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 unsigned int 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 unsigned int 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 +
3199 BUG_ON(root == fs_info->tree_root);
3200 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3202 clear_reserved_extent = false;
3203 btrfs_release_delalloc_bytes(fs_info,
3204 ordered_extent->disk_bytenr,
3205 ordered_extent->disk_num_bytes);
3208 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3209 ordered_extent->num_bytes, trans->transid);
3211 btrfs_abort_transaction(trans, ret);
3215 ret = add_pending_csums(trans, &ordered_extent->list);
3217 btrfs_abort_transaction(trans, ret);
3222 * If this is a new delalloc range, clear its new delalloc flag to
3223 * update the inode's number of bytes. This needs to be done first
3224 * before updating the inode item.
3226 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3227 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3228 clear_extent_bit(&inode->io_tree, start, end,
3229 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3230 0, 0, &cached_state);
3232 btrfs_inode_safe_disk_i_size_write(inode, 0);
3233 ret = btrfs_update_inode_fallback(trans, root, inode);
3234 if (ret) { /* -ENOMEM or corruption */
3235 btrfs_abort_transaction(trans, ret);
3240 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3241 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3245 btrfs_end_transaction(trans);
3247 if (ret || truncated) {
3248 u64 unwritten_start = start;
3251 * If we failed to finish this ordered extent for any reason we
3252 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3253 * extent, and mark the inode with the error if it wasn't
3254 * already set. Any error during writeback would have already
3255 * set the mapping error, so we need to set it if we're the ones
3256 * marking this ordered extent as failed.
3258 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3259 &ordered_extent->flags))
3260 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3263 unwritten_start += logical_len;
3264 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3266 /* Drop the cache for the part of the extent we didn't write. */
3267 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3270 * If the ordered extent had an IOERR or something else went
3271 * wrong we need to return the space for this ordered extent
3272 * back to the allocator. We only free the extent in the
3273 * truncated case if we didn't write out the extent at all.
3275 * If we made it past insert_reserved_file_extent before we
3276 * errored out then we don't need to do this as the accounting
3277 * has already been done.
3279 if ((ret || !logical_len) &&
3280 clear_reserved_extent &&
3281 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3282 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3284 * Discard the range before returning it back to the
3287 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3288 btrfs_discard_extent(fs_info,
3289 ordered_extent->disk_bytenr,
3290 ordered_extent->disk_num_bytes,
3292 btrfs_free_reserved_extent(fs_info,
3293 ordered_extent->disk_bytenr,
3294 ordered_extent->disk_num_bytes, 1);
3299 * This needs to be done to make sure anybody waiting knows we are done
3300 * updating everything for this ordered extent.
3302 btrfs_remove_ordered_extent(inode, ordered_extent);
3305 btrfs_put_ordered_extent(ordered_extent);
3306 /* once for the tree */
3307 btrfs_put_ordered_extent(ordered_extent);
3312 static void finish_ordered_fn(struct btrfs_work *work)
3314 struct btrfs_ordered_extent *ordered_extent;
3315 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3316 btrfs_finish_ordered_io(ordered_extent);
3319 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3320 struct page *page, u64 start,
3321 u64 end, bool uptodate)
3323 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3325 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3326 finish_ordered_fn, uptodate);
3330 * check_data_csum - verify checksum of one sector of uncompressed data
3332 * @io_bio: btrfs_io_bio which contains the csum
3333 * @bio_offset: offset to the beginning of the bio (in bytes)
3334 * @page: page where is the data to be verified
3335 * @pgoff: offset inside the page
3336 * @start: logical offset in the file
3338 * The length of such check is always one sector size.
3340 static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3341 u32 bio_offset, struct page *page, u32 pgoff,
3344 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3345 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3347 u32 len = fs_info->sectorsize;
3348 const u32 csum_size = fs_info->csum_size;
3349 unsigned int offset_sectors;
3351 u8 csum[BTRFS_CSUM_SIZE];
3353 ASSERT(pgoff + len <= PAGE_SIZE);
3355 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3356 csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size;
3358 kaddr = kmap_atomic(page);
3359 shash->tfm = fs_info->csum_shash;
3361 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3362 kunmap_atomic(kaddr);
3364 if (memcmp(csum, csum_expected, csum_size))
3369 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3372 btrfs_dev_stat_inc_and_print(bbio->device,
3373 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3374 memzero_page(page, pgoff, len);
3379 * When reads are done, we need to check csums to verify the data is correct.
3380 * if there's a match, we allow the bio to finish. If not, the code in
3381 * extent_io.c will try to find good copies for us.
3383 * @bio_offset: offset to the beginning of the bio (in bytes)
3384 * @start: file offset of the range start
3385 * @end: file offset of the range end (inclusive)
3387 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3390 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3391 u32 bio_offset, struct page *page,
3394 struct inode *inode = page->mapping->host;
3395 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3396 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3397 struct btrfs_root *root = BTRFS_I(inode)->root;
3398 const u32 sectorsize = root->fs_info->sectorsize;
3400 unsigned int result = 0;
3402 if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) {
3403 btrfs_page_clear_checked(fs_info, page, start, end + 1 - start);
3408 * This only happens for NODATASUM or compressed read.
3409 * Normally this should be covered by above check for compressed read
3410 * or the next check for NODATASUM. Just do a quicker exit here.
3412 if (bbio->csum == NULL)
3415 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3418 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3421 ASSERT(page_offset(page) <= start &&
3422 end <= page_offset(page) + PAGE_SIZE - 1);
3423 for (pg_off = offset_in_page(start);
3424 pg_off < offset_in_page(end);
3425 pg_off += sectorsize, bio_offset += sectorsize) {
3426 u64 file_offset = pg_off + page_offset(page);
3429 if (btrfs_is_data_reloc_root(root) &&
3430 test_range_bit(io_tree, file_offset,
3431 file_offset + sectorsize - 1,
3432 EXTENT_NODATASUM, 1, NULL)) {
3433 /* Skip the range without csum for data reloc inode */
3434 clear_extent_bits(io_tree, file_offset,
3435 file_offset + sectorsize - 1,
3439 ret = check_data_csum(inode, bbio, bio_offset, page, pg_off,
3440 page_offset(page) + pg_off);
3442 const int nr_bit = (pg_off - offset_in_page(start)) >>
3443 root->fs_info->sectorsize_bits;
3445 result |= (1U << nr_bit);
3452 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3454 * @inode: The inode we want to perform iput on
3456 * This function uses the generic vfs_inode::i_count to track whether we should
3457 * just decrement it (in case it's > 1) or if this is the last iput then link
3458 * the inode to the delayed iput machinery. Delayed iputs are processed at
3459 * transaction commit time/superblock commit/cleaner kthread.
3461 void btrfs_add_delayed_iput(struct inode *inode)
3463 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3464 struct btrfs_inode *binode = BTRFS_I(inode);
3466 if (atomic_add_unless(&inode->i_count, -1, 1))
3469 atomic_inc(&fs_info->nr_delayed_iputs);
3470 spin_lock(&fs_info->delayed_iput_lock);
3471 ASSERT(list_empty(&binode->delayed_iput));
3472 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3473 spin_unlock(&fs_info->delayed_iput_lock);
3474 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3475 wake_up_process(fs_info->cleaner_kthread);
3478 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3479 struct btrfs_inode *inode)
3481 list_del_init(&inode->delayed_iput);
3482 spin_unlock(&fs_info->delayed_iput_lock);
3483 iput(&inode->vfs_inode);
3484 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3485 wake_up(&fs_info->delayed_iputs_wait);
3486 spin_lock(&fs_info->delayed_iput_lock);
3489 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3490 struct btrfs_inode *inode)
3492 if (!list_empty(&inode->delayed_iput)) {
3493 spin_lock(&fs_info->delayed_iput_lock);
3494 if (!list_empty(&inode->delayed_iput))
3495 run_delayed_iput_locked(fs_info, inode);
3496 spin_unlock(&fs_info->delayed_iput_lock);
3500 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3503 spin_lock(&fs_info->delayed_iput_lock);
3504 while (!list_empty(&fs_info->delayed_iputs)) {
3505 struct btrfs_inode *inode;
3507 inode = list_first_entry(&fs_info->delayed_iputs,
3508 struct btrfs_inode, delayed_iput);
3509 run_delayed_iput_locked(fs_info, inode);
3510 cond_resched_lock(&fs_info->delayed_iput_lock);
3512 spin_unlock(&fs_info->delayed_iput_lock);
3516 * Wait for flushing all delayed iputs
3518 * @fs_info: the filesystem
3520 * This will wait on any delayed iputs that are currently running with KILLABLE
3521 * set. Once they are all done running we will return, unless we are killed in
3522 * which case we return EINTR. This helps in user operations like fallocate etc
3523 * that might get blocked on the iputs.
3525 * Return EINTR if we were killed, 0 if nothing's pending
3527 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3529 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3530 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3537 * This creates an orphan entry for the given inode in case something goes wrong
3538 * in the middle of an unlink.
3540 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3541 struct btrfs_inode *inode)
3545 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3546 if (ret && ret != -EEXIST) {
3547 btrfs_abort_transaction(trans, ret);
3555 * We have done the delete so we can go ahead and remove the orphan item for
3556 * this particular inode.
3558 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3559 struct btrfs_inode *inode)
3561 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3565 * this cleans up any orphans that may be left on the list from the last use
3568 int btrfs_orphan_cleanup(struct btrfs_root *root)
3570 struct btrfs_fs_info *fs_info = root->fs_info;
3571 struct btrfs_path *path;
3572 struct extent_buffer *leaf;
3573 struct btrfs_key key, found_key;
3574 struct btrfs_trans_handle *trans;
3575 struct inode *inode;
3576 u64 last_objectid = 0;
3577 int ret = 0, nr_unlink = 0;
3579 /* Bail out if the cleanup is already running. */
3580 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3583 path = btrfs_alloc_path();
3588 path->reada = READA_BACK;
3590 key.objectid = BTRFS_ORPHAN_OBJECTID;
3591 key.type = BTRFS_ORPHAN_ITEM_KEY;
3592 key.offset = (u64)-1;
3595 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3600 * if ret == 0 means we found what we were searching for, which
3601 * is weird, but possible, so only screw with path if we didn't
3602 * find the key and see if we have stuff that matches
3606 if (path->slots[0] == 0)
3611 /* pull out the item */
3612 leaf = path->nodes[0];
3613 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3615 /* make sure the item matches what we want */
3616 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3618 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3621 /* release the path since we're done with it */
3622 btrfs_release_path(path);
3625 * this is where we are basically btrfs_lookup, without the
3626 * crossing root thing. we store the inode number in the
3627 * offset of the orphan item.
3630 if (found_key.offset == last_objectid) {
3632 "Error removing orphan entry, stopping orphan cleanup");
3637 last_objectid = found_key.offset;
3639 found_key.objectid = found_key.offset;
3640 found_key.type = BTRFS_INODE_ITEM_KEY;
3641 found_key.offset = 0;
3642 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3643 ret = PTR_ERR_OR_ZERO(inode);
3644 if (ret && ret != -ENOENT)
3647 if (ret == -ENOENT && root == fs_info->tree_root) {
3648 struct btrfs_root *dead_root;
3649 int is_dead_root = 0;
3652 * This is an orphan in the tree root. Currently these
3653 * could come from 2 sources:
3654 * a) a root (snapshot/subvolume) deletion in progress
3655 * b) a free space cache inode
3656 * We need to distinguish those two, as the orphan item
3657 * for a root must not get deleted before the deletion
3658 * of the snapshot/subvolume's tree completes.
3660 * btrfs_find_orphan_roots() ran before us, which has
3661 * found all deleted roots and loaded them into
3662 * fs_info->fs_roots. So here we can find if an
3663 * orphan item corresponds to a deleted root by looking
3664 * up the root from that xarray.
3667 spin_lock(&fs_info->fs_roots_lock);
3668 dead_root = xa_load(&fs_info->fs_roots,
3669 (unsigned long)found_key.objectid);
3670 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3672 spin_unlock(&fs_info->fs_roots_lock);
3675 /* prevent this orphan from being found again */
3676 key.offset = found_key.objectid - 1;
3683 * If we have an inode with links, there are a couple of
3686 * 1. We were halfway through creating fsverity metadata for the
3687 * file. In that case, the orphan item represents incomplete
3688 * fsverity metadata which must be cleaned up with
3689 * btrfs_drop_verity_items and deleting the orphan item.
3691 * 2. Old kernels (before v3.12) used to create an
3692 * orphan item for truncate indicating that there were possibly
3693 * extent items past i_size that needed to be deleted. In v3.12,
3694 * truncate was changed to update i_size in sync with the extent
3695 * items, but the (useless) orphan item was still created. Since
3696 * v4.18, we don't create the orphan item for truncate at all.
3698 * So, this item could mean that we need to do a truncate, but
3699 * only if this filesystem was last used on a pre-v3.12 kernel
3700 * and was not cleanly unmounted. The odds of that are quite
3701 * slim, and it's a pain to do the truncate now, so just delete
3704 * It's also possible that this orphan item was supposed to be
3705 * deleted but wasn't. The inode number may have been reused,
3706 * but either way, we can delete the orphan item.
3708 if (ret == -ENOENT || inode->i_nlink) {
3710 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3715 trans = btrfs_start_transaction(root, 1);
3716 if (IS_ERR(trans)) {
3717 ret = PTR_ERR(trans);
3720 btrfs_debug(fs_info, "auto deleting %Lu",
3721 found_key.objectid);
3722 ret = btrfs_del_orphan_item(trans, root,
3723 found_key.objectid);
3724 btrfs_end_transaction(trans);
3732 /* this will do delete_inode and everything for us */
3735 /* release the path since we're done with it */
3736 btrfs_release_path(path);
3738 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3739 trans = btrfs_join_transaction(root);
3741 btrfs_end_transaction(trans);
3745 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3749 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3750 btrfs_free_path(path);
3755 * very simple check to peek ahead in the leaf looking for xattrs. If we
3756 * don't find any xattrs, we know there can't be any acls.
3758 * slot is the slot the inode is in, objectid is the objectid of the inode
3760 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3761 int slot, u64 objectid,
3762 int *first_xattr_slot)
3764 u32 nritems = btrfs_header_nritems(leaf);
3765 struct btrfs_key found_key;
3766 static u64 xattr_access = 0;
3767 static u64 xattr_default = 0;
3770 if (!xattr_access) {
3771 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3772 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3773 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3774 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3778 *first_xattr_slot = -1;
3779 while (slot < nritems) {
3780 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3782 /* we found a different objectid, there must not be acls */
3783 if (found_key.objectid != objectid)
3786 /* we found an xattr, assume we've got an acl */
3787 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3788 if (*first_xattr_slot == -1)
3789 *first_xattr_slot = slot;
3790 if (found_key.offset == xattr_access ||
3791 found_key.offset == xattr_default)
3796 * we found a key greater than an xattr key, there can't
3797 * be any acls later on
3799 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3806 * it goes inode, inode backrefs, xattrs, extents,
3807 * so if there are a ton of hard links to an inode there can
3808 * be a lot of backrefs. Don't waste time searching too hard,
3809 * this is just an optimization
3814 /* we hit the end of the leaf before we found an xattr or
3815 * something larger than an xattr. We have to assume the inode
3818 if (*first_xattr_slot == -1)
3819 *first_xattr_slot = slot;
3824 * read an inode from the btree into the in-memory inode
3826 static int btrfs_read_locked_inode(struct inode *inode,
3827 struct btrfs_path *in_path)
3829 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3830 struct btrfs_path *path = in_path;
3831 struct extent_buffer *leaf;
3832 struct btrfs_inode_item *inode_item;
3833 struct btrfs_root *root = BTRFS_I(inode)->root;
3834 struct btrfs_key location;
3839 bool filled = false;
3840 int first_xattr_slot;
3842 ret = btrfs_fill_inode(inode, &rdev);
3847 path = btrfs_alloc_path();
3852 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3854 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3856 if (path != in_path)
3857 btrfs_free_path(path);
3861 leaf = path->nodes[0];
3866 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3867 struct btrfs_inode_item);
3868 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3869 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3870 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3871 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3872 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3873 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3874 round_up(i_size_read(inode), fs_info->sectorsize));
3876 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3877 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3879 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3880 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3882 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3883 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3885 BTRFS_I(inode)->i_otime.tv_sec =
3886 btrfs_timespec_sec(leaf, &inode_item->otime);
3887 BTRFS_I(inode)->i_otime.tv_nsec =
3888 btrfs_timespec_nsec(leaf, &inode_item->otime);
3890 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3891 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3892 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3894 inode_set_iversion_queried(inode,
3895 btrfs_inode_sequence(leaf, inode_item));
3896 inode->i_generation = BTRFS_I(inode)->generation;
3898 rdev = btrfs_inode_rdev(leaf, inode_item);
3900 BTRFS_I(inode)->index_cnt = (u64)-1;
3901 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3902 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3906 * If we were modified in the current generation and evicted from memory
3907 * and then re-read we need to do a full sync since we don't have any
3908 * idea about which extents were modified before we were evicted from
3911 * This is required for both inode re-read from disk and delayed inode
3912 * in the delayed_nodes xarray.
3914 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3915 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3916 &BTRFS_I(inode)->runtime_flags);
3919 * We don't persist the id of the transaction where an unlink operation
3920 * against the inode was last made. So here we assume the inode might
3921 * have been evicted, and therefore the exact value of last_unlink_trans
3922 * lost, and set it to last_trans to avoid metadata inconsistencies
3923 * between the inode and its parent if the inode is fsync'ed and the log
3924 * replayed. For example, in the scenario:
3927 * ln mydir/foo mydir/bar
3930 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3931 * xfs_io -c fsync mydir/foo
3933 * mount fs, triggers fsync log replay
3935 * We must make sure that when we fsync our inode foo we also log its
3936 * parent inode, otherwise after log replay the parent still has the
3937 * dentry with the "bar" name but our inode foo has a link count of 1
3938 * and doesn't have an inode ref with the name "bar" anymore.
3940 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3941 * but it guarantees correctness at the expense of occasional full
3942 * transaction commits on fsync if our inode is a directory, or if our
3943 * inode is not a directory, logging its parent unnecessarily.
3945 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3948 * Same logic as for last_unlink_trans. We don't persist the generation
3949 * of the last transaction where this inode was used for a reflink
3950 * operation, so after eviction and reloading the inode we must be
3951 * pessimistic and assume the last transaction that modified the inode.
3953 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3956 if (inode->i_nlink != 1 ||
3957 path->slots[0] >= btrfs_header_nritems(leaf))
3960 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3961 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3964 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3965 if (location.type == BTRFS_INODE_REF_KEY) {
3966 struct btrfs_inode_ref *ref;
3968 ref = (struct btrfs_inode_ref *)ptr;
3969 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3970 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3971 struct btrfs_inode_extref *extref;
3973 extref = (struct btrfs_inode_extref *)ptr;
3974 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3979 * try to precache a NULL acl entry for files that don't have
3980 * any xattrs or acls
3982 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3983 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3984 if (first_xattr_slot != -1) {
3985 path->slots[0] = first_xattr_slot;
3986 ret = btrfs_load_inode_props(inode, path);
3989 "error loading props for ino %llu (root %llu): %d",
3990 btrfs_ino(BTRFS_I(inode)),
3991 root->root_key.objectid, ret);
3993 if (path != in_path)
3994 btrfs_free_path(path);
3997 cache_no_acl(inode);
3999 switch (inode->i_mode & S_IFMT) {
4001 inode->i_mapping->a_ops = &btrfs_aops;
4002 inode->i_fop = &btrfs_file_operations;
4003 inode->i_op = &btrfs_file_inode_operations;
4006 inode->i_fop = &btrfs_dir_file_operations;
4007 inode->i_op = &btrfs_dir_inode_operations;
4010 inode->i_op = &btrfs_symlink_inode_operations;
4011 inode_nohighmem(inode);
4012 inode->i_mapping->a_ops = &btrfs_aops;
4015 inode->i_op = &btrfs_special_inode_operations;
4016 init_special_inode(inode, inode->i_mode, rdev);
4020 btrfs_sync_inode_flags_to_i_flags(inode);
4025 * given a leaf and an inode, copy the inode fields into the leaf
4027 static void fill_inode_item(struct btrfs_trans_handle *trans,
4028 struct extent_buffer *leaf,
4029 struct btrfs_inode_item *item,
4030 struct inode *inode)
4032 struct btrfs_map_token token;
4035 btrfs_init_map_token(&token, leaf);
4037 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4038 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4039 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4040 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4041 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4043 btrfs_set_token_timespec_sec(&token, &item->atime,
4044 inode->i_atime.tv_sec);
4045 btrfs_set_token_timespec_nsec(&token, &item->atime,
4046 inode->i_atime.tv_nsec);
4048 btrfs_set_token_timespec_sec(&token, &item->mtime,
4049 inode->i_mtime.tv_sec);
4050 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4051 inode->i_mtime.tv_nsec);
4053 btrfs_set_token_timespec_sec(&token, &item->ctime,
4054 inode->i_ctime.tv_sec);
4055 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4056 inode->i_ctime.tv_nsec);
4058 btrfs_set_token_timespec_sec(&token, &item->otime,
4059 BTRFS_I(inode)->i_otime.tv_sec);
4060 btrfs_set_token_timespec_nsec(&token, &item->otime,
4061 BTRFS_I(inode)->i_otime.tv_nsec);
4063 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4064 btrfs_set_token_inode_generation(&token, item,
4065 BTRFS_I(inode)->generation);
4066 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4067 btrfs_set_token_inode_transid(&token, item, trans->transid);
4068 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4069 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4070 BTRFS_I(inode)->ro_flags);
4071 btrfs_set_token_inode_flags(&token, item, flags);
4072 btrfs_set_token_inode_block_group(&token, item, 0);
4076 * copy everything in the in-memory inode into the btree.
4078 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4079 struct btrfs_root *root,
4080 struct btrfs_inode *inode)
4082 struct btrfs_inode_item *inode_item;
4083 struct btrfs_path *path;
4084 struct extent_buffer *leaf;
4087 path = btrfs_alloc_path();
4091 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4098 leaf = path->nodes[0];
4099 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4100 struct btrfs_inode_item);
4102 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4103 btrfs_mark_buffer_dirty(leaf);
4104 btrfs_set_inode_last_trans(trans, inode);
4107 btrfs_free_path(path);
4112 * copy everything in the in-memory inode into the btree.
4114 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4115 struct btrfs_root *root,
4116 struct btrfs_inode *inode)
4118 struct btrfs_fs_info *fs_info = root->fs_info;
4122 * If the inode is a free space inode, we can deadlock during commit
4123 * if we put it into the delayed code.
4125 * The data relocation inode should also be directly updated
4128 if (!btrfs_is_free_space_inode(inode)
4129 && !btrfs_is_data_reloc_root(root)
4130 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4131 btrfs_update_root_times(trans, root);
4133 ret = btrfs_delayed_update_inode(trans, root, inode);
4135 btrfs_set_inode_last_trans(trans, inode);
4139 return btrfs_update_inode_item(trans, root, inode);
4142 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4143 struct btrfs_root *root, struct btrfs_inode *inode)
4147 ret = btrfs_update_inode(trans, root, inode);
4149 return btrfs_update_inode_item(trans, root, inode);
4154 * unlink helper that gets used here in inode.c and in the tree logging
4155 * recovery code. It remove a link in a directory with a given name, and
4156 * also drops the back refs in the inode to the directory
4158 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4159 struct btrfs_inode *dir,
4160 struct btrfs_inode *inode,
4161 const char *name, int name_len,
4162 struct btrfs_rename_ctx *rename_ctx)
4164 struct btrfs_root *root = dir->root;
4165 struct btrfs_fs_info *fs_info = root->fs_info;
4166 struct btrfs_path *path;
4168 struct btrfs_dir_item *di;
4170 u64 ino = btrfs_ino(inode);
4171 u64 dir_ino = btrfs_ino(dir);
4173 path = btrfs_alloc_path();
4179 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4180 name, name_len, -1);
4181 if (IS_ERR_OR_NULL(di)) {
4182 ret = di ? PTR_ERR(di) : -ENOENT;
4185 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4188 btrfs_release_path(path);
4191 * If we don't have dir index, we have to get it by looking up
4192 * the inode ref, since we get the inode ref, remove it directly,
4193 * it is unnecessary to do delayed deletion.
4195 * But if we have dir index, needn't search inode ref to get it.
4196 * Since the inode ref is close to the inode item, it is better
4197 * that we delay to delete it, and just do this deletion when
4198 * we update the inode item.
4200 if (inode->dir_index) {
4201 ret = btrfs_delayed_delete_inode_ref(inode);
4203 index = inode->dir_index;
4208 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4212 "failed to delete reference to %.*s, inode %llu parent %llu",
4213 name_len, name, ino, dir_ino);
4214 btrfs_abort_transaction(trans, ret);
4219 rename_ctx->index = index;
4221 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4223 btrfs_abort_transaction(trans, ret);
4228 * If we are in a rename context, we don't need to update anything in the
4229 * log. That will be done later during the rename by btrfs_log_new_name().
4230 * Besides that, doing it here would only cause extra unncessary btree
4231 * operations on the log tree, increasing latency for applications.
4234 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4236 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4241 * If we have a pending delayed iput we could end up with the final iput
4242 * being run in btrfs-cleaner context. If we have enough of these built
4243 * up we can end up burning a lot of time in btrfs-cleaner without any
4244 * way to throttle the unlinks. Since we're currently holding a ref on
4245 * the inode we can run the delayed iput here without any issues as the
4246 * final iput won't be done until after we drop the ref we're currently
4249 btrfs_run_delayed_iput(fs_info, inode);
4251 btrfs_free_path(path);
4255 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4256 inode_inc_iversion(&inode->vfs_inode);
4257 inode_inc_iversion(&dir->vfs_inode);
4258 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4259 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4260 ret = btrfs_update_inode(trans, root, dir);
4265 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4266 struct btrfs_inode *dir, struct btrfs_inode *inode,
4267 const char *name, int name_len)
4270 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4272 drop_nlink(&inode->vfs_inode);
4273 ret = btrfs_update_inode(trans, inode->root, inode);
4279 * helper to start transaction for unlink and rmdir.
4281 * unlink and rmdir are special in btrfs, they do not always free space, so
4282 * if we cannot make our reservations the normal way try and see if there is
4283 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4284 * allow the unlink to occur.
4286 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4288 struct btrfs_root *root = BTRFS_I(dir)->root;
4291 * 1 for the possible orphan item
4292 * 1 for the dir item
4293 * 1 for the dir index
4294 * 1 for the inode ref
4296 * 1 for the parent inode
4298 return btrfs_start_transaction_fallback_global_rsv(root, 6);
4301 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4303 struct btrfs_trans_handle *trans;
4304 struct inode *inode = d_inode(dentry);
4307 trans = __unlink_start_trans(dir);
4309 return PTR_ERR(trans);
4311 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4314 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4315 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4316 dentry->d_name.len);
4320 if (inode->i_nlink == 0) {
4321 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4327 btrfs_end_transaction(trans);
4328 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4332 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4333 struct inode *dir, struct dentry *dentry)
4335 struct btrfs_root *root = BTRFS_I(dir)->root;
4336 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4337 struct btrfs_path *path;
4338 struct extent_buffer *leaf;
4339 struct btrfs_dir_item *di;
4340 struct btrfs_key key;
4341 const char *name = dentry->d_name.name;
4342 int name_len = dentry->d_name.len;
4346 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4348 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4349 objectid = inode->root->root_key.objectid;
4350 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4351 objectid = inode->location.objectid;
4357 path = btrfs_alloc_path();
4361 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4362 name, name_len, -1);
4363 if (IS_ERR_OR_NULL(di)) {
4364 ret = di ? PTR_ERR(di) : -ENOENT;
4368 leaf = path->nodes[0];
4369 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4370 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4371 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4373 btrfs_abort_transaction(trans, ret);
4376 btrfs_release_path(path);
4379 * This is a placeholder inode for a subvolume we didn't have a
4380 * reference to at the time of the snapshot creation. In the meantime
4381 * we could have renamed the real subvol link into our snapshot, so
4382 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4383 * Instead simply lookup the dir_index_item for this entry so we can
4384 * remove it. Otherwise we know we have a ref to the root and we can
4385 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4387 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4388 di = btrfs_search_dir_index_item(root, path, dir_ino,
4390 if (IS_ERR_OR_NULL(di)) {
4395 btrfs_abort_transaction(trans, ret);
4399 leaf = path->nodes[0];
4400 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4402 btrfs_release_path(path);
4404 ret = btrfs_del_root_ref(trans, objectid,
4405 root->root_key.objectid, dir_ino,
4406 &index, name, name_len);
4408 btrfs_abort_transaction(trans, ret);
4413 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4415 btrfs_abort_transaction(trans, ret);
4419 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4420 inode_inc_iversion(dir);
4421 dir->i_mtime = dir->i_ctime = current_time(dir);
4422 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4424 btrfs_abort_transaction(trans, ret);
4426 btrfs_free_path(path);
4431 * Helper to check if the subvolume references other subvolumes or if it's
4434 static noinline int may_destroy_subvol(struct btrfs_root *root)
4436 struct btrfs_fs_info *fs_info = root->fs_info;
4437 struct btrfs_path *path;
4438 struct btrfs_dir_item *di;
4439 struct btrfs_key key;
4443 path = btrfs_alloc_path();
4447 /* Make sure this root isn't set as the default subvol */
4448 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4449 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4450 dir_id, "default", 7, 0);
4451 if (di && !IS_ERR(di)) {
4452 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4453 if (key.objectid == root->root_key.objectid) {
4456 "deleting default subvolume %llu is not allowed",
4460 btrfs_release_path(path);
4463 key.objectid = root->root_key.objectid;
4464 key.type = BTRFS_ROOT_REF_KEY;
4465 key.offset = (u64)-1;
4467 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4473 if (path->slots[0] > 0) {
4475 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4476 if (key.objectid == root->root_key.objectid &&
4477 key.type == BTRFS_ROOT_REF_KEY)
4481 btrfs_free_path(path);
4485 /* Delete all dentries for inodes belonging to the root */
4486 static void btrfs_prune_dentries(struct btrfs_root *root)
4488 struct btrfs_fs_info *fs_info = root->fs_info;
4489 struct rb_node *node;
4490 struct rb_node *prev;
4491 struct btrfs_inode *entry;
4492 struct inode *inode;
4495 if (!BTRFS_FS_ERROR(fs_info))
4496 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4498 spin_lock(&root->inode_lock);
4500 node = root->inode_tree.rb_node;
4504 entry = rb_entry(node, struct btrfs_inode, rb_node);
4506 if (objectid < btrfs_ino(entry))
4507 node = node->rb_left;
4508 else if (objectid > btrfs_ino(entry))
4509 node = node->rb_right;
4515 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4516 if (objectid <= btrfs_ino(entry)) {
4520 prev = rb_next(prev);
4524 entry = rb_entry(node, struct btrfs_inode, rb_node);
4525 objectid = btrfs_ino(entry) + 1;
4526 inode = igrab(&entry->vfs_inode);
4528 spin_unlock(&root->inode_lock);
4529 if (atomic_read(&inode->i_count) > 1)
4530 d_prune_aliases(inode);
4532 * btrfs_drop_inode will have it removed from the inode
4533 * cache when its usage count hits zero.
4537 spin_lock(&root->inode_lock);
4541 if (cond_resched_lock(&root->inode_lock))
4544 node = rb_next(node);
4546 spin_unlock(&root->inode_lock);
4549 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4551 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4552 struct btrfs_root *root = BTRFS_I(dir)->root;
4553 struct inode *inode = d_inode(dentry);
4554 struct btrfs_root *dest = BTRFS_I(inode)->root;
4555 struct btrfs_trans_handle *trans;
4556 struct btrfs_block_rsv block_rsv;
4561 * Don't allow to delete a subvolume with send in progress. This is
4562 * inside the inode lock so the error handling that has to drop the bit
4563 * again is not run concurrently.
4565 spin_lock(&dest->root_item_lock);
4566 if (dest->send_in_progress) {
4567 spin_unlock(&dest->root_item_lock);
4569 "attempt to delete subvolume %llu during send",
4570 dest->root_key.objectid);
4573 if (atomic_read(&dest->nr_swapfiles)) {
4574 spin_unlock(&dest->root_item_lock);
4576 "attempt to delete subvolume %llu with active swapfile",
4577 root->root_key.objectid);
4580 root_flags = btrfs_root_flags(&dest->root_item);
4581 btrfs_set_root_flags(&dest->root_item,
4582 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4583 spin_unlock(&dest->root_item_lock);
4585 down_write(&fs_info->subvol_sem);
4587 ret = may_destroy_subvol(dest);
4591 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4593 * One for dir inode,
4594 * two for dir entries,
4595 * two for root ref/backref.
4597 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4601 trans = btrfs_start_transaction(root, 0);
4602 if (IS_ERR(trans)) {
4603 ret = PTR_ERR(trans);
4606 trans->block_rsv = &block_rsv;
4607 trans->bytes_reserved = block_rsv.size;
4609 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4611 ret = btrfs_unlink_subvol(trans, dir, dentry);
4613 btrfs_abort_transaction(trans, ret);
4617 ret = btrfs_record_root_in_trans(trans, dest);
4619 btrfs_abort_transaction(trans, ret);
4623 memset(&dest->root_item.drop_progress, 0,
4624 sizeof(dest->root_item.drop_progress));
4625 btrfs_set_root_drop_level(&dest->root_item, 0);
4626 btrfs_set_root_refs(&dest->root_item, 0);
4628 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4629 ret = btrfs_insert_orphan_item(trans,
4631 dest->root_key.objectid);
4633 btrfs_abort_transaction(trans, ret);
4638 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4639 BTRFS_UUID_KEY_SUBVOL,
4640 dest->root_key.objectid);
4641 if (ret && ret != -ENOENT) {
4642 btrfs_abort_transaction(trans, ret);
4645 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4646 ret = btrfs_uuid_tree_remove(trans,
4647 dest->root_item.received_uuid,
4648 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4649 dest->root_key.objectid);
4650 if (ret && ret != -ENOENT) {
4651 btrfs_abort_transaction(trans, ret);
4656 free_anon_bdev(dest->anon_dev);
4659 trans->block_rsv = NULL;
4660 trans->bytes_reserved = 0;
4661 ret = btrfs_end_transaction(trans);
4662 inode->i_flags |= S_DEAD;
4664 btrfs_subvolume_release_metadata(root, &block_rsv);
4666 up_write(&fs_info->subvol_sem);
4668 spin_lock(&dest->root_item_lock);
4669 root_flags = btrfs_root_flags(&dest->root_item);
4670 btrfs_set_root_flags(&dest->root_item,
4671 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4672 spin_unlock(&dest->root_item_lock);
4674 d_invalidate(dentry);
4675 btrfs_prune_dentries(dest);
4676 ASSERT(dest->send_in_progress == 0);
4682 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4684 struct inode *inode = d_inode(dentry);
4685 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4687 struct btrfs_trans_handle *trans;
4688 u64 last_unlink_trans;
4690 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4692 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4693 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4695 "extent tree v2 doesn't support snapshot deletion yet");
4698 return btrfs_delete_subvolume(dir, dentry);
4701 trans = __unlink_start_trans(dir);
4703 return PTR_ERR(trans);
4705 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4706 err = btrfs_unlink_subvol(trans, dir, dentry);
4710 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4714 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4716 /* now the directory is empty */
4717 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4718 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4719 dentry->d_name.len);
4721 btrfs_i_size_write(BTRFS_I(inode), 0);
4723 * Propagate the last_unlink_trans value of the deleted dir to
4724 * its parent directory. This is to prevent an unrecoverable
4725 * log tree in the case we do something like this:
4727 * 2) create snapshot under dir foo
4728 * 3) delete the snapshot
4731 * 6) fsync foo or some file inside foo
4733 if (last_unlink_trans >= trans->transid)
4734 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4737 btrfs_end_transaction(trans);
4738 btrfs_btree_balance_dirty(fs_info);
4744 * btrfs_truncate_block - read, zero a chunk and write a block
4745 * @inode - inode that we're zeroing
4746 * @from - the offset to start zeroing
4747 * @len - the length to zero, 0 to zero the entire range respective to the
4749 * @front - zero up to the offset instead of from the offset on
4751 * This will find the block for the "from" offset and cow the block and zero the
4752 * part we want to zero. This is used with truncate and hole punching.
4754 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4757 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4758 struct address_space *mapping = inode->vfs_inode.i_mapping;
4759 struct extent_io_tree *io_tree = &inode->io_tree;
4760 struct btrfs_ordered_extent *ordered;
4761 struct extent_state *cached_state = NULL;
4762 struct extent_changeset *data_reserved = NULL;
4763 bool only_release_metadata = false;
4764 u32 blocksize = fs_info->sectorsize;
4765 pgoff_t index = from >> PAGE_SHIFT;
4766 unsigned offset = from & (blocksize - 1);
4768 gfp_t mask = btrfs_alloc_write_mask(mapping);
4769 size_t write_bytes = blocksize;
4774 if (IS_ALIGNED(offset, blocksize) &&
4775 (!len || IS_ALIGNED(len, blocksize)))
4778 block_start = round_down(from, blocksize);
4779 block_end = block_start + blocksize - 1;
4781 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4784 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4785 /* For nocow case, no need to reserve data space */
4786 only_release_metadata = true;
4791 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4793 if (!only_release_metadata)
4794 btrfs_free_reserved_data_space(inode, data_reserved,
4795 block_start, blocksize);
4799 page = find_or_create_page(mapping, index, mask);
4801 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4803 btrfs_delalloc_release_extents(inode, blocksize);
4807 ret = set_page_extent_mapped(page);
4811 if (!PageUptodate(page)) {
4812 ret = btrfs_read_folio(NULL, page_folio(page));
4814 if (page->mapping != mapping) {
4819 if (!PageUptodate(page)) {
4824 wait_on_page_writeback(page);
4826 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4828 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4830 unlock_extent_cached(io_tree, block_start, block_end,
4834 btrfs_start_ordered_extent(ordered, 1);
4835 btrfs_put_ordered_extent(ordered);
4839 clear_extent_bit(&inode->io_tree, block_start, block_end,
4840 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4841 0, 0, &cached_state);
4843 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4846 unlock_extent_cached(io_tree, block_start, block_end,
4851 if (offset != blocksize) {
4853 len = blocksize - offset;
4855 memzero_page(page, (block_start - page_offset(page)),
4858 memzero_page(page, (block_start - page_offset(page)) + offset,
4860 flush_dcache_page(page);
4862 btrfs_page_clear_checked(fs_info, page, block_start,
4863 block_end + 1 - block_start);
4864 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4865 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4867 if (only_release_metadata)
4868 set_extent_bit(&inode->io_tree, block_start, block_end,
4869 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4873 if (only_release_metadata)
4874 btrfs_delalloc_release_metadata(inode, blocksize, true);
4876 btrfs_delalloc_release_space(inode, data_reserved,
4877 block_start, blocksize, true);
4879 btrfs_delalloc_release_extents(inode, blocksize);
4883 if (only_release_metadata)
4884 btrfs_check_nocow_unlock(inode);
4885 extent_changeset_free(data_reserved);
4889 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4890 u64 offset, u64 len)
4892 struct btrfs_fs_info *fs_info = root->fs_info;
4893 struct btrfs_trans_handle *trans;
4894 struct btrfs_drop_extents_args drop_args = { 0 };
4898 * If NO_HOLES is enabled, we don't need to do anything.
4899 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4900 * or btrfs_update_inode() will be called, which guarantee that the next
4901 * fsync will know this inode was changed and needs to be logged.
4903 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4907 * 1 - for the one we're dropping
4908 * 1 - for the one we're adding
4909 * 1 - for updating the inode.
4911 trans = btrfs_start_transaction(root, 3);
4913 return PTR_ERR(trans);
4915 drop_args.start = offset;
4916 drop_args.end = offset + len;
4917 drop_args.drop_cache = true;
4919 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4921 btrfs_abort_transaction(trans, ret);
4922 btrfs_end_transaction(trans);
4926 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4927 offset, 0, 0, len, 0, len, 0, 0, 0);
4929 btrfs_abort_transaction(trans, ret);
4931 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4932 btrfs_update_inode(trans, root, inode);
4934 btrfs_end_transaction(trans);
4939 * This function puts in dummy file extents for the area we're creating a hole
4940 * for. So if we are truncating this file to a larger size we need to insert
4941 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4942 * the range between oldsize and size
4944 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4946 struct btrfs_root *root = inode->root;
4947 struct btrfs_fs_info *fs_info = root->fs_info;
4948 struct extent_io_tree *io_tree = &inode->io_tree;
4949 struct extent_map *em = NULL;
4950 struct extent_state *cached_state = NULL;
4951 struct extent_map_tree *em_tree = &inode->extent_tree;
4952 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4953 u64 block_end = ALIGN(size, fs_info->sectorsize);
4960 * If our size started in the middle of a block we need to zero out the
4961 * rest of the block before we expand the i_size, otherwise we could
4962 * expose stale data.
4964 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4968 if (size <= hole_start)
4971 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4973 cur_offset = hole_start;
4975 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4976 block_end - cur_offset);
4982 last_byte = min(extent_map_end(em), block_end);
4983 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4984 hole_size = last_byte - cur_offset;
4986 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4987 struct extent_map *hole_em;
4989 err = maybe_insert_hole(root, inode, cur_offset,
4994 err = btrfs_inode_set_file_extent_range(inode,
4995 cur_offset, hole_size);
4999 btrfs_drop_extent_cache(inode, cur_offset,
5000 cur_offset + hole_size - 1, 0);
5001 hole_em = alloc_extent_map();
5003 btrfs_set_inode_full_sync(inode);
5006 hole_em->start = cur_offset;
5007 hole_em->len = hole_size;
5008 hole_em->orig_start = cur_offset;
5010 hole_em->block_start = EXTENT_MAP_HOLE;
5011 hole_em->block_len = 0;
5012 hole_em->orig_block_len = 0;
5013 hole_em->ram_bytes = hole_size;
5014 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5015 hole_em->generation = fs_info->generation;
5018 write_lock(&em_tree->lock);
5019 err = add_extent_mapping(em_tree, hole_em, 1);
5020 write_unlock(&em_tree->lock);
5023 btrfs_drop_extent_cache(inode, cur_offset,
5027 free_extent_map(hole_em);
5029 err = btrfs_inode_set_file_extent_range(inode,
5030 cur_offset, hole_size);
5035 free_extent_map(em);
5037 cur_offset = last_byte;
5038 if (cur_offset >= block_end)
5041 free_extent_map(em);
5042 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5046 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5048 struct btrfs_root *root = BTRFS_I(inode)->root;
5049 struct btrfs_trans_handle *trans;
5050 loff_t oldsize = i_size_read(inode);
5051 loff_t newsize = attr->ia_size;
5052 int mask = attr->ia_valid;
5056 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5057 * special case where we need to update the times despite not having
5058 * these flags set. For all other operations the VFS set these flags
5059 * explicitly if it wants a timestamp update.
5061 if (newsize != oldsize) {
5062 inode_inc_iversion(inode);
5063 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5064 inode->i_ctime = inode->i_mtime =
5065 current_time(inode);
5068 if (newsize > oldsize) {
5070 * Don't do an expanding truncate while snapshotting is ongoing.
5071 * This is to ensure the snapshot captures a fully consistent
5072 * state of this file - if the snapshot captures this expanding
5073 * truncation, it must capture all writes that happened before
5076 btrfs_drew_write_lock(&root->snapshot_lock);
5077 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5079 btrfs_drew_write_unlock(&root->snapshot_lock);
5083 trans = btrfs_start_transaction(root, 1);
5084 if (IS_ERR(trans)) {
5085 btrfs_drew_write_unlock(&root->snapshot_lock);
5086 return PTR_ERR(trans);
5089 i_size_write(inode, newsize);
5090 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5091 pagecache_isize_extended(inode, oldsize, newsize);
5092 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5093 btrfs_drew_write_unlock(&root->snapshot_lock);
5094 btrfs_end_transaction(trans);
5096 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5098 if (btrfs_is_zoned(fs_info)) {
5099 ret = btrfs_wait_ordered_range(inode,
5100 ALIGN(newsize, fs_info->sectorsize),
5107 * We're truncating a file that used to have good data down to
5108 * zero. Make sure any new writes to the file get on disk
5112 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5113 &BTRFS_I(inode)->runtime_flags);
5115 truncate_setsize(inode, newsize);
5117 inode_dio_wait(inode);
5119 ret = btrfs_truncate(inode, newsize == oldsize);
5120 if (ret && inode->i_nlink) {
5124 * Truncate failed, so fix up the in-memory size. We
5125 * adjusted disk_i_size down as we removed extents, so
5126 * wait for disk_i_size to be stable and then update the
5127 * in-memory size to match.
5129 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5132 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5139 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5142 struct inode *inode = d_inode(dentry);
5143 struct btrfs_root *root = BTRFS_I(inode)->root;
5146 if (btrfs_root_readonly(root))
5149 err = setattr_prepare(mnt_userns, dentry, attr);
5153 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5154 err = btrfs_setsize(inode, attr);
5159 if (attr->ia_valid) {
5160 setattr_copy(mnt_userns, inode, attr);
5161 inode_inc_iversion(inode);
5162 err = btrfs_dirty_inode(inode);
5164 if (!err && attr->ia_valid & ATTR_MODE)
5165 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5172 * While truncating the inode pages during eviction, we get the VFS
5173 * calling btrfs_invalidate_folio() against each folio of the inode. This
5174 * is slow because the calls to btrfs_invalidate_folio() result in a
5175 * huge amount of calls to lock_extent_bits() and clear_extent_bit(),
5176 * which keep merging and splitting extent_state structures over and over,
5177 * wasting lots of time.
5179 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5180 * skip all those expensive operations on a per folio basis and do only
5181 * the ordered io finishing, while we release here the extent_map and
5182 * extent_state structures, without the excessive merging and splitting.
5184 static void evict_inode_truncate_pages(struct inode *inode)
5186 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5187 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5188 struct rb_node *node;
5190 ASSERT(inode->i_state & I_FREEING);
5191 truncate_inode_pages_final(&inode->i_data);
5193 write_lock(&map_tree->lock);
5194 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5195 struct extent_map *em;
5197 node = rb_first_cached(&map_tree->map);
5198 em = rb_entry(node, struct extent_map, rb_node);
5199 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5200 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5201 remove_extent_mapping(map_tree, em);
5202 free_extent_map(em);
5203 if (need_resched()) {
5204 write_unlock(&map_tree->lock);
5206 write_lock(&map_tree->lock);
5209 write_unlock(&map_tree->lock);
5212 * Keep looping until we have no more ranges in the io tree.
5213 * We can have ongoing bios started by readahead that have
5214 * their endio callback (extent_io.c:end_bio_extent_readpage)
5215 * still in progress (unlocked the pages in the bio but did not yet
5216 * unlocked the ranges in the io tree). Therefore this means some
5217 * ranges can still be locked and eviction started because before
5218 * submitting those bios, which are executed by a separate task (work
5219 * queue kthread), inode references (inode->i_count) were not taken
5220 * (which would be dropped in the end io callback of each bio).
5221 * Therefore here we effectively end up waiting for those bios and
5222 * anyone else holding locked ranges without having bumped the inode's
5223 * reference count - if we don't do it, when they access the inode's
5224 * io_tree to unlock a range it may be too late, leading to an
5225 * use-after-free issue.
5227 spin_lock(&io_tree->lock);
5228 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5229 struct extent_state *state;
5230 struct extent_state *cached_state = NULL;
5233 unsigned state_flags;
5235 node = rb_first(&io_tree->state);
5236 state = rb_entry(node, struct extent_state, rb_node);
5237 start = state->start;
5239 state_flags = state->state;
5240 spin_unlock(&io_tree->lock);
5242 lock_extent_bits(io_tree, start, end, &cached_state);
5245 * If still has DELALLOC flag, the extent didn't reach disk,
5246 * and its reserved space won't be freed by delayed_ref.
5247 * So we need to free its reserved space here.
5248 * (Refer to comment in btrfs_invalidate_folio, case 2)
5250 * Note, end is the bytenr of last byte, so we need + 1 here.
5252 if (state_flags & EXTENT_DELALLOC)
5253 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5256 clear_extent_bit(io_tree, start, end,
5257 EXTENT_LOCKED | EXTENT_DELALLOC |
5258 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5262 spin_lock(&io_tree->lock);
5264 spin_unlock(&io_tree->lock);
5267 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5268 struct btrfs_block_rsv *rsv)
5270 struct btrfs_fs_info *fs_info = root->fs_info;
5271 struct btrfs_trans_handle *trans;
5272 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5276 * Eviction should be taking place at some place safe because of our
5277 * delayed iputs. However the normal flushing code will run delayed
5278 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5280 * We reserve the delayed_refs_extra here again because we can't use
5281 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5282 * above. We reserve our extra bit here because we generate a ton of
5283 * delayed refs activity by truncating.
5285 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5286 * if we fail to make this reservation we can re-try without the
5287 * delayed_refs_extra so we can make some forward progress.
5289 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5290 BTRFS_RESERVE_FLUSH_EVICT);
5292 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5293 BTRFS_RESERVE_FLUSH_EVICT);
5296 "could not allocate space for delete; will truncate on mount");
5297 return ERR_PTR(-ENOSPC);
5299 delayed_refs_extra = 0;
5302 trans = btrfs_join_transaction(root);
5306 if (delayed_refs_extra) {
5307 trans->block_rsv = &fs_info->trans_block_rsv;
5308 trans->bytes_reserved = delayed_refs_extra;
5309 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5310 delayed_refs_extra, 1);
5315 void btrfs_evict_inode(struct inode *inode)
5317 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5318 struct btrfs_trans_handle *trans;
5319 struct btrfs_root *root = BTRFS_I(inode)->root;
5320 struct btrfs_block_rsv *rsv;
5323 trace_btrfs_inode_evict(inode);
5326 fsverity_cleanup_inode(inode);
5331 evict_inode_truncate_pages(inode);
5333 if (inode->i_nlink &&
5334 ((btrfs_root_refs(&root->root_item) != 0 &&
5335 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5336 btrfs_is_free_space_inode(BTRFS_I(inode))))
5339 if (is_bad_inode(inode))
5342 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5344 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5347 if (inode->i_nlink > 0) {
5348 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5349 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5354 * This makes sure the inode item in tree is uptodate and the space for
5355 * the inode update is released.
5357 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5362 * This drops any pending insert or delete operations we have for this
5363 * inode. We could have a delayed dir index deletion queued up, but
5364 * we're removing the inode completely so that'll be taken care of in
5367 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5369 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5372 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5375 btrfs_i_size_write(BTRFS_I(inode), 0);
5378 struct btrfs_truncate_control control = {
5379 .inode = BTRFS_I(inode),
5380 .ino = btrfs_ino(BTRFS_I(inode)),
5385 trans = evict_refill_and_join(root, rsv);
5389 trans->block_rsv = rsv;
5391 ret = btrfs_truncate_inode_items(trans, root, &control);
5392 trans->block_rsv = &fs_info->trans_block_rsv;
5393 btrfs_end_transaction(trans);
5394 btrfs_btree_balance_dirty(fs_info);
5395 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5402 * Errors here aren't a big deal, it just means we leave orphan items in
5403 * the tree. They will be cleaned up on the next mount. If the inode
5404 * number gets reused, cleanup deletes the orphan item without doing
5405 * anything, and unlink reuses the existing orphan item.
5407 * If it turns out that we are dropping too many of these, we might want
5408 * to add a mechanism for retrying these after a commit.
5410 trans = evict_refill_and_join(root, rsv);
5411 if (!IS_ERR(trans)) {
5412 trans->block_rsv = rsv;
5413 btrfs_orphan_del(trans, BTRFS_I(inode));
5414 trans->block_rsv = &fs_info->trans_block_rsv;
5415 btrfs_end_transaction(trans);
5419 btrfs_free_block_rsv(fs_info, rsv);
5422 * If we didn't successfully delete, the orphan item will still be in
5423 * the tree and we'll retry on the next mount. Again, we might also want
5424 * to retry these periodically in the future.
5426 btrfs_remove_delayed_node(BTRFS_I(inode));
5427 fsverity_cleanup_inode(inode);
5432 * Return the key found in the dir entry in the location pointer, fill @type
5433 * with BTRFS_FT_*, and return 0.
5435 * If no dir entries were found, returns -ENOENT.
5436 * If found a corrupted location in dir entry, returns -EUCLEAN.
5438 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5439 struct btrfs_key *location, u8 *type)
5441 const char *name = dentry->d_name.name;
5442 int namelen = dentry->d_name.len;
5443 struct btrfs_dir_item *di;
5444 struct btrfs_path *path;
5445 struct btrfs_root *root = BTRFS_I(dir)->root;
5448 path = btrfs_alloc_path();
5452 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5454 if (IS_ERR_OR_NULL(di)) {
5455 ret = di ? PTR_ERR(di) : -ENOENT;
5459 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5460 if (location->type != BTRFS_INODE_ITEM_KEY &&
5461 location->type != BTRFS_ROOT_ITEM_KEY) {
5463 btrfs_warn(root->fs_info,
5464 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5465 __func__, name, btrfs_ino(BTRFS_I(dir)),
5466 location->objectid, location->type, location->offset);
5469 *type = btrfs_dir_type(path->nodes[0], di);
5471 btrfs_free_path(path);
5476 * when we hit a tree root in a directory, the btrfs part of the inode
5477 * needs to be changed to reflect the root directory of the tree root. This
5478 * is kind of like crossing a mount point.
5480 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5482 struct dentry *dentry,
5483 struct btrfs_key *location,
5484 struct btrfs_root **sub_root)
5486 struct btrfs_path *path;
5487 struct btrfs_root *new_root;
5488 struct btrfs_root_ref *ref;
5489 struct extent_buffer *leaf;
5490 struct btrfs_key key;
5494 path = btrfs_alloc_path();
5501 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5502 key.type = BTRFS_ROOT_REF_KEY;
5503 key.offset = location->objectid;
5505 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5512 leaf = path->nodes[0];
5513 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5514 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5515 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5518 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5519 (unsigned long)(ref + 1),
5520 dentry->d_name.len);
5524 btrfs_release_path(path);
5526 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5527 if (IS_ERR(new_root)) {
5528 err = PTR_ERR(new_root);
5532 *sub_root = new_root;
5533 location->objectid = btrfs_root_dirid(&new_root->root_item);
5534 location->type = BTRFS_INODE_ITEM_KEY;
5535 location->offset = 0;
5538 btrfs_free_path(path);
5542 static void inode_tree_add(struct inode *inode)
5544 struct btrfs_root *root = BTRFS_I(inode)->root;
5545 struct btrfs_inode *entry;
5547 struct rb_node *parent;
5548 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5549 u64 ino = btrfs_ino(BTRFS_I(inode));
5551 if (inode_unhashed(inode))
5554 spin_lock(&root->inode_lock);
5555 p = &root->inode_tree.rb_node;
5558 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5560 if (ino < btrfs_ino(entry))
5561 p = &parent->rb_left;
5562 else if (ino > btrfs_ino(entry))
5563 p = &parent->rb_right;
5565 WARN_ON(!(entry->vfs_inode.i_state &
5566 (I_WILL_FREE | I_FREEING)));
5567 rb_replace_node(parent, new, &root->inode_tree);
5568 RB_CLEAR_NODE(parent);
5569 spin_unlock(&root->inode_lock);
5573 rb_link_node(new, parent, p);
5574 rb_insert_color(new, &root->inode_tree);
5575 spin_unlock(&root->inode_lock);
5578 static void inode_tree_del(struct btrfs_inode *inode)
5580 struct btrfs_root *root = inode->root;
5583 spin_lock(&root->inode_lock);
5584 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5585 rb_erase(&inode->rb_node, &root->inode_tree);
5586 RB_CLEAR_NODE(&inode->rb_node);
5587 empty = RB_EMPTY_ROOT(&root->inode_tree);
5589 spin_unlock(&root->inode_lock);
5591 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5592 spin_lock(&root->inode_lock);
5593 empty = RB_EMPTY_ROOT(&root->inode_tree);
5594 spin_unlock(&root->inode_lock);
5596 btrfs_add_dead_root(root);
5601 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5603 struct btrfs_iget_args *args = p;
5605 inode->i_ino = args->ino;
5606 BTRFS_I(inode)->location.objectid = args->ino;
5607 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5608 BTRFS_I(inode)->location.offset = 0;
5609 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5610 BUG_ON(args->root && !BTRFS_I(inode)->root);
5614 static int btrfs_find_actor(struct inode *inode, void *opaque)
5616 struct btrfs_iget_args *args = opaque;
5618 return args->ino == BTRFS_I(inode)->location.objectid &&
5619 args->root == BTRFS_I(inode)->root;
5622 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5623 struct btrfs_root *root)
5625 struct inode *inode;
5626 struct btrfs_iget_args args;
5627 unsigned long hashval = btrfs_inode_hash(ino, root);
5632 inode = iget5_locked(s, hashval, btrfs_find_actor,
5633 btrfs_init_locked_inode,
5639 * Get an inode object given its inode number and corresponding root.
5640 * Path can be preallocated to prevent recursing back to iget through
5641 * allocator. NULL is also valid but may require an additional allocation
5644 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5645 struct btrfs_root *root, struct btrfs_path *path)
5647 struct inode *inode;
5649 inode = btrfs_iget_locked(s, ino, root);
5651 return ERR_PTR(-ENOMEM);
5653 if (inode->i_state & I_NEW) {
5656 ret = btrfs_read_locked_inode(inode, path);
5658 inode_tree_add(inode);
5659 unlock_new_inode(inode);
5663 * ret > 0 can come from btrfs_search_slot called by
5664 * btrfs_read_locked_inode, this means the inode item
5669 inode = ERR_PTR(ret);
5676 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5678 return btrfs_iget_path(s, ino, root, NULL);
5681 static struct inode *new_simple_dir(struct super_block *s,
5682 struct btrfs_key *key,
5683 struct btrfs_root *root)
5685 struct inode *inode = new_inode(s);
5688 return ERR_PTR(-ENOMEM);
5690 BTRFS_I(inode)->root = btrfs_grab_root(root);
5691 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5692 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5694 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5696 * We only need lookup, the rest is read-only and there's no inode
5697 * associated with the dentry
5699 inode->i_op = &simple_dir_inode_operations;
5700 inode->i_opflags &= ~IOP_XATTR;
5701 inode->i_fop = &simple_dir_operations;
5702 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5703 inode->i_mtime = current_time(inode);
5704 inode->i_atime = inode->i_mtime;
5705 inode->i_ctime = inode->i_mtime;
5706 BTRFS_I(inode)->i_otime = inode->i_mtime;
5711 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5712 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5713 static_assert(BTRFS_FT_DIR == FT_DIR);
5714 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5715 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5716 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5717 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5718 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5720 static inline u8 btrfs_inode_type(struct inode *inode)
5722 return fs_umode_to_ftype(inode->i_mode);
5725 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5727 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5728 struct inode *inode;
5729 struct btrfs_root *root = BTRFS_I(dir)->root;
5730 struct btrfs_root *sub_root = root;
5731 struct btrfs_key location;
5735 if (dentry->d_name.len > BTRFS_NAME_LEN)
5736 return ERR_PTR(-ENAMETOOLONG);
5738 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5740 return ERR_PTR(ret);
5742 if (location.type == BTRFS_INODE_ITEM_KEY) {
5743 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5747 /* Do extra check against inode mode with di_type */
5748 if (btrfs_inode_type(inode) != di_type) {
5750 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5751 inode->i_mode, btrfs_inode_type(inode),
5754 return ERR_PTR(-EUCLEAN);
5759 ret = fixup_tree_root_location(fs_info, dir, dentry,
5760 &location, &sub_root);
5763 inode = ERR_PTR(ret);
5765 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5767 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5769 if (root != sub_root)
5770 btrfs_put_root(sub_root);
5772 if (!IS_ERR(inode) && root != sub_root) {
5773 down_read(&fs_info->cleanup_work_sem);
5774 if (!sb_rdonly(inode->i_sb))
5775 ret = btrfs_orphan_cleanup(sub_root);
5776 up_read(&fs_info->cleanup_work_sem);
5779 inode = ERR_PTR(ret);
5786 static int btrfs_dentry_delete(const struct dentry *dentry)
5788 struct btrfs_root *root;
5789 struct inode *inode = d_inode(dentry);
5791 if (!inode && !IS_ROOT(dentry))
5792 inode = d_inode(dentry->d_parent);
5795 root = BTRFS_I(inode)->root;
5796 if (btrfs_root_refs(&root->root_item) == 0)
5799 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5805 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5808 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5810 if (inode == ERR_PTR(-ENOENT))
5812 return d_splice_alias(inode, dentry);
5816 * All this infrastructure exists because dir_emit can fault, and we are holding
5817 * the tree lock when doing readdir. For now just allocate a buffer and copy
5818 * our information into that, and then dir_emit from the buffer. This is
5819 * similar to what NFS does, only we don't keep the buffer around in pagecache
5820 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5821 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5824 static int btrfs_opendir(struct inode *inode, struct file *file)
5826 struct btrfs_file_private *private;
5828 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5831 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5832 if (!private->filldir_buf) {
5836 file->private_data = private;
5847 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5850 struct dir_entry *entry = addr;
5851 char *name = (char *)(entry + 1);
5853 ctx->pos = get_unaligned(&entry->offset);
5854 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5855 get_unaligned(&entry->ino),
5856 get_unaligned(&entry->type)))
5858 addr += sizeof(struct dir_entry) +
5859 get_unaligned(&entry->name_len);
5865 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5867 struct inode *inode = file_inode(file);
5868 struct btrfs_root *root = BTRFS_I(inode)->root;
5869 struct btrfs_file_private *private = file->private_data;
5870 struct btrfs_dir_item *di;
5871 struct btrfs_key key;
5872 struct btrfs_key found_key;
5873 struct btrfs_path *path;
5875 struct list_head ins_list;
5876 struct list_head del_list;
5883 struct btrfs_key location;
5885 if (!dir_emit_dots(file, ctx))
5888 path = btrfs_alloc_path();
5892 addr = private->filldir_buf;
5893 path->reada = READA_FORWARD;
5895 INIT_LIST_HEAD(&ins_list);
5896 INIT_LIST_HEAD(&del_list);
5897 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5900 key.type = BTRFS_DIR_INDEX_KEY;
5901 key.offset = ctx->pos;
5902 key.objectid = btrfs_ino(BTRFS_I(inode));
5904 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5905 struct dir_entry *entry;
5906 struct extent_buffer *leaf = path->nodes[0];
5908 if (found_key.objectid != key.objectid)
5910 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5912 if (found_key.offset < ctx->pos)
5914 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5916 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5917 name_len = btrfs_dir_name_len(leaf, di);
5918 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5920 btrfs_release_path(path);
5921 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5924 addr = private->filldir_buf;
5931 put_unaligned(name_len, &entry->name_len);
5932 name_ptr = (char *)(entry + 1);
5933 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5935 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5937 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5938 put_unaligned(location.objectid, &entry->ino);
5939 put_unaligned(found_key.offset, &entry->offset);
5941 addr += sizeof(struct dir_entry) + name_len;
5942 total_len += sizeof(struct dir_entry) + name_len;
5944 /* Catch error encountered during iteration */
5948 btrfs_release_path(path);
5950 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5954 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5959 * Stop new entries from being returned after we return the last
5962 * New directory entries are assigned a strictly increasing
5963 * offset. This means that new entries created during readdir
5964 * are *guaranteed* to be seen in the future by that readdir.
5965 * This has broken buggy programs which operate on names as
5966 * they're returned by readdir. Until we re-use freed offsets
5967 * we have this hack to stop new entries from being returned
5968 * under the assumption that they'll never reach this huge
5971 * This is being careful not to overflow 32bit loff_t unless the
5972 * last entry requires it because doing so has broken 32bit apps
5975 if (ctx->pos >= INT_MAX)
5976 ctx->pos = LLONG_MAX;
5983 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5984 btrfs_free_path(path);
5989 * This is somewhat expensive, updating the tree every time the
5990 * inode changes. But, it is most likely to find the inode in cache.
5991 * FIXME, needs more benchmarking...there are no reasons other than performance
5992 * to keep or drop this code.
5994 static int btrfs_dirty_inode(struct inode *inode)
5996 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5997 struct btrfs_root *root = BTRFS_I(inode)->root;
5998 struct btrfs_trans_handle *trans;
6001 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6004 trans = btrfs_join_transaction(root);
6006 return PTR_ERR(trans);
6008 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6009 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6010 /* whoops, lets try again with the full transaction */
6011 btrfs_end_transaction(trans);
6012 trans = btrfs_start_transaction(root, 1);
6014 return PTR_ERR(trans);
6016 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6018 btrfs_end_transaction(trans);
6019 if (BTRFS_I(inode)->delayed_node)
6020 btrfs_balance_delayed_items(fs_info);
6026 * This is a copy of file_update_time. We need this so we can return error on
6027 * ENOSPC for updating the inode in the case of file write and mmap writes.
6029 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6032 struct btrfs_root *root = BTRFS_I(inode)->root;
6033 bool dirty = flags & ~S_VERSION;
6035 if (btrfs_root_readonly(root))
6038 if (flags & S_VERSION)
6039 dirty |= inode_maybe_inc_iversion(inode, dirty);
6040 if (flags & S_CTIME)
6041 inode->i_ctime = *now;
6042 if (flags & S_MTIME)
6043 inode->i_mtime = *now;
6044 if (flags & S_ATIME)
6045 inode->i_atime = *now;
6046 return dirty ? btrfs_dirty_inode(inode) : 0;
6050 * find the highest existing sequence number in a directory
6051 * and then set the in-memory index_cnt variable to reflect
6052 * free sequence numbers
6054 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6056 struct btrfs_root *root = inode->root;
6057 struct btrfs_key key, found_key;
6058 struct btrfs_path *path;
6059 struct extent_buffer *leaf;
6062 key.objectid = btrfs_ino(inode);
6063 key.type = BTRFS_DIR_INDEX_KEY;
6064 key.offset = (u64)-1;
6066 path = btrfs_alloc_path();
6070 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6073 /* FIXME: we should be able to handle this */
6078 if (path->slots[0] == 0) {
6079 inode->index_cnt = BTRFS_DIR_START_INDEX;
6085 leaf = path->nodes[0];
6086 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6088 if (found_key.objectid != btrfs_ino(inode) ||
6089 found_key.type != BTRFS_DIR_INDEX_KEY) {
6090 inode->index_cnt = BTRFS_DIR_START_INDEX;
6094 inode->index_cnt = found_key.offset + 1;
6096 btrfs_free_path(path);
6101 * helper to find a free sequence number in a given directory. This current
6102 * code is very simple, later versions will do smarter things in the btree
6104 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6108 if (dir->index_cnt == (u64)-1) {
6109 ret = btrfs_inode_delayed_dir_index_count(dir);
6111 ret = btrfs_set_inode_index_count(dir);
6117 *index = dir->index_cnt;
6123 static int btrfs_insert_inode_locked(struct inode *inode)
6125 struct btrfs_iget_args args;
6127 args.ino = BTRFS_I(inode)->location.objectid;
6128 args.root = BTRFS_I(inode)->root;
6130 return insert_inode_locked4(inode,
6131 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6132 btrfs_find_actor, &args);
6135 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6136 unsigned int *trans_num_items)
6138 struct inode *dir = args->dir;
6139 struct inode *inode = args->inode;
6142 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6146 /* 1 to add inode item */
6147 *trans_num_items = 1;
6148 /* 1 to add compression property */
6149 if (BTRFS_I(dir)->prop_compress)
6150 (*trans_num_items)++;
6151 /* 1 to add default ACL xattr */
6152 if (args->default_acl)
6153 (*trans_num_items)++;
6154 /* 1 to add access ACL xattr */
6156 (*trans_num_items)++;
6157 #ifdef CONFIG_SECURITY
6158 /* 1 to add LSM xattr */
6159 if (dir->i_security)
6160 (*trans_num_items)++;
6163 /* 1 to add orphan item */
6164 (*trans_num_items)++;
6168 * 1 to add dir index
6169 * 1 to update parent inode item
6171 * No need for 1 unit for the inode ref item because it is
6172 * inserted in a batch together with the inode item at
6173 * btrfs_create_new_inode().
6175 *trans_num_items += 3;
6180 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6182 posix_acl_release(args->acl);
6183 posix_acl_release(args->default_acl);
6187 * Inherit flags from the parent inode.
6189 * Currently only the compression flags and the cow flags are inherited.
6191 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6195 flags = BTRFS_I(dir)->flags;
6197 if (flags & BTRFS_INODE_NOCOMPRESS) {
6198 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6199 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6200 } else if (flags & BTRFS_INODE_COMPRESS) {
6201 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6202 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6205 if (flags & BTRFS_INODE_NODATACOW) {
6206 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6207 if (S_ISREG(inode->i_mode))
6208 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6211 btrfs_sync_inode_flags_to_i_flags(inode);
6214 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6215 struct btrfs_new_inode_args *args)
6217 struct inode *dir = args->dir;
6218 struct inode *inode = args->inode;
6219 const char *name = args->orphan ? NULL : args->dentry->d_name.name;
6220 int name_len = args->orphan ? 0 : args->dentry->d_name.len;
6221 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6222 struct btrfs_root *root;
6223 struct btrfs_inode_item *inode_item;
6224 struct btrfs_key *location;
6225 struct btrfs_path *path;
6227 struct btrfs_inode_ref *ref;
6228 struct btrfs_key key[2];
6230 struct btrfs_item_batch batch;
6234 path = btrfs_alloc_path();
6239 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6240 root = BTRFS_I(inode)->root;
6242 ret = btrfs_get_free_objectid(root, &objectid);
6245 inode->i_ino = objectid;
6249 * O_TMPFILE, set link count to 0, so that after this point, we
6250 * fill in an inode item with the correct link count.
6252 set_nlink(inode, 0);
6254 trace_btrfs_inode_request(dir);
6256 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6260 /* index_cnt is ignored for everything but a dir. */
6261 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6262 BTRFS_I(inode)->generation = trans->transid;
6263 inode->i_generation = BTRFS_I(inode)->generation;
6266 * Subvolumes don't inherit flags from their parent directory.
6267 * Originally this was probably by accident, but we probably can't
6268 * change it now without compatibility issues.
6271 btrfs_inherit_iflags(inode, dir);
6273 if (S_ISREG(inode->i_mode)) {
6274 if (btrfs_test_opt(fs_info, NODATASUM))
6275 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6276 if (btrfs_test_opt(fs_info, NODATACOW))
6277 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6278 BTRFS_INODE_NODATASUM;
6281 location = &BTRFS_I(inode)->location;
6282 location->objectid = objectid;
6283 location->offset = 0;
6284 location->type = BTRFS_INODE_ITEM_KEY;
6286 ret = btrfs_insert_inode_locked(inode);
6289 BTRFS_I(dir)->index_cnt--;
6294 * We could have gotten an inode number from somebody who was fsynced
6295 * and then removed in this same transaction, so let's just set full
6296 * sync since it will be a full sync anyway and this will blow away the
6297 * old info in the log.
6299 btrfs_set_inode_full_sync(BTRFS_I(inode));
6301 key[0].objectid = objectid;
6302 key[0].type = BTRFS_INODE_ITEM_KEY;
6305 sizes[0] = sizeof(struct btrfs_inode_item);
6307 if (!args->orphan) {
6309 * Start new inodes with an inode_ref. This is slightly more
6310 * efficient for small numbers of hard links since they will
6311 * be packed into one item. Extended refs will kick in if we
6312 * add more hard links than can fit in the ref item.
6314 key[1].objectid = objectid;
6315 key[1].type = BTRFS_INODE_REF_KEY;
6317 key[1].offset = objectid;
6318 sizes[1] = 2 + sizeof(*ref);
6320 key[1].offset = btrfs_ino(BTRFS_I(dir));
6321 sizes[1] = name_len + sizeof(*ref);
6325 batch.keys = &key[0];
6326 batch.data_sizes = &sizes[0];
6327 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6328 batch.nr = args->orphan ? 1 : 2;
6329 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6331 btrfs_abort_transaction(trans, ret);
6335 inode->i_mtime = current_time(inode);
6336 inode->i_atime = inode->i_mtime;
6337 inode->i_ctime = inode->i_mtime;
6338 BTRFS_I(inode)->i_otime = inode->i_mtime;
6341 * We're going to fill the inode item now, so at this point the inode
6342 * must be fully initialized.
6345 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6346 struct btrfs_inode_item);
6347 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6348 sizeof(*inode_item));
6349 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6351 if (!args->orphan) {
6352 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6353 struct btrfs_inode_ref);
6354 ptr = (unsigned long)(ref + 1);
6356 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6357 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6358 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6360 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6361 btrfs_set_inode_ref_index(path->nodes[0], ref,
6362 BTRFS_I(inode)->dir_index);
6363 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6367 btrfs_mark_buffer_dirty(path->nodes[0]);
6368 btrfs_release_path(path);
6371 struct inode *parent;
6374 * Subvolumes inherit properties from their parent subvolume,
6375 * not the directory they were created in.
6377 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6378 BTRFS_I(dir)->root);
6379 if (IS_ERR(parent)) {
6380 ret = PTR_ERR(parent);
6382 ret = btrfs_inode_inherit_props(trans, inode, parent);
6386 ret = btrfs_inode_inherit_props(trans, inode, dir);
6390 "error inheriting props for ino %llu (root %llu): %d",
6391 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6396 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6399 if (!args->subvol) {
6400 ret = btrfs_init_inode_security(trans, args);
6402 btrfs_abort_transaction(trans, ret);
6407 inode_tree_add(inode);
6409 trace_btrfs_inode_new(inode);
6410 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6412 btrfs_update_root_times(trans, root);
6415 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6417 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6418 name_len, 0, BTRFS_I(inode)->dir_index);
6421 btrfs_abort_transaction(trans, ret);
6430 * discard_new_inode() calls iput(), but the caller owns the reference
6434 discard_new_inode(inode);
6436 btrfs_free_path(path);
6441 * utility function to add 'inode' into 'parent_inode' with
6442 * a give name and a given sequence number.
6443 * if 'add_backref' is true, also insert a backref from the
6444 * inode to the parent directory.
6446 int btrfs_add_link(struct btrfs_trans_handle *trans,
6447 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6448 const char *name, int name_len, int add_backref, u64 index)
6451 struct btrfs_key key;
6452 struct btrfs_root *root = parent_inode->root;
6453 u64 ino = btrfs_ino(inode);
6454 u64 parent_ino = btrfs_ino(parent_inode);
6456 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6457 memcpy(&key, &inode->root->root_key, sizeof(key));
6460 key.type = BTRFS_INODE_ITEM_KEY;
6464 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6465 ret = btrfs_add_root_ref(trans, key.objectid,
6466 root->root_key.objectid, parent_ino,
6467 index, name, name_len);
6468 } else if (add_backref) {
6469 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6473 /* Nothing to clean up yet */
6477 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6478 btrfs_inode_type(&inode->vfs_inode), index);
6479 if (ret == -EEXIST || ret == -EOVERFLOW)
6482 btrfs_abort_transaction(trans, ret);
6486 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6488 inode_inc_iversion(&parent_inode->vfs_inode);
6490 * If we are replaying a log tree, we do not want to update the mtime
6491 * and ctime of the parent directory with the current time, since the
6492 * log replay procedure is responsible for setting them to their correct
6493 * values (the ones it had when the fsync was done).
6495 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6496 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6498 parent_inode->vfs_inode.i_mtime = now;
6499 parent_inode->vfs_inode.i_ctime = now;
6501 ret = btrfs_update_inode(trans, root, parent_inode);
6503 btrfs_abort_transaction(trans, ret);
6507 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6510 err = btrfs_del_root_ref(trans, key.objectid,
6511 root->root_key.objectid, parent_ino,
6512 &local_index, name, name_len);
6514 btrfs_abort_transaction(trans, err);
6515 } else if (add_backref) {
6519 err = btrfs_del_inode_ref(trans, root, name, name_len,
6520 ino, parent_ino, &local_index);
6522 btrfs_abort_transaction(trans, err);
6525 /* Return the original error code */
6529 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6530 struct inode *inode)
6532 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6533 struct btrfs_root *root = BTRFS_I(dir)->root;
6534 struct btrfs_new_inode_args new_inode_args = {
6539 unsigned int trans_num_items;
6540 struct btrfs_trans_handle *trans;
6543 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6547 trans = btrfs_start_transaction(root, trans_num_items);
6548 if (IS_ERR(trans)) {
6549 err = PTR_ERR(trans);
6550 goto out_new_inode_args;
6553 err = btrfs_create_new_inode(trans, &new_inode_args);
6555 d_instantiate_new(dentry, inode);
6557 btrfs_end_transaction(trans);
6558 btrfs_btree_balance_dirty(fs_info);
6560 btrfs_new_inode_args_destroy(&new_inode_args);
6567 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6568 struct dentry *dentry, umode_t mode, dev_t rdev)
6570 struct inode *inode;
6572 inode = new_inode(dir->i_sb);
6575 inode_init_owner(mnt_userns, inode, dir, mode);
6576 inode->i_op = &btrfs_special_inode_operations;
6577 init_special_inode(inode, inode->i_mode, rdev);
6578 return btrfs_create_common(dir, dentry, inode);
6581 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6582 struct dentry *dentry, umode_t mode, bool excl)
6584 struct inode *inode;
6586 inode = new_inode(dir->i_sb);
6589 inode_init_owner(mnt_userns, inode, dir, mode);
6590 inode->i_fop = &btrfs_file_operations;
6591 inode->i_op = &btrfs_file_inode_operations;
6592 inode->i_mapping->a_ops = &btrfs_aops;
6593 return btrfs_create_common(dir, dentry, inode);
6596 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6597 struct dentry *dentry)
6599 struct btrfs_trans_handle *trans = NULL;
6600 struct btrfs_root *root = BTRFS_I(dir)->root;
6601 struct inode *inode = d_inode(old_dentry);
6602 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6607 /* do not allow sys_link's with other subvols of the same device */
6608 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6611 if (inode->i_nlink >= BTRFS_LINK_MAX)
6614 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6619 * 2 items for inode and inode ref
6620 * 2 items for dir items
6621 * 1 item for parent inode
6622 * 1 item for orphan item deletion if O_TMPFILE
6624 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6625 if (IS_ERR(trans)) {
6626 err = PTR_ERR(trans);
6631 /* There are several dir indexes for this inode, clear the cache. */
6632 BTRFS_I(inode)->dir_index = 0ULL;
6634 inode_inc_iversion(inode);
6635 inode->i_ctime = current_time(inode);
6637 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6639 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6640 dentry->d_name.name, dentry->d_name.len, 1, index);
6645 struct dentry *parent = dentry->d_parent;
6647 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6650 if (inode->i_nlink == 1) {
6652 * If new hard link count is 1, it's a file created
6653 * with open(2) O_TMPFILE flag.
6655 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6659 d_instantiate(dentry, inode);
6660 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6665 btrfs_end_transaction(trans);
6667 inode_dec_link_count(inode);
6670 btrfs_btree_balance_dirty(fs_info);
6674 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6675 struct dentry *dentry, umode_t mode)
6677 struct inode *inode;
6679 inode = new_inode(dir->i_sb);
6682 inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode);
6683 inode->i_op = &btrfs_dir_inode_operations;
6684 inode->i_fop = &btrfs_dir_file_operations;
6685 return btrfs_create_common(dir, dentry, inode);
6688 static noinline int uncompress_inline(struct btrfs_path *path,
6690 size_t pg_offset, u64 extent_offset,
6691 struct btrfs_file_extent_item *item)
6694 struct extent_buffer *leaf = path->nodes[0];
6697 unsigned long inline_size;
6701 WARN_ON(pg_offset != 0);
6702 compress_type = btrfs_file_extent_compression(leaf, item);
6703 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6704 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6705 tmp = kmalloc(inline_size, GFP_NOFS);
6708 ptr = btrfs_file_extent_inline_start(item);
6710 read_extent_buffer(leaf, tmp, ptr, inline_size);
6712 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6713 ret = btrfs_decompress(compress_type, tmp, page,
6714 extent_offset, inline_size, max_size);
6717 * decompression code contains a memset to fill in any space between the end
6718 * of the uncompressed data and the end of max_size in case the decompressed
6719 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6720 * the end of an inline extent and the beginning of the next block, so we
6721 * cover that region here.
6724 if (max_size + pg_offset < PAGE_SIZE)
6725 memzero_page(page, pg_offset + max_size,
6726 PAGE_SIZE - max_size - pg_offset);
6732 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6733 * @inode: file to search in
6734 * @page: page to read extent data into if the extent is inline
6735 * @pg_offset: offset into @page to copy to
6736 * @start: file offset
6737 * @len: length of range starting at @start
6739 * This returns the first &struct extent_map which overlaps with the given
6740 * range, reading it from the B-tree and caching it if necessary. Note that
6741 * there may be more extents which overlap the given range after the returned
6744 * If @page is not NULL and the extent is inline, this also reads the extent
6745 * data directly into the page and marks the extent up to date in the io_tree.
6747 * Return: ERR_PTR on error, non-NULL extent_map on success.
6749 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6750 struct page *page, size_t pg_offset,
6753 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6755 u64 extent_start = 0;
6757 u64 objectid = btrfs_ino(inode);
6758 int extent_type = -1;
6759 struct btrfs_path *path = NULL;
6760 struct btrfs_root *root = inode->root;
6761 struct btrfs_file_extent_item *item;
6762 struct extent_buffer *leaf;
6763 struct btrfs_key found_key;
6764 struct extent_map *em = NULL;
6765 struct extent_map_tree *em_tree = &inode->extent_tree;
6766 struct extent_io_tree *io_tree = &inode->io_tree;
6768 read_lock(&em_tree->lock);
6769 em = lookup_extent_mapping(em_tree, start, len);
6770 read_unlock(&em_tree->lock);
6773 if (em->start > start || em->start + em->len <= start)
6774 free_extent_map(em);
6775 else if (em->block_start == EXTENT_MAP_INLINE && page)
6776 free_extent_map(em);
6780 em = alloc_extent_map();
6785 em->start = EXTENT_MAP_HOLE;
6786 em->orig_start = EXTENT_MAP_HOLE;
6788 em->block_len = (u64)-1;
6790 path = btrfs_alloc_path();
6796 /* Chances are we'll be called again, so go ahead and do readahead */
6797 path->reada = READA_FORWARD;
6800 * The same explanation in load_free_space_cache applies here as well,
6801 * we only read when we're loading the free space cache, and at that
6802 * point the commit_root has everything we need.
6804 if (btrfs_is_free_space_inode(inode)) {
6805 path->search_commit_root = 1;
6806 path->skip_locking = 1;
6809 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6812 } else if (ret > 0) {
6813 if (path->slots[0] == 0)
6819 leaf = path->nodes[0];
6820 item = btrfs_item_ptr(leaf, path->slots[0],
6821 struct btrfs_file_extent_item);
6822 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6823 if (found_key.objectid != objectid ||
6824 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6826 * If we backup past the first extent we want to move forward
6827 * and see if there is an extent in front of us, otherwise we'll
6828 * say there is a hole for our whole search range which can
6835 extent_type = btrfs_file_extent_type(leaf, item);
6836 extent_start = found_key.offset;
6837 extent_end = btrfs_file_extent_end(path);
6838 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6839 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6840 /* Only regular file could have regular/prealloc extent */
6841 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6844 "regular/prealloc extent found for non-regular inode %llu",
6848 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6850 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6851 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6856 if (start >= extent_end) {
6858 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6859 ret = btrfs_next_leaf(root, path);
6865 leaf = path->nodes[0];
6867 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6868 if (found_key.objectid != objectid ||
6869 found_key.type != BTRFS_EXTENT_DATA_KEY)
6871 if (start + len <= found_key.offset)
6873 if (start > found_key.offset)
6876 /* New extent overlaps with existing one */
6878 em->orig_start = start;
6879 em->len = found_key.offset - start;
6880 em->block_start = EXTENT_MAP_HOLE;
6884 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6886 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6887 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6889 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6893 size_t extent_offset;
6899 size = btrfs_file_extent_ram_bytes(leaf, item);
6900 extent_offset = page_offset(page) + pg_offset - extent_start;
6901 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6902 size - extent_offset);
6903 em->start = extent_start + extent_offset;
6904 em->len = ALIGN(copy_size, fs_info->sectorsize);
6905 em->orig_block_len = em->len;
6906 em->orig_start = em->start;
6907 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6909 if (!PageUptodate(page)) {
6910 if (btrfs_file_extent_compression(leaf, item) !=
6911 BTRFS_COMPRESS_NONE) {
6912 ret = uncompress_inline(path, page, pg_offset,
6913 extent_offset, item);
6917 map = kmap_local_page(page);
6918 read_extent_buffer(leaf, map + pg_offset, ptr,
6920 if (pg_offset + copy_size < PAGE_SIZE) {
6921 memset(map + pg_offset + copy_size, 0,
6922 PAGE_SIZE - pg_offset -
6927 flush_dcache_page(page);
6929 set_extent_uptodate(io_tree, em->start,
6930 extent_map_end(em) - 1, NULL, GFP_NOFS);
6935 em->orig_start = start;
6937 em->block_start = EXTENT_MAP_HOLE;
6940 btrfs_release_path(path);
6941 if (em->start > start || extent_map_end(em) <= start) {
6943 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6944 em->start, em->len, start, len);
6949 write_lock(&em_tree->lock);
6950 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6951 write_unlock(&em_tree->lock);
6953 btrfs_free_path(path);
6955 trace_btrfs_get_extent(root, inode, em);
6958 free_extent_map(em);
6959 return ERR_PTR(ret);
6964 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6967 struct extent_map *em;
6968 struct extent_map *hole_em = NULL;
6969 u64 delalloc_start = start;
6975 em = btrfs_get_extent(inode, NULL, 0, start, len);
6979 * If our em maps to:
6981 * - a pre-alloc extent,
6982 * there might actually be delalloc bytes behind it.
6984 if (em->block_start != EXTENT_MAP_HOLE &&
6985 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6990 /* check to see if we've wrapped (len == -1 or similar) */
6999 /* ok, we didn't find anything, lets look for delalloc */
7000 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7001 end, len, EXTENT_DELALLOC, 1);
7002 delalloc_end = delalloc_start + delalloc_len;
7003 if (delalloc_end < delalloc_start)
7004 delalloc_end = (u64)-1;
7007 * We didn't find anything useful, return the original results from
7010 if (delalloc_start > end || delalloc_end <= start) {
7017 * Adjust the delalloc_start to make sure it doesn't go backwards from
7018 * the start they passed in
7020 delalloc_start = max(start, delalloc_start);
7021 delalloc_len = delalloc_end - delalloc_start;
7023 if (delalloc_len > 0) {
7026 const u64 hole_end = extent_map_end(hole_em);
7028 em = alloc_extent_map();
7036 * When btrfs_get_extent can't find anything it returns one
7039 * Make sure what it found really fits our range, and adjust to
7040 * make sure it is based on the start from the caller
7042 if (hole_end <= start || hole_em->start > end) {
7043 free_extent_map(hole_em);
7046 hole_start = max(hole_em->start, start);
7047 hole_len = hole_end - hole_start;
7050 if (hole_em && delalloc_start > hole_start) {
7052 * Our hole starts before our delalloc, so we have to
7053 * return just the parts of the hole that go until the
7056 em->len = min(hole_len, delalloc_start - hole_start);
7057 em->start = hole_start;
7058 em->orig_start = hole_start;
7060 * Don't adjust block start at all, it is fixed at
7063 em->block_start = hole_em->block_start;
7064 em->block_len = hole_len;
7065 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7066 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7069 * Hole is out of passed range or it starts after
7072 em->start = delalloc_start;
7073 em->len = delalloc_len;
7074 em->orig_start = delalloc_start;
7075 em->block_start = EXTENT_MAP_DELALLOC;
7076 em->block_len = delalloc_len;
7083 free_extent_map(hole_em);
7085 free_extent_map(em);
7086 return ERR_PTR(err);
7091 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7094 const u64 orig_start,
7095 const u64 block_start,
7096 const u64 block_len,
7097 const u64 orig_block_len,
7098 const u64 ram_bytes,
7101 struct extent_map *em = NULL;
7104 if (type != BTRFS_ORDERED_NOCOW) {
7105 em = create_io_em(inode, start, len, orig_start, block_start,
7106 block_len, orig_block_len, ram_bytes,
7107 BTRFS_COMPRESS_NONE, /* compress_type */
7112 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7115 (1 << BTRFS_ORDERED_DIRECT),
7116 BTRFS_COMPRESS_NONE);
7119 free_extent_map(em);
7120 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7129 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7132 struct btrfs_root *root = inode->root;
7133 struct btrfs_fs_info *fs_info = root->fs_info;
7134 struct extent_map *em;
7135 struct btrfs_key ins;
7139 alloc_hint = get_extent_allocation_hint(inode, start, len);
7140 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7141 0, alloc_hint, &ins, 1, 1);
7143 return ERR_PTR(ret);
7145 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7146 ins.objectid, ins.offset, ins.offset,
7147 ins.offset, BTRFS_ORDERED_REGULAR);
7148 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7150 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7156 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7158 struct btrfs_block_group *block_group;
7159 bool readonly = false;
7161 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7162 if (!block_group || block_group->ro)
7165 btrfs_put_block_group(block_group);
7170 * Check if we can do nocow write into the range [@offset, @offset + @len)
7172 * @offset: File offset
7173 * @len: The length to write, will be updated to the nocow writeable
7175 * @orig_start: (optional) Return the original file offset of the file extent
7176 * @orig_len: (optional) Return the original on-disk length of the file extent
7177 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7178 * @strict: if true, omit optimizations that might force us into unnecessary
7179 * cow. e.g., don't trust generation number.
7182 * >0 and update @len if we can do nocow write
7183 * 0 if we can't do nocow write
7184 * <0 if error happened
7186 * NOTE: This only checks the file extents, caller is responsible to wait for
7187 * any ordered extents.
7189 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7190 u64 *orig_start, u64 *orig_block_len,
7191 u64 *ram_bytes, bool strict)
7193 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7194 struct can_nocow_file_extent_args nocow_args = { 0 };
7195 struct btrfs_path *path;
7197 struct extent_buffer *leaf;
7198 struct btrfs_root *root = BTRFS_I(inode)->root;
7199 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7200 struct btrfs_file_extent_item *fi;
7201 struct btrfs_key key;
7204 path = btrfs_alloc_path();
7208 ret = btrfs_lookup_file_extent(NULL, root, path,
7209 btrfs_ino(BTRFS_I(inode)), offset, 0);
7214 if (path->slots[0] == 0) {
7215 /* can't find the item, must cow */
7222 leaf = path->nodes[0];
7223 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7224 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7225 key.type != BTRFS_EXTENT_DATA_KEY) {
7226 /* not our file or wrong item type, must cow */
7230 if (key.offset > offset) {
7231 /* Wrong offset, must cow */
7235 if (btrfs_file_extent_end(path) <= offset)
7238 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7239 found_type = btrfs_file_extent_type(leaf, fi);
7241 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7243 nocow_args.start = offset;
7244 nocow_args.end = offset + *len - 1;
7245 nocow_args.strict = strict;
7246 nocow_args.free_path = true;
7248 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7249 /* can_nocow_file_extent() has freed the path. */
7253 /* Treat errors as not being able to NOCOW. */
7259 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7262 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7263 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7266 range_end = round_up(offset + nocow_args.num_bytes,
7267 root->fs_info->sectorsize) - 1;
7268 ret = test_range_bit(io_tree, offset, range_end,
7269 EXTENT_DELALLOC, 0, NULL);
7277 *orig_start = key.offset - nocow_args.extent_offset;
7279 *orig_block_len = nocow_args.disk_num_bytes;
7281 *len = nocow_args.num_bytes;
7284 btrfs_free_path(path);
7288 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7289 struct extent_state **cached_state,
7290 unsigned int iomap_flags)
7292 const bool writing = (iomap_flags & IOMAP_WRITE);
7293 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7294 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7295 struct btrfs_ordered_extent *ordered;
7300 if (!try_lock_extent(io_tree, lockstart, lockend))
7303 lock_extent_bits(io_tree, lockstart, lockend, cached_state);
7306 * We're concerned with the entire range that we're going to be
7307 * doing DIO to, so we need to make sure there's no ordered
7308 * extents in this range.
7310 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7311 lockend - lockstart + 1);
7314 * We need to make sure there are no buffered pages in this
7315 * range either, we could have raced between the invalidate in
7316 * generic_file_direct_write and locking the extent. The
7317 * invalidate needs to happen so that reads after a write do not
7321 (!writing || !filemap_range_has_page(inode->i_mapping,
7322 lockstart, lockend)))
7325 unlock_extent_cached(io_tree, lockstart, lockend, cached_state);
7329 btrfs_put_ordered_extent(ordered);
7334 * If we are doing a DIO read and the ordered extent we
7335 * found is for a buffered write, we can not wait for it
7336 * to complete and retry, because if we do so we can
7337 * deadlock with concurrent buffered writes on page
7338 * locks. This happens only if our DIO read covers more
7339 * than one extent map, if at this point has already
7340 * created an ordered extent for a previous extent map
7341 * and locked its range in the inode's io tree, and a
7342 * concurrent write against that previous extent map's
7343 * range and this range started (we unlock the ranges
7344 * in the io tree only when the bios complete and
7345 * buffered writes always lock pages before attempting
7346 * to lock range in the io tree).
7349 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7350 btrfs_start_ordered_extent(ordered, 1);
7352 ret = nowait ? -EAGAIN : -ENOTBLK;
7353 btrfs_put_ordered_extent(ordered);
7356 * We could trigger writeback for this range (and wait
7357 * for it to complete) and then invalidate the pages for
7358 * this range (through invalidate_inode_pages2_range()),
7359 * but that can lead us to a deadlock with a concurrent
7360 * call to readahead (a buffered read or a defrag call
7361 * triggered a readahead) on a page lock due to an
7362 * ordered dio extent we created before but did not have
7363 * yet a corresponding bio submitted (whence it can not
7364 * complete), which makes readahead wait for that
7365 * ordered extent to complete while holding a lock on
7368 ret = nowait ? -EAGAIN : -ENOTBLK;
7380 /* The callers of this must take lock_extent() */
7381 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7382 u64 len, u64 orig_start, u64 block_start,
7383 u64 block_len, u64 orig_block_len,
7384 u64 ram_bytes, int compress_type,
7387 struct extent_map_tree *em_tree;
7388 struct extent_map *em;
7391 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7392 type == BTRFS_ORDERED_COMPRESSED ||
7393 type == BTRFS_ORDERED_NOCOW ||
7394 type == BTRFS_ORDERED_REGULAR);
7396 em_tree = &inode->extent_tree;
7397 em = alloc_extent_map();
7399 return ERR_PTR(-ENOMEM);
7402 em->orig_start = orig_start;
7404 em->block_len = block_len;
7405 em->block_start = block_start;
7406 em->orig_block_len = orig_block_len;
7407 em->ram_bytes = ram_bytes;
7408 em->generation = -1;
7409 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7410 if (type == BTRFS_ORDERED_PREALLOC) {
7411 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7412 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7413 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7414 em->compress_type = compress_type;
7418 btrfs_drop_extent_cache(inode, em->start,
7419 em->start + em->len - 1, 0);
7420 write_lock(&em_tree->lock);
7421 ret = add_extent_mapping(em_tree, em, 1);
7422 write_unlock(&em_tree->lock);
7424 * The caller has taken lock_extent(), who could race with us
7427 } while (ret == -EEXIST);
7430 free_extent_map(em);
7431 return ERR_PTR(ret);
7434 /* em got 2 refs now, callers needs to do free_extent_map once. */
7439 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7440 struct inode *inode,
7441 struct btrfs_dio_data *dio_data,
7443 unsigned int iomap_flags)
7445 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7446 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7447 struct extent_map *em = *map;
7449 u64 block_start, orig_start, orig_block_len, ram_bytes;
7450 struct btrfs_block_group *bg;
7451 bool can_nocow = false;
7452 bool space_reserved = false;
7457 * We don't allocate a new extent in the following cases
7459 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7461 * 2) The extent is marked as PREALLOC. We're good to go here and can
7462 * just use the extent.
7465 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7466 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7467 em->block_start != EXTENT_MAP_HOLE)) {
7468 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7469 type = BTRFS_ORDERED_PREALLOC;
7471 type = BTRFS_ORDERED_NOCOW;
7472 len = min(len, em->len - (start - em->start));
7473 block_start = em->block_start + (start - em->start);
7475 if (can_nocow_extent(inode, start, &len, &orig_start,
7476 &orig_block_len, &ram_bytes, false) == 1) {
7477 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7485 struct extent_map *em2;
7487 /* We can NOCOW, so only need to reserve metadata space. */
7488 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7491 /* Our caller expects us to free the input extent map. */
7492 free_extent_map(em);
7494 btrfs_dec_nocow_writers(bg);
7495 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7499 space_reserved = true;
7501 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7502 orig_start, block_start,
7503 len, orig_block_len,
7505 btrfs_dec_nocow_writers(bg);
7506 if (type == BTRFS_ORDERED_PREALLOC) {
7507 free_extent_map(em);
7516 dio_data->nocow_done = true;
7518 /* Our caller expects us to free the input extent map. */
7519 free_extent_map(em);
7526 * If we could not allocate data space before locking the file
7527 * range and we can't do a NOCOW write, then we have to fail.
7529 if (!dio_data->data_space_reserved)
7533 * We have to COW and we have already reserved data space before,
7534 * so now we reserve only metadata.
7536 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7540 space_reserved = true;
7542 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7548 len = min(len, em->len - (start - em->start));
7550 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7551 prev_len - len, true);
7555 * We have created our ordered extent, so we can now release our reservation
7556 * for an outstanding extent.
7558 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7561 * Need to update the i_size under the extent lock so buffered
7562 * readers will get the updated i_size when we unlock.
7564 if (start + len > i_size_read(inode))
7565 i_size_write(inode, start + len);
7567 if (ret && space_reserved) {
7568 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7569 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7574 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7575 loff_t length, unsigned int flags, struct iomap *iomap,
7576 struct iomap *srcmap)
7578 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7579 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7580 struct extent_map *em;
7581 struct extent_state *cached_state = NULL;
7582 struct btrfs_dio_data *dio_data = iter->private;
7583 u64 lockstart, lockend;
7584 const bool write = !!(flags & IOMAP_WRITE);
7587 const u64 data_alloc_len = length;
7588 bool unlock_extents = false;
7591 len = min_t(u64, len, fs_info->sectorsize);
7594 lockend = start + len - 1;
7597 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7598 * enough if we've written compressed pages to this area, so we need to
7599 * flush the dirty pages again to make absolutely sure that any
7600 * outstanding dirty pages are on disk - the first flush only starts
7601 * compression on the data, while keeping the pages locked, so by the
7602 * time the second flush returns we know bios for the compressed pages
7603 * were submitted and finished, and the pages no longer under writeback.
7605 * If we have a NOWAIT request and we have any pages in the range that
7606 * are locked, likely due to compression still in progress, we don't want
7607 * to block on page locks. We also don't want to block on pages marked as
7608 * dirty or under writeback (same as for the non-compression case).
7609 * iomap_dio_rw() did the same check, but after that and before we got
7610 * here, mmap'ed writes may have happened or buffered reads started
7611 * (readpage() and readahead(), which lock pages), as we haven't locked
7612 * the file range yet.
7614 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7615 &BTRFS_I(inode)->runtime_flags)) {
7616 if (flags & IOMAP_NOWAIT) {
7617 if (filemap_range_needs_writeback(inode->i_mapping,
7618 lockstart, lockend))
7621 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7622 start + length - 1);
7628 memset(dio_data, 0, sizeof(*dio_data));
7631 * We always try to allocate data space and must do it before locking
7632 * the file range, to avoid deadlocks with concurrent writes to the same
7633 * range if the range has several extents and the writes don't expand the
7634 * current i_size (the inode lock is taken in shared mode). If we fail to
7635 * allocate data space here we continue and later, after locking the
7636 * file range, we fail with ENOSPC only if we figure out we can not do a
7639 if (write && !(flags & IOMAP_NOWAIT)) {
7640 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7641 &dio_data->data_reserved,
7642 start, data_alloc_len);
7644 dio_data->data_space_reserved = true;
7645 else if (ret && !(BTRFS_I(inode)->flags &
7646 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7651 * If this errors out it's because we couldn't invalidate pagecache for
7652 * this range and we need to fallback to buffered IO, or we are doing a
7653 * NOWAIT read/write and we need to block.
7655 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7659 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7666 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7667 * io. INLINE is special, and we could probably kludge it in here, but
7668 * it's still buffered so for safety lets just fall back to the generic
7671 * For COMPRESSED we _have_ to read the entire extent in so we can
7672 * decompress it, so there will be buffering required no matter what we
7673 * do, so go ahead and fallback to buffered.
7675 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7676 * to buffered IO. Don't blame me, this is the price we pay for using
7679 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7680 em->block_start == EXTENT_MAP_INLINE) {
7681 free_extent_map(em);
7686 len = min(len, em->len - (start - em->start));
7689 * If we have a NOWAIT request and the range contains multiple extents
7690 * (or a mix of extents and holes), then we return -EAGAIN to make the
7691 * caller fallback to a context where it can do a blocking (without
7692 * NOWAIT) request. This way we avoid doing partial IO and returning
7693 * success to the caller, which is not optimal for writes and for reads
7694 * it can result in unexpected behaviour for an application.
7696 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7697 * iomap_dio_rw(), we can end up returning less data then what the caller
7698 * asked for, resulting in an unexpected, and incorrect, short read.
7699 * That is, the caller asked to read N bytes and we return less than that,
7700 * which is wrong unless we are crossing EOF. This happens if we get a
7701 * page fault error when trying to fault in pages for the buffer that is
7702 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7703 * have previously submitted bios for other extents in the range, in
7704 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7705 * those bios have completed by the time we get the page fault error,
7706 * which we return back to our caller - we should only return EIOCBQUEUED
7707 * after we have submitted bios for all the extents in the range.
7709 if ((flags & IOMAP_NOWAIT) && len < length) {
7710 free_extent_map(em);
7716 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7720 unlock_extents = true;
7721 /* Recalc len in case the new em is smaller than requested */
7722 len = min(len, em->len - (start - em->start));
7723 if (dio_data->data_space_reserved) {
7725 u64 release_len = 0;
7727 if (dio_data->nocow_done) {
7728 release_offset = start;
7729 release_len = data_alloc_len;
7730 } else if (len < data_alloc_len) {
7731 release_offset = start + len;
7732 release_len = data_alloc_len - len;
7735 if (release_len > 0)
7736 btrfs_free_reserved_data_space(BTRFS_I(inode),
7737 dio_data->data_reserved,
7743 * We need to unlock only the end area that we aren't using.
7744 * The rest is going to be unlocked by the endio routine.
7746 lockstart = start + len;
7747 if (lockstart < lockend)
7748 unlock_extents = true;
7752 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7753 lockstart, lockend, &cached_state);
7755 free_extent_state(cached_state);
7758 * Translate extent map information to iomap.
7759 * We trim the extents (and move the addr) even though iomap code does
7760 * that, since we have locked only the parts we are performing I/O in.
7762 if ((em->block_start == EXTENT_MAP_HOLE) ||
7763 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7764 iomap->addr = IOMAP_NULL_ADDR;
7765 iomap->type = IOMAP_HOLE;
7767 iomap->addr = em->block_start + (start - em->start);
7768 iomap->type = IOMAP_MAPPED;
7770 iomap->offset = start;
7771 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7772 iomap->length = len;
7774 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7775 iomap->flags |= IOMAP_F_ZONE_APPEND;
7777 free_extent_map(em);
7782 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7785 if (dio_data->data_space_reserved) {
7786 btrfs_free_reserved_data_space(BTRFS_I(inode),
7787 dio_data->data_reserved,
7788 start, data_alloc_len);
7789 extent_changeset_free(dio_data->data_reserved);
7795 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7796 ssize_t written, unsigned int flags, struct iomap *iomap)
7798 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7799 struct btrfs_dio_data *dio_data = iter->private;
7800 size_t submitted = dio_data->submitted;
7801 const bool write = !!(flags & IOMAP_WRITE);
7804 if (!write && (iomap->type == IOMAP_HOLE)) {
7805 /* If reading from a hole, unlock and return */
7806 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7810 if (submitted < length) {
7812 length -= submitted;
7814 __endio_write_update_ordered(BTRFS_I(inode), pos,
7817 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7823 extent_changeset_free(dio_data->data_reserved);
7827 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7830 * This implies a barrier so that stores to dio_bio->bi_status before
7831 * this and loads of dio_bio->bi_status after this are fully ordered.
7833 if (!refcount_dec_and_test(&dip->refs))
7836 if (btrfs_op(&dip->bio) == BTRFS_MAP_WRITE) {
7837 __endio_write_update_ordered(BTRFS_I(dip->inode),
7840 !dip->bio.bi_status);
7842 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7844 dip->file_offset + dip->bytes - 1);
7848 bio_endio(&dip->bio);
7851 static void submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7853 enum btrfs_compression_type compress_type)
7855 struct btrfs_dio_private *dip = bio->bi_private;
7856 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7858 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7860 if (btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA))
7863 refcount_inc(&dip->refs);
7864 if (btrfs_map_bio(fs_info, bio, mirror_num))
7865 refcount_dec(&dip->refs);
7868 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7869 struct btrfs_bio *bbio,
7870 const bool uptodate)
7872 struct inode *inode = dip->inode;
7873 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7874 const u32 sectorsize = fs_info->sectorsize;
7875 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7876 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7877 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7878 struct bio_vec bvec;
7879 struct bvec_iter iter;
7881 blk_status_t err = BLK_STS_OK;
7883 __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) {
7884 unsigned int i, nr_sectors, pgoff;
7886 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7887 pgoff = bvec.bv_offset;
7888 for (i = 0; i < nr_sectors; i++) {
7889 u64 start = bbio->file_offset + bio_offset;
7891 ASSERT(pgoff < PAGE_SIZE);
7893 (!csum || !check_data_csum(inode, bbio,
7894 bio_offset, bvec.bv_page,
7896 clean_io_failure(fs_info, failure_tree, io_tree,
7897 start, bvec.bv_page,
7898 btrfs_ino(BTRFS_I(inode)),
7903 ret = btrfs_repair_one_sector(inode, &bbio->bio,
7904 bio_offset, bvec.bv_page, pgoff,
7905 start, bbio->mirror_num,
7906 submit_dio_repair_bio);
7908 err = errno_to_blk_status(ret);
7910 ASSERT(bio_offset + sectorsize > bio_offset);
7911 bio_offset += sectorsize;
7912 pgoff += sectorsize;
7918 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7919 const u64 offset, const u64 bytes,
7920 const bool uptodate)
7922 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7923 finish_ordered_fn, uptodate);
7926 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7928 u64 dio_file_offset)
7930 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
7933 static void btrfs_end_dio_bio(struct bio *bio)
7935 struct btrfs_dio_private *dip = bio->bi_private;
7936 struct btrfs_bio *bbio = btrfs_bio(bio);
7937 blk_status_t err = bio->bi_status;
7940 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7941 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7942 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7943 bio->bi_opf, bio->bi_iter.bi_sector,
7944 bio->bi_iter.bi_size, err);
7946 if (bio_op(bio) == REQ_OP_READ)
7947 err = btrfs_check_read_dio_bio(dip, bbio, !err);
7950 dip->bio.bi_status = err;
7952 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
7955 btrfs_dio_private_put(dip);
7958 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7959 struct inode *inode, u64 file_offset, int async_submit)
7961 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7962 struct btrfs_dio_private *dip = bio->bi_private;
7963 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7966 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7968 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7971 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7976 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7979 if (write && async_submit) {
7980 ret = btrfs_wq_submit_bio(inode, bio, 0, file_offset,
7981 btrfs_submit_bio_start_direct_io);
7985 * If we aren't doing async submit, calculate the csum of the
7988 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
7994 csum_offset = file_offset - dip->file_offset;
7995 csum_offset >>= fs_info->sectorsize_bits;
7996 csum_offset *= fs_info->csum_size;
7997 btrfs_bio(bio)->csum = dip->csums + csum_offset;
8000 ret = btrfs_map_bio(fs_info, bio, 0);
8005 static void btrfs_submit_direct(const struct iomap_iter *iter,
8006 struct bio *dio_bio, loff_t file_offset)
8008 struct btrfs_dio_private *dip =
8009 container_of(dio_bio, struct btrfs_dio_private, bio);
8010 struct inode *inode = iter->inode;
8011 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8012 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8013 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8014 BTRFS_BLOCK_GROUP_RAID56_MASK);
8017 int async_submit = 0;
8019 u64 clone_offset = 0;
8023 blk_status_t status;
8024 struct btrfs_io_geometry geom;
8025 struct btrfs_dio_data *dio_data = iter->private;
8026 struct extent_map *em = NULL;
8029 dip->file_offset = file_offset;
8030 dip->bytes = dio_bio->bi_iter.bi_size;
8031 refcount_set(&dip->refs, 1);
8034 if (!write && !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
8035 unsigned int nr_sectors =
8036 (dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits);
8039 * Load the csums up front to reduce csum tree searches and
8040 * contention when submitting bios.
8042 status = BLK_STS_RESOURCE;
8043 dip->csums = kcalloc(nr_sectors, fs_info->csum_size, GFP_NOFS);
8047 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8048 if (status != BLK_STS_OK)
8052 start_sector = dio_bio->bi_iter.bi_sector;
8053 submit_len = dio_bio->bi_iter.bi_size;
8056 logical = start_sector << 9;
8057 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8059 status = errno_to_blk_status(PTR_ERR(em));
8063 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8066 status = errno_to_blk_status(ret);
8070 clone_len = min(submit_len, geom.len);
8071 ASSERT(clone_len <= UINT_MAX);
8074 * This will never fail as it's passing GPF_NOFS and
8075 * the allocation is backed by btrfs_bioset.
8077 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8078 bio->bi_private = dip;
8079 bio->bi_end_io = btrfs_end_dio_bio;
8080 btrfs_bio(bio)->file_offset = file_offset;
8082 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8083 status = extract_ordered_extent(BTRFS_I(inode), bio,
8091 ASSERT(submit_len >= clone_len);
8092 submit_len -= clone_len;
8095 * Increase the count before we submit the bio so we know
8096 * the end IO handler won't happen before we increase the
8097 * count. Otherwise, the dip might get freed before we're
8098 * done setting it up.
8100 * We transfer the initial reference to the last bio, so we
8101 * don't need to increment the reference count for the last one.
8103 if (submit_len > 0) {
8104 refcount_inc(&dip->refs);
8106 * If we are submitting more than one bio, submit them
8107 * all asynchronously. The exception is RAID 5 or 6, as
8108 * asynchronous checksums make it difficult to collect
8109 * full stripe writes.
8115 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8120 refcount_dec(&dip->refs);
8124 dio_data->submitted += clone_len;
8125 clone_offset += clone_len;
8126 start_sector += clone_len >> 9;
8127 file_offset += clone_len;
8129 free_extent_map(em);
8130 } while (submit_len > 0);
8134 free_extent_map(em);
8136 dio_bio->bi_status = status;
8137 btrfs_dio_private_put(dip);
8140 static const struct iomap_ops btrfs_dio_iomap_ops = {
8141 .iomap_begin = btrfs_dio_iomap_begin,
8142 .iomap_end = btrfs_dio_iomap_end,
8145 static const struct iomap_dio_ops btrfs_dio_ops = {
8146 .submit_io = btrfs_submit_direct,
8147 .bio_set = &btrfs_dio_bioset,
8150 ssize_t btrfs_dio_rw(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
8152 struct btrfs_dio_data data;
8154 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
8155 IOMAP_DIO_PARTIAL, &data, done_before);
8158 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8163 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8167 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8170 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8172 struct inode *inode = page->mapping->host;
8175 if (current->flags & PF_MEMALLOC) {
8176 redirty_page_for_writepage(wbc, page);
8182 * If we are under memory pressure we will call this directly from the
8183 * VM, we need to make sure we have the inode referenced for the ordered
8184 * extent. If not just return like we didn't do anything.
8186 if (!igrab(inode)) {
8187 redirty_page_for_writepage(wbc, page);
8188 return AOP_WRITEPAGE_ACTIVATE;
8190 ret = extent_write_full_page(page, wbc);
8191 btrfs_add_delayed_iput(inode);
8195 static int btrfs_writepages(struct address_space *mapping,
8196 struct writeback_control *wbc)
8198 return extent_writepages(mapping, wbc);
8201 static void btrfs_readahead(struct readahead_control *rac)
8203 extent_readahead(rac);
8207 * For release_folio() and invalidate_folio() we have a race window where
8208 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8209 * If we continue to release/invalidate the page, we could cause use-after-free
8210 * for subpage spinlock. So this function is to spin and wait for subpage
8213 static void wait_subpage_spinlock(struct page *page)
8215 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8216 struct btrfs_subpage *subpage;
8218 if (!btrfs_is_subpage(fs_info, page))
8221 ASSERT(PagePrivate(page) && page->private);
8222 subpage = (struct btrfs_subpage *)page->private;
8225 * This may look insane as we just acquire the spinlock and release it,
8226 * without doing anything. But we just want to make sure no one is
8227 * still holding the subpage spinlock.
8228 * And since the page is not dirty nor writeback, and we have page
8229 * locked, the only possible way to hold a spinlock is from the endio
8230 * function to clear page writeback.
8232 * Here we just acquire the spinlock so that all existing callers
8233 * should exit and we're safe to release/invalidate the page.
8235 spin_lock_irq(&subpage->lock);
8236 spin_unlock_irq(&subpage->lock);
8239 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8241 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8244 wait_subpage_spinlock(&folio->page);
8245 clear_page_extent_mapped(&folio->page);
8250 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8252 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8254 return __btrfs_release_folio(folio, gfp_flags);
8257 #ifdef CONFIG_MIGRATION
8258 static int btrfs_migratepage(struct address_space *mapping,
8259 struct page *newpage, struct page *page,
8260 enum migrate_mode mode)
8264 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8265 if (ret != MIGRATEPAGE_SUCCESS)
8268 if (page_has_private(page))
8269 attach_page_private(newpage, detach_page_private(page));
8271 if (PageOrdered(page)) {
8272 ClearPageOrdered(page);
8273 SetPageOrdered(newpage);
8276 if (mode != MIGRATE_SYNC_NO_COPY)
8277 migrate_page_copy(newpage, page);
8279 migrate_page_states(newpage, page);
8280 return MIGRATEPAGE_SUCCESS;
8284 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8287 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8288 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8289 struct extent_io_tree *tree = &inode->io_tree;
8290 struct extent_state *cached_state = NULL;
8291 u64 page_start = folio_pos(folio);
8292 u64 page_end = page_start + folio_size(folio) - 1;
8294 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8297 * We have folio locked so no new ordered extent can be created on this
8298 * page, nor bio can be submitted for this folio.
8300 * But already submitted bio can still be finished on this folio.
8301 * Furthermore, endio function won't skip folio which has Ordered
8302 * (Private2) already cleared, so it's possible for endio and
8303 * invalidate_folio to do the same ordered extent accounting twice
8306 * So here we wait for any submitted bios to finish, so that we won't
8307 * do double ordered extent accounting on the same folio.
8309 folio_wait_writeback(folio);
8310 wait_subpage_spinlock(&folio->page);
8313 * For subpage case, we have call sites like
8314 * btrfs_punch_hole_lock_range() which passes range not aligned to
8316 * If the range doesn't cover the full folio, we don't need to and
8317 * shouldn't clear page extent mapped, as folio->private can still
8318 * record subpage dirty bits for other part of the range.
8320 * For cases that invalidate the full folio even the range doesn't
8321 * cover the full folio, like invalidating the last folio, we're
8322 * still safe to wait for ordered extent to finish.
8324 if (!(offset == 0 && length == folio_size(folio))) {
8325 btrfs_release_folio(folio, GFP_NOFS);
8329 if (!inode_evicting)
8330 lock_extent_bits(tree, page_start, page_end, &cached_state);
8333 while (cur < page_end) {
8334 struct btrfs_ordered_extent *ordered;
8339 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8340 page_end + 1 - cur);
8342 range_end = page_end;
8344 * No ordered extent covering this range, we are safe
8345 * to delete all extent states in the range.
8347 delete_states = true;
8350 if (ordered->file_offset > cur) {
8352 * There is a range between [cur, oe->file_offset) not
8353 * covered by any ordered extent.
8354 * We are safe to delete all extent states, and handle
8355 * the ordered extent in the next iteration.
8357 range_end = ordered->file_offset - 1;
8358 delete_states = true;
8362 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8364 ASSERT(range_end + 1 - cur < U32_MAX);
8365 range_len = range_end + 1 - cur;
8366 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8368 * If Ordered (Private2) is cleared, it means endio has
8369 * already been executed for the range.
8370 * We can't delete the extent states as
8371 * btrfs_finish_ordered_io() may still use some of them.
8373 delete_states = false;
8376 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8379 * IO on this page will never be started, so we need to account
8380 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8381 * here, must leave that up for the ordered extent completion.
8383 * This will also unlock the range for incoming
8384 * btrfs_finish_ordered_io().
8386 if (!inode_evicting)
8387 clear_extent_bit(tree, cur, range_end,
8389 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8390 EXTENT_DEFRAG, 1, 0, &cached_state);
8392 spin_lock_irq(&inode->ordered_tree.lock);
8393 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8394 ordered->truncated_len = min(ordered->truncated_len,
8395 cur - ordered->file_offset);
8396 spin_unlock_irq(&inode->ordered_tree.lock);
8398 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8399 cur, range_end + 1 - cur)) {
8400 btrfs_finish_ordered_io(ordered);
8402 * The ordered extent has finished, now we're again
8403 * safe to delete all extent states of the range.
8405 delete_states = true;
8408 * btrfs_finish_ordered_io() will get executed by endio
8409 * of other pages, thus we can't delete extent states
8412 delete_states = false;
8416 btrfs_put_ordered_extent(ordered);
8418 * Qgroup reserved space handler
8419 * Sector(s) here will be either:
8421 * 1) Already written to disk or bio already finished
8422 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8423 * Qgroup will be handled by its qgroup_record then.
8424 * btrfs_qgroup_free_data() call will do nothing here.
8426 * 2) Not written to disk yet
8427 * Then btrfs_qgroup_free_data() call will clear the
8428 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8429 * reserved data space.
8430 * Since the IO will never happen for this page.
8432 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8433 if (!inode_evicting) {
8434 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8435 EXTENT_DELALLOC | EXTENT_UPTODATE |
8436 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8437 delete_states, &cached_state);
8439 cur = range_end + 1;
8442 * We have iterated through all ordered extents of the page, the page
8443 * should not have Ordered (Private2) anymore, or the above iteration
8444 * did something wrong.
8446 ASSERT(!folio_test_ordered(folio));
8447 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8448 if (!inode_evicting)
8449 __btrfs_release_folio(folio, GFP_NOFS);
8450 clear_page_extent_mapped(&folio->page);
8454 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8455 * called from a page fault handler when a page is first dirtied. Hence we must
8456 * be careful to check for EOF conditions here. We set the page up correctly
8457 * for a written page which means we get ENOSPC checking when writing into
8458 * holes and correct delalloc and unwritten extent mapping on filesystems that
8459 * support these features.
8461 * We are not allowed to take the i_mutex here so we have to play games to
8462 * protect against truncate races as the page could now be beyond EOF. Because
8463 * truncate_setsize() writes the inode size before removing pages, once we have
8464 * the page lock we can determine safely if the page is beyond EOF. If it is not
8465 * beyond EOF, then the page is guaranteed safe against truncation until we
8468 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8470 struct page *page = vmf->page;
8471 struct inode *inode = file_inode(vmf->vma->vm_file);
8472 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8473 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8474 struct btrfs_ordered_extent *ordered;
8475 struct extent_state *cached_state = NULL;
8476 struct extent_changeset *data_reserved = NULL;
8477 unsigned long zero_start;
8487 reserved_space = PAGE_SIZE;
8489 sb_start_pagefault(inode->i_sb);
8490 page_start = page_offset(page);
8491 page_end = page_start + PAGE_SIZE - 1;
8495 * Reserving delalloc space after obtaining the page lock can lead to
8496 * deadlock. For example, if a dirty page is locked by this function
8497 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8498 * dirty page write out, then the btrfs_writepage() function could
8499 * end up waiting indefinitely to get a lock on the page currently
8500 * being processed by btrfs_page_mkwrite() function.
8502 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8503 page_start, reserved_space);
8505 ret2 = file_update_time(vmf->vma->vm_file);
8509 ret = vmf_error(ret2);
8515 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8517 down_read(&BTRFS_I(inode)->i_mmap_lock);
8519 size = i_size_read(inode);
8521 if ((page->mapping != inode->i_mapping) ||
8522 (page_start >= size)) {
8523 /* page got truncated out from underneath us */
8526 wait_on_page_writeback(page);
8528 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8529 ret2 = set_page_extent_mapped(page);
8531 ret = vmf_error(ret2);
8532 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8537 * we can't set the delalloc bits if there are pending ordered
8538 * extents. Drop our locks and wait for them to finish
8540 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8543 unlock_extent_cached(io_tree, page_start, page_end,
8546 up_read(&BTRFS_I(inode)->i_mmap_lock);
8547 btrfs_start_ordered_extent(ordered, 1);
8548 btrfs_put_ordered_extent(ordered);
8552 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8553 reserved_space = round_up(size - page_start,
8554 fs_info->sectorsize);
8555 if (reserved_space < PAGE_SIZE) {
8556 end = page_start + reserved_space - 1;
8557 btrfs_delalloc_release_space(BTRFS_I(inode),
8558 data_reserved, page_start,
8559 PAGE_SIZE - reserved_space, true);
8564 * page_mkwrite gets called when the page is firstly dirtied after it's
8565 * faulted in, but write(2) could also dirty a page and set delalloc
8566 * bits, thus in this case for space account reason, we still need to
8567 * clear any delalloc bits within this page range since we have to
8568 * reserve data&meta space before lock_page() (see above comments).
8570 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8571 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8572 EXTENT_DEFRAG, 0, 0, &cached_state);
8574 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8577 unlock_extent_cached(io_tree, page_start, page_end,
8579 ret = VM_FAULT_SIGBUS;
8583 /* page is wholly or partially inside EOF */
8584 if (page_start + PAGE_SIZE > size)
8585 zero_start = offset_in_page(size);
8587 zero_start = PAGE_SIZE;
8589 if (zero_start != PAGE_SIZE) {
8590 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8591 flush_dcache_page(page);
8593 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8594 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8595 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8597 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8599 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8600 up_read(&BTRFS_I(inode)->i_mmap_lock);
8602 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8603 sb_end_pagefault(inode->i_sb);
8604 extent_changeset_free(data_reserved);
8605 return VM_FAULT_LOCKED;
8609 up_read(&BTRFS_I(inode)->i_mmap_lock);
8611 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8612 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8613 reserved_space, (ret != 0));
8615 sb_end_pagefault(inode->i_sb);
8616 extent_changeset_free(data_reserved);
8620 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8622 struct btrfs_truncate_control control = {
8623 .inode = BTRFS_I(inode),
8624 .ino = btrfs_ino(BTRFS_I(inode)),
8625 .min_type = BTRFS_EXTENT_DATA_KEY,
8626 .clear_extent_range = true,
8628 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8629 struct btrfs_root *root = BTRFS_I(inode)->root;
8630 struct btrfs_block_rsv *rsv;
8632 struct btrfs_trans_handle *trans;
8633 u64 mask = fs_info->sectorsize - 1;
8634 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8636 if (!skip_writeback) {
8637 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8644 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8645 * things going on here:
8647 * 1) We need to reserve space to update our inode.
8649 * 2) We need to have something to cache all the space that is going to
8650 * be free'd up by the truncate operation, but also have some slack
8651 * space reserved in case it uses space during the truncate (thank you
8652 * very much snapshotting).
8654 * And we need these to be separate. The fact is we can use a lot of
8655 * space doing the truncate, and we have no earthly idea how much space
8656 * we will use, so we need the truncate reservation to be separate so it
8657 * doesn't end up using space reserved for updating the inode. We also
8658 * need to be able to stop the transaction and start a new one, which
8659 * means we need to be able to update the inode several times, and we
8660 * have no idea of knowing how many times that will be, so we can't just
8661 * reserve 1 item for the entirety of the operation, so that has to be
8662 * done separately as well.
8664 * So that leaves us with
8666 * 1) rsv - for the truncate reservation, which we will steal from the
8667 * transaction reservation.
8668 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8669 * updating the inode.
8671 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8674 rsv->size = min_size;
8678 * 1 for the truncate slack space
8679 * 1 for updating the inode.
8681 trans = btrfs_start_transaction(root, 2);
8682 if (IS_ERR(trans)) {
8683 ret = PTR_ERR(trans);
8687 /* Migrate the slack space for the truncate to our reserve */
8688 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8692 trans->block_rsv = rsv;
8695 struct extent_state *cached_state = NULL;
8696 const u64 new_size = inode->i_size;
8697 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8699 control.new_size = new_size;
8700 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8703 * We want to drop from the next block forward in case this new
8704 * size is not block aligned since we will be keeping the last
8705 * block of the extent just the way it is.
8707 btrfs_drop_extent_cache(BTRFS_I(inode),
8708 ALIGN(new_size, fs_info->sectorsize),
8711 ret = btrfs_truncate_inode_items(trans, root, &control);
8713 inode_sub_bytes(inode, control.sub_bytes);
8714 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8716 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8717 (u64)-1, &cached_state);
8719 trans->block_rsv = &fs_info->trans_block_rsv;
8720 if (ret != -ENOSPC && ret != -EAGAIN)
8723 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8727 btrfs_end_transaction(trans);
8728 btrfs_btree_balance_dirty(fs_info);
8730 trans = btrfs_start_transaction(root, 2);
8731 if (IS_ERR(trans)) {
8732 ret = PTR_ERR(trans);
8737 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8738 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8739 rsv, min_size, false);
8740 BUG_ON(ret); /* shouldn't happen */
8741 trans->block_rsv = rsv;
8745 * We can't call btrfs_truncate_block inside a trans handle as we could
8746 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8747 * know we've truncated everything except the last little bit, and can
8748 * do btrfs_truncate_block and then update the disk_i_size.
8750 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8751 btrfs_end_transaction(trans);
8752 btrfs_btree_balance_dirty(fs_info);
8754 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8757 trans = btrfs_start_transaction(root, 1);
8758 if (IS_ERR(trans)) {
8759 ret = PTR_ERR(trans);
8762 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8768 trans->block_rsv = &fs_info->trans_block_rsv;
8769 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8773 ret2 = btrfs_end_transaction(trans);
8776 btrfs_btree_balance_dirty(fs_info);
8779 btrfs_free_block_rsv(fs_info, rsv);
8781 * So if we truncate and then write and fsync we normally would just
8782 * write the extents that changed, which is a problem if we need to
8783 * first truncate that entire inode. So set this flag so we write out
8784 * all of the extents in the inode to the sync log so we're completely
8787 * If no extents were dropped or trimmed we don't need to force the next
8788 * fsync to truncate all the inode's items from the log and re-log them
8789 * all. This means the truncate operation did not change the file size,
8790 * or changed it to a smaller size but there was only an implicit hole
8791 * between the old i_size and the new i_size, and there were no prealloc
8792 * extents beyond i_size to drop.
8794 if (control.extents_found > 0)
8795 btrfs_set_inode_full_sync(BTRFS_I(inode));
8800 struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns,
8803 struct inode *inode;
8805 inode = new_inode(dir->i_sb);
8808 * Subvolumes don't inherit the sgid bit or the parent's gid if
8809 * the parent's sgid bit is set. This is probably a bug.
8811 inode_init_owner(mnt_userns, inode, NULL,
8812 S_IFDIR | (~current_umask() & S_IRWXUGO));
8813 inode->i_op = &btrfs_dir_inode_operations;
8814 inode->i_fop = &btrfs_dir_file_operations;
8819 struct inode *btrfs_alloc_inode(struct super_block *sb)
8821 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8822 struct btrfs_inode *ei;
8823 struct inode *inode;
8825 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8832 ei->last_sub_trans = 0;
8833 ei->logged_trans = 0;
8834 ei->delalloc_bytes = 0;
8835 ei->new_delalloc_bytes = 0;
8836 ei->defrag_bytes = 0;
8837 ei->disk_i_size = 0;
8841 ei->index_cnt = (u64)-1;
8843 ei->last_unlink_trans = 0;
8844 ei->last_reflink_trans = 0;
8845 ei->last_log_commit = 0;
8847 spin_lock_init(&ei->lock);
8848 ei->outstanding_extents = 0;
8849 if (sb->s_magic != BTRFS_TEST_MAGIC)
8850 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8851 BTRFS_BLOCK_RSV_DELALLOC);
8852 ei->runtime_flags = 0;
8853 ei->prop_compress = BTRFS_COMPRESS_NONE;
8854 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8856 ei->delayed_node = NULL;
8858 ei->i_otime.tv_sec = 0;
8859 ei->i_otime.tv_nsec = 0;
8861 inode = &ei->vfs_inode;
8862 extent_map_tree_init(&ei->extent_tree);
8863 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8864 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8865 IO_TREE_INODE_IO_FAILURE, inode);
8866 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8867 IO_TREE_INODE_FILE_EXTENT, inode);
8868 ei->io_tree.track_uptodate = true;
8869 ei->io_failure_tree.track_uptodate = true;
8870 atomic_set(&ei->sync_writers, 0);
8871 mutex_init(&ei->log_mutex);
8872 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8873 INIT_LIST_HEAD(&ei->delalloc_inodes);
8874 INIT_LIST_HEAD(&ei->delayed_iput);
8875 RB_CLEAR_NODE(&ei->rb_node);
8876 init_rwsem(&ei->i_mmap_lock);
8881 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8882 void btrfs_test_destroy_inode(struct inode *inode)
8884 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8885 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8889 void btrfs_free_inode(struct inode *inode)
8891 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8894 void btrfs_destroy_inode(struct inode *vfs_inode)
8896 struct btrfs_ordered_extent *ordered;
8897 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8898 struct btrfs_root *root = inode->root;
8900 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8901 WARN_ON(vfs_inode->i_data.nrpages);
8902 WARN_ON(inode->block_rsv.reserved);
8903 WARN_ON(inode->block_rsv.size);
8904 WARN_ON(inode->outstanding_extents);
8905 if (!S_ISDIR(vfs_inode->i_mode)) {
8906 WARN_ON(inode->delalloc_bytes);
8907 WARN_ON(inode->new_delalloc_bytes);
8909 WARN_ON(inode->csum_bytes);
8910 WARN_ON(inode->defrag_bytes);
8913 * This can happen where we create an inode, but somebody else also
8914 * created the same inode and we need to destroy the one we already
8921 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8925 btrfs_err(root->fs_info,
8926 "found ordered extent %llu %llu on inode cleanup",
8927 ordered->file_offset, ordered->num_bytes);
8928 btrfs_remove_ordered_extent(inode, ordered);
8929 btrfs_put_ordered_extent(ordered);
8930 btrfs_put_ordered_extent(ordered);
8933 btrfs_qgroup_check_reserved_leak(inode);
8934 inode_tree_del(inode);
8935 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8936 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8937 btrfs_put_root(inode->root);
8940 int btrfs_drop_inode(struct inode *inode)
8942 struct btrfs_root *root = BTRFS_I(inode)->root;
8947 /* the snap/subvol tree is on deleting */
8948 if (btrfs_root_refs(&root->root_item) == 0)
8951 return generic_drop_inode(inode);
8954 static void init_once(void *foo)
8956 struct btrfs_inode *ei = foo;
8958 inode_init_once(&ei->vfs_inode);
8961 void __cold btrfs_destroy_cachep(void)
8964 * Make sure all delayed rcu free inodes are flushed before we
8968 bioset_exit(&btrfs_dio_bioset);
8969 kmem_cache_destroy(btrfs_inode_cachep);
8970 kmem_cache_destroy(btrfs_trans_handle_cachep);
8971 kmem_cache_destroy(btrfs_path_cachep);
8972 kmem_cache_destroy(btrfs_free_space_cachep);
8973 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8976 int __init btrfs_init_cachep(void)
8978 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8979 sizeof(struct btrfs_inode), 0,
8980 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8982 if (!btrfs_inode_cachep)
8985 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8986 sizeof(struct btrfs_trans_handle), 0,
8987 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8988 if (!btrfs_trans_handle_cachep)
8991 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8992 sizeof(struct btrfs_path), 0,
8993 SLAB_MEM_SPREAD, NULL);
8994 if (!btrfs_path_cachep)
8997 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8998 sizeof(struct btrfs_free_space), 0,
8999 SLAB_MEM_SPREAD, NULL);
9000 if (!btrfs_free_space_cachep)
9003 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9004 PAGE_SIZE, PAGE_SIZE,
9005 SLAB_MEM_SPREAD, NULL);
9006 if (!btrfs_free_space_bitmap_cachep)
9009 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
9010 offsetof(struct btrfs_dio_private, bio),
9016 btrfs_destroy_cachep();
9020 static int btrfs_getattr(struct user_namespace *mnt_userns,
9021 const struct path *path, struct kstat *stat,
9022 u32 request_mask, unsigned int flags)
9026 struct inode *inode = d_inode(path->dentry);
9027 u32 blocksize = inode->i_sb->s_blocksize;
9028 u32 bi_flags = BTRFS_I(inode)->flags;
9029 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9031 stat->result_mask |= STATX_BTIME;
9032 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9033 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9034 if (bi_flags & BTRFS_INODE_APPEND)
9035 stat->attributes |= STATX_ATTR_APPEND;
9036 if (bi_flags & BTRFS_INODE_COMPRESS)
9037 stat->attributes |= STATX_ATTR_COMPRESSED;
9038 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9039 stat->attributes |= STATX_ATTR_IMMUTABLE;
9040 if (bi_flags & BTRFS_INODE_NODUMP)
9041 stat->attributes |= STATX_ATTR_NODUMP;
9042 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9043 stat->attributes |= STATX_ATTR_VERITY;
9045 stat->attributes_mask |= (STATX_ATTR_APPEND |
9046 STATX_ATTR_COMPRESSED |
9047 STATX_ATTR_IMMUTABLE |
9050 generic_fillattr(mnt_userns, inode, stat);
9051 stat->dev = BTRFS_I(inode)->root->anon_dev;
9053 spin_lock(&BTRFS_I(inode)->lock);
9054 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9055 inode_bytes = inode_get_bytes(inode);
9056 spin_unlock(&BTRFS_I(inode)->lock);
9057 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9058 ALIGN(delalloc_bytes, blocksize)) >> 9;
9062 static int btrfs_rename_exchange(struct inode *old_dir,
9063 struct dentry *old_dentry,
9064 struct inode *new_dir,
9065 struct dentry *new_dentry)
9067 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9068 struct btrfs_trans_handle *trans;
9069 unsigned int trans_num_items;
9070 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9071 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9072 struct inode *new_inode = new_dentry->d_inode;
9073 struct inode *old_inode = old_dentry->d_inode;
9074 struct timespec64 ctime = current_time(old_inode);
9075 struct btrfs_rename_ctx old_rename_ctx;
9076 struct btrfs_rename_ctx new_rename_ctx;
9077 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9078 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9083 bool need_abort = false;
9086 * For non-subvolumes allow exchange only within one subvolume, in the
9087 * same inode namespace. Two subvolumes (represented as directory) can
9088 * be exchanged as they're a logical link and have a fixed inode number.
9091 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9092 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9095 /* close the race window with snapshot create/destroy ioctl */
9096 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9097 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9098 down_read(&fs_info->subvol_sem);
9102 * 1 to remove old dir item
9103 * 1 to remove old dir index
9104 * 1 to add new dir item
9105 * 1 to add new dir index
9106 * 1 to update parent inode
9108 * If the parents are the same, we only need to account for one
9110 trans_num_items = (old_dir == new_dir ? 9 : 10);
9111 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9113 * 1 to remove old root ref
9114 * 1 to remove old root backref
9115 * 1 to add new root ref
9116 * 1 to add new root backref
9118 trans_num_items += 4;
9121 * 1 to update inode item
9122 * 1 to remove old inode ref
9123 * 1 to add new inode ref
9125 trans_num_items += 3;
9127 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9128 trans_num_items += 4;
9130 trans_num_items += 3;
9131 trans = btrfs_start_transaction(root, trans_num_items);
9132 if (IS_ERR(trans)) {
9133 ret = PTR_ERR(trans);
9138 ret = btrfs_record_root_in_trans(trans, dest);
9144 * We need to find a free sequence number both in the source and
9145 * in the destination directory for the exchange.
9147 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9150 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9154 BTRFS_I(old_inode)->dir_index = 0ULL;
9155 BTRFS_I(new_inode)->dir_index = 0ULL;
9157 /* Reference for the source. */
9158 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9159 /* force full log commit if subvolume involved. */
9160 btrfs_set_log_full_commit(trans);
9162 ret = btrfs_insert_inode_ref(trans, dest,
9163 new_dentry->d_name.name,
9164 new_dentry->d_name.len,
9166 btrfs_ino(BTRFS_I(new_dir)),
9173 /* And now for the dest. */
9174 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9175 /* force full log commit if subvolume involved. */
9176 btrfs_set_log_full_commit(trans);
9178 ret = btrfs_insert_inode_ref(trans, root,
9179 old_dentry->d_name.name,
9180 old_dentry->d_name.len,
9182 btrfs_ino(BTRFS_I(old_dir)),
9186 btrfs_abort_transaction(trans, ret);
9191 /* Update inode version and ctime/mtime. */
9192 inode_inc_iversion(old_dir);
9193 inode_inc_iversion(new_dir);
9194 inode_inc_iversion(old_inode);
9195 inode_inc_iversion(new_inode);
9196 old_dir->i_ctime = old_dir->i_mtime = ctime;
9197 new_dir->i_ctime = new_dir->i_mtime = ctime;
9198 old_inode->i_ctime = ctime;
9199 new_inode->i_ctime = ctime;
9201 if (old_dentry->d_parent != new_dentry->d_parent) {
9202 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9203 BTRFS_I(old_inode), 1);
9204 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9205 BTRFS_I(new_inode), 1);
9208 /* src is a subvolume */
9209 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9210 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9211 } else { /* src is an inode */
9212 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9213 BTRFS_I(old_dentry->d_inode),
9214 old_dentry->d_name.name,
9215 old_dentry->d_name.len,
9218 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9221 btrfs_abort_transaction(trans, ret);
9225 /* dest is a subvolume */
9226 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9227 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9228 } else { /* dest is an inode */
9229 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9230 BTRFS_I(new_dentry->d_inode),
9231 new_dentry->d_name.name,
9232 new_dentry->d_name.len,
9235 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9238 btrfs_abort_transaction(trans, ret);
9242 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9243 new_dentry->d_name.name,
9244 new_dentry->d_name.len, 0, old_idx);
9246 btrfs_abort_transaction(trans, ret);
9250 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9251 old_dentry->d_name.name,
9252 old_dentry->d_name.len, 0, new_idx);
9254 btrfs_abort_transaction(trans, ret);
9258 if (old_inode->i_nlink == 1)
9259 BTRFS_I(old_inode)->dir_index = old_idx;
9260 if (new_inode->i_nlink == 1)
9261 BTRFS_I(new_inode)->dir_index = new_idx;
9264 * Now pin the logs of the roots. We do it to ensure that no other task
9265 * can sync the logs while we are in progress with the rename, because
9266 * that could result in an inconsistency in case any of the inodes that
9267 * are part of this rename operation were logged before.
9269 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9270 btrfs_pin_log_trans(root);
9271 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9272 btrfs_pin_log_trans(dest);
9274 /* Do the log updates for all inodes. */
9275 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9276 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9277 old_rename_ctx.index, new_dentry->d_parent);
9278 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9279 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9280 new_rename_ctx.index, old_dentry->d_parent);
9282 /* Now unpin the logs. */
9283 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9284 btrfs_end_log_trans(root);
9285 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9286 btrfs_end_log_trans(dest);
9288 ret2 = btrfs_end_transaction(trans);
9289 ret = ret ? ret : ret2;
9291 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9292 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9293 up_read(&fs_info->subvol_sem);
9298 static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns,
9301 struct inode *inode;
9303 inode = new_inode(dir->i_sb);
9305 inode_init_owner(mnt_userns, inode, dir,
9306 S_IFCHR | WHITEOUT_MODE);
9307 inode->i_op = &btrfs_special_inode_operations;
9308 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9313 static int btrfs_rename(struct user_namespace *mnt_userns,
9314 struct inode *old_dir, struct dentry *old_dentry,
9315 struct inode *new_dir, struct dentry *new_dentry,
9318 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9319 struct btrfs_new_inode_args whiteout_args = {
9321 .dentry = old_dentry,
9323 struct btrfs_trans_handle *trans;
9324 unsigned int trans_num_items;
9325 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9326 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9327 struct inode *new_inode = d_inode(new_dentry);
9328 struct inode *old_inode = d_inode(old_dentry);
9329 struct btrfs_rename_ctx rename_ctx;
9333 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9335 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9338 /* we only allow rename subvolume link between subvolumes */
9339 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9342 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9343 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9346 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9347 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9351 /* check for collisions, even if the name isn't there */
9352 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9353 new_dentry->d_name.name,
9354 new_dentry->d_name.len);
9357 if (ret == -EEXIST) {
9359 * eexist without a new_inode */
9360 if (WARN_ON(!new_inode)) {
9364 /* maybe -EOVERFLOW */
9371 * we're using rename to replace one file with another. Start IO on it
9372 * now so we don't add too much work to the end of the transaction
9374 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9375 filemap_flush(old_inode->i_mapping);
9377 if (flags & RENAME_WHITEOUT) {
9378 whiteout_args.inode = new_whiteout_inode(mnt_userns, old_dir);
9379 if (!whiteout_args.inode)
9381 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9383 goto out_whiteout_inode;
9385 /* 1 to update the old parent inode. */
9386 trans_num_items = 1;
9389 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9390 /* Close the race window with snapshot create/destroy ioctl */
9391 down_read(&fs_info->subvol_sem);
9393 * 1 to remove old root ref
9394 * 1 to remove old root backref
9395 * 1 to add new root ref
9396 * 1 to add new root backref
9398 trans_num_items += 4;
9402 * 1 to remove old inode ref
9403 * 1 to add new inode ref
9405 trans_num_items += 3;
9408 * 1 to remove old dir item
9409 * 1 to remove old dir index
9410 * 1 to add new dir item
9411 * 1 to add new dir index
9413 trans_num_items += 4;
9414 /* 1 to update new parent inode if it's not the same as the old parent */
9415 if (new_dir != old_dir)
9420 * 1 to remove inode ref
9421 * 1 to remove dir item
9422 * 1 to remove dir index
9423 * 1 to possibly add orphan item
9425 trans_num_items += 5;
9427 trans = btrfs_start_transaction(root, trans_num_items);
9428 if (IS_ERR(trans)) {
9429 ret = PTR_ERR(trans);
9434 ret = btrfs_record_root_in_trans(trans, dest);
9439 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9443 BTRFS_I(old_inode)->dir_index = 0ULL;
9444 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9445 /* force full log commit if subvolume involved. */
9446 btrfs_set_log_full_commit(trans);
9448 ret = btrfs_insert_inode_ref(trans, dest,
9449 new_dentry->d_name.name,
9450 new_dentry->d_name.len,
9452 btrfs_ino(BTRFS_I(new_dir)), index);
9457 inode_inc_iversion(old_dir);
9458 inode_inc_iversion(new_dir);
9459 inode_inc_iversion(old_inode);
9460 old_dir->i_ctime = old_dir->i_mtime =
9461 new_dir->i_ctime = new_dir->i_mtime =
9462 old_inode->i_ctime = current_time(old_dir);
9464 if (old_dentry->d_parent != new_dentry->d_parent)
9465 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9466 BTRFS_I(old_inode), 1);
9468 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9469 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9471 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9472 BTRFS_I(d_inode(old_dentry)),
9473 old_dentry->d_name.name,
9474 old_dentry->d_name.len,
9477 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9480 btrfs_abort_transaction(trans, ret);
9485 inode_inc_iversion(new_inode);
9486 new_inode->i_ctime = current_time(new_inode);
9487 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9488 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9489 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9490 BUG_ON(new_inode->i_nlink == 0);
9492 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9493 BTRFS_I(d_inode(new_dentry)),
9494 new_dentry->d_name.name,
9495 new_dentry->d_name.len);
9497 if (!ret && new_inode->i_nlink == 0)
9498 ret = btrfs_orphan_add(trans,
9499 BTRFS_I(d_inode(new_dentry)));
9501 btrfs_abort_transaction(trans, ret);
9506 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9507 new_dentry->d_name.name,
9508 new_dentry->d_name.len, 0, index);
9510 btrfs_abort_transaction(trans, ret);
9514 if (old_inode->i_nlink == 1)
9515 BTRFS_I(old_inode)->dir_index = index;
9517 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9518 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9519 rename_ctx.index, new_dentry->d_parent);
9521 if (flags & RENAME_WHITEOUT) {
9522 ret = btrfs_create_new_inode(trans, &whiteout_args);
9524 btrfs_abort_transaction(trans, ret);
9527 unlock_new_inode(whiteout_args.inode);
9528 iput(whiteout_args.inode);
9529 whiteout_args.inode = NULL;
9533 ret2 = btrfs_end_transaction(trans);
9534 ret = ret ? ret : ret2;
9536 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9537 up_read(&fs_info->subvol_sem);
9538 if (flags & RENAME_WHITEOUT)
9539 btrfs_new_inode_args_destroy(&whiteout_args);
9541 if (flags & RENAME_WHITEOUT)
9542 iput(whiteout_args.inode);
9546 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9547 struct dentry *old_dentry, struct inode *new_dir,
9548 struct dentry *new_dentry, unsigned int flags)
9550 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9553 if (flags & RENAME_EXCHANGE)
9554 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9557 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9561 struct btrfs_delalloc_work {
9562 struct inode *inode;
9563 struct completion completion;
9564 struct list_head list;
9565 struct btrfs_work work;
9568 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9570 struct btrfs_delalloc_work *delalloc_work;
9571 struct inode *inode;
9573 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9575 inode = delalloc_work->inode;
9576 filemap_flush(inode->i_mapping);
9577 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9578 &BTRFS_I(inode)->runtime_flags))
9579 filemap_flush(inode->i_mapping);
9582 complete(&delalloc_work->completion);
9585 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9587 struct btrfs_delalloc_work *work;
9589 work = kmalloc(sizeof(*work), GFP_NOFS);
9593 init_completion(&work->completion);
9594 INIT_LIST_HEAD(&work->list);
9595 work->inode = inode;
9596 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9602 * some fairly slow code that needs optimization. This walks the list
9603 * of all the inodes with pending delalloc and forces them to disk.
9605 static int start_delalloc_inodes(struct btrfs_root *root,
9606 struct writeback_control *wbc, bool snapshot,
9607 bool in_reclaim_context)
9609 struct btrfs_inode *binode;
9610 struct inode *inode;
9611 struct btrfs_delalloc_work *work, *next;
9612 struct list_head works;
9613 struct list_head splice;
9615 bool full_flush = wbc->nr_to_write == LONG_MAX;
9617 INIT_LIST_HEAD(&works);
9618 INIT_LIST_HEAD(&splice);
9620 mutex_lock(&root->delalloc_mutex);
9621 spin_lock(&root->delalloc_lock);
9622 list_splice_init(&root->delalloc_inodes, &splice);
9623 while (!list_empty(&splice)) {
9624 binode = list_entry(splice.next, struct btrfs_inode,
9627 list_move_tail(&binode->delalloc_inodes,
9628 &root->delalloc_inodes);
9630 if (in_reclaim_context &&
9631 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9634 inode = igrab(&binode->vfs_inode);
9636 cond_resched_lock(&root->delalloc_lock);
9639 spin_unlock(&root->delalloc_lock);
9642 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9643 &binode->runtime_flags);
9645 work = btrfs_alloc_delalloc_work(inode);
9651 list_add_tail(&work->list, &works);
9652 btrfs_queue_work(root->fs_info->flush_workers,
9655 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9656 btrfs_add_delayed_iput(inode);
9657 if (ret || wbc->nr_to_write <= 0)
9661 spin_lock(&root->delalloc_lock);
9663 spin_unlock(&root->delalloc_lock);
9666 list_for_each_entry_safe(work, next, &works, list) {
9667 list_del_init(&work->list);
9668 wait_for_completion(&work->completion);
9672 if (!list_empty(&splice)) {
9673 spin_lock(&root->delalloc_lock);
9674 list_splice_tail(&splice, &root->delalloc_inodes);
9675 spin_unlock(&root->delalloc_lock);
9677 mutex_unlock(&root->delalloc_mutex);
9681 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9683 struct writeback_control wbc = {
9684 .nr_to_write = LONG_MAX,
9685 .sync_mode = WB_SYNC_NONE,
9687 .range_end = LLONG_MAX,
9689 struct btrfs_fs_info *fs_info = root->fs_info;
9691 if (BTRFS_FS_ERROR(fs_info))
9694 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9697 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9698 bool in_reclaim_context)
9700 struct writeback_control wbc = {
9702 .sync_mode = WB_SYNC_NONE,
9704 .range_end = LLONG_MAX,
9706 struct btrfs_root *root;
9707 struct list_head splice;
9710 if (BTRFS_FS_ERROR(fs_info))
9713 INIT_LIST_HEAD(&splice);
9715 mutex_lock(&fs_info->delalloc_root_mutex);
9716 spin_lock(&fs_info->delalloc_root_lock);
9717 list_splice_init(&fs_info->delalloc_roots, &splice);
9718 while (!list_empty(&splice)) {
9720 * Reset nr_to_write here so we know that we're doing a full
9724 wbc.nr_to_write = LONG_MAX;
9726 root = list_first_entry(&splice, struct btrfs_root,
9728 root = btrfs_grab_root(root);
9730 list_move_tail(&root->delalloc_root,
9731 &fs_info->delalloc_roots);
9732 spin_unlock(&fs_info->delalloc_root_lock);
9734 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9735 btrfs_put_root(root);
9736 if (ret < 0 || wbc.nr_to_write <= 0)
9738 spin_lock(&fs_info->delalloc_root_lock);
9740 spin_unlock(&fs_info->delalloc_root_lock);
9744 if (!list_empty(&splice)) {
9745 spin_lock(&fs_info->delalloc_root_lock);
9746 list_splice_tail(&splice, &fs_info->delalloc_roots);
9747 spin_unlock(&fs_info->delalloc_root_lock);
9749 mutex_unlock(&fs_info->delalloc_root_mutex);
9753 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9754 struct dentry *dentry, const char *symname)
9756 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9757 struct btrfs_trans_handle *trans;
9758 struct btrfs_root *root = BTRFS_I(dir)->root;
9759 struct btrfs_path *path;
9760 struct btrfs_key key;
9761 struct inode *inode;
9762 struct btrfs_new_inode_args new_inode_args = {
9766 unsigned int trans_num_items;
9771 struct btrfs_file_extent_item *ei;
9772 struct extent_buffer *leaf;
9774 name_len = strlen(symname);
9775 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9776 return -ENAMETOOLONG;
9778 inode = new_inode(dir->i_sb);
9781 inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO);
9782 inode->i_op = &btrfs_symlink_inode_operations;
9783 inode_nohighmem(inode);
9784 inode->i_mapping->a_ops = &btrfs_aops;
9785 btrfs_i_size_write(BTRFS_I(inode), name_len);
9786 inode_set_bytes(inode, name_len);
9788 new_inode_args.inode = inode;
9789 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9792 /* 1 additional item for the inline extent */
9795 trans = btrfs_start_transaction(root, trans_num_items);
9796 if (IS_ERR(trans)) {
9797 err = PTR_ERR(trans);
9798 goto out_new_inode_args;
9801 err = btrfs_create_new_inode(trans, &new_inode_args);
9805 path = btrfs_alloc_path();
9808 btrfs_abort_transaction(trans, err);
9809 discard_new_inode(inode);
9813 key.objectid = btrfs_ino(BTRFS_I(inode));
9815 key.type = BTRFS_EXTENT_DATA_KEY;
9816 datasize = btrfs_file_extent_calc_inline_size(name_len);
9817 err = btrfs_insert_empty_item(trans, root, path, &key,
9820 btrfs_abort_transaction(trans, err);
9821 btrfs_free_path(path);
9822 discard_new_inode(inode);
9826 leaf = path->nodes[0];
9827 ei = btrfs_item_ptr(leaf, path->slots[0],
9828 struct btrfs_file_extent_item);
9829 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9830 btrfs_set_file_extent_type(leaf, ei,
9831 BTRFS_FILE_EXTENT_INLINE);
9832 btrfs_set_file_extent_encryption(leaf, ei, 0);
9833 btrfs_set_file_extent_compression(leaf, ei, 0);
9834 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9835 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9837 ptr = btrfs_file_extent_inline_start(ei);
9838 write_extent_buffer(leaf, symname, ptr, name_len);
9839 btrfs_mark_buffer_dirty(leaf);
9840 btrfs_free_path(path);
9842 d_instantiate_new(dentry, inode);
9845 btrfs_end_transaction(trans);
9846 btrfs_btree_balance_dirty(fs_info);
9848 btrfs_new_inode_args_destroy(&new_inode_args);
9855 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9856 struct btrfs_trans_handle *trans_in,
9857 struct btrfs_inode *inode,
9858 struct btrfs_key *ins,
9861 struct btrfs_file_extent_item stack_fi;
9862 struct btrfs_replace_extent_info extent_info;
9863 struct btrfs_trans_handle *trans = trans_in;
9864 struct btrfs_path *path;
9865 u64 start = ins->objectid;
9866 u64 len = ins->offset;
9867 int qgroup_released;
9870 memset(&stack_fi, 0, sizeof(stack_fi));
9872 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9873 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9874 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9875 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9876 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9877 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9878 /* Encryption and other encoding is reserved and all 0 */
9880 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9881 if (qgroup_released < 0)
9882 return ERR_PTR(qgroup_released);
9885 ret = insert_reserved_file_extent(trans, inode,
9886 file_offset, &stack_fi,
9887 true, qgroup_released);
9893 extent_info.disk_offset = start;
9894 extent_info.disk_len = len;
9895 extent_info.data_offset = 0;
9896 extent_info.data_len = len;
9897 extent_info.file_offset = file_offset;
9898 extent_info.extent_buf = (char *)&stack_fi;
9899 extent_info.is_new_extent = true;
9900 extent_info.qgroup_reserved = qgroup_released;
9901 extent_info.insertions = 0;
9903 path = btrfs_alloc_path();
9909 ret = btrfs_replace_file_extents(inode, path, file_offset,
9910 file_offset + len - 1, &extent_info,
9912 btrfs_free_path(path);
9919 * We have released qgroup data range at the beginning of the function,
9920 * and normally qgroup_released bytes will be freed when committing
9922 * But if we error out early, we have to free what we have released
9923 * or we leak qgroup data reservation.
9925 btrfs_qgroup_free_refroot(inode->root->fs_info,
9926 inode->root->root_key.objectid, qgroup_released,
9927 BTRFS_QGROUP_RSV_DATA);
9928 return ERR_PTR(ret);
9931 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9932 u64 start, u64 num_bytes, u64 min_size,
9933 loff_t actual_len, u64 *alloc_hint,
9934 struct btrfs_trans_handle *trans)
9936 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9937 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9938 struct extent_map *em;
9939 struct btrfs_root *root = BTRFS_I(inode)->root;
9940 struct btrfs_key ins;
9941 u64 cur_offset = start;
9942 u64 clear_offset = start;
9945 u64 last_alloc = (u64)-1;
9947 bool own_trans = true;
9948 u64 end = start + num_bytes - 1;
9952 while (num_bytes > 0) {
9953 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9954 cur_bytes = max(cur_bytes, min_size);
9956 * If we are severely fragmented we could end up with really
9957 * small allocations, so if the allocator is returning small
9958 * chunks lets make its job easier by only searching for those
9961 cur_bytes = min(cur_bytes, last_alloc);
9962 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9963 min_size, 0, *alloc_hint, &ins, 1, 0);
9968 * We've reserved this space, and thus converted it from
9969 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9970 * from here on out we will only need to clear our reservation
9971 * for the remaining unreserved area, so advance our
9972 * clear_offset by our extent size.
9974 clear_offset += ins.offset;
9976 last_alloc = ins.offset;
9977 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9980 * Now that we inserted the prealloc extent we can finally
9981 * decrement the number of reservations in the block group.
9982 * If we did it before, we could race with relocation and have
9983 * relocation miss the reserved extent, making it fail later.
9985 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9986 if (IS_ERR(trans)) {
9987 ret = PTR_ERR(trans);
9988 btrfs_free_reserved_extent(fs_info, ins.objectid,
9993 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9994 cur_offset + ins.offset -1, 0);
9996 em = alloc_extent_map();
9998 btrfs_set_inode_full_sync(BTRFS_I(inode));
10002 em->start = cur_offset;
10003 em->orig_start = cur_offset;
10004 em->len = ins.offset;
10005 em->block_start = ins.objectid;
10006 em->block_len = ins.offset;
10007 em->orig_block_len = ins.offset;
10008 em->ram_bytes = ins.offset;
10009 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10010 em->generation = trans->transid;
10013 write_lock(&em_tree->lock);
10014 ret = add_extent_mapping(em_tree, em, 1);
10015 write_unlock(&em_tree->lock);
10016 if (ret != -EEXIST)
10018 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10019 cur_offset + ins.offset - 1,
10022 free_extent_map(em);
10024 num_bytes -= ins.offset;
10025 cur_offset += ins.offset;
10026 *alloc_hint = ins.objectid + ins.offset;
10028 inode_inc_iversion(inode);
10029 inode->i_ctime = current_time(inode);
10030 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10031 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10032 (actual_len > inode->i_size) &&
10033 (cur_offset > inode->i_size)) {
10034 if (cur_offset > actual_len)
10035 i_size = actual_len;
10037 i_size = cur_offset;
10038 i_size_write(inode, i_size);
10039 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10042 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10045 btrfs_abort_transaction(trans, ret);
10047 btrfs_end_transaction(trans);
10052 btrfs_end_transaction(trans);
10056 if (clear_offset < end)
10057 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10058 end - clear_offset + 1);
10062 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10063 u64 start, u64 num_bytes, u64 min_size,
10064 loff_t actual_len, u64 *alloc_hint)
10066 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10067 min_size, actual_len, alloc_hint,
10071 int btrfs_prealloc_file_range_trans(struct inode *inode,
10072 struct btrfs_trans_handle *trans, int mode,
10073 u64 start, u64 num_bytes, u64 min_size,
10074 loff_t actual_len, u64 *alloc_hint)
10076 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10077 min_size, actual_len, alloc_hint, trans);
10080 static int btrfs_permission(struct user_namespace *mnt_userns,
10081 struct inode *inode, int mask)
10083 struct btrfs_root *root = BTRFS_I(inode)->root;
10084 umode_t mode = inode->i_mode;
10086 if (mask & MAY_WRITE &&
10087 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10088 if (btrfs_root_readonly(root))
10090 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10093 return generic_permission(mnt_userns, inode, mask);
10096 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10097 struct dentry *dentry, umode_t mode)
10099 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10100 struct btrfs_trans_handle *trans;
10101 struct btrfs_root *root = BTRFS_I(dir)->root;
10102 struct inode *inode;
10103 struct btrfs_new_inode_args new_inode_args = {
10108 unsigned int trans_num_items;
10111 inode = new_inode(dir->i_sb);
10114 inode_init_owner(mnt_userns, inode, dir, mode);
10115 inode->i_fop = &btrfs_file_operations;
10116 inode->i_op = &btrfs_file_inode_operations;
10117 inode->i_mapping->a_ops = &btrfs_aops;
10119 new_inode_args.inode = inode;
10120 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
10124 trans = btrfs_start_transaction(root, trans_num_items);
10125 if (IS_ERR(trans)) {
10126 ret = PTR_ERR(trans);
10127 goto out_new_inode_args;
10130 ret = btrfs_create_new_inode(trans, &new_inode_args);
10133 * We set number of links to 0 in btrfs_create_new_inode(), and here we
10134 * set it to 1 because d_tmpfile() will issue a warning if the count is
10137 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10139 set_nlink(inode, 1);
10142 d_tmpfile(dentry, inode);
10143 unlock_new_inode(inode);
10144 mark_inode_dirty(inode);
10147 btrfs_end_transaction(trans);
10148 btrfs_btree_balance_dirty(fs_info);
10149 out_new_inode_args:
10150 btrfs_new_inode_args_destroy(&new_inode_args);
10157 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10159 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10160 unsigned long index = start >> PAGE_SHIFT;
10161 unsigned long end_index = end >> PAGE_SHIFT;
10165 ASSERT(end + 1 - start <= U32_MAX);
10166 len = end + 1 - start;
10167 while (index <= end_index) {
10168 page = find_get_page(inode->vfs_inode.i_mapping, index);
10169 ASSERT(page); /* Pages should be in the extent_io_tree */
10171 btrfs_page_set_writeback(fs_info, page, start, len);
10177 static int btrfs_encoded_io_compression_from_extent(
10178 struct btrfs_fs_info *fs_info,
10181 switch (compress_type) {
10182 case BTRFS_COMPRESS_NONE:
10183 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10184 case BTRFS_COMPRESS_ZLIB:
10185 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10186 case BTRFS_COMPRESS_LZO:
10188 * The LZO format depends on the sector size. 64K is the maximum
10189 * sector size that we support.
10191 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10193 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10194 (fs_info->sectorsize_bits - 12);
10195 case BTRFS_COMPRESS_ZSTD:
10196 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10202 static ssize_t btrfs_encoded_read_inline(
10203 struct kiocb *iocb,
10204 struct iov_iter *iter, u64 start,
10206 struct extent_state **cached_state,
10207 u64 extent_start, size_t count,
10208 struct btrfs_ioctl_encoded_io_args *encoded,
10211 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10212 struct btrfs_root *root = inode->root;
10213 struct btrfs_fs_info *fs_info = root->fs_info;
10214 struct extent_io_tree *io_tree = &inode->io_tree;
10215 struct btrfs_path *path;
10216 struct extent_buffer *leaf;
10217 struct btrfs_file_extent_item *item;
10223 path = btrfs_alloc_path();
10228 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10232 /* The extent item disappeared? */
10237 leaf = path->nodes[0];
10238 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10240 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10241 ptr = btrfs_file_extent_inline_start(item);
10243 encoded->len = min_t(u64, extent_start + ram_bytes,
10244 inode->vfs_inode.i_size) - iocb->ki_pos;
10245 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10246 btrfs_file_extent_compression(leaf, item));
10249 encoded->compression = ret;
10250 if (encoded->compression) {
10251 size_t inline_size;
10253 inline_size = btrfs_file_extent_inline_item_len(leaf,
10255 if (inline_size > count) {
10259 count = inline_size;
10260 encoded->unencoded_len = ram_bytes;
10261 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10263 count = min_t(u64, count, encoded->len);
10264 encoded->len = count;
10265 encoded->unencoded_len = count;
10266 ptr += iocb->ki_pos - extent_start;
10269 tmp = kmalloc(count, GFP_NOFS);
10274 read_extent_buffer(leaf, tmp, ptr, count);
10275 btrfs_release_path(path);
10276 unlock_extent_cached(io_tree, start, lockend, cached_state);
10277 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10280 ret = copy_to_iter(tmp, count, iter);
10285 btrfs_free_path(path);
10289 struct btrfs_encoded_read_private {
10290 struct btrfs_inode *inode;
10292 wait_queue_head_t wait;
10294 blk_status_t status;
10298 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10299 struct bio *bio, int mirror_num)
10301 struct btrfs_encoded_read_private *priv = bio->bi_private;
10302 struct btrfs_bio *bbio = btrfs_bio(bio);
10303 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10306 if (!priv->skip_csum) {
10307 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10312 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
10314 btrfs_bio_free_csum(bbio);
10318 atomic_inc(&priv->pending);
10319 ret = btrfs_map_bio(fs_info, bio, mirror_num);
10321 atomic_dec(&priv->pending);
10322 btrfs_bio_free_csum(bbio);
10327 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10329 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10330 struct btrfs_encoded_read_private *priv = bbio->bio.bi_private;
10331 struct btrfs_inode *inode = priv->inode;
10332 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10333 u32 sectorsize = fs_info->sectorsize;
10334 struct bio_vec *bvec;
10335 struct bvec_iter_all iter_all;
10336 u64 start = priv->file_offset;
10337 u32 bio_offset = 0;
10339 if (priv->skip_csum || !uptodate)
10340 return bbio->bio.bi_status;
10342 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10343 unsigned int i, nr_sectors, pgoff;
10345 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10346 pgoff = bvec->bv_offset;
10347 for (i = 0; i < nr_sectors; i++) {
10348 ASSERT(pgoff < PAGE_SIZE);
10349 if (check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10350 bvec->bv_page, pgoff, start))
10351 return BLK_STS_IOERR;
10352 start += sectorsize;
10353 bio_offset += sectorsize;
10354 pgoff += sectorsize;
10360 static void btrfs_encoded_read_endio(struct bio *bio)
10362 struct btrfs_encoded_read_private *priv = bio->bi_private;
10363 struct btrfs_bio *bbio = btrfs_bio(bio);
10364 blk_status_t status;
10366 status = btrfs_encoded_read_verify_csum(bbio);
10369 * The memory barrier implied by the atomic_dec_return() here
10370 * pairs with the memory barrier implied by the
10371 * atomic_dec_return() or io_wait_event() in
10372 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10373 * write is observed before the load of status in
10374 * btrfs_encoded_read_regular_fill_pages().
10376 WRITE_ONCE(priv->status, status);
10378 if (!atomic_dec_return(&priv->pending))
10379 wake_up(&priv->wait);
10380 btrfs_bio_free_csum(bbio);
10384 static int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10388 struct page **pages)
10390 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10391 struct btrfs_encoded_read_private priv = {
10393 .file_offset = file_offset,
10394 .pending = ATOMIC_INIT(1),
10395 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10397 unsigned long i = 0;
10401 init_waitqueue_head(&priv.wait);
10403 * Submit bios for the extent, splitting due to bio or stripe limits as
10406 while (cur < disk_io_size) {
10407 struct extent_map *em;
10408 struct btrfs_io_geometry geom;
10409 struct bio *bio = NULL;
10412 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10413 disk_io_size - cur);
10417 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10418 disk_bytenr + cur, &geom);
10419 free_extent_map(em);
10422 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10425 remaining = min(geom.len, disk_io_size - cur);
10426 while (bio || remaining) {
10427 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10430 bio = btrfs_bio_alloc(BIO_MAX_VECS);
10431 bio->bi_iter.bi_sector =
10432 (disk_bytenr + cur) >> SECTOR_SHIFT;
10433 bio->bi_end_io = btrfs_encoded_read_endio;
10434 bio->bi_private = &priv;
10435 bio->bi_opf = REQ_OP_READ;
10439 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10440 blk_status_t status;
10442 status = submit_encoded_read_bio(inode, bio, 0);
10444 WRITE_ONCE(priv.status, status);
10454 remaining -= bytes;
10459 if (atomic_dec_return(&priv.pending))
10460 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10461 /* See btrfs_encoded_read_endio() for ordering. */
10462 return blk_status_to_errno(READ_ONCE(priv.status));
10465 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10466 struct iov_iter *iter,
10467 u64 start, u64 lockend,
10468 struct extent_state **cached_state,
10469 u64 disk_bytenr, u64 disk_io_size,
10470 size_t count, bool compressed,
10473 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10474 struct extent_io_tree *io_tree = &inode->io_tree;
10475 struct page **pages;
10476 unsigned long nr_pages, i;
10478 size_t page_offset;
10481 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10482 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10485 ret = btrfs_alloc_page_array(nr_pages, pages);
10491 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10492 disk_io_size, pages);
10496 unlock_extent_cached(io_tree, start, lockend, cached_state);
10497 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10504 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10505 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10508 while (cur < count) {
10509 size_t bytes = min_t(size_t, count - cur,
10510 PAGE_SIZE - page_offset);
10512 if (copy_page_to_iter(pages[i], page_offset, bytes,
10523 for (i = 0; i < nr_pages; i++) {
10525 __free_page(pages[i]);
10531 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10532 struct btrfs_ioctl_encoded_io_args *encoded)
10534 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10535 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10536 struct extent_io_tree *io_tree = &inode->io_tree;
10538 size_t count = iov_iter_count(iter);
10539 u64 start, lockend, disk_bytenr, disk_io_size;
10540 struct extent_state *cached_state = NULL;
10541 struct extent_map *em;
10542 bool unlocked = false;
10544 file_accessed(iocb->ki_filp);
10546 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10548 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10549 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10552 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10554 * We don't know how long the extent containing iocb->ki_pos is, but if
10555 * it's compressed we know that it won't be longer than this.
10557 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10560 struct btrfs_ordered_extent *ordered;
10562 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10563 lockend - start + 1);
10565 goto out_unlock_inode;
10566 lock_extent_bits(io_tree, start, lockend, &cached_state);
10567 ordered = btrfs_lookup_ordered_range(inode, start,
10568 lockend - start + 1);
10571 btrfs_put_ordered_extent(ordered);
10572 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10576 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10579 goto out_unlock_extent;
10582 if (em->block_start == EXTENT_MAP_INLINE) {
10583 u64 extent_start = em->start;
10586 * For inline extents we get everything we need out of the
10589 free_extent_map(em);
10591 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10592 &cached_state, extent_start,
10593 count, encoded, &unlocked);
10598 * We only want to return up to EOF even if the extent extends beyond
10601 encoded->len = min_t(u64, extent_map_end(em),
10602 inode->vfs_inode.i_size) - iocb->ki_pos;
10603 if (em->block_start == EXTENT_MAP_HOLE ||
10604 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10605 disk_bytenr = EXTENT_MAP_HOLE;
10606 count = min_t(u64, count, encoded->len);
10607 encoded->len = count;
10608 encoded->unencoded_len = count;
10609 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10610 disk_bytenr = em->block_start;
10612 * Bail if the buffer isn't large enough to return the whole
10613 * compressed extent.
10615 if (em->block_len > count) {
10619 disk_io_size = count = em->block_len;
10620 encoded->unencoded_len = em->ram_bytes;
10621 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10622 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10623 em->compress_type);
10626 encoded->compression = ret;
10628 disk_bytenr = em->block_start + (start - em->start);
10629 if (encoded->len > count)
10630 encoded->len = count;
10632 * Don't read beyond what we locked. This also limits the page
10633 * allocations that we'll do.
10635 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10636 count = start + disk_io_size - iocb->ki_pos;
10637 encoded->len = count;
10638 encoded->unencoded_len = count;
10639 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10641 free_extent_map(em);
10644 if (disk_bytenr == EXTENT_MAP_HOLE) {
10645 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10646 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10648 ret = iov_iter_zero(count, iter);
10652 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10653 &cached_state, disk_bytenr,
10654 disk_io_size, count,
10655 encoded->compression,
10661 iocb->ki_pos += encoded->len;
10663 free_extent_map(em);
10666 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10669 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10673 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10674 const struct btrfs_ioctl_encoded_io_args *encoded)
10676 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10677 struct btrfs_root *root = inode->root;
10678 struct btrfs_fs_info *fs_info = root->fs_info;
10679 struct extent_io_tree *io_tree = &inode->io_tree;
10680 struct extent_changeset *data_reserved = NULL;
10681 struct extent_state *cached_state = NULL;
10685 u64 num_bytes, ram_bytes, disk_num_bytes;
10686 unsigned long nr_pages, i;
10687 struct page **pages;
10688 struct btrfs_key ins;
10689 bool extent_reserved = false;
10690 struct extent_map *em;
10693 switch (encoded->compression) {
10694 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10695 compression = BTRFS_COMPRESS_ZLIB;
10697 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10698 compression = BTRFS_COMPRESS_ZSTD;
10700 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10701 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10702 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10703 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10704 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10705 /* The sector size must match for LZO. */
10706 if (encoded->compression -
10707 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10708 fs_info->sectorsize_bits)
10710 compression = BTRFS_COMPRESS_LZO;
10715 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10718 orig_count = iov_iter_count(from);
10720 /* The extent size must be sane. */
10721 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10722 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10726 * The compressed data must be smaller than the decompressed data.
10728 * It's of course possible for data to compress to larger or the same
10729 * size, but the buffered I/O path falls back to no compression for such
10730 * data, and we don't want to break any assumptions by creating these
10733 * Note that this is less strict than the current check we have that the
10734 * compressed data must be at least one sector smaller than the
10735 * decompressed data. We only want to enforce the weaker requirement
10736 * from old kernels that it is at least one byte smaller.
10738 if (orig_count >= encoded->unencoded_len)
10741 /* The extent must start on a sector boundary. */
10742 start = iocb->ki_pos;
10743 if (!IS_ALIGNED(start, fs_info->sectorsize))
10747 * The extent must end on a sector boundary. However, we allow a write
10748 * which ends at or extends i_size to have an unaligned length; we round
10749 * up the extent size and set i_size to the unaligned end.
10751 if (start + encoded->len < inode->vfs_inode.i_size &&
10752 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10755 /* Finally, the offset in the unencoded data must be sector-aligned. */
10756 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10759 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10760 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10761 end = start + num_bytes - 1;
10764 * If the extent cannot be inline, the compressed data on disk must be
10765 * sector-aligned. For convenience, we extend it with zeroes if it
10768 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10769 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10770 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10773 for (i = 0; i < nr_pages; i++) {
10774 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10777 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10782 kaddr = kmap(pages[i]);
10783 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10788 if (bytes < PAGE_SIZE)
10789 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10794 struct btrfs_ordered_extent *ordered;
10796 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10799 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10800 start >> PAGE_SHIFT,
10801 end >> PAGE_SHIFT);
10804 lock_extent_bits(io_tree, start, end, &cached_state);
10805 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10807 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10810 btrfs_put_ordered_extent(ordered);
10811 unlock_extent_cached(io_tree, start, end, &cached_state);
10816 * We don't use the higher-level delalloc space functions because our
10817 * num_bytes and disk_num_bytes are different.
10819 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10822 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10824 goto out_free_data_space;
10825 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10828 goto out_qgroup_free_data;
10830 /* Try an inline extent first. */
10831 if (start == 0 && encoded->unencoded_len == encoded->len &&
10832 encoded->unencoded_offset == 0) {
10833 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10834 compression, pages, true);
10838 goto out_delalloc_release;
10842 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10843 disk_num_bytes, 0, 0, &ins, 1, 1);
10845 goto out_delalloc_release;
10846 extent_reserved = true;
10848 em = create_io_em(inode, start, num_bytes,
10849 start - encoded->unencoded_offset, ins.objectid,
10850 ins.offset, ins.offset, ram_bytes, compression,
10851 BTRFS_ORDERED_COMPRESSED);
10854 goto out_free_reserved;
10856 free_extent_map(em);
10858 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10859 ins.objectid, ins.offset,
10860 encoded->unencoded_offset,
10861 (1 << BTRFS_ORDERED_ENCODED) |
10862 (1 << BTRFS_ORDERED_COMPRESSED),
10865 btrfs_drop_extent_cache(inode, start, end, 0);
10866 goto out_free_reserved;
10868 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10870 if (start + encoded->len > inode->vfs_inode.i_size)
10871 i_size_write(&inode->vfs_inode, start + encoded->len);
10873 unlock_extent_cached(io_tree, start, end, &cached_state);
10875 btrfs_delalloc_release_extents(inode, num_bytes);
10877 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10878 ins.offset, pages, nr_pages, 0, NULL,
10880 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10888 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10889 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10890 out_delalloc_release:
10891 btrfs_delalloc_release_extents(inode, num_bytes);
10892 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10893 out_qgroup_free_data:
10895 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10896 out_free_data_space:
10898 * If btrfs_reserve_extent() succeeded, then we already decremented
10901 if (!extent_reserved)
10902 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10904 unlock_extent_cached(io_tree, start, end, &cached_state);
10906 for (i = 0; i < nr_pages; i++) {
10908 __free_page(pages[i]);
10913 iocb->ki_pos += encoded->len;
10919 * Add an entry indicating a block group or device which is pinned by a
10920 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10921 * negative errno on failure.
10923 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10924 bool is_block_group)
10926 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10927 struct btrfs_swapfile_pin *sp, *entry;
10928 struct rb_node **p;
10929 struct rb_node *parent = NULL;
10931 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10936 sp->is_block_group = is_block_group;
10937 sp->bg_extent_count = 1;
10939 spin_lock(&fs_info->swapfile_pins_lock);
10940 p = &fs_info->swapfile_pins.rb_node;
10943 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10944 if (sp->ptr < entry->ptr ||
10945 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10946 p = &(*p)->rb_left;
10947 } else if (sp->ptr > entry->ptr ||
10948 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10949 p = &(*p)->rb_right;
10951 if (is_block_group)
10952 entry->bg_extent_count++;
10953 spin_unlock(&fs_info->swapfile_pins_lock);
10958 rb_link_node(&sp->node, parent, p);
10959 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10960 spin_unlock(&fs_info->swapfile_pins_lock);
10964 /* Free all of the entries pinned by this swapfile. */
10965 static void btrfs_free_swapfile_pins(struct inode *inode)
10967 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10968 struct btrfs_swapfile_pin *sp;
10969 struct rb_node *node, *next;
10971 spin_lock(&fs_info->swapfile_pins_lock);
10972 node = rb_first(&fs_info->swapfile_pins);
10974 next = rb_next(node);
10975 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10976 if (sp->inode == inode) {
10977 rb_erase(&sp->node, &fs_info->swapfile_pins);
10978 if (sp->is_block_group) {
10979 btrfs_dec_block_group_swap_extents(sp->ptr,
10980 sp->bg_extent_count);
10981 btrfs_put_block_group(sp->ptr);
10987 spin_unlock(&fs_info->swapfile_pins_lock);
10990 struct btrfs_swap_info {
10996 unsigned long nr_pages;
11000 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
11001 struct btrfs_swap_info *bsi)
11003 unsigned long nr_pages;
11004 unsigned long max_pages;
11005 u64 first_ppage, first_ppage_reported, next_ppage;
11009 * Our swapfile may have had its size extended after the swap header was
11010 * written. In that case activating the swapfile should not go beyond
11011 * the max size set in the swap header.
11013 if (bsi->nr_pages >= sis->max)
11016 max_pages = sis->max - bsi->nr_pages;
11017 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
11018 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
11019 PAGE_SIZE) >> PAGE_SHIFT;
11021 if (first_ppage >= next_ppage)
11023 nr_pages = next_ppage - first_ppage;
11024 nr_pages = min(nr_pages, max_pages);
11026 first_ppage_reported = first_ppage;
11027 if (bsi->start == 0)
11028 first_ppage_reported++;
11029 if (bsi->lowest_ppage > first_ppage_reported)
11030 bsi->lowest_ppage = first_ppage_reported;
11031 if (bsi->highest_ppage < (next_ppage - 1))
11032 bsi->highest_ppage = next_ppage - 1;
11034 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
11037 bsi->nr_extents += ret;
11038 bsi->nr_pages += nr_pages;
11042 static void btrfs_swap_deactivate(struct file *file)
11044 struct inode *inode = file_inode(file);
11046 btrfs_free_swapfile_pins(inode);
11047 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
11050 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11053 struct inode *inode = file_inode(file);
11054 struct btrfs_root *root = BTRFS_I(inode)->root;
11055 struct btrfs_fs_info *fs_info = root->fs_info;
11056 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
11057 struct extent_state *cached_state = NULL;
11058 struct extent_map *em = NULL;
11059 struct btrfs_device *device = NULL;
11060 struct btrfs_swap_info bsi = {
11061 .lowest_ppage = (sector_t)-1ULL,
11068 * If the swap file was just created, make sure delalloc is done. If the
11069 * file changes again after this, the user is doing something stupid and
11070 * we don't really care.
11072 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
11077 * The inode is locked, so these flags won't change after we check them.
11079 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
11080 btrfs_warn(fs_info, "swapfile must not be compressed");
11083 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
11084 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
11087 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
11088 btrfs_warn(fs_info, "swapfile must not be checksummed");
11093 * Balance or device remove/replace/resize can move stuff around from
11094 * under us. The exclop protection makes sure they aren't running/won't
11095 * run concurrently while we are mapping the swap extents, and
11096 * fs_info->swapfile_pins prevents them from running while the swap
11097 * file is active and moving the extents. Note that this also prevents
11098 * a concurrent device add which isn't actually necessary, but it's not
11099 * really worth the trouble to allow it.
11101 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
11102 btrfs_warn(fs_info,
11103 "cannot activate swapfile while exclusive operation is running");
11108 * Prevent snapshot creation while we are activating the swap file.
11109 * We do not want to race with snapshot creation. If snapshot creation
11110 * already started before we bumped nr_swapfiles from 0 to 1 and
11111 * completes before the first write into the swap file after it is
11112 * activated, than that write would fallback to COW.
11114 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11115 btrfs_exclop_finish(fs_info);
11116 btrfs_warn(fs_info,
11117 "cannot activate swapfile because snapshot creation is in progress");
11121 * Snapshots can create extents which require COW even if NODATACOW is
11122 * set. We use this counter to prevent snapshots. We must increment it
11123 * before walking the extents because we don't want a concurrent
11124 * snapshot to run after we've already checked the extents.
11126 * It is possible that subvolume is marked for deletion but still not
11127 * removed yet. To prevent this race, we check the root status before
11128 * activating the swapfile.
11130 spin_lock(&root->root_item_lock);
11131 if (btrfs_root_dead(root)) {
11132 spin_unlock(&root->root_item_lock);
11134 btrfs_exclop_finish(fs_info);
11135 btrfs_warn(fs_info,
11136 "cannot activate swapfile because subvolume %llu is being deleted",
11137 root->root_key.objectid);
11140 atomic_inc(&root->nr_swapfiles);
11141 spin_unlock(&root->root_item_lock);
11143 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11145 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
11147 while (start < isize) {
11148 u64 logical_block_start, physical_block_start;
11149 struct btrfs_block_group *bg;
11150 u64 len = isize - start;
11152 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11158 if (em->block_start == EXTENT_MAP_HOLE) {
11159 btrfs_warn(fs_info, "swapfile must not have holes");
11163 if (em->block_start == EXTENT_MAP_INLINE) {
11165 * It's unlikely we'll ever actually find ourselves
11166 * here, as a file small enough to fit inline won't be
11167 * big enough to store more than the swap header, but in
11168 * case something changes in the future, let's catch it
11169 * here rather than later.
11171 btrfs_warn(fs_info, "swapfile must not be inline");
11175 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11176 btrfs_warn(fs_info, "swapfile must not be compressed");
11181 logical_block_start = em->block_start + (start - em->start);
11182 len = min(len, em->len - (start - em->start));
11183 free_extent_map(em);
11186 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
11192 btrfs_warn(fs_info,
11193 "swapfile must not be copy-on-write");
11198 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11204 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11205 btrfs_warn(fs_info,
11206 "swapfile must have single data profile");
11211 if (device == NULL) {
11212 device = em->map_lookup->stripes[0].dev;
11213 ret = btrfs_add_swapfile_pin(inode, device, false);
11218 } else if (device != em->map_lookup->stripes[0].dev) {
11219 btrfs_warn(fs_info, "swapfile must be on one device");
11224 physical_block_start = (em->map_lookup->stripes[0].physical +
11225 (logical_block_start - em->start));
11226 len = min(len, em->len - (logical_block_start - em->start));
11227 free_extent_map(em);
11230 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11232 btrfs_warn(fs_info,
11233 "could not find block group containing swapfile");
11238 if (!btrfs_inc_block_group_swap_extents(bg)) {
11239 btrfs_warn(fs_info,
11240 "block group for swapfile at %llu is read-only%s",
11242 atomic_read(&fs_info->scrubs_running) ?
11243 " (scrub running)" : "");
11244 btrfs_put_block_group(bg);
11249 ret = btrfs_add_swapfile_pin(inode, bg, true);
11251 btrfs_put_block_group(bg);
11258 if (bsi.block_len &&
11259 bsi.block_start + bsi.block_len == physical_block_start) {
11260 bsi.block_len += len;
11262 if (bsi.block_len) {
11263 ret = btrfs_add_swap_extent(sis, &bsi);
11268 bsi.block_start = physical_block_start;
11269 bsi.block_len = len;
11276 ret = btrfs_add_swap_extent(sis, &bsi);
11279 if (!IS_ERR_OR_NULL(em))
11280 free_extent_map(em);
11282 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
11285 btrfs_swap_deactivate(file);
11287 btrfs_drew_write_unlock(&root->snapshot_lock);
11289 btrfs_exclop_finish(fs_info);
11295 sis->bdev = device->bdev;
11296 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11297 sis->max = bsi.nr_pages;
11298 sis->pages = bsi.nr_pages - 1;
11299 sis->highest_bit = bsi.nr_pages - 1;
11300 return bsi.nr_extents;
11303 static void btrfs_swap_deactivate(struct file *file)
11307 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11310 return -EOPNOTSUPP;
11315 * Update the number of bytes used in the VFS' inode. When we replace extents in
11316 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11317 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11318 * always get a correct value.
11320 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11321 const u64 add_bytes,
11322 const u64 del_bytes)
11324 if (add_bytes == del_bytes)
11327 spin_lock(&inode->lock);
11329 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11331 inode_add_bytes(&inode->vfs_inode, add_bytes);
11332 spin_unlock(&inode->lock);
11336 * Verify that there are no ordered extents for a given file range.
11338 * @inode: The target inode.
11339 * @start: Start offset of the file range, should be sector size aligned.
11340 * @end: End offset (inclusive) of the file range, its value +1 should be
11341 * sector size aligned.
11343 * This should typically be used for cases where we locked an inode's VFS lock in
11344 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11345 * we have flushed all delalloc in the range, we have waited for all ordered
11346 * extents in the range to complete and finally we have locked the file range in
11347 * the inode's io_tree.
11349 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11351 struct btrfs_root *root = inode->root;
11352 struct btrfs_ordered_extent *ordered;
11354 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11357 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11359 btrfs_err(root->fs_info,
11360 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11361 start, end, btrfs_ino(inode), root->root_key.objectid,
11362 ordered->file_offset,
11363 ordered->file_offset + ordered->num_bytes - 1);
11364 btrfs_put_ordered_extent(ordered);
11367 ASSERT(ordered == NULL);
11370 static const struct inode_operations btrfs_dir_inode_operations = {
11371 .getattr = btrfs_getattr,
11372 .lookup = btrfs_lookup,
11373 .create = btrfs_create,
11374 .unlink = btrfs_unlink,
11375 .link = btrfs_link,
11376 .mkdir = btrfs_mkdir,
11377 .rmdir = btrfs_rmdir,
11378 .rename = btrfs_rename2,
11379 .symlink = btrfs_symlink,
11380 .setattr = btrfs_setattr,
11381 .mknod = btrfs_mknod,
11382 .listxattr = btrfs_listxattr,
11383 .permission = btrfs_permission,
11384 .get_acl = btrfs_get_acl,
11385 .set_acl = btrfs_set_acl,
11386 .update_time = btrfs_update_time,
11387 .tmpfile = btrfs_tmpfile,
11388 .fileattr_get = btrfs_fileattr_get,
11389 .fileattr_set = btrfs_fileattr_set,
11392 static const struct file_operations btrfs_dir_file_operations = {
11393 .llseek = generic_file_llseek,
11394 .read = generic_read_dir,
11395 .iterate_shared = btrfs_real_readdir,
11396 .open = btrfs_opendir,
11397 .unlocked_ioctl = btrfs_ioctl,
11398 #ifdef CONFIG_COMPAT
11399 .compat_ioctl = btrfs_compat_ioctl,
11401 .release = btrfs_release_file,
11402 .fsync = btrfs_sync_file,
11406 * btrfs doesn't support the bmap operation because swapfiles
11407 * use bmap to make a mapping of extents in the file. They assume
11408 * these extents won't change over the life of the file and they
11409 * use the bmap result to do IO directly to the drive.
11411 * the btrfs bmap call would return logical addresses that aren't
11412 * suitable for IO and they also will change frequently as COW
11413 * operations happen. So, swapfile + btrfs == corruption.
11415 * For now we're avoiding this by dropping bmap.
11417 static const struct address_space_operations btrfs_aops = {
11418 .read_folio = btrfs_read_folio,
11419 .writepage = btrfs_writepage,
11420 .writepages = btrfs_writepages,
11421 .readahead = btrfs_readahead,
11422 .direct_IO = noop_direct_IO,
11423 .invalidate_folio = btrfs_invalidate_folio,
11424 .release_folio = btrfs_release_folio,
11425 #ifdef CONFIG_MIGRATION
11426 .migratepage = btrfs_migratepage,
11428 .dirty_folio = filemap_dirty_folio,
11429 .error_remove_page = generic_error_remove_page,
11430 .swap_activate = btrfs_swap_activate,
11431 .swap_deactivate = btrfs_swap_deactivate,
11434 static const struct inode_operations btrfs_file_inode_operations = {
11435 .getattr = btrfs_getattr,
11436 .setattr = btrfs_setattr,
11437 .listxattr = btrfs_listxattr,
11438 .permission = btrfs_permission,
11439 .fiemap = btrfs_fiemap,
11440 .get_acl = btrfs_get_acl,
11441 .set_acl = btrfs_set_acl,
11442 .update_time = btrfs_update_time,
11443 .fileattr_get = btrfs_fileattr_get,
11444 .fileattr_set = btrfs_fileattr_set,
11446 static const struct inode_operations btrfs_special_inode_operations = {
11447 .getattr = btrfs_getattr,
11448 .setattr = btrfs_setattr,
11449 .permission = btrfs_permission,
11450 .listxattr = btrfs_listxattr,
11451 .get_acl = btrfs_get_acl,
11452 .set_acl = btrfs_set_acl,
11453 .update_time = btrfs_update_time,
11455 static const struct inode_operations btrfs_symlink_inode_operations = {
11456 .get_link = page_get_link,
11457 .getattr = btrfs_getattr,
11458 .setattr = btrfs_setattr,
11459 .permission = btrfs_permission,
11460 .listxattr = btrfs_listxattr,
11461 .update_time = btrfs_update_time,
11464 const struct dentry_operations btrfs_dentry_operations = {
11465 .d_delete = btrfs_dentry_delete,