOF, then the page is guaranteed safe against truncation until we
8542 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8544 struct page *page = vmf->page;
8545 struct inode *inode = file_inode(vmf->vma->vm_file);
8546 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8547 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8548 struct btrfs_ordered_extent *ordered;
8549 struct extent_state *cached_state = NULL;
8550 struct extent_changeset *data_reserved = NULL;
8551 unsigned long zero_start;
8561 reserved_space = PAGE_SIZE;
8563 sb_start_pagefault(inode->i_sb);
8564 page_start = page_offset(page);
8565 page_end = page_start + PAGE_SIZE - 1;
8569 * Reserving delalloc space after obtaining the page lock can lead to
8570 * deadlock. For example, if a dirty page is locked by this function
8571 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8572 * dirty page write out, then the btrfs_writepages() function could
8573 * end up waiting indefinitely to get a lock on the page currently
8574 * being processed by btrfs_page_mkwrite() function.
8576 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8577 page_start, reserved_space);
8579 ret2 = file_update_time(vmf->vma->vm_file);
8583 ret = vmf_error(ret2);
8589 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8591 down_read(&BTRFS_I(inode)->i_mmap_lock);
8593 size = i_size_read(inode);
8595 if ((page->mapping != inode->i_mapping) ||
8596 (page_start >= size)) {
8597 /* page got truncated out from underneath us */
8600 wait_on_page_writeback(page);
8602 lock_extent(io_tree, page_start, page_end, &cached_state);
8603 ret2 = set_page_extent_mapped(page);
8605 ret = vmf_error(ret2);
8606 unlock_extent(io_tree, page_start, page_end, &cached_state);
8611 * we can't set the delalloc bits if there are pending ordered
8612 * extents. Drop our locks and wait for them to finish
8614 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8617 unlock_extent(io_tree, page_start, page_end, &cached_state);
8619 up_read(&BTRFS_I(inode)->i_mmap_lock);
8620 btrfs_start_ordered_extent(ordered, 1);
8621 btrfs_put_ordered_extent(ordered);
8625 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8626 reserved_space = round_up(size - page_start,
8627 fs_info->sectorsize);
8628 if (reserved_space < PAGE_SIZE) {
8629 end = page_start + reserved_space - 1;
8630 btrfs_delalloc_release_space(BTRFS_I(inode),
8631 data_reserved, page_start,
8632 PAGE_SIZE - reserved_space, true);
8637 * page_mkwrite gets called when the page is firstly dirtied after it's
8638 * faulted in, but write(2) could also dirty a page and set delalloc
8639 * bits, thus in this case for space account reason, we still need to
8640 * clear any delalloc bits within this page range since we have to
8641 * reserve data&meta space before lock_page() (see above comments).
8643 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8644 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8645 EXTENT_DEFRAG, &cached_state);
8647 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8650 unlock_extent(io_tree, page_start, page_end, &cached_state);
8651 ret = VM_FAULT_SIGBUS;
8655 /* page is wholly or partially inside EOF */
8656 if (page_start + PAGE_SIZE > size)
8657 zero_start = offset_in_page(size);
8659 zero_start = PAGE_SIZE;
8661 if (zero_start != PAGE_SIZE)
8662 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8664 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8665 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8666 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8668 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8670 unlock_extent(io_tree, page_start, page_end, &cached_state);
8671 up_read(&BTRFS_I(inode)->i_mmap_lock);
8673 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8674 sb_end_pagefault(inode->i_sb);
8675 extent_changeset_free(data_reserved);
8676 return VM_FAULT_LOCKED;
8680 up_read(&BTRFS_I(inode)->i_mmap_lock);
8682 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8683 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8684 reserved_space, (ret != 0));
8686 sb_end_pagefault(inode->i_sb);
8687 extent_changeset_free(data_reserved);
8691 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8693 struct btrfs_truncate_control control = {
8694 .inode = BTRFS_I(inode),
8695 .ino = btrfs_ino(BTRFS_I(inode)),
8696 .min_type = BTRFS_EXTENT_DATA_KEY,
8697 .clear_extent_range = true,
8699 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8700 struct btrfs_root *root = BTRFS_I(inode)->root;
8701 struct btrfs_block_rsv *rsv;
8703 struct btrfs_trans_handle *trans;
8704 u64 mask = fs_info->sectorsize - 1;
8705 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8707 if (!skip_writeback) {
8708 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8715 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8716 * things going on here:
8718 * 1) We need to reserve space to update our inode.
8720 * 2) We need to have something to cache all the space that is going to
8721 * be free'd up by the truncate operation, but also have some slack
8722 * space reserved in case it uses space during the truncate (thank you
8723 * very much snapshotting).
8725 * And we need these to be separate. The fact is we can use a lot of
8726 * space doing the truncate, and we have no earthly idea how much space
8727 * we will use, so we need the truncate reservation to be separate so it
8728 * doesn't end up using space reserved for updating the inode. We also
8729 * need to be able to stop the transaction and start a new one, which
8730 * means we need to be able to update the inode several times, and we
8731 * have no idea of knowing how many times that will be, so we can't just
8732 * reserve 1 item for the entirety of the operation, so that has to be
8733 * done separately as well.
8735 * So that leaves us with
8737 * 1) rsv - for the truncate reservation, which we will steal from the
8738 * transaction reservation.
8739 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8740 * updating the inode.
8742 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8745 rsv->size = min_size;
8746 rsv->failfast = true;
8749 * 1 for the truncate slack space
8750 * 1 for updating the inode.
8752 trans = btrfs_start_transaction(root, 2);
8753 if (IS_ERR(trans)) {
8754 ret = PTR_ERR(trans);
8758 /* Migrate the slack space for the truncate to our reserve */
8759 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8763 trans->block_rsv = rsv;
8766 struct extent_state *cached_state = NULL;
8767 const u64 new_size = inode->i_size;
8768 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8770 control.new_size = new_size;
8771 lock_extent(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8774 * We want to drop from the next block forward in case this new
8775 * size is not block aligned since we will be keeping the last
8776 * block of the extent just the way it is.
8778 btrfs_drop_extent_map_range(BTRFS_I(inode),
8779 ALIGN(new_size, fs_info->sectorsize),
8782 ret = btrfs_truncate_inode_items(trans, root, &control);
8784 inode_sub_bytes(inode, control.sub_bytes);
8785 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8787 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8790 trans->block_rsv = &fs_info->trans_block_rsv;
8791 if (ret != -ENOSPC && ret != -EAGAIN)
8794 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8798 btrfs_end_transaction(trans);
8799 btrfs_btree_balance_dirty(fs_info);
8801 trans = btrfs_start_transaction(root, 2);
8802 if (IS_ERR(trans)) {
8803 ret = PTR_ERR(trans);
8808 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8809 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8810 rsv, min_size, false);
8811 BUG_ON(ret); /* shouldn't happen */
8812 trans->block_rsv = rsv;
8816 * We can't call btrfs_truncate_block inside a trans handle as we could
8817 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8818 * know we've truncated everything except the last little bit, and can
8819 * do btrfs_truncate_block and then update the disk_i_size.
8821 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8822 btrfs_end_transaction(trans);
8823 btrfs_btree_balance_dirty(fs_info);
8825 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8828 trans = btrfs_start_transaction(root, 1);
8829 if (IS_ERR(trans)) {
8830 ret = PTR_ERR(trans);
8833 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8839 trans->block_rsv = &fs_info->trans_block_rsv;
8840 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8844 ret2 = btrfs_end_transaction(trans);
8847 btrfs_btree_balance_dirty(fs_info);
8850 btrfs_free_block_rsv(fs_info, rsv);
8852 * So if we truncate and then write and fsync we normally would just
8853 * write the extents that changed, which is a problem if we need to
8854 * first truncate that entire inode. So set this flag so we write out
8855 * all of the extents in the inode to the sync log so we're completely
8858 * If no extents were dropped or trimmed we don't need to force the next
8859 * fsync to truncate all the inode's items from the log and re-log them
8860 * all. This means the truncate operation did not change the file size,
8861 * or changed it to a smaller size but there was only an implicit hole
8862 * between the old i_size and the new i_size, and there were no prealloc
8863 * extents beyond i_size to drop.
8865 if (control.extents_found > 0)
8866 btrfs_set_inode_full_sync(BTRFS_I(inode));
8871 struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns,
8874 struct inode *inode;
8876 inode = new_inode(dir->i_sb);
8879 * Subvolumes don't inherit the sgid bit or the parent's gid if
8880 * the parent's sgid bit is set. This is probably a bug.
8882 inode_init_owner(mnt_userns, inode, NULL,
8883 S_IFDIR | (~current_umask() & S_IRWXUGO));
8884 inode->i_op = &btrfs_dir_inode_operations;
8885 inode->i_fop = &btrfs_dir_file_operations;
8890 struct inode *btrfs_alloc_inode(struct super_block *sb)
8892 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8893 struct btrfs_inode *ei;
8894 struct inode *inode;
8896 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8903 ei->last_sub_trans = 0;
8904 ei->logged_trans = 0;
8905 ei->delalloc_bytes = 0;
8906 ei->new_delalloc_bytes = 0;
8907 ei->defrag_bytes = 0;
8908 ei->disk_i_size = 0;
8912 ei->index_cnt = (u64)-1;
8914 ei->last_unlink_trans = 0;
8915 ei->last_reflink_trans = 0;
8916 ei->last_log_commit = 0;
8918 spin_lock_init(&ei->lock);
8919 spin_lock_init(&ei->io_failure_lock);
8920 ei->outstanding_extents = 0;
8921 if (sb->s_magic != BTRFS_TEST_MAGIC)
8922 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8923 BTRFS_BLOCK_RSV_DELALLOC);
8924 ei->runtime_flags = 0;
8925 ei->prop_compress = BTRFS_COMPRESS_NONE;
8926 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8928 ei->delayed_node = NULL;
8930 ei->i_otime.tv_sec = 0;
8931 ei->i_otime.tv_nsec = 0;
8933 inode = &ei->vfs_inode;
8934 extent_map_tree_init(&ei->extent_tree);
8935 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8936 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8937 IO_TREE_INODE_FILE_EXTENT, NULL);
8938 ei->io_failure_tree = RB_ROOT;
8939 atomic_set(&ei->sync_writers, 0);
8940 mutex_init(&ei->log_mutex);
8941 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8942 INIT_LIST_HEAD(&ei->delalloc_inodes);
8943 INIT_LIST_HEAD(&ei->delayed_iput);
8944 RB_CLEAR_NODE(&ei->rb_node);
8945 init_rwsem(&ei->i_mmap_lock);
8950 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8951 void btrfs_test_destroy_inode(struct inode *inode)
8953 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8954 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8958 void btrfs_free_inode(struct inode *inode)
8960 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8963 void btrfs_destroy_inode(struct inode *vfs_inode)
8965 struct btrfs_ordered_extent *ordered;
8966 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8967 struct btrfs_root *root = inode->root;
8968 bool freespace_inode;
8970 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8971 WARN_ON(vfs_inode->i_data.nrpages);
8972 WARN_ON(inode->block_rsv.reserved);
8973 WARN_ON(inode->block_rsv.size);
8974 WARN_ON(inode->outstanding_extents);
8975 if (!S_ISDIR(vfs_inode->i_mode)) {
8976 WARN_ON(inode->delalloc_bytes);
8977 WARN_ON(inode->new_delalloc_bytes);
8979 WARN_ON(inode->csum_bytes);
8980 WARN_ON(inode->defrag_bytes);
8983 * This can happen where we create an inode, but somebody else also
8984 * created the same inode and we need to destroy the one we already
8991 * If this is a free space inode do not take the ordered extents lockdep
8994 freespace_inode = btrfs_is_free_space_inode(inode);
8997 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9001 btrfs_err(root->fs_info,
9002 "found ordered extent %llu %llu on inode cleanup",
9003 ordered->file_offset, ordered->num_bytes);
9005 if (!freespace_inode)
9006 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
9008 btrfs_remove_ordered_extent(inode, ordered);
9009 btrfs_put_ordered_extent(ordered);
9010 btrfs_put_ordered_extent(ordered);
9013 btrfs_qgroup_check_reserved_leak(inode);
9014 inode_tree_del(inode);
9015 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
9016 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
9017 btrfs_put_root(inode->root);
9020 int btrfs_drop_inode(struct inode *inode)
9022 struct btrfs_root *root = BTRFS_I(inode)->root;
9027 /* the snap/subvol tree is on deleting */
9028 if (btrfs_root_refs(&root->root_item) == 0)
9031 return generic_drop_inode(inode);
9034 static void init_once(void *foo)
9036 struct btrfs_inode *ei = foo;
9038 inode_init_once(&ei->vfs_inode);
9041 void __cold btrfs_destroy_cachep(void)
9044 * Make sure all delayed rcu free inodes are flushed before we
9048 bioset_exit(&btrfs_dio_bioset);
9049 kmem_cache_destroy(btrfs_inode_cachep);
9050 kmem_cache_destroy(btrfs_trans_handle_cachep);
9051 kmem_cache_destroy(btrfs_path_cachep);
9052 kmem_cache_destroy(btrfs_free_space_cachep);
9053 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9056 int __init btrfs_init_cachep(void)
9058 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9059 sizeof(struct btrfs_inode), 0,
9060 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9062 if (!btrfs_inode_cachep)
9065 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9066 sizeof(struct btrfs_trans_handle), 0,
9067 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9068 if (!btrfs_trans_handle_cachep)
9071 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9072 sizeof(struct btrfs_path), 0,
9073 SLAB_MEM_SPREAD, NULL);
9074 if (!btrfs_path_cachep)
9077 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9078 sizeof(struct btrfs_free_space), 0,
9079 SLAB_MEM_SPREAD, NULL);
9080 if (!btrfs_free_space_cachep)
9083 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9084 PAGE_SIZE, PAGE_SIZE,
9085 SLAB_MEM_SPREAD, NULL);
9086 if (!btrfs_free_space_bitmap_cachep)
9089 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
9090 offsetof(struct btrfs_dio_private, bio),
9096 btrfs_destroy_cachep();
9100 static int btrfs_getattr(struct user_namespace *mnt_userns,
9101 const struct path *path, struct kstat *stat,
9102 u32 request_mask, unsigned int flags)
9106 struct inode *inode = d_inode(path->dentry);
9107 u32 blocksize = inode->i_sb->s_blocksize;
9108 u32 bi_flags = BTRFS_I(inode)->flags;
9109 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9111 stat->result_mask |= STATX_BTIME;
9112 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9113 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9114 if (bi_flags & BTRFS_INODE_APPEND)
9115 stat->attributes |= STATX_ATTR_APPEND;
9116 if (bi_flags & BTRFS_INODE_COMPRESS)
9117 stat->attributes |= STATX_ATTR_COMPRESSED;
9118 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9119 stat->attributes |= STATX_ATTR_IMMUTABLE;
9120 if (bi_flags & BTRFS_INODE_NODUMP)
9121 stat->attributes |= STATX_ATTR_NODUMP;
9122 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9123 stat->attributes |= STATX_ATTR_VERITY;
9125 stat->attributes_mask |= (STATX_ATTR_APPEND |
9126 STATX_ATTR_COMPRESSED |
9127 STATX_ATTR_IMMUTABLE |
9130 generic_fillattr(mnt_userns, inode, stat);
9131 stat->dev = BTRFS_I(inode)->root->anon_dev;
9133 spin_lock(&BTRFS_I(inode)->lock);
9134 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9135 inode_bytes = inode_get_bytes(inode);
9136 spin_unlock(&BTRFS_I(inode)->lock);
9137 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9138 ALIGN(delalloc_bytes, blocksize)) >> 9;
9142 static int btrfs_rename_exchange(struct inode *old_dir,
9143 struct dentry *old_dentry,
9144 struct inode *new_dir,
9145 struct dentry *new_dentry)
9147 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9148 struct btrfs_trans_handle *trans;
9149 unsigned int trans_num_items;
9150 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9151 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9152 struct inode *new_inode = new_dentry->d_inode;
9153 struct inode *old_inode = old_dentry->d_inode;
9154 struct timespec64 ctime = current_time(old_inode);
9155 struct btrfs_rename_ctx old_rename_ctx;
9156 struct btrfs_rename_ctx new_rename_ctx;
9157 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9158 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9163 bool need_abort = false;
9164 struct fscrypt_name old_fname, new_fname;
9165 struct fscrypt_str *old_name, *new_name;
9168 * For non-subvolumes allow exchange only within one subvolume, in the
9169 * same inode namespace. Two subvolumes (represented as directory) can
9170 * be exchanged as they're a logical link and have a fixed inode number.
9173 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9174 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9177 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9181 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9183 fscrypt_free_filename(&old_fname);
9187 old_name = &old_fname.disk_name;
9188 new_name = &new_fname.disk_name;
9190 /* close the race window with snapshot create/destroy ioctl */
9191 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9192 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9193 down_read(&fs_info->subvol_sem);
9197 * 1 to remove old dir item
9198 * 1 to remove old dir index
9199 * 1 to add new dir item
9200 * 1 to add new dir index
9201 * 1 to update parent inode
9203 * If the parents are the same, we only need to account for one
9205 trans_num_items = (old_dir == new_dir ? 9 : 10);
9206 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9208 * 1 to remove old root ref
9209 * 1 to remove old root backref
9210 * 1 to add new root ref
9211 * 1 to add new root backref
9213 trans_num_items += 4;
9216 * 1 to update inode item
9217 * 1 to remove old inode ref
9218 * 1 to add new inode ref
9220 trans_num_items += 3;
9222 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9223 trans_num_items += 4;
9225 trans_num_items += 3;
9226 trans = btrfs_start_transaction(root, trans_num_items);
9227 if (IS_ERR(trans)) {
9228 ret = PTR_ERR(trans);
9233 ret = btrfs_record_root_in_trans(trans, dest);
9239 * We need to find a free sequence number both in the source and
9240 * in the destination directory for the exchange.
9242 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9245 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9249 BTRFS_I(old_inode)->dir_index = 0ULL;
9250 BTRFS_I(new_inode)->dir_index = 0ULL;
9252 /* Reference for the source. */
9253 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9254 /* force full log commit if subvolume involved. */
9255 btrfs_set_log_full_commit(trans);
9257 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
9258 btrfs_ino(BTRFS_I(new_dir)),
9265 /* And now for the dest. */
9266 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9267 /* force full log commit if subvolume involved. */
9268 btrfs_set_log_full_commit(trans);
9270 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
9271 btrfs_ino(BTRFS_I(old_dir)),
9275 btrfs_abort_transaction(trans, ret);
9280 /* Update inode version and ctime/mtime. */
9281 inode_inc_iversion(old_dir);
9282 inode_inc_iversion(new_dir);
9283 inode_inc_iversion(old_inode);
9284 inode_inc_iversion(new_inode);
9285 old_dir->i_mtime = ctime;
9286 old_dir->i_ctime = ctime;
9287 new_dir->i_mtime = ctime;
9288 new_dir->i_ctime = ctime;
9289 old_inode->i_ctime = ctime;
9290 new_inode->i_ctime = ctime;
9292 if (old_dentry->d_parent != new_dentry->d_parent) {
9293 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9294 BTRFS_I(old_inode), 1);
9295 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9296 BTRFS_I(new_inode), 1);
9299 /* src is a subvolume */
9300 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9301 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9302 } else { /* src is an inode */
9303 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9304 BTRFS_I(old_dentry->d_inode),
9305 old_name, &old_rename_ctx);
9307 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9310 btrfs_abort_transaction(trans, ret);
9314 /* dest is a subvolume */
9315 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9316 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9317 } else { /* dest is an inode */
9318 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9319 BTRFS_I(new_dentry->d_inode),
9320 new_name, &new_rename_ctx);
9322 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9325 btrfs_abort_transaction(trans, ret);
9329 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9330 new_name, 0, old_idx);
9332 btrfs_abort_transaction(trans, ret);
9336 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9337 old_name, 0, new_idx);
9339 btrfs_abort_transaction(trans, ret);
9343 if (old_inode->i_nlink == 1)
9344 BTRFS_I(old_inode)->dir_index = old_idx;
9345 if (new_inode->i_nlink == 1)
9346 BTRFS_I(new_inode)->dir_index = new_idx;
9349 * Now pin the logs of the roots. We do it to ensure that no other task
9350 * can sync the logs while we are in progress with the rename, because
9351 * that could result in an inconsistency in case any of the inodes that
9352 * are part of this rename operation were logged before.
9354 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9355 btrfs_pin_log_trans(root);
9356 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9357 btrfs_pin_log_trans(dest);
9359 /* Do the log updates for all inodes. */
9360 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9361 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9362 old_rename_ctx.index, new_dentry->d_parent);
9363 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9364 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9365 new_rename_ctx.index, old_dentry->d_parent);
9367 /* Now unpin the logs. */
9368 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9369 btrfs_end_log_trans(root);
9370 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9371 btrfs_end_log_trans(dest);
9373 ret2 = btrfs_end_transaction(trans);
9374 ret = ret ? ret : ret2;
9376 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9377 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9378 up_read(&fs_info->subvol_sem);
9380 fscrypt_free_filename(&new_fname);
9381 fscrypt_free_filename(&old_fname);
9385 static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns,
9388 struct inode *inode;
9390 inode = new_inode(dir->i_sb);
9392 inode_init_owner(mnt_userns, inode, dir,
9393 S_IFCHR | WHITEOUT_MODE);
9394 inode->i_op = &btrfs_special_inode_operations;
9395 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9400 static int btrfs_rename(struct user_namespace *mnt_userns,
9401 struct inode *old_dir, struct dentry *old_dentry,
9402 struct inode *new_dir, struct dentry *new_dentry,
9405 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9406 struct btrfs_new_inode_args whiteout_args = {
9408 .dentry = old_dentry,
9410 struct btrfs_trans_handle *trans;
9411 unsigned int trans_num_items;
9412 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9413 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9414 struct inode *new_inode = d_inode(new_dentry);
9415 struct inode *old_inode = d_inode(old_dentry);
9416 struct btrfs_rename_ctx rename_ctx;
9420 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9421 struct fscrypt_name old_fname, new_fname;
9423 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9426 /* we only allow rename subvolume link between subvolumes */
9427 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9430 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9431 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9434 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9435 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9438 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9442 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9444 fscrypt_free_filename(&old_fname);
9448 /* check for collisions, even if the name isn't there */
9449 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9451 if (ret == -EEXIST) {
9453 * eexist without a new_inode */
9454 if (WARN_ON(!new_inode)) {
9455 goto out_fscrypt_names;
9458 /* maybe -EOVERFLOW */
9459 goto out_fscrypt_names;
9465 * we're using rename to replace one file with another. Start IO on it
9466 * now so we don't add too much work to the end of the transaction
9468 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9469 filemap_flush(old_inode->i_mapping);
9471 if (flags & RENAME_WHITEOUT) {
9472 whiteout_args.inode = new_whiteout_inode(mnt_userns, old_dir);
9473 if (!whiteout_args.inode) {
9475 goto out_fscrypt_names;
9477 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9479 goto out_whiteout_inode;
9481 /* 1 to update the old parent inode. */
9482 trans_num_items = 1;
9485 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9486 /* Close the race window with snapshot create/destroy ioctl */
9487 down_read(&fs_info->subvol_sem);
9489 * 1 to remove old root ref
9490 * 1 to remove old root backref
9491 * 1 to add new root ref
9492 * 1 to add new root backref
9494 trans_num_items += 4;
9498 * 1 to remove old inode ref
9499 * 1 to add new inode ref
9501 trans_num_items += 3;
9504 * 1 to remove old dir item
9505 * 1 to remove old dir index
9506 * 1 to add new dir item
9507 * 1 to add new dir index
9509 trans_num_items += 4;
9510 /* 1 to update new parent inode if it's not the same as the old parent */
9511 if (new_dir != old_dir)
9516 * 1 to remove inode ref
9517 * 1 to remove dir item
9518 * 1 to remove dir index
9519 * 1 to possibly add orphan item
9521 trans_num_items += 5;
9523 trans = btrfs_start_transaction(root, trans_num_items);
9524 if (IS_ERR(trans)) {
9525 ret = PTR_ERR(trans);
9530 ret = btrfs_record_root_in_trans(trans, dest);
9535 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9539 BTRFS_I(old_inode)->dir_index = 0ULL;
9540 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9541 /* force full log commit if subvolume involved. */
9542 btrfs_set_log_full_commit(trans);
9544 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9545 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9551 inode_inc_iversion(old_dir);
9552 inode_inc_iversion(new_dir);
9553 inode_inc_iversion(old_inode);
9554 old_dir->i_mtime = current_time(old_dir);
9555 old_dir->i_ctime = old_dir->i_mtime;
9556 new_dir->i_mtime = old_dir->i_mtime;
9557 new_dir->i_ctime = old_dir->i_mtime;
9558 old_inode->i_ctime = old_dir->i_mtime;
9560 if (old_dentry->d_parent != new_dentry->d_parent)
9561 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9562 BTRFS_I(old_inode), 1);
9564 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9565 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9567 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9568 BTRFS_I(d_inode(old_dentry)),
9569 &old_fname.disk_name, &rename_ctx);
9571 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9574 btrfs_abort_transaction(trans, ret);
9579 inode_inc_iversion(new_inode);
9580 new_inode->i_ctime = current_time(new_inode);
9581 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9582 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9583 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9584 BUG_ON(new_inode->i_nlink == 0);
9586 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9587 BTRFS_I(d_inode(new_dentry)),
9588 &new_fname.disk_name);
9590 if (!ret && new_inode->i_nlink == 0)
9591 ret = btrfs_orphan_add(trans,
9592 BTRFS_I(d_inode(new_dentry)));
9594 btrfs_abort_transaction(trans, ret);
9599 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9600 &new_fname.disk_name, 0, index);
9602 btrfs_abort_transaction(trans, ret);
9606 if (old_inode->i_nlink == 1)
9607 BTRFS_I(old_inode)->dir_index = index;
9609 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9610 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9611 rename_ctx.index, new_dentry->d_parent);
9613 if (flags & RENAME_WHITEOUT) {
9614 ret = btrfs_create_new_inode(trans, &whiteout_args);
9616 btrfs_abort_transaction(trans, ret);
9619 unlock_new_inode(whiteout_args.inode);
9620 iput(whiteout_args.inode);
9621 whiteout_args.inode = NULL;
9625 ret2 = btrfs_end_transaction(trans);
9626 ret = ret ? ret : ret2;
9628 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9629 up_read(&fs_info->subvol_sem);
9630 if (flags & RENAME_WHITEOUT)
9631 btrfs_new_inode_args_destroy(&whiteout_args);
9633 if (flags & RENAME_WHITEOUT)
9634 iput(whiteout_args.inode);
9636 fscrypt_free_filename(&old_fname);
9637 fscrypt_free_filename(&new_fname);
9641 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9642 struct dentry *old_dentry, struct inode *new_dir,
9643 struct dentry *new_dentry, unsigned int flags)
9647 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9650 if (flags & RENAME_EXCHANGE)
9651 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9654 ret = btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9657 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9662 struct btrfs_delalloc_work {
9663 struct inode *inode;
9664 struct completion completion;
9665 struct list_head list;
9666 struct btrfs_work work;
9669 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9671 struct btrfs_delalloc_work *delalloc_work;
9672 struct inode *inode;
9674 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9676 inode = delalloc_work->inode;
9677 filemap_flush(inode->i_mapping);
9678 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9679 &BTRFS_I(inode)->runtime_flags))
9680 filemap_flush(inode->i_mapping);
9683 complete(&delalloc_work->completion);
9686 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9688 struct btrfs_delalloc_work *work;
9690 work = kmalloc(sizeof(*work), GFP_NOFS);
9694 init_completion(&work->completion);
9695 INIT_LIST_HEAD(&work->list);
9696 work->inode = inode;
9697 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9703 * some fairly slow code that needs optimization. This walks the list
9704 * of all the inodes with pending delalloc and forces them to disk.
9706 static int start_delalloc_inodes(struct btrfs_root *root,
9707 struct writeback_control *wbc, bool snapshot,
9708 bool in_reclaim_context)
9710 struct btrfs_inode *binode;
9711 struct inode *inode;
9712 struct btrfs_delalloc_work *work, *next;
9713 struct list_head works;
9714 struct list_head splice;
9716 bool full_flush = wbc->nr_to_write == LONG_MAX;
9718 INIT_LIST_HEAD(&works);
9719 INIT_LIST_HEAD(&splice);
9721 mutex_lock(&root->delalloc_mutex);
9722 spin_lock(&root->delalloc_lock);
9723 list_splice_init(&root->delalloc_inodes, &splice);
9724 while (!list_empty(&splice)) {
9725 binode = list_entry(splice.next, struct btrfs_inode,
9728 list_move_tail(&binode->delalloc_inodes,
9729 &root->delalloc_inodes);
9731 if (in_reclaim_context &&
9732 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9735 inode = igrab(&binode->vfs_inode);
9737 cond_resched_lock(&root->delalloc_lock);
9740 spin_unlock(&root->delalloc_lock);
9743 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9744 &binode->runtime_flags);
9746 work = btrfs_alloc_delalloc_work(inode);
9752 list_add_tail(&work->list, &works);
9753 btrfs_queue_work(root->fs_info->flush_workers,
9756 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9757 btrfs_add_delayed_iput(inode);
9758 if (ret || wbc->nr_to_write <= 0)
9762 spin_lock(&root->delalloc_lock);
9764 spin_unlock(&root->delalloc_lock);
9767 list_for_each_entry_safe(work, next, &works, list) {
9768 list_del_init(&work->list);
9769 wait_for_completion(&work->completion);
9773 if (!list_empty(&splice)) {
9774 spin_lock(&root->delalloc_lock);
9775 list_splice_tail(&splice, &root->delalloc_inodes);
9776 spin_unlock(&root->delalloc_lock);
9778 mutex_unlock(&root->delalloc_mutex);
9782 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9784 struct writeback_control wbc = {
9785 .nr_to_write = LONG_MAX,
9786 .sync_mode = WB_SYNC_NONE,
9788 .range_end = LLONG_MAX,
9790 struct btrfs_fs_info *fs_info = root->fs_info;
9792 if (BTRFS_FS_ERROR(fs_info))
9795 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9798 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9799 bool in_reclaim_context)
9801 struct writeback_control wbc = {
9803 .sync_mode = WB_SYNC_NONE,
9805 .range_end = LLONG_MAX,
9807 struct btrfs_root *root;
9808 struct list_head splice;
9811 if (BTRFS_FS_ERROR(fs_info))
9814 INIT_LIST_HEAD(&splice);
9816 mutex_lock(&fs_info->delalloc_root_mutex);
9817 spin_lock(&fs_info->delalloc_root_lock);
9818 list_splice_init(&fs_info->delalloc_roots, &splice);
9819 while (!list_empty(&splice)) {
9821 * Reset nr_to_write here so we know that we're doing a full
9825 wbc.nr_to_write = LONG_MAX;
9827 root = list_first_entry(&splice, struct btrfs_root,
9829 root = btrfs_grab_root(root);
9831 list_move_tail(&root->delalloc_root,
9832 &fs_info->delalloc_roots);
9833 spin_unlock(&fs_info->delalloc_root_lock);
9835 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9836 btrfs_put_root(root);
9837 if (ret < 0 || wbc.nr_to_write <= 0)
9839 spin_lock(&fs_info->delalloc_root_lock);
9841 spin_unlock(&fs_info->delalloc_root_lock);
9845 if (!list_empty(&splice)) {
9846 spin_lock(&fs_info->delalloc_root_lock);
9847 list_splice_tail(&splice, &fs_info->delalloc_roots);
9848 spin_unlock(&fs_info->delalloc_root_lock);
9850 mutex_unlock(&fs_info->delalloc_root_mutex);
9854 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9855 struct dentry *dentry, const char *symname)
9857 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9858 struct btrfs_trans_handle *trans;
9859 struct btrfs_root *root = BTRFS_I(dir)->root;
9860 struct btrfs_path *path;
9861 struct btrfs_key key;
9862 struct inode *inode;
9863 struct btrfs_new_inode_args new_inode_args = {
9867 unsigned int trans_num_items;
9872 struct btrfs_file_extent_item *ei;
9873 struct extent_buffer *leaf;
9875 name_len = strlen(symname);
9876 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9877 return -ENAMETOOLONG;
9879 inode = new_inode(dir->i_sb);
9882 inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO);
9883 inode->i_op = &btrfs_symlink_inode_operations;
9884 inode_nohighmem(inode);
9885 inode->i_mapping->a_ops = &btrfs_aops;
9886 btrfs_i_size_write(BTRFS_I(inode), name_len);
9887 inode_set_bytes(inode, name_len);
9889 new_inode_args.inode = inode;
9890 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9893 /* 1 additional item for the inline extent */
9896 trans = btrfs_start_transaction(root, trans_num_items);
9897 if (IS_ERR(trans)) {
9898 err = PTR_ERR(trans);
9899 goto out_new_inode_args;
9902 err = btrfs_create_new_inode(trans, &new_inode_args);
9906 path = btrfs_alloc_path();
9909 btrfs_abort_transaction(trans, err);
9910 discard_new_inode(inode);
9914 key.objectid = btrfs_ino(BTRFS_I(inode));
9916 key.type = BTRFS_EXTENT_DATA_KEY;
9917 datasize = btrfs_file_extent_calc_inline_size(name_len);
9918 err = btrfs_insert_empty_item(trans, root, path, &key,
9921 btrfs_abort_transaction(trans, err);
9922 btrfs_free_path(path);
9923 discard_new_inode(inode);
9927 leaf = path->nodes[0];
9928 ei = btrfs_item_ptr(leaf, path->slots[0],
9929 struct btrfs_file_extent_item);
9930 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9931 btrfs_set_file_extent_type(leaf, ei,
9932 BTRFS_FILE_EXTENT_INLINE);
9933 btrfs_set_file_extent_encryption(leaf, ei, 0);
9934 btrfs_set_file_extent_compression(leaf, ei, 0);
9935 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9936 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9938 ptr = btrfs_file_extent_inline_start(ei);
9939 write_extent_buffer(leaf, symname, ptr, name_len);
9940 btrfs_mark_buffer_dirty(leaf);
9941 btrfs_free_path(path);
9943 d_instantiate_new(dentry, inode);
9946 btrfs_end_transaction(trans);
9947 btrfs_btree_balance_dirty(fs_info);
9949 btrfs_new_inode_args_destroy(&new_inode_args);
9956 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9957 struct btrfs_trans_handle *trans_in,
9958 struct btrfs_inode *inode,
9959 struct btrfs_key *ins,
9962 struct btrfs_file_extent_item stack_fi;
9963 struct btrfs_replace_extent_info extent_info;
9964 struct btrfs_trans_handle *trans = trans_in;
9965 struct btrfs_path *path;
9966 u64 start = ins->objectid;
9967 u64 len = ins->offset;
9968 u64 qgroup_released = 0;
9971 memset(&stack_fi, 0, sizeof(stack_fi));
9973 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9974 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9975 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9976 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9977 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9978 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9979 /* Encryption and other encoding is reserved and all 0 */
9981 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9983 return ERR_PTR(ret);
9986 ret = insert_reserved_file_extent(trans, inode,
9987 file_offset, &stack_fi,
9988 true, qgroup_released);
9994 extent_info.disk_offset = start;
9995 extent_info.disk_len = len;
9996 extent_info.data_offset = 0;
9997 extent_info.data_len = len;
9998 extent_info.file_offset = file_offset;
9999 extent_info.extent_buf = (char *)&stack_fi;
10000 extent_info.is_new_extent = true;
10001 extent_info.update_times = true;
10002 extent_info.qgroup_reserved = qgroup_released;
10003 extent_info.insertions = 0;
10005 path = btrfs_alloc_path();
10011 ret = btrfs_replace_file_extents(inode, path, file_offset,
10012 file_offset + len - 1, &extent_info,
10014 btrfs_free_path(path);
10021 * We have released qgroup data range at the beginning of the function,
10022 * and normally qgroup_released bytes will be freed when committing
10024 * But if we error out early, we have to free what we have released
10025 * or we leak qgroup data reservation.
10027 btrfs_qgroup_free_refroot(inode->root->fs_info,
10028 inode->root->root_key.objectid, qgroup_released,
10029 BTRFS_QGROUP_RSV_DATA);
10030 return ERR_PTR(ret);
10033 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10034 u64 start, u64 num_bytes, u64 min_size,
10035 loff_t actual_len, u64 *alloc_hint,
10036 struct btrfs_trans_handle *trans)
10038 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10039 struct extent_map *em;
10040 struct btrfs_root *root = BTRFS_I(inode)->root;
10041 struct btrfs_key ins;
10042 u64 cur_offset = start;
10043 u64 clear_offset = start;
10046 u64 last_alloc = (u64)-1;
10048 bool own_trans = true;
10049 u64 end = start + num_bytes - 1;
10053 while (num_bytes > 0) {
10054 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10055 cur_bytes = max(cur_bytes, min_size);
10057 * If we are severely fragmented we could end up with really
10058 * small allocations, so if the allocator is returning small
10059 * chunks lets make its job easier by only searching for those
10062 cur_bytes = min(cur_bytes, last_alloc);
10063 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10064 min_size, 0, *alloc_hint, &ins, 1, 0);
10069 * We've reserved this space, and thus converted it from
10070 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10071 * from here on out we will only need to clear our reservation
10072 * for the remaining unreserved area, so advance our
10073 * clear_offset by our extent size.
10075 clear_offset += ins.offset;
10077 last_alloc = ins.offset;
10078 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10081 * Now that we inserted the prealloc extent we can finally
10082 * decrement the number of reservations in the block group.
10083 * If we did it before, we could race with relocation and have
10084 * relocation miss the reserved extent, making it fail later.
10086 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10087 if (IS_ERR(trans)) {
10088 ret = PTR_ERR(trans);
10089 btrfs_free_reserved_extent(fs_info, ins.objectid,
10094 em = alloc_extent_map();
10096 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
10097 cur_offset + ins.offset - 1, false);
10098 btrfs_set_inode_full_sync(BTRFS_I(inode));
10102 em->start = cur_offset;
10103 em->orig_start = cur_offset;
10104 em->len = ins.offset;
10105 em->block_start = ins.objectid;
10106 em->block_len = ins.offset;
10107 em->orig_block_len = ins.offset;
10108 em->ram_bytes = ins.offset;
10109 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10110 em->generation = trans->transid;
10112 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
10113 free_extent_map(em);
10115 num_bytes -= ins.offset;
10116 cur_offset += ins.offset;
10117 *alloc_hint = ins.objectid + ins.offset;
10119 inode_inc_iversion(inode);
10120 inode->i_ctime = current_time(inode);
10121 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10122 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10123 (actual_len > inode->i_size) &&
10124 (cur_offset > inode->i_size)) {
10125 if (cur_offset > actual_len)
10126 i_size = actual_len;
10128 i_size = cur_offset;
10129 i_size_write(inode, i_size);
10130 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10133 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10136 btrfs_abort_transaction(trans, ret);
10138 btrfs_end_transaction(trans);
10143 btrfs_end_transaction(trans);
10147 if (clear_offset < end)
10148 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10149 end - clear_offset + 1);
10153 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10154 u64 start, u64 num_bytes, u64 min_size,
10155 loff_t actual_len, u64 *alloc_hint)
10157 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10158 min_size, actual_len, alloc_hint,
10162 int btrfs_prealloc_file_range_trans(struct inode *inode,
10163 struct btrfs_trans_handle *trans, int mode,
10164 u64 start, u64 num_bytes, u64 min_size,
10165 loff_t actual_len, u64 *alloc_hint)
10167 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10168 min_size, actual_len, alloc_hint, trans);
10171 static int btrfs_permission(struct user_namespace *mnt_userns,
10172 struct inode *inode, int mask)
10174 struct btrfs_root *root = BTRFS_I(inode)->root;
10175 umode_t mode = inode->i_mode;
10177 if (mask & MAY_WRITE &&
10178 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10179 if (btrfs_root_readonly(root))
10181 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10184 return generic_permission(mnt_userns, inode, mask);
10187 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10188 struct file *file, umode_t mode)
10190 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10191 struct btrfs_trans_handle *trans;
10192 struct btrfs_root *root = BTRFS_I(dir)->root;
10193 struct inode *inode;
10194 struct btrfs_new_inode_args new_inode_args = {
10196 .dentry = file->f_path.dentry,
10199 unsigned int trans_num_items;
10202 inode = new_inode(dir->i_sb);
10205 inode_init_owner(mnt_userns, inode, dir, mode);
10206 inode->i_fop = &btrfs_file_operations;
10207 inode->i_op = &btrfs_file_inode_operations;
10208 inode->i_mapping->a_ops = &btrfs_aops;
10210 new_inode_args.inode = inode;
10211 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
10215 trans = btrfs_start_transaction(root, trans_num_items);
10216 if (IS_ERR(trans)) {
10217 ret = PTR_ERR(trans);
10218 goto out_new_inode_args;
10221 ret = btrfs_create_new_inode(trans, &new_inode_args);
10224 * We set number of links to 0 in btrfs_create_new_inode(), and here we
10225 * set it to 1 because d_tmpfile() will issue a warning if the count is
10228 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10230 set_nlink(inode, 1);
10233 d_tmpfile(file, inode);
10234 unlock_new_inode(inode);
10235 mark_inode_dirty(inode);
10238 btrfs_end_transaction(trans);
10239 btrfs_btree_balance_dirty(fs_info);
10240 out_new_inode_args:
10241 btrfs_new_inode_args_destroy(&new_inode_args);
10245 return finish_open_simple(file, ret);
10248 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10250 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10251 unsigned long index = start >> PAGE_SHIFT;
10252 unsigned long end_index = end >> PAGE_SHIFT;
10256 ASSERT(end + 1 - start <= U32_MAX);
10257 len = end + 1 - start;
10258 while (index <= end_index) {
10259 page = find_get_page(inode->vfs_inode.i_mapping, index);
10260 ASSERT(page); /* Pages should be in the extent_io_tree */
10262 btrfs_page_set_writeback(fs_info, page, start, len);
10268 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
10271 switch (compress_type) {
10272 case BTRFS_COMPRESS_NONE:
10273 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10274 case BTRFS_COMPRESS_ZLIB:
10275 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10276 case BTRFS_COMPRESS_LZO:
10278 * The LZO format depends on the sector size. 64K is the maximum
10279 * sector size that we support.
10281 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10283 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10284 (fs_info->sectorsize_bits - 12);
10285 case BTRFS_COMPRESS_ZSTD:
10286 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10292 static ssize_t btrfs_encoded_read_inline(
10293 struct kiocb *iocb,
10294 struct iov_iter *iter, u64 start,
10296 struct extent_state **cached_state,
10297 u64 extent_start, size_t count,
10298 struct btrfs_ioctl_encoded_io_args *encoded,
10301 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10302 struct btrfs_root *root = inode->root;
10303 struct btrfs_fs_info *fs_info = root->fs_info;
10304 struct extent_io_tree *io_tree = &inode->io_tree;
10305 struct btrfs_path *path;
10306 struct extent_buffer *leaf;
10307 struct btrfs_file_extent_item *item;
10313 path = btrfs_alloc_path();
10318 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10322 /* The extent item disappeared? */
10327 leaf = path->nodes[0];
10328 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10330 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10331 ptr = btrfs_file_extent_inline_start(item);
10333 encoded->len = min_t(u64, extent_start + ram_bytes,
10334 inode->vfs_inode.i_size) - iocb->ki_pos;
10335 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10336 btrfs_file_extent_compression(leaf, item));
10339 encoded->compression = ret;
10340 if (encoded->compression) {
10341 size_t inline_size;
10343 inline_size = btrfs_file_extent_inline_item_len(leaf,
10345 if (inline_size > count) {
10349 count = inline_size;
10350 encoded->unencoded_len = ram_bytes;
10351 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10353 count = min_t(u64, count, encoded->len);
10354 encoded->len = count;
10355 encoded->unencoded_len = count;
10356 ptr += iocb->ki_pos - extent_start;
10359 tmp = kmalloc(count, GFP_NOFS);
10364 read_extent_buffer(leaf, tmp, ptr, count);
10365 btrfs_release_path(path);
10366 unlock_extent(io_tree, start, lockend, cached_state);
10367 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10370 ret = copy_to_iter(tmp, count, iter);
10375 btrfs_free_path(path);
10379 struct btrfs_encoded_read_private {
10380 struct btrfs_inode *inode;
10382 wait_queue_head_t wait;
10384 blk_status_t status;
10388 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10389 struct bio *bio, int mirror_num)
10391 struct btrfs_encoded_read_private *priv = btrfs_bio(bio)->private;
10392 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10395 if (!priv->skip_csum) {
10396 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10401 atomic_inc(&priv->pending);
10402 btrfs_submit_bio(fs_info, bio, mirror_num);
10406 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10408 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10409 struct btrfs_encoded_read_private *priv = bbio->private;
10410 struct btrfs_inode *inode = priv->inode;
10411 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10412 u32 sectorsize = fs_info->sectorsize;
10413 struct bio_vec *bvec;
10414 struct bvec_iter_all iter_all;
10415 u32 bio_offset = 0;
10417 if (priv->skip_csum || !uptodate)
10418 return bbio->bio.bi_status;
10420 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10421 unsigned int i, nr_sectors, pgoff;
10423 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10424 pgoff = bvec->bv_offset;
10425 for (i = 0; i < nr_sectors; i++) {
10426 ASSERT(pgoff < PAGE_SIZE);
10427 if (btrfs_check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10428 bvec->bv_page, pgoff))
10429 return BLK_STS_IOERR;
10430 bio_offset += sectorsize;
10431 pgoff += sectorsize;
10437 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10439 struct btrfs_encoded_read_private *priv = bbio->private;
10440 blk_status_t status;
10442 status = btrfs_encoded_read_verify_csum(bbio);
10445 * The memory barrier implied by the atomic_dec_return() here
10446 * pairs with the memory barrier implied by the
10447 * atomic_dec_return() or io_wait_event() in
10448 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10449 * write is observed before the load of status in
10450 * btrfs_encoded_read_regular_fill_pages().
10452 WRITE_ONCE(priv->status, status);
10454 if (!atomic_dec_return(&priv->pending))
10455 wake_up(&priv->wait);
10456 btrfs_bio_free_csum(bbio);
10457 bio_put(&bbio->bio);
10460 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10461 u64 file_offset, u64 disk_bytenr,
10462 u64 disk_io_size, struct page **pages)
10464 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10465 struct btrfs_encoded_read_private priv = {
10467 .file_offset = file_offset,
10468 .pending = ATOMIC_INIT(1),
10469 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10471 unsigned long i = 0;
10475 init_waitqueue_head(&priv.wait);
10477 * Submit bios for the extent, splitting due to bio or stripe limits as
10480 while (cur < disk_io_size) {
10481 struct extent_map *em;
10482 struct btrfs_io_geometry geom;
10483 struct bio *bio = NULL;
10486 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10487 disk_io_size - cur);
10491 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10492 disk_bytenr + cur, &geom);
10493 free_extent_map(em);
10496 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10499 remaining = min(geom.len, disk_io_size - cur);
10500 while (bio || remaining) {
10501 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10504 bio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ,
10505 btrfs_encoded_read_endio,
10507 bio->bi_iter.bi_sector =
10508 (disk_bytenr + cur) >> SECTOR_SHIFT;
10512 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10513 blk_status_t status;
10515 status = submit_encoded_read_bio(inode, bio, 0);
10517 WRITE_ONCE(priv.status, status);
10527 remaining -= bytes;
10532 if (atomic_dec_return(&priv.pending))
10533 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10534 /* See btrfs_encoded_read_endio() for ordering. */
10535 return blk_status_to_errno(READ_ONCE(priv.status));
10538 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10539 struct iov_iter *iter,
10540 u64 start, u64 lockend,
10541 struct extent_state **cached_state,
10542 u64 disk_bytenr, u64 disk_io_size,
10543 size_t count, bool compressed,
10546 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10547 struct extent_io_tree *io_tree = &inode->io_tree;
10548 struct page **pages;
10549 unsigned long nr_pages, i;
10551 size_t page_offset;
10554 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10555 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10558 ret = btrfs_alloc_page_array(nr_pages, pages);
10564 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10565 disk_io_size, pages);
10569 unlock_extent(io_tree, start, lockend, cached_state);
10570 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10577 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10578 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10581 while (cur < count) {
10582 size_t bytes = min_t(size_t, count - cur,
10583 PAGE_SIZE - page_offset);
10585 if (copy_page_to_iter(pages[i], page_offset, bytes,
10596 for (i = 0; i < nr_pages; i++) {
10598 __free_page(pages[i]);
10604 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10605 struct btrfs_ioctl_encoded_io_args *encoded)
10607 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10608 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10609 struct extent_io_tree *io_tree = &inode->io_tree;
10611 size_t count = iov_iter_count(iter);
10612 u64 start, lockend, disk_bytenr, disk_io_size;
10613 struct extent_state *cached_state = NULL;
10614 struct extent_map *em;
10615 bool unlocked = false;
10617 file_accessed(iocb->ki_filp);
10619 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10621 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10622 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10625 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10627 * We don't know how long the extent containing iocb->ki_pos is, but if
10628 * it's compressed we know that it won't be longer than this.
10630 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10633 struct btrfs_ordered_extent *ordered;
10635 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10636 lockend - start + 1);
10638 goto out_unlock_inode;
10639 lock_extent(io_tree, start, lockend, &cached_state);
10640 ordered = btrfs_lookup_ordered_range(inode, start,
10641 lockend - start + 1);
10644 btrfs_put_ordered_extent(ordered);
10645 unlock_extent(io_tree, start, lockend, &cached_state);
10649 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10652 goto out_unlock_extent;
10655 if (em->block_start == EXTENT_MAP_INLINE) {
10656 u64 extent_start = em->start;
10659 * For inline extents we get everything we need out of the
10662 free_extent_map(em);
10664 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10665 &cached_state, extent_start,
10666 count, encoded, &unlocked);
10671 * We only want to return up to EOF even if the extent extends beyond
10674 encoded->len = min_t(u64, extent_map_end(em),
10675 inode->vfs_inode.i_size) - iocb->ki_pos;
10676 if (em->block_start == EXTENT_MAP_HOLE ||
10677 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10678 disk_bytenr = EXTENT_MAP_HOLE;
10679 count = min_t(u64, count, encoded->len);
10680 encoded->len = count;
10681 encoded->unencoded_len = count;
10682 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10683 disk_bytenr = em->block_start;
10685 * Bail if the buffer isn't large enough to return the whole
10686 * compressed extent.
10688 if (em->block_len > count) {
10692 disk_io_size = em->block_len;
10693 count = em->block_len;
10694 encoded->unencoded_len = em->ram_bytes;
10695 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10696 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10697 em->compress_type);
10700 encoded->compression = ret;
10702 disk_bytenr = em->block_start + (start - em->start);
10703 if (encoded->len > count)
10704 encoded->len = count;
10706 * Don't read beyond what we locked. This also limits the page
10707 * allocations that we'll do.
10709 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10710 count = start + disk_io_size - iocb->ki_pos;
10711 encoded->len = count;
10712 encoded->unencoded_len = count;
10713 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10715 free_extent_map(em);
10718 if (disk_bytenr == EXTENT_MAP_HOLE) {
10719 unlock_extent(io_tree, start, lockend, &cached_state);
10720 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10722 ret = iov_iter_zero(count, iter);
10726 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10727 &cached_state, disk_bytenr,
10728 disk_io_size, count,
10729 encoded->compression,
10735 iocb->ki_pos += encoded->len;
10737 free_extent_map(em);
10740 unlock_extent(io_tree, start, lockend, &cached_state);
10743 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10747 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10748 const struct btrfs_ioctl_encoded_io_args *encoded)
10750 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10751 struct btrfs_root *root = inode->root;
10752 struct btrfs_fs_info *fs_info = root->fs_info;
10753 struct extent_io_tree *io_tree = &inode->io_tree;
10754 struct extent_changeset *data_reserved = NULL;
10755 struct extent_state *cached_state = NULL;
10759 u64 num_bytes, ram_bytes, disk_num_bytes;
10760 unsigned long nr_pages, i;
10761 struct page **pages;
10762 struct btrfs_key ins;
10763 bool extent_reserved = false;
10764 struct extent_map *em;
10767 switch (encoded->compression) {
10768 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10769 compression = BTRFS_COMPRESS_ZLIB;
10771 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10772 compression = BTRFS_COMPRESS_ZSTD;
10774 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10775 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10776 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10777 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10778 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10779 /* The sector size must match for LZO. */
10780 if (encoded->compression -
10781 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10782 fs_info->sectorsize_bits)
10784 compression = BTRFS_COMPRESS_LZO;
10789 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10793 * Compressed extents should always have checksums, so error out if we
10794 * have a NOCOW file or inode was created while mounted with NODATASUM.
10796 if (inode->flags & BTRFS_INODE_NODATASUM)
10799 orig_count = iov_iter_count(from);
10801 /* The extent size must be sane. */
10802 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10803 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10807 * The compressed data must be smaller than the decompressed data.
10809 * It's of course possible for data to compress to larger or the same
10810 * size, but the buffered I/O path falls back to no compression for such
10811 * data, and we don't want to break any assumptions by creating these
10814 * Note that this is less strict than the current check we have that the
10815 * compressed data must be at least one sector smaller than the
10816 * decompressed data. We only want to enforce the weaker requirement
10817 * from old kernels that it is at least one byte smaller.
10819 if (orig_count >= encoded->unencoded_len)
10822 /* The extent must start on a sector boundary. */
10823 start = iocb->ki_pos;
10824 if (!IS_ALIGNED(start, fs_info->sectorsize))
10828 * The extent must end on a sector boundary. However, we allow a write
10829 * which ends at or extends i_size to have an unaligned length; we round
10830 * up the extent size and set i_size to the unaligned end.
10832 if (start + encoded->len < inode->vfs_inode.i_size &&
10833 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10836 /* Finally, the offset in the unencoded data must be sector-aligned. */
10837 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10840 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10841 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10842 end = start + num_bytes - 1;
10845 * If the extent cannot be inline, the compressed data on disk must be
10846 * sector-aligned. For convenience, we extend it with zeroes if it
10849 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10850 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10851 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10854 for (i = 0; i < nr_pages; i++) {
10855 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10858 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10863 kaddr = kmap_local_page(pages[i]);
10864 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10865 kunmap_local(kaddr);
10869 if (bytes < PAGE_SIZE)
10870 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10871 kunmap_local(kaddr);
10875 struct btrfs_ordered_extent *ordered;
10877 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10880 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10881 start >> PAGE_SHIFT,
10882 end >> PAGE_SHIFT);
10885 lock_extent(io_tree, start, end, &cached_state);
10886 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10888 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10891 btrfs_put_ordered_extent(ordered);
10892 unlock_extent(io_tree, start, end, &cached_state);
10897 * We don't use the higher-level delalloc space functions because our
10898 * num_bytes and disk_num_bytes are different.
10900 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10903 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10905 goto out_free_data_space;
10906 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10909 goto out_qgroup_free_data;
10911 /* Try an inline extent first. */
10912 if (start == 0 && encoded->unencoded_len == encoded->len &&
10913 encoded->unencoded_offset == 0) {
10914 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10915 compression, pages, true);
10919 goto out_delalloc_release;
10923 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10924 disk_num_bytes, 0, 0, &ins, 1, 1);
10926 goto out_delalloc_release;
10927 extent_reserved = true;
10929 em = create_io_em(inode, start, num_bytes,
10930 start - encoded->unencoded_offset, ins.objectid,
10931 ins.offset, ins.offset, ram_bytes, compression,
10932 BTRFS_ORDERED_COMPRESSED);
10935 goto out_free_reserved;
10937 free_extent_map(em);
10939 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10940 ins.objectid, ins.offset,
10941 encoded->unencoded_offset,
10942 (1 << BTRFS_ORDERED_ENCODED) |
10943 (1 << BTRFS_ORDERED_COMPRESSED),
10946 btrfs_drop_extent_map_range(inode, start, end, false);
10947 goto out_free_reserved;
10949 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10951 if (start + encoded->len > inode->vfs_inode.i_size)
10952 i_size_write(&inode->vfs_inode, start + encoded->len);
10954 unlock_extent(io_tree, start, end, &cached_state);
10956 btrfs_delalloc_release_extents(inode, num_bytes);
10958 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10959 ins.offset, pages, nr_pages, 0, NULL,
10961 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10969 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10970 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10971 out_delalloc_release:
10972 btrfs_delalloc_release_extents(inode, num_bytes);
10973 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10974 out_qgroup_free_data:
10976 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10977 out_free_data_space:
10979 * If btrfs_reserve_extent() succeeded, then we already decremented
10982 if (!extent_reserved)
10983 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10985 unlock_extent(io_tree, start, end, &cached_state);
10987 for (i = 0; i < nr_pages; i++) {
10989 __free_page(pages[i]);
10994 iocb->ki_pos += encoded->len;
11000 * Add an entry indicating a block group or device which is pinned by a
11001 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
11002 * negative errno on failure.
11004 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
11005 bool is_block_group)
11007 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
11008 struct btrfs_swapfile_pin *sp, *entry;
11009 struct rb_node **p;
11010 struct rb_node *parent = NULL;
11012 sp = kmalloc(sizeof(*sp), GFP_NOFS);
11017 sp->is_block_group = is_block_group;
11018 sp->bg_extent_count = 1;
11020 spin_lock(&fs_info->swapfile_pins_lock);
11021 p = &fs_info->swapfile_pins.rb_node;
11024 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
11025 if (sp->ptr < entry->ptr ||
11026 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
11027 p = &(*p)->rb_left;
11028 } else if (sp->ptr > entry->ptr ||
11029 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
11030 p = &(*p)->rb_right;
11032 if (is_block_group)
11033 entry->bg_extent_count++;
11034 spin_unlock(&fs_info->swapfile_pins_lock);
11039 rb_link_node(&sp->node, parent, p);
11040 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
11041 spin_unlock(&fs_info->swapfile_pins_lock);
11045 /* Free all of the entries pinned by this swapfile. */
11046 static void btrfs_free_swapfile_pins(struct inode *inode)
11048 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
11049 struct btrfs_swapfile_pin *sp;
11050 struct rb_node *node, *next;
11052 spin_lock(&fs_info->swapfile_pins_lock);
11053 node = rb_first(&fs_info->swapfile_pins);
11055 next = rb_next(node);
11056 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
11057 if (sp->inode == inode) {
11058 rb_erase(&sp->node, &fs_info->swapfile_pins);
11059 if (sp->is_block_group) {
11060 btrfs_dec_block_group_swap_extents(sp->ptr,
11061 sp->bg_extent_count);
11062 btrfs_put_block_group(sp->ptr);
11068 spin_unlock(&fs_info->swapfile_pins_lock);
11071 struct btrfs_swap_info {
11077 unsigned long nr_pages;
11081 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
11082 struct btrfs_swap_info *bsi)
11084 unsigned long nr_pages;
11085 unsigned long max_pages;
11086 u64 first_ppage, first_ppage_reported, next_ppage;
11090 * Our swapfile may have had its size extended after the swap header was
11091 * written. In that case activating the swapfile should not go beyond
11092 * the max size set in the swap header.
11094 if (bsi->nr_pages >= sis->max)
11097 max_pages = sis->max - bsi->nr_pages;
11098 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
11099 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
11100 PAGE_SIZE) >> PAGE_SHIFT;
11102 if (first_ppage >= next_ppage)
11104 nr_pages = next_ppage - first_ppage;
11105 nr_pages = min(nr_pages, max_pages);
11107 first_ppage_reported = first_ppage;
11108 if (bsi->start == 0)
11109 first_ppage_reported++;
11110 if (bsi->lowest_ppage > first_ppage_reported)
11111 bsi->lowest_ppage = first_ppage_reported;
11112 if (bsi->highest_ppage < (next_ppage - 1))
11113 bsi->highest_ppage = next_ppage - 1;
11115 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
11118 bsi->nr_extents += ret;
11119 bsi->nr_pages += nr_pages;
11123 static void btrfs_swap_deactivate(struct file *file)
11125 struct inode *inode = file_inode(file);
11127 btrfs_free_swapfile_pins(inode);
11128 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
11131 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11134 struct inode *inode = file_inode(file);
11135 struct btrfs_root *root = BTRFS_I(inode)->root;
11136 struct btrfs_fs_info *fs_info = root->fs_info;
11137 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
11138 struct extent_state *cached_state = NULL;
11139 struct extent_map *em = NULL;
11140 struct btrfs_device *device = NULL;
11141 struct btrfs_swap_info bsi = {
11142 .lowest_ppage = (sector_t)-1ULL,
11149 * If the swap file was just created, make sure delalloc is done. If the
11150 * file changes again after this, the user is doing something stupid and
11151 * we don't really care.
11153 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
11158 * The inode is locked, so these flags won't change after we check them.
11160 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
11161 btrfs_warn(fs_info, "swapfile must not be compressed");
11164 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
11165 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
11168 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
11169 btrfs_warn(fs_info, "swapfile must not be checksummed");
11174 * Balance or device remove/replace/resize can move stuff around from
11175 * under us. The exclop protection makes sure they aren't running/won't
11176 * run concurrently while we are mapping the swap extents, and
11177 * fs_info->swapfile_pins prevents them from running while the swap
11178 * file is active and moving the extents. Note that this also prevents
11179 * a concurrent device add which isn't actually necessary, but it's not
11180 * really worth the trouble to allow it.
11182 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
11183 btrfs_warn(fs_info,
11184 "cannot activate swapfile while exclusive operation is running");
11189 * Prevent snapshot creation while we are activating the swap file.
11190 * We do not want to race with snapshot creation. If snapshot creation
11191 * already started before we bumped nr_swapfiles from 0 to 1 and
11192 * completes before the first write into the swap file after it is
11193 * activated, than that write would fallback to COW.
11195 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11196 btrfs_exclop_finish(fs_info);
11197 btrfs_warn(fs_info,
11198 "cannot activate swapfile because snapshot creation is in progress");
11202 * Snapshots can create extents which require COW even if NODATACOW is
11203 * set. We use this counter to prevent snapshots. We must increment it
11204 * before walking the extents because we don't want a concurrent
11205 * snapshot to run after we've already checked the extents.
11207 * It is possible that subvolume is marked for deletion but still not
11208 * removed yet. To prevent this race, we check the root status before
11209 * activating the swapfile.
11211 spin_lock(&root->root_item_lock);
11212 if (btrfs_root_dead(root)) {
11213 spin_unlock(&root->root_item_lock);
11215 btrfs_exclop_finish(fs_info);
11216 btrfs_warn(fs_info,
11217 "cannot activate swapfile because subvolume %llu is being deleted",
11218 root->root_key.objectid);
11221 atomic_inc(&root->nr_swapfiles);
11222 spin_unlock(&root->root_item_lock);
11224 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11226 lock_extent(io_tree, 0, isize - 1, &cached_state);
11228 while (start < isize) {
11229 u64 logical_block_start, physical_block_start;
11230 struct btrfs_block_group *bg;
11231 u64 len = isize - start;
11233 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11239 if (em->block_start == EXTENT_MAP_HOLE) {
11240 btrfs_warn(fs_info, "swapfile must not have holes");
11244 if (em->block_start == EXTENT_MAP_INLINE) {
11246 * It's unlikely we'll ever actually find ourselves
11247 * here, as a file small enough to fit inline won't be
11248 * big enough to store more than the swap header, but in
11249 * case something changes in the future, let's catch it
11250 * here rather than later.
11252 btrfs_warn(fs_info, "swapfile must not be inline");
11256 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11257 btrfs_warn(fs_info, "swapfile must not be compressed");
11262 logical_block_start = em->block_start + (start - em->start);
11263 len = min(len, em->len - (start - em->start));
11264 free_extent_map(em);
11267 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
11273 btrfs_warn(fs_info,
11274 "swapfile must not be copy-on-write");
11279 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11285 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11286 btrfs_warn(fs_info,
11287 "swapfile must have single data profile");
11292 if (device == NULL) {
11293 device = em->map_lookup->stripes[0].dev;
11294 ret = btrfs_add_swapfile_pin(inode, device, false);
11299 } else if (device != em->map_lookup->stripes[0].dev) {
11300 btrfs_warn(fs_info, "swapfile must be on one device");
11305 physical_block_start = (em->map_lookup->stripes[0].physical +
11306 (logical_block_start - em->start));
11307 len = min(len, em->len - (logical_block_start - em->start));
11308 free_extent_map(em);
11311 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11313 btrfs_warn(fs_info,
11314 "could not find block group containing swapfile");
11319 if (!btrfs_inc_block_group_swap_extents(bg)) {
11320 btrfs_warn(fs_info,
11321 "block group for swapfile at %llu is read-only%s",
11323 atomic_read(&fs_info->scrubs_running) ?
11324 " (scrub running)" : "");
11325 btrfs_put_block_group(bg);
11330 ret = btrfs_add_swapfile_pin(inode, bg, true);
11332 btrfs_put_block_group(bg);
11339 if (bsi.block_len &&
11340 bsi.block_start + bsi.block_len == physical_block_start) {
11341 bsi.block_len += len;
11343 if (bsi.block_len) {
11344 ret = btrfs_add_swap_extent(sis, &bsi);
11349 bsi.block_start = physical_block_start;
11350 bsi.block_len = len;
11357 ret = btrfs_add_swap_extent(sis, &bsi);
11360 if (!IS_ERR_OR_NULL(em))
11361 free_extent_map(em);
11363 unlock_extent(io_tree, 0, isize - 1, &cached_state);
11366 btrfs_swap_deactivate(file);
11368 btrfs_drew_write_unlock(&root->snapshot_lock);
11370 btrfs_exclop_finish(fs_info);
11376 sis->bdev = device->bdev;
11377 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11378 sis->max = bsi.nr_pages;
11379 sis->pages = bsi.nr_pages - 1;
11380 sis->highest_bit = bsi.nr_pages - 1;
11381 return bsi.nr_extents;
11384 static void btrfs_swap_deactivate(struct file *file)
11388 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11391 return -EOPNOTSUPP;
11396 * Update the number of bytes used in the VFS' inode. When we replace extents in
11397 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11398 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11399 * always get a correct value.
11401 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11402 const u64 add_bytes,
11403 const u64 del_bytes)
11405 if (add_bytes == del_bytes)
11408 spin_lock(&inode->lock);
11410 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11412 inode_add_bytes(&inode->vfs_inode, add_bytes);
11413 spin_unlock(&inode->lock);
11417 * Verify that there are no ordered extents for a given file range.
11419 * @inode: The target inode.
11420 * @start: Start offset of the file range, should be sector size aligned.
11421 * @end: End offset (inclusive) of the file range, its value +1 should be
11422 * sector size aligned.
11424 * This should typically be used for cases where we locked an inode's VFS lock in
11425 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11426 * we have flushed all delalloc in the range, we have waited for all ordered
11427 * extents in the range to complete and finally we have locked the file range in
11428 * the inode's io_tree.
11430 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11432 struct btrfs_root *root = inode->root;
11433 struct btrfs_ordered_extent *ordered;
11435 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11438 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11440 btrfs_err(root->fs_info,
11441 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11442 start, end, btrfs_ino(inode), root->root_key.objectid,
11443 ordered->file_offset,
11444 ordered->file_offset + ordered->num_bytes - 1);
11445 btrfs_put_ordered_extent(ordered);
11448 ASSERT(ordered == NULL);
11451 static const struct inode_operations btrfs_dir_inode_operations = {
11452 .getattr = btrfs_getattr,
11453 .lookup = btrfs_lookup,
11454 .create = btrfs_create,
11455 .unlink = btrfs_unlink,
11456 .link = btrfs_link,
11457 .mkdir = btrfs_mkdir,
11458 .rmdir = btrfs_rmdir,
11459 .rename = btrfs_rename2,
11460 .symlink = btrfs_symlink,
11461 .setattr = btrfs_setattr,
11462 .mknod = btrfs_mknod,
11463 .listxattr = btrfs_listxattr,
11464 .permission = btrfs_permission,
11465 .get_acl = btrfs_get_acl,
11466 .set_acl = btrfs_set_acl,
11467 .update_time = btrfs_update_time,
11468 .tmpfile = btrfs_tmpfile,
11469 .fileattr_get = btrfs_fileattr_get,
11470 .fileattr_set = btrfs_fileattr_set,
11473 static const struct file_operations btrfs_dir_file_operations = {
11474 .llseek = btrfs_dir_llseek,
11475 .read = generic_read_dir,
11476 .iterate_shared = btrfs_real_readdir,
11477 .open = btrfs_opendir,
11478 .unlocked_ioctl = btrfs_ioctl,
11479 #ifdef CONFIG_COMPAT
11480 .compat_ioctl = btrfs_compat_ioctl,
11482 .release = btrfs_release_file,
11483 .fsync = btrfs_sync_file,
11487 * btrfs doesn't support the bmap operation because swapfiles
11488 * use bmap to make a mapping of extents in the file. They assume
11489 * these extents won't change over the life of the file and they
11490 * use the bmap result to do IO directly to the drive.
11492 * the btrfs bmap call would return logical addresses that aren't
11493 * suitable for IO and they also will change frequently as COW
11494 * operations happen. So, swapfile + btrfs == corruption.
11496 * For now we're avoiding this by dropping bmap.
11498 static const struct address_space_operations btrfs_aops = {
11499 .read_folio = btrfs_read_folio,
11500 .writepages = btrfs_writepages,
11501 .readahead = btrfs_readahead,
11502 .direct_IO = noop_direct_IO,
11503 .invalidate_folio = btrfs_invalidate_folio,
11504 .release_folio = btrfs_release_folio,
11505 .migrate_folio = btrfs_migrate_folio,
11506 .dirty_folio = filemap_dirty_folio,
11507 .error_remove_page = generic_error_remove_page,
11508 .swap_activate = btrfs_swap_activate,
11509 .swap_deactivate = btrfs_swap_deactivate,
11512 static const struct inode_operations btrfs_file_inode_operations = {
11513 .getattr = btrfs_getattr,
11514 .setattr = btrfs_setattr,
11515 .listxattr = btrfs_listxattr,
11516 .permission = btrfs_permission,
11517 .fiemap = btrfs_fiemap,
11518 .get_acl = btrfs_get_acl,
11519 .set_acl = btrfs_set_acl,
11520 .update_time = btrfs_update_time,
11521 .fileattr_get = btrfs_fileattr_get,
11522 .fileattr_set = btrfs_fileattr_set,
11524 static const struct inode_operations btrfs_special_inode_operations = {
11525 .getattr = btrfs_getattr,
11526 .setattr = btrfs_setattr,
11527 .permission = btrfs_permission,
11528 .listxattr = btrfs_listxattr,
11529 .get_acl = btrfs_get_acl,
11530 .set_acl = btrfs_set_acl,
11531 .update_time = btrfs_update_time,
11533 static const struct inode_operations btrfs_symlink_inode_operations = {
11534 .get_link = page_get_link,
11535 .getattr = btrfs_getattr,
11536 .setattr = btrfs_setattr,
11537 .permission = btrfs_permission,
11538 .listxattr = btrfs_listxattr,
11539 .update_time = btrfs_update_time,
11542 const struct dentry_operations btrfs_dentry_operations = {
11543 .d_delete = btrfs_dentry_delete,