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 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
118 static noinline int cow_file_range(struct btrfs_inode *inode,
119 struct page *locked_page,
120 u64 start, u64 end, int *page_started,
121 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,
128 static void __endio_write_update_ordered(struct btrfs_inode *inode,
129 const u64 offset, const u64 bytes,
130 const bool uptodate);
133 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
135 * ilock_flags can have the following bit set:
137 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
138 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
140 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
142 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
144 if (ilock_flags & BTRFS_ILOCK_SHARED) {
145 if (ilock_flags & BTRFS_ILOCK_TRY) {
146 if (!inode_trylock_shared(inode))
151 inode_lock_shared(inode);
153 if (ilock_flags & BTRFS_ILOCK_TRY) {
154 if (!inode_trylock(inode))
161 if (ilock_flags & BTRFS_ILOCK_MMAP)
162 down_write(&BTRFS_I(inode)->i_mmap_lock);
167 * btrfs_inode_unlock - unock inode i_rwsem
169 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
170 * to decide whether the lock acquired is shared or exclusive.
172 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
174 if (ilock_flags & BTRFS_ILOCK_MMAP)
175 up_write(&BTRFS_I(inode)->i_mmap_lock);
176 if (ilock_flags & BTRFS_ILOCK_SHARED)
177 inode_unlock_shared(inode);
183 * Cleanup all submitted ordered extents in specified range to handle errors
184 * from the btrfs_run_delalloc_range() callback.
186 * NOTE: caller must ensure that when an error happens, it can not call
187 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
188 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
189 * to be released, which we want to happen only when finishing the ordered
190 * extent (btrfs_finish_ordered_io()).
192 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
193 struct page *locked_page,
194 u64 offset, u64 bytes)
196 unsigned long index = offset >> PAGE_SHIFT;
197 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
198 u64 page_start = page_offset(locked_page);
199 u64 page_end = page_start + PAGE_SIZE - 1;
203 while (index <= end_index) {
205 * For locked page, we will call end_extent_writepage() on it
206 * in run_delalloc_range() for the error handling. That
207 * end_extent_writepage() function will call
208 * btrfs_mark_ordered_io_finished() to clear page Ordered and
209 * run the ordered extent accounting.
211 * Here we can't just clear the Ordered bit, or
212 * btrfs_mark_ordered_io_finished() would skip the accounting
213 * for the page range, and the ordered extent will never finish.
215 if (index == (page_offset(locked_page) >> PAGE_SHIFT)) {
219 page = find_get_page(inode->vfs_inode.i_mapping, index);
225 * Here we just clear all Ordered bits for every page in the
226 * range, then __endio_write_update_ordered() will handle
227 * the ordered extent accounting for the range.
229 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_offset(locked_page) + PAGE_SIZE)
238 * In case this page belongs to the delalloc range being instantiated
239 * then skip it, since the first page of a range is going to be
240 * properly cleaned up by the caller of run_delalloc_range
242 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
243 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
244 offset = page_offset(locked_page) + PAGE_SIZE;
247 return __endio_write_update_ordered(inode, offset, bytes, false);
250 static int btrfs_dirty_inode(struct inode *inode);
252 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
253 struct btrfs_new_inode_args *args)
257 if (args->default_acl) {
258 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
264 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
268 if (!args->default_acl && !args->acl)
269 cache_no_acl(args->inode);
270 return btrfs_xattr_security_init(trans, args->inode, args->dir,
271 &args->dentry->d_name);
275 * this does all the hard work for inserting an inline extent into
276 * the btree. The caller should have done a btrfs_drop_extents so that
277 * no overlapping inline items exist in the btree
279 static int insert_inline_extent(struct btrfs_trans_handle *trans,
280 struct btrfs_path *path,
281 struct btrfs_inode *inode, bool extent_inserted,
282 size_t size, size_t compressed_size,
284 struct page **compressed_pages,
287 struct btrfs_root *root = inode->root;
288 struct extent_buffer *leaf;
289 struct page *page = NULL;
292 struct btrfs_file_extent_item *ei;
294 size_t cur_size = size;
297 ASSERT((compressed_size > 0 && compressed_pages) ||
298 (compressed_size == 0 && !compressed_pages));
300 if (compressed_size && compressed_pages)
301 cur_size = compressed_size;
303 if (!extent_inserted) {
304 struct btrfs_key key;
307 key.objectid = btrfs_ino(inode);
309 key.type = BTRFS_EXTENT_DATA_KEY;
311 datasize = btrfs_file_extent_calc_inline_size(cur_size);
312 ret = btrfs_insert_empty_item(trans, root, path, &key,
317 leaf = path->nodes[0];
318 ei = btrfs_item_ptr(leaf, path->slots[0],
319 struct btrfs_file_extent_item);
320 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
321 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
322 btrfs_set_file_extent_encryption(leaf, ei, 0);
323 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
324 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
325 ptr = btrfs_file_extent_inline_start(ei);
327 if (compress_type != BTRFS_COMPRESS_NONE) {
330 while (compressed_size > 0) {
331 cpage = compressed_pages[i];
332 cur_size = min_t(unsigned long, compressed_size,
335 kaddr = kmap_atomic(cpage);
336 write_extent_buffer(leaf, kaddr, ptr, cur_size);
337 kunmap_atomic(kaddr);
341 compressed_size -= cur_size;
343 btrfs_set_file_extent_compression(leaf, ei,
346 page = find_get_page(inode->vfs_inode.i_mapping, 0);
347 btrfs_set_file_extent_compression(leaf, ei, 0);
348 kaddr = kmap_atomic(page);
349 write_extent_buffer(leaf, kaddr, ptr, size);
350 kunmap_atomic(kaddr);
353 btrfs_mark_buffer_dirty(leaf);
354 btrfs_release_path(path);
357 * We align size to sectorsize for inline extents just for simplicity
360 ret = btrfs_inode_set_file_extent_range(inode, 0,
361 ALIGN(size, root->fs_info->sectorsize));
366 * We're an inline extent, so nobody can extend the file past i_size
367 * without locking a page we already have locked.
369 * We must do any i_size and inode updates before we unlock the pages.
370 * Otherwise we could end up racing with unlink.
372 i_size = i_size_read(&inode->vfs_inode);
373 if (update_i_size && size > i_size) {
374 i_size_write(&inode->vfs_inode, size);
377 inode->disk_i_size = i_size;
385 * conditionally insert an inline extent into the file. This
386 * does the checks required to make sure the data is small enough
387 * to fit as an inline extent.
389 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
390 size_t compressed_size,
392 struct page **compressed_pages,
395 struct btrfs_drop_extents_args drop_args = { 0 };
396 struct btrfs_root *root = inode->root;
397 struct btrfs_fs_info *fs_info = root->fs_info;
398 struct btrfs_trans_handle *trans;
399 u64 data_len = (compressed_size ?: size);
401 struct btrfs_path *path;
404 * We can create an inline extent if it ends at or beyond the current
405 * i_size, is no larger than a sector (decompressed), and the (possibly
406 * compressed) data fits in a leaf and the configured maximum inline
409 if (size < i_size_read(&inode->vfs_inode) ||
410 size > fs_info->sectorsize ||
411 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
412 data_len > fs_info->max_inline)
415 path = btrfs_alloc_path();
419 trans = btrfs_join_transaction(root);
421 btrfs_free_path(path);
422 return PTR_ERR(trans);
424 trans->block_rsv = &inode->block_rsv;
426 drop_args.path = path;
428 drop_args.end = fs_info->sectorsize;
429 drop_args.drop_cache = true;
430 drop_args.replace_extent = true;
431 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
432 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
434 btrfs_abort_transaction(trans, ret);
438 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
439 size, compressed_size, compress_type,
440 compressed_pages, update_i_size);
441 if (ret && ret != -ENOSPC) {
442 btrfs_abort_transaction(trans, ret);
444 } else if (ret == -ENOSPC) {
449 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
450 ret = btrfs_update_inode(trans, root, inode);
451 if (ret && ret != -ENOSPC) {
452 btrfs_abort_transaction(trans, ret);
454 } else if (ret == -ENOSPC) {
459 btrfs_set_inode_full_sync(inode);
462 * Don't forget to free the reserved space, as for inlined extent
463 * it won't count as data extent, free them directly here.
464 * And at reserve time, it's always aligned to page size, so
465 * just free one page here.
467 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
468 btrfs_free_path(path);
469 btrfs_end_transaction(trans);
473 struct async_extent {
478 unsigned long nr_pages;
480 struct list_head list;
485 struct page *locked_page;
488 blk_opf_t write_flags;
489 struct list_head extents;
490 struct cgroup_subsys_state *blkcg_css;
491 struct btrfs_work work;
492 struct async_cow *async_cow;
497 struct async_chunk chunks[];
500 static noinline int add_async_extent(struct async_chunk *cow,
501 u64 start, u64 ram_size,
504 unsigned long nr_pages,
507 struct async_extent *async_extent;
509 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
510 BUG_ON(!async_extent); /* -ENOMEM */
511 async_extent->start = start;
512 async_extent->ram_size = ram_size;
513 async_extent->compressed_size = compressed_size;
514 async_extent->pages = pages;
515 async_extent->nr_pages = nr_pages;
516 async_extent->compress_type = compress_type;
517 list_add_tail(&async_extent->list, &cow->extents);
522 * Check if the inode needs to be submitted to compression, based on mount
523 * options, defragmentation, properties or heuristics.
525 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
528 struct btrfs_fs_info *fs_info = inode->root->fs_info;
530 if (!btrfs_inode_can_compress(inode)) {
531 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
532 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
537 * Special check for subpage.
539 * We lock the full page then run each delalloc range in the page, thus
540 * for the following case, we will hit some subpage specific corner case:
543 * | |///////| |///////|
546 * In above case, both range A and range B will try to unlock the full
547 * page [0, 64K), causing the one finished later will have page
548 * unlocked already, triggering various page lock requirement BUG_ON()s.
550 * So here we add an artificial limit that subpage compression can only
551 * if the range is fully page aligned.
553 * In theory we only need to ensure the first page is fully covered, but
554 * the tailing partial page will be locked until the full compression
555 * finishes, delaying the write of other range.
557 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
558 * first to prevent any submitted async extent to unlock the full page.
559 * By this, we can ensure for subpage case that only the last async_cow
560 * will unlock the full page.
562 if (fs_info->sectorsize < PAGE_SIZE) {
563 if (!IS_ALIGNED(start, PAGE_SIZE) ||
564 !IS_ALIGNED(end + 1, PAGE_SIZE))
569 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
572 if (inode->defrag_compress)
574 /* bad compression ratios */
575 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
577 if (btrfs_test_opt(fs_info, COMPRESS) ||
578 inode->flags & BTRFS_INODE_COMPRESS ||
579 inode->prop_compress)
580 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
584 static inline void inode_should_defrag(struct btrfs_inode *inode,
585 u64 start, u64 end, u64 num_bytes, u32 small_write)
587 /* If this is a small write inside eof, kick off a defrag */
588 if (num_bytes < small_write &&
589 (start > 0 || end + 1 < inode->disk_i_size))
590 btrfs_add_inode_defrag(NULL, inode, small_write);
594 * we create compressed extents in two phases. The first
595 * phase compresses a range of pages that have already been
596 * locked (both pages and state bits are locked).
598 * This is done inside an ordered work queue, and the compression
599 * is spread across many cpus. The actual IO submission is step
600 * two, and the ordered work queue takes care of making sure that
601 * happens in the same order things were put onto the queue by
602 * writepages and friends.
604 * If this code finds it can't get good compression, it puts an
605 * entry onto the work queue to write the uncompressed bytes. This
606 * makes sure that both compressed inodes and uncompressed inodes
607 * are written in the same order that the flusher thread sent them
610 static noinline int compress_file_range(struct async_chunk *async_chunk)
612 struct inode *inode = async_chunk->inode;
613 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
614 u64 blocksize = fs_info->sectorsize;
615 u64 start = async_chunk->start;
616 u64 end = async_chunk->end;
620 struct page **pages = NULL;
621 unsigned long nr_pages;
622 unsigned long total_compressed = 0;
623 unsigned long total_in = 0;
626 int compress_type = fs_info->compress_type;
627 int compressed_extents = 0;
630 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
634 * We need to save i_size before now because it could change in between
635 * us evaluating the size and assigning it. This is because we lock and
636 * unlock the page in truncate and fallocate, and then modify the i_size
639 * The barriers are to emulate READ_ONCE, remove that once i_size_read
643 i_size = i_size_read(inode);
645 actual_end = min_t(u64, i_size, end + 1);
648 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
649 nr_pages = min_t(unsigned long, nr_pages,
650 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
653 * we don't want to send crud past the end of i_size through
654 * compression, that's just a waste of CPU time. So, if the
655 * end of the file is before the start of our current
656 * requested range of bytes, we bail out to the uncompressed
657 * cleanup code that can deal with all of this.
659 * It isn't really the fastest way to fix things, but this is a
660 * very uncommon corner.
662 if (actual_end <= start)
663 goto cleanup_and_bail_uncompressed;
665 total_compressed = actual_end - start;
668 * Skip compression for a small file range(<=blocksize) that
669 * isn't an inline extent, since it doesn't save disk space at all.
671 if (total_compressed <= blocksize &&
672 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
673 goto cleanup_and_bail_uncompressed;
676 * For subpage case, we require full page alignment for the sector
678 * Thus we must also check against @actual_end, not just @end.
680 if (blocksize < PAGE_SIZE) {
681 if (!IS_ALIGNED(start, PAGE_SIZE) ||
682 !IS_ALIGNED(round_up(actual_end, blocksize), PAGE_SIZE))
683 goto cleanup_and_bail_uncompressed;
686 total_compressed = min_t(unsigned long, total_compressed,
687 BTRFS_MAX_UNCOMPRESSED);
692 * we do compression for mount -o compress and when the
693 * inode has not been flagged as nocompress. This flag can
694 * change at any time if we discover bad compression ratios.
696 if (inode_need_compress(BTRFS_I(inode), start, end)) {
698 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
700 /* just bail out to the uncompressed code */
705 if (BTRFS_I(inode)->defrag_compress)
706 compress_type = BTRFS_I(inode)->defrag_compress;
707 else if (BTRFS_I(inode)->prop_compress)
708 compress_type = BTRFS_I(inode)->prop_compress;
711 * we need to call clear_page_dirty_for_io on each
712 * page in the range. Otherwise applications with the file
713 * mmap'd can wander in and change the page contents while
714 * we are compressing them.
716 * If the compression fails for any reason, we set the pages
717 * dirty again later on.
719 * Note that the remaining part is redirtied, the start pointer
720 * has moved, the end is the original one.
723 extent_range_clear_dirty_for_io(inode, start, end);
727 /* Compression level is applied here and only here */
728 ret = btrfs_compress_pages(
729 compress_type | (fs_info->compress_level << 4),
730 inode->i_mapping, start,
737 unsigned long offset = offset_in_page(total_compressed);
738 struct page *page = pages[nr_pages - 1];
740 /* zero the tail end of the last page, we might be
741 * sending it down to disk
744 memzero_page(page, offset, PAGE_SIZE - offset);
750 * Check cow_file_range() for why we don't even try to create inline
751 * extent for subpage case.
753 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
754 /* lets try to make an inline extent */
755 if (ret || total_in < actual_end) {
756 /* we didn't compress the entire range, try
757 * to make an uncompressed inline extent.
759 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
760 0, BTRFS_COMPRESS_NONE,
763 /* try making a compressed inline extent */
764 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
766 compress_type, pages,
770 unsigned long clear_flags = EXTENT_DELALLOC |
771 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
772 EXTENT_DO_ACCOUNTING;
773 unsigned long page_error_op;
775 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
778 * inline extent creation worked or returned error,
779 * we don't need to create any more async work items.
780 * Unlock and free up our temp pages.
782 * We use DO_ACCOUNTING here because we need the
783 * delalloc_release_metadata to be done _after_ we drop
784 * our outstanding extent for clearing delalloc for this
787 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
791 PAGE_START_WRITEBACK |
796 * Ensure we only free the compressed pages if we have
797 * them allocated, as we can still reach here with
798 * inode_need_compress() == false.
801 for (i = 0; i < nr_pages; i++) {
802 WARN_ON(pages[i]->mapping);
813 * we aren't doing an inline extent round the compressed size
814 * up to a block size boundary so the allocator does sane
817 total_compressed = ALIGN(total_compressed, blocksize);
820 * one last check to make sure the compression is really a
821 * win, compare the page count read with the blocks on disk,
822 * compression must free at least one sector size
824 total_in = round_up(total_in, fs_info->sectorsize);
825 if (total_compressed + blocksize <= total_in) {
826 compressed_extents++;
829 * The async work queues will take care of doing actual
830 * allocation on disk for these compressed pages, and
831 * will submit them to the elevator.
833 add_async_extent(async_chunk, start, total_in,
834 total_compressed, pages, nr_pages,
837 if (start + total_in < end) {
843 return compressed_extents;
848 * the compression code ran but failed to make things smaller,
849 * free any pages it allocated and our page pointer array
851 for (i = 0; i < nr_pages; i++) {
852 WARN_ON(pages[i]->mapping);
857 total_compressed = 0;
860 /* flag the file so we don't compress in the future */
861 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
862 !(BTRFS_I(inode)->prop_compress)) {
863 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
866 cleanup_and_bail_uncompressed:
868 * No compression, but we still need to write the pages in the file
869 * we've been given so far. redirty the locked page if it corresponds
870 * to our extent and set things up for the async work queue to run
871 * cow_file_range to do the normal delalloc dance.
873 if (async_chunk->locked_page &&
874 (page_offset(async_chunk->locked_page) >= start &&
875 page_offset(async_chunk->locked_page)) <= end) {
876 __set_page_dirty_nobuffers(async_chunk->locked_page);
877 /* unlocked later on in the async handlers */
881 extent_range_redirty_for_io(inode, start, end);
882 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
883 BTRFS_COMPRESS_NONE);
884 compressed_extents++;
886 return compressed_extents;
889 static void free_async_extent_pages(struct async_extent *async_extent)
893 if (!async_extent->pages)
896 for (i = 0; i < async_extent->nr_pages; i++) {
897 WARN_ON(async_extent->pages[i]->mapping);
898 put_page(async_extent->pages[i]);
900 kfree(async_extent->pages);
901 async_extent->nr_pages = 0;
902 async_extent->pages = NULL;
905 static int submit_uncompressed_range(struct btrfs_inode *inode,
906 struct async_extent *async_extent,
907 struct page *locked_page)
909 u64 start = async_extent->start;
910 u64 end = async_extent->start + async_extent->ram_size - 1;
911 unsigned long nr_written = 0;
912 int page_started = 0;
916 * Call cow_file_range() to run the delalloc range directly, since we
917 * won't go to NOCOW or async path again.
919 * Also we call cow_file_range() with @unlock_page == 0, so that we
920 * can directly submit them without interruption.
922 ret = cow_file_range(inode, locked_page, start, end, &page_started,
924 /* Inline extent inserted, page gets unlocked and everything is done */
931 unlock_page(locked_page);
935 ret = extent_write_locked_range(&inode->vfs_inode, start, end);
936 /* All pages will be unlocked, including @locked_page */
942 static int submit_one_async_extent(struct btrfs_inode *inode,
943 struct async_chunk *async_chunk,
944 struct async_extent *async_extent,
947 struct extent_io_tree *io_tree = &inode->io_tree;
948 struct btrfs_root *root = inode->root;
949 struct btrfs_fs_info *fs_info = root->fs_info;
950 struct btrfs_key ins;
951 struct page *locked_page = NULL;
952 struct extent_map *em;
954 u64 start = async_extent->start;
955 u64 end = async_extent->start + async_extent->ram_size - 1;
958 * If async_chunk->locked_page is in the async_extent range, we need to
961 if (async_chunk->locked_page) {
962 u64 locked_page_start = page_offset(async_chunk->locked_page);
963 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
965 if (!(start >= locked_page_end || end <= locked_page_start))
966 locked_page = async_chunk->locked_page;
968 lock_extent(io_tree, start, end);
970 /* We have fall back to uncompressed write */
971 if (!async_extent->pages)
972 return submit_uncompressed_range(inode, async_extent, locked_page);
974 ret = btrfs_reserve_extent(root, async_extent->ram_size,
975 async_extent->compressed_size,
976 async_extent->compressed_size,
977 0, *alloc_hint, &ins, 1, 1);
979 free_async_extent_pages(async_extent);
981 * Here we used to try again by going back to non-compressed
982 * path for ENOSPC. But we can't reserve space even for
983 * compressed size, how could it work for uncompressed size
984 * which requires larger size? So here we directly go error
990 /* Here we're doing allocation and writeback of the compressed pages */
991 em = create_io_em(inode, start,
992 async_extent->ram_size, /* len */
993 start, /* orig_start */
994 ins.objectid, /* block_start */
995 ins.offset, /* block_len */
996 ins.offset, /* orig_block_len */
997 async_extent->ram_size, /* ram_bytes */
998 async_extent->compress_type,
999 BTRFS_ORDERED_COMPRESSED);
1002 goto out_free_reserve;
1004 free_extent_map(em);
1006 ret = btrfs_add_ordered_extent(inode, start, /* file_offset */
1007 async_extent->ram_size, /* num_bytes */
1008 async_extent->ram_size, /* ram_bytes */
1009 ins.objectid, /* disk_bytenr */
1010 ins.offset, /* disk_num_bytes */
1012 1 << BTRFS_ORDERED_COMPRESSED,
1013 async_extent->compress_type);
1015 btrfs_drop_extent_cache(inode, start, end, 0);
1016 goto out_free_reserve;
1018 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1020 /* Clear dirty, set writeback and unlock the pages. */
1021 extent_clear_unlock_delalloc(inode, start, end,
1022 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1023 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1024 if (btrfs_submit_compressed_write(inode, start, /* file_offset */
1025 async_extent->ram_size, /* num_bytes */
1026 ins.objectid, /* disk_bytenr */
1027 ins.offset, /* compressed_len */
1028 async_extent->pages, /* compressed_pages */
1029 async_extent->nr_pages,
1030 async_chunk->write_flags,
1031 async_chunk->blkcg_css, true)) {
1032 const u64 start = async_extent->start;
1033 const u64 end = start + async_extent->ram_size - 1;
1035 btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0);
1037 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
1038 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1039 free_async_extent_pages(async_extent);
1041 *alloc_hint = ins.objectid + ins.offset;
1042 kfree(async_extent);
1046 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1047 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1049 extent_clear_unlock_delalloc(inode, start, end,
1050 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1051 EXTENT_DELALLOC_NEW |
1052 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1053 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1054 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1055 free_async_extent_pages(async_extent);
1056 kfree(async_extent);
1061 * Phase two of compressed writeback. This is the ordered portion of the code,
1062 * which only gets called in the order the work was queued. We walk all the
1063 * async extents created by compress_file_range and send them down to the disk.
1065 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1067 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
1068 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1069 struct async_extent *async_extent;
1073 while (!list_empty(&async_chunk->extents)) {
1077 async_extent = list_entry(async_chunk->extents.next,
1078 struct async_extent, list);
1079 list_del(&async_extent->list);
1080 extent_start = async_extent->start;
1081 ram_size = async_extent->ram_size;
1083 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1085 btrfs_debug(fs_info,
1086 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1087 inode->root->root_key.objectid,
1088 btrfs_ino(inode), extent_start, ram_size, ret);
1092 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1095 struct extent_map_tree *em_tree = &inode->extent_tree;
1096 struct extent_map *em;
1099 read_lock(&em_tree->lock);
1100 em = search_extent_mapping(em_tree, start, num_bytes);
1103 * if block start isn't an actual block number then find the
1104 * first block in this inode and use that as a hint. If that
1105 * block is also bogus then just don't worry about it.
1107 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1108 free_extent_map(em);
1109 em = search_extent_mapping(em_tree, 0, 0);
1110 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1111 alloc_hint = em->block_start;
1113 free_extent_map(em);
1115 alloc_hint = em->block_start;
1116 free_extent_map(em);
1119 read_unlock(&em_tree->lock);
1125 * when extent_io.c finds a delayed allocation range in the file,
1126 * the call backs end up in this code. The basic idea is to
1127 * allocate extents on disk for the range, and create ordered data structs
1128 * in ram to track those extents.
1130 * locked_page is the page that writepage had locked already. We use
1131 * it to make sure we don't do extra locks or unlocks.
1133 * *page_started is set to one if we unlock locked_page and do everything
1134 * required to start IO on it. It may be clean and already done with
1135 * IO when we return.
1137 static noinline int cow_file_range(struct btrfs_inode *inode,
1138 struct page *locked_page,
1139 u64 start, u64 end, int *page_started,
1140 unsigned long *nr_written, int unlock)
1142 struct btrfs_root *root = inode->root;
1143 struct btrfs_fs_info *fs_info = root->fs_info;
1146 unsigned long ram_size;
1147 u64 cur_alloc_size = 0;
1149 u64 blocksize = fs_info->sectorsize;
1150 struct btrfs_key ins;
1151 struct extent_map *em;
1152 unsigned clear_bits;
1153 unsigned long page_ops;
1154 bool extent_reserved = false;
1157 if (btrfs_is_free_space_inode(inode)) {
1162 num_bytes = ALIGN(end - start + 1, blocksize);
1163 num_bytes = max(blocksize, num_bytes);
1164 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1166 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1169 * Due to the page size limit, for subpage we can only trigger the
1170 * writeback for the dirty sectors of page, that means data writeback
1171 * is doing more writeback than what we want.
1173 * This is especially unexpected for some call sites like fallocate,
1174 * where we only increase i_size after everything is done.
1175 * This means we can trigger inline extent even if we didn't want to.
1176 * So here we skip inline extent creation completely.
1178 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1179 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1182 /* lets try to make an inline extent */
1183 ret = cow_file_range_inline(inode, actual_end, 0,
1184 BTRFS_COMPRESS_NONE, NULL, false);
1187 * We use DO_ACCOUNTING here because we need the
1188 * delalloc_release_metadata to be run _after_ we drop
1189 * our outstanding extent for clearing delalloc for this
1192 extent_clear_unlock_delalloc(inode, start, end,
1194 EXTENT_LOCKED | EXTENT_DELALLOC |
1195 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1196 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1197 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1198 *nr_written = *nr_written +
1199 (end - start + PAGE_SIZE) / PAGE_SIZE;
1202 * locked_page is locked by the caller of
1203 * writepage_delalloc(), not locked by
1204 * __process_pages_contig().
1206 * We can't let __process_pages_contig() to unlock it,
1207 * as it doesn't have any subpage::writers recorded.
1209 * Here we manually unlock the page, since the caller
1210 * can't use page_started to determine if it's an
1211 * inline extent or a compressed extent.
1213 unlock_page(locked_page);
1215 } else if (ret < 0) {
1220 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1221 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1224 * Relocation relies on the relocated extents to have exactly the same
1225 * size as the original extents. Normally writeback for relocation data
1226 * extents follows a NOCOW path because relocation preallocates the
1227 * extents. However, due to an operation such as scrub turning a block
1228 * group to RO mode, it may fallback to COW mode, so we must make sure
1229 * an extent allocated during COW has exactly the requested size and can
1230 * not be split into smaller extents, otherwise relocation breaks and
1231 * fails during the stage where it updates the bytenr of file extent
1234 if (btrfs_is_data_reloc_root(root))
1235 min_alloc_size = num_bytes;
1237 min_alloc_size = fs_info->sectorsize;
1239 while (num_bytes > 0) {
1240 cur_alloc_size = num_bytes;
1241 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1242 min_alloc_size, 0, alloc_hint,
1246 cur_alloc_size = ins.offset;
1247 extent_reserved = true;
1249 ram_size = ins.offset;
1250 em = create_io_em(inode, start, ins.offset, /* len */
1251 start, /* orig_start */
1252 ins.objectid, /* block_start */
1253 ins.offset, /* block_len */
1254 ins.offset, /* orig_block_len */
1255 ram_size, /* ram_bytes */
1256 BTRFS_COMPRESS_NONE, /* compress_type */
1257 BTRFS_ORDERED_REGULAR /* type */);
1262 free_extent_map(em);
1264 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1265 ins.objectid, cur_alloc_size, 0,
1266 1 << BTRFS_ORDERED_REGULAR,
1267 BTRFS_COMPRESS_NONE);
1269 goto out_drop_extent_cache;
1271 if (btrfs_is_data_reloc_root(root)) {
1272 ret = btrfs_reloc_clone_csums(inode, start,
1275 * Only drop cache here, and process as normal.
1277 * We must not allow extent_clear_unlock_delalloc()
1278 * at out_unlock label to free meta of this ordered
1279 * extent, as its meta should be freed by
1280 * btrfs_finish_ordered_io().
1282 * So we must continue until @start is increased to
1283 * skip current ordered extent.
1286 btrfs_drop_extent_cache(inode, start,
1287 start + ram_size - 1, 0);
1290 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1293 * We're not doing compressed IO, don't unlock the first page
1294 * (which the caller expects to stay locked), don't clear any
1295 * dirty bits and don't set any writeback bits
1297 * Do set the Ordered (Private2) bit so we know this page was
1298 * properly setup for writepage.
1300 page_ops = unlock ? PAGE_UNLOCK : 0;
1301 page_ops |= PAGE_SET_ORDERED;
1303 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1305 EXTENT_LOCKED | EXTENT_DELALLOC,
1307 if (num_bytes < cur_alloc_size)
1310 num_bytes -= cur_alloc_size;
1311 alloc_hint = ins.objectid + ins.offset;
1312 start += cur_alloc_size;
1313 extent_reserved = false;
1316 * btrfs_reloc_clone_csums() error, since start is increased
1317 * extent_clear_unlock_delalloc() at out_unlock label won't
1318 * free metadata of current ordered extent, we're OK to exit.
1326 out_drop_extent_cache:
1327 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1329 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1330 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1332 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1333 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1334 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1336 * If we reserved an extent for our delalloc range (or a subrange) and
1337 * failed to create the respective ordered extent, then it means that
1338 * when we reserved the extent we decremented the extent's size from
1339 * the data space_info's bytes_may_use counter and incremented the
1340 * space_info's bytes_reserved counter by the same amount. We must make
1341 * sure extent_clear_unlock_delalloc() does not try to decrement again
1342 * the data space_info's bytes_may_use counter, therefore we do not pass
1343 * it the flag EXTENT_CLEAR_DATA_RESV.
1345 if (extent_reserved) {
1346 extent_clear_unlock_delalloc(inode, start,
1347 start + cur_alloc_size - 1,
1351 start += cur_alloc_size;
1355 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1356 clear_bits | EXTENT_CLEAR_DATA_RESV,
1362 * work queue call back to started compression on a file and pages
1364 static noinline void async_cow_start(struct btrfs_work *work)
1366 struct async_chunk *async_chunk;
1367 int compressed_extents;
1369 async_chunk = container_of(work, struct async_chunk, work);
1371 compressed_extents = compress_file_range(async_chunk);
1372 if (compressed_extents == 0) {
1373 btrfs_add_delayed_iput(async_chunk->inode);
1374 async_chunk->inode = NULL;
1379 * work queue call back to submit previously compressed pages
1381 static noinline void async_cow_submit(struct btrfs_work *work)
1383 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1385 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1386 unsigned long nr_pages;
1388 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1392 * ->inode could be NULL if async_chunk_start has failed to compress,
1393 * in which case we don't have anything to submit, yet we need to
1394 * always adjust ->async_delalloc_pages as its paired with the init
1395 * happening in cow_file_range_async
1397 if (async_chunk->inode)
1398 submit_compressed_extents(async_chunk);
1400 /* atomic_sub_return implies a barrier */
1401 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1403 cond_wake_up_nomb(&fs_info->async_submit_wait);
1406 static noinline void async_cow_free(struct btrfs_work *work)
1408 struct async_chunk *async_chunk;
1409 struct async_cow *async_cow;
1411 async_chunk = container_of(work, struct async_chunk, work);
1412 if (async_chunk->inode)
1413 btrfs_add_delayed_iput(async_chunk->inode);
1414 if (async_chunk->blkcg_css)
1415 css_put(async_chunk->blkcg_css);
1417 async_cow = async_chunk->async_cow;
1418 if (atomic_dec_and_test(&async_cow->num_chunks))
1422 static int cow_file_range_async(struct btrfs_inode *inode,
1423 struct writeback_control *wbc,
1424 struct page *locked_page,
1425 u64 start, u64 end, int *page_started,
1426 unsigned long *nr_written)
1428 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1429 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1430 struct async_cow *ctx;
1431 struct async_chunk *async_chunk;
1432 unsigned long nr_pages;
1434 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1436 bool should_compress;
1438 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1440 unlock_extent(&inode->io_tree, start, end);
1442 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1443 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1445 should_compress = false;
1447 should_compress = true;
1450 nofs_flag = memalloc_nofs_save();
1451 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1452 memalloc_nofs_restore(nofs_flag);
1455 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1456 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1457 EXTENT_DO_ACCOUNTING;
1458 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1459 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1461 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1462 clear_bits, page_ops);
1466 async_chunk = ctx->chunks;
1467 atomic_set(&ctx->num_chunks, num_chunks);
1469 for (i = 0; i < num_chunks; i++) {
1470 if (should_compress)
1471 cur_end = min(end, start + SZ_512K - 1);
1476 * igrab is called higher up in the call chain, take only the
1477 * lightweight reference for the callback lifetime
1479 ihold(&inode->vfs_inode);
1480 async_chunk[i].async_cow = ctx;
1481 async_chunk[i].inode = &inode->vfs_inode;
1482 async_chunk[i].start = start;
1483 async_chunk[i].end = cur_end;
1484 async_chunk[i].write_flags = write_flags;
1485 INIT_LIST_HEAD(&async_chunk[i].extents);
1488 * The locked_page comes all the way from writepage and its
1489 * the original page we were actually given. As we spread
1490 * this large delalloc region across multiple async_chunk
1491 * structs, only the first struct needs a pointer to locked_page
1493 * This way we don't need racey decisions about who is supposed
1498 * Depending on the compressibility, the pages might or
1499 * might not go through async. We want all of them to
1500 * be accounted against wbc once. Let's do it here
1501 * before the paths diverge. wbc accounting is used
1502 * only for foreign writeback detection and doesn't
1503 * need full accuracy. Just account the whole thing
1504 * against the first page.
1506 wbc_account_cgroup_owner(wbc, locked_page,
1508 async_chunk[i].locked_page = locked_page;
1511 async_chunk[i].locked_page = NULL;
1514 if (blkcg_css != blkcg_root_css) {
1516 async_chunk[i].blkcg_css = blkcg_css;
1518 async_chunk[i].blkcg_css = NULL;
1521 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1522 async_cow_submit, async_cow_free);
1524 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1525 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1527 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1529 *nr_written += nr_pages;
1530 start = cur_end + 1;
1536 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1537 struct page *locked_page, u64 start,
1538 u64 end, int *page_started,
1539 unsigned long *nr_written)
1543 ret = cow_file_range(inode, locked_page, start, end, page_started,
1551 __set_page_dirty_nobuffers(locked_page);
1552 account_page_redirty(locked_page);
1553 extent_write_locked_range(&inode->vfs_inode, start, end);
1559 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1560 u64 bytenr, u64 num_bytes)
1562 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1563 struct btrfs_ordered_sum *sums;
1567 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1568 bytenr + num_bytes - 1, &list, 0);
1569 if (ret == 0 && list_empty(&list))
1572 while (!list_empty(&list)) {
1573 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1574 list_del(&sums->list);
1582 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1583 const u64 start, const u64 end,
1584 int *page_started, unsigned long *nr_written)
1586 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1587 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1588 const u64 range_bytes = end + 1 - start;
1589 struct extent_io_tree *io_tree = &inode->io_tree;
1590 u64 range_start = start;
1594 * If EXTENT_NORESERVE is set it means that when the buffered write was
1595 * made we had not enough available data space and therefore we did not
1596 * reserve data space for it, since we though we could do NOCOW for the
1597 * respective file range (either there is prealloc extent or the inode
1598 * has the NOCOW bit set).
1600 * However when we need to fallback to COW mode (because for example the
1601 * block group for the corresponding extent was turned to RO mode by a
1602 * scrub or relocation) we need to do the following:
1604 * 1) We increment the bytes_may_use counter of the data space info.
1605 * If COW succeeds, it allocates a new data extent and after doing
1606 * that it decrements the space info's bytes_may_use counter and
1607 * increments its bytes_reserved counter by the same amount (we do
1608 * this at btrfs_add_reserved_bytes()). So we need to increment the
1609 * bytes_may_use counter to compensate (when space is reserved at
1610 * buffered write time, the bytes_may_use counter is incremented);
1612 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1613 * that if the COW path fails for any reason, it decrements (through
1614 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1615 * data space info, which we incremented in the step above.
1617 * If we need to fallback to cow and the inode corresponds to a free
1618 * space cache inode or an inode of the data relocation tree, we must
1619 * also increment bytes_may_use of the data space_info for the same
1620 * reason. Space caches and relocated data extents always get a prealloc
1621 * extent for them, however scrub or balance may have set the block
1622 * group that contains that extent to RO mode and therefore force COW
1623 * when starting writeback.
1625 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1626 EXTENT_NORESERVE, 0);
1627 if (count > 0 || is_space_ino || is_reloc_ino) {
1629 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1630 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1632 if (is_space_ino || is_reloc_ino)
1633 bytes = range_bytes;
1635 spin_lock(&sinfo->lock);
1636 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1637 spin_unlock(&sinfo->lock);
1640 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1644 return cow_file_range(inode, locked_page, start, end, page_started,
1648 struct can_nocow_file_extent_args {
1651 /* Start file offset of the range we want to NOCOW. */
1653 /* End file offset (inclusive) of the range we want to NOCOW. */
1655 bool writeback_path;
1658 * Free the path passed to can_nocow_file_extent() once it's not needed
1663 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1668 /* Number of bytes that can be written to in NOCOW mode. */
1673 * Check if we can NOCOW the file extent that the path points to.
1674 * This function may return with the path released, so the caller should check
1675 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1677 * Returns: < 0 on error
1678 * 0 if we can not NOCOW
1681 static int can_nocow_file_extent(struct btrfs_path *path,
1682 struct btrfs_key *key,
1683 struct btrfs_inode *inode,
1684 struct can_nocow_file_extent_args *args)
1686 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1687 struct extent_buffer *leaf = path->nodes[0];
1688 struct btrfs_root *root = inode->root;
1689 struct btrfs_file_extent_item *fi;
1695 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1696 extent_type = btrfs_file_extent_type(leaf, fi);
1698 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1701 /* Can't access these fields unless we know it's not an inline extent. */
1702 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1703 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1704 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1706 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1707 extent_type == BTRFS_FILE_EXTENT_REG)
1711 * If the extent was created before the generation where the last snapshot
1712 * for its subvolume was created, then this implies the extent is shared,
1713 * hence we must COW.
1715 if (!args->strict &&
1716 btrfs_file_extent_generation(leaf, fi) <=
1717 btrfs_root_last_snapshot(&root->root_item))
1720 /* An explicit hole, must COW. */
1721 if (args->disk_bytenr == 0)
1724 /* Compressed/encrypted/encoded extents must be COWed. */
1725 if (btrfs_file_extent_compression(leaf, fi) ||
1726 btrfs_file_extent_encryption(leaf, fi) ||
1727 btrfs_file_extent_other_encoding(leaf, fi))
1730 extent_end = btrfs_file_extent_end(path);
1733 * The following checks can be expensive, as they need to take other
1734 * locks and do btree or rbtree searches, so release the path to avoid
1735 * blocking other tasks for too long.
1737 btrfs_release_path(path);
1739 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1740 key->offset - args->extent_offset,
1741 args->disk_bytenr, false, path);
1742 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1746 if (args->free_path) {
1748 * We don't need the path anymore, plus through the
1749 * csum_exist_in_range() call below we will end up allocating
1750 * another path. So free the path to avoid unnecessary extra
1753 btrfs_free_path(path);
1757 /* If there are pending snapshots for this root, we must COW. */
1758 if (args->writeback_path && !is_freespace_inode &&
1759 atomic_read(&root->snapshot_force_cow))
1762 args->disk_bytenr += args->extent_offset;
1763 args->disk_bytenr += args->start - key->offset;
1764 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1767 * Force COW if csums exist in the range. This ensures that csums for a
1768 * given extent are either valid or do not exist.
1770 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes);
1771 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1777 if (args->free_path && path)
1778 btrfs_free_path(path);
1780 return ret < 0 ? ret : can_nocow;
1784 * when nowcow writeback call back. This checks for snapshots or COW copies
1785 * of the extents that exist in the file, and COWs the file as required.
1787 * If no cow copies or snapshots exist, we write directly to the existing
1790 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1791 struct page *locked_page,
1792 const u64 start, const u64 end,
1794 unsigned long *nr_written)
1796 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1797 struct btrfs_root *root = inode->root;
1798 struct btrfs_path *path;
1799 u64 cow_start = (u64)-1;
1800 u64 cur_offset = start;
1802 bool check_prev = true;
1803 u64 ino = btrfs_ino(inode);
1804 struct btrfs_block_group *bg;
1806 struct can_nocow_file_extent_args nocow_args = { 0 };
1808 path = btrfs_alloc_path();
1810 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1811 EXTENT_LOCKED | EXTENT_DELALLOC |
1812 EXTENT_DO_ACCOUNTING |
1813 EXTENT_DEFRAG, PAGE_UNLOCK |
1814 PAGE_START_WRITEBACK |
1815 PAGE_END_WRITEBACK);
1819 nocow_args.end = end;
1820 nocow_args.writeback_path = true;
1823 struct btrfs_key found_key;
1824 struct btrfs_file_extent_item *fi;
1825 struct extent_buffer *leaf;
1833 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1839 * If there is no extent for our range when doing the initial
1840 * search, then go back to the previous slot as it will be the
1841 * one containing the search offset
1843 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1844 leaf = path->nodes[0];
1845 btrfs_item_key_to_cpu(leaf, &found_key,
1846 path->slots[0] - 1);
1847 if (found_key.objectid == ino &&
1848 found_key.type == BTRFS_EXTENT_DATA_KEY)
1853 /* Go to next leaf if we have exhausted the current one */
1854 leaf = path->nodes[0];
1855 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1856 ret = btrfs_next_leaf(root, path);
1858 if (cow_start != (u64)-1)
1859 cur_offset = cow_start;
1864 leaf = path->nodes[0];
1867 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1869 /* Didn't find anything for our INO */
1870 if (found_key.objectid > ino)
1873 * Keep searching until we find an EXTENT_ITEM or there are no
1874 * more extents for this inode
1876 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1877 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1882 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1883 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1884 found_key.offset > end)
1888 * If the found extent starts after requested offset, then
1889 * adjust extent_end to be right before this extent begins
1891 if (found_key.offset > cur_offset) {
1892 extent_end = found_key.offset;
1898 * Found extent which begins before our range and potentially
1901 fi = btrfs_item_ptr(leaf, path->slots[0],
1902 struct btrfs_file_extent_item);
1903 extent_type = btrfs_file_extent_type(leaf, fi);
1904 /* If this is triggered then we have a memory corruption. */
1905 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
1906 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
1910 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1911 extent_end = btrfs_file_extent_end(path);
1914 * If the extent we got ends before our current offset, skip to
1917 if (extent_end <= cur_offset) {
1922 nocow_args.start = cur_offset;
1923 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
1925 if (cow_start != (u64)-1)
1926 cur_offset = cow_start;
1928 } else if (ret == 0) {
1933 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
1938 * If nocow is false then record the beginning of the range
1939 * that needs to be COWed
1942 if (cow_start == (u64)-1)
1943 cow_start = cur_offset;
1944 cur_offset = extent_end;
1945 if (cur_offset > end)
1947 if (!path->nodes[0])
1954 * COW range from cow_start to found_key.offset - 1. As the key
1955 * will contain the beginning of the first extent that can be
1956 * NOCOW, following one which needs to be COW'ed
1958 if (cow_start != (u64)-1) {
1959 ret = fallback_to_cow(inode, locked_page,
1960 cow_start, found_key.offset - 1,
1961 page_started, nr_written);
1964 cow_start = (u64)-1;
1967 nocow_end = cur_offset + nocow_args.num_bytes - 1;
1969 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1970 u64 orig_start = found_key.offset - nocow_args.extent_offset;
1971 struct extent_map *em;
1973 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
1975 nocow_args.disk_bytenr, /* block_start */
1976 nocow_args.num_bytes, /* block_len */
1977 nocow_args.disk_num_bytes, /* orig_block_len */
1978 ram_bytes, BTRFS_COMPRESS_NONE,
1979 BTRFS_ORDERED_PREALLOC);
1984 free_extent_map(em);
1985 ret = btrfs_add_ordered_extent(inode,
1986 cur_offset, nocow_args.num_bytes,
1987 nocow_args.num_bytes,
1988 nocow_args.disk_bytenr,
1989 nocow_args.num_bytes, 0,
1990 1 << BTRFS_ORDERED_PREALLOC,
1991 BTRFS_COMPRESS_NONE);
1993 btrfs_drop_extent_cache(inode, cur_offset,
1998 ret = btrfs_add_ordered_extent(inode, cur_offset,
1999 nocow_args.num_bytes,
2000 nocow_args.num_bytes,
2001 nocow_args.disk_bytenr,
2002 nocow_args.num_bytes,
2004 1 << BTRFS_ORDERED_NOCOW,
2005 BTRFS_COMPRESS_NONE);
2011 btrfs_dec_nocow_writers(bg);
2015 if (btrfs_is_data_reloc_root(root))
2017 * Error handled later, as we must prevent
2018 * extent_clear_unlock_delalloc() in error handler
2019 * from freeing metadata of created ordered extent.
2021 ret = btrfs_reloc_clone_csums(inode, cur_offset,
2022 nocow_args.num_bytes);
2024 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2025 locked_page, EXTENT_LOCKED |
2027 EXTENT_CLEAR_DATA_RESV,
2028 PAGE_UNLOCK | PAGE_SET_ORDERED);
2030 cur_offset = extent_end;
2033 * btrfs_reloc_clone_csums() error, now we're OK to call error
2034 * handler, as metadata for created ordered extent will only
2035 * be freed by btrfs_finish_ordered_io().
2039 if (cur_offset > end)
2042 btrfs_release_path(path);
2044 if (cur_offset <= end && cow_start == (u64)-1)
2045 cow_start = cur_offset;
2047 if (cow_start != (u64)-1) {
2049 ret = fallback_to_cow(inode, locked_page, cow_start, end,
2050 page_started, nr_written);
2057 btrfs_dec_nocow_writers(bg);
2059 if (ret && cur_offset < end)
2060 extent_clear_unlock_delalloc(inode, cur_offset, end,
2061 locked_page, EXTENT_LOCKED |
2062 EXTENT_DELALLOC | EXTENT_DEFRAG |
2063 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2064 PAGE_START_WRITEBACK |
2065 PAGE_END_WRITEBACK);
2066 btrfs_free_path(path);
2070 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2072 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2073 if (inode->defrag_bytes &&
2074 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2083 * Function to process delayed allocation (create CoW) for ranges which are
2084 * being touched for the first time.
2086 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2087 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2088 struct writeback_control *wbc)
2091 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2094 * The range must cover part of the @locked_page, or the returned
2095 * @page_started can confuse the caller.
2097 ASSERT(!(end <= page_offset(locked_page) ||
2098 start >= page_offset(locked_page) + PAGE_SIZE));
2100 if (should_nocow(inode, start, end)) {
2102 * Normally on a zoned device we're only doing COW writes, but
2103 * in case of relocation on a zoned filesystem we have taken
2104 * precaution, that we're only writing sequentially. It's safe
2105 * to use run_delalloc_nocow() here, like for regular
2106 * preallocated inodes.
2108 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2109 ret = run_delalloc_nocow(inode, locked_page, start, end,
2110 page_started, nr_written);
2111 } else if (!btrfs_inode_can_compress(inode) ||
2112 !inode_need_compress(inode, start, end)) {
2114 ret = run_delalloc_zoned(inode, locked_page, start, end,
2115 page_started, nr_written);
2117 ret = cow_file_range(inode, locked_page, start, end,
2118 page_started, nr_written, 1);
2120 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2121 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2122 page_started, nr_written);
2126 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2131 void btrfs_split_delalloc_extent(struct inode *inode,
2132 struct extent_state *orig, u64 split)
2136 /* not delalloc, ignore it */
2137 if (!(orig->state & EXTENT_DELALLOC))
2140 size = orig->end - orig->start + 1;
2141 if (size > BTRFS_MAX_EXTENT_SIZE) {
2146 * See the explanation in btrfs_merge_delalloc_extent, the same
2147 * applies here, just in reverse.
2149 new_size = orig->end - split + 1;
2150 num_extents = count_max_extents(new_size);
2151 new_size = split - orig->start;
2152 num_extents += count_max_extents(new_size);
2153 if (count_max_extents(size) >= num_extents)
2157 spin_lock(&BTRFS_I(inode)->lock);
2158 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2159 spin_unlock(&BTRFS_I(inode)->lock);
2163 * Handle merged delayed allocation extents so we can keep track of new extents
2164 * that are just merged onto old extents, such as when we are doing sequential
2165 * writes, so we can properly account for the metadata space we'll need.
2167 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2168 struct extent_state *other)
2170 u64 new_size, old_size;
2173 /* not delalloc, ignore it */
2174 if (!(other->state & EXTENT_DELALLOC))
2177 if (new->start > other->start)
2178 new_size = new->end - other->start + 1;
2180 new_size = other->end - new->start + 1;
2182 /* we're not bigger than the max, unreserve the space and go */
2183 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2184 spin_lock(&BTRFS_I(inode)->lock);
2185 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2186 spin_unlock(&BTRFS_I(inode)->lock);
2191 * We have to add up either side to figure out how many extents were
2192 * accounted for before we merged into one big extent. If the number of
2193 * extents we accounted for is <= the amount we need for the new range
2194 * then we can return, otherwise drop. Think of it like this
2198 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2199 * need 2 outstanding extents, on one side we have 1 and the other side
2200 * we have 1 so they are == and we can return. But in this case
2202 * [MAX_SIZE+4k][MAX_SIZE+4k]
2204 * Each range on their own accounts for 2 extents, but merged together
2205 * they are only 3 extents worth of accounting, so we need to drop in
2208 old_size = other->end - other->start + 1;
2209 num_extents = count_max_extents(old_size);
2210 old_size = new->end - new->start + 1;
2211 num_extents += count_max_extents(old_size);
2212 if (count_max_extents(new_size) >= num_extents)
2215 spin_lock(&BTRFS_I(inode)->lock);
2216 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2217 spin_unlock(&BTRFS_I(inode)->lock);
2220 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2221 struct inode *inode)
2223 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2225 spin_lock(&root->delalloc_lock);
2226 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2227 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2228 &root->delalloc_inodes);
2229 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2230 &BTRFS_I(inode)->runtime_flags);
2231 root->nr_delalloc_inodes++;
2232 if (root->nr_delalloc_inodes == 1) {
2233 spin_lock(&fs_info->delalloc_root_lock);
2234 BUG_ON(!list_empty(&root->delalloc_root));
2235 list_add_tail(&root->delalloc_root,
2236 &fs_info->delalloc_roots);
2237 spin_unlock(&fs_info->delalloc_root_lock);
2240 spin_unlock(&root->delalloc_lock);
2244 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2245 struct btrfs_inode *inode)
2247 struct btrfs_fs_info *fs_info = root->fs_info;
2249 if (!list_empty(&inode->delalloc_inodes)) {
2250 list_del_init(&inode->delalloc_inodes);
2251 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2252 &inode->runtime_flags);
2253 root->nr_delalloc_inodes--;
2254 if (!root->nr_delalloc_inodes) {
2255 ASSERT(list_empty(&root->delalloc_inodes));
2256 spin_lock(&fs_info->delalloc_root_lock);
2257 BUG_ON(list_empty(&root->delalloc_root));
2258 list_del_init(&root->delalloc_root);
2259 spin_unlock(&fs_info->delalloc_root_lock);
2264 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2265 struct btrfs_inode *inode)
2267 spin_lock(&root->delalloc_lock);
2268 __btrfs_del_delalloc_inode(root, inode);
2269 spin_unlock(&root->delalloc_lock);
2273 * Properly track delayed allocation bytes in the inode and to maintain the
2274 * list of inodes that have pending delalloc work to be done.
2276 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2279 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2281 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2284 * set_bit and clear bit hooks normally require _irqsave/restore
2285 * but in this case, we are only testing for the DELALLOC
2286 * bit, which is only set or cleared with irqs on
2288 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2289 struct btrfs_root *root = BTRFS_I(inode)->root;
2290 u64 len = state->end + 1 - state->start;
2291 u32 num_extents = count_max_extents(len);
2292 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2294 spin_lock(&BTRFS_I(inode)->lock);
2295 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2296 spin_unlock(&BTRFS_I(inode)->lock);
2298 /* For sanity tests */
2299 if (btrfs_is_testing(fs_info))
2302 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2303 fs_info->delalloc_batch);
2304 spin_lock(&BTRFS_I(inode)->lock);
2305 BTRFS_I(inode)->delalloc_bytes += len;
2306 if (*bits & EXTENT_DEFRAG)
2307 BTRFS_I(inode)->defrag_bytes += len;
2308 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2309 &BTRFS_I(inode)->runtime_flags))
2310 btrfs_add_delalloc_inodes(root, inode);
2311 spin_unlock(&BTRFS_I(inode)->lock);
2314 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2315 (*bits & EXTENT_DELALLOC_NEW)) {
2316 spin_lock(&BTRFS_I(inode)->lock);
2317 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2319 spin_unlock(&BTRFS_I(inode)->lock);
2324 * Once a range is no longer delalloc this function ensures that proper
2325 * accounting happens.
2327 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2328 struct extent_state *state, unsigned *bits)
2330 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2331 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2332 u64 len = state->end + 1 - state->start;
2333 u32 num_extents = count_max_extents(len);
2335 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2336 spin_lock(&inode->lock);
2337 inode->defrag_bytes -= len;
2338 spin_unlock(&inode->lock);
2342 * set_bit and clear bit hooks normally require _irqsave/restore
2343 * but in this case, we are only testing for the DELALLOC
2344 * bit, which is only set or cleared with irqs on
2346 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2347 struct btrfs_root *root = inode->root;
2348 bool do_list = !btrfs_is_free_space_inode(inode);
2350 spin_lock(&inode->lock);
2351 btrfs_mod_outstanding_extents(inode, -num_extents);
2352 spin_unlock(&inode->lock);
2355 * We don't reserve metadata space for space cache inodes so we
2356 * don't need to call delalloc_release_metadata if there is an
2359 if (*bits & EXTENT_CLEAR_META_RESV &&
2360 root != fs_info->tree_root)
2361 btrfs_delalloc_release_metadata(inode, len, false);
2363 /* For sanity tests. */
2364 if (btrfs_is_testing(fs_info))
2367 if (!btrfs_is_data_reloc_root(root) &&
2368 do_list && !(state->state & EXTENT_NORESERVE) &&
2369 (*bits & EXTENT_CLEAR_DATA_RESV))
2370 btrfs_free_reserved_data_space_noquota(fs_info, len);
2372 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2373 fs_info->delalloc_batch);
2374 spin_lock(&inode->lock);
2375 inode->delalloc_bytes -= len;
2376 if (do_list && inode->delalloc_bytes == 0 &&
2377 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2378 &inode->runtime_flags))
2379 btrfs_del_delalloc_inode(root, inode);
2380 spin_unlock(&inode->lock);
2383 if ((state->state & EXTENT_DELALLOC_NEW) &&
2384 (*bits & EXTENT_DELALLOC_NEW)) {
2385 spin_lock(&inode->lock);
2386 ASSERT(inode->new_delalloc_bytes >= len);
2387 inode->new_delalloc_bytes -= len;
2388 if (*bits & EXTENT_ADD_INODE_BYTES)
2389 inode_add_bytes(&inode->vfs_inode, len);
2390 spin_unlock(&inode->lock);
2395 * in order to insert checksums into the metadata in large chunks,
2396 * we wait until bio submission time. All the pages in the bio are
2397 * checksummed and sums are attached onto the ordered extent record.
2399 * At IO completion time the cums attached on the ordered extent record
2400 * are inserted into the btree
2402 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2403 u64 dio_file_offset)
2405 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2409 * Split an extent_map at [start, start + len]
2411 * This function is intended to be used only for extract_ordered_extent().
2413 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2416 struct extent_map_tree *em_tree = &inode->extent_tree;
2417 struct extent_map *em;
2418 struct extent_map *split_pre = NULL;
2419 struct extent_map *split_mid = NULL;
2420 struct extent_map *split_post = NULL;
2422 unsigned long flags;
2425 if (pre == 0 && post == 0)
2428 split_pre = alloc_extent_map();
2430 split_mid = alloc_extent_map();
2432 split_post = alloc_extent_map();
2433 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2438 ASSERT(pre + post < len);
2440 lock_extent(&inode->io_tree, start, start + len - 1);
2441 write_lock(&em_tree->lock);
2442 em = lookup_extent_mapping(em_tree, start, len);
2448 ASSERT(em->len == len);
2449 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2450 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2451 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2452 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2453 ASSERT(!list_empty(&em->list));
2456 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2458 /* First, replace the em with a new extent_map starting from * em->start */
2459 split_pre->start = em->start;
2460 split_pre->len = (pre ? pre : em->len - post);
2461 split_pre->orig_start = split_pre->start;
2462 split_pre->block_start = em->block_start;
2463 split_pre->block_len = split_pre->len;
2464 split_pre->orig_block_len = split_pre->block_len;
2465 split_pre->ram_bytes = split_pre->len;
2466 split_pre->flags = flags;
2467 split_pre->compress_type = em->compress_type;
2468 split_pre->generation = em->generation;
2470 replace_extent_mapping(em_tree, em, split_pre, 1);
2473 * Now we only have an extent_map at:
2474 * [em->start, em->start + pre] if pre != 0
2475 * [em->start, em->start + em->len - post] if pre == 0
2479 /* Insert the middle extent_map */
2480 split_mid->start = em->start + pre;
2481 split_mid->len = em->len - pre - post;
2482 split_mid->orig_start = split_mid->start;
2483 split_mid->block_start = em->block_start + pre;
2484 split_mid->block_len = split_mid->len;
2485 split_mid->orig_block_len = split_mid->block_len;
2486 split_mid->ram_bytes = split_mid->len;
2487 split_mid->flags = flags;
2488 split_mid->compress_type = em->compress_type;
2489 split_mid->generation = em->generation;
2490 add_extent_mapping(em_tree, split_mid, 1);
2494 split_post->start = em->start + em->len - post;
2495 split_post->len = post;
2496 split_post->orig_start = split_post->start;
2497 split_post->block_start = em->block_start + em->len - post;
2498 split_post->block_len = split_post->len;
2499 split_post->orig_block_len = split_post->block_len;
2500 split_post->ram_bytes = split_post->len;
2501 split_post->flags = flags;
2502 split_post->compress_type = em->compress_type;
2503 split_post->generation = em->generation;
2504 add_extent_mapping(em_tree, split_post, 1);
2508 free_extent_map(em);
2509 /* Once for the tree */
2510 free_extent_map(em);
2513 write_unlock(&em_tree->lock);
2514 unlock_extent(&inode->io_tree, start, start + len - 1);
2516 free_extent_map(split_pre);
2517 free_extent_map(split_mid);
2518 free_extent_map(split_post);
2523 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2524 struct bio *bio, loff_t file_offset)
2526 struct btrfs_ordered_extent *ordered;
2527 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2529 u64 len = bio->bi_iter.bi_size;
2530 u64 end = start + len;
2535 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2536 if (WARN_ON_ONCE(!ordered))
2537 return BLK_STS_IOERR;
2539 /* No need to split */
2540 if (ordered->disk_num_bytes == len)
2543 /* We cannot split once end_bio'd ordered extent */
2544 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2549 /* We cannot split a compressed ordered extent */
2550 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2555 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2556 /* bio must be in one ordered extent */
2557 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2562 /* Checksum list should be empty */
2563 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2568 file_len = ordered->num_bytes;
2569 pre = start - ordered->disk_bytenr;
2570 post = ordered_end - end;
2572 ret = btrfs_split_ordered_extent(ordered, pre, post);
2575 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2578 btrfs_put_ordered_extent(ordered);
2580 return errno_to_blk_status(ret);
2584 * extent_io.c submission hook. This does the right thing for csum calculation
2585 * on write, or reading the csums from the tree before a read.
2587 * Rules about async/sync submit,
2588 * a) read: sync submit
2590 * b) write without checksum: sync submit
2592 * c) write with checksum:
2593 * c-1) if bio is issued by fsync: sync submit
2594 * (sync_writers != 0)
2596 * c-2) if root is reloc root: sync submit
2597 * (only in case of buffered IO)
2599 * c-3) otherwise: async submit
2601 void btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2602 int mirror_num, enum btrfs_compression_type compress_type)
2604 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2605 struct btrfs_root *root = BTRFS_I(inode)->root;
2606 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2607 blk_status_t ret = 0;
2609 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2611 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2612 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2614 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2615 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2617 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2618 struct page *page = bio_first_bvec_all(bio)->bv_page;
2619 loff_t file_offset = page_offset(page);
2621 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2626 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2627 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2631 if (compress_type != BTRFS_COMPRESS_NONE) {
2633 * btrfs_submit_compressed_read will handle completing
2634 * the bio if there were any errors, so just return
2637 btrfs_submit_compressed_read(inode, bio, mirror_num);
2641 * Lookup bio sums does extra checks around whether we
2642 * need to csum or not, which is why we ignore skip_sum
2645 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2650 } else if (async && !skip_sum) {
2651 /* csum items have already been cloned */
2652 if (btrfs_is_data_reloc_root(root))
2654 /* we're doing a write, do the async checksumming */
2655 ret = btrfs_wq_submit_bio(inode, bio, mirror_num,
2656 0, btrfs_submit_bio_start);
2658 } else if (!skip_sum) {
2659 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2665 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2669 bio->bi_status = ret;
2675 * given a list of ordered sums record them in the inode. This happens
2676 * at IO completion time based on sums calculated at bio submission time.
2678 static int add_pending_csums(struct btrfs_trans_handle *trans,
2679 struct list_head *list)
2681 struct btrfs_ordered_sum *sum;
2682 struct btrfs_root *csum_root = NULL;
2685 list_for_each_entry(sum, list, list) {
2686 trans->adding_csums = true;
2688 csum_root = btrfs_csum_root(trans->fs_info,
2690 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2691 trans->adding_csums = false;
2698 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2701 struct extent_state **cached_state)
2703 u64 search_start = start;
2704 const u64 end = start + len - 1;
2706 while (search_start < end) {
2707 const u64 search_len = end - search_start + 1;
2708 struct extent_map *em;
2712 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2716 if (em->block_start != EXTENT_MAP_HOLE)
2720 if (em->start < search_start)
2721 em_len -= search_start - em->start;
2722 if (em_len > search_len)
2723 em_len = search_len;
2725 ret = set_extent_bit(&inode->io_tree, search_start,
2726 search_start + em_len - 1,
2727 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2730 search_start = extent_map_end(em);
2731 free_extent_map(em);
2738 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2739 unsigned int extra_bits,
2740 struct extent_state **cached_state)
2742 WARN_ON(PAGE_ALIGNED(end));
2744 if (start >= i_size_read(&inode->vfs_inode) &&
2745 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2747 * There can't be any extents following eof in this case so just
2748 * set the delalloc new bit for the range directly.
2750 extra_bits |= EXTENT_DELALLOC_NEW;
2754 ret = btrfs_find_new_delalloc_bytes(inode, start,
2761 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2765 /* see btrfs_writepage_start_hook for details on why this is required */
2766 struct btrfs_writepage_fixup {
2768 struct inode *inode;
2769 struct btrfs_work work;
2772 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2774 struct btrfs_writepage_fixup *fixup;
2775 struct btrfs_ordered_extent *ordered;
2776 struct extent_state *cached_state = NULL;
2777 struct extent_changeset *data_reserved = NULL;
2779 struct btrfs_inode *inode;
2783 bool free_delalloc_space = true;
2785 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2787 inode = BTRFS_I(fixup->inode);
2788 page_start = page_offset(page);
2789 page_end = page_offset(page) + PAGE_SIZE - 1;
2792 * This is similar to page_mkwrite, we need to reserve the space before
2793 * we take the page lock.
2795 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2801 * Before we queued this fixup, we took a reference on the page.
2802 * page->mapping may go NULL, but it shouldn't be moved to a different
2805 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2807 * Unfortunately this is a little tricky, either
2809 * 1) We got here and our page had already been dealt with and
2810 * we reserved our space, thus ret == 0, so we need to just
2811 * drop our space reservation and bail. This can happen the
2812 * first time we come into the fixup worker, or could happen
2813 * while waiting for the ordered extent.
2814 * 2) Our page was already dealt with, but we happened to get an
2815 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2816 * this case we obviously don't have anything to release, but
2817 * because the page was already dealt with we don't want to
2818 * mark the page with an error, so make sure we're resetting
2819 * ret to 0. This is why we have this check _before_ the ret
2820 * check, because we do not want to have a surprise ENOSPC
2821 * when the page was already properly dealt with.
2824 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2825 btrfs_delalloc_release_space(inode, data_reserved,
2826 page_start, PAGE_SIZE,
2834 * We can't mess with the page state unless it is locked, so now that
2835 * it is locked bail if we failed to make our space reservation.
2840 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2842 /* already ordered? We're done */
2843 if (PageOrdered(page))
2846 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2848 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2851 btrfs_start_ordered_extent(ordered, 1);
2852 btrfs_put_ordered_extent(ordered);
2856 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2862 * Everything went as planned, we're now the owner of a dirty page with
2863 * delayed allocation bits set and space reserved for our COW
2866 * The page was dirty when we started, nothing should have cleaned it.
2868 BUG_ON(!PageDirty(page));
2869 free_delalloc_space = false;
2871 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2872 if (free_delalloc_space)
2873 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2875 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2880 * We hit ENOSPC or other errors. Update the mapping and page
2881 * to reflect the errors and clean the page.
2883 mapping_set_error(page->mapping, ret);
2884 end_extent_writepage(page, ret, page_start, page_end);
2885 clear_page_dirty_for_io(page);
2888 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2892 extent_changeset_free(data_reserved);
2894 * As a precaution, do a delayed iput in case it would be the last iput
2895 * that could need flushing space. Recursing back to fixup worker would
2898 btrfs_add_delayed_iput(&inode->vfs_inode);
2902 * There are a few paths in the higher layers of the kernel that directly
2903 * set the page dirty bit without asking the filesystem if it is a
2904 * good idea. This causes problems because we want to make sure COW
2905 * properly happens and the data=ordered rules are followed.
2907 * In our case any range that doesn't have the ORDERED bit set
2908 * hasn't been properly setup for IO. We kick off an async process
2909 * to fix it up. The async helper will wait for ordered extents, set
2910 * the delalloc bit and make it safe to write the page.
2912 int btrfs_writepage_cow_fixup(struct page *page)
2914 struct inode *inode = page->mapping->host;
2915 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2916 struct btrfs_writepage_fixup *fixup;
2918 /* This page has ordered extent covering it already */
2919 if (PageOrdered(page))
2923 * PageChecked is set below when we create a fixup worker for this page,
2924 * don't try to create another one if we're already PageChecked()
2926 * The extent_io writepage code will redirty the page if we send back
2929 if (PageChecked(page))
2932 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2937 * We are already holding a reference to this inode from
2938 * write_cache_pages. We need to hold it because the space reservation
2939 * takes place outside of the page lock, and we can't trust
2940 * page->mapping outside of the page lock.
2943 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2945 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2947 fixup->inode = inode;
2948 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2953 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2954 struct btrfs_inode *inode, u64 file_pos,
2955 struct btrfs_file_extent_item *stack_fi,
2956 const bool update_inode_bytes,
2957 u64 qgroup_reserved)
2959 struct btrfs_root *root = inode->root;
2960 const u64 sectorsize = root->fs_info->sectorsize;
2961 struct btrfs_path *path;
2962 struct extent_buffer *leaf;
2963 struct btrfs_key ins;
2964 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2965 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2966 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2967 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2968 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2969 struct btrfs_drop_extents_args drop_args = { 0 };
2972 path = btrfs_alloc_path();
2977 * we may be replacing one extent in the tree with another.
2978 * The new extent is pinned in the extent map, and we don't want
2979 * to drop it from the cache until it is completely in the btree.
2981 * So, tell btrfs_drop_extents to leave this extent in the cache.
2982 * the caller is expected to unpin it and allow it to be merged
2985 drop_args.path = path;
2986 drop_args.start = file_pos;
2987 drop_args.end = file_pos + num_bytes;
2988 drop_args.replace_extent = true;
2989 drop_args.extent_item_size = sizeof(*stack_fi);
2990 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2994 if (!drop_args.extent_inserted) {
2995 ins.objectid = btrfs_ino(inode);
2996 ins.offset = file_pos;
2997 ins.type = BTRFS_EXTENT_DATA_KEY;
2999 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3004 leaf = path->nodes[0];
3005 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3006 write_extent_buffer(leaf, stack_fi,
3007 btrfs_item_ptr_offset(leaf, path->slots[0]),
3008 sizeof(struct btrfs_file_extent_item));
3010 btrfs_mark_buffer_dirty(leaf);
3011 btrfs_release_path(path);
3014 * If we dropped an inline extent here, we know the range where it is
3015 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3016 * number of bytes only for that range containing the inline extent.
3017 * The remaining of the range will be processed when clearning the
3018 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3020 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3021 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3023 inline_size = drop_args.bytes_found - inline_size;
3024 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3025 drop_args.bytes_found -= inline_size;
3026 num_bytes -= sectorsize;
3029 if (update_inode_bytes)
3030 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3032 ins.objectid = disk_bytenr;
3033 ins.offset = disk_num_bytes;
3034 ins.type = BTRFS_EXTENT_ITEM_KEY;
3036 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3040 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3042 qgroup_reserved, &ins);
3044 btrfs_free_path(path);
3049 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3052 struct btrfs_block_group *cache;
3054 cache = btrfs_lookup_block_group(fs_info, start);
3057 spin_lock(&cache->lock);
3058 cache->delalloc_bytes -= len;
3059 spin_unlock(&cache->lock);
3061 btrfs_put_block_group(cache);
3064 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3065 struct btrfs_ordered_extent *oe)
3067 struct btrfs_file_extent_item stack_fi;
3068 bool update_inode_bytes;
3069 u64 num_bytes = oe->num_bytes;
3070 u64 ram_bytes = oe->ram_bytes;
3072 memset(&stack_fi, 0, sizeof(stack_fi));
3073 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3074 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3075 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3076 oe->disk_num_bytes);
3077 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3078 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
3079 num_bytes = ram_bytes = oe->truncated_len;
3080 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3081 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3082 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3083 /* Encryption and other encoding is reserved and all 0 */
3086 * For delalloc, when completing an ordered extent we update the inode's
3087 * bytes when clearing the range in the inode's io tree, so pass false
3088 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3089 * except if the ordered extent was truncated.
3091 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3092 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3093 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3095 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3096 oe->file_offset, &stack_fi,
3097 update_inode_bytes, oe->qgroup_rsv);
3101 * As ordered data IO finishes, this gets called so we can finish
3102 * an ordered extent if the range of bytes in the file it covers are
3105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3107 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3108 struct btrfs_root *root = inode->root;
3109 struct btrfs_fs_info *fs_info = root->fs_info;
3110 struct btrfs_trans_handle *trans = NULL;
3111 struct extent_io_tree *io_tree = &inode->io_tree;
3112 struct extent_state *cached_state = NULL;
3114 int compress_type = 0;
3116 u64 logical_len = ordered_extent->num_bytes;
3117 bool freespace_inode;
3118 bool truncated = false;
3119 bool clear_reserved_extent = true;
3120 unsigned int clear_bits = EXTENT_DEFRAG;
3122 start = ordered_extent->file_offset;
3123 end = start + ordered_extent->num_bytes - 1;
3125 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3126 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3127 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3128 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3129 clear_bits |= EXTENT_DELALLOC_NEW;
3131 freespace_inode = btrfs_is_free_space_inode(inode);
3133 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3138 /* A valid bdev implies a write on a sequential zone */
3139 if (ordered_extent->bdev) {
3140 btrfs_rewrite_logical_zoned(ordered_extent);
3141 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3142 ordered_extent->disk_num_bytes);
3145 btrfs_free_io_failure_record(inode, start, end);
3147 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3149 logical_len = ordered_extent->truncated_len;
3150 /* Truncated the entire extent, don't bother adding */
3155 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3156 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3158 btrfs_inode_safe_disk_i_size_write(inode, 0);
3159 if (freespace_inode)
3160 trans = btrfs_join_transaction_spacecache(root);
3162 trans = btrfs_join_transaction(root);
3163 if (IS_ERR(trans)) {
3164 ret = PTR_ERR(trans);
3168 trans->block_rsv = &inode->block_rsv;
3169 ret = btrfs_update_inode_fallback(trans, root, inode);
3170 if (ret) /* -ENOMEM or corruption */
3171 btrfs_abort_transaction(trans, ret);
3175 clear_bits |= EXTENT_LOCKED;
3176 lock_extent_bits(io_tree, start, end, &cached_state);
3178 if (freespace_inode)
3179 trans = btrfs_join_transaction_spacecache(root);
3181 trans = btrfs_join_transaction(root);
3182 if (IS_ERR(trans)) {
3183 ret = PTR_ERR(trans);
3188 trans->block_rsv = &inode->block_rsv;
3190 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3191 compress_type = ordered_extent->compress_type;
3192 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3193 BUG_ON(compress_type);
3194 ret = btrfs_mark_extent_written(trans, inode,
3195 ordered_extent->file_offset,
3196 ordered_extent->file_offset +
3198 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3199 ordered_extent->disk_num_bytes);
3201 BUG_ON(root == fs_info->tree_root);
3202 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3204 clear_reserved_extent = false;
3205 btrfs_release_delalloc_bytes(fs_info,
3206 ordered_extent->disk_bytenr,
3207 ordered_extent->disk_num_bytes);
3210 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3211 ordered_extent->num_bytes, trans->transid);
3213 btrfs_abort_transaction(trans, ret);
3217 ret = add_pending_csums(trans, &ordered_extent->list);
3219 btrfs_abort_transaction(trans, ret);
3224 * If this is a new delalloc range, clear its new delalloc flag to
3225 * update the inode's number of bytes. This needs to be done first
3226 * before updating the inode item.
3228 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3229 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3230 clear_extent_bit(&inode->io_tree, start, end,
3231 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3232 0, 0, &cached_state);
3234 btrfs_inode_safe_disk_i_size_write(inode, 0);
3235 ret = btrfs_update_inode_fallback(trans, root, inode);
3236 if (ret) { /* -ENOMEM or corruption */
3237 btrfs_abort_transaction(trans, ret);
3242 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3243 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3247 btrfs_end_transaction(trans);
3249 if (ret || truncated) {
3250 u64 unwritten_start = start;
3253 * If we failed to finish this ordered extent for any reason we
3254 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3255 * extent, and mark the inode with the error if it wasn't
3256 * already set. Any error during writeback would have already
3257 * set the mapping error, so we need to set it if we're the ones
3258 * marking this ordered extent as failed.
3260 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3261 &ordered_extent->flags))
3262 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3265 unwritten_start += logical_len;
3266 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3268 /* Drop the cache for the part of the extent we didn't write. */
3269 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3272 * If the ordered extent had an IOERR or something else went
3273 * wrong we need to return the space for this ordered extent
3274 * back to the allocator. We only free the extent in the
3275 * truncated case if we didn't write out the extent at all.
3277 * If we made it past insert_reserved_file_extent before we
3278 * errored out then we don't need to do this as the accounting
3279 * has already been done.
3281 if ((ret || !logical_len) &&
3282 clear_reserved_extent &&
3283 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3284 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3286 * Discard the range before returning it back to the
3289 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3290 btrfs_discard_extent(fs_info,
3291 ordered_extent->disk_bytenr,
3292 ordered_extent->disk_num_bytes,
3294 btrfs_free_reserved_extent(fs_info,
3295 ordered_extent->disk_bytenr,
3296 ordered_extent->disk_num_bytes, 1);
3301 * This needs to be done to make sure anybody waiting knows we are done
3302 * updating everything for this ordered extent.
3304 btrfs_remove_ordered_extent(inode, ordered_extent);
3307 btrfs_put_ordered_extent(ordered_extent);
3308 /* once for the tree */
3309 btrfs_put_ordered_extent(ordered_extent);
3314 static void finish_ordered_fn(struct btrfs_work *work)
3316 struct btrfs_ordered_extent *ordered_extent;
3317 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3318 btrfs_finish_ordered_io(ordered_extent);
3321 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3322 struct page *page, u64 start,
3323 u64 end, bool uptodate)
3325 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3327 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3328 finish_ordered_fn, uptodate);
3332 * check_data_csum - verify checksum of one sector of uncompressed data
3334 * @io_bio: btrfs_io_bio which contains the csum
3335 * @bio_offset: offset to the beginning of the bio (in bytes)
3336 * @page: page where is the data to be verified
3337 * @pgoff: offset inside the page
3338 * @start: logical offset in the file
3340 * The length of such check is always one sector size.
3342 static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3343 u32 bio_offset, struct page *page, u32 pgoff,
3346 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3347 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3349 u32 len = fs_info->sectorsize;
3350 const u32 csum_size = fs_info->csum_size;
3351 unsigned int offset_sectors;
3353 u8 csum[BTRFS_CSUM_SIZE];
3355 ASSERT(pgoff + len <= PAGE_SIZE);
3357 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3358 csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size;
3360 kaddr = kmap_atomic(page);
3361 shash->tfm = fs_info->csum_shash;
3363 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3364 kunmap_atomic(kaddr);
3366 if (memcmp(csum, csum_expected, csum_size))
3371 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3374 btrfs_dev_stat_inc_and_print(bbio->device,
3375 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3376 memzero_page(page, pgoff, len);
3381 * When reads are done, we need to check csums to verify the data is correct.
3382 * if there's a match, we allow the bio to finish. If not, the code in
3383 * extent_io.c will try to find good copies for us.
3385 * @bio_offset: offset to the beginning of the bio (in bytes)
3386 * @start: file offset of the range start
3387 * @end: file offset of the range end (inclusive)
3389 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3392 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3393 u32 bio_offset, struct page *page,
3396 struct inode *inode = page->mapping->host;
3397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3398 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3399 struct btrfs_root *root = BTRFS_I(inode)->root;
3400 const u32 sectorsize = root->fs_info->sectorsize;
3402 unsigned int result = 0;
3404 if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) {
3405 btrfs_page_clear_checked(fs_info, page, start, end + 1 - start);
3410 * This only happens for NODATASUM or compressed read.
3411 * Normally this should be covered by above check for compressed read
3412 * or the next check for NODATASUM. Just do a quicker exit here.
3414 if (bbio->csum == NULL)
3417 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3420 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3423 ASSERT(page_offset(page) <= start &&
3424 end <= page_offset(page) + PAGE_SIZE - 1);
3425 for (pg_off = offset_in_page(start);
3426 pg_off < offset_in_page(end);
3427 pg_off += sectorsize, bio_offset += sectorsize) {
3428 u64 file_offset = pg_off + page_offset(page);
3431 if (btrfs_is_data_reloc_root(root) &&
3432 test_range_bit(io_tree, file_offset,
3433 file_offset + sectorsize - 1,
3434 EXTENT_NODATASUM, 1, NULL)) {
3435 /* Skip the range without csum for data reloc inode */
3436 clear_extent_bits(io_tree, file_offset,
3437 file_offset + sectorsize - 1,
3441 ret = check_data_csum(inode, bbio, bio_offset, page, pg_off,
3442 page_offset(page) + pg_off);
3444 const int nr_bit = (pg_off - offset_in_page(start)) >>
3445 root->fs_info->sectorsize_bits;
3447 result |= (1U << nr_bit);
3454 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3456 * @inode: The inode we want to perform iput on
3458 * This function uses the generic vfs_inode::i_count to track whether we should
3459 * just decrement it (in case it's > 1) or if this is the last iput then link
3460 * the inode to the delayed iput machinery. Delayed iputs are processed at
3461 * transaction commit time/superblock commit/cleaner kthread.
3463 void btrfs_add_delayed_iput(struct inode *inode)
3465 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3466 struct btrfs_inode *binode = BTRFS_I(inode);
3468 if (atomic_add_unless(&inode->i_count, -1, 1))
3471 atomic_inc(&fs_info->nr_delayed_iputs);
3472 spin_lock(&fs_info->delayed_iput_lock);
3473 ASSERT(list_empty(&binode->delayed_iput));
3474 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3475 spin_unlock(&fs_info->delayed_iput_lock);
3476 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3477 wake_up_process(fs_info->cleaner_kthread);
3480 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3481 struct btrfs_inode *inode)
3483 list_del_init(&inode->delayed_iput);
3484 spin_unlock(&fs_info->delayed_iput_lock);
3485 iput(&inode->vfs_inode);
3486 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3487 wake_up(&fs_info->delayed_iputs_wait);
3488 spin_lock(&fs_info->delayed_iput_lock);
3491 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3492 struct btrfs_inode *inode)
3494 if (!list_empty(&inode->delayed_iput)) {
3495 spin_lock(&fs_info->delayed_iput_lock);
3496 if (!list_empty(&inode->delayed_iput))
3497 run_delayed_iput_locked(fs_info, inode);
3498 spin_unlock(&fs_info->delayed_iput_lock);
3502 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3505 spin_lock(&fs_info->delayed_iput_lock);
3506 while (!list_empty(&fs_info->delayed_iputs)) {
3507 struct btrfs_inode *inode;
3509 inode = list_first_entry(&fs_info->delayed_iputs,
3510 struct btrfs_inode, delayed_iput);
3511 run_delayed_iput_locked(fs_info, inode);
3512 cond_resched_lock(&fs_info->delayed_iput_lock);
3514 spin_unlock(&fs_info->delayed_iput_lock);
3518 * Wait for flushing all delayed iputs
3520 * @fs_info: the filesystem
3522 * This will wait on any delayed iputs that are currently running with KILLABLE
3523 * set. Once they are all done running we will return, unless we are killed in
3524 * which case we return EINTR. This helps in user operations like fallocate etc
3525 * that might get blocked on the iputs.
3527 * Return EINTR if we were killed, 0 if nothing's pending
3529 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3531 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3532 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3539 * This creates an orphan entry for the given inode in case something goes wrong
3540 * in the middle of an unlink.
3542 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3543 struct btrfs_inode *inode)
3547 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3548 if (ret && ret != -EEXIST) {
3549 btrfs_abort_transaction(trans, ret);
3557 * We have done the delete so we can go ahead and remove the orphan item for
3558 * this particular inode.
3560 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3561 struct btrfs_inode *inode)
3563 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3567 * this cleans up any orphans that may be left on the list from the last use
3570 int btrfs_orphan_cleanup(struct btrfs_root *root)
3572 struct btrfs_fs_info *fs_info = root->fs_info;
3573 struct btrfs_path *path;
3574 struct extent_buffer *leaf;
3575 struct btrfs_key key, found_key;
3576 struct btrfs_trans_handle *trans;
3577 struct inode *inode;
3578 u64 last_objectid = 0;
3579 int ret = 0, nr_unlink = 0;
3581 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3584 path = btrfs_alloc_path();
3589 path->reada = READA_BACK;
3591 key.objectid = BTRFS_ORPHAN_OBJECTID;
3592 key.type = BTRFS_ORPHAN_ITEM_KEY;
3593 key.offset = (u64)-1;
3596 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3601 * if ret == 0 means we found what we were searching for, which
3602 * is weird, but possible, so only screw with path if we didn't
3603 * find the key and see if we have stuff that matches
3607 if (path->slots[0] == 0)
3612 /* pull out the item */
3613 leaf = path->nodes[0];
3614 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3616 /* make sure the item matches what we want */
3617 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3619 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3622 /* release the path since we're done with it */
3623 btrfs_release_path(path);
3626 * this is where we are basically btrfs_lookup, without the
3627 * crossing root thing. we store the inode number in the
3628 * offset of the orphan item.
3631 if (found_key.offset == last_objectid) {
3633 "Error removing orphan entry, stopping orphan cleanup");
3638 last_objectid = found_key.offset;
3640 found_key.objectid = found_key.offset;
3641 found_key.type = BTRFS_INODE_ITEM_KEY;
3642 found_key.offset = 0;
3643 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3644 ret = PTR_ERR_OR_ZERO(inode);
3645 if (ret && ret != -ENOENT)
3648 if (ret == -ENOENT && root == fs_info->tree_root) {
3649 struct btrfs_root *dead_root;
3650 int is_dead_root = 0;
3653 * This is an orphan in the tree root. Currently these
3654 * could come from 2 sources:
3655 * a) a root (snapshot/subvolume) deletion in progress
3656 * b) a free space cache inode
3657 * We need to distinguish those two, as the orphan item
3658 * for a root must not get deleted before the deletion
3659 * of the snapshot/subvolume's tree completes.
3661 * btrfs_find_orphan_roots() ran before us, which has
3662 * found all deleted roots and loaded them into
3663 * fs_info->fs_roots_radix. So here we can find if an
3664 * orphan item corresponds to a deleted root by looking
3665 * up the root from that radix tree.
3668 spin_lock(&fs_info->fs_roots_radix_lock);
3669 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3670 (unsigned long)found_key.objectid);
3671 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3673 spin_unlock(&fs_info->fs_roots_radix_lock);
3676 /* prevent this orphan from being found again */
3677 key.offset = found_key.objectid - 1;
3684 * If we have an inode with links, there are a couple of
3687 * 1. We were halfway through creating fsverity metadata for the
3688 * file. In that case, the orphan item represents incomplete
3689 * fsverity metadata which must be cleaned up with
3690 * btrfs_drop_verity_items and deleting the orphan item.
3692 * 2. Old kernels (before v3.12) used to create an
3693 * orphan item for truncate indicating that there were possibly
3694 * extent items past i_size that needed to be deleted. In v3.12,
3695 * truncate was changed to update i_size in sync with the extent
3696 * items, but the (useless) orphan item was still created. Since
3697 * v4.18, we don't create the orphan item for truncate at all.
3699 * So, this item could mean that we need to do a truncate, but
3700 * only if this filesystem was last used on a pre-v3.12 kernel
3701 * and was not cleanly unmounted. The odds of that are quite
3702 * slim, and it's a pain to do the truncate now, so just delete
3705 * It's also possible that this orphan item was supposed to be
3706 * deleted but wasn't. The inode number may have been reused,
3707 * but either way, we can delete the orphan item.
3709 if (ret == -ENOENT || inode->i_nlink) {
3711 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3716 trans = btrfs_start_transaction(root, 1);
3717 if (IS_ERR(trans)) {
3718 ret = PTR_ERR(trans);
3721 btrfs_debug(fs_info, "auto deleting %Lu",
3722 found_key.objectid);
3723 ret = btrfs_del_orphan_item(trans, root,
3724 found_key.objectid);
3725 btrfs_end_transaction(trans);
3733 /* this will do delete_inode and everything for us */
3736 /* release the path since we're done with it */
3737 btrfs_release_path(path);
3739 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3740 trans = btrfs_join_transaction(root);
3742 btrfs_end_transaction(trans);
3746 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3750 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3751 btrfs_free_path(path);
3756 * very simple check to peek ahead in the leaf looking for xattrs. If we
3757 * don't find any xattrs, we know there can't be any acls.
3759 * slot is the slot the inode is in, objectid is the objectid of the inode
3761 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3762 int slot, u64 objectid,
3763 int *first_xattr_slot)
3765 u32 nritems = btrfs_header_nritems(leaf);
3766 struct btrfs_key found_key;
3767 static u64 xattr_access = 0;
3768 static u64 xattr_default = 0;
3771 if (!xattr_access) {
3772 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3773 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3774 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3775 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3779 *first_xattr_slot = -1;
3780 while (slot < nritems) {
3781 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3783 /* we found a different objectid, there must not be acls */
3784 if (found_key.objectid != objectid)
3787 /* we found an xattr, assume we've got an acl */
3788 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3789 if (*first_xattr_slot == -1)
3790 *first_xattr_slot = slot;
3791 if (found_key.offset == xattr_access ||
3792 found_key.offset == xattr_default)
3797 * we found a key greater than an xattr key, there can't
3798 * be any acls later on
3800 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3807 * it goes inode, inode backrefs, xattrs, extents,
3808 * so if there are a ton of hard links to an inode there can
3809 * be a lot of backrefs. Don't waste time searching too hard,
3810 * this is just an optimization
3815 /* we hit the end of the leaf before we found an xattr or
3816 * something larger than an xattr. We have to assume the inode
3819 if (*first_xattr_slot == -1)
3820 *first_xattr_slot = slot;
3825 * read an inode from the btree into the in-memory inode
3827 static int btrfs_read_locked_inode(struct inode *inode,
3828 struct btrfs_path *in_path)
3830 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3831 struct btrfs_path *path = in_path;
3832 struct extent_buffer *leaf;
3833 struct btrfs_inode_item *inode_item;
3834 struct btrfs_root *root = BTRFS_I(inode)->root;
3835 struct btrfs_key location;
3840 bool filled = false;
3841 int first_xattr_slot;
3843 ret = btrfs_fill_inode(inode, &rdev);
3848 path = btrfs_alloc_path();
3853 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3855 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3857 if (path != in_path)
3858 btrfs_free_path(path);
3862 leaf = path->nodes[0];
3867 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3868 struct btrfs_inode_item);
3869 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3870 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3871 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3872 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3873 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3874 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3875 round_up(i_size_read(inode), fs_info->sectorsize));
3877 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3878 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3880 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3881 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3883 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3884 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3886 BTRFS_I(inode)->i_otime.tv_sec =
3887 btrfs_timespec_sec(leaf, &inode_item->otime);
3888 BTRFS_I(inode)->i_otime.tv_nsec =
3889 btrfs_timespec_nsec(leaf, &inode_item->otime);
3891 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3892 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3893 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3895 inode_set_iversion_queried(inode,
3896 btrfs_inode_sequence(leaf, inode_item));
3897 inode->i_generation = BTRFS_I(inode)->generation;
3899 rdev = btrfs_inode_rdev(leaf, inode_item);
3901 BTRFS_I(inode)->index_cnt = (u64)-1;
3902 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3903 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3907 * If we were modified in the current generation and evicted from memory
3908 * and then re-read we need to do a full sync since we don't have any
3909 * idea about which extents were modified before we were evicted from
3912 * This is required for both inode re-read from disk and delayed inode
3913 * in delayed_nodes_tree.
3915 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3916 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3917 &BTRFS_I(inode)->runtime_flags);
3920 * We don't persist the id of the transaction where an unlink operation
3921 * against the inode was last made. So here we assume the inode might
3922 * have been evicted, and therefore the exact value of last_unlink_trans
3923 * lost, and set it to last_trans to avoid metadata inconsistencies
3924 * between the inode and its parent if the inode is fsync'ed and the log
3925 * replayed. For example, in the scenario:
3928 * ln mydir/foo mydir/bar
3931 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3932 * xfs_io -c fsync mydir/foo
3934 * mount fs, triggers fsync log replay
3936 * We must make sure that when we fsync our inode foo we also log its
3937 * parent inode, otherwise after log replay the parent still has the
3938 * dentry with the "bar" name but our inode foo has a link count of 1
3939 * and doesn't have an inode ref with the name "bar" anymore.
3941 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3942 * but it guarantees correctness at the expense of occasional full
3943 * transaction commits on fsync if our inode is a directory, or if our
3944 * inode is not a directory, logging its parent unnecessarily.
3946 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3949 * Same logic as for last_unlink_trans. We don't persist the generation
3950 * of the last transaction where this inode was used for a reflink
3951 * operation, so after eviction and reloading the inode we must be
3952 * pessimistic and assume the last transaction that modified the inode.
3954 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3957 if (inode->i_nlink != 1 ||
3958 path->slots[0] >= btrfs_header_nritems(leaf))
3961 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3962 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3965 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3966 if (location.type == BTRFS_INODE_REF_KEY) {
3967 struct btrfs_inode_ref *ref;
3969 ref = (struct btrfs_inode_ref *)ptr;
3970 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3971 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3972 struct btrfs_inode_extref *extref;
3974 extref = (struct btrfs_inode_extref *)ptr;
3975 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3980 * try to precache a NULL acl entry for files that don't have
3981 * any xattrs or acls
3983 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3984 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3985 if (first_xattr_slot != -1) {
3986 path->slots[0] = first_xattr_slot;
3987 ret = btrfs_load_inode_props(inode, path);
3990 "error loading props for ino %llu (root %llu): %d",
3991 btrfs_ino(BTRFS_I(inode)),
3992 root->root_key.objectid, ret);
3994 if (path != in_path)
3995 btrfs_free_path(path);
3998 cache_no_acl(inode);
4000 switch (inode->i_mode & S_IFMT) {
4002 inode->i_mapping->a_ops = &btrfs_aops;
4003 inode->i_fop = &btrfs_file_operations;
4004 inode->i_op = &btrfs_file_inode_operations;
4007 inode->i_fop = &btrfs_dir_file_operations;
4008 inode->i_op = &btrfs_dir_inode_operations;
4011 inode->i_op = &btrfs_symlink_inode_operations;
4012 inode_nohighmem(inode);
4013 inode->i_mapping->a_ops = &btrfs_aops;
4016 inode->i_op = &btrfs_special_inode_operations;
4017 init_special_inode(inode, inode->i_mode, rdev);
4021 btrfs_sync_inode_flags_to_i_flags(inode);
4026 * given a leaf and an inode, copy the inode fields into the leaf
4028 static void fill_inode_item(struct btrfs_trans_handle *trans,
4029 struct extent_buffer *leaf,
4030 struct btrfs_inode_item *item,
4031 struct inode *inode)
4033 struct btrfs_map_token token;
4036 btrfs_init_map_token(&token, leaf);
4038 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4039 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4040 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4041 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4042 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4044 btrfs_set_token_timespec_sec(&token, &item->atime,
4045 inode->i_atime.tv_sec);
4046 btrfs_set_token_timespec_nsec(&token, &item->atime,
4047 inode->i_atime.tv_nsec);
4049 btrfs_set_token_timespec_sec(&token, &item->mtime,
4050 inode->i_mtime.tv_sec);
4051 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4052 inode->i_mtime.tv_nsec);
4054 btrfs_set_token_timespec_sec(&token, &item->ctime,
4055 inode->i_ctime.tv_sec);
4056 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4057 inode->i_ctime.tv_nsec);
4059 btrfs_set_token_timespec_sec(&token, &item->otime,
4060 BTRFS_I(inode)->i_otime.tv_sec);
4061 btrfs_set_token_timespec_nsec(&token, &item->otime,
4062 BTRFS_I(inode)->i_otime.tv_nsec);
4064 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4065 btrfs_set_token_inode_generation(&token, item,
4066 BTRFS_I(inode)->generation);
4067 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4068 btrfs_set_token_inode_transid(&token, item, trans->transid);
4069 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4070 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4071 BTRFS_I(inode)->ro_flags);
4072 btrfs_set_token_inode_flags(&token, item, flags);
4073 btrfs_set_token_inode_block_group(&token, item, 0);
4077 * copy everything in the in-memory inode into the btree.
4079 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4080 struct btrfs_root *root,
4081 struct btrfs_inode *inode)
4083 struct btrfs_inode_item *inode_item;
4084 struct btrfs_path *path;
4085 struct extent_buffer *leaf;
4088 path = btrfs_alloc_path();
4092 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4099 leaf = path->nodes[0];
4100 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4101 struct btrfs_inode_item);
4103 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4104 btrfs_mark_buffer_dirty(leaf);
4105 btrfs_set_inode_last_trans(trans, inode);
4108 btrfs_free_path(path);
4113 * copy everything in the in-memory inode into the btree.
4115 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4116 struct btrfs_root *root,
4117 struct btrfs_inode *inode)
4119 struct btrfs_fs_info *fs_info = root->fs_info;
4123 * If the inode is a free space inode, we can deadlock during commit
4124 * if we put it into the delayed code.
4126 * The data relocation inode should also be directly updated
4129 if (!btrfs_is_free_space_inode(inode)
4130 && !btrfs_is_data_reloc_root(root)
4131 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4132 btrfs_update_root_times(trans, root);
4134 ret = btrfs_delayed_update_inode(trans, root, inode);
4136 btrfs_set_inode_last_trans(trans, inode);
4140 return btrfs_update_inode_item(trans, root, inode);
4143 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4144 struct btrfs_root *root, struct btrfs_inode *inode)
4148 ret = btrfs_update_inode(trans, root, inode);
4150 return btrfs_update_inode_item(trans, root, inode);
4155 * unlink helper that gets used here in inode.c and in the tree logging
4156 * recovery code. It remove a link in a directory with a given name, and
4157 * also drops the back refs in the inode to the directory
4159 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4160 struct btrfs_inode *dir,
4161 struct btrfs_inode *inode,
4162 const char *name, int name_len,
4163 struct btrfs_rename_ctx *rename_ctx)
4165 struct btrfs_root *root = dir->root;
4166 struct btrfs_fs_info *fs_info = root->fs_info;
4167 struct btrfs_path *path;
4169 struct btrfs_dir_item *di;
4171 u64 ino = btrfs_ino(inode);
4172 u64 dir_ino = btrfs_ino(dir);
4174 path = btrfs_alloc_path();
4180 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4181 name, name_len, -1);
4182 if (IS_ERR_OR_NULL(di)) {
4183 ret = di ? PTR_ERR(di) : -ENOENT;
4186 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4189 btrfs_release_path(path);
4192 * If we don't have dir index, we have to get it by looking up
4193 * the inode ref, since we get the inode ref, remove it directly,
4194 * it is unnecessary to do delayed deletion.
4196 * But if we have dir index, needn't search inode ref to get it.
4197 * Since the inode ref is close to the inode item, it is better
4198 * that we delay to delete it, and just do this deletion when
4199 * we update the inode item.
4201 if (inode->dir_index) {
4202 ret = btrfs_delayed_delete_inode_ref(inode);
4204 index = inode->dir_index;
4209 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4213 "failed to delete reference to %.*s, inode %llu parent %llu",
4214 name_len, name, ino, dir_ino);
4215 btrfs_abort_transaction(trans, ret);
4220 rename_ctx->index = index;
4222 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4224 btrfs_abort_transaction(trans, ret);
4229 * If we are in a rename context, we don't need to update anything in the
4230 * log. That will be done later during the rename by btrfs_log_new_name().
4231 * Besides that, doing it here would only cause extra unncessary btree
4232 * operations on the log tree, increasing latency for applications.
4235 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4237 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4242 * If we have a pending delayed iput we could end up with the final iput
4243 * being run in btrfs-cleaner context. If we have enough of these built
4244 * up we can end up burning a lot of time in btrfs-cleaner without any
4245 * way to throttle the unlinks. Since we're currently holding a ref on
4246 * the inode we can run the delayed iput here without any issues as the
4247 * final iput won't be done until after we drop the ref we're currently
4250 btrfs_run_delayed_iput(fs_info, inode);
4252 btrfs_free_path(path);
4256 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4257 inode_inc_iversion(&inode->vfs_inode);
4258 inode_inc_iversion(&dir->vfs_inode);
4259 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4260 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4261 ret = btrfs_update_inode(trans, root, dir);
4266 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4267 struct btrfs_inode *dir, struct btrfs_inode *inode,
4268 const char *name, int name_len)
4271 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4273 drop_nlink(&inode->vfs_inode);
4274 ret = btrfs_update_inode(trans, inode->root, inode);
4280 * helper to start transaction for unlink and rmdir.
4282 * unlink and rmdir are special in btrfs, they do not always free space, so
4283 * if we cannot make our reservations the normal way try and see if there is
4284 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4285 * allow the unlink to occur.
4287 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4289 struct btrfs_root *root = BTRFS_I(dir)->root;
4292 * 1 for the possible orphan item
4293 * 1 for the dir item
4294 * 1 for the dir index
4295 * 1 for the inode ref
4297 * 1 for the parent inode
4299 return btrfs_start_transaction_fallback_global_rsv(root, 6);
4302 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4304 struct btrfs_trans_handle *trans;
4305 struct inode *inode = d_inode(dentry);
4308 trans = __unlink_start_trans(dir);
4310 return PTR_ERR(trans);
4312 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4315 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4316 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4317 dentry->d_name.len);
4321 if (inode->i_nlink == 0) {
4322 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4328 btrfs_end_transaction(trans);
4329 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4333 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4334 struct inode *dir, struct dentry *dentry)
4336 struct btrfs_root *root = BTRFS_I(dir)->root;
4337 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4338 struct btrfs_path *path;
4339 struct extent_buffer *leaf;
4340 struct btrfs_dir_item *di;
4341 struct btrfs_key key;
4342 const char *name = dentry->d_name.name;
4343 int name_len = dentry->d_name.len;
4347 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4349 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4350 objectid = inode->root->root_key.objectid;
4351 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4352 objectid = inode->location.objectid;
4358 path = btrfs_alloc_path();
4362 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4363 name, name_len, -1);
4364 if (IS_ERR_OR_NULL(di)) {
4365 ret = di ? PTR_ERR(di) : -ENOENT;
4369 leaf = path->nodes[0];
4370 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4371 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4372 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4374 btrfs_abort_transaction(trans, ret);
4377 btrfs_release_path(path);
4380 * This is a placeholder inode for a subvolume we didn't have a
4381 * reference to at the time of the snapshot creation. In the meantime
4382 * we could have renamed the real subvol link into our snapshot, so
4383 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4384 * Instead simply lookup the dir_index_item for this entry so we can
4385 * remove it. Otherwise we know we have a ref to the root and we can
4386 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4388 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4389 di = btrfs_search_dir_index_item(root, path, dir_ino,
4391 if (IS_ERR_OR_NULL(di)) {
4396 btrfs_abort_transaction(trans, ret);
4400 leaf = path->nodes[0];
4401 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4403 btrfs_release_path(path);
4405 ret = btrfs_del_root_ref(trans, objectid,
4406 root->root_key.objectid, dir_ino,
4407 &index, name, name_len);
4409 btrfs_abort_transaction(trans, ret);
4414 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4416 btrfs_abort_transaction(trans, ret);
4420 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4421 inode_inc_iversion(dir);
4422 dir->i_mtime = dir->i_ctime = current_time(dir);
4423 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4425 btrfs_abort_transaction(trans, ret);
4427 btrfs_free_path(path);
4432 * Helper to check if the subvolume references other subvolumes or if it's
4435 static noinline int may_destroy_subvol(struct btrfs_root *root)
4437 struct btrfs_fs_info *fs_info = root->fs_info;
4438 struct btrfs_path *path;
4439 struct btrfs_dir_item *di;
4440 struct btrfs_key key;
4444 path = btrfs_alloc_path();
4448 /* Make sure this root isn't set as the default subvol */
4449 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4450 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4451 dir_id, "default", 7, 0);
4452 if (di && !IS_ERR(di)) {
4453 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4454 if (key.objectid == root->root_key.objectid) {
4457 "deleting default subvolume %llu is not allowed",
4461 btrfs_release_path(path);
4464 key.objectid = root->root_key.objectid;
4465 key.type = BTRFS_ROOT_REF_KEY;
4466 key.offset = (u64)-1;
4468 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4474 if (path->slots[0] > 0) {
4476 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4477 if (key.objectid == root->root_key.objectid &&
4478 key.type == BTRFS_ROOT_REF_KEY)
4482 btrfs_free_path(path);
4486 /* Delete all dentries for inodes belonging to the root */
4487 static void btrfs_prune_dentries(struct btrfs_root *root)
4489 struct btrfs_fs_info *fs_info = root->fs_info;
4490 struct rb_node *node;
4491 struct rb_node *prev;
4492 struct btrfs_inode *entry;
4493 struct inode *inode;
4496 if (!BTRFS_FS_ERROR(fs_info))
4497 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4499 spin_lock(&root->inode_lock);
4501 node = root->inode_tree.rb_node;
4505 entry = rb_entry(node, struct btrfs_inode, rb_node);
4507 if (objectid < btrfs_ino(entry))
4508 node = node->rb_left;
4509 else if (objectid > btrfs_ino(entry))
4510 node = node->rb_right;
4516 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4517 if (objectid <= btrfs_ino(entry)) {
4521 prev = rb_next(prev);
4525 entry = rb_entry(node, struct btrfs_inode, rb_node);
4526 objectid = btrfs_ino(entry) + 1;
4527 inode = igrab(&entry->vfs_inode);
4529 spin_unlock(&root->inode_lock);
4530 if (atomic_read(&inode->i_count) > 1)
4531 d_prune_aliases(inode);
4533 * btrfs_drop_inode will have it removed from the inode
4534 * cache when its usage count hits zero.
4538 spin_lock(&root->inode_lock);
4542 if (cond_resched_lock(&root->inode_lock))
4545 node = rb_next(node);
4547 spin_unlock(&root->inode_lock);
4550 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4552 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4553 struct btrfs_root *root = BTRFS_I(dir)->root;
4554 struct inode *inode = d_inode(dentry);
4555 struct btrfs_root *dest = BTRFS_I(inode)->root;
4556 struct btrfs_trans_handle *trans;
4557 struct btrfs_block_rsv block_rsv;
4562 * Don't allow to delete a subvolume with send in progress. This is
4563 * inside the inode lock so the error handling that has to drop the bit
4564 * again is not run concurrently.
4566 spin_lock(&dest->root_item_lock);
4567 if (dest->send_in_progress) {
4568 spin_unlock(&dest->root_item_lock);
4570 "attempt to delete subvolume %llu during send",
4571 dest->root_key.objectid);
4574 if (atomic_read(&dest->nr_swapfiles)) {
4575 spin_unlock(&dest->root_item_lock);
4577 "attempt to delete subvolume %llu with active swapfile",
4578 root->root_key.objectid);
4581 root_flags = btrfs_root_flags(&dest->root_item);
4582 btrfs_set_root_flags(&dest->root_item,
4583 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4584 spin_unlock(&dest->root_item_lock);
4586 down_write(&fs_info->subvol_sem);
4588 ret = may_destroy_subvol(dest);
4592 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4594 * One for dir inode,
4595 * two for dir entries,
4596 * two for root ref/backref.
4598 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4602 trans = btrfs_start_transaction(root, 0);
4603 if (IS_ERR(trans)) {
4604 ret = PTR_ERR(trans);
4607 trans->block_rsv = &block_rsv;
4608 trans->bytes_reserved = block_rsv.size;
4610 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4612 ret = btrfs_unlink_subvol(trans, dir, dentry);
4614 btrfs_abort_transaction(trans, ret);
4618 ret = btrfs_record_root_in_trans(trans, dest);
4620 btrfs_abort_transaction(trans, ret);
4624 memset(&dest->root_item.drop_progress, 0,
4625 sizeof(dest->root_item.drop_progress));
4626 btrfs_set_root_drop_level(&dest->root_item, 0);
4627 btrfs_set_root_refs(&dest->root_item, 0);
4629 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4630 ret = btrfs_insert_orphan_item(trans,
4632 dest->root_key.objectid);
4634 btrfs_abort_transaction(trans, ret);
4639 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4640 BTRFS_UUID_KEY_SUBVOL,
4641 dest->root_key.objectid);
4642 if (ret && ret != -ENOENT) {
4643 btrfs_abort_transaction(trans, ret);
4646 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4647 ret = btrfs_uuid_tree_remove(trans,
4648 dest->root_item.received_uuid,
4649 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4650 dest->root_key.objectid);
4651 if (ret && ret != -ENOENT) {
4652 btrfs_abort_transaction(trans, ret);
4657 free_anon_bdev(dest->anon_dev);
4660 trans->block_rsv = NULL;
4661 trans->bytes_reserved = 0;
4662 ret = btrfs_end_transaction(trans);
4663 inode->i_flags |= S_DEAD;
4665 btrfs_subvolume_release_metadata(root, &block_rsv);
4667 up_write(&fs_info->subvol_sem);
4669 spin_lock(&dest->root_item_lock);
4670 root_flags = btrfs_root_flags(&dest->root_item);
4671 btrfs_set_root_flags(&dest->root_item,
4672 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4673 spin_unlock(&dest->root_item_lock);
4675 d_invalidate(dentry);
4676 btrfs_prune_dentries(dest);
4677 ASSERT(dest->send_in_progress == 0);
4683 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4685 struct inode *inode = d_inode(dentry);
4686 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4688 struct btrfs_trans_handle *trans;
4689 u64 last_unlink_trans;
4691 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4693 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4694 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4696 "extent tree v2 doesn't support snapshot deletion yet");
4699 return btrfs_delete_subvolume(dir, dentry);
4702 trans = __unlink_start_trans(dir);
4704 return PTR_ERR(trans);
4706 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4707 err = btrfs_unlink_subvol(trans, dir, dentry);
4711 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4715 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4717 /* now the directory is empty */
4718 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4719 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4720 dentry->d_name.len);
4722 btrfs_i_size_write(BTRFS_I(inode), 0);
4724 * Propagate the last_unlink_trans value of the deleted dir to
4725 * its parent directory. This is to prevent an unrecoverable
4726 * log tree in the case we do something like this:
4728 * 2) create snapshot under dir foo
4729 * 3) delete the snapshot
4732 * 6) fsync foo or some file inside foo
4734 if (last_unlink_trans >= trans->transid)
4735 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4738 btrfs_end_transaction(trans);
4739 btrfs_btree_balance_dirty(fs_info);
4745 * btrfs_truncate_block - read, zero a chunk and write a block
4746 * @inode - inode that we're zeroing
4747 * @from - the offset to start zeroing
4748 * @len - the length to zero, 0 to zero the entire range respective to the
4750 * @front - zero up to the offset instead of from the offset on
4752 * This will find the block for the "from" offset and cow the block and zero the
4753 * part we want to zero. This is used with truncate and hole punching.
4755 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4758 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4759 struct address_space *mapping = inode->vfs_inode.i_mapping;
4760 struct extent_io_tree *io_tree = &inode->io_tree;
4761 struct btrfs_ordered_extent *ordered;
4762 struct extent_state *cached_state = NULL;
4763 struct extent_changeset *data_reserved = NULL;
4764 bool only_release_metadata = false;
4765 u32 blocksize = fs_info->sectorsize;
4766 pgoff_t index = from >> PAGE_SHIFT;
4767 unsigned offset = from & (blocksize - 1);
4769 gfp_t mask = btrfs_alloc_write_mask(mapping);
4770 size_t write_bytes = blocksize;
4775 if (IS_ALIGNED(offset, blocksize) &&
4776 (!len || IS_ALIGNED(len, blocksize)))
4779 block_start = round_down(from, blocksize);
4780 block_end = block_start + blocksize - 1;
4782 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4785 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4786 /* For nocow case, no need to reserve data space */
4787 only_release_metadata = true;
4792 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4794 if (!only_release_metadata)
4795 btrfs_free_reserved_data_space(inode, data_reserved,
4796 block_start, blocksize);
4800 page = find_or_create_page(mapping, index, mask);
4802 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4804 btrfs_delalloc_release_extents(inode, blocksize);
4808 ret = set_page_extent_mapped(page);
4812 if (!PageUptodate(page)) {
4813 ret = btrfs_read_folio(NULL, page_folio(page));
4815 if (page->mapping != mapping) {
4820 if (!PageUptodate(page)) {
4825 wait_on_page_writeback(page);
4827 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4829 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4831 unlock_extent_cached(io_tree, block_start, block_end,
4835 btrfs_start_ordered_extent(ordered, 1);
4836 btrfs_put_ordered_extent(ordered);
4840 clear_extent_bit(&inode->io_tree, block_start, block_end,
4841 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4842 0, 0, &cached_state);
4844 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4847 unlock_extent_cached(io_tree, block_start, block_end,
4852 if (offset != blocksize) {
4854 len = blocksize - offset;
4856 memzero_page(page, (block_start - page_offset(page)),
4859 memzero_page(page, (block_start - page_offset(page)) + offset,
4861 flush_dcache_page(page);
4863 btrfs_page_clear_checked(fs_info, page, block_start,
4864 block_end + 1 - block_start);
4865 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4866 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4868 if (only_release_metadata)
4869 set_extent_bit(&inode->io_tree, block_start, block_end,
4870 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4874 if (only_release_metadata)
4875 btrfs_delalloc_release_metadata(inode, blocksize, true);
4877 btrfs_delalloc_release_space(inode, data_reserved,
4878 block_start, blocksize, true);
4880 btrfs_delalloc_release_extents(inode, blocksize);
4884 if (only_release_metadata)
4885 btrfs_check_nocow_unlock(inode);
4886 extent_changeset_free(data_reserved);
4890 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4891 u64 offset, u64 len)
4893 struct btrfs_fs_info *fs_info = root->fs_info;
4894 struct btrfs_trans_handle *trans;
4895 struct btrfs_drop_extents_args drop_args = { 0 };
4899 * If NO_HOLES is enabled, we don't need to do anything.
4900 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4901 * or btrfs_update_inode() will be called, which guarantee that the next
4902 * fsync will know this inode was changed and needs to be logged.
4904 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4908 * 1 - for the one we're dropping
4909 * 1 - for the one we're adding
4910 * 1 - for updating the inode.
4912 trans = btrfs_start_transaction(root, 3);
4914 return PTR_ERR(trans);
4916 drop_args.start = offset;
4917 drop_args.end = offset + len;
4918 drop_args.drop_cache = true;
4920 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4922 btrfs_abort_transaction(trans, ret);
4923 btrfs_end_transaction(trans);
4927 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4928 offset, 0, 0, len, 0, len, 0, 0, 0);
4930 btrfs_abort_transaction(trans, ret);
4932 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4933 btrfs_update_inode(trans, root, inode);
4935 btrfs_end_transaction(trans);
4940 * This function puts in dummy file extents for the area we're creating a hole
4941 * for. So if we are truncating this file to a larger size we need to insert
4942 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4943 * the range between oldsize and size
4945 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4947 struct btrfs_root *root = inode->root;
4948 struct btrfs_fs_info *fs_info = root->fs_info;
4949 struct extent_io_tree *io_tree = &inode->io_tree;
4950 struct extent_map *em = NULL;
4951 struct extent_state *cached_state = NULL;
4952 struct extent_map_tree *em_tree = &inode->extent_tree;
4953 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4954 u64 block_end = ALIGN(size, fs_info->sectorsize);
4961 * If our size started in the middle of a block we need to zero out the
4962 * rest of the block before we expand the i_size, otherwise we could
4963 * expose stale data.
4965 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4969 if (size <= hole_start)
4972 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4974 cur_offset = hole_start;
4976 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4977 block_end - cur_offset);
4983 last_byte = min(extent_map_end(em), block_end);
4984 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4985 hole_size = last_byte - cur_offset;
4987 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4988 struct extent_map *hole_em;
4990 err = maybe_insert_hole(root, inode, cur_offset,
4995 err = btrfs_inode_set_file_extent_range(inode,
4996 cur_offset, hole_size);
5000 btrfs_drop_extent_cache(inode, cur_offset,
5001 cur_offset + hole_size - 1, 0);
5002 hole_em = alloc_extent_map();
5004 btrfs_set_inode_full_sync(inode);
5007 hole_em->start = cur_offset;
5008 hole_em->len = hole_size;
5009 hole_em->orig_start = cur_offset;
5011 hole_em->block_start = EXTENT_MAP_HOLE;
5012 hole_em->block_len = 0;
5013 hole_em->orig_block_len = 0;
5014 hole_em->ram_bytes = hole_size;
5015 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5016 hole_em->generation = fs_info->generation;
5019 write_lock(&em_tree->lock);
5020 err = add_extent_mapping(em_tree, hole_em, 1);
5021 write_unlock(&em_tree->lock);
5024 btrfs_drop_extent_cache(inode, cur_offset,
5028 free_extent_map(hole_em);
5030 err = btrfs_inode_set_file_extent_range(inode,
5031 cur_offset, hole_size);
5036 free_extent_map(em);
5038 cur_offset = last_byte;
5039 if (cur_offset >= block_end)
5042 free_extent_map(em);
5043 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5047 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5049 struct btrfs_root *root = BTRFS_I(inode)->root;
5050 struct btrfs_trans_handle *trans;
5051 loff_t oldsize = i_size_read(inode);
5052 loff_t newsize = attr->ia_size;
5053 int mask = attr->ia_valid;
5057 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5058 * special case where we need to update the times despite not having
5059 * these flags set. For all other operations the VFS set these flags
5060 * explicitly if it wants a timestamp update.
5062 if (newsize != oldsize) {
5063 inode_inc_iversion(inode);
5064 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5065 inode->i_ctime = inode->i_mtime =
5066 current_time(inode);
5069 if (newsize > oldsize) {
5071 * Don't do an expanding truncate while snapshotting is ongoing.
5072 * This is to ensure the snapshot captures a fully consistent
5073 * state of this file - if the snapshot captures this expanding
5074 * truncation, it must capture all writes that happened before
5077 btrfs_drew_write_lock(&root->snapshot_lock);
5078 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5080 btrfs_drew_write_unlock(&root->snapshot_lock);
5084 trans = btrfs_start_transaction(root, 1);
5085 if (IS_ERR(trans)) {
5086 btrfs_drew_write_unlock(&root->snapshot_lock);
5087 return PTR_ERR(trans);
5090 i_size_write(inode, newsize);
5091 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5092 pagecache_isize_extended(inode, oldsize, newsize);
5093 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5094 btrfs_drew_write_unlock(&root->snapshot_lock);
5095 btrfs_end_transaction(trans);
5097 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5099 if (btrfs_is_zoned(fs_info)) {
5100 ret = btrfs_wait_ordered_range(inode,
5101 ALIGN(newsize, fs_info->sectorsize),
5108 * We're truncating a file that used to have good data down to
5109 * zero. Make sure any new writes to the file get on disk
5113 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5114 &BTRFS_I(inode)->runtime_flags);
5116 truncate_setsize(inode, newsize);
5118 inode_dio_wait(inode);
5120 ret = btrfs_truncate(inode, newsize == oldsize);
5121 if (ret && inode->i_nlink) {
5125 * Truncate failed, so fix up the in-memory size. We
5126 * adjusted disk_i_size down as we removed extents, so
5127 * wait for disk_i_size to be stable and then update the
5128 * in-memory size to match.
5130 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5133 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5140 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5143 struct inode *inode = d_inode(dentry);
5144 struct btrfs_root *root = BTRFS_I(inode)->root;
5147 if (btrfs_root_readonly(root))
5150 err = setattr_prepare(mnt_userns, dentry, attr);
5154 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5155 err = btrfs_setsize(inode, attr);
5160 if (attr->ia_valid) {
5161 setattr_copy(mnt_userns, inode, attr);
5162 inode_inc_iversion(inode);
5163 err = btrfs_dirty_inode(inode);
5165 if (!err && attr->ia_valid & ATTR_MODE)
5166 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5173 * While truncating the inode pages during eviction, we get the VFS
5174 * calling btrfs_invalidate_folio() against each folio of the inode. This
5175 * is slow because the calls to btrfs_invalidate_folio() result in a
5176 * huge amount of calls to lock_extent_bits() and clear_extent_bit(),
5177 * which keep merging and splitting extent_state structures over and over,
5178 * wasting lots of time.
5180 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5181 * skip all those expensive operations on a per folio basis and do only
5182 * the ordered io finishing, while we release here the extent_map and
5183 * extent_state structures, without the excessive merging and splitting.
5185 static void evict_inode_truncate_pages(struct inode *inode)
5187 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5188 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5189 struct rb_node *node;
5191 ASSERT(inode->i_state & I_FREEING);
5192 truncate_inode_pages_final(&inode->i_data);
5194 write_lock(&map_tree->lock);
5195 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5196 struct extent_map *em;
5198 node = rb_first_cached(&map_tree->map);
5199 em = rb_entry(node, struct extent_map, rb_node);
5200 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5201 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5202 remove_extent_mapping(map_tree, em);
5203 free_extent_map(em);
5204 if (need_resched()) {
5205 write_unlock(&map_tree->lock);
5207 write_lock(&map_tree->lock);
5210 write_unlock(&map_tree->lock);
5213 * Keep looping until we have no more ranges in the io tree.
5214 * We can have ongoing bios started by readahead that have
5215 * their endio callback (extent_io.c:end_bio_extent_readpage)
5216 * still in progress (unlocked the pages in the bio but did not yet
5217 * unlocked the ranges in the io tree). Therefore this means some
5218 * ranges can still be locked and eviction started because before
5219 * submitting those bios, which are executed by a separate task (work
5220 * queue kthread), inode references (inode->i_count) were not taken
5221 * (which would be dropped in the end io callback of each bio).
5222 * Therefore here we effectively end up waiting for those bios and
5223 * anyone else holding locked ranges without having bumped the inode's
5224 * reference count - if we don't do it, when they access the inode's
5225 * io_tree to unlock a range it may be too late, leading to an
5226 * use-after-free issue.
5228 spin_lock(&io_tree->lock);
5229 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5230 struct extent_state *state;
5231 struct extent_state *cached_state = NULL;
5234 unsigned state_flags;
5236 node = rb_first(&io_tree->state);
5237 state = rb_entry(node, struct extent_state, rb_node);
5238 start = state->start;
5240 state_flags = state->state;
5241 spin_unlock(&io_tree->lock);
5243 lock_extent_bits(io_tree, start, end, &cached_state);
5246 * If still has DELALLOC flag, the extent didn't reach disk,
5247 * and its reserved space won't be freed by delayed_ref.
5248 * So we need to free its reserved space here.
5249 * (Refer to comment in btrfs_invalidate_folio, case 2)
5251 * Note, end is the bytenr of last byte, so we need + 1 here.
5253 if (state_flags & EXTENT_DELALLOC)
5254 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5257 clear_extent_bit(io_tree, start, end,
5258 EXTENT_LOCKED | EXTENT_DELALLOC |
5259 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5263 spin_lock(&io_tree->lock);
5265 spin_unlock(&io_tree->lock);
5268 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5269 struct btrfs_block_rsv *rsv)
5271 struct btrfs_fs_info *fs_info = root->fs_info;
5272 struct btrfs_trans_handle *trans;
5273 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5277 * Eviction should be taking place at some place safe because of our
5278 * delayed iputs. However the normal flushing code will run delayed
5279 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5281 * We reserve the delayed_refs_extra here again because we can't use
5282 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5283 * above. We reserve our extra bit here because we generate a ton of
5284 * delayed refs activity by truncating.
5286 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5287 * if we fail to make this reservation we can re-try without the
5288 * delayed_refs_extra so we can make some forward progress.
5290 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5291 BTRFS_RESERVE_FLUSH_EVICT);
5293 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5294 BTRFS_RESERVE_FLUSH_EVICT);
5297 "could not allocate space for delete; will truncate on mount");
5298 return ERR_PTR(-ENOSPC);
5300 delayed_refs_extra = 0;
5303 trans = btrfs_join_transaction(root);
5307 if (delayed_refs_extra) {
5308 trans->block_rsv = &fs_info->trans_block_rsv;
5309 trans->bytes_reserved = delayed_refs_extra;
5310 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5311 delayed_refs_extra, 1);
5316 void btrfs_evict_inode(struct inode *inode)
5318 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5319 struct btrfs_trans_handle *trans;
5320 struct btrfs_root *root = BTRFS_I(inode)->root;
5321 struct btrfs_block_rsv *rsv;
5324 trace_btrfs_inode_evict(inode);
5327 fsverity_cleanup_inode(inode);
5332 evict_inode_truncate_pages(inode);
5334 if (inode->i_nlink &&
5335 ((btrfs_root_refs(&root->root_item) != 0 &&
5336 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5337 btrfs_is_free_space_inode(BTRFS_I(inode))))
5340 if (is_bad_inode(inode))
5343 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5345 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5348 if (inode->i_nlink > 0) {
5349 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5350 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5355 * This makes sure the inode item in tree is uptodate and the space for
5356 * the inode update is released.
5358 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5363 * This drops any pending insert or delete operations we have for this
5364 * inode. We could have a delayed dir index deletion queued up, but
5365 * we're removing the inode completely so that'll be taken care of in
5368 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5370 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5373 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5376 btrfs_i_size_write(BTRFS_I(inode), 0);
5379 struct btrfs_truncate_control control = {
5380 .inode = BTRFS_I(inode),
5381 .ino = btrfs_ino(BTRFS_I(inode)),
5386 trans = evict_refill_and_join(root, rsv);
5390 trans->block_rsv = rsv;
5392 ret = btrfs_truncate_inode_items(trans, root, &control);
5393 trans->block_rsv = &fs_info->trans_block_rsv;
5394 btrfs_end_transaction(trans);
5395 btrfs_btree_balance_dirty(fs_info);
5396 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5403 * Errors here aren't a big deal, it just means we leave orphan items in
5404 * the tree. They will be cleaned up on the next mount. If the inode
5405 * number gets reused, cleanup deletes the orphan item without doing
5406 * anything, and unlink reuses the existing orphan item.
5408 * If it turns out that we are dropping too many of these, we might want
5409 * to add a mechanism for retrying these after a commit.
5411 trans = evict_refill_and_join(root, rsv);
5412 if (!IS_ERR(trans)) {
5413 trans->block_rsv = rsv;
5414 btrfs_orphan_del(trans, BTRFS_I(inode));
5415 trans->block_rsv = &fs_info->trans_block_rsv;
5416 btrfs_end_transaction(trans);
5420 btrfs_free_block_rsv(fs_info, rsv);
5423 * If we didn't successfully delete, the orphan item will still be in
5424 * the tree and we'll retry on the next mount. Again, we might also want
5425 * to retry these periodically in the future.
5427 btrfs_remove_delayed_node(BTRFS_I(inode));
5428 fsverity_cleanup_inode(inode);
5433 * Return the key found in the dir entry in the location pointer, fill @type
5434 * with BTRFS_FT_*, and return 0.
5436 * If no dir entries were found, returns -ENOENT.
5437 * If found a corrupted location in dir entry, returns -EUCLEAN.
5439 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5440 struct btrfs_key *location, u8 *type)
5442 const char *name = dentry->d_name.name;
5443 int namelen = dentry->d_name.len;
5444 struct btrfs_dir_item *di;
5445 struct btrfs_path *path;
5446 struct btrfs_root *root = BTRFS_I(dir)->root;
5449 path = btrfs_alloc_path();
5453 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5455 if (IS_ERR_OR_NULL(di)) {
5456 ret = di ? PTR_ERR(di) : -ENOENT;
5460 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5461 if (location->type != BTRFS_INODE_ITEM_KEY &&
5462 location->type != BTRFS_ROOT_ITEM_KEY) {
5464 btrfs_warn(root->fs_info,
5465 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5466 __func__, name, btrfs_ino(BTRFS_I(dir)),
5467 location->objectid, location->type, location->offset);
5470 *type = btrfs_dir_type(path->nodes[0], di);
5472 btrfs_free_path(path);
5477 * when we hit a tree root in a directory, the btrfs part of the inode
5478 * needs to be changed to reflect the root directory of the tree root. This
5479 * is kind of like crossing a mount point.
5481 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5483 struct dentry *dentry,
5484 struct btrfs_key *location,
5485 struct btrfs_root **sub_root)
5487 struct btrfs_path *path;
5488 struct btrfs_root *new_root;
5489 struct btrfs_root_ref *ref;
5490 struct extent_buffer *leaf;
5491 struct btrfs_key key;
5495 path = btrfs_alloc_path();
5502 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5503 key.type = BTRFS_ROOT_REF_KEY;
5504 key.offset = location->objectid;
5506 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5513 leaf = path->nodes[0];
5514 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5515 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5516 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5519 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5520 (unsigned long)(ref + 1),
5521 dentry->d_name.len);
5525 btrfs_release_path(path);
5527 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5528 if (IS_ERR(new_root)) {
5529 err = PTR_ERR(new_root);
5533 *sub_root = new_root;
5534 location->objectid = btrfs_root_dirid(&new_root->root_item);
5535 location->type = BTRFS_INODE_ITEM_KEY;
5536 location->offset = 0;
5539 btrfs_free_path(path);
5543 static void inode_tree_add(struct inode *inode)
5545 struct btrfs_root *root = BTRFS_I(inode)->root;
5546 struct btrfs_inode *entry;
5548 struct rb_node *parent;
5549 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5550 u64 ino = btrfs_ino(BTRFS_I(inode));
5552 if (inode_unhashed(inode))
5555 spin_lock(&root->inode_lock);
5556 p = &root->inode_tree.rb_node;
5559 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5561 if (ino < btrfs_ino(entry))
5562 p = &parent->rb_left;
5563 else if (ino > btrfs_ino(entry))
5564 p = &parent->rb_right;
5566 WARN_ON(!(entry->vfs_inode.i_state &
5567 (I_WILL_FREE | I_FREEING)));
5568 rb_replace_node(parent, new, &root->inode_tree);
5569 RB_CLEAR_NODE(parent);
5570 spin_unlock(&root->inode_lock);
5574 rb_link_node(new, parent, p);
5575 rb_insert_color(new, &root->inode_tree);
5576 spin_unlock(&root->inode_lock);
5579 static void inode_tree_del(struct btrfs_inode *inode)
5581 struct btrfs_root *root = inode->root;
5584 spin_lock(&root->inode_lock);
5585 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5586 rb_erase(&inode->rb_node, &root->inode_tree);
5587 RB_CLEAR_NODE(&inode->rb_node);
5588 empty = RB_EMPTY_ROOT(&root->inode_tree);
5590 spin_unlock(&root->inode_lock);
5592 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5593 spin_lock(&root->inode_lock);
5594 empty = RB_EMPTY_ROOT(&root->inode_tree);
5595 spin_unlock(&root->inode_lock);
5597 btrfs_add_dead_root(root);
5602 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5604 struct btrfs_iget_args *args = p;
5606 inode->i_ino = args->ino;
5607 BTRFS_I(inode)->location.objectid = args->ino;
5608 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5609 BTRFS_I(inode)->location.offset = 0;
5610 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5611 BUG_ON(args->root && !BTRFS_I(inode)->root);
5615 static int btrfs_find_actor(struct inode *inode, void *opaque)
5617 struct btrfs_iget_args *args = opaque;
5619 return args->ino == BTRFS_I(inode)->location.objectid &&
5620 args->root == BTRFS_I(inode)->root;
5623 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5624 struct btrfs_root *root)
5626 struct inode *inode;
5627 struct btrfs_iget_args args;
5628 unsigned long hashval = btrfs_inode_hash(ino, root);
5633 inode = iget5_locked(s, hashval, btrfs_find_actor,
5634 btrfs_init_locked_inode,
5640 * Get an inode object given its inode number and corresponding root.
5641 * Path can be preallocated to prevent recursing back to iget through
5642 * allocator. NULL is also valid but may require an additional allocation
5645 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5646 struct btrfs_root *root, struct btrfs_path *path)
5648 struct inode *inode;
5650 inode = btrfs_iget_locked(s, ino, root);
5652 return ERR_PTR(-ENOMEM);
5654 if (inode->i_state & I_NEW) {
5657 ret = btrfs_read_locked_inode(inode, path);
5659 inode_tree_add(inode);
5660 unlock_new_inode(inode);
5664 * ret > 0 can come from btrfs_search_slot called by
5665 * btrfs_read_locked_inode, this means the inode item
5670 inode = ERR_PTR(ret);
5677 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5679 return btrfs_iget_path(s, ino, root, NULL);
5682 static struct inode *new_simple_dir(struct super_block *s,
5683 struct btrfs_key *key,
5684 struct btrfs_root *root)
5686 struct inode *inode = new_inode(s);
5689 return ERR_PTR(-ENOMEM);
5691 BTRFS_I(inode)->root = btrfs_grab_root(root);
5692 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5693 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5695 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5697 * We only need lookup, the rest is read-only and there's no inode
5698 * associated with the dentry
5700 inode->i_op = &simple_dir_inode_operations;
5701 inode->i_opflags &= ~IOP_XATTR;
5702 inode->i_fop = &simple_dir_operations;
5703 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5704 inode->i_mtime = current_time(inode);
5705 inode->i_atime = inode->i_mtime;
5706 inode->i_ctime = inode->i_mtime;
5707 BTRFS_I(inode)->i_otime = inode->i_mtime;
5712 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5713 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5714 static_assert(BTRFS_FT_DIR == FT_DIR);
5715 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5716 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5717 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5718 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5719 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5721 static inline u8 btrfs_inode_type(struct inode *inode)
5723 return fs_umode_to_ftype(inode->i_mode);
5726 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5728 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5729 struct inode *inode;
5730 struct btrfs_root *root = BTRFS_I(dir)->root;
5731 struct btrfs_root *sub_root = root;
5732 struct btrfs_key location;
5736 if (dentry->d_name.len > BTRFS_NAME_LEN)
5737 return ERR_PTR(-ENAMETOOLONG);
5739 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5741 return ERR_PTR(ret);
5743 if (location.type == BTRFS_INODE_ITEM_KEY) {
5744 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5748 /* Do extra check against inode mode with di_type */
5749 if (btrfs_inode_type(inode) != di_type) {
5751 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5752 inode->i_mode, btrfs_inode_type(inode),
5755 return ERR_PTR(-EUCLEAN);
5760 ret = fixup_tree_root_location(fs_info, dir, dentry,
5761 &location, &sub_root);
5764 inode = ERR_PTR(ret);
5766 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5768 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5770 if (root != sub_root)
5771 btrfs_put_root(sub_root);
5773 if (!IS_ERR(inode) && root != sub_root) {
5774 down_read(&fs_info->cleanup_work_sem);
5775 if (!sb_rdonly(inode->i_sb))
5776 ret = btrfs_orphan_cleanup(sub_root);
5777 up_read(&fs_info->cleanup_work_sem);
5780 inode = ERR_PTR(ret);
5787 static int btrfs_dentry_delete(const struct dentry *dentry)
5789 struct btrfs_root *root;
5790 struct inode *inode = d_inode(dentry);
5792 if (!inode && !IS_ROOT(dentry))
5793 inode = d_inode(dentry->d_parent);
5796 root = BTRFS_I(inode)->root;
5797 if (btrfs_root_refs(&root->root_item) == 0)
5800 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5806 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5809 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5811 if (inode == ERR_PTR(-ENOENT))
5813 return d_splice_alias(inode, dentry);
5817 * All this infrastructure exists because dir_emit can fault, and we are holding
5818 * the tree lock when doing readdir. For now just allocate a buffer and copy
5819 * our information into that, and then dir_emit from the buffer. This is
5820 * similar to what NFS does, only we don't keep the buffer around in pagecache
5821 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5822 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5825 static int btrfs_opendir(struct inode *inode, struct file *file)
5827 struct btrfs_file_private *private;
5829 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5832 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5833 if (!private->filldir_buf) {
5837 file->private_data = private;
5848 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5851 struct dir_entry *entry = addr;
5852 char *name = (char *)(entry + 1);
5854 ctx->pos = get_unaligned(&entry->offset);
5855 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5856 get_unaligned(&entry->ino),
5857 get_unaligned(&entry->type)))
5859 addr += sizeof(struct dir_entry) +
5860 get_unaligned(&entry->name_len);
5866 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5868 struct inode *inode = file_inode(file);
5869 struct btrfs_root *root = BTRFS_I(inode)->root;
5870 struct btrfs_file_private *private = file->private_data;
5871 struct btrfs_dir_item *di;
5872 struct btrfs_key key;
5873 struct btrfs_key found_key;
5874 struct btrfs_path *path;
5876 struct list_head ins_list;
5877 struct list_head del_list;
5884 struct btrfs_key location;
5886 if (!dir_emit_dots(file, ctx))
5889 path = btrfs_alloc_path();
5893 addr = private->filldir_buf;
5894 path->reada = READA_FORWARD;
5896 INIT_LIST_HEAD(&ins_list);
5897 INIT_LIST_HEAD(&del_list);
5898 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5901 key.type = BTRFS_DIR_INDEX_KEY;
5902 key.offset = ctx->pos;
5903 key.objectid = btrfs_ino(BTRFS_I(inode));
5905 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5906 struct dir_entry *entry;
5907 struct extent_buffer *leaf = path->nodes[0];
5909 if (found_key.objectid != key.objectid)
5911 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5913 if (found_key.offset < ctx->pos)
5915 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5917 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5918 name_len = btrfs_dir_name_len(leaf, di);
5919 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5921 btrfs_release_path(path);
5922 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5925 addr = private->filldir_buf;
5932 put_unaligned(name_len, &entry->name_len);
5933 name_ptr = (char *)(entry + 1);
5934 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5936 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5938 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5939 put_unaligned(location.objectid, &entry->ino);
5940 put_unaligned(found_key.offset, &entry->offset);
5942 addr += sizeof(struct dir_entry) + name_len;
5943 total_len += sizeof(struct dir_entry) + name_len;
5945 /* Catch error encountered during iteration */
5949 btrfs_release_path(path);
5951 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5955 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5960 * Stop new entries from being returned after we return the last
5963 * New directory entries are assigned a strictly increasing
5964 * offset. This means that new entries created during readdir
5965 * are *guaranteed* to be seen in the future by that readdir.
5966 * This has broken buggy programs which operate on names as
5967 * they're returned by readdir. Until we re-use freed offsets
5968 * we have this hack to stop new entries from being returned
5969 * under the assumption that they'll never reach this huge
5972 * This is being careful not to overflow 32bit loff_t unless the
5973 * last entry requires it because doing so has broken 32bit apps
5976 if (ctx->pos >= INT_MAX)
5977 ctx->pos = LLONG_MAX;
5984 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5985 btrfs_free_path(path);
5990 * This is somewhat expensive, updating the tree every time the
5991 * inode changes. But, it is most likely to find the inode in cache.
5992 * FIXME, needs more benchmarking...there are no reasons other than performance
5993 * to keep or drop this code.
5995 static int btrfs_dirty_inode(struct inode *inode)
5997 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5998 struct btrfs_root *root = BTRFS_I(inode)->root;
5999 struct btrfs_trans_handle *trans;
6002 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6005 trans = btrfs_join_transaction(root);
6007 return PTR_ERR(trans);
6009 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6010 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6011 /* whoops, lets try again with the full transaction */
6012 btrfs_end_transaction(trans);
6013 trans = btrfs_start_transaction(root, 1);
6015 return PTR_ERR(trans);
6017 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6019 btrfs_end_transaction(trans);
6020 if (BTRFS_I(inode)->delayed_node)
6021 btrfs_balance_delayed_items(fs_info);
6027 * This is a copy of file_update_time. We need this so we can return error on
6028 * ENOSPC for updating the inode in the case of file write and mmap writes.
6030 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6033 struct btrfs_root *root = BTRFS_I(inode)->root;
6034 bool dirty = flags & ~S_VERSION;
6036 if (btrfs_root_readonly(root))
6039 if (flags & S_VERSION)
6040 dirty |= inode_maybe_inc_iversion(inode, dirty);
6041 if (flags & S_CTIME)
6042 inode->i_ctime = *now;
6043 if (flags & S_MTIME)
6044 inode->i_mtime = *now;
6045 if (flags & S_ATIME)
6046 inode->i_atime = *now;
6047 return dirty ? btrfs_dirty_inode(inode) : 0;
6051 * find the highest existing sequence number in a directory
6052 * and then set the in-memory index_cnt variable to reflect
6053 * free sequence numbers
6055 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6057 struct btrfs_root *root = inode->root;
6058 struct btrfs_key key, found_key;
6059 struct btrfs_path *path;
6060 struct extent_buffer *leaf;
6063 key.objectid = btrfs_ino(inode);
6064 key.type = BTRFS_DIR_INDEX_KEY;
6065 key.offset = (u64)-1;
6067 path = btrfs_alloc_path();
6071 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6074 /* FIXME: we should be able to handle this */
6079 if (path->slots[0] == 0) {
6080 inode->index_cnt = BTRFS_DIR_START_INDEX;
6086 leaf = path->nodes[0];
6087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6089 if (found_key.objectid != btrfs_ino(inode) ||
6090 found_key.type != BTRFS_DIR_INDEX_KEY) {
6091 inode->index_cnt = BTRFS_DIR_START_INDEX;
6095 inode->index_cnt = found_key.offset + 1;
6097 btrfs_free_path(path);
6102 * helper to find a free sequence number in a given directory. This current
6103 * code is very simple, later versions will do smarter things in the btree
6105 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6109 if (dir->index_cnt == (u64)-1) {
6110 ret = btrfs_inode_delayed_dir_index_count(dir);
6112 ret = btrfs_set_inode_index_count(dir);
6118 *index = dir->index_cnt;
6124 static int btrfs_insert_inode_locked(struct inode *inode)
6126 struct btrfs_iget_args args;
6128 args.ino = BTRFS_I(inode)->location.objectid;
6129 args.root = BTRFS_I(inode)->root;
6131 return insert_inode_locked4(inode,
6132 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6133 btrfs_find_actor, &args);
6136 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6137 unsigned int *trans_num_items)
6139 struct inode *dir = args->dir;
6140 struct inode *inode = args->inode;
6143 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6147 /* 1 to add inode item */
6148 *trans_num_items = 1;
6149 /* 1 to add compression property */
6150 if (BTRFS_I(dir)->prop_compress)
6151 (*trans_num_items)++;
6152 /* 1 to add default ACL xattr */
6153 if (args->default_acl)
6154 (*trans_num_items)++;
6155 /* 1 to add access ACL xattr */
6157 (*trans_num_items)++;
6158 #ifdef CONFIG_SECURITY
6159 /* 1 to add LSM xattr */
6160 if (dir->i_security)
6161 (*trans_num_items)++;
6164 /* 1 to add orphan item */
6165 (*trans_num_items)++;
6169 * 1 to add dir index
6170 * 1 to update parent inode item
6172 * No need for 1 unit for the inode ref item because it is
6173 * inserted in a batch together with the inode item at
6174 * btrfs_create_new_inode().
6176 *trans_num_items += 3;
6181 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6183 posix_acl_release(args->acl);
6184 posix_acl_release(args->default_acl);
6188 * Inherit flags from the parent inode.
6190 * Currently only the compression flags and the cow flags are inherited.
6192 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6196 flags = BTRFS_I(dir)->flags;
6198 if (flags & BTRFS_INODE_NOCOMPRESS) {
6199 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6200 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6201 } else if (flags & BTRFS_INODE_COMPRESS) {
6202 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6203 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6206 if (flags & BTRFS_INODE_NODATACOW) {
6207 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6208 if (S_ISREG(inode->i_mode))
6209 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6212 btrfs_sync_inode_flags_to_i_flags(inode);
6215 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6216 struct btrfs_new_inode_args *args)
6218 struct inode *dir = args->dir;
6219 struct inode *inode = args->inode;
6220 const char *name = args->orphan ? NULL : args->dentry->d_name.name;
6221 int name_len = args->orphan ? 0 : args->dentry->d_name.len;
6222 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6223 struct btrfs_root *root;
6224 struct btrfs_inode_item *inode_item;
6225 struct btrfs_key *location;
6226 struct btrfs_path *path;
6228 struct btrfs_inode_ref *ref;
6229 struct btrfs_key key[2];
6231 struct btrfs_item_batch batch;
6235 path = btrfs_alloc_path();
6240 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6241 root = BTRFS_I(inode)->root;
6243 ret = btrfs_get_free_objectid(root, &objectid);
6246 inode->i_ino = objectid;
6250 * O_TMPFILE, set link count to 0, so that after this point, we
6251 * fill in an inode item with the correct link count.
6253 set_nlink(inode, 0);
6255 trace_btrfs_inode_request(dir);
6257 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6261 /* index_cnt is ignored for everything but a dir. */
6262 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6263 BTRFS_I(inode)->generation = trans->transid;
6264 inode->i_generation = BTRFS_I(inode)->generation;
6267 * Subvolumes don't inherit flags from their parent directory.
6268 * Originally this was probably by accident, but we probably can't
6269 * change it now without compatibility issues.
6272 btrfs_inherit_iflags(inode, dir);
6274 if (S_ISREG(inode->i_mode)) {
6275 if (btrfs_test_opt(fs_info, NODATASUM))
6276 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6277 if (btrfs_test_opt(fs_info, NODATACOW))
6278 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6279 BTRFS_INODE_NODATASUM;
6282 location = &BTRFS_I(inode)->location;
6283 location->objectid = objectid;
6284 location->offset = 0;
6285 location->type = BTRFS_INODE_ITEM_KEY;
6287 ret = btrfs_insert_inode_locked(inode);
6290 BTRFS_I(dir)->index_cnt--;
6295 * We could have gotten an inode number from somebody who was fsynced
6296 * and then removed in this same transaction, so let's just set full
6297 * sync since it will be a full sync anyway and this will blow away the
6298 * old info in the log.
6300 btrfs_set_inode_full_sync(BTRFS_I(inode));
6302 key[0].objectid = objectid;
6303 key[0].type = BTRFS_INODE_ITEM_KEY;
6306 sizes[0] = sizeof(struct btrfs_inode_item);
6308 if (!args->orphan) {
6310 * Start new inodes with an inode_ref. This is slightly more
6311 * efficient for small numbers of hard links since they will
6312 * be packed into one item. Extended refs will kick in if we
6313 * add more hard links than can fit in the ref item.
6315 key[1].objectid = objectid;
6316 key[1].type = BTRFS_INODE_REF_KEY;
6318 key[1].offset = objectid;
6319 sizes[1] = 2 + sizeof(*ref);
6321 key[1].offset = btrfs_ino(BTRFS_I(dir));
6322 sizes[1] = name_len + sizeof(*ref);
6326 batch.keys = &key[0];
6327 batch.data_sizes = &sizes[0];
6328 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6329 batch.nr = args->orphan ? 1 : 2;
6330 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6332 btrfs_abort_transaction(trans, ret);
6336 inode->i_mtime = current_time(inode);
6337 inode->i_atime = inode->i_mtime;
6338 inode->i_ctime = inode->i_mtime;
6339 BTRFS_I(inode)->i_otime = inode->i_mtime;
6342 * We're going to fill the inode item now, so at this point the inode
6343 * must be fully initialized.
6346 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6347 struct btrfs_inode_item);
6348 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6349 sizeof(*inode_item));
6350 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6352 if (!args->orphan) {
6353 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6354 struct btrfs_inode_ref);
6355 ptr = (unsigned long)(ref + 1);
6357 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6358 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6359 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6361 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6362 btrfs_set_inode_ref_index(path->nodes[0], ref,
6363 BTRFS_I(inode)->dir_index);
6364 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6368 btrfs_mark_buffer_dirty(path->nodes[0]);
6369 btrfs_release_path(path);
6372 struct inode *parent;
6375 * Subvolumes inherit properties from their parent subvolume,
6376 * not the directory they were created in.
6378 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6379 BTRFS_I(dir)->root);
6380 if (IS_ERR(parent)) {
6381 ret = PTR_ERR(parent);
6383 ret = btrfs_inode_inherit_props(trans, inode, parent);
6387 ret = btrfs_inode_inherit_props(trans, inode, dir);
6391 "error inheriting props for ino %llu (root %llu): %d",
6392 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6397 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6400 if (!args->subvol) {
6401 ret = btrfs_init_inode_security(trans, args);
6403 btrfs_abort_transaction(trans, ret);
6408 inode_tree_add(inode);
6410 trace_btrfs_inode_new(inode);
6411 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6413 btrfs_update_root_times(trans, root);
6416 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6418 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6419 name_len, 0, BTRFS_I(inode)->dir_index);
6422 btrfs_abort_transaction(trans, ret);
6431 * discard_new_inode() calls iput(), but the caller owns the reference
6435 discard_new_inode(inode);
6437 btrfs_free_path(path);
6442 * utility function to add 'inode' into 'parent_inode' with
6443 * a give name and a given sequence number.
6444 * if 'add_backref' is true, also insert a backref from the
6445 * inode to the parent directory.
6447 int btrfs_add_link(struct btrfs_trans_handle *trans,
6448 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6449 const char *name, int name_len, int add_backref, u64 index)
6452 struct btrfs_key key;
6453 struct btrfs_root *root = parent_inode->root;
6454 u64 ino = btrfs_ino(inode);
6455 u64 parent_ino = btrfs_ino(parent_inode);
6457 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6458 memcpy(&key, &inode->root->root_key, sizeof(key));
6461 key.type = BTRFS_INODE_ITEM_KEY;
6465 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6466 ret = btrfs_add_root_ref(trans, key.objectid,
6467 root->root_key.objectid, parent_ino,
6468 index, name, name_len);
6469 } else if (add_backref) {
6470 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6474 /* Nothing to clean up yet */
6478 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6479 btrfs_inode_type(&inode->vfs_inode), index);
6480 if (ret == -EEXIST || ret == -EOVERFLOW)
6483 btrfs_abort_transaction(trans, ret);
6487 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6489 inode_inc_iversion(&parent_inode->vfs_inode);
6491 * If we are replaying a log tree, we do not want to update the mtime
6492 * and ctime of the parent directory with the current time, since the
6493 * log replay procedure is responsible for setting them to their correct
6494 * values (the ones it had when the fsync was done).
6496 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6497 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6499 parent_inode->vfs_inode.i_mtime = now;
6500 parent_inode->vfs_inode.i_ctime = now;
6502 ret = btrfs_update_inode(trans, root, parent_inode);
6504 btrfs_abort_transaction(trans, ret);
6508 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6511 err = btrfs_del_root_ref(trans, key.objectid,
6512 root->root_key.objectid, parent_ino,
6513 &local_index, name, name_len);
6515 btrfs_abort_transaction(trans, err);
6516 } else if (add_backref) {
6520 err = btrfs_del_inode_ref(trans, root, name, name_len,
6521 ino, parent_ino, &local_index);
6523 btrfs_abort_transaction(trans, err);
6526 /* Return the original error code */
6530 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6531 struct inode *inode)
6533 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6534 struct btrfs_root *root = BTRFS_I(dir)->root;
6535 struct btrfs_new_inode_args new_inode_args = {
6540 unsigned int trans_num_items;
6541 struct btrfs_trans_handle *trans;
6544 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6548 trans = btrfs_start_transaction(root, trans_num_items);
6549 if (IS_ERR(trans)) {
6550 err = PTR_ERR(trans);
6551 goto out_new_inode_args;
6554 err = btrfs_create_new_inode(trans, &new_inode_args);
6556 d_instantiate_new(dentry, inode);
6558 btrfs_end_transaction(trans);
6559 btrfs_btree_balance_dirty(fs_info);
6561 btrfs_new_inode_args_destroy(&new_inode_args);
6568 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6569 struct dentry *dentry, umode_t mode, dev_t rdev)
6571 struct inode *inode;
6573 inode = new_inode(dir->i_sb);
6576 inode_init_owner(mnt_userns, inode, dir, mode);
6577 inode->i_op = &btrfs_special_inode_operations;
6578 init_special_inode(inode, inode->i_mode, rdev);
6579 return btrfs_create_common(dir, dentry, inode);
6582 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6583 struct dentry *dentry, umode_t mode, bool excl)
6585 struct inode *inode;
6587 inode = new_inode(dir->i_sb);
6590 inode_init_owner(mnt_userns, inode, dir, mode);
6591 inode->i_fop = &btrfs_file_operations;
6592 inode->i_op = &btrfs_file_inode_operations;
6593 inode->i_mapping->a_ops = &btrfs_aops;
6594 return btrfs_create_common(dir, dentry, inode);
6597 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6598 struct dentry *dentry)
6600 struct btrfs_trans_handle *trans = NULL;
6601 struct btrfs_root *root = BTRFS_I(dir)->root;
6602 struct inode *inode = d_inode(old_dentry);
6603 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6608 /* do not allow sys_link's with other subvols of the same device */
6609 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6612 if (inode->i_nlink >= BTRFS_LINK_MAX)
6615 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6620 * 2 items for inode and inode ref
6621 * 2 items for dir items
6622 * 1 item for parent inode
6623 * 1 item for orphan item deletion if O_TMPFILE
6625 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6626 if (IS_ERR(trans)) {
6627 err = PTR_ERR(trans);
6632 /* There are several dir indexes for this inode, clear the cache. */
6633 BTRFS_I(inode)->dir_index = 0ULL;
6635 inode_inc_iversion(inode);
6636 inode->i_ctime = current_time(inode);
6638 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6640 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6641 dentry->d_name.name, dentry->d_name.len, 1, index);
6646 struct dentry *parent = dentry->d_parent;
6648 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6651 if (inode->i_nlink == 1) {
6653 * If new hard link count is 1, it's a file created
6654 * with open(2) O_TMPFILE flag.
6656 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6660 d_instantiate(dentry, inode);
6661 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6666 btrfs_end_transaction(trans);
6668 inode_dec_link_count(inode);
6671 btrfs_btree_balance_dirty(fs_info);
6675 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6676 struct dentry *dentry, umode_t mode)
6678 struct inode *inode;
6680 inode = new_inode(dir->i_sb);
6683 inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode);
6684 inode->i_op = &btrfs_dir_inode_operations;
6685 inode->i_fop = &btrfs_dir_file_operations;
6686 return btrfs_create_common(dir, dentry, inode);
6689 static noinline int uncompress_inline(struct btrfs_path *path,
6691 size_t pg_offset, u64 extent_offset,
6692 struct btrfs_file_extent_item *item)
6695 struct extent_buffer *leaf = path->nodes[0];
6698 unsigned long inline_size;
6702 WARN_ON(pg_offset != 0);
6703 compress_type = btrfs_file_extent_compression(leaf, item);
6704 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6705 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6706 tmp = kmalloc(inline_size, GFP_NOFS);
6709 ptr = btrfs_file_extent_inline_start(item);
6711 read_extent_buffer(leaf, tmp, ptr, inline_size);
6713 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6714 ret = btrfs_decompress(compress_type, tmp, page,
6715 extent_offset, inline_size, max_size);
6718 * decompression code contains a memset to fill in any space between the end
6719 * of the uncompressed data and the end of max_size in case the decompressed
6720 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6721 * the end of an inline extent and the beginning of the next block, so we
6722 * cover that region here.
6725 if (max_size + pg_offset < PAGE_SIZE)
6726 memzero_page(page, pg_offset + max_size,
6727 PAGE_SIZE - max_size - pg_offset);
6733 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6734 * @inode: file to search in
6735 * @page: page to read extent data into if the extent is inline
6736 * @pg_offset: offset into @page to copy to
6737 * @start: file offset
6738 * @len: length of range starting at @start
6740 * This returns the first &struct extent_map which overlaps with the given
6741 * range, reading it from the B-tree and caching it if necessary. Note that
6742 * there may be more extents which overlap the given range after the returned
6745 * If @page is not NULL and the extent is inline, this also reads the extent
6746 * data directly into the page and marks the extent up to date in the io_tree.
6748 * Return: ERR_PTR on error, non-NULL extent_map on success.
6750 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6751 struct page *page, size_t pg_offset,
6754 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6756 u64 extent_start = 0;
6758 u64 objectid = btrfs_ino(inode);
6759 int extent_type = -1;
6760 struct btrfs_path *path = NULL;
6761 struct btrfs_root *root = inode->root;
6762 struct btrfs_file_extent_item *item;
6763 struct extent_buffer *leaf;
6764 struct btrfs_key found_key;
6765 struct extent_map *em = NULL;
6766 struct extent_map_tree *em_tree = &inode->extent_tree;
6767 struct extent_io_tree *io_tree = &inode->io_tree;
6769 read_lock(&em_tree->lock);
6770 em = lookup_extent_mapping(em_tree, start, len);
6771 read_unlock(&em_tree->lock);
6774 if (em->start > start || em->start + em->len <= start)
6775 free_extent_map(em);
6776 else if (em->block_start == EXTENT_MAP_INLINE && page)
6777 free_extent_map(em);
6781 em = alloc_extent_map();
6786 em->start = EXTENT_MAP_HOLE;
6787 em->orig_start = EXTENT_MAP_HOLE;
6789 em->block_len = (u64)-1;
6791 path = btrfs_alloc_path();
6797 /* Chances are we'll be called again, so go ahead and do readahead */
6798 path->reada = READA_FORWARD;
6801 * The same explanation in load_free_space_cache applies here as well,
6802 * we only read when we're loading the free space cache, and at that
6803 * point the commit_root has everything we need.
6805 if (btrfs_is_free_space_inode(inode)) {
6806 path->search_commit_root = 1;
6807 path->skip_locking = 1;
6810 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6813 } else if (ret > 0) {
6814 if (path->slots[0] == 0)
6820 leaf = path->nodes[0];
6821 item = btrfs_item_ptr(leaf, path->slots[0],
6822 struct btrfs_file_extent_item);
6823 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6824 if (found_key.objectid != objectid ||
6825 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6827 * If we backup past the first extent we want to move forward
6828 * and see if there is an extent in front of us, otherwise we'll
6829 * say there is a hole for our whole search range which can
6836 extent_type = btrfs_file_extent_type(leaf, item);
6837 extent_start = found_key.offset;
6838 extent_end = btrfs_file_extent_end(path);
6839 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6840 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6841 /* Only regular file could have regular/prealloc extent */
6842 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6845 "regular/prealloc extent found for non-regular inode %llu",
6849 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6851 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6852 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6857 if (start >= extent_end) {
6859 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6860 ret = btrfs_next_leaf(root, path);
6866 leaf = path->nodes[0];
6868 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6869 if (found_key.objectid != objectid ||
6870 found_key.type != BTRFS_EXTENT_DATA_KEY)
6872 if (start + len <= found_key.offset)
6874 if (start > found_key.offset)
6877 /* New extent overlaps with existing one */
6879 em->orig_start = start;
6880 em->len = found_key.offset - start;
6881 em->block_start = EXTENT_MAP_HOLE;
6885 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6887 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6888 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6890 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6894 size_t extent_offset;
6900 size = btrfs_file_extent_ram_bytes(leaf, item);
6901 extent_offset = page_offset(page) + pg_offset - extent_start;
6902 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6903 size - extent_offset);
6904 em->start = extent_start + extent_offset;
6905 em->len = ALIGN(copy_size, fs_info->sectorsize);
6906 em->orig_block_len = em->len;
6907 em->orig_start = em->start;
6908 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6910 if (!PageUptodate(page)) {
6911 if (btrfs_file_extent_compression(leaf, item) !=
6912 BTRFS_COMPRESS_NONE) {
6913 ret = uncompress_inline(path, page, pg_offset,
6914 extent_offset, item);
6918 map = kmap_local_page(page);
6919 read_extent_buffer(leaf, map + pg_offset, ptr,
6921 if (pg_offset + copy_size < PAGE_SIZE) {
6922 memset(map + pg_offset + copy_size, 0,
6923 PAGE_SIZE - pg_offset -
6928 flush_dcache_page(page);
6930 set_extent_uptodate(io_tree, em->start,
6931 extent_map_end(em) - 1, NULL, GFP_NOFS);
6936 em->orig_start = start;
6938 em->block_start = EXTENT_MAP_HOLE;
6941 btrfs_release_path(path);
6942 if (em->start > start || extent_map_end(em) <= start) {
6944 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6945 em->start, em->len, start, len);
6950 write_lock(&em_tree->lock);
6951 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6952 write_unlock(&em_tree->lock);
6954 btrfs_free_path(path);
6956 trace_btrfs_get_extent(root, inode, em);
6959 free_extent_map(em);
6960 return ERR_PTR(ret);
6965 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6968 struct extent_map *em;
6969 struct extent_map *hole_em = NULL;
6970 u64 delalloc_start = start;
6976 em = btrfs_get_extent(inode, NULL, 0, start, len);
6980 * If our em maps to:
6982 * - a pre-alloc extent,
6983 * there might actually be delalloc bytes behind it.
6985 if (em->block_start != EXTENT_MAP_HOLE &&
6986 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6991 /* check to see if we've wrapped (len == -1 or similar) */
7000 /* ok, we didn't find anything, lets look for delalloc */
7001 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7002 end, len, EXTENT_DELALLOC, 1);
7003 delalloc_end = delalloc_start + delalloc_len;
7004 if (delalloc_end < delalloc_start)
7005 delalloc_end = (u64)-1;
7008 * We didn't find anything useful, return the original results from
7011 if (delalloc_start > end || delalloc_end <= start) {
7018 * Adjust the delalloc_start to make sure it doesn't go backwards from
7019 * the start they passed in
7021 delalloc_start = max(start, delalloc_start);
7022 delalloc_len = delalloc_end - delalloc_start;
7024 if (delalloc_len > 0) {
7027 const u64 hole_end = extent_map_end(hole_em);
7029 em = alloc_extent_map();
7037 * When btrfs_get_extent can't find anything it returns one
7040 * Make sure what it found really fits our range, and adjust to
7041 * make sure it is based on the start from the caller
7043 if (hole_end <= start || hole_em->start > end) {
7044 free_extent_map(hole_em);
7047 hole_start = max(hole_em->start, start);
7048 hole_len = hole_end - hole_start;
7051 if (hole_em && delalloc_start > hole_start) {
7053 * Our hole starts before our delalloc, so we have to
7054 * return just the parts of the hole that go until the
7057 em->len = min(hole_len, delalloc_start - hole_start);
7058 em->start = hole_start;
7059 em->orig_start = hole_start;
7061 * Don't adjust block start at all, it is fixed at
7064 em->block_start = hole_em->block_start;
7065 em->block_len = hole_len;
7066 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7067 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7070 * Hole is out of passed range or it starts after
7073 em->start = delalloc_start;
7074 em->len = delalloc_len;
7075 em->orig_start = delalloc_start;
7076 em->block_start = EXTENT_MAP_DELALLOC;
7077 em->block_len = delalloc_len;
7084 free_extent_map(hole_em);
7086 free_extent_map(em);
7087 return ERR_PTR(err);
7092 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7095 const u64 orig_start,
7096 const u64 block_start,
7097 const u64 block_len,
7098 const u64 orig_block_len,
7099 const u64 ram_bytes,
7102 struct extent_map *em = NULL;
7105 if (type != BTRFS_ORDERED_NOCOW) {
7106 em = create_io_em(inode, start, len, orig_start, block_start,
7107 block_len, orig_block_len, ram_bytes,
7108 BTRFS_COMPRESS_NONE, /* compress_type */
7113 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7116 (1 << BTRFS_ORDERED_DIRECT),
7117 BTRFS_COMPRESS_NONE);
7120 free_extent_map(em);
7121 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7130 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7133 struct btrfs_root *root = inode->root;
7134 struct btrfs_fs_info *fs_info = root->fs_info;
7135 struct extent_map *em;
7136 struct btrfs_key ins;
7140 alloc_hint = get_extent_allocation_hint(inode, start, len);
7141 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7142 0, alloc_hint, &ins, 1, 1);
7144 return ERR_PTR(ret);
7146 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7147 ins.objectid, ins.offset, ins.offset,
7148 ins.offset, BTRFS_ORDERED_REGULAR);
7149 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7151 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7157 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7159 struct btrfs_block_group *block_group;
7160 bool readonly = false;
7162 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7163 if (!block_group || block_group->ro)
7166 btrfs_put_block_group(block_group);
7171 * Check if we can do nocow write into the range [@offset, @offset + @len)
7173 * @offset: File offset
7174 * @len: The length to write, will be updated to the nocow writeable
7176 * @orig_start: (optional) Return the original file offset of the file extent
7177 * @orig_len: (optional) Return the original on-disk length of the file extent
7178 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7179 * @strict: if true, omit optimizations that might force us into unnecessary
7180 * cow. e.g., don't trust generation number.
7183 * >0 and update @len if we can do nocow write
7184 * 0 if we can't do nocow write
7185 * <0 if error happened
7187 * NOTE: This only checks the file extents, caller is responsible to wait for
7188 * any ordered extents.
7190 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7191 u64 *orig_start, u64 *orig_block_len,
7192 u64 *ram_bytes, bool strict)
7194 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7195 struct can_nocow_file_extent_args nocow_args = { 0 };
7196 struct btrfs_path *path;
7198 struct extent_buffer *leaf;
7199 struct btrfs_root *root = BTRFS_I(inode)->root;
7200 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7201 struct btrfs_file_extent_item *fi;
7202 struct btrfs_key key;
7205 path = btrfs_alloc_path();
7209 ret = btrfs_lookup_file_extent(NULL, root, path,
7210 btrfs_ino(BTRFS_I(inode)), offset, 0);
7215 if (path->slots[0] == 0) {
7216 /* can't find the item, must cow */
7223 leaf = path->nodes[0];
7224 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7225 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7226 key.type != BTRFS_EXTENT_DATA_KEY) {
7227 /* not our file or wrong item type, must cow */
7231 if (key.offset > offset) {
7232 /* Wrong offset, must cow */
7236 if (btrfs_file_extent_end(path) <= offset)
7239 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7240 found_type = btrfs_file_extent_type(leaf, fi);
7242 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7244 nocow_args.start = offset;
7245 nocow_args.end = offset + *len - 1;
7246 nocow_args.strict = strict;
7247 nocow_args.free_path = true;
7249 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7250 /* can_nocow_file_extent() has freed the path. */
7254 /* Treat errors as not being able to NOCOW. */
7260 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7263 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7264 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7267 range_end = round_up(offset + nocow_args.num_bytes,
7268 root->fs_info->sectorsize) - 1;
7269 ret = test_range_bit(io_tree, offset, range_end,
7270 EXTENT_DELALLOC, 0, NULL);
7278 *orig_start = key.offset - nocow_args.extent_offset;
7280 *orig_block_len = nocow_args.disk_num_bytes;
7282 *len = nocow_args.num_bytes;
7285 btrfs_free_path(path);
7289 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7290 struct extent_state **cached_state,
7291 unsigned int iomap_flags)
7293 const bool writing = (iomap_flags & IOMAP_WRITE);
7294 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7295 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7296 struct btrfs_ordered_extent *ordered;
7301 if (!try_lock_extent(io_tree, lockstart, lockend))
7304 lock_extent_bits(io_tree, lockstart, lockend, cached_state);
7307 * We're concerned with the entire range that we're going to be
7308 * doing DIO to, so we need to make sure there's no ordered
7309 * extents in this range.
7311 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7312 lockend - lockstart + 1);
7315 * We need to make sure there are no buffered pages in this
7316 * range either, we could have raced between the invalidate in
7317 * generic_file_direct_write and locking the extent. The
7318 * invalidate needs to happen so that reads after a write do not
7322 (!writing || !filemap_range_has_page(inode->i_mapping,
7323 lockstart, lockend)))
7326 unlock_extent_cached(io_tree, lockstart, lockend, cached_state);
7330 btrfs_put_ordered_extent(ordered);
7335 * If we are doing a DIO read and the ordered extent we
7336 * found is for a buffered write, we can not wait for it
7337 * to complete and retry, because if we do so we can
7338 * deadlock with concurrent buffered writes on page
7339 * locks. This happens only if our DIO read covers more
7340 * than one extent map, if at this point has already
7341 * created an ordered extent for a previous extent map
7342 * and locked its range in the inode's io tree, and a
7343 * concurrent write against that previous extent map's
7344 * range and this range started (we unlock the ranges
7345 * in the io tree only when the bios complete and
7346 * buffered writes always lock pages before attempting
7347 * to lock range in the io tree).
7350 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7351 btrfs_start_ordered_extent(ordered, 1);
7353 ret = nowait ? -EAGAIN : -ENOTBLK;
7354 btrfs_put_ordered_extent(ordered);
7357 * We could trigger writeback for this range (and wait
7358 * for it to complete) and then invalidate the pages for
7359 * this range (through invalidate_inode_pages2_range()),
7360 * but that can lead us to a deadlock with a concurrent
7361 * call to readahead (a buffered read or a defrag call
7362 * triggered a readahead) on a page lock due to an
7363 * ordered dio extent we created before but did not have
7364 * yet a corresponding bio submitted (whence it can not
7365 * complete), which makes readahead wait for that
7366 * ordered extent to complete while holding a lock on
7369 ret = nowait ? -EAGAIN : -ENOTBLK;
7381 /* The callers of this must take lock_extent() */
7382 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7383 u64 len, u64 orig_start, u64 block_start,
7384 u64 block_len, u64 orig_block_len,
7385 u64 ram_bytes, int compress_type,
7388 struct extent_map_tree *em_tree;
7389 struct extent_map *em;
7392 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7393 type == BTRFS_ORDERED_COMPRESSED ||
7394 type == BTRFS_ORDERED_NOCOW ||
7395 type == BTRFS_ORDERED_REGULAR);
7397 em_tree = &inode->extent_tree;
7398 em = alloc_extent_map();
7400 return ERR_PTR(-ENOMEM);
7403 em->orig_start = orig_start;
7405 em->block_len = block_len;
7406 em->block_start = block_start;
7407 em->orig_block_len = orig_block_len;
7408 em->ram_bytes = ram_bytes;
7409 em->generation = -1;
7410 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7411 if (type == BTRFS_ORDERED_PREALLOC) {
7412 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7413 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7414 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7415 em->compress_type = compress_type;
7419 btrfs_drop_extent_cache(inode, em->start,
7420 em->start + em->len - 1, 0);
7421 write_lock(&em_tree->lock);
7422 ret = add_extent_mapping(em_tree, em, 1);
7423 write_unlock(&em_tree->lock);
7425 * The caller has taken lock_extent(), who could race with us
7428 } while (ret == -EEXIST);
7431 free_extent_map(em);
7432 return ERR_PTR(ret);
7435 /* em got 2 refs now, callers needs to do free_extent_map once. */
7440 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7441 struct inode *inode,
7442 struct btrfs_dio_data *dio_data,
7444 unsigned int iomap_flags)
7446 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7447 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7448 struct extent_map *em = *map;
7450 u64 block_start, orig_start, orig_block_len, ram_bytes;
7451 struct btrfs_block_group *bg;
7452 bool can_nocow = false;
7453 bool space_reserved = false;
7458 * We don't allocate a new extent in the following cases
7460 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7462 * 2) The extent is marked as PREALLOC. We're good to go here and can
7463 * just use the extent.
7466 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7467 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7468 em->block_start != EXTENT_MAP_HOLE)) {
7469 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7470 type = BTRFS_ORDERED_PREALLOC;
7472 type = BTRFS_ORDERED_NOCOW;
7473 len = min(len, em->len - (start - em->start));
7474 block_start = em->block_start + (start - em->start);
7476 if (can_nocow_extent(inode, start, &len, &orig_start,
7477 &orig_block_len, &ram_bytes, false) == 1) {
7478 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7486 struct extent_map *em2;
7488 /* We can NOCOW, so only need to reserve metadata space. */
7489 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7492 /* Our caller expects us to free the input extent map. */
7493 free_extent_map(em);
7495 btrfs_dec_nocow_writers(bg);
7496 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7500 space_reserved = true;
7502 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7503 orig_start, block_start,
7504 len, orig_block_len,
7506 btrfs_dec_nocow_writers(bg);
7507 if (type == BTRFS_ORDERED_PREALLOC) {
7508 free_extent_map(em);
7517 dio_data->nocow_done = true;
7519 /* Our caller expects us to free the input extent map. */
7520 free_extent_map(em);
7527 * If we could not allocate data space before locking the file
7528 * range and we can't do a NOCOW write, then we have to fail.
7530 if (!dio_data->data_space_reserved)
7534 * We have to COW and we have already reserved data space before,
7535 * so now we reserve only metadata.
7537 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7541 space_reserved = true;
7543 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7549 len = min(len, em->len - (start - em->start));
7551 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7552 prev_len - len, true);
7556 * We have created our ordered extent, so we can now release our reservation
7557 * for an outstanding extent.
7559 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7562 * Need to update the i_size under the extent lock so buffered
7563 * readers will get the updated i_size when we unlock.
7565 if (start + len > i_size_read(inode))
7566 i_size_write(inode, start + len);
7568 if (ret && space_reserved) {
7569 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7570 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7575 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7576 loff_t length, unsigned int flags, struct iomap *iomap,
7577 struct iomap *srcmap)
7579 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7580 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7581 struct extent_map *em;
7582 struct extent_state *cached_state = NULL;
7583 struct btrfs_dio_data *dio_data = iter->private;
7584 u64 lockstart, lockend;
7585 const bool write = !!(flags & IOMAP_WRITE);
7588 const u64 data_alloc_len = length;
7589 bool unlock_extents = false;
7592 len = min_t(u64, len, fs_info->sectorsize);
7595 lockend = start + len - 1;
7598 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7599 * enough if we've written compressed pages to this area, so we need to
7600 * flush the dirty pages again to make absolutely sure that any
7601 * outstanding dirty pages are on disk - the first flush only starts
7602 * compression on the data, while keeping the pages locked, so by the
7603 * time the second flush returns we know bios for the compressed pages
7604 * were submitted and finished, and the pages no longer under writeback.
7606 * If we have a NOWAIT request and we have any pages in the range that
7607 * are locked, likely due to compression still in progress, we don't want
7608 * to block on page locks. We also don't want to block on pages marked as
7609 * dirty or under writeback (same as for the non-compression case).
7610 * iomap_dio_rw() did the same check, but after that and before we got
7611 * here, mmap'ed writes may have happened or buffered reads started
7612 * (readpage() and readahead(), which lock pages), as we haven't locked
7613 * the file range yet.
7615 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7616 &BTRFS_I(inode)->runtime_flags)) {
7617 if (flags & IOMAP_NOWAIT) {
7618 if (filemap_range_needs_writeback(inode->i_mapping,
7619 lockstart, lockend))
7622 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7623 start + length - 1);
7629 memset(dio_data, 0, sizeof(*dio_data));
7632 * We always try to allocate data space and must do it before locking
7633 * the file range, to avoid deadlocks with concurrent writes to the same
7634 * range if the range has several extents and the writes don't expand the
7635 * current i_size (the inode lock is taken in shared mode). If we fail to
7636 * allocate data space here we continue and later, after locking the
7637 * file range, we fail with ENOSPC only if we figure out we can not do a
7640 if (write && !(flags & IOMAP_NOWAIT)) {
7641 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7642 &dio_data->data_reserved,
7643 start, data_alloc_len);
7645 dio_data->data_space_reserved = true;
7646 else if (ret && !(BTRFS_I(inode)->flags &
7647 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7652 * If this errors out it's because we couldn't invalidate pagecache for
7653 * this range and we need to fallback to buffered IO, or we are doing a
7654 * NOWAIT read/write and we need to block.
7656 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7660 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7667 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7668 * io. INLINE is special, and we could probably kludge it in here, but
7669 * it's still buffered so for safety lets just fall back to the generic
7672 * For COMPRESSED we _have_ to read the entire extent in so we can
7673 * decompress it, so there will be buffering required no matter what we
7674 * do, so go ahead and fallback to buffered.
7676 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7677 * to buffered IO. Don't blame me, this is the price we pay for using
7680 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7681 em->block_start == EXTENT_MAP_INLINE) {
7682 free_extent_map(em);
7684 * If we are in a NOWAIT context, return -EAGAIN in order to
7685 * fallback to buffered IO. This is not only because we can
7686 * block with buffered IO (no support for NOWAIT semantics at
7687 * the moment) but also to avoid returning short reads to user
7688 * space - this happens if we were able to read some data from
7689 * previous non-compressed extents and then when we fallback to
7690 * buffered IO, at btrfs_file_read_iter() by calling
7691 * filemap_read(), we fail to fault in pages for the read buffer,
7692 * in which case filemap_read() returns a short read (the number
7693 * of bytes previously read is > 0, so it does not return -EFAULT).
7695 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7699 len = min(len, em->len - (start - em->start));
7702 * If we have a NOWAIT request and the range contains multiple extents
7703 * (or a mix of extents and holes), then we return -EAGAIN to make the
7704 * caller fallback to a context where it can do a blocking (without
7705 * NOWAIT) request. This way we avoid doing partial IO and returning
7706 * success to the caller, which is not optimal for writes and for reads
7707 * it can result in unexpected behaviour for an application.
7709 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7710 * iomap_dio_rw(), we can end up returning less data then what the caller
7711 * asked for, resulting in an unexpected, and incorrect, short read.
7712 * That is, the caller asked to read N bytes and we return less than that,
7713 * which is wrong unless we are crossing EOF. This happens if we get a
7714 * page fault error when trying to fault in pages for the buffer that is
7715 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7716 * have previously submitted bios for other extents in the range, in
7717 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7718 * those bios have completed by the time we get the page fault error,
7719 * which we return back to our caller - we should only return EIOCBQUEUED
7720 * after we have submitted bios for all the extents in the range.
7722 if ((flags & IOMAP_NOWAIT) && len < length) {
7723 free_extent_map(em);
7729 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7733 unlock_extents = true;
7734 /* Recalc len in case the new em is smaller than requested */
7735 len = min(len, em->len - (start - em->start));
7736 if (dio_data->data_space_reserved) {
7738 u64 release_len = 0;
7740 if (dio_data->nocow_done) {
7741 release_offset = start;
7742 release_len = data_alloc_len;
7743 } else if (len < data_alloc_len) {
7744 release_offset = start + len;
7745 release_len = data_alloc_len - len;
7748 if (release_len > 0)
7749 btrfs_free_reserved_data_space(BTRFS_I(inode),
7750 dio_data->data_reserved,
7756 * We need to unlock only the end area that we aren't using.
7757 * The rest is going to be unlocked by the endio routine.
7759 lockstart = start + len;
7760 if (lockstart < lockend)
7761 unlock_extents = true;
7765 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7766 lockstart, lockend, &cached_state);
7768 free_extent_state(cached_state);
7771 * Translate extent map information to iomap.
7772 * We trim the extents (and move the addr) even though iomap code does
7773 * that, since we have locked only the parts we are performing I/O in.
7775 if ((em->block_start == EXTENT_MAP_HOLE) ||
7776 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7777 iomap->addr = IOMAP_NULL_ADDR;
7778 iomap->type = IOMAP_HOLE;
7780 iomap->addr = em->block_start + (start - em->start);
7781 iomap->type = IOMAP_MAPPED;
7783 iomap->offset = start;
7784 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7785 iomap->length = len;
7787 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7788 iomap->flags |= IOMAP_F_ZONE_APPEND;
7790 free_extent_map(em);
7795 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7798 if (dio_data->data_space_reserved) {
7799 btrfs_free_reserved_data_space(BTRFS_I(inode),
7800 dio_data->data_reserved,
7801 start, data_alloc_len);
7802 extent_changeset_free(dio_data->data_reserved);
7808 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7809 ssize_t written, unsigned int flags, struct iomap *iomap)
7811 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7812 struct btrfs_dio_data *dio_data = iter->private;
7813 size_t submitted = dio_data->submitted;
7814 const bool write = !!(flags & IOMAP_WRITE);
7817 if (!write && (iomap->type == IOMAP_HOLE)) {
7818 /* If reading from a hole, unlock and return */
7819 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7823 if (submitted < length) {
7825 length -= submitted;
7827 __endio_write_update_ordered(BTRFS_I(inode), pos,
7830 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7836 extent_changeset_free(dio_data->data_reserved);
7840 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7843 * This implies a barrier so that stores to dio_bio->bi_status before
7844 * this and loads of dio_bio->bi_status after this are fully ordered.
7846 if (!refcount_dec_and_test(&dip->refs))
7849 if (btrfs_op(&dip->bio) == BTRFS_MAP_WRITE) {
7850 __endio_write_update_ordered(BTRFS_I(dip->inode),
7853 !dip->bio.bi_status);
7855 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7857 dip->file_offset + dip->bytes - 1);
7861 bio_endio(&dip->bio);
7864 static void submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7866 enum btrfs_compression_type compress_type)
7868 struct btrfs_dio_private *dip = bio->bi_private;
7869 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7871 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7873 if (btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA))
7876 refcount_inc(&dip->refs);
7877 if (btrfs_map_bio(fs_info, bio, mirror_num))
7878 refcount_dec(&dip->refs);
7881 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7882 struct btrfs_bio *bbio,
7883 const bool uptodate)
7885 struct inode *inode = dip->inode;
7886 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7887 const u32 sectorsize = fs_info->sectorsize;
7888 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7889 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7890 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7891 struct bio_vec bvec;
7892 struct bvec_iter iter;
7894 blk_status_t err = BLK_STS_OK;
7896 __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) {
7897 unsigned int i, nr_sectors, pgoff;
7899 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7900 pgoff = bvec.bv_offset;
7901 for (i = 0; i < nr_sectors; i++) {
7902 u64 start = bbio->file_offset + bio_offset;
7904 ASSERT(pgoff < PAGE_SIZE);
7906 (!csum || !check_data_csum(inode, bbio,
7907 bio_offset, bvec.bv_page,
7909 clean_io_failure(fs_info, failure_tree, io_tree,
7910 start, bvec.bv_page,
7911 btrfs_ino(BTRFS_I(inode)),
7916 ret = btrfs_repair_one_sector(inode, &bbio->bio,
7917 bio_offset, bvec.bv_page, pgoff,
7918 start, bbio->mirror_num,
7919 submit_dio_repair_bio);
7921 err = errno_to_blk_status(ret);
7923 ASSERT(bio_offset + sectorsize > bio_offset);
7924 bio_offset += sectorsize;
7925 pgoff += sectorsize;
7931 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7932 const u64 offset, const u64 bytes,
7933 const bool uptodate)
7935 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7936 finish_ordered_fn, uptodate);
7939 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7941 u64 dio_file_offset)
7943 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
7946 static void btrfs_end_dio_bio(struct bio *bio)
7948 struct btrfs_dio_private *dip = bio->bi_private;
7949 struct btrfs_bio *bbio = btrfs_bio(bio);
7950 blk_status_t err = bio->bi_status;
7953 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7954 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7955 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7956 bio->bi_opf, bio->bi_iter.bi_sector,
7957 bio->bi_iter.bi_size, err);
7959 if (bio_op(bio) == REQ_OP_READ)
7960 err = btrfs_check_read_dio_bio(dip, bbio, !err);
7963 dip->bio.bi_status = err;
7965 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
7968 btrfs_dio_private_put(dip);
7971 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7972 struct inode *inode, u64 file_offset, int async_submit)
7974 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7975 struct btrfs_dio_private *dip = bio->bi_private;
7976 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7979 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7981 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7984 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7989 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7992 if (write && async_submit) {
7993 ret = btrfs_wq_submit_bio(inode, bio, 0, file_offset,
7994 btrfs_submit_bio_start_direct_io);
7998 * If we aren't doing async submit, calculate the csum of the
8001 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
8007 csum_offset = file_offset - dip->file_offset;
8008 csum_offset >>= fs_info->sectorsize_bits;
8009 csum_offset *= fs_info->csum_size;
8010 btrfs_bio(bio)->csum = dip->csums + csum_offset;
8013 ret = btrfs_map_bio(fs_info, bio, 0);
8018 static void btrfs_submit_direct(const struct iomap_iter *iter,
8019 struct bio *dio_bio, loff_t file_offset)
8021 struct btrfs_dio_private *dip =
8022 container_of(dio_bio, struct btrfs_dio_private, bio);
8023 struct inode *inode = iter->inode;
8024 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8025 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8026 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8027 BTRFS_BLOCK_GROUP_RAID56_MASK);
8030 int async_submit = 0;
8032 u64 clone_offset = 0;
8036 blk_status_t status;
8037 struct btrfs_io_geometry geom;
8038 struct btrfs_dio_data *dio_data = iter->private;
8039 struct extent_map *em = NULL;
8042 dip->file_offset = file_offset;
8043 dip->bytes = dio_bio->bi_iter.bi_size;
8044 refcount_set(&dip->refs, 1);
8047 if (!write && !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
8048 unsigned int nr_sectors =
8049 (dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits);
8052 * Load the csums up front to reduce csum tree searches and
8053 * contention when submitting bios.
8055 status = BLK_STS_RESOURCE;
8056 dip->csums = kcalloc(nr_sectors, fs_info->csum_size, GFP_NOFS);
8060 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8061 if (status != BLK_STS_OK)
8065 start_sector = dio_bio->bi_iter.bi_sector;
8066 submit_len = dio_bio->bi_iter.bi_size;
8069 logical = start_sector << 9;
8070 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8072 status = errno_to_blk_status(PTR_ERR(em));
8076 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8079 status = errno_to_blk_status(ret);
8083 clone_len = min(submit_len, geom.len);
8084 ASSERT(clone_len <= UINT_MAX);
8087 * This will never fail as it's passing GPF_NOFS and
8088 * the allocation is backed by btrfs_bioset.
8090 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8091 bio->bi_private = dip;
8092 bio->bi_end_io = btrfs_end_dio_bio;
8093 btrfs_bio(bio)->file_offset = file_offset;
8095 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8096 status = extract_ordered_extent(BTRFS_I(inode), bio,
8104 ASSERT(submit_len >= clone_len);
8105 submit_len -= clone_len;
8108 * Increase the count before we submit the bio so we know
8109 * the end IO handler won't happen before we increase the
8110 * count. Otherwise, the dip might get freed before we're
8111 * done setting it up.
8113 * We transfer the initial reference to the last bio, so we
8114 * don't need to increment the reference count for the last one.
8116 if (submit_len > 0) {
8117 refcount_inc(&dip->refs);
8119 * If we are submitting more than one bio, submit them
8120 * all asynchronously. The exception is RAID 5 or 6, as
8121 * asynchronous checksums make it difficult to collect
8122 * full stripe writes.
8128 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8133 refcount_dec(&dip->refs);
8137 dio_data->submitted += clone_len;
8138 clone_offset += clone_len;
8139 start_sector += clone_len >> 9;
8140 file_offset += clone_len;
8142 free_extent_map(em);
8143 } while (submit_len > 0);
8147 free_extent_map(em);
8149 dio_bio->bi_status = status;
8150 btrfs_dio_private_put(dip);
8153 static const struct iomap_ops btrfs_dio_iomap_ops = {
8154 .iomap_begin = btrfs_dio_iomap_begin,
8155 .iomap_end = btrfs_dio_iomap_end,
8158 static const struct iomap_dio_ops btrfs_dio_ops = {
8159 .submit_io = btrfs_submit_direct,
8160 .bio_set = &btrfs_dio_bioset,
8163 ssize_t btrfs_dio_rw(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
8165 struct btrfs_dio_data data;
8167 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
8168 IOMAP_DIO_PARTIAL, &data, done_before);
8171 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8176 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8180 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8183 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8185 struct inode *inode = page->mapping->host;
8188 if (current->flags & PF_MEMALLOC) {
8189 redirty_page_for_writepage(wbc, page);
8195 * If we are under memory pressure we will call this directly from the
8196 * VM, we need to make sure we have the inode referenced for the ordered
8197 * extent. If not just return like we didn't do anything.
8199 if (!igrab(inode)) {
8200 redirty_page_for_writepage(wbc, page);
8201 return AOP_WRITEPAGE_ACTIVATE;
8203 ret = extent_write_full_page(page, wbc);
8204 btrfs_add_delayed_iput(inode);
8208 static int btrfs_writepages(struct address_space *mapping,
8209 struct writeback_control *wbc)
8211 return extent_writepages(mapping, wbc);
8214 static void btrfs_readahead(struct readahead_control *rac)
8216 extent_readahead(rac);
8220 * For release_folio() and invalidate_folio() we have a race window where
8221 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8222 * If we continue to release/invalidate the page, we could cause use-after-free
8223 * for subpage spinlock. So this function is to spin and wait for subpage
8226 static void wait_subpage_spinlock(struct page *page)
8228 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8229 struct btrfs_subpage *subpage;
8231 if (!btrfs_is_subpage(fs_info, page))
8234 ASSERT(PagePrivate(page) && page->private);
8235 subpage = (struct btrfs_subpage *)page->private;
8238 * This may look insane as we just acquire the spinlock and release it,
8239 * without doing anything. But we just want to make sure no one is
8240 * still holding the subpage spinlock.
8241 * And since the page is not dirty nor writeback, and we have page
8242 * locked, the only possible way to hold a spinlock is from the endio
8243 * function to clear page writeback.
8245 * Here we just acquire the spinlock so that all existing callers
8246 * should exit and we're safe to release/invalidate the page.
8248 spin_lock_irq(&subpage->lock);
8249 spin_unlock_irq(&subpage->lock);
8252 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8254 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8257 wait_subpage_spinlock(&folio->page);
8258 clear_page_extent_mapped(&folio->page);
8263 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8265 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8267 return __btrfs_release_folio(folio, gfp_flags);
8270 #ifdef CONFIG_MIGRATION
8271 static int btrfs_migrate_folio(struct address_space *mapping,
8272 struct folio *dst, struct folio *src,
8273 enum migrate_mode mode)
8275 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8277 if (ret != MIGRATEPAGE_SUCCESS)
8280 if (folio_test_ordered(src)) {
8281 folio_clear_ordered(src);
8282 folio_set_ordered(dst);
8285 return MIGRATEPAGE_SUCCESS;
8288 #define btrfs_migrate_folio NULL
8291 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8294 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8295 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8296 struct extent_io_tree *tree = &inode->io_tree;
8297 struct extent_state *cached_state = NULL;
8298 u64 page_start = folio_pos(folio);
8299 u64 page_end = page_start + folio_size(folio) - 1;
8301 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8304 * We have folio locked so no new ordered extent can be created on this
8305 * page, nor bio can be submitted for this folio.
8307 * But already submitted bio can still be finished on this folio.
8308 * Furthermore, endio function won't skip folio which has Ordered
8309 * (Private2) already cleared, so it's possible for endio and
8310 * invalidate_folio to do the same ordered extent accounting twice
8313 * So here we wait for any submitted bios to finish, so that we won't
8314 * do double ordered extent accounting on the same folio.
8316 folio_wait_writeback(folio);
8317 wait_subpage_spinlock(&folio->page);
8320 * For subpage case, we have call sites like
8321 * btrfs_punch_hole_lock_range() which passes range not aligned to
8323 * If the range doesn't cover the full folio, we don't need to and
8324 * shouldn't clear page extent mapped, as folio->private can still
8325 * record subpage dirty bits for other part of the range.
8327 * For cases that invalidate the full folio even the range doesn't
8328 * cover the full folio, like invalidating the last folio, we're
8329 * still safe to wait for ordered extent to finish.
8331 if (!(offset == 0 && length == folio_size(folio))) {
8332 btrfs_release_folio(folio, GFP_NOFS);
8336 if (!inode_evicting)
8337 lock_extent_bits(tree, page_start, page_end, &cached_state);
8340 while (cur < page_end) {
8341 struct btrfs_ordered_extent *ordered;
8346 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8347 page_end + 1 - cur);
8349 range_end = page_end;
8351 * No ordered extent covering this range, we are safe
8352 * to delete all extent states in the range.
8354 delete_states = true;
8357 if (ordered->file_offset > cur) {
8359 * There is a range between [cur, oe->file_offset) not
8360 * covered by any ordered extent.
8361 * We are safe to delete all extent states, and handle
8362 * the ordered extent in the next iteration.
8364 range_end = ordered->file_offset - 1;
8365 delete_states = true;
8369 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8371 ASSERT(range_end + 1 - cur < U32_MAX);
8372 range_len = range_end + 1 - cur;
8373 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8375 * If Ordered (Private2) is cleared, it means endio has
8376 * already been executed for the range.
8377 * We can't delete the extent states as
8378 * btrfs_finish_ordered_io() may still use some of them.
8380 delete_states = false;
8383 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8386 * IO on this page will never be started, so we need to account
8387 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8388 * here, must leave that up for the ordered extent completion.
8390 * This will also unlock the range for incoming
8391 * btrfs_finish_ordered_io().
8393 if (!inode_evicting)
8394 clear_extent_bit(tree, cur, range_end,
8396 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8397 EXTENT_DEFRAG, 1, 0, &cached_state);
8399 spin_lock_irq(&inode->ordered_tree.lock);
8400 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8401 ordered->truncated_len = min(ordered->truncated_len,
8402 cur - ordered->file_offset);
8403 spin_unlock_irq(&inode->ordered_tree.lock);
8405 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8406 cur, range_end + 1 - cur)) {
8407 btrfs_finish_ordered_io(ordered);
8409 * The ordered extent has finished, now we're again
8410 * safe to delete all extent states of the range.
8412 delete_states = true;
8415 * btrfs_finish_ordered_io() will get executed by endio
8416 * of other pages, thus we can't delete extent states
8419 delete_states = false;
8423 btrfs_put_ordered_extent(ordered);
8425 * Qgroup reserved space handler
8426 * Sector(s) here will be either:
8428 * 1) Already written to disk or bio already finished
8429 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8430 * Qgroup will be handled by its qgroup_record then.
8431 * btrfs_qgroup_free_data() call will do nothing here.
8433 * 2) Not written to disk yet
8434 * Then btrfs_qgroup_free_data() call will clear the
8435 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8436 * reserved data space.
8437 * Since the IO will never happen for this page.
8439 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8440 if (!inode_evicting) {
8441 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8442 EXTENT_DELALLOC | EXTENT_UPTODATE |
8443 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8444 delete_states, &cached_state);
8446 cur = range_end + 1;
8449 * We have iterated through all ordered extents of the page, the page
8450 * should not have Ordered (Private2) anymore, or the above iteration
8451 * did something wrong.
8453 ASSERT(!folio_test_ordered(folio));
8454 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8455 if (!inode_evicting)
8456 __btrfs_release_folio(folio, GFP_NOFS);
8457 clear_page_extent_mapped(&folio->page);
8461 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8462 * called from a page fault handler when a page is first dirtied. Hence we must
8463 * be careful to check for EOF conditions here. We set the page up correctly
8464 * for a written page which means we get ENOSPC checking when writing into
8465 * holes and correct delalloc and unwritten extent mapping on filesystems that
8466 * support these features.
8468 * We are not allowed to take the i_mutex here so we have to play games to
8469 * protect against truncate races as the page could now be beyond EOF. Because
8470 * truncate_setsize() writes the inode size before removing pages, once we have
8471 * the page lock we can determine safely if the page is beyond EOF. If it is not
8472 * beyond EOF, then the page is guaranteed safe against truncation until we
8475 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8477 struct page *page = vmf->page;
8478 struct inode *inode = file_inode(vmf->vma->vm_file);
8479 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8480 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8481 struct btrfs_ordered_extent *ordered;
8482 struct extent_state *cached_state = NULL;
8483 struct extent_changeset *data_reserved = NULL;
8484 unsigned long zero_start;
8494 reserved_space = PAGE_SIZE;
8496 sb_start_pagefault(inode->i_sb);
8497 page_start = page_offset(page);
8498 page_end = page_start + PAGE_SIZE - 1;
8502 * Reserving delalloc space after obtaining the page lock can lead to
8503 * deadlock. For example, if a dirty page is locked by this function
8504 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8505 * dirty page write out, then the btrfs_writepage() function could
8506 * end up waiting indefinitely to get a lock on the page currently
8507 * being processed by btrfs_page_mkwrite() function.
8509 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8510 page_start, reserved_space);
8512 ret2 = file_update_time(vmf->vma->vm_file);
8516 ret = vmf_error(ret2);
8522 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8524 down_read(&BTRFS_I(inode)->i_mmap_lock);
8526 size = i_size_read(inode);
8528 if ((page->mapping != inode->i_mapping) ||
8529 (page_start >= size)) {
8530 /* page got truncated out from underneath us */
8533 wait_on_page_writeback(page);
8535 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8536 ret2 = set_page_extent_mapped(page);
8538 ret = vmf_error(ret2);
8539 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8544 * we can't set the delalloc bits if there are pending ordered
8545 * extents. Drop our locks and wait for them to finish
8547 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8550 unlock_extent_cached(io_tree, page_start, page_end,
8553 up_read(&BTRFS_I(inode)->i_mmap_lock);
8554 btrfs_start_ordered_extent(ordered, 1);
8555 btrfs_put_ordered_extent(ordered);
8559 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8560 reserved_space = round_up(size - page_start,
8561 fs_info->sectorsize);
8562 if (reserved_space < PAGE_SIZE) {
8563 end = page_start + reserved_space - 1;
8564 btrfs_delalloc_release_space(BTRFS_I(inode),
8565 data_reserved, page_start,
8566 PAGE_SIZE - reserved_space, true);
8571 * page_mkwrite gets called when the page is firstly dirtied after it's
8572 * faulted in, but write(2) could also dirty a page and set delalloc
8573 * bits, thus in this case for space account reason, we still need to
8574 * clear any delalloc bits within this page range since we have to
8575 * reserve data&meta space before lock_page() (see above comments).
8577 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8578 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8579 EXTENT_DEFRAG, 0, 0, &cached_state);
8581 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8584 unlock_extent_cached(io_tree, page_start, page_end,
8586 ret = VM_FAULT_SIGBUS;
8590 /* page is wholly or partially inside EOF */
8591 if (page_start + PAGE_SIZE > size)
8592 zero_start = offset_in_page(size);
8594 zero_start = PAGE_SIZE;
8596 if (zero_start != PAGE_SIZE) {
8597 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8598 flush_dcache_page(page);
8600 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8601 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8602 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8604 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8606 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8607 up_read(&BTRFS_I(inode)->i_mmap_lock);
8609 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8610 sb_end_pagefault(inode->i_sb);
8611 extent_changeset_free(data_reserved);
8612 return VM_FAULT_LOCKED;
8616 up_read(&BTRFS_I(inode)->i_mmap_lock);
8618 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8619 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8620 reserved_space, (ret != 0));
8622 sb_end_pagefault(inode->i_sb);
8623 extent_changeset_free(data_reserved);
8627 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8629 struct btrfs_truncate_control control = {
8630 .inode = BTRFS_I(inode),
8631 .ino = btrfs_ino(BTRFS_I(inode)),
8632 .min_type = BTRFS_EXTENT_DATA_KEY,
8633 .clear_extent_range = true,
8635 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8636 struct btrfs_root *root = BTRFS_I(inode)->root;
8637 struct btrfs_block_rsv *rsv;
8639 struct btrfs_trans_handle *trans;
8640 u64 mask = fs_info->sectorsize - 1;
8641 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8643 if (!skip_writeback) {
8644 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8651 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8652 * things going on here:
8654 * 1) We need to reserve space to update our inode.
8656 * 2) We need to have something to cache all the space that is going to
8657 * be free'd up by the truncate operation, but also have some slack
8658 * space reserved in case it uses space during the truncate (thank you
8659 * very much snapshotting).
8661 * And we need these to be separate. The fact is we can use a lot of
8662 * space doing the truncate, and we have no earthly idea how much space
8663 * we will use, so we need the truncate reservation to be separate so it
8664 * doesn't end up using space reserved for updating the inode. We also
8665 * need to be able to stop the transaction and start a new one, which
8666 * means we need to be able to update the inode several times, and we
8667 * have no idea of knowing how many times that will be, so we can't just
8668 * reserve 1 item for the entirety of the operation, so that has to be
8669 * done separately as well.
8671 * So that leaves us with
8673 * 1) rsv - for the truncate reservation, which we will steal from the
8674 * transaction reservation.
8675 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8676 * updating the inode.
8678 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8681 rsv->size = min_size;
8685 * 1 for the truncate slack space
8686 * 1 for updating the inode.
8688 trans = btrfs_start_transaction(root, 2);
8689 if (IS_ERR(trans)) {
8690 ret = PTR_ERR(trans);
8694 /* Migrate the slack space for the truncate to our reserve */
8695 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8699 trans->block_rsv = rsv;
8702 struct extent_state *cached_state = NULL;
8703 const u64 new_size = inode->i_size;
8704 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8706 control.new_size = new_size;
8707 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8710 * We want to drop from the next block forward in case this new
8711 * size is not block aligned since we will be keeping the last
8712 * block of the extent just the way it is.
8714 btrfs_drop_extent_cache(BTRFS_I(inode),
8715 ALIGN(new_size, fs_info->sectorsize),
8718 ret = btrfs_truncate_inode_items(trans, root, &control);
8720 inode_sub_bytes(inode, control.sub_bytes);
8721 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8723 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8724 (u64)-1, &cached_state);
8726 trans->block_rsv = &fs_info->trans_block_rsv;
8727 if (ret != -ENOSPC && ret != -EAGAIN)
8730 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8734 btrfs_end_transaction(trans);
8735 btrfs_btree_balance_dirty(fs_info);
8737 trans = btrfs_start_transaction(root, 2);
8738 if (IS_ERR(trans)) {
8739 ret = PTR_ERR(trans);
8744 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8745 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8746 rsv, min_size, false);
8747 BUG_ON(ret); /* shouldn't happen */
8748 trans->block_rsv = rsv;
8752 * We can't call btrfs_truncate_block inside a trans handle as we could
8753 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8754 * know we've truncated everything except the last little bit, and can
8755 * do btrfs_truncate_block and then update the disk_i_size.
8757 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8758 btrfs_end_transaction(trans);
8759 btrfs_btree_balance_dirty(fs_info);
8761 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8764 trans = btrfs_start_transaction(root, 1);
8765 if (IS_ERR(trans)) {
8766 ret = PTR_ERR(trans);
8769 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8775 trans->block_rsv = &fs_info->trans_block_rsv;
8776 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8780 ret2 = btrfs_end_transaction(trans);
8783 btrfs_btree_balance_dirty(fs_info);
8786 btrfs_free_block_rsv(fs_info, rsv);
8788 * So if we truncate and then write and fsync we normally would just
8789 * write the extents that changed, which is a problem if we need to
8790 * first truncate that entire inode. So set this flag so we write out
8791 * all of the extents in the inode to the sync log so we're completely
8794 * If no extents were dropped or trimmed we don't need to force the next
8795 * fsync to truncate all the inode's items from the log and re-log them
8796 * all. This means the truncate operation did not change the file size,
8797 * or changed it to a smaller size but there was only an implicit hole
8798 * between the old i_size and the new i_size, and there were no prealloc
8799 * extents beyond i_size to drop.
8801 if (control.extents_found > 0)
8802 btrfs_set_inode_full_sync(BTRFS_I(inode));
8807 struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns,
8810 struct inode *inode;
8812 inode = new_inode(dir->i_sb);
8815 * Subvolumes don't inherit the sgid bit or the parent's gid if
8816 * the parent's sgid bit is set. This is probably a bug.
8818 inode_init_owner(mnt_userns, inode, NULL,
8819 S_IFDIR | (~current_umask() & S_IRWXUGO));
8820 inode->i_op = &btrfs_dir_inode_operations;
8821 inode->i_fop = &btrfs_dir_file_operations;
8826 struct inode *btrfs_alloc_inode(struct super_block *sb)
8828 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8829 struct btrfs_inode *ei;
8830 struct inode *inode;
8832 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8839 ei->last_sub_trans = 0;
8840 ei->logged_trans = 0;
8841 ei->delalloc_bytes = 0;
8842 ei->new_delalloc_bytes = 0;
8843 ei->defrag_bytes = 0;
8844 ei->disk_i_size = 0;
8848 ei->index_cnt = (u64)-1;
8850 ei->last_unlink_trans = 0;
8851 ei->last_reflink_trans = 0;
8852 ei->last_log_commit = 0;
8854 spin_lock_init(&ei->lock);
8855 ei->outstanding_extents = 0;
8856 if (sb->s_magic != BTRFS_TEST_MAGIC)
8857 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8858 BTRFS_BLOCK_RSV_DELALLOC);
8859 ei->runtime_flags = 0;
8860 ei->prop_compress = BTRFS_COMPRESS_NONE;
8861 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8863 ei->delayed_node = NULL;
8865 ei->i_otime.tv_sec = 0;
8866 ei->i_otime.tv_nsec = 0;
8868 inode = &ei->vfs_inode;
8869 extent_map_tree_init(&ei->extent_tree);
8870 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8871 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8872 IO_TREE_INODE_IO_FAILURE, inode);
8873 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8874 IO_TREE_INODE_FILE_EXTENT, inode);
8875 ei->io_tree.track_uptodate = true;
8876 ei->io_failure_tree.track_uptodate = true;
8877 atomic_set(&ei->sync_writers, 0);
8878 mutex_init(&ei->log_mutex);
8879 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8880 INIT_LIST_HEAD(&ei->delalloc_inodes);
8881 INIT_LIST_HEAD(&ei->delayed_iput);
8882 RB_CLEAR_NODE(&ei->rb_node);
8883 init_rwsem(&ei->i_mmap_lock);
8888 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8889 void btrfs_test_destroy_inode(struct inode *inode)
8891 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8892 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8896 void btrfs_free_inode(struct inode *inode)
8898 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8901 void btrfs_destroy_inode(struct inode *vfs_inode)
8903 struct btrfs_ordered_extent *ordered;
8904 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8905 struct btrfs_root *root = inode->root;
8907 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8908 WARN_ON(vfs_inode->i_data.nrpages);
8909 WARN_ON(inode->block_rsv.reserved);
8910 WARN_ON(inode->block_rsv.size);
8911 WARN_ON(inode->outstanding_extents);
8912 if (!S_ISDIR(vfs_inode->i_mode)) {
8913 WARN_ON(inode->delalloc_bytes);
8914 WARN_ON(inode->new_delalloc_bytes);
8916 WARN_ON(inode->csum_bytes);
8917 WARN_ON(inode->defrag_bytes);
8920 * This can happen where we create an inode, but somebody else also
8921 * created the same inode and we need to destroy the one we already
8928 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8932 btrfs_err(root->fs_info,
8933 "found ordered extent %llu %llu on inode cleanup",
8934 ordered->file_offset, ordered->num_bytes);
8935 btrfs_remove_ordered_extent(inode, ordered);
8936 btrfs_put_ordered_extent(ordered);
8937 btrfs_put_ordered_extent(ordered);
8940 btrfs_qgroup_check_reserved_leak(inode);
8941 inode_tree_del(inode);
8942 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8943 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8944 btrfs_put_root(inode->root);
8947 int btrfs_drop_inode(struct inode *inode)
8949 struct btrfs_root *root = BTRFS_I(inode)->root;
8954 /* the snap/subvol tree is on deleting */
8955 if (btrfs_root_refs(&root->root_item) == 0)
8958 return generic_drop_inode(inode);
8961 static void init_once(void *foo)
8963 struct btrfs_inode *ei = foo;
8965 inode_init_once(&ei->vfs_inode);
8968 void __cold btrfs_destroy_cachep(void)
8971 * Make sure all delayed rcu free inodes are flushed before we
8975 bioset_exit(&btrfs_dio_bioset);
8976 kmem_cache_destroy(btrfs_inode_cachep);
8977 kmem_cache_destroy(btrfs_trans_handle_cachep);
8978 kmem_cache_destroy(btrfs_path_cachep);
8979 kmem_cache_destroy(btrfs_free_space_cachep);
8980 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8983 int __init btrfs_init_cachep(void)
8985 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8986 sizeof(struct btrfs_inode), 0,
8987 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8989 if (!btrfs_inode_cachep)
8992 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8993 sizeof(struct btrfs_trans_handle), 0,
8994 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8995 if (!btrfs_trans_handle_cachep)
8998 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8999 sizeof(struct btrfs_path), 0,
9000 SLAB_MEM_SPREAD, NULL);
9001 if (!btrfs_path_cachep)
9004 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9005 sizeof(struct btrfs_free_space), 0,
9006 SLAB_MEM_SPREAD, NULL);
9007 if (!btrfs_free_space_cachep)
9010 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9011 PAGE_SIZE, PAGE_SIZE,
9012 SLAB_MEM_SPREAD, NULL);
9013 if (!btrfs_free_space_bitmap_cachep)
9016 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
9017 offsetof(struct btrfs_dio_private, bio),
9023 btrfs_destroy_cachep();
9027 static int btrfs_getattr(struct user_namespace *mnt_userns,
9028 const struct path *path, struct kstat *stat,
9029 u32 request_mask, unsigned int flags)
9033 struct inode *inode = d_inode(path->dentry);
9034 u32 blocksize = inode->i_sb->s_blocksize;
9035 u32 bi_flags = BTRFS_I(inode)->flags;
9036 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9038 stat->result_mask |= STATX_BTIME;
9039 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9040 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9041 if (bi_flags & BTRFS_INODE_APPEND)
9042 stat->attributes |= STATX_ATTR_APPEND;
9043 if (bi_flags & BTRFS_INODE_COMPRESS)
9044 stat->attributes |= STATX_ATTR_COMPRESSED;
9045 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9046 stat->attributes |= STATX_ATTR_IMMUTABLE;
9047 if (bi_flags & BTRFS_INODE_NODUMP)
9048 stat->attributes |= STATX_ATTR_NODUMP;
9049 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9050 stat->attributes |= STATX_ATTR_VERITY;
9052 stat->attributes_mask |= (STATX_ATTR_APPEND |
9053 STATX_ATTR_COMPRESSED |
9054 STATX_ATTR_IMMUTABLE |
9057 generic_fillattr(mnt_userns, inode, stat);
9058 stat->dev = BTRFS_I(inode)->root->anon_dev;
9060 spin_lock(&BTRFS_I(inode)->lock);
9061 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9062 inode_bytes = inode_get_bytes(inode);
9063 spin_unlock(&BTRFS_I(inode)->lock);
9064 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9065 ALIGN(delalloc_bytes, blocksize)) >> 9;
9069 static int btrfs_rename_exchange(struct inode *old_dir,
9070 struct dentry *old_dentry,
9071 struct inode *new_dir,
9072 struct dentry *new_dentry)
9074 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9075 struct btrfs_trans_handle *trans;
9076 unsigned int trans_num_items;
9077 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9078 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9079 struct inode *new_inode = new_dentry->d_inode;
9080 struct inode *old_inode = old_dentry->d_inode;
9081 struct timespec64 ctime = current_time(old_inode);
9082 struct btrfs_rename_ctx old_rename_ctx;
9083 struct btrfs_rename_ctx new_rename_ctx;
9084 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9085 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9090 bool need_abort = false;
9093 * For non-subvolumes allow exchange only within one subvolume, in the
9094 * same inode namespace. Two subvolumes (represented as directory) can
9095 * be exchanged as they're a logical link and have a fixed inode number.
9098 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9099 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9102 /* close the race window with snapshot create/destroy ioctl */
9103 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9104 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9105 down_read(&fs_info->subvol_sem);
9109 * 1 to remove old dir item
9110 * 1 to remove old dir index
9111 * 1 to add new dir item
9112 * 1 to add new dir index
9113 * 1 to update parent inode
9115 * If the parents are the same, we only need to account for one
9117 trans_num_items = (old_dir == new_dir ? 9 : 10);
9118 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9120 * 1 to remove old root ref
9121 * 1 to remove old root backref
9122 * 1 to add new root ref
9123 * 1 to add new root backref
9125 trans_num_items += 4;
9128 * 1 to update inode item
9129 * 1 to remove old inode ref
9130 * 1 to add new inode ref
9132 trans_num_items += 3;
9134 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9135 trans_num_items += 4;
9137 trans_num_items += 3;
9138 trans = btrfs_start_transaction(root, trans_num_items);
9139 if (IS_ERR(trans)) {
9140 ret = PTR_ERR(trans);
9145 ret = btrfs_record_root_in_trans(trans, dest);
9151 * We need to find a free sequence number both in the source and
9152 * in the destination directory for the exchange.
9154 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9157 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9161 BTRFS_I(old_inode)->dir_index = 0ULL;
9162 BTRFS_I(new_inode)->dir_index = 0ULL;
9164 /* Reference for the source. */
9165 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9166 /* force full log commit if subvolume involved. */
9167 btrfs_set_log_full_commit(trans);
9169 ret = btrfs_insert_inode_ref(trans, dest,
9170 new_dentry->d_name.name,
9171 new_dentry->d_name.len,
9173 btrfs_ino(BTRFS_I(new_dir)),
9180 /* And now for the dest. */
9181 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9182 /* force full log commit if subvolume involved. */
9183 btrfs_set_log_full_commit(trans);
9185 ret = btrfs_insert_inode_ref(trans, root,
9186 old_dentry->d_name.name,
9187 old_dentry->d_name.len,
9189 btrfs_ino(BTRFS_I(old_dir)),
9193 btrfs_abort_transaction(trans, ret);
9198 /* Update inode version and ctime/mtime. */
9199 inode_inc_iversion(old_dir);
9200 inode_inc_iversion(new_dir);
9201 inode_inc_iversion(old_inode);
9202 inode_inc_iversion(new_inode);
9203 old_dir->i_ctime = old_dir->i_mtime = ctime;
9204 new_dir->i_ctime = new_dir->i_mtime = ctime;
9205 old_inode->i_ctime = ctime;
9206 new_inode->i_ctime = ctime;
9208 if (old_dentry->d_parent != new_dentry->d_parent) {
9209 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9210 BTRFS_I(old_inode), 1);
9211 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9212 BTRFS_I(new_inode), 1);
9215 /* src is a subvolume */
9216 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9217 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9218 } else { /* src is an inode */
9219 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9220 BTRFS_I(old_dentry->d_inode),
9221 old_dentry->d_name.name,
9222 old_dentry->d_name.len,
9225 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9228 btrfs_abort_transaction(trans, ret);
9232 /* dest is a subvolume */
9233 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9234 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9235 } else { /* dest is an inode */
9236 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9237 BTRFS_I(new_dentry->d_inode),
9238 new_dentry->d_name.name,
9239 new_dentry->d_name.len,
9242 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9245 btrfs_abort_transaction(trans, ret);
9249 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9250 new_dentry->d_name.name,
9251 new_dentry->d_name.len, 0, old_idx);
9253 btrfs_abort_transaction(trans, ret);
9257 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9258 old_dentry->d_name.name,
9259 old_dentry->d_name.len, 0, new_idx);
9261 btrfs_abort_transaction(trans, ret);
9265 if (old_inode->i_nlink == 1)
9266 BTRFS_I(old_inode)->dir_index = old_idx;
9267 if (new_inode->i_nlink == 1)
9268 BTRFS_I(new_inode)->dir_index = new_idx;
9271 * Now pin the logs of the roots. We do it to ensure that no other task
9272 * can sync the logs while we are in progress with the rename, because
9273 * that could result in an inconsistency in case any of the inodes that
9274 * are part of this rename operation were logged before.
9276 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9277 btrfs_pin_log_trans(root);
9278 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9279 btrfs_pin_log_trans(dest);
9281 /* Do the log updates for all inodes. */
9282 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9283 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9284 old_rename_ctx.index, new_dentry->d_parent);
9285 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9286 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9287 new_rename_ctx.index, old_dentry->d_parent);
9289 /* Now unpin the logs. */
9290 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9291 btrfs_end_log_trans(root);
9292 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9293 btrfs_end_log_trans(dest);
9295 ret2 = btrfs_end_transaction(trans);
9296 ret = ret ? ret : ret2;
9298 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9299 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9300 up_read(&fs_info->subvol_sem);
9305 static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns,
9308 struct inode *inode;
9310 inode = new_inode(dir->i_sb);
9312 inode_init_owner(mnt_userns, inode, dir,
9313 S_IFCHR | WHITEOUT_MODE);
9314 inode->i_op = &btrfs_special_inode_operations;
9315 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9320 static int btrfs_rename(struct user_namespace *mnt_userns,
9321 struct inode *old_dir, struct dentry *old_dentry,
9322 struct inode *new_dir, struct dentry *new_dentry,
9325 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9326 struct btrfs_new_inode_args whiteout_args = {
9328 .dentry = old_dentry,
9330 struct btrfs_trans_handle *trans;
9331 unsigned int trans_num_items;
9332 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9333 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9334 struct inode *new_inode = d_inode(new_dentry);
9335 struct inode *old_inode = d_inode(old_dentry);
9336 struct btrfs_rename_ctx rename_ctx;
9340 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9342 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9345 /* we only allow rename subvolume link between subvolumes */
9346 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9349 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9350 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9353 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9354 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9358 /* check for collisions, even if the name isn't there */
9359 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9360 new_dentry->d_name.name,
9361 new_dentry->d_name.len);
9364 if (ret == -EEXIST) {
9366 * eexist without a new_inode */
9367 if (WARN_ON(!new_inode)) {
9371 /* maybe -EOVERFLOW */
9378 * we're using rename to replace one file with another. Start IO on it
9379 * now so we don't add too much work to the end of the transaction
9381 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9382 filemap_flush(old_inode->i_mapping);
9384 if (flags & RENAME_WHITEOUT) {
9385 whiteout_args.inode = new_whiteout_inode(mnt_userns, old_dir);
9386 if (!whiteout_args.inode)
9388 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9390 goto out_whiteout_inode;
9392 /* 1 to update the old parent inode. */
9393 trans_num_items = 1;
9396 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9397 /* Close the race window with snapshot create/destroy ioctl */
9398 down_read(&fs_info->subvol_sem);
9400 * 1 to remove old root ref
9401 * 1 to remove old root backref
9402 * 1 to add new root ref
9403 * 1 to add new root backref
9405 trans_num_items += 4;
9409 * 1 to remove old inode ref
9410 * 1 to add new inode ref
9412 trans_num_items += 3;
9415 * 1 to remove old dir item
9416 * 1 to remove old dir index
9417 * 1 to add new dir item
9418 * 1 to add new dir index
9420 trans_num_items += 4;
9421 /* 1 to update new parent inode if it's not the same as the old parent */
9422 if (new_dir != old_dir)
9427 * 1 to remove inode ref
9428 * 1 to remove dir item
9429 * 1 to remove dir index
9430 * 1 to possibly add orphan item
9432 trans_num_items += 5;
9434 trans = btrfs_start_transaction(root, trans_num_items);
9435 if (IS_ERR(trans)) {
9436 ret = PTR_ERR(trans);
9441 ret = btrfs_record_root_in_trans(trans, dest);
9446 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9450 BTRFS_I(old_inode)->dir_index = 0ULL;
9451 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9452 /* force full log commit if subvolume involved. */
9453 btrfs_set_log_full_commit(trans);
9455 ret = btrfs_insert_inode_ref(trans, dest,
9456 new_dentry->d_name.name,
9457 new_dentry->d_name.len,
9459 btrfs_ino(BTRFS_I(new_dir)), index);
9464 inode_inc_iversion(old_dir);
9465 inode_inc_iversion(new_dir);
9466 inode_inc_iversion(old_inode);
9467 old_dir->i_ctime = old_dir->i_mtime =
9468 new_dir->i_ctime = new_dir->i_mtime =
9469 old_inode->i_ctime = current_time(old_dir);
9471 if (old_dentry->d_parent != new_dentry->d_parent)
9472 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9473 BTRFS_I(old_inode), 1);
9475 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9476 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9478 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9479 BTRFS_I(d_inode(old_dentry)),
9480 old_dentry->d_name.name,
9481 old_dentry->d_name.len,
9484 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9487 btrfs_abort_transaction(trans, ret);
9492 inode_inc_iversion(new_inode);
9493 new_inode->i_ctime = current_time(new_inode);
9494 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9495 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9496 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9497 BUG_ON(new_inode->i_nlink == 0);
9499 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9500 BTRFS_I(d_inode(new_dentry)),
9501 new_dentry->d_name.name,
9502 new_dentry->d_name.len);
9504 if (!ret && new_inode->i_nlink == 0)
9505 ret = btrfs_orphan_add(trans,
9506 BTRFS_I(d_inode(new_dentry)));
9508 btrfs_abort_transaction(trans, ret);
9513 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9514 new_dentry->d_name.name,
9515 new_dentry->d_name.len, 0, index);
9517 btrfs_abort_transaction(trans, ret);
9521 if (old_inode->i_nlink == 1)
9522 BTRFS_I(old_inode)->dir_index = index;
9524 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9525 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9526 rename_ctx.index, new_dentry->d_parent);
9528 if (flags & RENAME_WHITEOUT) {
9529 ret = btrfs_create_new_inode(trans, &whiteout_args);
9531 btrfs_abort_transaction(trans, ret);
9534 unlock_new_inode(whiteout_args.inode);
9535 iput(whiteout_args.inode);
9536 whiteout_args.inode = NULL;
9540 ret2 = btrfs_end_transaction(trans);
9541 ret = ret ? ret : ret2;
9543 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9544 up_read(&fs_info->subvol_sem);
9545 if (flags & RENAME_WHITEOUT)
9546 btrfs_new_inode_args_destroy(&whiteout_args);
9548 if (flags & RENAME_WHITEOUT)
9549 iput(whiteout_args.inode);
9553 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9554 struct dentry *old_dentry, struct inode *new_dir,
9555 struct dentry *new_dentry, unsigned int flags)
9557 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9560 if (flags & RENAME_EXCHANGE)
9561 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9564 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9568 struct btrfs_delalloc_work {
9569 struct inode *inode;
9570 struct completion completion;
9571 struct list_head list;
9572 struct btrfs_work work;
9575 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9577 struct btrfs_delalloc_work *delalloc_work;
9578 struct inode *inode;
9580 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9582 inode = delalloc_work->inode;
9583 filemap_flush(inode->i_mapping);
9584 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9585 &BTRFS_I(inode)->runtime_flags))
9586 filemap_flush(inode->i_mapping);
9589 complete(&delalloc_work->completion);
9592 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9594 struct btrfs_delalloc_work *work;
9596 work = kmalloc(sizeof(*work), GFP_NOFS);
9600 init_completion(&work->completion);
9601 INIT_LIST_HEAD(&work->list);
9602 work->inode = inode;
9603 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9609 * some fairly slow code that needs optimization. This walks the list
9610 * of all the inodes with pending delalloc and forces them to disk.
9612 static int start_delalloc_inodes(struct btrfs_root *root,
9613 struct writeback_control *wbc, bool snapshot,
9614 bool in_reclaim_context)
9616 struct btrfs_inode *binode;
9617 struct inode *inode;
9618 struct btrfs_delalloc_work *work, *next;
9619 struct list_head works;
9620 struct list_head splice;
9622 bool full_flush = wbc->nr_to_write == LONG_MAX;
9624 INIT_LIST_HEAD(&works);
9625 INIT_LIST_HEAD(&splice);
9627 mutex_lock(&root->delalloc_mutex);
9628 spin_lock(&root->delalloc_lock);
9629 list_splice_init(&root->delalloc_inodes, &splice);
9630 while (!list_empty(&splice)) {
9631 binode = list_entry(splice.next, struct btrfs_inode,
9634 list_move_tail(&binode->delalloc_inodes,
9635 &root->delalloc_inodes);
9637 if (in_reclaim_context &&
9638 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9641 inode = igrab(&binode->vfs_inode);
9643 cond_resched_lock(&root->delalloc_lock);
9646 spin_unlock(&root->delalloc_lock);
9649 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9650 &binode->runtime_flags);
9652 work = btrfs_alloc_delalloc_work(inode);
9658 list_add_tail(&work->list, &works);
9659 btrfs_queue_work(root->fs_info->flush_workers,
9662 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9663 btrfs_add_delayed_iput(inode);
9664 if (ret || wbc->nr_to_write <= 0)
9668 spin_lock(&root->delalloc_lock);
9670 spin_unlock(&root->delalloc_lock);
9673 list_for_each_entry_safe(work, next, &works, list) {
9674 list_del_init(&work->list);
9675 wait_for_completion(&work->completion);
9679 if (!list_empty(&splice)) {
9680 spin_lock(&root->delalloc_lock);
9681 list_splice_tail(&splice, &root->delalloc_inodes);
9682 spin_unlock(&root->delalloc_lock);
9684 mutex_unlock(&root->delalloc_mutex);
9688 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9690 struct writeback_control wbc = {
9691 .nr_to_write = LONG_MAX,
9692 .sync_mode = WB_SYNC_NONE,
9694 .range_end = LLONG_MAX,
9696 struct btrfs_fs_info *fs_info = root->fs_info;
9698 if (BTRFS_FS_ERROR(fs_info))
9701 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9704 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9705 bool in_reclaim_context)
9707 struct writeback_control wbc = {
9709 .sync_mode = WB_SYNC_NONE,
9711 .range_end = LLONG_MAX,
9713 struct btrfs_root *root;
9714 struct list_head splice;
9717 if (BTRFS_FS_ERROR(fs_info))
9720 INIT_LIST_HEAD(&splice);
9722 mutex_lock(&fs_info->delalloc_root_mutex);
9723 spin_lock(&fs_info->delalloc_root_lock);
9724 list_splice_init(&fs_info->delalloc_roots, &splice);
9725 while (!list_empty(&splice)) {
9727 * Reset nr_to_write here so we know that we're doing a full
9731 wbc.nr_to_write = LONG_MAX;
9733 root = list_first_entry(&splice, struct btrfs_root,
9735 root = btrfs_grab_root(root);
9737 list_move_tail(&root->delalloc_root,
9738 &fs_info->delalloc_roots);
9739 spin_unlock(&fs_info->delalloc_root_lock);
9741 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9742 btrfs_put_root(root);
9743 if (ret < 0 || wbc.nr_to_write <= 0)
9745 spin_lock(&fs_info->delalloc_root_lock);
9747 spin_unlock(&fs_info->delalloc_root_lock);
9751 if (!list_empty(&splice)) {
9752 spin_lock(&fs_info->delalloc_root_lock);
9753 list_splice_tail(&splice, &fs_info->delalloc_roots);
9754 spin_unlock(&fs_info->delalloc_root_lock);
9756 mutex_unlock(&fs_info->delalloc_root_mutex);
9760 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9761 struct dentry *dentry, const char *symname)
9763 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9764 struct btrfs_trans_handle *trans;
9765 struct btrfs_root *root = BTRFS_I(dir)->root;
9766 struct btrfs_path *path;
9767 struct btrfs_key key;
9768 struct inode *inode;
9769 struct btrfs_new_inode_args new_inode_args = {
9773 unsigned int trans_num_items;
9778 struct btrfs_file_extent_item *ei;
9779 struct extent_buffer *leaf;
9781 name_len = strlen(symname);
9782 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9783 return -ENAMETOOLONG;
9785 inode = new_inode(dir->i_sb);
9788 inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO);
9789 inode->i_op = &btrfs_symlink_inode_operations;
9790 inode_nohighmem(inode);
9791 inode->i_mapping->a_ops = &btrfs_aops;
9792 btrfs_i_size_write(BTRFS_I(inode), name_len);
9793 inode_set_bytes(inode, name_len);
9795 new_inode_args.inode = inode;
9796 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9799 /* 1 additional item for the inline extent */
9802 trans = btrfs_start_transaction(root, trans_num_items);
9803 if (IS_ERR(trans)) {
9804 err = PTR_ERR(trans);
9805 goto out_new_inode_args;
9808 err = btrfs_create_new_inode(trans, &new_inode_args);
9812 path = btrfs_alloc_path();
9815 btrfs_abort_transaction(trans, err);
9816 discard_new_inode(inode);
9820 key.objectid = btrfs_ino(BTRFS_I(inode));
9822 key.type = BTRFS_EXTENT_DATA_KEY;
9823 datasize = btrfs_file_extent_calc_inline_size(name_len);
9824 err = btrfs_insert_empty_item(trans, root, path, &key,
9827 btrfs_abort_transaction(trans, err);
9828 btrfs_free_path(path);
9829 discard_new_inode(inode);
9833 leaf = path->nodes[0];
9834 ei = btrfs_item_ptr(leaf, path->slots[0],
9835 struct btrfs_file_extent_item);
9836 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9837 btrfs_set_file_extent_type(leaf, ei,
9838 BTRFS_FILE_EXTENT_INLINE);
9839 btrfs_set_file_extent_encryption(leaf, ei, 0);
9840 btrfs_set_file_extent_compression(leaf, ei, 0);
9841 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9842 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9844 ptr = btrfs_file_extent_inline_start(ei);
9845 write_extent_buffer(leaf, symname, ptr, name_len);
9846 btrfs_mark_buffer_dirty(leaf);
9847 btrfs_free_path(path);
9849 d_instantiate_new(dentry, inode);
9852 btrfs_end_transaction(trans);
9853 btrfs_btree_balance_dirty(fs_info);
9855 btrfs_new_inode_args_destroy(&new_inode_args);
9862 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9863 struct btrfs_trans_handle *trans_in,
9864 struct btrfs_inode *inode,
9865 struct btrfs_key *ins,
9868 struct btrfs_file_extent_item stack_fi;
9869 struct btrfs_replace_extent_info extent_info;
9870 struct btrfs_trans_handle *trans = trans_in;
9871 struct btrfs_path *path;
9872 u64 start = ins->objectid;
9873 u64 len = ins->offset;
9874 int qgroup_released;
9877 memset(&stack_fi, 0, sizeof(stack_fi));
9879 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9880 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9881 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9882 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9883 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9884 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9885 /* Encryption and other encoding is reserved and all 0 */
9887 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9888 if (qgroup_released < 0)
9889 return ERR_PTR(qgroup_released);
9892 ret = insert_reserved_file_extent(trans, inode,
9893 file_offset, &stack_fi,
9894 true, qgroup_released);
9900 extent_info.disk_offset = start;
9901 extent_info.disk_len = len;
9902 extent_info.data_offset = 0;
9903 extent_info.data_len = len;
9904 extent_info.file_offset = file_offset;
9905 extent_info.extent_buf = (char *)&stack_fi;
9906 extent_info.is_new_extent = true;
9907 extent_info.update_times = true;
9908 extent_info.qgroup_reserved = qgroup_released;
9909 extent_info.insertions = 0;
9911 path = btrfs_alloc_path();
9917 ret = btrfs_replace_file_extents(inode, path, file_offset,
9918 file_offset + len - 1, &extent_info,
9920 btrfs_free_path(path);
9927 * We have released qgroup data range at the beginning of the function,
9928 * and normally qgroup_released bytes will be freed when committing
9930 * But if we error out early, we have to free what we have released
9931 * or we leak qgroup data reservation.
9933 btrfs_qgroup_free_refroot(inode->root->fs_info,
9934 inode->root->root_key.objectid, qgroup_released,
9935 BTRFS_QGROUP_RSV_DATA);
9936 return ERR_PTR(ret);
9939 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9940 u64 start, u64 num_bytes, u64 min_size,
9941 loff_t actual_len, u64 *alloc_hint,
9942 struct btrfs_trans_handle *trans)
9944 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9945 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9946 struct extent_map *em;
9947 struct btrfs_root *root = BTRFS_I(inode)->root;
9948 struct btrfs_key ins;
9949 u64 cur_offset = start;
9950 u64 clear_offset = start;
9953 u64 last_alloc = (u64)-1;
9955 bool own_trans = true;
9956 u64 end = start + num_bytes - 1;
9960 while (num_bytes > 0) {
9961 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9962 cur_bytes = max(cur_bytes, min_size);
9964 * If we are severely fragmented we could end up with really
9965 * small allocations, so if the allocator is returning small
9966 * chunks lets make its job easier by only searching for those
9969 cur_bytes = min(cur_bytes, last_alloc);
9970 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9971 min_size, 0, *alloc_hint, &ins, 1, 0);
9976 * We've reserved this space, and thus converted it from
9977 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9978 * from here on out we will only need to clear our reservation
9979 * for the remaining unreserved area, so advance our
9980 * clear_offset by our extent size.
9982 clear_offset += ins.offset;
9984 last_alloc = ins.offset;
9985 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9988 * Now that we inserted the prealloc extent we can finally
9989 * decrement the number of reservations in the block group.
9990 * If we did it before, we could race with relocation and have
9991 * relocation miss the reserved extent, making it fail later.
9993 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9994 if (IS_ERR(trans)) {
9995 ret = PTR_ERR(trans);
9996 btrfs_free_reserved_extent(fs_info, ins.objectid,
10001 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10002 cur_offset + ins.offset -1, 0);
10004 em = alloc_extent_map();
10006 btrfs_set_inode_full_sync(BTRFS_I(inode));
10010 em->start = cur_offset;
10011 em->orig_start = cur_offset;
10012 em->len = ins.offset;
10013 em->block_start = ins.objectid;
10014 em->block_len = ins.offset;
10015 em->orig_block_len = ins.offset;
10016 em->ram_bytes = ins.offset;
10017 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10018 em->generation = trans->transid;
10021 write_lock(&em_tree->lock);
10022 ret = add_extent_mapping(em_tree, em, 1);
10023 write_unlock(&em_tree->lock);
10024 if (ret != -EEXIST)
10026 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10027 cur_offset + ins.offset - 1,
10030 free_extent_map(em);
10032 num_bytes -= ins.offset;
10033 cur_offset += ins.offset;
10034 *alloc_hint = ins.objectid + ins.offset;
10036 inode_inc_iversion(inode);
10037 inode->i_ctime = current_time(inode);
10038 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10039 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10040 (actual_len > inode->i_size) &&
10041 (cur_offset > inode->i_size)) {
10042 if (cur_offset > actual_len)
10043 i_size = actual_len;
10045 i_size = cur_offset;
10046 i_size_write(inode, i_size);
10047 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10050 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10053 btrfs_abort_transaction(trans, ret);
10055 btrfs_end_transaction(trans);
10060 btrfs_end_transaction(trans);
10064 if (clear_offset < end)
10065 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10066 end - clear_offset + 1);
10070 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10071 u64 start, u64 num_bytes, u64 min_size,
10072 loff_t actual_len, u64 *alloc_hint)
10074 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10075 min_size, actual_len, alloc_hint,
10079 int btrfs_prealloc_file_range_trans(struct inode *inode,
10080 struct btrfs_trans_handle *trans, int mode,
10081 u64 start, u64 num_bytes, u64 min_size,
10082 loff_t actual_len, u64 *alloc_hint)
10084 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10085 min_size, actual_len, alloc_hint, trans);
10088 static int btrfs_permission(struct user_namespace *mnt_userns,
10089 struct inode *inode, int mask)
10091 struct btrfs_root *root = BTRFS_I(inode)->root;
10092 umode_t mode = inode->i_mode;
10094 if (mask & MAY_WRITE &&
10095 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10096 if (btrfs_root_readonly(root))
10098 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10101 return generic_permission(mnt_userns, inode, mask);
10104 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10105 struct dentry *dentry, umode_t mode)
10107 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10108 struct btrfs_trans_handle *trans;
10109 struct btrfs_root *root = BTRFS_I(dir)->root;
10110 struct inode *inode;
10111 struct btrfs_new_inode_args new_inode_args = {
10116 unsigned int trans_num_items;
10119 inode = new_inode(dir->i_sb);
10122 inode_init_owner(mnt_userns, inode, dir, mode);
10123 inode->i_fop = &btrfs_file_operations;
10124 inode->i_op = &btrfs_file_inode_operations;
10125 inode->i_mapping->a_ops = &btrfs_aops;
10127 new_inode_args.inode = inode;
10128 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
10132 trans = btrfs_start_transaction(root, trans_num_items);
10133 if (IS_ERR(trans)) {
10134 ret = PTR_ERR(trans);
10135 goto out_new_inode_args;
10138 ret = btrfs_create_new_inode(trans, &new_inode_args);
10141 * We set number of links to 0 in btrfs_create_new_inode(), and here we
10142 * set it to 1 because d_tmpfile() will issue a warning if the count is
10145 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10147 set_nlink(inode, 1);
10150 d_tmpfile(dentry, inode);
10151 unlock_new_inode(inode);
10152 mark_inode_dirty(inode);
10155 btrfs_end_transaction(trans);
10156 btrfs_btree_balance_dirty(fs_info);
10157 out_new_inode_args:
10158 btrfs_new_inode_args_destroy(&new_inode_args);
10165 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10167 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10168 unsigned long index = start >> PAGE_SHIFT;
10169 unsigned long end_index = end >> PAGE_SHIFT;
10173 ASSERT(end + 1 - start <= U32_MAX);
10174 len = end + 1 - start;
10175 while (index <= end_index) {
10176 page = find_get_page(inode->vfs_inode.i_mapping, index);
10177 ASSERT(page); /* Pages should be in the extent_io_tree */
10179 btrfs_page_set_writeback(fs_info, page, start, len);
10185 static int btrfs_encoded_io_compression_from_extent(
10186 struct btrfs_fs_info *fs_info,
10189 switch (compress_type) {
10190 case BTRFS_COMPRESS_NONE:
10191 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10192 case BTRFS_COMPRESS_ZLIB:
10193 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10194 case BTRFS_COMPRESS_LZO:
10196 * The LZO format depends on the sector size. 64K is the maximum
10197 * sector size that we support.
10199 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10201 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10202 (fs_info->sectorsize_bits - 12);
10203 case BTRFS_COMPRESS_ZSTD:
10204 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10210 static ssize_t btrfs_encoded_read_inline(
10211 struct kiocb *iocb,
10212 struct iov_iter *iter, u64 start,
10214 struct extent_state **cached_state,
10215 u64 extent_start, size_t count,
10216 struct btrfs_ioctl_encoded_io_args *encoded,
10219 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10220 struct btrfs_root *root = inode->root;
10221 struct btrfs_fs_info *fs_info = root->fs_info;
10222 struct extent_io_tree *io_tree = &inode->io_tree;
10223 struct btrfs_path *path;
10224 struct extent_buffer *leaf;
10225 struct btrfs_file_extent_item *item;
10231 path = btrfs_alloc_path();
10236 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10240 /* The extent item disappeared? */
10245 leaf = path->nodes[0];
10246 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10248 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10249 ptr = btrfs_file_extent_inline_start(item);
10251 encoded->len = min_t(u64, extent_start + ram_bytes,
10252 inode->vfs_inode.i_size) - iocb->ki_pos;
10253 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10254 btrfs_file_extent_compression(leaf, item));
10257 encoded->compression = ret;
10258 if (encoded->compression) {
10259 size_t inline_size;
10261 inline_size = btrfs_file_extent_inline_item_len(leaf,
10263 if (inline_size > count) {
10267 count = inline_size;
10268 encoded->unencoded_len = ram_bytes;
10269 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10271 count = min_t(u64, count, encoded->len);
10272 encoded->len = count;
10273 encoded->unencoded_len = count;
10274 ptr += iocb->ki_pos - extent_start;
10277 tmp = kmalloc(count, GFP_NOFS);
10282 read_extent_buffer(leaf, tmp, ptr, count);
10283 btrfs_release_path(path);
10284 unlock_extent_cached(io_tree, start, lockend, cached_state);
10285 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10288 ret = copy_to_iter(tmp, count, iter);
10293 btrfs_free_path(path);
10297 struct btrfs_encoded_read_private {
10298 struct btrfs_inode *inode;
10300 wait_queue_head_t wait;
10302 blk_status_t status;
10306 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10307 struct bio *bio, int mirror_num)
10309 struct btrfs_encoded_read_private *priv = bio->bi_private;
10310 struct btrfs_bio *bbio = btrfs_bio(bio);
10311 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10314 if (!priv->skip_csum) {
10315 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10320 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
10322 btrfs_bio_free_csum(bbio);
10326 atomic_inc(&priv->pending);
10327 ret = btrfs_map_bio(fs_info, bio, mirror_num);
10329 atomic_dec(&priv->pending);
10330 btrfs_bio_free_csum(bbio);
10335 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10337 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10338 struct btrfs_encoded_read_private *priv = bbio->bio.bi_private;
10339 struct btrfs_inode *inode = priv->inode;
10340 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10341 u32 sectorsize = fs_info->sectorsize;
10342 struct bio_vec *bvec;
10343 struct bvec_iter_all iter_all;
10344 u64 start = priv->file_offset;
10345 u32 bio_offset = 0;
10347 if (priv->skip_csum || !uptodate)
10348 return bbio->bio.bi_status;
10350 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10351 unsigned int i, nr_sectors, pgoff;
10353 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10354 pgoff = bvec->bv_offset;
10355 for (i = 0; i < nr_sectors; i++) {
10356 ASSERT(pgoff < PAGE_SIZE);
10357 if (check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10358 bvec->bv_page, pgoff, start))
10359 return BLK_STS_IOERR;
10360 start += sectorsize;
10361 bio_offset += sectorsize;
10362 pgoff += sectorsize;
10368 static void btrfs_encoded_read_endio(struct bio *bio)
10370 struct btrfs_encoded_read_private *priv = bio->bi_private;
10371 struct btrfs_bio *bbio = btrfs_bio(bio);
10372 blk_status_t status;
10374 status = btrfs_encoded_read_verify_csum(bbio);
10377 * The memory barrier implied by the atomic_dec_return() here
10378 * pairs with the memory barrier implied by the
10379 * atomic_dec_return() or io_wait_event() in
10380 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10381 * write is observed before the load of status in
10382 * btrfs_encoded_read_regular_fill_pages().
10384 WRITE_ONCE(priv->status, status);
10386 if (!atomic_dec_return(&priv->pending))
10387 wake_up(&priv->wait);
10388 btrfs_bio_free_csum(bbio);
10392 static int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10396 struct page **pages)
10398 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10399 struct btrfs_encoded_read_private priv = {
10401 .file_offset = file_offset,
10402 .pending = ATOMIC_INIT(1),
10403 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10405 unsigned long i = 0;
10409 init_waitqueue_head(&priv.wait);
10411 * Submit bios for the extent, splitting due to bio or stripe limits as
10414 while (cur < disk_io_size) {
10415 struct extent_map *em;
10416 struct btrfs_io_geometry geom;
10417 struct bio *bio = NULL;
10420 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10421 disk_io_size - cur);
10425 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10426 disk_bytenr + cur, &geom);
10427 free_extent_map(em);
10430 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10433 remaining = min(geom.len, disk_io_size - cur);
10434 while (bio || remaining) {
10435 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10438 bio = btrfs_bio_alloc(BIO_MAX_VECS);
10439 bio->bi_iter.bi_sector =
10440 (disk_bytenr + cur) >> SECTOR_SHIFT;
10441 bio->bi_end_io = btrfs_encoded_read_endio;
10442 bio->bi_private = &priv;
10443 bio->bi_opf = REQ_OP_READ;
10447 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10448 blk_status_t status;
10450 status = submit_encoded_read_bio(inode, bio, 0);
10452 WRITE_ONCE(priv.status, status);
10462 remaining -= bytes;
10467 if (atomic_dec_return(&priv.pending))
10468 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10469 /* See btrfs_encoded_read_endio() for ordering. */
10470 return blk_status_to_errno(READ_ONCE(priv.status));
10473 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10474 struct iov_iter *iter,
10475 u64 start, u64 lockend,
10476 struct extent_state **cached_state,
10477 u64 disk_bytenr, u64 disk_io_size,
10478 size_t count, bool compressed,
10481 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10482 struct extent_io_tree *io_tree = &inode->io_tree;
10483 struct page **pages;
10484 unsigned long nr_pages, i;
10486 size_t page_offset;
10489 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10490 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10493 ret = btrfs_alloc_page_array(nr_pages, pages);
10499 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10500 disk_io_size, pages);
10504 unlock_extent_cached(io_tree, start, lockend, cached_state);
10505 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10512 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10513 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10516 while (cur < count) {
10517 size_t bytes = min_t(size_t, count - cur,
10518 PAGE_SIZE - page_offset);
10520 if (copy_page_to_iter(pages[i], page_offset, bytes,
10531 for (i = 0; i < nr_pages; i++) {
10533 __free_page(pages[i]);
10539 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10540 struct btrfs_ioctl_encoded_io_args *encoded)
10542 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10543 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10544 struct extent_io_tree *io_tree = &inode->io_tree;
10546 size_t count = iov_iter_count(iter);
10547 u64 start, lockend, disk_bytenr, disk_io_size;
10548 struct extent_state *cached_state = NULL;
10549 struct extent_map *em;
10550 bool unlocked = false;
10552 file_accessed(iocb->ki_filp);
10554 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10556 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10557 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10560 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10562 * We don't know how long the extent containing iocb->ki_pos is, but if
10563 * it's compressed we know that it won't be longer than this.
10565 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10568 struct btrfs_ordered_extent *ordered;
10570 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10571 lockend - start + 1);
10573 goto out_unlock_inode;
10574 lock_extent_bits(io_tree, start, lockend, &cached_state);
10575 ordered = btrfs_lookup_ordered_range(inode, start,
10576 lockend - start + 1);
10579 btrfs_put_ordered_extent(ordered);
10580 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10584 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10587 goto out_unlock_extent;
10590 if (em->block_start == EXTENT_MAP_INLINE) {
10591 u64 extent_start = em->start;
10594 * For inline extents we get everything we need out of the
10597 free_extent_map(em);
10599 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10600 &cached_state, extent_start,
10601 count, encoded, &unlocked);
10606 * We only want to return up to EOF even if the extent extends beyond
10609 encoded->len = min_t(u64, extent_map_end(em),
10610 inode->vfs_inode.i_size) - iocb->ki_pos;
10611 if (em->block_start == EXTENT_MAP_HOLE ||
10612 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10613 disk_bytenr = EXTENT_MAP_HOLE;
10614 count = min_t(u64, count, encoded->len);
10615 encoded->len = count;
10616 encoded->unencoded_len = count;
10617 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10618 disk_bytenr = em->block_start;
10620 * Bail if the buffer isn't large enough to return the whole
10621 * compressed extent.
10623 if (em->block_len > count) {
10627 disk_io_size = count = em->block_len;
10628 encoded->unencoded_len = em->ram_bytes;
10629 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10630 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10631 em->compress_type);
10634 encoded->compression = ret;
10636 disk_bytenr = em->block_start + (start - em->start);
10637 if (encoded->len > count)
10638 encoded->len = count;
10640 * Don't read beyond what we locked. This also limits the page
10641 * allocations that we'll do.
10643 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10644 count = start + disk_io_size - iocb->ki_pos;
10645 encoded->len = count;
10646 encoded->unencoded_len = count;
10647 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10649 free_extent_map(em);
10652 if (disk_bytenr == EXTENT_MAP_HOLE) {
10653 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10654 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10656 ret = iov_iter_zero(count, iter);
10660 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10661 &cached_state, disk_bytenr,
10662 disk_io_size, count,
10663 encoded->compression,
10669 iocb->ki_pos += encoded->len;
10671 free_extent_map(em);
10674 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10677 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10681 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10682 const struct btrfs_ioctl_encoded_io_args *encoded)
10684 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10685 struct btrfs_root *root = inode->root;
10686 struct btrfs_fs_info *fs_info = root->fs_info;
10687 struct extent_io_tree *io_tree = &inode->io_tree;
10688 struct extent_changeset *data_reserved = NULL;
10689 struct extent_state *cached_state = NULL;
10693 u64 num_bytes, ram_bytes, disk_num_bytes;
10694 unsigned long nr_pages, i;
10695 struct page **pages;
10696 struct btrfs_key ins;
10697 bool extent_reserved = false;
10698 struct extent_map *em;
10701 switch (encoded->compression) {
10702 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10703 compression = BTRFS_COMPRESS_ZLIB;
10705 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10706 compression = BTRFS_COMPRESS_ZSTD;
10708 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10709 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10710 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10711 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10712 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10713 /* The sector size must match for LZO. */
10714 if (encoded->compression -
10715 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10716 fs_info->sectorsize_bits)
10718 compression = BTRFS_COMPRESS_LZO;
10723 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10726 orig_count = iov_iter_count(from);
10728 /* The extent size must be sane. */
10729 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10730 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10734 * The compressed data must be smaller than the decompressed data.
10736 * It's of course possible for data to compress to larger or the same
10737 * size, but the buffered I/O path falls back to no compression for such
10738 * data, and we don't want to break any assumptions by creating these
10741 * Note that this is less strict than the current check we have that the
10742 * compressed data must be at least one sector smaller than the
10743 * decompressed data. We only want to enforce the weaker requirement
10744 * from old kernels that it is at least one byte smaller.
10746 if (orig_count >= encoded->unencoded_len)
10749 /* The extent must start on a sector boundary. */
10750 start = iocb->ki_pos;
10751 if (!IS_ALIGNED(start, fs_info->sectorsize))
10755 * The extent must end on a sector boundary. However, we allow a write
10756 * which ends at or extends i_size to have an unaligned length; we round
10757 * up the extent size and set i_size to the unaligned end.
10759 if (start + encoded->len < inode->vfs_inode.i_size &&
10760 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10763 /* Finally, the offset in the unencoded data must be sector-aligned. */
10764 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10767 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10768 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10769 end = start + num_bytes - 1;
10772 * If the extent cannot be inline, the compressed data on disk must be
10773 * sector-aligned. For convenience, we extend it with zeroes if it
10776 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10777 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10778 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10781 for (i = 0; i < nr_pages; i++) {
10782 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10785 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10790 kaddr = kmap(pages[i]);
10791 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10796 if (bytes < PAGE_SIZE)
10797 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10802 struct btrfs_ordered_extent *ordered;
10804 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10807 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10808 start >> PAGE_SHIFT,
10809 end >> PAGE_SHIFT);
10812 lock_extent_bits(io_tree, start, end, &cached_state);
10813 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10815 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10818 btrfs_put_ordered_extent(ordered);
10819 unlock_extent_cached(io_tree, start, end, &cached_state);
10824 * We don't use the higher-level delalloc space functions because our
10825 * num_bytes and disk_num_bytes are different.
10827 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10830 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10832 goto out_free_data_space;
10833 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10836 goto out_qgroup_free_data;
10838 /* Try an inline extent first. */
10839 if (start == 0 && encoded->unencoded_len == encoded->len &&
10840 encoded->unencoded_offset == 0) {
10841 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10842 compression, pages, true);
10846 goto out_delalloc_release;
10850 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10851 disk_num_bytes, 0, 0, &ins, 1, 1);
10853 goto out_delalloc_release;
10854 extent_reserved = true;
10856 em = create_io_em(inode, start, num_bytes,
10857 start - encoded->unencoded_offset, ins.objectid,
10858 ins.offset, ins.offset, ram_bytes, compression,
10859 BTRFS_ORDERED_COMPRESSED);
10862 goto out_free_reserved;
10864 free_extent_map(em);
10866 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10867 ins.objectid, ins.offset,
10868 encoded->unencoded_offset,
10869 (1 << BTRFS_ORDERED_ENCODED) |
10870 (1 << BTRFS_ORDERED_COMPRESSED),
10873 btrfs_drop_extent_cache(inode, start, end, 0);
10874 goto out_free_reserved;
10876 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10878 if (start + encoded->len > inode->vfs_inode.i_size)
10879 i_size_write(&inode->vfs_inode, start + encoded->len);
10881 unlock_extent_cached(io_tree, start, end, &cached_state);
10883 btrfs_delalloc_release_extents(inode, num_bytes);
10885 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10886 ins.offset, pages, nr_pages, 0, NULL,
10888 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10896 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10897 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10898 out_delalloc_release:
10899 btrfs_delalloc_release_extents(inode, num_bytes);
10900 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10901 out_qgroup_free_data:
10903 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10904 out_free_data_space:
10906 * If btrfs_reserve_extent() succeeded, then we already decremented
10909 if (!extent_reserved)
10910 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10912 unlock_extent_cached(io_tree, start, end, &cached_state);
10914 for (i = 0; i < nr_pages; i++) {
10916 __free_page(pages[i]);
10921 iocb->ki_pos += encoded->len;
10927 * Add an entry indicating a block group or device which is pinned by a
10928 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10929 * negative errno on failure.
10931 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10932 bool is_block_group)
10934 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10935 struct btrfs_swapfile_pin *sp, *entry;
10936 struct rb_node **p;
10937 struct rb_node *parent = NULL;
10939 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10944 sp->is_block_group = is_block_group;
10945 sp->bg_extent_count = 1;
10947 spin_lock(&fs_info->swapfile_pins_lock);
10948 p = &fs_info->swapfile_pins.rb_node;
10951 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10952 if (sp->ptr < entry->ptr ||
10953 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10954 p = &(*p)->rb_left;
10955 } else if (sp->ptr > entry->ptr ||
10956 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10957 p = &(*p)->rb_right;
10959 if (is_block_group)
10960 entry->bg_extent_count++;
10961 spin_unlock(&fs_info->swapfile_pins_lock);
10966 rb_link_node(&sp->node, parent, p);
10967 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10968 spin_unlock(&fs_info->swapfile_pins_lock);
10972 /* Free all of the entries pinned by this swapfile. */
10973 static void btrfs_free_swapfile_pins(struct inode *inode)
10975 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10976 struct btrfs_swapfile_pin *sp;
10977 struct rb_node *node, *next;
10979 spin_lock(&fs_info->swapfile_pins_lock);
10980 node = rb_first(&fs_info->swapfile_pins);
10982 next = rb_next(node);
10983 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10984 if (sp->inode == inode) {
10985 rb_erase(&sp->node, &fs_info->swapfile_pins);
10986 if (sp->is_block_group) {
10987 btrfs_dec_block_group_swap_extents(sp->ptr,
10988 sp->bg_extent_count);
10989 btrfs_put_block_group(sp->ptr);
10995 spin_unlock(&fs_info->swapfile_pins_lock);
10998 struct btrfs_swap_info {
11004 unsigned long nr_pages;
11008 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
11009 struct btrfs_swap_info *bsi)
11011 unsigned long nr_pages;
11012 unsigned long max_pages;
11013 u64 first_ppage, first_ppage_reported, next_ppage;
11017 * Our swapfile may have had its size extended after the swap header was
11018 * written. In that case activating the swapfile should not go beyond
11019 * the max size set in the swap header.
11021 if (bsi->nr_pages >= sis->max)
11024 max_pages = sis->max - bsi->nr_pages;
11025 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
11026 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
11027 PAGE_SIZE) >> PAGE_SHIFT;
11029 if (first_ppage >= next_ppage)
11031 nr_pages = next_ppage - first_ppage;
11032 nr_pages = min(nr_pages, max_pages);
11034 first_ppage_reported = first_ppage;
11035 if (bsi->start == 0)
11036 first_ppage_reported++;
11037 if (bsi->lowest_ppage > first_ppage_reported)
11038 bsi->lowest_ppage = first_ppage_reported;
11039 if (bsi->highest_ppage < (next_ppage - 1))
11040 bsi->highest_ppage = next_ppage - 1;
11042 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
11045 bsi->nr_extents += ret;
11046 bsi->nr_pages += nr_pages;
11050 static void btrfs_swap_deactivate(struct file *file)
11052 struct inode *inode = file_inode(file);
11054 btrfs_free_swapfile_pins(inode);
11055 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
11058 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11061 struct inode *inode = file_inode(file);
11062 struct btrfs_root *root = BTRFS_I(inode)->root;
11063 struct btrfs_fs_info *fs_info = root->fs_info;
11064 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
11065 struct extent_state *cached_state = NULL;
11066 struct extent_map *em = NULL;
11067 struct btrfs_device *device = NULL;
11068 struct btrfs_swap_info bsi = {
11069 .lowest_ppage = (sector_t)-1ULL,
11076 * If the swap file was just created, make sure delalloc is done. If the
11077 * file changes again after this, the user is doing something stupid and
11078 * we don't really care.
11080 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
11085 * The inode is locked, so these flags won't change after we check them.
11087 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
11088 btrfs_warn(fs_info, "swapfile must not be compressed");
11091 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
11092 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
11095 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
11096 btrfs_warn(fs_info, "swapfile must not be checksummed");
11101 * Balance or device remove/replace/resize can move stuff around from
11102 * under us. The exclop protection makes sure they aren't running/won't
11103 * run concurrently while we are mapping the swap extents, and
11104 * fs_info->swapfile_pins prevents them from running while the swap
11105 * file is active and moving the extents. Note that this also prevents
11106 * a concurrent device add which isn't actually necessary, but it's not
11107 * really worth the trouble to allow it.
11109 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
11110 btrfs_warn(fs_info,
11111 "cannot activate swapfile while exclusive operation is running");
11116 * Prevent snapshot creation while we are activating the swap file.
11117 * We do not want to race with snapshot creation. If snapshot creation
11118 * already started before we bumped nr_swapfiles from 0 to 1 and
11119 * completes before the first write into the swap file after it is
11120 * activated, than that write would fallback to COW.
11122 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11123 btrfs_exclop_finish(fs_info);
11124 btrfs_warn(fs_info,
11125 "cannot activate swapfile because snapshot creation is in progress");
11129 * Snapshots can create extents which require COW even if NODATACOW is
11130 * set. We use this counter to prevent snapshots. We must increment it
11131 * before walking the extents because we don't want a concurrent
11132 * snapshot to run after we've already checked the extents.
11134 * It is possible that subvolume is marked for deletion but still not
11135 * removed yet. To prevent this race, we check the root status before
11136 * activating the swapfile.
11138 spin_lock(&root->root_item_lock);
11139 if (btrfs_root_dead(root)) {
11140 spin_unlock(&root->root_item_lock);
11142 btrfs_exclop_finish(fs_info);
11143 btrfs_warn(fs_info,
11144 "cannot activate swapfile because subvolume %llu is being deleted",
11145 root->root_key.objectid);
11148 atomic_inc(&root->nr_swapfiles);
11149 spin_unlock(&root->root_item_lock);
11151 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11153 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
11155 while (start < isize) {
11156 u64 logical_block_start, physical_block_start;
11157 struct btrfs_block_group *bg;
11158 u64 len = isize - start;
11160 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11166 if (em->block_start == EXTENT_MAP_HOLE) {
11167 btrfs_warn(fs_info, "swapfile must not have holes");
11171 if (em->block_start == EXTENT_MAP_INLINE) {
11173 * It's unlikely we'll ever actually find ourselves
11174 * here, as a file small enough to fit inline won't be
11175 * big enough to store more than the swap header, but in
11176 * case something changes in the future, let's catch it
11177 * here rather than later.
11179 btrfs_warn(fs_info, "swapfile must not be inline");
11183 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11184 btrfs_warn(fs_info, "swapfile must not be compressed");
11189 logical_block_start = em->block_start + (start - em->start);
11190 len = min(len, em->len - (start - em->start));
11191 free_extent_map(em);
11194 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
11200 btrfs_warn(fs_info,
11201 "swapfile must not be copy-on-write");
11206 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11212 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11213 btrfs_warn(fs_info,
11214 "swapfile must have single data profile");
11219 if (device == NULL) {
11220 device = em->map_lookup->stripes[0].dev;
11221 ret = btrfs_add_swapfile_pin(inode, device, false);
11226 } else if (device != em->map_lookup->stripes[0].dev) {
11227 btrfs_warn(fs_info, "swapfile must be on one device");
11232 physical_block_start = (em->map_lookup->stripes[0].physical +
11233 (logical_block_start - em->start));
11234 len = min(len, em->len - (logical_block_start - em->start));
11235 free_extent_map(em);
11238 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11240 btrfs_warn(fs_info,
11241 "could not find block group containing swapfile");
11246 if (!btrfs_inc_block_group_swap_extents(bg)) {
11247 btrfs_warn(fs_info,
11248 "block group for swapfile at %llu is read-only%s",
11250 atomic_read(&fs_info->scrubs_running) ?
11251 " (scrub running)" : "");
11252 btrfs_put_block_group(bg);
11257 ret = btrfs_add_swapfile_pin(inode, bg, true);
11259 btrfs_put_block_group(bg);
11266 if (bsi.block_len &&
11267 bsi.block_start + bsi.block_len == physical_block_start) {
11268 bsi.block_len += len;
11270 if (bsi.block_len) {
11271 ret = btrfs_add_swap_extent(sis, &bsi);
11276 bsi.block_start = physical_block_start;
11277 bsi.block_len = len;
11284 ret = btrfs_add_swap_extent(sis, &bsi);
11287 if (!IS_ERR_OR_NULL(em))
11288 free_extent_map(em);
11290 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
11293 btrfs_swap_deactivate(file);
11295 btrfs_drew_write_unlock(&root->snapshot_lock);
11297 btrfs_exclop_finish(fs_info);
11303 sis->bdev = device->bdev;
11304 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11305 sis->max = bsi.nr_pages;
11306 sis->pages = bsi.nr_pages - 1;
11307 sis->highest_bit = bsi.nr_pages - 1;
11308 return bsi.nr_extents;
11311 static void btrfs_swap_deactivate(struct file *file)
11315 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11318 return -EOPNOTSUPP;
11323 * Update the number of bytes used in the VFS' inode. When we replace extents in
11324 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11325 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11326 * always get a correct value.
11328 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11329 const u64 add_bytes,
11330 const u64 del_bytes)
11332 if (add_bytes == del_bytes)
11335 spin_lock(&inode->lock);
11337 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11339 inode_add_bytes(&inode->vfs_inode, add_bytes);
11340 spin_unlock(&inode->lock);
11344 * Verify that there are no ordered extents for a given file range.
11346 * @inode: The target inode.
11347 * @start: Start offset of the file range, should be sector size aligned.
11348 * @end: End offset (inclusive) of the file range, its value +1 should be
11349 * sector size aligned.
11351 * This should typically be used for cases where we locked an inode's VFS lock in
11352 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11353 * we have flushed all delalloc in the range, we have waited for all ordered
11354 * extents in the range to complete and finally we have locked the file range in
11355 * the inode's io_tree.
11357 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11359 struct btrfs_root *root = inode->root;
11360 struct btrfs_ordered_extent *ordered;
11362 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11365 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11367 btrfs_err(root->fs_info,
11368 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11369 start, end, btrfs_ino(inode), root->root_key.objectid,
11370 ordered->file_offset,
11371 ordered->file_offset + ordered->num_bytes - 1);
11372 btrfs_put_ordered_extent(ordered);
11375 ASSERT(ordered == NULL);
11378 static const struct inode_operations btrfs_dir_inode_operations = {
11379 .getattr = btrfs_getattr,
11380 .lookup = btrfs_lookup,
11381 .create = btrfs_create,
11382 .unlink = btrfs_unlink,
11383 .link = btrfs_link,
11384 .mkdir = btrfs_mkdir,
11385 .rmdir = btrfs_rmdir,
11386 .rename = btrfs_rename2,
11387 .symlink = btrfs_symlink,
11388 .setattr = btrfs_setattr,
11389 .mknod = btrfs_mknod,
11390 .listxattr = btrfs_listxattr,
11391 .permission = btrfs_permission,
11392 .get_acl = btrfs_get_acl,
11393 .set_acl = btrfs_set_acl,
11394 .update_time = btrfs_update_time,
11395 .tmpfile = btrfs_tmpfile,
11396 .fileattr_get = btrfs_fileattr_get,
11397 .fileattr_set = btrfs_fileattr_set,
11400 static const struct file_operations btrfs_dir_file_operations = {
11401 .llseek = generic_file_llseek,
11402 .read = generic_read_dir,
11403 .iterate_shared = btrfs_real_readdir,
11404 .open = btrfs_opendir,
11405 .unlocked_ioctl = btrfs_ioctl,
11406 #ifdef CONFIG_COMPAT
11407 .compat_ioctl = btrfs_compat_ioctl,
11409 .release = btrfs_release_file,
11410 .fsync = btrfs_sync_file,
11414 * btrfs doesn't support the bmap operation because swapfiles
11415 * use bmap to make a mapping of extents in the file. They assume
11416 * these extents won't change over the life of the file and they
11417 * use the bmap result to do IO directly to the drive.
11419 * the btrfs bmap call would return logical addresses that aren't
11420 * suitable for IO and they also will change frequently as COW
11421 * operations happen. So, swapfile + btrfs == corruption.
11423 * For now we're avoiding this by dropping bmap.
11425 static const struct address_space_operations btrfs_aops = {
11426 .read_folio = btrfs_read_folio,
11427 .writepage = btrfs_writepage,
11428 .writepages = btrfs_writepages,
11429 .readahead = btrfs_readahead,
11430 .direct_IO = noop_direct_IO,
11431 .invalidate_folio = btrfs_invalidate_folio,
11432 .release_folio = btrfs_release_folio,
11433 .migrate_folio = btrfs_migrate_folio,
11434 .dirty_folio = filemap_dirty_folio,
11435 .error_remove_page = generic_error_remove_page,
11436 .swap_activate = btrfs_swap_activate,
11437 .swap_deactivate = btrfs_swap_deactivate,
11440 static const struct inode_operations btrfs_file_inode_operations = {
11441 .getattr = btrfs_getattr,
11442 .setattr = btrfs_setattr,
11443 .listxattr = btrfs_listxattr,
11444 .permission = btrfs_permission,
11445 .fiemap = btrfs_fiemap,
11446 .get_acl = btrfs_get_acl,
11447 .set_acl = btrfs_set_acl,
11448 .update_time = btrfs_update_time,
11449 .fileattr_get = btrfs_fileattr_get,
11450 .fileattr_set = btrfs_fileattr_set,
11452 static const struct inode_operations btrfs_special_inode_operations = {
11453 .getattr = btrfs_getattr,
11454 .setattr = btrfs_setattr,
11455 .permission = btrfs_permission,
11456 .listxattr = btrfs_listxattr,
11457 .get_acl = btrfs_get_acl,
11458 .set_acl = btrfs_set_acl,
11459 .update_time = btrfs_update_time,
11461 static const struct inode_operations btrfs_symlink_inode_operations = {
11462 .get_link = page_get_link,
11463 .getattr = btrfs_getattr,
11464 .setattr = btrfs_setattr,
11465 .permission = btrfs_permission,
11466 .listxattr = btrfs_listxattr,
11467 .update_time = btrfs_update_time,
11470 const struct dentry_operations btrfs_dentry_operations = {
11471 .d_delete = btrfs_dentry_delete,