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;
69 struct btrfs_rename_ctx {
70 /* Output field. Stores the index number of the old directory entry. */
74 static const struct inode_operations btrfs_dir_inode_operations;
75 static const struct inode_operations btrfs_symlink_inode_operations;
76 static const struct inode_operations btrfs_special_inode_operations;
77 static const struct inode_operations btrfs_file_inode_operations;
78 static const struct address_space_operations btrfs_aops;
79 static const struct file_operations btrfs_dir_file_operations;
81 static struct kmem_cache *btrfs_inode_cachep;
82 struct kmem_cache *btrfs_trans_handle_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
85 struct kmem_cache *btrfs_free_space_bitmap_cachep;
87 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
88 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
89 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
90 static noinline int cow_file_range(struct btrfs_inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
94 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
95 u64 len, u64 orig_start, u64 block_start,
96 u64 block_len, u64 orig_block_len,
97 u64 ram_bytes, int compress_type,
100 static void __endio_write_update_ordered(struct btrfs_inode *inode,
101 const u64 offset, const u64 bytes,
102 const bool uptodate);
105 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
107 * ilock_flags can have the following bit set:
109 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
110 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
112 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
114 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
116 if (ilock_flags & BTRFS_ILOCK_SHARED) {
117 if (ilock_flags & BTRFS_ILOCK_TRY) {
118 if (!inode_trylock_shared(inode))
123 inode_lock_shared(inode);
125 if (ilock_flags & BTRFS_ILOCK_TRY) {
126 if (!inode_trylock(inode))
133 if (ilock_flags & BTRFS_ILOCK_MMAP)
134 down_write(&BTRFS_I(inode)->i_mmap_lock);
139 * btrfs_inode_unlock - unock inode i_rwsem
141 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
142 * to decide whether the lock acquired is shared or exclusive.
144 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
146 if (ilock_flags & BTRFS_ILOCK_MMAP)
147 up_write(&BTRFS_I(inode)->i_mmap_lock);
148 if (ilock_flags & BTRFS_ILOCK_SHARED)
149 inode_unlock_shared(inode);
155 * Cleanup all submitted ordered extents in specified range to handle errors
156 * from the btrfs_run_delalloc_range() callback.
158 * NOTE: caller must ensure that when an error happens, it can not call
159 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
160 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
161 * to be released, which we want to happen only when finishing the ordered
162 * extent (btrfs_finish_ordered_io()).
164 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
165 struct page *locked_page,
166 u64 offset, u64 bytes)
168 unsigned long index = offset >> PAGE_SHIFT;
169 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
170 u64 page_start = page_offset(locked_page);
171 u64 page_end = page_start + PAGE_SIZE - 1;
175 while (index <= end_index) {
177 * For locked page, we will call end_extent_writepage() on it
178 * in run_delalloc_range() for the error handling. That
179 * end_extent_writepage() function will call
180 * btrfs_mark_ordered_io_finished() to clear page Ordered and
181 * run the ordered extent accounting.
183 * Here we can't just clear the Ordered bit, or
184 * btrfs_mark_ordered_io_finished() would skip the accounting
185 * for the page range, and the ordered extent will never finish.
187 if (index == (page_offset(locked_page) >> PAGE_SHIFT)) {
191 page = find_get_page(inode->vfs_inode.i_mapping, index);
197 * Here we just clear all Ordered bits for every page in the
198 * range, then __endio_write_update_ordered() will handle
199 * the ordered extent accounting for the range.
201 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
206 /* The locked page covers the full range, nothing needs to be done */
207 if (bytes + offset <= page_offset(locked_page) + PAGE_SIZE)
210 * In case this page belongs to the delalloc range being instantiated
211 * then skip it, since the first page of a range is going to be
212 * properly cleaned up by the caller of run_delalloc_range
214 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
215 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
216 offset = page_offset(locked_page) + PAGE_SIZE;
219 return __endio_write_update_ordered(inode, offset, bytes, false);
222 static int btrfs_dirty_inode(struct inode *inode);
224 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
225 struct inode *inode, struct inode *dir,
226 const struct qstr *qstr)
230 err = btrfs_init_acl(trans, inode, dir);
232 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
237 * this does all the hard work for inserting an inline extent into
238 * the btree. The caller should have done a btrfs_drop_extents so that
239 * no overlapping inline items exist in the btree
241 static int insert_inline_extent(struct btrfs_trans_handle *trans,
242 struct btrfs_path *path,
243 struct btrfs_inode *inode, bool extent_inserted,
244 size_t size, size_t compressed_size,
246 struct page **compressed_pages,
249 struct btrfs_root *root = inode->root;
250 struct extent_buffer *leaf;
251 struct page *page = NULL;
254 struct btrfs_file_extent_item *ei;
256 size_t cur_size = size;
259 ASSERT((compressed_size > 0 && compressed_pages) ||
260 (compressed_size == 0 && !compressed_pages));
262 if (compressed_size && compressed_pages)
263 cur_size = compressed_size;
265 if (!extent_inserted) {
266 struct btrfs_key key;
269 key.objectid = btrfs_ino(inode);
271 key.type = BTRFS_EXTENT_DATA_KEY;
273 datasize = btrfs_file_extent_calc_inline_size(cur_size);
274 ret = btrfs_insert_empty_item(trans, root, path, &key,
279 leaf = path->nodes[0];
280 ei = btrfs_item_ptr(leaf, path->slots[0],
281 struct btrfs_file_extent_item);
282 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
283 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
284 btrfs_set_file_extent_encryption(leaf, ei, 0);
285 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
286 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
287 ptr = btrfs_file_extent_inline_start(ei);
289 if (compress_type != BTRFS_COMPRESS_NONE) {
292 while (compressed_size > 0) {
293 cpage = compressed_pages[i];
294 cur_size = min_t(unsigned long, compressed_size,
297 kaddr = kmap_atomic(cpage);
298 write_extent_buffer(leaf, kaddr, ptr, cur_size);
299 kunmap_atomic(kaddr);
303 compressed_size -= cur_size;
305 btrfs_set_file_extent_compression(leaf, ei,
308 page = find_get_page(inode->vfs_inode.i_mapping, 0);
309 btrfs_set_file_extent_compression(leaf, ei, 0);
310 kaddr = kmap_atomic(page);
311 write_extent_buffer(leaf, kaddr, ptr, size);
312 kunmap_atomic(kaddr);
315 btrfs_mark_buffer_dirty(leaf);
316 btrfs_release_path(path);
319 * We align size to sectorsize for inline extents just for simplicity
322 ret = btrfs_inode_set_file_extent_range(inode, 0,
323 ALIGN(size, root->fs_info->sectorsize));
328 * We're an inline extent, so nobody can extend the file past i_size
329 * without locking a page we already have locked.
331 * We must do any i_size and inode updates before we unlock the pages.
332 * Otherwise we could end up racing with unlink.
334 i_size = i_size_read(&inode->vfs_inode);
335 if (update_i_size && size > i_size) {
336 i_size_write(&inode->vfs_inode, size);
339 inode->disk_i_size = i_size;
347 * conditionally insert an inline extent into the file. This
348 * does the checks required to make sure the data is small enough
349 * to fit as an inline extent.
351 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
352 size_t compressed_size,
354 struct page **compressed_pages,
357 struct btrfs_drop_extents_args drop_args = { 0 };
358 struct btrfs_root *root = inode->root;
359 struct btrfs_fs_info *fs_info = root->fs_info;
360 struct btrfs_trans_handle *trans;
361 u64 data_len = (compressed_size ?: size);
363 struct btrfs_path *path;
366 * We can create an inline extent if it ends at or beyond the current
367 * i_size, is no larger than a sector (decompressed), and the (possibly
368 * compressed) data fits in a leaf and the configured maximum inline
371 if (size < i_size_read(&inode->vfs_inode) ||
372 size > fs_info->sectorsize ||
373 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
374 data_len > fs_info->max_inline)
377 path = btrfs_alloc_path();
381 trans = btrfs_join_transaction(root);
383 btrfs_free_path(path);
384 return PTR_ERR(trans);
386 trans->block_rsv = &inode->block_rsv;
388 drop_args.path = path;
390 drop_args.end = fs_info->sectorsize;
391 drop_args.drop_cache = true;
392 drop_args.replace_extent = true;
393 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
394 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
396 btrfs_abort_transaction(trans, ret);
400 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
401 size, compressed_size, compress_type,
402 compressed_pages, update_i_size);
403 if (ret && ret != -ENOSPC) {
404 btrfs_abort_transaction(trans, ret);
406 } else if (ret == -ENOSPC) {
411 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
412 ret = btrfs_update_inode(trans, root, inode);
413 if (ret && ret != -ENOSPC) {
414 btrfs_abort_transaction(trans, ret);
416 } else if (ret == -ENOSPC) {
421 btrfs_set_inode_full_sync(inode);
424 * Don't forget to free the reserved space, as for inlined extent
425 * it won't count as data extent, free them directly here.
426 * And at reserve time, it's always aligned to page size, so
427 * just free one page here.
429 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
430 btrfs_free_path(path);
431 btrfs_end_transaction(trans);
435 struct async_extent {
440 unsigned long nr_pages;
442 struct list_head list;
447 struct page *locked_page;
450 unsigned int write_flags;
451 struct list_head extents;
452 struct cgroup_subsys_state *blkcg_css;
453 struct btrfs_work work;
454 struct async_cow *async_cow;
459 struct async_chunk chunks[];
462 static noinline int add_async_extent(struct async_chunk *cow,
463 u64 start, u64 ram_size,
466 unsigned long nr_pages,
469 struct async_extent *async_extent;
471 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
472 BUG_ON(!async_extent); /* -ENOMEM */
473 async_extent->start = start;
474 async_extent->ram_size = ram_size;
475 async_extent->compressed_size = compressed_size;
476 async_extent->pages = pages;
477 async_extent->nr_pages = nr_pages;
478 async_extent->compress_type = compress_type;
479 list_add_tail(&async_extent->list, &cow->extents);
484 * Check if the inode needs to be submitted to compression, based on mount
485 * options, defragmentation, properties or heuristics.
487 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
490 struct btrfs_fs_info *fs_info = inode->root->fs_info;
492 if (!btrfs_inode_can_compress(inode)) {
493 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
494 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
499 * Special check for subpage.
501 * We lock the full page then run each delalloc range in the page, thus
502 * for the following case, we will hit some subpage specific corner case:
505 * | |///////| |///////|
508 * In above case, both range A and range B will try to unlock the full
509 * page [0, 64K), causing the one finished later will have page
510 * unlocked already, triggering various page lock requirement BUG_ON()s.
512 * So here we add an artificial limit that subpage compression can only
513 * if the range is fully page aligned.
515 * In theory we only need to ensure the first page is fully covered, but
516 * the tailing partial page will be locked until the full compression
517 * finishes, delaying the write of other range.
519 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
520 * first to prevent any submitted async extent to unlock the full page.
521 * By this, we can ensure for subpage case that only the last async_cow
522 * will unlock the full page.
524 if (fs_info->sectorsize < PAGE_SIZE) {
525 if (!IS_ALIGNED(start, PAGE_SIZE) ||
526 !IS_ALIGNED(end + 1, PAGE_SIZE))
531 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
534 if (inode->defrag_compress)
536 /* bad compression ratios */
537 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
539 if (btrfs_test_opt(fs_info, COMPRESS) ||
540 inode->flags & BTRFS_INODE_COMPRESS ||
541 inode->prop_compress)
542 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
546 static inline void inode_should_defrag(struct btrfs_inode *inode,
547 u64 start, u64 end, u64 num_bytes, u32 small_write)
549 /* If this is a small write inside eof, kick off a defrag */
550 if (num_bytes < small_write &&
551 (start > 0 || end + 1 < inode->disk_i_size))
552 btrfs_add_inode_defrag(NULL, inode, small_write);
556 * we create compressed extents in two phases. The first
557 * phase compresses a range of pages that have already been
558 * locked (both pages and state bits are locked).
560 * This is done inside an ordered work queue, and the compression
561 * is spread across many cpus. The actual IO submission is step
562 * two, and the ordered work queue takes care of making sure that
563 * happens in the same order things were put onto the queue by
564 * writepages and friends.
566 * If this code finds it can't get good compression, it puts an
567 * entry onto the work queue to write the uncompressed bytes. This
568 * makes sure that both compressed inodes and uncompressed inodes
569 * are written in the same order that the flusher thread sent them
572 static noinline int compress_file_range(struct async_chunk *async_chunk)
574 struct inode *inode = async_chunk->inode;
575 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
576 u64 blocksize = fs_info->sectorsize;
577 u64 start = async_chunk->start;
578 u64 end = async_chunk->end;
582 struct page **pages = NULL;
583 unsigned long nr_pages;
584 unsigned long total_compressed = 0;
585 unsigned long total_in = 0;
588 int compress_type = fs_info->compress_type;
589 int compressed_extents = 0;
592 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
596 * We need to save i_size before now because it could change in between
597 * us evaluating the size and assigning it. This is because we lock and
598 * unlock the page in truncate and fallocate, and then modify the i_size
601 * The barriers are to emulate READ_ONCE, remove that once i_size_read
605 i_size = i_size_read(inode);
607 actual_end = min_t(u64, i_size, end + 1);
610 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
611 nr_pages = min_t(unsigned long, nr_pages,
612 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
615 * we don't want to send crud past the end of i_size through
616 * compression, that's just a waste of CPU time. So, if the
617 * end of the file is before the start of our current
618 * requested range of bytes, we bail out to the uncompressed
619 * cleanup code that can deal with all of this.
621 * It isn't really the fastest way to fix things, but this is a
622 * very uncommon corner.
624 if (actual_end <= start)
625 goto cleanup_and_bail_uncompressed;
627 total_compressed = actual_end - start;
630 * Skip compression for a small file range(<=blocksize) that
631 * isn't an inline extent, since it doesn't save disk space at all.
633 if (total_compressed <= blocksize &&
634 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
635 goto cleanup_and_bail_uncompressed;
638 * For subpage case, we require full page alignment for the sector
640 * Thus we must also check against @actual_end, not just @end.
642 if (blocksize < PAGE_SIZE) {
643 if (!IS_ALIGNED(start, PAGE_SIZE) ||
644 !IS_ALIGNED(round_up(actual_end, blocksize), PAGE_SIZE))
645 goto cleanup_and_bail_uncompressed;
648 total_compressed = min_t(unsigned long, total_compressed,
649 BTRFS_MAX_UNCOMPRESSED);
654 * we do compression for mount -o compress and when the
655 * inode has not been flagged as nocompress. This flag can
656 * change at any time if we discover bad compression ratios.
658 if (inode_need_compress(BTRFS_I(inode), start, end)) {
660 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
662 /* just bail out to the uncompressed code */
667 if (BTRFS_I(inode)->defrag_compress)
668 compress_type = BTRFS_I(inode)->defrag_compress;
669 else if (BTRFS_I(inode)->prop_compress)
670 compress_type = BTRFS_I(inode)->prop_compress;
673 * we need to call clear_page_dirty_for_io on each
674 * page in the range. Otherwise applications with the file
675 * mmap'd can wander in and change the page contents while
676 * we are compressing them.
678 * If the compression fails for any reason, we set the pages
679 * dirty again later on.
681 * Note that the remaining part is redirtied, the start pointer
682 * has moved, the end is the original one.
685 extent_range_clear_dirty_for_io(inode, start, end);
689 /* Compression level is applied here and only here */
690 ret = btrfs_compress_pages(
691 compress_type | (fs_info->compress_level << 4),
692 inode->i_mapping, start,
699 unsigned long offset = offset_in_page(total_compressed);
700 struct page *page = pages[nr_pages - 1];
702 /* zero the tail end of the last page, we might be
703 * sending it down to disk
706 memzero_page(page, offset, PAGE_SIZE - offset);
712 * Check cow_file_range() for why we don't even try to create inline
713 * extent for subpage case.
715 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
716 /* lets try to make an inline extent */
717 if (ret || total_in < actual_end) {
718 /* we didn't compress the entire range, try
719 * to make an uncompressed inline extent.
721 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
722 0, BTRFS_COMPRESS_NONE,
725 /* try making a compressed inline extent */
726 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
728 compress_type, pages,
732 unsigned long clear_flags = EXTENT_DELALLOC |
733 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
734 EXTENT_DO_ACCOUNTING;
735 unsigned long page_error_op;
737 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
740 * inline extent creation worked or returned error,
741 * we don't need to create any more async work items.
742 * Unlock and free up our temp pages.
744 * We use DO_ACCOUNTING here because we need the
745 * delalloc_release_metadata to be done _after_ we drop
746 * our outstanding extent for clearing delalloc for this
749 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
753 PAGE_START_WRITEBACK |
758 * Ensure we only free the compressed pages if we have
759 * them allocated, as we can still reach here with
760 * inode_need_compress() == false.
763 for (i = 0; i < nr_pages; i++) {
764 WARN_ON(pages[i]->mapping);
775 * we aren't doing an inline extent round the compressed size
776 * up to a block size boundary so the allocator does sane
779 total_compressed = ALIGN(total_compressed, blocksize);
782 * one last check to make sure the compression is really a
783 * win, compare the page count read with the blocks on disk,
784 * compression must free at least one sector size
786 total_in = round_up(total_in, fs_info->sectorsize);
787 if (total_compressed + blocksize <= total_in) {
788 compressed_extents++;
791 * The async work queues will take care of doing actual
792 * allocation on disk for these compressed pages, and
793 * will submit them to the elevator.
795 add_async_extent(async_chunk, start, total_in,
796 total_compressed, pages, nr_pages,
799 if (start + total_in < end) {
805 return compressed_extents;
810 * the compression code ran but failed to make things smaller,
811 * free any pages it allocated and our page pointer array
813 for (i = 0; i < nr_pages; i++) {
814 WARN_ON(pages[i]->mapping);
819 total_compressed = 0;
822 /* flag the file so we don't compress in the future */
823 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
824 !(BTRFS_I(inode)->prop_compress)) {
825 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
828 cleanup_and_bail_uncompressed:
830 * No compression, but we still need to write the pages in the file
831 * we've been given so far. redirty the locked page if it corresponds
832 * to our extent and set things up for the async work queue to run
833 * cow_file_range to do the normal delalloc dance.
835 if (async_chunk->locked_page &&
836 (page_offset(async_chunk->locked_page) >= start &&
837 page_offset(async_chunk->locked_page)) <= end) {
838 __set_page_dirty_nobuffers(async_chunk->locked_page);
839 /* unlocked later on in the async handlers */
843 extent_range_redirty_for_io(inode, start, end);
844 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
845 BTRFS_COMPRESS_NONE);
846 compressed_extents++;
848 return compressed_extents;
851 static void free_async_extent_pages(struct async_extent *async_extent)
855 if (!async_extent->pages)
858 for (i = 0; i < async_extent->nr_pages; i++) {
859 WARN_ON(async_extent->pages[i]->mapping);
860 put_page(async_extent->pages[i]);
862 kfree(async_extent->pages);
863 async_extent->nr_pages = 0;
864 async_extent->pages = NULL;
867 static int submit_uncompressed_range(struct btrfs_inode *inode,
868 struct async_extent *async_extent,
869 struct page *locked_page)
871 u64 start = async_extent->start;
872 u64 end = async_extent->start + async_extent->ram_size - 1;
873 unsigned long nr_written = 0;
874 int page_started = 0;
878 * Call cow_file_range() to run the delalloc range directly, since we
879 * won't go to NOCOW or async path again.
881 * Also we call cow_file_range() with @unlock_page == 0, so that we
882 * can directly submit them without interruption.
884 ret = cow_file_range(inode, locked_page, start, end, &page_started,
886 /* Inline extent inserted, page gets unlocked and everything is done */
893 unlock_page(locked_page);
897 ret = extent_write_locked_range(&inode->vfs_inode, start, end);
898 /* All pages will be unlocked, including @locked_page */
904 static int submit_one_async_extent(struct btrfs_inode *inode,
905 struct async_chunk *async_chunk,
906 struct async_extent *async_extent,
909 struct extent_io_tree *io_tree = &inode->io_tree;
910 struct btrfs_root *root = inode->root;
911 struct btrfs_fs_info *fs_info = root->fs_info;
912 struct btrfs_key ins;
913 struct page *locked_page = NULL;
914 struct extent_map *em;
916 u64 start = async_extent->start;
917 u64 end = async_extent->start + async_extent->ram_size - 1;
920 * If async_chunk->locked_page is in the async_extent range, we need to
923 if (async_chunk->locked_page) {
924 u64 locked_page_start = page_offset(async_chunk->locked_page);
925 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
927 if (!(start >= locked_page_end || end <= locked_page_start))
928 locked_page = async_chunk->locked_page;
930 lock_extent(io_tree, start, end);
932 /* We have fall back to uncompressed write */
933 if (!async_extent->pages)
934 return submit_uncompressed_range(inode, async_extent, locked_page);
936 ret = btrfs_reserve_extent(root, async_extent->ram_size,
937 async_extent->compressed_size,
938 async_extent->compressed_size,
939 0, *alloc_hint, &ins, 1, 1);
941 free_async_extent_pages(async_extent);
943 * Here we used to try again by going back to non-compressed
944 * path for ENOSPC. But we can't reserve space even for
945 * compressed size, how could it work for uncompressed size
946 * which requires larger size? So here we directly go error
952 /* Here we're doing allocation and writeback of the compressed pages */
953 em = create_io_em(inode, start,
954 async_extent->ram_size, /* len */
955 start, /* orig_start */
956 ins.objectid, /* block_start */
957 ins.offset, /* block_len */
958 ins.offset, /* orig_block_len */
959 async_extent->ram_size, /* ram_bytes */
960 async_extent->compress_type,
961 BTRFS_ORDERED_COMPRESSED);
964 goto out_free_reserve;
968 ret = btrfs_add_ordered_extent(inode, start, /* file_offset */
969 async_extent->ram_size, /* num_bytes */
970 async_extent->ram_size, /* ram_bytes */
971 ins.objectid, /* disk_bytenr */
972 ins.offset, /* disk_num_bytes */
974 1 << BTRFS_ORDERED_COMPRESSED,
975 async_extent->compress_type);
977 btrfs_drop_extent_cache(inode, start, end, 0);
978 goto out_free_reserve;
980 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
982 /* Clear dirty, set writeback and unlock the pages. */
983 extent_clear_unlock_delalloc(inode, start, end,
984 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
985 PAGE_UNLOCK | PAGE_START_WRITEBACK);
986 if (btrfs_submit_compressed_write(inode, start, /* file_offset */
987 async_extent->ram_size, /* num_bytes */
988 ins.objectid, /* disk_bytenr */
989 ins.offset, /* compressed_len */
990 async_extent->pages, /* compressed_pages */
991 async_extent->nr_pages,
992 async_chunk->write_flags,
993 async_chunk->blkcg_css, true)) {
994 const u64 start = async_extent->start;
995 const u64 end = start + async_extent->ram_size - 1;
997 btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0);
999 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
1000 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1001 free_async_extent_pages(async_extent);
1003 *alloc_hint = ins.objectid + ins.offset;
1004 kfree(async_extent);
1008 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1009 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1011 extent_clear_unlock_delalloc(inode, start, end,
1012 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1013 EXTENT_DELALLOC_NEW |
1014 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1015 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1016 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1017 free_async_extent_pages(async_extent);
1018 kfree(async_extent);
1023 * Phase two of compressed writeback. This is the ordered portion of the code,
1024 * which only gets called in the order the work was queued. We walk all the
1025 * async extents created by compress_file_range and send them down to the disk.
1027 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1029 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
1030 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1031 struct async_extent *async_extent;
1035 while (!list_empty(&async_chunk->extents)) {
1039 async_extent = list_entry(async_chunk->extents.next,
1040 struct async_extent, list);
1041 list_del(&async_extent->list);
1042 extent_start = async_extent->start;
1043 ram_size = async_extent->ram_size;
1045 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1047 btrfs_debug(fs_info,
1048 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1049 inode->root->root_key.objectid,
1050 btrfs_ino(inode), extent_start, ram_size, ret);
1054 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1057 struct extent_map_tree *em_tree = &inode->extent_tree;
1058 struct extent_map *em;
1061 read_lock(&em_tree->lock);
1062 em = search_extent_mapping(em_tree, start, num_bytes);
1065 * if block start isn't an actual block number then find the
1066 * first block in this inode and use that as a hint. If that
1067 * block is also bogus then just don't worry about it.
1069 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1070 free_extent_map(em);
1071 em = search_extent_mapping(em_tree, 0, 0);
1072 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1073 alloc_hint = em->block_start;
1075 free_extent_map(em);
1077 alloc_hint = em->block_start;
1078 free_extent_map(em);
1081 read_unlock(&em_tree->lock);
1087 * when extent_io.c finds a delayed allocation range in the file,
1088 * the call backs end up in this code. The basic idea is to
1089 * allocate extents on disk for the range, and create ordered data structs
1090 * in ram to track those extents.
1092 * locked_page is the page that writepage had locked already. We use
1093 * it to make sure we don't do extra locks or unlocks.
1095 * *page_started is set to one if we unlock locked_page and do everything
1096 * required to start IO on it. It may be clean and already done with
1097 * IO when we return.
1099 static noinline int cow_file_range(struct btrfs_inode *inode,
1100 struct page *locked_page,
1101 u64 start, u64 end, int *page_started,
1102 unsigned long *nr_written, int unlock)
1104 struct btrfs_root *root = inode->root;
1105 struct btrfs_fs_info *fs_info = root->fs_info;
1108 unsigned long ram_size;
1109 u64 cur_alloc_size = 0;
1111 u64 blocksize = fs_info->sectorsize;
1112 struct btrfs_key ins;
1113 struct extent_map *em;
1114 unsigned clear_bits;
1115 unsigned long page_ops;
1116 bool extent_reserved = false;
1119 if (btrfs_is_free_space_inode(inode)) {
1124 num_bytes = ALIGN(end - start + 1, blocksize);
1125 num_bytes = max(blocksize, num_bytes);
1126 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1128 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1131 * Due to the page size limit, for subpage we can only trigger the
1132 * writeback for the dirty sectors of page, that means data writeback
1133 * is doing more writeback than what we want.
1135 * This is especially unexpected for some call sites like fallocate,
1136 * where we only increase i_size after everything is done.
1137 * This means we can trigger inline extent even if we didn't want to.
1138 * So here we skip inline extent creation completely.
1140 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1141 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1144 /* lets try to make an inline extent */
1145 ret = cow_file_range_inline(inode, actual_end, 0,
1146 BTRFS_COMPRESS_NONE, NULL, false);
1149 * We use DO_ACCOUNTING here because we need the
1150 * delalloc_release_metadata to be run _after_ we drop
1151 * our outstanding extent for clearing delalloc for this
1154 extent_clear_unlock_delalloc(inode, start, end,
1156 EXTENT_LOCKED | EXTENT_DELALLOC |
1157 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1158 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1159 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1160 *nr_written = *nr_written +
1161 (end - start + PAGE_SIZE) / PAGE_SIZE;
1164 * locked_page is locked by the caller of
1165 * writepage_delalloc(), not locked by
1166 * __process_pages_contig().
1168 * We can't let __process_pages_contig() to unlock it,
1169 * as it doesn't have any subpage::writers recorded.
1171 * Here we manually unlock the page, since the caller
1172 * can't use page_started to determine if it's an
1173 * inline extent or a compressed extent.
1175 unlock_page(locked_page);
1177 } else if (ret < 0) {
1182 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1183 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1186 * Relocation relies on the relocated extents to have exactly the same
1187 * size as the original extents. Normally writeback for relocation data
1188 * extents follows a NOCOW path because relocation preallocates the
1189 * extents. However, due to an operation such as scrub turning a block
1190 * group to RO mode, it may fallback to COW mode, so we must make sure
1191 * an extent allocated during COW has exactly the requested size and can
1192 * not be split into smaller extents, otherwise relocation breaks and
1193 * fails during the stage where it updates the bytenr of file extent
1196 if (btrfs_is_data_reloc_root(root))
1197 min_alloc_size = num_bytes;
1199 min_alloc_size = fs_info->sectorsize;
1201 while (num_bytes > 0) {
1202 cur_alloc_size = num_bytes;
1203 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1204 min_alloc_size, 0, alloc_hint,
1208 cur_alloc_size = ins.offset;
1209 extent_reserved = true;
1211 ram_size = ins.offset;
1212 em = create_io_em(inode, start, ins.offset, /* len */
1213 start, /* orig_start */
1214 ins.objectid, /* block_start */
1215 ins.offset, /* block_len */
1216 ins.offset, /* orig_block_len */
1217 ram_size, /* ram_bytes */
1218 BTRFS_COMPRESS_NONE, /* compress_type */
1219 BTRFS_ORDERED_REGULAR /* type */);
1224 free_extent_map(em);
1226 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1227 ins.objectid, cur_alloc_size, 0,
1228 1 << BTRFS_ORDERED_REGULAR,
1229 BTRFS_COMPRESS_NONE);
1231 goto out_drop_extent_cache;
1233 if (btrfs_is_data_reloc_root(root)) {
1234 ret = btrfs_reloc_clone_csums(inode, start,
1237 * Only drop cache here, and process as normal.
1239 * We must not allow extent_clear_unlock_delalloc()
1240 * at out_unlock label to free meta of this ordered
1241 * extent, as its meta should be freed by
1242 * btrfs_finish_ordered_io().
1244 * So we must continue until @start is increased to
1245 * skip current ordered extent.
1248 btrfs_drop_extent_cache(inode, start,
1249 start + ram_size - 1, 0);
1252 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1255 * We're not doing compressed IO, don't unlock the first page
1256 * (which the caller expects to stay locked), don't clear any
1257 * dirty bits and don't set any writeback bits
1259 * Do set the Ordered (Private2) bit so we know this page was
1260 * properly setup for writepage.
1262 page_ops = unlock ? PAGE_UNLOCK : 0;
1263 page_ops |= PAGE_SET_ORDERED;
1265 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1267 EXTENT_LOCKED | EXTENT_DELALLOC,
1269 if (num_bytes < cur_alloc_size)
1272 num_bytes -= cur_alloc_size;
1273 alloc_hint = ins.objectid + ins.offset;
1274 start += cur_alloc_size;
1275 extent_reserved = false;
1278 * btrfs_reloc_clone_csums() error, since start is increased
1279 * extent_clear_unlock_delalloc() at out_unlock label won't
1280 * free metadata of current ordered extent, we're OK to exit.
1288 out_drop_extent_cache:
1289 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1291 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1292 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1294 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1295 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1296 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1298 * If we reserved an extent for our delalloc range (or a subrange) and
1299 * failed to create the respective ordered extent, then it means that
1300 * when we reserved the extent we decremented the extent's size from
1301 * the data space_info's bytes_may_use counter and incremented the
1302 * space_info's bytes_reserved counter by the same amount. We must make
1303 * sure extent_clear_unlock_delalloc() does not try to decrement again
1304 * the data space_info's bytes_may_use counter, therefore we do not pass
1305 * it the flag EXTENT_CLEAR_DATA_RESV.
1307 if (extent_reserved) {
1308 extent_clear_unlock_delalloc(inode, start,
1309 start + cur_alloc_size - 1,
1313 start += cur_alloc_size;
1317 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1318 clear_bits | EXTENT_CLEAR_DATA_RESV,
1324 * work queue call back to started compression on a file and pages
1326 static noinline void async_cow_start(struct btrfs_work *work)
1328 struct async_chunk *async_chunk;
1329 int compressed_extents;
1331 async_chunk = container_of(work, struct async_chunk, work);
1333 compressed_extents = compress_file_range(async_chunk);
1334 if (compressed_extents == 0) {
1335 btrfs_add_delayed_iput(async_chunk->inode);
1336 async_chunk->inode = NULL;
1341 * work queue call back to submit previously compressed pages
1343 static noinline void async_cow_submit(struct btrfs_work *work)
1345 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1347 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1348 unsigned long nr_pages;
1350 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1354 * ->inode could be NULL if async_chunk_start has failed to compress,
1355 * in which case we don't have anything to submit, yet we need to
1356 * always adjust ->async_delalloc_pages as its paired with the init
1357 * happening in cow_file_range_async
1359 if (async_chunk->inode)
1360 submit_compressed_extents(async_chunk);
1362 /* atomic_sub_return implies a barrier */
1363 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1365 cond_wake_up_nomb(&fs_info->async_submit_wait);
1368 static noinline void async_cow_free(struct btrfs_work *work)
1370 struct async_chunk *async_chunk;
1371 struct async_cow *async_cow;
1373 async_chunk = container_of(work, struct async_chunk, work);
1374 if (async_chunk->inode)
1375 btrfs_add_delayed_iput(async_chunk->inode);
1376 if (async_chunk->blkcg_css)
1377 css_put(async_chunk->blkcg_css);
1379 async_cow = async_chunk->async_cow;
1380 if (atomic_dec_and_test(&async_cow->num_chunks))
1384 static int cow_file_range_async(struct btrfs_inode *inode,
1385 struct writeback_control *wbc,
1386 struct page *locked_page,
1387 u64 start, u64 end, int *page_started,
1388 unsigned long *nr_written)
1390 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1391 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1392 struct async_cow *ctx;
1393 struct async_chunk *async_chunk;
1394 unsigned long nr_pages;
1396 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1398 bool should_compress;
1400 const unsigned int write_flags = wbc_to_write_flags(wbc);
1402 unlock_extent(&inode->io_tree, start, end);
1404 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1405 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1407 should_compress = false;
1409 should_compress = true;
1412 nofs_flag = memalloc_nofs_save();
1413 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1414 memalloc_nofs_restore(nofs_flag);
1417 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1418 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1419 EXTENT_DO_ACCOUNTING;
1420 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1421 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1423 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1424 clear_bits, page_ops);
1428 async_chunk = ctx->chunks;
1429 atomic_set(&ctx->num_chunks, num_chunks);
1431 for (i = 0; i < num_chunks; i++) {
1432 if (should_compress)
1433 cur_end = min(end, start + SZ_512K - 1);
1438 * igrab is called higher up in the call chain, take only the
1439 * lightweight reference for the callback lifetime
1441 ihold(&inode->vfs_inode);
1442 async_chunk[i].async_cow = ctx;
1443 async_chunk[i].inode = &inode->vfs_inode;
1444 async_chunk[i].start = start;
1445 async_chunk[i].end = cur_end;
1446 async_chunk[i].write_flags = write_flags;
1447 INIT_LIST_HEAD(&async_chunk[i].extents);
1450 * The locked_page comes all the way from writepage and its
1451 * the original page we were actually given. As we spread
1452 * this large delalloc region across multiple async_chunk
1453 * structs, only the first struct needs a pointer to locked_page
1455 * This way we don't need racey decisions about who is supposed
1460 * Depending on the compressibility, the pages might or
1461 * might not go through async. We want all of them to
1462 * be accounted against wbc once. Let's do it here
1463 * before the paths diverge. wbc accounting is used
1464 * only for foreign writeback detection and doesn't
1465 * need full accuracy. Just account the whole thing
1466 * against the first page.
1468 wbc_account_cgroup_owner(wbc, locked_page,
1470 async_chunk[i].locked_page = locked_page;
1473 async_chunk[i].locked_page = NULL;
1476 if (blkcg_css != blkcg_root_css) {
1478 async_chunk[i].blkcg_css = blkcg_css;
1480 async_chunk[i].blkcg_css = NULL;
1483 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1484 async_cow_submit, async_cow_free);
1486 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1487 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1489 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1491 *nr_written += nr_pages;
1492 start = cur_end + 1;
1498 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1499 struct page *locked_page, u64 start,
1500 u64 end, int *page_started,
1501 unsigned long *nr_written)
1505 ret = cow_file_range(inode, locked_page, start, end, page_started,
1513 __set_page_dirty_nobuffers(locked_page);
1514 account_page_redirty(locked_page);
1515 extent_write_locked_range(&inode->vfs_inode, start, end);
1521 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1522 u64 bytenr, u64 num_bytes)
1524 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1525 struct btrfs_ordered_sum *sums;
1529 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1530 bytenr + num_bytes - 1, &list, 0);
1531 if (ret == 0 && list_empty(&list))
1534 while (!list_empty(&list)) {
1535 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1536 list_del(&sums->list);
1544 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1545 const u64 start, const u64 end,
1546 int *page_started, unsigned long *nr_written)
1548 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1549 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1550 const u64 range_bytes = end + 1 - start;
1551 struct extent_io_tree *io_tree = &inode->io_tree;
1552 u64 range_start = start;
1556 * If EXTENT_NORESERVE is set it means that when the buffered write was
1557 * made we had not enough available data space and therefore we did not
1558 * reserve data space for it, since we though we could do NOCOW for the
1559 * respective file range (either there is prealloc extent or the inode
1560 * has the NOCOW bit set).
1562 * However when we need to fallback to COW mode (because for example the
1563 * block group for the corresponding extent was turned to RO mode by a
1564 * scrub or relocation) we need to do the following:
1566 * 1) We increment the bytes_may_use counter of the data space info.
1567 * If COW succeeds, it allocates a new data extent and after doing
1568 * that it decrements the space info's bytes_may_use counter and
1569 * increments its bytes_reserved counter by the same amount (we do
1570 * this at btrfs_add_reserved_bytes()). So we need to increment the
1571 * bytes_may_use counter to compensate (when space is reserved at
1572 * buffered write time, the bytes_may_use counter is incremented);
1574 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1575 * that if the COW path fails for any reason, it decrements (through
1576 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1577 * data space info, which we incremented in the step above.
1579 * If we need to fallback to cow and the inode corresponds to a free
1580 * space cache inode or an inode of the data relocation tree, we must
1581 * also increment bytes_may_use of the data space_info for the same
1582 * reason. Space caches and relocated data extents always get a prealloc
1583 * extent for them, however scrub or balance may have set the block
1584 * group that contains that extent to RO mode and therefore force COW
1585 * when starting writeback.
1587 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1588 EXTENT_NORESERVE, 0);
1589 if (count > 0 || is_space_ino || is_reloc_ino) {
1591 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1592 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1594 if (is_space_ino || is_reloc_ino)
1595 bytes = range_bytes;
1597 spin_lock(&sinfo->lock);
1598 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1599 spin_unlock(&sinfo->lock);
1602 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1606 return cow_file_range(inode, locked_page, start, end, page_started,
1611 * when nowcow writeback call back. This checks for snapshots or COW copies
1612 * of the extents that exist in the file, and COWs the file as required.
1614 * If no cow copies or snapshots exist, we write directly to the existing
1617 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1618 struct page *locked_page,
1619 const u64 start, const u64 end,
1621 unsigned long *nr_written)
1623 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1624 struct btrfs_root *root = inode->root;
1625 struct btrfs_path *path;
1626 u64 cow_start = (u64)-1;
1627 u64 cur_offset = start;
1629 bool check_prev = true;
1630 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1631 u64 ino = btrfs_ino(inode);
1633 u64 disk_bytenr = 0;
1634 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1636 path = btrfs_alloc_path();
1638 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1639 EXTENT_LOCKED | EXTENT_DELALLOC |
1640 EXTENT_DO_ACCOUNTING |
1641 EXTENT_DEFRAG, PAGE_UNLOCK |
1642 PAGE_START_WRITEBACK |
1643 PAGE_END_WRITEBACK);
1648 struct btrfs_key found_key;
1649 struct btrfs_file_extent_item *fi;
1650 struct extent_buffer *leaf;
1660 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1666 * If there is no extent for our range when doing the initial
1667 * search, then go back to the previous slot as it will be the
1668 * one containing the search offset
1670 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1671 leaf = path->nodes[0];
1672 btrfs_item_key_to_cpu(leaf, &found_key,
1673 path->slots[0] - 1);
1674 if (found_key.objectid == ino &&
1675 found_key.type == BTRFS_EXTENT_DATA_KEY)
1680 /* Go to next leaf if we have exhausted the current one */
1681 leaf = path->nodes[0];
1682 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1683 ret = btrfs_next_leaf(root, path);
1685 if (cow_start != (u64)-1)
1686 cur_offset = cow_start;
1691 leaf = path->nodes[0];
1694 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1696 /* Didn't find anything for our INO */
1697 if (found_key.objectid > ino)
1700 * Keep searching until we find an EXTENT_ITEM or there are no
1701 * more extents for this inode
1703 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1704 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1709 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1710 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1711 found_key.offset > end)
1715 * If the found extent starts after requested offset, then
1716 * adjust extent_end to be right before this extent begins
1718 if (found_key.offset > cur_offset) {
1719 extent_end = found_key.offset;
1725 * Found extent which begins before our range and potentially
1728 fi = btrfs_item_ptr(leaf, path->slots[0],
1729 struct btrfs_file_extent_item);
1730 extent_type = btrfs_file_extent_type(leaf, fi);
1732 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1733 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1734 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1735 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1736 extent_offset = btrfs_file_extent_offset(leaf, fi);
1737 extent_end = found_key.offset +
1738 btrfs_file_extent_num_bytes(leaf, fi);
1740 btrfs_file_extent_disk_num_bytes(leaf, fi);
1742 * If the extent we got ends before our current offset,
1743 * skip to the next extent.
1745 if (extent_end <= cur_offset) {
1750 if (disk_bytenr == 0)
1752 /* Skip compressed/encrypted/encoded extents */
1753 if (btrfs_file_extent_compression(leaf, fi) ||
1754 btrfs_file_extent_encryption(leaf, fi) ||
1755 btrfs_file_extent_other_encoding(leaf, fi))
1758 * If extent is created before the last volume's snapshot
1759 * this implies the extent is shared, hence we can't do
1760 * nocow. This is the same check as in
1761 * btrfs_cross_ref_exist but without calling
1762 * btrfs_search_slot.
1764 if (!freespace_inode &&
1765 btrfs_file_extent_generation(leaf, fi) <=
1766 btrfs_root_last_snapshot(&root->root_item))
1768 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1772 * The following checks can be expensive, as they need to
1773 * take other locks and do btree or rbtree searches, so
1774 * release the path to avoid blocking other tasks for too
1777 btrfs_release_path(path);
1779 ret = btrfs_cross_ref_exist(root, ino,
1781 extent_offset, disk_bytenr, false);
1784 * ret could be -EIO if the above fails to read
1788 if (cow_start != (u64)-1)
1789 cur_offset = cow_start;
1793 WARN_ON_ONCE(freespace_inode);
1796 disk_bytenr += extent_offset;
1797 disk_bytenr += cur_offset - found_key.offset;
1798 num_bytes = min(end + 1, extent_end) - cur_offset;
1800 * If there are pending snapshots for this root, we
1801 * fall into common COW way
1803 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1806 * force cow if csum exists in the range.
1807 * this ensure that csum for a given extent are
1808 * either valid or do not exist.
1810 ret = csum_exist_in_range(fs_info, disk_bytenr,
1814 * ret could be -EIO if the above fails to read
1818 if (cow_start != (u64)-1)
1819 cur_offset = cow_start;
1822 WARN_ON_ONCE(freespace_inode);
1825 /* If the extent's block group is RO, we must COW */
1826 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1829 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1830 extent_end = found_key.offset + ram_bytes;
1831 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1832 /* Skip extents outside of our requested range */
1833 if (extent_end <= start) {
1838 /* If this triggers then we have a memory corruption */
1843 * If nocow is false then record the beginning of the range
1844 * that needs to be COWed
1847 if (cow_start == (u64)-1)
1848 cow_start = cur_offset;
1849 cur_offset = extent_end;
1850 if (cur_offset > end)
1852 if (!path->nodes[0])
1859 * COW range from cow_start to found_key.offset - 1. As the key
1860 * will contain the beginning of the first extent that can be
1861 * NOCOW, following one which needs to be COW'ed
1863 if (cow_start != (u64)-1) {
1864 ret = fallback_to_cow(inode, locked_page,
1865 cow_start, found_key.offset - 1,
1866 page_started, nr_written);
1869 cow_start = (u64)-1;
1872 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1873 u64 orig_start = found_key.offset - extent_offset;
1874 struct extent_map *em;
1876 em = create_io_em(inode, cur_offset, num_bytes,
1878 disk_bytenr, /* block_start */
1879 num_bytes, /* block_len */
1880 disk_num_bytes, /* orig_block_len */
1881 ram_bytes, BTRFS_COMPRESS_NONE,
1882 BTRFS_ORDERED_PREALLOC);
1887 free_extent_map(em);
1888 ret = btrfs_add_ordered_extent(inode,
1889 cur_offset, num_bytes, num_bytes,
1890 disk_bytenr, num_bytes, 0,
1891 1 << BTRFS_ORDERED_PREALLOC,
1892 BTRFS_COMPRESS_NONE);
1894 btrfs_drop_extent_cache(inode, cur_offset,
1895 cur_offset + num_bytes - 1,
1900 ret = btrfs_add_ordered_extent(inode, cur_offset,
1901 num_bytes, num_bytes,
1902 disk_bytenr, num_bytes,
1904 1 << BTRFS_ORDERED_NOCOW,
1905 BTRFS_COMPRESS_NONE);
1911 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1914 if (btrfs_is_data_reloc_root(root))
1916 * Error handled later, as we must prevent
1917 * extent_clear_unlock_delalloc() in error handler
1918 * from freeing metadata of created ordered extent.
1920 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1923 extent_clear_unlock_delalloc(inode, cur_offset,
1924 cur_offset + num_bytes - 1,
1925 locked_page, EXTENT_LOCKED |
1927 EXTENT_CLEAR_DATA_RESV,
1928 PAGE_UNLOCK | PAGE_SET_ORDERED);
1930 cur_offset = extent_end;
1933 * btrfs_reloc_clone_csums() error, now we're OK to call error
1934 * handler, as metadata for created ordered extent will only
1935 * be freed by btrfs_finish_ordered_io().
1939 if (cur_offset > end)
1942 btrfs_release_path(path);
1944 if (cur_offset <= end && cow_start == (u64)-1)
1945 cow_start = cur_offset;
1947 if (cow_start != (u64)-1) {
1949 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1950 page_started, nr_written);
1957 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1959 if (ret && cur_offset < end)
1960 extent_clear_unlock_delalloc(inode, cur_offset, end,
1961 locked_page, EXTENT_LOCKED |
1962 EXTENT_DELALLOC | EXTENT_DEFRAG |
1963 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1964 PAGE_START_WRITEBACK |
1965 PAGE_END_WRITEBACK);
1966 btrfs_free_path(path);
1970 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1972 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1973 if (inode->defrag_bytes &&
1974 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1983 * Function to process delayed allocation (create CoW) for ranges which are
1984 * being touched for the first time.
1986 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1987 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1988 struct writeback_control *wbc)
1991 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1994 * The range must cover part of the @locked_page, or the returned
1995 * @page_started can confuse the caller.
1997 ASSERT(!(end <= page_offset(locked_page) ||
1998 start >= page_offset(locked_page) + PAGE_SIZE));
2000 if (should_nocow(inode, start, end)) {
2002 * Normally on a zoned device we're only doing COW writes, but
2003 * in case of relocation on a zoned filesystem we have taken
2004 * precaution, that we're only writing sequentially. It's safe
2005 * to use run_delalloc_nocow() here, like for regular
2006 * preallocated inodes.
2008 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2009 ret = run_delalloc_nocow(inode, locked_page, start, end,
2010 page_started, nr_written);
2011 } else if (!btrfs_inode_can_compress(inode) ||
2012 !inode_need_compress(inode, start, end)) {
2014 ret = run_delalloc_zoned(inode, locked_page, start, end,
2015 page_started, nr_written);
2017 ret = cow_file_range(inode, locked_page, start, end,
2018 page_started, nr_written, 1);
2020 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2021 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2022 page_started, nr_written);
2026 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2031 void btrfs_split_delalloc_extent(struct inode *inode,
2032 struct extent_state *orig, u64 split)
2036 /* not delalloc, ignore it */
2037 if (!(orig->state & EXTENT_DELALLOC))
2040 size = orig->end - orig->start + 1;
2041 if (size > BTRFS_MAX_EXTENT_SIZE) {
2046 * See the explanation in btrfs_merge_delalloc_extent, the same
2047 * applies here, just in reverse.
2049 new_size = orig->end - split + 1;
2050 num_extents = count_max_extents(new_size);
2051 new_size = split - orig->start;
2052 num_extents += count_max_extents(new_size);
2053 if (count_max_extents(size) >= num_extents)
2057 spin_lock(&BTRFS_I(inode)->lock);
2058 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2059 spin_unlock(&BTRFS_I(inode)->lock);
2063 * Handle merged delayed allocation extents so we can keep track of new extents
2064 * that are just merged onto old extents, such as when we are doing sequential
2065 * writes, so we can properly account for the metadata space we'll need.
2067 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2068 struct extent_state *other)
2070 u64 new_size, old_size;
2073 /* not delalloc, ignore it */
2074 if (!(other->state & EXTENT_DELALLOC))
2077 if (new->start > other->start)
2078 new_size = new->end - other->start + 1;
2080 new_size = other->end - new->start + 1;
2082 /* we're not bigger than the max, unreserve the space and go */
2083 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2084 spin_lock(&BTRFS_I(inode)->lock);
2085 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2086 spin_unlock(&BTRFS_I(inode)->lock);
2091 * We have to add up either side to figure out how many extents were
2092 * accounted for before we merged into one big extent. If the number of
2093 * extents we accounted for is <= the amount we need for the new range
2094 * then we can return, otherwise drop. Think of it like this
2098 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2099 * need 2 outstanding extents, on one side we have 1 and the other side
2100 * we have 1 so they are == and we can return. But in this case
2102 * [MAX_SIZE+4k][MAX_SIZE+4k]
2104 * Each range on their own accounts for 2 extents, but merged together
2105 * they are only 3 extents worth of accounting, so we need to drop in
2108 old_size = other->end - other->start + 1;
2109 num_extents = count_max_extents(old_size);
2110 old_size = new->end - new->start + 1;
2111 num_extents += count_max_extents(old_size);
2112 if (count_max_extents(new_size) >= num_extents)
2115 spin_lock(&BTRFS_I(inode)->lock);
2116 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2117 spin_unlock(&BTRFS_I(inode)->lock);
2120 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2121 struct inode *inode)
2123 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2125 spin_lock(&root->delalloc_lock);
2126 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2127 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2128 &root->delalloc_inodes);
2129 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2130 &BTRFS_I(inode)->runtime_flags);
2131 root->nr_delalloc_inodes++;
2132 if (root->nr_delalloc_inodes == 1) {
2133 spin_lock(&fs_info->delalloc_root_lock);
2134 BUG_ON(!list_empty(&root->delalloc_root));
2135 list_add_tail(&root->delalloc_root,
2136 &fs_info->delalloc_roots);
2137 spin_unlock(&fs_info->delalloc_root_lock);
2140 spin_unlock(&root->delalloc_lock);
2144 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2145 struct btrfs_inode *inode)
2147 struct btrfs_fs_info *fs_info = root->fs_info;
2149 if (!list_empty(&inode->delalloc_inodes)) {
2150 list_del_init(&inode->delalloc_inodes);
2151 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2152 &inode->runtime_flags);
2153 root->nr_delalloc_inodes--;
2154 if (!root->nr_delalloc_inodes) {
2155 ASSERT(list_empty(&root->delalloc_inodes));
2156 spin_lock(&fs_info->delalloc_root_lock);
2157 BUG_ON(list_empty(&root->delalloc_root));
2158 list_del_init(&root->delalloc_root);
2159 spin_unlock(&fs_info->delalloc_root_lock);
2164 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2165 struct btrfs_inode *inode)
2167 spin_lock(&root->delalloc_lock);
2168 __btrfs_del_delalloc_inode(root, inode);
2169 spin_unlock(&root->delalloc_lock);
2173 * Properly track delayed allocation bytes in the inode and to maintain the
2174 * list of inodes that have pending delalloc work to be done.
2176 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2179 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2181 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2184 * set_bit and clear bit hooks normally require _irqsave/restore
2185 * but in this case, we are only testing for the DELALLOC
2186 * bit, which is only set or cleared with irqs on
2188 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2189 struct btrfs_root *root = BTRFS_I(inode)->root;
2190 u64 len = state->end + 1 - state->start;
2191 u32 num_extents = count_max_extents(len);
2192 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2194 spin_lock(&BTRFS_I(inode)->lock);
2195 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2196 spin_unlock(&BTRFS_I(inode)->lock);
2198 /* For sanity tests */
2199 if (btrfs_is_testing(fs_info))
2202 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2203 fs_info->delalloc_batch);
2204 spin_lock(&BTRFS_I(inode)->lock);
2205 BTRFS_I(inode)->delalloc_bytes += len;
2206 if (*bits & EXTENT_DEFRAG)
2207 BTRFS_I(inode)->defrag_bytes += len;
2208 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2209 &BTRFS_I(inode)->runtime_flags))
2210 btrfs_add_delalloc_inodes(root, inode);
2211 spin_unlock(&BTRFS_I(inode)->lock);
2214 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2215 (*bits & EXTENT_DELALLOC_NEW)) {
2216 spin_lock(&BTRFS_I(inode)->lock);
2217 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2219 spin_unlock(&BTRFS_I(inode)->lock);
2224 * Once a range is no longer delalloc this function ensures that proper
2225 * accounting happens.
2227 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2228 struct extent_state *state, unsigned *bits)
2230 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2231 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2232 u64 len = state->end + 1 - state->start;
2233 u32 num_extents = count_max_extents(len);
2235 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2236 spin_lock(&inode->lock);
2237 inode->defrag_bytes -= len;
2238 spin_unlock(&inode->lock);
2242 * set_bit and clear bit hooks normally require _irqsave/restore
2243 * but in this case, we are only testing for the DELALLOC
2244 * bit, which is only set or cleared with irqs on
2246 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2247 struct btrfs_root *root = inode->root;
2248 bool do_list = !btrfs_is_free_space_inode(inode);
2250 spin_lock(&inode->lock);
2251 btrfs_mod_outstanding_extents(inode, -num_extents);
2252 spin_unlock(&inode->lock);
2255 * We don't reserve metadata space for space cache inodes so we
2256 * don't need to call delalloc_release_metadata if there is an
2259 if (*bits & EXTENT_CLEAR_META_RESV &&
2260 root != fs_info->tree_root)
2261 btrfs_delalloc_release_metadata(inode, len, false);
2263 /* For sanity tests. */
2264 if (btrfs_is_testing(fs_info))
2267 if (!btrfs_is_data_reloc_root(root) &&
2268 do_list && !(state->state & EXTENT_NORESERVE) &&
2269 (*bits & EXTENT_CLEAR_DATA_RESV))
2270 btrfs_free_reserved_data_space_noquota(fs_info, len);
2272 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2273 fs_info->delalloc_batch);
2274 spin_lock(&inode->lock);
2275 inode->delalloc_bytes -= len;
2276 if (do_list && inode->delalloc_bytes == 0 &&
2277 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2278 &inode->runtime_flags))
2279 btrfs_del_delalloc_inode(root, inode);
2280 spin_unlock(&inode->lock);
2283 if ((state->state & EXTENT_DELALLOC_NEW) &&
2284 (*bits & EXTENT_DELALLOC_NEW)) {
2285 spin_lock(&inode->lock);
2286 ASSERT(inode->new_delalloc_bytes >= len);
2287 inode->new_delalloc_bytes -= len;
2288 if (*bits & EXTENT_ADD_INODE_BYTES)
2289 inode_add_bytes(&inode->vfs_inode, len);
2290 spin_unlock(&inode->lock);
2295 * in order to insert checksums into the metadata in large chunks,
2296 * we wait until bio submission time. All the pages in the bio are
2297 * checksummed and sums are attached onto the ordered extent record.
2299 * At IO completion time the cums attached on the ordered extent record
2300 * are inserted into the btree
2302 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2303 u64 dio_file_offset)
2305 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2309 * Split an extent_map at [start, start + len]
2311 * This function is intended to be used only for extract_ordered_extent().
2313 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2316 struct extent_map_tree *em_tree = &inode->extent_tree;
2317 struct extent_map *em;
2318 struct extent_map *split_pre = NULL;
2319 struct extent_map *split_mid = NULL;
2320 struct extent_map *split_post = NULL;
2322 unsigned long flags;
2325 if (pre == 0 && post == 0)
2328 split_pre = alloc_extent_map();
2330 split_mid = alloc_extent_map();
2332 split_post = alloc_extent_map();
2333 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2338 ASSERT(pre + post < len);
2340 lock_extent(&inode->io_tree, start, start + len - 1);
2341 write_lock(&em_tree->lock);
2342 em = lookup_extent_mapping(em_tree, start, len);
2348 ASSERT(em->len == len);
2349 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2350 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2351 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2352 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2353 ASSERT(!list_empty(&em->list));
2356 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2358 /* First, replace the em with a new extent_map starting from * em->start */
2359 split_pre->start = em->start;
2360 split_pre->len = (pre ? pre : em->len - post);
2361 split_pre->orig_start = split_pre->start;
2362 split_pre->block_start = em->block_start;
2363 split_pre->block_len = split_pre->len;
2364 split_pre->orig_block_len = split_pre->block_len;
2365 split_pre->ram_bytes = split_pre->len;
2366 split_pre->flags = flags;
2367 split_pre->compress_type = em->compress_type;
2368 split_pre->generation = em->generation;
2370 replace_extent_mapping(em_tree, em, split_pre, 1);
2373 * Now we only have an extent_map at:
2374 * [em->start, em->start + pre] if pre != 0
2375 * [em->start, em->start + em->len - post] if pre == 0
2379 /* Insert the middle extent_map */
2380 split_mid->start = em->start + pre;
2381 split_mid->len = em->len - pre - post;
2382 split_mid->orig_start = split_mid->start;
2383 split_mid->block_start = em->block_start + pre;
2384 split_mid->block_len = split_mid->len;
2385 split_mid->orig_block_len = split_mid->block_len;
2386 split_mid->ram_bytes = split_mid->len;
2387 split_mid->flags = flags;
2388 split_mid->compress_type = em->compress_type;
2389 split_mid->generation = em->generation;
2390 add_extent_mapping(em_tree, split_mid, 1);
2394 split_post->start = em->start + em->len - post;
2395 split_post->len = post;
2396 split_post->orig_start = split_post->start;
2397 split_post->block_start = em->block_start + em->len - post;
2398 split_post->block_len = split_post->len;
2399 split_post->orig_block_len = split_post->block_len;
2400 split_post->ram_bytes = split_post->len;
2401 split_post->flags = flags;
2402 split_post->compress_type = em->compress_type;
2403 split_post->generation = em->generation;
2404 add_extent_mapping(em_tree, split_post, 1);
2408 free_extent_map(em);
2409 /* Once for the tree */
2410 free_extent_map(em);
2413 write_unlock(&em_tree->lock);
2414 unlock_extent(&inode->io_tree, start, start + len - 1);
2416 free_extent_map(split_pre);
2417 free_extent_map(split_mid);
2418 free_extent_map(split_post);
2423 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2424 struct bio *bio, loff_t file_offset)
2426 struct btrfs_ordered_extent *ordered;
2427 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2429 u64 len = bio->bi_iter.bi_size;
2430 u64 end = start + len;
2435 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2436 if (WARN_ON_ONCE(!ordered))
2437 return BLK_STS_IOERR;
2439 /* No need to split */
2440 if (ordered->disk_num_bytes == len)
2443 /* We cannot split once end_bio'd ordered extent */
2444 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2449 /* We cannot split a compressed ordered extent */
2450 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2455 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2456 /* bio must be in one ordered extent */
2457 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2462 /* Checksum list should be empty */
2463 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2468 file_len = ordered->num_bytes;
2469 pre = start - ordered->disk_bytenr;
2470 post = ordered_end - end;
2472 ret = btrfs_split_ordered_extent(ordered, pre, post);
2475 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2478 btrfs_put_ordered_extent(ordered);
2480 return errno_to_blk_status(ret);
2484 * extent_io.c submission hook. This does the right thing for csum calculation
2485 * on write, or reading the csums from the tree before a read.
2487 * Rules about async/sync submit,
2488 * a) read: sync submit
2490 * b) write without checksum: sync submit
2492 * c) write with checksum:
2493 * c-1) if bio is issued by fsync: sync submit
2494 * (sync_writers != 0)
2496 * c-2) if root is reloc root: sync submit
2497 * (only in case of buffered IO)
2499 * c-3) otherwise: async submit
2501 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2502 int mirror_num, unsigned long bio_flags)
2505 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2506 struct btrfs_root *root = BTRFS_I(inode)->root;
2507 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2508 blk_status_t ret = 0;
2510 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2512 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2513 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2515 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2516 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2518 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2519 struct page *page = bio_first_bvec_all(bio)->bv_page;
2520 loff_t file_offset = page_offset(page);
2522 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2527 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2528 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2532 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2534 * btrfs_submit_compressed_read will handle completing
2535 * the bio if there were any errors, so just return
2538 ret = btrfs_submit_compressed_read(inode, bio,
2544 * Lookup bio sums does extra checks around whether we
2545 * need to csum or not, which is why we ignore skip_sum
2548 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2553 } else if (async && !skip_sum) {
2554 /* csum items have already been cloned */
2555 if (btrfs_is_data_reloc_root(root))
2557 /* we're doing a write, do the async checksumming */
2558 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2559 0, btrfs_submit_bio_start);
2561 } else if (!skip_sum) {
2562 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2568 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2572 bio->bi_status = ret;
2580 * given a list of ordered sums record them in the inode. This happens
2581 * at IO completion time based on sums calculated at bio submission time.
2583 static int add_pending_csums(struct btrfs_trans_handle *trans,
2584 struct list_head *list)
2586 struct btrfs_ordered_sum *sum;
2587 struct btrfs_root *csum_root = NULL;
2590 list_for_each_entry(sum, list, list) {
2591 trans->adding_csums = true;
2593 csum_root = btrfs_csum_root(trans->fs_info,
2595 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2596 trans->adding_csums = false;
2603 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2606 struct extent_state **cached_state)
2608 u64 search_start = start;
2609 const u64 end = start + len - 1;
2611 while (search_start < end) {
2612 const u64 search_len = end - search_start + 1;
2613 struct extent_map *em;
2617 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2621 if (em->block_start != EXTENT_MAP_HOLE)
2625 if (em->start < search_start)
2626 em_len -= search_start - em->start;
2627 if (em_len > search_len)
2628 em_len = search_len;
2630 ret = set_extent_bit(&inode->io_tree, search_start,
2631 search_start + em_len - 1,
2632 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2635 search_start = extent_map_end(em);
2636 free_extent_map(em);
2643 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2644 unsigned int extra_bits,
2645 struct extent_state **cached_state)
2647 WARN_ON(PAGE_ALIGNED(end));
2649 if (start >= i_size_read(&inode->vfs_inode) &&
2650 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2652 * There can't be any extents following eof in this case so just
2653 * set the delalloc new bit for the range directly.
2655 extra_bits |= EXTENT_DELALLOC_NEW;
2659 ret = btrfs_find_new_delalloc_bytes(inode, start,
2666 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2670 /* see btrfs_writepage_start_hook for details on why this is required */
2671 struct btrfs_writepage_fixup {
2673 struct inode *inode;
2674 struct btrfs_work work;
2677 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2679 struct btrfs_writepage_fixup *fixup;
2680 struct btrfs_ordered_extent *ordered;
2681 struct extent_state *cached_state = NULL;
2682 struct extent_changeset *data_reserved = NULL;
2684 struct btrfs_inode *inode;
2688 bool free_delalloc_space = true;
2690 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2692 inode = BTRFS_I(fixup->inode);
2693 page_start = page_offset(page);
2694 page_end = page_offset(page) + PAGE_SIZE - 1;
2697 * This is similar to page_mkwrite, we need to reserve the space before
2698 * we take the page lock.
2700 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2706 * Before we queued this fixup, we took a reference on the page.
2707 * page->mapping may go NULL, but it shouldn't be moved to a different
2710 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2712 * Unfortunately this is a little tricky, either
2714 * 1) We got here and our page had already been dealt with and
2715 * we reserved our space, thus ret == 0, so we need to just
2716 * drop our space reservation and bail. This can happen the
2717 * first time we come into the fixup worker, or could happen
2718 * while waiting for the ordered extent.
2719 * 2) Our page was already dealt with, but we happened to get an
2720 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2721 * this case we obviously don't have anything to release, but
2722 * because the page was already dealt with we don't want to
2723 * mark the page with an error, so make sure we're resetting
2724 * ret to 0. This is why we have this check _before_ the ret
2725 * check, because we do not want to have a surprise ENOSPC
2726 * when the page was already properly dealt with.
2729 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2730 btrfs_delalloc_release_space(inode, data_reserved,
2731 page_start, PAGE_SIZE,
2739 * We can't mess with the page state unless it is locked, so now that
2740 * it is locked bail if we failed to make our space reservation.
2745 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2747 /* already ordered? We're done */
2748 if (PageOrdered(page))
2751 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2753 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2756 btrfs_start_ordered_extent(ordered, 1);
2757 btrfs_put_ordered_extent(ordered);
2761 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2767 * Everything went as planned, we're now the owner of a dirty page with
2768 * delayed allocation bits set and space reserved for our COW
2771 * The page was dirty when we started, nothing should have cleaned it.
2773 BUG_ON(!PageDirty(page));
2774 free_delalloc_space = false;
2776 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2777 if (free_delalloc_space)
2778 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2780 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2785 * We hit ENOSPC or other errors. Update the mapping and page
2786 * to reflect the errors and clean the page.
2788 mapping_set_error(page->mapping, ret);
2789 end_extent_writepage(page, ret, page_start, page_end);
2790 clear_page_dirty_for_io(page);
2793 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2797 extent_changeset_free(data_reserved);
2799 * As a precaution, do a delayed iput in case it would be the last iput
2800 * that could need flushing space. Recursing back to fixup worker would
2803 btrfs_add_delayed_iput(&inode->vfs_inode);
2807 * There are a few paths in the higher layers of the kernel that directly
2808 * set the page dirty bit without asking the filesystem if it is a
2809 * good idea. This causes problems because we want to make sure COW
2810 * properly happens and the data=ordered rules are followed.
2812 * In our case any range that doesn't have the ORDERED bit set
2813 * hasn't been properly setup for IO. We kick off an async process
2814 * to fix it up. The async helper will wait for ordered extents, set
2815 * the delalloc bit and make it safe to write the page.
2817 int btrfs_writepage_cow_fixup(struct page *page)
2819 struct inode *inode = page->mapping->host;
2820 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2821 struct btrfs_writepage_fixup *fixup;
2823 /* This page has ordered extent covering it already */
2824 if (PageOrdered(page))
2828 * PageChecked is set below when we create a fixup worker for this page,
2829 * don't try to create another one if we're already PageChecked()
2831 * The extent_io writepage code will redirty the page if we send back
2834 if (PageChecked(page))
2837 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2842 * We are already holding a reference to this inode from
2843 * write_cache_pages. We need to hold it because the space reservation
2844 * takes place outside of the page lock, and we can't trust
2845 * page->mapping outside of the page lock.
2848 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2850 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2852 fixup->inode = inode;
2853 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2858 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2859 struct btrfs_inode *inode, u64 file_pos,
2860 struct btrfs_file_extent_item *stack_fi,
2861 const bool update_inode_bytes,
2862 u64 qgroup_reserved)
2864 struct btrfs_root *root = inode->root;
2865 const u64 sectorsize = root->fs_info->sectorsize;
2866 struct btrfs_path *path;
2867 struct extent_buffer *leaf;
2868 struct btrfs_key ins;
2869 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2870 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2871 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2872 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2873 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2874 struct btrfs_drop_extents_args drop_args = { 0 };
2877 path = btrfs_alloc_path();
2882 * we may be replacing one extent in the tree with another.
2883 * The new extent is pinned in the extent map, and we don't want
2884 * to drop it from the cache until it is completely in the btree.
2886 * So, tell btrfs_drop_extents to leave this extent in the cache.
2887 * the caller is expected to unpin it and allow it to be merged
2890 drop_args.path = path;
2891 drop_args.start = file_pos;
2892 drop_args.end = file_pos + num_bytes;
2893 drop_args.replace_extent = true;
2894 drop_args.extent_item_size = sizeof(*stack_fi);
2895 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2899 if (!drop_args.extent_inserted) {
2900 ins.objectid = btrfs_ino(inode);
2901 ins.offset = file_pos;
2902 ins.type = BTRFS_EXTENT_DATA_KEY;
2904 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2909 leaf = path->nodes[0];
2910 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2911 write_extent_buffer(leaf, stack_fi,
2912 btrfs_item_ptr_offset(leaf, path->slots[0]),
2913 sizeof(struct btrfs_file_extent_item));
2915 btrfs_mark_buffer_dirty(leaf);
2916 btrfs_release_path(path);
2919 * If we dropped an inline extent here, we know the range where it is
2920 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2921 * number of bytes only for that range containing the inline extent.
2922 * The remaining of the range will be processed when clearning the
2923 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2925 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2926 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2928 inline_size = drop_args.bytes_found - inline_size;
2929 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2930 drop_args.bytes_found -= inline_size;
2931 num_bytes -= sectorsize;
2934 if (update_inode_bytes)
2935 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2937 ins.objectid = disk_bytenr;
2938 ins.offset = disk_num_bytes;
2939 ins.type = BTRFS_EXTENT_ITEM_KEY;
2941 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2945 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2947 qgroup_reserved, &ins);
2949 btrfs_free_path(path);
2954 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2957 struct btrfs_block_group *cache;
2959 cache = btrfs_lookup_block_group(fs_info, start);
2962 spin_lock(&cache->lock);
2963 cache->delalloc_bytes -= len;
2964 spin_unlock(&cache->lock);
2966 btrfs_put_block_group(cache);
2969 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2970 struct btrfs_ordered_extent *oe)
2972 struct btrfs_file_extent_item stack_fi;
2973 bool update_inode_bytes;
2974 u64 num_bytes = oe->num_bytes;
2975 u64 ram_bytes = oe->ram_bytes;
2977 memset(&stack_fi, 0, sizeof(stack_fi));
2978 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2979 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2980 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2981 oe->disk_num_bytes);
2982 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2983 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2984 num_bytes = ram_bytes = oe->truncated_len;
2985 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2986 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
2987 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2988 /* Encryption and other encoding is reserved and all 0 */
2991 * For delalloc, when completing an ordered extent we update the inode's
2992 * bytes when clearing the range in the inode's io tree, so pass false
2993 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2994 * except if the ordered extent was truncated.
2996 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2997 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
2998 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3000 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3001 oe->file_offset, &stack_fi,
3002 update_inode_bytes, oe->qgroup_rsv);
3006 * As ordered data IO finishes, this gets called so we can finish
3007 * an ordered extent if the range of bytes in the file it covers are
3010 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3012 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3013 struct btrfs_root *root = inode->root;
3014 struct btrfs_fs_info *fs_info = root->fs_info;
3015 struct btrfs_trans_handle *trans = NULL;
3016 struct extent_io_tree *io_tree = &inode->io_tree;
3017 struct extent_state *cached_state = NULL;
3019 int compress_type = 0;
3021 u64 logical_len = ordered_extent->num_bytes;
3022 bool freespace_inode;
3023 bool truncated = false;
3024 bool clear_reserved_extent = true;
3025 unsigned int clear_bits = EXTENT_DEFRAG;
3027 start = ordered_extent->file_offset;
3028 end = start + ordered_extent->num_bytes - 1;
3030 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3031 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3032 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3033 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3034 clear_bits |= EXTENT_DELALLOC_NEW;
3036 freespace_inode = btrfs_is_free_space_inode(inode);
3038 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3043 /* A valid bdev implies a write on a sequential zone */
3044 if (ordered_extent->bdev) {
3045 btrfs_rewrite_logical_zoned(ordered_extent);
3046 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3047 ordered_extent->disk_num_bytes);
3050 btrfs_free_io_failure_record(inode, start, end);
3052 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3054 logical_len = ordered_extent->truncated_len;
3055 /* Truncated the entire extent, don't bother adding */
3060 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3061 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3063 btrfs_inode_safe_disk_i_size_write(inode, 0);
3064 if (freespace_inode)
3065 trans = btrfs_join_transaction_spacecache(root);
3067 trans = btrfs_join_transaction(root);
3068 if (IS_ERR(trans)) {
3069 ret = PTR_ERR(trans);
3073 trans->block_rsv = &inode->block_rsv;
3074 ret = btrfs_update_inode_fallback(trans, root, inode);
3075 if (ret) /* -ENOMEM or corruption */
3076 btrfs_abort_transaction(trans, ret);
3080 clear_bits |= EXTENT_LOCKED;
3081 lock_extent_bits(io_tree, start, end, &cached_state);
3083 if (freespace_inode)
3084 trans = btrfs_join_transaction_spacecache(root);
3086 trans = btrfs_join_transaction(root);
3087 if (IS_ERR(trans)) {
3088 ret = PTR_ERR(trans);
3093 trans->block_rsv = &inode->block_rsv;
3095 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3096 compress_type = ordered_extent->compress_type;
3097 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3098 BUG_ON(compress_type);
3099 ret = btrfs_mark_extent_written(trans, inode,
3100 ordered_extent->file_offset,
3101 ordered_extent->file_offset +
3104 BUG_ON(root == fs_info->tree_root);
3105 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3107 clear_reserved_extent = false;
3108 btrfs_release_delalloc_bytes(fs_info,
3109 ordered_extent->disk_bytenr,
3110 ordered_extent->disk_num_bytes);
3113 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3114 ordered_extent->num_bytes, trans->transid);
3116 btrfs_abort_transaction(trans, ret);
3120 ret = add_pending_csums(trans, &ordered_extent->list);
3122 btrfs_abort_transaction(trans, ret);
3127 * If this is a new delalloc range, clear its new delalloc flag to
3128 * update the inode's number of bytes. This needs to be done first
3129 * before updating the inode item.
3131 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3132 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3133 clear_extent_bit(&inode->io_tree, start, end,
3134 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3135 0, 0, &cached_state);
3137 btrfs_inode_safe_disk_i_size_write(inode, 0);
3138 ret = btrfs_update_inode_fallback(trans, root, inode);
3139 if (ret) { /* -ENOMEM or corruption */
3140 btrfs_abort_transaction(trans, ret);
3145 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3146 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3150 btrfs_end_transaction(trans);
3152 if (ret || truncated) {
3153 u64 unwritten_start = start;
3156 * If we failed to finish this ordered extent for any reason we
3157 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3158 * extent, and mark the inode with the error if it wasn't
3159 * already set. Any error during writeback would have already
3160 * set the mapping error, so we need to set it if we're the ones
3161 * marking this ordered extent as failed.
3163 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3164 &ordered_extent->flags))
3165 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3168 unwritten_start += logical_len;
3169 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3171 /* Drop the cache for the part of the extent we didn't write. */
3172 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3175 * If the ordered extent had an IOERR or something else went
3176 * wrong we need to return the space for this ordered extent
3177 * back to the allocator. We only free the extent in the
3178 * truncated case if we didn't write out the extent at all.
3180 * If we made it past insert_reserved_file_extent before we
3181 * errored out then we don't need to do this as the accounting
3182 * has already been done.
3184 if ((ret || !logical_len) &&
3185 clear_reserved_extent &&
3186 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3187 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3189 * Discard the range before returning it back to the
3192 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3193 btrfs_discard_extent(fs_info,
3194 ordered_extent->disk_bytenr,
3195 ordered_extent->disk_num_bytes,
3197 btrfs_free_reserved_extent(fs_info,
3198 ordered_extent->disk_bytenr,
3199 ordered_extent->disk_num_bytes, 1);
3204 * This needs to be done to make sure anybody waiting knows we are done
3205 * updating everything for this ordered extent.
3207 btrfs_remove_ordered_extent(inode, ordered_extent);
3210 btrfs_put_ordered_extent(ordered_extent);
3211 /* once for the tree */
3212 btrfs_put_ordered_extent(ordered_extent);
3217 static void finish_ordered_fn(struct btrfs_work *work)
3219 struct btrfs_ordered_extent *ordered_extent;
3220 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3221 btrfs_finish_ordered_io(ordered_extent);
3224 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3225 struct page *page, u64 start,
3226 u64 end, bool uptodate)
3228 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3230 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3231 finish_ordered_fn, uptodate);
3235 * check_data_csum - verify checksum of one sector of uncompressed data
3237 * @io_bio: btrfs_io_bio which contains the csum
3238 * @bio_offset: offset to the beginning of the bio (in bytes)
3239 * @page: page where is the data to be verified
3240 * @pgoff: offset inside the page
3241 * @start: logical offset in the file
3243 * The length of such check is always one sector size.
3245 static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3246 u32 bio_offset, struct page *page, u32 pgoff,
3249 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3250 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3252 u32 len = fs_info->sectorsize;
3253 const u32 csum_size = fs_info->csum_size;
3254 unsigned int offset_sectors;
3256 u8 csum[BTRFS_CSUM_SIZE];
3258 ASSERT(pgoff + len <= PAGE_SIZE);
3260 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3261 csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size;
3263 kaddr = kmap_atomic(page);
3264 shash->tfm = fs_info->csum_shash;
3266 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3268 if (memcmp(csum, csum_expected, csum_size))
3271 kunmap_atomic(kaddr);
3274 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3277 btrfs_dev_stat_inc_and_print(bbio->device,
3278 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3279 memset(kaddr + pgoff, 1, len);
3280 flush_dcache_page(page);
3281 kunmap_atomic(kaddr);
3286 * When reads are done, we need to check csums to verify the data is correct.
3287 * if there's a match, we allow the bio to finish. If not, the code in
3288 * extent_io.c will try to find good copies for us.
3290 * @bio_offset: offset to the beginning of the bio (in bytes)
3291 * @start: file offset of the range start
3292 * @end: file offset of the range end (inclusive)
3294 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3297 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3298 u32 bio_offset, struct page *page,
3301 struct inode *inode = page->mapping->host;
3302 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3303 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3304 struct btrfs_root *root = BTRFS_I(inode)->root;
3305 const u32 sectorsize = root->fs_info->sectorsize;
3307 unsigned int result = 0;
3309 if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) {
3310 btrfs_page_clear_checked(fs_info, page, start, end + 1 - start);
3315 * This only happens for NODATASUM or compressed read.
3316 * Normally this should be covered by above check for compressed read
3317 * or the next check for NODATASUM. Just do a quicker exit here.
3319 if (bbio->csum == NULL)
3322 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3325 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3328 ASSERT(page_offset(page) <= start &&
3329 end <= page_offset(page) + PAGE_SIZE - 1);
3330 for (pg_off = offset_in_page(start);
3331 pg_off < offset_in_page(end);
3332 pg_off += sectorsize, bio_offset += sectorsize) {
3333 u64 file_offset = pg_off + page_offset(page);
3336 if (btrfs_is_data_reloc_root(root) &&
3337 test_range_bit(io_tree, file_offset,
3338 file_offset + sectorsize - 1,
3339 EXTENT_NODATASUM, 1, NULL)) {
3340 /* Skip the range without csum for data reloc inode */
3341 clear_extent_bits(io_tree, file_offset,
3342 file_offset + sectorsize - 1,
3346 ret = check_data_csum(inode, bbio, bio_offset, page, pg_off,
3347 page_offset(page) + pg_off);
3349 const int nr_bit = (pg_off - offset_in_page(start)) >>
3350 root->fs_info->sectorsize_bits;
3352 result |= (1U << nr_bit);
3359 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3361 * @inode: The inode we want to perform iput on
3363 * This function uses the generic vfs_inode::i_count to track whether we should
3364 * just decrement it (in case it's > 1) or if this is the last iput then link
3365 * the inode to the delayed iput machinery. Delayed iputs are processed at
3366 * transaction commit time/superblock commit/cleaner kthread.
3368 void btrfs_add_delayed_iput(struct inode *inode)
3370 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3371 struct btrfs_inode *binode = BTRFS_I(inode);
3373 if (atomic_add_unless(&inode->i_count, -1, 1))
3376 atomic_inc(&fs_info->nr_delayed_iputs);
3377 spin_lock(&fs_info->delayed_iput_lock);
3378 ASSERT(list_empty(&binode->delayed_iput));
3379 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3380 spin_unlock(&fs_info->delayed_iput_lock);
3381 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3382 wake_up_process(fs_info->cleaner_kthread);
3385 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3386 struct btrfs_inode *inode)
3388 list_del_init(&inode->delayed_iput);
3389 spin_unlock(&fs_info->delayed_iput_lock);
3390 iput(&inode->vfs_inode);
3391 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3392 wake_up(&fs_info->delayed_iputs_wait);
3393 spin_lock(&fs_info->delayed_iput_lock);
3396 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3397 struct btrfs_inode *inode)
3399 if (!list_empty(&inode->delayed_iput)) {
3400 spin_lock(&fs_info->delayed_iput_lock);
3401 if (!list_empty(&inode->delayed_iput))
3402 run_delayed_iput_locked(fs_info, inode);
3403 spin_unlock(&fs_info->delayed_iput_lock);
3407 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3410 spin_lock(&fs_info->delayed_iput_lock);
3411 while (!list_empty(&fs_info->delayed_iputs)) {
3412 struct btrfs_inode *inode;
3414 inode = list_first_entry(&fs_info->delayed_iputs,
3415 struct btrfs_inode, delayed_iput);
3416 run_delayed_iput_locked(fs_info, inode);
3417 cond_resched_lock(&fs_info->delayed_iput_lock);
3419 spin_unlock(&fs_info->delayed_iput_lock);
3423 * Wait for flushing all delayed iputs
3425 * @fs_info: the filesystem
3427 * This will wait on any delayed iputs that are currently running with KILLABLE
3428 * set. Once they are all done running we will return, unless we are killed in
3429 * which case we return EINTR. This helps in user operations like fallocate etc
3430 * that might get blocked on the iputs.
3432 * Return EINTR if we were killed, 0 if nothing's pending
3434 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3436 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3437 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3444 * This creates an orphan entry for the given inode in case something goes wrong
3445 * in the middle of an unlink.
3447 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3448 struct btrfs_inode *inode)
3452 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3453 if (ret && ret != -EEXIST) {
3454 btrfs_abort_transaction(trans, ret);
3462 * We have done the delete so we can go ahead and remove the orphan item for
3463 * this particular inode.
3465 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3466 struct btrfs_inode *inode)
3468 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3472 * this cleans up any orphans that may be left on the list from the last use
3475 int btrfs_orphan_cleanup(struct btrfs_root *root)
3477 struct btrfs_fs_info *fs_info = root->fs_info;
3478 struct btrfs_path *path;
3479 struct extent_buffer *leaf;
3480 struct btrfs_key key, found_key;
3481 struct btrfs_trans_handle *trans;
3482 struct inode *inode;
3483 u64 last_objectid = 0;
3484 int ret = 0, nr_unlink = 0;
3486 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3489 path = btrfs_alloc_path();
3494 path->reada = READA_BACK;
3496 key.objectid = BTRFS_ORPHAN_OBJECTID;
3497 key.type = BTRFS_ORPHAN_ITEM_KEY;
3498 key.offset = (u64)-1;
3501 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3506 * if ret == 0 means we found what we were searching for, which
3507 * is weird, but possible, so only screw with path if we didn't
3508 * find the key and see if we have stuff that matches
3512 if (path->slots[0] == 0)
3517 /* pull out the item */
3518 leaf = path->nodes[0];
3519 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3521 /* make sure the item matches what we want */
3522 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3524 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3527 /* release the path since we're done with it */
3528 btrfs_release_path(path);
3531 * this is where we are basically btrfs_lookup, without the
3532 * crossing root thing. we store the inode number in the
3533 * offset of the orphan item.
3536 if (found_key.offset == last_objectid) {
3538 "Error removing orphan entry, stopping orphan cleanup");
3543 last_objectid = found_key.offset;
3545 found_key.objectid = found_key.offset;
3546 found_key.type = BTRFS_INODE_ITEM_KEY;
3547 found_key.offset = 0;
3548 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3549 ret = PTR_ERR_OR_ZERO(inode);
3550 if (ret && ret != -ENOENT)
3553 if (ret == -ENOENT && root == fs_info->tree_root) {
3554 struct btrfs_root *dead_root;
3555 int is_dead_root = 0;
3558 * This is an orphan in the tree root. Currently these
3559 * could come from 2 sources:
3560 * a) a root (snapshot/subvolume) deletion in progress
3561 * b) a free space cache inode
3562 * We need to distinguish those two, as the orphan item
3563 * for a root must not get deleted before the deletion
3564 * of the snapshot/subvolume's tree completes.
3566 * btrfs_find_orphan_roots() ran before us, which has
3567 * found all deleted roots and loaded them into
3568 * fs_info->fs_roots_radix. So here we can find if an
3569 * orphan item corresponds to a deleted root by looking
3570 * up the root from that radix tree.
3573 spin_lock(&fs_info->fs_roots_radix_lock);
3574 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3575 (unsigned long)found_key.objectid);
3576 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3578 spin_unlock(&fs_info->fs_roots_radix_lock);
3581 /* prevent this orphan from being found again */
3582 key.offset = found_key.objectid - 1;
3589 * If we have an inode with links, there are a couple of
3592 * 1. We were halfway through creating fsverity metadata for the
3593 * file. In that case, the orphan item represents incomplete
3594 * fsverity metadata which must be cleaned up with
3595 * btrfs_drop_verity_items and deleting the orphan item.
3597 * 2. Old kernels (before v3.12) used to create an
3598 * orphan item for truncate indicating that there were possibly
3599 * extent items past i_size that needed to be deleted. In v3.12,
3600 * truncate was changed to update i_size in sync with the extent
3601 * items, but the (useless) orphan item was still created. Since
3602 * v4.18, we don't create the orphan item for truncate at all.
3604 * So, this item could mean that we need to do a truncate, but
3605 * only if this filesystem was last used on a pre-v3.12 kernel
3606 * and was not cleanly unmounted. The odds of that are quite
3607 * slim, and it's a pain to do the truncate now, so just delete
3610 * It's also possible that this orphan item was supposed to be
3611 * deleted but wasn't. The inode number may have been reused,
3612 * but either way, we can delete the orphan item.
3614 if (ret == -ENOENT || inode->i_nlink) {
3616 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3621 trans = btrfs_start_transaction(root, 1);
3622 if (IS_ERR(trans)) {
3623 ret = PTR_ERR(trans);
3626 btrfs_debug(fs_info, "auto deleting %Lu",
3627 found_key.objectid);
3628 ret = btrfs_del_orphan_item(trans, root,
3629 found_key.objectid);
3630 btrfs_end_transaction(trans);
3638 /* this will do delete_inode and everything for us */
3641 /* release the path since we're done with it */
3642 btrfs_release_path(path);
3644 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3645 trans = btrfs_join_transaction(root);
3647 btrfs_end_transaction(trans);
3651 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3655 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3656 btrfs_free_path(path);
3661 * very simple check to peek ahead in the leaf looking for xattrs. If we
3662 * don't find any xattrs, we know there can't be any acls.
3664 * slot is the slot the inode is in, objectid is the objectid of the inode
3666 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3667 int slot, u64 objectid,
3668 int *first_xattr_slot)
3670 u32 nritems = btrfs_header_nritems(leaf);
3671 struct btrfs_key found_key;
3672 static u64 xattr_access = 0;
3673 static u64 xattr_default = 0;
3676 if (!xattr_access) {
3677 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3678 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3679 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3680 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3684 *first_xattr_slot = -1;
3685 while (slot < nritems) {
3686 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3688 /* we found a different objectid, there must not be acls */
3689 if (found_key.objectid != objectid)
3692 /* we found an xattr, assume we've got an acl */
3693 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3694 if (*first_xattr_slot == -1)
3695 *first_xattr_slot = slot;
3696 if (found_key.offset == xattr_access ||
3697 found_key.offset == xattr_default)
3702 * we found a key greater than an xattr key, there can't
3703 * be any acls later on
3705 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3712 * it goes inode, inode backrefs, xattrs, extents,
3713 * so if there are a ton of hard links to an inode there can
3714 * be a lot of backrefs. Don't waste time searching too hard,
3715 * this is just an optimization
3720 /* we hit the end of the leaf before we found an xattr or
3721 * something larger than an xattr. We have to assume the inode
3724 if (*first_xattr_slot == -1)
3725 *first_xattr_slot = slot;
3730 * read an inode from the btree into the in-memory inode
3732 static int btrfs_read_locked_inode(struct inode *inode,
3733 struct btrfs_path *in_path)
3735 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3736 struct btrfs_path *path = in_path;
3737 struct extent_buffer *leaf;
3738 struct btrfs_inode_item *inode_item;
3739 struct btrfs_root *root = BTRFS_I(inode)->root;
3740 struct btrfs_key location;
3745 bool filled = false;
3746 int first_xattr_slot;
3748 ret = btrfs_fill_inode(inode, &rdev);
3753 path = btrfs_alloc_path();
3758 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3760 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3762 if (path != in_path)
3763 btrfs_free_path(path);
3767 leaf = path->nodes[0];
3772 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3773 struct btrfs_inode_item);
3774 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3775 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3776 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3777 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3778 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3779 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3780 round_up(i_size_read(inode), fs_info->sectorsize));
3782 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3783 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3785 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3786 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3788 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3789 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3791 BTRFS_I(inode)->i_otime.tv_sec =
3792 btrfs_timespec_sec(leaf, &inode_item->otime);
3793 BTRFS_I(inode)->i_otime.tv_nsec =
3794 btrfs_timespec_nsec(leaf, &inode_item->otime);
3796 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3797 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3798 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3800 inode_set_iversion_queried(inode,
3801 btrfs_inode_sequence(leaf, inode_item));
3802 inode->i_generation = BTRFS_I(inode)->generation;
3804 rdev = btrfs_inode_rdev(leaf, inode_item);
3806 BTRFS_I(inode)->index_cnt = (u64)-1;
3807 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3808 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3812 * If we were modified in the current generation and evicted from memory
3813 * and then re-read we need to do a full sync since we don't have any
3814 * idea about which extents were modified before we were evicted from
3817 * This is required for both inode re-read from disk and delayed inode
3818 * in delayed_nodes_tree.
3820 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3821 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3822 &BTRFS_I(inode)->runtime_flags);
3825 * We don't persist the id of the transaction where an unlink operation
3826 * against the inode was last made. So here we assume the inode might
3827 * have been evicted, and therefore the exact value of last_unlink_trans
3828 * lost, and set it to last_trans to avoid metadata inconsistencies
3829 * between the inode and its parent if the inode is fsync'ed and the log
3830 * replayed. For example, in the scenario:
3833 * ln mydir/foo mydir/bar
3836 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3837 * xfs_io -c fsync mydir/foo
3839 * mount fs, triggers fsync log replay
3841 * We must make sure that when we fsync our inode foo we also log its
3842 * parent inode, otherwise after log replay the parent still has the
3843 * dentry with the "bar" name but our inode foo has a link count of 1
3844 * and doesn't have an inode ref with the name "bar" anymore.
3846 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3847 * but it guarantees correctness at the expense of occasional full
3848 * transaction commits on fsync if our inode is a directory, or if our
3849 * inode is not a directory, logging its parent unnecessarily.
3851 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3854 * Same logic as for last_unlink_trans. We don't persist the generation
3855 * of the last transaction where this inode was used for a reflink
3856 * operation, so after eviction and reloading the inode we must be
3857 * pessimistic and assume the last transaction that modified the inode.
3859 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3862 if (inode->i_nlink != 1 ||
3863 path->slots[0] >= btrfs_header_nritems(leaf))
3866 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3867 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3870 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3871 if (location.type == BTRFS_INODE_REF_KEY) {
3872 struct btrfs_inode_ref *ref;
3874 ref = (struct btrfs_inode_ref *)ptr;
3875 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3876 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3877 struct btrfs_inode_extref *extref;
3879 extref = (struct btrfs_inode_extref *)ptr;
3880 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3885 * try to precache a NULL acl entry for files that don't have
3886 * any xattrs or acls
3888 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3889 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3890 if (first_xattr_slot != -1) {
3891 path->slots[0] = first_xattr_slot;
3892 ret = btrfs_load_inode_props(inode, path);
3895 "error loading props for ino %llu (root %llu): %d",
3896 btrfs_ino(BTRFS_I(inode)),
3897 root->root_key.objectid, ret);
3899 if (path != in_path)
3900 btrfs_free_path(path);
3903 cache_no_acl(inode);
3905 switch (inode->i_mode & S_IFMT) {
3907 inode->i_mapping->a_ops = &btrfs_aops;
3908 inode->i_fop = &btrfs_file_operations;
3909 inode->i_op = &btrfs_file_inode_operations;
3912 inode->i_fop = &btrfs_dir_file_operations;
3913 inode->i_op = &btrfs_dir_inode_operations;
3916 inode->i_op = &btrfs_symlink_inode_operations;
3917 inode_nohighmem(inode);
3918 inode->i_mapping->a_ops = &btrfs_aops;
3921 inode->i_op = &btrfs_special_inode_operations;
3922 init_special_inode(inode, inode->i_mode, rdev);
3926 btrfs_sync_inode_flags_to_i_flags(inode);
3931 * given a leaf and an inode, copy the inode fields into the leaf
3933 static void fill_inode_item(struct btrfs_trans_handle *trans,
3934 struct extent_buffer *leaf,
3935 struct btrfs_inode_item *item,
3936 struct inode *inode)
3938 struct btrfs_map_token token;
3941 btrfs_init_map_token(&token, leaf);
3943 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3944 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3945 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3946 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3947 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3949 btrfs_set_token_timespec_sec(&token, &item->atime,
3950 inode->i_atime.tv_sec);
3951 btrfs_set_token_timespec_nsec(&token, &item->atime,
3952 inode->i_atime.tv_nsec);
3954 btrfs_set_token_timespec_sec(&token, &item->mtime,
3955 inode->i_mtime.tv_sec);
3956 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3957 inode->i_mtime.tv_nsec);
3959 btrfs_set_token_timespec_sec(&token, &item->ctime,
3960 inode->i_ctime.tv_sec);
3961 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3962 inode->i_ctime.tv_nsec);
3964 btrfs_set_token_timespec_sec(&token, &item->otime,
3965 BTRFS_I(inode)->i_otime.tv_sec);
3966 btrfs_set_token_timespec_nsec(&token, &item->otime,
3967 BTRFS_I(inode)->i_otime.tv_nsec);
3969 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3970 btrfs_set_token_inode_generation(&token, item,
3971 BTRFS_I(inode)->generation);
3972 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3973 btrfs_set_token_inode_transid(&token, item, trans->transid);
3974 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3975 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3976 BTRFS_I(inode)->ro_flags);
3977 btrfs_set_token_inode_flags(&token, item, flags);
3978 btrfs_set_token_inode_block_group(&token, item, 0);
3982 * copy everything in the in-memory inode into the btree.
3984 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3985 struct btrfs_root *root,
3986 struct btrfs_inode *inode)
3988 struct btrfs_inode_item *inode_item;
3989 struct btrfs_path *path;
3990 struct extent_buffer *leaf;
3993 path = btrfs_alloc_path();
3997 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4004 leaf = path->nodes[0];
4005 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4006 struct btrfs_inode_item);
4008 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4009 btrfs_mark_buffer_dirty(leaf);
4010 btrfs_set_inode_last_trans(trans, inode);
4013 btrfs_free_path(path);
4018 * copy everything in the in-memory inode into the btree.
4020 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4021 struct btrfs_root *root,
4022 struct btrfs_inode *inode)
4024 struct btrfs_fs_info *fs_info = root->fs_info;
4028 * If the inode is a free space inode, we can deadlock during commit
4029 * if we put it into the delayed code.
4031 * The data relocation inode should also be directly updated
4034 if (!btrfs_is_free_space_inode(inode)
4035 && !btrfs_is_data_reloc_root(root)
4036 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4037 btrfs_update_root_times(trans, root);
4039 ret = btrfs_delayed_update_inode(trans, root, inode);
4041 btrfs_set_inode_last_trans(trans, inode);
4045 return btrfs_update_inode_item(trans, root, inode);
4048 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4049 struct btrfs_root *root, struct btrfs_inode *inode)
4053 ret = btrfs_update_inode(trans, root, inode);
4055 return btrfs_update_inode_item(trans, root, inode);
4060 * unlink helper that gets used here in inode.c and in the tree logging
4061 * recovery code. It remove a link in a directory with a given name, and
4062 * also drops the back refs in the inode to the directory
4064 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4065 struct btrfs_inode *dir,
4066 struct btrfs_inode *inode,
4067 const char *name, int name_len,
4068 struct btrfs_rename_ctx *rename_ctx)
4070 struct btrfs_root *root = dir->root;
4071 struct btrfs_fs_info *fs_info = root->fs_info;
4072 struct btrfs_path *path;
4074 struct btrfs_dir_item *di;
4076 u64 ino = btrfs_ino(inode);
4077 u64 dir_ino = btrfs_ino(dir);
4079 path = btrfs_alloc_path();
4085 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4086 name, name_len, -1);
4087 if (IS_ERR_OR_NULL(di)) {
4088 ret = di ? PTR_ERR(di) : -ENOENT;
4091 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4094 btrfs_release_path(path);
4097 * If we don't have dir index, we have to get it by looking up
4098 * the inode ref, since we get the inode ref, remove it directly,
4099 * it is unnecessary to do delayed deletion.
4101 * But if we have dir index, needn't search inode ref to get it.
4102 * Since the inode ref is close to the inode item, it is better
4103 * that we delay to delete it, and just do this deletion when
4104 * we update the inode item.
4106 if (inode->dir_index) {
4107 ret = btrfs_delayed_delete_inode_ref(inode);
4109 index = inode->dir_index;
4114 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4118 "failed to delete reference to %.*s, inode %llu parent %llu",
4119 name_len, name, ino, dir_ino);
4120 btrfs_abort_transaction(trans, ret);
4125 rename_ctx->index = index;
4127 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4129 btrfs_abort_transaction(trans, ret);
4134 * If we are in a rename context, we don't need to update anything in the
4135 * log. That will be done later during the rename by btrfs_log_new_name().
4136 * Besides that, doing it here would only cause extra unncessary btree
4137 * operations on the log tree, increasing latency for applications.
4140 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4142 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4147 * If we have a pending delayed iput we could end up with the final iput
4148 * being run in btrfs-cleaner context. If we have enough of these built
4149 * up we can end up burning a lot of time in btrfs-cleaner without any
4150 * way to throttle the unlinks. Since we're currently holding a ref on
4151 * the inode we can run the delayed iput here without any issues as the
4152 * final iput won't be done until after we drop the ref we're currently
4155 btrfs_run_delayed_iput(fs_info, inode);
4157 btrfs_free_path(path);
4161 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4162 inode_inc_iversion(&inode->vfs_inode);
4163 inode_inc_iversion(&dir->vfs_inode);
4164 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4165 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4166 ret = btrfs_update_inode(trans, root, dir);
4171 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4172 struct btrfs_inode *dir, struct btrfs_inode *inode,
4173 const char *name, int name_len)
4176 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4178 drop_nlink(&inode->vfs_inode);
4179 ret = btrfs_update_inode(trans, inode->root, inode);
4185 * helper to start transaction for unlink and rmdir.
4187 * unlink and rmdir are special in btrfs, they do not always free space, so
4188 * if we cannot make our reservations the normal way try and see if there is
4189 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4190 * allow the unlink to occur.
4192 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4194 struct btrfs_root *root = BTRFS_I(dir)->root;
4197 * 1 for the possible orphan item
4198 * 1 for the dir item
4199 * 1 for the dir index
4200 * 1 for the inode ref
4202 * 1 for the parent inode
4204 return btrfs_start_transaction_fallback_global_rsv(root, 6);
4207 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4209 struct btrfs_trans_handle *trans;
4210 struct inode *inode = d_inode(dentry);
4213 trans = __unlink_start_trans(dir);
4215 return PTR_ERR(trans);
4217 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4220 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4221 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4222 dentry->d_name.len);
4226 if (inode->i_nlink == 0) {
4227 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4233 btrfs_end_transaction(trans);
4234 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4238 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4239 struct inode *dir, struct dentry *dentry)
4241 struct btrfs_root *root = BTRFS_I(dir)->root;
4242 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4243 struct btrfs_path *path;
4244 struct extent_buffer *leaf;
4245 struct btrfs_dir_item *di;
4246 struct btrfs_key key;
4247 const char *name = dentry->d_name.name;
4248 int name_len = dentry->d_name.len;
4252 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4254 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4255 objectid = inode->root->root_key.objectid;
4256 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4257 objectid = inode->location.objectid;
4263 path = btrfs_alloc_path();
4267 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4268 name, name_len, -1);
4269 if (IS_ERR_OR_NULL(di)) {
4270 ret = di ? PTR_ERR(di) : -ENOENT;
4274 leaf = path->nodes[0];
4275 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4276 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4277 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4279 btrfs_abort_transaction(trans, ret);
4282 btrfs_release_path(path);
4285 * This is a placeholder inode for a subvolume we didn't have a
4286 * reference to at the time of the snapshot creation. In the meantime
4287 * we could have renamed the real subvol link into our snapshot, so
4288 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4289 * Instead simply lookup the dir_index_item for this entry so we can
4290 * remove it. Otherwise we know we have a ref to the root and we can
4291 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4293 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4294 di = btrfs_search_dir_index_item(root, path, dir_ino,
4296 if (IS_ERR_OR_NULL(di)) {
4301 btrfs_abort_transaction(trans, ret);
4305 leaf = path->nodes[0];
4306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4308 btrfs_release_path(path);
4310 ret = btrfs_del_root_ref(trans, objectid,
4311 root->root_key.objectid, dir_ino,
4312 &index, name, name_len);
4314 btrfs_abort_transaction(trans, ret);
4319 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4321 btrfs_abort_transaction(trans, ret);
4325 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4326 inode_inc_iversion(dir);
4327 dir->i_mtime = dir->i_ctime = current_time(dir);
4328 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4330 btrfs_abort_transaction(trans, ret);
4332 btrfs_free_path(path);
4337 * Helper to check if the subvolume references other subvolumes or if it's
4340 static noinline int may_destroy_subvol(struct btrfs_root *root)
4342 struct btrfs_fs_info *fs_info = root->fs_info;
4343 struct btrfs_path *path;
4344 struct btrfs_dir_item *di;
4345 struct btrfs_key key;
4349 path = btrfs_alloc_path();
4353 /* Make sure this root isn't set as the default subvol */
4354 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4355 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4356 dir_id, "default", 7, 0);
4357 if (di && !IS_ERR(di)) {
4358 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4359 if (key.objectid == root->root_key.objectid) {
4362 "deleting default subvolume %llu is not allowed",
4366 btrfs_release_path(path);
4369 key.objectid = root->root_key.objectid;
4370 key.type = BTRFS_ROOT_REF_KEY;
4371 key.offset = (u64)-1;
4373 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4379 if (path->slots[0] > 0) {
4381 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4382 if (key.objectid == root->root_key.objectid &&
4383 key.type == BTRFS_ROOT_REF_KEY)
4387 btrfs_free_path(path);
4391 /* Delete all dentries for inodes belonging to the root */
4392 static void btrfs_prune_dentries(struct btrfs_root *root)
4394 struct btrfs_fs_info *fs_info = root->fs_info;
4395 struct rb_node *node;
4396 struct rb_node *prev;
4397 struct btrfs_inode *entry;
4398 struct inode *inode;
4401 if (!BTRFS_FS_ERROR(fs_info))
4402 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4404 spin_lock(&root->inode_lock);
4406 node = root->inode_tree.rb_node;
4410 entry = rb_entry(node, struct btrfs_inode, rb_node);
4412 if (objectid < btrfs_ino(entry))
4413 node = node->rb_left;
4414 else if (objectid > btrfs_ino(entry))
4415 node = node->rb_right;
4421 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4422 if (objectid <= btrfs_ino(entry)) {
4426 prev = rb_next(prev);
4430 entry = rb_entry(node, struct btrfs_inode, rb_node);
4431 objectid = btrfs_ino(entry) + 1;
4432 inode = igrab(&entry->vfs_inode);
4434 spin_unlock(&root->inode_lock);
4435 if (atomic_read(&inode->i_count) > 1)
4436 d_prune_aliases(inode);
4438 * btrfs_drop_inode will have it removed from the inode
4439 * cache when its usage count hits zero.
4443 spin_lock(&root->inode_lock);
4447 if (cond_resched_lock(&root->inode_lock))
4450 node = rb_next(node);
4452 spin_unlock(&root->inode_lock);
4455 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4457 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4458 struct btrfs_root *root = BTRFS_I(dir)->root;
4459 struct inode *inode = d_inode(dentry);
4460 struct btrfs_root *dest = BTRFS_I(inode)->root;
4461 struct btrfs_trans_handle *trans;
4462 struct btrfs_block_rsv block_rsv;
4467 * Don't allow to delete a subvolume with send in progress. This is
4468 * inside the inode lock so the error handling that has to drop the bit
4469 * again is not run concurrently.
4471 spin_lock(&dest->root_item_lock);
4472 if (dest->send_in_progress) {
4473 spin_unlock(&dest->root_item_lock);
4475 "attempt to delete subvolume %llu during send",
4476 dest->root_key.objectid);
4479 if (atomic_read(&dest->nr_swapfiles)) {
4480 spin_unlock(&dest->root_item_lock);
4482 "attempt to delete subvolume %llu with active swapfile",
4483 root->root_key.objectid);
4486 root_flags = btrfs_root_flags(&dest->root_item);
4487 btrfs_set_root_flags(&dest->root_item,
4488 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4489 spin_unlock(&dest->root_item_lock);
4491 down_write(&fs_info->subvol_sem);
4493 ret = may_destroy_subvol(dest);
4497 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4499 * One for dir inode,
4500 * two for dir entries,
4501 * two for root ref/backref.
4503 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4507 trans = btrfs_start_transaction(root, 0);
4508 if (IS_ERR(trans)) {
4509 ret = PTR_ERR(trans);
4512 trans->block_rsv = &block_rsv;
4513 trans->bytes_reserved = block_rsv.size;
4515 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4517 ret = btrfs_unlink_subvol(trans, dir, dentry);
4519 btrfs_abort_transaction(trans, ret);
4523 ret = btrfs_record_root_in_trans(trans, dest);
4525 btrfs_abort_transaction(trans, ret);
4529 memset(&dest->root_item.drop_progress, 0,
4530 sizeof(dest->root_item.drop_progress));
4531 btrfs_set_root_drop_level(&dest->root_item, 0);
4532 btrfs_set_root_refs(&dest->root_item, 0);
4534 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4535 ret = btrfs_insert_orphan_item(trans,
4537 dest->root_key.objectid);
4539 btrfs_abort_transaction(trans, ret);
4544 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4545 BTRFS_UUID_KEY_SUBVOL,
4546 dest->root_key.objectid);
4547 if (ret && ret != -ENOENT) {
4548 btrfs_abort_transaction(trans, ret);
4551 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4552 ret = btrfs_uuid_tree_remove(trans,
4553 dest->root_item.received_uuid,
4554 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4555 dest->root_key.objectid);
4556 if (ret && ret != -ENOENT) {
4557 btrfs_abort_transaction(trans, ret);
4562 free_anon_bdev(dest->anon_dev);
4565 trans->block_rsv = NULL;
4566 trans->bytes_reserved = 0;
4567 ret = btrfs_end_transaction(trans);
4568 inode->i_flags |= S_DEAD;
4570 btrfs_subvolume_release_metadata(root, &block_rsv);
4572 up_write(&fs_info->subvol_sem);
4574 spin_lock(&dest->root_item_lock);
4575 root_flags = btrfs_root_flags(&dest->root_item);
4576 btrfs_set_root_flags(&dest->root_item,
4577 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4578 spin_unlock(&dest->root_item_lock);
4580 d_invalidate(dentry);
4581 btrfs_prune_dentries(dest);
4582 ASSERT(dest->send_in_progress == 0);
4588 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4590 struct inode *inode = d_inode(dentry);
4591 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4593 struct btrfs_trans_handle *trans;
4594 u64 last_unlink_trans;
4596 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4598 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4599 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4601 "extent tree v2 doesn't support snapshot deletion yet");
4604 return btrfs_delete_subvolume(dir, dentry);
4607 trans = __unlink_start_trans(dir);
4609 return PTR_ERR(trans);
4611 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4612 err = btrfs_unlink_subvol(trans, dir, dentry);
4616 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4620 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4622 /* now the directory is empty */
4623 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4624 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4625 dentry->d_name.len);
4627 btrfs_i_size_write(BTRFS_I(inode), 0);
4629 * Propagate the last_unlink_trans value of the deleted dir to
4630 * its parent directory. This is to prevent an unrecoverable
4631 * log tree in the case we do something like this:
4633 * 2) create snapshot under dir foo
4634 * 3) delete the snapshot
4637 * 6) fsync foo or some file inside foo
4639 if (last_unlink_trans >= trans->transid)
4640 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4643 btrfs_end_transaction(trans);
4644 btrfs_btree_balance_dirty(fs_info);
4650 * btrfs_truncate_block - read, zero a chunk and write a block
4651 * @inode - inode that we're zeroing
4652 * @from - the offset to start zeroing
4653 * @len - the length to zero, 0 to zero the entire range respective to the
4655 * @front - zero up to the offset instead of from the offset on
4657 * This will find the block for the "from" offset and cow the block and zero the
4658 * part we want to zero. This is used with truncate and hole punching.
4660 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4663 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4664 struct address_space *mapping = inode->vfs_inode.i_mapping;
4665 struct extent_io_tree *io_tree = &inode->io_tree;
4666 struct btrfs_ordered_extent *ordered;
4667 struct extent_state *cached_state = NULL;
4668 struct extent_changeset *data_reserved = NULL;
4669 bool only_release_metadata = false;
4670 u32 blocksize = fs_info->sectorsize;
4671 pgoff_t index = from >> PAGE_SHIFT;
4672 unsigned offset = from & (blocksize - 1);
4674 gfp_t mask = btrfs_alloc_write_mask(mapping);
4675 size_t write_bytes = blocksize;
4680 if (IS_ALIGNED(offset, blocksize) &&
4681 (!len || IS_ALIGNED(len, blocksize)))
4684 block_start = round_down(from, blocksize);
4685 block_end = block_start + blocksize - 1;
4687 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4690 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4691 /* For nocow case, no need to reserve data space */
4692 only_release_metadata = true;
4697 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize);
4699 if (!only_release_metadata)
4700 btrfs_free_reserved_data_space(inode, data_reserved,
4701 block_start, blocksize);
4705 page = find_or_create_page(mapping, index, mask);
4707 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4709 btrfs_delalloc_release_extents(inode, blocksize);
4713 ret = set_page_extent_mapped(page);
4717 if (!PageUptodate(page)) {
4718 ret = btrfs_readpage(NULL, page);
4720 if (page->mapping != mapping) {
4725 if (!PageUptodate(page)) {
4730 wait_on_page_writeback(page);
4732 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4734 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4736 unlock_extent_cached(io_tree, block_start, block_end,
4740 btrfs_start_ordered_extent(ordered, 1);
4741 btrfs_put_ordered_extent(ordered);
4745 clear_extent_bit(&inode->io_tree, block_start, block_end,
4746 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4747 0, 0, &cached_state);
4749 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4752 unlock_extent_cached(io_tree, block_start, block_end,
4757 if (offset != blocksize) {
4759 len = blocksize - offset;
4761 memzero_page(page, (block_start - page_offset(page)),
4764 memzero_page(page, (block_start - page_offset(page)) + offset,
4766 flush_dcache_page(page);
4768 btrfs_page_clear_checked(fs_info, page, block_start,
4769 block_end + 1 - block_start);
4770 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4771 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4773 if (only_release_metadata)
4774 set_extent_bit(&inode->io_tree, block_start, block_end,
4775 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4779 if (only_release_metadata)
4780 btrfs_delalloc_release_metadata(inode, blocksize, true);
4782 btrfs_delalloc_release_space(inode, data_reserved,
4783 block_start, blocksize, true);
4785 btrfs_delalloc_release_extents(inode, blocksize);
4789 if (only_release_metadata)
4790 btrfs_check_nocow_unlock(inode);
4791 extent_changeset_free(data_reserved);
4795 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4796 u64 offset, u64 len)
4798 struct btrfs_fs_info *fs_info = root->fs_info;
4799 struct btrfs_trans_handle *trans;
4800 struct btrfs_drop_extents_args drop_args = { 0 };
4804 * If NO_HOLES is enabled, we don't need to do anything.
4805 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4806 * or btrfs_update_inode() will be called, which guarantee that the next
4807 * fsync will know this inode was changed and needs to be logged.
4809 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4813 * 1 - for the one we're dropping
4814 * 1 - for the one we're adding
4815 * 1 - for updating the inode.
4817 trans = btrfs_start_transaction(root, 3);
4819 return PTR_ERR(trans);
4821 drop_args.start = offset;
4822 drop_args.end = offset + len;
4823 drop_args.drop_cache = true;
4825 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4827 btrfs_abort_transaction(trans, ret);
4828 btrfs_end_transaction(trans);
4832 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4833 offset, 0, 0, len, 0, len, 0, 0, 0);
4835 btrfs_abort_transaction(trans, ret);
4837 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4838 btrfs_update_inode(trans, root, inode);
4840 btrfs_end_transaction(trans);
4845 * This function puts in dummy file extents for the area we're creating a hole
4846 * for. So if we are truncating this file to a larger size we need to insert
4847 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4848 * the range between oldsize and size
4850 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4852 struct btrfs_root *root = inode->root;
4853 struct btrfs_fs_info *fs_info = root->fs_info;
4854 struct extent_io_tree *io_tree = &inode->io_tree;
4855 struct extent_map *em = NULL;
4856 struct extent_state *cached_state = NULL;
4857 struct extent_map_tree *em_tree = &inode->extent_tree;
4858 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4859 u64 block_end = ALIGN(size, fs_info->sectorsize);
4866 * If our size started in the middle of a block we need to zero out the
4867 * rest of the block before we expand the i_size, otherwise we could
4868 * expose stale data.
4870 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4874 if (size <= hole_start)
4877 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4879 cur_offset = hole_start;
4881 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4882 block_end - cur_offset);
4888 last_byte = min(extent_map_end(em), block_end);
4889 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4890 hole_size = last_byte - cur_offset;
4892 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4893 struct extent_map *hole_em;
4895 err = maybe_insert_hole(root, inode, cur_offset,
4900 err = btrfs_inode_set_file_extent_range(inode,
4901 cur_offset, hole_size);
4905 btrfs_drop_extent_cache(inode, cur_offset,
4906 cur_offset + hole_size - 1, 0);
4907 hole_em = alloc_extent_map();
4909 btrfs_set_inode_full_sync(inode);
4912 hole_em->start = cur_offset;
4913 hole_em->len = hole_size;
4914 hole_em->orig_start = cur_offset;
4916 hole_em->block_start = EXTENT_MAP_HOLE;
4917 hole_em->block_len = 0;
4918 hole_em->orig_block_len = 0;
4919 hole_em->ram_bytes = hole_size;
4920 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4921 hole_em->generation = fs_info->generation;
4924 write_lock(&em_tree->lock);
4925 err = add_extent_mapping(em_tree, hole_em, 1);
4926 write_unlock(&em_tree->lock);
4929 btrfs_drop_extent_cache(inode, cur_offset,
4933 free_extent_map(hole_em);
4935 err = btrfs_inode_set_file_extent_range(inode,
4936 cur_offset, hole_size);
4941 free_extent_map(em);
4943 cur_offset = last_byte;
4944 if (cur_offset >= block_end)
4947 free_extent_map(em);
4948 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4952 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4954 struct btrfs_root *root = BTRFS_I(inode)->root;
4955 struct btrfs_trans_handle *trans;
4956 loff_t oldsize = i_size_read(inode);
4957 loff_t newsize = attr->ia_size;
4958 int mask = attr->ia_valid;
4962 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4963 * special case where we need to update the times despite not having
4964 * these flags set. For all other operations the VFS set these flags
4965 * explicitly if it wants a timestamp update.
4967 if (newsize != oldsize) {
4968 inode_inc_iversion(inode);
4969 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4970 inode->i_ctime = inode->i_mtime =
4971 current_time(inode);
4974 if (newsize > oldsize) {
4976 * Don't do an expanding truncate while snapshotting is ongoing.
4977 * This is to ensure the snapshot captures a fully consistent
4978 * state of this file - if the snapshot captures this expanding
4979 * truncation, it must capture all writes that happened before
4982 btrfs_drew_write_lock(&root->snapshot_lock);
4983 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4985 btrfs_drew_write_unlock(&root->snapshot_lock);
4989 trans = btrfs_start_transaction(root, 1);
4990 if (IS_ERR(trans)) {
4991 btrfs_drew_write_unlock(&root->snapshot_lock);
4992 return PTR_ERR(trans);
4995 i_size_write(inode, newsize);
4996 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
4997 pagecache_isize_extended(inode, oldsize, newsize);
4998 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
4999 btrfs_drew_write_unlock(&root->snapshot_lock);
5000 btrfs_end_transaction(trans);
5002 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5004 if (btrfs_is_zoned(fs_info)) {
5005 ret = btrfs_wait_ordered_range(inode,
5006 ALIGN(newsize, fs_info->sectorsize),
5013 * We're truncating a file that used to have good data down to
5014 * zero. Make sure any new writes to the file get on disk
5018 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5019 &BTRFS_I(inode)->runtime_flags);
5021 truncate_setsize(inode, newsize);
5023 inode_dio_wait(inode);
5025 ret = btrfs_truncate(inode, newsize == oldsize);
5026 if (ret && inode->i_nlink) {
5030 * Truncate failed, so fix up the in-memory size. We
5031 * adjusted disk_i_size down as we removed extents, so
5032 * wait for disk_i_size to be stable and then update the
5033 * in-memory size to match.
5035 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5038 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5045 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5048 struct inode *inode = d_inode(dentry);
5049 struct btrfs_root *root = BTRFS_I(inode)->root;
5052 if (btrfs_root_readonly(root))
5055 err = setattr_prepare(mnt_userns, dentry, attr);
5059 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5060 err = btrfs_setsize(inode, attr);
5065 if (attr->ia_valid) {
5066 setattr_copy(mnt_userns, inode, attr);
5067 inode_inc_iversion(inode);
5068 err = btrfs_dirty_inode(inode);
5070 if (!err && attr->ia_valid & ATTR_MODE)
5071 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5078 * While truncating the inode pages during eviction, we get the VFS
5079 * calling btrfs_invalidate_folio() against each folio of the inode. This
5080 * is slow because the calls to btrfs_invalidate_folio() result in a
5081 * huge amount of calls to lock_extent_bits() and clear_extent_bit(),
5082 * which keep merging and splitting extent_state structures over and over,
5083 * wasting lots of time.
5085 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5086 * skip all those expensive operations on a per folio basis and do only
5087 * the ordered io finishing, while we release here the extent_map and
5088 * extent_state structures, without the excessive merging and splitting.
5090 static void evict_inode_truncate_pages(struct inode *inode)
5092 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5093 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5094 struct rb_node *node;
5096 ASSERT(inode->i_state & I_FREEING);
5097 truncate_inode_pages_final(&inode->i_data);
5099 write_lock(&map_tree->lock);
5100 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5101 struct extent_map *em;
5103 node = rb_first_cached(&map_tree->map);
5104 em = rb_entry(node, struct extent_map, rb_node);
5105 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5106 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5107 remove_extent_mapping(map_tree, em);
5108 free_extent_map(em);
5109 if (need_resched()) {
5110 write_unlock(&map_tree->lock);
5112 write_lock(&map_tree->lock);
5115 write_unlock(&map_tree->lock);
5118 * Keep looping until we have no more ranges in the io tree.
5119 * We can have ongoing bios started by readahead that have
5120 * their endio callback (extent_io.c:end_bio_extent_readpage)
5121 * still in progress (unlocked the pages in the bio but did not yet
5122 * unlocked the ranges in the io tree). Therefore this means some
5123 * ranges can still be locked and eviction started because before
5124 * submitting those bios, which are executed by a separate task (work
5125 * queue kthread), inode references (inode->i_count) were not taken
5126 * (which would be dropped in the end io callback of each bio).
5127 * Therefore here we effectively end up waiting for those bios and
5128 * anyone else holding locked ranges without having bumped the inode's
5129 * reference count - if we don't do it, when they access the inode's
5130 * io_tree to unlock a range it may be too late, leading to an
5131 * use-after-free issue.
5133 spin_lock(&io_tree->lock);
5134 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5135 struct extent_state *state;
5136 struct extent_state *cached_state = NULL;
5139 unsigned state_flags;
5141 node = rb_first(&io_tree->state);
5142 state = rb_entry(node, struct extent_state, rb_node);
5143 start = state->start;
5145 state_flags = state->state;
5146 spin_unlock(&io_tree->lock);
5148 lock_extent_bits(io_tree, start, end, &cached_state);
5151 * If still has DELALLOC flag, the extent didn't reach disk,
5152 * and its reserved space won't be freed by delayed_ref.
5153 * So we need to free its reserved space here.
5154 * (Refer to comment in btrfs_invalidate_folio, case 2)
5156 * Note, end is the bytenr of last byte, so we need + 1 here.
5158 if (state_flags & EXTENT_DELALLOC)
5159 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5162 clear_extent_bit(io_tree, start, end,
5163 EXTENT_LOCKED | EXTENT_DELALLOC |
5164 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5168 spin_lock(&io_tree->lock);
5170 spin_unlock(&io_tree->lock);
5173 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5174 struct btrfs_block_rsv *rsv)
5176 struct btrfs_fs_info *fs_info = root->fs_info;
5177 struct btrfs_trans_handle *trans;
5178 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5182 * Eviction should be taking place at some place safe because of our
5183 * delayed iputs. However the normal flushing code will run delayed
5184 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5186 * We reserve the delayed_refs_extra here again because we can't use
5187 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5188 * above. We reserve our extra bit here because we generate a ton of
5189 * delayed refs activity by truncating.
5191 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5192 * if we fail to make this reservation we can re-try without the
5193 * delayed_refs_extra so we can make some forward progress.
5195 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5196 BTRFS_RESERVE_FLUSH_EVICT);
5198 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5199 BTRFS_RESERVE_FLUSH_EVICT);
5202 "could not allocate space for delete; will truncate on mount");
5203 return ERR_PTR(-ENOSPC);
5205 delayed_refs_extra = 0;
5208 trans = btrfs_join_transaction(root);
5212 if (delayed_refs_extra) {
5213 trans->block_rsv = &fs_info->trans_block_rsv;
5214 trans->bytes_reserved = delayed_refs_extra;
5215 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5216 delayed_refs_extra, 1);
5221 void btrfs_evict_inode(struct inode *inode)
5223 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5224 struct btrfs_trans_handle *trans;
5225 struct btrfs_root *root = BTRFS_I(inode)->root;
5226 struct btrfs_block_rsv *rsv;
5229 trace_btrfs_inode_evict(inode);
5232 fsverity_cleanup_inode(inode);
5237 evict_inode_truncate_pages(inode);
5239 if (inode->i_nlink &&
5240 ((btrfs_root_refs(&root->root_item) != 0 &&
5241 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5242 btrfs_is_free_space_inode(BTRFS_I(inode))))
5245 if (is_bad_inode(inode))
5248 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5250 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5253 if (inode->i_nlink > 0) {
5254 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5255 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5260 * This makes sure the inode item in tree is uptodate and the space for
5261 * the inode update is released.
5263 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5268 * This drops any pending insert or delete operations we have for this
5269 * inode. We could have a delayed dir index deletion queued up, but
5270 * we're removing the inode completely so that'll be taken care of in
5273 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5275 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5278 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5281 btrfs_i_size_write(BTRFS_I(inode), 0);
5284 struct btrfs_truncate_control control = {
5285 .inode = BTRFS_I(inode),
5286 .ino = btrfs_ino(BTRFS_I(inode)),
5291 trans = evict_refill_and_join(root, rsv);
5295 trans->block_rsv = rsv;
5297 ret = btrfs_truncate_inode_items(trans, root, &control);
5298 trans->block_rsv = &fs_info->trans_block_rsv;
5299 btrfs_end_transaction(trans);
5300 btrfs_btree_balance_dirty(fs_info);
5301 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5308 * Errors here aren't a big deal, it just means we leave orphan items in
5309 * the tree. They will be cleaned up on the next mount. If the inode
5310 * number gets reused, cleanup deletes the orphan item without doing
5311 * anything, and unlink reuses the existing orphan item.
5313 * If it turns out that we are dropping too many of these, we might want
5314 * to add a mechanism for retrying these after a commit.
5316 trans = evict_refill_and_join(root, rsv);
5317 if (!IS_ERR(trans)) {
5318 trans->block_rsv = rsv;
5319 btrfs_orphan_del(trans, BTRFS_I(inode));
5320 trans->block_rsv = &fs_info->trans_block_rsv;
5321 btrfs_end_transaction(trans);
5325 btrfs_free_block_rsv(fs_info, rsv);
5328 * If we didn't successfully delete, the orphan item will still be in
5329 * the tree and we'll retry on the next mount. Again, we might also want
5330 * to retry these periodically in the future.
5332 btrfs_remove_delayed_node(BTRFS_I(inode));
5333 fsverity_cleanup_inode(inode);
5338 * Return the key found in the dir entry in the location pointer, fill @type
5339 * with BTRFS_FT_*, and return 0.
5341 * If no dir entries were found, returns -ENOENT.
5342 * If found a corrupted location in dir entry, returns -EUCLEAN.
5344 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5345 struct btrfs_key *location, u8 *type)
5347 const char *name = dentry->d_name.name;
5348 int namelen = dentry->d_name.len;
5349 struct btrfs_dir_item *di;
5350 struct btrfs_path *path;
5351 struct btrfs_root *root = BTRFS_I(dir)->root;
5354 path = btrfs_alloc_path();
5358 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5360 if (IS_ERR_OR_NULL(di)) {
5361 ret = di ? PTR_ERR(di) : -ENOENT;
5365 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5366 if (location->type != BTRFS_INODE_ITEM_KEY &&
5367 location->type != BTRFS_ROOT_ITEM_KEY) {
5369 btrfs_warn(root->fs_info,
5370 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5371 __func__, name, btrfs_ino(BTRFS_I(dir)),
5372 location->objectid, location->type, location->offset);
5375 *type = btrfs_dir_type(path->nodes[0], di);
5377 btrfs_free_path(path);
5382 * when we hit a tree root in a directory, the btrfs part of the inode
5383 * needs to be changed to reflect the root directory of the tree root. This
5384 * is kind of like crossing a mount point.
5386 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5388 struct dentry *dentry,
5389 struct btrfs_key *location,
5390 struct btrfs_root **sub_root)
5392 struct btrfs_path *path;
5393 struct btrfs_root *new_root;
5394 struct btrfs_root_ref *ref;
5395 struct extent_buffer *leaf;
5396 struct btrfs_key key;
5400 path = btrfs_alloc_path();
5407 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5408 key.type = BTRFS_ROOT_REF_KEY;
5409 key.offset = location->objectid;
5411 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5418 leaf = path->nodes[0];
5419 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5420 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5421 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5424 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5425 (unsigned long)(ref + 1),
5426 dentry->d_name.len);
5430 btrfs_release_path(path);
5432 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5433 if (IS_ERR(new_root)) {
5434 err = PTR_ERR(new_root);
5438 *sub_root = new_root;
5439 location->objectid = btrfs_root_dirid(&new_root->root_item);
5440 location->type = BTRFS_INODE_ITEM_KEY;
5441 location->offset = 0;
5444 btrfs_free_path(path);
5448 static void inode_tree_add(struct inode *inode)
5450 struct btrfs_root *root = BTRFS_I(inode)->root;
5451 struct btrfs_inode *entry;
5453 struct rb_node *parent;
5454 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5455 u64 ino = btrfs_ino(BTRFS_I(inode));
5457 if (inode_unhashed(inode))
5460 spin_lock(&root->inode_lock);
5461 p = &root->inode_tree.rb_node;
5464 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5466 if (ino < btrfs_ino(entry))
5467 p = &parent->rb_left;
5468 else if (ino > btrfs_ino(entry))
5469 p = &parent->rb_right;
5471 WARN_ON(!(entry->vfs_inode.i_state &
5472 (I_WILL_FREE | I_FREEING)));
5473 rb_replace_node(parent, new, &root->inode_tree);
5474 RB_CLEAR_NODE(parent);
5475 spin_unlock(&root->inode_lock);
5479 rb_link_node(new, parent, p);
5480 rb_insert_color(new, &root->inode_tree);
5481 spin_unlock(&root->inode_lock);
5484 static void inode_tree_del(struct btrfs_inode *inode)
5486 struct btrfs_root *root = inode->root;
5489 spin_lock(&root->inode_lock);
5490 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5491 rb_erase(&inode->rb_node, &root->inode_tree);
5492 RB_CLEAR_NODE(&inode->rb_node);
5493 empty = RB_EMPTY_ROOT(&root->inode_tree);
5495 spin_unlock(&root->inode_lock);
5497 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5498 spin_lock(&root->inode_lock);
5499 empty = RB_EMPTY_ROOT(&root->inode_tree);
5500 spin_unlock(&root->inode_lock);
5502 btrfs_add_dead_root(root);
5507 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5509 struct btrfs_iget_args *args = p;
5511 inode->i_ino = args->ino;
5512 BTRFS_I(inode)->location.objectid = args->ino;
5513 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5514 BTRFS_I(inode)->location.offset = 0;
5515 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5516 BUG_ON(args->root && !BTRFS_I(inode)->root);
5520 static int btrfs_find_actor(struct inode *inode, void *opaque)
5522 struct btrfs_iget_args *args = opaque;
5524 return args->ino == BTRFS_I(inode)->location.objectid &&
5525 args->root == BTRFS_I(inode)->root;
5528 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5529 struct btrfs_root *root)
5531 struct inode *inode;
5532 struct btrfs_iget_args args;
5533 unsigned long hashval = btrfs_inode_hash(ino, root);
5538 inode = iget5_locked(s, hashval, btrfs_find_actor,
5539 btrfs_init_locked_inode,
5545 * Get an inode object given its inode number and corresponding root.
5546 * Path can be preallocated to prevent recursing back to iget through
5547 * allocator. NULL is also valid but may require an additional allocation
5550 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5551 struct btrfs_root *root, struct btrfs_path *path)
5553 struct inode *inode;
5555 inode = btrfs_iget_locked(s, ino, root);
5557 return ERR_PTR(-ENOMEM);
5559 if (inode->i_state & I_NEW) {
5562 ret = btrfs_read_locked_inode(inode, path);
5564 inode_tree_add(inode);
5565 unlock_new_inode(inode);
5569 * ret > 0 can come from btrfs_search_slot called by
5570 * btrfs_read_locked_inode, this means the inode item
5575 inode = ERR_PTR(ret);
5582 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5584 return btrfs_iget_path(s, ino, root, NULL);
5587 static struct inode *new_simple_dir(struct super_block *s,
5588 struct btrfs_key *key,
5589 struct btrfs_root *root)
5591 struct inode *inode = new_inode(s);
5594 return ERR_PTR(-ENOMEM);
5596 BTRFS_I(inode)->root = btrfs_grab_root(root);
5597 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5598 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5600 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5602 * We only need lookup, the rest is read-only and there's no inode
5603 * associated with the dentry
5605 inode->i_op = &simple_dir_inode_operations;
5606 inode->i_opflags &= ~IOP_XATTR;
5607 inode->i_fop = &simple_dir_operations;
5608 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5609 inode->i_mtime = current_time(inode);
5610 inode->i_atime = inode->i_mtime;
5611 inode->i_ctime = inode->i_mtime;
5612 BTRFS_I(inode)->i_otime = inode->i_mtime;
5617 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5618 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5619 static_assert(BTRFS_FT_DIR == FT_DIR);
5620 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5621 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5622 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5623 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5624 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5626 static inline u8 btrfs_inode_type(struct inode *inode)
5628 return fs_umode_to_ftype(inode->i_mode);
5631 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5633 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5634 struct inode *inode;
5635 struct btrfs_root *root = BTRFS_I(dir)->root;
5636 struct btrfs_root *sub_root = root;
5637 struct btrfs_key location;
5641 if (dentry->d_name.len > BTRFS_NAME_LEN)
5642 return ERR_PTR(-ENAMETOOLONG);
5644 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5646 return ERR_PTR(ret);
5648 if (location.type == BTRFS_INODE_ITEM_KEY) {
5649 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5653 /* Do extra check against inode mode with di_type */
5654 if (btrfs_inode_type(inode) != di_type) {
5656 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5657 inode->i_mode, btrfs_inode_type(inode),
5660 return ERR_PTR(-EUCLEAN);
5665 ret = fixup_tree_root_location(fs_info, dir, dentry,
5666 &location, &sub_root);
5669 inode = ERR_PTR(ret);
5671 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5673 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5675 if (root != sub_root)
5676 btrfs_put_root(sub_root);
5678 if (!IS_ERR(inode) && root != sub_root) {
5679 down_read(&fs_info->cleanup_work_sem);
5680 if (!sb_rdonly(inode->i_sb))
5681 ret = btrfs_orphan_cleanup(sub_root);
5682 up_read(&fs_info->cleanup_work_sem);
5685 inode = ERR_PTR(ret);
5692 static int btrfs_dentry_delete(const struct dentry *dentry)
5694 struct btrfs_root *root;
5695 struct inode *inode = d_inode(dentry);
5697 if (!inode && !IS_ROOT(dentry))
5698 inode = d_inode(dentry->d_parent);
5701 root = BTRFS_I(inode)->root;
5702 if (btrfs_root_refs(&root->root_item) == 0)
5705 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5711 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5714 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5716 if (inode == ERR_PTR(-ENOENT))
5718 return d_splice_alias(inode, dentry);
5722 * All this infrastructure exists because dir_emit can fault, and we are holding
5723 * the tree lock when doing readdir. For now just allocate a buffer and copy
5724 * our information into that, and then dir_emit from the buffer. This is
5725 * similar to what NFS does, only we don't keep the buffer around in pagecache
5726 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5727 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5730 static int btrfs_opendir(struct inode *inode, struct file *file)
5732 struct btrfs_file_private *private;
5734 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5737 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5738 if (!private->filldir_buf) {
5742 file->private_data = private;
5753 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5756 struct dir_entry *entry = addr;
5757 char *name = (char *)(entry + 1);
5759 ctx->pos = get_unaligned(&entry->offset);
5760 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5761 get_unaligned(&entry->ino),
5762 get_unaligned(&entry->type)))
5764 addr += sizeof(struct dir_entry) +
5765 get_unaligned(&entry->name_len);
5771 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5773 struct inode *inode = file_inode(file);
5774 struct btrfs_root *root = BTRFS_I(inode)->root;
5775 struct btrfs_file_private *private = file->private_data;
5776 struct btrfs_dir_item *di;
5777 struct btrfs_key key;
5778 struct btrfs_key found_key;
5779 struct btrfs_path *path;
5781 struct list_head ins_list;
5782 struct list_head del_list;
5789 struct btrfs_key location;
5791 if (!dir_emit_dots(file, ctx))
5794 path = btrfs_alloc_path();
5798 addr = private->filldir_buf;
5799 path->reada = READA_FORWARD;
5801 INIT_LIST_HEAD(&ins_list);
5802 INIT_LIST_HEAD(&del_list);
5803 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5806 key.type = BTRFS_DIR_INDEX_KEY;
5807 key.offset = ctx->pos;
5808 key.objectid = btrfs_ino(BTRFS_I(inode));
5810 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5811 struct dir_entry *entry;
5812 struct extent_buffer *leaf = path->nodes[0];
5814 if (found_key.objectid != key.objectid)
5816 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5818 if (found_key.offset < ctx->pos)
5820 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5822 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5823 name_len = btrfs_dir_name_len(leaf, di);
5824 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5826 btrfs_release_path(path);
5827 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5830 addr = private->filldir_buf;
5837 put_unaligned(name_len, &entry->name_len);
5838 name_ptr = (char *)(entry + 1);
5839 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5841 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5843 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5844 put_unaligned(location.objectid, &entry->ino);
5845 put_unaligned(found_key.offset, &entry->offset);
5847 addr += sizeof(struct dir_entry) + name_len;
5848 total_len += sizeof(struct dir_entry) + name_len;
5850 /* Catch error encountered during iteration */
5854 btrfs_release_path(path);
5856 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5860 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5865 * Stop new entries from being returned after we return the last
5868 * New directory entries are assigned a strictly increasing
5869 * offset. This means that new entries created during readdir
5870 * are *guaranteed* to be seen in the future by that readdir.
5871 * This has broken buggy programs which operate on names as
5872 * they're returned by readdir. Until we re-use freed offsets
5873 * we have this hack to stop new entries from being returned
5874 * under the assumption that they'll never reach this huge
5877 * This is being careful not to overflow 32bit loff_t unless the
5878 * last entry requires it because doing so has broken 32bit apps
5881 if (ctx->pos >= INT_MAX)
5882 ctx->pos = LLONG_MAX;
5889 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5890 btrfs_free_path(path);
5895 * This is somewhat expensive, updating the tree every time the
5896 * inode changes. But, it is most likely to find the inode in cache.
5897 * FIXME, needs more benchmarking...there are no reasons other than performance
5898 * to keep or drop this code.
5900 static int btrfs_dirty_inode(struct inode *inode)
5902 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5903 struct btrfs_root *root = BTRFS_I(inode)->root;
5904 struct btrfs_trans_handle *trans;
5907 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5910 trans = btrfs_join_transaction(root);
5912 return PTR_ERR(trans);
5914 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5915 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
5916 /* whoops, lets try again with the full transaction */
5917 btrfs_end_transaction(trans);
5918 trans = btrfs_start_transaction(root, 1);
5920 return PTR_ERR(trans);
5922 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5924 btrfs_end_transaction(trans);
5925 if (BTRFS_I(inode)->delayed_node)
5926 btrfs_balance_delayed_items(fs_info);
5932 * This is a copy of file_update_time. We need this so we can return error on
5933 * ENOSPC for updating the inode in the case of file write and mmap writes.
5935 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5938 struct btrfs_root *root = BTRFS_I(inode)->root;
5939 bool dirty = flags & ~S_VERSION;
5941 if (btrfs_root_readonly(root))
5944 if (flags & S_VERSION)
5945 dirty |= inode_maybe_inc_iversion(inode, dirty);
5946 if (flags & S_CTIME)
5947 inode->i_ctime = *now;
5948 if (flags & S_MTIME)
5949 inode->i_mtime = *now;
5950 if (flags & S_ATIME)
5951 inode->i_atime = *now;
5952 return dirty ? btrfs_dirty_inode(inode) : 0;
5956 * find the highest existing sequence number in a directory
5957 * and then set the in-memory index_cnt variable to reflect
5958 * free sequence numbers
5960 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5962 struct btrfs_root *root = inode->root;
5963 struct btrfs_key key, found_key;
5964 struct btrfs_path *path;
5965 struct extent_buffer *leaf;
5968 key.objectid = btrfs_ino(inode);
5969 key.type = BTRFS_DIR_INDEX_KEY;
5970 key.offset = (u64)-1;
5972 path = btrfs_alloc_path();
5976 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5979 /* FIXME: we should be able to handle this */
5984 if (path->slots[0] == 0) {
5985 inode->index_cnt = BTRFS_DIR_START_INDEX;
5991 leaf = path->nodes[0];
5992 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5994 if (found_key.objectid != btrfs_ino(inode) ||
5995 found_key.type != BTRFS_DIR_INDEX_KEY) {
5996 inode->index_cnt = BTRFS_DIR_START_INDEX;
6000 inode->index_cnt = found_key.offset + 1;
6002 btrfs_free_path(path);
6007 * helper to find a free sequence number in a given directory. This current
6008 * code is very simple, later versions will do smarter things in the btree
6010 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6014 if (dir->index_cnt == (u64)-1) {
6015 ret = btrfs_inode_delayed_dir_index_count(dir);
6017 ret = btrfs_set_inode_index_count(dir);
6023 *index = dir->index_cnt;
6029 static int btrfs_insert_inode_locked(struct inode *inode)
6031 struct btrfs_iget_args args;
6033 args.ino = BTRFS_I(inode)->location.objectid;
6034 args.root = BTRFS_I(inode)->root;
6036 return insert_inode_locked4(inode,
6037 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6038 btrfs_find_actor, &args);
6042 * Inherit flags from the parent inode.
6044 * Currently only the compression flags and the cow flags are inherited.
6046 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6053 flags = BTRFS_I(dir)->flags;
6055 if (flags & BTRFS_INODE_NOCOMPRESS) {
6056 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6057 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6058 } else if (flags & BTRFS_INODE_COMPRESS) {
6059 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6060 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6063 if (flags & BTRFS_INODE_NODATACOW) {
6064 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6065 if (S_ISREG(inode->i_mode))
6066 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6069 btrfs_sync_inode_flags_to_i_flags(inode);
6072 static int btrfs_new_inode(struct btrfs_trans_handle *trans,
6073 struct btrfs_root *root, struct inode *inode,
6074 struct inode *dir, const char *name, int name_len,
6077 struct btrfs_fs_info *fs_info = root->fs_info;
6078 struct btrfs_inode_item *inode_item;
6079 struct btrfs_key *location;
6080 struct btrfs_path *path;
6082 struct btrfs_inode_ref *ref;
6083 struct btrfs_key key[2];
6085 struct btrfs_item_batch batch;
6089 path = btrfs_alloc_path();
6094 * O_TMPFILE, set link count to 0, so that after this point,
6095 * we fill in an inode item with the correct link count.
6098 set_nlink(inode, 0);
6100 ret = btrfs_get_free_objectid(root, &objectid);
6102 btrfs_free_path(path);
6105 inode->i_ino = objectid;
6108 trace_btrfs_inode_request(dir);
6110 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6112 btrfs_free_path(path);
6119 * index_cnt is ignored for everything but a dir,
6120 * btrfs_set_inode_index_count has an explanation for the magic
6123 BTRFS_I(inode)->index_cnt = 2;
6124 BTRFS_I(inode)->dir_index = *index;
6125 if (!BTRFS_I(inode)->root)
6126 BTRFS_I(inode)->root = btrfs_grab_root(root);
6127 BTRFS_I(inode)->generation = trans->transid;
6128 inode->i_generation = BTRFS_I(inode)->generation;
6130 btrfs_inherit_iflags(inode, dir);
6132 if (S_ISREG(inode->i_mode)) {
6133 if (btrfs_test_opt(fs_info, NODATASUM))
6134 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6135 if (btrfs_test_opt(fs_info, NODATACOW))
6136 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6137 BTRFS_INODE_NODATASUM;
6141 * We could have gotten an inode number from somebody who was fsynced
6142 * and then removed in this same transaction, so let's just set full
6143 * sync since it will be a full sync anyway and this will blow away the
6144 * old info in the log.
6146 btrfs_set_inode_full_sync(BTRFS_I(inode));
6148 key[0].objectid = objectid;
6149 key[0].type = BTRFS_INODE_ITEM_KEY;
6152 sizes[0] = sizeof(struct btrfs_inode_item);
6156 * Start new inodes with an inode_ref. This is slightly more
6157 * efficient for small numbers of hard links since they will
6158 * be packed into one item. Extended refs will kick in if we
6159 * add more hard links than can fit in the ref item.
6161 key[1].objectid = objectid;
6162 key[1].type = BTRFS_INODE_REF_KEY;
6164 key[1].offset = btrfs_ino(BTRFS_I(dir));
6166 key[1].offset = objectid;
6168 sizes[1] = name_len + sizeof(*ref);
6171 location = &BTRFS_I(inode)->location;
6172 location->objectid = objectid;
6173 location->offset = 0;
6174 location->type = BTRFS_INODE_ITEM_KEY;
6176 ret = btrfs_insert_inode_locked(inode);
6180 batch.keys = &key[0];
6181 batch.data_sizes = &sizes[0];
6182 batch.total_data_size = sizes[0] + (name ? sizes[1] : 0);
6183 batch.nr = name ? 2 : 1;
6184 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6188 inode->i_mtime = current_time(inode);
6189 inode->i_atime = inode->i_mtime;
6190 inode->i_ctime = inode->i_mtime;
6191 BTRFS_I(inode)->i_otime = inode->i_mtime;
6193 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6194 struct btrfs_inode_item);
6195 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6196 sizeof(*inode_item));
6197 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6200 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6201 struct btrfs_inode_ref);
6202 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6203 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6204 ptr = (unsigned long)(ref + 1);
6205 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6208 btrfs_mark_buffer_dirty(path->nodes[0]);
6209 btrfs_free_path(path);
6211 inode_tree_add(inode);
6213 trace_btrfs_inode_new(inode);
6214 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6216 btrfs_update_root_times(trans, root);
6218 ret = btrfs_inode_inherit_props(trans, inode, dir);
6221 "error inheriting props for ino %llu (root %llu): %d",
6222 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6228 * discard_new_inode() calls iput(), but the caller owns the reference
6232 discard_new_inode(inode);
6235 BTRFS_I(dir)->index_cnt--;
6236 btrfs_free_path(path);
6241 * utility function to add 'inode' into 'parent_inode' with
6242 * a give name and a given sequence number.
6243 * if 'add_backref' is true, also insert a backref from the
6244 * inode to the parent directory.
6246 int btrfs_add_link(struct btrfs_trans_handle *trans,
6247 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6248 const char *name, int name_len, int add_backref, u64 index)
6251 struct btrfs_key key;
6252 struct btrfs_root *root = parent_inode->root;
6253 u64 ino = btrfs_ino(inode);
6254 u64 parent_ino = btrfs_ino(parent_inode);
6256 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6257 memcpy(&key, &inode->root->root_key, sizeof(key));
6260 key.type = BTRFS_INODE_ITEM_KEY;
6264 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6265 ret = btrfs_add_root_ref(trans, key.objectid,
6266 root->root_key.objectid, parent_ino,
6267 index, name, name_len);
6268 } else if (add_backref) {
6269 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6273 /* Nothing to clean up yet */
6277 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6278 btrfs_inode_type(&inode->vfs_inode), index);
6279 if (ret == -EEXIST || ret == -EOVERFLOW)
6282 btrfs_abort_transaction(trans, ret);
6286 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6288 inode_inc_iversion(&parent_inode->vfs_inode);
6290 * If we are replaying a log tree, we do not want to update the mtime
6291 * and ctime of the parent directory with the current time, since the
6292 * log replay procedure is responsible for setting them to their correct
6293 * values (the ones it had when the fsync was done).
6295 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6296 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6298 parent_inode->vfs_inode.i_mtime = now;
6299 parent_inode->vfs_inode.i_ctime = now;
6301 ret = btrfs_update_inode(trans, root, parent_inode);
6303 btrfs_abort_transaction(trans, ret);
6307 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6310 err = btrfs_del_root_ref(trans, key.objectid,
6311 root->root_key.objectid, parent_ino,
6312 &local_index, name, name_len);
6314 btrfs_abort_transaction(trans, err);
6315 } else if (add_backref) {
6319 err = btrfs_del_inode_ref(trans, root, name, name_len,
6320 ino, parent_ino, &local_index);
6322 btrfs_abort_transaction(trans, err);
6325 /* Return the original error code */
6329 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6330 struct inode *inode)
6332 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6333 struct btrfs_root *root = BTRFS_I(dir)->root;
6334 struct btrfs_trans_handle *trans;
6339 * 2 for inode item and ref
6341 * 1 for xattr if selinux is on
6343 trans = btrfs_start_transaction(root, 5);
6344 if (IS_ERR(trans)) {
6346 return PTR_ERR(trans);
6349 err = btrfs_new_inode(trans, root, inode, dir, dentry->d_name.name,
6350 dentry->d_name.len, &index);
6357 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6361 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6365 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6366 dentry->d_name.name, dentry->d_name.len, 0, index);
6370 d_instantiate_new(dentry, inode);
6373 btrfs_end_transaction(trans);
6375 inode_dec_link_count(inode);
6376 discard_new_inode(inode);
6378 btrfs_btree_balance_dirty(fs_info);
6382 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6383 struct dentry *dentry, umode_t mode, dev_t rdev)
6385 struct inode *inode;
6387 inode = new_inode(dir->i_sb);
6390 inode_init_owner(mnt_userns, inode, dir, mode);
6391 inode->i_op = &btrfs_special_inode_operations;
6392 init_special_inode(inode, inode->i_mode, rdev);
6393 return btrfs_create_common(dir, dentry, inode);
6396 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6397 struct dentry *dentry, umode_t mode, bool excl)
6399 struct inode *inode;
6401 inode = new_inode(dir->i_sb);
6404 inode_init_owner(mnt_userns, inode, dir, mode);
6405 inode->i_fop = &btrfs_file_operations;
6406 inode->i_op = &btrfs_file_inode_operations;
6407 inode->i_mapping->a_ops = &btrfs_aops;
6408 return btrfs_create_common(dir, dentry, inode);
6411 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6412 struct dentry *dentry)
6414 struct btrfs_trans_handle *trans = NULL;
6415 struct btrfs_root *root = BTRFS_I(dir)->root;
6416 struct inode *inode = d_inode(old_dentry);
6417 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6422 /* do not allow sys_link's with other subvols of the same device */
6423 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6426 if (inode->i_nlink >= BTRFS_LINK_MAX)
6429 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6434 * 2 items for inode and inode ref
6435 * 2 items for dir items
6436 * 1 item for parent inode
6437 * 1 item for orphan item deletion if O_TMPFILE
6439 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6440 if (IS_ERR(trans)) {
6441 err = PTR_ERR(trans);
6446 /* There are several dir indexes for this inode, clear the cache. */
6447 BTRFS_I(inode)->dir_index = 0ULL;
6449 inode_inc_iversion(inode);
6450 inode->i_ctime = current_time(inode);
6452 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6454 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6455 dentry->d_name.name, dentry->d_name.len, 1, index);
6460 struct dentry *parent = dentry->d_parent;
6462 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6465 if (inode->i_nlink == 1) {
6467 * If new hard link count is 1, it's a file created
6468 * with open(2) O_TMPFILE flag.
6470 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6474 d_instantiate(dentry, inode);
6475 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6480 btrfs_end_transaction(trans);
6482 inode_dec_link_count(inode);
6485 btrfs_btree_balance_dirty(fs_info);
6489 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6490 struct dentry *dentry, umode_t mode)
6492 struct inode *inode;
6494 inode = new_inode(dir->i_sb);
6497 inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode);
6498 inode->i_op = &btrfs_dir_inode_operations;
6499 inode->i_fop = &btrfs_dir_file_operations;
6500 return btrfs_create_common(dir, dentry, inode);
6503 static noinline int uncompress_inline(struct btrfs_path *path,
6505 size_t pg_offset, u64 extent_offset,
6506 struct btrfs_file_extent_item *item)
6509 struct extent_buffer *leaf = path->nodes[0];
6512 unsigned long inline_size;
6516 WARN_ON(pg_offset != 0);
6517 compress_type = btrfs_file_extent_compression(leaf, item);
6518 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6519 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6520 tmp = kmalloc(inline_size, GFP_NOFS);
6523 ptr = btrfs_file_extent_inline_start(item);
6525 read_extent_buffer(leaf, tmp, ptr, inline_size);
6527 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6528 ret = btrfs_decompress(compress_type, tmp, page,
6529 extent_offset, inline_size, max_size);
6532 * decompression code contains a memset to fill in any space between the end
6533 * of the uncompressed data and the end of max_size in case the decompressed
6534 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6535 * the end of an inline extent and the beginning of the next block, so we
6536 * cover that region here.
6539 if (max_size + pg_offset < PAGE_SIZE)
6540 memzero_page(page, pg_offset + max_size,
6541 PAGE_SIZE - max_size - pg_offset);
6547 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6548 * @inode: file to search in
6549 * @page: page to read extent data into if the extent is inline
6550 * @pg_offset: offset into @page to copy to
6551 * @start: file offset
6552 * @len: length of range starting at @start
6554 * This returns the first &struct extent_map which overlaps with the given
6555 * range, reading it from the B-tree and caching it if necessary. Note that
6556 * there may be more extents which overlap the given range after the returned
6559 * If @page is not NULL and the extent is inline, this also reads the extent
6560 * data directly into the page and marks the extent up to date in the io_tree.
6562 * Return: ERR_PTR on error, non-NULL extent_map on success.
6564 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6565 struct page *page, size_t pg_offset,
6568 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6570 u64 extent_start = 0;
6572 u64 objectid = btrfs_ino(inode);
6573 int extent_type = -1;
6574 struct btrfs_path *path = NULL;
6575 struct btrfs_root *root = inode->root;
6576 struct btrfs_file_extent_item *item;
6577 struct extent_buffer *leaf;
6578 struct btrfs_key found_key;
6579 struct extent_map *em = NULL;
6580 struct extent_map_tree *em_tree = &inode->extent_tree;
6581 struct extent_io_tree *io_tree = &inode->io_tree;
6583 read_lock(&em_tree->lock);
6584 em = lookup_extent_mapping(em_tree, start, len);
6585 read_unlock(&em_tree->lock);
6588 if (em->start > start || em->start + em->len <= start)
6589 free_extent_map(em);
6590 else if (em->block_start == EXTENT_MAP_INLINE && page)
6591 free_extent_map(em);
6595 em = alloc_extent_map();
6600 em->start = EXTENT_MAP_HOLE;
6601 em->orig_start = EXTENT_MAP_HOLE;
6603 em->block_len = (u64)-1;
6605 path = btrfs_alloc_path();
6611 /* Chances are we'll be called again, so go ahead and do readahead */
6612 path->reada = READA_FORWARD;
6615 * The same explanation in load_free_space_cache applies here as well,
6616 * we only read when we're loading the free space cache, and at that
6617 * point the commit_root has everything we need.
6619 if (btrfs_is_free_space_inode(inode)) {
6620 path->search_commit_root = 1;
6621 path->skip_locking = 1;
6624 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6627 } else if (ret > 0) {
6628 if (path->slots[0] == 0)
6634 leaf = path->nodes[0];
6635 item = btrfs_item_ptr(leaf, path->slots[0],
6636 struct btrfs_file_extent_item);
6637 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6638 if (found_key.objectid != objectid ||
6639 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6641 * If we backup past the first extent we want to move forward
6642 * and see if there is an extent in front of us, otherwise we'll
6643 * say there is a hole for our whole search range which can
6650 extent_type = btrfs_file_extent_type(leaf, item);
6651 extent_start = found_key.offset;
6652 extent_end = btrfs_file_extent_end(path);
6653 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6654 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6655 /* Only regular file could have regular/prealloc extent */
6656 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6659 "regular/prealloc extent found for non-regular inode %llu",
6663 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6665 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6666 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6671 if (start >= extent_end) {
6673 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6674 ret = btrfs_next_leaf(root, path);
6680 leaf = path->nodes[0];
6682 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6683 if (found_key.objectid != objectid ||
6684 found_key.type != BTRFS_EXTENT_DATA_KEY)
6686 if (start + len <= found_key.offset)
6688 if (start > found_key.offset)
6691 /* New extent overlaps with existing one */
6693 em->orig_start = start;
6694 em->len = found_key.offset - start;
6695 em->block_start = EXTENT_MAP_HOLE;
6699 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6701 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6702 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6704 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6708 size_t extent_offset;
6714 size = btrfs_file_extent_ram_bytes(leaf, item);
6715 extent_offset = page_offset(page) + pg_offset - extent_start;
6716 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6717 size - extent_offset);
6718 em->start = extent_start + extent_offset;
6719 em->len = ALIGN(copy_size, fs_info->sectorsize);
6720 em->orig_block_len = em->len;
6721 em->orig_start = em->start;
6722 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6724 if (!PageUptodate(page)) {
6725 if (btrfs_file_extent_compression(leaf, item) !=
6726 BTRFS_COMPRESS_NONE) {
6727 ret = uncompress_inline(path, page, pg_offset,
6728 extent_offset, item);
6732 map = kmap_local_page(page);
6733 read_extent_buffer(leaf, map + pg_offset, ptr,
6735 if (pg_offset + copy_size < PAGE_SIZE) {
6736 memset(map + pg_offset + copy_size, 0,
6737 PAGE_SIZE - pg_offset -
6742 flush_dcache_page(page);
6744 set_extent_uptodate(io_tree, em->start,
6745 extent_map_end(em) - 1, NULL, GFP_NOFS);
6750 em->orig_start = start;
6752 em->block_start = EXTENT_MAP_HOLE;
6755 btrfs_release_path(path);
6756 if (em->start > start || extent_map_end(em) <= start) {
6758 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6759 em->start, em->len, start, len);
6764 write_lock(&em_tree->lock);
6765 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6766 write_unlock(&em_tree->lock);
6768 btrfs_free_path(path);
6770 trace_btrfs_get_extent(root, inode, em);
6773 free_extent_map(em);
6774 return ERR_PTR(ret);
6779 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6782 struct extent_map *em;
6783 struct extent_map *hole_em = NULL;
6784 u64 delalloc_start = start;
6790 em = btrfs_get_extent(inode, NULL, 0, start, len);
6794 * If our em maps to:
6796 * - a pre-alloc extent,
6797 * there might actually be delalloc bytes behind it.
6799 if (em->block_start != EXTENT_MAP_HOLE &&
6800 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6805 /* check to see if we've wrapped (len == -1 or similar) */
6814 /* ok, we didn't find anything, lets look for delalloc */
6815 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6816 end, len, EXTENT_DELALLOC, 1);
6817 delalloc_end = delalloc_start + delalloc_len;
6818 if (delalloc_end < delalloc_start)
6819 delalloc_end = (u64)-1;
6822 * We didn't find anything useful, return the original results from
6825 if (delalloc_start > end || delalloc_end <= start) {
6832 * Adjust the delalloc_start to make sure it doesn't go backwards from
6833 * the start they passed in
6835 delalloc_start = max(start, delalloc_start);
6836 delalloc_len = delalloc_end - delalloc_start;
6838 if (delalloc_len > 0) {
6841 const u64 hole_end = extent_map_end(hole_em);
6843 em = alloc_extent_map();
6851 * When btrfs_get_extent can't find anything it returns one
6854 * Make sure what it found really fits our range, and adjust to
6855 * make sure it is based on the start from the caller
6857 if (hole_end <= start || hole_em->start > end) {
6858 free_extent_map(hole_em);
6861 hole_start = max(hole_em->start, start);
6862 hole_len = hole_end - hole_start;
6865 if (hole_em && delalloc_start > hole_start) {
6867 * Our hole starts before our delalloc, so we have to
6868 * return just the parts of the hole that go until the
6871 em->len = min(hole_len, delalloc_start - hole_start);
6872 em->start = hole_start;
6873 em->orig_start = hole_start;
6875 * Don't adjust block start at all, it is fixed at
6878 em->block_start = hole_em->block_start;
6879 em->block_len = hole_len;
6880 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6881 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6884 * Hole is out of passed range or it starts after
6887 em->start = delalloc_start;
6888 em->len = delalloc_len;
6889 em->orig_start = delalloc_start;
6890 em->block_start = EXTENT_MAP_DELALLOC;
6891 em->block_len = delalloc_len;
6898 free_extent_map(hole_em);
6900 free_extent_map(em);
6901 return ERR_PTR(err);
6906 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6909 const u64 orig_start,
6910 const u64 block_start,
6911 const u64 block_len,
6912 const u64 orig_block_len,
6913 const u64 ram_bytes,
6916 struct extent_map *em = NULL;
6919 if (type != BTRFS_ORDERED_NOCOW) {
6920 em = create_io_em(inode, start, len, orig_start, block_start,
6921 block_len, orig_block_len, ram_bytes,
6922 BTRFS_COMPRESS_NONE, /* compress_type */
6927 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
6930 (1 << BTRFS_ORDERED_DIRECT),
6931 BTRFS_COMPRESS_NONE);
6934 free_extent_map(em);
6935 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
6944 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6947 struct btrfs_root *root = inode->root;
6948 struct btrfs_fs_info *fs_info = root->fs_info;
6949 struct extent_map *em;
6950 struct btrfs_key ins;
6954 alloc_hint = get_extent_allocation_hint(inode, start, len);
6955 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6956 0, alloc_hint, &ins, 1, 1);
6958 return ERR_PTR(ret);
6960 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
6961 ins.objectid, ins.offset, ins.offset,
6962 ins.offset, BTRFS_ORDERED_REGULAR);
6963 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6965 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
6971 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
6973 struct btrfs_block_group *block_group;
6974 bool readonly = false;
6976 block_group = btrfs_lookup_block_group(fs_info, bytenr);
6977 if (!block_group || block_group->ro)
6980 btrfs_put_block_group(block_group);
6985 * Check if we can do nocow write into the range [@offset, @offset + @len)
6987 * @offset: File offset
6988 * @len: The length to write, will be updated to the nocow writeable
6990 * @orig_start: (optional) Return the original file offset of the file extent
6991 * @orig_len: (optional) Return the original on-disk length of the file extent
6992 * @ram_bytes: (optional) Return the ram_bytes of the file extent
6993 * @strict: if true, omit optimizations that might force us into unnecessary
6994 * cow. e.g., don't trust generation number.
6997 * >0 and update @len if we can do nocow write
6998 * 0 if we can't do nocow write
6999 * <0 if error happened
7001 * NOTE: This only checks the file extents, caller is responsible to wait for
7002 * any ordered extents.
7004 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7005 u64 *orig_start, u64 *orig_block_len,
7006 u64 *ram_bytes, bool strict)
7008 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7009 struct btrfs_path *path;
7011 struct extent_buffer *leaf;
7012 struct btrfs_root *root = BTRFS_I(inode)->root;
7013 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7014 struct btrfs_file_extent_item *fi;
7015 struct btrfs_key key;
7022 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7024 path = btrfs_alloc_path();
7028 ret = btrfs_lookup_file_extent(NULL, root, path,
7029 btrfs_ino(BTRFS_I(inode)), offset, 0);
7033 slot = path->slots[0];
7036 /* can't find the item, must cow */
7043 leaf = path->nodes[0];
7044 btrfs_item_key_to_cpu(leaf, &key, slot);
7045 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7046 key.type != BTRFS_EXTENT_DATA_KEY) {
7047 /* not our file or wrong item type, must cow */
7051 if (key.offset > offset) {
7052 /* Wrong offset, must cow */
7056 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7057 found_type = btrfs_file_extent_type(leaf, fi);
7058 if (found_type != BTRFS_FILE_EXTENT_REG &&
7059 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7060 /* not a regular extent, must cow */
7064 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7067 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7068 if (extent_end <= offset)
7071 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7072 if (disk_bytenr == 0)
7075 if (btrfs_file_extent_compression(leaf, fi) ||
7076 btrfs_file_extent_encryption(leaf, fi) ||
7077 btrfs_file_extent_other_encoding(leaf, fi))
7081 * Do the same check as in btrfs_cross_ref_exist but without the
7082 * unnecessary search.
7085 (btrfs_file_extent_generation(leaf, fi) <=
7086 btrfs_root_last_snapshot(&root->root_item)))
7089 backref_offset = btrfs_file_extent_offset(leaf, fi);
7092 *orig_start = key.offset - backref_offset;
7093 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7094 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7097 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7100 num_bytes = min(offset + *len, extent_end) - offset;
7101 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7104 range_end = round_up(offset + num_bytes,
7105 root->fs_info->sectorsize) - 1;
7106 ret = test_range_bit(io_tree, offset, range_end,
7107 EXTENT_DELALLOC, 0, NULL);
7114 btrfs_release_path(path);
7117 * look for other files referencing this extent, if we
7118 * find any we must cow
7121 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7122 key.offset - backref_offset, disk_bytenr,
7130 * adjust disk_bytenr and num_bytes to cover just the bytes
7131 * in this extent we are about to write. If there
7132 * are any csums in that range we have to cow in order
7133 * to keep the csums correct
7135 disk_bytenr += backref_offset;
7136 disk_bytenr += offset - key.offset;
7137 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7140 * all of the above have passed, it is safe to overwrite this extent
7146 btrfs_free_path(path);
7150 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7151 struct extent_state **cached_state, bool writing)
7153 struct btrfs_ordered_extent *ordered;
7157 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7160 * We're concerned with the entire range that we're going to be
7161 * doing DIO to, so we need to make sure there's no ordered
7162 * extents in this range.
7164 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7165 lockend - lockstart + 1);
7168 * We need to make sure there are no buffered pages in this
7169 * range either, we could have raced between the invalidate in
7170 * generic_file_direct_write and locking the extent. The
7171 * invalidate needs to happen so that reads after a write do not
7175 (!writing || !filemap_range_has_page(inode->i_mapping,
7176 lockstart, lockend)))
7179 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7184 * If we are doing a DIO read and the ordered extent we
7185 * found is for a buffered write, we can not wait for it
7186 * to complete and retry, because if we do so we can
7187 * deadlock with concurrent buffered writes on page
7188 * locks. This happens only if our DIO read covers more
7189 * than one extent map, if at this point has already
7190 * created an ordered extent for a previous extent map
7191 * and locked its range in the inode's io tree, and a
7192 * concurrent write against that previous extent map's
7193 * range and this range started (we unlock the ranges
7194 * in the io tree only when the bios complete and
7195 * buffered writes always lock pages before attempting
7196 * to lock range in the io tree).
7199 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7200 btrfs_start_ordered_extent(ordered, 1);
7203 btrfs_put_ordered_extent(ordered);
7206 * We could trigger writeback for this range (and wait
7207 * for it to complete) and then invalidate the pages for
7208 * this range (through invalidate_inode_pages2_range()),
7209 * but that can lead us to a deadlock with a concurrent
7210 * call to readahead (a buffered read or a defrag call
7211 * triggered a readahead) on a page lock due to an
7212 * ordered dio extent we created before but did not have
7213 * yet a corresponding bio submitted (whence it can not
7214 * complete), which makes readahead wait for that
7215 * ordered extent to complete while holding a lock on
7230 /* The callers of this must take lock_extent() */
7231 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7232 u64 len, u64 orig_start, u64 block_start,
7233 u64 block_len, u64 orig_block_len,
7234 u64 ram_bytes, int compress_type,
7237 struct extent_map_tree *em_tree;
7238 struct extent_map *em;
7241 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7242 type == BTRFS_ORDERED_COMPRESSED ||
7243 type == BTRFS_ORDERED_NOCOW ||
7244 type == BTRFS_ORDERED_REGULAR);
7246 em_tree = &inode->extent_tree;
7247 em = alloc_extent_map();
7249 return ERR_PTR(-ENOMEM);
7252 em->orig_start = orig_start;
7254 em->block_len = block_len;
7255 em->block_start = block_start;
7256 em->orig_block_len = orig_block_len;
7257 em->ram_bytes = ram_bytes;
7258 em->generation = -1;
7259 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7260 if (type == BTRFS_ORDERED_PREALLOC) {
7261 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7262 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7263 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7264 em->compress_type = compress_type;
7268 btrfs_drop_extent_cache(inode, em->start,
7269 em->start + em->len - 1, 0);
7270 write_lock(&em_tree->lock);
7271 ret = add_extent_mapping(em_tree, em, 1);
7272 write_unlock(&em_tree->lock);
7274 * The caller has taken lock_extent(), who could race with us
7277 } while (ret == -EEXIST);
7280 free_extent_map(em);
7281 return ERR_PTR(ret);
7284 /* em got 2 refs now, callers needs to do free_extent_map once. */
7289 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7290 struct inode *inode,
7291 struct btrfs_dio_data *dio_data,
7294 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7295 struct extent_map *em = *map;
7297 u64 block_start, orig_start, orig_block_len, ram_bytes;
7298 bool can_nocow = false;
7299 bool space_reserved = false;
7304 * We don't allocate a new extent in the following cases
7306 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7308 * 2) The extent is marked as PREALLOC. We're good to go here and can
7309 * just use the extent.
7312 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7313 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7314 em->block_start != EXTENT_MAP_HOLE)) {
7315 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7316 type = BTRFS_ORDERED_PREALLOC;
7318 type = BTRFS_ORDERED_NOCOW;
7319 len = min(len, em->len - (start - em->start));
7320 block_start = em->block_start + (start - em->start);
7322 if (can_nocow_extent(inode, start, &len, &orig_start,
7323 &orig_block_len, &ram_bytes, false) == 1 &&
7324 btrfs_inc_nocow_writers(fs_info, block_start))
7330 struct extent_map *em2;
7332 /* We can NOCOW, so only need to reserve metadata space. */
7333 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len);
7335 /* Our caller expects us to free the input extent map. */
7336 free_extent_map(em);
7338 btrfs_dec_nocow_writers(fs_info, block_start);
7341 space_reserved = true;
7343 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7344 orig_start, block_start,
7345 len, orig_block_len,
7347 btrfs_dec_nocow_writers(fs_info, block_start);
7348 if (type == BTRFS_ORDERED_PREALLOC) {
7349 free_extent_map(em);
7358 /* Our caller expects us to free the input extent map. */
7359 free_extent_map(em);
7362 /* We have to COW, so need to reserve metadata and data space. */
7363 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7364 &dio_data->data_reserved,
7368 space_reserved = true;
7370 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7376 len = min(len, em->len - (start - em->start));
7378 btrfs_delalloc_release_space(BTRFS_I(inode),
7379 dio_data->data_reserved,
7380 start + len, prev_len - len,
7385 * We have created our ordered extent, so we can now release our reservation
7386 * for an outstanding extent.
7388 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7391 * Need to update the i_size under the extent lock so buffered
7392 * readers will get the updated i_size when we unlock.
7394 if (start + len > i_size_read(inode))
7395 i_size_write(inode, start + len);
7397 if (ret && space_reserved) {
7398 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7400 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7402 btrfs_delalloc_release_space(BTRFS_I(inode),
7403 dio_data->data_reserved,
7405 extent_changeset_free(dio_data->data_reserved);
7406 dio_data->data_reserved = NULL;
7412 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7413 loff_t length, unsigned int flags, struct iomap *iomap,
7414 struct iomap *srcmap)
7416 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7417 struct extent_map *em;
7418 struct extent_state *cached_state = NULL;
7419 struct btrfs_dio_data *dio_data = NULL;
7420 u64 lockstart, lockend;
7421 const bool write = !!(flags & IOMAP_WRITE);
7424 bool unlock_extents = false;
7427 len = min_t(u64, len, fs_info->sectorsize);
7430 lockend = start + len - 1;
7433 * The generic stuff only does filemap_write_and_wait_range, which
7434 * isn't enough if we've written compressed pages to this area, so we
7435 * need to flush the dirty pages again to make absolutely sure that any
7436 * outstanding dirty pages are on disk.
7438 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7439 &BTRFS_I(inode)->runtime_flags)) {
7440 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7441 start + length - 1);
7446 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7450 iomap->private = dio_data;
7454 * If this errors out it's because we couldn't invalidate pagecache for
7455 * this range and we need to fallback to buffered.
7457 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7462 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7469 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7470 * io. INLINE is special, and we could probably kludge it in here, but
7471 * it's still buffered so for safety lets just fall back to the generic
7474 * For COMPRESSED we _have_ to read the entire extent in so we can
7475 * decompress it, so there will be buffering required no matter what we
7476 * do, so go ahead and fallback to buffered.
7478 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7479 * to buffered IO. Don't blame me, this is the price we pay for using
7482 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7483 em->block_start == EXTENT_MAP_INLINE) {
7484 free_extent_map(em);
7489 len = min(len, em->len - (start - em->start));
7492 * If we have a NOWAIT request and the range contains multiple extents
7493 * (or a mix of extents and holes), then we return -EAGAIN to make the
7494 * caller fallback to a context where it can do a blocking (without
7495 * NOWAIT) request. This way we avoid doing partial IO and returning
7496 * success to the caller, which is not optimal for writes and for reads
7497 * it can result in unexpected behaviour for an application.
7499 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7500 * iomap_dio_rw(), we can end up returning less data then what the caller
7501 * asked for, resulting in an unexpected, and incorrect, short read.
7502 * That is, the caller asked to read N bytes and we return less than that,
7503 * which is wrong unless we are crossing EOF. This happens if we get a
7504 * page fault error when trying to fault in pages for the buffer that is
7505 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7506 * have previously submitted bios for other extents in the range, in
7507 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7508 * those bios have completed by the time we get the page fault error,
7509 * which we return back to our caller - we should only return EIOCBQUEUED
7510 * after we have submitted bios for all the extents in the range.
7512 if ((flags & IOMAP_NOWAIT) && len < length) {
7513 free_extent_map(em);
7519 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7523 unlock_extents = true;
7524 /* Recalc len in case the new em is smaller than requested */
7525 len = min(len, em->len - (start - em->start));
7528 * We need to unlock only the end area that we aren't using.
7529 * The rest is going to be unlocked by the endio routine.
7531 lockstart = start + len;
7532 if (lockstart < lockend)
7533 unlock_extents = true;
7537 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7538 lockstart, lockend, &cached_state);
7540 free_extent_state(cached_state);
7543 * Translate extent map information to iomap.
7544 * We trim the extents (and move the addr) even though iomap code does
7545 * that, since we have locked only the parts we are performing I/O in.
7547 if ((em->block_start == EXTENT_MAP_HOLE) ||
7548 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7549 iomap->addr = IOMAP_NULL_ADDR;
7550 iomap->type = IOMAP_HOLE;
7552 iomap->addr = em->block_start + (start - em->start);
7553 iomap->type = IOMAP_MAPPED;
7555 iomap->offset = start;
7556 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7557 iomap->length = len;
7559 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7560 iomap->flags |= IOMAP_F_ZONE_APPEND;
7562 free_extent_map(em);
7567 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7575 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7576 ssize_t written, unsigned int flags, struct iomap *iomap)
7579 struct btrfs_dio_data *dio_data = iomap->private;
7580 size_t submitted = dio_data->submitted;
7581 const bool write = !!(flags & IOMAP_WRITE);
7583 if (!write && (iomap->type == IOMAP_HOLE)) {
7584 /* If reading from a hole, unlock and return */
7585 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7589 if (submitted < length) {
7591 length -= submitted;
7593 __endio_write_update_ordered(BTRFS_I(inode), pos,
7596 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7602 extent_changeset_free(dio_data->data_reserved);
7605 iomap->private = NULL;
7610 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7613 * This implies a barrier so that stores to dio_bio->bi_status before
7614 * this and loads of dio_bio->bi_status after this are fully ordered.
7616 if (!refcount_dec_and_test(&dip->refs))
7619 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7620 __endio_write_update_ordered(BTRFS_I(dip->inode),
7623 !dip->dio_bio->bi_status);
7625 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7627 dip->file_offset + dip->bytes - 1);
7630 bio_endio(dip->dio_bio);
7634 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7636 unsigned long bio_flags)
7638 struct btrfs_dio_private *dip = bio->bi_private;
7639 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7642 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7644 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7648 refcount_inc(&dip->refs);
7649 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7651 refcount_dec(&dip->refs);
7655 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7656 struct btrfs_bio *bbio,
7657 const bool uptodate)
7659 struct inode *inode = dip->inode;
7660 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7661 const u32 sectorsize = fs_info->sectorsize;
7662 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7663 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7664 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7665 struct bio_vec bvec;
7666 struct bvec_iter iter;
7668 blk_status_t err = BLK_STS_OK;
7670 __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) {
7671 unsigned int i, nr_sectors, pgoff;
7673 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7674 pgoff = bvec.bv_offset;
7675 for (i = 0; i < nr_sectors; i++) {
7676 u64 start = bbio->file_offset + bio_offset;
7678 ASSERT(pgoff < PAGE_SIZE);
7680 (!csum || !check_data_csum(inode, bbio,
7681 bio_offset, bvec.bv_page,
7683 clean_io_failure(fs_info, failure_tree, io_tree,
7684 start, bvec.bv_page,
7685 btrfs_ino(BTRFS_I(inode)),
7690 ret = btrfs_repair_one_sector(inode, &bbio->bio,
7691 bio_offset, bvec.bv_page, pgoff,
7692 start, bbio->mirror_num,
7693 submit_dio_repair_bio);
7695 err = errno_to_blk_status(ret);
7697 ASSERT(bio_offset + sectorsize > bio_offset);
7698 bio_offset += sectorsize;
7699 pgoff += sectorsize;
7705 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7706 const u64 offset, const u64 bytes,
7707 const bool uptodate)
7709 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7710 finish_ordered_fn, uptodate);
7713 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7715 u64 dio_file_offset)
7717 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
7720 static void btrfs_end_dio_bio(struct bio *bio)
7722 struct btrfs_dio_private *dip = bio->bi_private;
7723 struct btrfs_bio *bbio = btrfs_bio(bio);
7724 blk_status_t err = bio->bi_status;
7727 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7728 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7729 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7730 bio->bi_opf, bio->bi_iter.bi_sector,
7731 bio->bi_iter.bi_size, err);
7733 if (bio_op(bio) == REQ_OP_READ)
7734 err = btrfs_check_read_dio_bio(dip, bbio, !err);
7737 dip->dio_bio->bi_status = err;
7739 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
7742 btrfs_dio_private_put(dip);
7745 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7746 struct inode *inode, u64 file_offset, int async_submit)
7748 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7749 struct btrfs_dio_private *dip = bio->bi_private;
7750 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7753 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7755 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7758 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7763 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7766 if (write && async_submit) {
7767 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7768 btrfs_submit_bio_start_direct_io);
7772 * If we aren't doing async submit, calculate the csum of the
7775 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
7781 csum_offset = file_offset - dip->file_offset;
7782 csum_offset >>= fs_info->sectorsize_bits;
7783 csum_offset *= fs_info->csum_size;
7784 btrfs_bio(bio)->csum = dip->csums + csum_offset;
7787 ret = btrfs_map_bio(fs_info, bio, 0);
7793 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7794 * or ordered extents whether or not we submit any bios.
7796 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7797 struct inode *inode,
7800 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7801 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7803 struct btrfs_dio_private *dip;
7805 dip_size = sizeof(*dip);
7806 if (!write && csum) {
7807 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7810 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7811 dip_size += fs_info->csum_size * nblocks;
7814 dip = kzalloc(dip_size, GFP_NOFS);
7819 dip->file_offset = file_offset;
7820 dip->bytes = dio_bio->bi_iter.bi_size;
7821 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
7822 dip->dio_bio = dio_bio;
7823 refcount_set(&dip->refs, 1);
7827 static void btrfs_submit_direct(const struct iomap_iter *iter,
7828 struct bio *dio_bio, loff_t file_offset)
7830 struct inode *inode = iter->inode;
7831 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7832 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7833 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7834 BTRFS_BLOCK_GROUP_RAID56_MASK);
7835 struct btrfs_dio_private *dip;
7838 int async_submit = 0;
7840 u64 clone_offset = 0;
7844 blk_status_t status;
7845 struct btrfs_io_geometry geom;
7846 struct btrfs_dio_data *dio_data = iter->iomap.private;
7847 struct extent_map *em = NULL;
7849 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7852 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7853 file_offset + dio_bio->bi_iter.bi_size - 1);
7855 dio_bio->bi_status = BLK_STS_RESOURCE;
7862 * Load the csums up front to reduce csum tree searches and
7863 * contention when submitting bios.
7865 * If we have csums disabled this will do nothing.
7867 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
7868 if (status != BLK_STS_OK)
7872 start_sector = dio_bio->bi_iter.bi_sector;
7873 submit_len = dio_bio->bi_iter.bi_size;
7876 logical = start_sector << 9;
7877 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
7879 status = errno_to_blk_status(PTR_ERR(em));
7883 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
7886 status = errno_to_blk_status(ret);
7890 clone_len = min(submit_len, geom.len);
7891 ASSERT(clone_len <= UINT_MAX);
7894 * This will never fail as it's passing GPF_NOFS and
7895 * the allocation is backed by btrfs_bioset.
7897 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7898 bio->bi_private = dip;
7899 bio->bi_end_io = btrfs_end_dio_bio;
7900 btrfs_bio(bio)->file_offset = file_offset;
7902 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
7903 status = extract_ordered_extent(BTRFS_I(inode), bio,
7911 ASSERT(submit_len >= clone_len);
7912 submit_len -= clone_len;
7915 * Increase the count before we submit the bio so we know
7916 * the end IO handler won't happen before we increase the
7917 * count. Otherwise, the dip might get freed before we're
7918 * done setting it up.
7920 * We transfer the initial reference to the last bio, so we
7921 * don't need to increment the reference count for the last one.
7923 if (submit_len > 0) {
7924 refcount_inc(&dip->refs);
7926 * If we are submitting more than one bio, submit them
7927 * all asynchronously. The exception is RAID 5 or 6, as
7928 * asynchronous checksums make it difficult to collect
7929 * full stripe writes.
7935 status = btrfs_submit_dio_bio(bio, inode, file_offset,
7940 refcount_dec(&dip->refs);
7944 dio_data->submitted += clone_len;
7945 clone_offset += clone_len;
7946 start_sector += clone_len >> 9;
7947 file_offset += clone_len;
7949 free_extent_map(em);
7950 } while (submit_len > 0);
7954 free_extent_map(em);
7956 dip->dio_bio->bi_status = status;
7957 btrfs_dio_private_put(dip);
7960 const struct iomap_ops btrfs_dio_iomap_ops = {
7961 .iomap_begin = btrfs_dio_iomap_begin,
7962 .iomap_end = btrfs_dio_iomap_end,
7965 const struct iomap_dio_ops btrfs_dio_ops = {
7966 .submit_io = btrfs_submit_direct,
7969 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7974 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7978 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7981 int btrfs_readpage(struct file *file, struct page *page)
7983 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
7984 u64 start = page_offset(page);
7985 u64 end = start + PAGE_SIZE - 1;
7986 struct btrfs_bio_ctrl bio_ctrl = { 0 };
7989 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
7991 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
7995 ret2 = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8002 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8004 struct inode *inode = page->mapping->host;
8007 if (current->flags & PF_MEMALLOC) {
8008 redirty_page_for_writepage(wbc, page);
8014 * If we are under memory pressure we will call this directly from the
8015 * VM, we need to make sure we have the inode referenced for the ordered
8016 * extent. If not just return like we didn't do anything.
8018 if (!igrab(inode)) {
8019 redirty_page_for_writepage(wbc, page);
8020 return AOP_WRITEPAGE_ACTIVATE;
8022 ret = extent_write_full_page(page, wbc);
8023 btrfs_add_delayed_iput(inode);
8027 static int btrfs_writepages(struct address_space *mapping,
8028 struct writeback_control *wbc)
8030 return extent_writepages(mapping, wbc);
8033 static void btrfs_readahead(struct readahead_control *rac)
8035 extent_readahead(rac);
8039 * For releasepage() and invalidate_folio() we have a race window where
8040 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8041 * If we continue to release/invalidate the page, we could cause use-after-free
8042 * for subpage spinlock. So this function is to spin and wait for subpage
8045 static void wait_subpage_spinlock(struct page *page)
8047 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8048 struct btrfs_subpage *subpage;
8050 if (fs_info->sectorsize == PAGE_SIZE)
8053 ASSERT(PagePrivate(page) && page->private);
8054 subpage = (struct btrfs_subpage *)page->private;
8057 * This may look insane as we just acquire the spinlock and release it,
8058 * without doing anything. But we just want to make sure no one is
8059 * still holding the subpage spinlock.
8060 * And since the page is not dirty nor writeback, and we have page
8061 * locked, the only possible way to hold a spinlock is from the endio
8062 * function to clear page writeback.
8064 * Here we just acquire the spinlock so that all existing callers
8065 * should exit and we're safe to release/invalidate the page.
8067 spin_lock_irq(&subpage->lock);
8068 spin_unlock_irq(&subpage->lock);
8071 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8073 int ret = try_release_extent_mapping(page, gfp_flags);
8076 wait_subpage_spinlock(page);
8077 clear_page_extent_mapped(page);
8082 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8084 if (PageWriteback(page) || PageDirty(page))
8086 return __btrfs_releasepage(page, gfp_flags);
8089 #ifdef CONFIG_MIGRATION
8090 static int btrfs_migratepage(struct address_space *mapping,
8091 struct page *newpage, struct page *page,
8092 enum migrate_mode mode)
8096 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8097 if (ret != MIGRATEPAGE_SUCCESS)
8100 if (page_has_private(page))
8101 attach_page_private(newpage, detach_page_private(page));
8103 if (PageOrdered(page)) {
8104 ClearPageOrdered(page);
8105 SetPageOrdered(newpage);
8108 if (mode != MIGRATE_SYNC_NO_COPY)
8109 migrate_page_copy(newpage, page);
8111 migrate_page_states(newpage, page);
8112 return MIGRATEPAGE_SUCCESS;
8116 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8119 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8120 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8121 struct extent_io_tree *tree = &inode->io_tree;
8122 struct extent_state *cached_state = NULL;
8123 u64 page_start = folio_pos(folio);
8124 u64 page_end = page_start + folio_size(folio) - 1;
8126 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8129 * We have folio locked so no new ordered extent can be created on this
8130 * page, nor bio can be submitted for this folio.
8132 * But already submitted bio can still be finished on this folio.
8133 * Furthermore, endio function won't skip folio which has Ordered
8134 * (Private2) already cleared, so it's possible for endio and
8135 * invalidate_folio to do the same ordered extent accounting twice
8138 * So here we wait for any submitted bios to finish, so that we won't
8139 * do double ordered extent accounting on the same folio.
8141 folio_wait_writeback(folio);
8142 wait_subpage_spinlock(&folio->page);
8145 * For subpage case, we have call sites like
8146 * btrfs_punch_hole_lock_range() which passes range not aligned to
8148 * If the range doesn't cover the full folio, we don't need to and
8149 * shouldn't clear page extent mapped, as folio->private can still
8150 * record subpage dirty bits for other part of the range.
8152 * For cases that invalidate the full folio even the range doesn't
8153 * cover the full folio, like invalidating the last folio, we're
8154 * still safe to wait for ordered extent to finish.
8156 if (!(offset == 0 && length == folio_size(folio))) {
8157 btrfs_releasepage(&folio->page, GFP_NOFS);
8161 if (!inode_evicting)
8162 lock_extent_bits(tree, page_start, page_end, &cached_state);
8165 while (cur < page_end) {
8166 struct btrfs_ordered_extent *ordered;
8171 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8172 page_end + 1 - cur);
8174 range_end = page_end;
8176 * No ordered extent covering this range, we are safe
8177 * to delete all extent states in the range.
8179 delete_states = true;
8182 if (ordered->file_offset > cur) {
8184 * There is a range between [cur, oe->file_offset) not
8185 * covered by any ordered extent.
8186 * We are safe to delete all extent states, and handle
8187 * the ordered extent in the next iteration.
8189 range_end = ordered->file_offset - 1;
8190 delete_states = true;
8194 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8196 ASSERT(range_end + 1 - cur < U32_MAX);
8197 range_len = range_end + 1 - cur;
8198 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8200 * If Ordered (Private2) is cleared, it means endio has
8201 * already been executed for the range.
8202 * We can't delete the extent states as
8203 * btrfs_finish_ordered_io() may still use some of them.
8205 delete_states = false;
8208 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8211 * IO on this page will never be started, so we need to account
8212 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8213 * here, must leave that up for the ordered extent completion.
8215 * This will also unlock the range for incoming
8216 * btrfs_finish_ordered_io().
8218 if (!inode_evicting)
8219 clear_extent_bit(tree, cur, range_end,
8221 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8222 EXTENT_DEFRAG, 1, 0, &cached_state);
8224 spin_lock_irq(&inode->ordered_tree.lock);
8225 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8226 ordered->truncated_len = min(ordered->truncated_len,
8227 cur - ordered->file_offset);
8228 spin_unlock_irq(&inode->ordered_tree.lock);
8230 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8231 cur, range_end + 1 - cur)) {
8232 btrfs_finish_ordered_io(ordered);
8234 * The ordered extent has finished, now we're again
8235 * safe to delete all extent states of the range.
8237 delete_states = true;
8240 * btrfs_finish_ordered_io() will get executed by endio
8241 * of other pages, thus we can't delete extent states
8244 delete_states = false;
8248 btrfs_put_ordered_extent(ordered);
8250 * Qgroup reserved space handler
8251 * Sector(s) here will be either:
8253 * 1) Already written to disk or bio already finished
8254 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8255 * Qgroup will be handled by its qgroup_record then.
8256 * btrfs_qgroup_free_data() call will do nothing here.
8258 * 2) Not written to disk yet
8259 * Then btrfs_qgroup_free_data() call will clear the
8260 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8261 * reserved data space.
8262 * Since the IO will never happen for this page.
8264 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8265 if (!inode_evicting) {
8266 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8267 EXTENT_DELALLOC | EXTENT_UPTODATE |
8268 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8269 delete_states, &cached_state);
8271 cur = range_end + 1;
8274 * We have iterated through all ordered extents of the page, the page
8275 * should not have Ordered (Private2) anymore, or the above iteration
8276 * did something wrong.
8278 ASSERT(!folio_test_ordered(folio));
8279 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8280 if (!inode_evicting)
8281 __btrfs_releasepage(&folio->page, GFP_NOFS);
8282 clear_page_extent_mapped(&folio->page);
8286 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8287 * called from a page fault handler when a page is first dirtied. Hence we must
8288 * be careful to check for EOF conditions here. We set the page up correctly
8289 * for a written page which means we get ENOSPC checking when writing into
8290 * holes and correct delalloc and unwritten extent mapping on filesystems that
8291 * support these features.
8293 * We are not allowed to take the i_mutex here so we have to play games to
8294 * protect against truncate races as the page could now be beyond EOF. Because
8295 * truncate_setsize() writes the inode size before removing pages, once we have
8296 * the page lock we can determine safely if the page is beyond EOF. If it is not
8297 * beyond EOF, then the page is guaranteed safe against truncation until we
8300 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8302 struct page *page = vmf->page;
8303 struct inode *inode = file_inode(vmf->vma->vm_file);
8304 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8305 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8306 struct btrfs_ordered_extent *ordered;
8307 struct extent_state *cached_state = NULL;
8308 struct extent_changeset *data_reserved = NULL;
8309 unsigned long zero_start;
8319 reserved_space = PAGE_SIZE;
8321 sb_start_pagefault(inode->i_sb);
8322 page_start = page_offset(page);
8323 page_end = page_start + PAGE_SIZE - 1;
8327 * Reserving delalloc space after obtaining the page lock can lead to
8328 * deadlock. For example, if a dirty page is locked by this function
8329 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8330 * dirty page write out, then the btrfs_writepage() function could
8331 * end up waiting indefinitely to get a lock on the page currently
8332 * being processed by btrfs_page_mkwrite() function.
8334 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8335 page_start, reserved_space);
8337 ret2 = file_update_time(vmf->vma->vm_file);
8341 ret = vmf_error(ret2);
8347 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8349 down_read(&BTRFS_I(inode)->i_mmap_lock);
8351 size = i_size_read(inode);
8353 if ((page->mapping != inode->i_mapping) ||
8354 (page_start >= size)) {
8355 /* page got truncated out from underneath us */
8358 wait_on_page_writeback(page);
8360 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8361 ret2 = set_page_extent_mapped(page);
8363 ret = vmf_error(ret2);
8364 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8369 * we can't set the delalloc bits if there are pending ordered
8370 * extents. Drop our locks and wait for them to finish
8372 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8375 unlock_extent_cached(io_tree, page_start, page_end,
8378 up_read(&BTRFS_I(inode)->i_mmap_lock);
8379 btrfs_start_ordered_extent(ordered, 1);
8380 btrfs_put_ordered_extent(ordered);
8384 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8385 reserved_space = round_up(size - page_start,
8386 fs_info->sectorsize);
8387 if (reserved_space < PAGE_SIZE) {
8388 end = page_start + reserved_space - 1;
8389 btrfs_delalloc_release_space(BTRFS_I(inode),
8390 data_reserved, page_start,
8391 PAGE_SIZE - reserved_space, true);
8396 * page_mkwrite gets called when the page is firstly dirtied after it's
8397 * faulted in, but write(2) could also dirty a page and set delalloc
8398 * bits, thus in this case for space account reason, we still need to
8399 * clear any delalloc bits within this page range since we have to
8400 * reserve data&meta space before lock_page() (see above comments).
8402 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8403 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8404 EXTENT_DEFRAG, 0, 0, &cached_state);
8406 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8409 unlock_extent_cached(io_tree, page_start, page_end,
8411 ret = VM_FAULT_SIGBUS;
8415 /* page is wholly or partially inside EOF */
8416 if (page_start + PAGE_SIZE > size)
8417 zero_start = offset_in_page(size);
8419 zero_start = PAGE_SIZE;
8421 if (zero_start != PAGE_SIZE) {
8422 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8423 flush_dcache_page(page);
8425 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8426 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8427 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8429 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8431 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8432 up_read(&BTRFS_I(inode)->i_mmap_lock);
8434 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8435 sb_end_pagefault(inode->i_sb);
8436 extent_changeset_free(data_reserved);
8437 return VM_FAULT_LOCKED;
8441 up_read(&BTRFS_I(inode)->i_mmap_lock);
8443 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8444 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8445 reserved_space, (ret != 0));
8447 sb_end_pagefault(inode->i_sb);
8448 extent_changeset_free(data_reserved);
8452 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8454 struct btrfs_truncate_control control = {
8455 .inode = BTRFS_I(inode),
8456 .ino = btrfs_ino(BTRFS_I(inode)),
8457 .min_type = BTRFS_EXTENT_DATA_KEY,
8458 .clear_extent_range = true,
8460 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8461 struct btrfs_root *root = BTRFS_I(inode)->root;
8462 struct btrfs_block_rsv *rsv;
8464 struct btrfs_trans_handle *trans;
8465 u64 mask = fs_info->sectorsize - 1;
8466 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8468 if (!skip_writeback) {
8469 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8476 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8477 * things going on here:
8479 * 1) We need to reserve space to update our inode.
8481 * 2) We need to have something to cache all the space that is going to
8482 * be free'd up by the truncate operation, but also have some slack
8483 * space reserved in case it uses space during the truncate (thank you
8484 * very much snapshotting).
8486 * And we need these to be separate. The fact is we can use a lot of
8487 * space doing the truncate, and we have no earthly idea how much space
8488 * we will use, so we need the truncate reservation to be separate so it
8489 * doesn't end up using space reserved for updating the inode. We also
8490 * need to be able to stop the transaction and start a new one, which
8491 * means we need to be able to update the inode several times, and we
8492 * have no idea of knowing how many times that will be, so we can't just
8493 * reserve 1 item for the entirety of the operation, so that has to be
8494 * done separately as well.
8496 * So that leaves us with
8498 * 1) rsv - for the truncate reservation, which we will steal from the
8499 * transaction reservation.
8500 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8501 * updating the inode.
8503 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8506 rsv->size = min_size;
8510 * 1 for the truncate slack space
8511 * 1 for updating the inode.
8513 trans = btrfs_start_transaction(root, 2);
8514 if (IS_ERR(trans)) {
8515 ret = PTR_ERR(trans);
8519 /* Migrate the slack space for the truncate to our reserve */
8520 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8524 trans->block_rsv = rsv;
8527 struct extent_state *cached_state = NULL;
8528 const u64 new_size = inode->i_size;
8529 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8531 control.new_size = new_size;
8532 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8535 * We want to drop from the next block forward in case this new
8536 * size is not block aligned since we will be keeping the last
8537 * block of the extent just the way it is.
8539 btrfs_drop_extent_cache(BTRFS_I(inode),
8540 ALIGN(new_size, fs_info->sectorsize),
8543 ret = btrfs_truncate_inode_items(trans, root, &control);
8545 inode_sub_bytes(inode, control.sub_bytes);
8546 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8548 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8549 (u64)-1, &cached_state);
8551 trans->block_rsv = &fs_info->trans_block_rsv;
8552 if (ret != -ENOSPC && ret != -EAGAIN)
8555 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8559 btrfs_end_transaction(trans);
8560 btrfs_btree_balance_dirty(fs_info);
8562 trans = btrfs_start_transaction(root, 2);
8563 if (IS_ERR(trans)) {
8564 ret = PTR_ERR(trans);
8569 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8570 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8571 rsv, min_size, false);
8572 BUG_ON(ret); /* shouldn't happen */
8573 trans->block_rsv = rsv;
8577 * We can't call btrfs_truncate_block inside a trans handle as we could
8578 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8579 * know we've truncated everything except the last little bit, and can
8580 * do btrfs_truncate_block and then update the disk_i_size.
8582 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8583 btrfs_end_transaction(trans);
8584 btrfs_btree_balance_dirty(fs_info);
8586 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8589 trans = btrfs_start_transaction(root, 1);
8590 if (IS_ERR(trans)) {
8591 ret = PTR_ERR(trans);
8594 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8600 trans->block_rsv = &fs_info->trans_block_rsv;
8601 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8605 ret2 = btrfs_end_transaction(trans);
8608 btrfs_btree_balance_dirty(fs_info);
8611 btrfs_free_block_rsv(fs_info, rsv);
8613 * So if we truncate and then write and fsync we normally would just
8614 * write the extents that changed, which is a problem if we need to
8615 * first truncate that entire inode. So set this flag so we write out
8616 * all of the extents in the inode to the sync log so we're completely
8619 * If no extents were dropped or trimmed we don't need to force the next
8620 * fsync to truncate all the inode's items from the log and re-log them
8621 * all. This means the truncate operation did not change the file size,
8622 * or changed it to a smaller size but there was only an implicit hole
8623 * between the old i_size and the new i_size, and there were no prealloc
8624 * extents beyond i_size to drop.
8626 if (control.extents_found > 0)
8627 btrfs_set_inode_full_sync(BTRFS_I(inode));
8632 struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns,
8635 struct inode *inode;
8637 inode = new_inode(dir->i_sb);
8640 * Subvolumes don't inherit the sgid bit or the parent's gid if
8641 * the parent's sgid bit is set. This is probably a bug.
8643 inode_init_owner(mnt_userns, inode, NULL,
8644 S_IFDIR | (~current_umask() & S_IRWXUGO));
8645 inode->i_op = &btrfs_dir_inode_operations;
8646 inode->i_fop = &btrfs_dir_file_operations;
8652 * create a new subvolume directory/inode (helper for the ioctl).
8654 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8655 struct btrfs_root *parent_root,
8656 struct inode *inode)
8658 struct btrfs_root *new_root = BTRFS_I(inode)->root;
8662 err = btrfs_new_inode(trans, new_root, inode, NULL, "..", 2, &index);
8666 unlock_new_inode(inode);
8668 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8670 btrfs_err(new_root->fs_info,
8671 "error inheriting subvolume %llu properties: %d",
8672 new_root->root_key.objectid, err);
8674 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8678 struct inode *btrfs_alloc_inode(struct super_block *sb)
8680 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8681 struct btrfs_inode *ei;
8682 struct inode *inode;
8684 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8691 ei->last_sub_trans = 0;
8692 ei->logged_trans = 0;
8693 ei->delalloc_bytes = 0;
8694 ei->new_delalloc_bytes = 0;
8695 ei->defrag_bytes = 0;
8696 ei->disk_i_size = 0;
8700 ei->index_cnt = (u64)-1;
8702 ei->last_unlink_trans = 0;
8703 ei->last_reflink_trans = 0;
8704 ei->last_log_commit = 0;
8706 spin_lock_init(&ei->lock);
8707 ei->outstanding_extents = 0;
8708 if (sb->s_magic != BTRFS_TEST_MAGIC)
8709 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8710 BTRFS_BLOCK_RSV_DELALLOC);
8711 ei->runtime_flags = 0;
8712 ei->prop_compress = BTRFS_COMPRESS_NONE;
8713 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8715 ei->delayed_node = NULL;
8717 ei->i_otime.tv_sec = 0;
8718 ei->i_otime.tv_nsec = 0;
8720 inode = &ei->vfs_inode;
8721 extent_map_tree_init(&ei->extent_tree);
8722 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8723 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8724 IO_TREE_INODE_IO_FAILURE, inode);
8725 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8726 IO_TREE_INODE_FILE_EXTENT, inode);
8727 ei->io_tree.track_uptodate = true;
8728 ei->io_failure_tree.track_uptodate = true;
8729 atomic_set(&ei->sync_writers, 0);
8730 mutex_init(&ei->log_mutex);
8731 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8732 INIT_LIST_HEAD(&ei->delalloc_inodes);
8733 INIT_LIST_HEAD(&ei->delayed_iput);
8734 RB_CLEAR_NODE(&ei->rb_node);
8735 init_rwsem(&ei->i_mmap_lock);
8740 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8741 void btrfs_test_destroy_inode(struct inode *inode)
8743 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8744 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8748 void btrfs_free_inode(struct inode *inode)
8750 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8753 void btrfs_destroy_inode(struct inode *vfs_inode)
8755 struct btrfs_ordered_extent *ordered;
8756 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8757 struct btrfs_root *root = inode->root;
8759 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8760 WARN_ON(vfs_inode->i_data.nrpages);
8761 WARN_ON(inode->block_rsv.reserved);
8762 WARN_ON(inode->block_rsv.size);
8763 WARN_ON(inode->outstanding_extents);
8764 if (!S_ISDIR(vfs_inode->i_mode)) {
8765 WARN_ON(inode->delalloc_bytes);
8766 WARN_ON(inode->new_delalloc_bytes);
8768 WARN_ON(inode->csum_bytes);
8769 WARN_ON(inode->defrag_bytes);
8772 * This can happen where we create an inode, but somebody else also
8773 * created the same inode and we need to destroy the one we already
8780 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8784 btrfs_err(root->fs_info,
8785 "found ordered extent %llu %llu on inode cleanup",
8786 ordered->file_offset, ordered->num_bytes);
8787 btrfs_remove_ordered_extent(inode, ordered);
8788 btrfs_put_ordered_extent(ordered);
8789 btrfs_put_ordered_extent(ordered);
8792 btrfs_qgroup_check_reserved_leak(inode);
8793 inode_tree_del(inode);
8794 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8795 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8796 btrfs_put_root(inode->root);
8799 int btrfs_drop_inode(struct inode *inode)
8801 struct btrfs_root *root = BTRFS_I(inode)->root;
8806 /* the snap/subvol tree is on deleting */
8807 if (btrfs_root_refs(&root->root_item) == 0)
8810 return generic_drop_inode(inode);
8813 static void init_once(void *foo)
8815 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8817 inode_init_once(&ei->vfs_inode);
8820 void __cold btrfs_destroy_cachep(void)
8823 * Make sure all delayed rcu free inodes are flushed before we
8827 kmem_cache_destroy(btrfs_inode_cachep);
8828 kmem_cache_destroy(btrfs_trans_handle_cachep);
8829 kmem_cache_destroy(btrfs_path_cachep);
8830 kmem_cache_destroy(btrfs_free_space_cachep);
8831 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8834 int __init btrfs_init_cachep(void)
8836 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8837 sizeof(struct btrfs_inode), 0,
8838 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8840 if (!btrfs_inode_cachep)
8843 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8844 sizeof(struct btrfs_trans_handle), 0,
8845 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8846 if (!btrfs_trans_handle_cachep)
8849 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8850 sizeof(struct btrfs_path), 0,
8851 SLAB_MEM_SPREAD, NULL);
8852 if (!btrfs_path_cachep)
8855 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8856 sizeof(struct btrfs_free_space), 0,
8857 SLAB_MEM_SPREAD, NULL);
8858 if (!btrfs_free_space_cachep)
8861 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8862 PAGE_SIZE, PAGE_SIZE,
8863 SLAB_MEM_SPREAD, NULL);
8864 if (!btrfs_free_space_bitmap_cachep)
8869 btrfs_destroy_cachep();
8873 static int btrfs_getattr(struct user_namespace *mnt_userns,
8874 const struct path *path, struct kstat *stat,
8875 u32 request_mask, unsigned int flags)
8879 struct inode *inode = d_inode(path->dentry);
8880 u32 blocksize = inode->i_sb->s_blocksize;
8881 u32 bi_flags = BTRFS_I(inode)->flags;
8882 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8884 stat->result_mask |= STATX_BTIME;
8885 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8886 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8887 if (bi_flags & BTRFS_INODE_APPEND)
8888 stat->attributes |= STATX_ATTR_APPEND;
8889 if (bi_flags & BTRFS_INODE_COMPRESS)
8890 stat->attributes |= STATX_ATTR_COMPRESSED;
8891 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8892 stat->attributes |= STATX_ATTR_IMMUTABLE;
8893 if (bi_flags & BTRFS_INODE_NODUMP)
8894 stat->attributes |= STATX_ATTR_NODUMP;
8895 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8896 stat->attributes |= STATX_ATTR_VERITY;
8898 stat->attributes_mask |= (STATX_ATTR_APPEND |
8899 STATX_ATTR_COMPRESSED |
8900 STATX_ATTR_IMMUTABLE |
8903 generic_fillattr(mnt_userns, inode, stat);
8904 stat->dev = BTRFS_I(inode)->root->anon_dev;
8906 spin_lock(&BTRFS_I(inode)->lock);
8907 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8908 inode_bytes = inode_get_bytes(inode);
8909 spin_unlock(&BTRFS_I(inode)->lock);
8910 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8911 ALIGN(delalloc_bytes, blocksize)) >> 9;
8915 static int btrfs_rename_exchange(struct inode *old_dir,
8916 struct dentry *old_dentry,
8917 struct inode *new_dir,
8918 struct dentry *new_dentry)
8920 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8921 struct btrfs_trans_handle *trans;
8922 unsigned int trans_num_items;
8923 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8924 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8925 struct inode *new_inode = new_dentry->d_inode;
8926 struct inode *old_inode = old_dentry->d_inode;
8927 struct timespec64 ctime = current_time(old_inode);
8928 struct btrfs_rename_ctx old_rename_ctx;
8929 struct btrfs_rename_ctx new_rename_ctx;
8930 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8931 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8936 bool need_abort = false;
8939 * For non-subvolumes allow exchange only within one subvolume, in the
8940 * same inode namespace. Two subvolumes (represented as directory) can
8941 * be exchanged as they're a logical link and have a fixed inode number.
8944 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8945 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8948 /* close the race window with snapshot create/destroy ioctl */
8949 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8950 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8951 down_read(&fs_info->subvol_sem);
8955 * 1 to remove old dir item
8956 * 1 to remove old dir index
8957 * 1 to add new dir item
8958 * 1 to add new dir index
8959 * 1 to update parent inode
8961 * If the parents are the same, we only need to account for one
8963 trans_num_items = (old_dir == new_dir ? 9 : 10);
8964 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8966 * 1 to remove old root ref
8967 * 1 to remove old root backref
8968 * 1 to add new root ref
8969 * 1 to add new root backref
8971 trans_num_items += 4;
8974 * 1 to update inode item
8975 * 1 to remove old inode ref
8976 * 1 to add new inode ref
8978 trans_num_items += 3;
8980 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8981 trans_num_items += 4;
8983 trans_num_items += 3;
8984 trans = btrfs_start_transaction(root, trans_num_items);
8985 if (IS_ERR(trans)) {
8986 ret = PTR_ERR(trans);
8991 ret = btrfs_record_root_in_trans(trans, dest);
8997 * We need to find a free sequence number both in the source and
8998 * in the destination directory for the exchange.
9000 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9003 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9007 BTRFS_I(old_inode)->dir_index = 0ULL;
9008 BTRFS_I(new_inode)->dir_index = 0ULL;
9010 /* Reference for the source. */
9011 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9012 /* force full log commit if subvolume involved. */
9013 btrfs_set_log_full_commit(trans);
9015 ret = btrfs_insert_inode_ref(trans, dest,
9016 new_dentry->d_name.name,
9017 new_dentry->d_name.len,
9019 btrfs_ino(BTRFS_I(new_dir)),
9026 /* And now for the dest. */
9027 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9028 /* force full log commit if subvolume involved. */
9029 btrfs_set_log_full_commit(trans);
9031 ret = btrfs_insert_inode_ref(trans, root,
9032 old_dentry->d_name.name,
9033 old_dentry->d_name.len,
9035 btrfs_ino(BTRFS_I(old_dir)),
9039 btrfs_abort_transaction(trans, ret);
9044 /* Update inode version and ctime/mtime. */
9045 inode_inc_iversion(old_dir);
9046 inode_inc_iversion(new_dir);
9047 inode_inc_iversion(old_inode);
9048 inode_inc_iversion(new_inode);
9049 old_dir->i_ctime = old_dir->i_mtime = ctime;
9050 new_dir->i_ctime = new_dir->i_mtime = ctime;
9051 old_inode->i_ctime = ctime;
9052 new_inode->i_ctime = ctime;
9054 if (old_dentry->d_parent != new_dentry->d_parent) {
9055 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9056 BTRFS_I(old_inode), 1);
9057 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9058 BTRFS_I(new_inode), 1);
9061 /* src is a subvolume */
9062 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9063 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9064 } else { /* src is an inode */
9065 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9066 BTRFS_I(old_dentry->d_inode),
9067 old_dentry->d_name.name,
9068 old_dentry->d_name.len,
9071 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9074 btrfs_abort_transaction(trans, ret);
9078 /* dest is a subvolume */
9079 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9080 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9081 } else { /* dest is an inode */
9082 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9083 BTRFS_I(new_dentry->d_inode),
9084 new_dentry->d_name.name,
9085 new_dentry->d_name.len,
9088 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9091 btrfs_abort_transaction(trans, ret);
9095 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9096 new_dentry->d_name.name,
9097 new_dentry->d_name.len, 0, old_idx);
9099 btrfs_abort_transaction(trans, ret);
9103 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9104 old_dentry->d_name.name,
9105 old_dentry->d_name.len, 0, new_idx);
9107 btrfs_abort_transaction(trans, ret);
9111 if (old_inode->i_nlink == 1)
9112 BTRFS_I(old_inode)->dir_index = old_idx;
9113 if (new_inode->i_nlink == 1)
9114 BTRFS_I(new_inode)->dir_index = new_idx;
9117 * Now pin the logs of the roots. We do it to ensure that no other task
9118 * can sync the logs while we are in progress with the rename, because
9119 * that could result in an inconsistency in case any of the inodes that
9120 * are part of this rename operation were logged before.
9122 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9123 btrfs_pin_log_trans(root);
9124 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9125 btrfs_pin_log_trans(dest);
9127 /* Do the log updates for all inodes. */
9128 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9129 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9130 old_rename_ctx.index, new_dentry->d_parent);
9131 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9132 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9133 new_rename_ctx.index, old_dentry->d_parent);
9135 /* Now unpin the logs. */
9136 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9137 btrfs_end_log_trans(root);
9138 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9139 btrfs_end_log_trans(dest);
9141 ret2 = btrfs_end_transaction(trans);
9142 ret = ret ? ret : ret2;
9144 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9145 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9146 up_read(&fs_info->subvol_sem);
9151 static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns,
9154 struct inode *inode;
9156 inode = new_inode(dir->i_sb);
9158 inode_init_owner(mnt_userns, inode, dir,
9159 S_IFCHR | WHITEOUT_MODE);
9160 inode->i_op = &btrfs_special_inode_operations;
9161 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9166 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9167 struct btrfs_root *root,
9168 struct inode *inode, struct inode *dir,
9169 struct dentry *dentry)
9174 ret = btrfs_new_inode(trans, root, inode, dir, dentry->d_name.name,
9175 dentry->d_name.len, &index);
9181 ret = btrfs_init_inode_security(trans, inode, dir,
9186 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
9187 dentry->d_name.name, dentry->d_name.len, 0, index);
9191 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9193 unlock_new_inode(inode);
9195 inode_dec_link_count(inode);
9201 static int btrfs_rename(struct user_namespace *mnt_userns,
9202 struct inode *old_dir, struct dentry *old_dentry,
9203 struct inode *new_dir, struct dentry *new_dentry,
9206 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9207 struct inode *whiteout_inode;
9208 struct btrfs_trans_handle *trans;
9209 unsigned int trans_num_items;
9210 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9211 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9212 struct inode *new_inode = d_inode(new_dentry);
9213 struct inode *old_inode = d_inode(old_dentry);
9214 struct btrfs_rename_ctx rename_ctx;
9218 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9220 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9223 /* we only allow rename subvolume link between subvolumes */
9224 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9227 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9228 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9231 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9232 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9236 /* check for collisions, even if the name isn't there */
9237 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9238 new_dentry->d_name.name,
9239 new_dentry->d_name.len);
9242 if (ret == -EEXIST) {
9244 * eexist without a new_inode */
9245 if (WARN_ON(!new_inode)) {
9249 /* maybe -EOVERFLOW */
9256 * we're using rename to replace one file with another. Start IO on it
9257 * now so we don't add too much work to the end of the transaction
9259 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9260 filemap_flush(old_inode->i_mapping);
9262 if (flags & RENAME_WHITEOUT) {
9263 whiteout_inode = new_whiteout_inode(mnt_userns, old_dir);
9264 if (!whiteout_inode)
9268 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9269 /* Close the race window with snapshot create/destroy ioctl */
9270 down_read(&fs_info->subvol_sem);
9272 * 1 to remove old root ref
9273 * 1 to remove old root backref
9274 * 1 to add new root ref
9275 * 1 to add new root backref
9277 trans_num_items = 4;
9281 * 1 to remove old inode ref
9282 * 1 to add new inode ref
9284 trans_num_items = 3;
9287 * 1 to remove old dir item
9288 * 1 to remove old dir index
9289 * 1 to update old parent inode
9290 * 1 to add new dir item
9291 * 1 to add new dir index
9292 * 1 to update new parent inode (if it's not the same as the old parent)
9294 trans_num_items += 6;
9295 if (new_dir != old_dir)
9300 * 1 to remove inode ref
9301 * 1 to remove dir item
9302 * 1 to remove dir index
9303 * 1 to possibly add orphan item
9305 trans_num_items += 5;
9307 if (flags & RENAME_WHITEOUT)
9308 trans_num_items += 5;
9309 trans = btrfs_start_transaction(root, trans_num_items);
9310 if (IS_ERR(trans)) {
9311 ret = PTR_ERR(trans);
9316 ret = btrfs_record_root_in_trans(trans, dest);
9321 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9325 BTRFS_I(old_inode)->dir_index = 0ULL;
9326 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9327 /* force full log commit if subvolume involved. */
9328 btrfs_set_log_full_commit(trans);
9330 ret = btrfs_insert_inode_ref(trans, dest,
9331 new_dentry->d_name.name,
9332 new_dentry->d_name.len,
9334 btrfs_ino(BTRFS_I(new_dir)), index);
9339 inode_inc_iversion(old_dir);
9340 inode_inc_iversion(new_dir);
9341 inode_inc_iversion(old_inode);
9342 old_dir->i_ctime = old_dir->i_mtime =
9343 new_dir->i_ctime = new_dir->i_mtime =
9344 old_inode->i_ctime = current_time(old_dir);
9346 if (old_dentry->d_parent != new_dentry->d_parent)
9347 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9348 BTRFS_I(old_inode), 1);
9350 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9351 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9353 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9354 BTRFS_I(d_inode(old_dentry)),
9355 old_dentry->d_name.name,
9356 old_dentry->d_name.len,
9359 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9362 btrfs_abort_transaction(trans, ret);
9367 inode_inc_iversion(new_inode);
9368 new_inode->i_ctime = current_time(new_inode);
9369 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9370 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9371 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9372 BUG_ON(new_inode->i_nlink == 0);
9374 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9375 BTRFS_I(d_inode(new_dentry)),
9376 new_dentry->d_name.name,
9377 new_dentry->d_name.len);
9379 if (!ret && new_inode->i_nlink == 0)
9380 ret = btrfs_orphan_add(trans,
9381 BTRFS_I(d_inode(new_dentry)));
9383 btrfs_abort_transaction(trans, ret);
9388 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9389 new_dentry->d_name.name,
9390 new_dentry->d_name.len, 0, index);
9392 btrfs_abort_transaction(trans, ret);
9396 if (old_inode->i_nlink == 1)
9397 BTRFS_I(old_inode)->dir_index = index;
9399 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9400 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9401 rename_ctx.index, new_dentry->d_parent);
9403 if (flags & RENAME_WHITEOUT) {
9404 ret = btrfs_whiteout_for_rename(trans, root, whiteout_inode,
9405 old_dir, old_dentry);
9406 whiteout_inode = NULL;
9408 btrfs_abort_transaction(trans, ret);
9413 ret2 = btrfs_end_transaction(trans);
9414 ret = ret ? ret : ret2;
9416 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9417 up_read(&fs_info->subvol_sem);
9418 if (flags & RENAME_WHITEOUT)
9419 iput(whiteout_inode);
9423 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9424 struct dentry *old_dentry, struct inode *new_dir,
9425 struct dentry *new_dentry, unsigned int flags)
9427 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9430 if (flags & RENAME_EXCHANGE)
9431 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9434 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9438 struct btrfs_delalloc_work {
9439 struct inode *inode;
9440 struct completion completion;
9441 struct list_head list;
9442 struct btrfs_work work;
9445 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9447 struct btrfs_delalloc_work *delalloc_work;
9448 struct inode *inode;
9450 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9452 inode = delalloc_work->inode;
9453 filemap_flush(inode->i_mapping);
9454 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9455 &BTRFS_I(inode)->runtime_flags))
9456 filemap_flush(inode->i_mapping);
9459 complete(&delalloc_work->completion);
9462 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9464 struct btrfs_delalloc_work *work;
9466 work = kmalloc(sizeof(*work), GFP_NOFS);
9470 init_completion(&work->completion);
9471 INIT_LIST_HEAD(&work->list);
9472 work->inode = inode;
9473 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9479 * some fairly slow code that needs optimization. This walks the list
9480 * of all the inodes with pending delalloc and forces them to disk.
9482 static int start_delalloc_inodes(struct btrfs_root *root,
9483 struct writeback_control *wbc, bool snapshot,
9484 bool in_reclaim_context)
9486 struct btrfs_inode *binode;
9487 struct inode *inode;
9488 struct btrfs_delalloc_work *work, *next;
9489 struct list_head works;
9490 struct list_head splice;
9492 bool full_flush = wbc->nr_to_write == LONG_MAX;
9494 INIT_LIST_HEAD(&works);
9495 INIT_LIST_HEAD(&splice);
9497 mutex_lock(&root->delalloc_mutex);
9498 spin_lock(&root->delalloc_lock);
9499 list_splice_init(&root->delalloc_inodes, &splice);
9500 while (!list_empty(&splice)) {
9501 binode = list_entry(splice.next, struct btrfs_inode,
9504 list_move_tail(&binode->delalloc_inodes,
9505 &root->delalloc_inodes);
9507 if (in_reclaim_context &&
9508 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9511 inode = igrab(&binode->vfs_inode);
9513 cond_resched_lock(&root->delalloc_lock);
9516 spin_unlock(&root->delalloc_lock);
9519 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9520 &binode->runtime_flags);
9522 work = btrfs_alloc_delalloc_work(inode);
9528 list_add_tail(&work->list, &works);
9529 btrfs_queue_work(root->fs_info->flush_workers,
9532 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9533 btrfs_add_delayed_iput(inode);
9534 if (ret || wbc->nr_to_write <= 0)
9538 spin_lock(&root->delalloc_lock);
9540 spin_unlock(&root->delalloc_lock);
9543 list_for_each_entry_safe(work, next, &works, list) {
9544 list_del_init(&work->list);
9545 wait_for_completion(&work->completion);
9549 if (!list_empty(&splice)) {
9550 spin_lock(&root->delalloc_lock);
9551 list_splice_tail(&splice, &root->delalloc_inodes);
9552 spin_unlock(&root->delalloc_lock);
9554 mutex_unlock(&root->delalloc_mutex);
9558 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9560 struct writeback_control wbc = {
9561 .nr_to_write = LONG_MAX,
9562 .sync_mode = WB_SYNC_NONE,
9564 .range_end = LLONG_MAX,
9566 struct btrfs_fs_info *fs_info = root->fs_info;
9568 if (BTRFS_FS_ERROR(fs_info))
9571 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9574 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9575 bool in_reclaim_context)
9577 struct writeback_control wbc = {
9579 .sync_mode = WB_SYNC_NONE,
9581 .range_end = LLONG_MAX,
9583 struct btrfs_root *root;
9584 struct list_head splice;
9587 if (BTRFS_FS_ERROR(fs_info))
9590 INIT_LIST_HEAD(&splice);
9592 mutex_lock(&fs_info->delalloc_root_mutex);
9593 spin_lock(&fs_info->delalloc_root_lock);
9594 list_splice_init(&fs_info->delalloc_roots, &splice);
9595 while (!list_empty(&splice)) {
9597 * Reset nr_to_write here so we know that we're doing a full
9601 wbc.nr_to_write = LONG_MAX;
9603 root = list_first_entry(&splice, struct btrfs_root,
9605 root = btrfs_grab_root(root);
9607 list_move_tail(&root->delalloc_root,
9608 &fs_info->delalloc_roots);
9609 spin_unlock(&fs_info->delalloc_root_lock);
9611 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9612 btrfs_put_root(root);
9613 if (ret < 0 || wbc.nr_to_write <= 0)
9615 spin_lock(&fs_info->delalloc_root_lock);
9617 spin_unlock(&fs_info->delalloc_root_lock);
9621 if (!list_empty(&splice)) {
9622 spin_lock(&fs_info->delalloc_root_lock);
9623 list_splice_tail(&splice, &fs_info->delalloc_roots);
9624 spin_unlock(&fs_info->delalloc_root_lock);
9626 mutex_unlock(&fs_info->delalloc_root_mutex);
9630 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9631 struct dentry *dentry, const char *symname)
9633 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9634 struct btrfs_trans_handle *trans;
9635 struct btrfs_root *root = BTRFS_I(dir)->root;
9636 struct btrfs_path *path;
9637 struct btrfs_key key;
9638 struct inode *inode;
9644 struct btrfs_file_extent_item *ei;
9645 struct extent_buffer *leaf;
9647 name_len = strlen(symname);
9648 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9649 return -ENAMETOOLONG;
9651 inode = new_inode(dir->i_sb);
9654 inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO);
9655 inode->i_op = &btrfs_symlink_inode_operations;
9656 inode_nohighmem(inode);
9657 inode->i_mapping->a_ops = &btrfs_aops;
9660 * 2 items for inode item and ref
9661 * 2 items for dir items
9662 * 1 item for updating parent inode item
9663 * 1 item for the inline extent item
9664 * 1 item for xattr if selinux is on
9666 trans = btrfs_start_transaction(root, 7);
9667 if (IS_ERR(trans)) {
9669 return PTR_ERR(trans);
9672 err = btrfs_new_inode(trans, root, inode, dir, dentry->d_name.name,
9673 dentry->d_name.len, &index);
9680 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9684 path = btrfs_alloc_path();
9689 key.objectid = btrfs_ino(BTRFS_I(inode));
9691 key.type = BTRFS_EXTENT_DATA_KEY;
9692 datasize = btrfs_file_extent_calc_inline_size(name_len);
9693 err = btrfs_insert_empty_item(trans, root, path, &key,
9696 btrfs_free_path(path);
9699 leaf = path->nodes[0];
9700 ei = btrfs_item_ptr(leaf, path->slots[0],
9701 struct btrfs_file_extent_item);
9702 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9703 btrfs_set_file_extent_type(leaf, ei,
9704 BTRFS_FILE_EXTENT_INLINE);
9705 btrfs_set_file_extent_encryption(leaf, ei, 0);
9706 btrfs_set_file_extent_compression(leaf, ei, 0);
9707 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9708 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9710 ptr = btrfs_file_extent_inline_start(ei);
9711 write_extent_buffer(leaf, symname, ptr, name_len);
9712 btrfs_mark_buffer_dirty(leaf);
9713 btrfs_free_path(path);
9715 inode_set_bytes(inode, name_len);
9716 btrfs_i_size_write(BTRFS_I(inode), name_len);
9717 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9719 * Last step, add directory indexes for our symlink inode. This is the
9720 * last step to avoid extra cleanup of these indexes if an error happens
9724 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
9725 dentry->d_name.name, dentry->d_name.len, 0,
9730 d_instantiate_new(dentry, inode);
9733 btrfs_end_transaction(trans);
9735 inode_dec_link_count(inode);
9736 discard_new_inode(inode);
9738 btrfs_btree_balance_dirty(fs_info);
9742 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9743 struct btrfs_trans_handle *trans_in,
9744 struct btrfs_inode *inode,
9745 struct btrfs_key *ins,
9748 struct btrfs_file_extent_item stack_fi;
9749 struct btrfs_replace_extent_info extent_info;
9750 struct btrfs_trans_handle *trans = trans_in;
9751 struct btrfs_path *path;
9752 u64 start = ins->objectid;
9753 u64 len = ins->offset;
9754 int qgroup_released;
9757 memset(&stack_fi, 0, sizeof(stack_fi));
9759 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9760 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9761 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9762 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9763 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9764 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9765 /* Encryption and other encoding is reserved and all 0 */
9767 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9768 if (qgroup_released < 0)
9769 return ERR_PTR(qgroup_released);
9772 ret = insert_reserved_file_extent(trans, inode,
9773 file_offset, &stack_fi,
9774 true, qgroup_released);
9780 extent_info.disk_offset = start;
9781 extent_info.disk_len = len;
9782 extent_info.data_offset = 0;
9783 extent_info.data_len = len;
9784 extent_info.file_offset = file_offset;
9785 extent_info.extent_buf = (char *)&stack_fi;
9786 extent_info.is_new_extent = true;
9787 extent_info.qgroup_reserved = qgroup_released;
9788 extent_info.insertions = 0;
9790 path = btrfs_alloc_path();
9796 ret = btrfs_replace_file_extents(inode, path, file_offset,
9797 file_offset + len - 1, &extent_info,
9799 btrfs_free_path(path);
9806 * We have released qgroup data range at the beginning of the function,
9807 * and normally qgroup_released bytes will be freed when committing
9809 * But if we error out early, we have to free what we have released
9810 * or we leak qgroup data reservation.
9812 btrfs_qgroup_free_refroot(inode->root->fs_info,
9813 inode->root->root_key.objectid, qgroup_released,
9814 BTRFS_QGROUP_RSV_DATA);
9815 return ERR_PTR(ret);
9818 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9819 u64 start, u64 num_bytes, u64 min_size,
9820 loff_t actual_len, u64 *alloc_hint,
9821 struct btrfs_trans_handle *trans)
9823 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9824 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9825 struct extent_map *em;
9826 struct btrfs_root *root = BTRFS_I(inode)->root;
9827 struct btrfs_key ins;
9828 u64 cur_offset = start;
9829 u64 clear_offset = start;
9832 u64 last_alloc = (u64)-1;
9834 bool own_trans = true;
9835 u64 end = start + num_bytes - 1;
9839 while (num_bytes > 0) {
9840 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9841 cur_bytes = max(cur_bytes, min_size);
9843 * If we are severely fragmented we could end up with really
9844 * small allocations, so if the allocator is returning small
9845 * chunks lets make its job easier by only searching for those
9848 cur_bytes = min(cur_bytes, last_alloc);
9849 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9850 min_size, 0, *alloc_hint, &ins, 1, 0);
9855 * We've reserved this space, and thus converted it from
9856 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9857 * from here on out we will only need to clear our reservation
9858 * for the remaining unreserved area, so advance our
9859 * clear_offset by our extent size.
9861 clear_offset += ins.offset;
9863 last_alloc = ins.offset;
9864 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9867 * Now that we inserted the prealloc extent we can finally
9868 * decrement the number of reservations in the block group.
9869 * If we did it before, we could race with relocation and have
9870 * relocation miss the reserved extent, making it fail later.
9872 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9873 if (IS_ERR(trans)) {
9874 ret = PTR_ERR(trans);
9875 btrfs_free_reserved_extent(fs_info, ins.objectid,
9880 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9881 cur_offset + ins.offset -1, 0);
9883 em = alloc_extent_map();
9885 btrfs_set_inode_full_sync(BTRFS_I(inode));
9889 em->start = cur_offset;
9890 em->orig_start = cur_offset;
9891 em->len = ins.offset;
9892 em->block_start = ins.objectid;
9893 em->block_len = ins.offset;
9894 em->orig_block_len = ins.offset;
9895 em->ram_bytes = ins.offset;
9896 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9897 em->generation = trans->transid;
9900 write_lock(&em_tree->lock);
9901 ret = add_extent_mapping(em_tree, em, 1);
9902 write_unlock(&em_tree->lock);
9905 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9906 cur_offset + ins.offset - 1,
9909 free_extent_map(em);
9911 num_bytes -= ins.offset;
9912 cur_offset += ins.offset;
9913 *alloc_hint = ins.objectid + ins.offset;
9915 inode_inc_iversion(inode);
9916 inode->i_ctime = current_time(inode);
9917 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9918 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9919 (actual_len > inode->i_size) &&
9920 (cur_offset > inode->i_size)) {
9921 if (cur_offset > actual_len)
9922 i_size = actual_len;
9924 i_size = cur_offset;
9925 i_size_write(inode, i_size);
9926 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9929 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9932 btrfs_abort_transaction(trans, ret);
9934 btrfs_end_transaction(trans);
9939 btrfs_end_transaction(trans);
9943 if (clear_offset < end)
9944 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9945 end - clear_offset + 1);
9949 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9950 u64 start, u64 num_bytes, u64 min_size,
9951 loff_t actual_len, u64 *alloc_hint)
9953 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9954 min_size, actual_len, alloc_hint,
9958 int btrfs_prealloc_file_range_trans(struct inode *inode,
9959 struct btrfs_trans_handle *trans, int mode,
9960 u64 start, u64 num_bytes, u64 min_size,
9961 loff_t actual_len, u64 *alloc_hint)
9963 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9964 min_size, actual_len, alloc_hint, trans);
9967 static int btrfs_permission(struct user_namespace *mnt_userns,
9968 struct inode *inode, int mask)
9970 struct btrfs_root *root = BTRFS_I(inode)->root;
9971 umode_t mode = inode->i_mode;
9973 if (mask & MAY_WRITE &&
9974 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9975 if (btrfs_root_readonly(root))
9977 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9980 return generic_permission(mnt_userns, inode, mask);
9983 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
9984 struct dentry *dentry, umode_t mode)
9986 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9987 struct btrfs_trans_handle *trans;
9988 struct btrfs_root *root = BTRFS_I(dir)->root;
9989 struct inode *inode;
9993 inode = new_inode(dir->i_sb);
9996 inode_init_owner(mnt_userns, inode, dir, mode);
9997 inode->i_fop = &btrfs_file_operations;
9998 inode->i_op = &btrfs_file_inode_operations;
9999 inode->i_mapping->a_ops = &btrfs_aops;
10002 * 5 units required for adding orphan entry
10004 trans = btrfs_start_transaction(root, 5);
10005 if (IS_ERR(trans)) {
10007 return PTR_ERR(trans);
10010 ret = btrfs_new_inode(trans, root, inode, dir, NULL, 0, &index);
10017 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10021 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10024 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10029 * We set number of links to 0 in btrfs_new_inode(), and here we set
10030 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10033 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10035 set_nlink(inode, 1);
10036 d_tmpfile(dentry, inode);
10037 unlock_new_inode(inode);
10038 mark_inode_dirty(inode);
10040 btrfs_end_transaction(trans);
10042 discard_new_inode(inode);
10043 btrfs_btree_balance_dirty(fs_info);
10047 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10049 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10050 unsigned long index = start >> PAGE_SHIFT;
10051 unsigned long end_index = end >> PAGE_SHIFT;
10055 ASSERT(end + 1 - start <= U32_MAX);
10056 len = end + 1 - start;
10057 while (index <= end_index) {
10058 page = find_get_page(inode->vfs_inode.i_mapping, index);
10059 ASSERT(page); /* Pages should be in the extent_io_tree */
10061 btrfs_page_set_writeback(fs_info, page, start, len);
10067 static int btrfs_encoded_io_compression_from_extent(
10068 struct btrfs_fs_info *fs_info,
10071 switch (compress_type) {
10072 case BTRFS_COMPRESS_NONE:
10073 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10074 case BTRFS_COMPRESS_ZLIB:
10075 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10076 case BTRFS_COMPRESS_LZO:
10078 * The LZO format depends on the sector size. 64K is the maximum
10079 * sector size that we support.
10081 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10083 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10084 (fs_info->sectorsize_bits - 12);
10085 case BTRFS_COMPRESS_ZSTD:
10086 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10092 static ssize_t btrfs_encoded_read_inline(
10093 struct kiocb *iocb,
10094 struct iov_iter *iter, u64 start,
10096 struct extent_state **cached_state,
10097 u64 extent_start, size_t count,
10098 struct btrfs_ioctl_encoded_io_args *encoded,
10101 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10102 struct btrfs_root *root = inode->root;
10103 struct btrfs_fs_info *fs_info = root->fs_info;
10104 struct extent_io_tree *io_tree = &inode->io_tree;
10105 struct btrfs_path *path;
10106 struct extent_buffer *leaf;
10107 struct btrfs_file_extent_item *item;
10113 path = btrfs_alloc_path();
10118 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10122 /* The extent item disappeared? */
10127 leaf = path->nodes[0];
10128 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10130 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10131 ptr = btrfs_file_extent_inline_start(item);
10133 encoded->len = min_t(u64, extent_start + ram_bytes,
10134 inode->vfs_inode.i_size) - iocb->ki_pos;
10135 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10136 btrfs_file_extent_compression(leaf, item));
10139 encoded->compression = ret;
10140 if (encoded->compression) {
10141 size_t inline_size;
10143 inline_size = btrfs_file_extent_inline_item_len(leaf,
10145 if (inline_size > count) {
10149 count = inline_size;
10150 encoded->unencoded_len = ram_bytes;
10151 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10153 count = min_t(u64, count, encoded->len);
10154 encoded->len = count;
10155 encoded->unencoded_len = count;
10156 ptr += iocb->ki_pos - extent_start;
10159 tmp = kmalloc(count, GFP_NOFS);
10164 read_extent_buffer(leaf, tmp, ptr, count);
10165 btrfs_release_path(path);
10166 unlock_extent_cached(io_tree, start, lockend, cached_state);
10167 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10170 ret = copy_to_iter(tmp, count, iter);
10175 btrfs_free_path(path);
10179 struct btrfs_encoded_read_private {
10180 struct btrfs_inode *inode;
10182 wait_queue_head_t wait;
10184 blk_status_t status;
10188 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10189 struct bio *bio, int mirror_num)
10191 struct btrfs_encoded_read_private *priv = bio->bi_private;
10192 struct btrfs_bio *bbio = btrfs_bio(bio);
10193 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10196 if (!priv->skip_csum) {
10197 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10202 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
10204 btrfs_bio_free_csum(bbio);
10208 atomic_inc(&priv->pending);
10209 ret = btrfs_map_bio(fs_info, bio, mirror_num);
10211 atomic_dec(&priv->pending);
10212 btrfs_bio_free_csum(bbio);
10217 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10219 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10220 struct btrfs_encoded_read_private *priv = bbio->bio.bi_private;
10221 struct btrfs_inode *inode = priv->inode;
10222 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10223 u32 sectorsize = fs_info->sectorsize;
10224 struct bio_vec *bvec;
10225 struct bvec_iter_all iter_all;
10226 u64 start = priv->file_offset;
10227 u32 bio_offset = 0;
10229 if (priv->skip_csum || !uptodate)
10230 return bbio->bio.bi_status;
10232 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10233 unsigned int i, nr_sectors, pgoff;
10235 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10236 pgoff = bvec->bv_offset;
10237 for (i = 0; i < nr_sectors; i++) {
10238 ASSERT(pgoff < PAGE_SIZE);
10239 if (check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10240 bvec->bv_page, pgoff, start))
10241 return BLK_STS_IOERR;
10242 start += sectorsize;
10243 bio_offset += sectorsize;
10244 pgoff += sectorsize;
10250 static void btrfs_encoded_read_endio(struct bio *bio)
10252 struct btrfs_encoded_read_private *priv = bio->bi_private;
10253 struct btrfs_bio *bbio = btrfs_bio(bio);
10254 blk_status_t status;
10256 status = btrfs_encoded_read_verify_csum(bbio);
10259 * The memory barrier implied by the atomic_dec_return() here
10260 * pairs with the memory barrier implied by the
10261 * atomic_dec_return() or io_wait_event() in
10262 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10263 * write is observed before the load of status in
10264 * btrfs_encoded_read_regular_fill_pages().
10266 WRITE_ONCE(priv->status, status);
10268 if (!atomic_dec_return(&priv->pending))
10269 wake_up(&priv->wait);
10270 btrfs_bio_free_csum(bbio);
10274 static int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10278 struct page **pages)
10280 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10281 struct btrfs_encoded_read_private priv = {
10283 .file_offset = file_offset,
10284 .pending = ATOMIC_INIT(1),
10285 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10287 unsigned long i = 0;
10291 init_waitqueue_head(&priv.wait);
10293 * Submit bios for the extent, splitting due to bio or stripe limits as
10296 while (cur < disk_io_size) {
10297 struct extent_map *em;
10298 struct btrfs_io_geometry geom;
10299 struct bio *bio = NULL;
10302 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10303 disk_io_size - cur);
10307 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10308 disk_bytenr + cur, &geom);
10309 free_extent_map(em);
10312 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10315 remaining = min(geom.len, disk_io_size - cur);
10316 while (bio || remaining) {
10317 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10320 bio = btrfs_bio_alloc(BIO_MAX_VECS);
10321 bio->bi_iter.bi_sector =
10322 (disk_bytenr + cur) >> SECTOR_SHIFT;
10323 bio->bi_end_io = btrfs_encoded_read_endio;
10324 bio->bi_private = &priv;
10325 bio->bi_opf = REQ_OP_READ;
10329 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10330 blk_status_t status;
10332 status = submit_encoded_read_bio(inode, bio, 0);
10334 WRITE_ONCE(priv.status, status);
10344 remaining -= bytes;
10349 if (atomic_dec_return(&priv.pending))
10350 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10351 /* See btrfs_encoded_read_endio() for ordering. */
10352 return blk_status_to_errno(READ_ONCE(priv.status));
10355 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10356 struct iov_iter *iter,
10357 u64 start, u64 lockend,
10358 struct extent_state **cached_state,
10359 u64 disk_bytenr, u64 disk_io_size,
10360 size_t count, bool compressed,
10363 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10364 struct extent_io_tree *io_tree = &inode->io_tree;
10365 struct page **pages;
10366 unsigned long nr_pages, i;
10368 size_t page_offset;
10371 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10372 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10375 for (i = 0; i < nr_pages; i++) {
10376 pages[i] = alloc_page(GFP_NOFS);
10383 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10384 disk_io_size, pages);
10388 unlock_extent_cached(io_tree, start, lockend, cached_state);
10389 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10396 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10397 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10400 while (cur < count) {
10401 size_t bytes = min_t(size_t, count - cur,
10402 PAGE_SIZE - page_offset);
10404 if (copy_page_to_iter(pages[i], page_offset, bytes,
10415 for (i = 0; i < nr_pages; i++) {
10417 __free_page(pages[i]);
10423 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10424 struct btrfs_ioctl_encoded_io_args *encoded)
10426 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10427 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10428 struct extent_io_tree *io_tree = &inode->io_tree;
10430 size_t count = iov_iter_count(iter);
10431 u64 start, lockend, disk_bytenr, disk_io_size;
10432 struct extent_state *cached_state = NULL;
10433 struct extent_map *em;
10434 bool unlocked = false;
10436 file_accessed(iocb->ki_filp);
10438 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10440 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10441 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10444 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10446 * We don't know how long the extent containing iocb->ki_pos is, but if
10447 * it's compressed we know that it won't be longer than this.
10449 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10452 struct btrfs_ordered_extent *ordered;
10454 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10455 lockend - start + 1);
10457 goto out_unlock_inode;
10458 lock_extent_bits(io_tree, start, lockend, &cached_state);
10459 ordered = btrfs_lookup_ordered_range(inode, start,
10460 lockend - start + 1);
10463 btrfs_put_ordered_extent(ordered);
10464 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10468 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10471 goto out_unlock_extent;
10474 if (em->block_start == EXTENT_MAP_INLINE) {
10475 u64 extent_start = em->start;
10478 * For inline extents we get everything we need out of the
10481 free_extent_map(em);
10483 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10484 &cached_state, extent_start,
10485 count, encoded, &unlocked);
10490 * We only want to return up to EOF even if the extent extends beyond
10493 encoded->len = min_t(u64, extent_map_end(em),
10494 inode->vfs_inode.i_size) - iocb->ki_pos;
10495 if (em->block_start == EXTENT_MAP_HOLE ||
10496 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10497 disk_bytenr = EXTENT_MAP_HOLE;
10498 count = min_t(u64, count, encoded->len);
10499 encoded->len = count;
10500 encoded->unencoded_len = count;
10501 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10502 disk_bytenr = em->block_start;
10504 * Bail if the buffer isn't large enough to return the whole
10505 * compressed extent.
10507 if (em->block_len > count) {
10511 disk_io_size = count = em->block_len;
10512 encoded->unencoded_len = em->ram_bytes;
10513 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10514 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10515 em->compress_type);
10518 encoded->compression = ret;
10520 disk_bytenr = em->block_start + (start - em->start);
10521 if (encoded->len > count)
10522 encoded->len = count;
10524 * Don't read beyond what we locked. This also limits the page
10525 * allocations that we'll do.
10527 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10528 count = start + disk_io_size - iocb->ki_pos;
10529 encoded->len = count;
10530 encoded->unencoded_len = count;
10531 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10533 free_extent_map(em);
10536 if (disk_bytenr == EXTENT_MAP_HOLE) {
10537 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10538 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10540 ret = iov_iter_zero(count, iter);
10544 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10545 &cached_state, disk_bytenr,
10546 disk_io_size, count,
10547 encoded->compression,
10553 iocb->ki_pos += encoded->len;
10555 free_extent_map(em);
10558 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10561 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10565 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10566 const struct btrfs_ioctl_encoded_io_args *encoded)
10568 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10569 struct btrfs_root *root = inode->root;
10570 struct btrfs_fs_info *fs_info = root->fs_info;
10571 struct extent_io_tree *io_tree = &inode->io_tree;
10572 struct extent_changeset *data_reserved = NULL;
10573 struct extent_state *cached_state = NULL;
10577 u64 num_bytes, ram_bytes, disk_num_bytes;
10578 unsigned long nr_pages, i;
10579 struct page **pages;
10580 struct btrfs_key ins;
10581 bool extent_reserved = false;
10582 struct extent_map *em;
10585 switch (encoded->compression) {
10586 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10587 compression = BTRFS_COMPRESS_ZLIB;
10589 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10590 compression = BTRFS_COMPRESS_ZSTD;
10592 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10593 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10594 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10595 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10596 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10597 /* The sector size must match for LZO. */
10598 if (encoded->compression -
10599 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10600 fs_info->sectorsize_bits)
10602 compression = BTRFS_COMPRESS_LZO;
10607 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10610 orig_count = iov_iter_count(from);
10612 /* The extent size must be sane. */
10613 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10614 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10618 * The compressed data must be smaller than the decompressed data.
10620 * It's of course possible for data to compress to larger or the same
10621 * size, but the buffered I/O path falls back to no compression for such
10622 * data, and we don't want to break any assumptions by creating these
10625 * Note that this is less strict than the current check we have that the
10626 * compressed data must be at least one sector smaller than the
10627 * decompressed data. We only want to enforce the weaker requirement
10628 * from old kernels that it is at least one byte smaller.
10630 if (orig_count >= encoded->unencoded_len)
10633 /* The extent must start on a sector boundary. */
10634 start = iocb->ki_pos;
10635 if (!IS_ALIGNED(start, fs_info->sectorsize))
10639 * The extent must end on a sector boundary. However, we allow a write
10640 * which ends at or extends i_size to have an unaligned length; we round
10641 * up the extent size and set i_size to the unaligned end.
10643 if (start + encoded->len < inode->vfs_inode.i_size &&
10644 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10647 /* Finally, the offset in the unencoded data must be sector-aligned. */
10648 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10651 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10652 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10653 end = start + num_bytes - 1;
10656 * If the extent cannot be inline, the compressed data on disk must be
10657 * sector-aligned. For convenience, we extend it with zeroes if it
10660 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10661 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10662 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10665 for (i = 0; i < nr_pages; i++) {
10666 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10669 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10674 kaddr = kmap(pages[i]);
10675 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10680 if (bytes < PAGE_SIZE)
10681 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10686 struct btrfs_ordered_extent *ordered;
10688 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10691 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10692 start >> PAGE_SHIFT,
10693 end >> PAGE_SHIFT);
10696 lock_extent_bits(io_tree, start, end, &cached_state);
10697 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10699 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10702 btrfs_put_ordered_extent(ordered);
10703 unlock_extent_cached(io_tree, start, end, &cached_state);
10708 * We don't use the higher-level delalloc space functions because our
10709 * num_bytes and disk_num_bytes are different.
10711 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10714 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10716 goto out_free_data_space;
10717 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes);
10719 goto out_qgroup_free_data;
10721 /* Try an inline extent first. */
10722 if (start == 0 && encoded->unencoded_len == encoded->len &&
10723 encoded->unencoded_offset == 0) {
10724 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10725 compression, pages, true);
10729 goto out_delalloc_release;
10733 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10734 disk_num_bytes, 0, 0, &ins, 1, 1);
10736 goto out_delalloc_release;
10737 extent_reserved = true;
10739 em = create_io_em(inode, start, num_bytes,
10740 start - encoded->unencoded_offset, ins.objectid,
10741 ins.offset, ins.offset, ram_bytes, compression,
10742 BTRFS_ORDERED_COMPRESSED);
10745 goto out_free_reserved;
10747 free_extent_map(em);
10749 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10750 ins.objectid, ins.offset,
10751 encoded->unencoded_offset,
10752 (1 << BTRFS_ORDERED_ENCODED) |
10753 (1 << BTRFS_ORDERED_COMPRESSED),
10756 btrfs_drop_extent_cache(inode, start, end, 0);
10757 goto out_free_reserved;
10759 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10761 if (start + encoded->len > inode->vfs_inode.i_size)
10762 i_size_write(&inode->vfs_inode, start + encoded->len);
10764 unlock_extent_cached(io_tree, start, end, &cached_state);
10766 btrfs_delalloc_release_extents(inode, num_bytes);
10768 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10769 ins.offset, pages, nr_pages, 0, NULL,
10771 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10779 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10780 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10781 out_delalloc_release:
10782 btrfs_delalloc_release_extents(inode, num_bytes);
10783 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10784 out_qgroup_free_data:
10786 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10787 out_free_data_space:
10789 * If btrfs_reserve_extent() succeeded, then we already decremented
10792 if (!extent_reserved)
10793 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10795 unlock_extent_cached(io_tree, start, end, &cached_state);
10797 for (i = 0; i < nr_pages; i++) {
10799 __free_page(pages[i]);
10804 iocb->ki_pos += encoded->len;
10810 * Add an entry indicating a block group or device which is pinned by a
10811 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10812 * negative errno on failure.
10814 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10815 bool is_block_group)
10817 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10818 struct btrfs_swapfile_pin *sp, *entry;
10819 struct rb_node **p;
10820 struct rb_node *parent = NULL;
10822 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10827 sp->is_block_group = is_block_group;
10828 sp->bg_extent_count = 1;
10830 spin_lock(&fs_info->swapfile_pins_lock);
10831 p = &fs_info->swapfile_pins.rb_node;
10834 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10835 if (sp->ptr < entry->ptr ||
10836 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10837 p = &(*p)->rb_left;
10838 } else if (sp->ptr > entry->ptr ||
10839 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10840 p = &(*p)->rb_right;
10842 if (is_block_group)
10843 entry->bg_extent_count++;
10844 spin_unlock(&fs_info->swapfile_pins_lock);
10849 rb_link_node(&sp->node, parent, p);
10850 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10851 spin_unlock(&fs_info->swapfile_pins_lock);
10855 /* Free all of the entries pinned by this swapfile. */
10856 static void btrfs_free_swapfile_pins(struct inode *inode)
10858 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10859 struct btrfs_swapfile_pin *sp;
10860 struct rb_node *node, *next;
10862 spin_lock(&fs_info->swapfile_pins_lock);
10863 node = rb_first(&fs_info->swapfile_pins);
10865 next = rb_next(node);
10866 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10867 if (sp->inode == inode) {
10868 rb_erase(&sp->node, &fs_info->swapfile_pins);
10869 if (sp->is_block_group) {
10870 btrfs_dec_block_group_swap_extents(sp->ptr,
10871 sp->bg_extent_count);
10872 btrfs_put_block_group(sp->ptr);
10878 spin_unlock(&fs_info->swapfile_pins_lock);
10881 struct btrfs_swap_info {
10887 unsigned long nr_pages;
10891 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10892 struct btrfs_swap_info *bsi)
10894 unsigned long nr_pages;
10895 unsigned long max_pages;
10896 u64 first_ppage, first_ppage_reported, next_ppage;
10900 * Our swapfile may have had its size extended after the swap header was
10901 * written. In that case activating the swapfile should not go beyond
10902 * the max size set in the swap header.
10904 if (bsi->nr_pages >= sis->max)
10907 max_pages = sis->max - bsi->nr_pages;
10908 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10909 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10910 PAGE_SIZE) >> PAGE_SHIFT;
10912 if (first_ppage >= next_ppage)
10914 nr_pages = next_ppage - first_ppage;
10915 nr_pages = min(nr_pages, max_pages);
10917 first_ppage_reported = first_ppage;
10918 if (bsi->start == 0)
10919 first_ppage_reported++;
10920 if (bsi->lowest_ppage > first_ppage_reported)
10921 bsi->lowest_ppage = first_ppage_reported;
10922 if (bsi->highest_ppage < (next_ppage - 1))
10923 bsi->highest_ppage = next_ppage - 1;
10925 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10928 bsi->nr_extents += ret;
10929 bsi->nr_pages += nr_pages;
10933 static void btrfs_swap_deactivate(struct file *file)
10935 struct inode *inode = file_inode(file);
10937 btrfs_free_swapfile_pins(inode);
10938 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10941 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10944 struct inode *inode = file_inode(file);
10945 struct btrfs_root *root = BTRFS_I(inode)->root;
10946 struct btrfs_fs_info *fs_info = root->fs_info;
10947 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10948 struct extent_state *cached_state = NULL;
10949 struct extent_map *em = NULL;
10950 struct btrfs_device *device = NULL;
10951 struct btrfs_swap_info bsi = {
10952 .lowest_ppage = (sector_t)-1ULL,
10959 * If the swap file was just created, make sure delalloc is done. If the
10960 * file changes again after this, the user is doing something stupid and
10961 * we don't really care.
10963 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10968 * The inode is locked, so these flags won't change after we check them.
10970 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10971 btrfs_warn(fs_info, "swapfile must not be compressed");
10974 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10975 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10978 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10979 btrfs_warn(fs_info, "swapfile must not be checksummed");
10984 * Balance or device remove/replace/resize can move stuff around from
10985 * under us. The exclop protection makes sure they aren't running/won't
10986 * run concurrently while we are mapping the swap extents, and
10987 * fs_info->swapfile_pins prevents them from running while the swap
10988 * file is active and moving the extents. Note that this also prevents
10989 * a concurrent device add which isn't actually necessary, but it's not
10990 * really worth the trouble to allow it.
10992 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10993 btrfs_warn(fs_info,
10994 "cannot activate swapfile while exclusive operation is running");
10999 * Prevent snapshot creation while we are activating the swap file.
11000 * We do not want to race with snapshot creation. If snapshot creation
11001 * already started before we bumped nr_swapfiles from 0 to 1 and
11002 * completes before the first write into the swap file after it is
11003 * activated, than that write would fallback to COW.
11005 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11006 btrfs_exclop_finish(fs_info);
11007 btrfs_warn(fs_info,
11008 "cannot activate swapfile because snapshot creation is in progress");
11012 * Snapshots can create extents which require COW even if NODATACOW is
11013 * set. We use this counter to prevent snapshots. We must increment it
11014 * before walking the extents because we don't want a concurrent
11015 * snapshot to run after we've already checked the extents.
11017 * It is possible that subvolume is marked for deletion but still not
11018 * removed yet. To prevent this race, we check the root status before
11019 * activating the swapfile.
11021 spin_lock(&root->root_item_lock);
11022 if (btrfs_root_dead(root)) {
11023 spin_unlock(&root->root_item_lock);
11025 btrfs_exclop_finish(fs_info);
11026 btrfs_warn(fs_info,
11027 "cannot activate swapfile because subvolume %llu is being deleted",
11028 root->root_key.objectid);
11031 atomic_inc(&root->nr_swapfiles);
11032 spin_unlock(&root->root_item_lock);
11034 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11036 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
11038 while (start < isize) {
11039 u64 logical_block_start, physical_block_start;
11040 struct btrfs_block_group *bg;
11041 u64 len = isize - start;
11043 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11049 if (em->block_start == EXTENT_MAP_HOLE) {
11050 btrfs_warn(fs_info, "swapfile must not have holes");
11054 if (em->block_start == EXTENT_MAP_INLINE) {
11056 * It's unlikely we'll ever actually find ourselves
11057 * here, as a file small enough to fit inline won't be
11058 * big enough to store more than the swap header, but in
11059 * case something changes in the future, let's catch it
11060 * here rather than later.
11062 btrfs_warn(fs_info, "swapfile must not be inline");
11066 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11067 btrfs_warn(fs_info, "swapfile must not be compressed");
11072 logical_block_start = em->block_start + (start - em->start);
11073 len = min(len, em->len - (start - em->start));
11074 free_extent_map(em);
11077 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
11083 btrfs_warn(fs_info,
11084 "swapfile must not be copy-on-write");
11089 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11095 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11096 btrfs_warn(fs_info,
11097 "swapfile must have single data profile");
11102 if (device == NULL) {
11103 device = em->map_lookup->stripes[0].dev;
11104 ret = btrfs_add_swapfile_pin(inode, device, false);
11109 } else if (device != em->map_lookup->stripes[0].dev) {
11110 btrfs_warn(fs_info, "swapfile must be on one device");
11115 physical_block_start = (em->map_lookup->stripes[0].physical +
11116 (logical_block_start - em->start));
11117 len = min(len, em->len - (logical_block_start - em->start));
11118 free_extent_map(em);
11121 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11123 btrfs_warn(fs_info,
11124 "could not find block group containing swapfile");
11129 if (!btrfs_inc_block_group_swap_extents(bg)) {
11130 btrfs_warn(fs_info,
11131 "block group for swapfile at %llu is read-only%s",
11133 atomic_read(&fs_info->scrubs_running) ?
11134 " (scrub running)" : "");
11135 btrfs_put_block_group(bg);
11140 ret = btrfs_add_swapfile_pin(inode, bg, true);
11142 btrfs_put_block_group(bg);
11149 if (bsi.block_len &&
11150 bsi.block_start + bsi.block_len == physical_block_start) {
11151 bsi.block_len += len;
11153 if (bsi.block_len) {
11154 ret = btrfs_add_swap_extent(sis, &bsi);
11159 bsi.block_start = physical_block_start;
11160 bsi.block_len = len;
11167 ret = btrfs_add_swap_extent(sis, &bsi);
11170 if (!IS_ERR_OR_NULL(em))
11171 free_extent_map(em);
11173 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
11176 btrfs_swap_deactivate(file);
11178 btrfs_drew_write_unlock(&root->snapshot_lock);
11180 btrfs_exclop_finish(fs_info);
11186 sis->bdev = device->bdev;
11187 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11188 sis->max = bsi.nr_pages;
11189 sis->pages = bsi.nr_pages - 1;
11190 sis->highest_bit = bsi.nr_pages - 1;
11191 return bsi.nr_extents;
11194 static void btrfs_swap_deactivate(struct file *file)
11198 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11201 return -EOPNOTSUPP;
11206 * Update the number of bytes used in the VFS' inode. When we replace extents in
11207 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11208 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11209 * always get a correct value.
11211 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11212 const u64 add_bytes,
11213 const u64 del_bytes)
11215 if (add_bytes == del_bytes)
11218 spin_lock(&inode->lock);
11220 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11222 inode_add_bytes(&inode->vfs_inode, add_bytes);
11223 spin_unlock(&inode->lock);
11226 static const struct inode_operations btrfs_dir_inode_operations = {
11227 .getattr = btrfs_getattr,
11228 .lookup = btrfs_lookup,
11229 .create = btrfs_create,
11230 .unlink = btrfs_unlink,
11231 .link = btrfs_link,
11232 .mkdir = btrfs_mkdir,
11233 .rmdir = btrfs_rmdir,
11234 .rename = btrfs_rename2,
11235 .symlink = btrfs_symlink,
11236 .setattr = btrfs_setattr,
11237 .mknod = btrfs_mknod,
11238 .listxattr = btrfs_listxattr,
11239 .permission = btrfs_permission,
11240 .get_acl = btrfs_get_acl,
11241 .set_acl = btrfs_set_acl,
11242 .update_time = btrfs_update_time,
11243 .tmpfile = btrfs_tmpfile,
11244 .fileattr_get = btrfs_fileattr_get,
11245 .fileattr_set = btrfs_fileattr_set,
11248 static const struct file_operations btrfs_dir_file_operations = {
11249 .llseek = generic_file_llseek,
11250 .read = generic_read_dir,
11251 .iterate_shared = btrfs_real_readdir,
11252 .open = btrfs_opendir,
11253 .unlocked_ioctl = btrfs_ioctl,
11254 #ifdef CONFIG_COMPAT
11255 .compat_ioctl = btrfs_compat_ioctl,
11257 .release = btrfs_release_file,
11258 .fsync = btrfs_sync_file,
11262 * btrfs doesn't support the bmap operation because swapfiles
11263 * use bmap to make a mapping of extents in the file. They assume
11264 * these extents won't change over the life of the file and they
11265 * use the bmap result to do IO directly to the drive.
11267 * the btrfs bmap call would return logical addresses that aren't
11268 * suitable for IO and they also will change frequently as COW
11269 * operations happen. So, swapfile + btrfs == corruption.
11271 * For now we're avoiding this by dropping bmap.
11273 static const struct address_space_operations btrfs_aops = {
11274 .readpage = btrfs_readpage,
11275 .writepage = btrfs_writepage,
11276 .writepages = btrfs_writepages,
11277 .readahead = btrfs_readahead,
11278 .direct_IO = noop_direct_IO,
11279 .invalidate_folio = btrfs_invalidate_folio,
11280 .releasepage = btrfs_releasepage,
11281 #ifdef CONFIG_MIGRATION
11282 .migratepage = btrfs_migratepage,
11284 .dirty_folio = filemap_dirty_folio,
11285 .error_remove_page = generic_error_remove_page,
11286 .swap_activate = btrfs_swap_activate,
11287 .swap_deactivate = btrfs_swap_deactivate,
11290 static const struct inode_operations btrfs_file_inode_operations = {
11291 .getattr = btrfs_getattr,
11292 .setattr = btrfs_setattr,
11293 .listxattr = btrfs_listxattr,
11294 .permission = btrfs_permission,
11295 .fiemap = btrfs_fiemap,
11296 .get_acl = btrfs_get_acl,
11297 .set_acl = btrfs_set_acl,
11298 .update_time = btrfs_update_time,
11299 .fileattr_get = btrfs_fileattr_get,
11300 .fileattr_set = btrfs_fileattr_set,
11302 static const struct inode_operations btrfs_special_inode_operations = {
11303 .getattr = btrfs_getattr,
11304 .setattr = btrfs_setattr,
11305 .permission = btrfs_permission,
11306 .listxattr = btrfs_listxattr,
11307 .get_acl = btrfs_get_acl,
11308 .set_acl = btrfs_set_acl,
11309 .update_time = btrfs_update_time,
11311 static const struct inode_operations btrfs_symlink_inode_operations = {
11312 .get_link = page_get_link,
11313 .getattr = btrfs_getattr,
11314 .setattr = btrfs_setattr,
11315 .permission = btrfs_permission,
11316 .listxattr = btrfs_listxattr,
11317 .update_time = btrfs_update_time,
11320 const struct dentry_operations btrfs_dentry_operations = {
11321 .d_delete = btrfs_dentry_delete,