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];
651 /* zero the tail end of the last page, we might be
652 * sending it down to disk
655 kaddr = kmap_atomic(page);
656 memset(kaddr + offset, 0,
658 kunmap_atomic(kaddr);
665 /* lets try to make an inline extent */
666 if (ret || total_in < actual_end) {
667 /* we didn't compress the entire range, try
668 * to make an uncompressed inline extent.
670 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
671 0, BTRFS_COMPRESS_NONE,
674 /* try making a compressed inline extent */
675 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
677 compress_type, pages);
680 unsigned long clear_flags = EXTENT_DELALLOC |
681 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
682 EXTENT_DO_ACCOUNTING;
683 unsigned long page_error_op;
685 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
688 * inline extent creation worked or returned error,
689 * we don't need to create any more async work items.
690 * Unlock and free up our temp pages.
692 * We use DO_ACCOUNTING here because we need the
693 * delalloc_release_metadata to be done _after_ we drop
694 * our outstanding extent for clearing delalloc for this
697 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
701 PAGE_START_WRITEBACK |
706 * Ensure we only free the compressed pages if we have
707 * them allocated, as we can still reach here with
708 * inode_need_compress() == false.
711 for (i = 0; i < nr_pages; i++) {
712 WARN_ON(pages[i]->mapping);
723 * we aren't doing an inline extent round the compressed size
724 * up to a block size boundary so the allocator does sane
727 total_compressed = ALIGN(total_compressed, blocksize);
730 * one last check to make sure the compression is really a
731 * win, compare the page count read with the blocks on disk,
732 * compression must free at least one sector size
734 total_in = ALIGN(total_in, PAGE_SIZE);
735 if (total_compressed + blocksize <= total_in) {
736 compressed_extents++;
739 * The async work queues will take care of doing actual
740 * allocation on disk for these compressed pages, and
741 * will submit them to the elevator.
743 add_async_extent(async_chunk, start, total_in,
744 total_compressed, pages, nr_pages,
747 if (start + total_in < end) {
753 return compressed_extents;
758 * the compression code ran but failed to make things smaller,
759 * free any pages it allocated and our page pointer array
761 for (i = 0; i < nr_pages; i++) {
762 WARN_ON(pages[i]->mapping);
767 total_compressed = 0;
770 /* flag the file so we don't compress in the future */
771 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
772 !(BTRFS_I(inode)->prop_compress)) {
773 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
776 cleanup_and_bail_uncompressed:
778 * No compression, but we still need to write the pages in the file
779 * we've been given so far. redirty the locked page if it corresponds
780 * to our extent and set things up for the async work queue to run
781 * cow_file_range to do the normal delalloc dance.
783 if (async_chunk->locked_page &&
784 (page_offset(async_chunk->locked_page) >= start &&
785 page_offset(async_chunk->locked_page)) <= end) {
786 __set_page_dirty_nobuffers(async_chunk->locked_page);
787 /* unlocked later on in the async handlers */
791 extent_range_redirty_for_io(inode, start, end);
792 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
793 BTRFS_COMPRESS_NONE);
794 compressed_extents++;
796 return compressed_extents;
799 static void free_async_extent_pages(struct async_extent *async_extent)
803 if (!async_extent->pages)
806 for (i = 0; i < async_extent->nr_pages; i++) {
807 WARN_ON(async_extent->pages[i]->mapping);
808 put_page(async_extent->pages[i]);
810 kfree(async_extent->pages);
811 async_extent->nr_pages = 0;
812 async_extent->pages = NULL;
816 * phase two of compressed writeback. This is the ordered portion
817 * of the code, which only gets called in the order the work was
818 * queued. We walk all the async extents created by compress_file_range
819 * and send them down to the disk.
821 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
823 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
824 struct btrfs_fs_info *fs_info = inode->root->fs_info;
825 struct async_extent *async_extent;
827 struct btrfs_key ins;
828 struct extent_map *em;
829 struct btrfs_root *root = inode->root;
830 struct extent_io_tree *io_tree = &inode->io_tree;
834 while (!list_empty(&async_chunk->extents)) {
835 async_extent = list_entry(async_chunk->extents.next,
836 struct async_extent, list);
837 list_del(&async_extent->list);
840 lock_extent(io_tree, async_extent->start,
841 async_extent->start + async_extent->ram_size - 1);
842 /* did the compression code fall back to uncompressed IO? */
843 if (!async_extent->pages) {
844 int page_started = 0;
845 unsigned long nr_written = 0;
847 /* allocate blocks */
848 ret = cow_file_range(inode, async_chunk->locked_page,
850 async_extent->start +
851 async_extent->ram_size - 1,
852 &page_started, &nr_written, 0);
857 * if page_started, cow_file_range inserted an
858 * inline extent and took care of all the unlocking
859 * and IO for us. Otherwise, we need to submit
860 * all those pages down to the drive.
862 if (!page_started && !ret)
863 extent_write_locked_range(&inode->vfs_inode,
865 async_extent->start +
866 async_extent->ram_size - 1,
868 else if (ret && async_chunk->locked_page)
869 unlock_page(async_chunk->locked_page);
875 ret = btrfs_reserve_extent(root, async_extent->ram_size,
876 async_extent->compressed_size,
877 async_extent->compressed_size,
878 0, alloc_hint, &ins, 1, 1);
880 free_async_extent_pages(async_extent);
882 if (ret == -ENOSPC) {
883 unlock_extent(io_tree, async_extent->start,
884 async_extent->start +
885 async_extent->ram_size - 1);
888 * we need to redirty the pages if we decide to
889 * fallback to uncompressed IO, otherwise we
890 * will not submit these pages down to lower
893 extent_range_redirty_for_io(&inode->vfs_inode,
895 async_extent->start +
896 async_extent->ram_size - 1);
903 * here we're doing allocation and writeback of the
906 em = create_io_em(inode, async_extent->start,
907 async_extent->ram_size, /* len */
908 async_extent->start, /* orig_start */
909 ins.objectid, /* block_start */
910 ins.offset, /* block_len */
911 ins.offset, /* orig_block_len */
912 async_extent->ram_size, /* ram_bytes */
913 async_extent->compress_type,
914 BTRFS_ORDERED_COMPRESSED);
916 /* ret value is not necessary due to void function */
917 goto out_free_reserve;
920 ret = btrfs_add_ordered_extent_compress(inode,
923 async_extent->ram_size,
925 async_extent->compress_type);
927 btrfs_drop_extent_cache(inode, async_extent->start,
928 async_extent->start +
929 async_extent->ram_size - 1, 0);
930 goto out_free_reserve;
932 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
935 * clear dirty, set writeback and unlock the pages.
937 extent_clear_unlock_delalloc(inode, async_extent->start,
938 async_extent->start +
939 async_extent->ram_size - 1,
940 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
941 PAGE_UNLOCK | PAGE_START_WRITEBACK);
942 if (btrfs_submit_compressed_write(inode, async_extent->start,
943 async_extent->ram_size,
945 ins.offset, async_extent->pages,
946 async_extent->nr_pages,
947 async_chunk->write_flags,
948 async_chunk->blkcg_css)) {
949 struct page *p = async_extent->pages[0];
950 const u64 start = async_extent->start;
951 const u64 end = start + async_extent->ram_size - 1;
953 p->mapping = inode->vfs_inode.i_mapping;
954 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
957 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
960 free_async_extent_pages(async_extent);
962 alloc_hint = ins.objectid + ins.offset;
968 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
969 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
971 extent_clear_unlock_delalloc(inode, async_extent->start,
972 async_extent->start +
973 async_extent->ram_size - 1,
974 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
975 EXTENT_DELALLOC_NEW |
976 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
977 PAGE_UNLOCK | PAGE_START_WRITEBACK |
978 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
979 free_async_extent_pages(async_extent);
984 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
987 struct extent_map_tree *em_tree = &inode->extent_tree;
988 struct extent_map *em;
991 read_lock(&em_tree->lock);
992 em = search_extent_mapping(em_tree, start, num_bytes);
995 * if block start isn't an actual block number then find the
996 * first block in this inode and use that as a hint. If that
997 * block is also bogus then just don't worry about it.
999 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1000 free_extent_map(em);
1001 em = search_extent_mapping(em_tree, 0, 0);
1002 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1003 alloc_hint = em->block_start;
1005 free_extent_map(em);
1007 alloc_hint = em->block_start;
1008 free_extent_map(em);
1011 read_unlock(&em_tree->lock);
1017 * when extent_io.c finds a delayed allocation range in the file,
1018 * the call backs end up in this code. The basic idea is to
1019 * allocate extents on disk for the range, and create ordered data structs
1020 * in ram to track those extents.
1022 * locked_page is the page that writepage had locked already. We use
1023 * it to make sure we don't do extra locks or unlocks.
1025 * *page_started is set to one if we unlock locked_page and do everything
1026 * required to start IO on it. It may be clean and already done with
1027 * IO when we return.
1029 static noinline int cow_file_range(struct btrfs_inode *inode,
1030 struct page *locked_page,
1031 u64 start, u64 end, int *page_started,
1032 unsigned long *nr_written, int unlock)
1034 struct btrfs_root *root = inode->root;
1035 struct btrfs_fs_info *fs_info = root->fs_info;
1038 unsigned long ram_size;
1039 u64 cur_alloc_size = 0;
1041 u64 blocksize = fs_info->sectorsize;
1042 struct btrfs_key ins;
1043 struct extent_map *em;
1044 unsigned clear_bits;
1045 unsigned long page_ops;
1046 bool extent_reserved = false;
1049 if (btrfs_is_free_space_inode(inode)) {
1055 num_bytes = ALIGN(end - start + 1, blocksize);
1056 num_bytes = max(blocksize, num_bytes);
1057 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1059 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1062 /* lets try to make an inline extent */
1063 ret = cow_file_range_inline(inode, start, end, 0,
1064 BTRFS_COMPRESS_NONE, NULL);
1067 * We use DO_ACCOUNTING here because we need the
1068 * delalloc_release_metadata to be run _after_ we drop
1069 * our outstanding extent for clearing delalloc for this
1072 extent_clear_unlock_delalloc(inode, start, end, NULL,
1073 EXTENT_LOCKED | EXTENT_DELALLOC |
1074 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1075 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1076 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1077 *nr_written = *nr_written +
1078 (end - start + PAGE_SIZE) / PAGE_SIZE;
1081 } else if (ret < 0) {
1086 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1087 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1090 * Relocation relies on the relocated extents to have exactly the same
1091 * size as the original extents. Normally writeback for relocation data
1092 * extents follows a NOCOW path because relocation preallocates the
1093 * extents. However, due to an operation such as scrub turning a block
1094 * group to RO mode, it may fallback to COW mode, so we must make sure
1095 * an extent allocated during COW has exactly the requested size and can
1096 * not be split into smaller extents, otherwise relocation breaks and
1097 * fails during the stage where it updates the bytenr of file extent
1100 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1101 min_alloc_size = num_bytes;
1103 min_alloc_size = fs_info->sectorsize;
1105 while (num_bytes > 0) {
1106 cur_alloc_size = num_bytes;
1107 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1108 min_alloc_size, 0, alloc_hint,
1112 cur_alloc_size = ins.offset;
1113 extent_reserved = true;
1115 ram_size = ins.offset;
1116 em = create_io_em(inode, start, ins.offset, /* len */
1117 start, /* orig_start */
1118 ins.objectid, /* block_start */
1119 ins.offset, /* block_len */
1120 ins.offset, /* orig_block_len */
1121 ram_size, /* ram_bytes */
1122 BTRFS_COMPRESS_NONE, /* compress_type */
1123 BTRFS_ORDERED_REGULAR /* type */);
1128 free_extent_map(em);
1130 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1131 ram_size, cur_alloc_size,
1132 BTRFS_ORDERED_REGULAR);
1134 goto out_drop_extent_cache;
1136 if (root->root_key.objectid ==
1137 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1138 ret = btrfs_reloc_clone_csums(inode, start,
1141 * Only drop cache here, and process as normal.
1143 * We must not allow extent_clear_unlock_delalloc()
1144 * at out_unlock label to free meta of this ordered
1145 * extent, as its meta should be freed by
1146 * btrfs_finish_ordered_io().
1148 * So we must continue until @start is increased to
1149 * skip current ordered extent.
1152 btrfs_drop_extent_cache(inode, start,
1153 start + ram_size - 1, 0);
1156 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1158 /* we're not doing compressed IO, don't unlock the first
1159 * page (which the caller expects to stay locked), don't
1160 * clear any dirty bits and don't set any writeback bits
1162 * Do set the Private2 bit so we know this page was properly
1163 * setup for writepage
1165 page_ops = unlock ? PAGE_UNLOCK : 0;
1166 page_ops |= PAGE_SET_PRIVATE2;
1168 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1170 EXTENT_LOCKED | EXTENT_DELALLOC,
1172 if (num_bytes < cur_alloc_size)
1175 num_bytes -= cur_alloc_size;
1176 alloc_hint = ins.objectid + ins.offset;
1177 start += cur_alloc_size;
1178 extent_reserved = false;
1181 * btrfs_reloc_clone_csums() error, since start is increased
1182 * extent_clear_unlock_delalloc() at out_unlock label won't
1183 * free metadata of current ordered extent, we're OK to exit.
1191 out_drop_extent_cache:
1192 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1194 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1195 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1197 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1198 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1199 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1201 * If we reserved an extent for our delalloc range (or a subrange) and
1202 * failed to create the respective ordered extent, then it means that
1203 * when we reserved the extent we decremented the extent's size from
1204 * the data space_info's bytes_may_use counter and incremented the
1205 * space_info's bytes_reserved counter by the same amount. We must make
1206 * sure extent_clear_unlock_delalloc() does not try to decrement again
1207 * the data space_info's bytes_may_use counter, therefore we do not pass
1208 * it the flag EXTENT_CLEAR_DATA_RESV.
1210 if (extent_reserved) {
1211 extent_clear_unlock_delalloc(inode, start,
1212 start + cur_alloc_size - 1,
1216 start += cur_alloc_size;
1220 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1221 clear_bits | EXTENT_CLEAR_DATA_RESV,
1227 * work queue call back to started compression on a file and pages
1229 static noinline void async_cow_start(struct btrfs_work *work)
1231 struct async_chunk *async_chunk;
1232 int compressed_extents;
1234 async_chunk = container_of(work, struct async_chunk, work);
1236 compressed_extents = compress_file_range(async_chunk);
1237 if (compressed_extents == 0) {
1238 btrfs_add_delayed_iput(async_chunk->inode);
1239 async_chunk->inode = NULL;
1244 * work queue call back to submit previously compressed pages
1246 static noinline void async_cow_submit(struct btrfs_work *work)
1248 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1250 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1251 unsigned long nr_pages;
1253 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1256 /* atomic_sub_return implies a barrier */
1257 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1259 cond_wake_up_nomb(&fs_info->async_submit_wait);
1262 * ->inode could be NULL if async_chunk_start has failed to compress,
1263 * in which case we don't have anything to submit, yet we need to
1264 * always adjust ->async_delalloc_pages as its paired with the init
1265 * happening in cow_file_range_async
1267 if (async_chunk->inode)
1268 submit_compressed_extents(async_chunk);
1271 static noinline void async_cow_free(struct btrfs_work *work)
1273 struct async_chunk *async_chunk;
1275 async_chunk = container_of(work, struct async_chunk, work);
1276 if (async_chunk->inode)
1277 btrfs_add_delayed_iput(async_chunk->inode);
1278 if (async_chunk->blkcg_css)
1279 css_put(async_chunk->blkcg_css);
1281 * Since the pointer to 'pending' is at the beginning of the array of
1282 * async_chunk's, freeing it ensures the whole array has been freed.
1284 if (atomic_dec_and_test(async_chunk->pending))
1285 kvfree(async_chunk->pending);
1288 static int cow_file_range_async(struct btrfs_inode *inode,
1289 struct writeback_control *wbc,
1290 struct page *locked_page,
1291 u64 start, u64 end, int *page_started,
1292 unsigned long *nr_written)
1294 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1295 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1296 struct async_cow *ctx;
1297 struct async_chunk *async_chunk;
1298 unsigned long nr_pages;
1300 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1302 bool should_compress;
1304 const unsigned int write_flags = wbc_to_write_flags(wbc);
1306 unlock_extent(&inode->io_tree, start, end);
1308 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1309 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1311 should_compress = false;
1313 should_compress = true;
1316 nofs_flag = memalloc_nofs_save();
1317 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1318 memalloc_nofs_restore(nofs_flag);
1321 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1322 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1323 EXTENT_DO_ACCOUNTING;
1324 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1325 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1327 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1328 clear_bits, page_ops);
1332 async_chunk = ctx->chunks;
1333 atomic_set(&ctx->num_chunks, num_chunks);
1335 for (i = 0; i < num_chunks; i++) {
1336 if (should_compress)
1337 cur_end = min(end, start + SZ_512K - 1);
1342 * igrab is called higher up in the call chain, take only the
1343 * lightweight reference for the callback lifetime
1345 ihold(&inode->vfs_inode);
1346 async_chunk[i].pending = &ctx->num_chunks;
1347 async_chunk[i].inode = &inode->vfs_inode;
1348 async_chunk[i].start = start;
1349 async_chunk[i].end = cur_end;
1350 async_chunk[i].write_flags = write_flags;
1351 INIT_LIST_HEAD(&async_chunk[i].extents);
1354 * The locked_page comes all the way from writepage and its
1355 * the original page we were actually given. As we spread
1356 * this large delalloc region across multiple async_chunk
1357 * structs, only the first struct needs a pointer to locked_page
1359 * This way we don't need racey decisions about who is supposed
1364 * Depending on the compressibility, the pages might or
1365 * might not go through async. We want all of them to
1366 * be accounted against wbc once. Let's do it here
1367 * before the paths diverge. wbc accounting is used
1368 * only for foreign writeback detection and doesn't
1369 * need full accuracy. Just account the whole thing
1370 * against the first page.
1372 wbc_account_cgroup_owner(wbc, locked_page,
1374 async_chunk[i].locked_page = locked_page;
1377 async_chunk[i].locked_page = NULL;
1380 if (blkcg_css != blkcg_root_css) {
1382 async_chunk[i].blkcg_css = blkcg_css;
1384 async_chunk[i].blkcg_css = NULL;
1387 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1388 async_cow_submit, async_cow_free);
1390 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1391 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1393 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1395 *nr_written += nr_pages;
1396 start = cur_end + 1;
1402 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1403 struct page *locked_page, u64 start,
1404 u64 end, int *page_started,
1405 unsigned long *nr_written)
1409 ret = cow_file_range(inode, locked_page, start, end, page_started,
1417 __set_page_dirty_nobuffers(locked_page);
1418 account_page_redirty(locked_page);
1419 extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL);
1425 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1426 u64 bytenr, u64 num_bytes)
1429 struct btrfs_ordered_sum *sums;
1432 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1433 bytenr + num_bytes - 1, &list, 0);
1434 if (ret == 0 && list_empty(&list))
1437 while (!list_empty(&list)) {
1438 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1439 list_del(&sums->list);
1447 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1448 const u64 start, const u64 end,
1449 int *page_started, unsigned long *nr_written)
1451 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1452 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1453 BTRFS_DATA_RELOC_TREE_OBJECTID);
1454 const u64 range_bytes = end + 1 - start;
1455 struct extent_io_tree *io_tree = &inode->io_tree;
1456 u64 range_start = start;
1460 * If EXTENT_NORESERVE is set it means that when the buffered write was
1461 * made we had not enough available data space and therefore we did not
1462 * reserve data space for it, since we though we could do NOCOW for the
1463 * respective file range (either there is prealloc extent or the inode
1464 * has the NOCOW bit set).
1466 * However when we need to fallback to COW mode (because for example the
1467 * block group for the corresponding extent was turned to RO mode by a
1468 * scrub or relocation) we need to do the following:
1470 * 1) We increment the bytes_may_use counter of the data space info.
1471 * If COW succeeds, it allocates a new data extent and after doing
1472 * that it decrements the space info's bytes_may_use counter and
1473 * increments its bytes_reserved counter by the same amount (we do
1474 * this at btrfs_add_reserved_bytes()). So we need to increment the
1475 * bytes_may_use counter to compensate (when space is reserved at
1476 * buffered write time, the bytes_may_use counter is incremented);
1478 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1479 * that if the COW path fails for any reason, it decrements (through
1480 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1481 * data space info, which we incremented in the step above.
1483 * If we need to fallback to cow and the inode corresponds to a free
1484 * space cache inode or an inode of the data relocation tree, we must
1485 * also increment bytes_may_use of the data space_info for the same
1486 * reason. Space caches and relocated data extents always get a prealloc
1487 * extent for them, however scrub or balance may have set the block
1488 * group that contains that extent to RO mode and therefore force COW
1489 * when starting writeback.
1491 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1492 EXTENT_NORESERVE, 0);
1493 if (count > 0 || is_space_ino || is_reloc_ino) {
1495 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1496 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1498 if (is_space_ino || is_reloc_ino)
1499 bytes = range_bytes;
1501 spin_lock(&sinfo->lock);
1502 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1503 spin_unlock(&sinfo->lock);
1506 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1510 return cow_file_range(inode, locked_page, start, end, page_started,
1515 * when nowcow writeback call back. This checks for snapshots or COW copies
1516 * of the extents that exist in the file, and COWs the file as required.
1518 * If no cow copies or snapshots exist, we write directly to the existing
1521 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1522 struct page *locked_page,
1523 const u64 start, const u64 end,
1525 unsigned long *nr_written)
1527 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1528 struct btrfs_root *root = inode->root;
1529 struct btrfs_path *path;
1530 u64 cow_start = (u64)-1;
1531 u64 cur_offset = start;
1533 bool check_prev = true;
1534 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1535 u64 ino = btrfs_ino(inode);
1537 u64 disk_bytenr = 0;
1538 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1540 path = btrfs_alloc_path();
1542 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1543 EXTENT_LOCKED | EXTENT_DELALLOC |
1544 EXTENT_DO_ACCOUNTING |
1545 EXTENT_DEFRAG, PAGE_UNLOCK |
1546 PAGE_START_WRITEBACK |
1547 PAGE_END_WRITEBACK);
1552 struct btrfs_key found_key;
1553 struct btrfs_file_extent_item *fi;
1554 struct extent_buffer *leaf;
1564 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1570 * If there is no extent for our range when doing the initial
1571 * search, then go back to the previous slot as it will be the
1572 * one containing the search offset
1574 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1575 leaf = path->nodes[0];
1576 btrfs_item_key_to_cpu(leaf, &found_key,
1577 path->slots[0] - 1);
1578 if (found_key.objectid == ino &&
1579 found_key.type == BTRFS_EXTENT_DATA_KEY)
1584 /* Go to next leaf if we have exhausted the current one */
1585 leaf = path->nodes[0];
1586 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1587 ret = btrfs_next_leaf(root, path);
1589 if (cow_start != (u64)-1)
1590 cur_offset = cow_start;
1595 leaf = path->nodes[0];
1598 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1600 /* Didn't find anything for our INO */
1601 if (found_key.objectid > ino)
1604 * Keep searching until we find an EXTENT_ITEM or there are no
1605 * more extents for this inode
1607 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1608 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1613 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1614 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1615 found_key.offset > end)
1619 * If the found extent starts after requested offset, then
1620 * adjust extent_end to be right before this extent begins
1622 if (found_key.offset > cur_offset) {
1623 extent_end = found_key.offset;
1629 * Found extent which begins before our range and potentially
1632 fi = btrfs_item_ptr(leaf, path->slots[0],
1633 struct btrfs_file_extent_item);
1634 extent_type = btrfs_file_extent_type(leaf, fi);
1636 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1637 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1638 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1639 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1640 extent_offset = btrfs_file_extent_offset(leaf, fi);
1641 extent_end = found_key.offset +
1642 btrfs_file_extent_num_bytes(leaf, fi);
1644 btrfs_file_extent_disk_num_bytes(leaf, fi);
1646 * If the extent we got ends before our current offset,
1647 * skip to the next extent.
1649 if (extent_end <= cur_offset) {
1654 if (disk_bytenr == 0)
1656 /* Skip compressed/encrypted/encoded extents */
1657 if (btrfs_file_extent_compression(leaf, fi) ||
1658 btrfs_file_extent_encryption(leaf, fi) ||
1659 btrfs_file_extent_other_encoding(leaf, fi))
1662 * If extent is created before the last volume's snapshot
1663 * this implies the extent is shared, hence we can't do
1664 * nocow. This is the same check as in
1665 * btrfs_cross_ref_exist but without calling
1666 * btrfs_search_slot.
1668 if (!freespace_inode &&
1669 btrfs_file_extent_generation(leaf, fi) <=
1670 btrfs_root_last_snapshot(&root->root_item))
1672 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1676 * The following checks can be expensive, as they need to
1677 * take other locks and do btree or rbtree searches, so
1678 * release the path to avoid blocking other tasks for too
1681 btrfs_release_path(path);
1683 ret = btrfs_cross_ref_exist(root, ino,
1685 extent_offset, disk_bytenr, false);
1688 * ret could be -EIO if the above fails to read
1692 if (cow_start != (u64)-1)
1693 cur_offset = cow_start;
1697 WARN_ON_ONCE(freespace_inode);
1700 disk_bytenr += extent_offset;
1701 disk_bytenr += cur_offset - found_key.offset;
1702 num_bytes = min(end + 1, extent_end) - cur_offset;
1704 * If there are pending snapshots for this root, we
1705 * fall into common COW way
1707 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1710 * force cow if csum exists in the range.
1711 * this ensure that csum for a given extent are
1712 * either valid or do not exist.
1714 ret = csum_exist_in_range(fs_info, disk_bytenr,
1718 * ret could be -EIO if the above fails to read
1722 if (cow_start != (u64)-1)
1723 cur_offset = cow_start;
1726 WARN_ON_ONCE(freespace_inode);
1729 /* If the extent's block group is RO, we must COW */
1730 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1733 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1734 extent_end = found_key.offset + ram_bytes;
1735 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1736 /* Skip extents outside of our requested range */
1737 if (extent_end <= start) {
1742 /* If this triggers then we have a memory corruption */
1747 * If nocow is false then record the beginning of the range
1748 * that needs to be COWed
1751 if (cow_start == (u64)-1)
1752 cow_start = cur_offset;
1753 cur_offset = extent_end;
1754 if (cur_offset > end)
1756 if (!path->nodes[0])
1763 * COW range from cow_start to found_key.offset - 1. As the key
1764 * will contain the beginning of the first extent that can be
1765 * NOCOW, following one which needs to be COW'ed
1767 if (cow_start != (u64)-1) {
1768 ret = fallback_to_cow(inode, locked_page,
1769 cow_start, found_key.offset - 1,
1770 page_started, nr_written);
1773 cow_start = (u64)-1;
1776 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1777 u64 orig_start = found_key.offset - extent_offset;
1778 struct extent_map *em;
1780 em = create_io_em(inode, cur_offset, num_bytes,
1782 disk_bytenr, /* block_start */
1783 num_bytes, /* block_len */
1784 disk_num_bytes, /* orig_block_len */
1785 ram_bytes, BTRFS_COMPRESS_NONE,
1786 BTRFS_ORDERED_PREALLOC);
1791 free_extent_map(em);
1792 ret = btrfs_add_ordered_extent(inode, cur_offset,
1793 disk_bytenr, num_bytes,
1795 BTRFS_ORDERED_PREALLOC);
1797 btrfs_drop_extent_cache(inode, cur_offset,
1798 cur_offset + num_bytes - 1,
1803 ret = btrfs_add_ordered_extent(inode, cur_offset,
1804 disk_bytenr, num_bytes,
1806 BTRFS_ORDERED_NOCOW);
1812 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1815 if (root->root_key.objectid ==
1816 BTRFS_DATA_RELOC_TREE_OBJECTID)
1818 * Error handled later, as we must prevent
1819 * extent_clear_unlock_delalloc() in error handler
1820 * from freeing metadata of created ordered extent.
1822 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1825 extent_clear_unlock_delalloc(inode, cur_offset,
1826 cur_offset + num_bytes - 1,
1827 locked_page, EXTENT_LOCKED |
1829 EXTENT_CLEAR_DATA_RESV,
1830 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1832 cur_offset = extent_end;
1835 * btrfs_reloc_clone_csums() error, now we're OK to call error
1836 * handler, as metadata for created ordered extent will only
1837 * be freed by btrfs_finish_ordered_io().
1841 if (cur_offset > end)
1844 btrfs_release_path(path);
1846 if (cur_offset <= end && cow_start == (u64)-1)
1847 cow_start = cur_offset;
1849 if (cow_start != (u64)-1) {
1851 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1852 page_started, nr_written);
1859 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1861 if (ret && cur_offset < end)
1862 extent_clear_unlock_delalloc(inode, cur_offset, end,
1863 locked_page, EXTENT_LOCKED |
1864 EXTENT_DELALLOC | EXTENT_DEFRAG |
1865 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1866 PAGE_START_WRITEBACK |
1867 PAGE_END_WRITEBACK);
1868 btrfs_free_path(path);
1872 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1874 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1875 if (inode->defrag_bytes &&
1876 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1885 * Function to process delayed allocation (create CoW) for ranges which are
1886 * being touched for the first time.
1888 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1889 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1890 struct writeback_control *wbc)
1893 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1895 if (should_nocow(inode, start, end)) {
1897 ret = run_delalloc_nocow(inode, locked_page, start, end,
1898 page_started, nr_written);
1899 } else if (!inode_can_compress(inode) ||
1900 !inode_need_compress(inode, start, end)) {
1902 ret = run_delalloc_zoned(inode, locked_page, start, end,
1903 page_started, nr_written);
1905 ret = cow_file_range(inode, locked_page, start, end,
1906 page_started, nr_written, 1);
1908 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1909 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1910 page_started, nr_written);
1913 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1918 void btrfs_split_delalloc_extent(struct inode *inode,
1919 struct extent_state *orig, u64 split)
1923 /* not delalloc, ignore it */
1924 if (!(orig->state & EXTENT_DELALLOC))
1927 size = orig->end - orig->start + 1;
1928 if (size > BTRFS_MAX_EXTENT_SIZE) {
1933 * See the explanation in btrfs_merge_delalloc_extent, the same
1934 * applies here, just in reverse.
1936 new_size = orig->end - split + 1;
1937 num_extents = count_max_extents(new_size);
1938 new_size = split - orig->start;
1939 num_extents += count_max_extents(new_size);
1940 if (count_max_extents(size) >= num_extents)
1944 spin_lock(&BTRFS_I(inode)->lock);
1945 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1946 spin_unlock(&BTRFS_I(inode)->lock);
1950 * Handle merged delayed allocation extents so we can keep track of new extents
1951 * that are just merged onto old extents, such as when we are doing sequential
1952 * writes, so we can properly account for the metadata space we'll need.
1954 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1955 struct extent_state *other)
1957 u64 new_size, old_size;
1960 /* not delalloc, ignore it */
1961 if (!(other->state & EXTENT_DELALLOC))
1964 if (new->start > other->start)
1965 new_size = new->end - other->start + 1;
1967 new_size = other->end - new->start + 1;
1969 /* we're not bigger than the max, unreserve the space and go */
1970 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1971 spin_lock(&BTRFS_I(inode)->lock);
1972 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1973 spin_unlock(&BTRFS_I(inode)->lock);
1978 * We have to add up either side to figure out how many extents were
1979 * accounted for before we merged into one big extent. If the number of
1980 * extents we accounted for is <= the amount we need for the new range
1981 * then we can return, otherwise drop. Think of it like this
1985 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1986 * need 2 outstanding extents, on one side we have 1 and the other side
1987 * we have 1 so they are == and we can return. But in this case
1989 * [MAX_SIZE+4k][MAX_SIZE+4k]
1991 * Each range on their own accounts for 2 extents, but merged together
1992 * they are only 3 extents worth of accounting, so we need to drop in
1995 old_size = other->end - other->start + 1;
1996 num_extents = count_max_extents(old_size);
1997 old_size = new->end - new->start + 1;
1998 num_extents += count_max_extents(old_size);
1999 if (count_max_extents(new_size) >= num_extents)
2002 spin_lock(&BTRFS_I(inode)->lock);
2003 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
2004 spin_unlock(&BTRFS_I(inode)->lock);
2007 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2008 struct inode *inode)
2010 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2012 spin_lock(&root->delalloc_lock);
2013 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2014 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2015 &root->delalloc_inodes);
2016 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2017 &BTRFS_I(inode)->runtime_flags);
2018 root->nr_delalloc_inodes++;
2019 if (root->nr_delalloc_inodes == 1) {
2020 spin_lock(&fs_info->delalloc_root_lock);
2021 BUG_ON(!list_empty(&root->delalloc_root));
2022 list_add_tail(&root->delalloc_root,
2023 &fs_info->delalloc_roots);
2024 spin_unlock(&fs_info->delalloc_root_lock);
2027 spin_unlock(&root->delalloc_lock);
2031 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2032 struct btrfs_inode *inode)
2034 struct btrfs_fs_info *fs_info = root->fs_info;
2036 if (!list_empty(&inode->delalloc_inodes)) {
2037 list_del_init(&inode->delalloc_inodes);
2038 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2039 &inode->runtime_flags);
2040 root->nr_delalloc_inodes--;
2041 if (!root->nr_delalloc_inodes) {
2042 ASSERT(list_empty(&root->delalloc_inodes));
2043 spin_lock(&fs_info->delalloc_root_lock);
2044 BUG_ON(list_empty(&root->delalloc_root));
2045 list_del_init(&root->delalloc_root);
2046 spin_unlock(&fs_info->delalloc_root_lock);
2051 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2052 struct btrfs_inode *inode)
2054 spin_lock(&root->delalloc_lock);
2055 __btrfs_del_delalloc_inode(root, inode);
2056 spin_unlock(&root->delalloc_lock);
2060 * Properly track delayed allocation bytes in the inode and to maintain the
2061 * list of inodes that have pending delalloc work to be done.
2063 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2066 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2068 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2071 * set_bit and clear bit hooks normally require _irqsave/restore
2072 * but in this case, we are only testing for the DELALLOC
2073 * bit, which is only set or cleared with irqs on
2075 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2076 struct btrfs_root *root = BTRFS_I(inode)->root;
2077 u64 len = state->end + 1 - state->start;
2078 u32 num_extents = count_max_extents(len);
2079 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2081 spin_lock(&BTRFS_I(inode)->lock);
2082 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2083 spin_unlock(&BTRFS_I(inode)->lock);
2085 /* For sanity tests */
2086 if (btrfs_is_testing(fs_info))
2089 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2090 fs_info->delalloc_batch);
2091 spin_lock(&BTRFS_I(inode)->lock);
2092 BTRFS_I(inode)->delalloc_bytes += len;
2093 if (*bits & EXTENT_DEFRAG)
2094 BTRFS_I(inode)->defrag_bytes += len;
2095 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2096 &BTRFS_I(inode)->runtime_flags))
2097 btrfs_add_delalloc_inodes(root, inode);
2098 spin_unlock(&BTRFS_I(inode)->lock);
2101 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2102 (*bits & EXTENT_DELALLOC_NEW)) {
2103 spin_lock(&BTRFS_I(inode)->lock);
2104 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2106 spin_unlock(&BTRFS_I(inode)->lock);
2111 * Once a range is no longer delalloc this function ensures that proper
2112 * accounting happens.
2114 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2115 struct extent_state *state, unsigned *bits)
2117 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2118 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2119 u64 len = state->end + 1 - state->start;
2120 u32 num_extents = count_max_extents(len);
2122 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2123 spin_lock(&inode->lock);
2124 inode->defrag_bytes -= len;
2125 spin_unlock(&inode->lock);
2129 * set_bit and clear bit hooks normally require _irqsave/restore
2130 * but in this case, we are only testing for the DELALLOC
2131 * bit, which is only set or cleared with irqs on
2133 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2134 struct btrfs_root *root = inode->root;
2135 bool do_list = !btrfs_is_free_space_inode(inode);
2137 spin_lock(&inode->lock);
2138 btrfs_mod_outstanding_extents(inode, -num_extents);
2139 spin_unlock(&inode->lock);
2142 * We don't reserve metadata space for space cache inodes so we
2143 * don't need to call delalloc_release_metadata if there is an
2146 if (*bits & EXTENT_CLEAR_META_RESV &&
2147 root != fs_info->tree_root)
2148 btrfs_delalloc_release_metadata(inode, len, false);
2150 /* For sanity tests. */
2151 if (btrfs_is_testing(fs_info))
2154 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2155 do_list && !(state->state & EXTENT_NORESERVE) &&
2156 (*bits & EXTENT_CLEAR_DATA_RESV))
2157 btrfs_free_reserved_data_space_noquota(fs_info, len);
2159 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2160 fs_info->delalloc_batch);
2161 spin_lock(&inode->lock);
2162 inode->delalloc_bytes -= len;
2163 if (do_list && inode->delalloc_bytes == 0 &&
2164 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2165 &inode->runtime_flags))
2166 btrfs_del_delalloc_inode(root, inode);
2167 spin_unlock(&inode->lock);
2170 if ((state->state & EXTENT_DELALLOC_NEW) &&
2171 (*bits & EXTENT_DELALLOC_NEW)) {
2172 spin_lock(&inode->lock);
2173 ASSERT(inode->new_delalloc_bytes >= len);
2174 inode->new_delalloc_bytes -= len;
2175 if (*bits & EXTENT_ADD_INODE_BYTES)
2176 inode_add_bytes(&inode->vfs_inode, len);
2177 spin_unlock(&inode->lock);
2182 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2183 * in a chunk's stripe. This function ensures that bios do not span a
2186 * @page - The page we are about to add to the bio
2187 * @size - size we want to add to the bio
2188 * @bio - bio we want to ensure is smaller than a stripe
2189 * @bio_flags - flags of the bio
2191 * return 1 if page cannot be added to the bio
2192 * return 0 if page can be added to the bio
2193 * return error otherwise
2195 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2196 unsigned long bio_flags)
2198 struct inode *inode = page->mapping->host;
2199 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2200 u64 logical = bio->bi_iter.bi_sector << 9;
2201 struct extent_map *em;
2205 struct btrfs_io_geometry geom;
2207 if (bio_flags & EXTENT_BIO_COMPRESSED)
2210 length = bio->bi_iter.bi_size;
2211 map_length = length;
2212 em = btrfs_get_chunk_map(fs_info, logical, map_length);
2215 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical,
2220 if (geom.len < length + size)
2223 free_extent_map(em);
2228 * in order to insert checksums into the metadata in large chunks,
2229 * we wait until bio submission time. All the pages in the bio are
2230 * checksummed and sums are attached onto the ordered extent record.
2232 * At IO completion time the cums attached on the ordered extent record
2233 * are inserted into the btree
2235 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2236 u64 dio_file_offset)
2238 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2241 bool btrfs_bio_fits_in_ordered_extent(struct page *page, struct bio *bio,
2244 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2245 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2246 struct btrfs_ordered_extent *ordered;
2247 u64 len = bio->bi_iter.bi_size + size;
2250 ASSERT(btrfs_is_zoned(fs_info));
2251 ASSERT(fs_info->max_zone_append_size > 0);
2252 ASSERT(bio_op(bio) == REQ_OP_ZONE_APPEND);
2254 /* Ordered extent not yet created, so we're good */
2255 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
2259 if ((bio->bi_iter.bi_sector << SECTOR_SHIFT) + len >
2260 ordered->disk_bytenr + ordered->disk_num_bytes)
2263 btrfs_put_ordered_extent(ordered);
2268 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2269 struct bio *bio, loff_t file_offset)
2271 struct btrfs_ordered_extent *ordered;
2272 struct extent_map *em = NULL, *em_new = NULL;
2273 struct extent_map_tree *em_tree = &inode->extent_tree;
2274 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2275 u64 len = bio->bi_iter.bi_size;
2276 u64 end = start + len;
2281 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2282 if (WARN_ON_ONCE(!ordered))
2283 return BLK_STS_IOERR;
2285 /* No need to split */
2286 if (ordered->disk_num_bytes == len)
2289 /* We cannot split once end_bio'd ordered extent */
2290 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2295 /* We cannot split a compressed ordered extent */
2296 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2301 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2302 /* bio must be in one ordered extent */
2303 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2308 /* Checksum list should be empty */
2309 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2314 pre = start - ordered->disk_bytenr;
2315 post = ordered_end - end;
2317 ret = btrfs_split_ordered_extent(ordered, pre, post);
2321 read_lock(&em_tree->lock);
2322 em = lookup_extent_mapping(em_tree, ordered->file_offset, len);
2324 read_unlock(&em_tree->lock);
2328 read_unlock(&em_tree->lock);
2330 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2332 * We cannot reuse em_new here but have to create a new one, as
2333 * unpin_extent_cache() expects the start of the extent map to be the
2334 * logical offset of the file, which does not hold true anymore after
2337 em_new = create_io_em(inode, em->start + pre, len,
2338 em->start + pre, em->block_start + pre, len,
2339 len, len, BTRFS_COMPRESS_NONE,
2340 BTRFS_ORDERED_REGULAR);
2341 if (IS_ERR(em_new)) {
2342 ret = PTR_ERR(em_new);
2345 free_extent_map(em_new);
2348 free_extent_map(em);
2349 btrfs_put_ordered_extent(ordered);
2351 return errno_to_blk_status(ret);
2355 * extent_io.c submission hook. This does the right thing for csum calculation
2356 * on write, or reading the csums from the tree before a read.
2358 * Rules about async/sync submit,
2359 * a) read: sync submit
2361 * b) write without checksum: sync submit
2363 * c) write with checksum:
2364 * c-1) if bio is issued by fsync: sync submit
2365 * (sync_writers != 0)
2367 * c-2) if root is reloc root: sync submit
2368 * (only in case of buffered IO)
2370 * c-3) otherwise: async submit
2372 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2373 int mirror_num, unsigned long bio_flags)
2376 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2377 struct btrfs_root *root = BTRFS_I(inode)->root;
2378 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2379 blk_status_t ret = 0;
2381 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2383 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2384 !fs_info->csum_root;
2386 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2387 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2389 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2390 struct page *page = bio_first_bvec_all(bio)->bv_page;
2391 loff_t file_offset = page_offset(page);
2393 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2398 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2399 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2403 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2404 ret = btrfs_submit_compressed_read(inode, bio,
2410 * Lookup bio sums does extra checks around whether we
2411 * need to csum or not, which is why we ignore skip_sum
2414 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2419 } else if (async && !skip_sum) {
2420 /* csum items have already been cloned */
2421 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2423 /* we're doing a write, do the async checksumming */
2424 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2425 0, btrfs_submit_bio_start);
2427 } else if (!skip_sum) {
2428 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2434 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2438 bio->bi_status = ret;
2445 * given a list of ordered sums record them in the inode. This happens
2446 * at IO completion time based on sums calculated at bio submission time.
2448 static int add_pending_csums(struct btrfs_trans_handle *trans,
2449 struct list_head *list)
2451 struct btrfs_ordered_sum *sum;
2454 list_for_each_entry(sum, list, list) {
2455 trans->adding_csums = true;
2456 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2457 trans->adding_csums = false;
2464 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2467 struct extent_state **cached_state)
2469 u64 search_start = start;
2470 const u64 end = start + len - 1;
2472 while (search_start < end) {
2473 const u64 search_len = end - search_start + 1;
2474 struct extent_map *em;
2478 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2482 if (em->block_start != EXTENT_MAP_HOLE)
2486 if (em->start < search_start)
2487 em_len -= search_start - em->start;
2488 if (em_len > search_len)
2489 em_len = search_len;
2491 ret = set_extent_bit(&inode->io_tree, search_start,
2492 search_start + em_len - 1,
2493 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2496 search_start = extent_map_end(em);
2497 free_extent_map(em);
2504 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2505 unsigned int extra_bits,
2506 struct extent_state **cached_state)
2508 WARN_ON(PAGE_ALIGNED(end));
2510 if (start >= i_size_read(&inode->vfs_inode) &&
2511 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2513 * There can't be any extents following eof in this case so just
2514 * set the delalloc new bit for the range directly.
2516 extra_bits |= EXTENT_DELALLOC_NEW;
2520 ret = btrfs_find_new_delalloc_bytes(inode, start,
2527 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2531 /* see btrfs_writepage_start_hook for details on why this is required */
2532 struct btrfs_writepage_fixup {
2534 struct inode *inode;
2535 struct btrfs_work work;
2538 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2540 struct btrfs_writepage_fixup *fixup;
2541 struct btrfs_ordered_extent *ordered;
2542 struct extent_state *cached_state = NULL;
2543 struct extent_changeset *data_reserved = NULL;
2545 struct btrfs_inode *inode;
2549 bool free_delalloc_space = true;
2551 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2553 inode = BTRFS_I(fixup->inode);
2554 page_start = page_offset(page);
2555 page_end = page_offset(page) + PAGE_SIZE - 1;
2558 * This is similar to page_mkwrite, we need to reserve the space before
2559 * we take the page lock.
2561 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2567 * Before we queued this fixup, we took a reference on the page.
2568 * page->mapping may go NULL, but it shouldn't be moved to a different
2571 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2573 * Unfortunately this is a little tricky, either
2575 * 1) We got here and our page had already been dealt with and
2576 * we reserved our space, thus ret == 0, so we need to just
2577 * drop our space reservation and bail. This can happen the
2578 * first time we come into the fixup worker, or could happen
2579 * while waiting for the ordered extent.
2580 * 2) Our page was already dealt with, but we happened to get an
2581 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2582 * this case we obviously don't have anything to release, but
2583 * because the page was already dealt with we don't want to
2584 * mark the page with an error, so make sure we're resetting
2585 * ret to 0. This is why we have this check _before_ the ret
2586 * check, because we do not want to have a surprise ENOSPC
2587 * when the page was already properly dealt with.
2590 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2591 btrfs_delalloc_release_space(inode, data_reserved,
2592 page_start, PAGE_SIZE,
2600 * We can't mess with the page state unless it is locked, so now that
2601 * it is locked bail if we failed to make our space reservation.
2606 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2608 /* already ordered? We're done */
2609 if (PagePrivate2(page))
2612 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2614 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2617 btrfs_start_ordered_extent(ordered, 1);
2618 btrfs_put_ordered_extent(ordered);
2622 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2628 * Everything went as planned, we're now the owner of a dirty page with
2629 * delayed allocation bits set and space reserved for our COW
2632 * The page was dirty when we started, nothing should have cleaned it.
2634 BUG_ON(!PageDirty(page));
2635 free_delalloc_space = false;
2637 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2638 if (free_delalloc_space)
2639 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2641 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2646 * We hit ENOSPC or other errors. Update the mapping and page
2647 * to reflect the errors and clean the page.
2649 mapping_set_error(page->mapping, ret);
2650 end_extent_writepage(page, ret, page_start, page_end);
2651 clear_page_dirty_for_io(page);
2654 ClearPageChecked(page);
2658 extent_changeset_free(data_reserved);
2660 * As a precaution, do a delayed iput in case it would be the last iput
2661 * that could need flushing space. Recursing back to fixup worker would
2664 btrfs_add_delayed_iput(&inode->vfs_inode);
2668 * There are a few paths in the higher layers of the kernel that directly
2669 * set the page dirty bit without asking the filesystem if it is a
2670 * good idea. This causes problems because we want to make sure COW
2671 * properly happens and the data=ordered rules are followed.
2673 * In our case any range that doesn't have the ORDERED bit set
2674 * hasn't been properly setup for IO. We kick off an async process
2675 * to fix it up. The async helper will wait for ordered extents, set
2676 * the delalloc bit and make it safe to write the page.
2678 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2680 struct inode *inode = page->mapping->host;
2681 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2682 struct btrfs_writepage_fixup *fixup;
2684 /* this page is properly in the ordered list */
2685 if (TestClearPagePrivate2(page))
2689 * PageChecked is set below when we create a fixup worker for this page,
2690 * don't try to create another one if we're already PageChecked()
2692 * The extent_io writepage code will redirty the page if we send back
2695 if (PageChecked(page))
2698 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2703 * We are already holding a reference to this inode from
2704 * write_cache_pages. We need to hold it because the space reservation
2705 * takes place outside of the page lock, and we can't trust
2706 * page->mapping outside of the page lock.
2709 SetPageChecked(page);
2711 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2713 fixup->inode = inode;
2714 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2719 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2720 struct btrfs_inode *inode, u64 file_pos,
2721 struct btrfs_file_extent_item *stack_fi,
2722 const bool update_inode_bytes,
2723 u64 qgroup_reserved)
2725 struct btrfs_root *root = inode->root;
2726 const u64 sectorsize = root->fs_info->sectorsize;
2727 struct btrfs_path *path;
2728 struct extent_buffer *leaf;
2729 struct btrfs_key ins;
2730 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2731 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2732 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2733 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2734 struct btrfs_drop_extents_args drop_args = { 0 };
2737 path = btrfs_alloc_path();
2742 * we may be replacing one extent in the tree with another.
2743 * The new extent is pinned in the extent map, and we don't want
2744 * to drop it from the cache until it is completely in the btree.
2746 * So, tell btrfs_drop_extents to leave this extent in the cache.
2747 * the caller is expected to unpin it and allow it to be merged
2750 drop_args.path = path;
2751 drop_args.start = file_pos;
2752 drop_args.end = file_pos + num_bytes;
2753 drop_args.replace_extent = true;
2754 drop_args.extent_item_size = sizeof(*stack_fi);
2755 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2759 if (!drop_args.extent_inserted) {
2760 ins.objectid = btrfs_ino(inode);
2761 ins.offset = file_pos;
2762 ins.type = BTRFS_EXTENT_DATA_KEY;
2764 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2769 leaf = path->nodes[0];
2770 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2771 write_extent_buffer(leaf, stack_fi,
2772 btrfs_item_ptr_offset(leaf, path->slots[0]),
2773 sizeof(struct btrfs_file_extent_item));
2775 btrfs_mark_buffer_dirty(leaf);
2776 btrfs_release_path(path);
2779 * If we dropped an inline extent here, we know the range where it is
2780 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2781 * number of bytes only for that range contaning the inline extent.
2782 * The remaining of the range will be processed when clearning the
2783 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2785 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2786 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2788 inline_size = drop_args.bytes_found - inline_size;
2789 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2790 drop_args.bytes_found -= inline_size;
2791 num_bytes -= sectorsize;
2794 if (update_inode_bytes)
2795 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2797 ins.objectid = disk_bytenr;
2798 ins.offset = disk_num_bytes;
2799 ins.type = BTRFS_EXTENT_ITEM_KEY;
2801 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2805 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2806 file_pos, qgroup_reserved, &ins);
2808 btrfs_free_path(path);
2813 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2816 struct btrfs_block_group *cache;
2818 cache = btrfs_lookup_block_group(fs_info, start);
2821 spin_lock(&cache->lock);
2822 cache->delalloc_bytes -= len;
2823 spin_unlock(&cache->lock);
2825 btrfs_put_block_group(cache);
2828 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2829 struct btrfs_ordered_extent *oe)
2831 struct btrfs_file_extent_item stack_fi;
2833 bool update_inode_bytes;
2835 memset(&stack_fi, 0, sizeof(stack_fi));
2836 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2837 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2838 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2839 oe->disk_num_bytes);
2840 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2841 logical_len = oe->truncated_len;
2843 logical_len = oe->num_bytes;
2844 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2845 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2846 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2847 /* Encryption and other encoding is reserved and all 0 */
2850 * For delalloc, when completing an ordered extent we update the inode's
2851 * bytes when clearing the range in the inode's io tree, so pass false
2852 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2853 * except if the ordered extent was truncated.
2855 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2856 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2858 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2859 oe->file_offset, &stack_fi,
2860 update_inode_bytes, oe->qgroup_rsv);
2864 * As ordered data IO finishes, this gets called so we can finish
2865 * an ordered extent if the range of bytes in the file it covers are
2868 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2870 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2871 struct btrfs_root *root = inode->root;
2872 struct btrfs_fs_info *fs_info = root->fs_info;
2873 struct btrfs_trans_handle *trans = NULL;
2874 struct extent_io_tree *io_tree = &inode->io_tree;
2875 struct extent_state *cached_state = NULL;
2877 int compress_type = 0;
2879 u64 logical_len = ordered_extent->num_bytes;
2880 bool freespace_inode;
2881 bool truncated = false;
2882 bool clear_reserved_extent = true;
2883 unsigned int clear_bits = EXTENT_DEFRAG;
2885 start = ordered_extent->file_offset;
2886 end = start + ordered_extent->num_bytes - 1;
2888 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2889 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2890 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2891 clear_bits |= EXTENT_DELALLOC_NEW;
2893 freespace_inode = btrfs_is_free_space_inode(inode);
2895 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2900 if (ordered_extent->disk)
2901 btrfs_rewrite_logical_zoned(ordered_extent);
2903 btrfs_free_io_failure_record(inode, start, end);
2905 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2907 logical_len = ordered_extent->truncated_len;
2908 /* Truncated the entire extent, don't bother adding */
2913 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2914 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2916 btrfs_inode_safe_disk_i_size_write(inode, 0);
2917 if (freespace_inode)
2918 trans = btrfs_join_transaction_spacecache(root);
2920 trans = btrfs_join_transaction(root);
2921 if (IS_ERR(trans)) {
2922 ret = PTR_ERR(trans);
2926 trans->block_rsv = &inode->block_rsv;
2927 ret = btrfs_update_inode_fallback(trans, root, inode);
2928 if (ret) /* -ENOMEM or corruption */
2929 btrfs_abort_transaction(trans, ret);
2933 clear_bits |= EXTENT_LOCKED;
2934 lock_extent_bits(io_tree, start, end, &cached_state);
2936 if (freespace_inode)
2937 trans = btrfs_join_transaction_spacecache(root);
2939 trans = btrfs_join_transaction(root);
2940 if (IS_ERR(trans)) {
2941 ret = PTR_ERR(trans);
2946 trans->block_rsv = &inode->block_rsv;
2948 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2949 compress_type = ordered_extent->compress_type;
2950 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2951 BUG_ON(compress_type);
2952 ret = btrfs_mark_extent_written(trans, inode,
2953 ordered_extent->file_offset,
2954 ordered_extent->file_offset +
2957 BUG_ON(root == fs_info->tree_root);
2958 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2960 clear_reserved_extent = false;
2961 btrfs_release_delalloc_bytes(fs_info,
2962 ordered_extent->disk_bytenr,
2963 ordered_extent->disk_num_bytes);
2966 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
2967 ordered_extent->num_bytes, trans->transid);
2969 btrfs_abort_transaction(trans, ret);
2973 ret = add_pending_csums(trans, &ordered_extent->list);
2975 btrfs_abort_transaction(trans, ret);
2980 * If this is a new delalloc range, clear its new delalloc flag to
2981 * update the inode's number of bytes. This needs to be done first
2982 * before updating the inode item.
2984 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
2985 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
2986 clear_extent_bit(&inode->io_tree, start, end,
2987 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
2988 0, 0, &cached_state);
2990 btrfs_inode_safe_disk_i_size_write(inode, 0);
2991 ret = btrfs_update_inode_fallback(trans, root, inode);
2992 if (ret) { /* -ENOMEM or corruption */
2993 btrfs_abort_transaction(trans, ret);
2998 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
2999 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3003 btrfs_end_transaction(trans);
3005 if (ret || truncated) {
3006 u64 unwritten_start = start;
3009 * If we failed to finish this ordered extent for any reason we
3010 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3011 * extent, and mark the inode with the error if it wasn't
3012 * already set. Any error during writeback would have already
3013 * set the mapping error, so we need to set it if we're the ones
3014 * marking this ordered extent as failed.
3016 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3017 &ordered_extent->flags))
3018 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3021 unwritten_start += logical_len;
3022 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3024 /* Drop the cache for the part of the extent we didn't write. */
3025 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3028 * If the ordered extent had an IOERR or something else went
3029 * wrong we need to return the space for this ordered extent
3030 * back to the allocator. We only free the extent in the
3031 * truncated case if we didn't write out the extent at all.
3033 * If we made it past insert_reserved_file_extent before we
3034 * errored out then we don't need to do this as the accounting
3035 * has already been done.
3037 if ((ret || !logical_len) &&
3038 clear_reserved_extent &&
3039 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3040 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3042 * Discard the range before returning it back to the
3045 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3046 btrfs_discard_extent(fs_info,
3047 ordered_extent->disk_bytenr,
3048 ordered_extent->disk_num_bytes,
3050 btrfs_free_reserved_extent(fs_info,
3051 ordered_extent->disk_bytenr,
3052 ordered_extent->disk_num_bytes, 1);
3057 * This needs to be done to make sure anybody waiting knows we are done
3058 * updating everything for this ordered extent.
3060 btrfs_remove_ordered_extent(inode, ordered_extent);
3063 btrfs_put_ordered_extent(ordered_extent);
3064 /* once for the tree */
3065 btrfs_put_ordered_extent(ordered_extent);
3070 static void finish_ordered_fn(struct btrfs_work *work)
3072 struct btrfs_ordered_extent *ordered_extent;
3073 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3074 btrfs_finish_ordered_io(ordered_extent);
3077 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
3078 u64 end, int uptodate)
3080 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3081 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3082 struct btrfs_ordered_extent *ordered_extent = NULL;
3083 struct btrfs_workqueue *wq;
3085 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3087 ClearPagePrivate2(page);
3088 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3089 end - start + 1, uptodate))
3092 if (btrfs_is_free_space_inode(inode))
3093 wq = fs_info->endio_freespace_worker;
3095 wq = fs_info->endio_write_workers;
3097 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
3098 btrfs_queue_work(wq, &ordered_extent->work);
3102 * check_data_csum - verify checksum of one sector of uncompressed data
3104 * @io_bio: btrfs_io_bio which contains the csum
3105 * @bio_offset: offset to the beginning of the bio (in bytes)
3106 * @page: page where is the data to be verified
3107 * @pgoff: offset inside the page
3108 * @start: logical offset in the file
3110 * The length of such check is always one sector size.
3112 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
3113 u32 bio_offset, struct page *page, u32 pgoff,
3116 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3117 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3119 u32 len = fs_info->sectorsize;
3120 const u32 csum_size = fs_info->csum_size;
3121 unsigned int offset_sectors;
3123 u8 csum[BTRFS_CSUM_SIZE];
3125 ASSERT(pgoff + len <= PAGE_SIZE);
3127 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3128 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
3130 kaddr = kmap_atomic(page);
3131 shash->tfm = fs_info->csum_shash;
3133 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3135 if (memcmp(csum, csum_expected, csum_size))
3138 kunmap_atomic(kaddr);
3141 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3142 io_bio->mirror_num);
3144 btrfs_dev_stat_inc_and_print(io_bio->device,
3145 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3146 memset(kaddr + pgoff, 1, len);
3147 flush_dcache_page(page);
3148 kunmap_atomic(kaddr);
3153 * When reads are done, we need to check csums to verify the data is correct.
3154 * if there's a match, we allow the bio to finish. If not, the code in
3155 * extent_io.c will try to find good copies for us.
3157 * @bio_offset: offset to the beginning of the bio (in bytes)
3158 * @start: file offset of the range start
3159 * @end: file offset of the range end (inclusive)
3161 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3162 struct page *page, u64 start, u64 end)
3164 struct inode *inode = page->mapping->host;
3165 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3166 struct btrfs_root *root = BTRFS_I(inode)->root;
3167 const u32 sectorsize = root->fs_info->sectorsize;
3170 if (PageChecked(page)) {
3171 ClearPageChecked(page);
3175 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3178 if (!root->fs_info->csum_root)
3181 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3182 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3183 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3187 ASSERT(page_offset(page) <= start &&
3188 end <= page_offset(page) + PAGE_SIZE - 1);
3189 for (pg_off = offset_in_page(start);
3190 pg_off < offset_in_page(end);
3191 pg_off += sectorsize, bio_offset += sectorsize) {
3194 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off,
3195 page_offset(page) + pg_off);
3203 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3205 * @inode: The inode we want to perform iput on
3207 * This function uses the generic vfs_inode::i_count to track whether we should
3208 * just decrement it (in case it's > 1) or if this is the last iput then link
3209 * the inode to the delayed iput machinery. Delayed iputs are processed at
3210 * transaction commit time/superblock commit/cleaner kthread.
3212 void btrfs_add_delayed_iput(struct inode *inode)
3214 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3215 struct btrfs_inode *binode = BTRFS_I(inode);
3217 if (atomic_add_unless(&inode->i_count, -1, 1))
3220 atomic_inc(&fs_info->nr_delayed_iputs);
3221 spin_lock(&fs_info->delayed_iput_lock);
3222 ASSERT(list_empty(&binode->delayed_iput));
3223 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3224 spin_unlock(&fs_info->delayed_iput_lock);
3225 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3226 wake_up_process(fs_info->cleaner_kthread);
3229 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3230 struct btrfs_inode *inode)
3232 list_del_init(&inode->delayed_iput);
3233 spin_unlock(&fs_info->delayed_iput_lock);
3234 iput(&inode->vfs_inode);
3235 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3236 wake_up(&fs_info->delayed_iputs_wait);
3237 spin_lock(&fs_info->delayed_iput_lock);
3240 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3241 struct btrfs_inode *inode)
3243 if (!list_empty(&inode->delayed_iput)) {
3244 spin_lock(&fs_info->delayed_iput_lock);
3245 if (!list_empty(&inode->delayed_iput))
3246 run_delayed_iput_locked(fs_info, inode);
3247 spin_unlock(&fs_info->delayed_iput_lock);
3251 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3254 spin_lock(&fs_info->delayed_iput_lock);
3255 while (!list_empty(&fs_info->delayed_iputs)) {
3256 struct btrfs_inode *inode;
3258 inode = list_first_entry(&fs_info->delayed_iputs,
3259 struct btrfs_inode, delayed_iput);
3260 run_delayed_iput_locked(fs_info, inode);
3261 cond_resched_lock(&fs_info->delayed_iput_lock);
3263 spin_unlock(&fs_info->delayed_iput_lock);
3267 * Wait for flushing all delayed iputs
3269 * @fs_info: the filesystem
3271 * This will wait on any delayed iputs that are currently running with KILLABLE
3272 * set. Once they are all done running we will return, unless we are killed in
3273 * which case we return EINTR. This helps in user operations like fallocate etc
3274 * that might get blocked on the iputs.
3276 * Return EINTR if we were killed, 0 if nothing's pending
3278 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3280 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3281 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3288 * This creates an orphan entry for the given inode in case something goes wrong
3289 * in the middle of an unlink.
3291 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3292 struct btrfs_inode *inode)
3296 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3297 if (ret && ret != -EEXIST) {
3298 btrfs_abort_transaction(trans, ret);
3306 * We have done the delete so we can go ahead and remove the orphan item for
3307 * this particular inode.
3309 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3310 struct btrfs_inode *inode)
3312 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3316 * this cleans up any orphans that may be left on the list from the last use
3319 int btrfs_orphan_cleanup(struct btrfs_root *root)
3321 struct btrfs_fs_info *fs_info = root->fs_info;
3322 struct btrfs_path *path;
3323 struct extent_buffer *leaf;
3324 struct btrfs_key key, found_key;
3325 struct btrfs_trans_handle *trans;
3326 struct inode *inode;
3327 u64 last_objectid = 0;
3328 int ret = 0, nr_unlink = 0;
3330 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3333 path = btrfs_alloc_path();
3338 path->reada = READA_BACK;
3340 key.objectid = BTRFS_ORPHAN_OBJECTID;
3341 key.type = BTRFS_ORPHAN_ITEM_KEY;
3342 key.offset = (u64)-1;
3345 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3350 * if ret == 0 means we found what we were searching for, which
3351 * is weird, but possible, so only screw with path if we didn't
3352 * find the key and see if we have stuff that matches
3356 if (path->slots[0] == 0)
3361 /* pull out the item */
3362 leaf = path->nodes[0];
3363 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3365 /* make sure the item matches what we want */
3366 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3368 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3371 /* release the path since we're done with it */
3372 btrfs_release_path(path);
3375 * this is where we are basically btrfs_lookup, without the
3376 * crossing root thing. we store the inode number in the
3377 * offset of the orphan item.
3380 if (found_key.offset == last_objectid) {
3382 "Error removing orphan entry, stopping orphan cleanup");
3387 last_objectid = found_key.offset;
3389 found_key.objectid = found_key.offset;
3390 found_key.type = BTRFS_INODE_ITEM_KEY;
3391 found_key.offset = 0;
3392 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3393 ret = PTR_ERR_OR_ZERO(inode);
3394 if (ret && ret != -ENOENT)
3397 if (ret == -ENOENT && root == fs_info->tree_root) {
3398 struct btrfs_root *dead_root;
3399 int is_dead_root = 0;
3402 * This is an orphan in the tree root. Currently these
3403 * could come from 2 sources:
3404 * a) a root (snapshot/subvolume) deletion in progress
3405 * b) a free space cache inode
3406 * We need to distinguish those two, as the orphan item
3407 * for a root must not get deleted before the deletion
3408 * of the snapshot/subvolume's tree completes.
3410 * btrfs_find_orphan_roots() ran before us, which has
3411 * found all deleted roots and loaded them into
3412 * fs_info->fs_roots_radix. So here we can find if an
3413 * orphan item corresponds to a deleted root by looking
3414 * up the root from that radix tree.
3417 spin_lock(&fs_info->fs_roots_radix_lock);
3418 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3419 (unsigned long)found_key.objectid);
3420 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3422 spin_unlock(&fs_info->fs_roots_radix_lock);
3425 /* prevent this orphan from being found again */
3426 key.offset = found_key.objectid - 1;
3433 * If we have an inode with links, there are a couple of
3434 * possibilities. Old kernels (before v3.12) used to create an
3435 * orphan item for truncate indicating that there were possibly
3436 * extent items past i_size that needed to be deleted. In v3.12,
3437 * truncate was changed to update i_size in sync with the extent
3438 * items, but the (useless) orphan item was still created. Since
3439 * v4.18, we don't create the orphan item for truncate at all.
3441 * So, this item could mean that we need to do a truncate, but
3442 * only if this filesystem was last used on a pre-v3.12 kernel
3443 * and was not cleanly unmounted. The odds of that are quite
3444 * slim, and it's a pain to do the truncate now, so just delete
3447 * It's also possible that this orphan item was supposed to be
3448 * deleted but wasn't. The inode number may have been reused,
3449 * but either way, we can delete the orphan item.
3451 if (ret == -ENOENT || inode->i_nlink) {
3454 trans = btrfs_start_transaction(root, 1);
3455 if (IS_ERR(trans)) {
3456 ret = PTR_ERR(trans);
3459 btrfs_debug(fs_info, "auto deleting %Lu",
3460 found_key.objectid);
3461 ret = btrfs_del_orphan_item(trans, root,
3462 found_key.objectid);
3463 btrfs_end_transaction(trans);
3471 /* this will do delete_inode and everything for us */
3474 /* release the path since we're done with it */
3475 btrfs_release_path(path);
3477 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3479 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3480 trans = btrfs_join_transaction(root);
3482 btrfs_end_transaction(trans);
3486 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3490 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3491 btrfs_free_path(path);
3496 * very simple check to peek ahead in the leaf looking for xattrs. If we
3497 * don't find any xattrs, we know there can't be any acls.
3499 * slot is the slot the inode is in, objectid is the objectid of the inode
3501 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3502 int slot, u64 objectid,
3503 int *first_xattr_slot)
3505 u32 nritems = btrfs_header_nritems(leaf);
3506 struct btrfs_key found_key;
3507 static u64 xattr_access = 0;
3508 static u64 xattr_default = 0;
3511 if (!xattr_access) {
3512 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3513 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3514 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3515 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3519 *first_xattr_slot = -1;
3520 while (slot < nritems) {
3521 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3523 /* we found a different objectid, there must not be acls */
3524 if (found_key.objectid != objectid)
3527 /* we found an xattr, assume we've got an acl */
3528 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3529 if (*first_xattr_slot == -1)
3530 *first_xattr_slot = slot;
3531 if (found_key.offset == xattr_access ||
3532 found_key.offset == xattr_default)
3537 * we found a key greater than an xattr key, there can't
3538 * be any acls later on
3540 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3547 * it goes inode, inode backrefs, xattrs, extents,
3548 * so if there are a ton of hard links to an inode there can
3549 * be a lot of backrefs. Don't waste time searching too hard,
3550 * this is just an optimization
3555 /* we hit the end of the leaf before we found an xattr or
3556 * something larger than an xattr. We have to assume the inode
3559 if (*first_xattr_slot == -1)
3560 *first_xattr_slot = slot;
3565 * read an inode from the btree into the in-memory inode
3567 static int btrfs_read_locked_inode(struct inode *inode,
3568 struct btrfs_path *in_path)
3570 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3571 struct btrfs_path *path = in_path;
3572 struct extent_buffer *leaf;
3573 struct btrfs_inode_item *inode_item;
3574 struct btrfs_root *root = BTRFS_I(inode)->root;
3575 struct btrfs_key location;
3580 bool filled = false;
3581 int first_xattr_slot;
3583 ret = btrfs_fill_inode(inode, &rdev);
3588 path = btrfs_alloc_path();
3593 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3595 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3597 if (path != in_path)
3598 btrfs_free_path(path);
3602 leaf = path->nodes[0];
3607 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3608 struct btrfs_inode_item);
3609 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3610 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3611 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3612 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3613 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3614 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3615 round_up(i_size_read(inode), fs_info->sectorsize));
3617 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3620 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3623 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3626 BTRFS_I(inode)->i_otime.tv_sec =
3627 btrfs_timespec_sec(leaf, &inode_item->otime);
3628 BTRFS_I(inode)->i_otime.tv_nsec =
3629 btrfs_timespec_nsec(leaf, &inode_item->otime);
3631 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3635 inode_set_iversion_queried(inode,
3636 btrfs_inode_sequence(leaf, inode_item));
3637 inode->i_generation = BTRFS_I(inode)->generation;
3639 rdev = btrfs_inode_rdev(leaf, inode_item);
3641 BTRFS_I(inode)->index_cnt = (u64)-1;
3642 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3646 * If we were modified in the current generation and evicted from memory
3647 * and then re-read we need to do a full sync since we don't have any
3648 * idea about which extents were modified before we were evicted from
3651 * This is required for both inode re-read from disk and delayed inode
3652 * in delayed_nodes_tree.
3654 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3655 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3656 &BTRFS_I(inode)->runtime_flags);
3659 * We don't persist the id of the transaction where an unlink operation
3660 * against the inode was last made. So here we assume the inode might
3661 * have been evicted, and therefore the exact value of last_unlink_trans
3662 * lost, and set it to last_trans to avoid metadata inconsistencies
3663 * between the inode and its parent if the inode is fsync'ed and the log
3664 * replayed. For example, in the scenario:
3667 * ln mydir/foo mydir/bar
3670 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3671 * xfs_io -c fsync mydir/foo
3673 * mount fs, triggers fsync log replay
3675 * We must make sure that when we fsync our inode foo we also log its
3676 * parent inode, otherwise after log replay the parent still has the
3677 * dentry with the "bar" name but our inode foo has a link count of 1
3678 * and doesn't have an inode ref with the name "bar" anymore.
3680 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3681 * but it guarantees correctness at the expense of occasional full
3682 * transaction commits on fsync if our inode is a directory, or if our
3683 * inode is not a directory, logging its parent unnecessarily.
3685 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3688 * Same logic as for last_unlink_trans. We don't persist the generation
3689 * of the last transaction where this inode was used for a reflink
3690 * operation, so after eviction and reloading the inode we must be
3691 * pessimistic and assume the last transaction that modified the inode.
3693 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3696 if (inode->i_nlink != 1 ||
3697 path->slots[0] >= btrfs_header_nritems(leaf))
3700 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3701 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3704 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3705 if (location.type == BTRFS_INODE_REF_KEY) {
3706 struct btrfs_inode_ref *ref;
3708 ref = (struct btrfs_inode_ref *)ptr;
3709 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3710 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3711 struct btrfs_inode_extref *extref;
3713 extref = (struct btrfs_inode_extref *)ptr;
3714 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3719 * try to precache a NULL acl entry for files that don't have
3720 * any xattrs or acls
3722 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3723 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3724 if (first_xattr_slot != -1) {
3725 path->slots[0] = first_xattr_slot;
3726 ret = btrfs_load_inode_props(inode, path);
3729 "error loading props for ino %llu (root %llu): %d",
3730 btrfs_ino(BTRFS_I(inode)),
3731 root->root_key.objectid, ret);
3733 if (path != in_path)
3734 btrfs_free_path(path);
3737 cache_no_acl(inode);
3739 switch (inode->i_mode & S_IFMT) {
3741 inode->i_mapping->a_ops = &btrfs_aops;
3742 inode->i_fop = &btrfs_file_operations;
3743 inode->i_op = &btrfs_file_inode_operations;
3746 inode->i_fop = &btrfs_dir_file_operations;
3747 inode->i_op = &btrfs_dir_inode_operations;
3750 inode->i_op = &btrfs_symlink_inode_operations;
3751 inode_nohighmem(inode);
3752 inode->i_mapping->a_ops = &btrfs_aops;
3755 inode->i_op = &btrfs_special_inode_operations;
3756 init_special_inode(inode, inode->i_mode, rdev);
3760 btrfs_sync_inode_flags_to_i_flags(inode);
3765 * given a leaf and an inode, copy the inode fields into the leaf
3767 static void fill_inode_item(struct btrfs_trans_handle *trans,
3768 struct extent_buffer *leaf,
3769 struct btrfs_inode_item *item,
3770 struct inode *inode)
3772 struct btrfs_map_token token;
3774 btrfs_init_map_token(&token, leaf);
3776 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3777 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3778 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3779 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3780 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3782 btrfs_set_token_timespec_sec(&token, &item->atime,
3783 inode->i_atime.tv_sec);
3784 btrfs_set_token_timespec_nsec(&token, &item->atime,
3785 inode->i_atime.tv_nsec);
3787 btrfs_set_token_timespec_sec(&token, &item->mtime,
3788 inode->i_mtime.tv_sec);
3789 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3790 inode->i_mtime.tv_nsec);
3792 btrfs_set_token_timespec_sec(&token, &item->ctime,
3793 inode->i_ctime.tv_sec);
3794 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3795 inode->i_ctime.tv_nsec);
3797 btrfs_set_token_timespec_sec(&token, &item->otime,
3798 BTRFS_I(inode)->i_otime.tv_sec);
3799 btrfs_set_token_timespec_nsec(&token, &item->otime,
3800 BTRFS_I(inode)->i_otime.tv_nsec);
3802 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3803 btrfs_set_token_inode_generation(&token, item,
3804 BTRFS_I(inode)->generation);
3805 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3806 btrfs_set_token_inode_transid(&token, item, trans->transid);
3807 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3808 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3809 btrfs_set_token_inode_block_group(&token, item, 0);
3813 * copy everything in the in-memory inode into the btree.
3815 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3816 struct btrfs_root *root,
3817 struct btrfs_inode *inode)
3819 struct btrfs_inode_item *inode_item;
3820 struct btrfs_path *path;
3821 struct extent_buffer *leaf;
3824 path = btrfs_alloc_path();
3828 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3835 leaf = path->nodes[0];
3836 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3837 struct btrfs_inode_item);
3839 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3840 btrfs_mark_buffer_dirty(leaf);
3841 btrfs_set_inode_last_trans(trans, inode);
3844 btrfs_free_path(path);
3849 * copy everything in the in-memory inode into the btree.
3851 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3852 struct btrfs_root *root,
3853 struct btrfs_inode *inode)
3855 struct btrfs_fs_info *fs_info = root->fs_info;
3859 * If the inode is a free space inode, we can deadlock during commit
3860 * if we put it into the delayed code.
3862 * The data relocation inode should also be directly updated
3865 if (!btrfs_is_free_space_inode(inode)
3866 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3867 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3868 btrfs_update_root_times(trans, root);
3870 ret = btrfs_delayed_update_inode(trans, root, inode);
3872 btrfs_set_inode_last_trans(trans, inode);
3876 return btrfs_update_inode_item(trans, root, inode);
3879 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3880 struct btrfs_root *root, struct btrfs_inode *inode)
3884 ret = btrfs_update_inode(trans, root, inode);
3886 return btrfs_update_inode_item(trans, root, inode);
3891 * unlink helper that gets used here in inode.c and in the tree logging
3892 * recovery code. It remove a link in a directory with a given name, and
3893 * also drops the back refs in the inode to the directory
3895 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3896 struct btrfs_root *root,
3897 struct btrfs_inode *dir,
3898 struct btrfs_inode *inode,
3899 const char *name, int name_len)
3901 struct btrfs_fs_info *fs_info = root->fs_info;
3902 struct btrfs_path *path;
3904 struct btrfs_dir_item *di;
3906 u64 ino = btrfs_ino(inode);
3907 u64 dir_ino = btrfs_ino(dir);
3909 path = btrfs_alloc_path();
3915 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3916 name, name_len, -1);
3917 if (IS_ERR_OR_NULL(di)) {
3918 ret = di ? PTR_ERR(di) : -ENOENT;
3921 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3924 btrfs_release_path(path);
3927 * If we don't have dir index, we have to get it by looking up
3928 * the inode ref, since we get the inode ref, remove it directly,
3929 * it is unnecessary to do delayed deletion.
3931 * But if we have dir index, needn't search inode ref to get it.
3932 * Since the inode ref is close to the inode item, it is better
3933 * that we delay to delete it, and just do this deletion when
3934 * we update the inode item.
3936 if (inode->dir_index) {
3937 ret = btrfs_delayed_delete_inode_ref(inode);
3939 index = inode->dir_index;
3944 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3948 "failed to delete reference to %.*s, inode %llu parent %llu",
3949 name_len, name, ino, dir_ino);
3950 btrfs_abort_transaction(trans, ret);
3954 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3956 btrfs_abort_transaction(trans, ret);
3960 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3962 if (ret != 0 && ret != -ENOENT) {
3963 btrfs_abort_transaction(trans, ret);
3967 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3972 btrfs_abort_transaction(trans, ret);
3975 * If we have a pending delayed iput we could end up with the final iput
3976 * being run in btrfs-cleaner context. If we have enough of these built
3977 * up we can end up burning a lot of time in btrfs-cleaner without any
3978 * way to throttle the unlinks. Since we're currently holding a ref on
3979 * the inode we can run the delayed iput here without any issues as the
3980 * final iput won't be done until after we drop the ref we're currently
3983 btrfs_run_delayed_iput(fs_info, inode);
3985 btrfs_free_path(path);
3989 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3990 inode_inc_iversion(&inode->vfs_inode);
3991 inode_inc_iversion(&dir->vfs_inode);
3992 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3993 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3994 ret = btrfs_update_inode(trans, root, dir);
3999 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4000 struct btrfs_root *root,
4001 struct btrfs_inode *dir, struct btrfs_inode *inode,
4002 const char *name, int name_len)
4005 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4007 drop_nlink(&inode->vfs_inode);
4008 ret = btrfs_update_inode(trans, root, inode);
4014 * helper to start transaction for unlink and rmdir.
4016 * unlink and rmdir are special in btrfs, they do not always free space, so
4017 * if we cannot make our reservations the normal way try and see if there is
4018 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4019 * allow the unlink to occur.
4021 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4023 struct btrfs_root *root = BTRFS_I(dir)->root;
4026 * 1 for the possible orphan item
4027 * 1 for the dir item
4028 * 1 for the dir index
4029 * 1 for the inode ref
4032 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4035 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4037 struct btrfs_root *root = BTRFS_I(dir)->root;
4038 struct btrfs_trans_handle *trans;
4039 struct inode *inode = d_inode(dentry);
4042 trans = __unlink_start_trans(dir);
4044 return PTR_ERR(trans);
4046 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4049 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4050 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4051 dentry->d_name.len);
4055 if (inode->i_nlink == 0) {
4056 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4062 btrfs_end_transaction(trans);
4063 btrfs_btree_balance_dirty(root->fs_info);
4067 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4068 struct inode *dir, struct dentry *dentry)
4070 struct btrfs_root *root = BTRFS_I(dir)->root;
4071 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4072 struct btrfs_path *path;
4073 struct extent_buffer *leaf;
4074 struct btrfs_dir_item *di;
4075 struct btrfs_key key;
4076 const char *name = dentry->d_name.name;
4077 int name_len = dentry->d_name.len;
4081 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4083 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4084 objectid = inode->root->root_key.objectid;
4085 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4086 objectid = inode->location.objectid;
4092 path = btrfs_alloc_path();
4096 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4097 name, name_len, -1);
4098 if (IS_ERR_OR_NULL(di)) {
4099 ret = di ? PTR_ERR(di) : -ENOENT;
4103 leaf = path->nodes[0];
4104 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4105 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4106 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4108 btrfs_abort_transaction(trans, ret);
4111 btrfs_release_path(path);
4114 * This is a placeholder inode for a subvolume we didn't have a
4115 * reference to at the time of the snapshot creation. In the meantime
4116 * we could have renamed the real subvol link into our snapshot, so
4117 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
4118 * Instead simply lookup the dir_index_item for this entry so we can
4119 * remove it. Otherwise we know we have a ref to the root and we can
4120 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4122 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4123 di = btrfs_search_dir_index_item(root, path, dir_ino,
4125 if (IS_ERR_OR_NULL(di)) {
4130 btrfs_abort_transaction(trans, ret);
4134 leaf = path->nodes[0];
4135 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4137 btrfs_release_path(path);
4139 ret = btrfs_del_root_ref(trans, objectid,
4140 root->root_key.objectid, dir_ino,
4141 &index, name, name_len);
4143 btrfs_abort_transaction(trans, ret);
4148 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4150 btrfs_abort_transaction(trans, ret);
4154 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4155 inode_inc_iversion(dir);
4156 dir->i_mtime = dir->i_ctime = current_time(dir);
4157 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4159 btrfs_abort_transaction(trans, ret);
4161 btrfs_free_path(path);
4166 * Helper to check if the subvolume references other subvolumes or if it's
4169 static noinline int may_destroy_subvol(struct btrfs_root *root)
4171 struct btrfs_fs_info *fs_info = root->fs_info;
4172 struct btrfs_path *path;
4173 struct btrfs_dir_item *di;
4174 struct btrfs_key key;
4178 path = btrfs_alloc_path();
4182 /* Make sure this root isn't set as the default subvol */
4183 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4184 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4185 dir_id, "default", 7, 0);
4186 if (di && !IS_ERR(di)) {
4187 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4188 if (key.objectid == root->root_key.objectid) {
4191 "deleting default subvolume %llu is not allowed",
4195 btrfs_release_path(path);
4198 key.objectid = root->root_key.objectid;
4199 key.type = BTRFS_ROOT_REF_KEY;
4200 key.offset = (u64)-1;
4202 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4208 if (path->slots[0] > 0) {
4210 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4211 if (key.objectid == root->root_key.objectid &&
4212 key.type == BTRFS_ROOT_REF_KEY)
4216 btrfs_free_path(path);
4220 /* Delete all dentries for inodes belonging to the root */
4221 static void btrfs_prune_dentries(struct btrfs_root *root)
4223 struct btrfs_fs_info *fs_info = root->fs_info;
4224 struct rb_node *node;
4225 struct rb_node *prev;
4226 struct btrfs_inode *entry;
4227 struct inode *inode;
4230 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4231 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4233 spin_lock(&root->inode_lock);
4235 node = root->inode_tree.rb_node;
4239 entry = rb_entry(node, struct btrfs_inode, rb_node);
4241 if (objectid < btrfs_ino(entry))
4242 node = node->rb_left;
4243 else if (objectid > btrfs_ino(entry))
4244 node = node->rb_right;
4250 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4251 if (objectid <= btrfs_ino(entry)) {
4255 prev = rb_next(prev);
4259 entry = rb_entry(node, struct btrfs_inode, rb_node);
4260 objectid = btrfs_ino(entry) + 1;
4261 inode = igrab(&entry->vfs_inode);
4263 spin_unlock(&root->inode_lock);
4264 if (atomic_read(&inode->i_count) > 1)
4265 d_prune_aliases(inode);
4267 * btrfs_drop_inode will have it removed from the inode
4268 * cache when its usage count hits zero.
4272 spin_lock(&root->inode_lock);
4276 if (cond_resched_lock(&root->inode_lock))
4279 node = rb_next(node);
4281 spin_unlock(&root->inode_lock);
4284 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4286 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4287 struct btrfs_root *root = BTRFS_I(dir)->root;
4288 struct inode *inode = d_inode(dentry);
4289 struct btrfs_root *dest = BTRFS_I(inode)->root;
4290 struct btrfs_trans_handle *trans;
4291 struct btrfs_block_rsv block_rsv;
4296 * Don't allow to delete a subvolume with send in progress. This is
4297 * inside the inode lock so the error handling that has to drop the bit
4298 * again is not run concurrently.
4300 spin_lock(&dest->root_item_lock);
4301 if (dest->send_in_progress) {
4302 spin_unlock(&dest->root_item_lock);
4304 "attempt to delete subvolume %llu during send",
4305 dest->root_key.objectid);
4308 root_flags = btrfs_root_flags(&dest->root_item);
4309 btrfs_set_root_flags(&dest->root_item,
4310 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4311 spin_unlock(&dest->root_item_lock);
4313 down_write(&fs_info->subvol_sem);
4315 ret = may_destroy_subvol(dest);
4319 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4321 * One for dir inode,
4322 * two for dir entries,
4323 * two for root ref/backref.
4325 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4329 trans = btrfs_start_transaction(root, 0);
4330 if (IS_ERR(trans)) {
4331 ret = PTR_ERR(trans);
4334 trans->block_rsv = &block_rsv;
4335 trans->bytes_reserved = block_rsv.size;
4337 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4339 ret = btrfs_unlink_subvol(trans, dir, dentry);
4341 btrfs_abort_transaction(trans, ret);
4345 ret = btrfs_record_root_in_trans(trans, dest);
4347 btrfs_abort_transaction(trans, ret);
4351 memset(&dest->root_item.drop_progress, 0,
4352 sizeof(dest->root_item.drop_progress));
4353 btrfs_set_root_drop_level(&dest->root_item, 0);
4354 btrfs_set_root_refs(&dest->root_item, 0);
4356 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4357 ret = btrfs_insert_orphan_item(trans,
4359 dest->root_key.objectid);
4361 btrfs_abort_transaction(trans, ret);
4366 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4367 BTRFS_UUID_KEY_SUBVOL,
4368 dest->root_key.objectid);
4369 if (ret && ret != -ENOENT) {
4370 btrfs_abort_transaction(trans, ret);
4373 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4374 ret = btrfs_uuid_tree_remove(trans,
4375 dest->root_item.received_uuid,
4376 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4377 dest->root_key.objectid);
4378 if (ret && ret != -ENOENT) {
4379 btrfs_abort_transaction(trans, ret);
4384 free_anon_bdev(dest->anon_dev);
4387 trans->block_rsv = NULL;
4388 trans->bytes_reserved = 0;
4389 ret = btrfs_end_transaction(trans);
4390 inode->i_flags |= S_DEAD;
4392 btrfs_subvolume_release_metadata(root, &block_rsv);
4394 up_write(&fs_info->subvol_sem);
4396 spin_lock(&dest->root_item_lock);
4397 root_flags = btrfs_root_flags(&dest->root_item);
4398 btrfs_set_root_flags(&dest->root_item,
4399 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4400 spin_unlock(&dest->root_item_lock);
4402 d_invalidate(dentry);
4403 btrfs_prune_dentries(dest);
4404 ASSERT(dest->send_in_progress == 0);
4410 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4412 struct inode *inode = d_inode(dentry);
4414 struct btrfs_root *root = BTRFS_I(dir)->root;
4415 struct btrfs_trans_handle *trans;
4416 u64 last_unlink_trans;
4418 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4420 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4421 return btrfs_delete_subvolume(dir, dentry);
4423 trans = __unlink_start_trans(dir);
4425 return PTR_ERR(trans);
4427 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4428 err = btrfs_unlink_subvol(trans, dir, dentry);
4432 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4436 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4438 /* now the directory is empty */
4439 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4440 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4441 dentry->d_name.len);
4443 btrfs_i_size_write(BTRFS_I(inode), 0);
4445 * Propagate the last_unlink_trans value of the deleted dir to
4446 * its parent directory. This is to prevent an unrecoverable
4447 * log tree in the case we do something like this:
4449 * 2) create snapshot under dir foo
4450 * 3) delete the snapshot
4453 * 6) fsync foo or some file inside foo
4455 if (last_unlink_trans >= trans->transid)
4456 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4459 btrfs_end_transaction(trans);
4460 btrfs_btree_balance_dirty(root->fs_info);
4466 * Return this if we need to call truncate_block for the last bit of the
4469 #define NEED_TRUNCATE_BLOCK 1
4472 * this can truncate away extent items, csum items and directory items.
4473 * It starts at a high offset and removes keys until it can't find
4474 * any higher than new_size
4476 * csum items that cross the new i_size are truncated to the new size
4479 * min_type is the minimum key type to truncate down to. If set to 0, this
4480 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4482 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4483 struct btrfs_root *root,
4484 struct btrfs_inode *inode,
4485 u64 new_size, u32 min_type)
4487 struct btrfs_fs_info *fs_info = root->fs_info;
4488 struct btrfs_path *path;
4489 struct extent_buffer *leaf;
4490 struct btrfs_file_extent_item *fi;
4491 struct btrfs_key key;
4492 struct btrfs_key found_key;
4493 u64 extent_start = 0;
4494 u64 extent_num_bytes = 0;
4495 u64 extent_offset = 0;
4497 u64 last_size = new_size;
4498 u32 found_type = (u8)-1;
4501 int pending_del_nr = 0;
4502 int pending_del_slot = 0;
4503 int extent_type = -1;
4505 u64 ino = btrfs_ino(inode);
4506 u64 bytes_deleted = 0;
4507 bool be_nice = false;
4508 bool should_throttle = false;
4509 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4510 struct extent_state *cached_state = NULL;
4512 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4515 * For non-free space inodes and non-shareable roots, we want to back
4516 * off from time to time. This means all inodes in subvolume roots,
4517 * reloc roots, and data reloc roots.
4519 if (!btrfs_is_free_space_inode(inode) &&
4520 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4523 path = btrfs_alloc_path();
4526 path->reada = READA_BACK;
4528 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4529 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4533 * We want to drop from the next block forward in case this
4534 * new size is not block aligned since we will be keeping the
4535 * last block of the extent just the way it is.
4537 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4538 fs_info->sectorsize),
4543 * This function is also used to drop the items in the log tree before
4544 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4545 * it is used to drop the logged items. So we shouldn't kill the delayed
4548 if (min_type == 0 && root == inode->root)
4549 btrfs_kill_delayed_inode_items(inode);
4552 key.offset = (u64)-1;
4557 * with a 16K leaf size and 128MB extents, you can actually queue
4558 * up a huge file in a single leaf. Most of the time that
4559 * bytes_deleted is > 0, it will be huge by the time we get here
4561 if (be_nice && bytes_deleted > SZ_32M &&
4562 btrfs_should_end_transaction(trans)) {
4567 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4573 /* there are no items in the tree for us to truncate, we're
4576 if (path->slots[0] == 0)
4582 u64 clear_start = 0, clear_len = 0;
4585 leaf = path->nodes[0];
4586 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4587 found_type = found_key.type;
4589 if (found_key.objectid != ino)
4592 if (found_type < min_type)
4595 item_end = found_key.offset;
4596 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4597 fi = btrfs_item_ptr(leaf, path->slots[0],
4598 struct btrfs_file_extent_item);
4599 extent_type = btrfs_file_extent_type(leaf, fi);
4600 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4602 btrfs_file_extent_num_bytes(leaf, fi);
4604 trace_btrfs_truncate_show_fi_regular(
4605 inode, leaf, fi, found_key.offset);
4606 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4607 item_end += btrfs_file_extent_ram_bytes(leaf,
4610 trace_btrfs_truncate_show_fi_inline(
4611 inode, leaf, fi, path->slots[0],
4616 if (found_type > min_type) {
4619 if (item_end < new_size)
4621 if (found_key.offset >= new_size)
4627 /* FIXME, shrink the extent if the ref count is only 1 */
4628 if (found_type != BTRFS_EXTENT_DATA_KEY)
4631 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4634 clear_start = found_key.offset;
4635 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4637 u64 orig_num_bytes =
4638 btrfs_file_extent_num_bytes(leaf, fi);
4639 extent_num_bytes = ALIGN(new_size -
4641 fs_info->sectorsize);
4642 clear_start = ALIGN(new_size, fs_info->sectorsize);
4643 btrfs_set_file_extent_num_bytes(leaf, fi,
4645 num_dec = (orig_num_bytes -
4647 if (test_bit(BTRFS_ROOT_SHAREABLE,
4650 inode_sub_bytes(&inode->vfs_inode,
4652 btrfs_mark_buffer_dirty(leaf);
4655 btrfs_file_extent_disk_num_bytes(leaf,
4657 extent_offset = found_key.offset -
4658 btrfs_file_extent_offset(leaf, fi);
4660 /* FIXME blocksize != 4096 */
4661 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4662 if (extent_start != 0) {
4664 if (test_bit(BTRFS_ROOT_SHAREABLE,
4666 inode_sub_bytes(&inode->vfs_inode,
4670 clear_len = num_dec;
4671 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4673 * we can't truncate inline items that have had
4677 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4678 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4679 btrfs_file_extent_compression(leaf, fi) == 0) {
4680 u32 size = (u32)(new_size - found_key.offset);
4682 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4683 size = btrfs_file_extent_calc_inline_size(size);
4684 btrfs_truncate_item(path, size, 1);
4685 } else if (!del_item) {
4687 * We have to bail so the last_size is set to
4688 * just before this extent.
4690 ret = NEED_TRUNCATE_BLOCK;
4694 * Inline extents are special, we just treat
4695 * them as a full sector worth in the file
4696 * extent tree just for simplicity sake.
4698 clear_len = fs_info->sectorsize;
4701 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4702 inode_sub_bytes(&inode->vfs_inode,
4703 item_end + 1 - new_size);
4707 * We use btrfs_truncate_inode_items() to clean up log trees for
4708 * multiple fsyncs, and in this case we don't want to clear the
4709 * file extent range because it's just the log.
4711 if (root == inode->root) {
4712 ret = btrfs_inode_clear_file_extent_range(inode,
4713 clear_start, clear_len);
4715 btrfs_abort_transaction(trans, ret);
4721 last_size = found_key.offset;
4723 last_size = new_size;
4725 if (!pending_del_nr) {
4726 /* no pending yet, add ourselves */
4727 pending_del_slot = path->slots[0];
4729 } else if (pending_del_nr &&
4730 path->slots[0] + 1 == pending_del_slot) {
4731 /* hop on the pending chunk */
4733 pending_del_slot = path->slots[0];
4740 should_throttle = false;
4743 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4744 struct btrfs_ref ref = { 0 };
4746 bytes_deleted += extent_num_bytes;
4748 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4749 extent_start, extent_num_bytes, 0);
4750 ref.real_root = root->root_key.objectid;
4751 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4752 ino, extent_offset);
4753 ret = btrfs_free_extent(trans, &ref);
4755 btrfs_abort_transaction(trans, ret);
4759 if (btrfs_should_throttle_delayed_refs(trans))
4760 should_throttle = true;
4764 if (found_type == BTRFS_INODE_ITEM_KEY)
4767 if (path->slots[0] == 0 ||
4768 path->slots[0] != pending_del_slot ||
4770 if (pending_del_nr) {
4771 ret = btrfs_del_items(trans, root, path,
4775 btrfs_abort_transaction(trans, ret);
4780 btrfs_release_path(path);
4783 * We can generate a lot of delayed refs, so we need to
4784 * throttle every once and a while and make sure we're
4785 * adding enough space to keep up with the work we are
4786 * generating. Since we hold a transaction here we
4787 * can't flush, and we don't want to FLUSH_LIMIT because
4788 * we could have generated too many delayed refs to
4789 * actually allocate, so just bail if we're short and
4790 * let the normal reservation dance happen higher up.
4792 if (should_throttle) {
4793 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4794 BTRFS_RESERVE_NO_FLUSH);
4806 if (ret >= 0 && pending_del_nr) {
4809 err = btrfs_del_items(trans, root, path, pending_del_slot,
4812 btrfs_abort_transaction(trans, err);
4816 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4817 ASSERT(last_size >= new_size);
4818 if (!ret && last_size > new_size)
4819 last_size = new_size;
4820 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4821 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4825 btrfs_free_path(path);
4830 * btrfs_truncate_block - read, zero a chunk and write a block
4831 * @inode - inode that we're zeroing
4832 * @from - the offset to start zeroing
4833 * @len - the length to zero, 0 to zero the entire range respective to the
4835 * @front - zero up to the offset instead of from the offset on
4837 * This will find the block for the "from" offset and cow the block and zero the
4838 * part we want to zero. This is used with truncate and hole punching.
4840 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4843 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4844 struct address_space *mapping = inode->vfs_inode.i_mapping;
4845 struct extent_io_tree *io_tree = &inode->io_tree;
4846 struct btrfs_ordered_extent *ordered;
4847 struct extent_state *cached_state = NULL;
4848 struct extent_changeset *data_reserved = NULL;
4850 bool only_release_metadata = false;
4851 u32 blocksize = fs_info->sectorsize;
4852 pgoff_t index = from >> PAGE_SHIFT;
4853 unsigned offset = from & (blocksize - 1);
4855 gfp_t mask = btrfs_alloc_write_mask(mapping);
4856 size_t write_bytes = blocksize;
4861 if (IS_ALIGNED(offset, blocksize) &&
4862 (!len || IS_ALIGNED(len, blocksize)))
4865 block_start = round_down(from, blocksize);
4866 block_end = block_start + blocksize - 1;
4868 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4871 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4872 /* For nocow case, no need to reserve data space */
4873 only_release_metadata = true;
4878 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4880 if (!only_release_metadata)
4881 btrfs_free_reserved_data_space(inode, data_reserved,
4882 block_start, blocksize);
4886 page = find_or_create_page(mapping, index, mask);
4888 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4890 btrfs_delalloc_release_extents(inode, blocksize);
4894 ret = set_page_extent_mapped(page);
4898 if (!PageUptodate(page)) {
4899 ret = btrfs_readpage(NULL, page);
4901 if (page->mapping != mapping) {
4906 if (!PageUptodate(page)) {
4911 wait_on_page_writeback(page);
4913 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4915 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4917 unlock_extent_cached(io_tree, block_start, block_end,
4921 btrfs_start_ordered_extent(ordered, 1);
4922 btrfs_put_ordered_extent(ordered);
4926 clear_extent_bit(&inode->io_tree, block_start, block_end,
4927 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4928 0, 0, &cached_state);
4930 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4933 unlock_extent_cached(io_tree, block_start, block_end,
4938 if (offset != blocksize) {
4940 len = blocksize - offset;
4943 memset(kaddr + (block_start - page_offset(page)),
4946 memset(kaddr + (block_start - page_offset(page)) + offset,
4948 flush_dcache_page(page);
4951 ClearPageChecked(page);
4952 set_page_dirty(page);
4953 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4955 if (only_release_metadata)
4956 set_extent_bit(&inode->io_tree, block_start, block_end,
4957 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4961 if (only_release_metadata)
4962 btrfs_delalloc_release_metadata(inode, blocksize, true);
4964 btrfs_delalloc_release_space(inode, data_reserved,
4965 block_start, blocksize, true);
4967 btrfs_delalloc_release_extents(inode, blocksize);
4971 if (only_release_metadata)
4972 btrfs_check_nocow_unlock(inode);
4973 extent_changeset_free(data_reserved);
4977 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4978 u64 offset, u64 len)
4980 struct btrfs_fs_info *fs_info = root->fs_info;
4981 struct btrfs_trans_handle *trans;
4982 struct btrfs_drop_extents_args drop_args = { 0 };
4986 * Still need to make sure the inode looks like it's been updated so
4987 * that any holes get logged if we fsync.
4989 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4990 inode->last_trans = fs_info->generation;
4991 inode->last_sub_trans = root->log_transid;
4992 inode->last_log_commit = root->last_log_commit;
4997 * 1 - for the one we're dropping
4998 * 1 - for the one we're adding
4999 * 1 - for updating the inode.
5001 trans = btrfs_start_transaction(root, 3);
5003 return PTR_ERR(trans);
5005 drop_args.start = offset;
5006 drop_args.end = offset + len;
5007 drop_args.drop_cache = true;
5009 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5011 btrfs_abort_transaction(trans, ret);
5012 btrfs_end_transaction(trans);
5016 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5017 offset, 0, 0, len, 0, len, 0, 0, 0);
5019 btrfs_abort_transaction(trans, ret);
5021 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5022 btrfs_update_inode(trans, root, inode);
5024 btrfs_end_transaction(trans);
5029 * This function puts in dummy file extents for the area we're creating a hole
5030 * for. So if we are truncating this file to a larger size we need to insert
5031 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5032 * the range between oldsize and size
5034 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5036 struct btrfs_root *root = inode->root;
5037 struct btrfs_fs_info *fs_info = root->fs_info;
5038 struct extent_io_tree *io_tree = &inode->io_tree;
5039 struct extent_map *em = NULL;
5040 struct extent_state *cached_state = NULL;
5041 struct extent_map_tree *em_tree = &inode->extent_tree;
5042 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5043 u64 block_end = ALIGN(size, fs_info->sectorsize);
5050 * If our size started in the middle of a block we need to zero out the
5051 * rest of the block before we expand the i_size, otherwise we could
5052 * expose stale data.
5054 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5058 if (size <= hole_start)
5061 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5063 cur_offset = hole_start;
5065 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5066 block_end - cur_offset);
5072 last_byte = min(extent_map_end(em), block_end);
5073 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5074 hole_size = last_byte - cur_offset;
5076 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5077 struct extent_map *hole_em;
5079 err = maybe_insert_hole(root, inode, cur_offset,
5084 err = btrfs_inode_set_file_extent_range(inode,
5085 cur_offset, hole_size);
5089 btrfs_drop_extent_cache(inode, cur_offset,
5090 cur_offset + hole_size - 1, 0);
5091 hole_em = alloc_extent_map();
5093 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5094 &inode->runtime_flags);
5097 hole_em->start = cur_offset;
5098 hole_em->len = hole_size;
5099 hole_em->orig_start = cur_offset;
5101 hole_em->block_start = EXTENT_MAP_HOLE;
5102 hole_em->block_len = 0;
5103 hole_em->orig_block_len = 0;
5104 hole_em->ram_bytes = hole_size;
5105 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5106 hole_em->generation = fs_info->generation;
5109 write_lock(&em_tree->lock);
5110 err = add_extent_mapping(em_tree, hole_em, 1);
5111 write_unlock(&em_tree->lock);
5114 btrfs_drop_extent_cache(inode, cur_offset,
5118 free_extent_map(hole_em);
5120 err = btrfs_inode_set_file_extent_range(inode,
5121 cur_offset, hole_size);
5126 free_extent_map(em);
5128 cur_offset = last_byte;
5129 if (cur_offset >= block_end)
5132 free_extent_map(em);
5133 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5137 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5139 struct btrfs_root *root = BTRFS_I(inode)->root;
5140 struct btrfs_trans_handle *trans;
5141 loff_t oldsize = i_size_read(inode);
5142 loff_t newsize = attr->ia_size;
5143 int mask = attr->ia_valid;
5147 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5148 * special case where we need to update the times despite not having
5149 * these flags set. For all other operations the VFS set these flags
5150 * explicitly if it wants a timestamp update.
5152 if (newsize != oldsize) {
5153 inode_inc_iversion(inode);
5154 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5155 inode->i_ctime = inode->i_mtime =
5156 current_time(inode);
5159 if (newsize > oldsize) {
5161 * Don't do an expanding truncate while snapshotting is ongoing.
5162 * This is to ensure the snapshot captures a fully consistent
5163 * state of this file - if the snapshot captures this expanding
5164 * truncation, it must capture all writes that happened before
5167 btrfs_drew_write_lock(&root->snapshot_lock);
5168 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5170 btrfs_drew_write_unlock(&root->snapshot_lock);
5174 trans = btrfs_start_transaction(root, 1);
5175 if (IS_ERR(trans)) {
5176 btrfs_drew_write_unlock(&root->snapshot_lock);
5177 return PTR_ERR(trans);
5180 i_size_write(inode, newsize);
5181 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5182 pagecache_isize_extended(inode, oldsize, newsize);
5183 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5184 btrfs_drew_write_unlock(&root->snapshot_lock);
5185 btrfs_end_transaction(trans);
5187 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5189 if (btrfs_is_zoned(fs_info)) {
5190 ret = btrfs_wait_ordered_range(inode,
5191 ALIGN(newsize, fs_info->sectorsize),
5198 * We're truncating a file that used to have good data down to
5199 * zero. Make sure any new writes to the file get on disk
5203 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5204 &BTRFS_I(inode)->runtime_flags);
5206 truncate_setsize(inode, newsize);
5208 inode_dio_wait(inode);
5210 ret = btrfs_truncate(inode, newsize == oldsize);
5211 if (ret && inode->i_nlink) {
5215 * Truncate failed, so fix up the in-memory size. We
5216 * adjusted disk_i_size down as we removed extents, so
5217 * wait for disk_i_size to be stable and then update the
5218 * in-memory size to match.
5220 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5223 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5230 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5233 struct inode *inode = d_inode(dentry);
5234 struct btrfs_root *root = BTRFS_I(inode)->root;
5237 if (btrfs_root_readonly(root))
5240 err = setattr_prepare(&init_user_ns, dentry, attr);
5244 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5245 err = btrfs_setsize(inode, attr);
5250 if (attr->ia_valid) {
5251 setattr_copy(&init_user_ns, inode, attr);
5252 inode_inc_iversion(inode);
5253 err = btrfs_dirty_inode(inode);
5255 if (!err && attr->ia_valid & ATTR_MODE)
5256 err = posix_acl_chmod(&init_user_ns, inode,
5264 * While truncating the inode pages during eviction, we get the VFS calling
5265 * btrfs_invalidatepage() against each page of the inode. This is slow because
5266 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5267 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5268 * extent_state structures over and over, wasting lots of time.
5270 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5271 * those expensive operations on a per page basis and do only the ordered io
5272 * finishing, while we release here the extent_map and extent_state structures,
5273 * without the excessive merging and splitting.
5275 static void evict_inode_truncate_pages(struct inode *inode)
5277 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5278 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5279 struct rb_node *node;
5281 ASSERT(inode->i_state & I_FREEING);
5282 truncate_inode_pages_final(&inode->i_data);
5284 write_lock(&map_tree->lock);
5285 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5286 struct extent_map *em;
5288 node = rb_first_cached(&map_tree->map);
5289 em = rb_entry(node, struct extent_map, rb_node);
5290 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5291 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5292 remove_extent_mapping(map_tree, em);
5293 free_extent_map(em);
5294 if (need_resched()) {
5295 write_unlock(&map_tree->lock);
5297 write_lock(&map_tree->lock);
5300 write_unlock(&map_tree->lock);
5303 * Keep looping until we have no more ranges in the io tree.
5304 * We can have ongoing bios started by readahead that have
5305 * their endio callback (extent_io.c:end_bio_extent_readpage)
5306 * still in progress (unlocked the pages in the bio but did not yet
5307 * unlocked the ranges in the io tree). Therefore this means some
5308 * ranges can still be locked and eviction started because before
5309 * submitting those bios, which are executed by a separate task (work
5310 * queue kthread), inode references (inode->i_count) were not taken
5311 * (which would be dropped in the end io callback of each bio).
5312 * Therefore here we effectively end up waiting for those bios and
5313 * anyone else holding locked ranges without having bumped the inode's
5314 * reference count - if we don't do it, when they access the inode's
5315 * io_tree to unlock a range it may be too late, leading to an
5316 * use-after-free issue.
5318 spin_lock(&io_tree->lock);
5319 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5320 struct extent_state *state;
5321 struct extent_state *cached_state = NULL;
5324 unsigned state_flags;
5326 node = rb_first(&io_tree->state);
5327 state = rb_entry(node, struct extent_state, rb_node);
5328 start = state->start;
5330 state_flags = state->state;
5331 spin_unlock(&io_tree->lock);
5333 lock_extent_bits(io_tree, start, end, &cached_state);
5336 * If still has DELALLOC flag, the extent didn't reach disk,
5337 * and its reserved space won't be freed by delayed_ref.
5338 * So we need to free its reserved space here.
5339 * (Refer to comment in btrfs_invalidatepage, case 2)
5341 * Note, end is the bytenr of last byte, so we need + 1 here.
5343 if (state_flags & EXTENT_DELALLOC)
5344 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5347 clear_extent_bit(io_tree, start, end,
5348 EXTENT_LOCKED | EXTENT_DELALLOC |
5349 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5353 spin_lock(&io_tree->lock);
5355 spin_unlock(&io_tree->lock);
5358 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5359 struct btrfs_block_rsv *rsv)
5361 struct btrfs_fs_info *fs_info = root->fs_info;
5362 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5363 struct btrfs_trans_handle *trans;
5364 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5368 * Eviction should be taking place at some place safe because of our
5369 * delayed iputs. However the normal flushing code will run delayed
5370 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5372 * We reserve the delayed_refs_extra here again because we can't use
5373 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5374 * above. We reserve our extra bit here because we generate a ton of
5375 * delayed refs activity by truncating.
5377 * If we cannot make our reservation we'll attempt to steal from the
5378 * global reserve, because we really want to be able to free up space.
5380 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5381 BTRFS_RESERVE_FLUSH_EVICT);
5384 * Try to steal from the global reserve if there is space for
5387 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5388 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5390 "could not allocate space for delete; will truncate on mount");
5391 return ERR_PTR(-ENOSPC);
5393 delayed_refs_extra = 0;
5396 trans = btrfs_join_transaction(root);
5400 if (delayed_refs_extra) {
5401 trans->block_rsv = &fs_info->trans_block_rsv;
5402 trans->bytes_reserved = delayed_refs_extra;
5403 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5404 delayed_refs_extra, 1);
5409 void btrfs_evict_inode(struct inode *inode)
5411 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5412 struct btrfs_trans_handle *trans;
5413 struct btrfs_root *root = BTRFS_I(inode)->root;
5414 struct btrfs_block_rsv *rsv;
5417 trace_btrfs_inode_evict(inode);
5424 evict_inode_truncate_pages(inode);
5426 if (inode->i_nlink &&
5427 ((btrfs_root_refs(&root->root_item) != 0 &&
5428 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5429 btrfs_is_free_space_inode(BTRFS_I(inode))))
5432 if (is_bad_inode(inode))
5435 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5437 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5440 if (inode->i_nlink > 0) {
5441 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5442 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5446 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5450 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5453 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5456 btrfs_i_size_write(BTRFS_I(inode), 0);
5459 trans = evict_refill_and_join(root, rsv);
5463 trans->block_rsv = rsv;
5465 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5467 trans->block_rsv = &fs_info->trans_block_rsv;
5468 btrfs_end_transaction(trans);
5469 btrfs_btree_balance_dirty(fs_info);
5470 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5477 * Errors here aren't a big deal, it just means we leave orphan items in
5478 * the tree. They will be cleaned up on the next mount. If the inode
5479 * number gets reused, cleanup deletes the orphan item without doing
5480 * anything, and unlink reuses the existing orphan item.
5482 * If it turns out that we are dropping too many of these, we might want
5483 * to add a mechanism for retrying these after a commit.
5485 trans = evict_refill_and_join(root, rsv);
5486 if (!IS_ERR(trans)) {
5487 trans->block_rsv = rsv;
5488 btrfs_orphan_del(trans, BTRFS_I(inode));
5489 trans->block_rsv = &fs_info->trans_block_rsv;
5490 btrfs_end_transaction(trans);
5494 btrfs_free_block_rsv(fs_info, rsv);
5497 * If we didn't successfully delete, the orphan item will still be in
5498 * the tree and we'll retry on the next mount. Again, we might also want
5499 * to retry these periodically in the future.
5501 btrfs_remove_delayed_node(BTRFS_I(inode));
5506 * Return the key found in the dir entry in the location pointer, fill @type
5507 * with BTRFS_FT_*, and return 0.
5509 * If no dir entries were found, returns -ENOENT.
5510 * If found a corrupted location in dir entry, returns -EUCLEAN.
5512 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5513 struct btrfs_key *location, u8 *type)
5515 const char *name = dentry->d_name.name;
5516 int namelen = dentry->d_name.len;
5517 struct btrfs_dir_item *di;
5518 struct btrfs_path *path;
5519 struct btrfs_root *root = BTRFS_I(dir)->root;
5522 path = btrfs_alloc_path();
5526 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5528 if (IS_ERR_OR_NULL(di)) {
5529 ret = di ? PTR_ERR(di) : -ENOENT;
5533 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5534 if (location->type != BTRFS_INODE_ITEM_KEY &&
5535 location->type != BTRFS_ROOT_ITEM_KEY) {
5537 btrfs_warn(root->fs_info,
5538 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5539 __func__, name, btrfs_ino(BTRFS_I(dir)),
5540 location->objectid, location->type, location->offset);
5543 *type = btrfs_dir_type(path->nodes[0], di);
5545 btrfs_free_path(path);
5550 * when we hit a tree root in a directory, the btrfs part of the inode
5551 * needs to be changed to reflect the root directory of the tree root. This
5552 * is kind of like crossing a mount point.
5554 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5556 struct dentry *dentry,
5557 struct btrfs_key *location,
5558 struct btrfs_root **sub_root)
5560 struct btrfs_path *path;
5561 struct btrfs_root *new_root;
5562 struct btrfs_root_ref *ref;
5563 struct extent_buffer *leaf;
5564 struct btrfs_key key;
5568 path = btrfs_alloc_path();
5575 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5576 key.type = BTRFS_ROOT_REF_KEY;
5577 key.offset = location->objectid;
5579 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5586 leaf = path->nodes[0];
5587 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5588 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5589 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5592 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5593 (unsigned long)(ref + 1),
5594 dentry->d_name.len);
5598 btrfs_release_path(path);
5600 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5601 if (IS_ERR(new_root)) {
5602 err = PTR_ERR(new_root);
5606 *sub_root = new_root;
5607 location->objectid = btrfs_root_dirid(&new_root->root_item);
5608 location->type = BTRFS_INODE_ITEM_KEY;
5609 location->offset = 0;
5612 btrfs_free_path(path);
5616 static void inode_tree_add(struct inode *inode)
5618 struct btrfs_root *root = BTRFS_I(inode)->root;
5619 struct btrfs_inode *entry;
5621 struct rb_node *parent;
5622 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5623 u64 ino = btrfs_ino(BTRFS_I(inode));
5625 if (inode_unhashed(inode))
5628 spin_lock(&root->inode_lock);
5629 p = &root->inode_tree.rb_node;
5632 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5634 if (ino < btrfs_ino(entry))
5635 p = &parent->rb_left;
5636 else if (ino > btrfs_ino(entry))
5637 p = &parent->rb_right;
5639 WARN_ON(!(entry->vfs_inode.i_state &
5640 (I_WILL_FREE | I_FREEING)));
5641 rb_replace_node(parent, new, &root->inode_tree);
5642 RB_CLEAR_NODE(parent);
5643 spin_unlock(&root->inode_lock);
5647 rb_link_node(new, parent, p);
5648 rb_insert_color(new, &root->inode_tree);
5649 spin_unlock(&root->inode_lock);
5652 static void inode_tree_del(struct btrfs_inode *inode)
5654 struct btrfs_root *root = inode->root;
5657 spin_lock(&root->inode_lock);
5658 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5659 rb_erase(&inode->rb_node, &root->inode_tree);
5660 RB_CLEAR_NODE(&inode->rb_node);
5661 empty = RB_EMPTY_ROOT(&root->inode_tree);
5663 spin_unlock(&root->inode_lock);
5665 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5666 spin_lock(&root->inode_lock);
5667 empty = RB_EMPTY_ROOT(&root->inode_tree);
5668 spin_unlock(&root->inode_lock);
5670 btrfs_add_dead_root(root);
5675 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5677 struct btrfs_iget_args *args = p;
5679 inode->i_ino = args->ino;
5680 BTRFS_I(inode)->location.objectid = args->ino;
5681 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5682 BTRFS_I(inode)->location.offset = 0;
5683 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5684 BUG_ON(args->root && !BTRFS_I(inode)->root);
5688 static int btrfs_find_actor(struct inode *inode, void *opaque)
5690 struct btrfs_iget_args *args = opaque;
5692 return args->ino == BTRFS_I(inode)->location.objectid &&
5693 args->root == BTRFS_I(inode)->root;
5696 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5697 struct btrfs_root *root)
5699 struct inode *inode;
5700 struct btrfs_iget_args args;
5701 unsigned long hashval = btrfs_inode_hash(ino, root);
5706 inode = iget5_locked(s, hashval, btrfs_find_actor,
5707 btrfs_init_locked_inode,
5713 * Get an inode object given its inode number and corresponding root.
5714 * Path can be preallocated to prevent recursing back to iget through
5715 * allocator. NULL is also valid but may require an additional allocation
5718 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5719 struct btrfs_root *root, struct btrfs_path *path)
5721 struct inode *inode;
5723 inode = btrfs_iget_locked(s, ino, root);
5725 return ERR_PTR(-ENOMEM);
5727 if (inode->i_state & I_NEW) {
5730 ret = btrfs_read_locked_inode(inode, path);
5732 inode_tree_add(inode);
5733 unlock_new_inode(inode);
5737 * ret > 0 can come from btrfs_search_slot called by
5738 * btrfs_read_locked_inode, this means the inode item
5743 inode = ERR_PTR(ret);
5750 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5752 return btrfs_iget_path(s, ino, root, NULL);
5755 static struct inode *new_simple_dir(struct super_block *s,
5756 struct btrfs_key *key,
5757 struct btrfs_root *root)
5759 struct inode *inode = new_inode(s);
5762 return ERR_PTR(-ENOMEM);
5764 BTRFS_I(inode)->root = btrfs_grab_root(root);
5765 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5766 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5768 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5770 * We only need lookup, the rest is read-only and there's no inode
5771 * associated with the dentry
5773 inode->i_op = &simple_dir_inode_operations;
5774 inode->i_opflags &= ~IOP_XATTR;
5775 inode->i_fop = &simple_dir_operations;
5776 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5777 inode->i_mtime = current_time(inode);
5778 inode->i_atime = inode->i_mtime;
5779 inode->i_ctime = inode->i_mtime;
5780 BTRFS_I(inode)->i_otime = inode->i_mtime;
5785 static inline u8 btrfs_inode_type(struct inode *inode)
5788 * Compile-time asserts that generic FT_* types still match
5791 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5792 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5793 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5794 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5795 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5796 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5797 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5798 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5800 return fs_umode_to_ftype(inode->i_mode);
5803 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5805 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5806 struct inode *inode;
5807 struct btrfs_root *root = BTRFS_I(dir)->root;
5808 struct btrfs_root *sub_root = root;
5809 struct btrfs_key location;
5813 if (dentry->d_name.len > BTRFS_NAME_LEN)
5814 return ERR_PTR(-ENAMETOOLONG);
5816 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5818 return ERR_PTR(ret);
5820 if (location.type == BTRFS_INODE_ITEM_KEY) {
5821 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5825 /* Do extra check against inode mode with di_type */
5826 if (btrfs_inode_type(inode) != di_type) {
5828 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5829 inode->i_mode, btrfs_inode_type(inode),
5832 return ERR_PTR(-EUCLEAN);
5837 ret = fixup_tree_root_location(fs_info, dir, dentry,
5838 &location, &sub_root);
5841 inode = ERR_PTR(ret);
5843 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5845 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5847 if (root != sub_root)
5848 btrfs_put_root(sub_root);
5850 if (!IS_ERR(inode) && root != sub_root) {
5851 down_read(&fs_info->cleanup_work_sem);
5852 if (!sb_rdonly(inode->i_sb))
5853 ret = btrfs_orphan_cleanup(sub_root);
5854 up_read(&fs_info->cleanup_work_sem);
5857 inode = ERR_PTR(ret);
5864 static int btrfs_dentry_delete(const struct dentry *dentry)
5866 struct btrfs_root *root;
5867 struct inode *inode = d_inode(dentry);
5869 if (!inode && !IS_ROOT(dentry))
5870 inode = d_inode(dentry->d_parent);
5873 root = BTRFS_I(inode)->root;
5874 if (btrfs_root_refs(&root->root_item) == 0)
5877 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5883 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5886 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5888 if (inode == ERR_PTR(-ENOENT))
5890 return d_splice_alias(inode, dentry);
5894 * All this infrastructure exists because dir_emit can fault, and we are holding
5895 * the tree lock when doing readdir. For now just allocate a buffer and copy
5896 * our information into that, and then dir_emit from the buffer. This is
5897 * similar to what NFS does, only we don't keep the buffer around in pagecache
5898 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5899 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5902 static int btrfs_opendir(struct inode *inode, struct file *file)
5904 struct btrfs_file_private *private;
5906 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5909 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5910 if (!private->filldir_buf) {
5914 file->private_data = private;
5925 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5928 struct dir_entry *entry = addr;
5929 char *name = (char *)(entry + 1);
5931 ctx->pos = get_unaligned(&entry->offset);
5932 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5933 get_unaligned(&entry->ino),
5934 get_unaligned(&entry->type)))
5936 addr += sizeof(struct dir_entry) +
5937 get_unaligned(&entry->name_len);
5943 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5945 struct inode *inode = file_inode(file);
5946 struct btrfs_root *root = BTRFS_I(inode)->root;
5947 struct btrfs_file_private *private = file->private_data;
5948 struct btrfs_dir_item *di;
5949 struct btrfs_key key;
5950 struct btrfs_key found_key;
5951 struct btrfs_path *path;
5953 struct list_head ins_list;
5954 struct list_head del_list;
5956 struct extent_buffer *leaf;
5963 struct btrfs_key location;
5965 if (!dir_emit_dots(file, ctx))
5968 path = btrfs_alloc_path();
5972 addr = private->filldir_buf;
5973 path->reada = READA_FORWARD;
5975 INIT_LIST_HEAD(&ins_list);
5976 INIT_LIST_HEAD(&del_list);
5977 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5980 key.type = BTRFS_DIR_INDEX_KEY;
5981 key.offset = ctx->pos;
5982 key.objectid = btrfs_ino(BTRFS_I(inode));
5984 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5989 struct dir_entry *entry;
5991 leaf = path->nodes[0];
5992 slot = path->slots[0];
5993 if (slot >= btrfs_header_nritems(leaf)) {
5994 ret = btrfs_next_leaf(root, path);
6002 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6004 if (found_key.objectid != key.objectid)
6006 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6008 if (found_key.offset < ctx->pos)
6010 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6012 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6013 name_len = btrfs_dir_name_len(leaf, di);
6014 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6016 btrfs_release_path(path);
6017 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6020 addr = private->filldir_buf;
6027 put_unaligned(name_len, &entry->name_len);
6028 name_ptr = (char *)(entry + 1);
6029 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6031 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6033 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6034 put_unaligned(location.objectid, &entry->ino);
6035 put_unaligned(found_key.offset, &entry->offset);
6037 addr += sizeof(struct dir_entry) + name_len;
6038 total_len += sizeof(struct dir_entry) + name_len;
6042 btrfs_release_path(path);
6044 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6048 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6053 * Stop new entries from being returned after we return the last
6056 * New directory entries are assigned a strictly increasing
6057 * offset. This means that new entries created during readdir
6058 * are *guaranteed* to be seen in the future by that readdir.
6059 * This has broken buggy programs which operate on names as
6060 * they're returned by readdir. Until we re-use freed offsets
6061 * we have this hack to stop new entries from being returned
6062 * under the assumption that they'll never reach this huge
6065 * This is being careful not to overflow 32bit loff_t unless the
6066 * last entry requires it because doing so has broken 32bit apps
6069 if (ctx->pos >= INT_MAX)
6070 ctx->pos = LLONG_MAX;
6077 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6078 btrfs_free_path(path);
6083 * This is somewhat expensive, updating the tree every time the
6084 * inode changes. But, it is most likely to find the inode in cache.
6085 * FIXME, needs more benchmarking...there are no reasons other than performance
6086 * to keep or drop this code.
6088 static int btrfs_dirty_inode(struct inode *inode)
6090 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6091 struct btrfs_root *root = BTRFS_I(inode)->root;
6092 struct btrfs_trans_handle *trans;
6095 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6098 trans = btrfs_join_transaction(root);
6100 return PTR_ERR(trans);
6102 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6103 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6104 /* whoops, lets try again with the full transaction */
6105 btrfs_end_transaction(trans);
6106 trans = btrfs_start_transaction(root, 1);
6108 return PTR_ERR(trans);
6110 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6112 btrfs_end_transaction(trans);
6113 if (BTRFS_I(inode)->delayed_node)
6114 btrfs_balance_delayed_items(fs_info);
6120 * This is a copy of file_update_time. We need this so we can return error on
6121 * ENOSPC for updating the inode in the case of file write and mmap writes.
6123 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6126 struct btrfs_root *root = BTRFS_I(inode)->root;
6127 bool dirty = flags & ~S_VERSION;
6129 if (btrfs_root_readonly(root))
6132 if (flags & S_VERSION)
6133 dirty |= inode_maybe_inc_iversion(inode, dirty);
6134 if (flags & S_CTIME)
6135 inode->i_ctime = *now;
6136 if (flags & S_MTIME)
6137 inode->i_mtime = *now;
6138 if (flags & S_ATIME)
6139 inode->i_atime = *now;
6140 return dirty ? btrfs_dirty_inode(inode) : 0;
6144 * find the highest existing sequence number in a directory
6145 * and then set the in-memory index_cnt variable to reflect
6146 * free sequence numbers
6148 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6150 struct btrfs_root *root = inode->root;
6151 struct btrfs_key key, found_key;
6152 struct btrfs_path *path;
6153 struct extent_buffer *leaf;
6156 key.objectid = btrfs_ino(inode);
6157 key.type = BTRFS_DIR_INDEX_KEY;
6158 key.offset = (u64)-1;
6160 path = btrfs_alloc_path();
6164 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6167 /* FIXME: we should be able to handle this */
6173 * MAGIC NUMBER EXPLANATION:
6174 * since we search a directory based on f_pos we have to start at 2
6175 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6176 * else has to start at 2
6178 if (path->slots[0] == 0) {
6179 inode->index_cnt = 2;
6185 leaf = path->nodes[0];
6186 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6188 if (found_key.objectid != btrfs_ino(inode) ||
6189 found_key.type != BTRFS_DIR_INDEX_KEY) {
6190 inode->index_cnt = 2;
6194 inode->index_cnt = found_key.offset + 1;
6196 btrfs_free_path(path);
6201 * helper to find a free sequence number in a given directory. This current
6202 * code is very simple, later versions will do smarter things in the btree
6204 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6208 if (dir->index_cnt == (u64)-1) {
6209 ret = btrfs_inode_delayed_dir_index_count(dir);
6211 ret = btrfs_set_inode_index_count(dir);
6217 *index = dir->index_cnt;
6223 static int btrfs_insert_inode_locked(struct inode *inode)
6225 struct btrfs_iget_args args;
6227 args.ino = BTRFS_I(inode)->location.objectid;
6228 args.root = BTRFS_I(inode)->root;
6230 return insert_inode_locked4(inode,
6231 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6232 btrfs_find_actor, &args);
6236 * Inherit flags from the parent inode.
6238 * Currently only the compression flags and the cow flags are inherited.
6240 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6247 flags = BTRFS_I(dir)->flags;
6249 if (flags & BTRFS_INODE_NOCOMPRESS) {
6250 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6251 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6252 } else if (flags & BTRFS_INODE_COMPRESS) {
6253 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6254 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6257 if (flags & BTRFS_INODE_NODATACOW) {
6258 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6259 if (S_ISREG(inode->i_mode))
6260 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6263 btrfs_sync_inode_flags_to_i_flags(inode);
6266 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6267 struct btrfs_root *root,
6269 const char *name, int name_len,
6270 u64 ref_objectid, u64 objectid,
6271 umode_t mode, u64 *index)
6273 struct btrfs_fs_info *fs_info = root->fs_info;
6274 struct inode *inode;
6275 struct btrfs_inode_item *inode_item;
6276 struct btrfs_key *location;
6277 struct btrfs_path *path;
6278 struct btrfs_inode_ref *ref;
6279 struct btrfs_key key[2];
6281 int nitems = name ? 2 : 1;
6283 unsigned int nofs_flag;
6286 path = btrfs_alloc_path();
6288 return ERR_PTR(-ENOMEM);
6290 nofs_flag = memalloc_nofs_save();
6291 inode = new_inode(fs_info->sb);
6292 memalloc_nofs_restore(nofs_flag);
6294 btrfs_free_path(path);
6295 return ERR_PTR(-ENOMEM);
6299 * O_TMPFILE, set link count to 0, so that after this point,
6300 * we fill in an inode item with the correct link count.
6303 set_nlink(inode, 0);
6306 * we have to initialize this early, so we can reclaim the inode
6307 * number if we fail afterwards in this function.
6309 inode->i_ino = objectid;
6312 trace_btrfs_inode_request(dir);
6314 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6316 btrfs_free_path(path);
6318 return ERR_PTR(ret);
6324 * index_cnt is ignored for everything but a dir,
6325 * btrfs_set_inode_index_count has an explanation for the magic
6328 BTRFS_I(inode)->index_cnt = 2;
6329 BTRFS_I(inode)->dir_index = *index;
6330 BTRFS_I(inode)->root = btrfs_grab_root(root);
6331 BTRFS_I(inode)->generation = trans->transid;
6332 inode->i_generation = BTRFS_I(inode)->generation;
6335 * We could have gotten an inode number from somebody who was fsynced
6336 * and then removed in this same transaction, so let's just set full
6337 * sync since it will be a full sync anyway and this will blow away the
6338 * old info in the log.
6340 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6342 key[0].objectid = objectid;
6343 key[0].type = BTRFS_INODE_ITEM_KEY;
6346 sizes[0] = sizeof(struct btrfs_inode_item);
6350 * Start new inodes with an inode_ref. This is slightly more
6351 * efficient for small numbers of hard links since they will
6352 * be packed into one item. Extended refs will kick in if we
6353 * add more hard links than can fit in the ref item.
6355 key[1].objectid = objectid;
6356 key[1].type = BTRFS_INODE_REF_KEY;
6357 key[1].offset = ref_objectid;
6359 sizes[1] = name_len + sizeof(*ref);
6362 location = &BTRFS_I(inode)->location;
6363 location->objectid = objectid;
6364 location->offset = 0;
6365 location->type = BTRFS_INODE_ITEM_KEY;
6367 ret = btrfs_insert_inode_locked(inode);
6373 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6377 inode_init_owner(&init_user_ns, inode, dir, mode);
6378 inode_set_bytes(inode, 0);
6380 inode->i_mtime = current_time(inode);
6381 inode->i_atime = inode->i_mtime;
6382 inode->i_ctime = inode->i_mtime;
6383 BTRFS_I(inode)->i_otime = inode->i_mtime;
6385 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6386 struct btrfs_inode_item);
6387 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6388 sizeof(*inode_item));
6389 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6392 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6393 struct btrfs_inode_ref);
6394 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6395 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6396 ptr = (unsigned long)(ref + 1);
6397 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6400 btrfs_mark_buffer_dirty(path->nodes[0]);
6401 btrfs_free_path(path);
6403 btrfs_inherit_iflags(inode, dir);
6405 if (S_ISREG(mode)) {
6406 if (btrfs_test_opt(fs_info, NODATASUM))
6407 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6408 if (btrfs_test_opt(fs_info, NODATACOW))
6409 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6410 BTRFS_INODE_NODATASUM;
6413 inode_tree_add(inode);
6415 trace_btrfs_inode_new(inode);
6416 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6418 btrfs_update_root_times(trans, root);
6420 ret = btrfs_inode_inherit_props(trans, inode, dir);
6423 "error inheriting props for ino %llu (root %llu): %d",
6424 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6429 discard_new_inode(inode);
6432 BTRFS_I(dir)->index_cnt--;
6433 btrfs_free_path(path);
6434 return ERR_PTR(ret);
6438 * utility function to add 'inode' into 'parent_inode' with
6439 * a give name and a given sequence number.
6440 * if 'add_backref' is true, also insert a backref from the
6441 * inode to the parent directory.
6443 int btrfs_add_link(struct btrfs_trans_handle *trans,
6444 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6445 const char *name, int name_len, int add_backref, u64 index)
6448 struct btrfs_key key;
6449 struct btrfs_root *root = parent_inode->root;
6450 u64 ino = btrfs_ino(inode);
6451 u64 parent_ino = btrfs_ino(parent_inode);
6453 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6454 memcpy(&key, &inode->root->root_key, sizeof(key));
6457 key.type = BTRFS_INODE_ITEM_KEY;
6461 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6462 ret = btrfs_add_root_ref(trans, key.objectid,
6463 root->root_key.objectid, parent_ino,
6464 index, name, name_len);
6465 } else if (add_backref) {
6466 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6470 /* Nothing to clean up yet */
6474 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6475 btrfs_inode_type(&inode->vfs_inode), index);
6476 if (ret == -EEXIST || ret == -EOVERFLOW)
6479 btrfs_abort_transaction(trans, ret);
6483 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6485 inode_inc_iversion(&parent_inode->vfs_inode);
6487 * If we are replaying a log tree, we do not want to update the mtime
6488 * and ctime of the parent directory with the current time, since the
6489 * log replay procedure is responsible for setting them to their correct
6490 * values (the ones it had when the fsync was done).
6492 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6493 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6495 parent_inode->vfs_inode.i_mtime = now;
6496 parent_inode->vfs_inode.i_ctime = now;
6498 ret = btrfs_update_inode(trans, root, parent_inode);
6500 btrfs_abort_transaction(trans, ret);
6504 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6507 err = btrfs_del_root_ref(trans, key.objectid,
6508 root->root_key.objectid, parent_ino,
6509 &local_index, name, name_len);
6511 btrfs_abort_transaction(trans, err);
6512 } else if (add_backref) {
6516 err = btrfs_del_inode_ref(trans, root, name, name_len,
6517 ino, parent_ino, &local_index);
6519 btrfs_abort_transaction(trans, err);
6522 /* Return the original error code */
6526 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6527 struct btrfs_inode *dir, struct dentry *dentry,
6528 struct btrfs_inode *inode, int backref, u64 index)
6530 int err = btrfs_add_link(trans, dir, inode,
6531 dentry->d_name.name, dentry->d_name.len,
6538 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6539 struct dentry *dentry, umode_t mode, dev_t rdev)
6541 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6542 struct btrfs_trans_handle *trans;
6543 struct btrfs_root *root = BTRFS_I(dir)->root;
6544 struct inode *inode = NULL;
6550 * 2 for inode item and ref
6552 * 1 for xattr if selinux is on
6554 trans = btrfs_start_transaction(root, 5);
6556 return PTR_ERR(trans);
6558 err = btrfs_get_free_objectid(root, &objectid);
6562 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6563 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6565 if (IS_ERR(inode)) {
6566 err = PTR_ERR(inode);
6572 * If the active LSM wants to access the inode during
6573 * d_instantiate it needs these. Smack checks to see
6574 * if the filesystem supports xattrs by looking at the
6577 inode->i_op = &btrfs_special_inode_operations;
6578 init_special_inode(inode, inode->i_mode, rdev);
6580 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6584 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6589 btrfs_update_inode(trans, root, BTRFS_I(inode));
6590 d_instantiate_new(dentry, inode);
6593 btrfs_end_transaction(trans);
6594 btrfs_btree_balance_dirty(fs_info);
6596 inode_dec_link_count(inode);
6597 discard_new_inode(inode);
6602 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6603 struct dentry *dentry, umode_t mode, bool excl)
6605 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6606 struct btrfs_trans_handle *trans;
6607 struct btrfs_root *root = BTRFS_I(dir)->root;
6608 struct inode *inode = NULL;
6614 * 2 for inode item and ref
6616 * 1 for xattr if selinux is on
6618 trans = btrfs_start_transaction(root, 5);
6620 return PTR_ERR(trans);
6622 err = btrfs_get_free_objectid(root, &objectid);
6626 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6627 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6629 if (IS_ERR(inode)) {
6630 err = PTR_ERR(inode);
6635 * If the active LSM wants to access the inode during
6636 * d_instantiate it needs these. Smack checks to see
6637 * if the filesystem supports xattrs by looking at the
6640 inode->i_fop = &btrfs_file_operations;
6641 inode->i_op = &btrfs_file_inode_operations;
6642 inode->i_mapping->a_ops = &btrfs_aops;
6644 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6648 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6652 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6657 d_instantiate_new(dentry, inode);
6660 btrfs_end_transaction(trans);
6662 inode_dec_link_count(inode);
6663 discard_new_inode(inode);
6665 btrfs_btree_balance_dirty(fs_info);
6669 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6670 struct dentry *dentry)
6672 struct btrfs_trans_handle *trans = NULL;
6673 struct btrfs_root *root = BTRFS_I(dir)->root;
6674 struct inode *inode = d_inode(old_dentry);
6675 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6680 /* do not allow sys_link's with other subvols of the same device */
6681 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6684 if (inode->i_nlink >= BTRFS_LINK_MAX)
6687 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6692 * 2 items for inode and inode ref
6693 * 2 items for dir items
6694 * 1 item for parent inode
6695 * 1 item for orphan item deletion if O_TMPFILE
6697 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6698 if (IS_ERR(trans)) {
6699 err = PTR_ERR(trans);
6704 /* There are several dir indexes for this inode, clear the cache. */
6705 BTRFS_I(inode)->dir_index = 0ULL;
6707 inode_inc_iversion(inode);
6708 inode->i_ctime = current_time(inode);
6710 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6712 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6718 struct dentry *parent = dentry->d_parent;
6720 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6723 if (inode->i_nlink == 1) {
6725 * If new hard link count is 1, it's a file created
6726 * with open(2) O_TMPFILE flag.
6728 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6732 d_instantiate(dentry, inode);
6733 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6738 btrfs_end_transaction(trans);
6740 inode_dec_link_count(inode);
6743 btrfs_btree_balance_dirty(fs_info);
6747 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6748 struct dentry *dentry, umode_t mode)
6750 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6751 struct inode *inode = NULL;
6752 struct btrfs_trans_handle *trans;
6753 struct btrfs_root *root = BTRFS_I(dir)->root;
6759 * 2 items for inode and ref
6760 * 2 items for dir items
6761 * 1 for xattr if selinux is on
6763 trans = btrfs_start_transaction(root, 5);
6765 return PTR_ERR(trans);
6767 err = btrfs_get_free_objectid(root, &objectid);
6771 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6772 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6773 S_IFDIR | mode, &index);
6774 if (IS_ERR(inode)) {
6775 err = PTR_ERR(inode);
6780 /* these must be set before we unlock the inode */
6781 inode->i_op = &btrfs_dir_inode_operations;
6782 inode->i_fop = &btrfs_dir_file_operations;
6784 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6788 btrfs_i_size_write(BTRFS_I(inode), 0);
6789 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6793 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6794 dentry->d_name.name,
6795 dentry->d_name.len, 0, index);
6799 d_instantiate_new(dentry, inode);
6802 btrfs_end_transaction(trans);
6804 inode_dec_link_count(inode);
6805 discard_new_inode(inode);
6807 btrfs_btree_balance_dirty(fs_info);
6811 static noinline int uncompress_inline(struct btrfs_path *path,
6813 size_t pg_offset, u64 extent_offset,
6814 struct btrfs_file_extent_item *item)
6817 struct extent_buffer *leaf = path->nodes[0];
6820 unsigned long inline_size;
6824 WARN_ON(pg_offset != 0);
6825 compress_type = btrfs_file_extent_compression(leaf, item);
6826 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6827 inline_size = btrfs_file_extent_inline_item_len(leaf,
6828 btrfs_item_nr(path->slots[0]));
6829 tmp = kmalloc(inline_size, GFP_NOFS);
6832 ptr = btrfs_file_extent_inline_start(item);
6834 read_extent_buffer(leaf, tmp, ptr, inline_size);
6836 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6837 ret = btrfs_decompress(compress_type, tmp, page,
6838 extent_offset, inline_size, max_size);
6841 * decompression code contains a memset to fill in any space between the end
6842 * of the uncompressed data and the end of max_size in case the decompressed
6843 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6844 * the end of an inline extent and the beginning of the next block, so we
6845 * cover that region here.
6848 if (max_size + pg_offset < PAGE_SIZE) {
6849 char *map = kmap(page);
6850 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6858 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6859 * @inode: file to search in
6860 * @page: page to read extent data into if the extent is inline
6861 * @pg_offset: offset into @page to copy to
6862 * @start: file offset
6863 * @len: length of range starting at @start
6865 * This returns the first &struct extent_map which overlaps with the given
6866 * range, reading it from the B-tree and caching it if necessary. Note that
6867 * there may be more extents which overlap the given range after the returned
6870 * If @page is not NULL and the extent is inline, this also reads the extent
6871 * data directly into the page and marks the extent up to date in the io_tree.
6873 * Return: ERR_PTR on error, non-NULL extent_map on success.
6875 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6876 struct page *page, size_t pg_offset,
6879 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6881 u64 extent_start = 0;
6883 u64 objectid = btrfs_ino(inode);
6884 int extent_type = -1;
6885 struct btrfs_path *path = NULL;
6886 struct btrfs_root *root = inode->root;
6887 struct btrfs_file_extent_item *item;
6888 struct extent_buffer *leaf;
6889 struct btrfs_key found_key;
6890 struct extent_map *em = NULL;
6891 struct extent_map_tree *em_tree = &inode->extent_tree;
6892 struct extent_io_tree *io_tree = &inode->io_tree;
6894 read_lock(&em_tree->lock);
6895 em = lookup_extent_mapping(em_tree, start, len);
6896 read_unlock(&em_tree->lock);
6899 if (em->start > start || em->start + em->len <= start)
6900 free_extent_map(em);
6901 else if (em->block_start == EXTENT_MAP_INLINE && page)
6902 free_extent_map(em);
6906 em = alloc_extent_map();
6911 em->start = EXTENT_MAP_HOLE;
6912 em->orig_start = EXTENT_MAP_HOLE;
6914 em->block_len = (u64)-1;
6916 path = btrfs_alloc_path();
6922 /* Chances are we'll be called again, so go ahead and do readahead */
6923 path->reada = READA_FORWARD;
6926 * The same explanation in load_free_space_cache applies here as well,
6927 * we only read when we're loading the free space cache, and at that
6928 * point the commit_root has everything we need.
6930 if (btrfs_is_free_space_inode(inode)) {
6931 path->search_commit_root = 1;
6932 path->skip_locking = 1;
6935 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6938 } else if (ret > 0) {
6939 if (path->slots[0] == 0)
6945 leaf = path->nodes[0];
6946 item = btrfs_item_ptr(leaf, path->slots[0],
6947 struct btrfs_file_extent_item);
6948 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6949 if (found_key.objectid != objectid ||
6950 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6952 * If we backup past the first extent we want to move forward
6953 * and see if there is an extent in front of us, otherwise we'll
6954 * say there is a hole for our whole search range which can
6961 extent_type = btrfs_file_extent_type(leaf, item);
6962 extent_start = found_key.offset;
6963 extent_end = btrfs_file_extent_end(path);
6964 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6965 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6966 /* Only regular file could have regular/prealloc extent */
6967 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6970 "regular/prealloc extent found for non-regular inode %llu",
6974 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6976 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6977 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6982 if (start >= extent_end) {
6984 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6985 ret = btrfs_next_leaf(root, path);
6991 leaf = path->nodes[0];
6993 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6994 if (found_key.objectid != objectid ||
6995 found_key.type != BTRFS_EXTENT_DATA_KEY)
6997 if (start + len <= found_key.offset)
6999 if (start > found_key.offset)
7002 /* New extent overlaps with existing one */
7004 em->orig_start = start;
7005 em->len = found_key.offset - start;
7006 em->block_start = EXTENT_MAP_HOLE;
7010 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
7012 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7013 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7015 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7019 size_t extent_offset;
7025 size = btrfs_file_extent_ram_bytes(leaf, item);
7026 extent_offset = page_offset(page) + pg_offset - extent_start;
7027 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7028 size - extent_offset);
7029 em->start = extent_start + extent_offset;
7030 em->len = ALIGN(copy_size, fs_info->sectorsize);
7031 em->orig_block_len = em->len;
7032 em->orig_start = em->start;
7033 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7035 if (!PageUptodate(page)) {
7036 if (btrfs_file_extent_compression(leaf, item) !=
7037 BTRFS_COMPRESS_NONE) {
7038 ret = uncompress_inline(path, page, pg_offset,
7039 extent_offset, item);
7043 map = kmap_local_page(page);
7044 read_extent_buffer(leaf, map + pg_offset, ptr,
7046 if (pg_offset + copy_size < PAGE_SIZE) {
7047 memset(map + pg_offset + copy_size, 0,
7048 PAGE_SIZE - pg_offset -
7053 flush_dcache_page(page);
7055 set_extent_uptodate(io_tree, em->start,
7056 extent_map_end(em) - 1, NULL, GFP_NOFS);
7061 em->orig_start = start;
7063 em->block_start = EXTENT_MAP_HOLE;
7066 btrfs_release_path(path);
7067 if (em->start > start || extent_map_end(em) <= start) {
7069 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7070 em->start, em->len, start, len);
7075 write_lock(&em_tree->lock);
7076 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7077 write_unlock(&em_tree->lock);
7079 btrfs_free_path(path);
7081 trace_btrfs_get_extent(root, inode, em);
7084 free_extent_map(em);
7085 return ERR_PTR(ret);
7090 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7093 struct extent_map *em;
7094 struct extent_map *hole_em = NULL;
7095 u64 delalloc_start = start;
7101 em = btrfs_get_extent(inode, NULL, 0, start, len);
7105 * If our em maps to:
7107 * - a pre-alloc extent,
7108 * there might actually be delalloc bytes behind it.
7110 if (em->block_start != EXTENT_MAP_HOLE &&
7111 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7116 /* check to see if we've wrapped (len == -1 or similar) */
7125 /* ok, we didn't find anything, lets look for delalloc */
7126 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7127 end, len, EXTENT_DELALLOC, 1);
7128 delalloc_end = delalloc_start + delalloc_len;
7129 if (delalloc_end < delalloc_start)
7130 delalloc_end = (u64)-1;
7133 * We didn't find anything useful, return the original results from
7136 if (delalloc_start > end || delalloc_end <= start) {
7143 * Adjust the delalloc_start to make sure it doesn't go backwards from
7144 * the start they passed in
7146 delalloc_start = max(start, delalloc_start);
7147 delalloc_len = delalloc_end - delalloc_start;
7149 if (delalloc_len > 0) {
7152 const u64 hole_end = extent_map_end(hole_em);
7154 em = alloc_extent_map();
7162 * When btrfs_get_extent can't find anything it returns one
7165 * Make sure what it found really fits our range, and adjust to
7166 * make sure it is based on the start from the caller
7168 if (hole_end <= start || hole_em->start > end) {
7169 free_extent_map(hole_em);
7172 hole_start = max(hole_em->start, start);
7173 hole_len = hole_end - hole_start;
7176 if (hole_em && delalloc_start > hole_start) {
7178 * Our hole starts before our delalloc, so we have to
7179 * return just the parts of the hole that go until the
7182 em->len = min(hole_len, delalloc_start - hole_start);
7183 em->start = hole_start;
7184 em->orig_start = hole_start;
7186 * Don't adjust block start at all, it is fixed at
7189 em->block_start = hole_em->block_start;
7190 em->block_len = hole_len;
7191 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7192 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7195 * Hole is out of passed range or it starts after
7198 em->start = delalloc_start;
7199 em->len = delalloc_len;
7200 em->orig_start = delalloc_start;
7201 em->block_start = EXTENT_MAP_DELALLOC;
7202 em->block_len = delalloc_len;
7209 free_extent_map(hole_em);
7211 free_extent_map(em);
7212 return ERR_PTR(err);
7217 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7220 const u64 orig_start,
7221 const u64 block_start,
7222 const u64 block_len,
7223 const u64 orig_block_len,
7224 const u64 ram_bytes,
7227 struct extent_map *em = NULL;
7230 if (type != BTRFS_ORDERED_NOCOW) {
7231 em = create_io_em(inode, start, len, orig_start, block_start,
7232 block_len, orig_block_len, ram_bytes,
7233 BTRFS_COMPRESS_NONE, /* compress_type */
7238 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7242 free_extent_map(em);
7243 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7252 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7255 struct btrfs_root *root = inode->root;
7256 struct btrfs_fs_info *fs_info = root->fs_info;
7257 struct extent_map *em;
7258 struct btrfs_key ins;
7262 alloc_hint = get_extent_allocation_hint(inode, start, len);
7263 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7264 0, alloc_hint, &ins, 1, 1);
7266 return ERR_PTR(ret);
7268 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7269 ins.objectid, ins.offset, ins.offset,
7270 ins.offset, BTRFS_ORDERED_REGULAR);
7271 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7273 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7279 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7281 struct btrfs_block_group *block_group;
7282 bool readonly = false;
7284 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7285 if (!block_group || block_group->ro)
7288 btrfs_put_block_group(block_group);
7293 * Check if we can do nocow write into the range [@offset, @offset + @len)
7295 * @offset: File offset
7296 * @len: The length to write, will be updated to the nocow writeable
7298 * @orig_start: (optional) Return the original file offset of the file extent
7299 * @orig_len: (optional) Return the original on-disk length of the file extent
7300 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7301 * @strict: if true, omit optimizations that might force us into unnecessary
7302 * cow. e.g., don't trust generation number.
7305 * >0 and update @len if we can do nocow write
7306 * 0 if we can't do nocow write
7307 * <0 if error happened
7309 * NOTE: This only checks the file extents, caller is responsible to wait for
7310 * any ordered extents.
7312 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7313 u64 *orig_start, u64 *orig_block_len,
7314 u64 *ram_bytes, bool strict)
7316 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7317 struct btrfs_path *path;
7319 struct extent_buffer *leaf;
7320 struct btrfs_root *root = BTRFS_I(inode)->root;
7321 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7322 struct btrfs_file_extent_item *fi;
7323 struct btrfs_key key;
7330 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7332 path = btrfs_alloc_path();
7336 ret = btrfs_lookup_file_extent(NULL, root, path,
7337 btrfs_ino(BTRFS_I(inode)), offset, 0);
7341 slot = path->slots[0];
7344 /* can't find the item, must cow */
7351 leaf = path->nodes[0];
7352 btrfs_item_key_to_cpu(leaf, &key, slot);
7353 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7354 key.type != BTRFS_EXTENT_DATA_KEY) {
7355 /* not our file or wrong item type, must cow */
7359 if (key.offset > offset) {
7360 /* Wrong offset, must cow */
7364 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7365 found_type = btrfs_file_extent_type(leaf, fi);
7366 if (found_type != BTRFS_FILE_EXTENT_REG &&
7367 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7368 /* not a regular extent, must cow */
7372 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7375 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7376 if (extent_end <= offset)
7379 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7380 if (disk_bytenr == 0)
7383 if (btrfs_file_extent_compression(leaf, fi) ||
7384 btrfs_file_extent_encryption(leaf, fi) ||
7385 btrfs_file_extent_other_encoding(leaf, fi))
7389 * Do the same check as in btrfs_cross_ref_exist but without the
7390 * unnecessary search.
7393 (btrfs_file_extent_generation(leaf, fi) <=
7394 btrfs_root_last_snapshot(&root->root_item)))
7397 backref_offset = btrfs_file_extent_offset(leaf, fi);
7400 *orig_start = key.offset - backref_offset;
7401 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7402 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7405 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7408 num_bytes = min(offset + *len, extent_end) - offset;
7409 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7412 range_end = round_up(offset + num_bytes,
7413 root->fs_info->sectorsize) - 1;
7414 ret = test_range_bit(io_tree, offset, range_end,
7415 EXTENT_DELALLOC, 0, NULL);
7422 btrfs_release_path(path);
7425 * look for other files referencing this extent, if we
7426 * find any we must cow
7429 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7430 key.offset - backref_offset, disk_bytenr,
7438 * adjust disk_bytenr and num_bytes to cover just the bytes
7439 * in this extent we are about to write. If there
7440 * are any csums in that range we have to cow in order
7441 * to keep the csums correct
7443 disk_bytenr += backref_offset;
7444 disk_bytenr += offset - key.offset;
7445 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7448 * all of the above have passed, it is safe to overwrite this extent
7454 btrfs_free_path(path);
7458 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7459 struct extent_state **cached_state, bool writing)
7461 struct btrfs_ordered_extent *ordered;
7465 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7468 * We're concerned with the entire range that we're going to be
7469 * doing DIO to, so we need to make sure there's no ordered
7470 * extents in this range.
7472 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7473 lockend - lockstart + 1);
7476 * We need to make sure there are no buffered pages in this
7477 * range either, we could have raced between the invalidate in
7478 * generic_file_direct_write and locking the extent. The
7479 * invalidate needs to happen so that reads after a write do not
7483 (!writing || !filemap_range_has_page(inode->i_mapping,
7484 lockstart, lockend)))
7487 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7492 * If we are doing a DIO read and the ordered extent we
7493 * found is for a buffered write, we can not wait for it
7494 * to complete and retry, because if we do so we can
7495 * deadlock with concurrent buffered writes on page
7496 * locks. This happens only if our DIO read covers more
7497 * than one extent map, if at this point has already
7498 * created an ordered extent for a previous extent map
7499 * and locked its range in the inode's io tree, and a
7500 * concurrent write against that previous extent map's
7501 * range and this range started (we unlock the ranges
7502 * in the io tree only when the bios complete and
7503 * buffered writes always lock pages before attempting
7504 * to lock range in the io tree).
7507 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7508 btrfs_start_ordered_extent(ordered, 1);
7511 btrfs_put_ordered_extent(ordered);
7514 * We could trigger writeback for this range (and wait
7515 * for it to complete) and then invalidate the pages for
7516 * this range (through invalidate_inode_pages2_range()),
7517 * but that can lead us to a deadlock with a concurrent
7518 * call to readahead (a buffered read or a defrag call
7519 * triggered a readahead) on a page lock due to an
7520 * ordered dio extent we created before but did not have
7521 * yet a corresponding bio submitted (whence it can not
7522 * complete), which makes readahead wait for that
7523 * ordered extent to complete while holding a lock on
7538 /* The callers of this must take lock_extent() */
7539 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7540 u64 len, u64 orig_start, u64 block_start,
7541 u64 block_len, u64 orig_block_len,
7542 u64 ram_bytes, int compress_type,
7545 struct extent_map_tree *em_tree;
7546 struct extent_map *em;
7549 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7550 type == BTRFS_ORDERED_COMPRESSED ||
7551 type == BTRFS_ORDERED_NOCOW ||
7552 type == BTRFS_ORDERED_REGULAR);
7554 em_tree = &inode->extent_tree;
7555 em = alloc_extent_map();
7557 return ERR_PTR(-ENOMEM);
7560 em->orig_start = orig_start;
7562 em->block_len = block_len;
7563 em->block_start = block_start;
7564 em->orig_block_len = orig_block_len;
7565 em->ram_bytes = ram_bytes;
7566 em->generation = -1;
7567 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7568 if (type == BTRFS_ORDERED_PREALLOC) {
7569 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7570 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7571 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7572 em->compress_type = compress_type;
7576 btrfs_drop_extent_cache(inode, em->start,
7577 em->start + em->len - 1, 0);
7578 write_lock(&em_tree->lock);
7579 ret = add_extent_mapping(em_tree, em, 1);
7580 write_unlock(&em_tree->lock);
7582 * The caller has taken lock_extent(), who could race with us
7585 } while (ret == -EEXIST);
7588 free_extent_map(em);
7589 return ERR_PTR(ret);
7592 /* em got 2 refs now, callers needs to do free_extent_map once. */
7597 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7598 struct inode *inode,
7599 struct btrfs_dio_data *dio_data,
7602 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7603 struct extent_map *em = *map;
7607 * We don't allocate a new extent in the following cases
7609 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7611 * 2) The extent is marked as PREALLOC. We're good to go here and can
7612 * just use the extent.
7615 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7616 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7617 em->block_start != EXTENT_MAP_HOLE)) {
7619 u64 block_start, orig_start, orig_block_len, ram_bytes;
7621 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7622 type = BTRFS_ORDERED_PREALLOC;
7624 type = BTRFS_ORDERED_NOCOW;
7625 len = min(len, em->len - (start - em->start));
7626 block_start = em->block_start + (start - em->start);
7628 if (can_nocow_extent(inode, start, &len, &orig_start,
7629 &orig_block_len, &ram_bytes, false) == 1 &&
7630 btrfs_inc_nocow_writers(fs_info, block_start)) {
7631 struct extent_map *em2;
7633 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7634 orig_start, block_start,
7635 len, orig_block_len,
7637 btrfs_dec_nocow_writers(fs_info, block_start);
7638 if (type == BTRFS_ORDERED_PREALLOC) {
7639 free_extent_map(em);
7643 if (em2 && IS_ERR(em2)) {
7648 * For inode marked NODATACOW or extent marked PREALLOC,
7649 * use the existing or preallocated extent, so does not
7650 * need to adjust btrfs_space_info's bytes_may_use.
7652 btrfs_free_reserved_data_space_noquota(fs_info, len);
7657 /* this will cow the extent */
7658 free_extent_map(em);
7659 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7665 len = min(len, em->len - (start - em->start));
7669 * Need to update the i_size under the extent lock so buffered
7670 * readers will get the updated i_size when we unlock.
7672 if (start + len > i_size_read(inode))
7673 i_size_write(inode, start + len);
7675 dio_data->reserve -= len;
7680 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7681 loff_t length, unsigned int flags, struct iomap *iomap,
7682 struct iomap *srcmap)
7684 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7685 struct extent_map *em;
7686 struct extent_state *cached_state = NULL;
7687 struct btrfs_dio_data *dio_data = NULL;
7688 u64 lockstart, lockend;
7689 const bool write = !!(flags & IOMAP_WRITE);
7692 bool unlock_extents = false;
7695 len = min_t(u64, len, fs_info->sectorsize);
7698 lockend = start + len - 1;
7701 * The generic stuff only does filemap_write_and_wait_range, which
7702 * isn't enough if we've written compressed pages to this area, so we
7703 * need to flush the dirty pages again to make absolutely sure that any
7704 * outstanding dirty pages are on disk.
7706 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7707 &BTRFS_I(inode)->runtime_flags)) {
7708 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7709 start + length - 1);
7714 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7718 dio_data->length = length;
7720 dio_data->reserve = round_up(length, fs_info->sectorsize);
7721 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7722 &dio_data->data_reserved,
7723 start, dio_data->reserve);
7725 extent_changeset_free(dio_data->data_reserved);
7730 iomap->private = dio_data;
7734 * If this errors out it's because we couldn't invalidate pagecache for
7735 * this range and we need to fallback to buffered.
7737 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7742 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7749 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7750 * io. INLINE is special, and we could probably kludge it in here, but
7751 * it's still buffered so for safety lets just fall back to the generic
7754 * For COMPRESSED we _have_ to read the entire extent in so we can
7755 * decompress it, so there will be buffering required no matter what we
7756 * do, so go ahead and fallback to buffered.
7758 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7759 * to buffered IO. Don't blame me, this is the price we pay for using
7762 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7763 em->block_start == EXTENT_MAP_INLINE) {
7764 free_extent_map(em);
7769 len = min(len, em->len - (start - em->start));
7771 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7775 unlock_extents = true;
7776 /* Recalc len in case the new em is smaller than requested */
7777 len = min(len, em->len - (start - em->start));
7780 * We need to unlock only the end area that we aren't using.
7781 * The rest is going to be unlocked by the endio routine.
7783 lockstart = start + len;
7784 if (lockstart < lockend)
7785 unlock_extents = true;
7789 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7790 lockstart, lockend, &cached_state);
7792 free_extent_state(cached_state);
7795 * Translate extent map information to iomap.
7796 * We trim the extents (and move the addr) even though iomap code does
7797 * that, since we have locked only the parts we are performing I/O in.
7799 if ((em->block_start == EXTENT_MAP_HOLE) ||
7800 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7801 iomap->addr = IOMAP_NULL_ADDR;
7802 iomap->type = IOMAP_HOLE;
7804 iomap->addr = em->block_start + (start - em->start);
7805 iomap->type = IOMAP_MAPPED;
7807 iomap->offset = start;
7808 iomap->bdev = fs_info->fs_devices->latest_bdev;
7809 iomap->length = len;
7811 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7812 iomap->flags |= IOMAP_F_ZONE_APPEND;
7814 free_extent_map(em);
7819 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7823 btrfs_delalloc_release_space(BTRFS_I(inode),
7824 dio_data->data_reserved, start,
7825 dio_data->reserve, true);
7826 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7827 extent_changeset_free(dio_data->data_reserved);
7833 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7834 ssize_t written, unsigned int flags, struct iomap *iomap)
7837 struct btrfs_dio_data *dio_data = iomap->private;
7838 size_t submitted = dio_data->submitted;
7839 const bool write = !!(flags & IOMAP_WRITE);
7841 if (!write && (iomap->type == IOMAP_HOLE)) {
7842 /* If reading from a hole, unlock and return */
7843 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7847 if (submitted < length) {
7849 length -= submitted;
7851 __endio_write_update_ordered(BTRFS_I(inode), pos,
7854 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7860 if (dio_data->reserve)
7861 btrfs_delalloc_release_space(BTRFS_I(inode),
7862 dio_data->data_reserved, pos,
7863 dio_data->reserve, true);
7864 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7865 extent_changeset_free(dio_data->data_reserved);
7869 iomap->private = NULL;
7874 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7877 * This implies a barrier so that stores to dio_bio->bi_status before
7878 * this and loads of dio_bio->bi_status after this are fully ordered.
7880 if (!refcount_dec_and_test(&dip->refs))
7883 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7884 __endio_write_update_ordered(BTRFS_I(dip->inode),
7885 dip->logical_offset,
7887 !dip->dio_bio->bi_status);
7889 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7890 dip->logical_offset,
7891 dip->logical_offset + dip->bytes - 1);
7894 bio_endio(dip->dio_bio);
7898 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7900 unsigned long bio_flags)
7902 struct btrfs_dio_private *dip = bio->bi_private;
7903 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7906 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7908 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7912 refcount_inc(&dip->refs);
7913 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7915 refcount_dec(&dip->refs);
7919 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7920 struct btrfs_io_bio *io_bio,
7921 const bool uptodate)
7923 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7924 const u32 sectorsize = fs_info->sectorsize;
7925 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7926 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7927 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7928 struct bio_vec bvec;
7929 struct bvec_iter iter;
7930 u64 start = io_bio->logical;
7932 blk_status_t err = BLK_STS_OK;
7934 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7935 unsigned int i, nr_sectors, pgoff;
7937 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7938 pgoff = bvec.bv_offset;
7939 for (i = 0; i < nr_sectors; i++) {
7940 ASSERT(pgoff < PAGE_SIZE);
7942 (!csum || !check_data_csum(inode, io_bio,
7943 bio_offset, bvec.bv_page,
7945 clean_io_failure(fs_info, failure_tree, io_tree,
7946 start, bvec.bv_page,
7947 btrfs_ino(BTRFS_I(inode)),
7950 blk_status_t status;
7952 ASSERT((start - io_bio->logical) < UINT_MAX);
7953 status = btrfs_submit_read_repair(inode,
7955 start - io_bio->logical,
7956 bvec.bv_page, pgoff,
7958 start + sectorsize - 1,
7960 submit_dio_repair_bio);
7964 start += sectorsize;
7965 ASSERT(bio_offset + sectorsize > bio_offset);
7966 bio_offset += sectorsize;
7967 pgoff += sectorsize;
7973 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7974 const u64 offset, const u64 bytes,
7975 const bool uptodate)
7977 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7978 struct btrfs_ordered_extent *ordered = NULL;
7979 struct btrfs_workqueue *wq;
7980 u64 ordered_offset = offset;
7981 u64 ordered_bytes = bytes;
7984 if (btrfs_is_free_space_inode(inode))
7985 wq = fs_info->endio_freespace_worker;
7987 wq = fs_info->endio_write_workers;
7989 while (ordered_offset < offset + bytes) {
7990 last_offset = ordered_offset;
7991 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7995 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7997 btrfs_queue_work(wq, &ordered->work);
8000 /* No ordered extent found in the range, exit */
8001 if (ordered_offset == last_offset)
8004 * Our bio might span multiple ordered extents. In this case
8005 * we keep going until we have accounted the whole dio.
8007 if (ordered_offset < offset + bytes) {
8008 ordered_bytes = offset + bytes - ordered_offset;
8014 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8016 u64 dio_file_offset)
8018 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
8021 static void btrfs_end_dio_bio(struct bio *bio)
8023 struct btrfs_dio_private *dip = bio->bi_private;
8024 blk_status_t err = bio->bi_status;
8027 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8028 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8029 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8030 bio->bi_opf, bio->bi_iter.bi_sector,
8031 bio->bi_iter.bi_size, err);
8033 if (bio_op(bio) == REQ_OP_READ) {
8034 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
8039 dip->dio_bio->bi_status = err;
8041 btrfs_record_physical_zoned(dip->inode, dip->logical_offset, bio);
8044 btrfs_dio_private_put(dip);
8047 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8048 struct inode *inode, u64 file_offset, int async_submit)
8050 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8051 struct btrfs_dio_private *dip = bio->bi_private;
8052 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
8055 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8057 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8060 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8065 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8068 if (write && async_submit) {
8069 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
8070 btrfs_submit_bio_start_direct_io);
8074 * If we aren't doing async submit, calculate the csum of the
8077 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
8083 csum_offset = file_offset - dip->logical_offset;
8084 csum_offset >>= fs_info->sectorsize_bits;
8085 csum_offset *= fs_info->csum_size;
8086 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
8089 ret = btrfs_map_bio(fs_info, bio, 0);
8095 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
8096 * or ordered extents whether or not we submit any bios.
8098 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
8099 struct inode *inode,
8102 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8103 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8105 struct btrfs_dio_private *dip;
8107 dip_size = sizeof(*dip);
8108 if (!write && csum) {
8109 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8112 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
8113 dip_size += fs_info->csum_size * nblocks;
8116 dip = kzalloc(dip_size, GFP_NOFS);
8121 dip->logical_offset = file_offset;
8122 dip->bytes = dio_bio->bi_iter.bi_size;
8123 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
8124 dip->dio_bio = dio_bio;
8125 refcount_set(&dip->refs, 1);
8129 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
8130 struct bio *dio_bio, loff_t file_offset)
8132 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8133 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8134 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8135 BTRFS_BLOCK_GROUP_RAID56_MASK);
8136 struct btrfs_dio_private *dip;
8139 int async_submit = 0;
8141 int clone_offset = 0;
8145 blk_status_t status;
8146 struct btrfs_io_geometry geom;
8147 struct btrfs_dio_data *dio_data = iomap->private;
8148 struct extent_map *em = NULL;
8150 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8153 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8154 file_offset + dio_bio->bi_iter.bi_size - 1);
8156 dio_bio->bi_status = BLK_STS_RESOURCE;
8158 return BLK_QC_T_NONE;
8163 * Load the csums up front to reduce csum tree searches and
8164 * contention when submitting bios.
8166 * If we have csums disabled this will do nothing.
8168 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8169 if (status != BLK_STS_OK)
8173 start_sector = dio_bio->bi_iter.bi_sector;
8174 submit_len = dio_bio->bi_iter.bi_size;
8177 logical = start_sector << 9;
8178 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8180 status = errno_to_blk_status(PTR_ERR(em));
8184 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8185 logical, submit_len, &geom);
8187 status = errno_to_blk_status(ret);
8190 ASSERT(geom.len <= INT_MAX);
8192 clone_len = min_t(int, submit_len, geom.len);
8195 * This will never fail as it's passing GPF_NOFS and
8196 * the allocation is backed by btrfs_bioset.
8198 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8199 bio->bi_private = dip;
8200 bio->bi_end_io = btrfs_end_dio_bio;
8201 btrfs_io_bio(bio)->logical = file_offset;
8203 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8204 status = extract_ordered_extent(BTRFS_I(inode), bio,
8212 ASSERT(submit_len >= clone_len);
8213 submit_len -= clone_len;
8216 * Increase the count before we submit the bio so we know
8217 * the end IO handler won't happen before we increase the
8218 * count. Otherwise, the dip might get freed before we're
8219 * done setting it up.
8221 * We transfer the initial reference to the last bio, so we
8222 * don't need to increment the reference count for the last one.
8224 if (submit_len > 0) {
8225 refcount_inc(&dip->refs);
8227 * If we are submitting more than one bio, submit them
8228 * all asynchronously. The exception is RAID 5 or 6, as
8229 * asynchronous checksums make it difficult to collect
8230 * full stripe writes.
8236 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8241 refcount_dec(&dip->refs);
8245 dio_data->submitted += clone_len;
8246 clone_offset += clone_len;
8247 start_sector += clone_len >> 9;
8248 file_offset += clone_len;
8250 free_extent_map(em);
8251 } while (submit_len > 0);
8252 return BLK_QC_T_NONE;
8255 free_extent_map(em);
8257 dip->dio_bio->bi_status = status;
8258 btrfs_dio_private_put(dip);
8260 return BLK_QC_T_NONE;
8263 const struct iomap_ops btrfs_dio_iomap_ops = {
8264 .iomap_begin = btrfs_dio_iomap_begin,
8265 .iomap_end = btrfs_dio_iomap_end,
8268 const struct iomap_dio_ops btrfs_dio_ops = {
8269 .submit_io = btrfs_submit_direct,
8272 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8277 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8281 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8284 int btrfs_readpage(struct file *file, struct page *page)
8286 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8287 u64 start = page_offset(page);
8288 u64 end = start + PAGE_SIZE - 1;
8289 unsigned long bio_flags = 0;
8290 struct bio *bio = NULL;
8293 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8295 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8297 ret = submit_one_bio(bio, 0, bio_flags);
8301 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8303 struct inode *inode = page->mapping->host;
8306 if (current->flags & PF_MEMALLOC) {
8307 redirty_page_for_writepage(wbc, page);
8313 * If we are under memory pressure we will call this directly from the
8314 * VM, we need to make sure we have the inode referenced for the ordered
8315 * extent. If not just return like we didn't do anything.
8317 if (!igrab(inode)) {
8318 redirty_page_for_writepage(wbc, page);
8319 return AOP_WRITEPAGE_ACTIVATE;
8321 ret = extent_write_full_page(page, wbc);
8322 btrfs_add_delayed_iput(inode);
8326 static int btrfs_writepages(struct address_space *mapping,
8327 struct writeback_control *wbc)
8329 return extent_writepages(mapping, wbc);
8332 static void btrfs_readahead(struct readahead_control *rac)
8334 extent_readahead(rac);
8337 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8339 int ret = try_release_extent_mapping(page, gfp_flags);
8341 clear_page_extent_mapped(page);
8345 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8347 if (PageWriteback(page) || PageDirty(page))
8349 return __btrfs_releasepage(page, gfp_flags);
8352 #ifdef CONFIG_MIGRATION
8353 static int btrfs_migratepage(struct address_space *mapping,
8354 struct page *newpage, struct page *page,
8355 enum migrate_mode mode)
8359 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8360 if (ret != MIGRATEPAGE_SUCCESS)
8363 if (page_has_private(page))
8364 attach_page_private(newpage, detach_page_private(page));
8366 if (PagePrivate2(page)) {
8367 ClearPagePrivate2(page);
8368 SetPagePrivate2(newpage);
8371 if (mode != MIGRATE_SYNC_NO_COPY)
8372 migrate_page_copy(newpage, page);
8374 migrate_page_states(newpage, page);
8375 return MIGRATEPAGE_SUCCESS;
8379 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8380 unsigned int length)
8382 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8383 struct extent_io_tree *tree = &inode->io_tree;
8384 struct btrfs_ordered_extent *ordered;
8385 struct extent_state *cached_state = NULL;
8386 u64 page_start = page_offset(page);
8387 u64 page_end = page_start + PAGE_SIZE - 1;
8390 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8391 bool found_ordered = false;
8392 bool completed_ordered = false;
8395 * we have the page locked, so new writeback can't start,
8396 * and the dirty bit won't be cleared while we are here.
8398 * Wait for IO on this page so that we can safely clear
8399 * the PagePrivate2 bit and do ordered accounting
8401 wait_on_page_writeback(page);
8404 btrfs_releasepage(page, GFP_NOFS);
8408 if (!inode_evicting)
8409 lock_extent_bits(tree, page_start, page_end, &cached_state);
8413 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8415 found_ordered = true;
8417 ordered->file_offset + ordered->num_bytes - 1);
8419 * IO on this page will never be started, so we need to account
8420 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8421 * here, must leave that up for the ordered extent completion.
8423 if (!inode_evicting)
8424 clear_extent_bit(tree, start, end,
8426 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8427 EXTENT_DEFRAG, 1, 0, &cached_state);
8429 * whoever cleared the private bit is responsible
8430 * for the finish_ordered_io
8432 if (TestClearPagePrivate2(page)) {
8433 spin_lock_irq(&inode->ordered_tree.lock);
8434 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8435 ordered->truncated_len = min(ordered->truncated_len,
8436 start - ordered->file_offset);
8437 spin_unlock_irq(&inode->ordered_tree.lock);
8439 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8441 end - start + 1, 1)) {
8442 btrfs_finish_ordered_io(ordered);
8443 completed_ordered = true;
8446 btrfs_put_ordered_extent(ordered);
8447 if (!inode_evicting) {
8448 cached_state = NULL;
8449 lock_extent_bits(tree, start, end,
8454 if (start < page_end)
8459 * Qgroup reserved space handler
8460 * Page here will be either
8461 * 1) Already written to disk or ordered extent already submitted
8462 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8463 * Qgroup will be handled by its qgroup_record then.
8464 * btrfs_qgroup_free_data() call will do nothing here.
8466 * 2) Not written to disk yet
8467 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8468 * bit of its io_tree, and free the qgroup reserved data space.
8469 * Since the IO will never happen for this page.
8471 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8472 if (!inode_evicting) {
8476 * If there's an ordered extent for this range and we have not
8477 * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set
8478 * in the range for the ordered extent completion. We must also
8479 * not delete the range, otherwise we would lose that bit (and
8480 * any other bits set in the range). Make sure EXTENT_UPTODATE
8481 * is cleared if we don't delete, otherwise it can lead to
8482 * corruptions if the i_size is extented later.
8484 if (found_ordered && !completed_ordered)
8486 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8487 EXTENT_DELALLOC | EXTENT_UPTODATE |
8488 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8489 delete, &cached_state);
8491 __btrfs_releasepage(page, GFP_NOFS);
8494 ClearPageChecked(page);
8495 clear_page_extent_mapped(page);
8499 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8500 * called from a page fault handler when a page is first dirtied. Hence we must
8501 * be careful to check for EOF conditions here. We set the page up correctly
8502 * for a written page which means we get ENOSPC checking when writing into
8503 * holes and correct delalloc and unwritten extent mapping on filesystems that
8504 * support these features.
8506 * We are not allowed to take the i_mutex here so we have to play games to
8507 * protect against truncate races as the page could now be beyond EOF. Because
8508 * truncate_setsize() writes the inode size before removing pages, once we have
8509 * the page lock we can determine safely if the page is beyond EOF. If it is not
8510 * beyond EOF, then the page is guaranteed safe against truncation until we
8513 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8515 struct page *page = vmf->page;
8516 struct inode *inode = file_inode(vmf->vma->vm_file);
8517 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8518 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8519 struct btrfs_ordered_extent *ordered;
8520 struct extent_state *cached_state = NULL;
8521 struct extent_changeset *data_reserved = NULL;
8523 unsigned long zero_start;
8533 reserved_space = PAGE_SIZE;
8535 sb_start_pagefault(inode->i_sb);
8536 page_start = page_offset(page);
8537 page_end = page_start + PAGE_SIZE - 1;
8541 * Reserving delalloc space after obtaining the page lock can lead to
8542 * deadlock. For example, if a dirty page is locked by this function
8543 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8544 * dirty page write out, then the btrfs_writepage() function could
8545 * end up waiting indefinitely to get a lock on the page currently
8546 * being processed by btrfs_page_mkwrite() function.
8548 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8549 page_start, reserved_space);
8551 ret2 = file_update_time(vmf->vma->vm_file);
8555 ret = vmf_error(ret2);
8561 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8563 down_read(&BTRFS_I(inode)->i_mmap_lock);
8565 size = i_size_read(inode);
8567 if ((page->mapping != inode->i_mapping) ||
8568 (page_start >= size)) {
8569 /* page got truncated out from underneath us */
8572 wait_on_page_writeback(page);
8574 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8575 ret2 = set_page_extent_mapped(page);
8577 ret = vmf_error(ret2);
8578 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8583 * we can't set the delalloc bits if there are pending ordered
8584 * extents. Drop our locks and wait for them to finish
8586 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8589 unlock_extent_cached(io_tree, page_start, page_end,
8592 up_read(&BTRFS_I(inode)->i_mmap_lock);
8593 btrfs_start_ordered_extent(ordered, 1);
8594 btrfs_put_ordered_extent(ordered);
8598 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8599 reserved_space = round_up(size - page_start,
8600 fs_info->sectorsize);
8601 if (reserved_space < PAGE_SIZE) {
8602 end = page_start + reserved_space - 1;
8603 btrfs_delalloc_release_space(BTRFS_I(inode),
8604 data_reserved, page_start,
8605 PAGE_SIZE - reserved_space, true);
8610 * page_mkwrite gets called when the page is firstly dirtied after it's
8611 * faulted in, but write(2) could also dirty a page and set delalloc
8612 * bits, thus in this case for space account reason, we still need to
8613 * clear any delalloc bits within this page range since we have to
8614 * reserve data&meta space before lock_page() (see above comments).
8616 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8617 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8618 EXTENT_DEFRAG, 0, 0, &cached_state);
8620 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8623 unlock_extent_cached(io_tree, page_start, page_end,
8625 ret = VM_FAULT_SIGBUS;
8629 /* page is wholly or partially inside EOF */
8630 if (page_start + PAGE_SIZE > size)
8631 zero_start = offset_in_page(size);
8633 zero_start = PAGE_SIZE;
8635 if (zero_start != PAGE_SIZE) {
8637 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8638 flush_dcache_page(page);
8641 ClearPageChecked(page);
8642 set_page_dirty(page);
8643 SetPageUptodate(page);
8645 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8647 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8648 up_read(&BTRFS_I(inode)->i_mmap_lock);
8650 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8651 sb_end_pagefault(inode->i_sb);
8652 extent_changeset_free(data_reserved);
8653 return VM_FAULT_LOCKED;
8657 up_read(&BTRFS_I(inode)->i_mmap_lock);
8659 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8660 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8661 reserved_space, (ret != 0));
8663 sb_end_pagefault(inode->i_sb);
8664 extent_changeset_free(data_reserved);
8668 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8670 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8671 struct btrfs_root *root = BTRFS_I(inode)->root;
8672 struct btrfs_block_rsv *rsv;
8674 struct btrfs_trans_handle *trans;
8675 u64 mask = fs_info->sectorsize - 1;
8676 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8678 if (!skip_writeback) {
8679 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8686 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8687 * things going on here:
8689 * 1) We need to reserve space to update our inode.
8691 * 2) We need to have something to cache all the space that is going to
8692 * be free'd up by the truncate operation, but also have some slack
8693 * space reserved in case it uses space during the truncate (thank you
8694 * very much snapshotting).
8696 * And we need these to be separate. The fact is we can use a lot of
8697 * space doing the truncate, and we have no earthly idea how much space
8698 * we will use, so we need the truncate reservation to be separate so it
8699 * doesn't end up using space reserved for updating the inode. We also
8700 * need to be able to stop the transaction and start a new one, which
8701 * means we need to be able to update the inode several times, and we
8702 * have no idea of knowing how many times that will be, so we can't just
8703 * reserve 1 item for the entirety of the operation, so that has to be
8704 * done separately as well.
8706 * So that leaves us with
8708 * 1) rsv - for the truncate reservation, which we will steal from the
8709 * transaction reservation.
8710 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8711 * updating the inode.
8713 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8716 rsv->size = min_size;
8720 * 1 for the truncate slack space
8721 * 1 for updating the inode.
8723 trans = btrfs_start_transaction(root, 2);
8724 if (IS_ERR(trans)) {
8725 ret = PTR_ERR(trans);
8729 /* Migrate the slack space for the truncate to our reserve */
8730 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8735 * So if we truncate and then write and fsync we normally would just
8736 * write the extents that changed, which is a problem if we need to
8737 * first truncate that entire inode. So set this flag so we write out
8738 * all of the extents in the inode to the sync log so we're completely
8741 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8742 trans->block_rsv = rsv;
8745 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8747 BTRFS_EXTENT_DATA_KEY);
8748 trans->block_rsv = &fs_info->trans_block_rsv;
8749 if (ret != -ENOSPC && ret != -EAGAIN)
8752 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8756 btrfs_end_transaction(trans);
8757 btrfs_btree_balance_dirty(fs_info);
8759 trans = btrfs_start_transaction(root, 2);
8760 if (IS_ERR(trans)) {
8761 ret = PTR_ERR(trans);
8766 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8767 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8768 rsv, min_size, false);
8769 BUG_ON(ret); /* shouldn't happen */
8770 trans->block_rsv = rsv;
8774 * We can't call btrfs_truncate_block inside a trans handle as we could
8775 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8776 * we've truncated everything except the last little bit, and can do
8777 * btrfs_truncate_block and then update the disk_i_size.
8779 if (ret == NEED_TRUNCATE_BLOCK) {
8780 btrfs_end_transaction(trans);
8781 btrfs_btree_balance_dirty(fs_info);
8783 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8786 trans = btrfs_start_transaction(root, 1);
8787 if (IS_ERR(trans)) {
8788 ret = PTR_ERR(trans);
8791 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8797 trans->block_rsv = &fs_info->trans_block_rsv;
8798 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8802 ret2 = btrfs_end_transaction(trans);
8805 btrfs_btree_balance_dirty(fs_info);
8808 btrfs_free_block_rsv(fs_info, rsv);
8814 * create a new subvolume directory/inode (helper for the ioctl).
8816 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8817 struct btrfs_root *new_root,
8818 struct btrfs_root *parent_root)
8820 struct inode *inode;
8825 err = btrfs_get_free_objectid(new_root, &ino);
8829 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8830 S_IFDIR | (~current_umask() & S_IRWXUGO),
8833 return PTR_ERR(inode);
8834 inode->i_op = &btrfs_dir_inode_operations;
8835 inode->i_fop = &btrfs_dir_file_operations;
8837 set_nlink(inode, 1);
8838 btrfs_i_size_write(BTRFS_I(inode), 0);
8839 unlock_new_inode(inode);
8841 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8843 btrfs_err(new_root->fs_info,
8844 "error inheriting subvolume %llu properties: %d",
8845 new_root->root_key.objectid, err);
8847 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8853 struct inode *btrfs_alloc_inode(struct super_block *sb)
8855 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8856 struct btrfs_inode *ei;
8857 struct inode *inode;
8859 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8866 ei->last_sub_trans = 0;
8867 ei->logged_trans = 0;
8868 ei->delalloc_bytes = 0;
8869 ei->new_delalloc_bytes = 0;
8870 ei->defrag_bytes = 0;
8871 ei->disk_i_size = 0;
8874 ei->index_cnt = (u64)-1;
8876 ei->last_unlink_trans = 0;
8877 ei->last_reflink_trans = 0;
8878 ei->last_log_commit = 0;
8880 spin_lock_init(&ei->lock);
8881 ei->outstanding_extents = 0;
8882 if (sb->s_magic != BTRFS_TEST_MAGIC)
8883 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8884 BTRFS_BLOCK_RSV_DELALLOC);
8885 ei->runtime_flags = 0;
8886 ei->prop_compress = BTRFS_COMPRESS_NONE;
8887 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8889 ei->delayed_node = NULL;
8891 ei->i_otime.tv_sec = 0;
8892 ei->i_otime.tv_nsec = 0;
8894 inode = &ei->vfs_inode;
8895 extent_map_tree_init(&ei->extent_tree);
8896 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8897 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8898 IO_TREE_INODE_IO_FAILURE, inode);
8899 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8900 IO_TREE_INODE_FILE_EXTENT, inode);
8901 ei->io_tree.track_uptodate = true;
8902 ei->io_failure_tree.track_uptodate = true;
8903 atomic_set(&ei->sync_writers, 0);
8904 mutex_init(&ei->log_mutex);
8905 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8906 INIT_LIST_HEAD(&ei->delalloc_inodes);
8907 INIT_LIST_HEAD(&ei->delayed_iput);
8908 RB_CLEAR_NODE(&ei->rb_node);
8909 init_rwsem(&ei->i_mmap_lock);
8914 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8915 void btrfs_test_destroy_inode(struct inode *inode)
8917 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8918 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8922 void btrfs_free_inode(struct inode *inode)
8924 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8927 void btrfs_destroy_inode(struct inode *vfs_inode)
8929 struct btrfs_ordered_extent *ordered;
8930 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8931 struct btrfs_root *root = inode->root;
8933 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8934 WARN_ON(vfs_inode->i_data.nrpages);
8935 WARN_ON(inode->block_rsv.reserved);
8936 WARN_ON(inode->block_rsv.size);
8937 WARN_ON(inode->outstanding_extents);
8938 WARN_ON(inode->delalloc_bytes);
8939 WARN_ON(inode->new_delalloc_bytes);
8940 WARN_ON(inode->csum_bytes);
8941 WARN_ON(inode->defrag_bytes);
8944 * This can happen where we create an inode, but somebody else also
8945 * created the same inode and we need to destroy the one we already
8952 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8956 btrfs_err(root->fs_info,
8957 "found ordered extent %llu %llu on inode cleanup",
8958 ordered->file_offset, ordered->num_bytes);
8959 btrfs_remove_ordered_extent(inode, ordered);
8960 btrfs_put_ordered_extent(ordered);
8961 btrfs_put_ordered_extent(ordered);
8964 btrfs_qgroup_check_reserved_leak(inode);
8965 inode_tree_del(inode);
8966 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8967 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8968 btrfs_put_root(inode->root);
8971 int btrfs_drop_inode(struct inode *inode)
8973 struct btrfs_root *root = BTRFS_I(inode)->root;
8978 /* the snap/subvol tree is on deleting */
8979 if (btrfs_root_refs(&root->root_item) == 0)
8982 return generic_drop_inode(inode);
8985 static void init_once(void *foo)
8987 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8989 inode_init_once(&ei->vfs_inode);
8992 void __cold btrfs_destroy_cachep(void)
8995 * Make sure all delayed rcu free inodes are flushed before we
8999 kmem_cache_destroy(btrfs_inode_cachep);
9000 kmem_cache_destroy(btrfs_trans_handle_cachep);
9001 kmem_cache_destroy(btrfs_path_cachep);
9002 kmem_cache_destroy(btrfs_free_space_cachep);
9003 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9006 int __init btrfs_init_cachep(void)
9008 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9009 sizeof(struct btrfs_inode), 0,
9010 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9012 if (!btrfs_inode_cachep)
9015 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9016 sizeof(struct btrfs_trans_handle), 0,
9017 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9018 if (!btrfs_trans_handle_cachep)
9021 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9022 sizeof(struct btrfs_path), 0,
9023 SLAB_MEM_SPREAD, NULL);
9024 if (!btrfs_path_cachep)
9027 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9028 sizeof(struct btrfs_free_space), 0,
9029 SLAB_MEM_SPREAD, NULL);
9030 if (!btrfs_free_space_cachep)
9033 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9034 PAGE_SIZE, PAGE_SIZE,
9035 SLAB_MEM_SPREAD, NULL);
9036 if (!btrfs_free_space_bitmap_cachep)
9041 btrfs_destroy_cachep();
9045 static int btrfs_getattr(struct user_namespace *mnt_userns,
9046 const struct path *path, struct kstat *stat,
9047 u32 request_mask, unsigned int flags)
9051 struct inode *inode = d_inode(path->dentry);
9052 u32 blocksize = inode->i_sb->s_blocksize;
9053 u32 bi_flags = BTRFS_I(inode)->flags;
9055 stat->result_mask |= STATX_BTIME;
9056 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9057 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9058 if (bi_flags & BTRFS_INODE_APPEND)
9059 stat->attributes |= STATX_ATTR_APPEND;
9060 if (bi_flags & BTRFS_INODE_COMPRESS)
9061 stat->attributes |= STATX_ATTR_COMPRESSED;
9062 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9063 stat->attributes |= STATX_ATTR_IMMUTABLE;
9064 if (bi_flags & BTRFS_INODE_NODUMP)
9065 stat->attributes |= STATX_ATTR_NODUMP;
9067 stat->attributes_mask |= (STATX_ATTR_APPEND |
9068 STATX_ATTR_COMPRESSED |
9069 STATX_ATTR_IMMUTABLE |
9072 generic_fillattr(&init_user_ns, inode, stat);
9073 stat->dev = BTRFS_I(inode)->root->anon_dev;
9075 spin_lock(&BTRFS_I(inode)->lock);
9076 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9077 inode_bytes = inode_get_bytes(inode);
9078 spin_unlock(&BTRFS_I(inode)->lock);
9079 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9080 ALIGN(delalloc_bytes, blocksize)) >> 9;
9084 static int btrfs_rename_exchange(struct inode *old_dir,
9085 struct dentry *old_dentry,
9086 struct inode *new_dir,
9087 struct dentry *new_dentry)
9089 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9090 struct btrfs_trans_handle *trans;
9091 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9092 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9093 struct inode *new_inode = new_dentry->d_inode;
9094 struct inode *old_inode = old_dentry->d_inode;
9095 struct timespec64 ctime = current_time(old_inode);
9096 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9097 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9102 bool root_log_pinned = false;
9103 bool dest_log_pinned = false;
9105 /* we only allow rename subvolume link between subvolumes */
9106 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9109 /* close the race window with snapshot create/destroy ioctl */
9110 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9111 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9112 down_read(&fs_info->subvol_sem);
9115 * We want to reserve the absolute worst case amount of items. So if
9116 * both inodes are subvols and we need to unlink them then that would
9117 * require 4 item modifications, but if they are both normal inodes it
9118 * would require 5 item modifications, so we'll assume their normal
9119 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9120 * should cover the worst case number of items we'll modify.
9122 trans = btrfs_start_transaction(root, 12);
9123 if (IS_ERR(trans)) {
9124 ret = PTR_ERR(trans);
9129 ret = btrfs_record_root_in_trans(trans, dest);
9135 * We need to find a free sequence number both in the source and
9136 * in the destination directory for the exchange.
9138 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9141 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9145 BTRFS_I(old_inode)->dir_index = 0ULL;
9146 BTRFS_I(new_inode)->dir_index = 0ULL;
9148 /* Reference for the source. */
9149 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9150 /* force full log commit if subvolume involved. */
9151 btrfs_set_log_full_commit(trans);
9153 btrfs_pin_log_trans(root);
9154 root_log_pinned = true;
9155 ret = btrfs_insert_inode_ref(trans, dest,
9156 new_dentry->d_name.name,
9157 new_dentry->d_name.len,
9159 btrfs_ino(BTRFS_I(new_dir)),
9165 /* And now for the dest. */
9166 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9167 /* force full log commit if subvolume involved. */
9168 btrfs_set_log_full_commit(trans);
9170 btrfs_pin_log_trans(dest);
9171 dest_log_pinned = true;
9172 ret = btrfs_insert_inode_ref(trans, root,
9173 old_dentry->d_name.name,
9174 old_dentry->d_name.len,
9176 btrfs_ino(BTRFS_I(old_dir)),
9182 /* Update inode version and ctime/mtime. */
9183 inode_inc_iversion(old_dir);
9184 inode_inc_iversion(new_dir);
9185 inode_inc_iversion(old_inode);
9186 inode_inc_iversion(new_inode);
9187 old_dir->i_ctime = old_dir->i_mtime = ctime;
9188 new_dir->i_ctime = new_dir->i_mtime = ctime;
9189 old_inode->i_ctime = ctime;
9190 new_inode->i_ctime = ctime;
9192 if (old_dentry->d_parent != new_dentry->d_parent) {
9193 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9194 BTRFS_I(old_inode), 1);
9195 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9196 BTRFS_I(new_inode), 1);
9199 /* src is a subvolume */
9200 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9201 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9202 } else { /* src is an inode */
9203 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9204 BTRFS_I(old_dentry->d_inode),
9205 old_dentry->d_name.name,
9206 old_dentry->d_name.len);
9208 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9211 btrfs_abort_transaction(trans, ret);
9215 /* dest is a subvolume */
9216 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9217 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9218 } else { /* dest is an inode */
9219 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9220 BTRFS_I(new_dentry->d_inode),
9221 new_dentry->d_name.name,
9222 new_dentry->d_name.len);
9224 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9227 btrfs_abort_transaction(trans, ret);
9231 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9232 new_dentry->d_name.name,
9233 new_dentry->d_name.len, 0, old_idx);
9235 btrfs_abort_transaction(trans, ret);
9239 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9240 old_dentry->d_name.name,
9241 old_dentry->d_name.len, 0, new_idx);
9243 btrfs_abort_transaction(trans, ret);
9247 if (old_inode->i_nlink == 1)
9248 BTRFS_I(old_inode)->dir_index = old_idx;
9249 if (new_inode->i_nlink == 1)
9250 BTRFS_I(new_inode)->dir_index = new_idx;
9252 if (root_log_pinned) {
9253 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9254 new_dentry->d_parent);
9255 btrfs_end_log_trans(root);
9256 root_log_pinned = false;
9258 if (dest_log_pinned) {
9259 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9260 old_dentry->d_parent);
9261 btrfs_end_log_trans(dest);
9262 dest_log_pinned = false;
9266 * If we have pinned a log and an error happened, we unpin tasks
9267 * trying to sync the log and force them to fallback to a transaction
9268 * commit if the log currently contains any of the inodes involved in
9269 * this rename operation (to ensure we do not persist a log with an
9270 * inconsistent state for any of these inodes or leading to any
9271 * inconsistencies when replayed). If the transaction was aborted, the
9272 * abortion reason is propagated to userspace when attempting to commit
9273 * the transaction. If the log does not contain any of these inodes, we
9274 * allow the tasks to sync it.
9276 if (ret && (root_log_pinned || dest_log_pinned)) {
9277 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9278 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9279 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9281 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9282 btrfs_set_log_full_commit(trans);
9284 if (root_log_pinned) {
9285 btrfs_end_log_trans(root);
9286 root_log_pinned = false;
9288 if (dest_log_pinned) {
9289 btrfs_end_log_trans(dest);
9290 dest_log_pinned = false;
9293 ret2 = btrfs_end_transaction(trans);
9294 ret = ret ? ret : ret2;
9296 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9297 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9298 up_read(&fs_info->subvol_sem);
9303 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9304 struct btrfs_root *root,
9306 struct dentry *dentry)
9309 struct inode *inode;
9313 ret = btrfs_get_free_objectid(root, &objectid);
9317 inode = btrfs_new_inode(trans, root, dir,
9318 dentry->d_name.name,
9320 btrfs_ino(BTRFS_I(dir)),
9322 S_IFCHR | WHITEOUT_MODE,
9325 if (IS_ERR(inode)) {
9326 ret = PTR_ERR(inode);
9330 inode->i_op = &btrfs_special_inode_operations;
9331 init_special_inode(inode, inode->i_mode,
9334 ret = btrfs_init_inode_security(trans, inode, dir,
9339 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9340 BTRFS_I(inode), 0, index);
9344 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9346 unlock_new_inode(inode);
9348 inode_dec_link_count(inode);
9354 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9355 struct inode *new_dir, struct dentry *new_dentry,
9358 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9359 struct btrfs_trans_handle *trans;
9360 unsigned int trans_num_items;
9361 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9362 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9363 struct inode *new_inode = d_inode(new_dentry);
9364 struct inode *old_inode = d_inode(old_dentry);
9368 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9369 bool log_pinned = false;
9371 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9374 /* we only allow rename subvolume link between subvolumes */
9375 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9378 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9379 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9382 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9383 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9387 /* check for collisions, even if the name isn't there */
9388 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9389 new_dentry->d_name.name,
9390 new_dentry->d_name.len);
9393 if (ret == -EEXIST) {
9395 * eexist without a new_inode */
9396 if (WARN_ON(!new_inode)) {
9400 /* maybe -EOVERFLOW */
9407 * we're using rename to replace one file with another. Start IO on it
9408 * now so we don't add too much work to the end of the transaction
9410 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9411 filemap_flush(old_inode->i_mapping);
9413 /* close the racy window with snapshot create/destroy ioctl */
9414 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9415 down_read(&fs_info->subvol_sem);
9417 * We want to reserve the absolute worst case amount of items. So if
9418 * both inodes are subvols and we need to unlink them then that would
9419 * require 4 item modifications, but if they are both normal inodes it
9420 * would require 5 item modifications, so we'll assume they are normal
9421 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9422 * should cover the worst case number of items we'll modify.
9423 * If our rename has the whiteout flag, we need more 5 units for the
9424 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9425 * when selinux is enabled).
9427 trans_num_items = 11;
9428 if (flags & RENAME_WHITEOUT)
9429 trans_num_items += 5;
9430 trans = btrfs_start_transaction(root, trans_num_items);
9431 if (IS_ERR(trans)) {
9432 ret = PTR_ERR(trans);
9437 ret = btrfs_record_root_in_trans(trans, dest);
9442 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9446 BTRFS_I(old_inode)->dir_index = 0ULL;
9447 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9448 /* force full log commit if subvolume involved. */
9449 btrfs_set_log_full_commit(trans);
9451 btrfs_pin_log_trans(root);
9453 ret = btrfs_insert_inode_ref(trans, dest,
9454 new_dentry->d_name.name,
9455 new_dentry->d_name.len,
9457 btrfs_ino(BTRFS_I(new_dir)), index);
9462 inode_inc_iversion(old_dir);
9463 inode_inc_iversion(new_dir);
9464 inode_inc_iversion(old_inode);
9465 old_dir->i_ctime = old_dir->i_mtime =
9466 new_dir->i_ctime = new_dir->i_mtime =
9467 old_inode->i_ctime = current_time(old_dir);
9469 if (old_dentry->d_parent != new_dentry->d_parent)
9470 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9471 BTRFS_I(old_inode), 1);
9473 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9474 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9476 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9477 BTRFS_I(d_inode(old_dentry)),
9478 old_dentry->d_name.name,
9479 old_dentry->d_name.len);
9481 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9484 btrfs_abort_transaction(trans, ret);
9489 inode_inc_iversion(new_inode);
9490 new_inode->i_ctime = current_time(new_inode);
9491 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9492 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9493 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9494 BUG_ON(new_inode->i_nlink == 0);
9496 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9497 BTRFS_I(d_inode(new_dentry)),
9498 new_dentry->d_name.name,
9499 new_dentry->d_name.len);
9501 if (!ret && new_inode->i_nlink == 0)
9502 ret = btrfs_orphan_add(trans,
9503 BTRFS_I(d_inode(new_dentry)));
9505 btrfs_abort_transaction(trans, ret);
9510 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9511 new_dentry->d_name.name,
9512 new_dentry->d_name.len, 0, index);
9514 btrfs_abort_transaction(trans, ret);
9518 if (old_inode->i_nlink == 1)
9519 BTRFS_I(old_inode)->dir_index = index;
9522 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9523 new_dentry->d_parent);
9524 btrfs_end_log_trans(root);
9528 if (flags & RENAME_WHITEOUT) {
9529 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9533 btrfs_abort_transaction(trans, ret);
9539 * If we have pinned the log and an error happened, we unpin tasks
9540 * trying to sync the log and force them to fallback to a transaction
9541 * commit if the log currently contains any of the inodes involved in
9542 * this rename operation (to ensure we do not persist a log with an
9543 * inconsistent state for any of these inodes or leading to any
9544 * inconsistencies when replayed). If the transaction was aborted, the
9545 * abortion reason is propagated to userspace when attempting to commit
9546 * the transaction. If the log does not contain any of these inodes, we
9547 * allow the tasks to sync it.
9549 if (ret && log_pinned) {
9550 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9551 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9552 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9554 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9555 btrfs_set_log_full_commit(trans);
9557 btrfs_end_log_trans(root);
9560 ret2 = btrfs_end_transaction(trans);
9561 ret = ret ? ret : ret2;
9563 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9564 up_read(&fs_info->subvol_sem);
9569 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9570 struct dentry *old_dentry, struct inode *new_dir,
9571 struct dentry *new_dentry, unsigned int flags)
9573 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9576 if (flags & RENAME_EXCHANGE)
9577 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9580 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9583 struct btrfs_delalloc_work {
9584 struct inode *inode;
9585 struct completion completion;
9586 struct list_head list;
9587 struct btrfs_work work;
9590 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9592 struct btrfs_delalloc_work *delalloc_work;
9593 struct inode *inode;
9595 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9597 inode = delalloc_work->inode;
9598 filemap_flush(inode->i_mapping);
9599 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9600 &BTRFS_I(inode)->runtime_flags))
9601 filemap_flush(inode->i_mapping);
9604 complete(&delalloc_work->completion);
9607 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9609 struct btrfs_delalloc_work *work;
9611 work = kmalloc(sizeof(*work), GFP_NOFS);
9615 init_completion(&work->completion);
9616 INIT_LIST_HEAD(&work->list);
9617 work->inode = inode;
9618 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9624 * some fairly slow code that needs optimization. This walks the list
9625 * of all the inodes with pending delalloc and forces them to disk.
9627 static int start_delalloc_inodes(struct btrfs_root *root,
9628 struct writeback_control *wbc, bool snapshot,
9629 bool in_reclaim_context)
9631 struct btrfs_inode *binode;
9632 struct inode *inode;
9633 struct btrfs_delalloc_work *work, *next;
9634 struct list_head works;
9635 struct list_head splice;
9637 bool full_flush = wbc->nr_to_write == LONG_MAX;
9639 INIT_LIST_HEAD(&works);
9640 INIT_LIST_HEAD(&splice);
9642 mutex_lock(&root->delalloc_mutex);
9643 spin_lock(&root->delalloc_lock);
9644 list_splice_init(&root->delalloc_inodes, &splice);
9645 while (!list_empty(&splice)) {
9646 binode = list_entry(splice.next, struct btrfs_inode,
9649 list_move_tail(&binode->delalloc_inodes,
9650 &root->delalloc_inodes);
9652 if (in_reclaim_context &&
9653 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9656 inode = igrab(&binode->vfs_inode);
9658 cond_resched_lock(&root->delalloc_lock);
9661 spin_unlock(&root->delalloc_lock);
9664 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9665 &binode->runtime_flags);
9667 work = btrfs_alloc_delalloc_work(inode);
9673 list_add_tail(&work->list, &works);
9674 btrfs_queue_work(root->fs_info->flush_workers,
9677 ret = sync_inode(inode, wbc);
9679 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9680 &BTRFS_I(inode)->runtime_flags))
9681 ret = sync_inode(inode, wbc);
9682 btrfs_add_delayed_iput(inode);
9683 if (ret || wbc->nr_to_write <= 0)
9687 spin_lock(&root->delalloc_lock);
9689 spin_unlock(&root->delalloc_lock);
9692 list_for_each_entry_safe(work, next, &works, list) {
9693 list_del_init(&work->list);
9694 wait_for_completion(&work->completion);
9698 if (!list_empty(&splice)) {
9699 spin_lock(&root->delalloc_lock);
9700 list_splice_tail(&splice, &root->delalloc_inodes);
9701 spin_unlock(&root->delalloc_lock);
9703 mutex_unlock(&root->delalloc_mutex);
9707 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9709 struct writeback_control wbc = {
9710 .nr_to_write = LONG_MAX,
9711 .sync_mode = WB_SYNC_NONE,
9713 .range_end = LLONG_MAX,
9715 struct btrfs_fs_info *fs_info = root->fs_info;
9717 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9720 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9723 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9724 bool in_reclaim_context)
9726 struct writeback_control wbc = {
9728 .sync_mode = WB_SYNC_NONE,
9730 .range_end = LLONG_MAX,
9732 struct btrfs_root *root;
9733 struct list_head splice;
9736 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9739 INIT_LIST_HEAD(&splice);
9741 mutex_lock(&fs_info->delalloc_root_mutex);
9742 spin_lock(&fs_info->delalloc_root_lock);
9743 list_splice_init(&fs_info->delalloc_roots, &splice);
9744 while (!list_empty(&splice)) {
9746 * Reset nr_to_write here so we know that we're doing a full
9750 wbc.nr_to_write = LONG_MAX;
9752 root = list_first_entry(&splice, struct btrfs_root,
9754 root = btrfs_grab_root(root);
9756 list_move_tail(&root->delalloc_root,
9757 &fs_info->delalloc_roots);
9758 spin_unlock(&fs_info->delalloc_root_lock);
9760 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9761 btrfs_put_root(root);
9762 if (ret < 0 || wbc.nr_to_write <= 0)
9764 spin_lock(&fs_info->delalloc_root_lock);
9766 spin_unlock(&fs_info->delalloc_root_lock);
9770 if (!list_empty(&splice)) {
9771 spin_lock(&fs_info->delalloc_root_lock);
9772 list_splice_tail(&splice, &fs_info->delalloc_roots);
9773 spin_unlock(&fs_info->delalloc_root_lock);
9775 mutex_unlock(&fs_info->delalloc_root_mutex);
9779 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9780 struct dentry *dentry, const char *symname)
9782 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9783 struct btrfs_trans_handle *trans;
9784 struct btrfs_root *root = BTRFS_I(dir)->root;
9785 struct btrfs_path *path;
9786 struct btrfs_key key;
9787 struct inode *inode = NULL;
9794 struct btrfs_file_extent_item *ei;
9795 struct extent_buffer *leaf;
9797 name_len = strlen(symname);
9798 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9799 return -ENAMETOOLONG;
9802 * 2 items for inode item and ref
9803 * 2 items for dir items
9804 * 1 item for updating parent inode item
9805 * 1 item for the inline extent item
9806 * 1 item for xattr if selinux is on
9808 trans = btrfs_start_transaction(root, 7);
9810 return PTR_ERR(trans);
9812 err = btrfs_get_free_objectid(root, &objectid);
9816 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9817 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9818 objectid, S_IFLNK|S_IRWXUGO, &index);
9819 if (IS_ERR(inode)) {
9820 err = PTR_ERR(inode);
9826 * If the active LSM wants to access the inode during
9827 * d_instantiate it needs these. Smack checks to see
9828 * if the filesystem supports xattrs by looking at the
9831 inode->i_fop = &btrfs_file_operations;
9832 inode->i_op = &btrfs_file_inode_operations;
9833 inode->i_mapping->a_ops = &btrfs_aops;
9835 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9839 path = btrfs_alloc_path();
9844 key.objectid = btrfs_ino(BTRFS_I(inode));
9846 key.type = BTRFS_EXTENT_DATA_KEY;
9847 datasize = btrfs_file_extent_calc_inline_size(name_len);
9848 err = btrfs_insert_empty_item(trans, root, path, &key,
9851 btrfs_free_path(path);
9854 leaf = path->nodes[0];
9855 ei = btrfs_item_ptr(leaf, path->slots[0],
9856 struct btrfs_file_extent_item);
9857 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9858 btrfs_set_file_extent_type(leaf, ei,
9859 BTRFS_FILE_EXTENT_INLINE);
9860 btrfs_set_file_extent_encryption(leaf, ei, 0);
9861 btrfs_set_file_extent_compression(leaf, ei, 0);
9862 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9863 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9865 ptr = btrfs_file_extent_inline_start(ei);
9866 write_extent_buffer(leaf, symname, ptr, name_len);
9867 btrfs_mark_buffer_dirty(leaf);
9868 btrfs_free_path(path);
9870 inode->i_op = &btrfs_symlink_inode_operations;
9871 inode_nohighmem(inode);
9872 inode_set_bytes(inode, name_len);
9873 btrfs_i_size_write(BTRFS_I(inode), name_len);
9874 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9876 * Last step, add directory indexes for our symlink inode. This is the
9877 * last step to avoid extra cleanup of these indexes if an error happens
9881 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9882 BTRFS_I(inode), 0, index);
9886 d_instantiate_new(dentry, inode);
9889 btrfs_end_transaction(trans);
9891 inode_dec_link_count(inode);
9892 discard_new_inode(inode);
9894 btrfs_btree_balance_dirty(fs_info);
9898 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9899 struct btrfs_trans_handle *trans_in,
9900 struct btrfs_inode *inode,
9901 struct btrfs_key *ins,
9904 struct btrfs_file_extent_item stack_fi;
9905 struct btrfs_replace_extent_info extent_info;
9906 struct btrfs_trans_handle *trans = trans_in;
9907 struct btrfs_path *path;
9908 u64 start = ins->objectid;
9909 u64 len = ins->offset;
9910 int qgroup_released;
9913 memset(&stack_fi, 0, sizeof(stack_fi));
9915 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9916 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9917 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9918 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9919 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9920 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9921 /* Encryption and other encoding is reserved and all 0 */
9923 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9924 if (qgroup_released < 0)
9925 return ERR_PTR(qgroup_released);
9928 ret = insert_reserved_file_extent(trans, inode,
9929 file_offset, &stack_fi,
9930 true, qgroup_released);
9936 extent_info.disk_offset = start;
9937 extent_info.disk_len = len;
9938 extent_info.data_offset = 0;
9939 extent_info.data_len = len;
9940 extent_info.file_offset = file_offset;
9941 extent_info.extent_buf = (char *)&stack_fi;
9942 extent_info.is_new_extent = true;
9943 extent_info.qgroup_reserved = qgroup_released;
9944 extent_info.insertions = 0;
9946 path = btrfs_alloc_path();
9952 ret = btrfs_replace_file_extents(inode, path, file_offset,
9953 file_offset + len - 1, &extent_info,
9955 btrfs_free_path(path);
9962 * We have released qgroup data range at the beginning of the function,
9963 * and normally qgroup_released bytes will be freed when committing
9965 * But if we error out early, we have to free what we have released
9966 * or we leak qgroup data reservation.
9968 btrfs_qgroup_free_refroot(inode->root->fs_info,
9969 inode->root->root_key.objectid, qgroup_released,
9970 BTRFS_QGROUP_RSV_DATA);
9971 return ERR_PTR(ret);
9974 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9975 u64 start, u64 num_bytes, u64 min_size,
9976 loff_t actual_len, u64 *alloc_hint,
9977 struct btrfs_trans_handle *trans)
9979 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9980 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9981 struct extent_map *em;
9982 struct btrfs_root *root = BTRFS_I(inode)->root;
9983 struct btrfs_key ins;
9984 u64 cur_offset = start;
9985 u64 clear_offset = start;
9988 u64 last_alloc = (u64)-1;
9990 bool own_trans = true;
9991 u64 end = start + num_bytes - 1;
9995 while (num_bytes > 0) {
9996 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9997 cur_bytes = max(cur_bytes, min_size);
9999 * If we are severely fragmented we could end up with really
10000 * small allocations, so if the allocator is returning small
10001 * chunks lets make its job easier by only searching for those
10004 cur_bytes = min(cur_bytes, last_alloc);
10005 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10006 min_size, 0, *alloc_hint, &ins, 1, 0);
10011 * We've reserved this space, and thus converted it from
10012 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10013 * from here on out we will only need to clear our reservation
10014 * for the remaining unreserved area, so advance our
10015 * clear_offset by our extent size.
10017 clear_offset += ins.offset;
10019 last_alloc = ins.offset;
10020 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10023 * Now that we inserted the prealloc extent we can finally
10024 * decrement the number of reservations in the block group.
10025 * If we did it before, we could race with relocation and have
10026 * relocation miss the reserved extent, making it fail later.
10028 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10029 if (IS_ERR(trans)) {
10030 ret = PTR_ERR(trans);
10031 btrfs_free_reserved_extent(fs_info, ins.objectid,
10036 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10037 cur_offset + ins.offset -1, 0);
10039 em = alloc_extent_map();
10041 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10042 &BTRFS_I(inode)->runtime_flags);
10046 em->start = cur_offset;
10047 em->orig_start = cur_offset;
10048 em->len = ins.offset;
10049 em->block_start = ins.objectid;
10050 em->block_len = ins.offset;
10051 em->orig_block_len = ins.offset;
10052 em->ram_bytes = ins.offset;
10053 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10054 em->generation = trans->transid;
10057 write_lock(&em_tree->lock);
10058 ret = add_extent_mapping(em_tree, em, 1);
10059 write_unlock(&em_tree->lock);
10060 if (ret != -EEXIST)
10062 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10063 cur_offset + ins.offset - 1,
10066 free_extent_map(em);
10068 num_bytes -= ins.offset;
10069 cur_offset += ins.offset;
10070 *alloc_hint = ins.objectid + ins.offset;
10072 inode_inc_iversion(inode);
10073 inode->i_ctime = current_time(inode);
10074 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10075 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10076 (actual_len > inode->i_size) &&
10077 (cur_offset > inode->i_size)) {
10078 if (cur_offset > actual_len)
10079 i_size = actual_len;
10081 i_size = cur_offset;
10082 i_size_write(inode, i_size);
10083 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10086 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10089 btrfs_abort_transaction(trans, ret);
10091 btrfs_end_transaction(trans);
10096 btrfs_end_transaction(trans);
10100 if (clear_offset < end)
10101 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10102 end - clear_offset + 1);
10106 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10107 u64 start, u64 num_bytes, u64 min_size,
10108 loff_t actual_len, u64 *alloc_hint)
10110 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10111 min_size, actual_len, alloc_hint,
10115 int btrfs_prealloc_file_range_trans(struct inode *inode,
10116 struct btrfs_trans_handle *trans, int mode,
10117 u64 start, u64 num_bytes, u64 min_size,
10118 loff_t actual_len, u64 *alloc_hint)
10120 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10121 min_size, actual_len, alloc_hint, trans);
10124 static int btrfs_set_page_dirty(struct page *page)
10126 return __set_page_dirty_nobuffers(page);
10129 static int btrfs_permission(struct user_namespace *mnt_userns,
10130 struct inode *inode, int mask)
10132 struct btrfs_root *root = BTRFS_I(inode)->root;
10133 umode_t mode = inode->i_mode;
10135 if (mask & MAY_WRITE &&
10136 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10137 if (btrfs_root_readonly(root))
10139 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10142 return generic_permission(&init_user_ns, inode, mask);
10145 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10146 struct dentry *dentry, umode_t mode)
10148 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10149 struct btrfs_trans_handle *trans;
10150 struct btrfs_root *root = BTRFS_I(dir)->root;
10151 struct inode *inode = NULL;
10157 * 5 units required for adding orphan entry
10159 trans = btrfs_start_transaction(root, 5);
10161 return PTR_ERR(trans);
10163 ret = btrfs_get_free_objectid(root, &objectid);
10167 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10168 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10169 if (IS_ERR(inode)) {
10170 ret = PTR_ERR(inode);
10175 inode->i_fop = &btrfs_file_operations;
10176 inode->i_op = &btrfs_file_inode_operations;
10178 inode->i_mapping->a_ops = &btrfs_aops;
10180 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10184 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10187 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10192 * We set number of links to 0 in btrfs_new_inode(), and here we set
10193 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10196 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10198 set_nlink(inode, 1);
10199 d_tmpfile(dentry, inode);
10200 unlock_new_inode(inode);
10201 mark_inode_dirty(inode);
10203 btrfs_end_transaction(trans);
10205 discard_new_inode(inode);
10206 btrfs_btree_balance_dirty(fs_info);
10210 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10212 struct inode *inode = tree->private_data;
10213 unsigned long index = start >> PAGE_SHIFT;
10214 unsigned long end_index = end >> PAGE_SHIFT;
10217 while (index <= end_index) {
10218 page = find_get_page(inode->i_mapping, index);
10219 ASSERT(page); /* Pages should be in the extent_io_tree */
10220 set_page_writeback(page);
10228 * Add an entry indicating a block group or device which is pinned by a
10229 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10230 * negative errno on failure.
10232 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10233 bool is_block_group)
10235 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10236 struct btrfs_swapfile_pin *sp, *entry;
10237 struct rb_node **p;
10238 struct rb_node *parent = NULL;
10240 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10245 sp->is_block_group = is_block_group;
10246 sp->bg_extent_count = 1;
10248 spin_lock(&fs_info->swapfile_pins_lock);
10249 p = &fs_info->swapfile_pins.rb_node;
10252 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10253 if (sp->ptr < entry->ptr ||
10254 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10255 p = &(*p)->rb_left;
10256 } else if (sp->ptr > entry->ptr ||
10257 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10258 p = &(*p)->rb_right;
10260 if (is_block_group)
10261 entry->bg_extent_count++;
10262 spin_unlock(&fs_info->swapfile_pins_lock);
10267 rb_link_node(&sp->node, parent, p);
10268 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10269 spin_unlock(&fs_info->swapfile_pins_lock);
10273 /* Free all of the entries pinned by this swapfile. */
10274 static void btrfs_free_swapfile_pins(struct inode *inode)
10276 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10277 struct btrfs_swapfile_pin *sp;
10278 struct rb_node *node, *next;
10280 spin_lock(&fs_info->swapfile_pins_lock);
10281 node = rb_first(&fs_info->swapfile_pins);
10283 next = rb_next(node);
10284 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10285 if (sp->inode == inode) {
10286 rb_erase(&sp->node, &fs_info->swapfile_pins);
10287 if (sp->is_block_group) {
10288 btrfs_dec_block_group_swap_extents(sp->ptr,
10289 sp->bg_extent_count);
10290 btrfs_put_block_group(sp->ptr);
10296 spin_unlock(&fs_info->swapfile_pins_lock);
10299 struct btrfs_swap_info {
10305 unsigned long nr_pages;
10309 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10310 struct btrfs_swap_info *bsi)
10312 unsigned long nr_pages;
10313 u64 first_ppage, first_ppage_reported, next_ppage;
10316 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10317 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10318 PAGE_SIZE) >> PAGE_SHIFT;
10320 if (first_ppage >= next_ppage)
10322 nr_pages = next_ppage - first_ppage;
10324 first_ppage_reported = first_ppage;
10325 if (bsi->start == 0)
10326 first_ppage_reported++;
10327 if (bsi->lowest_ppage > first_ppage_reported)
10328 bsi->lowest_ppage = first_ppage_reported;
10329 if (bsi->highest_ppage < (next_ppage - 1))
10330 bsi->highest_ppage = next_ppage - 1;
10332 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10335 bsi->nr_extents += ret;
10336 bsi->nr_pages += nr_pages;
10340 static void btrfs_swap_deactivate(struct file *file)
10342 struct inode *inode = file_inode(file);
10344 btrfs_free_swapfile_pins(inode);
10345 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10348 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10351 struct inode *inode = file_inode(file);
10352 struct btrfs_root *root = BTRFS_I(inode)->root;
10353 struct btrfs_fs_info *fs_info = root->fs_info;
10354 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10355 struct extent_state *cached_state = NULL;
10356 struct extent_map *em = NULL;
10357 struct btrfs_device *device = NULL;
10358 struct btrfs_swap_info bsi = {
10359 .lowest_ppage = (sector_t)-1ULL,
10366 * If the swap file was just created, make sure delalloc is done. If the
10367 * file changes again after this, the user is doing something stupid and
10368 * we don't really care.
10370 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10375 * The inode is locked, so these flags won't change after we check them.
10377 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10378 btrfs_warn(fs_info, "swapfile must not be compressed");
10381 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10382 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10385 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10386 btrfs_warn(fs_info, "swapfile must not be checksummed");
10391 * Balance or device remove/replace/resize can move stuff around from
10392 * under us. The exclop protection makes sure they aren't running/won't
10393 * run concurrently while we are mapping the swap extents, and
10394 * fs_info->swapfile_pins prevents them from running while the swap
10395 * file is active and moving the extents. Note that this also prevents
10396 * a concurrent device add which isn't actually necessary, but it's not
10397 * really worth the trouble to allow it.
10399 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10400 btrfs_warn(fs_info,
10401 "cannot activate swapfile while exclusive operation is running");
10406 * Prevent snapshot creation while we are activating the swap file.
10407 * We do not want to race with snapshot creation. If snapshot creation
10408 * already started before we bumped nr_swapfiles from 0 to 1 and
10409 * completes before the first write into the swap file after it is
10410 * activated, than that write would fallback to COW.
10412 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10413 btrfs_exclop_finish(fs_info);
10414 btrfs_warn(fs_info,
10415 "cannot activate swapfile because snapshot creation is in progress");
10419 * Snapshots can create extents which require COW even if NODATACOW is
10420 * set. We use this counter to prevent snapshots. We must increment it
10421 * before walking the extents because we don't want a concurrent
10422 * snapshot to run after we've already checked the extents.
10424 atomic_inc(&root->nr_swapfiles);
10426 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10428 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10430 while (start < isize) {
10431 u64 logical_block_start, physical_block_start;
10432 struct btrfs_block_group *bg;
10433 u64 len = isize - start;
10435 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10441 if (em->block_start == EXTENT_MAP_HOLE) {
10442 btrfs_warn(fs_info, "swapfile must not have holes");
10446 if (em->block_start == EXTENT_MAP_INLINE) {
10448 * It's unlikely we'll ever actually find ourselves
10449 * here, as a file small enough to fit inline won't be
10450 * big enough to store more than the swap header, but in
10451 * case something changes in the future, let's catch it
10452 * here rather than later.
10454 btrfs_warn(fs_info, "swapfile must not be inline");
10458 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10459 btrfs_warn(fs_info, "swapfile must not be compressed");
10464 logical_block_start = em->block_start + (start - em->start);
10465 len = min(len, em->len - (start - em->start));
10466 free_extent_map(em);
10469 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10475 btrfs_warn(fs_info,
10476 "swapfile must not be copy-on-write");
10481 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10487 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10488 btrfs_warn(fs_info,
10489 "swapfile must have single data profile");
10494 if (device == NULL) {
10495 device = em->map_lookup->stripes[0].dev;
10496 ret = btrfs_add_swapfile_pin(inode, device, false);
10501 } else if (device != em->map_lookup->stripes[0].dev) {
10502 btrfs_warn(fs_info, "swapfile must be on one device");
10507 physical_block_start = (em->map_lookup->stripes[0].physical +
10508 (logical_block_start - em->start));
10509 len = min(len, em->len - (logical_block_start - em->start));
10510 free_extent_map(em);
10513 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10515 btrfs_warn(fs_info,
10516 "could not find block group containing swapfile");
10521 if (!btrfs_inc_block_group_swap_extents(bg)) {
10522 btrfs_warn(fs_info,
10523 "block group for swapfile at %llu is read-only%s",
10525 atomic_read(&fs_info->scrubs_running) ?
10526 " (scrub running)" : "");
10527 btrfs_put_block_group(bg);
10532 ret = btrfs_add_swapfile_pin(inode, bg, true);
10534 btrfs_put_block_group(bg);
10541 if (bsi.block_len &&
10542 bsi.block_start + bsi.block_len == physical_block_start) {
10543 bsi.block_len += len;
10545 if (bsi.block_len) {
10546 ret = btrfs_add_swap_extent(sis, &bsi);
10551 bsi.block_start = physical_block_start;
10552 bsi.block_len = len;
10559 ret = btrfs_add_swap_extent(sis, &bsi);
10562 if (!IS_ERR_OR_NULL(em))
10563 free_extent_map(em);
10565 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10568 btrfs_swap_deactivate(file);
10570 btrfs_drew_write_unlock(&root->snapshot_lock);
10572 btrfs_exclop_finish(fs_info);
10578 sis->bdev = device->bdev;
10579 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10580 sis->max = bsi.nr_pages;
10581 sis->pages = bsi.nr_pages - 1;
10582 sis->highest_bit = bsi.nr_pages - 1;
10583 return bsi.nr_extents;
10586 static void btrfs_swap_deactivate(struct file *file)
10590 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10593 return -EOPNOTSUPP;
10598 * Update the number of bytes used in the VFS' inode. When we replace extents in
10599 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10600 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10601 * always get a correct value.
10603 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10604 const u64 add_bytes,
10605 const u64 del_bytes)
10607 if (add_bytes == del_bytes)
10610 spin_lock(&inode->lock);
10612 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10614 inode_add_bytes(&inode->vfs_inode, add_bytes);
10615 spin_unlock(&inode->lock);
10618 static const struct inode_operations btrfs_dir_inode_operations = {
10619 .getattr = btrfs_getattr,
10620 .lookup = btrfs_lookup,
10621 .create = btrfs_create,
10622 .unlink = btrfs_unlink,
10623 .link = btrfs_link,
10624 .mkdir = btrfs_mkdir,
10625 .rmdir = btrfs_rmdir,
10626 .rename = btrfs_rename2,
10627 .symlink = btrfs_symlink,
10628 .setattr = btrfs_setattr,
10629 .mknod = btrfs_mknod,
10630 .listxattr = btrfs_listxattr,
10631 .permission = btrfs_permission,
10632 .get_acl = btrfs_get_acl,
10633 .set_acl = btrfs_set_acl,
10634 .update_time = btrfs_update_time,
10635 .tmpfile = btrfs_tmpfile,
10638 static const struct file_operations btrfs_dir_file_operations = {
10639 .llseek = generic_file_llseek,
10640 .read = generic_read_dir,
10641 .iterate_shared = btrfs_real_readdir,
10642 .open = btrfs_opendir,
10643 .unlocked_ioctl = btrfs_ioctl,
10644 #ifdef CONFIG_COMPAT
10645 .compat_ioctl = btrfs_compat_ioctl,
10647 .release = btrfs_release_file,
10648 .fsync = btrfs_sync_file,
10652 * btrfs doesn't support the bmap operation because swapfiles
10653 * use bmap to make a mapping of extents in the file. They assume
10654 * these extents won't change over the life of the file and they
10655 * use the bmap result to do IO directly to the drive.
10657 * the btrfs bmap call would return logical addresses that aren't
10658 * suitable for IO and they also will change frequently as COW
10659 * operations happen. So, swapfile + btrfs == corruption.
10661 * For now we're avoiding this by dropping bmap.
10663 static const struct address_space_operations btrfs_aops = {
10664 .readpage = btrfs_readpage,
10665 .writepage = btrfs_writepage,
10666 .writepages = btrfs_writepages,
10667 .readahead = btrfs_readahead,
10668 .direct_IO = noop_direct_IO,
10669 .invalidatepage = btrfs_invalidatepage,
10670 .releasepage = btrfs_releasepage,
10671 #ifdef CONFIG_MIGRATION
10672 .migratepage = btrfs_migratepage,
10674 .set_page_dirty = btrfs_set_page_dirty,
10675 .error_remove_page = generic_error_remove_page,
10676 .swap_activate = btrfs_swap_activate,
10677 .swap_deactivate = btrfs_swap_deactivate,
10680 static const struct inode_operations btrfs_file_inode_operations = {
10681 .getattr = btrfs_getattr,
10682 .setattr = btrfs_setattr,
10683 .listxattr = btrfs_listxattr,
10684 .permission = btrfs_permission,
10685 .fiemap = btrfs_fiemap,
10686 .get_acl = btrfs_get_acl,
10687 .set_acl = btrfs_set_acl,
10688 .update_time = btrfs_update_time,
10690 static const struct inode_operations btrfs_special_inode_operations = {
10691 .getattr = btrfs_getattr,
10692 .setattr = btrfs_setattr,
10693 .permission = btrfs_permission,
10694 .listxattr = btrfs_listxattr,
10695 .get_acl = btrfs_get_acl,
10696 .set_acl = btrfs_set_acl,
10697 .update_time = btrfs_update_time,
10699 static const struct inode_operations btrfs_symlink_inode_operations = {
10700 .get_link = page_get_link,
10701 .getattr = btrfs_getattr,
10702 .setattr = btrfs_setattr,
10703 .permission = btrfs_permission,
10704 .listxattr = btrfs_listxattr,
10705 .update_time = btrfs_update_time,
10708 const struct dentry_operations btrfs_dentry_operations = {
10709 .d_delete = btrfs_dentry_delete,