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 btrfs_new_inode_args *args)
229 if (args->default_acl) {
230 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
236 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
240 if (!args->default_acl && !args->acl)
241 cache_no_acl(args->inode);
242 return btrfs_xattr_security_init(trans, args->inode, args->dir,
243 &args->dentry->d_name);
247 * this does all the hard work for inserting an inline extent into
248 * the btree. The caller should have done a btrfs_drop_extents so that
249 * no overlapping inline items exist in the btree
251 static int insert_inline_extent(struct btrfs_trans_handle *trans,
252 struct btrfs_path *path,
253 struct btrfs_inode *inode, bool extent_inserted,
254 size_t size, size_t compressed_size,
256 struct page **compressed_pages,
259 struct btrfs_root *root = inode->root;
260 struct extent_buffer *leaf;
261 struct page *page = NULL;
264 struct btrfs_file_extent_item *ei;
266 size_t cur_size = size;
269 ASSERT((compressed_size > 0 && compressed_pages) ||
270 (compressed_size == 0 && !compressed_pages));
272 if (compressed_size && compressed_pages)
273 cur_size = compressed_size;
275 if (!extent_inserted) {
276 struct btrfs_key key;
279 key.objectid = btrfs_ino(inode);
281 key.type = BTRFS_EXTENT_DATA_KEY;
283 datasize = btrfs_file_extent_calc_inline_size(cur_size);
284 ret = btrfs_insert_empty_item(trans, root, path, &key,
289 leaf = path->nodes[0];
290 ei = btrfs_item_ptr(leaf, path->slots[0],
291 struct btrfs_file_extent_item);
292 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
293 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
294 btrfs_set_file_extent_encryption(leaf, ei, 0);
295 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
296 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
297 ptr = btrfs_file_extent_inline_start(ei);
299 if (compress_type != BTRFS_COMPRESS_NONE) {
302 while (compressed_size > 0) {
303 cpage = compressed_pages[i];
304 cur_size = min_t(unsigned long, compressed_size,
307 kaddr = kmap_atomic(cpage);
308 write_extent_buffer(leaf, kaddr, ptr, cur_size);
309 kunmap_atomic(kaddr);
313 compressed_size -= cur_size;
315 btrfs_set_file_extent_compression(leaf, ei,
318 page = find_get_page(inode->vfs_inode.i_mapping, 0);
319 btrfs_set_file_extent_compression(leaf, ei, 0);
320 kaddr = kmap_atomic(page);
321 write_extent_buffer(leaf, kaddr, ptr, size);
322 kunmap_atomic(kaddr);
325 btrfs_mark_buffer_dirty(leaf);
326 btrfs_release_path(path);
329 * We align size to sectorsize for inline extents just for simplicity
332 ret = btrfs_inode_set_file_extent_range(inode, 0,
333 ALIGN(size, root->fs_info->sectorsize));
338 * We're an inline extent, so nobody can extend the file past i_size
339 * without locking a page we already have locked.
341 * We must do any i_size and inode updates before we unlock the pages.
342 * Otherwise we could end up racing with unlink.
344 i_size = i_size_read(&inode->vfs_inode);
345 if (update_i_size && size > i_size) {
346 i_size_write(&inode->vfs_inode, size);
349 inode->disk_i_size = i_size;
357 * conditionally insert an inline extent into the file. This
358 * does the checks required to make sure the data is small enough
359 * to fit as an inline extent.
361 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
362 size_t compressed_size,
364 struct page **compressed_pages,
367 struct btrfs_drop_extents_args drop_args = { 0 };
368 struct btrfs_root *root = inode->root;
369 struct btrfs_fs_info *fs_info = root->fs_info;
370 struct btrfs_trans_handle *trans;
371 u64 data_len = (compressed_size ?: size);
373 struct btrfs_path *path;
376 * We can create an inline extent if it ends at or beyond the current
377 * i_size, is no larger than a sector (decompressed), and the (possibly
378 * compressed) data fits in a leaf and the configured maximum inline
381 if (size < i_size_read(&inode->vfs_inode) ||
382 size > fs_info->sectorsize ||
383 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
384 data_len > fs_info->max_inline)
387 path = btrfs_alloc_path();
391 trans = btrfs_join_transaction(root);
393 btrfs_free_path(path);
394 return PTR_ERR(trans);
396 trans->block_rsv = &inode->block_rsv;
398 drop_args.path = path;
400 drop_args.end = fs_info->sectorsize;
401 drop_args.drop_cache = true;
402 drop_args.replace_extent = true;
403 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
404 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
406 btrfs_abort_transaction(trans, ret);
410 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
411 size, compressed_size, compress_type,
412 compressed_pages, update_i_size);
413 if (ret && ret != -ENOSPC) {
414 btrfs_abort_transaction(trans, ret);
416 } else if (ret == -ENOSPC) {
421 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
422 ret = btrfs_update_inode(trans, root, inode);
423 if (ret && ret != -ENOSPC) {
424 btrfs_abort_transaction(trans, ret);
426 } else if (ret == -ENOSPC) {
431 btrfs_set_inode_full_sync(inode);
434 * Don't forget to free the reserved space, as for inlined extent
435 * it won't count as data extent, free them directly here.
436 * And at reserve time, it's always aligned to page size, so
437 * just free one page here.
439 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
440 btrfs_free_path(path);
441 btrfs_end_transaction(trans);
445 struct async_extent {
450 unsigned long nr_pages;
452 struct list_head list;
457 struct page *locked_page;
460 unsigned int write_flags;
461 struct list_head extents;
462 struct cgroup_subsys_state *blkcg_css;
463 struct btrfs_work work;
464 struct async_cow *async_cow;
469 struct async_chunk chunks[];
472 static noinline int add_async_extent(struct async_chunk *cow,
473 u64 start, u64 ram_size,
476 unsigned long nr_pages,
479 struct async_extent *async_extent;
481 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
482 BUG_ON(!async_extent); /* -ENOMEM */
483 async_extent->start = start;
484 async_extent->ram_size = ram_size;
485 async_extent->compressed_size = compressed_size;
486 async_extent->pages = pages;
487 async_extent->nr_pages = nr_pages;
488 async_extent->compress_type = compress_type;
489 list_add_tail(&async_extent->list, &cow->extents);
494 * Check if the inode needs to be submitted to compression, based on mount
495 * options, defragmentation, properties or heuristics.
497 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
500 struct btrfs_fs_info *fs_info = inode->root->fs_info;
502 if (!btrfs_inode_can_compress(inode)) {
503 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
504 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
509 * Special check for subpage.
511 * We lock the full page then run each delalloc range in the page, thus
512 * for the following case, we will hit some subpage specific corner case:
515 * | |///////| |///////|
518 * In above case, both range A and range B will try to unlock the full
519 * page [0, 64K), causing the one finished later will have page
520 * unlocked already, triggering various page lock requirement BUG_ON()s.
522 * So here we add an artificial limit that subpage compression can only
523 * if the range is fully page aligned.
525 * In theory we only need to ensure the first page is fully covered, but
526 * the tailing partial page will be locked until the full compression
527 * finishes, delaying the write of other range.
529 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
530 * first to prevent any submitted async extent to unlock the full page.
531 * By this, we can ensure for subpage case that only the last async_cow
532 * will unlock the full page.
534 if (fs_info->sectorsize < PAGE_SIZE) {
535 if (!IS_ALIGNED(start, PAGE_SIZE) ||
536 !IS_ALIGNED(end + 1, PAGE_SIZE))
541 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
544 if (inode->defrag_compress)
546 /* bad compression ratios */
547 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
549 if (btrfs_test_opt(fs_info, COMPRESS) ||
550 inode->flags & BTRFS_INODE_COMPRESS ||
551 inode->prop_compress)
552 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
556 static inline void inode_should_defrag(struct btrfs_inode *inode,
557 u64 start, u64 end, u64 num_bytes, u32 small_write)
559 /* If this is a small write inside eof, kick off a defrag */
560 if (num_bytes < small_write &&
561 (start > 0 || end + 1 < inode->disk_i_size))
562 btrfs_add_inode_defrag(NULL, inode, small_write);
566 * we create compressed extents in two phases. The first
567 * phase compresses a range of pages that have already been
568 * locked (both pages and state bits are locked).
570 * This is done inside an ordered work queue, and the compression
571 * is spread across many cpus. The actual IO submission is step
572 * two, and the ordered work queue takes care of making sure that
573 * happens in the same order things were put onto the queue by
574 * writepages and friends.
576 * If this code finds it can't get good compression, it puts an
577 * entry onto the work queue to write the uncompressed bytes. This
578 * makes sure that both compressed inodes and uncompressed inodes
579 * are written in the same order that the flusher thread sent them
582 static noinline int compress_file_range(struct async_chunk *async_chunk)
584 struct inode *inode = async_chunk->inode;
585 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
586 u64 blocksize = fs_info->sectorsize;
587 u64 start = async_chunk->start;
588 u64 end = async_chunk->end;
592 struct page **pages = NULL;
593 unsigned long nr_pages;
594 unsigned long total_compressed = 0;
595 unsigned long total_in = 0;
598 int compress_type = fs_info->compress_type;
599 int compressed_extents = 0;
602 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
606 * We need to save i_size before now because it could change in between
607 * us evaluating the size and assigning it. This is because we lock and
608 * unlock the page in truncate and fallocate, and then modify the i_size
611 * The barriers are to emulate READ_ONCE, remove that once i_size_read
615 i_size = i_size_read(inode);
617 actual_end = min_t(u64, i_size, end + 1);
620 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
621 nr_pages = min_t(unsigned long, nr_pages,
622 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
625 * we don't want to send crud past the end of i_size through
626 * compression, that's just a waste of CPU time. So, if the
627 * end of the file is before the start of our current
628 * requested range of bytes, we bail out to the uncompressed
629 * cleanup code that can deal with all of this.
631 * It isn't really the fastest way to fix things, but this is a
632 * very uncommon corner.
634 if (actual_end <= start)
635 goto cleanup_and_bail_uncompressed;
637 total_compressed = actual_end - start;
640 * Skip compression for a small file range(<=blocksize) that
641 * isn't an inline extent, since it doesn't save disk space at all.
643 if (total_compressed <= blocksize &&
644 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
645 goto cleanup_and_bail_uncompressed;
648 * For subpage case, we require full page alignment for the sector
650 * Thus we must also check against @actual_end, not just @end.
652 if (blocksize < PAGE_SIZE) {
653 if (!IS_ALIGNED(start, PAGE_SIZE) ||
654 !IS_ALIGNED(round_up(actual_end, blocksize), PAGE_SIZE))
655 goto cleanup_and_bail_uncompressed;
658 total_compressed = min_t(unsigned long, total_compressed,
659 BTRFS_MAX_UNCOMPRESSED);
664 * we do compression for mount -o compress and when the
665 * inode has not been flagged as nocompress. This flag can
666 * change at any time if we discover bad compression ratios.
668 if (inode_need_compress(BTRFS_I(inode), start, end)) {
670 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
672 /* just bail out to the uncompressed code */
677 if (BTRFS_I(inode)->defrag_compress)
678 compress_type = BTRFS_I(inode)->defrag_compress;
679 else if (BTRFS_I(inode)->prop_compress)
680 compress_type = BTRFS_I(inode)->prop_compress;
683 * we need to call clear_page_dirty_for_io on each
684 * page in the range. Otherwise applications with the file
685 * mmap'd can wander in and change the page contents while
686 * we are compressing them.
688 * If the compression fails for any reason, we set the pages
689 * dirty again later on.
691 * Note that the remaining part is redirtied, the start pointer
692 * has moved, the end is the original one.
695 extent_range_clear_dirty_for_io(inode, start, end);
699 /* Compression level is applied here and only here */
700 ret = btrfs_compress_pages(
701 compress_type | (fs_info->compress_level << 4),
702 inode->i_mapping, start,
709 unsigned long offset = offset_in_page(total_compressed);
710 struct page *page = pages[nr_pages - 1];
712 /* zero the tail end of the last page, we might be
713 * sending it down to disk
716 memzero_page(page, offset, PAGE_SIZE - offset);
722 * Check cow_file_range() for why we don't even try to create inline
723 * extent for subpage case.
725 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
726 /* lets try to make an inline extent */
727 if (ret || total_in < actual_end) {
728 /* we didn't compress the entire range, try
729 * to make an uncompressed inline extent.
731 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
732 0, BTRFS_COMPRESS_NONE,
735 /* try making a compressed inline extent */
736 ret = cow_file_range_inline(BTRFS_I(inode), actual_end,
738 compress_type, pages,
742 unsigned long clear_flags = EXTENT_DELALLOC |
743 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
744 EXTENT_DO_ACCOUNTING;
745 unsigned long page_error_op;
747 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
750 * inline extent creation worked or returned error,
751 * we don't need to create any more async work items.
752 * Unlock and free up our temp pages.
754 * We use DO_ACCOUNTING here because we need the
755 * delalloc_release_metadata to be done _after_ we drop
756 * our outstanding extent for clearing delalloc for this
759 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
763 PAGE_START_WRITEBACK |
768 * Ensure we only free the compressed pages if we have
769 * them allocated, as we can still reach here with
770 * inode_need_compress() == false.
773 for (i = 0; i < nr_pages; i++) {
774 WARN_ON(pages[i]->mapping);
785 * we aren't doing an inline extent round the compressed size
786 * up to a block size boundary so the allocator does sane
789 total_compressed = ALIGN(total_compressed, blocksize);
792 * one last check to make sure the compression is really a
793 * win, compare the page count read with the blocks on disk,
794 * compression must free at least one sector size
796 total_in = round_up(total_in, fs_info->sectorsize);
797 if (total_compressed + blocksize <= total_in) {
798 compressed_extents++;
801 * The async work queues will take care of doing actual
802 * allocation on disk for these compressed pages, and
803 * will submit them to the elevator.
805 add_async_extent(async_chunk, start, total_in,
806 total_compressed, pages, nr_pages,
809 if (start + total_in < end) {
815 return compressed_extents;
820 * the compression code ran but failed to make things smaller,
821 * free any pages it allocated and our page pointer array
823 for (i = 0; i < nr_pages; i++) {
824 WARN_ON(pages[i]->mapping);
829 total_compressed = 0;
832 /* flag the file so we don't compress in the future */
833 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
834 !(BTRFS_I(inode)->prop_compress)) {
835 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
838 cleanup_and_bail_uncompressed:
840 * No compression, but we still need to write the pages in the file
841 * we've been given so far. redirty the locked page if it corresponds
842 * to our extent and set things up for the async work queue to run
843 * cow_file_range to do the normal delalloc dance.
845 if (async_chunk->locked_page &&
846 (page_offset(async_chunk->locked_page) >= start &&
847 page_offset(async_chunk->locked_page)) <= end) {
848 __set_page_dirty_nobuffers(async_chunk->locked_page);
849 /* unlocked later on in the async handlers */
853 extent_range_redirty_for_io(inode, start, end);
854 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
855 BTRFS_COMPRESS_NONE);
856 compressed_extents++;
858 return compressed_extents;
861 static void free_async_extent_pages(struct async_extent *async_extent)
865 if (!async_extent->pages)
868 for (i = 0; i < async_extent->nr_pages; i++) {
869 WARN_ON(async_extent->pages[i]->mapping);
870 put_page(async_extent->pages[i]);
872 kfree(async_extent->pages);
873 async_extent->nr_pages = 0;
874 async_extent->pages = NULL;
877 static int submit_uncompressed_range(struct btrfs_inode *inode,
878 struct async_extent *async_extent,
879 struct page *locked_page)
881 u64 start = async_extent->start;
882 u64 end = async_extent->start + async_extent->ram_size - 1;
883 unsigned long nr_written = 0;
884 int page_started = 0;
888 * Call cow_file_range() to run the delalloc range directly, since we
889 * won't go to NOCOW or async path again.
891 * Also we call cow_file_range() with @unlock_page == 0, so that we
892 * can directly submit them without interruption.
894 ret = cow_file_range(inode, locked_page, start, end, &page_started,
896 /* Inline extent inserted, page gets unlocked and everything is done */
903 unlock_page(locked_page);
907 ret = extent_write_locked_range(&inode->vfs_inode, start, end);
908 /* All pages will be unlocked, including @locked_page */
914 static int submit_one_async_extent(struct btrfs_inode *inode,
915 struct async_chunk *async_chunk,
916 struct async_extent *async_extent,
919 struct extent_io_tree *io_tree = &inode->io_tree;
920 struct btrfs_root *root = inode->root;
921 struct btrfs_fs_info *fs_info = root->fs_info;
922 struct btrfs_key ins;
923 struct page *locked_page = NULL;
924 struct extent_map *em;
926 u64 start = async_extent->start;
927 u64 end = async_extent->start + async_extent->ram_size - 1;
930 * If async_chunk->locked_page is in the async_extent range, we need to
933 if (async_chunk->locked_page) {
934 u64 locked_page_start = page_offset(async_chunk->locked_page);
935 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
937 if (!(start >= locked_page_end || end <= locked_page_start))
938 locked_page = async_chunk->locked_page;
940 lock_extent(io_tree, start, end);
942 /* We have fall back to uncompressed write */
943 if (!async_extent->pages)
944 return submit_uncompressed_range(inode, async_extent, locked_page);
946 ret = btrfs_reserve_extent(root, async_extent->ram_size,
947 async_extent->compressed_size,
948 async_extent->compressed_size,
949 0, *alloc_hint, &ins, 1, 1);
951 free_async_extent_pages(async_extent);
953 * Here we used to try again by going back to non-compressed
954 * path for ENOSPC. But we can't reserve space even for
955 * compressed size, how could it work for uncompressed size
956 * which requires larger size? So here we directly go error
962 /* Here we're doing allocation and writeback of the compressed pages */
963 em = create_io_em(inode, start,
964 async_extent->ram_size, /* len */
965 start, /* orig_start */
966 ins.objectid, /* block_start */
967 ins.offset, /* block_len */
968 ins.offset, /* orig_block_len */
969 async_extent->ram_size, /* ram_bytes */
970 async_extent->compress_type,
971 BTRFS_ORDERED_COMPRESSED);
974 goto out_free_reserve;
978 ret = btrfs_add_ordered_extent(inode, start, /* file_offset */
979 async_extent->ram_size, /* num_bytes */
980 async_extent->ram_size, /* ram_bytes */
981 ins.objectid, /* disk_bytenr */
982 ins.offset, /* disk_num_bytes */
984 1 << BTRFS_ORDERED_COMPRESSED,
985 async_extent->compress_type);
987 btrfs_drop_extent_cache(inode, start, end, 0);
988 goto out_free_reserve;
990 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
992 /* Clear dirty, set writeback and unlock the pages. */
993 extent_clear_unlock_delalloc(inode, start, end,
994 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
995 PAGE_UNLOCK | PAGE_START_WRITEBACK);
996 if (btrfs_submit_compressed_write(inode, start, /* file_offset */
997 async_extent->ram_size, /* num_bytes */
998 ins.objectid, /* disk_bytenr */
999 ins.offset, /* compressed_len */
1000 async_extent->pages, /* compressed_pages */
1001 async_extent->nr_pages,
1002 async_chunk->write_flags,
1003 async_chunk->blkcg_css, true)) {
1004 const u64 start = async_extent->start;
1005 const u64 end = start + async_extent->ram_size - 1;
1007 btrfs_writepage_endio_finish_ordered(inode, NULL, start, end, 0);
1009 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
1010 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1011 free_async_extent_pages(async_extent);
1013 *alloc_hint = ins.objectid + ins.offset;
1014 kfree(async_extent);
1018 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1019 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1021 extent_clear_unlock_delalloc(inode, start, end,
1022 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1023 EXTENT_DELALLOC_NEW |
1024 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1025 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1026 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1027 free_async_extent_pages(async_extent);
1028 kfree(async_extent);
1033 * Phase two of compressed writeback. This is the ordered portion of the code,
1034 * which only gets called in the order the work was queued. We walk all the
1035 * async extents created by compress_file_range and send them down to the disk.
1037 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1039 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
1040 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1041 struct async_extent *async_extent;
1045 while (!list_empty(&async_chunk->extents)) {
1049 async_extent = list_entry(async_chunk->extents.next,
1050 struct async_extent, list);
1051 list_del(&async_extent->list);
1052 extent_start = async_extent->start;
1053 ram_size = async_extent->ram_size;
1055 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1057 btrfs_debug(fs_info,
1058 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1059 inode->root->root_key.objectid,
1060 btrfs_ino(inode), extent_start, ram_size, ret);
1064 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1067 struct extent_map_tree *em_tree = &inode->extent_tree;
1068 struct extent_map *em;
1071 read_lock(&em_tree->lock);
1072 em = search_extent_mapping(em_tree, start, num_bytes);
1075 * if block start isn't an actual block number then find the
1076 * first block in this inode and use that as a hint. If that
1077 * block is also bogus then just don't worry about it.
1079 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1080 free_extent_map(em);
1081 em = search_extent_mapping(em_tree, 0, 0);
1082 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1083 alloc_hint = em->block_start;
1085 free_extent_map(em);
1087 alloc_hint = em->block_start;
1088 free_extent_map(em);
1091 read_unlock(&em_tree->lock);
1097 * when extent_io.c finds a delayed allocation range in the file,
1098 * the call backs end up in this code. The basic idea is to
1099 * allocate extents on disk for the range, and create ordered data structs
1100 * in ram to track those extents.
1102 * locked_page is the page that writepage had locked already. We use
1103 * it to make sure we don't do extra locks or unlocks.
1105 * *page_started is set to one if we unlock locked_page and do everything
1106 * required to start IO on it. It may be clean and already done with
1107 * IO when we return.
1109 static noinline int cow_file_range(struct btrfs_inode *inode,
1110 struct page *locked_page,
1111 u64 start, u64 end, int *page_started,
1112 unsigned long *nr_written, int unlock)
1114 struct btrfs_root *root = inode->root;
1115 struct btrfs_fs_info *fs_info = root->fs_info;
1118 unsigned long ram_size;
1119 u64 cur_alloc_size = 0;
1121 u64 blocksize = fs_info->sectorsize;
1122 struct btrfs_key ins;
1123 struct extent_map *em;
1124 unsigned clear_bits;
1125 unsigned long page_ops;
1126 bool extent_reserved = false;
1129 if (btrfs_is_free_space_inode(inode)) {
1134 num_bytes = ALIGN(end - start + 1, blocksize);
1135 num_bytes = max(blocksize, num_bytes);
1136 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1138 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1141 * Due to the page size limit, for subpage we can only trigger the
1142 * writeback for the dirty sectors of page, that means data writeback
1143 * is doing more writeback than what we want.
1145 * This is especially unexpected for some call sites like fallocate,
1146 * where we only increase i_size after everything is done.
1147 * This means we can trigger inline extent even if we didn't want to.
1148 * So here we skip inline extent creation completely.
1150 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1151 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1154 /* lets try to make an inline extent */
1155 ret = cow_file_range_inline(inode, actual_end, 0,
1156 BTRFS_COMPRESS_NONE, NULL, false);
1159 * We use DO_ACCOUNTING here because we need the
1160 * delalloc_release_metadata to be run _after_ we drop
1161 * our outstanding extent for clearing delalloc for this
1164 extent_clear_unlock_delalloc(inode, start, end,
1166 EXTENT_LOCKED | EXTENT_DELALLOC |
1167 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1168 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1169 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1170 *nr_written = *nr_written +
1171 (end - start + PAGE_SIZE) / PAGE_SIZE;
1174 * locked_page is locked by the caller of
1175 * writepage_delalloc(), not locked by
1176 * __process_pages_contig().
1178 * We can't let __process_pages_contig() to unlock it,
1179 * as it doesn't have any subpage::writers recorded.
1181 * Here we manually unlock the page, since the caller
1182 * can't use page_started to determine if it's an
1183 * inline extent or a compressed extent.
1185 unlock_page(locked_page);
1187 } else if (ret < 0) {
1192 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1193 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1196 * Relocation relies on the relocated extents to have exactly the same
1197 * size as the original extents. Normally writeback for relocation data
1198 * extents follows a NOCOW path because relocation preallocates the
1199 * extents. However, due to an operation such as scrub turning a block
1200 * group to RO mode, it may fallback to COW mode, so we must make sure
1201 * an extent allocated during COW has exactly the requested size and can
1202 * not be split into smaller extents, otherwise relocation breaks and
1203 * fails during the stage where it updates the bytenr of file extent
1206 if (btrfs_is_data_reloc_root(root))
1207 min_alloc_size = num_bytes;
1209 min_alloc_size = fs_info->sectorsize;
1211 while (num_bytes > 0) {
1212 cur_alloc_size = num_bytes;
1213 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1214 min_alloc_size, 0, alloc_hint,
1218 cur_alloc_size = ins.offset;
1219 extent_reserved = true;
1221 ram_size = ins.offset;
1222 em = create_io_em(inode, start, ins.offset, /* len */
1223 start, /* orig_start */
1224 ins.objectid, /* block_start */
1225 ins.offset, /* block_len */
1226 ins.offset, /* orig_block_len */
1227 ram_size, /* ram_bytes */
1228 BTRFS_COMPRESS_NONE, /* compress_type */
1229 BTRFS_ORDERED_REGULAR /* type */);
1234 free_extent_map(em);
1236 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1237 ins.objectid, cur_alloc_size, 0,
1238 1 << BTRFS_ORDERED_REGULAR,
1239 BTRFS_COMPRESS_NONE);
1241 goto out_drop_extent_cache;
1243 if (btrfs_is_data_reloc_root(root)) {
1244 ret = btrfs_reloc_clone_csums(inode, start,
1247 * Only drop cache here, and process as normal.
1249 * We must not allow extent_clear_unlock_delalloc()
1250 * at out_unlock label to free meta of this ordered
1251 * extent, as its meta should be freed by
1252 * btrfs_finish_ordered_io().
1254 * So we must continue until @start is increased to
1255 * skip current ordered extent.
1258 btrfs_drop_extent_cache(inode, start,
1259 start + ram_size - 1, 0);
1262 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1265 * We're not doing compressed IO, don't unlock the first page
1266 * (which the caller expects to stay locked), don't clear any
1267 * dirty bits and don't set any writeback bits
1269 * Do set the Ordered (Private2) bit so we know this page was
1270 * properly setup for writepage.
1272 page_ops = unlock ? PAGE_UNLOCK : 0;
1273 page_ops |= PAGE_SET_ORDERED;
1275 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1277 EXTENT_LOCKED | EXTENT_DELALLOC,
1279 if (num_bytes < cur_alloc_size)
1282 num_bytes -= cur_alloc_size;
1283 alloc_hint = ins.objectid + ins.offset;
1284 start += cur_alloc_size;
1285 extent_reserved = false;
1288 * btrfs_reloc_clone_csums() error, since start is increased
1289 * extent_clear_unlock_delalloc() at out_unlock label won't
1290 * free metadata of current ordered extent, we're OK to exit.
1298 out_drop_extent_cache:
1299 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1301 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1302 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1304 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1305 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1306 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1308 * If we reserved an extent for our delalloc range (or a subrange) and
1309 * failed to create the respective ordered extent, then it means that
1310 * when we reserved the extent we decremented the extent's size from
1311 * the data space_info's bytes_may_use counter and incremented the
1312 * space_info's bytes_reserved counter by the same amount. We must make
1313 * sure extent_clear_unlock_delalloc() does not try to decrement again
1314 * the data space_info's bytes_may_use counter, therefore we do not pass
1315 * it the flag EXTENT_CLEAR_DATA_RESV.
1317 if (extent_reserved) {
1318 extent_clear_unlock_delalloc(inode, start,
1319 start + cur_alloc_size - 1,
1323 start += cur_alloc_size;
1327 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1328 clear_bits | EXTENT_CLEAR_DATA_RESV,
1334 * work queue call back to started compression on a file and pages
1336 static noinline void async_cow_start(struct btrfs_work *work)
1338 struct async_chunk *async_chunk;
1339 int compressed_extents;
1341 async_chunk = container_of(work, struct async_chunk, work);
1343 compressed_extents = compress_file_range(async_chunk);
1344 if (compressed_extents == 0) {
1345 btrfs_add_delayed_iput(async_chunk->inode);
1346 async_chunk->inode = NULL;
1351 * work queue call back to submit previously compressed pages
1353 static noinline void async_cow_submit(struct btrfs_work *work)
1355 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1357 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1358 unsigned long nr_pages;
1360 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1364 * ->inode could be NULL if async_chunk_start has failed to compress,
1365 * in which case we don't have anything to submit, yet we need to
1366 * always adjust ->async_delalloc_pages as its paired with the init
1367 * happening in cow_file_range_async
1369 if (async_chunk->inode)
1370 submit_compressed_extents(async_chunk);
1372 /* atomic_sub_return implies a barrier */
1373 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1375 cond_wake_up_nomb(&fs_info->async_submit_wait);
1378 static noinline void async_cow_free(struct btrfs_work *work)
1380 struct async_chunk *async_chunk;
1381 struct async_cow *async_cow;
1383 async_chunk = container_of(work, struct async_chunk, work);
1384 if (async_chunk->inode)
1385 btrfs_add_delayed_iput(async_chunk->inode);
1386 if (async_chunk->blkcg_css)
1387 css_put(async_chunk->blkcg_css);
1389 async_cow = async_chunk->async_cow;
1390 if (atomic_dec_and_test(&async_cow->num_chunks))
1394 static int cow_file_range_async(struct btrfs_inode *inode,
1395 struct writeback_control *wbc,
1396 struct page *locked_page,
1397 u64 start, u64 end, int *page_started,
1398 unsigned long *nr_written)
1400 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1401 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1402 struct async_cow *ctx;
1403 struct async_chunk *async_chunk;
1404 unsigned long nr_pages;
1406 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1408 bool should_compress;
1410 const unsigned int write_flags = wbc_to_write_flags(wbc);
1412 unlock_extent(&inode->io_tree, start, end);
1414 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1415 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1417 should_compress = false;
1419 should_compress = true;
1422 nofs_flag = memalloc_nofs_save();
1423 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1424 memalloc_nofs_restore(nofs_flag);
1427 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1428 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1429 EXTENT_DO_ACCOUNTING;
1430 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1431 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1433 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1434 clear_bits, page_ops);
1438 async_chunk = ctx->chunks;
1439 atomic_set(&ctx->num_chunks, num_chunks);
1441 for (i = 0; i < num_chunks; i++) {
1442 if (should_compress)
1443 cur_end = min(end, start + SZ_512K - 1);
1448 * igrab is called higher up in the call chain, take only the
1449 * lightweight reference for the callback lifetime
1451 ihold(&inode->vfs_inode);
1452 async_chunk[i].async_cow = ctx;
1453 async_chunk[i].inode = &inode->vfs_inode;
1454 async_chunk[i].start = start;
1455 async_chunk[i].end = cur_end;
1456 async_chunk[i].write_flags = write_flags;
1457 INIT_LIST_HEAD(&async_chunk[i].extents);
1460 * The locked_page comes all the way from writepage and its
1461 * the original page we were actually given. As we spread
1462 * this large delalloc region across multiple async_chunk
1463 * structs, only the first struct needs a pointer to locked_page
1465 * This way we don't need racey decisions about who is supposed
1470 * Depending on the compressibility, the pages might or
1471 * might not go through async. We want all of them to
1472 * be accounted against wbc once. Let's do it here
1473 * before the paths diverge. wbc accounting is used
1474 * only for foreign writeback detection and doesn't
1475 * need full accuracy. Just account the whole thing
1476 * against the first page.
1478 wbc_account_cgroup_owner(wbc, locked_page,
1480 async_chunk[i].locked_page = locked_page;
1483 async_chunk[i].locked_page = NULL;
1486 if (blkcg_css != blkcg_root_css) {
1488 async_chunk[i].blkcg_css = blkcg_css;
1490 async_chunk[i].blkcg_css = NULL;
1493 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1494 async_cow_submit, async_cow_free);
1496 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1497 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1499 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1501 *nr_written += nr_pages;
1502 start = cur_end + 1;
1508 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1509 struct page *locked_page, u64 start,
1510 u64 end, int *page_started,
1511 unsigned long *nr_written)
1515 ret = cow_file_range(inode, locked_page, start, end, page_started,
1523 __set_page_dirty_nobuffers(locked_page);
1524 account_page_redirty(locked_page);
1525 extent_write_locked_range(&inode->vfs_inode, start, end);
1531 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1532 u64 bytenr, u64 num_bytes)
1534 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1535 struct btrfs_ordered_sum *sums;
1539 ret = btrfs_lookup_csums_range(csum_root, bytenr,
1540 bytenr + num_bytes - 1, &list, 0);
1541 if (ret == 0 && list_empty(&list))
1544 while (!list_empty(&list)) {
1545 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1546 list_del(&sums->list);
1554 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1555 const u64 start, const u64 end,
1556 int *page_started, unsigned long *nr_written)
1558 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1559 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1560 const u64 range_bytes = end + 1 - start;
1561 struct extent_io_tree *io_tree = &inode->io_tree;
1562 u64 range_start = start;
1566 * If EXTENT_NORESERVE is set it means that when the buffered write was
1567 * made we had not enough available data space and therefore we did not
1568 * reserve data space for it, since we though we could do NOCOW for the
1569 * respective file range (either there is prealloc extent or the inode
1570 * has the NOCOW bit set).
1572 * However when we need to fallback to COW mode (because for example the
1573 * block group for the corresponding extent was turned to RO mode by a
1574 * scrub or relocation) we need to do the following:
1576 * 1) We increment the bytes_may_use counter of the data space info.
1577 * If COW succeeds, it allocates a new data extent and after doing
1578 * that it decrements the space info's bytes_may_use counter and
1579 * increments its bytes_reserved counter by the same amount (we do
1580 * this at btrfs_add_reserved_bytes()). So we need to increment the
1581 * bytes_may_use counter to compensate (when space is reserved at
1582 * buffered write time, the bytes_may_use counter is incremented);
1584 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1585 * that if the COW path fails for any reason, it decrements (through
1586 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1587 * data space info, which we incremented in the step above.
1589 * If we need to fallback to cow and the inode corresponds to a free
1590 * space cache inode or an inode of the data relocation tree, we must
1591 * also increment bytes_may_use of the data space_info for the same
1592 * reason. Space caches and relocated data extents always get a prealloc
1593 * extent for them, however scrub or balance may have set the block
1594 * group that contains that extent to RO mode and therefore force COW
1595 * when starting writeback.
1597 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1598 EXTENT_NORESERVE, 0);
1599 if (count > 0 || is_space_ino || is_reloc_ino) {
1601 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1602 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1604 if (is_space_ino || is_reloc_ino)
1605 bytes = range_bytes;
1607 spin_lock(&sinfo->lock);
1608 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1609 spin_unlock(&sinfo->lock);
1612 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1616 return cow_file_range(inode, locked_page, start, end, page_started,
1621 * when nowcow writeback call back. This checks for snapshots or COW copies
1622 * of the extents that exist in the file, and COWs the file as required.
1624 * If no cow copies or snapshots exist, we write directly to the existing
1627 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1628 struct page *locked_page,
1629 const u64 start, const u64 end,
1631 unsigned long *nr_written)
1633 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1634 struct btrfs_root *root = inode->root;
1635 struct btrfs_path *path;
1636 u64 cow_start = (u64)-1;
1637 u64 cur_offset = start;
1639 bool check_prev = true;
1640 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1641 u64 ino = btrfs_ino(inode);
1643 u64 disk_bytenr = 0;
1644 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1646 path = btrfs_alloc_path();
1648 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1649 EXTENT_LOCKED | EXTENT_DELALLOC |
1650 EXTENT_DO_ACCOUNTING |
1651 EXTENT_DEFRAG, PAGE_UNLOCK |
1652 PAGE_START_WRITEBACK |
1653 PAGE_END_WRITEBACK);
1658 struct btrfs_key found_key;
1659 struct btrfs_file_extent_item *fi;
1660 struct extent_buffer *leaf;
1670 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1676 * If there is no extent for our range when doing the initial
1677 * search, then go back to the previous slot as it will be the
1678 * one containing the search offset
1680 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1681 leaf = path->nodes[0];
1682 btrfs_item_key_to_cpu(leaf, &found_key,
1683 path->slots[0] - 1);
1684 if (found_key.objectid == ino &&
1685 found_key.type == BTRFS_EXTENT_DATA_KEY)
1690 /* Go to next leaf if we have exhausted the current one */
1691 leaf = path->nodes[0];
1692 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1693 ret = btrfs_next_leaf(root, path);
1695 if (cow_start != (u64)-1)
1696 cur_offset = cow_start;
1701 leaf = path->nodes[0];
1704 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1706 /* Didn't find anything for our INO */
1707 if (found_key.objectid > ino)
1710 * Keep searching until we find an EXTENT_ITEM or there are no
1711 * more extents for this inode
1713 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1714 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1719 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1720 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1721 found_key.offset > end)
1725 * If the found extent starts after requested offset, then
1726 * adjust extent_end to be right before this extent begins
1728 if (found_key.offset > cur_offset) {
1729 extent_end = found_key.offset;
1735 * Found extent which begins before our range and potentially
1738 fi = btrfs_item_ptr(leaf, path->slots[0],
1739 struct btrfs_file_extent_item);
1740 extent_type = btrfs_file_extent_type(leaf, fi);
1742 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1743 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1744 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1745 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1746 extent_offset = btrfs_file_extent_offset(leaf, fi);
1747 extent_end = found_key.offset +
1748 btrfs_file_extent_num_bytes(leaf, fi);
1750 btrfs_file_extent_disk_num_bytes(leaf, fi);
1752 * If the extent we got ends before our current offset,
1753 * skip to the next extent.
1755 if (extent_end <= cur_offset) {
1760 if (disk_bytenr == 0)
1762 /* Skip compressed/encrypted/encoded extents */
1763 if (btrfs_file_extent_compression(leaf, fi) ||
1764 btrfs_file_extent_encryption(leaf, fi) ||
1765 btrfs_file_extent_other_encoding(leaf, fi))
1768 * If extent is created before the last volume's snapshot
1769 * this implies the extent is shared, hence we can't do
1770 * nocow. This is the same check as in
1771 * btrfs_cross_ref_exist but without calling
1772 * btrfs_search_slot.
1774 if (!freespace_inode &&
1775 btrfs_file_extent_generation(leaf, fi) <=
1776 btrfs_root_last_snapshot(&root->root_item))
1778 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1782 * The following checks can be expensive, as they need to
1783 * take other locks and do btree or rbtree searches, so
1784 * release the path to avoid blocking other tasks for too
1787 btrfs_release_path(path);
1789 ret = btrfs_cross_ref_exist(root, ino,
1791 extent_offset, disk_bytenr,
1795 * ret could be -EIO if the above fails to read
1799 if (cow_start != (u64)-1)
1800 cur_offset = cow_start;
1804 WARN_ON_ONCE(freespace_inode);
1807 disk_bytenr += extent_offset;
1808 disk_bytenr += cur_offset - found_key.offset;
1809 num_bytes = min(end + 1, extent_end) - cur_offset;
1811 * If there are pending snapshots for this root, we
1812 * fall into common COW way
1814 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1817 * force cow if csum exists in the range.
1818 * this ensure that csum for a given extent are
1819 * either valid or do not exist.
1821 ret = csum_exist_in_range(fs_info, disk_bytenr,
1825 * ret could be -EIO if the above fails to read
1829 if (cow_start != (u64)-1)
1830 cur_offset = cow_start;
1833 WARN_ON_ONCE(freespace_inode);
1836 /* If the extent's block group is RO, we must COW */
1837 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1840 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1841 extent_end = found_key.offset + ram_bytes;
1842 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1843 /* Skip extents outside of our requested range */
1844 if (extent_end <= start) {
1849 /* If this triggers then we have a memory corruption */
1854 * If nocow is false then record the beginning of the range
1855 * that needs to be COWed
1858 if (cow_start == (u64)-1)
1859 cow_start = cur_offset;
1860 cur_offset = extent_end;
1861 if (cur_offset > end)
1863 if (!path->nodes[0])
1870 * COW range from cow_start to found_key.offset - 1. As the key
1871 * will contain the beginning of the first extent that can be
1872 * NOCOW, following one which needs to be COW'ed
1874 if (cow_start != (u64)-1) {
1875 ret = fallback_to_cow(inode, locked_page,
1876 cow_start, found_key.offset - 1,
1877 page_started, nr_written);
1880 cow_start = (u64)-1;
1883 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1884 u64 orig_start = found_key.offset - extent_offset;
1885 struct extent_map *em;
1887 em = create_io_em(inode, cur_offset, num_bytes,
1889 disk_bytenr, /* block_start */
1890 num_bytes, /* block_len */
1891 disk_num_bytes, /* orig_block_len */
1892 ram_bytes, BTRFS_COMPRESS_NONE,
1893 BTRFS_ORDERED_PREALLOC);
1898 free_extent_map(em);
1899 ret = btrfs_add_ordered_extent(inode,
1900 cur_offset, num_bytes, num_bytes,
1901 disk_bytenr, num_bytes, 0,
1902 1 << BTRFS_ORDERED_PREALLOC,
1903 BTRFS_COMPRESS_NONE);
1905 btrfs_drop_extent_cache(inode, cur_offset,
1906 cur_offset + num_bytes - 1,
1911 ret = btrfs_add_ordered_extent(inode, cur_offset,
1912 num_bytes, num_bytes,
1913 disk_bytenr, num_bytes,
1915 1 << BTRFS_ORDERED_NOCOW,
1916 BTRFS_COMPRESS_NONE);
1922 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1925 if (btrfs_is_data_reloc_root(root))
1927 * Error handled later, as we must prevent
1928 * extent_clear_unlock_delalloc() in error handler
1929 * from freeing metadata of created ordered extent.
1931 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1934 extent_clear_unlock_delalloc(inode, cur_offset,
1935 cur_offset + num_bytes - 1,
1936 locked_page, EXTENT_LOCKED |
1938 EXTENT_CLEAR_DATA_RESV,
1939 PAGE_UNLOCK | PAGE_SET_ORDERED);
1941 cur_offset = extent_end;
1944 * btrfs_reloc_clone_csums() error, now we're OK to call error
1945 * handler, as metadata for created ordered extent will only
1946 * be freed by btrfs_finish_ordered_io().
1950 if (cur_offset > end)
1953 btrfs_release_path(path);
1955 if (cur_offset <= end && cow_start == (u64)-1)
1956 cow_start = cur_offset;
1958 if (cow_start != (u64)-1) {
1960 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1961 page_started, nr_written);
1968 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1970 if (ret && cur_offset < end)
1971 extent_clear_unlock_delalloc(inode, cur_offset, end,
1972 locked_page, EXTENT_LOCKED |
1973 EXTENT_DELALLOC | EXTENT_DEFRAG |
1974 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1975 PAGE_START_WRITEBACK |
1976 PAGE_END_WRITEBACK);
1977 btrfs_free_path(path);
1981 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1983 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1984 if (inode->defrag_bytes &&
1985 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1994 * Function to process delayed allocation (create CoW) for ranges which are
1995 * being touched for the first time.
1997 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1998 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1999 struct writeback_control *wbc)
2002 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2005 * The range must cover part of the @locked_page, or the returned
2006 * @page_started can confuse the caller.
2008 ASSERT(!(end <= page_offset(locked_page) ||
2009 start >= page_offset(locked_page) + PAGE_SIZE));
2011 if (should_nocow(inode, start, end)) {
2013 * Normally on a zoned device we're only doing COW writes, but
2014 * in case of relocation on a zoned filesystem we have taken
2015 * precaution, that we're only writing sequentially. It's safe
2016 * to use run_delalloc_nocow() here, like for regular
2017 * preallocated inodes.
2019 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2020 ret = run_delalloc_nocow(inode, locked_page, start, end,
2021 page_started, nr_written);
2022 } else if (!btrfs_inode_can_compress(inode) ||
2023 !inode_need_compress(inode, start, end)) {
2025 ret = run_delalloc_zoned(inode, locked_page, start, end,
2026 page_started, nr_written);
2028 ret = cow_file_range(inode, locked_page, start, end,
2029 page_started, nr_written, 1);
2031 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2032 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2033 page_started, nr_written);
2037 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2042 void btrfs_split_delalloc_extent(struct inode *inode,
2043 struct extent_state *orig, u64 split)
2047 /* not delalloc, ignore it */
2048 if (!(orig->state & EXTENT_DELALLOC))
2051 size = orig->end - orig->start + 1;
2052 if (size > BTRFS_MAX_EXTENT_SIZE) {
2057 * See the explanation in btrfs_merge_delalloc_extent, the same
2058 * applies here, just in reverse.
2060 new_size = orig->end - split + 1;
2061 num_extents = count_max_extents(new_size);
2062 new_size = split - orig->start;
2063 num_extents += count_max_extents(new_size);
2064 if (count_max_extents(size) >= num_extents)
2068 spin_lock(&BTRFS_I(inode)->lock);
2069 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
2070 spin_unlock(&BTRFS_I(inode)->lock);
2074 * Handle merged delayed allocation extents so we can keep track of new extents
2075 * that are just merged onto old extents, such as when we are doing sequential
2076 * writes, so we can properly account for the metadata space we'll need.
2078 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
2079 struct extent_state *other)
2081 u64 new_size, old_size;
2084 /* not delalloc, ignore it */
2085 if (!(other->state & EXTENT_DELALLOC))
2088 if (new->start > other->start)
2089 new_size = new->end - other->start + 1;
2091 new_size = other->end - new->start + 1;
2093 /* we're not bigger than the max, unreserve the space and go */
2094 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
2095 spin_lock(&BTRFS_I(inode)->lock);
2096 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2097 spin_unlock(&BTRFS_I(inode)->lock);
2102 * We have to add up either side to figure out how many extents were
2103 * accounted for before we merged into one big extent. If the number of
2104 * extents we accounted for is <= the amount we need for the new range
2105 * then we can return, otherwise drop. Think of it like this
2109 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2110 * need 2 outstanding extents, on one side we have 1 and the other side
2111 * we have 1 so they are == and we can return. But in this case
2113 * [MAX_SIZE+4k][MAX_SIZE+4k]
2115 * Each range on their own accounts for 2 extents, but merged together
2116 * they are only 3 extents worth of accounting, so we need to drop in
2119 old_size = other->end - other->start + 1;
2120 num_extents = count_max_extents(old_size);
2121 old_size = new->end - new->start + 1;
2122 num_extents += count_max_extents(old_size);
2123 if (count_max_extents(new_size) >= num_extents)
2126 spin_lock(&BTRFS_I(inode)->lock);
2127 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2128 spin_unlock(&BTRFS_I(inode)->lock);
2131 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2132 struct inode *inode)
2134 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2136 spin_lock(&root->delalloc_lock);
2137 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2138 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2139 &root->delalloc_inodes);
2140 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2141 &BTRFS_I(inode)->runtime_flags);
2142 root->nr_delalloc_inodes++;
2143 if (root->nr_delalloc_inodes == 1) {
2144 spin_lock(&fs_info->delalloc_root_lock);
2145 BUG_ON(!list_empty(&root->delalloc_root));
2146 list_add_tail(&root->delalloc_root,
2147 &fs_info->delalloc_roots);
2148 spin_unlock(&fs_info->delalloc_root_lock);
2151 spin_unlock(&root->delalloc_lock);
2155 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2156 struct btrfs_inode *inode)
2158 struct btrfs_fs_info *fs_info = root->fs_info;
2160 if (!list_empty(&inode->delalloc_inodes)) {
2161 list_del_init(&inode->delalloc_inodes);
2162 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2163 &inode->runtime_flags);
2164 root->nr_delalloc_inodes--;
2165 if (!root->nr_delalloc_inodes) {
2166 ASSERT(list_empty(&root->delalloc_inodes));
2167 spin_lock(&fs_info->delalloc_root_lock);
2168 BUG_ON(list_empty(&root->delalloc_root));
2169 list_del_init(&root->delalloc_root);
2170 spin_unlock(&fs_info->delalloc_root_lock);
2175 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2176 struct btrfs_inode *inode)
2178 spin_lock(&root->delalloc_lock);
2179 __btrfs_del_delalloc_inode(root, inode);
2180 spin_unlock(&root->delalloc_lock);
2184 * Properly track delayed allocation bytes in the inode and to maintain the
2185 * list of inodes that have pending delalloc work to be done.
2187 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2190 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2192 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2195 * set_bit and clear bit hooks normally require _irqsave/restore
2196 * but in this case, we are only testing for the DELALLOC
2197 * bit, which is only set or cleared with irqs on
2199 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2200 struct btrfs_root *root = BTRFS_I(inode)->root;
2201 u64 len = state->end + 1 - state->start;
2202 u32 num_extents = count_max_extents(len);
2203 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2205 spin_lock(&BTRFS_I(inode)->lock);
2206 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2207 spin_unlock(&BTRFS_I(inode)->lock);
2209 /* For sanity tests */
2210 if (btrfs_is_testing(fs_info))
2213 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2214 fs_info->delalloc_batch);
2215 spin_lock(&BTRFS_I(inode)->lock);
2216 BTRFS_I(inode)->delalloc_bytes += len;
2217 if (*bits & EXTENT_DEFRAG)
2218 BTRFS_I(inode)->defrag_bytes += len;
2219 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2220 &BTRFS_I(inode)->runtime_flags))
2221 btrfs_add_delalloc_inodes(root, inode);
2222 spin_unlock(&BTRFS_I(inode)->lock);
2225 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2226 (*bits & EXTENT_DELALLOC_NEW)) {
2227 spin_lock(&BTRFS_I(inode)->lock);
2228 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2230 spin_unlock(&BTRFS_I(inode)->lock);
2235 * Once a range is no longer delalloc this function ensures that proper
2236 * accounting happens.
2238 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2239 struct extent_state *state, unsigned *bits)
2241 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2242 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2243 u64 len = state->end + 1 - state->start;
2244 u32 num_extents = count_max_extents(len);
2246 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2247 spin_lock(&inode->lock);
2248 inode->defrag_bytes -= len;
2249 spin_unlock(&inode->lock);
2253 * set_bit and clear bit hooks normally require _irqsave/restore
2254 * but in this case, we are only testing for the DELALLOC
2255 * bit, which is only set or cleared with irqs on
2257 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2258 struct btrfs_root *root = inode->root;
2259 bool do_list = !btrfs_is_free_space_inode(inode);
2261 spin_lock(&inode->lock);
2262 btrfs_mod_outstanding_extents(inode, -num_extents);
2263 spin_unlock(&inode->lock);
2266 * We don't reserve metadata space for space cache inodes so we
2267 * don't need to call delalloc_release_metadata if there is an
2270 if (*bits & EXTENT_CLEAR_META_RESV &&
2271 root != fs_info->tree_root)
2272 btrfs_delalloc_release_metadata(inode, len, false);
2274 /* For sanity tests. */
2275 if (btrfs_is_testing(fs_info))
2278 if (!btrfs_is_data_reloc_root(root) &&
2279 do_list && !(state->state & EXTENT_NORESERVE) &&
2280 (*bits & EXTENT_CLEAR_DATA_RESV))
2281 btrfs_free_reserved_data_space_noquota(fs_info, len);
2283 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2284 fs_info->delalloc_batch);
2285 spin_lock(&inode->lock);
2286 inode->delalloc_bytes -= len;
2287 if (do_list && inode->delalloc_bytes == 0 &&
2288 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2289 &inode->runtime_flags))
2290 btrfs_del_delalloc_inode(root, inode);
2291 spin_unlock(&inode->lock);
2294 if ((state->state & EXTENT_DELALLOC_NEW) &&
2295 (*bits & EXTENT_DELALLOC_NEW)) {
2296 spin_lock(&inode->lock);
2297 ASSERT(inode->new_delalloc_bytes >= len);
2298 inode->new_delalloc_bytes -= len;
2299 if (*bits & EXTENT_ADD_INODE_BYTES)
2300 inode_add_bytes(&inode->vfs_inode, len);
2301 spin_unlock(&inode->lock);
2306 * in order to insert checksums into the metadata in large chunks,
2307 * we wait until bio submission time. All the pages in the bio are
2308 * checksummed and sums are attached onto the ordered extent record.
2310 * At IO completion time the cums attached on the ordered extent record
2311 * are inserted into the btree
2313 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2314 u64 dio_file_offset)
2316 return btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2320 * Split an extent_map at [start, start + len]
2322 * This function is intended to be used only for extract_ordered_extent().
2324 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2327 struct extent_map_tree *em_tree = &inode->extent_tree;
2328 struct extent_map *em;
2329 struct extent_map *split_pre = NULL;
2330 struct extent_map *split_mid = NULL;
2331 struct extent_map *split_post = NULL;
2333 unsigned long flags;
2336 if (pre == 0 && post == 0)
2339 split_pre = alloc_extent_map();
2341 split_mid = alloc_extent_map();
2343 split_post = alloc_extent_map();
2344 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2349 ASSERT(pre + post < len);
2351 lock_extent(&inode->io_tree, start, start + len - 1);
2352 write_lock(&em_tree->lock);
2353 em = lookup_extent_mapping(em_tree, start, len);
2359 ASSERT(em->len == len);
2360 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2361 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2362 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2363 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2364 ASSERT(!list_empty(&em->list));
2367 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2369 /* First, replace the em with a new extent_map starting from * em->start */
2370 split_pre->start = em->start;
2371 split_pre->len = (pre ? pre : em->len - post);
2372 split_pre->orig_start = split_pre->start;
2373 split_pre->block_start = em->block_start;
2374 split_pre->block_len = split_pre->len;
2375 split_pre->orig_block_len = split_pre->block_len;
2376 split_pre->ram_bytes = split_pre->len;
2377 split_pre->flags = flags;
2378 split_pre->compress_type = em->compress_type;
2379 split_pre->generation = em->generation;
2381 replace_extent_mapping(em_tree, em, split_pre, 1);
2384 * Now we only have an extent_map at:
2385 * [em->start, em->start + pre] if pre != 0
2386 * [em->start, em->start + em->len - post] if pre == 0
2390 /* Insert the middle extent_map */
2391 split_mid->start = em->start + pre;
2392 split_mid->len = em->len - pre - post;
2393 split_mid->orig_start = split_mid->start;
2394 split_mid->block_start = em->block_start + pre;
2395 split_mid->block_len = split_mid->len;
2396 split_mid->orig_block_len = split_mid->block_len;
2397 split_mid->ram_bytes = split_mid->len;
2398 split_mid->flags = flags;
2399 split_mid->compress_type = em->compress_type;
2400 split_mid->generation = em->generation;
2401 add_extent_mapping(em_tree, split_mid, 1);
2405 split_post->start = em->start + em->len - post;
2406 split_post->len = post;
2407 split_post->orig_start = split_post->start;
2408 split_post->block_start = em->block_start + em->len - post;
2409 split_post->block_len = split_post->len;
2410 split_post->orig_block_len = split_post->block_len;
2411 split_post->ram_bytes = split_post->len;
2412 split_post->flags = flags;
2413 split_post->compress_type = em->compress_type;
2414 split_post->generation = em->generation;
2415 add_extent_mapping(em_tree, split_post, 1);
2419 free_extent_map(em);
2420 /* Once for the tree */
2421 free_extent_map(em);
2424 write_unlock(&em_tree->lock);
2425 unlock_extent(&inode->io_tree, start, start + len - 1);
2427 free_extent_map(split_pre);
2428 free_extent_map(split_mid);
2429 free_extent_map(split_post);
2434 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2435 struct bio *bio, loff_t file_offset)
2437 struct btrfs_ordered_extent *ordered;
2438 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2440 u64 len = bio->bi_iter.bi_size;
2441 u64 end = start + len;
2446 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2447 if (WARN_ON_ONCE(!ordered))
2448 return BLK_STS_IOERR;
2450 /* No need to split */
2451 if (ordered->disk_num_bytes == len)
2454 /* We cannot split once end_bio'd ordered extent */
2455 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2460 /* We cannot split a compressed ordered extent */
2461 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2466 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2467 /* bio must be in one ordered extent */
2468 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2473 /* Checksum list should be empty */
2474 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2479 file_len = ordered->num_bytes;
2480 pre = start - ordered->disk_bytenr;
2481 post = ordered_end - end;
2483 ret = btrfs_split_ordered_extent(ordered, pre, post);
2486 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2489 btrfs_put_ordered_extent(ordered);
2491 return errno_to_blk_status(ret);
2495 * extent_io.c submission hook. This does the right thing for csum calculation
2496 * on write, or reading the csums from the tree before a read.
2498 * Rules about async/sync submit,
2499 * a) read: sync submit
2501 * b) write without checksum: sync submit
2503 * c) write with checksum:
2504 * c-1) if bio is issued by fsync: sync submit
2505 * (sync_writers != 0)
2507 * c-2) if root is reloc root: sync submit
2508 * (only in case of buffered IO)
2510 * c-3) otherwise: async submit
2512 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2513 int mirror_num, unsigned long bio_flags)
2516 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2517 struct btrfs_root *root = BTRFS_I(inode)->root;
2518 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2519 blk_status_t ret = 0;
2521 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2523 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2524 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2526 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2527 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2529 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2530 struct page *page = bio_first_bvec_all(bio)->bv_page;
2531 loff_t file_offset = page_offset(page);
2533 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2538 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2539 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2543 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2545 * btrfs_submit_compressed_read will handle completing
2546 * the bio if there were any errors, so just return
2549 ret = btrfs_submit_compressed_read(inode, bio,
2555 * Lookup bio sums does extra checks around whether we
2556 * need to csum or not, which is why we ignore skip_sum
2559 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2564 } else if (async && !skip_sum) {
2565 /* csum items have already been cloned */
2566 if (btrfs_is_data_reloc_root(root))
2568 /* we're doing a write, do the async checksumming */
2569 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2570 0, btrfs_submit_bio_start);
2572 } else if (!skip_sum) {
2573 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
2579 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2583 bio->bi_status = ret;
2591 * given a list of ordered sums record them in the inode. This happens
2592 * at IO completion time based on sums calculated at bio submission time.
2594 static int add_pending_csums(struct btrfs_trans_handle *trans,
2595 struct list_head *list)
2597 struct btrfs_ordered_sum *sum;
2598 struct btrfs_root *csum_root = NULL;
2601 list_for_each_entry(sum, list, list) {
2602 trans->adding_csums = true;
2604 csum_root = btrfs_csum_root(trans->fs_info,
2606 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2607 trans->adding_csums = false;
2614 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2617 struct extent_state **cached_state)
2619 u64 search_start = start;
2620 const u64 end = start + len - 1;
2622 while (search_start < end) {
2623 const u64 search_len = end - search_start + 1;
2624 struct extent_map *em;
2628 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2632 if (em->block_start != EXTENT_MAP_HOLE)
2636 if (em->start < search_start)
2637 em_len -= search_start - em->start;
2638 if (em_len > search_len)
2639 em_len = search_len;
2641 ret = set_extent_bit(&inode->io_tree, search_start,
2642 search_start + em_len - 1,
2643 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2646 search_start = extent_map_end(em);
2647 free_extent_map(em);
2654 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2655 unsigned int extra_bits,
2656 struct extent_state **cached_state)
2658 WARN_ON(PAGE_ALIGNED(end));
2660 if (start >= i_size_read(&inode->vfs_inode) &&
2661 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2663 * There can't be any extents following eof in this case so just
2664 * set the delalloc new bit for the range directly.
2666 extra_bits |= EXTENT_DELALLOC_NEW;
2670 ret = btrfs_find_new_delalloc_bytes(inode, start,
2677 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2681 /* see btrfs_writepage_start_hook for details on why this is required */
2682 struct btrfs_writepage_fixup {
2684 struct inode *inode;
2685 struct btrfs_work work;
2688 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2690 struct btrfs_writepage_fixup *fixup;
2691 struct btrfs_ordered_extent *ordered;
2692 struct extent_state *cached_state = NULL;
2693 struct extent_changeset *data_reserved = NULL;
2695 struct btrfs_inode *inode;
2699 bool free_delalloc_space = true;
2701 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2703 inode = BTRFS_I(fixup->inode);
2704 page_start = page_offset(page);
2705 page_end = page_offset(page) + PAGE_SIZE - 1;
2708 * This is similar to page_mkwrite, we need to reserve the space before
2709 * we take the page lock.
2711 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2717 * Before we queued this fixup, we took a reference on the page.
2718 * page->mapping may go NULL, but it shouldn't be moved to a different
2721 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2723 * Unfortunately this is a little tricky, either
2725 * 1) We got here and our page had already been dealt with and
2726 * we reserved our space, thus ret == 0, so we need to just
2727 * drop our space reservation and bail. This can happen the
2728 * first time we come into the fixup worker, or could happen
2729 * while waiting for the ordered extent.
2730 * 2) Our page was already dealt with, but we happened to get an
2731 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2732 * this case we obviously don't have anything to release, but
2733 * because the page was already dealt with we don't want to
2734 * mark the page with an error, so make sure we're resetting
2735 * ret to 0. This is why we have this check _before_ the ret
2736 * check, because we do not want to have a surprise ENOSPC
2737 * when the page was already properly dealt with.
2740 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2741 btrfs_delalloc_release_space(inode, data_reserved,
2742 page_start, PAGE_SIZE,
2750 * We can't mess with the page state unless it is locked, so now that
2751 * it is locked bail if we failed to make our space reservation.
2756 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2758 /* already ordered? We're done */
2759 if (PageOrdered(page))
2762 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2764 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2767 btrfs_start_ordered_extent(ordered, 1);
2768 btrfs_put_ordered_extent(ordered);
2772 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2778 * Everything went as planned, we're now the owner of a dirty page with
2779 * delayed allocation bits set and space reserved for our COW
2782 * The page was dirty when we started, nothing should have cleaned it.
2784 BUG_ON(!PageDirty(page));
2785 free_delalloc_space = false;
2787 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2788 if (free_delalloc_space)
2789 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2791 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2796 * We hit ENOSPC or other errors. Update the mapping and page
2797 * to reflect the errors and clean the page.
2799 mapping_set_error(page->mapping, ret);
2800 end_extent_writepage(page, ret, page_start, page_end);
2801 clear_page_dirty_for_io(page);
2804 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2808 extent_changeset_free(data_reserved);
2810 * As a precaution, do a delayed iput in case it would be the last iput
2811 * that could need flushing space. Recursing back to fixup worker would
2814 btrfs_add_delayed_iput(&inode->vfs_inode);
2818 * There are a few paths in the higher layers of the kernel that directly
2819 * set the page dirty bit without asking the filesystem if it is a
2820 * good idea. This causes problems because we want to make sure COW
2821 * properly happens and the data=ordered rules are followed.
2823 * In our case any range that doesn't have the ORDERED bit set
2824 * hasn't been properly setup for IO. We kick off an async process
2825 * to fix it up. The async helper will wait for ordered extents, set
2826 * the delalloc bit and make it safe to write the page.
2828 int btrfs_writepage_cow_fixup(struct page *page)
2830 struct inode *inode = page->mapping->host;
2831 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2832 struct btrfs_writepage_fixup *fixup;
2834 /* This page has ordered extent covering it already */
2835 if (PageOrdered(page))
2839 * PageChecked is set below when we create a fixup worker for this page,
2840 * don't try to create another one if we're already PageChecked()
2842 * The extent_io writepage code will redirty the page if we send back
2845 if (PageChecked(page))
2848 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2853 * We are already holding a reference to this inode from
2854 * write_cache_pages. We need to hold it because the space reservation
2855 * takes place outside of the page lock, and we can't trust
2856 * page->mapping outside of the page lock.
2859 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2861 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2863 fixup->inode = inode;
2864 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2869 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2870 struct btrfs_inode *inode, u64 file_pos,
2871 struct btrfs_file_extent_item *stack_fi,
2872 const bool update_inode_bytes,
2873 u64 qgroup_reserved)
2875 struct btrfs_root *root = inode->root;
2876 const u64 sectorsize = root->fs_info->sectorsize;
2877 struct btrfs_path *path;
2878 struct extent_buffer *leaf;
2879 struct btrfs_key ins;
2880 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2881 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2882 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2883 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2884 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2885 struct btrfs_drop_extents_args drop_args = { 0 };
2888 path = btrfs_alloc_path();
2893 * we may be replacing one extent in the tree with another.
2894 * The new extent is pinned in the extent map, and we don't want
2895 * to drop it from the cache until it is completely in the btree.
2897 * So, tell btrfs_drop_extents to leave this extent in the cache.
2898 * the caller is expected to unpin it and allow it to be merged
2901 drop_args.path = path;
2902 drop_args.start = file_pos;
2903 drop_args.end = file_pos + num_bytes;
2904 drop_args.replace_extent = true;
2905 drop_args.extent_item_size = sizeof(*stack_fi);
2906 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2910 if (!drop_args.extent_inserted) {
2911 ins.objectid = btrfs_ino(inode);
2912 ins.offset = file_pos;
2913 ins.type = BTRFS_EXTENT_DATA_KEY;
2915 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2920 leaf = path->nodes[0];
2921 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2922 write_extent_buffer(leaf, stack_fi,
2923 btrfs_item_ptr_offset(leaf, path->slots[0]),
2924 sizeof(struct btrfs_file_extent_item));
2926 btrfs_mark_buffer_dirty(leaf);
2927 btrfs_release_path(path);
2930 * If we dropped an inline extent here, we know the range where it is
2931 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2932 * number of bytes only for that range containing the inline extent.
2933 * The remaining of the range will be processed when clearning the
2934 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2936 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2937 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2939 inline_size = drop_args.bytes_found - inline_size;
2940 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2941 drop_args.bytes_found -= inline_size;
2942 num_bytes -= sectorsize;
2945 if (update_inode_bytes)
2946 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2948 ins.objectid = disk_bytenr;
2949 ins.offset = disk_num_bytes;
2950 ins.type = BTRFS_EXTENT_ITEM_KEY;
2952 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2956 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2958 qgroup_reserved, &ins);
2960 btrfs_free_path(path);
2965 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2968 struct btrfs_block_group *cache;
2970 cache = btrfs_lookup_block_group(fs_info, start);
2973 spin_lock(&cache->lock);
2974 cache->delalloc_bytes -= len;
2975 spin_unlock(&cache->lock);
2977 btrfs_put_block_group(cache);
2980 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2981 struct btrfs_ordered_extent *oe)
2983 struct btrfs_file_extent_item stack_fi;
2984 bool update_inode_bytes;
2985 u64 num_bytes = oe->num_bytes;
2986 u64 ram_bytes = oe->ram_bytes;
2988 memset(&stack_fi, 0, sizeof(stack_fi));
2989 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2990 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2991 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2992 oe->disk_num_bytes);
2993 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2994 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2995 num_bytes = ram_bytes = oe->truncated_len;
2996 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2997 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
2998 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2999 /* Encryption and other encoding is reserved and all 0 */
3002 * For delalloc, when completing an ordered extent we update the inode's
3003 * bytes when clearing the range in the inode's io tree, so pass false
3004 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3005 * except if the ordered extent was truncated.
3007 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3008 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3009 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3011 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3012 oe->file_offset, &stack_fi,
3013 update_inode_bytes, oe->qgroup_rsv);
3017 * As ordered data IO finishes, this gets called so we can finish
3018 * an ordered extent if the range of bytes in the file it covers are
3021 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3023 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3024 struct btrfs_root *root = inode->root;
3025 struct btrfs_fs_info *fs_info = root->fs_info;
3026 struct btrfs_trans_handle *trans = NULL;
3027 struct extent_io_tree *io_tree = &inode->io_tree;
3028 struct extent_state *cached_state = NULL;
3030 int compress_type = 0;
3032 u64 logical_len = ordered_extent->num_bytes;
3033 bool freespace_inode;
3034 bool truncated = false;
3035 bool clear_reserved_extent = true;
3036 unsigned int clear_bits = EXTENT_DEFRAG;
3038 start = ordered_extent->file_offset;
3039 end = start + ordered_extent->num_bytes - 1;
3041 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3042 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3043 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3044 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3045 clear_bits |= EXTENT_DELALLOC_NEW;
3047 freespace_inode = btrfs_is_free_space_inode(inode);
3049 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3054 /* A valid bdev implies a write on a sequential zone */
3055 if (ordered_extent->bdev) {
3056 btrfs_rewrite_logical_zoned(ordered_extent);
3057 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3058 ordered_extent->disk_num_bytes);
3061 btrfs_free_io_failure_record(inode, start, end);
3063 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3065 logical_len = ordered_extent->truncated_len;
3066 /* Truncated the entire extent, don't bother adding */
3071 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3072 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3074 btrfs_inode_safe_disk_i_size_write(inode, 0);
3075 if (freespace_inode)
3076 trans = btrfs_join_transaction_spacecache(root);
3078 trans = btrfs_join_transaction(root);
3079 if (IS_ERR(trans)) {
3080 ret = PTR_ERR(trans);
3084 trans->block_rsv = &inode->block_rsv;
3085 ret = btrfs_update_inode_fallback(trans, root, inode);
3086 if (ret) /* -ENOMEM or corruption */
3087 btrfs_abort_transaction(trans, ret);
3091 clear_bits |= EXTENT_LOCKED;
3092 lock_extent_bits(io_tree, start, end, &cached_state);
3094 if (freespace_inode)
3095 trans = btrfs_join_transaction_spacecache(root);
3097 trans = btrfs_join_transaction(root);
3098 if (IS_ERR(trans)) {
3099 ret = PTR_ERR(trans);
3104 trans->block_rsv = &inode->block_rsv;
3106 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3107 compress_type = ordered_extent->compress_type;
3108 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3109 BUG_ON(compress_type);
3110 ret = btrfs_mark_extent_written(trans, inode,
3111 ordered_extent->file_offset,
3112 ordered_extent->file_offset +
3115 BUG_ON(root == fs_info->tree_root);
3116 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3118 clear_reserved_extent = false;
3119 btrfs_release_delalloc_bytes(fs_info,
3120 ordered_extent->disk_bytenr,
3121 ordered_extent->disk_num_bytes);
3124 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3125 ordered_extent->num_bytes, trans->transid);
3127 btrfs_abort_transaction(trans, ret);
3131 ret = add_pending_csums(trans, &ordered_extent->list);
3133 btrfs_abort_transaction(trans, ret);
3138 * If this is a new delalloc range, clear its new delalloc flag to
3139 * update the inode's number of bytes. This needs to be done first
3140 * before updating the inode item.
3142 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3143 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3144 clear_extent_bit(&inode->io_tree, start, end,
3145 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3146 0, 0, &cached_state);
3148 btrfs_inode_safe_disk_i_size_write(inode, 0);
3149 ret = btrfs_update_inode_fallback(trans, root, inode);
3150 if (ret) { /* -ENOMEM or corruption */
3151 btrfs_abort_transaction(trans, ret);
3156 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3157 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3161 btrfs_end_transaction(trans);
3163 if (ret || truncated) {
3164 u64 unwritten_start = start;
3167 * If we failed to finish this ordered extent for any reason we
3168 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3169 * extent, and mark the inode with the error if it wasn't
3170 * already set. Any error during writeback would have already
3171 * set the mapping error, so we need to set it if we're the ones
3172 * marking this ordered extent as failed.
3174 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3175 &ordered_extent->flags))
3176 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3179 unwritten_start += logical_len;
3180 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3182 /* Drop the cache for the part of the extent we didn't write. */
3183 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3186 * If the ordered extent had an IOERR or something else went
3187 * wrong we need to return the space for this ordered extent
3188 * back to the allocator. We only free the extent in the
3189 * truncated case if we didn't write out the extent at all.
3191 * If we made it past insert_reserved_file_extent before we
3192 * errored out then we don't need to do this as the accounting
3193 * has already been done.
3195 if ((ret || !logical_len) &&
3196 clear_reserved_extent &&
3197 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3198 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3200 * Discard the range before returning it back to the
3203 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3204 btrfs_discard_extent(fs_info,
3205 ordered_extent->disk_bytenr,
3206 ordered_extent->disk_num_bytes,
3208 btrfs_free_reserved_extent(fs_info,
3209 ordered_extent->disk_bytenr,
3210 ordered_extent->disk_num_bytes, 1);
3215 * This needs to be done to make sure anybody waiting knows we are done
3216 * updating everything for this ordered extent.
3218 btrfs_remove_ordered_extent(inode, ordered_extent);
3221 btrfs_put_ordered_extent(ordered_extent);
3222 /* once for the tree */
3223 btrfs_put_ordered_extent(ordered_extent);
3228 static void finish_ordered_fn(struct btrfs_work *work)
3230 struct btrfs_ordered_extent *ordered_extent;
3231 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3232 btrfs_finish_ordered_io(ordered_extent);
3235 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3236 struct page *page, u64 start,
3237 u64 end, bool uptodate)
3239 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3241 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start,
3242 finish_ordered_fn, uptodate);
3246 * check_data_csum - verify checksum of one sector of uncompressed data
3248 * @io_bio: btrfs_io_bio which contains the csum
3249 * @bio_offset: offset to the beginning of the bio (in bytes)
3250 * @page: page where is the data to be verified
3251 * @pgoff: offset inside the page
3252 * @start: logical offset in the file
3254 * The length of such check is always one sector size.
3256 static int check_data_csum(struct inode *inode, struct btrfs_bio *bbio,
3257 u32 bio_offset, struct page *page, u32 pgoff,
3260 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3261 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3263 u32 len = fs_info->sectorsize;
3264 const u32 csum_size = fs_info->csum_size;
3265 unsigned int offset_sectors;
3267 u8 csum[BTRFS_CSUM_SIZE];
3269 ASSERT(pgoff + len <= PAGE_SIZE);
3271 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3272 csum_expected = ((u8 *)bbio->csum) + offset_sectors * csum_size;
3274 kaddr = kmap_atomic(page);
3275 shash->tfm = fs_info->csum_shash;
3277 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3278 kunmap_atomic(kaddr);
3280 if (memcmp(csum, csum_expected, csum_size))
3285 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3288 btrfs_dev_stat_inc_and_print(bbio->device,
3289 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3290 memzero_page(page, pgoff, len);
3295 * When reads are done, we need to check csums to verify the data is correct.
3296 * if there's a match, we allow the bio to finish. If not, the code in
3297 * extent_io.c will try to find good copies for us.
3299 * @bio_offset: offset to the beginning of the bio (in bytes)
3300 * @start: file offset of the range start
3301 * @end: file offset of the range end (inclusive)
3303 * Return a bitmap where bit set means a csum mismatch, and bit not set means
3306 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
3307 u32 bio_offset, struct page *page,
3310 struct inode *inode = page->mapping->host;
3311 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3312 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3313 struct btrfs_root *root = BTRFS_I(inode)->root;
3314 const u32 sectorsize = root->fs_info->sectorsize;
3316 unsigned int result = 0;
3318 if (btrfs_page_test_checked(fs_info, page, start, end + 1 - start)) {
3319 btrfs_page_clear_checked(fs_info, page, start, end + 1 - start);
3324 * This only happens for NODATASUM or compressed read.
3325 * Normally this should be covered by above check for compressed read
3326 * or the next check for NODATASUM. Just do a quicker exit here.
3328 if (bbio->csum == NULL)
3331 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3334 if (unlikely(test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)))
3337 ASSERT(page_offset(page) <= start &&
3338 end <= page_offset(page) + PAGE_SIZE - 1);
3339 for (pg_off = offset_in_page(start);
3340 pg_off < offset_in_page(end);
3341 pg_off += sectorsize, bio_offset += sectorsize) {
3342 u64 file_offset = pg_off + page_offset(page);
3345 if (btrfs_is_data_reloc_root(root) &&
3346 test_range_bit(io_tree, file_offset,
3347 file_offset + sectorsize - 1,
3348 EXTENT_NODATASUM, 1, NULL)) {
3349 /* Skip the range without csum for data reloc inode */
3350 clear_extent_bits(io_tree, file_offset,
3351 file_offset + sectorsize - 1,
3355 ret = check_data_csum(inode, bbio, bio_offset, page, pg_off,
3356 page_offset(page) + pg_off);
3358 const int nr_bit = (pg_off - offset_in_page(start)) >>
3359 root->fs_info->sectorsize_bits;
3361 result |= (1U << nr_bit);
3368 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3370 * @inode: The inode we want to perform iput on
3372 * This function uses the generic vfs_inode::i_count to track whether we should
3373 * just decrement it (in case it's > 1) or if this is the last iput then link
3374 * the inode to the delayed iput machinery. Delayed iputs are processed at
3375 * transaction commit time/superblock commit/cleaner kthread.
3377 void btrfs_add_delayed_iput(struct inode *inode)
3379 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3380 struct btrfs_inode *binode = BTRFS_I(inode);
3382 if (atomic_add_unless(&inode->i_count, -1, 1))
3385 atomic_inc(&fs_info->nr_delayed_iputs);
3386 spin_lock(&fs_info->delayed_iput_lock);
3387 ASSERT(list_empty(&binode->delayed_iput));
3388 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3389 spin_unlock(&fs_info->delayed_iput_lock);
3390 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3391 wake_up_process(fs_info->cleaner_kthread);
3394 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3395 struct btrfs_inode *inode)
3397 list_del_init(&inode->delayed_iput);
3398 spin_unlock(&fs_info->delayed_iput_lock);
3399 iput(&inode->vfs_inode);
3400 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3401 wake_up(&fs_info->delayed_iputs_wait);
3402 spin_lock(&fs_info->delayed_iput_lock);
3405 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3406 struct btrfs_inode *inode)
3408 if (!list_empty(&inode->delayed_iput)) {
3409 spin_lock(&fs_info->delayed_iput_lock);
3410 if (!list_empty(&inode->delayed_iput))
3411 run_delayed_iput_locked(fs_info, inode);
3412 spin_unlock(&fs_info->delayed_iput_lock);
3416 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3419 spin_lock(&fs_info->delayed_iput_lock);
3420 while (!list_empty(&fs_info->delayed_iputs)) {
3421 struct btrfs_inode *inode;
3423 inode = list_first_entry(&fs_info->delayed_iputs,
3424 struct btrfs_inode, delayed_iput);
3425 run_delayed_iput_locked(fs_info, inode);
3426 cond_resched_lock(&fs_info->delayed_iput_lock);
3428 spin_unlock(&fs_info->delayed_iput_lock);
3432 * Wait for flushing all delayed iputs
3434 * @fs_info: the filesystem
3436 * This will wait on any delayed iputs that are currently running with KILLABLE
3437 * set. Once they are all done running we will return, unless we are killed in
3438 * which case we return EINTR. This helps in user operations like fallocate etc
3439 * that might get blocked on the iputs.
3441 * Return EINTR if we were killed, 0 if nothing's pending
3443 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3445 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3446 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3453 * This creates an orphan entry for the given inode in case something goes wrong
3454 * in the middle of an unlink.
3456 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3457 struct btrfs_inode *inode)
3461 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3462 if (ret && ret != -EEXIST) {
3463 btrfs_abort_transaction(trans, ret);
3471 * We have done the delete so we can go ahead and remove the orphan item for
3472 * this particular inode.
3474 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3475 struct btrfs_inode *inode)
3477 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3481 * this cleans up any orphans that may be left on the list from the last use
3484 int btrfs_orphan_cleanup(struct btrfs_root *root)
3486 struct btrfs_fs_info *fs_info = root->fs_info;
3487 struct btrfs_path *path;
3488 struct extent_buffer *leaf;
3489 struct btrfs_key key, found_key;
3490 struct btrfs_trans_handle *trans;
3491 struct inode *inode;
3492 u64 last_objectid = 0;
3493 int ret = 0, nr_unlink = 0;
3495 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3498 path = btrfs_alloc_path();
3503 path->reada = READA_BACK;
3505 key.objectid = BTRFS_ORPHAN_OBJECTID;
3506 key.type = BTRFS_ORPHAN_ITEM_KEY;
3507 key.offset = (u64)-1;
3510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3515 * if ret == 0 means we found what we were searching for, which
3516 * is weird, but possible, so only screw with path if we didn't
3517 * find the key and see if we have stuff that matches
3521 if (path->slots[0] == 0)
3526 /* pull out the item */
3527 leaf = path->nodes[0];
3528 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3530 /* make sure the item matches what we want */
3531 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3533 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3536 /* release the path since we're done with it */
3537 btrfs_release_path(path);
3540 * this is where we are basically btrfs_lookup, without the
3541 * crossing root thing. we store the inode number in the
3542 * offset of the orphan item.
3545 if (found_key.offset == last_objectid) {
3547 "Error removing orphan entry, stopping orphan cleanup");
3552 last_objectid = found_key.offset;
3554 found_key.objectid = found_key.offset;
3555 found_key.type = BTRFS_INODE_ITEM_KEY;
3556 found_key.offset = 0;
3557 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3558 ret = PTR_ERR_OR_ZERO(inode);
3559 if (ret && ret != -ENOENT)
3562 if (ret == -ENOENT && root == fs_info->tree_root) {
3563 struct btrfs_root *dead_root;
3564 int is_dead_root = 0;
3567 * This is an orphan in the tree root. Currently these
3568 * could come from 2 sources:
3569 * a) a root (snapshot/subvolume) deletion in progress
3570 * b) a free space cache inode
3571 * We need to distinguish those two, as the orphan item
3572 * for a root must not get deleted before the deletion
3573 * of the snapshot/subvolume's tree completes.
3575 * btrfs_find_orphan_roots() ran before us, which has
3576 * found all deleted roots and loaded them into
3577 * fs_info->fs_roots_radix. So here we can find if an
3578 * orphan item corresponds to a deleted root by looking
3579 * up the root from that radix tree.
3582 spin_lock(&fs_info->fs_roots_radix_lock);
3583 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3584 (unsigned long)found_key.objectid);
3585 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3587 spin_unlock(&fs_info->fs_roots_radix_lock);
3590 /* prevent this orphan from being found again */
3591 key.offset = found_key.objectid - 1;
3598 * If we have an inode with links, there are a couple of
3601 * 1. We were halfway through creating fsverity metadata for the
3602 * file. In that case, the orphan item represents incomplete
3603 * fsverity metadata which must be cleaned up with
3604 * btrfs_drop_verity_items and deleting the orphan item.
3606 * 2. Old kernels (before v3.12) used to create an
3607 * orphan item for truncate indicating that there were possibly
3608 * extent items past i_size that needed to be deleted. In v3.12,
3609 * truncate was changed to update i_size in sync with the extent
3610 * items, but the (useless) orphan item was still created. Since
3611 * v4.18, we don't create the orphan item for truncate at all.
3613 * So, this item could mean that we need to do a truncate, but
3614 * only if this filesystem was last used on a pre-v3.12 kernel
3615 * and was not cleanly unmounted. The odds of that are quite
3616 * slim, and it's a pain to do the truncate now, so just delete
3619 * It's also possible that this orphan item was supposed to be
3620 * deleted but wasn't. The inode number may have been reused,
3621 * but either way, we can delete the orphan item.
3623 if (ret == -ENOENT || inode->i_nlink) {
3625 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3630 trans = btrfs_start_transaction(root, 1);
3631 if (IS_ERR(trans)) {
3632 ret = PTR_ERR(trans);
3635 btrfs_debug(fs_info, "auto deleting %Lu",
3636 found_key.objectid);
3637 ret = btrfs_del_orphan_item(trans, root,
3638 found_key.objectid);
3639 btrfs_end_transaction(trans);
3647 /* this will do delete_inode and everything for us */
3650 /* release the path since we're done with it */
3651 btrfs_release_path(path);
3653 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3654 trans = btrfs_join_transaction(root);
3656 btrfs_end_transaction(trans);
3660 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3664 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3665 btrfs_free_path(path);
3670 * very simple check to peek ahead in the leaf looking for xattrs. If we
3671 * don't find any xattrs, we know there can't be any acls.
3673 * slot is the slot the inode is in, objectid is the objectid of the inode
3675 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3676 int slot, u64 objectid,
3677 int *first_xattr_slot)
3679 u32 nritems = btrfs_header_nritems(leaf);
3680 struct btrfs_key found_key;
3681 static u64 xattr_access = 0;
3682 static u64 xattr_default = 0;
3685 if (!xattr_access) {
3686 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3687 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3688 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3689 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3693 *first_xattr_slot = -1;
3694 while (slot < nritems) {
3695 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3697 /* we found a different objectid, there must not be acls */
3698 if (found_key.objectid != objectid)
3701 /* we found an xattr, assume we've got an acl */
3702 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3703 if (*first_xattr_slot == -1)
3704 *first_xattr_slot = slot;
3705 if (found_key.offset == xattr_access ||
3706 found_key.offset == xattr_default)
3711 * we found a key greater than an xattr key, there can't
3712 * be any acls later on
3714 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3721 * it goes inode, inode backrefs, xattrs, extents,
3722 * so if there are a ton of hard links to an inode there can
3723 * be a lot of backrefs. Don't waste time searching too hard,
3724 * this is just an optimization
3729 /* we hit the end of the leaf before we found an xattr or
3730 * something larger than an xattr. We have to assume the inode
3733 if (*first_xattr_slot == -1)
3734 *first_xattr_slot = slot;
3739 * read an inode from the btree into the in-memory inode
3741 static int btrfs_read_locked_inode(struct inode *inode,
3742 struct btrfs_path *in_path)
3744 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3745 struct btrfs_path *path = in_path;
3746 struct extent_buffer *leaf;
3747 struct btrfs_inode_item *inode_item;
3748 struct btrfs_root *root = BTRFS_I(inode)->root;
3749 struct btrfs_key location;
3754 bool filled = false;
3755 int first_xattr_slot;
3757 ret = btrfs_fill_inode(inode, &rdev);
3762 path = btrfs_alloc_path();
3767 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3769 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3771 if (path != in_path)
3772 btrfs_free_path(path);
3776 leaf = path->nodes[0];
3781 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3782 struct btrfs_inode_item);
3783 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3784 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3785 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3786 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3787 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3788 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3789 round_up(i_size_read(inode), fs_info->sectorsize));
3791 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3792 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3794 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3795 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3797 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3798 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3800 BTRFS_I(inode)->i_otime.tv_sec =
3801 btrfs_timespec_sec(leaf, &inode_item->otime);
3802 BTRFS_I(inode)->i_otime.tv_nsec =
3803 btrfs_timespec_nsec(leaf, &inode_item->otime);
3805 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3806 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3807 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3809 inode_set_iversion_queried(inode,
3810 btrfs_inode_sequence(leaf, inode_item));
3811 inode->i_generation = BTRFS_I(inode)->generation;
3813 rdev = btrfs_inode_rdev(leaf, inode_item);
3815 BTRFS_I(inode)->index_cnt = (u64)-1;
3816 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3817 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3821 * If we were modified in the current generation and evicted from memory
3822 * and then re-read we need to do a full sync since we don't have any
3823 * idea about which extents were modified before we were evicted from
3826 * This is required for both inode re-read from disk and delayed inode
3827 * in delayed_nodes_tree.
3829 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3830 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3831 &BTRFS_I(inode)->runtime_flags);
3834 * We don't persist the id of the transaction where an unlink operation
3835 * against the inode was last made. So here we assume the inode might
3836 * have been evicted, and therefore the exact value of last_unlink_trans
3837 * lost, and set it to last_trans to avoid metadata inconsistencies
3838 * between the inode and its parent if the inode is fsync'ed and the log
3839 * replayed. For example, in the scenario:
3842 * ln mydir/foo mydir/bar
3845 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3846 * xfs_io -c fsync mydir/foo
3848 * mount fs, triggers fsync log replay
3850 * We must make sure that when we fsync our inode foo we also log its
3851 * parent inode, otherwise after log replay the parent still has the
3852 * dentry with the "bar" name but our inode foo has a link count of 1
3853 * and doesn't have an inode ref with the name "bar" anymore.
3855 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3856 * but it guarantees correctness at the expense of occasional full
3857 * transaction commits on fsync if our inode is a directory, or if our
3858 * inode is not a directory, logging its parent unnecessarily.
3860 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3863 * Same logic as for last_unlink_trans. We don't persist the generation
3864 * of the last transaction where this inode was used for a reflink
3865 * operation, so after eviction and reloading the inode we must be
3866 * pessimistic and assume the last transaction that modified the inode.
3868 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3871 if (inode->i_nlink != 1 ||
3872 path->slots[0] >= btrfs_header_nritems(leaf))
3875 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3876 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3879 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3880 if (location.type == BTRFS_INODE_REF_KEY) {
3881 struct btrfs_inode_ref *ref;
3883 ref = (struct btrfs_inode_ref *)ptr;
3884 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3885 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3886 struct btrfs_inode_extref *extref;
3888 extref = (struct btrfs_inode_extref *)ptr;
3889 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3894 * try to precache a NULL acl entry for files that don't have
3895 * any xattrs or acls
3897 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3898 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3899 if (first_xattr_slot != -1) {
3900 path->slots[0] = first_xattr_slot;
3901 ret = btrfs_load_inode_props(inode, path);
3904 "error loading props for ino %llu (root %llu): %d",
3905 btrfs_ino(BTRFS_I(inode)),
3906 root->root_key.objectid, ret);
3908 if (path != in_path)
3909 btrfs_free_path(path);
3912 cache_no_acl(inode);
3914 switch (inode->i_mode & S_IFMT) {
3916 inode->i_mapping->a_ops = &btrfs_aops;
3917 inode->i_fop = &btrfs_file_operations;
3918 inode->i_op = &btrfs_file_inode_operations;
3921 inode->i_fop = &btrfs_dir_file_operations;
3922 inode->i_op = &btrfs_dir_inode_operations;
3925 inode->i_op = &btrfs_symlink_inode_operations;
3926 inode_nohighmem(inode);
3927 inode->i_mapping->a_ops = &btrfs_aops;
3930 inode->i_op = &btrfs_special_inode_operations;
3931 init_special_inode(inode, inode->i_mode, rdev);
3935 btrfs_sync_inode_flags_to_i_flags(inode);
3940 * given a leaf and an inode, copy the inode fields into the leaf
3942 static void fill_inode_item(struct btrfs_trans_handle *trans,
3943 struct extent_buffer *leaf,
3944 struct btrfs_inode_item *item,
3945 struct inode *inode)
3947 struct btrfs_map_token token;
3950 btrfs_init_map_token(&token, leaf);
3952 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3953 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3954 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3955 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3956 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3958 btrfs_set_token_timespec_sec(&token, &item->atime,
3959 inode->i_atime.tv_sec);
3960 btrfs_set_token_timespec_nsec(&token, &item->atime,
3961 inode->i_atime.tv_nsec);
3963 btrfs_set_token_timespec_sec(&token, &item->mtime,
3964 inode->i_mtime.tv_sec);
3965 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3966 inode->i_mtime.tv_nsec);
3968 btrfs_set_token_timespec_sec(&token, &item->ctime,
3969 inode->i_ctime.tv_sec);
3970 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3971 inode->i_ctime.tv_nsec);
3973 btrfs_set_token_timespec_sec(&token, &item->otime,
3974 BTRFS_I(inode)->i_otime.tv_sec);
3975 btrfs_set_token_timespec_nsec(&token, &item->otime,
3976 BTRFS_I(inode)->i_otime.tv_nsec);
3978 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3979 btrfs_set_token_inode_generation(&token, item,
3980 BTRFS_I(inode)->generation);
3981 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3982 btrfs_set_token_inode_transid(&token, item, trans->transid);
3983 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3984 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3985 BTRFS_I(inode)->ro_flags);
3986 btrfs_set_token_inode_flags(&token, item, flags);
3987 btrfs_set_token_inode_block_group(&token, item, 0);
3991 * copy everything in the in-memory inode into the btree.
3993 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3994 struct btrfs_root *root,
3995 struct btrfs_inode *inode)
3997 struct btrfs_inode_item *inode_item;
3998 struct btrfs_path *path;
3999 struct extent_buffer *leaf;
4002 path = btrfs_alloc_path();
4006 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4013 leaf = path->nodes[0];
4014 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4015 struct btrfs_inode_item);
4017 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4018 btrfs_mark_buffer_dirty(leaf);
4019 btrfs_set_inode_last_trans(trans, inode);
4022 btrfs_free_path(path);
4027 * copy everything in the in-memory inode into the btree.
4029 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4030 struct btrfs_root *root,
4031 struct btrfs_inode *inode)
4033 struct btrfs_fs_info *fs_info = root->fs_info;
4037 * If the inode is a free space inode, we can deadlock during commit
4038 * if we put it into the delayed code.
4040 * The data relocation inode should also be directly updated
4043 if (!btrfs_is_free_space_inode(inode)
4044 && !btrfs_is_data_reloc_root(root)
4045 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4046 btrfs_update_root_times(trans, root);
4048 ret = btrfs_delayed_update_inode(trans, root, inode);
4050 btrfs_set_inode_last_trans(trans, inode);
4054 return btrfs_update_inode_item(trans, root, inode);
4057 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4058 struct btrfs_root *root, struct btrfs_inode *inode)
4062 ret = btrfs_update_inode(trans, root, inode);
4064 return btrfs_update_inode_item(trans, root, inode);
4069 * unlink helper that gets used here in inode.c and in the tree logging
4070 * recovery code. It remove a link in a directory with a given name, and
4071 * also drops the back refs in the inode to the directory
4073 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4074 struct btrfs_inode *dir,
4075 struct btrfs_inode *inode,
4076 const char *name, int name_len,
4077 struct btrfs_rename_ctx *rename_ctx)
4079 struct btrfs_root *root = dir->root;
4080 struct btrfs_fs_info *fs_info = root->fs_info;
4081 struct btrfs_path *path;
4083 struct btrfs_dir_item *di;
4085 u64 ino = btrfs_ino(inode);
4086 u64 dir_ino = btrfs_ino(dir);
4088 path = btrfs_alloc_path();
4094 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4095 name, name_len, -1);
4096 if (IS_ERR_OR_NULL(di)) {
4097 ret = di ? PTR_ERR(di) : -ENOENT;
4100 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4103 btrfs_release_path(path);
4106 * If we don't have dir index, we have to get it by looking up
4107 * the inode ref, since we get the inode ref, remove it directly,
4108 * it is unnecessary to do delayed deletion.
4110 * But if we have dir index, needn't search inode ref to get it.
4111 * Since the inode ref is close to the inode item, it is better
4112 * that we delay to delete it, and just do this deletion when
4113 * we update the inode item.
4115 if (inode->dir_index) {
4116 ret = btrfs_delayed_delete_inode_ref(inode);
4118 index = inode->dir_index;
4123 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4127 "failed to delete reference to %.*s, inode %llu parent %llu",
4128 name_len, name, ino, dir_ino);
4129 btrfs_abort_transaction(trans, ret);
4134 rename_ctx->index = index;
4136 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4138 btrfs_abort_transaction(trans, ret);
4143 * If we are in a rename context, we don't need to update anything in the
4144 * log. That will be done later during the rename by btrfs_log_new_name().
4145 * Besides that, doing it here would only cause extra unncessary btree
4146 * operations on the log tree, increasing latency for applications.
4149 btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4151 btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4156 * If we have a pending delayed iput we could end up with the final iput
4157 * being run in btrfs-cleaner context. If we have enough of these built
4158 * up we can end up burning a lot of time in btrfs-cleaner without any
4159 * way to throttle the unlinks. Since we're currently holding a ref on
4160 * the inode we can run the delayed iput here without any issues as the
4161 * final iput won't be done until after we drop the ref we're currently
4164 btrfs_run_delayed_iput(fs_info, inode);
4166 btrfs_free_path(path);
4170 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4171 inode_inc_iversion(&inode->vfs_inode);
4172 inode_inc_iversion(&dir->vfs_inode);
4173 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4174 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4175 ret = btrfs_update_inode(trans, root, dir);
4180 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4181 struct btrfs_inode *dir, struct btrfs_inode *inode,
4182 const char *name, int name_len)
4185 ret = __btrfs_unlink_inode(trans, dir, inode, name, name_len, NULL);
4187 drop_nlink(&inode->vfs_inode);
4188 ret = btrfs_update_inode(trans, inode->root, inode);
4194 * helper to start transaction for unlink and rmdir.
4196 * unlink and rmdir are special in btrfs, they do not always free space, so
4197 * if we cannot make our reservations the normal way try and see if there is
4198 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4199 * allow the unlink to occur.
4201 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4203 struct btrfs_root *root = BTRFS_I(dir)->root;
4206 * 1 for the possible orphan item
4207 * 1 for the dir item
4208 * 1 for the dir index
4209 * 1 for the inode ref
4211 * 1 for the parent inode
4213 return btrfs_start_transaction_fallback_global_rsv(root, 6);
4216 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4218 struct btrfs_trans_handle *trans;
4219 struct inode *inode = d_inode(dentry);
4222 trans = __unlink_start_trans(dir);
4224 return PTR_ERR(trans);
4226 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4229 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
4230 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4231 dentry->d_name.len);
4235 if (inode->i_nlink == 0) {
4236 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4242 btrfs_end_transaction(trans);
4243 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4247 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4248 struct inode *dir, struct dentry *dentry)
4250 struct btrfs_root *root = BTRFS_I(dir)->root;
4251 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4252 struct btrfs_path *path;
4253 struct extent_buffer *leaf;
4254 struct btrfs_dir_item *di;
4255 struct btrfs_key key;
4256 const char *name = dentry->d_name.name;
4257 int name_len = dentry->d_name.len;
4261 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4263 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4264 objectid = inode->root->root_key.objectid;
4265 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4266 objectid = inode->location.objectid;
4272 path = btrfs_alloc_path();
4276 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4277 name, name_len, -1);
4278 if (IS_ERR_OR_NULL(di)) {
4279 ret = di ? PTR_ERR(di) : -ENOENT;
4283 leaf = path->nodes[0];
4284 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4285 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4286 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4288 btrfs_abort_transaction(trans, ret);
4291 btrfs_release_path(path);
4294 * This is a placeholder inode for a subvolume we didn't have a
4295 * reference to at the time of the snapshot creation. In the meantime
4296 * we could have renamed the real subvol link into our snapshot, so
4297 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4298 * Instead simply lookup the dir_index_item for this entry so we can
4299 * remove it. Otherwise we know we have a ref to the root and we can
4300 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4302 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4303 di = btrfs_search_dir_index_item(root, path, dir_ino,
4305 if (IS_ERR_OR_NULL(di)) {
4310 btrfs_abort_transaction(trans, ret);
4314 leaf = path->nodes[0];
4315 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4317 btrfs_release_path(path);
4319 ret = btrfs_del_root_ref(trans, objectid,
4320 root->root_key.objectid, dir_ino,
4321 &index, name, name_len);
4323 btrfs_abort_transaction(trans, ret);
4328 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4330 btrfs_abort_transaction(trans, ret);
4334 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4335 inode_inc_iversion(dir);
4336 dir->i_mtime = dir->i_ctime = current_time(dir);
4337 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4339 btrfs_abort_transaction(trans, ret);
4341 btrfs_free_path(path);
4346 * Helper to check if the subvolume references other subvolumes or if it's
4349 static noinline int may_destroy_subvol(struct btrfs_root *root)
4351 struct btrfs_fs_info *fs_info = root->fs_info;
4352 struct btrfs_path *path;
4353 struct btrfs_dir_item *di;
4354 struct btrfs_key key;
4358 path = btrfs_alloc_path();
4362 /* Make sure this root isn't set as the default subvol */
4363 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4364 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4365 dir_id, "default", 7, 0);
4366 if (di && !IS_ERR(di)) {
4367 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4368 if (key.objectid == root->root_key.objectid) {
4371 "deleting default subvolume %llu is not allowed",
4375 btrfs_release_path(path);
4378 key.objectid = root->root_key.objectid;
4379 key.type = BTRFS_ROOT_REF_KEY;
4380 key.offset = (u64)-1;
4382 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4388 if (path->slots[0] > 0) {
4390 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4391 if (key.objectid == root->root_key.objectid &&
4392 key.type == BTRFS_ROOT_REF_KEY)
4396 btrfs_free_path(path);
4400 /* Delete all dentries for inodes belonging to the root */
4401 static void btrfs_prune_dentries(struct btrfs_root *root)
4403 struct btrfs_fs_info *fs_info = root->fs_info;
4404 struct rb_node *node;
4405 struct rb_node *prev;
4406 struct btrfs_inode *entry;
4407 struct inode *inode;
4410 if (!BTRFS_FS_ERROR(fs_info))
4411 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4413 spin_lock(&root->inode_lock);
4415 node = root->inode_tree.rb_node;
4419 entry = rb_entry(node, struct btrfs_inode, rb_node);
4421 if (objectid < btrfs_ino(entry))
4422 node = node->rb_left;
4423 else if (objectid > btrfs_ino(entry))
4424 node = node->rb_right;
4430 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4431 if (objectid <= btrfs_ino(entry)) {
4435 prev = rb_next(prev);
4439 entry = rb_entry(node, struct btrfs_inode, rb_node);
4440 objectid = btrfs_ino(entry) + 1;
4441 inode = igrab(&entry->vfs_inode);
4443 spin_unlock(&root->inode_lock);
4444 if (atomic_read(&inode->i_count) > 1)
4445 d_prune_aliases(inode);
4447 * btrfs_drop_inode will have it removed from the inode
4448 * cache when its usage count hits zero.
4452 spin_lock(&root->inode_lock);
4456 if (cond_resched_lock(&root->inode_lock))
4459 node = rb_next(node);
4461 spin_unlock(&root->inode_lock);
4464 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4466 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4467 struct btrfs_root *root = BTRFS_I(dir)->root;
4468 struct inode *inode = d_inode(dentry);
4469 struct btrfs_root *dest = BTRFS_I(inode)->root;
4470 struct btrfs_trans_handle *trans;
4471 struct btrfs_block_rsv block_rsv;
4476 * Don't allow to delete a subvolume with send in progress. This is
4477 * inside the inode lock so the error handling that has to drop the bit
4478 * again is not run concurrently.
4480 spin_lock(&dest->root_item_lock);
4481 if (dest->send_in_progress) {
4482 spin_unlock(&dest->root_item_lock);
4484 "attempt to delete subvolume %llu during send",
4485 dest->root_key.objectid);
4488 if (atomic_read(&dest->nr_swapfiles)) {
4489 spin_unlock(&dest->root_item_lock);
4491 "attempt to delete subvolume %llu with active swapfile",
4492 root->root_key.objectid);
4495 root_flags = btrfs_root_flags(&dest->root_item);
4496 btrfs_set_root_flags(&dest->root_item,
4497 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4498 spin_unlock(&dest->root_item_lock);
4500 down_write(&fs_info->subvol_sem);
4502 ret = may_destroy_subvol(dest);
4506 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4508 * One for dir inode,
4509 * two for dir entries,
4510 * two for root ref/backref.
4512 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4516 trans = btrfs_start_transaction(root, 0);
4517 if (IS_ERR(trans)) {
4518 ret = PTR_ERR(trans);
4521 trans->block_rsv = &block_rsv;
4522 trans->bytes_reserved = block_rsv.size;
4524 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4526 ret = btrfs_unlink_subvol(trans, dir, dentry);
4528 btrfs_abort_transaction(trans, ret);
4532 ret = btrfs_record_root_in_trans(trans, dest);
4534 btrfs_abort_transaction(trans, ret);
4538 memset(&dest->root_item.drop_progress, 0,
4539 sizeof(dest->root_item.drop_progress));
4540 btrfs_set_root_drop_level(&dest->root_item, 0);
4541 btrfs_set_root_refs(&dest->root_item, 0);
4543 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4544 ret = btrfs_insert_orphan_item(trans,
4546 dest->root_key.objectid);
4548 btrfs_abort_transaction(trans, ret);
4553 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4554 BTRFS_UUID_KEY_SUBVOL,
4555 dest->root_key.objectid);
4556 if (ret && ret != -ENOENT) {
4557 btrfs_abort_transaction(trans, ret);
4560 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4561 ret = btrfs_uuid_tree_remove(trans,
4562 dest->root_item.received_uuid,
4563 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4564 dest->root_key.objectid);
4565 if (ret && ret != -ENOENT) {
4566 btrfs_abort_transaction(trans, ret);
4571 free_anon_bdev(dest->anon_dev);
4574 trans->block_rsv = NULL;
4575 trans->bytes_reserved = 0;
4576 ret = btrfs_end_transaction(trans);
4577 inode->i_flags |= S_DEAD;
4579 btrfs_subvolume_release_metadata(root, &block_rsv);
4581 up_write(&fs_info->subvol_sem);
4583 spin_lock(&dest->root_item_lock);
4584 root_flags = btrfs_root_flags(&dest->root_item);
4585 btrfs_set_root_flags(&dest->root_item,
4586 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4587 spin_unlock(&dest->root_item_lock);
4589 d_invalidate(dentry);
4590 btrfs_prune_dentries(dest);
4591 ASSERT(dest->send_in_progress == 0);
4597 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4599 struct inode *inode = d_inode(dentry);
4600 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4602 struct btrfs_trans_handle *trans;
4603 u64 last_unlink_trans;
4605 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4607 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4608 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4610 "extent tree v2 doesn't support snapshot deletion yet");
4613 return btrfs_delete_subvolume(dir, dentry);
4616 trans = __unlink_start_trans(dir);
4618 return PTR_ERR(trans);
4620 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4621 err = btrfs_unlink_subvol(trans, dir, dentry);
4625 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4629 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4631 /* now the directory is empty */
4632 err = btrfs_unlink_inode(trans, BTRFS_I(dir),
4633 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4634 dentry->d_name.len);
4636 btrfs_i_size_write(BTRFS_I(inode), 0);
4638 * Propagate the last_unlink_trans value of the deleted dir to
4639 * its parent directory. This is to prevent an unrecoverable
4640 * log tree in the case we do something like this:
4642 * 2) create snapshot under dir foo
4643 * 3) delete the snapshot
4646 * 6) fsync foo or some file inside foo
4648 if (last_unlink_trans >= trans->transid)
4649 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4652 btrfs_end_transaction(trans);
4653 btrfs_btree_balance_dirty(fs_info);
4659 * btrfs_truncate_block - read, zero a chunk and write a block
4660 * @inode - inode that we're zeroing
4661 * @from - the offset to start zeroing
4662 * @len - the length to zero, 0 to zero the entire range respective to the
4664 * @front - zero up to the offset instead of from the offset on
4666 * This will find the block for the "from" offset and cow the block and zero the
4667 * part we want to zero. This is used with truncate and hole punching.
4669 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4672 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4673 struct address_space *mapping = inode->vfs_inode.i_mapping;
4674 struct extent_io_tree *io_tree = &inode->io_tree;
4675 struct btrfs_ordered_extent *ordered;
4676 struct extent_state *cached_state = NULL;
4677 struct extent_changeset *data_reserved = NULL;
4678 bool only_release_metadata = false;
4679 u32 blocksize = fs_info->sectorsize;
4680 pgoff_t index = from >> PAGE_SHIFT;
4681 unsigned offset = from & (blocksize - 1);
4683 gfp_t mask = btrfs_alloc_write_mask(mapping);
4684 size_t write_bytes = blocksize;
4689 if (IS_ALIGNED(offset, blocksize) &&
4690 (!len || IS_ALIGNED(len, blocksize)))
4693 block_start = round_down(from, blocksize);
4694 block_end = block_start + blocksize - 1;
4696 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4699 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4700 /* For nocow case, no need to reserve data space */
4701 only_release_metadata = true;
4706 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4708 if (!only_release_metadata)
4709 btrfs_free_reserved_data_space(inode, data_reserved,
4710 block_start, blocksize);
4714 page = find_or_create_page(mapping, index, mask);
4716 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4718 btrfs_delalloc_release_extents(inode, blocksize);
4722 ret = set_page_extent_mapped(page);
4726 if (!PageUptodate(page)) {
4727 ret = btrfs_readpage(NULL, page);
4729 if (page->mapping != mapping) {
4734 if (!PageUptodate(page)) {
4739 wait_on_page_writeback(page);
4741 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4743 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4745 unlock_extent_cached(io_tree, block_start, block_end,
4749 btrfs_start_ordered_extent(ordered, 1);
4750 btrfs_put_ordered_extent(ordered);
4754 clear_extent_bit(&inode->io_tree, block_start, block_end,
4755 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4756 0, 0, &cached_state);
4758 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4761 unlock_extent_cached(io_tree, block_start, block_end,
4766 if (offset != blocksize) {
4768 len = blocksize - offset;
4770 memzero_page(page, (block_start - page_offset(page)),
4773 memzero_page(page, (block_start - page_offset(page)) + offset,
4775 flush_dcache_page(page);
4777 btrfs_page_clear_checked(fs_info, page, block_start,
4778 block_end + 1 - block_start);
4779 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4780 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4782 if (only_release_metadata)
4783 set_extent_bit(&inode->io_tree, block_start, block_end,
4784 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4788 if (only_release_metadata)
4789 btrfs_delalloc_release_metadata(inode, blocksize, true);
4791 btrfs_delalloc_release_space(inode, data_reserved,
4792 block_start, blocksize, true);
4794 btrfs_delalloc_release_extents(inode, blocksize);
4798 if (only_release_metadata)
4799 btrfs_check_nocow_unlock(inode);
4800 extent_changeset_free(data_reserved);
4804 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4805 u64 offset, u64 len)
4807 struct btrfs_fs_info *fs_info = root->fs_info;
4808 struct btrfs_trans_handle *trans;
4809 struct btrfs_drop_extents_args drop_args = { 0 };
4813 * If NO_HOLES is enabled, we don't need to do anything.
4814 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4815 * or btrfs_update_inode() will be called, which guarantee that the next
4816 * fsync will know this inode was changed and needs to be logged.
4818 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4822 * 1 - for the one we're dropping
4823 * 1 - for the one we're adding
4824 * 1 - for updating the inode.
4826 trans = btrfs_start_transaction(root, 3);
4828 return PTR_ERR(trans);
4830 drop_args.start = offset;
4831 drop_args.end = offset + len;
4832 drop_args.drop_cache = true;
4834 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4836 btrfs_abort_transaction(trans, ret);
4837 btrfs_end_transaction(trans);
4841 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4842 offset, 0, 0, len, 0, len, 0, 0, 0);
4844 btrfs_abort_transaction(trans, ret);
4846 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4847 btrfs_update_inode(trans, root, inode);
4849 btrfs_end_transaction(trans);
4854 * This function puts in dummy file extents for the area we're creating a hole
4855 * for. So if we are truncating this file to a larger size we need to insert
4856 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4857 * the range between oldsize and size
4859 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4861 struct btrfs_root *root = inode->root;
4862 struct btrfs_fs_info *fs_info = root->fs_info;
4863 struct extent_io_tree *io_tree = &inode->io_tree;
4864 struct extent_map *em = NULL;
4865 struct extent_state *cached_state = NULL;
4866 struct extent_map_tree *em_tree = &inode->extent_tree;
4867 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4868 u64 block_end = ALIGN(size, fs_info->sectorsize);
4875 * If our size started in the middle of a block we need to zero out the
4876 * rest of the block before we expand the i_size, otherwise we could
4877 * expose stale data.
4879 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4883 if (size <= hole_start)
4886 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4888 cur_offset = hole_start;
4890 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4891 block_end - cur_offset);
4897 last_byte = min(extent_map_end(em), block_end);
4898 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4899 hole_size = last_byte - cur_offset;
4901 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4902 struct extent_map *hole_em;
4904 err = maybe_insert_hole(root, inode, cur_offset,
4909 err = btrfs_inode_set_file_extent_range(inode,
4910 cur_offset, hole_size);
4914 btrfs_drop_extent_cache(inode, cur_offset,
4915 cur_offset + hole_size - 1, 0);
4916 hole_em = alloc_extent_map();
4918 btrfs_set_inode_full_sync(inode);
4921 hole_em->start = cur_offset;
4922 hole_em->len = hole_size;
4923 hole_em->orig_start = cur_offset;
4925 hole_em->block_start = EXTENT_MAP_HOLE;
4926 hole_em->block_len = 0;
4927 hole_em->orig_block_len = 0;
4928 hole_em->ram_bytes = hole_size;
4929 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4930 hole_em->generation = fs_info->generation;
4933 write_lock(&em_tree->lock);
4934 err = add_extent_mapping(em_tree, hole_em, 1);
4935 write_unlock(&em_tree->lock);
4938 btrfs_drop_extent_cache(inode, cur_offset,
4942 free_extent_map(hole_em);
4944 err = btrfs_inode_set_file_extent_range(inode,
4945 cur_offset, hole_size);
4950 free_extent_map(em);
4952 cur_offset = last_byte;
4953 if (cur_offset >= block_end)
4956 free_extent_map(em);
4957 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4961 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4963 struct btrfs_root *root = BTRFS_I(inode)->root;
4964 struct btrfs_trans_handle *trans;
4965 loff_t oldsize = i_size_read(inode);
4966 loff_t newsize = attr->ia_size;
4967 int mask = attr->ia_valid;
4971 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4972 * special case where we need to update the times despite not having
4973 * these flags set. For all other operations the VFS set these flags
4974 * explicitly if it wants a timestamp update.
4976 if (newsize != oldsize) {
4977 inode_inc_iversion(inode);
4978 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4979 inode->i_ctime = inode->i_mtime =
4980 current_time(inode);
4983 if (newsize > oldsize) {
4985 * Don't do an expanding truncate while snapshotting is ongoing.
4986 * This is to ensure the snapshot captures a fully consistent
4987 * state of this file - if the snapshot captures this expanding
4988 * truncation, it must capture all writes that happened before
4991 btrfs_drew_write_lock(&root->snapshot_lock);
4992 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4994 btrfs_drew_write_unlock(&root->snapshot_lock);
4998 trans = btrfs_start_transaction(root, 1);
4999 if (IS_ERR(trans)) {
5000 btrfs_drew_write_unlock(&root->snapshot_lock);
5001 return PTR_ERR(trans);
5004 i_size_write(inode, newsize);
5005 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5006 pagecache_isize_extended(inode, oldsize, newsize);
5007 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5008 btrfs_drew_write_unlock(&root->snapshot_lock);
5009 btrfs_end_transaction(trans);
5011 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5013 if (btrfs_is_zoned(fs_info)) {
5014 ret = btrfs_wait_ordered_range(inode,
5015 ALIGN(newsize, fs_info->sectorsize),
5022 * We're truncating a file that used to have good data down to
5023 * zero. Make sure any new writes to the file get on disk
5027 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5028 &BTRFS_I(inode)->runtime_flags);
5030 truncate_setsize(inode, newsize);
5032 inode_dio_wait(inode);
5034 ret = btrfs_truncate(inode, newsize == oldsize);
5035 if (ret && inode->i_nlink) {
5039 * Truncate failed, so fix up the in-memory size. We
5040 * adjusted disk_i_size down as we removed extents, so
5041 * wait for disk_i_size to be stable and then update the
5042 * in-memory size to match.
5044 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5047 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5054 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5057 struct inode *inode = d_inode(dentry);
5058 struct btrfs_root *root = BTRFS_I(inode)->root;
5061 if (btrfs_root_readonly(root))
5064 err = setattr_prepare(mnt_userns, dentry, attr);
5068 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5069 err = btrfs_setsize(inode, attr);
5074 if (attr->ia_valid) {
5075 setattr_copy(mnt_userns, inode, attr);
5076 inode_inc_iversion(inode);
5077 err = btrfs_dirty_inode(inode);
5079 if (!err && attr->ia_valid & ATTR_MODE)
5080 err = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5087 * While truncating the inode pages during eviction, we get the VFS
5088 * calling btrfs_invalidate_folio() against each folio of the inode. This
5089 * is slow because the calls to btrfs_invalidate_folio() result in a
5090 * huge amount of calls to lock_extent_bits() and clear_extent_bit(),
5091 * which keep merging and splitting extent_state structures over and over,
5092 * wasting lots of time.
5094 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5095 * skip all those expensive operations on a per folio basis and do only
5096 * the ordered io finishing, while we release here the extent_map and
5097 * extent_state structures, without the excessive merging and splitting.
5099 static void evict_inode_truncate_pages(struct inode *inode)
5101 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5102 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5103 struct rb_node *node;
5105 ASSERT(inode->i_state & I_FREEING);
5106 truncate_inode_pages_final(&inode->i_data);
5108 write_lock(&map_tree->lock);
5109 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5110 struct extent_map *em;
5112 node = rb_first_cached(&map_tree->map);
5113 em = rb_entry(node, struct extent_map, rb_node);
5114 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5115 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5116 remove_extent_mapping(map_tree, em);
5117 free_extent_map(em);
5118 if (need_resched()) {
5119 write_unlock(&map_tree->lock);
5121 write_lock(&map_tree->lock);
5124 write_unlock(&map_tree->lock);
5127 * Keep looping until we have no more ranges in the io tree.
5128 * We can have ongoing bios started by readahead that have
5129 * their endio callback (extent_io.c:end_bio_extent_readpage)
5130 * still in progress (unlocked the pages in the bio but did not yet
5131 * unlocked the ranges in the io tree). Therefore this means some
5132 * ranges can still be locked and eviction started because before
5133 * submitting those bios, which are executed by a separate task (work
5134 * queue kthread), inode references (inode->i_count) were not taken
5135 * (which would be dropped in the end io callback of each bio).
5136 * Therefore here we effectively end up waiting for those bios and
5137 * anyone else holding locked ranges without having bumped the inode's
5138 * reference count - if we don't do it, when they access the inode's
5139 * io_tree to unlock a range it may be too late, leading to an
5140 * use-after-free issue.
5142 spin_lock(&io_tree->lock);
5143 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5144 struct extent_state *state;
5145 struct extent_state *cached_state = NULL;
5148 unsigned state_flags;
5150 node = rb_first(&io_tree->state);
5151 state = rb_entry(node, struct extent_state, rb_node);
5152 start = state->start;
5154 state_flags = state->state;
5155 spin_unlock(&io_tree->lock);
5157 lock_extent_bits(io_tree, start, end, &cached_state);
5160 * If still has DELALLOC flag, the extent didn't reach disk,
5161 * and its reserved space won't be freed by delayed_ref.
5162 * So we need to free its reserved space here.
5163 * (Refer to comment in btrfs_invalidate_folio, case 2)
5165 * Note, end is the bytenr of last byte, so we need + 1 here.
5167 if (state_flags & EXTENT_DELALLOC)
5168 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5171 clear_extent_bit(io_tree, start, end,
5172 EXTENT_LOCKED | EXTENT_DELALLOC |
5173 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5177 spin_lock(&io_tree->lock);
5179 spin_unlock(&io_tree->lock);
5182 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5183 struct btrfs_block_rsv *rsv)
5185 struct btrfs_fs_info *fs_info = root->fs_info;
5186 struct btrfs_trans_handle *trans;
5187 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5191 * Eviction should be taking place at some place safe because of our
5192 * delayed iputs. However the normal flushing code will run delayed
5193 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5195 * We reserve the delayed_refs_extra here again because we can't use
5196 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5197 * above. We reserve our extra bit here because we generate a ton of
5198 * delayed refs activity by truncating.
5200 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5201 * if we fail to make this reservation we can re-try without the
5202 * delayed_refs_extra so we can make some forward progress.
5204 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5205 BTRFS_RESERVE_FLUSH_EVICT);
5207 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5208 BTRFS_RESERVE_FLUSH_EVICT);
5211 "could not allocate space for delete; will truncate on mount");
5212 return ERR_PTR(-ENOSPC);
5214 delayed_refs_extra = 0;
5217 trans = btrfs_join_transaction(root);
5221 if (delayed_refs_extra) {
5222 trans->block_rsv = &fs_info->trans_block_rsv;
5223 trans->bytes_reserved = delayed_refs_extra;
5224 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5225 delayed_refs_extra, 1);
5230 void btrfs_evict_inode(struct inode *inode)
5232 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5233 struct btrfs_trans_handle *trans;
5234 struct btrfs_root *root = BTRFS_I(inode)->root;
5235 struct btrfs_block_rsv *rsv;
5238 trace_btrfs_inode_evict(inode);
5241 fsverity_cleanup_inode(inode);
5246 evict_inode_truncate_pages(inode);
5248 if (inode->i_nlink &&
5249 ((btrfs_root_refs(&root->root_item) != 0 &&
5250 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5251 btrfs_is_free_space_inode(BTRFS_I(inode))))
5254 if (is_bad_inode(inode))
5257 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5259 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5262 if (inode->i_nlink > 0) {
5263 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5264 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5269 * This makes sure the inode item in tree is uptodate and the space for
5270 * the inode update is released.
5272 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5277 * This drops any pending insert or delete operations we have for this
5278 * inode. We could have a delayed dir index deletion queued up, but
5279 * we're removing the inode completely so that'll be taken care of in
5282 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5284 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5287 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5290 btrfs_i_size_write(BTRFS_I(inode), 0);
5293 struct btrfs_truncate_control control = {
5294 .inode = BTRFS_I(inode),
5295 .ino = btrfs_ino(BTRFS_I(inode)),
5300 trans = evict_refill_and_join(root, rsv);
5304 trans->block_rsv = rsv;
5306 ret = btrfs_truncate_inode_items(trans, root, &control);
5307 trans->block_rsv = &fs_info->trans_block_rsv;
5308 btrfs_end_transaction(trans);
5309 btrfs_btree_balance_dirty(fs_info);
5310 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5317 * Errors here aren't a big deal, it just means we leave orphan items in
5318 * the tree. They will be cleaned up on the next mount. If the inode
5319 * number gets reused, cleanup deletes the orphan item without doing
5320 * anything, and unlink reuses the existing orphan item.
5322 * If it turns out that we are dropping too many of these, we might want
5323 * to add a mechanism for retrying these after a commit.
5325 trans = evict_refill_and_join(root, rsv);
5326 if (!IS_ERR(trans)) {
5327 trans->block_rsv = rsv;
5328 btrfs_orphan_del(trans, BTRFS_I(inode));
5329 trans->block_rsv = &fs_info->trans_block_rsv;
5330 btrfs_end_transaction(trans);
5334 btrfs_free_block_rsv(fs_info, rsv);
5337 * If we didn't successfully delete, the orphan item will still be in
5338 * the tree and we'll retry on the next mount. Again, we might also want
5339 * to retry these periodically in the future.
5341 btrfs_remove_delayed_node(BTRFS_I(inode));
5342 fsverity_cleanup_inode(inode);
5347 * Return the key found in the dir entry in the location pointer, fill @type
5348 * with BTRFS_FT_*, and return 0.
5350 * If no dir entries were found, returns -ENOENT.
5351 * If found a corrupted location in dir entry, returns -EUCLEAN.
5353 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5354 struct btrfs_key *location, u8 *type)
5356 const char *name = dentry->d_name.name;
5357 int namelen = dentry->d_name.len;
5358 struct btrfs_dir_item *di;
5359 struct btrfs_path *path;
5360 struct btrfs_root *root = BTRFS_I(dir)->root;
5363 path = btrfs_alloc_path();
5367 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5369 if (IS_ERR_OR_NULL(di)) {
5370 ret = di ? PTR_ERR(di) : -ENOENT;
5374 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5375 if (location->type != BTRFS_INODE_ITEM_KEY &&
5376 location->type != BTRFS_ROOT_ITEM_KEY) {
5378 btrfs_warn(root->fs_info,
5379 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5380 __func__, name, btrfs_ino(BTRFS_I(dir)),
5381 location->objectid, location->type, location->offset);
5384 *type = btrfs_dir_type(path->nodes[0], di);
5386 btrfs_free_path(path);
5391 * when we hit a tree root in a directory, the btrfs part of the inode
5392 * needs to be changed to reflect the root directory of the tree root. This
5393 * is kind of like crossing a mount point.
5395 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5397 struct dentry *dentry,
5398 struct btrfs_key *location,
5399 struct btrfs_root **sub_root)
5401 struct btrfs_path *path;
5402 struct btrfs_root *new_root;
5403 struct btrfs_root_ref *ref;
5404 struct extent_buffer *leaf;
5405 struct btrfs_key key;
5409 path = btrfs_alloc_path();
5416 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5417 key.type = BTRFS_ROOT_REF_KEY;
5418 key.offset = location->objectid;
5420 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5427 leaf = path->nodes[0];
5428 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5429 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5430 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5433 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5434 (unsigned long)(ref + 1),
5435 dentry->d_name.len);
5439 btrfs_release_path(path);
5441 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5442 if (IS_ERR(new_root)) {
5443 err = PTR_ERR(new_root);
5447 *sub_root = new_root;
5448 location->objectid = btrfs_root_dirid(&new_root->root_item);
5449 location->type = BTRFS_INODE_ITEM_KEY;
5450 location->offset = 0;
5453 btrfs_free_path(path);
5457 static void inode_tree_add(struct inode *inode)
5459 struct btrfs_root *root = BTRFS_I(inode)->root;
5460 struct btrfs_inode *entry;
5462 struct rb_node *parent;
5463 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5464 u64 ino = btrfs_ino(BTRFS_I(inode));
5466 if (inode_unhashed(inode))
5469 spin_lock(&root->inode_lock);
5470 p = &root->inode_tree.rb_node;
5473 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5475 if (ino < btrfs_ino(entry))
5476 p = &parent->rb_left;
5477 else if (ino > btrfs_ino(entry))
5478 p = &parent->rb_right;
5480 WARN_ON(!(entry->vfs_inode.i_state &
5481 (I_WILL_FREE | I_FREEING)));
5482 rb_replace_node(parent, new, &root->inode_tree);
5483 RB_CLEAR_NODE(parent);
5484 spin_unlock(&root->inode_lock);
5488 rb_link_node(new, parent, p);
5489 rb_insert_color(new, &root->inode_tree);
5490 spin_unlock(&root->inode_lock);
5493 static void inode_tree_del(struct btrfs_inode *inode)
5495 struct btrfs_root *root = inode->root;
5498 spin_lock(&root->inode_lock);
5499 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5500 rb_erase(&inode->rb_node, &root->inode_tree);
5501 RB_CLEAR_NODE(&inode->rb_node);
5502 empty = RB_EMPTY_ROOT(&root->inode_tree);
5504 spin_unlock(&root->inode_lock);
5506 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5507 spin_lock(&root->inode_lock);
5508 empty = RB_EMPTY_ROOT(&root->inode_tree);
5509 spin_unlock(&root->inode_lock);
5511 btrfs_add_dead_root(root);
5516 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5518 struct btrfs_iget_args *args = p;
5520 inode->i_ino = args->ino;
5521 BTRFS_I(inode)->location.objectid = args->ino;
5522 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5523 BTRFS_I(inode)->location.offset = 0;
5524 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5525 BUG_ON(args->root && !BTRFS_I(inode)->root);
5529 static int btrfs_find_actor(struct inode *inode, void *opaque)
5531 struct btrfs_iget_args *args = opaque;
5533 return args->ino == BTRFS_I(inode)->location.objectid &&
5534 args->root == BTRFS_I(inode)->root;
5537 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5538 struct btrfs_root *root)
5540 struct inode *inode;
5541 struct btrfs_iget_args args;
5542 unsigned long hashval = btrfs_inode_hash(ino, root);
5547 inode = iget5_locked(s, hashval, btrfs_find_actor,
5548 btrfs_init_locked_inode,
5554 * Get an inode object given its inode number and corresponding root.
5555 * Path can be preallocated to prevent recursing back to iget through
5556 * allocator. NULL is also valid but may require an additional allocation
5559 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5560 struct btrfs_root *root, struct btrfs_path *path)
5562 struct inode *inode;
5564 inode = btrfs_iget_locked(s, ino, root);
5566 return ERR_PTR(-ENOMEM);
5568 if (inode->i_state & I_NEW) {
5571 ret = btrfs_read_locked_inode(inode, path);
5573 inode_tree_add(inode);
5574 unlock_new_inode(inode);
5578 * ret > 0 can come from btrfs_search_slot called by
5579 * btrfs_read_locked_inode, this means the inode item
5584 inode = ERR_PTR(ret);
5591 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5593 return btrfs_iget_path(s, ino, root, NULL);
5596 static struct inode *new_simple_dir(struct super_block *s,
5597 struct btrfs_key *key,
5598 struct btrfs_root *root)
5600 struct inode *inode = new_inode(s);
5603 return ERR_PTR(-ENOMEM);
5605 BTRFS_I(inode)->root = btrfs_grab_root(root);
5606 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5607 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5609 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5611 * We only need lookup, the rest is read-only and there's no inode
5612 * associated with the dentry
5614 inode->i_op = &simple_dir_inode_operations;
5615 inode->i_opflags &= ~IOP_XATTR;
5616 inode->i_fop = &simple_dir_operations;
5617 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5618 inode->i_mtime = current_time(inode);
5619 inode->i_atime = inode->i_mtime;
5620 inode->i_ctime = inode->i_mtime;
5621 BTRFS_I(inode)->i_otime = inode->i_mtime;
5626 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5627 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5628 static_assert(BTRFS_FT_DIR == FT_DIR);
5629 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5630 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5631 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5632 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5633 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5635 static inline u8 btrfs_inode_type(struct inode *inode)
5637 return fs_umode_to_ftype(inode->i_mode);
5640 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5642 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5643 struct inode *inode;
5644 struct btrfs_root *root = BTRFS_I(dir)->root;
5645 struct btrfs_root *sub_root = root;
5646 struct btrfs_key location;
5650 if (dentry->d_name.len > BTRFS_NAME_LEN)
5651 return ERR_PTR(-ENAMETOOLONG);
5653 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5655 return ERR_PTR(ret);
5657 if (location.type == BTRFS_INODE_ITEM_KEY) {
5658 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5662 /* Do extra check against inode mode with di_type */
5663 if (btrfs_inode_type(inode) != di_type) {
5665 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5666 inode->i_mode, btrfs_inode_type(inode),
5669 return ERR_PTR(-EUCLEAN);
5674 ret = fixup_tree_root_location(fs_info, dir, dentry,
5675 &location, &sub_root);
5678 inode = ERR_PTR(ret);
5680 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5682 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5684 if (root != sub_root)
5685 btrfs_put_root(sub_root);
5687 if (!IS_ERR(inode) && root != sub_root) {
5688 down_read(&fs_info->cleanup_work_sem);
5689 if (!sb_rdonly(inode->i_sb))
5690 ret = btrfs_orphan_cleanup(sub_root);
5691 up_read(&fs_info->cleanup_work_sem);
5694 inode = ERR_PTR(ret);
5701 static int btrfs_dentry_delete(const struct dentry *dentry)
5703 struct btrfs_root *root;
5704 struct inode *inode = d_inode(dentry);
5706 if (!inode && !IS_ROOT(dentry))
5707 inode = d_inode(dentry->d_parent);
5710 root = BTRFS_I(inode)->root;
5711 if (btrfs_root_refs(&root->root_item) == 0)
5714 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5720 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5723 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5725 if (inode == ERR_PTR(-ENOENT))
5727 return d_splice_alias(inode, dentry);
5731 * All this infrastructure exists because dir_emit can fault, and we are holding
5732 * the tree lock when doing readdir. For now just allocate a buffer and copy
5733 * our information into that, and then dir_emit from the buffer. This is
5734 * similar to what NFS does, only we don't keep the buffer around in pagecache
5735 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5736 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5739 static int btrfs_opendir(struct inode *inode, struct file *file)
5741 struct btrfs_file_private *private;
5743 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5746 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5747 if (!private->filldir_buf) {
5751 file->private_data = private;
5762 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5765 struct dir_entry *entry = addr;
5766 char *name = (char *)(entry + 1);
5768 ctx->pos = get_unaligned(&entry->offset);
5769 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5770 get_unaligned(&entry->ino),
5771 get_unaligned(&entry->type)))
5773 addr += sizeof(struct dir_entry) +
5774 get_unaligned(&entry->name_len);
5780 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5782 struct inode *inode = file_inode(file);
5783 struct btrfs_root *root = BTRFS_I(inode)->root;
5784 struct btrfs_file_private *private = file->private_data;
5785 struct btrfs_dir_item *di;
5786 struct btrfs_key key;
5787 struct btrfs_key found_key;
5788 struct btrfs_path *path;
5790 struct list_head ins_list;
5791 struct list_head del_list;
5798 struct btrfs_key location;
5800 if (!dir_emit_dots(file, ctx))
5803 path = btrfs_alloc_path();
5807 addr = private->filldir_buf;
5808 path->reada = READA_FORWARD;
5810 INIT_LIST_HEAD(&ins_list);
5811 INIT_LIST_HEAD(&del_list);
5812 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5815 key.type = BTRFS_DIR_INDEX_KEY;
5816 key.offset = ctx->pos;
5817 key.objectid = btrfs_ino(BTRFS_I(inode));
5819 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5820 struct dir_entry *entry;
5821 struct extent_buffer *leaf = path->nodes[0];
5823 if (found_key.objectid != key.objectid)
5825 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5827 if (found_key.offset < ctx->pos)
5829 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5831 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5832 name_len = btrfs_dir_name_len(leaf, di);
5833 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5835 btrfs_release_path(path);
5836 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5839 addr = private->filldir_buf;
5846 put_unaligned(name_len, &entry->name_len);
5847 name_ptr = (char *)(entry + 1);
5848 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5850 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5852 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5853 put_unaligned(location.objectid, &entry->ino);
5854 put_unaligned(found_key.offset, &entry->offset);
5856 addr += sizeof(struct dir_entry) + name_len;
5857 total_len += sizeof(struct dir_entry) + name_len;
5859 /* Catch error encountered during iteration */
5863 btrfs_release_path(path);
5865 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5869 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5874 * Stop new entries from being returned after we return the last
5877 * New directory entries are assigned a strictly increasing
5878 * offset. This means that new entries created during readdir
5879 * are *guaranteed* to be seen in the future by that readdir.
5880 * This has broken buggy programs which operate on names as
5881 * they're returned by readdir. Until we re-use freed offsets
5882 * we have this hack to stop new entries from being returned
5883 * under the assumption that they'll never reach this huge
5886 * This is being careful not to overflow 32bit loff_t unless the
5887 * last entry requires it because doing so has broken 32bit apps
5890 if (ctx->pos >= INT_MAX)
5891 ctx->pos = LLONG_MAX;
5898 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5899 btrfs_free_path(path);
5904 * This is somewhat expensive, updating the tree every time the
5905 * inode changes. But, it is most likely to find the inode in cache.
5906 * FIXME, needs more benchmarking...there are no reasons other than performance
5907 * to keep or drop this code.
5909 static int btrfs_dirty_inode(struct inode *inode)
5911 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5912 struct btrfs_root *root = BTRFS_I(inode)->root;
5913 struct btrfs_trans_handle *trans;
5916 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5919 trans = btrfs_join_transaction(root);
5921 return PTR_ERR(trans);
5923 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5924 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
5925 /* whoops, lets try again with the full transaction */
5926 btrfs_end_transaction(trans);
5927 trans = btrfs_start_transaction(root, 1);
5929 return PTR_ERR(trans);
5931 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5933 btrfs_end_transaction(trans);
5934 if (BTRFS_I(inode)->delayed_node)
5935 btrfs_balance_delayed_items(fs_info);
5941 * This is a copy of file_update_time. We need this so we can return error on
5942 * ENOSPC for updating the inode in the case of file write and mmap writes.
5944 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5947 struct btrfs_root *root = BTRFS_I(inode)->root;
5948 bool dirty = flags & ~S_VERSION;
5950 if (btrfs_root_readonly(root))
5953 if (flags & S_VERSION)
5954 dirty |= inode_maybe_inc_iversion(inode, dirty);
5955 if (flags & S_CTIME)
5956 inode->i_ctime = *now;
5957 if (flags & S_MTIME)
5958 inode->i_mtime = *now;
5959 if (flags & S_ATIME)
5960 inode->i_atime = *now;
5961 return dirty ? btrfs_dirty_inode(inode) : 0;
5965 * find the highest existing sequence number in a directory
5966 * and then set the in-memory index_cnt variable to reflect
5967 * free sequence numbers
5969 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5971 struct btrfs_root *root = inode->root;
5972 struct btrfs_key key, found_key;
5973 struct btrfs_path *path;
5974 struct extent_buffer *leaf;
5977 key.objectid = btrfs_ino(inode);
5978 key.type = BTRFS_DIR_INDEX_KEY;
5979 key.offset = (u64)-1;
5981 path = btrfs_alloc_path();
5985 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5988 /* FIXME: we should be able to handle this */
5993 if (path->slots[0] == 0) {
5994 inode->index_cnt = BTRFS_DIR_START_INDEX;
6000 leaf = path->nodes[0];
6001 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6003 if (found_key.objectid != btrfs_ino(inode) ||
6004 found_key.type != BTRFS_DIR_INDEX_KEY) {
6005 inode->index_cnt = BTRFS_DIR_START_INDEX;
6009 inode->index_cnt = found_key.offset + 1;
6011 btrfs_free_path(path);
6016 * helper to find a free sequence number in a given directory. This current
6017 * code is very simple, later versions will do smarter things in the btree
6019 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6023 if (dir->index_cnt == (u64)-1) {
6024 ret = btrfs_inode_delayed_dir_index_count(dir);
6026 ret = btrfs_set_inode_index_count(dir);
6032 *index = dir->index_cnt;
6038 static int btrfs_insert_inode_locked(struct inode *inode)
6040 struct btrfs_iget_args args;
6042 args.ino = BTRFS_I(inode)->location.objectid;
6043 args.root = BTRFS_I(inode)->root;
6045 return insert_inode_locked4(inode,
6046 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6047 btrfs_find_actor, &args);
6050 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6051 unsigned int *trans_num_items)
6053 struct inode *dir = args->dir;
6054 struct inode *inode = args->inode;
6057 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6061 /* 1 to add inode item */
6062 *trans_num_items = 1;
6063 /* 1 to add compression property */
6064 if (BTRFS_I(dir)->prop_compress)
6065 (*trans_num_items)++;
6066 /* 1 to add default ACL xattr */
6067 if (args->default_acl)
6068 (*trans_num_items)++;
6069 /* 1 to add access ACL xattr */
6071 (*trans_num_items)++;
6072 #ifdef CONFIG_SECURITY
6073 /* 1 to add LSM xattr */
6074 if (dir->i_security)
6075 (*trans_num_items)++;
6078 /* 1 to add orphan item */
6079 (*trans_num_items)++;
6082 * 1 to add inode ref
6084 * 1 to add dir index
6085 * 1 to update parent inode item
6087 *trans_num_items += 4;
6092 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6094 posix_acl_release(args->acl);
6095 posix_acl_release(args->default_acl);
6099 * Inherit flags from the parent inode.
6101 * Currently only the compression flags and the cow flags are inherited.
6103 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6107 flags = BTRFS_I(dir)->flags;
6109 if (flags & BTRFS_INODE_NOCOMPRESS) {
6110 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6111 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6112 } else if (flags & BTRFS_INODE_COMPRESS) {
6113 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6114 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6117 if (flags & BTRFS_INODE_NODATACOW) {
6118 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6119 if (S_ISREG(inode->i_mode))
6120 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6123 btrfs_sync_inode_flags_to_i_flags(inode);
6126 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6127 struct btrfs_new_inode_args *args)
6129 struct inode *dir = args->dir;
6130 struct inode *inode = args->inode;
6131 const char *name = args->orphan ? NULL : args->dentry->d_name.name;
6132 int name_len = args->orphan ? 0 : args->dentry->d_name.len;
6133 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6134 struct btrfs_root *root;
6135 struct btrfs_inode_item *inode_item;
6136 struct btrfs_key *location;
6137 struct btrfs_path *path;
6139 struct btrfs_inode_ref *ref;
6140 struct btrfs_key key[2];
6142 struct btrfs_item_batch batch;
6146 path = btrfs_alloc_path();
6151 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6152 root = BTRFS_I(inode)->root;
6154 ret = btrfs_get_free_objectid(root, &objectid);
6157 inode->i_ino = objectid;
6161 * O_TMPFILE, set link count to 0, so that after this point, we
6162 * fill in an inode item with the correct link count.
6164 set_nlink(inode, 0);
6166 trace_btrfs_inode_request(dir);
6168 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6173 * index_cnt is ignored for everything but a dir,
6174 * btrfs_set_inode_index_count has an explanation for the magic
6177 BTRFS_I(inode)->index_cnt = 2;
6178 BTRFS_I(inode)->generation = trans->transid;
6179 inode->i_generation = BTRFS_I(inode)->generation;
6182 * Subvolumes don't inherit flags from their parent directory.
6183 * Originally this was probably by accident, but we probably can't
6184 * change it now without compatibility issues.
6187 btrfs_inherit_iflags(inode, dir);
6189 if (S_ISREG(inode->i_mode)) {
6190 if (btrfs_test_opt(fs_info, NODATASUM))
6191 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6192 if (btrfs_test_opt(fs_info, NODATACOW))
6193 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6194 BTRFS_INODE_NODATASUM;
6197 location = &BTRFS_I(inode)->location;
6198 location->objectid = objectid;
6199 location->offset = 0;
6200 location->type = BTRFS_INODE_ITEM_KEY;
6202 ret = btrfs_insert_inode_locked(inode);
6205 BTRFS_I(dir)->index_cnt--;
6210 * We could have gotten an inode number from somebody who was fsynced
6211 * and then removed in this same transaction, so let's just set full
6212 * sync since it will be a full sync anyway and this will blow away the
6213 * old info in the log.
6215 btrfs_set_inode_full_sync(BTRFS_I(inode));
6217 key[0].objectid = objectid;
6218 key[0].type = BTRFS_INODE_ITEM_KEY;
6221 sizes[0] = sizeof(struct btrfs_inode_item);
6223 if (!args->orphan) {
6225 * Start new inodes with an inode_ref. This is slightly more
6226 * efficient for small numbers of hard links since they will
6227 * be packed into one item. Extended refs will kick in if we
6228 * add more hard links than can fit in the ref item.
6230 key[1].objectid = objectid;
6231 key[1].type = BTRFS_INODE_REF_KEY;
6233 key[1].offset = objectid;
6234 sizes[1] = 2 + sizeof(*ref);
6236 key[1].offset = btrfs_ino(BTRFS_I(dir));
6237 sizes[1] = name_len + sizeof(*ref);
6241 batch.keys = &key[0];
6242 batch.data_sizes = &sizes[0];
6243 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6244 batch.nr = args->orphan ? 1 : 2;
6245 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6247 btrfs_abort_transaction(trans, ret);
6251 inode->i_mtime = current_time(inode);
6252 inode->i_atime = inode->i_mtime;
6253 inode->i_ctime = inode->i_mtime;
6254 BTRFS_I(inode)->i_otime = inode->i_mtime;
6257 * We're going to fill the inode item now, so at this point the inode
6258 * must be fully initialized.
6261 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6262 struct btrfs_inode_item);
6263 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6264 sizeof(*inode_item));
6265 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6267 if (!args->orphan) {
6268 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6269 struct btrfs_inode_ref);
6270 ptr = (unsigned long)(ref + 1);
6272 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6273 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6274 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6276 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6277 btrfs_set_inode_ref_index(path->nodes[0], ref,
6278 BTRFS_I(inode)->dir_index);
6279 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6283 btrfs_mark_buffer_dirty(path->nodes[0]);
6284 btrfs_release_path(path);
6287 struct inode *parent;
6290 * Subvolumes inherit properties from their parent subvolume,
6291 * not the directory they were created in.
6293 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6294 BTRFS_I(dir)->root);
6295 if (IS_ERR(parent)) {
6296 ret = PTR_ERR(parent);
6298 ret = btrfs_inode_inherit_props(trans, inode, parent);
6302 ret = btrfs_inode_inherit_props(trans, inode, dir);
6306 "error inheriting props for ino %llu (root %llu): %d",
6307 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6312 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6315 if (!args->subvol) {
6316 ret = btrfs_init_inode_security(trans, args);
6318 btrfs_abort_transaction(trans, ret);
6323 inode_tree_add(inode);
6325 trace_btrfs_inode_new(inode);
6326 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6328 btrfs_update_root_times(trans, root);
6331 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6333 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6334 name_len, 0, BTRFS_I(inode)->dir_index);
6337 btrfs_abort_transaction(trans, ret);
6346 * discard_new_inode() calls iput(), but the caller owns the reference
6350 discard_new_inode(inode);
6352 btrfs_free_path(path);
6357 * utility function to add 'inode' into 'parent_inode' with
6358 * a give name and a given sequence number.
6359 * if 'add_backref' is true, also insert a backref from the
6360 * inode to the parent directory.
6362 int btrfs_add_link(struct btrfs_trans_handle *trans,
6363 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6364 const char *name, int name_len, int add_backref, u64 index)
6367 struct btrfs_key key;
6368 struct btrfs_root *root = parent_inode->root;
6369 u64 ino = btrfs_ino(inode);
6370 u64 parent_ino = btrfs_ino(parent_inode);
6372 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6373 memcpy(&key, &inode->root->root_key, sizeof(key));
6376 key.type = BTRFS_INODE_ITEM_KEY;
6380 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6381 ret = btrfs_add_root_ref(trans, key.objectid,
6382 root->root_key.objectid, parent_ino,
6383 index, name, name_len);
6384 } else if (add_backref) {
6385 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6389 /* Nothing to clean up yet */
6393 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6394 btrfs_inode_type(&inode->vfs_inode), index);
6395 if (ret == -EEXIST || ret == -EOVERFLOW)
6398 btrfs_abort_transaction(trans, ret);
6402 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6404 inode_inc_iversion(&parent_inode->vfs_inode);
6406 * If we are replaying a log tree, we do not want to update the mtime
6407 * and ctime of the parent directory with the current time, since the
6408 * log replay procedure is responsible for setting them to their correct
6409 * values (the ones it had when the fsync was done).
6411 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6412 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6414 parent_inode->vfs_inode.i_mtime = now;
6415 parent_inode->vfs_inode.i_ctime = now;
6417 ret = btrfs_update_inode(trans, root, parent_inode);
6419 btrfs_abort_transaction(trans, ret);
6423 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6426 err = btrfs_del_root_ref(trans, key.objectid,
6427 root->root_key.objectid, parent_ino,
6428 &local_index, name, name_len);
6430 btrfs_abort_transaction(trans, err);
6431 } else if (add_backref) {
6435 err = btrfs_del_inode_ref(trans, root, name, name_len,
6436 ino, parent_ino, &local_index);
6438 btrfs_abort_transaction(trans, err);
6441 /* Return the original error code */
6445 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6446 struct inode *inode)
6448 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6449 struct btrfs_root *root = BTRFS_I(dir)->root;
6450 struct btrfs_new_inode_args new_inode_args = {
6455 unsigned int trans_num_items;
6456 struct btrfs_trans_handle *trans;
6459 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6463 trans = btrfs_start_transaction(root, trans_num_items);
6464 if (IS_ERR(trans)) {
6465 err = PTR_ERR(trans);
6466 goto out_new_inode_args;
6469 err = btrfs_create_new_inode(trans, &new_inode_args);
6471 d_instantiate_new(dentry, inode);
6473 btrfs_end_transaction(trans);
6474 btrfs_btree_balance_dirty(fs_info);
6476 btrfs_new_inode_args_destroy(&new_inode_args);
6483 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6484 struct dentry *dentry, umode_t mode, dev_t rdev)
6486 struct inode *inode;
6488 inode = new_inode(dir->i_sb);
6491 inode_init_owner(mnt_userns, inode, dir, mode);
6492 inode->i_op = &btrfs_special_inode_operations;
6493 init_special_inode(inode, inode->i_mode, rdev);
6494 return btrfs_create_common(dir, dentry, inode);
6497 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6498 struct dentry *dentry, umode_t mode, bool excl)
6500 struct inode *inode;
6502 inode = new_inode(dir->i_sb);
6505 inode_init_owner(mnt_userns, inode, dir, mode);
6506 inode->i_fop = &btrfs_file_operations;
6507 inode->i_op = &btrfs_file_inode_operations;
6508 inode->i_mapping->a_ops = &btrfs_aops;
6509 return btrfs_create_common(dir, dentry, inode);
6512 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6513 struct dentry *dentry)
6515 struct btrfs_trans_handle *trans = NULL;
6516 struct btrfs_root *root = BTRFS_I(dir)->root;
6517 struct inode *inode = d_inode(old_dentry);
6518 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6523 /* do not allow sys_link's with other subvols of the same device */
6524 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6527 if (inode->i_nlink >= BTRFS_LINK_MAX)
6530 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6535 * 2 items for inode and inode ref
6536 * 2 items for dir items
6537 * 1 item for parent inode
6538 * 1 item for orphan item deletion if O_TMPFILE
6540 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6541 if (IS_ERR(trans)) {
6542 err = PTR_ERR(trans);
6547 /* There are several dir indexes for this inode, clear the cache. */
6548 BTRFS_I(inode)->dir_index = 0ULL;
6550 inode_inc_iversion(inode);
6551 inode->i_ctime = current_time(inode);
6553 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6555 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6556 dentry->d_name.name, dentry->d_name.len, 1, index);
6561 struct dentry *parent = dentry->d_parent;
6563 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6566 if (inode->i_nlink == 1) {
6568 * If new hard link count is 1, it's a file created
6569 * with open(2) O_TMPFILE flag.
6571 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6575 d_instantiate(dentry, inode);
6576 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6581 btrfs_end_transaction(trans);
6583 inode_dec_link_count(inode);
6586 btrfs_btree_balance_dirty(fs_info);
6590 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6591 struct dentry *dentry, umode_t mode)
6593 struct inode *inode;
6595 inode = new_inode(dir->i_sb);
6598 inode_init_owner(mnt_userns, inode, dir, S_IFDIR | mode);
6599 inode->i_op = &btrfs_dir_inode_operations;
6600 inode->i_fop = &btrfs_dir_file_operations;
6601 return btrfs_create_common(dir, dentry, inode);
6604 static noinline int uncompress_inline(struct btrfs_path *path,
6606 size_t pg_offset, u64 extent_offset,
6607 struct btrfs_file_extent_item *item)
6610 struct extent_buffer *leaf = path->nodes[0];
6613 unsigned long inline_size;
6617 WARN_ON(pg_offset != 0);
6618 compress_type = btrfs_file_extent_compression(leaf, item);
6619 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6620 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6621 tmp = kmalloc(inline_size, GFP_NOFS);
6624 ptr = btrfs_file_extent_inline_start(item);
6626 read_extent_buffer(leaf, tmp, ptr, inline_size);
6628 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6629 ret = btrfs_decompress(compress_type, tmp, page,
6630 extent_offset, inline_size, max_size);
6633 * decompression code contains a memset to fill in any space between the end
6634 * of the uncompressed data and the end of max_size in case the decompressed
6635 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6636 * the end of an inline extent and the beginning of the next block, so we
6637 * cover that region here.
6640 if (max_size + pg_offset < PAGE_SIZE)
6641 memzero_page(page, pg_offset + max_size,
6642 PAGE_SIZE - max_size - pg_offset);
6648 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6649 * @inode: file to search in
6650 * @page: page to read extent data into if the extent is inline
6651 * @pg_offset: offset into @page to copy to
6652 * @start: file offset
6653 * @len: length of range starting at @start
6655 * This returns the first &struct extent_map which overlaps with the given
6656 * range, reading it from the B-tree and caching it if necessary. Note that
6657 * there may be more extents which overlap the given range after the returned
6660 * If @page is not NULL and the extent is inline, this also reads the extent
6661 * data directly into the page and marks the extent up to date in the io_tree.
6663 * Return: ERR_PTR on error, non-NULL extent_map on success.
6665 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6666 struct page *page, size_t pg_offset,
6669 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6671 u64 extent_start = 0;
6673 u64 objectid = btrfs_ino(inode);
6674 int extent_type = -1;
6675 struct btrfs_path *path = NULL;
6676 struct btrfs_root *root = inode->root;
6677 struct btrfs_file_extent_item *item;
6678 struct extent_buffer *leaf;
6679 struct btrfs_key found_key;
6680 struct extent_map *em = NULL;
6681 struct extent_map_tree *em_tree = &inode->extent_tree;
6682 struct extent_io_tree *io_tree = &inode->io_tree;
6684 read_lock(&em_tree->lock);
6685 em = lookup_extent_mapping(em_tree, start, len);
6686 read_unlock(&em_tree->lock);
6689 if (em->start > start || em->start + em->len <= start)
6690 free_extent_map(em);
6691 else if (em->block_start == EXTENT_MAP_INLINE && page)
6692 free_extent_map(em);
6696 em = alloc_extent_map();
6701 em->start = EXTENT_MAP_HOLE;
6702 em->orig_start = EXTENT_MAP_HOLE;
6704 em->block_len = (u64)-1;
6706 path = btrfs_alloc_path();
6712 /* Chances are we'll be called again, so go ahead and do readahead */
6713 path->reada = READA_FORWARD;
6716 * The same explanation in load_free_space_cache applies here as well,
6717 * we only read when we're loading the free space cache, and at that
6718 * point the commit_root has everything we need.
6720 if (btrfs_is_free_space_inode(inode)) {
6721 path->search_commit_root = 1;
6722 path->skip_locking = 1;
6725 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6728 } else if (ret > 0) {
6729 if (path->slots[0] == 0)
6735 leaf = path->nodes[0];
6736 item = btrfs_item_ptr(leaf, path->slots[0],
6737 struct btrfs_file_extent_item);
6738 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6739 if (found_key.objectid != objectid ||
6740 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6742 * If we backup past the first extent we want to move forward
6743 * and see if there is an extent in front of us, otherwise we'll
6744 * say there is a hole for our whole search range which can
6751 extent_type = btrfs_file_extent_type(leaf, item);
6752 extent_start = found_key.offset;
6753 extent_end = btrfs_file_extent_end(path);
6754 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6755 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6756 /* Only regular file could have regular/prealloc extent */
6757 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6760 "regular/prealloc extent found for non-regular inode %llu",
6764 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6766 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6767 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6772 if (start >= extent_end) {
6774 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6775 ret = btrfs_next_leaf(root, path);
6781 leaf = path->nodes[0];
6783 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6784 if (found_key.objectid != objectid ||
6785 found_key.type != BTRFS_EXTENT_DATA_KEY)
6787 if (start + len <= found_key.offset)
6789 if (start > found_key.offset)
6792 /* New extent overlaps with existing one */
6794 em->orig_start = start;
6795 em->len = found_key.offset - start;
6796 em->block_start = EXTENT_MAP_HOLE;
6800 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6802 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6803 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6805 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6809 size_t extent_offset;
6815 size = btrfs_file_extent_ram_bytes(leaf, item);
6816 extent_offset = page_offset(page) + pg_offset - extent_start;
6817 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6818 size - extent_offset);
6819 em->start = extent_start + extent_offset;
6820 em->len = ALIGN(copy_size, fs_info->sectorsize);
6821 em->orig_block_len = em->len;
6822 em->orig_start = em->start;
6823 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6825 if (!PageUptodate(page)) {
6826 if (btrfs_file_extent_compression(leaf, item) !=
6827 BTRFS_COMPRESS_NONE) {
6828 ret = uncompress_inline(path, page, pg_offset,
6829 extent_offset, item);
6833 map = kmap_local_page(page);
6834 read_extent_buffer(leaf, map + pg_offset, ptr,
6836 if (pg_offset + copy_size < PAGE_SIZE) {
6837 memset(map + pg_offset + copy_size, 0,
6838 PAGE_SIZE - pg_offset -
6843 flush_dcache_page(page);
6845 set_extent_uptodate(io_tree, em->start,
6846 extent_map_end(em) - 1, NULL, GFP_NOFS);
6851 em->orig_start = start;
6853 em->block_start = EXTENT_MAP_HOLE;
6856 btrfs_release_path(path);
6857 if (em->start > start || extent_map_end(em) <= start) {
6859 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6860 em->start, em->len, start, len);
6865 write_lock(&em_tree->lock);
6866 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6867 write_unlock(&em_tree->lock);
6869 btrfs_free_path(path);
6871 trace_btrfs_get_extent(root, inode, em);
6874 free_extent_map(em);
6875 return ERR_PTR(ret);
6880 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6883 struct extent_map *em;
6884 struct extent_map *hole_em = NULL;
6885 u64 delalloc_start = start;
6891 em = btrfs_get_extent(inode, NULL, 0, start, len);
6895 * If our em maps to:
6897 * - a pre-alloc extent,
6898 * there might actually be delalloc bytes behind it.
6900 if (em->block_start != EXTENT_MAP_HOLE &&
6901 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6906 /* check to see if we've wrapped (len == -1 or similar) */
6915 /* ok, we didn't find anything, lets look for delalloc */
6916 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6917 end, len, EXTENT_DELALLOC, 1);
6918 delalloc_end = delalloc_start + delalloc_len;
6919 if (delalloc_end < delalloc_start)
6920 delalloc_end = (u64)-1;
6923 * We didn't find anything useful, return the original results from
6926 if (delalloc_start > end || delalloc_end <= start) {
6933 * Adjust the delalloc_start to make sure it doesn't go backwards from
6934 * the start they passed in
6936 delalloc_start = max(start, delalloc_start);
6937 delalloc_len = delalloc_end - delalloc_start;
6939 if (delalloc_len > 0) {
6942 const u64 hole_end = extent_map_end(hole_em);
6944 em = alloc_extent_map();
6952 * When btrfs_get_extent can't find anything it returns one
6955 * Make sure what it found really fits our range, and adjust to
6956 * make sure it is based on the start from the caller
6958 if (hole_end <= start || hole_em->start > end) {
6959 free_extent_map(hole_em);
6962 hole_start = max(hole_em->start, start);
6963 hole_len = hole_end - hole_start;
6966 if (hole_em && delalloc_start > hole_start) {
6968 * Our hole starts before our delalloc, so we have to
6969 * return just the parts of the hole that go until the
6972 em->len = min(hole_len, delalloc_start - hole_start);
6973 em->start = hole_start;
6974 em->orig_start = hole_start;
6976 * Don't adjust block start at all, it is fixed at
6979 em->block_start = hole_em->block_start;
6980 em->block_len = hole_len;
6981 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6982 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6985 * Hole is out of passed range or it starts after
6988 em->start = delalloc_start;
6989 em->len = delalloc_len;
6990 em->orig_start = delalloc_start;
6991 em->block_start = EXTENT_MAP_DELALLOC;
6992 em->block_len = delalloc_len;
6999 free_extent_map(hole_em);
7001 free_extent_map(em);
7002 return ERR_PTR(err);
7007 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7010 const u64 orig_start,
7011 const u64 block_start,
7012 const u64 block_len,
7013 const u64 orig_block_len,
7014 const u64 ram_bytes,
7017 struct extent_map *em = NULL;
7020 if (type != BTRFS_ORDERED_NOCOW) {
7021 em = create_io_em(inode, start, len, orig_start, block_start,
7022 block_len, orig_block_len, ram_bytes,
7023 BTRFS_COMPRESS_NONE, /* compress_type */
7028 ret = btrfs_add_ordered_extent(inode, start, len, len, block_start,
7031 (1 << BTRFS_ORDERED_DIRECT),
7032 BTRFS_COMPRESS_NONE);
7035 free_extent_map(em);
7036 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7045 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7048 struct btrfs_root *root = inode->root;
7049 struct btrfs_fs_info *fs_info = root->fs_info;
7050 struct extent_map *em;
7051 struct btrfs_key ins;
7055 alloc_hint = get_extent_allocation_hint(inode, start, len);
7056 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7057 0, alloc_hint, &ins, 1, 1);
7059 return ERR_PTR(ret);
7061 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7062 ins.objectid, ins.offset, ins.offset,
7063 ins.offset, BTRFS_ORDERED_REGULAR);
7064 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7066 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7072 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7074 struct btrfs_block_group *block_group;
7075 bool readonly = false;
7077 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7078 if (!block_group || block_group->ro)
7081 btrfs_put_block_group(block_group);
7086 * Check if we can do nocow write into the range [@offset, @offset + @len)
7088 * @offset: File offset
7089 * @len: The length to write, will be updated to the nocow writeable
7091 * @orig_start: (optional) Return the original file offset of the file extent
7092 * @orig_len: (optional) Return the original on-disk length of the file extent
7093 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7094 * @strict: if true, omit optimizations that might force us into unnecessary
7095 * cow. e.g., don't trust generation number.
7098 * >0 and update @len if we can do nocow write
7099 * 0 if we can't do nocow write
7100 * <0 if error happened
7102 * NOTE: This only checks the file extents, caller is responsible to wait for
7103 * any ordered extents.
7105 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7106 u64 *orig_start, u64 *orig_block_len,
7107 u64 *ram_bytes, bool strict)
7109 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7110 struct btrfs_path *path;
7112 struct extent_buffer *leaf;
7113 struct btrfs_root *root = BTRFS_I(inode)->root;
7114 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7115 struct btrfs_file_extent_item *fi;
7116 struct btrfs_key key;
7123 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7125 path = btrfs_alloc_path();
7129 ret = btrfs_lookup_file_extent(NULL, root, path,
7130 btrfs_ino(BTRFS_I(inode)), offset, 0);
7134 slot = path->slots[0];
7137 /* can't find the item, must cow */
7144 leaf = path->nodes[0];
7145 btrfs_item_key_to_cpu(leaf, &key, slot);
7146 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7147 key.type != BTRFS_EXTENT_DATA_KEY) {
7148 /* not our file or wrong item type, must cow */
7152 if (key.offset > offset) {
7153 /* Wrong offset, must cow */
7157 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7158 found_type = btrfs_file_extent_type(leaf, fi);
7159 if (found_type != BTRFS_FILE_EXTENT_REG &&
7160 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7161 /* not a regular extent, must cow */
7165 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7168 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7169 if (extent_end <= offset)
7172 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7173 if (disk_bytenr == 0)
7176 if (btrfs_file_extent_compression(leaf, fi) ||
7177 btrfs_file_extent_encryption(leaf, fi) ||
7178 btrfs_file_extent_other_encoding(leaf, fi))
7182 * Do the same check as in btrfs_cross_ref_exist but without the
7183 * unnecessary search.
7186 (btrfs_file_extent_generation(leaf, fi) <=
7187 btrfs_root_last_snapshot(&root->root_item)))
7190 backref_offset = btrfs_file_extent_offset(leaf, fi);
7193 *orig_start = key.offset - backref_offset;
7194 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7195 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7198 btrfs_release_path(path);
7200 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7203 num_bytes = min(offset + *len, extent_end) - offset;
7204 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7207 range_end = round_up(offset + num_bytes,
7208 root->fs_info->sectorsize) - 1;
7209 ret = test_range_bit(io_tree, offset, range_end,
7210 EXTENT_DELALLOC, 0, NULL);
7218 * look for other files referencing this extent, if we
7219 * find any we must cow
7222 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7223 key.offset - backref_offset, disk_bytenr,
7231 * We don't need the path anymore, plus through the csum_exist_in_range()
7232 * call below we will end up allocating another path. So free the path
7233 * to avoid unnecessary extra memory usage.
7235 btrfs_free_path(path);
7239 * adjust disk_bytenr and num_bytes to cover just the bytes
7240 * in this extent we are about to write. If there
7241 * are any csums in that range we have to cow in order
7242 * to keep the csums correct
7244 disk_bytenr += backref_offset;
7245 disk_bytenr += offset - key.offset;
7246 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7249 * all of the above have passed, it is safe to overwrite this extent
7255 btrfs_free_path(path);
7259 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7260 struct extent_state **cached_state,
7261 unsigned int iomap_flags)
7263 const bool writing = (iomap_flags & IOMAP_WRITE);
7264 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7265 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7266 struct btrfs_ordered_extent *ordered;
7271 if (!try_lock_extent(io_tree, lockstart, lockend))
7274 lock_extent_bits(io_tree, lockstart, lockend, cached_state);
7277 * We're concerned with the entire range that we're going to be
7278 * doing DIO to, so we need to make sure there's no ordered
7279 * extents in this range.
7281 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7282 lockend - lockstart + 1);
7285 * We need to make sure there are no buffered pages in this
7286 * range either, we could have raced between the invalidate in
7287 * generic_file_direct_write and locking the extent. The
7288 * invalidate needs to happen so that reads after a write do not
7292 (!writing || !filemap_range_has_page(inode->i_mapping,
7293 lockstart, lockend)))
7296 unlock_extent_cached(io_tree, lockstart, lockend, cached_state);
7300 btrfs_put_ordered_extent(ordered);
7305 * If we are doing a DIO read and the ordered extent we
7306 * found is for a buffered write, we can not wait for it
7307 * to complete and retry, because if we do so we can
7308 * deadlock with concurrent buffered writes on page
7309 * locks. This happens only if our DIO read covers more
7310 * than one extent map, if at this point has already
7311 * created an ordered extent for a previous extent map
7312 * and locked its range in the inode's io tree, and a
7313 * concurrent write against that previous extent map's
7314 * range and this range started (we unlock the ranges
7315 * in the io tree only when the bios complete and
7316 * buffered writes always lock pages before attempting
7317 * to lock range in the io tree).
7320 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7321 btrfs_start_ordered_extent(ordered, 1);
7323 ret = nowait ? -EAGAIN : -ENOTBLK;
7324 btrfs_put_ordered_extent(ordered);
7327 * We could trigger writeback for this range (and wait
7328 * for it to complete) and then invalidate the pages for
7329 * this range (through invalidate_inode_pages2_range()),
7330 * but that can lead us to a deadlock with a concurrent
7331 * call to readahead (a buffered read or a defrag call
7332 * triggered a readahead) on a page lock due to an
7333 * ordered dio extent we created before but did not have
7334 * yet a corresponding bio submitted (whence it can not
7335 * complete), which makes readahead wait for that
7336 * ordered extent to complete while holding a lock on
7339 ret = nowait ? -EAGAIN : -ENOTBLK;
7351 /* The callers of this must take lock_extent() */
7352 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7353 u64 len, u64 orig_start, u64 block_start,
7354 u64 block_len, u64 orig_block_len,
7355 u64 ram_bytes, int compress_type,
7358 struct extent_map_tree *em_tree;
7359 struct extent_map *em;
7362 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7363 type == BTRFS_ORDERED_COMPRESSED ||
7364 type == BTRFS_ORDERED_NOCOW ||
7365 type == BTRFS_ORDERED_REGULAR);
7367 em_tree = &inode->extent_tree;
7368 em = alloc_extent_map();
7370 return ERR_PTR(-ENOMEM);
7373 em->orig_start = orig_start;
7375 em->block_len = block_len;
7376 em->block_start = block_start;
7377 em->orig_block_len = orig_block_len;
7378 em->ram_bytes = ram_bytes;
7379 em->generation = -1;
7380 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7381 if (type == BTRFS_ORDERED_PREALLOC) {
7382 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7383 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7384 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7385 em->compress_type = compress_type;
7389 btrfs_drop_extent_cache(inode, em->start,
7390 em->start + em->len - 1, 0);
7391 write_lock(&em_tree->lock);
7392 ret = add_extent_mapping(em_tree, em, 1);
7393 write_unlock(&em_tree->lock);
7395 * The caller has taken lock_extent(), who could race with us
7398 } while (ret == -EEXIST);
7401 free_extent_map(em);
7402 return ERR_PTR(ret);
7405 /* em got 2 refs now, callers needs to do free_extent_map once. */
7410 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7411 struct inode *inode,
7412 struct btrfs_dio_data *dio_data,
7414 unsigned int iomap_flags)
7416 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7417 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7418 struct extent_map *em = *map;
7420 u64 block_start, orig_start, orig_block_len, ram_bytes;
7421 bool can_nocow = false;
7422 bool space_reserved = false;
7427 * We don't allocate a new extent in the following cases
7429 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7431 * 2) The extent is marked as PREALLOC. We're good to go here and can
7432 * just use the extent.
7435 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7436 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7437 em->block_start != EXTENT_MAP_HOLE)) {
7438 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7439 type = BTRFS_ORDERED_PREALLOC;
7441 type = BTRFS_ORDERED_NOCOW;
7442 len = min(len, em->len - (start - em->start));
7443 block_start = em->block_start + (start - em->start);
7445 if (can_nocow_extent(inode, start, &len, &orig_start,
7446 &orig_block_len, &ram_bytes, false) == 1 &&
7447 btrfs_inc_nocow_writers(fs_info, block_start))
7453 struct extent_map *em2;
7455 /* We can NOCOW, so only need to reserve metadata space. */
7456 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7459 /* Our caller expects us to free the input extent map. */
7460 free_extent_map(em);
7462 btrfs_dec_nocow_writers(fs_info, block_start);
7463 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7467 space_reserved = true;
7469 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7470 orig_start, block_start,
7471 len, orig_block_len,
7473 btrfs_dec_nocow_writers(fs_info, block_start);
7474 if (type == BTRFS_ORDERED_PREALLOC) {
7475 free_extent_map(em);
7484 /* Our caller expects us to free the input extent map. */
7485 free_extent_map(em);
7491 /* We have to COW, so need to reserve metadata and data space. */
7492 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7493 &dio_data->data_reserved,
7497 space_reserved = true;
7499 em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7505 len = min(len, em->len - (start - em->start));
7507 btrfs_delalloc_release_space(BTRFS_I(inode),
7508 dio_data->data_reserved,
7509 start + len, prev_len - len,
7514 * We have created our ordered extent, so we can now release our reservation
7515 * for an outstanding extent.
7517 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7520 * Need to update the i_size under the extent lock so buffered
7521 * readers will get the updated i_size when we unlock.
7523 if (start + len > i_size_read(inode))
7524 i_size_write(inode, start + len);
7526 if (ret && space_reserved) {
7527 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7529 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7531 btrfs_delalloc_release_space(BTRFS_I(inode),
7532 dio_data->data_reserved,
7534 extent_changeset_free(dio_data->data_reserved);
7535 dio_data->data_reserved = NULL;
7541 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7542 loff_t length, unsigned int flags, struct iomap *iomap,
7543 struct iomap *srcmap)
7545 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7546 struct extent_map *em;
7547 struct extent_state *cached_state = NULL;
7548 struct btrfs_dio_data *dio_data = NULL;
7549 u64 lockstart, lockend;
7550 const bool write = !!(flags & IOMAP_WRITE);
7553 bool unlock_extents = false;
7556 len = min_t(u64, len, fs_info->sectorsize);
7559 lockend = start + len - 1;
7562 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7563 * enough if we've written compressed pages to this area, so we need to
7564 * flush the dirty pages again to make absolutely sure that any
7565 * outstanding dirty pages are on disk - the first flush only starts
7566 * compression on the data, while keeping the pages locked, so by the
7567 * time the second flush returns we know bios for the compressed pages
7568 * were submitted and finished, and the pages no longer under writeback.
7570 * If we have a NOWAIT request and we have any pages in the range that
7571 * are locked, likely due to compression still in progress, we don't want
7572 * to block on page locks. We also don't want to block on pages marked as
7573 * dirty or under writeback (same as for the non-compression case).
7574 * iomap_dio_rw() did the same check, but after that and before we got
7575 * here, mmap'ed writes may have happened or buffered reads started
7576 * (readpage() and readahead(), which lock pages), as we haven't locked
7577 * the file range yet.
7579 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7580 &BTRFS_I(inode)->runtime_flags)) {
7581 if (flags & IOMAP_NOWAIT) {
7582 if (filemap_range_needs_writeback(inode->i_mapping,
7583 lockstart, lockend))
7586 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7587 start + length - 1);
7593 if (flags & IOMAP_NOWAIT) {
7594 dio_data = kzalloc(sizeof(*dio_data), GFP_NOWAIT);
7598 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7603 iomap->private = dio_data;
7607 * If this errors out it's because we couldn't invalidate pagecache for
7608 * this range and we need to fallback to buffered IO, or we are doing a
7609 * NOWAIT read/write and we need to block.
7611 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7615 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7622 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7623 * io. INLINE is special, and we could probably kludge it in here, but
7624 * it's still buffered so for safety lets just fall back to the generic
7627 * For COMPRESSED we _have_ to read the entire extent in so we can
7628 * decompress it, so there will be buffering required no matter what we
7629 * do, so go ahead and fallback to buffered.
7631 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7632 * to buffered IO. Don't blame me, this is the price we pay for using
7635 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7636 em->block_start == EXTENT_MAP_INLINE) {
7637 free_extent_map(em);
7642 len = min(len, em->len - (start - em->start));
7645 * If we have a NOWAIT request and the range contains multiple extents
7646 * (or a mix of extents and holes), then we return -EAGAIN to make the
7647 * caller fallback to a context where it can do a blocking (without
7648 * NOWAIT) request. This way we avoid doing partial IO and returning
7649 * success to the caller, which is not optimal for writes and for reads
7650 * it can result in unexpected behaviour for an application.
7652 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7653 * iomap_dio_rw(), we can end up returning less data then what the caller
7654 * asked for, resulting in an unexpected, and incorrect, short read.
7655 * That is, the caller asked to read N bytes and we return less than that,
7656 * which is wrong unless we are crossing EOF. This happens if we get a
7657 * page fault error when trying to fault in pages for the buffer that is
7658 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7659 * have previously submitted bios for other extents in the range, in
7660 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7661 * those bios have completed by the time we get the page fault error,
7662 * which we return back to our caller - we should only return EIOCBQUEUED
7663 * after we have submitted bios for all the extents in the range.
7665 if ((flags & IOMAP_NOWAIT) && len < length) {
7666 free_extent_map(em);
7672 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7676 unlock_extents = true;
7677 /* Recalc len in case the new em is smaller than requested */
7678 len = min(len, em->len - (start - em->start));
7681 * We need to unlock only the end area that we aren't using.
7682 * The rest is going to be unlocked by the endio routine.
7684 lockstart = start + len;
7685 if (lockstart < lockend)
7686 unlock_extents = true;
7690 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7691 lockstart, lockend, &cached_state);
7693 free_extent_state(cached_state);
7696 * Translate extent map information to iomap.
7697 * We trim the extents (and move the addr) even though iomap code does
7698 * that, since we have locked only the parts we are performing I/O in.
7700 if ((em->block_start == EXTENT_MAP_HOLE) ||
7701 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7702 iomap->addr = IOMAP_NULL_ADDR;
7703 iomap->type = IOMAP_HOLE;
7705 iomap->addr = em->block_start + (start - em->start);
7706 iomap->type = IOMAP_MAPPED;
7708 iomap->offset = start;
7709 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7710 iomap->length = len;
7712 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7713 iomap->flags |= IOMAP_F_ZONE_APPEND;
7715 free_extent_map(em);
7720 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7728 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7729 ssize_t written, unsigned int flags, struct iomap *iomap)
7732 struct btrfs_dio_data *dio_data = iomap->private;
7733 size_t submitted = dio_data->submitted;
7734 const bool write = !!(flags & IOMAP_WRITE);
7736 if (!write && (iomap->type == IOMAP_HOLE)) {
7737 /* If reading from a hole, unlock and return */
7738 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7742 if (submitted < length) {
7744 length -= submitted;
7746 __endio_write_update_ordered(BTRFS_I(inode), pos,
7749 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7755 extent_changeset_free(dio_data->data_reserved);
7758 iomap->private = NULL;
7763 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7766 * This implies a barrier so that stores to dio_bio->bi_status before
7767 * this and loads of dio_bio->bi_status after this are fully ordered.
7769 if (!refcount_dec_and_test(&dip->refs))
7772 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7773 __endio_write_update_ordered(BTRFS_I(dip->inode),
7776 !dip->dio_bio->bi_status);
7778 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7780 dip->file_offset + dip->bytes - 1);
7783 bio_endio(dip->dio_bio);
7787 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7789 unsigned long bio_flags)
7791 struct btrfs_dio_private *dip = bio->bi_private;
7792 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7795 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7797 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7801 refcount_inc(&dip->refs);
7802 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7804 refcount_dec(&dip->refs);
7808 static blk_status_t btrfs_check_read_dio_bio(struct btrfs_dio_private *dip,
7809 struct btrfs_bio *bbio,
7810 const bool uptodate)
7812 struct inode *inode = dip->inode;
7813 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7814 const u32 sectorsize = fs_info->sectorsize;
7815 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7816 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7817 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7818 struct bio_vec bvec;
7819 struct bvec_iter iter;
7821 blk_status_t err = BLK_STS_OK;
7823 __bio_for_each_segment(bvec, &bbio->bio, iter, bbio->iter) {
7824 unsigned int i, nr_sectors, pgoff;
7826 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7827 pgoff = bvec.bv_offset;
7828 for (i = 0; i < nr_sectors; i++) {
7829 u64 start = bbio->file_offset + bio_offset;
7831 ASSERT(pgoff < PAGE_SIZE);
7833 (!csum || !check_data_csum(inode, bbio,
7834 bio_offset, bvec.bv_page,
7836 clean_io_failure(fs_info, failure_tree, io_tree,
7837 start, bvec.bv_page,
7838 btrfs_ino(BTRFS_I(inode)),
7843 ret = btrfs_repair_one_sector(inode, &bbio->bio,
7844 bio_offset, bvec.bv_page, pgoff,
7845 start, bbio->mirror_num,
7846 submit_dio_repair_bio);
7848 err = errno_to_blk_status(ret);
7850 ASSERT(bio_offset + sectorsize > bio_offset);
7851 bio_offset += sectorsize;
7852 pgoff += sectorsize;
7858 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7859 const u64 offset, const u64 bytes,
7860 const bool uptodate)
7862 btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes,
7863 finish_ordered_fn, uptodate);
7866 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7868 u64 dio_file_offset)
7870 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, false);
7873 static void btrfs_end_dio_bio(struct bio *bio)
7875 struct btrfs_dio_private *dip = bio->bi_private;
7876 struct btrfs_bio *bbio = btrfs_bio(bio);
7877 blk_status_t err = bio->bi_status;
7880 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7881 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7882 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7883 bio->bi_opf, bio->bi_iter.bi_sector,
7884 bio->bi_iter.bi_size, err);
7886 if (bio_op(bio) == REQ_OP_READ)
7887 err = btrfs_check_read_dio_bio(dip, bbio, !err);
7890 dip->dio_bio->bi_status = err;
7892 btrfs_record_physical_zoned(dip->inode, bbio->file_offset, bio);
7895 btrfs_dio_private_put(dip);
7898 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7899 struct inode *inode, u64 file_offset, int async_submit)
7901 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7902 struct btrfs_dio_private *dip = bio->bi_private;
7903 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
7906 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7908 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7911 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7916 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7919 if (write && async_submit) {
7920 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7921 btrfs_submit_bio_start_direct_io);
7925 * If we aren't doing async submit, calculate the csum of the
7928 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
7934 csum_offset = file_offset - dip->file_offset;
7935 csum_offset >>= fs_info->sectorsize_bits;
7936 csum_offset *= fs_info->csum_size;
7937 btrfs_bio(bio)->csum = dip->csums + csum_offset;
7940 ret = btrfs_map_bio(fs_info, bio, 0);
7946 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7947 * or ordered extents whether or not we submit any bios.
7949 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7950 struct inode *inode,
7953 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7954 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7956 struct btrfs_dio_private *dip;
7958 dip_size = sizeof(*dip);
7959 if (!write && csum) {
7960 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7963 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7964 dip_size += fs_info->csum_size * nblocks;
7967 dip = kzalloc(dip_size, GFP_NOFS);
7972 dip->file_offset = file_offset;
7973 dip->bytes = dio_bio->bi_iter.bi_size;
7974 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
7975 dip->dio_bio = dio_bio;
7976 refcount_set(&dip->refs, 1);
7980 static void btrfs_submit_direct(const struct iomap_iter *iter,
7981 struct bio *dio_bio, loff_t file_offset)
7983 struct inode *inode = iter->inode;
7984 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
7985 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7986 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7987 BTRFS_BLOCK_GROUP_RAID56_MASK);
7988 struct btrfs_dio_private *dip;
7991 int async_submit = 0;
7993 u64 clone_offset = 0;
7997 blk_status_t status;
7998 struct btrfs_io_geometry geom;
7999 struct btrfs_dio_data *dio_data = iter->iomap.private;
8000 struct extent_map *em = NULL;
8002 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8005 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8006 file_offset + dio_bio->bi_iter.bi_size - 1);
8008 dio_bio->bi_status = BLK_STS_RESOURCE;
8015 * Load the csums up front to reduce csum tree searches and
8016 * contention when submitting bios.
8018 * If we have csums disabled this will do nothing.
8020 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8021 if (status != BLK_STS_OK)
8025 start_sector = dio_bio->bi_iter.bi_sector;
8026 submit_len = dio_bio->bi_iter.bi_size;
8029 logical = start_sector << 9;
8030 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8032 status = errno_to_blk_status(PTR_ERR(em));
8036 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8039 status = errno_to_blk_status(ret);
8043 clone_len = min(submit_len, geom.len);
8044 ASSERT(clone_len <= UINT_MAX);
8047 * This will never fail as it's passing GPF_NOFS and
8048 * the allocation is backed by btrfs_bioset.
8050 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8051 bio->bi_private = dip;
8052 bio->bi_end_io = btrfs_end_dio_bio;
8053 btrfs_bio(bio)->file_offset = file_offset;
8055 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8056 status = extract_ordered_extent(BTRFS_I(inode), bio,
8064 ASSERT(submit_len >= clone_len);
8065 submit_len -= clone_len;
8068 * Increase the count before we submit the bio so we know
8069 * the end IO handler won't happen before we increase the
8070 * count. Otherwise, the dip might get freed before we're
8071 * done setting it up.
8073 * We transfer the initial reference to the last bio, so we
8074 * don't need to increment the reference count for the last one.
8076 if (submit_len > 0) {
8077 refcount_inc(&dip->refs);
8079 * If we are submitting more than one bio, submit them
8080 * all asynchronously. The exception is RAID 5 or 6, as
8081 * asynchronous checksums make it difficult to collect
8082 * full stripe writes.
8088 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8093 refcount_dec(&dip->refs);
8097 dio_data->submitted += clone_len;
8098 clone_offset += clone_len;
8099 start_sector += clone_len >> 9;
8100 file_offset += clone_len;
8102 free_extent_map(em);
8103 } while (submit_len > 0);
8107 free_extent_map(em);
8109 dip->dio_bio->bi_status = status;
8110 btrfs_dio_private_put(dip);
8113 const struct iomap_ops btrfs_dio_iomap_ops = {
8114 .iomap_begin = btrfs_dio_iomap_begin,
8115 .iomap_end = btrfs_dio_iomap_end,
8118 const struct iomap_dio_ops btrfs_dio_ops = {
8119 .submit_io = btrfs_submit_direct,
8122 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8127 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8131 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8134 int btrfs_readpage(struct file *file, struct page *page)
8136 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8137 u64 start = page_offset(page);
8138 u64 end = start + PAGE_SIZE - 1;
8139 struct btrfs_bio_ctrl bio_ctrl = { 0 };
8142 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8144 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
8148 ret2 = submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags);
8155 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8157 struct inode *inode = page->mapping->host;
8160 if (current->flags & PF_MEMALLOC) {
8161 redirty_page_for_writepage(wbc, page);
8167 * If we are under memory pressure we will call this directly from the
8168 * VM, we need to make sure we have the inode referenced for the ordered
8169 * extent. If not just return like we didn't do anything.
8171 if (!igrab(inode)) {
8172 redirty_page_for_writepage(wbc, page);
8173 return AOP_WRITEPAGE_ACTIVATE;
8175 ret = extent_write_full_page(page, wbc);
8176 btrfs_add_delayed_iput(inode);
8180 static int btrfs_writepages(struct address_space *mapping,
8181 struct writeback_control *wbc)
8183 return extent_writepages(mapping, wbc);
8186 static void btrfs_readahead(struct readahead_control *rac)
8188 extent_readahead(rac);
8192 * For releasepage() and invalidate_folio() we have a race window where
8193 * folio_end_writeback() is called but the subpage spinlock is not yet released.
8194 * If we continue to release/invalidate the page, we could cause use-after-free
8195 * for subpage spinlock. So this function is to spin and wait for subpage
8198 static void wait_subpage_spinlock(struct page *page)
8200 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
8201 struct btrfs_subpage *subpage;
8203 if (!btrfs_is_subpage(fs_info, page))
8206 ASSERT(PagePrivate(page) && page->private);
8207 subpage = (struct btrfs_subpage *)page->private;
8210 * This may look insane as we just acquire the spinlock and release it,
8211 * without doing anything. But we just want to make sure no one is
8212 * still holding the subpage spinlock.
8213 * And since the page is not dirty nor writeback, and we have page
8214 * locked, the only possible way to hold a spinlock is from the endio
8215 * function to clear page writeback.
8217 * Here we just acquire the spinlock so that all existing callers
8218 * should exit and we're safe to release/invalidate the page.
8220 spin_lock_irq(&subpage->lock);
8221 spin_unlock_irq(&subpage->lock);
8224 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8226 int ret = try_release_extent_mapping(page, gfp_flags);
8229 wait_subpage_spinlock(page);
8230 clear_page_extent_mapped(page);
8235 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8237 if (PageWriteback(page) || PageDirty(page))
8239 return __btrfs_releasepage(page, gfp_flags);
8242 #ifdef CONFIG_MIGRATION
8243 static int btrfs_migratepage(struct address_space *mapping,
8244 struct page *newpage, struct page *page,
8245 enum migrate_mode mode)
8249 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8250 if (ret != MIGRATEPAGE_SUCCESS)
8253 if (page_has_private(page))
8254 attach_page_private(newpage, detach_page_private(page));
8256 if (PageOrdered(page)) {
8257 ClearPageOrdered(page);
8258 SetPageOrdered(newpage);
8261 if (mode != MIGRATE_SYNC_NO_COPY)
8262 migrate_page_copy(newpage, page);
8264 migrate_page_states(newpage, page);
8265 return MIGRATEPAGE_SUCCESS;
8269 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8272 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8273 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8274 struct extent_io_tree *tree = &inode->io_tree;
8275 struct extent_state *cached_state = NULL;
8276 u64 page_start = folio_pos(folio);
8277 u64 page_end = page_start + folio_size(folio) - 1;
8279 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8282 * We have folio locked so no new ordered extent can be created on this
8283 * page, nor bio can be submitted for this folio.
8285 * But already submitted bio can still be finished on this folio.
8286 * Furthermore, endio function won't skip folio which has Ordered
8287 * (Private2) already cleared, so it's possible for endio and
8288 * invalidate_folio to do the same ordered extent accounting twice
8291 * So here we wait for any submitted bios to finish, so that we won't
8292 * do double ordered extent accounting on the same folio.
8294 folio_wait_writeback(folio);
8295 wait_subpage_spinlock(&folio->page);
8298 * For subpage case, we have call sites like
8299 * btrfs_punch_hole_lock_range() which passes range not aligned to
8301 * If the range doesn't cover the full folio, we don't need to and
8302 * shouldn't clear page extent mapped, as folio->private can still
8303 * record subpage dirty bits for other part of the range.
8305 * For cases that invalidate the full folio even the range doesn't
8306 * cover the full folio, like invalidating the last folio, we're
8307 * still safe to wait for ordered extent to finish.
8309 if (!(offset == 0 && length == folio_size(folio))) {
8310 btrfs_releasepage(&folio->page, GFP_NOFS);
8314 if (!inode_evicting)
8315 lock_extent_bits(tree, page_start, page_end, &cached_state);
8318 while (cur < page_end) {
8319 struct btrfs_ordered_extent *ordered;
8324 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8325 page_end + 1 - cur);
8327 range_end = page_end;
8329 * No ordered extent covering this range, we are safe
8330 * to delete all extent states in the range.
8332 delete_states = true;
8335 if (ordered->file_offset > cur) {
8337 * There is a range between [cur, oe->file_offset) not
8338 * covered by any ordered extent.
8339 * We are safe to delete all extent states, and handle
8340 * the ordered extent in the next iteration.
8342 range_end = ordered->file_offset - 1;
8343 delete_states = true;
8347 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8349 ASSERT(range_end + 1 - cur < U32_MAX);
8350 range_len = range_end + 1 - cur;
8351 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8353 * If Ordered (Private2) is cleared, it means endio has
8354 * already been executed for the range.
8355 * We can't delete the extent states as
8356 * btrfs_finish_ordered_io() may still use some of them.
8358 delete_states = false;
8361 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8364 * IO on this page will never be started, so we need to account
8365 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8366 * here, must leave that up for the ordered extent completion.
8368 * This will also unlock the range for incoming
8369 * btrfs_finish_ordered_io().
8371 if (!inode_evicting)
8372 clear_extent_bit(tree, cur, range_end,
8374 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8375 EXTENT_DEFRAG, 1, 0, &cached_state);
8377 spin_lock_irq(&inode->ordered_tree.lock);
8378 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8379 ordered->truncated_len = min(ordered->truncated_len,
8380 cur - ordered->file_offset);
8381 spin_unlock_irq(&inode->ordered_tree.lock);
8383 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8384 cur, range_end + 1 - cur)) {
8385 btrfs_finish_ordered_io(ordered);
8387 * The ordered extent has finished, now we're again
8388 * safe to delete all extent states of the range.
8390 delete_states = true;
8393 * btrfs_finish_ordered_io() will get executed by endio
8394 * of other pages, thus we can't delete extent states
8397 delete_states = false;
8401 btrfs_put_ordered_extent(ordered);
8403 * Qgroup reserved space handler
8404 * Sector(s) here will be either:
8406 * 1) Already written to disk or bio already finished
8407 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8408 * Qgroup will be handled by its qgroup_record then.
8409 * btrfs_qgroup_free_data() call will do nothing here.
8411 * 2) Not written to disk yet
8412 * Then btrfs_qgroup_free_data() call will clear the
8413 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8414 * reserved data space.
8415 * Since the IO will never happen for this page.
8417 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8418 if (!inode_evicting) {
8419 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8420 EXTENT_DELALLOC | EXTENT_UPTODATE |
8421 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8422 delete_states, &cached_state);
8424 cur = range_end + 1;
8427 * We have iterated through all ordered extents of the page, the page
8428 * should not have Ordered (Private2) anymore, or the above iteration
8429 * did something wrong.
8431 ASSERT(!folio_test_ordered(folio));
8432 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8433 if (!inode_evicting)
8434 __btrfs_releasepage(&folio->page, GFP_NOFS);
8435 clear_page_extent_mapped(&folio->page);
8439 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8440 * called from a page fault handler when a page is first dirtied. Hence we must
8441 * be careful to check for EOF conditions here. We set the page up correctly
8442 * for a written page which means we get ENOSPC checking when writing into
8443 * holes and correct delalloc and unwritten extent mapping on filesystems that
8444 * support these features.
8446 * We are not allowed to take the i_mutex here so we have to play games to
8447 * protect against truncate races as the page could now be beyond EOF. Because
8448 * truncate_setsize() writes the inode size before removing pages, once we have
8449 * the page lock we can determine safely if the page is beyond EOF. If it is not
8450 * beyond EOF, then the page is guaranteed safe against truncation until we
8453 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8455 struct page *page = vmf->page;
8456 struct inode *inode = file_inode(vmf->vma->vm_file);
8457 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8458 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8459 struct btrfs_ordered_extent *ordered;
8460 struct extent_state *cached_state = NULL;
8461 struct extent_changeset *data_reserved = NULL;
8462 unsigned long zero_start;
8472 reserved_space = PAGE_SIZE;
8474 sb_start_pagefault(inode->i_sb);
8475 page_start = page_offset(page);
8476 page_end = page_start + PAGE_SIZE - 1;
8480 * Reserving delalloc space after obtaining the page lock can lead to
8481 * deadlock. For example, if a dirty page is locked by this function
8482 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8483 * dirty page write out, then the btrfs_writepage() function could
8484 * end up waiting indefinitely to get a lock on the page currently
8485 * being processed by btrfs_page_mkwrite() function.
8487 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8488 page_start, reserved_space);
8490 ret2 = file_update_time(vmf->vma->vm_file);
8494 ret = vmf_error(ret2);
8500 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8502 down_read(&BTRFS_I(inode)->i_mmap_lock);
8504 size = i_size_read(inode);
8506 if ((page->mapping != inode->i_mapping) ||
8507 (page_start >= size)) {
8508 /* page got truncated out from underneath us */
8511 wait_on_page_writeback(page);
8513 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8514 ret2 = set_page_extent_mapped(page);
8516 ret = vmf_error(ret2);
8517 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8522 * we can't set the delalloc bits if there are pending ordered
8523 * extents. Drop our locks and wait for them to finish
8525 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8528 unlock_extent_cached(io_tree, page_start, page_end,
8531 up_read(&BTRFS_I(inode)->i_mmap_lock);
8532 btrfs_start_ordered_extent(ordered, 1);
8533 btrfs_put_ordered_extent(ordered);
8537 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8538 reserved_space = round_up(size - page_start,
8539 fs_info->sectorsize);
8540 if (reserved_space < PAGE_SIZE) {
8541 end = page_start + reserved_space - 1;
8542 btrfs_delalloc_release_space(BTRFS_I(inode),
8543 data_reserved, page_start,
8544 PAGE_SIZE - reserved_space, true);
8549 * page_mkwrite gets called when the page is firstly dirtied after it's
8550 * faulted in, but write(2) could also dirty a page and set delalloc
8551 * bits, thus in this case for space account reason, we still need to
8552 * clear any delalloc bits within this page range since we have to
8553 * reserve data&meta space before lock_page() (see above comments).
8555 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8556 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8557 EXTENT_DEFRAG, 0, 0, &cached_state);
8559 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8562 unlock_extent_cached(io_tree, page_start, page_end,
8564 ret = VM_FAULT_SIGBUS;
8568 /* page is wholly or partially inside EOF */
8569 if (page_start + PAGE_SIZE > size)
8570 zero_start = offset_in_page(size);
8572 zero_start = PAGE_SIZE;
8574 if (zero_start != PAGE_SIZE) {
8575 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8576 flush_dcache_page(page);
8578 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8579 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8580 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8582 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8584 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8585 up_read(&BTRFS_I(inode)->i_mmap_lock);
8587 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8588 sb_end_pagefault(inode->i_sb);
8589 extent_changeset_free(data_reserved);
8590 return VM_FAULT_LOCKED;
8594 up_read(&BTRFS_I(inode)->i_mmap_lock);
8596 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8597 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8598 reserved_space, (ret != 0));
8600 sb_end_pagefault(inode->i_sb);
8601 extent_changeset_free(data_reserved);
8605 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8607 struct btrfs_truncate_control control = {
8608 .inode = BTRFS_I(inode),
8609 .ino = btrfs_ino(BTRFS_I(inode)),
8610 .min_type = BTRFS_EXTENT_DATA_KEY,
8611 .clear_extent_range = true,
8613 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8614 struct btrfs_root *root = BTRFS_I(inode)->root;
8615 struct btrfs_block_rsv *rsv;
8617 struct btrfs_trans_handle *trans;
8618 u64 mask = fs_info->sectorsize - 1;
8619 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8621 if (!skip_writeback) {
8622 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8629 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8630 * things going on here:
8632 * 1) We need to reserve space to update our inode.
8634 * 2) We need to have something to cache all the space that is going to
8635 * be free'd up by the truncate operation, but also have some slack
8636 * space reserved in case it uses space during the truncate (thank you
8637 * very much snapshotting).
8639 * And we need these to be separate. The fact is we can use a lot of
8640 * space doing the truncate, and we have no earthly idea how much space
8641 * we will use, so we need the truncate reservation to be separate so it
8642 * doesn't end up using space reserved for updating the inode. We also
8643 * need to be able to stop the transaction and start a new one, which
8644 * means we need to be able to update the inode several times, and we
8645 * have no idea of knowing how many times that will be, so we can't just
8646 * reserve 1 item for the entirety of the operation, so that has to be
8647 * done separately as well.
8649 * So that leaves us with
8651 * 1) rsv - for the truncate reservation, which we will steal from the
8652 * transaction reservation.
8653 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8654 * updating the inode.
8656 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8659 rsv->size = min_size;
8663 * 1 for the truncate slack space
8664 * 1 for updating the inode.
8666 trans = btrfs_start_transaction(root, 2);
8667 if (IS_ERR(trans)) {
8668 ret = PTR_ERR(trans);
8672 /* Migrate the slack space for the truncate to our reserve */
8673 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8677 trans->block_rsv = rsv;
8680 struct extent_state *cached_state = NULL;
8681 const u64 new_size = inode->i_size;
8682 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8684 control.new_size = new_size;
8685 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
8688 * We want to drop from the next block forward in case this new
8689 * size is not block aligned since we will be keeping the last
8690 * block of the extent just the way it is.
8692 btrfs_drop_extent_cache(BTRFS_I(inode),
8693 ALIGN(new_size, fs_info->sectorsize),
8696 ret = btrfs_truncate_inode_items(trans, root, &control);
8698 inode_sub_bytes(inode, control.sub_bytes);
8699 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), control.last_size);
8701 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
8702 (u64)-1, &cached_state);
8704 trans->block_rsv = &fs_info->trans_block_rsv;
8705 if (ret != -ENOSPC && ret != -EAGAIN)
8708 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8712 btrfs_end_transaction(trans);
8713 btrfs_btree_balance_dirty(fs_info);
8715 trans = btrfs_start_transaction(root, 2);
8716 if (IS_ERR(trans)) {
8717 ret = PTR_ERR(trans);
8722 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8723 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8724 rsv, min_size, false);
8725 BUG_ON(ret); /* shouldn't happen */
8726 trans->block_rsv = rsv;
8730 * We can't call btrfs_truncate_block inside a trans handle as we could
8731 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8732 * know we've truncated everything except the last little bit, and can
8733 * do btrfs_truncate_block and then update the disk_i_size.
8735 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8736 btrfs_end_transaction(trans);
8737 btrfs_btree_balance_dirty(fs_info);
8739 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8742 trans = btrfs_start_transaction(root, 1);
8743 if (IS_ERR(trans)) {
8744 ret = PTR_ERR(trans);
8747 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8753 trans->block_rsv = &fs_info->trans_block_rsv;
8754 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8758 ret2 = btrfs_end_transaction(trans);
8761 btrfs_btree_balance_dirty(fs_info);
8764 btrfs_free_block_rsv(fs_info, rsv);
8766 * So if we truncate and then write and fsync we normally would just
8767 * write the extents that changed, which is a problem if we need to
8768 * first truncate that entire inode. So set this flag so we write out
8769 * all of the extents in the inode to the sync log so we're completely
8772 * If no extents were dropped or trimmed we don't need to force the next
8773 * fsync to truncate all the inode's items from the log and re-log them
8774 * all. This means the truncate operation did not change the file size,
8775 * or changed it to a smaller size but there was only an implicit hole
8776 * between the old i_size and the new i_size, and there were no prealloc
8777 * extents beyond i_size to drop.
8779 if (control.extents_found > 0)
8780 btrfs_set_inode_full_sync(BTRFS_I(inode));
8785 struct inode *btrfs_new_subvol_inode(struct user_namespace *mnt_userns,
8788 struct inode *inode;
8790 inode = new_inode(dir->i_sb);
8793 * Subvolumes don't inherit the sgid bit or the parent's gid if
8794 * the parent's sgid bit is set. This is probably a bug.
8796 inode_init_owner(mnt_userns, inode, NULL,
8797 S_IFDIR | (~current_umask() & S_IRWXUGO));
8798 inode->i_op = &btrfs_dir_inode_operations;
8799 inode->i_fop = &btrfs_dir_file_operations;
8804 struct inode *btrfs_alloc_inode(struct super_block *sb)
8806 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8807 struct btrfs_inode *ei;
8808 struct inode *inode;
8810 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8817 ei->last_sub_trans = 0;
8818 ei->logged_trans = 0;
8819 ei->delalloc_bytes = 0;
8820 ei->new_delalloc_bytes = 0;
8821 ei->defrag_bytes = 0;
8822 ei->disk_i_size = 0;
8826 ei->index_cnt = (u64)-1;
8828 ei->last_unlink_trans = 0;
8829 ei->last_reflink_trans = 0;
8830 ei->last_log_commit = 0;
8832 spin_lock_init(&ei->lock);
8833 ei->outstanding_extents = 0;
8834 if (sb->s_magic != BTRFS_TEST_MAGIC)
8835 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8836 BTRFS_BLOCK_RSV_DELALLOC);
8837 ei->runtime_flags = 0;
8838 ei->prop_compress = BTRFS_COMPRESS_NONE;
8839 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8841 ei->delayed_node = NULL;
8843 ei->i_otime.tv_sec = 0;
8844 ei->i_otime.tv_nsec = 0;
8846 inode = &ei->vfs_inode;
8847 extent_map_tree_init(&ei->extent_tree);
8848 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8849 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8850 IO_TREE_INODE_IO_FAILURE, inode);
8851 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8852 IO_TREE_INODE_FILE_EXTENT, inode);
8853 ei->io_tree.track_uptodate = true;
8854 ei->io_failure_tree.track_uptodate = true;
8855 atomic_set(&ei->sync_writers, 0);
8856 mutex_init(&ei->log_mutex);
8857 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8858 INIT_LIST_HEAD(&ei->delalloc_inodes);
8859 INIT_LIST_HEAD(&ei->delayed_iput);
8860 RB_CLEAR_NODE(&ei->rb_node);
8861 init_rwsem(&ei->i_mmap_lock);
8866 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8867 void btrfs_test_destroy_inode(struct inode *inode)
8869 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8870 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8874 void btrfs_free_inode(struct inode *inode)
8876 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8879 void btrfs_destroy_inode(struct inode *vfs_inode)
8881 struct btrfs_ordered_extent *ordered;
8882 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8883 struct btrfs_root *root = inode->root;
8885 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8886 WARN_ON(vfs_inode->i_data.nrpages);
8887 WARN_ON(inode->block_rsv.reserved);
8888 WARN_ON(inode->block_rsv.size);
8889 WARN_ON(inode->outstanding_extents);
8890 if (!S_ISDIR(vfs_inode->i_mode)) {
8891 WARN_ON(inode->delalloc_bytes);
8892 WARN_ON(inode->new_delalloc_bytes);
8894 WARN_ON(inode->csum_bytes);
8895 WARN_ON(inode->defrag_bytes);
8898 * This can happen where we create an inode, but somebody else also
8899 * created the same inode and we need to destroy the one we already
8906 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8910 btrfs_err(root->fs_info,
8911 "found ordered extent %llu %llu on inode cleanup",
8912 ordered->file_offset, ordered->num_bytes);
8913 btrfs_remove_ordered_extent(inode, ordered);
8914 btrfs_put_ordered_extent(ordered);
8915 btrfs_put_ordered_extent(ordered);
8918 btrfs_qgroup_check_reserved_leak(inode);
8919 inode_tree_del(inode);
8920 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8921 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8922 btrfs_put_root(inode->root);
8925 int btrfs_drop_inode(struct inode *inode)
8927 struct btrfs_root *root = BTRFS_I(inode)->root;
8932 /* the snap/subvol tree is on deleting */
8933 if (btrfs_root_refs(&root->root_item) == 0)
8936 return generic_drop_inode(inode);
8939 static void init_once(void *foo)
8941 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8943 inode_init_once(&ei->vfs_inode);
8946 void __cold btrfs_destroy_cachep(void)
8949 * Make sure all delayed rcu free inodes are flushed before we
8953 kmem_cache_destroy(btrfs_inode_cachep);
8954 kmem_cache_destroy(btrfs_trans_handle_cachep);
8955 kmem_cache_destroy(btrfs_path_cachep);
8956 kmem_cache_destroy(btrfs_free_space_cachep);
8957 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8960 int __init btrfs_init_cachep(void)
8962 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8963 sizeof(struct btrfs_inode), 0,
8964 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8966 if (!btrfs_inode_cachep)
8969 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8970 sizeof(struct btrfs_trans_handle), 0,
8971 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8972 if (!btrfs_trans_handle_cachep)
8975 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8976 sizeof(struct btrfs_path), 0,
8977 SLAB_MEM_SPREAD, NULL);
8978 if (!btrfs_path_cachep)
8981 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8982 sizeof(struct btrfs_free_space), 0,
8983 SLAB_MEM_SPREAD, NULL);
8984 if (!btrfs_free_space_cachep)
8987 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8988 PAGE_SIZE, PAGE_SIZE,
8989 SLAB_MEM_SPREAD, NULL);
8990 if (!btrfs_free_space_bitmap_cachep)
8995 btrfs_destroy_cachep();
8999 static int btrfs_getattr(struct user_namespace *mnt_userns,
9000 const struct path *path, struct kstat *stat,
9001 u32 request_mask, unsigned int flags)
9005 struct inode *inode = d_inode(path->dentry);
9006 u32 blocksize = inode->i_sb->s_blocksize;
9007 u32 bi_flags = BTRFS_I(inode)->flags;
9008 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
9010 stat->result_mask |= STATX_BTIME;
9011 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9012 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9013 if (bi_flags & BTRFS_INODE_APPEND)
9014 stat->attributes |= STATX_ATTR_APPEND;
9015 if (bi_flags & BTRFS_INODE_COMPRESS)
9016 stat->attributes |= STATX_ATTR_COMPRESSED;
9017 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9018 stat->attributes |= STATX_ATTR_IMMUTABLE;
9019 if (bi_flags & BTRFS_INODE_NODUMP)
9020 stat->attributes |= STATX_ATTR_NODUMP;
9021 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
9022 stat->attributes |= STATX_ATTR_VERITY;
9024 stat->attributes_mask |= (STATX_ATTR_APPEND |
9025 STATX_ATTR_COMPRESSED |
9026 STATX_ATTR_IMMUTABLE |
9029 generic_fillattr(mnt_userns, inode, stat);
9030 stat->dev = BTRFS_I(inode)->root->anon_dev;
9032 spin_lock(&BTRFS_I(inode)->lock);
9033 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9034 inode_bytes = inode_get_bytes(inode);
9035 spin_unlock(&BTRFS_I(inode)->lock);
9036 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9037 ALIGN(delalloc_bytes, blocksize)) >> 9;
9041 static int btrfs_rename_exchange(struct inode *old_dir,
9042 struct dentry *old_dentry,
9043 struct inode *new_dir,
9044 struct dentry *new_dentry)
9046 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9047 struct btrfs_trans_handle *trans;
9048 unsigned int trans_num_items;
9049 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9050 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9051 struct inode *new_inode = new_dentry->d_inode;
9052 struct inode *old_inode = old_dentry->d_inode;
9053 struct timespec64 ctime = current_time(old_inode);
9054 struct btrfs_rename_ctx old_rename_ctx;
9055 struct btrfs_rename_ctx new_rename_ctx;
9056 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9057 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9062 bool need_abort = false;
9065 * For non-subvolumes allow exchange only within one subvolume, in the
9066 * same inode namespace. Two subvolumes (represented as directory) can
9067 * be exchanged as they're a logical link and have a fixed inode number.
9070 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9071 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9074 /* close the race window with snapshot create/destroy ioctl */
9075 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9076 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9077 down_read(&fs_info->subvol_sem);
9081 * 1 to remove old dir item
9082 * 1 to remove old dir index
9083 * 1 to add new dir item
9084 * 1 to add new dir index
9085 * 1 to update parent inode
9087 * If the parents are the same, we only need to account for one
9089 trans_num_items = (old_dir == new_dir ? 9 : 10);
9090 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9092 * 1 to remove old root ref
9093 * 1 to remove old root backref
9094 * 1 to add new root ref
9095 * 1 to add new root backref
9097 trans_num_items += 4;
9100 * 1 to update inode item
9101 * 1 to remove old inode ref
9102 * 1 to add new inode ref
9104 trans_num_items += 3;
9106 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9107 trans_num_items += 4;
9109 trans_num_items += 3;
9110 trans = btrfs_start_transaction(root, trans_num_items);
9111 if (IS_ERR(trans)) {
9112 ret = PTR_ERR(trans);
9117 ret = btrfs_record_root_in_trans(trans, dest);
9123 * We need to find a free sequence number both in the source and
9124 * in the destination directory for the exchange.
9126 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9129 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9133 BTRFS_I(old_inode)->dir_index = 0ULL;
9134 BTRFS_I(new_inode)->dir_index = 0ULL;
9136 /* Reference for the source. */
9137 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9138 /* force full log commit if subvolume involved. */
9139 btrfs_set_log_full_commit(trans);
9141 ret = btrfs_insert_inode_ref(trans, dest,
9142 new_dentry->d_name.name,
9143 new_dentry->d_name.len,
9145 btrfs_ino(BTRFS_I(new_dir)),
9152 /* And now for the dest. */
9153 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9154 /* force full log commit if subvolume involved. */
9155 btrfs_set_log_full_commit(trans);
9157 ret = btrfs_insert_inode_ref(trans, root,
9158 old_dentry->d_name.name,
9159 old_dentry->d_name.len,
9161 btrfs_ino(BTRFS_I(old_dir)),
9165 btrfs_abort_transaction(trans, ret);
9170 /* Update inode version and ctime/mtime. */
9171 inode_inc_iversion(old_dir);
9172 inode_inc_iversion(new_dir);
9173 inode_inc_iversion(old_inode);
9174 inode_inc_iversion(new_inode);
9175 old_dir->i_ctime = old_dir->i_mtime = ctime;
9176 new_dir->i_ctime = new_dir->i_mtime = ctime;
9177 old_inode->i_ctime = ctime;
9178 new_inode->i_ctime = ctime;
9180 if (old_dentry->d_parent != new_dentry->d_parent) {
9181 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9182 BTRFS_I(old_inode), 1);
9183 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9184 BTRFS_I(new_inode), 1);
9187 /* src is a subvolume */
9188 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9189 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9190 } else { /* src is an inode */
9191 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9192 BTRFS_I(old_dentry->d_inode),
9193 old_dentry->d_name.name,
9194 old_dentry->d_name.len,
9197 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9200 btrfs_abort_transaction(trans, ret);
9204 /* dest is a subvolume */
9205 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9206 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9207 } else { /* dest is an inode */
9208 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9209 BTRFS_I(new_dentry->d_inode),
9210 new_dentry->d_name.name,
9211 new_dentry->d_name.len,
9214 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9217 btrfs_abort_transaction(trans, ret);
9221 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9222 new_dentry->d_name.name,
9223 new_dentry->d_name.len, 0, old_idx);
9225 btrfs_abort_transaction(trans, ret);
9229 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9230 old_dentry->d_name.name,
9231 old_dentry->d_name.len, 0, new_idx);
9233 btrfs_abort_transaction(trans, ret);
9237 if (old_inode->i_nlink == 1)
9238 BTRFS_I(old_inode)->dir_index = old_idx;
9239 if (new_inode->i_nlink == 1)
9240 BTRFS_I(new_inode)->dir_index = new_idx;
9243 * Now pin the logs of the roots. We do it to ensure that no other task
9244 * can sync the logs while we are in progress with the rename, because
9245 * that could result in an inconsistency in case any of the inodes that
9246 * are part of this rename operation were logged before.
9248 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9249 btrfs_pin_log_trans(root);
9250 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9251 btrfs_pin_log_trans(dest);
9253 /* Do the log updates for all inodes. */
9254 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9255 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9256 old_rename_ctx.index, new_dentry->d_parent);
9257 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9258 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9259 new_rename_ctx.index, old_dentry->d_parent);
9261 /* Now unpin the logs. */
9262 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9263 btrfs_end_log_trans(root);
9264 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9265 btrfs_end_log_trans(dest);
9267 ret2 = btrfs_end_transaction(trans);
9268 ret = ret ? ret : ret2;
9270 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9271 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9272 up_read(&fs_info->subvol_sem);
9277 static struct inode *new_whiteout_inode(struct user_namespace *mnt_userns,
9280 struct inode *inode;
9282 inode = new_inode(dir->i_sb);
9284 inode_init_owner(mnt_userns, inode, dir,
9285 S_IFCHR | WHITEOUT_MODE);
9286 inode->i_op = &btrfs_special_inode_operations;
9287 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9292 static int btrfs_rename(struct user_namespace *mnt_userns,
9293 struct inode *old_dir, struct dentry *old_dentry,
9294 struct inode *new_dir, struct dentry *new_dentry,
9297 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9298 struct btrfs_new_inode_args whiteout_args = {
9300 .dentry = old_dentry,
9302 struct btrfs_trans_handle *trans;
9303 unsigned int trans_num_items;
9304 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9305 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9306 struct inode *new_inode = d_inode(new_dentry);
9307 struct inode *old_inode = d_inode(old_dentry);
9308 struct btrfs_rename_ctx rename_ctx;
9312 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9314 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9317 /* we only allow rename subvolume link between subvolumes */
9318 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9321 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9322 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9325 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9326 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9330 /* check for collisions, even if the name isn't there */
9331 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9332 new_dentry->d_name.name,
9333 new_dentry->d_name.len);
9336 if (ret == -EEXIST) {
9338 * eexist without a new_inode */
9339 if (WARN_ON(!new_inode)) {
9343 /* maybe -EOVERFLOW */
9350 * we're using rename to replace one file with another. Start IO on it
9351 * now so we don't add too much work to the end of the transaction
9353 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9354 filemap_flush(old_inode->i_mapping);
9356 if (flags & RENAME_WHITEOUT) {
9357 whiteout_args.inode = new_whiteout_inode(mnt_userns, old_dir);
9358 if (!whiteout_args.inode)
9360 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9362 goto out_whiteout_inode;
9364 /* 1 to update the old parent inode. */
9365 trans_num_items = 1;
9368 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9369 /* Close the race window with snapshot create/destroy ioctl */
9370 down_read(&fs_info->subvol_sem);
9372 * 1 to remove old root ref
9373 * 1 to remove old root backref
9374 * 1 to add new root ref
9375 * 1 to add new root backref
9377 trans_num_items += 4;
9381 * 1 to remove old inode ref
9382 * 1 to add new inode ref
9384 trans_num_items += 3;
9387 * 1 to remove old dir item
9388 * 1 to remove old dir index
9389 * 1 to add new dir item
9390 * 1 to add new dir index
9392 trans_num_items += 4;
9393 /* 1 to update new parent inode if it's not the same as the old parent */
9394 if (new_dir != old_dir)
9399 * 1 to remove inode ref
9400 * 1 to remove dir item
9401 * 1 to remove dir index
9402 * 1 to possibly add orphan item
9404 trans_num_items += 5;
9406 trans = btrfs_start_transaction(root, trans_num_items);
9407 if (IS_ERR(trans)) {
9408 ret = PTR_ERR(trans);
9413 ret = btrfs_record_root_in_trans(trans, dest);
9418 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9422 BTRFS_I(old_inode)->dir_index = 0ULL;
9423 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9424 /* force full log commit if subvolume involved. */
9425 btrfs_set_log_full_commit(trans);
9427 ret = btrfs_insert_inode_ref(trans, dest,
9428 new_dentry->d_name.name,
9429 new_dentry->d_name.len,
9431 btrfs_ino(BTRFS_I(new_dir)), index);
9436 inode_inc_iversion(old_dir);
9437 inode_inc_iversion(new_dir);
9438 inode_inc_iversion(old_inode);
9439 old_dir->i_ctime = old_dir->i_mtime =
9440 new_dir->i_ctime = new_dir->i_mtime =
9441 old_inode->i_ctime = current_time(old_dir);
9443 if (old_dentry->d_parent != new_dentry->d_parent)
9444 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9445 BTRFS_I(old_inode), 1);
9447 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9448 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9450 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9451 BTRFS_I(d_inode(old_dentry)),
9452 old_dentry->d_name.name,
9453 old_dentry->d_name.len,
9456 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9459 btrfs_abort_transaction(trans, ret);
9464 inode_inc_iversion(new_inode);
9465 new_inode->i_ctime = current_time(new_inode);
9466 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9467 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9468 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9469 BUG_ON(new_inode->i_nlink == 0);
9471 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9472 BTRFS_I(d_inode(new_dentry)),
9473 new_dentry->d_name.name,
9474 new_dentry->d_name.len);
9476 if (!ret && new_inode->i_nlink == 0)
9477 ret = btrfs_orphan_add(trans,
9478 BTRFS_I(d_inode(new_dentry)));
9480 btrfs_abort_transaction(trans, ret);
9485 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9486 new_dentry->d_name.name,
9487 new_dentry->d_name.len, 0, index);
9489 btrfs_abort_transaction(trans, ret);
9493 if (old_inode->i_nlink == 1)
9494 BTRFS_I(old_inode)->dir_index = index;
9496 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9497 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9498 rename_ctx.index, new_dentry->d_parent);
9500 if (flags & RENAME_WHITEOUT) {
9501 ret = btrfs_create_new_inode(trans, &whiteout_args);
9503 btrfs_abort_transaction(trans, ret);
9506 unlock_new_inode(whiteout_args.inode);
9507 iput(whiteout_args.inode);
9508 whiteout_args.inode = NULL;
9512 ret2 = btrfs_end_transaction(trans);
9513 ret = ret ? ret : ret2;
9515 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9516 up_read(&fs_info->subvol_sem);
9517 if (flags & RENAME_WHITEOUT)
9518 btrfs_new_inode_args_destroy(&whiteout_args);
9520 if (flags & RENAME_WHITEOUT)
9521 iput(whiteout_args.inode);
9525 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9526 struct dentry *old_dentry, struct inode *new_dir,
9527 struct dentry *new_dentry, unsigned int flags)
9529 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9532 if (flags & RENAME_EXCHANGE)
9533 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9536 return btrfs_rename(mnt_userns, old_dir, old_dentry, new_dir,
9540 struct btrfs_delalloc_work {
9541 struct inode *inode;
9542 struct completion completion;
9543 struct list_head list;
9544 struct btrfs_work work;
9547 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9549 struct btrfs_delalloc_work *delalloc_work;
9550 struct inode *inode;
9552 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9554 inode = delalloc_work->inode;
9555 filemap_flush(inode->i_mapping);
9556 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9557 &BTRFS_I(inode)->runtime_flags))
9558 filemap_flush(inode->i_mapping);
9561 complete(&delalloc_work->completion);
9564 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9566 struct btrfs_delalloc_work *work;
9568 work = kmalloc(sizeof(*work), GFP_NOFS);
9572 init_completion(&work->completion);
9573 INIT_LIST_HEAD(&work->list);
9574 work->inode = inode;
9575 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9581 * some fairly slow code that needs optimization. This walks the list
9582 * of all the inodes with pending delalloc and forces them to disk.
9584 static int start_delalloc_inodes(struct btrfs_root *root,
9585 struct writeback_control *wbc, bool snapshot,
9586 bool in_reclaim_context)
9588 struct btrfs_inode *binode;
9589 struct inode *inode;
9590 struct btrfs_delalloc_work *work, *next;
9591 struct list_head works;
9592 struct list_head splice;
9594 bool full_flush = wbc->nr_to_write == LONG_MAX;
9596 INIT_LIST_HEAD(&works);
9597 INIT_LIST_HEAD(&splice);
9599 mutex_lock(&root->delalloc_mutex);
9600 spin_lock(&root->delalloc_lock);
9601 list_splice_init(&root->delalloc_inodes, &splice);
9602 while (!list_empty(&splice)) {
9603 binode = list_entry(splice.next, struct btrfs_inode,
9606 list_move_tail(&binode->delalloc_inodes,
9607 &root->delalloc_inodes);
9609 if (in_reclaim_context &&
9610 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9613 inode = igrab(&binode->vfs_inode);
9615 cond_resched_lock(&root->delalloc_lock);
9618 spin_unlock(&root->delalloc_lock);
9621 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9622 &binode->runtime_flags);
9624 work = btrfs_alloc_delalloc_work(inode);
9630 list_add_tail(&work->list, &works);
9631 btrfs_queue_work(root->fs_info->flush_workers,
9634 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9635 btrfs_add_delayed_iput(inode);
9636 if (ret || wbc->nr_to_write <= 0)
9640 spin_lock(&root->delalloc_lock);
9642 spin_unlock(&root->delalloc_lock);
9645 list_for_each_entry_safe(work, next, &works, list) {
9646 list_del_init(&work->list);
9647 wait_for_completion(&work->completion);
9651 if (!list_empty(&splice)) {
9652 spin_lock(&root->delalloc_lock);
9653 list_splice_tail(&splice, &root->delalloc_inodes);
9654 spin_unlock(&root->delalloc_lock);
9656 mutex_unlock(&root->delalloc_mutex);
9660 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9662 struct writeback_control wbc = {
9663 .nr_to_write = LONG_MAX,
9664 .sync_mode = WB_SYNC_NONE,
9666 .range_end = LLONG_MAX,
9668 struct btrfs_fs_info *fs_info = root->fs_info;
9670 if (BTRFS_FS_ERROR(fs_info))
9673 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9676 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9677 bool in_reclaim_context)
9679 struct writeback_control wbc = {
9681 .sync_mode = WB_SYNC_NONE,
9683 .range_end = LLONG_MAX,
9685 struct btrfs_root *root;
9686 struct list_head splice;
9689 if (BTRFS_FS_ERROR(fs_info))
9692 INIT_LIST_HEAD(&splice);
9694 mutex_lock(&fs_info->delalloc_root_mutex);
9695 spin_lock(&fs_info->delalloc_root_lock);
9696 list_splice_init(&fs_info->delalloc_roots, &splice);
9697 while (!list_empty(&splice)) {
9699 * Reset nr_to_write here so we know that we're doing a full
9703 wbc.nr_to_write = LONG_MAX;
9705 root = list_first_entry(&splice, struct btrfs_root,
9707 root = btrfs_grab_root(root);
9709 list_move_tail(&root->delalloc_root,
9710 &fs_info->delalloc_roots);
9711 spin_unlock(&fs_info->delalloc_root_lock);
9713 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9714 btrfs_put_root(root);
9715 if (ret < 0 || wbc.nr_to_write <= 0)
9717 spin_lock(&fs_info->delalloc_root_lock);
9719 spin_unlock(&fs_info->delalloc_root_lock);
9723 if (!list_empty(&splice)) {
9724 spin_lock(&fs_info->delalloc_root_lock);
9725 list_splice_tail(&splice, &fs_info->delalloc_roots);
9726 spin_unlock(&fs_info->delalloc_root_lock);
9728 mutex_unlock(&fs_info->delalloc_root_mutex);
9732 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9733 struct dentry *dentry, const char *symname)
9735 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9736 struct btrfs_trans_handle *trans;
9737 struct btrfs_root *root = BTRFS_I(dir)->root;
9738 struct btrfs_path *path;
9739 struct btrfs_key key;
9740 struct inode *inode;
9741 struct btrfs_new_inode_args new_inode_args = {
9745 unsigned int trans_num_items;
9750 struct btrfs_file_extent_item *ei;
9751 struct extent_buffer *leaf;
9753 name_len = strlen(symname);
9754 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9755 return -ENAMETOOLONG;
9757 inode = new_inode(dir->i_sb);
9760 inode_init_owner(mnt_userns, inode, dir, S_IFLNK | S_IRWXUGO);
9761 inode->i_op = &btrfs_symlink_inode_operations;
9762 inode_nohighmem(inode);
9763 inode->i_mapping->a_ops = &btrfs_aops;
9764 btrfs_i_size_write(BTRFS_I(inode), name_len);
9765 inode_set_bytes(inode, name_len);
9767 new_inode_args.inode = inode;
9768 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9771 /* 1 additional item for the inline extent */
9774 trans = btrfs_start_transaction(root, trans_num_items);
9775 if (IS_ERR(trans)) {
9776 err = PTR_ERR(trans);
9777 goto out_new_inode_args;
9780 err = btrfs_create_new_inode(trans, &new_inode_args);
9784 path = btrfs_alloc_path();
9787 btrfs_abort_transaction(trans, err);
9788 discard_new_inode(inode);
9792 key.objectid = btrfs_ino(BTRFS_I(inode));
9794 key.type = BTRFS_EXTENT_DATA_KEY;
9795 datasize = btrfs_file_extent_calc_inline_size(name_len);
9796 err = btrfs_insert_empty_item(trans, root, path, &key,
9799 btrfs_abort_transaction(trans, err);
9800 btrfs_free_path(path);
9801 discard_new_inode(inode);
9805 leaf = path->nodes[0];
9806 ei = btrfs_item_ptr(leaf, path->slots[0],
9807 struct btrfs_file_extent_item);
9808 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9809 btrfs_set_file_extent_type(leaf, ei,
9810 BTRFS_FILE_EXTENT_INLINE);
9811 btrfs_set_file_extent_encryption(leaf, ei, 0);
9812 btrfs_set_file_extent_compression(leaf, ei, 0);
9813 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9814 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9816 ptr = btrfs_file_extent_inline_start(ei);
9817 write_extent_buffer(leaf, symname, ptr, name_len);
9818 btrfs_mark_buffer_dirty(leaf);
9819 btrfs_free_path(path);
9821 d_instantiate_new(dentry, inode);
9824 btrfs_end_transaction(trans);
9825 btrfs_btree_balance_dirty(fs_info);
9827 btrfs_new_inode_args_destroy(&new_inode_args);
9834 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9835 struct btrfs_trans_handle *trans_in,
9836 struct btrfs_inode *inode,
9837 struct btrfs_key *ins,
9840 struct btrfs_file_extent_item stack_fi;
9841 struct btrfs_replace_extent_info extent_info;
9842 struct btrfs_trans_handle *trans = trans_in;
9843 struct btrfs_path *path;
9844 u64 start = ins->objectid;
9845 u64 len = ins->offset;
9846 int qgroup_released;
9849 memset(&stack_fi, 0, sizeof(stack_fi));
9851 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9852 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9853 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9854 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9855 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9856 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9857 /* Encryption and other encoding is reserved and all 0 */
9859 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9860 if (qgroup_released < 0)
9861 return ERR_PTR(qgroup_released);
9864 ret = insert_reserved_file_extent(trans, inode,
9865 file_offset, &stack_fi,
9866 true, qgroup_released);
9872 extent_info.disk_offset = start;
9873 extent_info.disk_len = len;
9874 extent_info.data_offset = 0;
9875 extent_info.data_len = len;
9876 extent_info.file_offset = file_offset;
9877 extent_info.extent_buf = (char *)&stack_fi;
9878 extent_info.is_new_extent = true;
9879 extent_info.qgroup_reserved = qgroup_released;
9880 extent_info.insertions = 0;
9882 path = btrfs_alloc_path();
9888 ret = btrfs_replace_file_extents(inode, path, file_offset,
9889 file_offset + len - 1, &extent_info,
9891 btrfs_free_path(path);
9898 * We have released qgroup data range at the beginning of the function,
9899 * and normally qgroup_released bytes will be freed when committing
9901 * But if we error out early, we have to free what we have released
9902 * or we leak qgroup data reservation.
9904 btrfs_qgroup_free_refroot(inode->root->fs_info,
9905 inode->root->root_key.objectid, qgroup_released,
9906 BTRFS_QGROUP_RSV_DATA);
9907 return ERR_PTR(ret);
9910 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9911 u64 start, u64 num_bytes, u64 min_size,
9912 loff_t actual_len, u64 *alloc_hint,
9913 struct btrfs_trans_handle *trans)
9915 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9916 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9917 struct extent_map *em;
9918 struct btrfs_root *root = BTRFS_I(inode)->root;
9919 struct btrfs_key ins;
9920 u64 cur_offset = start;
9921 u64 clear_offset = start;
9924 u64 last_alloc = (u64)-1;
9926 bool own_trans = true;
9927 u64 end = start + num_bytes - 1;
9931 while (num_bytes > 0) {
9932 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9933 cur_bytes = max(cur_bytes, min_size);
9935 * If we are severely fragmented we could end up with really
9936 * small allocations, so if the allocator is returning small
9937 * chunks lets make its job easier by only searching for those
9940 cur_bytes = min(cur_bytes, last_alloc);
9941 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9942 min_size, 0, *alloc_hint, &ins, 1, 0);
9947 * We've reserved this space, and thus converted it from
9948 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9949 * from here on out we will only need to clear our reservation
9950 * for the remaining unreserved area, so advance our
9951 * clear_offset by our extent size.
9953 clear_offset += ins.offset;
9955 last_alloc = ins.offset;
9956 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9959 * Now that we inserted the prealloc extent we can finally
9960 * decrement the number of reservations in the block group.
9961 * If we did it before, we could race with relocation and have
9962 * relocation miss the reserved extent, making it fail later.
9964 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9965 if (IS_ERR(trans)) {
9966 ret = PTR_ERR(trans);
9967 btrfs_free_reserved_extent(fs_info, ins.objectid,
9972 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9973 cur_offset + ins.offset -1, 0);
9975 em = alloc_extent_map();
9977 btrfs_set_inode_full_sync(BTRFS_I(inode));
9981 em->start = cur_offset;
9982 em->orig_start = cur_offset;
9983 em->len = ins.offset;
9984 em->block_start = ins.objectid;
9985 em->block_len = ins.offset;
9986 em->orig_block_len = ins.offset;
9987 em->ram_bytes = ins.offset;
9988 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9989 em->generation = trans->transid;
9992 write_lock(&em_tree->lock);
9993 ret = add_extent_mapping(em_tree, em, 1);
9994 write_unlock(&em_tree->lock);
9997 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9998 cur_offset + ins.offset - 1,
10001 free_extent_map(em);
10003 num_bytes -= ins.offset;
10004 cur_offset += ins.offset;
10005 *alloc_hint = ins.objectid + ins.offset;
10007 inode_inc_iversion(inode);
10008 inode->i_ctime = current_time(inode);
10009 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10010 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10011 (actual_len > inode->i_size) &&
10012 (cur_offset > inode->i_size)) {
10013 if (cur_offset > actual_len)
10014 i_size = actual_len;
10016 i_size = cur_offset;
10017 i_size_write(inode, i_size);
10018 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10021 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10024 btrfs_abort_transaction(trans, ret);
10026 btrfs_end_transaction(trans);
10031 btrfs_end_transaction(trans);
10035 if (clear_offset < end)
10036 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10037 end - clear_offset + 1);
10041 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10042 u64 start, u64 num_bytes, u64 min_size,
10043 loff_t actual_len, u64 *alloc_hint)
10045 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10046 min_size, actual_len, alloc_hint,
10050 int btrfs_prealloc_file_range_trans(struct inode *inode,
10051 struct btrfs_trans_handle *trans, int mode,
10052 u64 start, u64 num_bytes, u64 min_size,
10053 loff_t actual_len, u64 *alloc_hint)
10055 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10056 min_size, actual_len, alloc_hint, trans);
10059 static int btrfs_permission(struct user_namespace *mnt_userns,
10060 struct inode *inode, int mask)
10062 struct btrfs_root *root = BTRFS_I(inode)->root;
10063 umode_t mode = inode->i_mode;
10065 if (mask & MAY_WRITE &&
10066 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10067 if (btrfs_root_readonly(root))
10069 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10072 return generic_permission(mnt_userns, inode, mask);
10075 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10076 struct dentry *dentry, umode_t mode)
10078 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10079 struct btrfs_trans_handle *trans;
10080 struct btrfs_root *root = BTRFS_I(dir)->root;
10081 struct inode *inode;
10082 struct btrfs_new_inode_args new_inode_args = {
10087 unsigned int trans_num_items;
10090 inode = new_inode(dir->i_sb);
10093 inode_init_owner(mnt_userns, inode, dir, mode);
10094 inode->i_fop = &btrfs_file_operations;
10095 inode->i_op = &btrfs_file_inode_operations;
10096 inode->i_mapping->a_ops = &btrfs_aops;
10098 new_inode_args.inode = inode;
10099 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
10103 trans = btrfs_start_transaction(root, trans_num_items);
10104 if (IS_ERR(trans)) {
10105 ret = PTR_ERR(trans);
10106 goto out_new_inode_args;
10109 ret = btrfs_create_new_inode(trans, &new_inode_args);
10112 * We set number of links to 0 in btrfs_create_new_inode(), and here we
10113 * set it to 1 because d_tmpfile() will issue a warning if the count is
10116 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10118 set_nlink(inode, 1);
10121 d_tmpfile(dentry, inode);
10122 unlock_new_inode(inode);
10123 mark_inode_dirty(inode);
10126 btrfs_end_transaction(trans);
10127 btrfs_btree_balance_dirty(fs_info);
10128 out_new_inode_args:
10129 btrfs_new_inode_args_destroy(&new_inode_args);
10136 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
10138 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10139 unsigned long index = start >> PAGE_SHIFT;
10140 unsigned long end_index = end >> PAGE_SHIFT;
10144 ASSERT(end + 1 - start <= U32_MAX);
10145 len = end + 1 - start;
10146 while (index <= end_index) {
10147 page = find_get_page(inode->vfs_inode.i_mapping, index);
10148 ASSERT(page); /* Pages should be in the extent_io_tree */
10150 btrfs_page_set_writeback(fs_info, page, start, len);
10156 static int btrfs_encoded_io_compression_from_extent(
10157 struct btrfs_fs_info *fs_info,
10160 switch (compress_type) {
10161 case BTRFS_COMPRESS_NONE:
10162 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
10163 case BTRFS_COMPRESS_ZLIB:
10164 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
10165 case BTRFS_COMPRESS_LZO:
10167 * The LZO format depends on the sector size. 64K is the maximum
10168 * sector size that we support.
10170 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
10172 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
10173 (fs_info->sectorsize_bits - 12);
10174 case BTRFS_COMPRESS_ZSTD:
10175 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
10181 static ssize_t btrfs_encoded_read_inline(
10182 struct kiocb *iocb,
10183 struct iov_iter *iter, u64 start,
10185 struct extent_state **cached_state,
10186 u64 extent_start, size_t count,
10187 struct btrfs_ioctl_encoded_io_args *encoded,
10190 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10191 struct btrfs_root *root = inode->root;
10192 struct btrfs_fs_info *fs_info = root->fs_info;
10193 struct extent_io_tree *io_tree = &inode->io_tree;
10194 struct btrfs_path *path;
10195 struct extent_buffer *leaf;
10196 struct btrfs_file_extent_item *item;
10202 path = btrfs_alloc_path();
10207 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10211 /* The extent item disappeared? */
10216 leaf = path->nodes[0];
10217 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10219 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10220 ptr = btrfs_file_extent_inline_start(item);
10222 encoded->len = min_t(u64, extent_start + ram_bytes,
10223 inode->vfs_inode.i_size) - iocb->ki_pos;
10224 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10225 btrfs_file_extent_compression(leaf, item));
10228 encoded->compression = ret;
10229 if (encoded->compression) {
10230 size_t inline_size;
10232 inline_size = btrfs_file_extent_inline_item_len(leaf,
10234 if (inline_size > count) {
10238 count = inline_size;
10239 encoded->unencoded_len = ram_bytes;
10240 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10242 count = min_t(u64, count, encoded->len);
10243 encoded->len = count;
10244 encoded->unencoded_len = count;
10245 ptr += iocb->ki_pos - extent_start;
10248 tmp = kmalloc(count, GFP_NOFS);
10253 read_extent_buffer(leaf, tmp, ptr, count);
10254 btrfs_release_path(path);
10255 unlock_extent_cached(io_tree, start, lockend, cached_state);
10256 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10259 ret = copy_to_iter(tmp, count, iter);
10264 btrfs_free_path(path);
10268 struct btrfs_encoded_read_private {
10269 struct btrfs_inode *inode;
10271 wait_queue_head_t wait;
10273 blk_status_t status;
10277 static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
10278 struct bio *bio, int mirror_num)
10280 struct btrfs_encoded_read_private *priv = bio->bi_private;
10281 struct btrfs_bio *bbio = btrfs_bio(bio);
10282 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10285 if (!priv->skip_csum) {
10286 ret = btrfs_lookup_bio_sums(&inode->vfs_inode, bio, NULL);
10291 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
10293 btrfs_bio_free_csum(bbio);
10297 atomic_inc(&priv->pending);
10298 ret = btrfs_map_bio(fs_info, bio, mirror_num);
10300 atomic_dec(&priv->pending);
10301 btrfs_bio_free_csum(bbio);
10306 static blk_status_t btrfs_encoded_read_verify_csum(struct btrfs_bio *bbio)
10308 const bool uptodate = (bbio->bio.bi_status == BLK_STS_OK);
10309 struct btrfs_encoded_read_private *priv = bbio->bio.bi_private;
10310 struct btrfs_inode *inode = priv->inode;
10311 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10312 u32 sectorsize = fs_info->sectorsize;
10313 struct bio_vec *bvec;
10314 struct bvec_iter_all iter_all;
10315 u64 start = priv->file_offset;
10316 u32 bio_offset = 0;
10318 if (priv->skip_csum || !uptodate)
10319 return bbio->bio.bi_status;
10321 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
10322 unsigned int i, nr_sectors, pgoff;
10324 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
10325 pgoff = bvec->bv_offset;
10326 for (i = 0; i < nr_sectors; i++) {
10327 ASSERT(pgoff < PAGE_SIZE);
10328 if (check_data_csum(&inode->vfs_inode, bbio, bio_offset,
10329 bvec->bv_page, pgoff, start))
10330 return BLK_STS_IOERR;
10331 start += sectorsize;
10332 bio_offset += sectorsize;
10333 pgoff += sectorsize;
10339 static void btrfs_encoded_read_endio(struct bio *bio)
10341 struct btrfs_encoded_read_private *priv = bio->bi_private;
10342 struct btrfs_bio *bbio = btrfs_bio(bio);
10343 blk_status_t status;
10345 status = btrfs_encoded_read_verify_csum(bbio);
10348 * The memory barrier implied by the atomic_dec_return() here
10349 * pairs with the memory barrier implied by the
10350 * atomic_dec_return() or io_wait_event() in
10351 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10352 * write is observed before the load of status in
10353 * btrfs_encoded_read_regular_fill_pages().
10355 WRITE_ONCE(priv->status, status);
10357 if (!atomic_dec_return(&priv->pending))
10358 wake_up(&priv->wait);
10359 btrfs_bio_free_csum(bbio);
10363 static int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10367 struct page **pages)
10369 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10370 struct btrfs_encoded_read_private priv = {
10372 .file_offset = file_offset,
10373 .pending = ATOMIC_INIT(1),
10374 .skip_csum = (inode->flags & BTRFS_INODE_NODATASUM),
10376 unsigned long i = 0;
10380 init_waitqueue_head(&priv.wait);
10382 * Submit bios for the extent, splitting due to bio or stripe limits as
10385 while (cur < disk_io_size) {
10386 struct extent_map *em;
10387 struct btrfs_io_geometry geom;
10388 struct bio *bio = NULL;
10391 em = btrfs_get_chunk_map(fs_info, disk_bytenr + cur,
10392 disk_io_size - cur);
10396 ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
10397 disk_bytenr + cur, &geom);
10398 free_extent_map(em);
10401 WRITE_ONCE(priv.status, errno_to_blk_status(ret));
10404 remaining = min(geom.len, disk_io_size - cur);
10405 while (bio || remaining) {
10406 size_t bytes = min_t(u64, remaining, PAGE_SIZE);
10409 bio = btrfs_bio_alloc(BIO_MAX_VECS);
10410 bio->bi_iter.bi_sector =
10411 (disk_bytenr + cur) >> SECTOR_SHIFT;
10412 bio->bi_end_io = btrfs_encoded_read_endio;
10413 bio->bi_private = &priv;
10414 bio->bi_opf = REQ_OP_READ;
10418 bio_add_page(bio, pages[i], bytes, 0) < bytes) {
10419 blk_status_t status;
10421 status = submit_encoded_read_bio(inode, bio, 0);
10423 WRITE_ONCE(priv.status, status);
10433 remaining -= bytes;
10438 if (atomic_dec_return(&priv.pending))
10439 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10440 /* See btrfs_encoded_read_endio() for ordering. */
10441 return blk_status_to_errno(READ_ONCE(priv.status));
10444 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10445 struct iov_iter *iter,
10446 u64 start, u64 lockend,
10447 struct extent_state **cached_state,
10448 u64 disk_bytenr, u64 disk_io_size,
10449 size_t count, bool compressed,
10452 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10453 struct extent_io_tree *io_tree = &inode->io_tree;
10454 struct page **pages;
10455 unsigned long nr_pages, i;
10457 size_t page_offset;
10460 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10461 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10464 for (i = 0; i < nr_pages; i++) {
10465 pages[i] = alloc_page(GFP_NOFS);
10472 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10473 disk_io_size, pages);
10477 unlock_extent_cached(io_tree, start, lockend, cached_state);
10478 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10485 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10486 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10489 while (cur < count) {
10490 size_t bytes = min_t(size_t, count - cur,
10491 PAGE_SIZE - page_offset);
10493 if (copy_page_to_iter(pages[i], page_offset, bytes,
10504 for (i = 0; i < nr_pages; i++) {
10506 __free_page(pages[i]);
10512 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10513 struct btrfs_ioctl_encoded_io_args *encoded)
10515 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10516 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10517 struct extent_io_tree *io_tree = &inode->io_tree;
10519 size_t count = iov_iter_count(iter);
10520 u64 start, lockend, disk_bytenr, disk_io_size;
10521 struct extent_state *cached_state = NULL;
10522 struct extent_map *em;
10523 bool unlocked = false;
10525 file_accessed(iocb->ki_filp);
10527 btrfs_inode_lock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10529 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10530 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10533 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10535 * We don't know how long the extent containing iocb->ki_pos is, but if
10536 * it's compressed we know that it won't be longer than this.
10538 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10541 struct btrfs_ordered_extent *ordered;
10543 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10544 lockend - start + 1);
10546 goto out_unlock_inode;
10547 lock_extent_bits(io_tree, start, lockend, &cached_state);
10548 ordered = btrfs_lookup_ordered_range(inode, start,
10549 lockend - start + 1);
10552 btrfs_put_ordered_extent(ordered);
10553 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10557 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10560 goto out_unlock_extent;
10563 if (em->block_start == EXTENT_MAP_INLINE) {
10564 u64 extent_start = em->start;
10567 * For inline extents we get everything we need out of the
10570 free_extent_map(em);
10572 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10573 &cached_state, extent_start,
10574 count, encoded, &unlocked);
10579 * We only want to return up to EOF even if the extent extends beyond
10582 encoded->len = min_t(u64, extent_map_end(em),
10583 inode->vfs_inode.i_size) - iocb->ki_pos;
10584 if (em->block_start == EXTENT_MAP_HOLE ||
10585 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10586 disk_bytenr = EXTENT_MAP_HOLE;
10587 count = min_t(u64, count, encoded->len);
10588 encoded->len = count;
10589 encoded->unencoded_len = count;
10590 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10591 disk_bytenr = em->block_start;
10593 * Bail if the buffer isn't large enough to return the whole
10594 * compressed extent.
10596 if (em->block_len > count) {
10600 disk_io_size = count = em->block_len;
10601 encoded->unencoded_len = em->ram_bytes;
10602 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10603 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10604 em->compress_type);
10607 encoded->compression = ret;
10609 disk_bytenr = em->block_start + (start - em->start);
10610 if (encoded->len > count)
10611 encoded->len = count;
10613 * Don't read beyond what we locked. This also limits the page
10614 * allocations that we'll do.
10616 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10617 count = start + disk_io_size - iocb->ki_pos;
10618 encoded->len = count;
10619 encoded->unencoded_len = count;
10620 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10622 free_extent_map(em);
10625 if (disk_bytenr == EXTENT_MAP_HOLE) {
10626 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10627 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10629 ret = iov_iter_zero(count, iter);
10633 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10634 &cached_state, disk_bytenr,
10635 disk_io_size, count,
10636 encoded->compression,
10642 iocb->ki_pos += encoded->len;
10644 free_extent_map(em);
10647 unlock_extent_cached(io_tree, start, lockend, &cached_state);
10650 btrfs_inode_unlock(&inode->vfs_inode, BTRFS_ILOCK_SHARED);
10654 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10655 const struct btrfs_ioctl_encoded_io_args *encoded)
10657 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10658 struct btrfs_root *root = inode->root;
10659 struct btrfs_fs_info *fs_info = root->fs_info;
10660 struct extent_io_tree *io_tree = &inode->io_tree;
10661 struct extent_changeset *data_reserved = NULL;
10662 struct extent_state *cached_state = NULL;
10666 u64 num_bytes, ram_bytes, disk_num_bytes;
10667 unsigned long nr_pages, i;
10668 struct page **pages;
10669 struct btrfs_key ins;
10670 bool extent_reserved = false;
10671 struct extent_map *em;
10674 switch (encoded->compression) {
10675 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10676 compression = BTRFS_COMPRESS_ZLIB;
10678 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10679 compression = BTRFS_COMPRESS_ZSTD;
10681 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10682 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10683 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10684 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10685 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10686 /* The sector size must match for LZO. */
10687 if (encoded->compression -
10688 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10689 fs_info->sectorsize_bits)
10691 compression = BTRFS_COMPRESS_LZO;
10696 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10699 orig_count = iov_iter_count(from);
10701 /* The extent size must be sane. */
10702 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10703 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10707 * The compressed data must be smaller than the decompressed data.
10709 * It's of course possible for data to compress to larger or the same
10710 * size, but the buffered I/O path falls back to no compression for such
10711 * data, and we don't want to break any assumptions by creating these
10714 * Note that this is less strict than the current check we have that the
10715 * compressed data must be at least one sector smaller than the
10716 * decompressed data. We only want to enforce the weaker requirement
10717 * from old kernels that it is at least one byte smaller.
10719 if (orig_count >= encoded->unencoded_len)
10722 /* The extent must start on a sector boundary. */
10723 start = iocb->ki_pos;
10724 if (!IS_ALIGNED(start, fs_info->sectorsize))
10728 * The extent must end on a sector boundary. However, we allow a write
10729 * which ends at or extends i_size to have an unaligned length; we round
10730 * up the extent size and set i_size to the unaligned end.
10732 if (start + encoded->len < inode->vfs_inode.i_size &&
10733 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10736 /* Finally, the offset in the unencoded data must be sector-aligned. */
10737 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10740 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10741 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10742 end = start + num_bytes - 1;
10745 * If the extent cannot be inline, the compressed data on disk must be
10746 * sector-aligned. For convenience, we extend it with zeroes if it
10749 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10750 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10751 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10754 for (i = 0; i < nr_pages; i++) {
10755 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10758 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10763 kaddr = kmap(pages[i]);
10764 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10769 if (bytes < PAGE_SIZE)
10770 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10775 struct btrfs_ordered_extent *ordered;
10777 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10780 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10781 start >> PAGE_SHIFT,
10782 end >> PAGE_SHIFT);
10785 lock_extent_bits(io_tree, start, end, &cached_state);
10786 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10788 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10791 btrfs_put_ordered_extent(ordered);
10792 unlock_extent_cached(io_tree, start, end, &cached_state);
10797 * We don't use the higher-level delalloc space functions because our
10798 * num_bytes and disk_num_bytes are different.
10800 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10803 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10805 goto out_free_data_space;
10806 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10809 goto out_qgroup_free_data;
10811 /* Try an inline extent first. */
10812 if (start == 0 && encoded->unencoded_len == encoded->len &&
10813 encoded->unencoded_offset == 0) {
10814 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10815 compression, pages, true);
10819 goto out_delalloc_release;
10823 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10824 disk_num_bytes, 0, 0, &ins, 1, 1);
10826 goto out_delalloc_release;
10827 extent_reserved = true;
10829 em = create_io_em(inode, start, num_bytes,
10830 start - encoded->unencoded_offset, ins.objectid,
10831 ins.offset, ins.offset, ram_bytes, compression,
10832 BTRFS_ORDERED_COMPRESSED);
10835 goto out_free_reserved;
10837 free_extent_map(em);
10839 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10840 ins.objectid, ins.offset,
10841 encoded->unencoded_offset,
10842 (1 << BTRFS_ORDERED_ENCODED) |
10843 (1 << BTRFS_ORDERED_COMPRESSED),
10846 btrfs_drop_extent_cache(inode, start, end, 0);
10847 goto out_free_reserved;
10849 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10851 if (start + encoded->len > inode->vfs_inode.i_size)
10852 i_size_write(&inode->vfs_inode, start + encoded->len);
10854 unlock_extent_cached(io_tree, start, end, &cached_state);
10856 btrfs_delalloc_release_extents(inode, num_bytes);
10858 if (btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10859 ins.offset, pages, nr_pages, 0, NULL,
10861 btrfs_writepage_endio_finish_ordered(inode, pages[0], start, end, 0);
10869 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10870 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10871 out_delalloc_release:
10872 btrfs_delalloc_release_extents(inode, num_bytes);
10873 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10874 out_qgroup_free_data:
10876 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10877 out_free_data_space:
10879 * If btrfs_reserve_extent() succeeded, then we already decremented
10882 if (!extent_reserved)
10883 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10885 unlock_extent_cached(io_tree, start, end, &cached_state);
10887 for (i = 0; i < nr_pages; i++) {
10889 __free_page(pages[i]);
10894 iocb->ki_pos += encoded->len;
10900 * Add an entry indicating a block group or device which is pinned by a
10901 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10902 * negative errno on failure.
10904 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10905 bool is_block_group)
10907 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10908 struct btrfs_swapfile_pin *sp, *entry;
10909 struct rb_node **p;
10910 struct rb_node *parent = NULL;
10912 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10917 sp->is_block_group = is_block_group;
10918 sp->bg_extent_count = 1;
10920 spin_lock(&fs_info->swapfile_pins_lock);
10921 p = &fs_info->swapfile_pins.rb_node;
10924 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10925 if (sp->ptr < entry->ptr ||
10926 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10927 p = &(*p)->rb_left;
10928 } else if (sp->ptr > entry->ptr ||
10929 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10930 p = &(*p)->rb_right;
10932 if (is_block_group)
10933 entry->bg_extent_count++;
10934 spin_unlock(&fs_info->swapfile_pins_lock);
10939 rb_link_node(&sp->node, parent, p);
10940 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10941 spin_unlock(&fs_info->swapfile_pins_lock);
10945 /* Free all of the entries pinned by this swapfile. */
10946 static void btrfs_free_swapfile_pins(struct inode *inode)
10948 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10949 struct btrfs_swapfile_pin *sp;
10950 struct rb_node *node, *next;
10952 spin_lock(&fs_info->swapfile_pins_lock);
10953 node = rb_first(&fs_info->swapfile_pins);
10955 next = rb_next(node);
10956 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10957 if (sp->inode == inode) {
10958 rb_erase(&sp->node, &fs_info->swapfile_pins);
10959 if (sp->is_block_group) {
10960 btrfs_dec_block_group_swap_extents(sp->ptr,
10961 sp->bg_extent_count);
10962 btrfs_put_block_group(sp->ptr);
10968 spin_unlock(&fs_info->swapfile_pins_lock);
10971 struct btrfs_swap_info {
10977 unsigned long nr_pages;
10981 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10982 struct btrfs_swap_info *bsi)
10984 unsigned long nr_pages;
10985 unsigned long max_pages;
10986 u64 first_ppage, first_ppage_reported, next_ppage;
10990 * Our swapfile may have had its size extended after the swap header was
10991 * written. In that case activating the swapfile should not go beyond
10992 * the max size set in the swap header.
10994 if (bsi->nr_pages >= sis->max)
10997 max_pages = sis->max - bsi->nr_pages;
10998 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10999 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
11000 PAGE_SIZE) >> PAGE_SHIFT;
11002 if (first_ppage >= next_ppage)
11004 nr_pages = next_ppage - first_ppage;
11005 nr_pages = min(nr_pages, max_pages);
11007 first_ppage_reported = first_ppage;
11008 if (bsi->start == 0)
11009 first_ppage_reported++;
11010 if (bsi->lowest_ppage > first_ppage_reported)
11011 bsi->lowest_ppage = first_ppage_reported;
11012 if (bsi->highest_ppage < (next_ppage - 1))
11013 bsi->highest_ppage = next_ppage - 1;
11015 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
11018 bsi->nr_extents += ret;
11019 bsi->nr_pages += nr_pages;
11023 static void btrfs_swap_deactivate(struct file *file)
11025 struct inode *inode = file_inode(file);
11027 btrfs_free_swapfile_pins(inode);
11028 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
11031 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11034 struct inode *inode = file_inode(file);
11035 struct btrfs_root *root = BTRFS_I(inode)->root;
11036 struct btrfs_fs_info *fs_info = root->fs_info;
11037 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
11038 struct extent_state *cached_state = NULL;
11039 struct extent_map *em = NULL;
11040 struct btrfs_device *device = NULL;
11041 struct btrfs_swap_info bsi = {
11042 .lowest_ppage = (sector_t)-1ULL,
11049 * If the swap file was just created, make sure delalloc is done. If the
11050 * file changes again after this, the user is doing something stupid and
11051 * we don't really care.
11053 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
11058 * The inode is locked, so these flags won't change after we check them.
11060 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
11061 btrfs_warn(fs_info, "swapfile must not be compressed");
11064 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
11065 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
11068 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
11069 btrfs_warn(fs_info, "swapfile must not be checksummed");
11074 * Balance or device remove/replace/resize can move stuff around from
11075 * under us. The exclop protection makes sure they aren't running/won't
11076 * run concurrently while we are mapping the swap extents, and
11077 * fs_info->swapfile_pins prevents them from running while the swap
11078 * file is active and moving the extents. Note that this also prevents
11079 * a concurrent device add which isn't actually necessary, but it's not
11080 * really worth the trouble to allow it.
11082 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
11083 btrfs_warn(fs_info,
11084 "cannot activate swapfile while exclusive operation is running");
11089 * Prevent snapshot creation while we are activating the swap file.
11090 * We do not want to race with snapshot creation. If snapshot creation
11091 * already started before we bumped nr_swapfiles from 0 to 1 and
11092 * completes before the first write into the swap file after it is
11093 * activated, than that write would fallback to COW.
11095 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
11096 btrfs_exclop_finish(fs_info);
11097 btrfs_warn(fs_info,
11098 "cannot activate swapfile because snapshot creation is in progress");
11102 * Snapshots can create extents which require COW even if NODATACOW is
11103 * set. We use this counter to prevent snapshots. We must increment it
11104 * before walking the extents because we don't want a concurrent
11105 * snapshot to run after we've already checked the extents.
11107 * It is possible that subvolume is marked for deletion but still not
11108 * removed yet. To prevent this race, we check the root status before
11109 * activating the swapfile.
11111 spin_lock(&root->root_item_lock);
11112 if (btrfs_root_dead(root)) {
11113 spin_unlock(&root->root_item_lock);
11115 btrfs_exclop_finish(fs_info);
11116 btrfs_warn(fs_info,
11117 "cannot activate swapfile because subvolume %llu is being deleted",
11118 root->root_key.objectid);
11121 atomic_inc(&root->nr_swapfiles);
11122 spin_unlock(&root->root_item_lock);
11124 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
11126 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
11128 while (start < isize) {
11129 u64 logical_block_start, physical_block_start;
11130 struct btrfs_block_group *bg;
11131 u64 len = isize - start;
11133 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
11139 if (em->block_start == EXTENT_MAP_HOLE) {
11140 btrfs_warn(fs_info, "swapfile must not have holes");
11144 if (em->block_start == EXTENT_MAP_INLINE) {
11146 * It's unlikely we'll ever actually find ourselves
11147 * here, as a file small enough to fit inline won't be
11148 * big enough to store more than the swap header, but in
11149 * case something changes in the future, let's catch it
11150 * here rather than later.
11152 btrfs_warn(fs_info, "swapfile must not be inline");
11156 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
11157 btrfs_warn(fs_info, "swapfile must not be compressed");
11162 logical_block_start = em->block_start + (start - em->start);
11163 len = min(len, em->len - (start - em->start));
11164 free_extent_map(em);
11167 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
11173 btrfs_warn(fs_info,
11174 "swapfile must not be copy-on-write");
11179 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
11185 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
11186 btrfs_warn(fs_info,
11187 "swapfile must have single data profile");
11192 if (device == NULL) {
11193 device = em->map_lookup->stripes[0].dev;
11194 ret = btrfs_add_swapfile_pin(inode, device, false);
11199 } else if (device != em->map_lookup->stripes[0].dev) {
11200 btrfs_warn(fs_info, "swapfile must be on one device");
11205 physical_block_start = (em->map_lookup->stripes[0].physical +
11206 (logical_block_start - em->start));
11207 len = min(len, em->len - (logical_block_start - em->start));
11208 free_extent_map(em);
11211 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
11213 btrfs_warn(fs_info,
11214 "could not find block group containing swapfile");
11219 if (!btrfs_inc_block_group_swap_extents(bg)) {
11220 btrfs_warn(fs_info,
11221 "block group for swapfile at %llu is read-only%s",
11223 atomic_read(&fs_info->scrubs_running) ?
11224 " (scrub running)" : "");
11225 btrfs_put_block_group(bg);
11230 ret = btrfs_add_swapfile_pin(inode, bg, true);
11232 btrfs_put_block_group(bg);
11239 if (bsi.block_len &&
11240 bsi.block_start + bsi.block_len == physical_block_start) {
11241 bsi.block_len += len;
11243 if (bsi.block_len) {
11244 ret = btrfs_add_swap_extent(sis, &bsi);
11249 bsi.block_start = physical_block_start;
11250 bsi.block_len = len;
11257 ret = btrfs_add_swap_extent(sis, &bsi);
11260 if (!IS_ERR_OR_NULL(em))
11261 free_extent_map(em);
11263 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
11266 btrfs_swap_deactivate(file);
11268 btrfs_drew_write_unlock(&root->snapshot_lock);
11270 btrfs_exclop_finish(fs_info);
11276 sis->bdev = device->bdev;
11277 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
11278 sis->max = bsi.nr_pages;
11279 sis->pages = bsi.nr_pages - 1;
11280 sis->highest_bit = bsi.nr_pages - 1;
11281 return bsi.nr_extents;
11284 static void btrfs_swap_deactivate(struct file *file)
11288 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
11291 return -EOPNOTSUPP;
11296 * Update the number of bytes used in the VFS' inode. When we replace extents in
11297 * a range (clone, dedupe, fallocate's zero range), we must update the number of
11298 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
11299 * always get a correct value.
11301 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
11302 const u64 add_bytes,
11303 const u64 del_bytes)
11305 if (add_bytes == del_bytes)
11308 spin_lock(&inode->lock);
11310 inode_sub_bytes(&inode->vfs_inode, del_bytes);
11312 inode_add_bytes(&inode->vfs_inode, add_bytes);
11313 spin_unlock(&inode->lock);
11317 * Verify that there are no ordered extents for a given file range.
11319 * @inode: The target inode.
11320 * @start: Start offset of the file range, should be sector size aligned.
11321 * @end: End offset (inclusive) of the file range, its value +1 should be
11322 * sector size aligned.
11324 * This should typically be used for cases where we locked an inode's VFS lock in
11325 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11326 * we have flushed all delalloc in the range, we have waited for all ordered
11327 * extents in the range to complete and finally we have locked the file range in
11328 * the inode's io_tree.
11330 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11332 struct btrfs_root *root = inode->root;
11333 struct btrfs_ordered_extent *ordered;
11335 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11338 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11340 btrfs_err(root->fs_info,
11341 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11342 start, end, btrfs_ino(inode), root->root_key.objectid,
11343 ordered->file_offset,
11344 ordered->file_offset + ordered->num_bytes - 1);
11345 btrfs_put_ordered_extent(ordered);
11348 ASSERT(ordered == NULL);
11351 static const struct inode_operations btrfs_dir_inode_operations = {
11352 .getattr = btrfs_getattr,
11353 .lookup = btrfs_lookup,
11354 .create = btrfs_create,
11355 .unlink = btrfs_unlink,
11356 .link = btrfs_link,
11357 .mkdir = btrfs_mkdir,
11358 .rmdir = btrfs_rmdir,
11359 .rename = btrfs_rename2,
11360 .symlink = btrfs_symlink,
11361 .setattr = btrfs_setattr,
11362 .mknod = btrfs_mknod,
11363 .listxattr = btrfs_listxattr,
11364 .permission = btrfs_permission,
11365 .get_acl = btrfs_get_acl,
11366 .set_acl = btrfs_set_acl,
11367 .update_time = btrfs_update_time,
11368 .tmpfile = btrfs_tmpfile,
11369 .fileattr_get = btrfs_fileattr_get,
11370 .fileattr_set = btrfs_fileattr_set,
11373 static const struct file_operations btrfs_dir_file_operations = {
11374 .llseek = generic_file_llseek,
11375 .read = generic_read_dir,
11376 .iterate_shared = btrfs_real_readdir,
11377 .open = btrfs_opendir,
11378 .unlocked_ioctl = btrfs_ioctl,
11379 #ifdef CONFIG_COMPAT
11380 .compat_ioctl = btrfs_compat_ioctl,
11382 .release = btrfs_release_file,
11383 .fsync = btrfs_sync_file,
11387 * btrfs doesn't support the bmap operation because swapfiles
11388 * use bmap to make a mapping of extents in the file. They assume
11389 * these extents won't change over the life of the file and they
11390 * use the bmap result to do IO directly to the drive.
11392 * the btrfs bmap call would return logical addresses that aren't
11393 * suitable for IO and they also will change frequently as COW
11394 * operations happen. So, swapfile + btrfs == corruption.
11396 * For now we're avoiding this by dropping bmap.
11398 static const struct address_space_operations btrfs_aops = {
11399 .readpage = btrfs_readpage,
11400 .writepage = btrfs_writepage,
11401 .writepages = btrfs_writepages,
11402 .readahead = btrfs_readahead,
11403 .direct_IO = noop_direct_IO,
11404 .invalidate_folio = btrfs_invalidate_folio,
11405 .releasepage = btrfs_releasepage,
11406 #ifdef CONFIG_MIGRATION
11407 .migratepage = btrfs_migratepage,
11409 .dirty_folio = filemap_dirty_folio,
11410 .error_remove_page = generic_error_remove_page,
11411 .swap_activate = btrfs_swap_activate,
11412 .swap_deactivate = btrfs_swap_deactivate,
11415 static const struct inode_operations btrfs_file_inode_operations = {
11416 .getattr = btrfs_getattr,
11417 .setattr = btrfs_setattr,
11418 .listxattr = btrfs_listxattr,
11419 .permission = btrfs_permission,
11420 .fiemap = btrfs_fiemap,
11421 .get_acl = btrfs_get_acl,
11422 .set_acl = btrfs_set_acl,
11423 .update_time = btrfs_update_time,
11424 .fileattr_get = btrfs_fileattr_get,
11425 .fileattr_set = btrfs_fileattr_set,
11427 static const struct inode_operations btrfs_special_inode_operations = {
11428 .getattr = btrfs_getattr,
11429 .setattr = btrfs_setattr,
11430 .permission = btrfs_permission,
11431 .listxattr = btrfs_listxattr,
11432 .get_acl = btrfs_get_acl,
11433 .set_acl = btrfs_set_acl,
11434 .update_time = btrfs_update_time,
11436 static const struct inode_operations btrfs_symlink_inode_operations = {
11437 .get_link = page_get_link,
11438 .getattr = btrfs_getattr,
11439 .setattr = btrfs_setattr,
11440 .permission = btrfs_permission,
11441 .listxattr = btrfs_listxattr,
11442 .update_time = btrfs_update_time,
11445 const struct dentry_operations btrfs_dentry_operations = {
11446 .d_delete = btrfs_dentry_delete,