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/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <linux/migrate.h>
32 #include <linux/sched/mm.h>
33 #include <linux/iomap.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "ordered-data.h"
45 #include "compression.h"
47 #include "free-space-cache.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 struct btrfs_dio_data {
64 struct extent_changeset *data_reserved;
67 static const struct inode_operations btrfs_dir_inode_operations;
68 static const struct inode_operations btrfs_symlink_inode_operations;
69 static const struct inode_operations btrfs_special_inode_operations;
70 static const struct inode_operations btrfs_file_inode_operations;
71 static const struct address_space_operations btrfs_aops;
72 static const struct file_operations btrfs_dir_file_operations;
74 static struct kmem_cache *btrfs_inode_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
78 struct kmem_cache *btrfs_free_space_bitmap_cachep;
80 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
81 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
82 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
83 static noinline int cow_file_range(struct btrfs_inode *inode,
84 struct page *locked_page,
85 u64 start, u64 end, int *page_started,
86 unsigned long *nr_written, int unlock);
87 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
88 u64 len, u64 orig_start, u64 block_start,
89 u64 block_len, u64 orig_block_len,
90 u64 ram_bytes, int compress_type,
93 static void __endio_write_update_ordered(struct btrfs_inode *inode,
94 const u64 offset, const u64 bytes,
98 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
100 * ilock_flags can have the following bit set:
102 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
103 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
105 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
107 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
109 if (ilock_flags & BTRFS_ILOCK_SHARED) {
110 if (ilock_flags & BTRFS_ILOCK_TRY) {
111 if (!inode_trylock_shared(inode))
116 inode_lock_shared(inode);
118 if (ilock_flags & BTRFS_ILOCK_TRY) {
119 if (!inode_trylock(inode))
126 if (ilock_flags & BTRFS_ILOCK_MMAP)
127 down_write(&BTRFS_I(inode)->i_mmap_lock);
132 * btrfs_inode_unlock - unock inode i_rwsem
134 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
135 * to decide whether the lock acquired is shared or exclusive.
137 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
139 if (ilock_flags & BTRFS_ILOCK_MMAP)
140 up_write(&BTRFS_I(inode)->i_mmap_lock);
141 if (ilock_flags & BTRFS_ILOCK_SHARED)
142 inode_unlock_shared(inode);
148 * Cleanup all submitted ordered extents in specified range to handle errors
149 * from the btrfs_run_delalloc_range() callback.
151 * NOTE: caller must ensure that when an error happens, it can not call
152 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
153 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
154 * to be released, which we want to happen only when finishing the ordered
155 * extent (btrfs_finish_ordered_io()).
157 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
158 struct page *locked_page,
159 u64 offset, u64 bytes)
161 unsigned long index = offset >> PAGE_SHIFT;
162 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
163 u64 page_start = page_offset(locked_page);
164 u64 page_end = page_start + PAGE_SIZE - 1;
168 while (index <= end_index) {
169 page = find_get_page(inode->vfs_inode.i_mapping, index);
173 ClearPagePrivate2(page);
178 * In case this page belongs to the delalloc range being instantiated
179 * then skip it, since the first page of a range is going to be
180 * properly cleaned up by the caller of run_delalloc_range
182 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
187 return __endio_write_update_ordered(inode, offset, bytes, false);
190 static int btrfs_dirty_inode(struct inode *inode);
192 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
193 struct inode *inode, struct inode *dir,
194 const struct qstr *qstr)
198 err = btrfs_init_acl(trans, inode, dir);
200 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
205 * this does all the hard work for inserting an inline extent into
206 * the btree. The caller should have done a btrfs_drop_extents so that
207 * no overlapping inline items exist in the btree
209 static int insert_inline_extent(struct btrfs_trans_handle *trans,
210 struct btrfs_path *path, bool extent_inserted,
211 struct btrfs_root *root, struct inode *inode,
212 u64 start, size_t size, size_t compressed_size,
214 struct page **compressed_pages)
216 struct extent_buffer *leaf;
217 struct page *page = NULL;
220 struct btrfs_file_extent_item *ei;
222 size_t cur_size = size;
223 unsigned long offset;
225 ASSERT((compressed_size > 0 && compressed_pages) ||
226 (compressed_size == 0 && !compressed_pages));
228 if (compressed_size && compressed_pages)
229 cur_size = compressed_size;
231 if (!extent_inserted) {
232 struct btrfs_key key;
235 key.objectid = btrfs_ino(BTRFS_I(inode));
237 key.type = BTRFS_EXTENT_DATA_KEY;
239 datasize = btrfs_file_extent_calc_inline_size(cur_size);
240 ret = btrfs_insert_empty_item(trans, root, path, &key,
245 leaf = path->nodes[0];
246 ei = btrfs_item_ptr(leaf, path->slots[0],
247 struct btrfs_file_extent_item);
248 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
249 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
250 btrfs_set_file_extent_encryption(leaf, ei, 0);
251 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
252 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
253 ptr = btrfs_file_extent_inline_start(ei);
255 if (compress_type != BTRFS_COMPRESS_NONE) {
258 while (compressed_size > 0) {
259 cpage = compressed_pages[i];
260 cur_size = min_t(unsigned long, compressed_size,
263 kaddr = kmap_atomic(cpage);
264 write_extent_buffer(leaf, kaddr, ptr, cur_size);
265 kunmap_atomic(kaddr);
269 compressed_size -= cur_size;
271 btrfs_set_file_extent_compression(leaf, ei,
274 page = find_get_page(inode->i_mapping,
275 start >> PAGE_SHIFT);
276 btrfs_set_file_extent_compression(leaf, ei, 0);
277 kaddr = kmap_atomic(page);
278 offset = offset_in_page(start);
279 write_extent_buffer(leaf, kaddr + offset, ptr, size);
280 kunmap_atomic(kaddr);
283 btrfs_mark_buffer_dirty(leaf);
284 btrfs_release_path(path);
287 * We align size to sectorsize for inline extents just for simplicity
290 size = ALIGN(size, root->fs_info->sectorsize);
291 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
296 * we're an inline extent, so nobody can
297 * extend the file past i_size without locking
298 * a page we already have locked.
300 * We must do any isize and inode updates
301 * before we unlock the pages. Otherwise we
302 * could end up racing with unlink.
304 BTRFS_I(inode)->disk_i_size = inode->i_size;
311 * conditionally insert an inline extent into the file. This
312 * does the checks required to make sure the data is small enough
313 * to fit as an inline extent.
315 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
316 u64 end, size_t compressed_size,
318 struct page **compressed_pages)
320 struct btrfs_drop_extents_args drop_args = { 0 };
321 struct btrfs_root *root = inode->root;
322 struct btrfs_fs_info *fs_info = root->fs_info;
323 struct btrfs_trans_handle *trans;
324 u64 isize = i_size_read(&inode->vfs_inode);
325 u64 actual_end = min(end + 1, isize);
326 u64 inline_len = actual_end - start;
327 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
328 u64 data_len = inline_len;
330 struct btrfs_path *path;
333 data_len = compressed_size;
336 actual_end > fs_info->sectorsize ||
337 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
339 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
341 data_len > fs_info->max_inline) {
345 path = btrfs_alloc_path();
349 trans = btrfs_join_transaction(root);
351 btrfs_free_path(path);
352 return PTR_ERR(trans);
354 trans->block_rsv = &inode->block_rsv;
356 drop_args.path = path;
357 drop_args.start = start;
358 drop_args.end = aligned_end;
359 drop_args.drop_cache = true;
360 drop_args.replace_extent = true;
362 if (compressed_size && compressed_pages)
363 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
366 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
369 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
371 btrfs_abort_transaction(trans, ret);
375 if (isize > actual_end)
376 inline_len = min_t(u64, isize, actual_end);
377 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
378 root, &inode->vfs_inode, start,
379 inline_len, compressed_size,
380 compress_type, compressed_pages);
381 if (ret && ret != -ENOSPC) {
382 btrfs_abort_transaction(trans, ret);
384 } else if (ret == -ENOSPC) {
389 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
390 ret = btrfs_update_inode(trans, root, inode);
391 if (ret && ret != -ENOSPC) {
392 btrfs_abort_transaction(trans, ret);
394 } else if (ret == -ENOSPC) {
399 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
402 * Don't forget to free the reserved space, as for inlined extent
403 * it won't count as data extent, free them directly here.
404 * And at reserve time, it's always aligned to page size, so
405 * just free one page here.
407 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
408 btrfs_free_path(path);
409 btrfs_end_transaction(trans);
413 struct async_extent {
418 unsigned long nr_pages;
420 struct list_head list;
425 struct page *locked_page;
428 unsigned int write_flags;
429 struct list_head extents;
430 struct cgroup_subsys_state *blkcg_css;
431 struct btrfs_work work;
436 /* Number of chunks in flight; must be first in the structure */
438 struct async_chunk chunks[];
441 static noinline int add_async_extent(struct async_chunk *cow,
442 u64 start, u64 ram_size,
445 unsigned long nr_pages,
448 struct async_extent *async_extent;
450 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
451 BUG_ON(!async_extent); /* -ENOMEM */
452 async_extent->start = start;
453 async_extent->ram_size = ram_size;
454 async_extent->compressed_size = compressed_size;
455 async_extent->pages = pages;
456 async_extent->nr_pages = nr_pages;
457 async_extent->compress_type = compress_type;
458 list_add_tail(&async_extent->list, &cow->extents);
463 * Check if the inode has flags compatible with compression
465 static inline bool inode_can_compress(struct btrfs_inode *inode)
467 if (inode->flags & BTRFS_INODE_NODATACOW ||
468 inode->flags & BTRFS_INODE_NODATASUM)
474 * Check if the inode needs to be submitted to compression, based on mount
475 * options, defragmentation, properties or heuristics.
477 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
480 struct btrfs_fs_info *fs_info = inode->root->fs_info;
482 if (!inode_can_compress(inode)) {
483 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
484 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
489 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
492 if (inode->defrag_compress)
494 /* bad compression ratios */
495 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
497 if (btrfs_test_opt(fs_info, COMPRESS) ||
498 inode->flags & BTRFS_INODE_COMPRESS ||
499 inode->prop_compress)
500 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
504 static inline void inode_should_defrag(struct btrfs_inode *inode,
505 u64 start, u64 end, u64 num_bytes, u64 small_write)
507 /* If this is a small write inside eof, kick off a defrag */
508 if (num_bytes < small_write &&
509 (start > 0 || end + 1 < inode->disk_i_size))
510 btrfs_add_inode_defrag(NULL, inode);
514 * we create compressed extents in two phases. The first
515 * phase compresses a range of pages that have already been
516 * locked (both pages and state bits are locked).
518 * This is done inside an ordered work queue, and the compression
519 * is spread across many cpus. The actual IO submission is step
520 * two, and the ordered work queue takes care of making sure that
521 * happens in the same order things were put onto the queue by
522 * writepages and friends.
524 * If this code finds it can't get good compression, it puts an
525 * entry onto the work queue to write the uncompressed bytes. This
526 * makes sure that both compressed inodes and uncompressed inodes
527 * are written in the same order that the flusher thread sent them
530 static noinline int compress_file_range(struct async_chunk *async_chunk)
532 struct inode *inode = async_chunk->inode;
533 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
534 u64 blocksize = fs_info->sectorsize;
535 u64 start = async_chunk->start;
536 u64 end = async_chunk->end;
540 struct page **pages = NULL;
541 unsigned long nr_pages;
542 unsigned long total_compressed = 0;
543 unsigned long total_in = 0;
546 int compress_type = fs_info->compress_type;
547 int compressed_extents = 0;
550 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
554 * We need to save i_size before now because it could change in between
555 * us evaluating the size and assigning it. This is because we lock and
556 * unlock the page in truncate and fallocate, and then modify the i_size
559 * The barriers are to emulate READ_ONCE, remove that once i_size_read
563 i_size = i_size_read(inode);
565 actual_end = min_t(u64, i_size, end + 1);
568 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
569 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
570 nr_pages = min_t(unsigned long, nr_pages,
571 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
574 * we don't want to send crud past the end of i_size through
575 * compression, that's just a waste of CPU time. So, if the
576 * end of the file is before the start of our current
577 * requested range of bytes, we bail out to the uncompressed
578 * cleanup code that can deal with all of this.
580 * It isn't really the fastest way to fix things, but this is a
581 * very uncommon corner.
583 if (actual_end <= start)
584 goto cleanup_and_bail_uncompressed;
586 total_compressed = actual_end - start;
589 * skip compression for a small file range(<=blocksize) that
590 * isn't an inline extent, since it doesn't save disk space at all.
592 if (total_compressed <= blocksize &&
593 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
594 goto cleanup_and_bail_uncompressed;
596 total_compressed = min_t(unsigned long, total_compressed,
597 BTRFS_MAX_UNCOMPRESSED);
602 * we do compression for mount -o compress and when the
603 * inode has not been flagged as nocompress. This flag can
604 * change at any time if we discover bad compression ratios.
606 if (inode_need_compress(BTRFS_I(inode), start, end)) {
608 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
610 /* just bail out to the uncompressed code */
615 if (BTRFS_I(inode)->defrag_compress)
616 compress_type = BTRFS_I(inode)->defrag_compress;
617 else if (BTRFS_I(inode)->prop_compress)
618 compress_type = BTRFS_I(inode)->prop_compress;
621 * we need to call clear_page_dirty_for_io on each
622 * page in the range. Otherwise applications with the file
623 * mmap'd can wander in and change the page contents while
624 * we are compressing them.
626 * If the compression fails for any reason, we set the pages
627 * dirty again later on.
629 * Note that the remaining part is redirtied, the start pointer
630 * has moved, the end is the original one.
633 extent_range_clear_dirty_for_io(inode, start, end);
637 /* Compression level is applied here and only here */
638 ret = btrfs_compress_pages(
639 compress_type | (fs_info->compress_level << 4),
640 inode->i_mapping, start,
647 unsigned long offset = offset_in_page(total_compressed);
648 struct page *page = pages[nr_pages - 1];
650 /* zero the tail end of the last page, we might be
651 * sending it down to disk
654 memzero_page(page, offset, PAGE_SIZE - offset);
660 /* lets try to make an inline extent */
661 if (ret || total_in < actual_end) {
662 /* we didn't compress the entire range, try
663 * to make an uncompressed inline extent.
665 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
666 0, BTRFS_COMPRESS_NONE,
669 /* try making a compressed inline extent */
670 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
672 compress_type, pages);
675 unsigned long clear_flags = EXTENT_DELALLOC |
676 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
677 EXTENT_DO_ACCOUNTING;
678 unsigned long page_error_op;
680 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
683 * inline extent creation worked or returned error,
684 * we don't need to create any more async work items.
685 * Unlock and free up our temp pages.
687 * We use DO_ACCOUNTING here because we need the
688 * delalloc_release_metadata to be done _after_ we drop
689 * our outstanding extent for clearing delalloc for this
692 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
696 PAGE_START_WRITEBACK |
701 * Ensure we only free the compressed pages if we have
702 * them allocated, as we can still reach here with
703 * inode_need_compress() == false.
706 for (i = 0; i < nr_pages; i++) {
707 WARN_ON(pages[i]->mapping);
718 * we aren't doing an inline extent round the compressed size
719 * up to a block size boundary so the allocator does sane
722 total_compressed = ALIGN(total_compressed, blocksize);
725 * one last check to make sure the compression is really a
726 * win, compare the page count read with the blocks on disk,
727 * compression must free at least one sector size
729 total_in = ALIGN(total_in, PAGE_SIZE);
730 if (total_compressed + blocksize <= total_in) {
731 compressed_extents++;
734 * The async work queues will take care of doing actual
735 * allocation on disk for these compressed pages, and
736 * will submit them to the elevator.
738 add_async_extent(async_chunk, start, total_in,
739 total_compressed, pages, nr_pages,
742 if (start + total_in < end) {
748 return compressed_extents;
753 * the compression code ran but failed to make things smaller,
754 * free any pages it allocated and our page pointer array
756 for (i = 0; i < nr_pages; i++) {
757 WARN_ON(pages[i]->mapping);
762 total_compressed = 0;
765 /* flag the file so we don't compress in the future */
766 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
767 !(BTRFS_I(inode)->prop_compress)) {
768 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
771 cleanup_and_bail_uncompressed:
773 * No compression, but we still need to write the pages in the file
774 * we've been given so far. redirty the locked page if it corresponds
775 * to our extent and set things up for the async work queue to run
776 * cow_file_range to do the normal delalloc dance.
778 if (async_chunk->locked_page &&
779 (page_offset(async_chunk->locked_page) >= start &&
780 page_offset(async_chunk->locked_page)) <= end) {
781 __set_page_dirty_nobuffers(async_chunk->locked_page);
782 /* unlocked later on in the async handlers */
786 extent_range_redirty_for_io(inode, start, end);
787 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
788 BTRFS_COMPRESS_NONE);
789 compressed_extents++;
791 return compressed_extents;
794 static void free_async_extent_pages(struct async_extent *async_extent)
798 if (!async_extent->pages)
801 for (i = 0; i < async_extent->nr_pages; i++) {
802 WARN_ON(async_extent->pages[i]->mapping);
803 put_page(async_extent->pages[i]);
805 kfree(async_extent->pages);
806 async_extent->nr_pages = 0;
807 async_extent->pages = NULL;
811 * phase two of compressed writeback. This is the ordered portion
812 * of the code, which only gets called in the order the work was
813 * queued. We walk all the async extents created by compress_file_range
814 * and send them down to the disk.
816 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
818 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
819 struct btrfs_fs_info *fs_info = inode->root->fs_info;
820 struct async_extent *async_extent;
822 struct btrfs_key ins;
823 struct extent_map *em;
824 struct btrfs_root *root = inode->root;
825 struct extent_io_tree *io_tree = &inode->io_tree;
829 while (!list_empty(&async_chunk->extents)) {
830 async_extent = list_entry(async_chunk->extents.next,
831 struct async_extent, list);
832 list_del(&async_extent->list);
835 lock_extent(io_tree, async_extent->start,
836 async_extent->start + async_extent->ram_size - 1);
837 /* did the compression code fall back to uncompressed IO? */
838 if (!async_extent->pages) {
839 int page_started = 0;
840 unsigned long nr_written = 0;
842 /* allocate blocks */
843 ret = cow_file_range(inode, async_chunk->locked_page,
845 async_extent->start +
846 async_extent->ram_size - 1,
847 &page_started, &nr_written, 0);
852 * if page_started, cow_file_range inserted an
853 * inline extent and took care of all the unlocking
854 * and IO for us. Otherwise, we need to submit
855 * all those pages down to the drive.
857 if (!page_started && !ret)
858 extent_write_locked_range(&inode->vfs_inode,
860 async_extent->start +
861 async_extent->ram_size - 1,
863 else if (ret && async_chunk->locked_page)
864 unlock_page(async_chunk->locked_page);
870 ret = btrfs_reserve_extent(root, async_extent->ram_size,
871 async_extent->compressed_size,
872 async_extent->compressed_size,
873 0, alloc_hint, &ins, 1, 1);
875 free_async_extent_pages(async_extent);
877 if (ret == -ENOSPC) {
878 unlock_extent(io_tree, async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1);
883 * we need to redirty the pages if we decide to
884 * fallback to uncompressed IO, otherwise we
885 * will not submit these pages down to lower
888 extent_range_redirty_for_io(&inode->vfs_inode,
890 async_extent->start +
891 async_extent->ram_size - 1);
898 * here we're doing allocation and writeback of the
901 em = create_io_em(inode, async_extent->start,
902 async_extent->ram_size, /* len */
903 async_extent->start, /* orig_start */
904 ins.objectid, /* block_start */
905 ins.offset, /* block_len */
906 ins.offset, /* orig_block_len */
907 async_extent->ram_size, /* ram_bytes */
908 async_extent->compress_type,
909 BTRFS_ORDERED_COMPRESSED);
911 /* ret value is not necessary due to void function */
912 goto out_free_reserve;
915 ret = btrfs_add_ordered_extent_compress(inode,
918 async_extent->ram_size,
920 async_extent->compress_type);
922 btrfs_drop_extent_cache(inode, async_extent->start,
923 async_extent->start +
924 async_extent->ram_size - 1, 0);
925 goto out_free_reserve;
927 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
930 * clear dirty, set writeback and unlock the pages.
932 extent_clear_unlock_delalloc(inode, async_extent->start,
933 async_extent->start +
934 async_extent->ram_size - 1,
935 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
936 PAGE_UNLOCK | PAGE_START_WRITEBACK);
937 if (btrfs_submit_compressed_write(inode, async_extent->start,
938 async_extent->ram_size,
940 ins.offset, async_extent->pages,
941 async_extent->nr_pages,
942 async_chunk->write_flags,
943 async_chunk->blkcg_css)) {
944 struct page *p = async_extent->pages[0];
945 const u64 start = async_extent->start;
946 const u64 end = start + async_extent->ram_size - 1;
948 p->mapping = inode->vfs_inode.i_mapping;
949 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
952 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
955 free_async_extent_pages(async_extent);
957 alloc_hint = ins.objectid + ins.offset;
963 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
964 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
966 extent_clear_unlock_delalloc(inode, async_extent->start,
967 async_extent->start +
968 async_extent->ram_size - 1,
969 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
970 EXTENT_DELALLOC_NEW |
971 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
972 PAGE_UNLOCK | PAGE_START_WRITEBACK |
973 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
974 free_async_extent_pages(async_extent);
979 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
982 struct extent_map_tree *em_tree = &inode->extent_tree;
983 struct extent_map *em;
986 read_lock(&em_tree->lock);
987 em = search_extent_mapping(em_tree, start, num_bytes);
990 * if block start isn't an actual block number then find the
991 * first block in this inode and use that as a hint. If that
992 * block is also bogus then just don't worry about it.
994 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
996 em = search_extent_mapping(em_tree, 0, 0);
997 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
998 alloc_hint = em->block_start;
1000 free_extent_map(em);
1002 alloc_hint = em->block_start;
1003 free_extent_map(em);
1006 read_unlock(&em_tree->lock);
1012 * when extent_io.c finds a delayed allocation range in the file,
1013 * the call backs end up in this code. The basic idea is to
1014 * allocate extents on disk for the range, and create ordered data structs
1015 * in ram to track those extents.
1017 * locked_page is the page that writepage had locked already. We use
1018 * it to make sure we don't do extra locks or unlocks.
1020 * *page_started is set to one if we unlock locked_page and do everything
1021 * required to start IO on it. It may be clean and already done with
1022 * IO when we return.
1024 static noinline int cow_file_range(struct btrfs_inode *inode,
1025 struct page *locked_page,
1026 u64 start, u64 end, int *page_started,
1027 unsigned long *nr_written, int unlock)
1029 struct btrfs_root *root = inode->root;
1030 struct btrfs_fs_info *fs_info = root->fs_info;
1033 unsigned long ram_size;
1034 u64 cur_alloc_size = 0;
1036 u64 blocksize = fs_info->sectorsize;
1037 struct btrfs_key ins;
1038 struct extent_map *em;
1039 unsigned clear_bits;
1040 unsigned long page_ops;
1041 bool extent_reserved = false;
1044 if (btrfs_is_free_space_inode(inode)) {
1050 num_bytes = ALIGN(end - start + 1, blocksize);
1051 num_bytes = max(blocksize, num_bytes);
1052 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1054 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1057 /* lets try to make an inline extent */
1058 ret = cow_file_range_inline(inode, start, end, 0,
1059 BTRFS_COMPRESS_NONE, NULL);
1062 * We use DO_ACCOUNTING here because we need the
1063 * delalloc_release_metadata to be run _after_ we drop
1064 * our outstanding extent for clearing delalloc for this
1067 extent_clear_unlock_delalloc(inode, start, end, NULL,
1068 EXTENT_LOCKED | EXTENT_DELALLOC |
1069 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1070 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1071 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1072 *nr_written = *nr_written +
1073 (end - start + PAGE_SIZE) / PAGE_SIZE;
1076 } else if (ret < 0) {
1081 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1082 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1085 * Relocation relies on the relocated extents to have exactly the same
1086 * size as the original extents. Normally writeback for relocation data
1087 * extents follows a NOCOW path because relocation preallocates the
1088 * extents. However, due to an operation such as scrub turning a block
1089 * group to RO mode, it may fallback to COW mode, so we must make sure
1090 * an extent allocated during COW has exactly the requested size and can
1091 * not be split into smaller extents, otherwise relocation breaks and
1092 * fails during the stage where it updates the bytenr of file extent
1095 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1096 min_alloc_size = num_bytes;
1098 min_alloc_size = fs_info->sectorsize;
1100 while (num_bytes > 0) {
1101 cur_alloc_size = num_bytes;
1102 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1103 min_alloc_size, 0, alloc_hint,
1107 cur_alloc_size = ins.offset;
1108 extent_reserved = true;
1110 ram_size = ins.offset;
1111 em = create_io_em(inode, start, ins.offset, /* len */
1112 start, /* orig_start */
1113 ins.objectid, /* block_start */
1114 ins.offset, /* block_len */
1115 ins.offset, /* orig_block_len */
1116 ram_size, /* ram_bytes */
1117 BTRFS_COMPRESS_NONE, /* compress_type */
1118 BTRFS_ORDERED_REGULAR /* type */);
1123 free_extent_map(em);
1125 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1126 ram_size, cur_alloc_size,
1127 BTRFS_ORDERED_REGULAR);
1129 goto out_drop_extent_cache;
1131 if (root->root_key.objectid ==
1132 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1133 ret = btrfs_reloc_clone_csums(inode, start,
1136 * Only drop cache here, and process as normal.
1138 * We must not allow extent_clear_unlock_delalloc()
1139 * at out_unlock label to free meta of this ordered
1140 * extent, as its meta should be freed by
1141 * btrfs_finish_ordered_io().
1143 * So we must continue until @start is increased to
1144 * skip current ordered extent.
1147 btrfs_drop_extent_cache(inode, start,
1148 start + ram_size - 1, 0);
1151 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1153 /* we're not doing compressed IO, don't unlock the first
1154 * page (which the caller expects to stay locked), don't
1155 * clear any dirty bits and don't set any writeback bits
1157 * Do set the Private2 bit so we know this page was properly
1158 * setup for writepage
1160 page_ops = unlock ? PAGE_UNLOCK : 0;
1161 page_ops |= PAGE_SET_PRIVATE2;
1163 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1165 EXTENT_LOCKED | EXTENT_DELALLOC,
1167 if (num_bytes < cur_alloc_size)
1170 num_bytes -= cur_alloc_size;
1171 alloc_hint = ins.objectid + ins.offset;
1172 start += cur_alloc_size;
1173 extent_reserved = false;
1176 * btrfs_reloc_clone_csums() error, since start is increased
1177 * extent_clear_unlock_delalloc() at out_unlock label won't
1178 * free metadata of current ordered extent, we're OK to exit.
1186 out_drop_extent_cache:
1187 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1189 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1190 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1192 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1193 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1194 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1196 * If we reserved an extent for our delalloc range (or a subrange) and
1197 * failed to create the respective ordered extent, then it means that
1198 * when we reserved the extent we decremented the extent's size from
1199 * the data space_info's bytes_may_use counter and incremented the
1200 * space_info's bytes_reserved counter by the same amount. We must make
1201 * sure extent_clear_unlock_delalloc() does not try to decrement again
1202 * the data space_info's bytes_may_use counter, therefore we do not pass
1203 * it the flag EXTENT_CLEAR_DATA_RESV.
1205 if (extent_reserved) {
1206 extent_clear_unlock_delalloc(inode, start,
1207 start + cur_alloc_size - 1,
1211 start += cur_alloc_size;
1215 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1216 clear_bits | EXTENT_CLEAR_DATA_RESV,
1222 * work queue call back to started compression on a file and pages
1224 static noinline void async_cow_start(struct btrfs_work *work)
1226 struct async_chunk *async_chunk;
1227 int compressed_extents;
1229 async_chunk = container_of(work, struct async_chunk, work);
1231 compressed_extents = compress_file_range(async_chunk);
1232 if (compressed_extents == 0) {
1233 btrfs_add_delayed_iput(async_chunk->inode);
1234 async_chunk->inode = NULL;
1239 * work queue call back to submit previously compressed pages
1241 static noinline void async_cow_submit(struct btrfs_work *work)
1243 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1245 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1246 unsigned long nr_pages;
1248 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1251 /* atomic_sub_return implies a barrier */
1252 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1254 cond_wake_up_nomb(&fs_info->async_submit_wait);
1257 * ->inode could be NULL if async_chunk_start has failed to compress,
1258 * in which case we don't have anything to submit, yet we need to
1259 * always adjust ->async_delalloc_pages as its paired with the init
1260 * happening in cow_file_range_async
1262 if (async_chunk->inode)
1263 submit_compressed_extents(async_chunk);
1266 static noinline void async_cow_free(struct btrfs_work *work)
1268 struct async_chunk *async_chunk;
1270 async_chunk = container_of(work, struct async_chunk, work);
1271 if (async_chunk->inode)
1272 btrfs_add_delayed_iput(async_chunk->inode);
1273 if (async_chunk->blkcg_css)
1274 css_put(async_chunk->blkcg_css);
1276 * Since the pointer to 'pending' is at the beginning of the array of
1277 * async_chunk's, freeing it ensures the whole array has been freed.
1279 if (atomic_dec_and_test(async_chunk->pending))
1280 kvfree(async_chunk->pending);
1283 static int cow_file_range_async(struct btrfs_inode *inode,
1284 struct writeback_control *wbc,
1285 struct page *locked_page,
1286 u64 start, u64 end, int *page_started,
1287 unsigned long *nr_written)
1289 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1290 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1291 struct async_cow *ctx;
1292 struct async_chunk *async_chunk;
1293 unsigned long nr_pages;
1295 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1297 bool should_compress;
1299 const unsigned int write_flags = wbc_to_write_flags(wbc);
1301 unlock_extent(&inode->io_tree, start, end);
1303 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1304 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1306 should_compress = false;
1308 should_compress = true;
1311 nofs_flag = memalloc_nofs_save();
1312 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1313 memalloc_nofs_restore(nofs_flag);
1316 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1317 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1318 EXTENT_DO_ACCOUNTING;
1319 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1320 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1322 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1323 clear_bits, page_ops);
1327 async_chunk = ctx->chunks;
1328 atomic_set(&ctx->num_chunks, num_chunks);
1330 for (i = 0; i < num_chunks; i++) {
1331 if (should_compress)
1332 cur_end = min(end, start + SZ_512K - 1);
1337 * igrab is called higher up in the call chain, take only the
1338 * lightweight reference for the callback lifetime
1340 ihold(&inode->vfs_inode);
1341 async_chunk[i].pending = &ctx->num_chunks;
1342 async_chunk[i].inode = &inode->vfs_inode;
1343 async_chunk[i].start = start;
1344 async_chunk[i].end = cur_end;
1345 async_chunk[i].write_flags = write_flags;
1346 INIT_LIST_HEAD(&async_chunk[i].extents);
1349 * The locked_page comes all the way from writepage and its
1350 * the original page we were actually given. As we spread
1351 * this large delalloc region across multiple async_chunk
1352 * structs, only the first struct needs a pointer to locked_page
1354 * This way we don't need racey decisions about who is supposed
1359 * Depending on the compressibility, the pages might or
1360 * might not go through async. We want all of them to
1361 * be accounted against wbc once. Let's do it here
1362 * before the paths diverge. wbc accounting is used
1363 * only for foreign writeback detection and doesn't
1364 * need full accuracy. Just account the whole thing
1365 * against the first page.
1367 wbc_account_cgroup_owner(wbc, locked_page,
1369 async_chunk[i].locked_page = locked_page;
1372 async_chunk[i].locked_page = NULL;
1375 if (blkcg_css != blkcg_root_css) {
1377 async_chunk[i].blkcg_css = blkcg_css;
1379 async_chunk[i].blkcg_css = NULL;
1382 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1383 async_cow_submit, async_cow_free);
1385 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1386 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1388 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1390 *nr_written += nr_pages;
1391 start = cur_end + 1;
1397 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1398 struct page *locked_page, u64 start,
1399 u64 end, int *page_started,
1400 unsigned long *nr_written)
1404 ret = cow_file_range(inode, locked_page, start, end, page_started,
1412 __set_page_dirty_nobuffers(locked_page);
1413 account_page_redirty(locked_page);
1414 extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL);
1420 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1421 u64 bytenr, u64 num_bytes)
1424 struct btrfs_ordered_sum *sums;
1427 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1428 bytenr + num_bytes - 1, &list, 0);
1429 if (ret == 0 && list_empty(&list))
1432 while (!list_empty(&list)) {
1433 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1434 list_del(&sums->list);
1442 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1443 const u64 start, const u64 end,
1444 int *page_started, unsigned long *nr_written)
1446 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1447 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1448 BTRFS_DATA_RELOC_TREE_OBJECTID);
1449 const u64 range_bytes = end + 1 - start;
1450 struct extent_io_tree *io_tree = &inode->io_tree;
1451 u64 range_start = start;
1455 * If EXTENT_NORESERVE is set it means that when the buffered write was
1456 * made we had not enough available data space and therefore we did not
1457 * reserve data space for it, since we though we could do NOCOW for the
1458 * respective file range (either there is prealloc extent or the inode
1459 * has the NOCOW bit set).
1461 * However when we need to fallback to COW mode (because for example the
1462 * block group for the corresponding extent was turned to RO mode by a
1463 * scrub or relocation) we need to do the following:
1465 * 1) We increment the bytes_may_use counter of the data space info.
1466 * If COW succeeds, it allocates a new data extent and after doing
1467 * that it decrements the space info's bytes_may_use counter and
1468 * increments its bytes_reserved counter by the same amount (we do
1469 * this at btrfs_add_reserved_bytes()). So we need to increment the
1470 * bytes_may_use counter to compensate (when space is reserved at
1471 * buffered write time, the bytes_may_use counter is incremented);
1473 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1474 * that if the COW path fails for any reason, it decrements (through
1475 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1476 * data space info, which we incremented in the step above.
1478 * If we need to fallback to cow and the inode corresponds to a free
1479 * space cache inode or an inode of the data relocation tree, we must
1480 * also increment bytes_may_use of the data space_info for the same
1481 * reason. Space caches and relocated data extents always get a prealloc
1482 * extent for them, however scrub or balance may have set the block
1483 * group that contains that extent to RO mode and therefore force COW
1484 * when starting writeback.
1486 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1487 EXTENT_NORESERVE, 0);
1488 if (count > 0 || is_space_ino || is_reloc_ino) {
1490 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1491 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1493 if (is_space_ino || is_reloc_ino)
1494 bytes = range_bytes;
1496 spin_lock(&sinfo->lock);
1497 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1498 spin_unlock(&sinfo->lock);
1501 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1505 return cow_file_range(inode, locked_page, start, end, page_started,
1510 * when nowcow writeback call back. This checks for snapshots or COW copies
1511 * of the extents that exist in the file, and COWs the file as required.
1513 * If no cow copies or snapshots exist, we write directly to the existing
1516 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1517 struct page *locked_page,
1518 const u64 start, const u64 end,
1520 unsigned long *nr_written)
1522 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1523 struct btrfs_root *root = inode->root;
1524 struct btrfs_path *path;
1525 u64 cow_start = (u64)-1;
1526 u64 cur_offset = start;
1528 bool check_prev = true;
1529 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1530 u64 ino = btrfs_ino(inode);
1532 u64 disk_bytenr = 0;
1533 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1535 path = btrfs_alloc_path();
1537 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1538 EXTENT_LOCKED | EXTENT_DELALLOC |
1539 EXTENT_DO_ACCOUNTING |
1540 EXTENT_DEFRAG, PAGE_UNLOCK |
1541 PAGE_START_WRITEBACK |
1542 PAGE_END_WRITEBACK);
1547 struct btrfs_key found_key;
1548 struct btrfs_file_extent_item *fi;
1549 struct extent_buffer *leaf;
1559 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1565 * If there is no extent for our range when doing the initial
1566 * search, then go back to the previous slot as it will be the
1567 * one containing the search offset
1569 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1570 leaf = path->nodes[0];
1571 btrfs_item_key_to_cpu(leaf, &found_key,
1572 path->slots[0] - 1);
1573 if (found_key.objectid == ino &&
1574 found_key.type == BTRFS_EXTENT_DATA_KEY)
1579 /* Go to next leaf if we have exhausted the current one */
1580 leaf = path->nodes[0];
1581 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1582 ret = btrfs_next_leaf(root, path);
1584 if (cow_start != (u64)-1)
1585 cur_offset = cow_start;
1590 leaf = path->nodes[0];
1593 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1595 /* Didn't find anything for our INO */
1596 if (found_key.objectid > ino)
1599 * Keep searching until we find an EXTENT_ITEM or there are no
1600 * more extents for this inode
1602 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1603 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1608 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1609 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1610 found_key.offset > end)
1614 * If the found extent starts after requested offset, then
1615 * adjust extent_end to be right before this extent begins
1617 if (found_key.offset > cur_offset) {
1618 extent_end = found_key.offset;
1624 * Found extent which begins before our range and potentially
1627 fi = btrfs_item_ptr(leaf, path->slots[0],
1628 struct btrfs_file_extent_item);
1629 extent_type = btrfs_file_extent_type(leaf, fi);
1631 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1632 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1633 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1634 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1635 extent_offset = btrfs_file_extent_offset(leaf, fi);
1636 extent_end = found_key.offset +
1637 btrfs_file_extent_num_bytes(leaf, fi);
1639 btrfs_file_extent_disk_num_bytes(leaf, fi);
1641 * If the extent we got ends before our current offset,
1642 * skip to the next extent.
1644 if (extent_end <= cur_offset) {
1649 if (disk_bytenr == 0)
1651 /* Skip compressed/encrypted/encoded extents */
1652 if (btrfs_file_extent_compression(leaf, fi) ||
1653 btrfs_file_extent_encryption(leaf, fi) ||
1654 btrfs_file_extent_other_encoding(leaf, fi))
1657 * If extent is created before the last volume's snapshot
1658 * this implies the extent is shared, hence we can't do
1659 * nocow. This is the same check as in
1660 * btrfs_cross_ref_exist but without calling
1661 * btrfs_search_slot.
1663 if (!freespace_inode &&
1664 btrfs_file_extent_generation(leaf, fi) <=
1665 btrfs_root_last_snapshot(&root->root_item))
1667 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1671 * The following checks can be expensive, as they need to
1672 * take other locks and do btree or rbtree searches, so
1673 * release the path to avoid blocking other tasks for too
1676 btrfs_release_path(path);
1678 ret = btrfs_cross_ref_exist(root, ino,
1680 extent_offset, disk_bytenr, false);
1683 * ret could be -EIO if the above fails to read
1687 if (cow_start != (u64)-1)
1688 cur_offset = cow_start;
1692 WARN_ON_ONCE(freespace_inode);
1695 disk_bytenr += extent_offset;
1696 disk_bytenr += cur_offset - found_key.offset;
1697 num_bytes = min(end + 1, extent_end) - cur_offset;
1699 * If there are pending snapshots for this root, we
1700 * fall into common COW way
1702 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1705 * force cow if csum exists in the range.
1706 * this ensure that csum for a given extent are
1707 * either valid or do not exist.
1709 ret = csum_exist_in_range(fs_info, disk_bytenr,
1713 * ret could be -EIO if the above fails to read
1717 if (cow_start != (u64)-1)
1718 cur_offset = cow_start;
1721 WARN_ON_ONCE(freespace_inode);
1724 /* If the extent's block group is RO, we must COW */
1725 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1728 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1729 extent_end = found_key.offset + ram_bytes;
1730 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1731 /* Skip extents outside of our requested range */
1732 if (extent_end <= start) {
1737 /* If this triggers then we have a memory corruption */
1742 * If nocow is false then record the beginning of the range
1743 * that needs to be COWed
1746 if (cow_start == (u64)-1)
1747 cow_start = cur_offset;
1748 cur_offset = extent_end;
1749 if (cur_offset > end)
1751 if (!path->nodes[0])
1758 * COW range from cow_start to found_key.offset - 1. As the key
1759 * will contain the beginning of the first extent that can be
1760 * NOCOW, following one which needs to be COW'ed
1762 if (cow_start != (u64)-1) {
1763 ret = fallback_to_cow(inode, locked_page,
1764 cow_start, found_key.offset - 1,
1765 page_started, nr_written);
1768 cow_start = (u64)-1;
1771 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1772 u64 orig_start = found_key.offset - extent_offset;
1773 struct extent_map *em;
1775 em = create_io_em(inode, cur_offset, num_bytes,
1777 disk_bytenr, /* block_start */
1778 num_bytes, /* block_len */
1779 disk_num_bytes, /* orig_block_len */
1780 ram_bytes, BTRFS_COMPRESS_NONE,
1781 BTRFS_ORDERED_PREALLOC);
1786 free_extent_map(em);
1787 ret = btrfs_add_ordered_extent(inode, cur_offset,
1788 disk_bytenr, num_bytes,
1790 BTRFS_ORDERED_PREALLOC);
1792 btrfs_drop_extent_cache(inode, cur_offset,
1793 cur_offset + num_bytes - 1,
1798 ret = btrfs_add_ordered_extent(inode, cur_offset,
1799 disk_bytenr, num_bytes,
1801 BTRFS_ORDERED_NOCOW);
1807 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1810 if (root->root_key.objectid ==
1811 BTRFS_DATA_RELOC_TREE_OBJECTID)
1813 * Error handled later, as we must prevent
1814 * extent_clear_unlock_delalloc() in error handler
1815 * from freeing metadata of created ordered extent.
1817 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1820 extent_clear_unlock_delalloc(inode, cur_offset,
1821 cur_offset + num_bytes - 1,
1822 locked_page, EXTENT_LOCKED |
1824 EXTENT_CLEAR_DATA_RESV,
1825 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1827 cur_offset = extent_end;
1830 * btrfs_reloc_clone_csums() error, now we're OK to call error
1831 * handler, as metadata for created ordered extent will only
1832 * be freed by btrfs_finish_ordered_io().
1836 if (cur_offset > end)
1839 btrfs_release_path(path);
1841 if (cur_offset <= end && cow_start == (u64)-1)
1842 cow_start = cur_offset;
1844 if (cow_start != (u64)-1) {
1846 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1847 page_started, nr_written);
1854 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1856 if (ret && cur_offset < end)
1857 extent_clear_unlock_delalloc(inode, cur_offset, end,
1858 locked_page, EXTENT_LOCKED |
1859 EXTENT_DELALLOC | EXTENT_DEFRAG |
1860 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1861 PAGE_START_WRITEBACK |
1862 PAGE_END_WRITEBACK);
1863 btrfs_free_path(path);
1867 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1869 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1870 if (inode->defrag_bytes &&
1871 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1880 * Function to process delayed allocation (create CoW) for ranges which are
1881 * being touched for the first time.
1883 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1884 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1885 struct writeback_control *wbc)
1888 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1890 if (should_nocow(inode, start, end)) {
1892 ret = run_delalloc_nocow(inode, locked_page, start, end,
1893 page_started, nr_written);
1894 } else if (!inode_can_compress(inode) ||
1895 !inode_need_compress(inode, start, end)) {
1897 ret = run_delalloc_zoned(inode, locked_page, start, end,
1898 page_started, nr_written);
1900 ret = cow_file_range(inode, locked_page, start, end,
1901 page_started, nr_written, 1);
1903 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1904 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1905 page_started, nr_written);
1908 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1913 void btrfs_split_delalloc_extent(struct inode *inode,
1914 struct extent_state *orig, u64 split)
1918 /* not delalloc, ignore it */
1919 if (!(orig->state & EXTENT_DELALLOC))
1922 size = orig->end - orig->start + 1;
1923 if (size > BTRFS_MAX_EXTENT_SIZE) {
1928 * See the explanation in btrfs_merge_delalloc_extent, the same
1929 * applies here, just in reverse.
1931 new_size = orig->end - split + 1;
1932 num_extents = count_max_extents(new_size);
1933 new_size = split - orig->start;
1934 num_extents += count_max_extents(new_size);
1935 if (count_max_extents(size) >= num_extents)
1939 spin_lock(&BTRFS_I(inode)->lock);
1940 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1941 spin_unlock(&BTRFS_I(inode)->lock);
1945 * Handle merged delayed allocation extents so we can keep track of new extents
1946 * that are just merged onto old extents, such as when we are doing sequential
1947 * writes, so we can properly account for the metadata space we'll need.
1949 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1950 struct extent_state *other)
1952 u64 new_size, old_size;
1955 /* not delalloc, ignore it */
1956 if (!(other->state & EXTENT_DELALLOC))
1959 if (new->start > other->start)
1960 new_size = new->end - other->start + 1;
1962 new_size = other->end - new->start + 1;
1964 /* we're not bigger than the max, unreserve the space and go */
1965 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1966 spin_lock(&BTRFS_I(inode)->lock);
1967 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1968 spin_unlock(&BTRFS_I(inode)->lock);
1973 * We have to add up either side to figure out how many extents were
1974 * accounted for before we merged into one big extent. If the number of
1975 * extents we accounted for is <= the amount we need for the new range
1976 * then we can return, otherwise drop. Think of it like this
1980 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1981 * need 2 outstanding extents, on one side we have 1 and the other side
1982 * we have 1 so they are == and we can return. But in this case
1984 * [MAX_SIZE+4k][MAX_SIZE+4k]
1986 * Each range on their own accounts for 2 extents, but merged together
1987 * they are only 3 extents worth of accounting, so we need to drop in
1990 old_size = other->end - other->start + 1;
1991 num_extents = count_max_extents(old_size);
1992 old_size = new->end - new->start + 1;
1993 num_extents += count_max_extents(old_size);
1994 if (count_max_extents(new_size) >= num_extents)
1997 spin_lock(&BTRFS_I(inode)->lock);
1998 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1999 spin_unlock(&BTRFS_I(inode)->lock);
2002 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2003 struct inode *inode)
2005 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2007 spin_lock(&root->delalloc_lock);
2008 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2009 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2010 &root->delalloc_inodes);
2011 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2012 &BTRFS_I(inode)->runtime_flags);
2013 root->nr_delalloc_inodes++;
2014 if (root->nr_delalloc_inodes == 1) {
2015 spin_lock(&fs_info->delalloc_root_lock);
2016 BUG_ON(!list_empty(&root->delalloc_root));
2017 list_add_tail(&root->delalloc_root,
2018 &fs_info->delalloc_roots);
2019 spin_unlock(&fs_info->delalloc_root_lock);
2022 spin_unlock(&root->delalloc_lock);
2026 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2027 struct btrfs_inode *inode)
2029 struct btrfs_fs_info *fs_info = root->fs_info;
2031 if (!list_empty(&inode->delalloc_inodes)) {
2032 list_del_init(&inode->delalloc_inodes);
2033 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2034 &inode->runtime_flags);
2035 root->nr_delalloc_inodes--;
2036 if (!root->nr_delalloc_inodes) {
2037 ASSERT(list_empty(&root->delalloc_inodes));
2038 spin_lock(&fs_info->delalloc_root_lock);
2039 BUG_ON(list_empty(&root->delalloc_root));
2040 list_del_init(&root->delalloc_root);
2041 spin_unlock(&fs_info->delalloc_root_lock);
2046 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2047 struct btrfs_inode *inode)
2049 spin_lock(&root->delalloc_lock);
2050 __btrfs_del_delalloc_inode(root, inode);
2051 spin_unlock(&root->delalloc_lock);
2055 * Properly track delayed allocation bytes in the inode and to maintain the
2056 * list of inodes that have pending delalloc work to be done.
2058 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2061 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2063 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2066 * set_bit and clear bit hooks normally require _irqsave/restore
2067 * but in this case, we are only testing for the DELALLOC
2068 * bit, which is only set or cleared with irqs on
2070 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2071 struct btrfs_root *root = BTRFS_I(inode)->root;
2072 u64 len = state->end + 1 - state->start;
2073 u32 num_extents = count_max_extents(len);
2074 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2076 spin_lock(&BTRFS_I(inode)->lock);
2077 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2078 spin_unlock(&BTRFS_I(inode)->lock);
2080 /* For sanity tests */
2081 if (btrfs_is_testing(fs_info))
2084 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2085 fs_info->delalloc_batch);
2086 spin_lock(&BTRFS_I(inode)->lock);
2087 BTRFS_I(inode)->delalloc_bytes += len;
2088 if (*bits & EXTENT_DEFRAG)
2089 BTRFS_I(inode)->defrag_bytes += len;
2090 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2091 &BTRFS_I(inode)->runtime_flags))
2092 btrfs_add_delalloc_inodes(root, inode);
2093 spin_unlock(&BTRFS_I(inode)->lock);
2096 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2097 (*bits & EXTENT_DELALLOC_NEW)) {
2098 spin_lock(&BTRFS_I(inode)->lock);
2099 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2101 spin_unlock(&BTRFS_I(inode)->lock);
2106 * Once a range is no longer delalloc this function ensures that proper
2107 * accounting happens.
2109 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2110 struct extent_state *state, unsigned *bits)
2112 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2113 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2114 u64 len = state->end + 1 - state->start;
2115 u32 num_extents = count_max_extents(len);
2117 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2118 spin_lock(&inode->lock);
2119 inode->defrag_bytes -= len;
2120 spin_unlock(&inode->lock);
2124 * set_bit and clear bit hooks normally require _irqsave/restore
2125 * but in this case, we are only testing for the DELALLOC
2126 * bit, which is only set or cleared with irqs on
2128 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2129 struct btrfs_root *root = inode->root;
2130 bool do_list = !btrfs_is_free_space_inode(inode);
2132 spin_lock(&inode->lock);
2133 btrfs_mod_outstanding_extents(inode, -num_extents);
2134 spin_unlock(&inode->lock);
2137 * We don't reserve metadata space for space cache inodes so we
2138 * don't need to call delalloc_release_metadata if there is an
2141 if (*bits & EXTENT_CLEAR_META_RESV &&
2142 root != fs_info->tree_root)
2143 btrfs_delalloc_release_metadata(inode, len, false);
2145 /* For sanity tests. */
2146 if (btrfs_is_testing(fs_info))
2149 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2150 do_list && !(state->state & EXTENT_NORESERVE) &&
2151 (*bits & EXTENT_CLEAR_DATA_RESV))
2152 btrfs_free_reserved_data_space_noquota(fs_info, len);
2154 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2155 fs_info->delalloc_batch);
2156 spin_lock(&inode->lock);
2157 inode->delalloc_bytes -= len;
2158 if (do_list && inode->delalloc_bytes == 0 &&
2159 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2160 &inode->runtime_flags))
2161 btrfs_del_delalloc_inode(root, inode);
2162 spin_unlock(&inode->lock);
2165 if ((state->state & EXTENT_DELALLOC_NEW) &&
2166 (*bits & EXTENT_DELALLOC_NEW)) {
2167 spin_lock(&inode->lock);
2168 ASSERT(inode->new_delalloc_bytes >= len);
2169 inode->new_delalloc_bytes -= len;
2170 if (*bits & EXTENT_ADD_INODE_BYTES)
2171 inode_add_bytes(&inode->vfs_inode, len);
2172 spin_unlock(&inode->lock);
2177 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2178 * in a chunk's stripe. This function ensures that bios do not span a
2181 * @page - The page we are about to add to the bio
2182 * @size - size we want to add to the bio
2183 * @bio - bio we want to ensure is smaller than a stripe
2184 * @bio_flags - flags of the bio
2186 * return 1 if page cannot be added to the bio
2187 * return 0 if page can be added to the bio
2188 * return error otherwise
2190 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2191 unsigned long bio_flags)
2193 struct inode *inode = page->mapping->host;
2194 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2195 u64 logical = bio->bi_iter.bi_sector << 9;
2196 struct extent_map *em;
2200 struct btrfs_io_geometry geom;
2202 if (bio_flags & EXTENT_BIO_COMPRESSED)
2205 length = bio->bi_iter.bi_size;
2206 map_length = length;
2207 em = btrfs_get_chunk_map(fs_info, logical, map_length);
2210 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical,
2215 if (geom.len < length + size)
2218 free_extent_map(em);
2223 * in order to insert checksums into the metadata in large chunks,
2224 * we wait until bio submission time. All the pages in the bio are
2225 * checksummed and sums are attached onto the ordered extent record.
2227 * At IO completion time the cums attached on the ordered extent record
2228 * are inserted into the btree
2230 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2231 u64 dio_file_offset)
2233 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2236 bool btrfs_bio_fits_in_ordered_extent(struct page *page, struct bio *bio,
2239 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2240 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2241 struct btrfs_ordered_extent *ordered;
2242 u64 len = bio->bi_iter.bi_size + size;
2245 ASSERT(btrfs_is_zoned(fs_info));
2246 ASSERT(fs_info->max_zone_append_size > 0);
2247 ASSERT(bio_op(bio) == REQ_OP_ZONE_APPEND);
2249 /* Ordered extent not yet created, so we're good */
2250 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
2254 if ((bio->bi_iter.bi_sector << SECTOR_SHIFT) + len >
2255 ordered->disk_bytenr + ordered->disk_num_bytes)
2258 btrfs_put_ordered_extent(ordered);
2263 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2264 struct bio *bio, loff_t file_offset)
2266 struct btrfs_ordered_extent *ordered;
2267 struct extent_map *em = NULL, *em_new = NULL;
2268 struct extent_map_tree *em_tree = &inode->extent_tree;
2269 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2270 u64 len = bio->bi_iter.bi_size;
2271 u64 end = start + len;
2276 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2277 if (WARN_ON_ONCE(!ordered))
2278 return BLK_STS_IOERR;
2280 /* No need to split */
2281 if (ordered->disk_num_bytes == len)
2284 /* We cannot split once end_bio'd ordered extent */
2285 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2290 /* We cannot split a compressed ordered extent */
2291 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2296 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2297 /* bio must be in one ordered extent */
2298 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2303 /* Checksum list should be empty */
2304 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2309 pre = start - ordered->disk_bytenr;
2310 post = ordered_end - end;
2312 ret = btrfs_split_ordered_extent(ordered, pre, post);
2316 read_lock(&em_tree->lock);
2317 em = lookup_extent_mapping(em_tree, ordered->file_offset, len);
2319 read_unlock(&em_tree->lock);
2323 read_unlock(&em_tree->lock);
2325 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2327 * We cannot reuse em_new here but have to create a new one, as
2328 * unpin_extent_cache() expects the start of the extent map to be the
2329 * logical offset of the file, which does not hold true anymore after
2332 em_new = create_io_em(inode, em->start + pre, len,
2333 em->start + pre, em->block_start + pre, len,
2334 len, len, BTRFS_COMPRESS_NONE,
2335 BTRFS_ORDERED_REGULAR);
2336 if (IS_ERR(em_new)) {
2337 ret = PTR_ERR(em_new);
2340 free_extent_map(em_new);
2343 free_extent_map(em);
2344 btrfs_put_ordered_extent(ordered);
2346 return errno_to_blk_status(ret);
2350 * extent_io.c submission hook. This does the right thing for csum calculation
2351 * on write, or reading the csums from the tree before a read.
2353 * Rules about async/sync submit,
2354 * a) read: sync submit
2356 * b) write without checksum: sync submit
2358 * c) write with checksum:
2359 * c-1) if bio is issued by fsync: sync submit
2360 * (sync_writers != 0)
2362 * c-2) if root is reloc root: sync submit
2363 * (only in case of buffered IO)
2365 * c-3) otherwise: async submit
2367 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2368 int mirror_num, unsigned long bio_flags)
2371 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2372 struct btrfs_root *root = BTRFS_I(inode)->root;
2373 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2374 blk_status_t ret = 0;
2376 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2378 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2379 !fs_info->csum_root;
2381 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2382 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2384 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2385 struct page *page = bio_first_bvec_all(bio)->bv_page;
2386 loff_t file_offset = page_offset(page);
2388 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2393 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2394 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2398 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2399 ret = btrfs_submit_compressed_read(inode, bio,
2405 * Lookup bio sums does extra checks around whether we
2406 * need to csum or not, which is why we ignore skip_sum
2409 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2414 } else if (async && !skip_sum) {
2415 /* csum items have already been cloned */
2416 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2418 /* we're doing a write, do the async checksumming */
2419 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2420 0, btrfs_submit_bio_start);
2422 } else if (!skip_sum) {
2423 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2429 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2433 bio->bi_status = ret;
2440 * given a list of ordered sums record them in the inode. This happens
2441 * at IO completion time based on sums calculated at bio submission time.
2443 static int add_pending_csums(struct btrfs_trans_handle *trans,
2444 struct list_head *list)
2446 struct btrfs_ordered_sum *sum;
2449 list_for_each_entry(sum, list, list) {
2450 trans->adding_csums = true;
2451 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2452 trans->adding_csums = false;
2459 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2462 struct extent_state **cached_state)
2464 u64 search_start = start;
2465 const u64 end = start + len - 1;
2467 while (search_start < end) {
2468 const u64 search_len = end - search_start + 1;
2469 struct extent_map *em;
2473 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2477 if (em->block_start != EXTENT_MAP_HOLE)
2481 if (em->start < search_start)
2482 em_len -= search_start - em->start;
2483 if (em_len > search_len)
2484 em_len = search_len;
2486 ret = set_extent_bit(&inode->io_tree, search_start,
2487 search_start + em_len - 1,
2488 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2491 search_start = extent_map_end(em);
2492 free_extent_map(em);
2499 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2500 unsigned int extra_bits,
2501 struct extent_state **cached_state)
2503 WARN_ON(PAGE_ALIGNED(end));
2505 if (start >= i_size_read(&inode->vfs_inode) &&
2506 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2508 * There can't be any extents following eof in this case so just
2509 * set the delalloc new bit for the range directly.
2511 extra_bits |= EXTENT_DELALLOC_NEW;
2515 ret = btrfs_find_new_delalloc_bytes(inode, start,
2522 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2526 /* see btrfs_writepage_start_hook for details on why this is required */
2527 struct btrfs_writepage_fixup {
2529 struct inode *inode;
2530 struct btrfs_work work;
2533 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2535 struct btrfs_writepage_fixup *fixup;
2536 struct btrfs_ordered_extent *ordered;
2537 struct extent_state *cached_state = NULL;
2538 struct extent_changeset *data_reserved = NULL;
2540 struct btrfs_inode *inode;
2544 bool free_delalloc_space = true;
2546 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2548 inode = BTRFS_I(fixup->inode);
2549 page_start = page_offset(page);
2550 page_end = page_offset(page) + PAGE_SIZE - 1;
2553 * This is similar to page_mkwrite, we need to reserve the space before
2554 * we take the page lock.
2556 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2562 * Before we queued this fixup, we took a reference on the page.
2563 * page->mapping may go NULL, but it shouldn't be moved to a different
2566 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2568 * Unfortunately this is a little tricky, either
2570 * 1) We got here and our page had already been dealt with and
2571 * we reserved our space, thus ret == 0, so we need to just
2572 * drop our space reservation and bail. This can happen the
2573 * first time we come into the fixup worker, or could happen
2574 * while waiting for the ordered extent.
2575 * 2) Our page was already dealt with, but we happened to get an
2576 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2577 * this case we obviously don't have anything to release, but
2578 * because the page was already dealt with we don't want to
2579 * mark the page with an error, so make sure we're resetting
2580 * ret to 0. This is why we have this check _before_ the ret
2581 * check, because we do not want to have a surprise ENOSPC
2582 * when the page was already properly dealt with.
2585 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2586 btrfs_delalloc_release_space(inode, data_reserved,
2587 page_start, PAGE_SIZE,
2595 * We can't mess with the page state unless it is locked, so now that
2596 * it is locked bail if we failed to make our space reservation.
2601 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2603 /* already ordered? We're done */
2604 if (PagePrivate2(page))
2607 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2609 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2612 btrfs_start_ordered_extent(ordered, 1);
2613 btrfs_put_ordered_extent(ordered);
2617 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2623 * Everything went as planned, we're now the owner of a dirty page with
2624 * delayed allocation bits set and space reserved for our COW
2627 * The page was dirty when we started, nothing should have cleaned it.
2629 BUG_ON(!PageDirty(page));
2630 free_delalloc_space = false;
2632 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2633 if (free_delalloc_space)
2634 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2636 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2641 * We hit ENOSPC or other errors. Update the mapping and page
2642 * to reflect the errors and clean the page.
2644 mapping_set_error(page->mapping, ret);
2645 end_extent_writepage(page, ret, page_start, page_end);
2646 clear_page_dirty_for_io(page);
2649 ClearPageChecked(page);
2653 extent_changeset_free(data_reserved);
2655 * As a precaution, do a delayed iput in case it would be the last iput
2656 * that could need flushing space. Recursing back to fixup worker would
2659 btrfs_add_delayed_iput(&inode->vfs_inode);
2663 * There are a few paths in the higher layers of the kernel that directly
2664 * set the page dirty bit without asking the filesystem if it is a
2665 * good idea. This causes problems because we want to make sure COW
2666 * properly happens and the data=ordered rules are followed.
2668 * In our case any range that doesn't have the ORDERED bit set
2669 * hasn't been properly setup for IO. We kick off an async process
2670 * to fix it up. The async helper will wait for ordered extents, set
2671 * the delalloc bit and make it safe to write the page.
2673 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2675 struct inode *inode = page->mapping->host;
2676 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2677 struct btrfs_writepage_fixup *fixup;
2679 /* this page is properly in the ordered list */
2680 if (TestClearPagePrivate2(page))
2684 * PageChecked is set below when we create a fixup worker for this page,
2685 * don't try to create another one if we're already PageChecked()
2687 * The extent_io writepage code will redirty the page if we send back
2690 if (PageChecked(page))
2693 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2698 * We are already holding a reference to this inode from
2699 * write_cache_pages. We need to hold it because the space reservation
2700 * takes place outside of the page lock, and we can't trust
2701 * page->mapping outside of the page lock.
2704 SetPageChecked(page);
2706 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2708 fixup->inode = inode;
2709 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2714 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2715 struct btrfs_inode *inode, u64 file_pos,
2716 struct btrfs_file_extent_item *stack_fi,
2717 const bool update_inode_bytes,
2718 u64 qgroup_reserved)
2720 struct btrfs_root *root = inode->root;
2721 const u64 sectorsize = root->fs_info->sectorsize;
2722 struct btrfs_path *path;
2723 struct extent_buffer *leaf;
2724 struct btrfs_key ins;
2725 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2726 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2727 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2728 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2729 struct btrfs_drop_extents_args drop_args = { 0 };
2732 path = btrfs_alloc_path();
2737 * we may be replacing one extent in the tree with another.
2738 * The new extent is pinned in the extent map, and we don't want
2739 * to drop it from the cache until it is completely in the btree.
2741 * So, tell btrfs_drop_extents to leave this extent in the cache.
2742 * the caller is expected to unpin it and allow it to be merged
2745 drop_args.path = path;
2746 drop_args.start = file_pos;
2747 drop_args.end = file_pos + num_bytes;
2748 drop_args.replace_extent = true;
2749 drop_args.extent_item_size = sizeof(*stack_fi);
2750 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2754 if (!drop_args.extent_inserted) {
2755 ins.objectid = btrfs_ino(inode);
2756 ins.offset = file_pos;
2757 ins.type = BTRFS_EXTENT_DATA_KEY;
2759 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2764 leaf = path->nodes[0];
2765 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2766 write_extent_buffer(leaf, stack_fi,
2767 btrfs_item_ptr_offset(leaf, path->slots[0]),
2768 sizeof(struct btrfs_file_extent_item));
2770 btrfs_mark_buffer_dirty(leaf);
2771 btrfs_release_path(path);
2774 * If we dropped an inline extent here, we know the range where it is
2775 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2776 * number of bytes only for that range contaning the inline extent.
2777 * The remaining of the range will be processed when clearning the
2778 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2780 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2781 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2783 inline_size = drop_args.bytes_found - inline_size;
2784 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2785 drop_args.bytes_found -= inline_size;
2786 num_bytes -= sectorsize;
2789 if (update_inode_bytes)
2790 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2792 ins.objectid = disk_bytenr;
2793 ins.offset = disk_num_bytes;
2794 ins.type = BTRFS_EXTENT_ITEM_KEY;
2796 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2800 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2801 file_pos, qgroup_reserved, &ins);
2803 btrfs_free_path(path);
2808 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2811 struct btrfs_block_group *cache;
2813 cache = btrfs_lookup_block_group(fs_info, start);
2816 spin_lock(&cache->lock);
2817 cache->delalloc_bytes -= len;
2818 spin_unlock(&cache->lock);
2820 btrfs_put_block_group(cache);
2823 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2824 struct btrfs_ordered_extent *oe)
2826 struct btrfs_file_extent_item stack_fi;
2828 bool update_inode_bytes;
2830 memset(&stack_fi, 0, sizeof(stack_fi));
2831 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2832 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2833 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2834 oe->disk_num_bytes);
2835 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2836 logical_len = oe->truncated_len;
2838 logical_len = oe->num_bytes;
2839 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2840 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2841 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2842 /* Encryption and other encoding is reserved and all 0 */
2845 * For delalloc, when completing an ordered extent we update the inode's
2846 * bytes when clearing the range in the inode's io tree, so pass false
2847 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2848 * except if the ordered extent was truncated.
2850 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2851 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2853 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2854 oe->file_offset, &stack_fi,
2855 update_inode_bytes, oe->qgroup_rsv);
2859 * As ordered data IO finishes, this gets called so we can finish
2860 * an ordered extent if the range of bytes in the file it covers are
2863 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2865 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2866 struct btrfs_root *root = inode->root;
2867 struct btrfs_fs_info *fs_info = root->fs_info;
2868 struct btrfs_trans_handle *trans = NULL;
2869 struct extent_io_tree *io_tree = &inode->io_tree;
2870 struct extent_state *cached_state = NULL;
2872 int compress_type = 0;
2874 u64 logical_len = ordered_extent->num_bytes;
2875 bool freespace_inode;
2876 bool truncated = false;
2877 bool clear_reserved_extent = true;
2878 unsigned int clear_bits = EXTENT_DEFRAG;
2880 start = ordered_extent->file_offset;
2881 end = start + ordered_extent->num_bytes - 1;
2883 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2884 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2885 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2886 clear_bits |= EXTENT_DELALLOC_NEW;
2888 freespace_inode = btrfs_is_free_space_inode(inode);
2890 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2895 if (ordered_extent->disk)
2896 btrfs_rewrite_logical_zoned(ordered_extent);
2898 btrfs_free_io_failure_record(inode, start, end);
2900 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2902 logical_len = ordered_extent->truncated_len;
2903 /* Truncated the entire extent, don't bother adding */
2908 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2909 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2911 btrfs_inode_safe_disk_i_size_write(inode, 0);
2912 if (freespace_inode)
2913 trans = btrfs_join_transaction_spacecache(root);
2915 trans = btrfs_join_transaction(root);
2916 if (IS_ERR(trans)) {
2917 ret = PTR_ERR(trans);
2921 trans->block_rsv = &inode->block_rsv;
2922 ret = btrfs_update_inode_fallback(trans, root, inode);
2923 if (ret) /* -ENOMEM or corruption */
2924 btrfs_abort_transaction(trans, ret);
2928 clear_bits |= EXTENT_LOCKED;
2929 lock_extent_bits(io_tree, start, end, &cached_state);
2931 if (freespace_inode)
2932 trans = btrfs_join_transaction_spacecache(root);
2934 trans = btrfs_join_transaction(root);
2935 if (IS_ERR(trans)) {
2936 ret = PTR_ERR(trans);
2941 trans->block_rsv = &inode->block_rsv;
2943 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2944 compress_type = ordered_extent->compress_type;
2945 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2946 BUG_ON(compress_type);
2947 ret = btrfs_mark_extent_written(trans, inode,
2948 ordered_extent->file_offset,
2949 ordered_extent->file_offset +
2952 BUG_ON(root == fs_info->tree_root);
2953 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2955 clear_reserved_extent = false;
2956 btrfs_release_delalloc_bytes(fs_info,
2957 ordered_extent->disk_bytenr,
2958 ordered_extent->disk_num_bytes);
2961 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
2962 ordered_extent->num_bytes, trans->transid);
2964 btrfs_abort_transaction(trans, ret);
2968 ret = add_pending_csums(trans, &ordered_extent->list);
2970 btrfs_abort_transaction(trans, ret);
2975 * If this is a new delalloc range, clear its new delalloc flag to
2976 * update the inode's number of bytes. This needs to be done first
2977 * before updating the inode item.
2979 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
2980 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
2981 clear_extent_bit(&inode->io_tree, start, end,
2982 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
2983 0, 0, &cached_state);
2985 btrfs_inode_safe_disk_i_size_write(inode, 0);
2986 ret = btrfs_update_inode_fallback(trans, root, inode);
2987 if (ret) { /* -ENOMEM or corruption */
2988 btrfs_abort_transaction(trans, ret);
2993 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
2994 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2998 btrfs_end_transaction(trans);
3000 if (ret || truncated) {
3001 u64 unwritten_start = start;
3004 * If we failed to finish this ordered extent for any reason we
3005 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3006 * extent, and mark the inode with the error if it wasn't
3007 * already set. Any error during writeback would have already
3008 * set the mapping error, so we need to set it if we're the ones
3009 * marking this ordered extent as failed.
3011 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3012 &ordered_extent->flags))
3013 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3016 unwritten_start += logical_len;
3017 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3019 /* Drop the cache for the part of the extent we didn't write. */
3020 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3023 * If the ordered extent had an IOERR or something else went
3024 * wrong we need to return the space for this ordered extent
3025 * back to the allocator. We only free the extent in the
3026 * truncated case if we didn't write out the extent at all.
3028 * If we made it past insert_reserved_file_extent before we
3029 * errored out then we don't need to do this as the accounting
3030 * has already been done.
3032 if ((ret || !logical_len) &&
3033 clear_reserved_extent &&
3034 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3035 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3037 * Discard the range before returning it back to the
3040 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3041 btrfs_discard_extent(fs_info,
3042 ordered_extent->disk_bytenr,
3043 ordered_extent->disk_num_bytes,
3045 btrfs_free_reserved_extent(fs_info,
3046 ordered_extent->disk_bytenr,
3047 ordered_extent->disk_num_bytes, 1);
3052 * This needs to be done to make sure anybody waiting knows we are done
3053 * updating everything for this ordered extent.
3055 btrfs_remove_ordered_extent(inode, ordered_extent);
3058 btrfs_put_ordered_extent(ordered_extent);
3059 /* once for the tree */
3060 btrfs_put_ordered_extent(ordered_extent);
3065 static void finish_ordered_fn(struct btrfs_work *work)
3067 struct btrfs_ordered_extent *ordered_extent;
3068 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3069 btrfs_finish_ordered_io(ordered_extent);
3072 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
3073 u64 end, int uptodate)
3075 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3076 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3077 struct btrfs_ordered_extent *ordered_extent = NULL;
3078 struct btrfs_workqueue *wq;
3080 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3082 ClearPagePrivate2(page);
3083 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3084 end - start + 1, uptodate))
3087 if (btrfs_is_free_space_inode(inode))
3088 wq = fs_info->endio_freespace_worker;
3090 wq = fs_info->endio_write_workers;
3092 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
3093 btrfs_queue_work(wq, &ordered_extent->work);
3097 * check_data_csum - verify checksum of one sector of uncompressed data
3099 * @io_bio: btrfs_io_bio which contains the csum
3100 * @bio_offset: offset to the beginning of the bio (in bytes)
3101 * @page: page where is the data to be verified
3102 * @pgoff: offset inside the page
3103 * @start: logical offset in the file
3105 * The length of such check is always one sector size.
3107 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
3108 u32 bio_offset, struct page *page, u32 pgoff,
3111 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3112 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3114 u32 len = fs_info->sectorsize;
3115 const u32 csum_size = fs_info->csum_size;
3116 unsigned int offset_sectors;
3118 u8 csum[BTRFS_CSUM_SIZE];
3120 ASSERT(pgoff + len <= PAGE_SIZE);
3122 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3123 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
3125 kaddr = kmap_atomic(page);
3126 shash->tfm = fs_info->csum_shash;
3128 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3130 if (memcmp(csum, csum_expected, csum_size))
3133 kunmap_atomic(kaddr);
3136 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3137 io_bio->mirror_num);
3139 btrfs_dev_stat_inc_and_print(io_bio->device,
3140 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3141 memset(kaddr + pgoff, 1, len);
3142 flush_dcache_page(page);
3143 kunmap_atomic(kaddr);
3148 * When reads are done, we need to check csums to verify the data is correct.
3149 * if there's a match, we allow the bio to finish. If not, the code in
3150 * extent_io.c will try to find good copies for us.
3152 * @bio_offset: offset to the beginning of the bio (in bytes)
3153 * @start: file offset of the range start
3154 * @end: file offset of the range end (inclusive)
3156 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3157 struct page *page, u64 start, u64 end)
3159 struct inode *inode = page->mapping->host;
3160 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3161 struct btrfs_root *root = BTRFS_I(inode)->root;
3162 const u32 sectorsize = root->fs_info->sectorsize;
3165 if (PageChecked(page)) {
3166 ClearPageChecked(page);
3170 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3173 if (!root->fs_info->csum_root)
3176 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3177 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3178 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3182 ASSERT(page_offset(page) <= start &&
3183 end <= page_offset(page) + PAGE_SIZE - 1);
3184 for (pg_off = offset_in_page(start);
3185 pg_off < offset_in_page(end);
3186 pg_off += sectorsize, bio_offset += sectorsize) {
3189 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off,
3190 page_offset(page) + pg_off);
3198 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3200 * @inode: The inode we want to perform iput on
3202 * This function uses the generic vfs_inode::i_count to track whether we should
3203 * just decrement it (in case it's > 1) or if this is the last iput then link
3204 * the inode to the delayed iput machinery. Delayed iputs are processed at
3205 * transaction commit time/superblock commit/cleaner kthread.
3207 void btrfs_add_delayed_iput(struct inode *inode)
3209 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3210 struct btrfs_inode *binode = BTRFS_I(inode);
3212 if (atomic_add_unless(&inode->i_count, -1, 1))
3215 atomic_inc(&fs_info->nr_delayed_iputs);
3216 spin_lock(&fs_info->delayed_iput_lock);
3217 ASSERT(list_empty(&binode->delayed_iput));
3218 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3219 spin_unlock(&fs_info->delayed_iput_lock);
3220 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3221 wake_up_process(fs_info->cleaner_kthread);
3224 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3225 struct btrfs_inode *inode)
3227 list_del_init(&inode->delayed_iput);
3228 spin_unlock(&fs_info->delayed_iput_lock);
3229 iput(&inode->vfs_inode);
3230 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3231 wake_up(&fs_info->delayed_iputs_wait);
3232 spin_lock(&fs_info->delayed_iput_lock);
3235 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3236 struct btrfs_inode *inode)
3238 if (!list_empty(&inode->delayed_iput)) {
3239 spin_lock(&fs_info->delayed_iput_lock);
3240 if (!list_empty(&inode->delayed_iput))
3241 run_delayed_iput_locked(fs_info, inode);
3242 spin_unlock(&fs_info->delayed_iput_lock);
3246 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3249 spin_lock(&fs_info->delayed_iput_lock);
3250 while (!list_empty(&fs_info->delayed_iputs)) {
3251 struct btrfs_inode *inode;
3253 inode = list_first_entry(&fs_info->delayed_iputs,
3254 struct btrfs_inode, delayed_iput);
3255 run_delayed_iput_locked(fs_info, inode);
3256 cond_resched_lock(&fs_info->delayed_iput_lock);
3258 spin_unlock(&fs_info->delayed_iput_lock);
3262 * Wait for flushing all delayed iputs
3264 * @fs_info: the filesystem
3266 * This will wait on any delayed iputs that are currently running with KILLABLE
3267 * set. Once they are all done running we will return, unless we are killed in
3268 * which case we return EINTR. This helps in user operations like fallocate etc
3269 * that might get blocked on the iputs.
3271 * Return EINTR if we were killed, 0 if nothing's pending
3273 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3275 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3276 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3283 * This creates an orphan entry for the given inode in case something goes wrong
3284 * in the middle of an unlink.
3286 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3287 struct btrfs_inode *inode)
3291 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3292 if (ret && ret != -EEXIST) {
3293 btrfs_abort_transaction(trans, ret);
3301 * We have done the delete so we can go ahead and remove the orphan item for
3302 * this particular inode.
3304 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3305 struct btrfs_inode *inode)
3307 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3311 * this cleans up any orphans that may be left on the list from the last use
3314 int btrfs_orphan_cleanup(struct btrfs_root *root)
3316 struct btrfs_fs_info *fs_info = root->fs_info;
3317 struct btrfs_path *path;
3318 struct extent_buffer *leaf;
3319 struct btrfs_key key, found_key;
3320 struct btrfs_trans_handle *trans;
3321 struct inode *inode;
3322 u64 last_objectid = 0;
3323 int ret = 0, nr_unlink = 0;
3325 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3328 path = btrfs_alloc_path();
3333 path->reada = READA_BACK;
3335 key.objectid = BTRFS_ORPHAN_OBJECTID;
3336 key.type = BTRFS_ORPHAN_ITEM_KEY;
3337 key.offset = (u64)-1;
3340 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3345 * if ret == 0 means we found what we were searching for, which
3346 * is weird, but possible, so only screw with path if we didn't
3347 * find the key and see if we have stuff that matches
3351 if (path->slots[0] == 0)
3356 /* pull out the item */
3357 leaf = path->nodes[0];
3358 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3360 /* make sure the item matches what we want */
3361 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3363 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3366 /* release the path since we're done with it */
3367 btrfs_release_path(path);
3370 * this is where we are basically btrfs_lookup, without the
3371 * crossing root thing. we store the inode number in the
3372 * offset of the orphan item.
3375 if (found_key.offset == last_objectid) {
3377 "Error removing orphan entry, stopping orphan cleanup");
3382 last_objectid = found_key.offset;
3384 found_key.objectid = found_key.offset;
3385 found_key.type = BTRFS_INODE_ITEM_KEY;
3386 found_key.offset = 0;
3387 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3388 ret = PTR_ERR_OR_ZERO(inode);
3389 if (ret && ret != -ENOENT)
3392 if (ret == -ENOENT && root == fs_info->tree_root) {
3393 struct btrfs_root *dead_root;
3394 int is_dead_root = 0;
3397 * This is an orphan in the tree root. Currently these
3398 * could come from 2 sources:
3399 * a) a root (snapshot/subvolume) deletion in progress
3400 * b) a free space cache inode
3401 * We need to distinguish those two, as the orphan item
3402 * for a root must not get deleted before the deletion
3403 * of the snapshot/subvolume's tree completes.
3405 * btrfs_find_orphan_roots() ran before us, which has
3406 * found all deleted roots and loaded them into
3407 * fs_info->fs_roots_radix. So here we can find if an
3408 * orphan item corresponds to a deleted root by looking
3409 * up the root from that radix tree.
3412 spin_lock(&fs_info->fs_roots_radix_lock);
3413 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3414 (unsigned long)found_key.objectid);
3415 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3417 spin_unlock(&fs_info->fs_roots_radix_lock);
3420 /* prevent this orphan from being found again */
3421 key.offset = found_key.objectid - 1;
3428 * If we have an inode with links, there are a couple of
3429 * possibilities. Old kernels (before v3.12) used to create an
3430 * orphan item for truncate indicating that there were possibly
3431 * extent items past i_size that needed to be deleted. In v3.12,
3432 * truncate was changed to update i_size in sync with the extent
3433 * items, but the (useless) orphan item was still created. Since
3434 * v4.18, we don't create the orphan item for truncate at all.
3436 * So, this item could mean that we need to do a truncate, but
3437 * only if this filesystem was last used on a pre-v3.12 kernel
3438 * and was not cleanly unmounted. The odds of that are quite
3439 * slim, and it's a pain to do the truncate now, so just delete
3442 * It's also possible that this orphan item was supposed to be
3443 * deleted but wasn't. The inode number may have been reused,
3444 * but either way, we can delete the orphan item.
3446 if (ret == -ENOENT || inode->i_nlink) {
3449 trans = btrfs_start_transaction(root, 1);
3450 if (IS_ERR(trans)) {
3451 ret = PTR_ERR(trans);
3454 btrfs_debug(fs_info, "auto deleting %Lu",
3455 found_key.objectid);
3456 ret = btrfs_del_orphan_item(trans, root,
3457 found_key.objectid);
3458 btrfs_end_transaction(trans);
3466 /* this will do delete_inode and everything for us */
3469 /* release the path since we're done with it */
3470 btrfs_release_path(path);
3472 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3474 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3475 trans = btrfs_join_transaction(root);
3477 btrfs_end_transaction(trans);
3481 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3485 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3486 btrfs_free_path(path);
3491 * very simple check to peek ahead in the leaf looking for xattrs. If we
3492 * don't find any xattrs, we know there can't be any acls.
3494 * slot is the slot the inode is in, objectid is the objectid of the inode
3496 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3497 int slot, u64 objectid,
3498 int *first_xattr_slot)
3500 u32 nritems = btrfs_header_nritems(leaf);
3501 struct btrfs_key found_key;
3502 static u64 xattr_access = 0;
3503 static u64 xattr_default = 0;
3506 if (!xattr_access) {
3507 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3508 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3509 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3510 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3514 *first_xattr_slot = -1;
3515 while (slot < nritems) {
3516 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3518 /* we found a different objectid, there must not be acls */
3519 if (found_key.objectid != objectid)
3522 /* we found an xattr, assume we've got an acl */
3523 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3524 if (*first_xattr_slot == -1)
3525 *first_xattr_slot = slot;
3526 if (found_key.offset == xattr_access ||
3527 found_key.offset == xattr_default)
3532 * we found a key greater than an xattr key, there can't
3533 * be any acls later on
3535 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3542 * it goes inode, inode backrefs, xattrs, extents,
3543 * so if there are a ton of hard links to an inode there can
3544 * be a lot of backrefs. Don't waste time searching too hard,
3545 * this is just an optimization
3550 /* we hit the end of the leaf before we found an xattr or
3551 * something larger than an xattr. We have to assume the inode
3554 if (*first_xattr_slot == -1)
3555 *first_xattr_slot = slot;
3560 * read an inode from the btree into the in-memory inode
3562 static int btrfs_read_locked_inode(struct inode *inode,
3563 struct btrfs_path *in_path)
3565 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3566 struct btrfs_path *path = in_path;
3567 struct extent_buffer *leaf;
3568 struct btrfs_inode_item *inode_item;
3569 struct btrfs_root *root = BTRFS_I(inode)->root;
3570 struct btrfs_key location;
3575 bool filled = false;
3576 int first_xattr_slot;
3578 ret = btrfs_fill_inode(inode, &rdev);
3583 path = btrfs_alloc_path();
3588 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3590 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3592 if (path != in_path)
3593 btrfs_free_path(path);
3597 leaf = path->nodes[0];
3602 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3603 struct btrfs_inode_item);
3604 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3605 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3606 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3607 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3608 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3609 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3610 round_up(i_size_read(inode), fs_info->sectorsize));
3612 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3613 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3615 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3616 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3618 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3619 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3621 BTRFS_I(inode)->i_otime.tv_sec =
3622 btrfs_timespec_sec(leaf, &inode_item->otime);
3623 BTRFS_I(inode)->i_otime.tv_nsec =
3624 btrfs_timespec_nsec(leaf, &inode_item->otime);
3626 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3627 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3628 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3630 inode_set_iversion_queried(inode,
3631 btrfs_inode_sequence(leaf, inode_item));
3632 inode->i_generation = BTRFS_I(inode)->generation;
3634 rdev = btrfs_inode_rdev(leaf, inode_item);
3636 BTRFS_I(inode)->index_cnt = (u64)-1;
3637 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3641 * If we were modified in the current generation and evicted from memory
3642 * and then re-read we need to do a full sync since we don't have any
3643 * idea about which extents were modified before we were evicted from
3646 * This is required for both inode re-read from disk and delayed inode
3647 * in delayed_nodes_tree.
3649 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3650 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3651 &BTRFS_I(inode)->runtime_flags);
3654 * We don't persist the id of the transaction where an unlink operation
3655 * against the inode was last made. So here we assume the inode might
3656 * have been evicted, and therefore the exact value of last_unlink_trans
3657 * lost, and set it to last_trans to avoid metadata inconsistencies
3658 * between the inode and its parent if the inode is fsync'ed and the log
3659 * replayed. For example, in the scenario:
3662 * ln mydir/foo mydir/bar
3665 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3666 * xfs_io -c fsync mydir/foo
3668 * mount fs, triggers fsync log replay
3670 * We must make sure that when we fsync our inode foo we also log its
3671 * parent inode, otherwise after log replay the parent still has the
3672 * dentry with the "bar" name but our inode foo has a link count of 1
3673 * and doesn't have an inode ref with the name "bar" anymore.
3675 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3676 * but it guarantees correctness at the expense of occasional full
3677 * transaction commits on fsync if our inode is a directory, or if our
3678 * inode is not a directory, logging its parent unnecessarily.
3680 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3683 * Same logic as for last_unlink_trans. We don't persist the generation
3684 * of the last transaction where this inode was used for a reflink
3685 * operation, so after eviction and reloading the inode we must be
3686 * pessimistic and assume the last transaction that modified the inode.
3688 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3691 if (inode->i_nlink != 1 ||
3692 path->slots[0] >= btrfs_header_nritems(leaf))
3695 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3696 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3699 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3700 if (location.type == BTRFS_INODE_REF_KEY) {
3701 struct btrfs_inode_ref *ref;
3703 ref = (struct btrfs_inode_ref *)ptr;
3704 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3705 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3706 struct btrfs_inode_extref *extref;
3708 extref = (struct btrfs_inode_extref *)ptr;
3709 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3714 * try to precache a NULL acl entry for files that don't have
3715 * any xattrs or acls
3717 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3718 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3719 if (first_xattr_slot != -1) {
3720 path->slots[0] = first_xattr_slot;
3721 ret = btrfs_load_inode_props(inode, path);
3724 "error loading props for ino %llu (root %llu): %d",
3725 btrfs_ino(BTRFS_I(inode)),
3726 root->root_key.objectid, ret);
3728 if (path != in_path)
3729 btrfs_free_path(path);
3732 cache_no_acl(inode);
3734 switch (inode->i_mode & S_IFMT) {
3736 inode->i_mapping->a_ops = &btrfs_aops;
3737 inode->i_fop = &btrfs_file_operations;
3738 inode->i_op = &btrfs_file_inode_operations;
3741 inode->i_fop = &btrfs_dir_file_operations;
3742 inode->i_op = &btrfs_dir_inode_operations;
3745 inode->i_op = &btrfs_symlink_inode_operations;
3746 inode_nohighmem(inode);
3747 inode->i_mapping->a_ops = &btrfs_aops;
3750 inode->i_op = &btrfs_special_inode_operations;
3751 init_special_inode(inode, inode->i_mode, rdev);
3755 btrfs_sync_inode_flags_to_i_flags(inode);
3760 * given a leaf and an inode, copy the inode fields into the leaf
3762 static void fill_inode_item(struct btrfs_trans_handle *trans,
3763 struct extent_buffer *leaf,
3764 struct btrfs_inode_item *item,
3765 struct inode *inode)
3767 struct btrfs_map_token token;
3769 btrfs_init_map_token(&token, leaf);
3771 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3772 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3773 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3774 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3775 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3777 btrfs_set_token_timespec_sec(&token, &item->atime,
3778 inode->i_atime.tv_sec);
3779 btrfs_set_token_timespec_nsec(&token, &item->atime,
3780 inode->i_atime.tv_nsec);
3782 btrfs_set_token_timespec_sec(&token, &item->mtime,
3783 inode->i_mtime.tv_sec);
3784 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3785 inode->i_mtime.tv_nsec);
3787 btrfs_set_token_timespec_sec(&token, &item->ctime,
3788 inode->i_ctime.tv_sec);
3789 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3790 inode->i_ctime.tv_nsec);
3792 btrfs_set_token_timespec_sec(&token, &item->otime,
3793 BTRFS_I(inode)->i_otime.tv_sec);
3794 btrfs_set_token_timespec_nsec(&token, &item->otime,
3795 BTRFS_I(inode)->i_otime.tv_nsec);
3797 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3798 btrfs_set_token_inode_generation(&token, item,
3799 BTRFS_I(inode)->generation);
3800 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3801 btrfs_set_token_inode_transid(&token, item, trans->transid);
3802 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3803 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3804 btrfs_set_token_inode_block_group(&token, item, 0);
3808 * copy everything in the in-memory inode into the btree.
3810 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3811 struct btrfs_root *root,
3812 struct btrfs_inode *inode)
3814 struct btrfs_inode_item *inode_item;
3815 struct btrfs_path *path;
3816 struct extent_buffer *leaf;
3819 path = btrfs_alloc_path();
3823 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3830 leaf = path->nodes[0];
3831 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3832 struct btrfs_inode_item);
3834 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3835 btrfs_mark_buffer_dirty(leaf);
3836 btrfs_set_inode_last_trans(trans, inode);
3839 btrfs_free_path(path);
3844 * copy everything in the in-memory inode into the btree.
3846 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3847 struct btrfs_root *root,
3848 struct btrfs_inode *inode)
3850 struct btrfs_fs_info *fs_info = root->fs_info;
3854 * If the inode is a free space inode, we can deadlock during commit
3855 * if we put it into the delayed code.
3857 * The data relocation inode should also be directly updated
3860 if (!btrfs_is_free_space_inode(inode)
3861 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3862 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3863 btrfs_update_root_times(trans, root);
3865 ret = btrfs_delayed_update_inode(trans, root, inode);
3867 btrfs_set_inode_last_trans(trans, inode);
3871 return btrfs_update_inode_item(trans, root, inode);
3874 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3875 struct btrfs_root *root, struct btrfs_inode *inode)
3879 ret = btrfs_update_inode(trans, root, inode);
3881 return btrfs_update_inode_item(trans, root, inode);
3886 * unlink helper that gets used here in inode.c and in the tree logging
3887 * recovery code. It remove a link in a directory with a given name, and
3888 * also drops the back refs in the inode to the directory
3890 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3891 struct btrfs_root *root,
3892 struct btrfs_inode *dir,
3893 struct btrfs_inode *inode,
3894 const char *name, int name_len)
3896 struct btrfs_fs_info *fs_info = root->fs_info;
3897 struct btrfs_path *path;
3899 struct btrfs_dir_item *di;
3901 u64 ino = btrfs_ino(inode);
3902 u64 dir_ino = btrfs_ino(dir);
3904 path = btrfs_alloc_path();
3910 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3911 name, name_len, -1);
3912 if (IS_ERR_OR_NULL(di)) {
3913 ret = di ? PTR_ERR(di) : -ENOENT;
3916 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3919 btrfs_release_path(path);
3922 * If we don't have dir index, we have to get it by looking up
3923 * the inode ref, since we get the inode ref, remove it directly,
3924 * it is unnecessary to do delayed deletion.
3926 * But if we have dir index, needn't search inode ref to get it.
3927 * Since the inode ref is close to the inode item, it is better
3928 * that we delay to delete it, and just do this deletion when
3929 * we update the inode item.
3931 if (inode->dir_index) {
3932 ret = btrfs_delayed_delete_inode_ref(inode);
3934 index = inode->dir_index;
3939 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3943 "failed to delete reference to %.*s, inode %llu parent %llu",
3944 name_len, name, ino, dir_ino);
3945 btrfs_abort_transaction(trans, ret);
3949 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3951 btrfs_abort_transaction(trans, ret);
3955 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3957 if (ret != 0 && ret != -ENOENT) {
3958 btrfs_abort_transaction(trans, ret);
3962 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3967 btrfs_abort_transaction(trans, ret);
3970 * If we have a pending delayed iput we could end up with the final iput
3971 * being run in btrfs-cleaner context. If we have enough of these built
3972 * up we can end up burning a lot of time in btrfs-cleaner without any
3973 * way to throttle the unlinks. Since we're currently holding a ref on
3974 * the inode we can run the delayed iput here without any issues as the
3975 * final iput won't be done until after we drop the ref we're currently
3978 btrfs_run_delayed_iput(fs_info, inode);
3980 btrfs_free_path(path);
3984 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3985 inode_inc_iversion(&inode->vfs_inode);
3986 inode_inc_iversion(&dir->vfs_inode);
3987 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3988 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3989 ret = btrfs_update_inode(trans, root, dir);
3994 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3995 struct btrfs_root *root,
3996 struct btrfs_inode *dir, struct btrfs_inode *inode,
3997 const char *name, int name_len)
4000 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4002 drop_nlink(&inode->vfs_inode);
4003 ret = btrfs_update_inode(trans, root, inode);
4009 * helper to start transaction for unlink and rmdir.
4011 * unlink and rmdir are special in btrfs, they do not always free space, so
4012 * if we cannot make our reservations the normal way try and see if there is
4013 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4014 * allow the unlink to occur.
4016 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4018 struct btrfs_root *root = BTRFS_I(dir)->root;
4021 * 1 for the possible orphan item
4022 * 1 for the dir item
4023 * 1 for the dir index
4024 * 1 for the inode ref
4027 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4030 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4032 struct btrfs_root *root = BTRFS_I(dir)->root;
4033 struct btrfs_trans_handle *trans;
4034 struct inode *inode = d_inode(dentry);
4037 trans = __unlink_start_trans(dir);
4039 return PTR_ERR(trans);
4041 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4044 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4045 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4046 dentry->d_name.len);
4050 if (inode->i_nlink == 0) {
4051 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4057 btrfs_end_transaction(trans);
4058 btrfs_btree_balance_dirty(root->fs_info);
4062 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4063 struct inode *dir, struct dentry *dentry)
4065 struct btrfs_root *root = BTRFS_I(dir)->root;
4066 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4067 struct btrfs_path *path;
4068 struct extent_buffer *leaf;
4069 struct btrfs_dir_item *di;
4070 struct btrfs_key key;
4071 const char *name = dentry->d_name.name;
4072 int name_len = dentry->d_name.len;
4076 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4078 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4079 objectid = inode->root->root_key.objectid;
4080 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4081 objectid = inode->location.objectid;
4087 path = btrfs_alloc_path();
4091 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4092 name, name_len, -1);
4093 if (IS_ERR_OR_NULL(di)) {
4094 ret = di ? PTR_ERR(di) : -ENOENT;
4098 leaf = path->nodes[0];
4099 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4100 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4101 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4103 btrfs_abort_transaction(trans, ret);
4106 btrfs_release_path(path);
4109 * This is a placeholder inode for a subvolume we didn't have a
4110 * reference to at the time of the snapshot creation. In the meantime
4111 * we could have renamed the real subvol link into our snapshot, so
4112 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
4113 * Instead simply lookup the dir_index_item for this entry so we can
4114 * remove it. Otherwise we know we have a ref to the root and we can
4115 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4117 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4118 di = btrfs_search_dir_index_item(root, path, dir_ino,
4120 if (IS_ERR_OR_NULL(di)) {
4125 btrfs_abort_transaction(trans, ret);
4129 leaf = path->nodes[0];
4130 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4132 btrfs_release_path(path);
4134 ret = btrfs_del_root_ref(trans, objectid,
4135 root->root_key.objectid, dir_ino,
4136 &index, name, name_len);
4138 btrfs_abort_transaction(trans, ret);
4143 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4145 btrfs_abort_transaction(trans, ret);
4149 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4150 inode_inc_iversion(dir);
4151 dir->i_mtime = dir->i_ctime = current_time(dir);
4152 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4154 btrfs_abort_transaction(trans, ret);
4156 btrfs_free_path(path);
4161 * Helper to check if the subvolume references other subvolumes or if it's
4164 static noinline int may_destroy_subvol(struct btrfs_root *root)
4166 struct btrfs_fs_info *fs_info = root->fs_info;
4167 struct btrfs_path *path;
4168 struct btrfs_dir_item *di;
4169 struct btrfs_key key;
4173 path = btrfs_alloc_path();
4177 /* Make sure this root isn't set as the default subvol */
4178 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4179 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4180 dir_id, "default", 7, 0);
4181 if (di && !IS_ERR(di)) {
4182 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4183 if (key.objectid == root->root_key.objectid) {
4186 "deleting default subvolume %llu is not allowed",
4190 btrfs_release_path(path);
4193 key.objectid = root->root_key.objectid;
4194 key.type = BTRFS_ROOT_REF_KEY;
4195 key.offset = (u64)-1;
4197 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4203 if (path->slots[0] > 0) {
4205 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4206 if (key.objectid == root->root_key.objectid &&
4207 key.type == BTRFS_ROOT_REF_KEY)
4211 btrfs_free_path(path);
4215 /* Delete all dentries for inodes belonging to the root */
4216 static void btrfs_prune_dentries(struct btrfs_root *root)
4218 struct btrfs_fs_info *fs_info = root->fs_info;
4219 struct rb_node *node;
4220 struct rb_node *prev;
4221 struct btrfs_inode *entry;
4222 struct inode *inode;
4225 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4226 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4228 spin_lock(&root->inode_lock);
4230 node = root->inode_tree.rb_node;
4234 entry = rb_entry(node, struct btrfs_inode, rb_node);
4236 if (objectid < btrfs_ino(entry))
4237 node = node->rb_left;
4238 else if (objectid > btrfs_ino(entry))
4239 node = node->rb_right;
4245 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4246 if (objectid <= btrfs_ino(entry)) {
4250 prev = rb_next(prev);
4254 entry = rb_entry(node, struct btrfs_inode, rb_node);
4255 objectid = btrfs_ino(entry) + 1;
4256 inode = igrab(&entry->vfs_inode);
4258 spin_unlock(&root->inode_lock);
4259 if (atomic_read(&inode->i_count) > 1)
4260 d_prune_aliases(inode);
4262 * btrfs_drop_inode will have it removed from the inode
4263 * cache when its usage count hits zero.
4267 spin_lock(&root->inode_lock);
4271 if (cond_resched_lock(&root->inode_lock))
4274 node = rb_next(node);
4276 spin_unlock(&root->inode_lock);
4279 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4281 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4282 struct btrfs_root *root = BTRFS_I(dir)->root;
4283 struct inode *inode = d_inode(dentry);
4284 struct btrfs_root *dest = BTRFS_I(inode)->root;
4285 struct btrfs_trans_handle *trans;
4286 struct btrfs_block_rsv block_rsv;
4291 * Don't allow to delete a subvolume with send in progress. This is
4292 * inside the inode lock so the error handling that has to drop the bit
4293 * again is not run concurrently.
4295 spin_lock(&dest->root_item_lock);
4296 if (dest->send_in_progress) {
4297 spin_unlock(&dest->root_item_lock);
4299 "attempt to delete subvolume %llu during send",
4300 dest->root_key.objectid);
4303 root_flags = btrfs_root_flags(&dest->root_item);
4304 btrfs_set_root_flags(&dest->root_item,
4305 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4306 spin_unlock(&dest->root_item_lock);
4308 down_write(&fs_info->subvol_sem);
4310 ret = may_destroy_subvol(dest);
4314 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4316 * One for dir inode,
4317 * two for dir entries,
4318 * two for root ref/backref.
4320 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4324 trans = btrfs_start_transaction(root, 0);
4325 if (IS_ERR(trans)) {
4326 ret = PTR_ERR(trans);
4329 trans->block_rsv = &block_rsv;
4330 trans->bytes_reserved = block_rsv.size;
4332 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4334 ret = btrfs_unlink_subvol(trans, dir, dentry);
4336 btrfs_abort_transaction(trans, ret);
4340 ret = btrfs_record_root_in_trans(trans, dest);
4342 btrfs_abort_transaction(trans, ret);
4346 memset(&dest->root_item.drop_progress, 0,
4347 sizeof(dest->root_item.drop_progress));
4348 btrfs_set_root_drop_level(&dest->root_item, 0);
4349 btrfs_set_root_refs(&dest->root_item, 0);
4351 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4352 ret = btrfs_insert_orphan_item(trans,
4354 dest->root_key.objectid);
4356 btrfs_abort_transaction(trans, ret);
4361 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4362 BTRFS_UUID_KEY_SUBVOL,
4363 dest->root_key.objectid);
4364 if (ret && ret != -ENOENT) {
4365 btrfs_abort_transaction(trans, ret);
4368 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4369 ret = btrfs_uuid_tree_remove(trans,
4370 dest->root_item.received_uuid,
4371 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4372 dest->root_key.objectid);
4373 if (ret && ret != -ENOENT) {
4374 btrfs_abort_transaction(trans, ret);
4379 free_anon_bdev(dest->anon_dev);
4382 trans->block_rsv = NULL;
4383 trans->bytes_reserved = 0;
4384 ret = btrfs_end_transaction(trans);
4385 inode->i_flags |= S_DEAD;
4387 btrfs_subvolume_release_metadata(root, &block_rsv);
4389 up_write(&fs_info->subvol_sem);
4391 spin_lock(&dest->root_item_lock);
4392 root_flags = btrfs_root_flags(&dest->root_item);
4393 btrfs_set_root_flags(&dest->root_item,
4394 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4395 spin_unlock(&dest->root_item_lock);
4397 d_invalidate(dentry);
4398 btrfs_prune_dentries(dest);
4399 ASSERT(dest->send_in_progress == 0);
4405 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4407 struct inode *inode = d_inode(dentry);
4409 struct btrfs_root *root = BTRFS_I(dir)->root;
4410 struct btrfs_trans_handle *trans;
4411 u64 last_unlink_trans;
4413 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4415 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4416 return btrfs_delete_subvolume(dir, dentry);
4418 trans = __unlink_start_trans(dir);
4420 return PTR_ERR(trans);
4422 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4423 err = btrfs_unlink_subvol(trans, dir, dentry);
4427 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4431 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4433 /* now the directory is empty */
4434 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4435 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4436 dentry->d_name.len);
4438 btrfs_i_size_write(BTRFS_I(inode), 0);
4440 * Propagate the last_unlink_trans value of the deleted dir to
4441 * its parent directory. This is to prevent an unrecoverable
4442 * log tree in the case we do something like this:
4444 * 2) create snapshot under dir foo
4445 * 3) delete the snapshot
4448 * 6) fsync foo or some file inside foo
4450 if (last_unlink_trans >= trans->transid)
4451 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4454 btrfs_end_transaction(trans);
4455 btrfs_btree_balance_dirty(root->fs_info);
4461 * Return this if we need to call truncate_block for the last bit of the
4464 #define NEED_TRUNCATE_BLOCK 1
4467 * this can truncate away extent items, csum items and directory items.
4468 * It starts at a high offset and removes keys until it can't find
4469 * any higher than new_size
4471 * csum items that cross the new i_size are truncated to the new size
4474 * min_type is the minimum key type to truncate down to. If set to 0, this
4475 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4477 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4478 struct btrfs_root *root,
4479 struct btrfs_inode *inode,
4480 u64 new_size, u32 min_type)
4482 struct btrfs_fs_info *fs_info = root->fs_info;
4483 struct btrfs_path *path;
4484 struct extent_buffer *leaf;
4485 struct btrfs_file_extent_item *fi;
4486 struct btrfs_key key;
4487 struct btrfs_key found_key;
4488 u64 extent_start = 0;
4489 u64 extent_num_bytes = 0;
4490 u64 extent_offset = 0;
4492 u64 last_size = new_size;
4493 u32 found_type = (u8)-1;
4496 int pending_del_nr = 0;
4497 int pending_del_slot = 0;
4498 int extent_type = -1;
4500 u64 ino = btrfs_ino(inode);
4501 u64 bytes_deleted = 0;
4502 bool be_nice = false;
4503 bool should_throttle = false;
4504 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4505 struct extent_state *cached_state = NULL;
4507 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4510 * For non-free space inodes and non-shareable roots, we want to back
4511 * off from time to time. This means all inodes in subvolume roots,
4512 * reloc roots, and data reloc roots.
4514 if (!btrfs_is_free_space_inode(inode) &&
4515 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4518 path = btrfs_alloc_path();
4521 path->reada = READA_BACK;
4523 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4524 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4528 * We want to drop from the next block forward in case this
4529 * new size is not block aligned since we will be keeping the
4530 * last block of the extent just the way it is.
4532 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4533 fs_info->sectorsize),
4538 * This function is also used to drop the items in the log tree before
4539 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4540 * it is used to drop the logged items. So we shouldn't kill the delayed
4543 if (min_type == 0 && root == inode->root)
4544 btrfs_kill_delayed_inode_items(inode);
4547 key.offset = (u64)-1;
4552 * with a 16K leaf size and 128MB extents, you can actually queue
4553 * up a huge file in a single leaf. Most of the time that
4554 * bytes_deleted is > 0, it will be huge by the time we get here
4556 if (be_nice && bytes_deleted > SZ_32M &&
4557 btrfs_should_end_transaction(trans)) {
4562 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4568 /* there are no items in the tree for us to truncate, we're
4571 if (path->slots[0] == 0)
4577 u64 clear_start = 0, clear_len = 0;
4580 leaf = path->nodes[0];
4581 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4582 found_type = found_key.type;
4584 if (found_key.objectid != ino)
4587 if (found_type < min_type)
4590 item_end = found_key.offset;
4591 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4592 fi = btrfs_item_ptr(leaf, path->slots[0],
4593 struct btrfs_file_extent_item);
4594 extent_type = btrfs_file_extent_type(leaf, fi);
4595 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4597 btrfs_file_extent_num_bytes(leaf, fi);
4599 trace_btrfs_truncate_show_fi_regular(
4600 inode, leaf, fi, found_key.offset);
4601 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4602 item_end += btrfs_file_extent_ram_bytes(leaf,
4605 trace_btrfs_truncate_show_fi_inline(
4606 inode, leaf, fi, path->slots[0],
4611 if (found_type > min_type) {
4614 if (item_end < new_size)
4616 if (found_key.offset >= new_size)
4622 /* FIXME, shrink the extent if the ref count is only 1 */
4623 if (found_type != BTRFS_EXTENT_DATA_KEY)
4626 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4629 clear_start = found_key.offset;
4630 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4632 u64 orig_num_bytes =
4633 btrfs_file_extent_num_bytes(leaf, fi);
4634 extent_num_bytes = ALIGN(new_size -
4636 fs_info->sectorsize);
4637 clear_start = ALIGN(new_size, fs_info->sectorsize);
4638 btrfs_set_file_extent_num_bytes(leaf, fi,
4640 num_dec = (orig_num_bytes -
4642 if (test_bit(BTRFS_ROOT_SHAREABLE,
4645 inode_sub_bytes(&inode->vfs_inode,
4647 btrfs_mark_buffer_dirty(leaf);
4650 btrfs_file_extent_disk_num_bytes(leaf,
4652 extent_offset = found_key.offset -
4653 btrfs_file_extent_offset(leaf, fi);
4655 /* FIXME blocksize != 4096 */
4656 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4657 if (extent_start != 0) {
4659 if (test_bit(BTRFS_ROOT_SHAREABLE,
4661 inode_sub_bytes(&inode->vfs_inode,
4665 clear_len = num_dec;
4666 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4668 * we can't truncate inline items that have had
4672 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4673 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4674 btrfs_file_extent_compression(leaf, fi) == 0) {
4675 u32 size = (u32)(new_size - found_key.offset);
4677 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4678 size = btrfs_file_extent_calc_inline_size(size);
4679 btrfs_truncate_item(path, size, 1);
4680 } else if (!del_item) {
4682 * We have to bail so the last_size is set to
4683 * just before this extent.
4685 ret = NEED_TRUNCATE_BLOCK;
4689 * Inline extents are special, we just treat
4690 * them as a full sector worth in the file
4691 * extent tree just for simplicity sake.
4693 clear_len = fs_info->sectorsize;
4696 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4697 inode_sub_bytes(&inode->vfs_inode,
4698 item_end + 1 - new_size);
4702 * We use btrfs_truncate_inode_items() to clean up log trees for
4703 * multiple fsyncs, and in this case we don't want to clear the
4704 * file extent range because it's just the log.
4706 if (root == inode->root) {
4707 ret = btrfs_inode_clear_file_extent_range(inode,
4708 clear_start, clear_len);
4710 btrfs_abort_transaction(trans, ret);
4716 last_size = found_key.offset;
4718 last_size = new_size;
4720 if (!pending_del_nr) {
4721 /* no pending yet, add ourselves */
4722 pending_del_slot = path->slots[0];
4724 } else if (pending_del_nr &&
4725 path->slots[0] + 1 == pending_del_slot) {
4726 /* hop on the pending chunk */
4728 pending_del_slot = path->slots[0];
4735 should_throttle = false;
4738 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4739 struct btrfs_ref ref = { 0 };
4741 bytes_deleted += extent_num_bytes;
4743 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4744 extent_start, extent_num_bytes, 0);
4745 ref.real_root = root->root_key.objectid;
4746 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4747 ino, extent_offset);
4748 ret = btrfs_free_extent(trans, &ref);
4750 btrfs_abort_transaction(trans, ret);
4754 if (btrfs_should_throttle_delayed_refs(trans))
4755 should_throttle = true;
4759 if (found_type == BTRFS_INODE_ITEM_KEY)
4762 if (path->slots[0] == 0 ||
4763 path->slots[0] != pending_del_slot ||
4765 if (pending_del_nr) {
4766 ret = btrfs_del_items(trans, root, path,
4770 btrfs_abort_transaction(trans, ret);
4775 btrfs_release_path(path);
4778 * We can generate a lot of delayed refs, so we need to
4779 * throttle every once and a while and make sure we're
4780 * adding enough space to keep up with the work we are
4781 * generating. Since we hold a transaction here we
4782 * can't flush, and we don't want to FLUSH_LIMIT because
4783 * we could have generated too many delayed refs to
4784 * actually allocate, so just bail if we're short and
4785 * let the normal reservation dance happen higher up.
4787 if (should_throttle) {
4788 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4789 BTRFS_RESERVE_NO_FLUSH);
4801 if (ret >= 0 && pending_del_nr) {
4804 err = btrfs_del_items(trans, root, path, pending_del_slot,
4807 btrfs_abort_transaction(trans, err);
4811 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4812 ASSERT(last_size >= new_size);
4813 if (!ret && last_size > new_size)
4814 last_size = new_size;
4815 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4816 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4820 btrfs_free_path(path);
4825 * btrfs_truncate_block - read, zero a chunk and write a block
4826 * @inode - inode that we're zeroing
4827 * @from - the offset to start zeroing
4828 * @len - the length to zero, 0 to zero the entire range respective to the
4830 * @front - zero up to the offset instead of from the offset on
4832 * This will find the block for the "from" offset and cow the block and zero the
4833 * part we want to zero. This is used with truncate and hole punching.
4835 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4838 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4839 struct address_space *mapping = inode->vfs_inode.i_mapping;
4840 struct extent_io_tree *io_tree = &inode->io_tree;
4841 struct btrfs_ordered_extent *ordered;
4842 struct extent_state *cached_state = NULL;
4843 struct extent_changeset *data_reserved = NULL;
4844 bool only_release_metadata = false;
4845 u32 blocksize = fs_info->sectorsize;
4846 pgoff_t index = from >> PAGE_SHIFT;
4847 unsigned offset = from & (blocksize - 1);
4849 gfp_t mask = btrfs_alloc_write_mask(mapping);
4850 size_t write_bytes = blocksize;
4855 if (IS_ALIGNED(offset, blocksize) &&
4856 (!len || IS_ALIGNED(len, blocksize)))
4859 block_start = round_down(from, blocksize);
4860 block_end = block_start + blocksize - 1;
4862 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4865 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4866 /* For nocow case, no need to reserve data space */
4867 only_release_metadata = true;
4872 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4874 if (!only_release_metadata)
4875 btrfs_free_reserved_data_space(inode, data_reserved,
4876 block_start, blocksize);
4880 page = find_or_create_page(mapping, index, mask);
4882 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4884 btrfs_delalloc_release_extents(inode, blocksize);
4888 ret = set_page_extent_mapped(page);
4892 if (!PageUptodate(page)) {
4893 ret = btrfs_readpage(NULL, page);
4895 if (page->mapping != mapping) {
4900 if (!PageUptodate(page)) {
4905 wait_on_page_writeback(page);
4907 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4909 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4911 unlock_extent_cached(io_tree, block_start, block_end,
4915 btrfs_start_ordered_extent(ordered, 1);
4916 btrfs_put_ordered_extent(ordered);
4920 clear_extent_bit(&inode->io_tree, block_start, block_end,
4921 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4922 0, 0, &cached_state);
4924 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4927 unlock_extent_cached(io_tree, block_start, block_end,
4932 if (offset != blocksize) {
4934 len = blocksize - offset;
4936 memzero_page(page, (block_start - page_offset(page)),
4939 memzero_page(page, (block_start - page_offset(page)) + offset,
4941 flush_dcache_page(page);
4943 ClearPageChecked(page);
4944 set_page_dirty(page);
4945 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4947 if (only_release_metadata)
4948 set_extent_bit(&inode->io_tree, block_start, block_end,
4949 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4953 if (only_release_metadata)
4954 btrfs_delalloc_release_metadata(inode, blocksize, true);
4956 btrfs_delalloc_release_space(inode, data_reserved,
4957 block_start, blocksize, true);
4959 btrfs_delalloc_release_extents(inode, blocksize);
4963 if (only_release_metadata)
4964 btrfs_check_nocow_unlock(inode);
4965 extent_changeset_free(data_reserved);
4969 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4970 u64 offset, u64 len)
4972 struct btrfs_fs_info *fs_info = root->fs_info;
4973 struct btrfs_trans_handle *trans;
4974 struct btrfs_drop_extents_args drop_args = { 0 };
4978 * Still need to make sure the inode looks like it's been updated so
4979 * that any holes get logged if we fsync.
4981 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4982 inode->last_trans = fs_info->generation;
4983 inode->last_sub_trans = root->log_transid;
4984 inode->last_log_commit = root->last_log_commit;
4989 * 1 - for the one we're dropping
4990 * 1 - for the one we're adding
4991 * 1 - for updating the inode.
4993 trans = btrfs_start_transaction(root, 3);
4995 return PTR_ERR(trans);
4997 drop_args.start = offset;
4998 drop_args.end = offset + len;
4999 drop_args.drop_cache = true;
5001 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5003 btrfs_abort_transaction(trans, ret);
5004 btrfs_end_transaction(trans);
5008 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5009 offset, 0, 0, len, 0, len, 0, 0, 0);
5011 btrfs_abort_transaction(trans, ret);
5013 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5014 btrfs_update_inode(trans, root, inode);
5016 btrfs_end_transaction(trans);
5021 * This function puts in dummy file extents for the area we're creating a hole
5022 * for. So if we are truncating this file to a larger size we need to insert
5023 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5024 * the range between oldsize and size
5026 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5028 struct btrfs_root *root = inode->root;
5029 struct btrfs_fs_info *fs_info = root->fs_info;
5030 struct extent_io_tree *io_tree = &inode->io_tree;
5031 struct extent_map *em = NULL;
5032 struct extent_state *cached_state = NULL;
5033 struct extent_map_tree *em_tree = &inode->extent_tree;
5034 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5035 u64 block_end = ALIGN(size, fs_info->sectorsize);
5042 * If our size started in the middle of a block we need to zero out the
5043 * rest of the block before we expand the i_size, otherwise we could
5044 * expose stale data.
5046 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5050 if (size <= hole_start)
5053 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5055 cur_offset = hole_start;
5057 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5058 block_end - cur_offset);
5064 last_byte = min(extent_map_end(em), block_end);
5065 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5066 hole_size = last_byte - cur_offset;
5068 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5069 struct extent_map *hole_em;
5071 err = maybe_insert_hole(root, inode, cur_offset,
5076 err = btrfs_inode_set_file_extent_range(inode,
5077 cur_offset, hole_size);
5081 btrfs_drop_extent_cache(inode, cur_offset,
5082 cur_offset + hole_size - 1, 0);
5083 hole_em = alloc_extent_map();
5085 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5086 &inode->runtime_flags);
5089 hole_em->start = cur_offset;
5090 hole_em->len = hole_size;
5091 hole_em->orig_start = cur_offset;
5093 hole_em->block_start = EXTENT_MAP_HOLE;
5094 hole_em->block_len = 0;
5095 hole_em->orig_block_len = 0;
5096 hole_em->ram_bytes = hole_size;
5097 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5098 hole_em->generation = fs_info->generation;
5101 write_lock(&em_tree->lock);
5102 err = add_extent_mapping(em_tree, hole_em, 1);
5103 write_unlock(&em_tree->lock);
5106 btrfs_drop_extent_cache(inode, cur_offset,
5110 free_extent_map(hole_em);
5112 err = btrfs_inode_set_file_extent_range(inode,
5113 cur_offset, hole_size);
5118 free_extent_map(em);
5120 cur_offset = last_byte;
5121 if (cur_offset >= block_end)
5124 free_extent_map(em);
5125 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5129 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5131 struct btrfs_root *root = BTRFS_I(inode)->root;
5132 struct btrfs_trans_handle *trans;
5133 loff_t oldsize = i_size_read(inode);
5134 loff_t newsize = attr->ia_size;
5135 int mask = attr->ia_valid;
5139 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5140 * special case where we need to update the times despite not having
5141 * these flags set. For all other operations the VFS set these flags
5142 * explicitly if it wants a timestamp update.
5144 if (newsize != oldsize) {
5145 inode_inc_iversion(inode);
5146 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5147 inode->i_ctime = inode->i_mtime =
5148 current_time(inode);
5151 if (newsize > oldsize) {
5153 * Don't do an expanding truncate while snapshotting is ongoing.
5154 * This is to ensure the snapshot captures a fully consistent
5155 * state of this file - if the snapshot captures this expanding
5156 * truncation, it must capture all writes that happened before
5159 btrfs_drew_write_lock(&root->snapshot_lock);
5160 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5162 btrfs_drew_write_unlock(&root->snapshot_lock);
5166 trans = btrfs_start_transaction(root, 1);
5167 if (IS_ERR(trans)) {
5168 btrfs_drew_write_unlock(&root->snapshot_lock);
5169 return PTR_ERR(trans);
5172 i_size_write(inode, newsize);
5173 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5174 pagecache_isize_extended(inode, oldsize, newsize);
5175 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5176 btrfs_drew_write_unlock(&root->snapshot_lock);
5177 btrfs_end_transaction(trans);
5179 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5181 if (btrfs_is_zoned(fs_info)) {
5182 ret = btrfs_wait_ordered_range(inode,
5183 ALIGN(newsize, fs_info->sectorsize),
5190 * We're truncating a file that used to have good data down to
5191 * zero. Make sure any new writes to the file get on disk
5195 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5196 &BTRFS_I(inode)->runtime_flags);
5198 truncate_setsize(inode, newsize);
5200 inode_dio_wait(inode);
5202 ret = btrfs_truncate(inode, newsize == oldsize);
5203 if (ret && inode->i_nlink) {
5207 * Truncate failed, so fix up the in-memory size. We
5208 * adjusted disk_i_size down as we removed extents, so
5209 * wait for disk_i_size to be stable and then update the
5210 * in-memory size to match.
5212 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5215 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5222 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5225 struct inode *inode = d_inode(dentry);
5226 struct btrfs_root *root = BTRFS_I(inode)->root;
5229 if (btrfs_root_readonly(root))
5232 err = setattr_prepare(&init_user_ns, dentry, attr);
5236 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5237 err = btrfs_setsize(inode, attr);
5242 if (attr->ia_valid) {
5243 setattr_copy(&init_user_ns, inode, attr);
5244 inode_inc_iversion(inode);
5245 err = btrfs_dirty_inode(inode);
5247 if (!err && attr->ia_valid & ATTR_MODE)
5248 err = posix_acl_chmod(&init_user_ns, inode,
5256 * While truncating the inode pages during eviction, we get the VFS calling
5257 * btrfs_invalidatepage() against each page of the inode. This is slow because
5258 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5259 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5260 * extent_state structures over and over, wasting lots of time.
5262 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5263 * those expensive operations on a per page basis and do only the ordered io
5264 * finishing, while we release here the extent_map and extent_state structures,
5265 * without the excessive merging and splitting.
5267 static void evict_inode_truncate_pages(struct inode *inode)
5269 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5270 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5271 struct rb_node *node;
5273 ASSERT(inode->i_state & I_FREEING);
5274 truncate_inode_pages_final(&inode->i_data);
5276 write_lock(&map_tree->lock);
5277 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5278 struct extent_map *em;
5280 node = rb_first_cached(&map_tree->map);
5281 em = rb_entry(node, struct extent_map, rb_node);
5282 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5283 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5284 remove_extent_mapping(map_tree, em);
5285 free_extent_map(em);
5286 if (need_resched()) {
5287 write_unlock(&map_tree->lock);
5289 write_lock(&map_tree->lock);
5292 write_unlock(&map_tree->lock);
5295 * Keep looping until we have no more ranges in the io tree.
5296 * We can have ongoing bios started by readahead that have
5297 * their endio callback (extent_io.c:end_bio_extent_readpage)
5298 * still in progress (unlocked the pages in the bio but did not yet
5299 * unlocked the ranges in the io tree). Therefore this means some
5300 * ranges can still be locked and eviction started because before
5301 * submitting those bios, which are executed by a separate task (work
5302 * queue kthread), inode references (inode->i_count) were not taken
5303 * (which would be dropped in the end io callback of each bio).
5304 * Therefore here we effectively end up waiting for those bios and
5305 * anyone else holding locked ranges without having bumped the inode's
5306 * reference count - if we don't do it, when they access the inode's
5307 * io_tree to unlock a range it may be too late, leading to an
5308 * use-after-free issue.
5310 spin_lock(&io_tree->lock);
5311 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5312 struct extent_state *state;
5313 struct extent_state *cached_state = NULL;
5316 unsigned state_flags;
5318 node = rb_first(&io_tree->state);
5319 state = rb_entry(node, struct extent_state, rb_node);
5320 start = state->start;
5322 state_flags = state->state;
5323 spin_unlock(&io_tree->lock);
5325 lock_extent_bits(io_tree, start, end, &cached_state);
5328 * If still has DELALLOC flag, the extent didn't reach disk,
5329 * and its reserved space won't be freed by delayed_ref.
5330 * So we need to free its reserved space here.
5331 * (Refer to comment in btrfs_invalidatepage, case 2)
5333 * Note, end is the bytenr of last byte, so we need + 1 here.
5335 if (state_flags & EXTENT_DELALLOC)
5336 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5339 clear_extent_bit(io_tree, start, end,
5340 EXTENT_LOCKED | EXTENT_DELALLOC |
5341 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5345 spin_lock(&io_tree->lock);
5347 spin_unlock(&io_tree->lock);
5350 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5351 struct btrfs_block_rsv *rsv)
5353 struct btrfs_fs_info *fs_info = root->fs_info;
5354 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5355 struct btrfs_trans_handle *trans;
5356 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5360 * Eviction should be taking place at some place safe because of our
5361 * delayed iputs. However the normal flushing code will run delayed
5362 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5364 * We reserve the delayed_refs_extra here again because we can't use
5365 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5366 * above. We reserve our extra bit here because we generate a ton of
5367 * delayed refs activity by truncating.
5369 * If we cannot make our reservation we'll attempt to steal from the
5370 * global reserve, because we really want to be able to free up space.
5372 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5373 BTRFS_RESERVE_FLUSH_EVICT);
5376 * Try to steal from the global reserve if there is space for
5379 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5380 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5382 "could not allocate space for delete; will truncate on mount");
5383 return ERR_PTR(-ENOSPC);
5385 delayed_refs_extra = 0;
5388 trans = btrfs_join_transaction(root);
5392 if (delayed_refs_extra) {
5393 trans->block_rsv = &fs_info->trans_block_rsv;
5394 trans->bytes_reserved = delayed_refs_extra;
5395 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5396 delayed_refs_extra, 1);
5401 void btrfs_evict_inode(struct inode *inode)
5403 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5404 struct btrfs_trans_handle *trans;
5405 struct btrfs_root *root = BTRFS_I(inode)->root;
5406 struct btrfs_block_rsv *rsv;
5409 trace_btrfs_inode_evict(inode);
5416 evict_inode_truncate_pages(inode);
5418 if (inode->i_nlink &&
5419 ((btrfs_root_refs(&root->root_item) != 0 &&
5420 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5421 btrfs_is_free_space_inode(BTRFS_I(inode))))
5424 if (is_bad_inode(inode))
5427 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5429 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5432 if (inode->i_nlink > 0) {
5433 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5434 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5438 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5442 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5445 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5448 btrfs_i_size_write(BTRFS_I(inode), 0);
5451 trans = evict_refill_and_join(root, rsv);
5455 trans->block_rsv = rsv;
5457 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5459 trans->block_rsv = &fs_info->trans_block_rsv;
5460 btrfs_end_transaction(trans);
5461 btrfs_btree_balance_dirty(fs_info);
5462 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5469 * Errors here aren't a big deal, it just means we leave orphan items in
5470 * the tree. They will be cleaned up on the next mount. If the inode
5471 * number gets reused, cleanup deletes the orphan item without doing
5472 * anything, and unlink reuses the existing orphan item.
5474 * If it turns out that we are dropping too many of these, we might want
5475 * to add a mechanism for retrying these after a commit.
5477 trans = evict_refill_and_join(root, rsv);
5478 if (!IS_ERR(trans)) {
5479 trans->block_rsv = rsv;
5480 btrfs_orphan_del(trans, BTRFS_I(inode));
5481 trans->block_rsv = &fs_info->trans_block_rsv;
5482 btrfs_end_transaction(trans);
5486 btrfs_free_block_rsv(fs_info, rsv);
5489 * If we didn't successfully delete, the orphan item will still be in
5490 * the tree and we'll retry on the next mount. Again, we might also want
5491 * to retry these periodically in the future.
5493 btrfs_remove_delayed_node(BTRFS_I(inode));
5498 * Return the key found in the dir entry in the location pointer, fill @type
5499 * with BTRFS_FT_*, and return 0.
5501 * If no dir entries were found, returns -ENOENT.
5502 * If found a corrupted location in dir entry, returns -EUCLEAN.
5504 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5505 struct btrfs_key *location, u8 *type)
5507 const char *name = dentry->d_name.name;
5508 int namelen = dentry->d_name.len;
5509 struct btrfs_dir_item *di;
5510 struct btrfs_path *path;
5511 struct btrfs_root *root = BTRFS_I(dir)->root;
5514 path = btrfs_alloc_path();
5518 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5520 if (IS_ERR_OR_NULL(di)) {
5521 ret = di ? PTR_ERR(di) : -ENOENT;
5525 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5526 if (location->type != BTRFS_INODE_ITEM_KEY &&
5527 location->type != BTRFS_ROOT_ITEM_KEY) {
5529 btrfs_warn(root->fs_info,
5530 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5531 __func__, name, btrfs_ino(BTRFS_I(dir)),
5532 location->objectid, location->type, location->offset);
5535 *type = btrfs_dir_type(path->nodes[0], di);
5537 btrfs_free_path(path);
5542 * when we hit a tree root in a directory, the btrfs part of the inode
5543 * needs to be changed to reflect the root directory of the tree root. This
5544 * is kind of like crossing a mount point.
5546 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5548 struct dentry *dentry,
5549 struct btrfs_key *location,
5550 struct btrfs_root **sub_root)
5552 struct btrfs_path *path;
5553 struct btrfs_root *new_root;
5554 struct btrfs_root_ref *ref;
5555 struct extent_buffer *leaf;
5556 struct btrfs_key key;
5560 path = btrfs_alloc_path();
5567 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5568 key.type = BTRFS_ROOT_REF_KEY;
5569 key.offset = location->objectid;
5571 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5578 leaf = path->nodes[0];
5579 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5580 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5581 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5584 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5585 (unsigned long)(ref + 1),
5586 dentry->d_name.len);
5590 btrfs_release_path(path);
5592 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5593 if (IS_ERR(new_root)) {
5594 err = PTR_ERR(new_root);
5598 *sub_root = new_root;
5599 location->objectid = btrfs_root_dirid(&new_root->root_item);
5600 location->type = BTRFS_INODE_ITEM_KEY;
5601 location->offset = 0;
5604 btrfs_free_path(path);
5608 static void inode_tree_add(struct inode *inode)
5610 struct btrfs_root *root = BTRFS_I(inode)->root;
5611 struct btrfs_inode *entry;
5613 struct rb_node *parent;
5614 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5615 u64 ino = btrfs_ino(BTRFS_I(inode));
5617 if (inode_unhashed(inode))
5620 spin_lock(&root->inode_lock);
5621 p = &root->inode_tree.rb_node;
5624 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5626 if (ino < btrfs_ino(entry))
5627 p = &parent->rb_left;
5628 else if (ino > btrfs_ino(entry))
5629 p = &parent->rb_right;
5631 WARN_ON(!(entry->vfs_inode.i_state &
5632 (I_WILL_FREE | I_FREEING)));
5633 rb_replace_node(parent, new, &root->inode_tree);
5634 RB_CLEAR_NODE(parent);
5635 spin_unlock(&root->inode_lock);
5639 rb_link_node(new, parent, p);
5640 rb_insert_color(new, &root->inode_tree);
5641 spin_unlock(&root->inode_lock);
5644 static void inode_tree_del(struct btrfs_inode *inode)
5646 struct btrfs_root *root = inode->root;
5649 spin_lock(&root->inode_lock);
5650 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5651 rb_erase(&inode->rb_node, &root->inode_tree);
5652 RB_CLEAR_NODE(&inode->rb_node);
5653 empty = RB_EMPTY_ROOT(&root->inode_tree);
5655 spin_unlock(&root->inode_lock);
5657 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5658 spin_lock(&root->inode_lock);
5659 empty = RB_EMPTY_ROOT(&root->inode_tree);
5660 spin_unlock(&root->inode_lock);
5662 btrfs_add_dead_root(root);
5667 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5669 struct btrfs_iget_args *args = p;
5671 inode->i_ino = args->ino;
5672 BTRFS_I(inode)->location.objectid = args->ino;
5673 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5674 BTRFS_I(inode)->location.offset = 0;
5675 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5676 BUG_ON(args->root && !BTRFS_I(inode)->root);
5680 static int btrfs_find_actor(struct inode *inode, void *opaque)
5682 struct btrfs_iget_args *args = opaque;
5684 return args->ino == BTRFS_I(inode)->location.objectid &&
5685 args->root == BTRFS_I(inode)->root;
5688 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5689 struct btrfs_root *root)
5691 struct inode *inode;
5692 struct btrfs_iget_args args;
5693 unsigned long hashval = btrfs_inode_hash(ino, root);
5698 inode = iget5_locked(s, hashval, btrfs_find_actor,
5699 btrfs_init_locked_inode,
5705 * Get an inode object given its inode number and corresponding root.
5706 * Path can be preallocated to prevent recursing back to iget through
5707 * allocator. NULL is also valid but may require an additional allocation
5710 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5711 struct btrfs_root *root, struct btrfs_path *path)
5713 struct inode *inode;
5715 inode = btrfs_iget_locked(s, ino, root);
5717 return ERR_PTR(-ENOMEM);
5719 if (inode->i_state & I_NEW) {
5722 ret = btrfs_read_locked_inode(inode, path);
5724 inode_tree_add(inode);
5725 unlock_new_inode(inode);
5729 * ret > 0 can come from btrfs_search_slot called by
5730 * btrfs_read_locked_inode, this means the inode item
5735 inode = ERR_PTR(ret);
5742 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5744 return btrfs_iget_path(s, ino, root, NULL);
5747 static struct inode *new_simple_dir(struct super_block *s,
5748 struct btrfs_key *key,
5749 struct btrfs_root *root)
5751 struct inode *inode = new_inode(s);
5754 return ERR_PTR(-ENOMEM);
5756 BTRFS_I(inode)->root = btrfs_grab_root(root);
5757 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5758 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5760 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5762 * We only need lookup, the rest is read-only and there's no inode
5763 * associated with the dentry
5765 inode->i_op = &simple_dir_inode_operations;
5766 inode->i_opflags &= ~IOP_XATTR;
5767 inode->i_fop = &simple_dir_operations;
5768 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5769 inode->i_mtime = current_time(inode);
5770 inode->i_atime = inode->i_mtime;
5771 inode->i_ctime = inode->i_mtime;
5772 BTRFS_I(inode)->i_otime = inode->i_mtime;
5777 static inline u8 btrfs_inode_type(struct inode *inode)
5780 * Compile-time asserts that generic FT_* types still match
5783 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5784 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5785 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5786 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5787 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5788 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5789 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5790 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5792 return fs_umode_to_ftype(inode->i_mode);
5795 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5797 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5798 struct inode *inode;
5799 struct btrfs_root *root = BTRFS_I(dir)->root;
5800 struct btrfs_root *sub_root = root;
5801 struct btrfs_key location;
5805 if (dentry->d_name.len > BTRFS_NAME_LEN)
5806 return ERR_PTR(-ENAMETOOLONG);
5808 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5810 return ERR_PTR(ret);
5812 if (location.type == BTRFS_INODE_ITEM_KEY) {
5813 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5817 /* Do extra check against inode mode with di_type */
5818 if (btrfs_inode_type(inode) != di_type) {
5820 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5821 inode->i_mode, btrfs_inode_type(inode),
5824 return ERR_PTR(-EUCLEAN);
5829 ret = fixup_tree_root_location(fs_info, dir, dentry,
5830 &location, &sub_root);
5833 inode = ERR_PTR(ret);
5835 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5837 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5839 if (root != sub_root)
5840 btrfs_put_root(sub_root);
5842 if (!IS_ERR(inode) && root != sub_root) {
5843 down_read(&fs_info->cleanup_work_sem);
5844 if (!sb_rdonly(inode->i_sb))
5845 ret = btrfs_orphan_cleanup(sub_root);
5846 up_read(&fs_info->cleanup_work_sem);
5849 inode = ERR_PTR(ret);
5856 static int btrfs_dentry_delete(const struct dentry *dentry)
5858 struct btrfs_root *root;
5859 struct inode *inode = d_inode(dentry);
5861 if (!inode && !IS_ROOT(dentry))
5862 inode = d_inode(dentry->d_parent);
5865 root = BTRFS_I(inode)->root;
5866 if (btrfs_root_refs(&root->root_item) == 0)
5869 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5875 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5878 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5880 if (inode == ERR_PTR(-ENOENT))
5882 return d_splice_alias(inode, dentry);
5886 * All this infrastructure exists because dir_emit can fault, and we are holding
5887 * the tree lock when doing readdir. For now just allocate a buffer and copy
5888 * our information into that, and then dir_emit from the buffer. This is
5889 * similar to what NFS does, only we don't keep the buffer around in pagecache
5890 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5891 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5894 static int btrfs_opendir(struct inode *inode, struct file *file)
5896 struct btrfs_file_private *private;
5898 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5901 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5902 if (!private->filldir_buf) {
5906 file->private_data = private;
5917 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5920 struct dir_entry *entry = addr;
5921 char *name = (char *)(entry + 1);
5923 ctx->pos = get_unaligned(&entry->offset);
5924 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5925 get_unaligned(&entry->ino),
5926 get_unaligned(&entry->type)))
5928 addr += sizeof(struct dir_entry) +
5929 get_unaligned(&entry->name_len);
5935 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5937 struct inode *inode = file_inode(file);
5938 struct btrfs_root *root = BTRFS_I(inode)->root;
5939 struct btrfs_file_private *private = file->private_data;
5940 struct btrfs_dir_item *di;
5941 struct btrfs_key key;
5942 struct btrfs_key found_key;
5943 struct btrfs_path *path;
5945 struct list_head ins_list;
5946 struct list_head del_list;
5948 struct extent_buffer *leaf;
5955 struct btrfs_key location;
5957 if (!dir_emit_dots(file, ctx))
5960 path = btrfs_alloc_path();
5964 addr = private->filldir_buf;
5965 path->reada = READA_FORWARD;
5967 INIT_LIST_HEAD(&ins_list);
5968 INIT_LIST_HEAD(&del_list);
5969 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5972 key.type = BTRFS_DIR_INDEX_KEY;
5973 key.offset = ctx->pos;
5974 key.objectid = btrfs_ino(BTRFS_I(inode));
5976 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5981 struct dir_entry *entry;
5983 leaf = path->nodes[0];
5984 slot = path->slots[0];
5985 if (slot >= btrfs_header_nritems(leaf)) {
5986 ret = btrfs_next_leaf(root, path);
5994 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5996 if (found_key.objectid != key.objectid)
5998 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6000 if (found_key.offset < ctx->pos)
6002 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6004 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6005 name_len = btrfs_dir_name_len(leaf, di);
6006 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6008 btrfs_release_path(path);
6009 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6012 addr = private->filldir_buf;
6019 put_unaligned(name_len, &entry->name_len);
6020 name_ptr = (char *)(entry + 1);
6021 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6023 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6025 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6026 put_unaligned(location.objectid, &entry->ino);
6027 put_unaligned(found_key.offset, &entry->offset);
6029 addr += sizeof(struct dir_entry) + name_len;
6030 total_len += sizeof(struct dir_entry) + name_len;
6034 btrfs_release_path(path);
6036 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6040 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6045 * Stop new entries from being returned after we return the last
6048 * New directory entries are assigned a strictly increasing
6049 * offset. This means that new entries created during readdir
6050 * are *guaranteed* to be seen in the future by that readdir.
6051 * This has broken buggy programs which operate on names as
6052 * they're returned by readdir. Until we re-use freed offsets
6053 * we have this hack to stop new entries from being returned
6054 * under the assumption that they'll never reach this huge
6057 * This is being careful not to overflow 32bit loff_t unless the
6058 * last entry requires it because doing so has broken 32bit apps
6061 if (ctx->pos >= INT_MAX)
6062 ctx->pos = LLONG_MAX;
6069 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6070 btrfs_free_path(path);
6075 * This is somewhat expensive, updating the tree every time the
6076 * inode changes. But, it is most likely to find the inode in cache.
6077 * FIXME, needs more benchmarking...there are no reasons other than performance
6078 * to keep or drop this code.
6080 static int btrfs_dirty_inode(struct inode *inode)
6082 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6083 struct btrfs_root *root = BTRFS_I(inode)->root;
6084 struct btrfs_trans_handle *trans;
6087 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6090 trans = btrfs_join_transaction(root);
6092 return PTR_ERR(trans);
6094 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6095 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6096 /* whoops, lets try again with the full transaction */
6097 btrfs_end_transaction(trans);
6098 trans = btrfs_start_transaction(root, 1);
6100 return PTR_ERR(trans);
6102 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6104 btrfs_end_transaction(trans);
6105 if (BTRFS_I(inode)->delayed_node)
6106 btrfs_balance_delayed_items(fs_info);
6112 * This is a copy of file_update_time. We need this so we can return error on
6113 * ENOSPC for updating the inode in the case of file write and mmap writes.
6115 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6118 struct btrfs_root *root = BTRFS_I(inode)->root;
6119 bool dirty = flags & ~S_VERSION;
6121 if (btrfs_root_readonly(root))
6124 if (flags & S_VERSION)
6125 dirty |= inode_maybe_inc_iversion(inode, dirty);
6126 if (flags & S_CTIME)
6127 inode->i_ctime = *now;
6128 if (flags & S_MTIME)
6129 inode->i_mtime = *now;
6130 if (flags & S_ATIME)
6131 inode->i_atime = *now;
6132 return dirty ? btrfs_dirty_inode(inode) : 0;
6136 * find the highest existing sequence number in a directory
6137 * and then set the in-memory index_cnt variable to reflect
6138 * free sequence numbers
6140 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6142 struct btrfs_root *root = inode->root;
6143 struct btrfs_key key, found_key;
6144 struct btrfs_path *path;
6145 struct extent_buffer *leaf;
6148 key.objectid = btrfs_ino(inode);
6149 key.type = BTRFS_DIR_INDEX_KEY;
6150 key.offset = (u64)-1;
6152 path = btrfs_alloc_path();
6156 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6159 /* FIXME: we should be able to handle this */
6165 * MAGIC NUMBER EXPLANATION:
6166 * since we search a directory based on f_pos we have to start at 2
6167 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6168 * else has to start at 2
6170 if (path->slots[0] == 0) {
6171 inode->index_cnt = 2;
6177 leaf = path->nodes[0];
6178 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6180 if (found_key.objectid != btrfs_ino(inode) ||
6181 found_key.type != BTRFS_DIR_INDEX_KEY) {
6182 inode->index_cnt = 2;
6186 inode->index_cnt = found_key.offset + 1;
6188 btrfs_free_path(path);
6193 * helper to find a free sequence number in a given directory. This current
6194 * code is very simple, later versions will do smarter things in the btree
6196 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6200 if (dir->index_cnt == (u64)-1) {
6201 ret = btrfs_inode_delayed_dir_index_count(dir);
6203 ret = btrfs_set_inode_index_count(dir);
6209 *index = dir->index_cnt;
6215 static int btrfs_insert_inode_locked(struct inode *inode)
6217 struct btrfs_iget_args args;
6219 args.ino = BTRFS_I(inode)->location.objectid;
6220 args.root = BTRFS_I(inode)->root;
6222 return insert_inode_locked4(inode,
6223 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6224 btrfs_find_actor, &args);
6228 * Inherit flags from the parent inode.
6230 * Currently only the compression flags and the cow flags are inherited.
6232 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6239 flags = BTRFS_I(dir)->flags;
6241 if (flags & BTRFS_INODE_NOCOMPRESS) {
6242 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6243 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6244 } else if (flags & BTRFS_INODE_COMPRESS) {
6245 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6246 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6249 if (flags & BTRFS_INODE_NODATACOW) {
6250 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6251 if (S_ISREG(inode->i_mode))
6252 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6255 btrfs_sync_inode_flags_to_i_flags(inode);
6258 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6259 struct btrfs_root *root,
6261 const char *name, int name_len,
6262 u64 ref_objectid, u64 objectid,
6263 umode_t mode, u64 *index)
6265 struct btrfs_fs_info *fs_info = root->fs_info;
6266 struct inode *inode;
6267 struct btrfs_inode_item *inode_item;
6268 struct btrfs_key *location;
6269 struct btrfs_path *path;
6270 struct btrfs_inode_ref *ref;
6271 struct btrfs_key key[2];
6273 int nitems = name ? 2 : 1;
6275 unsigned int nofs_flag;
6278 path = btrfs_alloc_path();
6280 return ERR_PTR(-ENOMEM);
6282 nofs_flag = memalloc_nofs_save();
6283 inode = new_inode(fs_info->sb);
6284 memalloc_nofs_restore(nofs_flag);
6286 btrfs_free_path(path);
6287 return ERR_PTR(-ENOMEM);
6291 * O_TMPFILE, set link count to 0, so that after this point,
6292 * we fill in an inode item with the correct link count.
6295 set_nlink(inode, 0);
6298 * we have to initialize this early, so we can reclaim the inode
6299 * number if we fail afterwards in this function.
6301 inode->i_ino = objectid;
6304 trace_btrfs_inode_request(dir);
6306 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6308 btrfs_free_path(path);
6310 return ERR_PTR(ret);
6316 * index_cnt is ignored for everything but a dir,
6317 * btrfs_set_inode_index_count has an explanation for the magic
6320 BTRFS_I(inode)->index_cnt = 2;
6321 BTRFS_I(inode)->dir_index = *index;
6322 BTRFS_I(inode)->root = btrfs_grab_root(root);
6323 BTRFS_I(inode)->generation = trans->transid;
6324 inode->i_generation = BTRFS_I(inode)->generation;
6327 * We could have gotten an inode number from somebody who was fsynced
6328 * and then removed in this same transaction, so let's just set full
6329 * sync since it will be a full sync anyway and this will blow away the
6330 * old info in the log.
6332 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6334 key[0].objectid = objectid;
6335 key[0].type = BTRFS_INODE_ITEM_KEY;
6338 sizes[0] = sizeof(struct btrfs_inode_item);
6342 * Start new inodes with an inode_ref. This is slightly more
6343 * efficient for small numbers of hard links since they will
6344 * be packed into one item. Extended refs will kick in if we
6345 * add more hard links than can fit in the ref item.
6347 key[1].objectid = objectid;
6348 key[1].type = BTRFS_INODE_REF_KEY;
6349 key[1].offset = ref_objectid;
6351 sizes[1] = name_len + sizeof(*ref);
6354 location = &BTRFS_I(inode)->location;
6355 location->objectid = objectid;
6356 location->offset = 0;
6357 location->type = BTRFS_INODE_ITEM_KEY;
6359 ret = btrfs_insert_inode_locked(inode);
6365 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6369 inode_init_owner(&init_user_ns, inode, dir, mode);
6370 inode_set_bytes(inode, 0);
6372 inode->i_mtime = current_time(inode);
6373 inode->i_atime = inode->i_mtime;
6374 inode->i_ctime = inode->i_mtime;
6375 BTRFS_I(inode)->i_otime = inode->i_mtime;
6377 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6378 struct btrfs_inode_item);
6379 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6380 sizeof(*inode_item));
6381 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6384 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6385 struct btrfs_inode_ref);
6386 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6387 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6388 ptr = (unsigned long)(ref + 1);
6389 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6392 btrfs_mark_buffer_dirty(path->nodes[0]);
6393 btrfs_free_path(path);
6395 btrfs_inherit_iflags(inode, dir);
6397 if (S_ISREG(mode)) {
6398 if (btrfs_test_opt(fs_info, NODATASUM))
6399 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6400 if (btrfs_test_opt(fs_info, NODATACOW))
6401 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6402 BTRFS_INODE_NODATASUM;
6405 inode_tree_add(inode);
6407 trace_btrfs_inode_new(inode);
6408 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6410 btrfs_update_root_times(trans, root);
6412 ret = btrfs_inode_inherit_props(trans, inode, dir);
6415 "error inheriting props for ino %llu (root %llu): %d",
6416 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6421 discard_new_inode(inode);
6424 BTRFS_I(dir)->index_cnt--;
6425 btrfs_free_path(path);
6426 return ERR_PTR(ret);
6430 * utility function to add 'inode' into 'parent_inode' with
6431 * a give name and a given sequence number.
6432 * if 'add_backref' is true, also insert a backref from the
6433 * inode to the parent directory.
6435 int btrfs_add_link(struct btrfs_trans_handle *trans,
6436 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6437 const char *name, int name_len, int add_backref, u64 index)
6440 struct btrfs_key key;
6441 struct btrfs_root *root = parent_inode->root;
6442 u64 ino = btrfs_ino(inode);
6443 u64 parent_ino = btrfs_ino(parent_inode);
6445 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6446 memcpy(&key, &inode->root->root_key, sizeof(key));
6449 key.type = BTRFS_INODE_ITEM_KEY;
6453 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6454 ret = btrfs_add_root_ref(trans, key.objectid,
6455 root->root_key.objectid, parent_ino,
6456 index, name, name_len);
6457 } else if (add_backref) {
6458 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6462 /* Nothing to clean up yet */
6466 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6467 btrfs_inode_type(&inode->vfs_inode), index);
6468 if (ret == -EEXIST || ret == -EOVERFLOW)
6471 btrfs_abort_transaction(trans, ret);
6475 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6477 inode_inc_iversion(&parent_inode->vfs_inode);
6479 * If we are replaying a log tree, we do not want to update the mtime
6480 * and ctime of the parent directory with the current time, since the
6481 * log replay procedure is responsible for setting them to their correct
6482 * values (the ones it had when the fsync was done).
6484 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6485 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6487 parent_inode->vfs_inode.i_mtime = now;
6488 parent_inode->vfs_inode.i_ctime = now;
6490 ret = btrfs_update_inode(trans, root, parent_inode);
6492 btrfs_abort_transaction(trans, ret);
6496 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6499 err = btrfs_del_root_ref(trans, key.objectid,
6500 root->root_key.objectid, parent_ino,
6501 &local_index, name, name_len);
6503 btrfs_abort_transaction(trans, err);
6504 } else if (add_backref) {
6508 err = btrfs_del_inode_ref(trans, root, name, name_len,
6509 ino, parent_ino, &local_index);
6511 btrfs_abort_transaction(trans, err);
6514 /* Return the original error code */
6518 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6519 struct btrfs_inode *dir, struct dentry *dentry,
6520 struct btrfs_inode *inode, int backref, u64 index)
6522 int err = btrfs_add_link(trans, dir, inode,
6523 dentry->d_name.name, dentry->d_name.len,
6530 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6531 struct dentry *dentry, umode_t mode, dev_t rdev)
6533 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6534 struct btrfs_trans_handle *trans;
6535 struct btrfs_root *root = BTRFS_I(dir)->root;
6536 struct inode *inode = NULL;
6542 * 2 for inode item and ref
6544 * 1 for xattr if selinux is on
6546 trans = btrfs_start_transaction(root, 5);
6548 return PTR_ERR(trans);
6550 err = btrfs_get_free_objectid(root, &objectid);
6554 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6555 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6557 if (IS_ERR(inode)) {
6558 err = PTR_ERR(inode);
6564 * If the active LSM wants to access the inode during
6565 * d_instantiate it needs these. Smack checks to see
6566 * if the filesystem supports xattrs by looking at the
6569 inode->i_op = &btrfs_special_inode_operations;
6570 init_special_inode(inode, inode->i_mode, rdev);
6572 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6576 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6581 btrfs_update_inode(trans, root, BTRFS_I(inode));
6582 d_instantiate_new(dentry, inode);
6585 btrfs_end_transaction(trans);
6586 btrfs_btree_balance_dirty(fs_info);
6588 inode_dec_link_count(inode);
6589 discard_new_inode(inode);
6594 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6595 struct dentry *dentry, umode_t mode, bool excl)
6597 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6598 struct btrfs_trans_handle *trans;
6599 struct btrfs_root *root = BTRFS_I(dir)->root;
6600 struct inode *inode = NULL;
6606 * 2 for inode item and ref
6608 * 1 for xattr if selinux is on
6610 trans = btrfs_start_transaction(root, 5);
6612 return PTR_ERR(trans);
6614 err = btrfs_get_free_objectid(root, &objectid);
6618 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6619 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6621 if (IS_ERR(inode)) {
6622 err = PTR_ERR(inode);
6627 * If the active LSM wants to access the inode during
6628 * d_instantiate it needs these. Smack checks to see
6629 * if the filesystem supports xattrs by looking at the
6632 inode->i_fop = &btrfs_file_operations;
6633 inode->i_op = &btrfs_file_inode_operations;
6634 inode->i_mapping->a_ops = &btrfs_aops;
6636 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6640 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6644 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6649 d_instantiate_new(dentry, inode);
6652 btrfs_end_transaction(trans);
6654 inode_dec_link_count(inode);
6655 discard_new_inode(inode);
6657 btrfs_btree_balance_dirty(fs_info);
6661 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6662 struct dentry *dentry)
6664 struct btrfs_trans_handle *trans = NULL;
6665 struct btrfs_root *root = BTRFS_I(dir)->root;
6666 struct inode *inode = d_inode(old_dentry);
6667 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6672 /* do not allow sys_link's with other subvols of the same device */
6673 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6676 if (inode->i_nlink >= BTRFS_LINK_MAX)
6679 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6684 * 2 items for inode and inode ref
6685 * 2 items for dir items
6686 * 1 item for parent inode
6687 * 1 item for orphan item deletion if O_TMPFILE
6689 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6690 if (IS_ERR(trans)) {
6691 err = PTR_ERR(trans);
6696 /* There are several dir indexes for this inode, clear the cache. */
6697 BTRFS_I(inode)->dir_index = 0ULL;
6699 inode_inc_iversion(inode);
6700 inode->i_ctime = current_time(inode);
6702 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6704 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6710 struct dentry *parent = dentry->d_parent;
6712 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6715 if (inode->i_nlink == 1) {
6717 * If new hard link count is 1, it's a file created
6718 * with open(2) O_TMPFILE flag.
6720 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6724 d_instantiate(dentry, inode);
6725 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6730 btrfs_end_transaction(trans);
6732 inode_dec_link_count(inode);
6735 btrfs_btree_balance_dirty(fs_info);
6739 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6740 struct dentry *dentry, umode_t mode)
6742 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6743 struct inode *inode = NULL;
6744 struct btrfs_trans_handle *trans;
6745 struct btrfs_root *root = BTRFS_I(dir)->root;
6751 * 2 items for inode and ref
6752 * 2 items for dir items
6753 * 1 for xattr if selinux is on
6755 trans = btrfs_start_transaction(root, 5);
6757 return PTR_ERR(trans);
6759 err = btrfs_get_free_objectid(root, &objectid);
6763 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6764 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6765 S_IFDIR | mode, &index);
6766 if (IS_ERR(inode)) {
6767 err = PTR_ERR(inode);
6772 /* these must be set before we unlock the inode */
6773 inode->i_op = &btrfs_dir_inode_operations;
6774 inode->i_fop = &btrfs_dir_file_operations;
6776 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6780 btrfs_i_size_write(BTRFS_I(inode), 0);
6781 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6785 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6786 dentry->d_name.name,
6787 dentry->d_name.len, 0, index);
6791 d_instantiate_new(dentry, inode);
6794 btrfs_end_transaction(trans);
6796 inode_dec_link_count(inode);
6797 discard_new_inode(inode);
6799 btrfs_btree_balance_dirty(fs_info);
6803 static noinline int uncompress_inline(struct btrfs_path *path,
6805 size_t pg_offset, u64 extent_offset,
6806 struct btrfs_file_extent_item *item)
6809 struct extent_buffer *leaf = path->nodes[0];
6812 unsigned long inline_size;
6816 WARN_ON(pg_offset != 0);
6817 compress_type = btrfs_file_extent_compression(leaf, item);
6818 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6819 inline_size = btrfs_file_extent_inline_item_len(leaf,
6820 btrfs_item_nr(path->slots[0]));
6821 tmp = kmalloc(inline_size, GFP_NOFS);
6824 ptr = btrfs_file_extent_inline_start(item);
6826 read_extent_buffer(leaf, tmp, ptr, inline_size);
6828 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6829 ret = btrfs_decompress(compress_type, tmp, page,
6830 extent_offset, inline_size, max_size);
6833 * decompression code contains a memset to fill in any space between the end
6834 * of the uncompressed data and the end of max_size in case the decompressed
6835 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6836 * the end of an inline extent and the beginning of the next block, so we
6837 * cover that region here.
6840 if (max_size + pg_offset < PAGE_SIZE)
6841 memzero_page(page, pg_offset + max_size,
6842 PAGE_SIZE - max_size - pg_offset);
6848 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6849 * @inode: file to search in
6850 * @page: page to read extent data into if the extent is inline
6851 * @pg_offset: offset into @page to copy to
6852 * @start: file offset
6853 * @len: length of range starting at @start
6855 * This returns the first &struct extent_map which overlaps with the given
6856 * range, reading it from the B-tree and caching it if necessary. Note that
6857 * there may be more extents which overlap the given range after the returned
6860 * If @page is not NULL and the extent is inline, this also reads the extent
6861 * data directly into the page and marks the extent up to date in the io_tree.
6863 * Return: ERR_PTR on error, non-NULL extent_map on success.
6865 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6866 struct page *page, size_t pg_offset,
6869 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6871 u64 extent_start = 0;
6873 u64 objectid = btrfs_ino(inode);
6874 int extent_type = -1;
6875 struct btrfs_path *path = NULL;
6876 struct btrfs_root *root = inode->root;
6877 struct btrfs_file_extent_item *item;
6878 struct extent_buffer *leaf;
6879 struct btrfs_key found_key;
6880 struct extent_map *em = NULL;
6881 struct extent_map_tree *em_tree = &inode->extent_tree;
6882 struct extent_io_tree *io_tree = &inode->io_tree;
6884 read_lock(&em_tree->lock);
6885 em = lookup_extent_mapping(em_tree, start, len);
6886 read_unlock(&em_tree->lock);
6889 if (em->start > start || em->start + em->len <= start)
6890 free_extent_map(em);
6891 else if (em->block_start == EXTENT_MAP_INLINE && page)
6892 free_extent_map(em);
6896 em = alloc_extent_map();
6901 em->start = EXTENT_MAP_HOLE;
6902 em->orig_start = EXTENT_MAP_HOLE;
6904 em->block_len = (u64)-1;
6906 path = btrfs_alloc_path();
6912 /* Chances are we'll be called again, so go ahead and do readahead */
6913 path->reada = READA_FORWARD;
6916 * The same explanation in load_free_space_cache applies here as well,
6917 * we only read when we're loading the free space cache, and at that
6918 * point the commit_root has everything we need.
6920 if (btrfs_is_free_space_inode(inode)) {
6921 path->search_commit_root = 1;
6922 path->skip_locking = 1;
6925 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6928 } else if (ret > 0) {
6929 if (path->slots[0] == 0)
6935 leaf = path->nodes[0];
6936 item = btrfs_item_ptr(leaf, path->slots[0],
6937 struct btrfs_file_extent_item);
6938 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6939 if (found_key.objectid != objectid ||
6940 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6942 * If we backup past the first extent we want to move forward
6943 * and see if there is an extent in front of us, otherwise we'll
6944 * say there is a hole for our whole search range which can
6951 extent_type = btrfs_file_extent_type(leaf, item);
6952 extent_start = found_key.offset;
6953 extent_end = btrfs_file_extent_end(path);
6954 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6955 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6956 /* Only regular file could have regular/prealloc extent */
6957 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6960 "regular/prealloc extent found for non-regular inode %llu",
6964 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6966 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6967 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6972 if (start >= extent_end) {
6974 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6975 ret = btrfs_next_leaf(root, path);
6981 leaf = path->nodes[0];
6983 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6984 if (found_key.objectid != objectid ||
6985 found_key.type != BTRFS_EXTENT_DATA_KEY)
6987 if (start + len <= found_key.offset)
6989 if (start > found_key.offset)
6992 /* New extent overlaps with existing one */
6994 em->orig_start = start;
6995 em->len = found_key.offset - start;
6996 em->block_start = EXTENT_MAP_HOLE;
7000 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
7002 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7003 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7005 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7009 size_t extent_offset;
7015 size = btrfs_file_extent_ram_bytes(leaf, item);
7016 extent_offset = page_offset(page) + pg_offset - extent_start;
7017 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7018 size - extent_offset);
7019 em->start = extent_start + extent_offset;
7020 em->len = ALIGN(copy_size, fs_info->sectorsize);
7021 em->orig_block_len = em->len;
7022 em->orig_start = em->start;
7023 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7025 if (!PageUptodate(page)) {
7026 if (btrfs_file_extent_compression(leaf, item) !=
7027 BTRFS_COMPRESS_NONE) {
7028 ret = uncompress_inline(path, page, pg_offset,
7029 extent_offset, item);
7033 map = kmap_local_page(page);
7034 read_extent_buffer(leaf, map + pg_offset, ptr,
7036 if (pg_offset + copy_size < PAGE_SIZE) {
7037 memset(map + pg_offset + copy_size, 0,
7038 PAGE_SIZE - pg_offset -
7043 flush_dcache_page(page);
7045 set_extent_uptodate(io_tree, em->start,
7046 extent_map_end(em) - 1, NULL, GFP_NOFS);
7051 em->orig_start = start;
7053 em->block_start = EXTENT_MAP_HOLE;
7056 btrfs_release_path(path);
7057 if (em->start > start || extent_map_end(em) <= start) {
7059 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7060 em->start, em->len, start, len);
7065 write_lock(&em_tree->lock);
7066 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7067 write_unlock(&em_tree->lock);
7069 btrfs_free_path(path);
7071 trace_btrfs_get_extent(root, inode, em);
7074 free_extent_map(em);
7075 return ERR_PTR(ret);
7080 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7083 struct extent_map *em;
7084 struct extent_map *hole_em = NULL;
7085 u64 delalloc_start = start;
7091 em = btrfs_get_extent(inode, NULL, 0, start, len);
7095 * If our em maps to:
7097 * - a pre-alloc extent,
7098 * there might actually be delalloc bytes behind it.
7100 if (em->block_start != EXTENT_MAP_HOLE &&
7101 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7106 /* check to see if we've wrapped (len == -1 or similar) */
7115 /* ok, we didn't find anything, lets look for delalloc */
7116 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7117 end, len, EXTENT_DELALLOC, 1);
7118 delalloc_end = delalloc_start + delalloc_len;
7119 if (delalloc_end < delalloc_start)
7120 delalloc_end = (u64)-1;
7123 * We didn't find anything useful, return the original results from
7126 if (delalloc_start > end || delalloc_end <= start) {
7133 * Adjust the delalloc_start to make sure it doesn't go backwards from
7134 * the start they passed in
7136 delalloc_start = max(start, delalloc_start);
7137 delalloc_len = delalloc_end - delalloc_start;
7139 if (delalloc_len > 0) {
7142 const u64 hole_end = extent_map_end(hole_em);
7144 em = alloc_extent_map();
7152 * When btrfs_get_extent can't find anything it returns one
7155 * Make sure what it found really fits our range, and adjust to
7156 * make sure it is based on the start from the caller
7158 if (hole_end <= start || hole_em->start > end) {
7159 free_extent_map(hole_em);
7162 hole_start = max(hole_em->start, start);
7163 hole_len = hole_end - hole_start;
7166 if (hole_em && delalloc_start > hole_start) {
7168 * Our hole starts before our delalloc, so we have to
7169 * return just the parts of the hole that go until the
7172 em->len = min(hole_len, delalloc_start - hole_start);
7173 em->start = hole_start;
7174 em->orig_start = hole_start;
7176 * Don't adjust block start at all, it is fixed at
7179 em->block_start = hole_em->block_start;
7180 em->block_len = hole_len;
7181 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7182 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7185 * Hole is out of passed range or it starts after
7188 em->start = delalloc_start;
7189 em->len = delalloc_len;
7190 em->orig_start = delalloc_start;
7191 em->block_start = EXTENT_MAP_DELALLOC;
7192 em->block_len = delalloc_len;
7199 free_extent_map(hole_em);
7201 free_extent_map(em);
7202 return ERR_PTR(err);
7207 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7210 const u64 orig_start,
7211 const u64 block_start,
7212 const u64 block_len,
7213 const u64 orig_block_len,
7214 const u64 ram_bytes,
7217 struct extent_map *em = NULL;
7220 if (type != BTRFS_ORDERED_NOCOW) {
7221 em = create_io_em(inode, start, len, orig_start, block_start,
7222 block_len, orig_block_len, ram_bytes,
7223 BTRFS_COMPRESS_NONE, /* compress_type */
7228 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7232 free_extent_map(em);
7233 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7242 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7245 struct btrfs_root *root = inode->root;
7246 struct btrfs_fs_info *fs_info = root->fs_info;
7247 struct extent_map *em;
7248 struct btrfs_key ins;
7252 alloc_hint = get_extent_allocation_hint(inode, start, len);
7253 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7254 0, alloc_hint, &ins, 1, 1);
7256 return ERR_PTR(ret);
7258 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7259 ins.objectid, ins.offset, ins.offset,
7260 ins.offset, BTRFS_ORDERED_REGULAR);
7261 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7263 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7269 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7271 struct btrfs_block_group *block_group;
7272 bool readonly = false;
7274 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7275 if (!block_group || block_group->ro)
7278 btrfs_put_block_group(block_group);
7283 * Check if we can do nocow write into the range [@offset, @offset + @len)
7285 * @offset: File offset
7286 * @len: The length to write, will be updated to the nocow writeable
7288 * @orig_start: (optional) Return the original file offset of the file extent
7289 * @orig_len: (optional) Return the original on-disk length of the file extent
7290 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7291 * @strict: if true, omit optimizations that might force us into unnecessary
7292 * cow. e.g., don't trust generation number.
7295 * >0 and update @len if we can do nocow write
7296 * 0 if we can't do nocow write
7297 * <0 if error happened
7299 * NOTE: This only checks the file extents, caller is responsible to wait for
7300 * any ordered extents.
7302 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7303 u64 *orig_start, u64 *orig_block_len,
7304 u64 *ram_bytes, bool strict)
7306 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7307 struct btrfs_path *path;
7309 struct extent_buffer *leaf;
7310 struct btrfs_root *root = BTRFS_I(inode)->root;
7311 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7312 struct btrfs_file_extent_item *fi;
7313 struct btrfs_key key;
7320 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7322 path = btrfs_alloc_path();
7326 ret = btrfs_lookup_file_extent(NULL, root, path,
7327 btrfs_ino(BTRFS_I(inode)), offset, 0);
7331 slot = path->slots[0];
7334 /* can't find the item, must cow */
7341 leaf = path->nodes[0];
7342 btrfs_item_key_to_cpu(leaf, &key, slot);
7343 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7344 key.type != BTRFS_EXTENT_DATA_KEY) {
7345 /* not our file or wrong item type, must cow */
7349 if (key.offset > offset) {
7350 /* Wrong offset, must cow */
7354 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7355 found_type = btrfs_file_extent_type(leaf, fi);
7356 if (found_type != BTRFS_FILE_EXTENT_REG &&
7357 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7358 /* not a regular extent, must cow */
7362 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7365 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7366 if (extent_end <= offset)
7369 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7370 if (disk_bytenr == 0)
7373 if (btrfs_file_extent_compression(leaf, fi) ||
7374 btrfs_file_extent_encryption(leaf, fi) ||
7375 btrfs_file_extent_other_encoding(leaf, fi))
7379 * Do the same check as in btrfs_cross_ref_exist but without the
7380 * unnecessary search.
7383 (btrfs_file_extent_generation(leaf, fi) <=
7384 btrfs_root_last_snapshot(&root->root_item)))
7387 backref_offset = btrfs_file_extent_offset(leaf, fi);
7390 *orig_start = key.offset - backref_offset;
7391 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7392 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7395 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7398 num_bytes = min(offset + *len, extent_end) - offset;
7399 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7402 range_end = round_up(offset + num_bytes,
7403 root->fs_info->sectorsize) - 1;
7404 ret = test_range_bit(io_tree, offset, range_end,
7405 EXTENT_DELALLOC, 0, NULL);
7412 btrfs_release_path(path);
7415 * look for other files referencing this extent, if we
7416 * find any we must cow
7419 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7420 key.offset - backref_offset, disk_bytenr,
7428 * adjust disk_bytenr and num_bytes to cover just the bytes
7429 * in this extent we are about to write. If there
7430 * are any csums in that range we have to cow in order
7431 * to keep the csums correct
7433 disk_bytenr += backref_offset;
7434 disk_bytenr += offset - key.offset;
7435 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7438 * all of the above have passed, it is safe to overwrite this extent
7444 btrfs_free_path(path);
7448 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7449 struct extent_state **cached_state, bool writing)
7451 struct btrfs_ordered_extent *ordered;
7455 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7458 * We're concerned with the entire range that we're going to be
7459 * doing DIO to, so we need to make sure there's no ordered
7460 * extents in this range.
7462 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7463 lockend - lockstart + 1);
7466 * We need to make sure there are no buffered pages in this
7467 * range either, we could have raced between the invalidate in
7468 * generic_file_direct_write and locking the extent. The
7469 * invalidate needs to happen so that reads after a write do not
7473 (!writing || !filemap_range_has_page(inode->i_mapping,
7474 lockstart, lockend)))
7477 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7482 * If we are doing a DIO read and the ordered extent we
7483 * found is for a buffered write, we can not wait for it
7484 * to complete and retry, because if we do so we can
7485 * deadlock with concurrent buffered writes on page
7486 * locks. This happens only if our DIO read covers more
7487 * than one extent map, if at this point has already
7488 * created an ordered extent for a previous extent map
7489 * and locked its range in the inode's io tree, and a
7490 * concurrent write against that previous extent map's
7491 * range and this range started (we unlock the ranges
7492 * in the io tree only when the bios complete and
7493 * buffered writes always lock pages before attempting
7494 * to lock range in the io tree).
7497 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7498 btrfs_start_ordered_extent(ordered, 1);
7501 btrfs_put_ordered_extent(ordered);
7504 * We could trigger writeback for this range (and wait
7505 * for it to complete) and then invalidate the pages for
7506 * this range (through invalidate_inode_pages2_range()),
7507 * but that can lead us to a deadlock with a concurrent
7508 * call to readahead (a buffered read or a defrag call
7509 * triggered a readahead) on a page lock due to an
7510 * ordered dio extent we created before but did not have
7511 * yet a corresponding bio submitted (whence it can not
7512 * complete), which makes readahead wait for that
7513 * ordered extent to complete while holding a lock on
7528 /* The callers of this must take lock_extent() */
7529 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7530 u64 len, u64 orig_start, u64 block_start,
7531 u64 block_len, u64 orig_block_len,
7532 u64 ram_bytes, int compress_type,
7535 struct extent_map_tree *em_tree;
7536 struct extent_map *em;
7539 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7540 type == BTRFS_ORDERED_COMPRESSED ||
7541 type == BTRFS_ORDERED_NOCOW ||
7542 type == BTRFS_ORDERED_REGULAR);
7544 em_tree = &inode->extent_tree;
7545 em = alloc_extent_map();
7547 return ERR_PTR(-ENOMEM);
7550 em->orig_start = orig_start;
7552 em->block_len = block_len;
7553 em->block_start = block_start;
7554 em->orig_block_len = orig_block_len;
7555 em->ram_bytes = ram_bytes;
7556 em->generation = -1;
7557 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7558 if (type == BTRFS_ORDERED_PREALLOC) {
7559 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7560 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7561 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7562 em->compress_type = compress_type;
7566 btrfs_drop_extent_cache(inode, em->start,
7567 em->start + em->len - 1, 0);
7568 write_lock(&em_tree->lock);
7569 ret = add_extent_mapping(em_tree, em, 1);
7570 write_unlock(&em_tree->lock);
7572 * The caller has taken lock_extent(), who could race with us
7575 } while (ret == -EEXIST);
7578 free_extent_map(em);
7579 return ERR_PTR(ret);
7582 /* em got 2 refs now, callers needs to do free_extent_map once. */
7587 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7588 struct inode *inode,
7589 struct btrfs_dio_data *dio_data,
7592 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7593 struct extent_map *em = *map;
7597 * We don't allocate a new extent in the following cases
7599 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7601 * 2) The extent is marked as PREALLOC. We're good to go here and can
7602 * just use the extent.
7605 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7606 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7607 em->block_start != EXTENT_MAP_HOLE)) {
7609 u64 block_start, orig_start, orig_block_len, ram_bytes;
7611 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7612 type = BTRFS_ORDERED_PREALLOC;
7614 type = BTRFS_ORDERED_NOCOW;
7615 len = min(len, em->len - (start - em->start));
7616 block_start = em->block_start + (start - em->start);
7618 if (can_nocow_extent(inode, start, &len, &orig_start,
7619 &orig_block_len, &ram_bytes, false) == 1 &&
7620 btrfs_inc_nocow_writers(fs_info, block_start)) {
7621 struct extent_map *em2;
7623 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7624 orig_start, block_start,
7625 len, orig_block_len,
7627 btrfs_dec_nocow_writers(fs_info, block_start);
7628 if (type == BTRFS_ORDERED_PREALLOC) {
7629 free_extent_map(em);
7633 if (em2 && IS_ERR(em2)) {
7638 * For inode marked NODATACOW or extent marked PREALLOC,
7639 * use the existing or preallocated extent, so does not
7640 * need to adjust btrfs_space_info's bytes_may_use.
7642 btrfs_free_reserved_data_space_noquota(fs_info, len);
7647 /* this will cow the extent */
7648 free_extent_map(em);
7649 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7655 len = min(len, em->len - (start - em->start));
7659 * Need to update the i_size under the extent lock so buffered
7660 * readers will get the updated i_size when we unlock.
7662 if (start + len > i_size_read(inode))
7663 i_size_write(inode, start + len);
7665 dio_data->reserve -= len;
7670 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7671 loff_t length, unsigned int flags, struct iomap *iomap,
7672 struct iomap *srcmap)
7674 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7675 struct extent_map *em;
7676 struct extent_state *cached_state = NULL;
7677 struct btrfs_dio_data *dio_data = NULL;
7678 u64 lockstart, lockend;
7679 const bool write = !!(flags & IOMAP_WRITE);
7682 bool unlock_extents = false;
7685 len = min_t(u64, len, fs_info->sectorsize);
7688 lockend = start + len - 1;
7691 * The generic stuff only does filemap_write_and_wait_range, which
7692 * isn't enough if we've written compressed pages to this area, so we
7693 * need to flush the dirty pages again to make absolutely sure that any
7694 * outstanding dirty pages are on disk.
7696 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7697 &BTRFS_I(inode)->runtime_flags)) {
7698 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7699 start + length - 1);
7704 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7708 dio_data->length = length;
7710 dio_data->reserve = round_up(length, fs_info->sectorsize);
7711 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7712 &dio_data->data_reserved,
7713 start, dio_data->reserve);
7715 extent_changeset_free(dio_data->data_reserved);
7720 iomap->private = dio_data;
7724 * If this errors out it's because we couldn't invalidate pagecache for
7725 * this range and we need to fallback to buffered.
7727 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7732 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7739 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7740 * io. INLINE is special, and we could probably kludge it in here, but
7741 * it's still buffered so for safety lets just fall back to the generic
7744 * For COMPRESSED we _have_ to read the entire extent in so we can
7745 * decompress it, so there will be buffering required no matter what we
7746 * do, so go ahead and fallback to buffered.
7748 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7749 * to buffered IO. Don't blame me, this is the price we pay for using
7752 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7753 em->block_start == EXTENT_MAP_INLINE) {
7754 free_extent_map(em);
7759 len = min(len, em->len - (start - em->start));
7761 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7765 unlock_extents = true;
7766 /* Recalc len in case the new em is smaller than requested */
7767 len = min(len, em->len - (start - em->start));
7770 * We need to unlock only the end area that we aren't using.
7771 * The rest is going to be unlocked by the endio routine.
7773 lockstart = start + len;
7774 if (lockstart < lockend)
7775 unlock_extents = true;
7779 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7780 lockstart, lockend, &cached_state);
7782 free_extent_state(cached_state);
7785 * Translate extent map information to iomap.
7786 * We trim the extents (and move the addr) even though iomap code does
7787 * that, since we have locked only the parts we are performing I/O in.
7789 if ((em->block_start == EXTENT_MAP_HOLE) ||
7790 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7791 iomap->addr = IOMAP_NULL_ADDR;
7792 iomap->type = IOMAP_HOLE;
7794 iomap->addr = em->block_start + (start - em->start);
7795 iomap->type = IOMAP_MAPPED;
7797 iomap->offset = start;
7798 iomap->bdev = fs_info->fs_devices->latest_bdev;
7799 iomap->length = len;
7801 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7802 iomap->flags |= IOMAP_F_ZONE_APPEND;
7804 free_extent_map(em);
7809 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7813 btrfs_delalloc_release_space(BTRFS_I(inode),
7814 dio_data->data_reserved, start,
7815 dio_data->reserve, true);
7816 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7817 extent_changeset_free(dio_data->data_reserved);
7823 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7824 ssize_t written, unsigned int flags, struct iomap *iomap)
7827 struct btrfs_dio_data *dio_data = iomap->private;
7828 size_t submitted = dio_data->submitted;
7829 const bool write = !!(flags & IOMAP_WRITE);
7831 if (!write && (iomap->type == IOMAP_HOLE)) {
7832 /* If reading from a hole, unlock and return */
7833 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7837 if (submitted < length) {
7839 length -= submitted;
7841 __endio_write_update_ordered(BTRFS_I(inode), pos,
7844 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7850 if (dio_data->reserve)
7851 btrfs_delalloc_release_space(BTRFS_I(inode),
7852 dio_data->data_reserved, pos,
7853 dio_data->reserve, true);
7854 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7855 extent_changeset_free(dio_data->data_reserved);
7859 iomap->private = NULL;
7864 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7867 * This implies a barrier so that stores to dio_bio->bi_status before
7868 * this and loads of dio_bio->bi_status after this are fully ordered.
7870 if (!refcount_dec_and_test(&dip->refs))
7873 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7874 __endio_write_update_ordered(BTRFS_I(dip->inode),
7875 dip->logical_offset,
7877 !dip->dio_bio->bi_status);
7879 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7880 dip->logical_offset,
7881 dip->logical_offset + dip->bytes - 1);
7884 bio_endio(dip->dio_bio);
7888 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7890 unsigned long bio_flags)
7892 struct btrfs_dio_private *dip = bio->bi_private;
7893 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7896 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7898 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7902 refcount_inc(&dip->refs);
7903 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7905 refcount_dec(&dip->refs);
7909 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7910 struct btrfs_io_bio *io_bio,
7911 const bool uptodate)
7913 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7914 const u32 sectorsize = fs_info->sectorsize;
7915 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7916 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7917 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7918 struct bio_vec bvec;
7919 struct bvec_iter iter;
7920 u64 start = io_bio->logical;
7922 blk_status_t err = BLK_STS_OK;
7924 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7925 unsigned int i, nr_sectors, pgoff;
7927 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7928 pgoff = bvec.bv_offset;
7929 for (i = 0; i < nr_sectors; i++) {
7930 ASSERT(pgoff < PAGE_SIZE);
7932 (!csum || !check_data_csum(inode, io_bio,
7933 bio_offset, bvec.bv_page,
7935 clean_io_failure(fs_info, failure_tree, io_tree,
7936 start, bvec.bv_page,
7937 btrfs_ino(BTRFS_I(inode)),
7940 blk_status_t status;
7942 ASSERT((start - io_bio->logical) < UINT_MAX);
7943 status = btrfs_submit_read_repair(inode,
7945 start - io_bio->logical,
7946 bvec.bv_page, pgoff,
7948 start + sectorsize - 1,
7950 submit_dio_repair_bio);
7954 start += sectorsize;
7955 ASSERT(bio_offset + sectorsize > bio_offset);
7956 bio_offset += sectorsize;
7957 pgoff += sectorsize;
7963 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7964 const u64 offset, const u64 bytes,
7965 const bool uptodate)
7967 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7968 struct btrfs_ordered_extent *ordered = NULL;
7969 struct btrfs_workqueue *wq;
7970 u64 ordered_offset = offset;
7971 u64 ordered_bytes = bytes;
7974 if (btrfs_is_free_space_inode(inode))
7975 wq = fs_info->endio_freespace_worker;
7977 wq = fs_info->endio_write_workers;
7979 while (ordered_offset < offset + bytes) {
7980 last_offset = ordered_offset;
7981 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7985 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7987 btrfs_queue_work(wq, &ordered->work);
7990 /* No ordered extent found in the range, exit */
7991 if (ordered_offset == last_offset)
7994 * Our bio might span multiple ordered extents. In this case
7995 * we keep going until we have accounted the whole dio.
7997 if (ordered_offset < offset + bytes) {
7998 ordered_bytes = offset + bytes - ordered_offset;
8004 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8006 u64 dio_file_offset)
8008 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
8011 static void btrfs_end_dio_bio(struct bio *bio)
8013 struct btrfs_dio_private *dip = bio->bi_private;
8014 blk_status_t err = bio->bi_status;
8017 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8018 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8019 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8020 bio->bi_opf, bio->bi_iter.bi_sector,
8021 bio->bi_iter.bi_size, err);
8023 if (bio_op(bio) == REQ_OP_READ) {
8024 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
8029 dip->dio_bio->bi_status = err;
8031 btrfs_record_physical_zoned(dip->inode, dip->logical_offset, bio);
8034 btrfs_dio_private_put(dip);
8037 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8038 struct inode *inode, u64 file_offset, int async_submit)
8040 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8041 struct btrfs_dio_private *dip = bio->bi_private;
8042 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
8045 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8047 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8050 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8055 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8058 if (write && async_submit) {
8059 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
8060 btrfs_submit_bio_start_direct_io);
8064 * If we aren't doing async submit, calculate the csum of the
8067 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
8073 csum_offset = file_offset - dip->logical_offset;
8074 csum_offset >>= fs_info->sectorsize_bits;
8075 csum_offset *= fs_info->csum_size;
8076 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
8079 ret = btrfs_map_bio(fs_info, bio, 0);
8085 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
8086 * or ordered extents whether or not we submit any bios.
8088 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
8089 struct inode *inode,
8092 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8093 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8095 struct btrfs_dio_private *dip;
8097 dip_size = sizeof(*dip);
8098 if (!write && csum) {
8099 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8102 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
8103 dip_size += fs_info->csum_size * nblocks;
8106 dip = kzalloc(dip_size, GFP_NOFS);
8111 dip->logical_offset = file_offset;
8112 dip->bytes = dio_bio->bi_iter.bi_size;
8113 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
8114 dip->dio_bio = dio_bio;
8115 refcount_set(&dip->refs, 1);
8119 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
8120 struct bio *dio_bio, loff_t file_offset)
8122 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8123 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8124 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8125 BTRFS_BLOCK_GROUP_RAID56_MASK);
8126 struct btrfs_dio_private *dip;
8129 int async_submit = 0;
8131 int clone_offset = 0;
8135 blk_status_t status;
8136 struct btrfs_io_geometry geom;
8137 struct btrfs_dio_data *dio_data = iomap->private;
8138 struct extent_map *em = NULL;
8140 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8143 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8144 file_offset + dio_bio->bi_iter.bi_size - 1);
8146 dio_bio->bi_status = BLK_STS_RESOURCE;
8148 return BLK_QC_T_NONE;
8153 * Load the csums up front to reduce csum tree searches and
8154 * contention when submitting bios.
8156 * If we have csums disabled this will do nothing.
8158 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8159 if (status != BLK_STS_OK)
8163 start_sector = dio_bio->bi_iter.bi_sector;
8164 submit_len = dio_bio->bi_iter.bi_size;
8167 logical = start_sector << 9;
8168 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8170 status = errno_to_blk_status(PTR_ERR(em));
8174 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8175 logical, submit_len, &geom);
8177 status = errno_to_blk_status(ret);
8180 ASSERT(geom.len <= INT_MAX);
8182 clone_len = min_t(int, submit_len, geom.len);
8185 * This will never fail as it's passing GPF_NOFS and
8186 * the allocation is backed by btrfs_bioset.
8188 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8189 bio->bi_private = dip;
8190 bio->bi_end_io = btrfs_end_dio_bio;
8191 btrfs_io_bio(bio)->logical = file_offset;
8193 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8194 status = extract_ordered_extent(BTRFS_I(inode), bio,
8202 ASSERT(submit_len >= clone_len);
8203 submit_len -= clone_len;
8206 * Increase the count before we submit the bio so we know
8207 * the end IO handler won't happen before we increase the
8208 * count. Otherwise, the dip might get freed before we're
8209 * done setting it up.
8211 * We transfer the initial reference to the last bio, so we
8212 * don't need to increment the reference count for the last one.
8214 if (submit_len > 0) {
8215 refcount_inc(&dip->refs);
8217 * If we are submitting more than one bio, submit them
8218 * all asynchronously. The exception is RAID 5 or 6, as
8219 * asynchronous checksums make it difficult to collect
8220 * full stripe writes.
8226 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8231 refcount_dec(&dip->refs);
8235 dio_data->submitted += clone_len;
8236 clone_offset += clone_len;
8237 start_sector += clone_len >> 9;
8238 file_offset += clone_len;
8240 free_extent_map(em);
8241 } while (submit_len > 0);
8242 return BLK_QC_T_NONE;
8245 free_extent_map(em);
8247 dip->dio_bio->bi_status = status;
8248 btrfs_dio_private_put(dip);
8250 return BLK_QC_T_NONE;
8253 const struct iomap_ops btrfs_dio_iomap_ops = {
8254 .iomap_begin = btrfs_dio_iomap_begin,
8255 .iomap_end = btrfs_dio_iomap_end,
8258 const struct iomap_dio_ops btrfs_dio_ops = {
8259 .submit_io = btrfs_submit_direct,
8262 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8267 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8271 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8274 int btrfs_readpage(struct file *file, struct page *page)
8276 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8277 u64 start = page_offset(page);
8278 u64 end = start + PAGE_SIZE - 1;
8279 unsigned long bio_flags = 0;
8280 struct bio *bio = NULL;
8283 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8285 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8287 ret = submit_one_bio(bio, 0, bio_flags);
8291 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8293 struct inode *inode = page->mapping->host;
8296 if (current->flags & PF_MEMALLOC) {
8297 redirty_page_for_writepage(wbc, page);
8303 * If we are under memory pressure we will call this directly from the
8304 * VM, we need to make sure we have the inode referenced for the ordered
8305 * extent. If not just return like we didn't do anything.
8307 if (!igrab(inode)) {
8308 redirty_page_for_writepage(wbc, page);
8309 return AOP_WRITEPAGE_ACTIVATE;
8311 ret = extent_write_full_page(page, wbc);
8312 btrfs_add_delayed_iput(inode);
8316 static int btrfs_writepages(struct address_space *mapping,
8317 struct writeback_control *wbc)
8319 return extent_writepages(mapping, wbc);
8322 static void btrfs_readahead(struct readahead_control *rac)
8324 extent_readahead(rac);
8327 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8329 int ret = try_release_extent_mapping(page, gfp_flags);
8331 clear_page_extent_mapped(page);
8335 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8337 if (PageWriteback(page) || PageDirty(page))
8339 return __btrfs_releasepage(page, gfp_flags);
8342 #ifdef CONFIG_MIGRATION
8343 static int btrfs_migratepage(struct address_space *mapping,
8344 struct page *newpage, struct page *page,
8345 enum migrate_mode mode)
8349 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8350 if (ret != MIGRATEPAGE_SUCCESS)
8353 if (page_has_private(page))
8354 attach_page_private(newpage, detach_page_private(page));
8356 if (PagePrivate2(page)) {
8357 ClearPagePrivate2(page);
8358 SetPagePrivate2(newpage);
8361 if (mode != MIGRATE_SYNC_NO_COPY)
8362 migrate_page_copy(newpage, page);
8364 migrate_page_states(newpage, page);
8365 return MIGRATEPAGE_SUCCESS;
8369 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8370 unsigned int length)
8372 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8373 struct extent_io_tree *tree = &inode->io_tree;
8374 struct btrfs_ordered_extent *ordered;
8375 struct extent_state *cached_state = NULL;
8376 u64 page_start = page_offset(page);
8377 u64 page_end = page_start + PAGE_SIZE - 1;
8380 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8381 bool found_ordered = false;
8382 bool completed_ordered = false;
8385 * we have the page locked, so new writeback can't start,
8386 * and the dirty bit won't be cleared while we are here.
8388 * Wait for IO on this page so that we can safely clear
8389 * the PagePrivate2 bit and do ordered accounting
8391 wait_on_page_writeback(page);
8394 btrfs_releasepage(page, GFP_NOFS);
8398 if (!inode_evicting)
8399 lock_extent_bits(tree, page_start, page_end, &cached_state);
8403 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8405 found_ordered = true;
8407 ordered->file_offset + ordered->num_bytes - 1);
8409 * IO on this page will never be started, so we need to account
8410 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8411 * here, must leave that up for the ordered extent completion.
8413 if (!inode_evicting)
8414 clear_extent_bit(tree, start, end,
8416 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8417 EXTENT_DEFRAG, 1, 0, &cached_state);
8419 * whoever cleared the private bit is responsible
8420 * for the finish_ordered_io
8422 if (TestClearPagePrivate2(page)) {
8423 spin_lock_irq(&inode->ordered_tree.lock);
8424 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8425 ordered->truncated_len = min(ordered->truncated_len,
8426 start - ordered->file_offset);
8427 spin_unlock_irq(&inode->ordered_tree.lock);
8429 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8431 end - start + 1, 1)) {
8432 btrfs_finish_ordered_io(ordered);
8433 completed_ordered = true;
8436 btrfs_put_ordered_extent(ordered);
8437 if (!inode_evicting) {
8438 cached_state = NULL;
8439 lock_extent_bits(tree, start, end,
8444 if (start < page_end)
8449 * Qgroup reserved space handler
8450 * Page here will be either
8451 * 1) Already written to disk or ordered extent already submitted
8452 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8453 * Qgroup will be handled by its qgroup_record then.
8454 * btrfs_qgroup_free_data() call will do nothing here.
8456 * 2) Not written to disk yet
8457 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8458 * bit of its io_tree, and free the qgroup reserved data space.
8459 * Since the IO will never happen for this page.
8461 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8462 if (!inode_evicting) {
8466 * If there's an ordered extent for this range and we have not
8467 * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set
8468 * in the range for the ordered extent completion. We must also
8469 * not delete the range, otherwise we would lose that bit (and
8470 * any other bits set in the range). Make sure EXTENT_UPTODATE
8471 * is cleared if we don't delete, otherwise it can lead to
8472 * corruptions if the i_size is extented later.
8474 if (found_ordered && !completed_ordered)
8476 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8477 EXTENT_DELALLOC | EXTENT_UPTODATE |
8478 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8479 delete, &cached_state);
8481 __btrfs_releasepage(page, GFP_NOFS);
8484 ClearPageChecked(page);
8485 clear_page_extent_mapped(page);
8489 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8490 * called from a page fault handler when a page is first dirtied. Hence we must
8491 * be careful to check for EOF conditions here. We set the page up correctly
8492 * for a written page which means we get ENOSPC checking when writing into
8493 * holes and correct delalloc and unwritten extent mapping on filesystems that
8494 * support these features.
8496 * We are not allowed to take the i_mutex here so we have to play games to
8497 * protect against truncate races as the page could now be beyond EOF. Because
8498 * truncate_setsize() writes the inode size before removing pages, once we have
8499 * the page lock we can determine safely if the page is beyond EOF. If it is not
8500 * beyond EOF, then the page is guaranteed safe against truncation until we
8503 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8505 struct page *page = vmf->page;
8506 struct inode *inode = file_inode(vmf->vma->vm_file);
8507 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8508 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8509 struct btrfs_ordered_extent *ordered;
8510 struct extent_state *cached_state = NULL;
8511 struct extent_changeset *data_reserved = NULL;
8512 unsigned long zero_start;
8522 reserved_space = PAGE_SIZE;
8524 sb_start_pagefault(inode->i_sb);
8525 page_start = page_offset(page);
8526 page_end = page_start + PAGE_SIZE - 1;
8530 * Reserving delalloc space after obtaining the page lock can lead to
8531 * deadlock. For example, if a dirty page is locked by this function
8532 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8533 * dirty page write out, then the btrfs_writepage() function could
8534 * end up waiting indefinitely to get a lock on the page currently
8535 * being processed by btrfs_page_mkwrite() function.
8537 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8538 page_start, reserved_space);
8540 ret2 = file_update_time(vmf->vma->vm_file);
8544 ret = vmf_error(ret2);
8550 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8552 down_read(&BTRFS_I(inode)->i_mmap_lock);
8554 size = i_size_read(inode);
8556 if ((page->mapping != inode->i_mapping) ||
8557 (page_start >= size)) {
8558 /* page got truncated out from underneath us */
8561 wait_on_page_writeback(page);
8563 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8564 ret2 = set_page_extent_mapped(page);
8566 ret = vmf_error(ret2);
8567 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8572 * we can't set the delalloc bits if there are pending ordered
8573 * extents. Drop our locks and wait for them to finish
8575 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8578 unlock_extent_cached(io_tree, page_start, page_end,
8581 up_read(&BTRFS_I(inode)->i_mmap_lock);
8582 btrfs_start_ordered_extent(ordered, 1);
8583 btrfs_put_ordered_extent(ordered);
8587 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8588 reserved_space = round_up(size - page_start,
8589 fs_info->sectorsize);
8590 if (reserved_space < PAGE_SIZE) {
8591 end = page_start + reserved_space - 1;
8592 btrfs_delalloc_release_space(BTRFS_I(inode),
8593 data_reserved, page_start,
8594 PAGE_SIZE - reserved_space, true);
8599 * page_mkwrite gets called when the page is firstly dirtied after it's
8600 * faulted in, but write(2) could also dirty a page and set delalloc
8601 * bits, thus in this case for space account reason, we still need to
8602 * clear any delalloc bits within this page range since we have to
8603 * reserve data&meta space before lock_page() (see above comments).
8605 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8606 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8607 EXTENT_DEFRAG, 0, 0, &cached_state);
8609 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8612 unlock_extent_cached(io_tree, page_start, page_end,
8614 ret = VM_FAULT_SIGBUS;
8618 /* page is wholly or partially inside EOF */
8619 if (page_start + PAGE_SIZE > size)
8620 zero_start = offset_in_page(size);
8622 zero_start = PAGE_SIZE;
8624 if (zero_start != PAGE_SIZE) {
8625 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8626 flush_dcache_page(page);
8628 ClearPageChecked(page);
8629 set_page_dirty(page);
8630 SetPageUptodate(page);
8632 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8634 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8635 up_read(&BTRFS_I(inode)->i_mmap_lock);
8637 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8638 sb_end_pagefault(inode->i_sb);
8639 extent_changeset_free(data_reserved);
8640 return VM_FAULT_LOCKED;
8644 up_read(&BTRFS_I(inode)->i_mmap_lock);
8646 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8647 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8648 reserved_space, (ret != 0));
8650 sb_end_pagefault(inode->i_sb);
8651 extent_changeset_free(data_reserved);
8655 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8657 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8658 struct btrfs_root *root = BTRFS_I(inode)->root;
8659 struct btrfs_block_rsv *rsv;
8661 struct btrfs_trans_handle *trans;
8662 u64 mask = fs_info->sectorsize - 1;
8663 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8665 if (!skip_writeback) {
8666 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8673 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8674 * things going on here:
8676 * 1) We need to reserve space to update our inode.
8678 * 2) We need to have something to cache all the space that is going to
8679 * be free'd up by the truncate operation, but also have some slack
8680 * space reserved in case it uses space during the truncate (thank you
8681 * very much snapshotting).
8683 * And we need these to be separate. The fact is we can use a lot of
8684 * space doing the truncate, and we have no earthly idea how much space
8685 * we will use, so we need the truncate reservation to be separate so it
8686 * doesn't end up using space reserved for updating the inode. We also
8687 * need to be able to stop the transaction and start a new one, which
8688 * means we need to be able to update the inode several times, and we
8689 * have no idea of knowing how many times that will be, so we can't just
8690 * reserve 1 item for the entirety of the operation, so that has to be
8691 * done separately as well.
8693 * So that leaves us with
8695 * 1) rsv - for the truncate reservation, which we will steal from the
8696 * transaction reservation.
8697 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8698 * updating the inode.
8700 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8703 rsv->size = min_size;
8707 * 1 for the truncate slack space
8708 * 1 for updating the inode.
8710 trans = btrfs_start_transaction(root, 2);
8711 if (IS_ERR(trans)) {
8712 ret = PTR_ERR(trans);
8716 /* Migrate the slack space for the truncate to our reserve */
8717 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8722 * So if we truncate and then write and fsync we normally would just
8723 * write the extents that changed, which is a problem if we need to
8724 * first truncate that entire inode. So set this flag so we write out
8725 * all of the extents in the inode to the sync log so we're completely
8728 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8729 trans->block_rsv = rsv;
8732 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8734 BTRFS_EXTENT_DATA_KEY);
8735 trans->block_rsv = &fs_info->trans_block_rsv;
8736 if (ret != -ENOSPC && ret != -EAGAIN)
8739 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8743 btrfs_end_transaction(trans);
8744 btrfs_btree_balance_dirty(fs_info);
8746 trans = btrfs_start_transaction(root, 2);
8747 if (IS_ERR(trans)) {
8748 ret = PTR_ERR(trans);
8753 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8754 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8755 rsv, min_size, false);
8756 BUG_ON(ret); /* shouldn't happen */
8757 trans->block_rsv = rsv;
8761 * We can't call btrfs_truncate_block inside a trans handle as we could
8762 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8763 * we've truncated everything except the last little bit, and can do
8764 * btrfs_truncate_block and then update the disk_i_size.
8766 if (ret == NEED_TRUNCATE_BLOCK) {
8767 btrfs_end_transaction(trans);
8768 btrfs_btree_balance_dirty(fs_info);
8770 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8773 trans = btrfs_start_transaction(root, 1);
8774 if (IS_ERR(trans)) {
8775 ret = PTR_ERR(trans);
8778 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8784 trans->block_rsv = &fs_info->trans_block_rsv;
8785 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8789 ret2 = btrfs_end_transaction(trans);
8792 btrfs_btree_balance_dirty(fs_info);
8795 btrfs_free_block_rsv(fs_info, rsv);
8801 * create a new subvolume directory/inode (helper for the ioctl).
8803 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8804 struct btrfs_root *new_root,
8805 struct btrfs_root *parent_root)
8807 struct inode *inode;
8812 err = btrfs_get_free_objectid(new_root, &ino);
8816 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8817 S_IFDIR | (~current_umask() & S_IRWXUGO),
8820 return PTR_ERR(inode);
8821 inode->i_op = &btrfs_dir_inode_operations;
8822 inode->i_fop = &btrfs_dir_file_operations;
8824 set_nlink(inode, 1);
8825 btrfs_i_size_write(BTRFS_I(inode), 0);
8826 unlock_new_inode(inode);
8828 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8830 btrfs_err(new_root->fs_info,
8831 "error inheriting subvolume %llu properties: %d",
8832 new_root->root_key.objectid, err);
8834 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8840 struct inode *btrfs_alloc_inode(struct super_block *sb)
8842 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8843 struct btrfs_inode *ei;
8844 struct inode *inode;
8846 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8853 ei->last_sub_trans = 0;
8854 ei->logged_trans = 0;
8855 ei->delalloc_bytes = 0;
8856 ei->new_delalloc_bytes = 0;
8857 ei->defrag_bytes = 0;
8858 ei->disk_i_size = 0;
8861 ei->index_cnt = (u64)-1;
8863 ei->last_unlink_trans = 0;
8864 ei->last_reflink_trans = 0;
8865 ei->last_log_commit = 0;
8867 spin_lock_init(&ei->lock);
8868 ei->outstanding_extents = 0;
8869 if (sb->s_magic != BTRFS_TEST_MAGIC)
8870 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8871 BTRFS_BLOCK_RSV_DELALLOC);
8872 ei->runtime_flags = 0;
8873 ei->prop_compress = BTRFS_COMPRESS_NONE;
8874 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8876 ei->delayed_node = NULL;
8878 ei->i_otime.tv_sec = 0;
8879 ei->i_otime.tv_nsec = 0;
8881 inode = &ei->vfs_inode;
8882 extent_map_tree_init(&ei->extent_tree);
8883 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8884 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8885 IO_TREE_INODE_IO_FAILURE, inode);
8886 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8887 IO_TREE_INODE_FILE_EXTENT, inode);
8888 ei->io_tree.track_uptodate = true;
8889 ei->io_failure_tree.track_uptodate = true;
8890 atomic_set(&ei->sync_writers, 0);
8891 mutex_init(&ei->log_mutex);
8892 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8893 INIT_LIST_HEAD(&ei->delalloc_inodes);
8894 INIT_LIST_HEAD(&ei->delayed_iput);
8895 RB_CLEAR_NODE(&ei->rb_node);
8896 init_rwsem(&ei->i_mmap_lock);
8901 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8902 void btrfs_test_destroy_inode(struct inode *inode)
8904 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8905 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8909 void btrfs_free_inode(struct inode *inode)
8911 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8914 void btrfs_destroy_inode(struct inode *vfs_inode)
8916 struct btrfs_ordered_extent *ordered;
8917 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8918 struct btrfs_root *root = inode->root;
8920 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8921 WARN_ON(vfs_inode->i_data.nrpages);
8922 WARN_ON(inode->block_rsv.reserved);
8923 WARN_ON(inode->block_rsv.size);
8924 WARN_ON(inode->outstanding_extents);
8925 WARN_ON(inode->delalloc_bytes);
8926 WARN_ON(inode->new_delalloc_bytes);
8927 WARN_ON(inode->csum_bytes);
8928 WARN_ON(inode->defrag_bytes);
8931 * This can happen where we create an inode, but somebody else also
8932 * created the same inode and we need to destroy the one we already
8939 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8943 btrfs_err(root->fs_info,
8944 "found ordered extent %llu %llu on inode cleanup",
8945 ordered->file_offset, ordered->num_bytes);
8946 btrfs_remove_ordered_extent(inode, ordered);
8947 btrfs_put_ordered_extent(ordered);
8948 btrfs_put_ordered_extent(ordered);
8951 btrfs_qgroup_check_reserved_leak(inode);
8952 inode_tree_del(inode);
8953 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8954 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8955 btrfs_put_root(inode->root);
8958 int btrfs_drop_inode(struct inode *inode)
8960 struct btrfs_root *root = BTRFS_I(inode)->root;
8965 /* the snap/subvol tree is on deleting */
8966 if (btrfs_root_refs(&root->root_item) == 0)
8969 return generic_drop_inode(inode);
8972 static void init_once(void *foo)
8974 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8976 inode_init_once(&ei->vfs_inode);
8979 void __cold btrfs_destroy_cachep(void)
8982 * Make sure all delayed rcu free inodes are flushed before we
8986 kmem_cache_destroy(btrfs_inode_cachep);
8987 kmem_cache_destroy(btrfs_trans_handle_cachep);
8988 kmem_cache_destroy(btrfs_path_cachep);
8989 kmem_cache_destroy(btrfs_free_space_cachep);
8990 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8993 int __init btrfs_init_cachep(void)
8995 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8996 sizeof(struct btrfs_inode), 0,
8997 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8999 if (!btrfs_inode_cachep)
9002 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9003 sizeof(struct btrfs_trans_handle), 0,
9004 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9005 if (!btrfs_trans_handle_cachep)
9008 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9009 sizeof(struct btrfs_path), 0,
9010 SLAB_MEM_SPREAD, NULL);
9011 if (!btrfs_path_cachep)
9014 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9015 sizeof(struct btrfs_free_space), 0,
9016 SLAB_MEM_SPREAD, NULL);
9017 if (!btrfs_free_space_cachep)
9020 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9021 PAGE_SIZE, PAGE_SIZE,
9022 SLAB_MEM_SPREAD, NULL);
9023 if (!btrfs_free_space_bitmap_cachep)
9028 btrfs_destroy_cachep();
9032 static int btrfs_getattr(struct user_namespace *mnt_userns,
9033 const struct path *path, struct kstat *stat,
9034 u32 request_mask, unsigned int flags)
9038 struct inode *inode = d_inode(path->dentry);
9039 u32 blocksize = inode->i_sb->s_blocksize;
9040 u32 bi_flags = BTRFS_I(inode)->flags;
9042 stat->result_mask |= STATX_BTIME;
9043 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9044 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9045 if (bi_flags & BTRFS_INODE_APPEND)
9046 stat->attributes |= STATX_ATTR_APPEND;
9047 if (bi_flags & BTRFS_INODE_COMPRESS)
9048 stat->attributes |= STATX_ATTR_COMPRESSED;
9049 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9050 stat->attributes |= STATX_ATTR_IMMUTABLE;
9051 if (bi_flags & BTRFS_INODE_NODUMP)
9052 stat->attributes |= STATX_ATTR_NODUMP;
9054 stat->attributes_mask |= (STATX_ATTR_APPEND |
9055 STATX_ATTR_COMPRESSED |
9056 STATX_ATTR_IMMUTABLE |
9059 generic_fillattr(&init_user_ns, inode, stat);
9060 stat->dev = BTRFS_I(inode)->root->anon_dev;
9062 spin_lock(&BTRFS_I(inode)->lock);
9063 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9064 inode_bytes = inode_get_bytes(inode);
9065 spin_unlock(&BTRFS_I(inode)->lock);
9066 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9067 ALIGN(delalloc_bytes, blocksize)) >> 9;
9071 static int btrfs_rename_exchange(struct inode *old_dir,
9072 struct dentry *old_dentry,
9073 struct inode *new_dir,
9074 struct dentry *new_dentry)
9076 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9077 struct btrfs_trans_handle *trans;
9078 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9079 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9080 struct inode *new_inode = new_dentry->d_inode;
9081 struct inode *old_inode = old_dentry->d_inode;
9082 struct timespec64 ctime = current_time(old_inode);
9083 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9084 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9089 bool root_log_pinned = false;
9090 bool dest_log_pinned = false;
9091 bool need_abort = false;
9093 /* we only allow rename subvolume link between subvolumes */
9094 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9097 /* close the race window with snapshot create/destroy ioctl */
9098 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9099 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9100 down_read(&fs_info->subvol_sem);
9103 * We want to reserve the absolute worst case amount of items. So if
9104 * both inodes are subvols and we need to unlink them then that would
9105 * require 4 item modifications, but if they are both normal inodes it
9106 * would require 5 item modifications, so we'll assume their normal
9107 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9108 * should cover the worst case number of items we'll modify.
9110 trans = btrfs_start_transaction(root, 12);
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 btrfs_pin_log_trans(root);
9142 root_log_pinned = true;
9143 ret = btrfs_insert_inode_ref(trans, dest,
9144 new_dentry->d_name.name,
9145 new_dentry->d_name.len,
9147 btrfs_ino(BTRFS_I(new_dir)),
9154 /* And now for the dest. */
9155 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9156 /* force full log commit if subvolume involved. */
9157 btrfs_set_log_full_commit(trans);
9159 btrfs_pin_log_trans(dest);
9160 dest_log_pinned = true;
9161 ret = btrfs_insert_inode_ref(trans, root,
9162 old_dentry->d_name.name,
9163 old_dentry->d_name.len,
9165 btrfs_ino(BTRFS_I(old_dir)),
9169 btrfs_abort_transaction(trans, ret);
9174 /* Update inode version and ctime/mtime. */
9175 inode_inc_iversion(old_dir);
9176 inode_inc_iversion(new_dir);
9177 inode_inc_iversion(old_inode);
9178 inode_inc_iversion(new_inode);
9179 old_dir->i_ctime = old_dir->i_mtime = ctime;
9180 new_dir->i_ctime = new_dir->i_mtime = ctime;
9181 old_inode->i_ctime = ctime;
9182 new_inode->i_ctime = ctime;
9184 if (old_dentry->d_parent != new_dentry->d_parent) {
9185 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9186 BTRFS_I(old_inode), 1);
9187 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9188 BTRFS_I(new_inode), 1);
9191 /* src is a subvolume */
9192 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9193 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9194 } else { /* src is an inode */
9195 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9196 BTRFS_I(old_dentry->d_inode),
9197 old_dentry->d_name.name,
9198 old_dentry->d_name.len);
9200 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9203 btrfs_abort_transaction(trans, ret);
9207 /* dest is a subvolume */
9208 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9209 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9210 } else { /* dest is an inode */
9211 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9212 BTRFS_I(new_dentry->d_inode),
9213 new_dentry->d_name.name,
9214 new_dentry->d_name.len);
9216 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9219 btrfs_abort_transaction(trans, ret);
9223 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9224 new_dentry->d_name.name,
9225 new_dentry->d_name.len, 0, old_idx);
9227 btrfs_abort_transaction(trans, ret);
9231 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9232 old_dentry->d_name.name,
9233 old_dentry->d_name.len, 0, new_idx);
9235 btrfs_abort_transaction(trans, ret);
9239 if (old_inode->i_nlink == 1)
9240 BTRFS_I(old_inode)->dir_index = old_idx;
9241 if (new_inode->i_nlink == 1)
9242 BTRFS_I(new_inode)->dir_index = new_idx;
9244 if (root_log_pinned) {
9245 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9246 new_dentry->d_parent);
9247 btrfs_end_log_trans(root);
9248 root_log_pinned = false;
9250 if (dest_log_pinned) {
9251 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9252 old_dentry->d_parent);
9253 btrfs_end_log_trans(dest);
9254 dest_log_pinned = false;
9258 * If we have pinned a log and an error happened, we unpin tasks
9259 * trying to sync the log and force them to fallback to a transaction
9260 * commit if the log currently contains any of the inodes involved in
9261 * this rename operation (to ensure we do not persist a log with an
9262 * inconsistent state for any of these inodes or leading to any
9263 * inconsistencies when replayed). If the transaction was aborted, the
9264 * abortion reason is propagated to userspace when attempting to commit
9265 * the transaction. If the log does not contain any of these inodes, we
9266 * allow the tasks to sync it.
9268 if (ret && (root_log_pinned || dest_log_pinned)) {
9269 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9270 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9271 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9273 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9274 btrfs_set_log_full_commit(trans);
9276 if (root_log_pinned) {
9277 btrfs_end_log_trans(root);
9278 root_log_pinned = false;
9280 if (dest_log_pinned) {
9281 btrfs_end_log_trans(dest);
9282 dest_log_pinned = false;
9285 ret2 = btrfs_end_transaction(trans);
9286 ret = ret ? ret : ret2;
9288 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9289 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9290 up_read(&fs_info->subvol_sem);
9295 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9296 struct btrfs_root *root,
9298 struct dentry *dentry)
9301 struct inode *inode;
9305 ret = btrfs_get_free_objectid(root, &objectid);
9309 inode = btrfs_new_inode(trans, root, dir,
9310 dentry->d_name.name,
9312 btrfs_ino(BTRFS_I(dir)),
9314 S_IFCHR | WHITEOUT_MODE,
9317 if (IS_ERR(inode)) {
9318 ret = PTR_ERR(inode);
9322 inode->i_op = &btrfs_special_inode_operations;
9323 init_special_inode(inode, inode->i_mode,
9326 ret = btrfs_init_inode_security(trans, inode, dir,
9331 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9332 BTRFS_I(inode), 0, index);
9336 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9338 unlock_new_inode(inode);
9340 inode_dec_link_count(inode);
9346 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9347 struct inode *new_dir, struct dentry *new_dentry,
9350 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9351 struct btrfs_trans_handle *trans;
9352 unsigned int trans_num_items;
9353 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9354 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9355 struct inode *new_inode = d_inode(new_dentry);
9356 struct inode *old_inode = d_inode(old_dentry);
9360 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9361 bool log_pinned = false;
9363 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9366 /* we only allow rename subvolume link between subvolumes */
9367 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9370 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9371 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9374 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9375 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9379 /* check for collisions, even if the name isn't there */
9380 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9381 new_dentry->d_name.name,
9382 new_dentry->d_name.len);
9385 if (ret == -EEXIST) {
9387 * eexist without a new_inode */
9388 if (WARN_ON(!new_inode)) {
9392 /* maybe -EOVERFLOW */
9399 * we're using rename to replace one file with another. Start IO on it
9400 * now so we don't add too much work to the end of the transaction
9402 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9403 filemap_flush(old_inode->i_mapping);
9405 /* close the racy window with snapshot create/destroy ioctl */
9406 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9407 down_read(&fs_info->subvol_sem);
9409 * We want to reserve the absolute worst case amount of items. So if
9410 * both inodes are subvols and we need to unlink them then that would
9411 * require 4 item modifications, but if they are both normal inodes it
9412 * would require 5 item modifications, so we'll assume they are normal
9413 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9414 * should cover the worst case number of items we'll modify.
9415 * If our rename has the whiteout flag, we need more 5 units for the
9416 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9417 * when selinux is enabled).
9419 trans_num_items = 11;
9420 if (flags & RENAME_WHITEOUT)
9421 trans_num_items += 5;
9422 trans = btrfs_start_transaction(root, trans_num_items);
9423 if (IS_ERR(trans)) {
9424 ret = PTR_ERR(trans);
9429 ret = btrfs_record_root_in_trans(trans, dest);
9434 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9438 BTRFS_I(old_inode)->dir_index = 0ULL;
9439 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9440 /* force full log commit if subvolume involved. */
9441 btrfs_set_log_full_commit(trans);
9443 btrfs_pin_log_trans(root);
9445 ret = btrfs_insert_inode_ref(trans, dest,
9446 new_dentry->d_name.name,
9447 new_dentry->d_name.len,
9449 btrfs_ino(BTRFS_I(new_dir)), index);
9454 inode_inc_iversion(old_dir);
9455 inode_inc_iversion(new_dir);
9456 inode_inc_iversion(old_inode);
9457 old_dir->i_ctime = old_dir->i_mtime =
9458 new_dir->i_ctime = new_dir->i_mtime =
9459 old_inode->i_ctime = current_time(old_dir);
9461 if (old_dentry->d_parent != new_dentry->d_parent)
9462 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9463 BTRFS_I(old_inode), 1);
9465 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9466 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9468 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9469 BTRFS_I(d_inode(old_dentry)),
9470 old_dentry->d_name.name,
9471 old_dentry->d_name.len);
9473 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9476 btrfs_abort_transaction(trans, ret);
9481 inode_inc_iversion(new_inode);
9482 new_inode->i_ctime = current_time(new_inode);
9483 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9484 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9485 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9486 BUG_ON(new_inode->i_nlink == 0);
9488 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9489 BTRFS_I(d_inode(new_dentry)),
9490 new_dentry->d_name.name,
9491 new_dentry->d_name.len);
9493 if (!ret && new_inode->i_nlink == 0)
9494 ret = btrfs_orphan_add(trans,
9495 BTRFS_I(d_inode(new_dentry)));
9497 btrfs_abort_transaction(trans, ret);
9502 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9503 new_dentry->d_name.name,
9504 new_dentry->d_name.len, 0, index);
9506 btrfs_abort_transaction(trans, ret);
9510 if (old_inode->i_nlink == 1)
9511 BTRFS_I(old_inode)->dir_index = index;
9514 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9515 new_dentry->d_parent);
9516 btrfs_end_log_trans(root);
9520 if (flags & RENAME_WHITEOUT) {
9521 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9525 btrfs_abort_transaction(trans, ret);
9531 * If we have pinned the log and an error happened, we unpin tasks
9532 * trying to sync the log and force them to fallback to a transaction
9533 * commit if the log currently contains any of the inodes involved in
9534 * this rename operation (to ensure we do not persist a log with an
9535 * inconsistent state for any of these inodes or leading to any
9536 * inconsistencies when replayed). If the transaction was aborted, the
9537 * abortion reason is propagated to userspace when attempting to commit
9538 * the transaction. If the log does not contain any of these inodes, we
9539 * allow the tasks to sync it.
9541 if (ret && log_pinned) {
9542 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9543 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9544 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9546 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9547 btrfs_set_log_full_commit(trans);
9549 btrfs_end_log_trans(root);
9552 ret2 = btrfs_end_transaction(trans);
9553 ret = ret ? ret : ret2;
9555 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9556 up_read(&fs_info->subvol_sem);
9561 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9562 struct dentry *old_dentry, struct inode *new_dir,
9563 struct dentry *new_dentry, unsigned int flags)
9565 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9568 if (flags & RENAME_EXCHANGE)
9569 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9572 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9575 struct btrfs_delalloc_work {
9576 struct inode *inode;
9577 struct completion completion;
9578 struct list_head list;
9579 struct btrfs_work work;
9582 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9584 struct btrfs_delalloc_work *delalloc_work;
9585 struct inode *inode;
9587 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9589 inode = delalloc_work->inode;
9590 filemap_flush(inode->i_mapping);
9591 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9592 &BTRFS_I(inode)->runtime_flags))
9593 filemap_flush(inode->i_mapping);
9596 complete(&delalloc_work->completion);
9599 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9601 struct btrfs_delalloc_work *work;
9603 work = kmalloc(sizeof(*work), GFP_NOFS);
9607 init_completion(&work->completion);
9608 INIT_LIST_HEAD(&work->list);
9609 work->inode = inode;
9610 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9616 * some fairly slow code that needs optimization. This walks the list
9617 * of all the inodes with pending delalloc and forces them to disk.
9619 static int start_delalloc_inodes(struct btrfs_root *root,
9620 struct writeback_control *wbc, bool snapshot,
9621 bool in_reclaim_context)
9623 struct btrfs_inode *binode;
9624 struct inode *inode;
9625 struct btrfs_delalloc_work *work, *next;
9626 struct list_head works;
9627 struct list_head splice;
9629 bool full_flush = wbc->nr_to_write == LONG_MAX;
9631 INIT_LIST_HEAD(&works);
9632 INIT_LIST_HEAD(&splice);
9634 mutex_lock(&root->delalloc_mutex);
9635 spin_lock(&root->delalloc_lock);
9636 list_splice_init(&root->delalloc_inodes, &splice);
9637 while (!list_empty(&splice)) {
9638 binode = list_entry(splice.next, struct btrfs_inode,
9641 list_move_tail(&binode->delalloc_inodes,
9642 &root->delalloc_inodes);
9644 if (in_reclaim_context &&
9645 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9648 inode = igrab(&binode->vfs_inode);
9650 cond_resched_lock(&root->delalloc_lock);
9653 spin_unlock(&root->delalloc_lock);
9656 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9657 &binode->runtime_flags);
9659 work = btrfs_alloc_delalloc_work(inode);
9665 list_add_tail(&work->list, &works);
9666 btrfs_queue_work(root->fs_info->flush_workers,
9669 ret = sync_inode(inode, wbc);
9671 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9672 &BTRFS_I(inode)->runtime_flags))
9673 ret = sync_inode(inode, wbc);
9674 btrfs_add_delayed_iput(inode);
9675 if (ret || wbc->nr_to_write <= 0)
9679 spin_lock(&root->delalloc_lock);
9681 spin_unlock(&root->delalloc_lock);
9684 list_for_each_entry_safe(work, next, &works, list) {
9685 list_del_init(&work->list);
9686 wait_for_completion(&work->completion);
9690 if (!list_empty(&splice)) {
9691 spin_lock(&root->delalloc_lock);
9692 list_splice_tail(&splice, &root->delalloc_inodes);
9693 spin_unlock(&root->delalloc_lock);
9695 mutex_unlock(&root->delalloc_mutex);
9699 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9701 struct writeback_control wbc = {
9702 .nr_to_write = LONG_MAX,
9703 .sync_mode = WB_SYNC_NONE,
9705 .range_end = LLONG_MAX,
9707 struct btrfs_fs_info *fs_info = root->fs_info;
9709 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9712 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9715 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9716 bool in_reclaim_context)
9718 struct writeback_control wbc = {
9720 .sync_mode = WB_SYNC_NONE,
9722 .range_end = LLONG_MAX,
9724 struct btrfs_root *root;
9725 struct list_head splice;
9728 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9731 INIT_LIST_HEAD(&splice);
9733 mutex_lock(&fs_info->delalloc_root_mutex);
9734 spin_lock(&fs_info->delalloc_root_lock);
9735 list_splice_init(&fs_info->delalloc_roots, &splice);
9736 while (!list_empty(&splice)) {
9738 * Reset nr_to_write here so we know that we're doing a full
9742 wbc.nr_to_write = LONG_MAX;
9744 root = list_first_entry(&splice, struct btrfs_root,
9746 root = btrfs_grab_root(root);
9748 list_move_tail(&root->delalloc_root,
9749 &fs_info->delalloc_roots);
9750 spin_unlock(&fs_info->delalloc_root_lock);
9752 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9753 btrfs_put_root(root);
9754 if (ret < 0 || wbc.nr_to_write <= 0)
9756 spin_lock(&fs_info->delalloc_root_lock);
9758 spin_unlock(&fs_info->delalloc_root_lock);
9762 if (!list_empty(&splice)) {
9763 spin_lock(&fs_info->delalloc_root_lock);
9764 list_splice_tail(&splice, &fs_info->delalloc_roots);
9765 spin_unlock(&fs_info->delalloc_root_lock);
9767 mutex_unlock(&fs_info->delalloc_root_mutex);
9771 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9772 struct dentry *dentry, const char *symname)
9774 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9775 struct btrfs_trans_handle *trans;
9776 struct btrfs_root *root = BTRFS_I(dir)->root;
9777 struct btrfs_path *path;
9778 struct btrfs_key key;
9779 struct inode *inode = NULL;
9786 struct btrfs_file_extent_item *ei;
9787 struct extent_buffer *leaf;
9789 name_len = strlen(symname);
9790 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9791 return -ENAMETOOLONG;
9794 * 2 items for inode item and ref
9795 * 2 items for dir items
9796 * 1 item for updating parent inode item
9797 * 1 item for the inline extent item
9798 * 1 item for xattr if selinux is on
9800 trans = btrfs_start_transaction(root, 7);
9802 return PTR_ERR(trans);
9804 err = btrfs_get_free_objectid(root, &objectid);
9808 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9809 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9810 objectid, S_IFLNK|S_IRWXUGO, &index);
9811 if (IS_ERR(inode)) {
9812 err = PTR_ERR(inode);
9818 * If the active LSM wants to access the inode during
9819 * d_instantiate it needs these. Smack checks to see
9820 * if the filesystem supports xattrs by looking at the
9823 inode->i_fop = &btrfs_file_operations;
9824 inode->i_op = &btrfs_file_inode_operations;
9825 inode->i_mapping->a_ops = &btrfs_aops;
9827 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9831 path = btrfs_alloc_path();
9836 key.objectid = btrfs_ino(BTRFS_I(inode));
9838 key.type = BTRFS_EXTENT_DATA_KEY;
9839 datasize = btrfs_file_extent_calc_inline_size(name_len);
9840 err = btrfs_insert_empty_item(trans, root, path, &key,
9843 btrfs_free_path(path);
9846 leaf = path->nodes[0];
9847 ei = btrfs_item_ptr(leaf, path->slots[0],
9848 struct btrfs_file_extent_item);
9849 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9850 btrfs_set_file_extent_type(leaf, ei,
9851 BTRFS_FILE_EXTENT_INLINE);
9852 btrfs_set_file_extent_encryption(leaf, ei, 0);
9853 btrfs_set_file_extent_compression(leaf, ei, 0);
9854 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9855 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9857 ptr = btrfs_file_extent_inline_start(ei);
9858 write_extent_buffer(leaf, symname, ptr, name_len);
9859 btrfs_mark_buffer_dirty(leaf);
9860 btrfs_free_path(path);
9862 inode->i_op = &btrfs_symlink_inode_operations;
9863 inode_nohighmem(inode);
9864 inode_set_bytes(inode, name_len);
9865 btrfs_i_size_write(BTRFS_I(inode), name_len);
9866 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9868 * Last step, add directory indexes for our symlink inode. This is the
9869 * last step to avoid extra cleanup of these indexes if an error happens
9873 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9874 BTRFS_I(inode), 0, index);
9878 d_instantiate_new(dentry, inode);
9881 btrfs_end_transaction(trans);
9883 inode_dec_link_count(inode);
9884 discard_new_inode(inode);
9886 btrfs_btree_balance_dirty(fs_info);
9890 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9891 struct btrfs_trans_handle *trans_in,
9892 struct btrfs_inode *inode,
9893 struct btrfs_key *ins,
9896 struct btrfs_file_extent_item stack_fi;
9897 struct btrfs_replace_extent_info extent_info;
9898 struct btrfs_trans_handle *trans = trans_in;
9899 struct btrfs_path *path;
9900 u64 start = ins->objectid;
9901 u64 len = ins->offset;
9902 int qgroup_released;
9905 memset(&stack_fi, 0, sizeof(stack_fi));
9907 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9908 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9909 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9910 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9911 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9912 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9913 /* Encryption and other encoding is reserved and all 0 */
9915 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9916 if (qgroup_released < 0)
9917 return ERR_PTR(qgroup_released);
9920 ret = insert_reserved_file_extent(trans, inode,
9921 file_offset, &stack_fi,
9922 true, qgroup_released);
9928 extent_info.disk_offset = start;
9929 extent_info.disk_len = len;
9930 extent_info.data_offset = 0;
9931 extent_info.data_len = len;
9932 extent_info.file_offset = file_offset;
9933 extent_info.extent_buf = (char *)&stack_fi;
9934 extent_info.is_new_extent = true;
9935 extent_info.qgroup_reserved = qgroup_released;
9936 extent_info.insertions = 0;
9938 path = btrfs_alloc_path();
9944 ret = btrfs_replace_file_extents(inode, path, file_offset,
9945 file_offset + len - 1, &extent_info,
9947 btrfs_free_path(path);
9954 * We have released qgroup data range at the beginning of the function,
9955 * and normally qgroup_released bytes will be freed when committing
9957 * But if we error out early, we have to free what we have released
9958 * or we leak qgroup data reservation.
9960 btrfs_qgroup_free_refroot(inode->root->fs_info,
9961 inode->root->root_key.objectid, qgroup_released,
9962 BTRFS_QGROUP_RSV_DATA);
9963 return ERR_PTR(ret);
9966 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9967 u64 start, u64 num_bytes, u64 min_size,
9968 loff_t actual_len, u64 *alloc_hint,
9969 struct btrfs_trans_handle *trans)
9971 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9972 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9973 struct extent_map *em;
9974 struct btrfs_root *root = BTRFS_I(inode)->root;
9975 struct btrfs_key ins;
9976 u64 cur_offset = start;
9977 u64 clear_offset = start;
9980 u64 last_alloc = (u64)-1;
9982 bool own_trans = true;
9983 u64 end = start + num_bytes - 1;
9987 while (num_bytes > 0) {
9988 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9989 cur_bytes = max(cur_bytes, min_size);
9991 * If we are severely fragmented we could end up with really
9992 * small allocations, so if the allocator is returning small
9993 * chunks lets make its job easier by only searching for those
9996 cur_bytes = min(cur_bytes, last_alloc);
9997 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9998 min_size, 0, *alloc_hint, &ins, 1, 0);
10003 * We've reserved this space, and thus converted it from
10004 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10005 * from here on out we will only need to clear our reservation
10006 * for the remaining unreserved area, so advance our
10007 * clear_offset by our extent size.
10009 clear_offset += ins.offset;
10011 last_alloc = ins.offset;
10012 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10015 * Now that we inserted the prealloc extent we can finally
10016 * decrement the number of reservations in the block group.
10017 * If we did it before, we could race with relocation and have
10018 * relocation miss the reserved extent, making it fail later.
10020 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10021 if (IS_ERR(trans)) {
10022 ret = PTR_ERR(trans);
10023 btrfs_free_reserved_extent(fs_info, ins.objectid,
10028 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10029 cur_offset + ins.offset -1, 0);
10031 em = alloc_extent_map();
10033 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10034 &BTRFS_I(inode)->runtime_flags);
10038 em->start = cur_offset;
10039 em->orig_start = cur_offset;
10040 em->len = ins.offset;
10041 em->block_start = ins.objectid;
10042 em->block_len = ins.offset;
10043 em->orig_block_len = ins.offset;
10044 em->ram_bytes = ins.offset;
10045 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10046 em->generation = trans->transid;
10049 write_lock(&em_tree->lock);
10050 ret = add_extent_mapping(em_tree, em, 1);
10051 write_unlock(&em_tree->lock);
10052 if (ret != -EEXIST)
10054 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10055 cur_offset + ins.offset - 1,
10058 free_extent_map(em);
10060 num_bytes -= ins.offset;
10061 cur_offset += ins.offset;
10062 *alloc_hint = ins.objectid + ins.offset;
10064 inode_inc_iversion(inode);
10065 inode->i_ctime = current_time(inode);
10066 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10067 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10068 (actual_len > inode->i_size) &&
10069 (cur_offset > inode->i_size)) {
10070 if (cur_offset > actual_len)
10071 i_size = actual_len;
10073 i_size = cur_offset;
10074 i_size_write(inode, i_size);
10075 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10078 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10081 btrfs_abort_transaction(trans, ret);
10083 btrfs_end_transaction(trans);
10088 btrfs_end_transaction(trans);
10092 if (clear_offset < end)
10093 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10094 end - clear_offset + 1);
10098 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10099 u64 start, u64 num_bytes, u64 min_size,
10100 loff_t actual_len, u64 *alloc_hint)
10102 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10103 min_size, actual_len, alloc_hint,
10107 int btrfs_prealloc_file_range_trans(struct inode *inode,
10108 struct btrfs_trans_handle *trans, int mode,
10109 u64 start, u64 num_bytes, u64 min_size,
10110 loff_t actual_len, u64 *alloc_hint)
10112 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10113 min_size, actual_len, alloc_hint, trans);
10116 static int btrfs_set_page_dirty(struct page *page)
10118 return __set_page_dirty_nobuffers(page);
10121 static int btrfs_permission(struct user_namespace *mnt_userns,
10122 struct inode *inode, int mask)
10124 struct btrfs_root *root = BTRFS_I(inode)->root;
10125 umode_t mode = inode->i_mode;
10127 if (mask & MAY_WRITE &&
10128 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10129 if (btrfs_root_readonly(root))
10131 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10134 return generic_permission(&init_user_ns, inode, mask);
10137 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10138 struct dentry *dentry, umode_t mode)
10140 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10141 struct btrfs_trans_handle *trans;
10142 struct btrfs_root *root = BTRFS_I(dir)->root;
10143 struct inode *inode = NULL;
10149 * 5 units required for adding orphan entry
10151 trans = btrfs_start_transaction(root, 5);
10153 return PTR_ERR(trans);
10155 ret = btrfs_get_free_objectid(root, &objectid);
10159 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10160 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10161 if (IS_ERR(inode)) {
10162 ret = PTR_ERR(inode);
10167 inode->i_fop = &btrfs_file_operations;
10168 inode->i_op = &btrfs_file_inode_operations;
10170 inode->i_mapping->a_ops = &btrfs_aops;
10172 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10176 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10179 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10184 * We set number of links to 0 in btrfs_new_inode(), and here we set
10185 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10188 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10190 set_nlink(inode, 1);
10191 d_tmpfile(dentry, inode);
10192 unlock_new_inode(inode);
10193 mark_inode_dirty(inode);
10195 btrfs_end_transaction(trans);
10197 discard_new_inode(inode);
10198 btrfs_btree_balance_dirty(fs_info);
10202 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10204 struct inode *inode = tree->private_data;
10205 unsigned long index = start >> PAGE_SHIFT;
10206 unsigned long end_index = end >> PAGE_SHIFT;
10209 while (index <= end_index) {
10210 page = find_get_page(inode->i_mapping, index);
10211 ASSERT(page); /* Pages should be in the extent_io_tree */
10212 set_page_writeback(page);
10220 * Add an entry indicating a block group or device which is pinned by a
10221 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10222 * negative errno on failure.
10224 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10225 bool is_block_group)
10227 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10228 struct btrfs_swapfile_pin *sp, *entry;
10229 struct rb_node **p;
10230 struct rb_node *parent = NULL;
10232 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10237 sp->is_block_group = is_block_group;
10238 sp->bg_extent_count = 1;
10240 spin_lock(&fs_info->swapfile_pins_lock);
10241 p = &fs_info->swapfile_pins.rb_node;
10244 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10245 if (sp->ptr < entry->ptr ||
10246 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10247 p = &(*p)->rb_left;
10248 } else if (sp->ptr > entry->ptr ||
10249 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10250 p = &(*p)->rb_right;
10252 if (is_block_group)
10253 entry->bg_extent_count++;
10254 spin_unlock(&fs_info->swapfile_pins_lock);
10259 rb_link_node(&sp->node, parent, p);
10260 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10261 spin_unlock(&fs_info->swapfile_pins_lock);
10265 /* Free all of the entries pinned by this swapfile. */
10266 static void btrfs_free_swapfile_pins(struct inode *inode)
10268 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10269 struct btrfs_swapfile_pin *sp;
10270 struct rb_node *node, *next;
10272 spin_lock(&fs_info->swapfile_pins_lock);
10273 node = rb_first(&fs_info->swapfile_pins);
10275 next = rb_next(node);
10276 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10277 if (sp->inode == inode) {
10278 rb_erase(&sp->node, &fs_info->swapfile_pins);
10279 if (sp->is_block_group) {
10280 btrfs_dec_block_group_swap_extents(sp->ptr,
10281 sp->bg_extent_count);
10282 btrfs_put_block_group(sp->ptr);
10288 spin_unlock(&fs_info->swapfile_pins_lock);
10291 struct btrfs_swap_info {
10297 unsigned long nr_pages;
10301 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10302 struct btrfs_swap_info *bsi)
10304 unsigned long nr_pages;
10305 u64 first_ppage, first_ppage_reported, next_ppage;
10308 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10309 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10310 PAGE_SIZE) >> PAGE_SHIFT;
10312 if (first_ppage >= next_ppage)
10314 nr_pages = next_ppage - first_ppage;
10316 first_ppage_reported = first_ppage;
10317 if (bsi->start == 0)
10318 first_ppage_reported++;
10319 if (bsi->lowest_ppage > first_ppage_reported)
10320 bsi->lowest_ppage = first_ppage_reported;
10321 if (bsi->highest_ppage < (next_ppage - 1))
10322 bsi->highest_ppage = next_ppage - 1;
10324 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10327 bsi->nr_extents += ret;
10328 bsi->nr_pages += nr_pages;
10332 static void btrfs_swap_deactivate(struct file *file)
10334 struct inode *inode = file_inode(file);
10336 btrfs_free_swapfile_pins(inode);
10337 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10340 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10343 struct inode *inode = file_inode(file);
10344 struct btrfs_root *root = BTRFS_I(inode)->root;
10345 struct btrfs_fs_info *fs_info = root->fs_info;
10346 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10347 struct extent_state *cached_state = NULL;
10348 struct extent_map *em = NULL;
10349 struct btrfs_device *device = NULL;
10350 struct btrfs_swap_info bsi = {
10351 .lowest_ppage = (sector_t)-1ULL,
10358 * If the swap file was just created, make sure delalloc is done. If the
10359 * file changes again after this, the user is doing something stupid and
10360 * we don't really care.
10362 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10367 * The inode is locked, so these flags won't change after we check them.
10369 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10370 btrfs_warn(fs_info, "swapfile must not be compressed");
10373 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10374 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10377 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10378 btrfs_warn(fs_info, "swapfile must not be checksummed");
10383 * Balance or device remove/replace/resize can move stuff around from
10384 * under us. The exclop protection makes sure they aren't running/won't
10385 * run concurrently while we are mapping the swap extents, and
10386 * fs_info->swapfile_pins prevents them from running while the swap
10387 * file is active and moving the extents. Note that this also prevents
10388 * a concurrent device add which isn't actually necessary, but it's not
10389 * really worth the trouble to allow it.
10391 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10392 btrfs_warn(fs_info,
10393 "cannot activate swapfile while exclusive operation is running");
10398 * Prevent snapshot creation while we are activating the swap file.
10399 * We do not want to race with snapshot creation. If snapshot creation
10400 * already started before we bumped nr_swapfiles from 0 to 1 and
10401 * completes before the first write into the swap file after it is
10402 * activated, than that write would fallback to COW.
10404 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10405 btrfs_exclop_finish(fs_info);
10406 btrfs_warn(fs_info,
10407 "cannot activate swapfile because snapshot creation is in progress");
10411 * Snapshots can create extents which require COW even if NODATACOW is
10412 * set. We use this counter to prevent snapshots. We must increment it
10413 * before walking the extents because we don't want a concurrent
10414 * snapshot to run after we've already checked the extents.
10416 atomic_inc(&root->nr_swapfiles);
10418 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10420 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10422 while (start < isize) {
10423 u64 logical_block_start, physical_block_start;
10424 struct btrfs_block_group *bg;
10425 u64 len = isize - start;
10427 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10433 if (em->block_start == EXTENT_MAP_HOLE) {
10434 btrfs_warn(fs_info, "swapfile must not have holes");
10438 if (em->block_start == EXTENT_MAP_INLINE) {
10440 * It's unlikely we'll ever actually find ourselves
10441 * here, as a file small enough to fit inline won't be
10442 * big enough to store more than the swap header, but in
10443 * case something changes in the future, let's catch it
10444 * here rather than later.
10446 btrfs_warn(fs_info, "swapfile must not be inline");
10450 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10451 btrfs_warn(fs_info, "swapfile must not be compressed");
10456 logical_block_start = em->block_start + (start - em->start);
10457 len = min(len, em->len - (start - em->start));
10458 free_extent_map(em);
10461 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10467 btrfs_warn(fs_info,
10468 "swapfile must not be copy-on-write");
10473 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10479 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10480 btrfs_warn(fs_info,
10481 "swapfile must have single data profile");
10486 if (device == NULL) {
10487 device = em->map_lookup->stripes[0].dev;
10488 ret = btrfs_add_swapfile_pin(inode, device, false);
10493 } else if (device != em->map_lookup->stripes[0].dev) {
10494 btrfs_warn(fs_info, "swapfile must be on one device");
10499 physical_block_start = (em->map_lookup->stripes[0].physical +
10500 (logical_block_start - em->start));
10501 len = min(len, em->len - (logical_block_start - em->start));
10502 free_extent_map(em);
10505 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10507 btrfs_warn(fs_info,
10508 "could not find block group containing swapfile");
10513 if (!btrfs_inc_block_group_swap_extents(bg)) {
10514 btrfs_warn(fs_info,
10515 "block group for swapfile at %llu is read-only%s",
10517 atomic_read(&fs_info->scrubs_running) ?
10518 " (scrub running)" : "");
10519 btrfs_put_block_group(bg);
10524 ret = btrfs_add_swapfile_pin(inode, bg, true);
10526 btrfs_put_block_group(bg);
10533 if (bsi.block_len &&
10534 bsi.block_start + bsi.block_len == physical_block_start) {
10535 bsi.block_len += len;
10537 if (bsi.block_len) {
10538 ret = btrfs_add_swap_extent(sis, &bsi);
10543 bsi.block_start = physical_block_start;
10544 bsi.block_len = len;
10551 ret = btrfs_add_swap_extent(sis, &bsi);
10554 if (!IS_ERR_OR_NULL(em))
10555 free_extent_map(em);
10557 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10560 btrfs_swap_deactivate(file);
10562 btrfs_drew_write_unlock(&root->snapshot_lock);
10564 btrfs_exclop_finish(fs_info);
10570 sis->bdev = device->bdev;
10571 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10572 sis->max = bsi.nr_pages;
10573 sis->pages = bsi.nr_pages - 1;
10574 sis->highest_bit = bsi.nr_pages - 1;
10575 return bsi.nr_extents;
10578 static void btrfs_swap_deactivate(struct file *file)
10582 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10585 return -EOPNOTSUPP;
10590 * Update the number of bytes used in the VFS' inode. When we replace extents in
10591 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10592 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10593 * always get a correct value.
10595 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10596 const u64 add_bytes,
10597 const u64 del_bytes)
10599 if (add_bytes == del_bytes)
10602 spin_lock(&inode->lock);
10604 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10606 inode_add_bytes(&inode->vfs_inode, add_bytes);
10607 spin_unlock(&inode->lock);
10610 static const struct inode_operations btrfs_dir_inode_operations = {
10611 .getattr = btrfs_getattr,
10612 .lookup = btrfs_lookup,
10613 .create = btrfs_create,
10614 .unlink = btrfs_unlink,
10615 .link = btrfs_link,
10616 .mkdir = btrfs_mkdir,
10617 .rmdir = btrfs_rmdir,
10618 .rename = btrfs_rename2,
10619 .symlink = btrfs_symlink,
10620 .setattr = btrfs_setattr,
10621 .mknod = btrfs_mknod,
10622 .listxattr = btrfs_listxattr,
10623 .permission = btrfs_permission,
10624 .get_acl = btrfs_get_acl,
10625 .set_acl = btrfs_set_acl,
10626 .update_time = btrfs_update_time,
10627 .tmpfile = btrfs_tmpfile,
10628 .fileattr_get = btrfs_fileattr_get,
10629 .fileattr_set = btrfs_fileattr_set,
10632 static const struct file_operations btrfs_dir_file_operations = {
10633 .llseek = generic_file_llseek,
10634 .read = generic_read_dir,
10635 .iterate_shared = btrfs_real_readdir,
10636 .open = btrfs_opendir,
10637 .unlocked_ioctl = btrfs_ioctl,
10638 #ifdef CONFIG_COMPAT
10639 .compat_ioctl = btrfs_compat_ioctl,
10641 .release = btrfs_release_file,
10642 .fsync = btrfs_sync_file,
10646 * btrfs doesn't support the bmap operation because swapfiles
10647 * use bmap to make a mapping of extents in the file. They assume
10648 * these extents won't change over the life of the file and they
10649 * use the bmap result to do IO directly to the drive.
10651 * the btrfs bmap call would return logical addresses that aren't
10652 * suitable for IO and they also will change frequently as COW
10653 * operations happen. So, swapfile + btrfs == corruption.
10655 * For now we're avoiding this by dropping bmap.
10657 static const struct address_space_operations btrfs_aops = {
10658 .readpage = btrfs_readpage,
10659 .writepage = btrfs_writepage,
10660 .writepages = btrfs_writepages,
10661 .readahead = btrfs_readahead,
10662 .direct_IO = noop_direct_IO,
10663 .invalidatepage = btrfs_invalidatepage,
10664 .releasepage = btrfs_releasepage,
10665 #ifdef CONFIG_MIGRATION
10666 .migratepage = btrfs_migratepage,
10668 .set_page_dirty = btrfs_set_page_dirty,
10669 .error_remove_page = generic_error_remove_page,
10670 .swap_activate = btrfs_swap_activate,
10671 .swap_deactivate = btrfs_swap_deactivate,
10674 static const struct inode_operations btrfs_file_inode_operations = {
10675 .getattr = btrfs_getattr,
10676 .setattr = btrfs_setattr,
10677 .listxattr = btrfs_listxattr,
10678 .permission = btrfs_permission,
10679 .fiemap = btrfs_fiemap,
10680 .get_acl = btrfs_get_acl,
10681 .set_acl = btrfs_set_acl,
10682 .update_time = btrfs_update_time,
10683 .fileattr_get = btrfs_fileattr_get,
10684 .fileattr_set = btrfs_fileattr_set,
10686 static const struct inode_operations btrfs_special_inode_operations = {
10687 .getattr = btrfs_getattr,
10688 .setattr = btrfs_setattr,
10689 .permission = btrfs_permission,
10690 .listxattr = btrfs_listxattr,
10691 .get_acl = btrfs_get_acl,
10692 .set_acl = btrfs_set_acl,
10693 .update_time = btrfs_update_time,
10695 static const struct inode_operations btrfs_symlink_inode_operations = {
10696 .get_link = page_get_link,
10697 .getattr = btrfs_getattr,
10698 .setattr = btrfs_setattr,
10699 .permission = btrfs_permission,
10700 .listxattr = btrfs_listxattr,
10701 .update_time = btrfs_update_time,
10704 const struct dentry_operations btrfs_dentry_operations = {
10705 .d_delete = btrfs_dentry_delete,