1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <linux/migrate.h>
32 #include <linux/sched/mm.h>
33 #include <linux/iomap.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "ordered-data.h"
45 #include "compression.h"
47 #include "free-space-cache.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 struct btrfs_dio_data {
64 struct extent_changeset *data_reserved;
67 static const struct inode_operations btrfs_dir_inode_operations;
68 static const struct inode_operations btrfs_symlink_inode_operations;
69 static const struct inode_operations btrfs_special_inode_operations;
70 static const struct inode_operations btrfs_file_inode_operations;
71 static const struct address_space_operations btrfs_aops;
72 static const struct file_operations btrfs_dir_file_operations;
74 static struct kmem_cache *btrfs_inode_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
78 struct kmem_cache *btrfs_free_space_bitmap_cachep;
80 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
81 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
82 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
83 static noinline int cow_file_range(struct btrfs_inode *inode,
84 struct page *locked_page,
85 u64 start, u64 end, int *page_started,
86 unsigned long *nr_written, int unlock);
87 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
88 u64 len, u64 orig_start, u64 block_start,
89 u64 block_len, u64 orig_block_len,
90 u64 ram_bytes, int compress_type,
93 static void __endio_write_update_ordered(struct btrfs_inode *inode,
94 const u64 offset, const u64 bytes,
98 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
100 * ilock_flags can have the following bit set:
102 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
103 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
105 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
107 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
109 if (ilock_flags & BTRFS_ILOCK_SHARED) {
110 if (ilock_flags & BTRFS_ILOCK_TRY) {
111 if (!inode_trylock_shared(inode))
116 inode_lock_shared(inode);
118 if (ilock_flags & BTRFS_ILOCK_TRY) {
119 if (!inode_trylock(inode))
126 if (ilock_flags & BTRFS_ILOCK_MMAP)
127 down_write(&BTRFS_I(inode)->i_mmap_lock);
132 * btrfs_inode_unlock - unock inode i_rwsem
134 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
135 * to decide whether the lock acquired is shared or exclusive.
137 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
139 if (ilock_flags & BTRFS_ILOCK_MMAP)
140 up_write(&BTRFS_I(inode)->i_mmap_lock);
141 if (ilock_flags & BTRFS_ILOCK_SHARED)
142 inode_unlock_shared(inode);
148 * Cleanup all submitted ordered extents in specified range to handle errors
149 * from the btrfs_run_delalloc_range() callback.
151 * NOTE: caller must ensure that when an error happens, it can not call
152 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
153 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
154 * to be released, which we want to happen only when finishing the ordered
155 * extent (btrfs_finish_ordered_io()).
157 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
158 struct page *locked_page,
159 u64 offset, u64 bytes)
161 unsigned long index = offset >> PAGE_SHIFT;
162 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
163 u64 page_start = page_offset(locked_page);
164 u64 page_end = page_start + PAGE_SIZE - 1;
168 while (index <= end_index) {
169 page = find_get_page(inode->vfs_inode.i_mapping, index);
173 ClearPagePrivate2(page);
178 * In case this page belongs to the delalloc range being instantiated
179 * then skip it, since the first page of a range is going to be
180 * properly cleaned up by the caller of run_delalloc_range
182 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
187 return __endio_write_update_ordered(inode, offset, bytes, false);
190 static int btrfs_dirty_inode(struct inode *inode);
192 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
193 struct inode *inode, struct inode *dir,
194 const struct qstr *qstr)
198 err = btrfs_init_acl(trans, inode, dir);
200 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
205 * this does all the hard work for inserting an inline extent into
206 * the btree. The caller should have done a btrfs_drop_extents so that
207 * no overlapping inline items exist in the btree
209 static int insert_inline_extent(struct btrfs_trans_handle *trans,
210 struct btrfs_path *path, bool extent_inserted,
211 struct btrfs_root *root, struct inode *inode,
212 u64 start, size_t size, size_t compressed_size,
214 struct page **compressed_pages)
216 struct extent_buffer *leaf;
217 struct page *page = NULL;
220 struct btrfs_file_extent_item *ei;
222 size_t cur_size = size;
223 unsigned long offset;
225 ASSERT((compressed_size > 0 && compressed_pages) ||
226 (compressed_size == 0 && !compressed_pages));
228 if (compressed_size && compressed_pages)
229 cur_size = compressed_size;
231 if (!extent_inserted) {
232 struct btrfs_key key;
235 key.objectid = btrfs_ino(BTRFS_I(inode));
237 key.type = BTRFS_EXTENT_DATA_KEY;
239 datasize = btrfs_file_extent_calc_inline_size(cur_size);
240 ret = btrfs_insert_empty_item(trans, root, path, &key,
245 leaf = path->nodes[0];
246 ei = btrfs_item_ptr(leaf, path->slots[0],
247 struct btrfs_file_extent_item);
248 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
249 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
250 btrfs_set_file_extent_encryption(leaf, ei, 0);
251 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
252 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
253 ptr = btrfs_file_extent_inline_start(ei);
255 if (compress_type != BTRFS_COMPRESS_NONE) {
258 while (compressed_size > 0) {
259 cpage = compressed_pages[i];
260 cur_size = min_t(unsigned long, compressed_size,
263 kaddr = kmap_atomic(cpage);
264 write_extent_buffer(leaf, kaddr, ptr, cur_size);
265 kunmap_atomic(kaddr);
269 compressed_size -= cur_size;
271 btrfs_set_file_extent_compression(leaf, ei,
274 page = find_get_page(inode->i_mapping,
275 start >> PAGE_SHIFT);
276 btrfs_set_file_extent_compression(leaf, ei, 0);
277 kaddr = kmap_atomic(page);
278 offset = offset_in_page(start);
279 write_extent_buffer(leaf, kaddr + offset, ptr, size);
280 kunmap_atomic(kaddr);
283 btrfs_mark_buffer_dirty(leaf);
284 btrfs_release_path(path);
287 * We align size to sectorsize for inline extents just for simplicity
290 size = ALIGN(size, root->fs_info->sectorsize);
291 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
296 * we're an inline extent, so nobody can
297 * extend the file past i_size without locking
298 * a page we already have locked.
300 * We must do any isize and inode updates
301 * before we unlock the pages. Otherwise we
302 * could end up racing with unlink.
304 BTRFS_I(inode)->disk_i_size = inode->i_size;
311 * conditionally insert an inline extent into the file. This
312 * does the checks required to make sure the data is small enough
313 * to fit as an inline extent.
315 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
316 u64 end, size_t compressed_size,
318 struct page **compressed_pages)
320 struct btrfs_drop_extents_args drop_args = { 0 };
321 struct btrfs_root *root = inode->root;
322 struct btrfs_fs_info *fs_info = root->fs_info;
323 struct btrfs_trans_handle *trans;
324 u64 isize = i_size_read(&inode->vfs_inode);
325 u64 actual_end = min(end + 1, isize);
326 u64 inline_len = actual_end - start;
327 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
328 u64 data_len = inline_len;
330 struct btrfs_path *path;
333 data_len = compressed_size;
336 actual_end > fs_info->sectorsize ||
337 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
339 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
341 data_len > fs_info->max_inline) {
345 path = btrfs_alloc_path();
349 trans = btrfs_join_transaction(root);
351 btrfs_free_path(path);
352 return PTR_ERR(trans);
354 trans->block_rsv = &inode->block_rsv;
356 drop_args.path = path;
357 drop_args.start = start;
358 drop_args.end = aligned_end;
359 drop_args.drop_cache = true;
360 drop_args.replace_extent = true;
362 if (compressed_size && compressed_pages)
363 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
366 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
369 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
371 btrfs_abort_transaction(trans, ret);
375 if (isize > actual_end)
376 inline_len = min_t(u64, isize, actual_end);
377 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
378 root, &inode->vfs_inode, start,
379 inline_len, compressed_size,
380 compress_type, compressed_pages);
381 if (ret && ret != -ENOSPC) {
382 btrfs_abort_transaction(trans, ret);
384 } else if (ret == -ENOSPC) {
389 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
390 ret = btrfs_update_inode(trans, root, inode);
391 if (ret && ret != -ENOSPC) {
392 btrfs_abort_transaction(trans, ret);
394 } else if (ret == -ENOSPC) {
399 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
402 * Don't forget to free the reserved space, as for inlined extent
403 * it won't count as data extent, free them directly here.
404 * And at reserve time, it's always aligned to page size, so
405 * just free one page here.
407 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
408 btrfs_free_path(path);
409 btrfs_end_transaction(trans);
413 struct async_extent {
418 unsigned long nr_pages;
420 struct list_head list;
425 struct page *locked_page;
428 unsigned int write_flags;
429 struct list_head extents;
430 struct cgroup_subsys_state *blkcg_css;
431 struct btrfs_work work;
436 /* Number of chunks in flight; must be first in the structure */
438 struct async_chunk chunks[];
441 static noinline int add_async_extent(struct async_chunk *cow,
442 u64 start, u64 ram_size,
445 unsigned long nr_pages,
448 struct async_extent *async_extent;
450 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
451 BUG_ON(!async_extent); /* -ENOMEM */
452 async_extent->start = start;
453 async_extent->ram_size = ram_size;
454 async_extent->compressed_size = compressed_size;
455 async_extent->pages = pages;
456 async_extent->nr_pages = nr_pages;
457 async_extent->compress_type = compress_type;
458 list_add_tail(&async_extent->list, &cow->extents);
463 * Check if the inode has flags compatible with compression
465 static inline bool inode_can_compress(struct btrfs_inode *inode)
467 if (inode->flags & BTRFS_INODE_NODATACOW ||
468 inode->flags & BTRFS_INODE_NODATASUM)
474 * Check if the inode needs to be submitted to compression, based on mount
475 * options, defragmentation, properties or heuristics.
477 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
480 struct btrfs_fs_info *fs_info = inode->root->fs_info;
482 if (!inode_can_compress(inode)) {
483 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
484 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
489 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
492 if (inode->defrag_compress)
494 /* bad compression ratios */
495 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
497 if (btrfs_test_opt(fs_info, COMPRESS) ||
498 inode->flags & BTRFS_INODE_COMPRESS ||
499 inode->prop_compress)
500 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
504 static inline void inode_should_defrag(struct btrfs_inode *inode,
505 u64 start, u64 end, u64 num_bytes, u64 small_write)
507 /* If this is a small write inside eof, kick off a defrag */
508 if (num_bytes < small_write &&
509 (start > 0 || end + 1 < inode->disk_i_size))
510 btrfs_add_inode_defrag(NULL, inode);
514 * we create compressed extents in two phases. The first
515 * phase compresses a range of pages that have already been
516 * locked (both pages and state bits are locked).
518 * This is done inside an ordered work queue, and the compression
519 * is spread across many cpus. The actual IO submission is step
520 * two, and the ordered work queue takes care of making sure that
521 * happens in the same order things were put onto the queue by
522 * writepages and friends.
524 * If this code finds it can't get good compression, it puts an
525 * entry onto the work queue to write the uncompressed bytes. This
526 * makes sure that both compressed inodes and uncompressed inodes
527 * are written in the same order that the flusher thread sent them
530 static noinline int compress_file_range(struct async_chunk *async_chunk)
532 struct inode *inode = async_chunk->inode;
533 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
534 u64 blocksize = fs_info->sectorsize;
535 u64 start = async_chunk->start;
536 u64 end = async_chunk->end;
540 struct page **pages = NULL;
541 unsigned long nr_pages;
542 unsigned long total_compressed = 0;
543 unsigned long total_in = 0;
546 int compress_type = fs_info->compress_type;
547 int compressed_extents = 0;
550 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
554 * We need to save i_size before now because it could change in between
555 * us evaluating the size and assigning it. This is because we lock and
556 * unlock the page in truncate and fallocate, and then modify the i_size
559 * The barriers are to emulate READ_ONCE, remove that once i_size_read
563 i_size = i_size_read(inode);
565 actual_end = min_t(u64, i_size, end + 1);
568 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
569 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
570 nr_pages = min_t(unsigned long, nr_pages,
571 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
574 * we don't want to send crud past the end of i_size through
575 * compression, that's just a waste of CPU time. So, if the
576 * end of the file is before the start of our current
577 * requested range of bytes, we bail out to the uncompressed
578 * cleanup code that can deal with all of this.
580 * It isn't really the fastest way to fix things, but this is a
581 * very uncommon corner.
583 if (actual_end <= start)
584 goto cleanup_and_bail_uncompressed;
586 total_compressed = actual_end - start;
589 * skip compression for a small file range(<=blocksize) that
590 * isn't an inline extent, since it doesn't save disk space at all.
592 if (total_compressed <= blocksize &&
593 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
594 goto cleanup_and_bail_uncompressed;
596 total_compressed = min_t(unsigned long, total_compressed,
597 BTRFS_MAX_UNCOMPRESSED);
602 * we do compression for mount -o compress and when the
603 * inode has not been flagged as nocompress. This flag can
604 * change at any time if we discover bad compression ratios.
606 if (inode_need_compress(BTRFS_I(inode), start, end)) {
608 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
610 /* just bail out to the uncompressed code */
615 if (BTRFS_I(inode)->defrag_compress)
616 compress_type = BTRFS_I(inode)->defrag_compress;
617 else if (BTRFS_I(inode)->prop_compress)
618 compress_type = BTRFS_I(inode)->prop_compress;
621 * we need to call clear_page_dirty_for_io on each
622 * page in the range. Otherwise applications with the file
623 * mmap'd can wander in and change the page contents while
624 * we are compressing them.
626 * If the compression fails for any reason, we set the pages
627 * dirty again later on.
629 * Note that the remaining part is redirtied, the start pointer
630 * has moved, the end is the original one.
633 extent_range_clear_dirty_for_io(inode, start, end);
637 /* Compression level is applied here and only here */
638 ret = btrfs_compress_pages(
639 compress_type | (fs_info->compress_level << 4),
640 inode->i_mapping, start,
647 unsigned long offset = offset_in_page(total_compressed);
648 struct page *page = pages[nr_pages - 1];
650 /* zero the tail end of the last page, we might be
651 * sending it down to disk
654 memzero_page(page, offset, PAGE_SIZE - offset);
660 /* lets try to make an inline extent */
661 if (ret || total_in < actual_end) {
662 /* we didn't compress the entire range, try
663 * to make an uncompressed inline extent.
665 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
666 0, BTRFS_COMPRESS_NONE,
669 /* try making a compressed inline extent */
670 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
672 compress_type, pages);
675 unsigned long clear_flags = EXTENT_DELALLOC |
676 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
677 EXTENT_DO_ACCOUNTING;
678 unsigned long page_error_op;
680 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
683 * inline extent creation worked or returned error,
684 * we don't need to create any more async work items.
685 * Unlock and free up our temp pages.
687 * We use DO_ACCOUNTING here because we need the
688 * delalloc_release_metadata to be done _after_ we drop
689 * our outstanding extent for clearing delalloc for this
692 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
696 PAGE_START_WRITEBACK |
701 * Ensure we only free the compressed pages if we have
702 * them allocated, as we can still reach here with
703 * inode_need_compress() == false.
706 for (i = 0; i < nr_pages; i++) {
707 WARN_ON(pages[i]->mapping);
718 * we aren't doing an inline extent round the compressed size
719 * up to a block size boundary so the allocator does sane
722 total_compressed = ALIGN(total_compressed, blocksize);
725 * one last check to make sure the compression is really a
726 * win, compare the page count read with the blocks on disk,
727 * compression must free at least one sector size
729 total_in = ALIGN(total_in, PAGE_SIZE);
730 if (total_compressed + blocksize <= total_in) {
731 compressed_extents++;
734 * The async work queues will take care of doing actual
735 * allocation on disk for these compressed pages, and
736 * will submit them to the elevator.
738 add_async_extent(async_chunk, start, total_in,
739 total_compressed, pages, nr_pages,
742 if (start + total_in < end) {
748 return compressed_extents;
753 * the compression code ran but failed to make things smaller,
754 * free any pages it allocated and our page pointer array
756 for (i = 0; i < nr_pages; i++) {
757 WARN_ON(pages[i]->mapping);
762 total_compressed = 0;
765 /* flag the file so we don't compress in the future */
766 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
767 !(BTRFS_I(inode)->prop_compress)) {
768 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
771 cleanup_and_bail_uncompressed:
773 * No compression, but we still need to write the pages in the file
774 * we've been given so far. redirty the locked page if it corresponds
775 * to our extent and set things up for the async work queue to run
776 * cow_file_range to do the normal delalloc dance.
778 if (async_chunk->locked_page &&
779 (page_offset(async_chunk->locked_page) >= start &&
780 page_offset(async_chunk->locked_page)) <= end) {
781 __set_page_dirty_nobuffers(async_chunk->locked_page);
782 /* unlocked later on in the async handlers */
786 extent_range_redirty_for_io(inode, start, end);
787 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
788 BTRFS_COMPRESS_NONE);
789 compressed_extents++;
791 return compressed_extents;
794 static void free_async_extent_pages(struct async_extent *async_extent)
798 if (!async_extent->pages)
801 for (i = 0; i < async_extent->nr_pages; i++) {
802 WARN_ON(async_extent->pages[i]->mapping);
803 put_page(async_extent->pages[i]);
805 kfree(async_extent->pages);
806 async_extent->nr_pages = 0;
807 async_extent->pages = NULL;
811 * phase two of compressed writeback. This is the ordered portion
812 * of the code, which only gets called in the order the work was
813 * queued. We walk all the async extents created by compress_file_range
814 * and send them down to the disk.
816 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
818 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
819 struct btrfs_fs_info *fs_info = inode->root->fs_info;
820 struct async_extent *async_extent;
822 struct btrfs_key ins;
823 struct extent_map *em;
824 struct btrfs_root *root = inode->root;
825 struct extent_io_tree *io_tree = &inode->io_tree;
829 while (!list_empty(&async_chunk->extents)) {
830 async_extent = list_entry(async_chunk->extents.next,
831 struct async_extent, list);
832 list_del(&async_extent->list);
835 lock_extent(io_tree, async_extent->start,
836 async_extent->start + async_extent->ram_size - 1);
837 /* did the compression code fall back to uncompressed IO? */
838 if (!async_extent->pages) {
839 int page_started = 0;
840 unsigned long nr_written = 0;
842 /* allocate blocks */
843 ret = cow_file_range(inode, async_chunk->locked_page,
845 async_extent->start +
846 async_extent->ram_size - 1,
847 &page_started, &nr_written, 0);
852 * if page_started, cow_file_range inserted an
853 * inline extent and took care of all the unlocking
854 * and IO for us. Otherwise, we need to submit
855 * all those pages down to the drive.
857 if (!page_started && !ret)
858 extent_write_locked_range(&inode->vfs_inode,
860 async_extent->start +
861 async_extent->ram_size - 1,
863 else if (ret && async_chunk->locked_page)
864 unlock_page(async_chunk->locked_page);
870 ret = btrfs_reserve_extent(root, async_extent->ram_size,
871 async_extent->compressed_size,
872 async_extent->compressed_size,
873 0, alloc_hint, &ins, 1, 1);
875 free_async_extent_pages(async_extent);
877 if (ret == -ENOSPC) {
878 unlock_extent(io_tree, async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1);
883 * we need to redirty the pages if we decide to
884 * fallback to uncompressed IO, otherwise we
885 * will not submit these pages down to lower
888 extent_range_redirty_for_io(&inode->vfs_inode,
890 async_extent->start +
891 async_extent->ram_size - 1);
898 * here we're doing allocation and writeback of the
901 em = create_io_em(inode, async_extent->start,
902 async_extent->ram_size, /* len */
903 async_extent->start, /* orig_start */
904 ins.objectid, /* block_start */
905 ins.offset, /* block_len */
906 ins.offset, /* orig_block_len */
907 async_extent->ram_size, /* ram_bytes */
908 async_extent->compress_type,
909 BTRFS_ORDERED_COMPRESSED);
911 /* ret value is not necessary due to void function */
912 goto out_free_reserve;
915 ret = btrfs_add_ordered_extent_compress(inode,
918 async_extent->ram_size,
920 async_extent->compress_type);
922 btrfs_drop_extent_cache(inode, async_extent->start,
923 async_extent->start +
924 async_extent->ram_size - 1, 0);
925 goto out_free_reserve;
927 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
930 * clear dirty, set writeback and unlock the pages.
932 extent_clear_unlock_delalloc(inode, async_extent->start,
933 async_extent->start +
934 async_extent->ram_size - 1,
935 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
936 PAGE_UNLOCK | PAGE_START_WRITEBACK);
937 if (btrfs_submit_compressed_write(inode, async_extent->start,
938 async_extent->ram_size,
940 ins.offset, async_extent->pages,
941 async_extent->nr_pages,
942 async_chunk->write_flags,
943 async_chunk->blkcg_css)) {
944 struct page *p = async_extent->pages[0];
945 const u64 start = async_extent->start;
946 const u64 end = start + async_extent->ram_size - 1;
948 p->mapping = inode->vfs_inode.i_mapping;
949 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
952 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
955 free_async_extent_pages(async_extent);
957 alloc_hint = ins.objectid + ins.offset;
963 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
964 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
966 extent_clear_unlock_delalloc(inode, async_extent->start,
967 async_extent->start +
968 async_extent->ram_size - 1,
969 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
970 EXTENT_DELALLOC_NEW |
971 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
972 PAGE_UNLOCK | PAGE_START_WRITEBACK |
973 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
974 free_async_extent_pages(async_extent);
979 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
982 struct extent_map_tree *em_tree = &inode->extent_tree;
983 struct extent_map *em;
986 read_lock(&em_tree->lock);
987 em = search_extent_mapping(em_tree, start, num_bytes);
990 * if block start isn't an actual block number then find the
991 * first block in this inode and use that as a hint. If that
992 * block is also bogus then just don't worry about it.
994 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
996 em = search_extent_mapping(em_tree, 0, 0);
997 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
998 alloc_hint = em->block_start;
1000 free_extent_map(em);
1002 alloc_hint = em->block_start;
1003 free_extent_map(em);
1006 read_unlock(&em_tree->lock);
1012 * when extent_io.c finds a delayed allocation range in the file,
1013 * the call backs end up in this code. The basic idea is to
1014 * allocate extents on disk for the range, and create ordered data structs
1015 * in ram to track those extents.
1017 * locked_page is the page that writepage had locked already. We use
1018 * it to make sure we don't do extra locks or unlocks.
1020 * *page_started is set to one if we unlock locked_page and do everything
1021 * required to start IO on it. It may be clean and already done with
1022 * IO when we return.
1024 static noinline int cow_file_range(struct btrfs_inode *inode,
1025 struct page *locked_page,
1026 u64 start, u64 end, int *page_started,
1027 unsigned long *nr_written, int unlock)
1029 struct btrfs_root *root = inode->root;
1030 struct btrfs_fs_info *fs_info = root->fs_info;
1033 unsigned long ram_size;
1034 u64 cur_alloc_size = 0;
1036 u64 blocksize = fs_info->sectorsize;
1037 struct btrfs_key ins;
1038 struct extent_map *em;
1039 unsigned clear_bits;
1040 unsigned long page_ops;
1041 bool extent_reserved = false;
1044 if (btrfs_is_free_space_inode(inode)) {
1050 num_bytes = ALIGN(end - start + 1, blocksize);
1051 num_bytes = max(blocksize, num_bytes);
1052 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1054 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1057 /* lets try to make an inline extent */
1058 ret = cow_file_range_inline(inode, start, end, 0,
1059 BTRFS_COMPRESS_NONE, NULL);
1062 * We use DO_ACCOUNTING here because we need the
1063 * delalloc_release_metadata to be run _after_ we drop
1064 * our outstanding extent for clearing delalloc for this
1067 extent_clear_unlock_delalloc(inode, start, end, NULL,
1068 EXTENT_LOCKED | EXTENT_DELALLOC |
1069 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1070 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1071 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1072 *nr_written = *nr_written +
1073 (end - start + PAGE_SIZE) / PAGE_SIZE;
1076 } else if (ret < 0) {
1081 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1082 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1085 * Relocation relies on the relocated extents to have exactly the same
1086 * size as the original extents. Normally writeback for relocation data
1087 * extents follows a NOCOW path because relocation preallocates the
1088 * extents. However, due to an operation such as scrub turning a block
1089 * group to RO mode, it may fallback to COW mode, so we must make sure
1090 * an extent allocated during COW has exactly the requested size and can
1091 * not be split into smaller extents, otherwise relocation breaks and
1092 * fails during the stage where it updates the bytenr of file extent
1095 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1096 min_alloc_size = num_bytes;
1098 min_alloc_size = fs_info->sectorsize;
1100 while (num_bytes > 0) {
1101 cur_alloc_size = num_bytes;
1102 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1103 min_alloc_size, 0, alloc_hint,
1107 cur_alloc_size = ins.offset;
1108 extent_reserved = true;
1110 ram_size = ins.offset;
1111 em = create_io_em(inode, start, ins.offset, /* len */
1112 start, /* orig_start */
1113 ins.objectid, /* block_start */
1114 ins.offset, /* block_len */
1115 ins.offset, /* orig_block_len */
1116 ram_size, /* ram_bytes */
1117 BTRFS_COMPRESS_NONE, /* compress_type */
1118 BTRFS_ORDERED_REGULAR /* type */);
1123 free_extent_map(em);
1125 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1126 ram_size, cur_alloc_size,
1127 BTRFS_ORDERED_REGULAR);
1129 goto out_drop_extent_cache;
1131 if (root->root_key.objectid ==
1132 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1133 ret = btrfs_reloc_clone_csums(inode, start,
1136 * Only drop cache here, and process as normal.
1138 * We must not allow extent_clear_unlock_delalloc()
1139 * at out_unlock label to free meta of this ordered
1140 * extent, as its meta should be freed by
1141 * btrfs_finish_ordered_io().
1143 * So we must continue until @start is increased to
1144 * skip current ordered extent.
1147 btrfs_drop_extent_cache(inode, start,
1148 start + ram_size - 1, 0);
1151 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1153 /* we're not doing compressed IO, don't unlock the first
1154 * page (which the caller expects to stay locked), don't
1155 * clear any dirty bits and don't set any writeback bits
1157 * Do set the Private2 bit so we know this page was properly
1158 * setup for writepage
1160 page_ops = unlock ? PAGE_UNLOCK : 0;
1161 page_ops |= PAGE_SET_PRIVATE2;
1163 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1165 EXTENT_LOCKED | EXTENT_DELALLOC,
1167 if (num_bytes < cur_alloc_size)
1170 num_bytes -= cur_alloc_size;
1171 alloc_hint = ins.objectid + ins.offset;
1172 start += cur_alloc_size;
1173 extent_reserved = false;
1176 * btrfs_reloc_clone_csums() error, since start is increased
1177 * extent_clear_unlock_delalloc() at out_unlock label won't
1178 * free metadata of current ordered extent, we're OK to exit.
1186 out_drop_extent_cache:
1187 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1189 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1190 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1192 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1193 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1194 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1196 * If we reserved an extent for our delalloc range (or a subrange) and
1197 * failed to create the respective ordered extent, then it means that
1198 * when we reserved the extent we decremented the extent's size from
1199 * the data space_info's bytes_may_use counter and incremented the
1200 * space_info's bytes_reserved counter by the same amount. We must make
1201 * sure extent_clear_unlock_delalloc() does not try to decrement again
1202 * the data space_info's bytes_may_use counter, therefore we do not pass
1203 * it the flag EXTENT_CLEAR_DATA_RESV.
1205 if (extent_reserved) {
1206 extent_clear_unlock_delalloc(inode, start,
1207 start + cur_alloc_size - 1,
1211 start += cur_alloc_size;
1215 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1216 clear_bits | EXTENT_CLEAR_DATA_RESV,
1222 * work queue call back to started compression on a file and pages
1224 static noinline void async_cow_start(struct btrfs_work *work)
1226 struct async_chunk *async_chunk;
1227 int compressed_extents;
1229 async_chunk = container_of(work, struct async_chunk, work);
1231 compressed_extents = compress_file_range(async_chunk);
1232 if (compressed_extents == 0) {
1233 btrfs_add_delayed_iput(async_chunk->inode);
1234 async_chunk->inode = NULL;
1239 * work queue call back to submit previously compressed pages
1241 static noinline void async_cow_submit(struct btrfs_work *work)
1243 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1245 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1246 unsigned long nr_pages;
1248 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1251 /* atomic_sub_return implies a barrier */
1252 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1254 cond_wake_up_nomb(&fs_info->async_submit_wait);
1257 * ->inode could be NULL if async_chunk_start has failed to compress,
1258 * in which case we don't have anything to submit, yet we need to
1259 * always adjust ->async_delalloc_pages as its paired with the init
1260 * happening in cow_file_range_async
1262 if (async_chunk->inode)
1263 submit_compressed_extents(async_chunk);
1266 static noinline void async_cow_free(struct btrfs_work *work)
1268 struct async_chunk *async_chunk;
1270 async_chunk = container_of(work, struct async_chunk, work);
1271 if (async_chunk->inode)
1272 btrfs_add_delayed_iput(async_chunk->inode);
1273 if (async_chunk->blkcg_css)
1274 css_put(async_chunk->blkcg_css);
1276 * Since the pointer to 'pending' is at the beginning of the array of
1277 * async_chunk's, freeing it ensures the whole array has been freed.
1279 if (atomic_dec_and_test(async_chunk->pending))
1280 kvfree(async_chunk->pending);
1283 static int cow_file_range_async(struct btrfs_inode *inode,
1284 struct writeback_control *wbc,
1285 struct page *locked_page,
1286 u64 start, u64 end, int *page_started,
1287 unsigned long *nr_written)
1289 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1290 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1291 struct async_cow *ctx;
1292 struct async_chunk *async_chunk;
1293 unsigned long nr_pages;
1295 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1297 bool should_compress;
1299 const unsigned int write_flags = wbc_to_write_flags(wbc);
1301 unlock_extent(&inode->io_tree, start, end);
1303 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1304 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1306 should_compress = false;
1308 should_compress = true;
1311 nofs_flag = memalloc_nofs_save();
1312 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1313 memalloc_nofs_restore(nofs_flag);
1316 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1317 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1318 EXTENT_DO_ACCOUNTING;
1319 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1320 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1322 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1323 clear_bits, page_ops);
1327 async_chunk = ctx->chunks;
1328 atomic_set(&ctx->num_chunks, num_chunks);
1330 for (i = 0; i < num_chunks; i++) {
1331 if (should_compress)
1332 cur_end = min(end, start + SZ_512K - 1);
1337 * igrab is called higher up in the call chain, take only the
1338 * lightweight reference for the callback lifetime
1340 ihold(&inode->vfs_inode);
1341 async_chunk[i].pending = &ctx->num_chunks;
1342 async_chunk[i].inode = &inode->vfs_inode;
1343 async_chunk[i].start = start;
1344 async_chunk[i].end = cur_end;
1345 async_chunk[i].write_flags = write_flags;
1346 INIT_LIST_HEAD(&async_chunk[i].extents);
1349 * The locked_page comes all the way from writepage and its
1350 * the original page we were actually given. As we spread
1351 * this large delalloc region across multiple async_chunk
1352 * structs, only the first struct needs a pointer to locked_page
1354 * This way we don't need racey decisions about who is supposed
1359 * Depending on the compressibility, the pages might or
1360 * might not go through async. We want all of them to
1361 * be accounted against wbc once. Let's do it here
1362 * before the paths diverge. wbc accounting is used
1363 * only for foreign writeback detection and doesn't
1364 * need full accuracy. Just account the whole thing
1365 * against the first page.
1367 wbc_account_cgroup_owner(wbc, locked_page,
1369 async_chunk[i].locked_page = locked_page;
1372 async_chunk[i].locked_page = NULL;
1375 if (blkcg_css != blkcg_root_css) {
1377 async_chunk[i].blkcg_css = blkcg_css;
1379 async_chunk[i].blkcg_css = NULL;
1382 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1383 async_cow_submit, async_cow_free);
1385 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1386 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1388 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1390 *nr_written += nr_pages;
1391 start = cur_end + 1;
1397 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1398 struct page *locked_page, u64 start,
1399 u64 end, int *page_started,
1400 unsigned long *nr_written)
1404 ret = cow_file_range(inode, locked_page, start, end, page_started,
1412 __set_page_dirty_nobuffers(locked_page);
1413 account_page_redirty(locked_page);
1414 extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL);
1420 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1421 u64 bytenr, u64 num_bytes)
1424 struct btrfs_ordered_sum *sums;
1427 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1428 bytenr + num_bytes - 1, &list, 0);
1429 if (ret == 0 && list_empty(&list))
1432 while (!list_empty(&list)) {
1433 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1434 list_del(&sums->list);
1442 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1443 const u64 start, const u64 end,
1444 int *page_started, unsigned long *nr_written)
1446 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1447 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1448 BTRFS_DATA_RELOC_TREE_OBJECTID);
1449 const u64 range_bytes = end + 1 - start;
1450 struct extent_io_tree *io_tree = &inode->io_tree;
1451 u64 range_start = start;
1455 * If EXTENT_NORESERVE is set it means that when the buffered write was
1456 * made we had not enough available data space and therefore we did not
1457 * reserve data space for it, since we though we could do NOCOW for the
1458 * respective file range (either there is prealloc extent or the inode
1459 * has the NOCOW bit set).
1461 * However when we need to fallback to COW mode (because for example the
1462 * block group for the corresponding extent was turned to RO mode by a
1463 * scrub or relocation) we need to do the following:
1465 * 1) We increment the bytes_may_use counter of the data space info.
1466 * If COW succeeds, it allocates a new data extent and after doing
1467 * that it decrements the space info's bytes_may_use counter and
1468 * increments its bytes_reserved counter by the same amount (we do
1469 * this at btrfs_add_reserved_bytes()). So we need to increment the
1470 * bytes_may_use counter to compensate (when space is reserved at
1471 * buffered write time, the bytes_may_use counter is incremented);
1473 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1474 * that if the COW path fails for any reason, it decrements (through
1475 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1476 * data space info, which we incremented in the step above.
1478 * If we need to fallback to cow and the inode corresponds to a free
1479 * space cache inode or an inode of the data relocation tree, we must
1480 * also increment bytes_may_use of the data space_info for the same
1481 * reason. Space caches and relocated data extents always get a prealloc
1482 * extent for them, however scrub or balance may have set the block
1483 * group that contains that extent to RO mode and therefore force COW
1484 * when starting writeback.
1486 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1487 EXTENT_NORESERVE, 0);
1488 if (count > 0 || is_space_ino || is_reloc_ino) {
1490 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1491 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1493 if (is_space_ino || is_reloc_ino)
1494 bytes = range_bytes;
1496 spin_lock(&sinfo->lock);
1497 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1498 spin_unlock(&sinfo->lock);
1501 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1505 return cow_file_range(inode, locked_page, start, end, page_started,
1510 * when nowcow writeback call back. This checks for snapshots or COW copies
1511 * of the extents that exist in the file, and COWs the file as required.
1513 * If no cow copies or snapshots exist, we write directly to the existing
1516 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1517 struct page *locked_page,
1518 const u64 start, const u64 end,
1520 unsigned long *nr_written)
1522 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1523 struct btrfs_root *root = inode->root;
1524 struct btrfs_path *path;
1525 u64 cow_start = (u64)-1;
1526 u64 cur_offset = start;
1528 bool check_prev = true;
1529 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1530 u64 ino = btrfs_ino(inode);
1532 u64 disk_bytenr = 0;
1533 const bool force = inode->flags & BTRFS_INODE_NODATACOW;
1535 path = btrfs_alloc_path();
1537 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1538 EXTENT_LOCKED | EXTENT_DELALLOC |
1539 EXTENT_DO_ACCOUNTING |
1540 EXTENT_DEFRAG, PAGE_UNLOCK |
1541 PAGE_START_WRITEBACK |
1542 PAGE_END_WRITEBACK);
1547 struct btrfs_key found_key;
1548 struct btrfs_file_extent_item *fi;
1549 struct extent_buffer *leaf;
1559 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1565 * If there is no extent for our range when doing the initial
1566 * search, then go back to the previous slot as it will be the
1567 * one containing the search offset
1569 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1570 leaf = path->nodes[0];
1571 btrfs_item_key_to_cpu(leaf, &found_key,
1572 path->slots[0] - 1);
1573 if (found_key.objectid == ino &&
1574 found_key.type == BTRFS_EXTENT_DATA_KEY)
1579 /* Go to next leaf if we have exhausted the current one */
1580 leaf = path->nodes[0];
1581 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1582 ret = btrfs_next_leaf(root, path);
1584 if (cow_start != (u64)-1)
1585 cur_offset = cow_start;
1590 leaf = path->nodes[0];
1593 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1595 /* Didn't find anything for our INO */
1596 if (found_key.objectid > ino)
1599 * Keep searching until we find an EXTENT_ITEM or there are no
1600 * more extents for this inode
1602 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1603 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1608 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1609 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1610 found_key.offset > end)
1614 * If the found extent starts after requested offset, then
1615 * adjust extent_end to be right before this extent begins
1617 if (found_key.offset > cur_offset) {
1618 extent_end = found_key.offset;
1624 * Found extent which begins before our range and potentially
1627 fi = btrfs_item_ptr(leaf, path->slots[0],
1628 struct btrfs_file_extent_item);
1629 extent_type = btrfs_file_extent_type(leaf, fi);
1631 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1632 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1633 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1634 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1635 extent_offset = btrfs_file_extent_offset(leaf, fi);
1636 extent_end = found_key.offset +
1637 btrfs_file_extent_num_bytes(leaf, fi);
1639 btrfs_file_extent_disk_num_bytes(leaf, fi);
1641 * If the extent we got ends before our current offset,
1642 * skip to the next extent.
1644 if (extent_end <= cur_offset) {
1649 if (disk_bytenr == 0)
1651 /* Skip compressed/encrypted/encoded extents */
1652 if (btrfs_file_extent_compression(leaf, fi) ||
1653 btrfs_file_extent_encryption(leaf, fi) ||
1654 btrfs_file_extent_other_encoding(leaf, fi))
1657 * If extent is created before the last volume's snapshot
1658 * this implies the extent is shared, hence we can't do
1659 * nocow. This is the same check as in
1660 * btrfs_cross_ref_exist but without calling
1661 * btrfs_search_slot.
1663 if (!freespace_inode &&
1664 btrfs_file_extent_generation(leaf, fi) <=
1665 btrfs_root_last_snapshot(&root->root_item))
1667 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1671 * The following checks can be expensive, as they need to
1672 * take other locks and do btree or rbtree searches, so
1673 * release the path to avoid blocking other tasks for too
1676 btrfs_release_path(path);
1678 ret = btrfs_cross_ref_exist(root, ino,
1680 extent_offset, disk_bytenr, false);
1683 * ret could be -EIO if the above fails to read
1687 if (cow_start != (u64)-1)
1688 cur_offset = cow_start;
1692 WARN_ON_ONCE(freespace_inode);
1695 disk_bytenr += extent_offset;
1696 disk_bytenr += cur_offset - found_key.offset;
1697 num_bytes = min(end + 1, extent_end) - cur_offset;
1699 * If there are pending snapshots for this root, we
1700 * fall into common COW way
1702 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1705 * force cow if csum exists in the range.
1706 * this ensure that csum for a given extent are
1707 * either valid or do not exist.
1709 ret = csum_exist_in_range(fs_info, disk_bytenr,
1713 * ret could be -EIO if the above fails to read
1717 if (cow_start != (u64)-1)
1718 cur_offset = cow_start;
1721 WARN_ON_ONCE(freespace_inode);
1724 /* If the extent's block group is RO, we must COW */
1725 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1728 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1729 extent_end = found_key.offset + ram_bytes;
1730 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1731 /* Skip extents outside of our requested range */
1732 if (extent_end <= start) {
1737 /* If this triggers then we have a memory corruption */
1742 * If nocow is false then record the beginning of the range
1743 * that needs to be COWed
1746 if (cow_start == (u64)-1)
1747 cow_start = cur_offset;
1748 cur_offset = extent_end;
1749 if (cur_offset > end)
1751 if (!path->nodes[0])
1758 * COW range from cow_start to found_key.offset - 1. As the key
1759 * will contain the beginning of the first extent that can be
1760 * NOCOW, following one which needs to be COW'ed
1762 if (cow_start != (u64)-1) {
1763 ret = fallback_to_cow(inode, locked_page,
1764 cow_start, found_key.offset - 1,
1765 page_started, nr_written);
1768 cow_start = (u64)-1;
1771 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1772 u64 orig_start = found_key.offset - extent_offset;
1773 struct extent_map *em;
1775 em = create_io_em(inode, cur_offset, num_bytes,
1777 disk_bytenr, /* block_start */
1778 num_bytes, /* block_len */
1779 disk_num_bytes, /* orig_block_len */
1780 ram_bytes, BTRFS_COMPRESS_NONE,
1781 BTRFS_ORDERED_PREALLOC);
1786 free_extent_map(em);
1787 ret = btrfs_add_ordered_extent(inode, cur_offset,
1788 disk_bytenr, num_bytes,
1790 BTRFS_ORDERED_PREALLOC);
1792 btrfs_drop_extent_cache(inode, cur_offset,
1793 cur_offset + num_bytes - 1,
1798 ret = btrfs_add_ordered_extent(inode, cur_offset,
1799 disk_bytenr, num_bytes,
1801 BTRFS_ORDERED_NOCOW);
1807 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1810 if (root->root_key.objectid ==
1811 BTRFS_DATA_RELOC_TREE_OBJECTID)
1813 * Error handled later, as we must prevent
1814 * extent_clear_unlock_delalloc() in error handler
1815 * from freeing metadata of created ordered extent.
1817 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1820 extent_clear_unlock_delalloc(inode, cur_offset,
1821 cur_offset + num_bytes - 1,
1822 locked_page, EXTENT_LOCKED |
1824 EXTENT_CLEAR_DATA_RESV,
1825 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1827 cur_offset = extent_end;
1830 * btrfs_reloc_clone_csums() error, now we're OK to call error
1831 * handler, as metadata for created ordered extent will only
1832 * be freed by btrfs_finish_ordered_io().
1836 if (cur_offset > end)
1839 btrfs_release_path(path);
1841 if (cur_offset <= end && cow_start == (u64)-1)
1842 cow_start = cur_offset;
1844 if (cow_start != (u64)-1) {
1846 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1847 page_started, nr_written);
1854 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1856 if (ret && cur_offset < end)
1857 extent_clear_unlock_delalloc(inode, cur_offset, end,
1858 locked_page, EXTENT_LOCKED |
1859 EXTENT_DELALLOC | EXTENT_DEFRAG |
1860 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1861 PAGE_START_WRITEBACK |
1862 PAGE_END_WRITEBACK);
1863 btrfs_free_path(path);
1867 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
1869 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
1870 if (inode->defrag_bytes &&
1871 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
1880 * Function to process delayed allocation (create CoW) for ranges which are
1881 * being touched for the first time.
1883 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1884 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1885 struct writeback_control *wbc)
1888 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
1890 if (should_nocow(inode, start, end)) {
1892 ret = run_delalloc_nocow(inode, locked_page, start, end,
1893 page_started, nr_written);
1894 } else if (!inode_can_compress(inode) ||
1895 !inode_need_compress(inode, start, end)) {
1897 ret = run_delalloc_zoned(inode, locked_page, start, end,
1898 page_started, nr_written);
1900 ret = cow_file_range(inode, locked_page, start, end,
1901 page_started, nr_written, 1);
1903 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1904 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1905 page_started, nr_written);
1908 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1913 void btrfs_split_delalloc_extent(struct inode *inode,
1914 struct extent_state *orig, u64 split)
1918 /* not delalloc, ignore it */
1919 if (!(orig->state & EXTENT_DELALLOC))
1922 size = orig->end - orig->start + 1;
1923 if (size > BTRFS_MAX_EXTENT_SIZE) {
1928 * See the explanation in btrfs_merge_delalloc_extent, the same
1929 * applies here, just in reverse.
1931 new_size = orig->end - split + 1;
1932 num_extents = count_max_extents(new_size);
1933 new_size = split - orig->start;
1934 num_extents += count_max_extents(new_size);
1935 if (count_max_extents(size) >= num_extents)
1939 spin_lock(&BTRFS_I(inode)->lock);
1940 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1941 spin_unlock(&BTRFS_I(inode)->lock);
1945 * Handle merged delayed allocation extents so we can keep track of new extents
1946 * that are just merged onto old extents, such as when we are doing sequential
1947 * writes, so we can properly account for the metadata space we'll need.
1949 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1950 struct extent_state *other)
1952 u64 new_size, old_size;
1955 /* not delalloc, ignore it */
1956 if (!(other->state & EXTENT_DELALLOC))
1959 if (new->start > other->start)
1960 new_size = new->end - other->start + 1;
1962 new_size = other->end - new->start + 1;
1964 /* we're not bigger than the max, unreserve the space and go */
1965 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1966 spin_lock(&BTRFS_I(inode)->lock);
1967 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1968 spin_unlock(&BTRFS_I(inode)->lock);
1973 * We have to add up either side to figure out how many extents were
1974 * accounted for before we merged into one big extent. If the number of
1975 * extents we accounted for is <= the amount we need for the new range
1976 * then we can return, otherwise drop. Think of it like this
1980 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1981 * need 2 outstanding extents, on one side we have 1 and the other side
1982 * we have 1 so they are == and we can return. But in this case
1984 * [MAX_SIZE+4k][MAX_SIZE+4k]
1986 * Each range on their own accounts for 2 extents, but merged together
1987 * they are only 3 extents worth of accounting, so we need to drop in
1990 old_size = other->end - other->start + 1;
1991 num_extents = count_max_extents(old_size);
1992 old_size = new->end - new->start + 1;
1993 num_extents += count_max_extents(old_size);
1994 if (count_max_extents(new_size) >= num_extents)
1997 spin_lock(&BTRFS_I(inode)->lock);
1998 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1999 spin_unlock(&BTRFS_I(inode)->lock);
2002 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2003 struct inode *inode)
2005 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2007 spin_lock(&root->delalloc_lock);
2008 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
2009 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2010 &root->delalloc_inodes);
2011 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2012 &BTRFS_I(inode)->runtime_flags);
2013 root->nr_delalloc_inodes++;
2014 if (root->nr_delalloc_inodes == 1) {
2015 spin_lock(&fs_info->delalloc_root_lock);
2016 BUG_ON(!list_empty(&root->delalloc_root));
2017 list_add_tail(&root->delalloc_root,
2018 &fs_info->delalloc_roots);
2019 spin_unlock(&fs_info->delalloc_root_lock);
2022 spin_unlock(&root->delalloc_lock);
2026 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2027 struct btrfs_inode *inode)
2029 struct btrfs_fs_info *fs_info = root->fs_info;
2031 if (!list_empty(&inode->delalloc_inodes)) {
2032 list_del_init(&inode->delalloc_inodes);
2033 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2034 &inode->runtime_flags);
2035 root->nr_delalloc_inodes--;
2036 if (!root->nr_delalloc_inodes) {
2037 ASSERT(list_empty(&root->delalloc_inodes));
2038 spin_lock(&fs_info->delalloc_root_lock);
2039 BUG_ON(list_empty(&root->delalloc_root));
2040 list_del_init(&root->delalloc_root);
2041 spin_unlock(&fs_info->delalloc_root_lock);
2046 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2047 struct btrfs_inode *inode)
2049 spin_lock(&root->delalloc_lock);
2050 __btrfs_del_delalloc_inode(root, inode);
2051 spin_unlock(&root->delalloc_lock);
2055 * Properly track delayed allocation bytes in the inode and to maintain the
2056 * list of inodes that have pending delalloc work to be done.
2058 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2061 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2063 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2066 * set_bit and clear bit hooks normally require _irqsave/restore
2067 * but in this case, we are only testing for the DELALLOC
2068 * bit, which is only set or cleared with irqs on
2070 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2071 struct btrfs_root *root = BTRFS_I(inode)->root;
2072 u64 len = state->end + 1 - state->start;
2073 u32 num_extents = count_max_extents(len);
2074 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2076 spin_lock(&BTRFS_I(inode)->lock);
2077 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2078 spin_unlock(&BTRFS_I(inode)->lock);
2080 /* For sanity tests */
2081 if (btrfs_is_testing(fs_info))
2084 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2085 fs_info->delalloc_batch);
2086 spin_lock(&BTRFS_I(inode)->lock);
2087 BTRFS_I(inode)->delalloc_bytes += len;
2088 if (*bits & EXTENT_DEFRAG)
2089 BTRFS_I(inode)->defrag_bytes += len;
2090 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2091 &BTRFS_I(inode)->runtime_flags))
2092 btrfs_add_delalloc_inodes(root, inode);
2093 spin_unlock(&BTRFS_I(inode)->lock);
2096 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2097 (*bits & EXTENT_DELALLOC_NEW)) {
2098 spin_lock(&BTRFS_I(inode)->lock);
2099 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2101 spin_unlock(&BTRFS_I(inode)->lock);
2106 * Once a range is no longer delalloc this function ensures that proper
2107 * accounting happens.
2109 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2110 struct extent_state *state, unsigned *bits)
2112 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2113 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2114 u64 len = state->end + 1 - state->start;
2115 u32 num_extents = count_max_extents(len);
2117 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2118 spin_lock(&inode->lock);
2119 inode->defrag_bytes -= len;
2120 spin_unlock(&inode->lock);
2124 * set_bit and clear bit hooks normally require _irqsave/restore
2125 * but in this case, we are only testing for the DELALLOC
2126 * bit, which is only set or cleared with irqs on
2128 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2129 struct btrfs_root *root = inode->root;
2130 bool do_list = !btrfs_is_free_space_inode(inode);
2132 spin_lock(&inode->lock);
2133 btrfs_mod_outstanding_extents(inode, -num_extents);
2134 spin_unlock(&inode->lock);
2137 * We don't reserve metadata space for space cache inodes so we
2138 * don't need to call delalloc_release_metadata if there is an
2141 if (*bits & EXTENT_CLEAR_META_RESV &&
2142 root != fs_info->tree_root)
2143 btrfs_delalloc_release_metadata(inode, len, false);
2145 /* For sanity tests. */
2146 if (btrfs_is_testing(fs_info))
2149 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2150 do_list && !(state->state & EXTENT_NORESERVE) &&
2151 (*bits & EXTENT_CLEAR_DATA_RESV))
2152 btrfs_free_reserved_data_space_noquota(fs_info, len);
2154 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2155 fs_info->delalloc_batch);
2156 spin_lock(&inode->lock);
2157 inode->delalloc_bytes -= len;
2158 if (do_list && inode->delalloc_bytes == 0 &&
2159 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2160 &inode->runtime_flags))
2161 btrfs_del_delalloc_inode(root, inode);
2162 spin_unlock(&inode->lock);
2165 if ((state->state & EXTENT_DELALLOC_NEW) &&
2166 (*bits & EXTENT_DELALLOC_NEW)) {
2167 spin_lock(&inode->lock);
2168 ASSERT(inode->new_delalloc_bytes >= len);
2169 inode->new_delalloc_bytes -= len;
2170 if (*bits & EXTENT_ADD_INODE_BYTES)
2171 inode_add_bytes(&inode->vfs_inode, len);
2172 spin_unlock(&inode->lock);
2177 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2178 * in a chunk's stripe. This function ensures that bios do not span a
2181 * @page - The page we are about to add to the bio
2182 * @size - size we want to add to the bio
2183 * @bio - bio we want to ensure is smaller than a stripe
2184 * @bio_flags - flags of the bio
2186 * return 1 if page cannot be added to the bio
2187 * return 0 if page can be added to the bio
2188 * return error otherwise
2190 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2191 unsigned long bio_flags)
2193 struct inode *inode = page->mapping->host;
2194 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2195 u64 logical = bio->bi_iter.bi_sector << 9;
2196 struct extent_map *em;
2200 struct btrfs_io_geometry geom;
2202 if (bio_flags & EXTENT_BIO_COMPRESSED)
2205 length = bio->bi_iter.bi_size;
2206 map_length = length;
2207 em = btrfs_get_chunk_map(fs_info, logical, map_length);
2210 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical,
2215 if (geom.len < length + size)
2218 free_extent_map(em);
2223 * in order to insert checksums into the metadata in large chunks,
2224 * we wait until bio submission time. All the pages in the bio are
2225 * checksummed and sums are attached onto the ordered extent record.
2227 * At IO completion time the cums attached on the ordered extent record
2228 * are inserted into the btree
2230 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2231 u64 dio_file_offset)
2233 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2236 bool btrfs_bio_fits_in_ordered_extent(struct page *page, struct bio *bio,
2239 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2240 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2241 struct btrfs_ordered_extent *ordered;
2242 u64 len = bio->bi_iter.bi_size + size;
2245 ASSERT(btrfs_is_zoned(fs_info));
2246 ASSERT(fs_info->max_zone_append_size > 0);
2247 ASSERT(bio_op(bio) == REQ_OP_ZONE_APPEND);
2249 /* Ordered extent not yet created, so we're good */
2250 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
2254 if ((bio->bi_iter.bi_sector << SECTOR_SHIFT) + len >
2255 ordered->disk_bytenr + ordered->disk_num_bytes)
2258 btrfs_put_ordered_extent(ordered);
2264 * Split an extent_map at [start, start + len]
2266 * This function is intended to be used only for extract_ordered_extent().
2268 static int split_zoned_em(struct btrfs_inode *inode, u64 start, u64 len,
2271 struct extent_map_tree *em_tree = &inode->extent_tree;
2272 struct extent_map *em;
2273 struct extent_map *split_pre = NULL;
2274 struct extent_map *split_mid = NULL;
2275 struct extent_map *split_post = NULL;
2278 unsigned long flags;
2281 if (pre == 0 && post == 0)
2284 split_pre = alloc_extent_map();
2286 split_mid = alloc_extent_map();
2288 split_post = alloc_extent_map();
2289 if (!split_pre || (pre && !split_mid) || (post && !split_post)) {
2294 ASSERT(pre + post < len);
2296 lock_extent(&inode->io_tree, start, start + len - 1);
2297 write_lock(&em_tree->lock);
2298 em = lookup_extent_mapping(em_tree, start, len);
2304 ASSERT(em->len == len);
2305 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2306 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2309 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2310 clear_bit(EXTENT_FLAG_LOGGING, &flags);
2311 modified = !list_empty(&em->list);
2313 /* First, replace the em with a new extent_map starting from * em->start */
2314 split_pre->start = em->start;
2315 split_pre->len = (pre ? pre : em->len - post);
2316 split_pre->orig_start = split_pre->start;
2317 split_pre->block_start = em->block_start;
2318 split_pre->block_len = split_pre->len;
2319 split_pre->orig_block_len = split_pre->block_len;
2320 split_pre->ram_bytes = split_pre->len;
2321 split_pre->flags = flags;
2322 split_pre->compress_type = em->compress_type;
2323 split_pre->generation = em->generation;
2325 replace_extent_mapping(em_tree, em, split_pre, modified);
2328 * Now we only have an extent_map at:
2329 * [em->start, em->start + pre] if pre != 0
2330 * [em->start, em->start + em->len - post] if pre == 0
2334 /* Insert the middle extent_map */
2335 split_mid->start = em->start + pre;
2336 split_mid->len = em->len - pre - post;
2337 split_mid->orig_start = split_mid->start;
2338 split_mid->block_start = em->block_start + pre;
2339 split_mid->block_len = split_mid->len;
2340 split_mid->orig_block_len = split_mid->block_len;
2341 split_mid->ram_bytes = split_mid->len;
2342 split_mid->flags = flags;
2343 split_mid->compress_type = em->compress_type;
2344 split_mid->generation = em->generation;
2345 add_extent_mapping(em_tree, split_mid, modified);
2349 split_post->start = em->start + em->len - post;
2350 split_post->len = post;
2351 split_post->orig_start = split_post->start;
2352 split_post->block_start = em->block_start + em->len - post;
2353 split_post->block_len = split_post->len;
2354 split_post->orig_block_len = split_post->block_len;
2355 split_post->ram_bytes = split_post->len;
2356 split_post->flags = flags;
2357 split_post->compress_type = em->compress_type;
2358 split_post->generation = em->generation;
2359 add_extent_mapping(em_tree, split_post, modified);
2363 free_extent_map(em);
2364 /* Once for the tree */
2365 free_extent_map(em);
2368 write_unlock(&em_tree->lock);
2369 unlock_extent(&inode->io_tree, start, start + len - 1);
2371 free_extent_map(split_pre);
2372 free_extent_map(split_mid);
2373 free_extent_map(split_post);
2378 static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
2379 struct bio *bio, loff_t file_offset)
2381 struct btrfs_ordered_extent *ordered;
2382 u64 start = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
2384 u64 len = bio->bi_iter.bi_size;
2385 u64 end = start + len;
2390 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
2391 if (WARN_ON_ONCE(!ordered))
2392 return BLK_STS_IOERR;
2394 /* No need to split */
2395 if (ordered->disk_num_bytes == len)
2398 /* We cannot split once end_bio'd ordered extent */
2399 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes)) {
2404 /* We cannot split a compressed ordered extent */
2405 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes)) {
2410 ordered_end = ordered->disk_bytenr + ordered->disk_num_bytes;
2411 /* bio must be in one ordered extent */
2412 if (WARN_ON_ONCE(start < ordered->disk_bytenr || end > ordered_end)) {
2417 /* Checksum list should be empty */
2418 if (WARN_ON_ONCE(!list_empty(&ordered->list))) {
2423 file_len = ordered->num_bytes;
2424 pre = start - ordered->disk_bytenr;
2425 post = ordered_end - end;
2427 ret = btrfs_split_ordered_extent(ordered, pre, post);
2430 ret = split_zoned_em(inode, file_offset, file_len, pre, post);
2433 btrfs_put_ordered_extent(ordered);
2435 return errno_to_blk_status(ret);
2439 * extent_io.c submission hook. This does the right thing for csum calculation
2440 * on write, or reading the csums from the tree before a read.
2442 * Rules about async/sync submit,
2443 * a) read: sync submit
2445 * b) write without checksum: sync submit
2447 * c) write with checksum:
2448 * c-1) if bio is issued by fsync: sync submit
2449 * (sync_writers != 0)
2451 * c-2) if root is reloc root: sync submit
2452 * (only in case of buffered IO)
2454 * c-3) otherwise: async submit
2456 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2457 int mirror_num, unsigned long bio_flags)
2460 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2461 struct btrfs_root *root = BTRFS_I(inode)->root;
2462 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2463 blk_status_t ret = 0;
2465 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2467 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2468 !fs_info->csum_root;
2470 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2471 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2473 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
2474 struct page *page = bio_first_bvec_all(bio)->bv_page;
2475 loff_t file_offset = page_offset(page);
2477 ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
2482 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
2483 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2487 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2488 ret = btrfs_submit_compressed_read(inode, bio,
2494 * Lookup bio sums does extra checks around whether we
2495 * need to csum or not, which is why we ignore skip_sum
2498 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2503 } else if (async && !skip_sum) {
2504 /* csum items have already been cloned */
2505 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2507 /* we're doing a write, do the async checksumming */
2508 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2509 0, btrfs_submit_bio_start);
2511 } else if (!skip_sum) {
2512 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2518 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2522 bio->bi_status = ret;
2529 * given a list of ordered sums record them in the inode. This happens
2530 * at IO completion time based on sums calculated at bio submission time.
2532 static int add_pending_csums(struct btrfs_trans_handle *trans,
2533 struct list_head *list)
2535 struct btrfs_ordered_sum *sum;
2538 list_for_each_entry(sum, list, list) {
2539 trans->adding_csums = true;
2540 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2541 trans->adding_csums = false;
2548 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2551 struct extent_state **cached_state)
2553 u64 search_start = start;
2554 const u64 end = start + len - 1;
2556 while (search_start < end) {
2557 const u64 search_len = end - search_start + 1;
2558 struct extent_map *em;
2562 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2566 if (em->block_start != EXTENT_MAP_HOLE)
2570 if (em->start < search_start)
2571 em_len -= search_start - em->start;
2572 if (em_len > search_len)
2573 em_len = search_len;
2575 ret = set_extent_bit(&inode->io_tree, search_start,
2576 search_start + em_len - 1,
2577 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2580 search_start = extent_map_end(em);
2581 free_extent_map(em);
2588 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2589 unsigned int extra_bits,
2590 struct extent_state **cached_state)
2592 WARN_ON(PAGE_ALIGNED(end));
2594 if (start >= i_size_read(&inode->vfs_inode) &&
2595 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2597 * There can't be any extents following eof in this case so just
2598 * set the delalloc new bit for the range directly.
2600 extra_bits |= EXTENT_DELALLOC_NEW;
2604 ret = btrfs_find_new_delalloc_bytes(inode, start,
2611 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2615 /* see btrfs_writepage_start_hook for details on why this is required */
2616 struct btrfs_writepage_fixup {
2618 struct inode *inode;
2619 struct btrfs_work work;
2622 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2624 struct btrfs_writepage_fixup *fixup;
2625 struct btrfs_ordered_extent *ordered;
2626 struct extent_state *cached_state = NULL;
2627 struct extent_changeset *data_reserved = NULL;
2629 struct btrfs_inode *inode;
2633 bool free_delalloc_space = true;
2635 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2637 inode = BTRFS_I(fixup->inode);
2638 page_start = page_offset(page);
2639 page_end = page_offset(page) + PAGE_SIZE - 1;
2642 * This is similar to page_mkwrite, we need to reserve the space before
2643 * we take the page lock.
2645 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2651 * Before we queued this fixup, we took a reference on the page.
2652 * page->mapping may go NULL, but it shouldn't be moved to a different
2655 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2657 * Unfortunately this is a little tricky, either
2659 * 1) We got here and our page had already been dealt with and
2660 * we reserved our space, thus ret == 0, so we need to just
2661 * drop our space reservation and bail. This can happen the
2662 * first time we come into the fixup worker, or could happen
2663 * while waiting for the ordered extent.
2664 * 2) Our page was already dealt with, but we happened to get an
2665 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2666 * this case we obviously don't have anything to release, but
2667 * because the page was already dealt with we don't want to
2668 * mark the page with an error, so make sure we're resetting
2669 * ret to 0. This is why we have this check _before_ the ret
2670 * check, because we do not want to have a surprise ENOSPC
2671 * when the page was already properly dealt with.
2674 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2675 btrfs_delalloc_release_space(inode, data_reserved,
2676 page_start, PAGE_SIZE,
2684 * We can't mess with the page state unless it is locked, so now that
2685 * it is locked bail if we failed to make our space reservation.
2690 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2692 /* already ordered? We're done */
2693 if (PagePrivate2(page))
2696 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2698 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2701 btrfs_start_ordered_extent(ordered, 1);
2702 btrfs_put_ordered_extent(ordered);
2706 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2712 * Everything went as planned, we're now the owner of a dirty page with
2713 * delayed allocation bits set and space reserved for our COW
2716 * The page was dirty when we started, nothing should have cleaned it.
2718 BUG_ON(!PageDirty(page));
2719 free_delalloc_space = false;
2721 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2722 if (free_delalloc_space)
2723 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2725 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2730 * We hit ENOSPC or other errors. Update the mapping and page
2731 * to reflect the errors and clean the page.
2733 mapping_set_error(page->mapping, ret);
2734 end_extent_writepage(page, ret, page_start, page_end);
2735 clear_page_dirty_for_io(page);
2738 ClearPageChecked(page);
2742 extent_changeset_free(data_reserved);
2744 * As a precaution, do a delayed iput in case it would be the last iput
2745 * that could need flushing space. Recursing back to fixup worker would
2748 btrfs_add_delayed_iput(&inode->vfs_inode);
2752 * There are a few paths in the higher layers of the kernel that directly
2753 * set the page dirty bit without asking the filesystem if it is a
2754 * good idea. This causes problems because we want to make sure COW
2755 * properly happens and the data=ordered rules are followed.
2757 * In our case any range that doesn't have the ORDERED bit set
2758 * hasn't been properly setup for IO. We kick off an async process
2759 * to fix it up. The async helper will wait for ordered extents, set
2760 * the delalloc bit and make it safe to write the page.
2762 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2764 struct inode *inode = page->mapping->host;
2765 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2766 struct btrfs_writepage_fixup *fixup;
2768 /* this page is properly in the ordered list */
2769 if (TestClearPagePrivate2(page))
2773 * PageChecked is set below when we create a fixup worker for this page,
2774 * don't try to create another one if we're already PageChecked()
2776 * The extent_io writepage code will redirty the page if we send back
2779 if (PageChecked(page))
2782 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2787 * We are already holding a reference to this inode from
2788 * write_cache_pages. We need to hold it because the space reservation
2789 * takes place outside of the page lock, and we can't trust
2790 * page->mapping outside of the page lock.
2793 SetPageChecked(page);
2795 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2797 fixup->inode = inode;
2798 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2803 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2804 struct btrfs_inode *inode, u64 file_pos,
2805 struct btrfs_file_extent_item *stack_fi,
2806 const bool update_inode_bytes,
2807 u64 qgroup_reserved)
2809 struct btrfs_root *root = inode->root;
2810 const u64 sectorsize = root->fs_info->sectorsize;
2811 struct btrfs_path *path;
2812 struct extent_buffer *leaf;
2813 struct btrfs_key ins;
2814 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2815 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2816 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2817 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2818 struct btrfs_drop_extents_args drop_args = { 0 };
2821 path = btrfs_alloc_path();
2826 * we may be replacing one extent in the tree with another.
2827 * The new extent is pinned in the extent map, and we don't want
2828 * to drop it from the cache until it is completely in the btree.
2830 * So, tell btrfs_drop_extents to leave this extent in the cache.
2831 * the caller is expected to unpin it and allow it to be merged
2834 drop_args.path = path;
2835 drop_args.start = file_pos;
2836 drop_args.end = file_pos + num_bytes;
2837 drop_args.replace_extent = true;
2838 drop_args.extent_item_size = sizeof(*stack_fi);
2839 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2843 if (!drop_args.extent_inserted) {
2844 ins.objectid = btrfs_ino(inode);
2845 ins.offset = file_pos;
2846 ins.type = BTRFS_EXTENT_DATA_KEY;
2848 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2853 leaf = path->nodes[0];
2854 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2855 write_extent_buffer(leaf, stack_fi,
2856 btrfs_item_ptr_offset(leaf, path->slots[0]),
2857 sizeof(struct btrfs_file_extent_item));
2859 btrfs_mark_buffer_dirty(leaf);
2860 btrfs_release_path(path);
2863 * If we dropped an inline extent here, we know the range where it is
2864 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2865 * number of bytes only for that range contaning the inline extent.
2866 * The remaining of the range will be processed when clearning the
2867 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2869 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2870 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2872 inline_size = drop_args.bytes_found - inline_size;
2873 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2874 drop_args.bytes_found -= inline_size;
2875 num_bytes -= sectorsize;
2878 if (update_inode_bytes)
2879 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2881 ins.objectid = disk_bytenr;
2882 ins.offset = disk_num_bytes;
2883 ins.type = BTRFS_EXTENT_ITEM_KEY;
2885 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2889 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2890 file_pos, qgroup_reserved, &ins);
2892 btrfs_free_path(path);
2897 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2900 struct btrfs_block_group *cache;
2902 cache = btrfs_lookup_block_group(fs_info, start);
2905 spin_lock(&cache->lock);
2906 cache->delalloc_bytes -= len;
2907 spin_unlock(&cache->lock);
2909 btrfs_put_block_group(cache);
2912 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2913 struct btrfs_ordered_extent *oe)
2915 struct btrfs_file_extent_item stack_fi;
2917 bool update_inode_bytes;
2919 memset(&stack_fi, 0, sizeof(stack_fi));
2920 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2921 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2922 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2923 oe->disk_num_bytes);
2924 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2925 logical_len = oe->truncated_len;
2927 logical_len = oe->num_bytes;
2928 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2929 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2930 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2931 /* Encryption and other encoding is reserved and all 0 */
2934 * For delalloc, when completing an ordered extent we update the inode's
2935 * bytes when clearing the range in the inode's io tree, so pass false
2936 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2937 * except if the ordered extent was truncated.
2939 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2940 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2942 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2943 oe->file_offset, &stack_fi,
2944 update_inode_bytes, oe->qgroup_rsv);
2948 * As ordered data IO finishes, this gets called so we can finish
2949 * an ordered extent if the range of bytes in the file it covers are
2952 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2954 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2955 struct btrfs_root *root = inode->root;
2956 struct btrfs_fs_info *fs_info = root->fs_info;
2957 struct btrfs_trans_handle *trans = NULL;
2958 struct extent_io_tree *io_tree = &inode->io_tree;
2959 struct extent_state *cached_state = NULL;
2961 int compress_type = 0;
2963 u64 logical_len = ordered_extent->num_bytes;
2964 bool freespace_inode;
2965 bool truncated = false;
2966 bool clear_reserved_extent = true;
2967 unsigned int clear_bits = EXTENT_DEFRAG;
2969 start = ordered_extent->file_offset;
2970 end = start + ordered_extent->num_bytes - 1;
2972 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2973 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2974 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2975 clear_bits |= EXTENT_DELALLOC_NEW;
2977 freespace_inode = btrfs_is_free_space_inode(inode);
2979 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2984 if (ordered_extent->disk)
2985 btrfs_rewrite_logical_zoned(ordered_extent);
2987 btrfs_free_io_failure_record(inode, start, end);
2989 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2991 logical_len = ordered_extent->truncated_len;
2992 /* Truncated the entire extent, don't bother adding */
2997 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2998 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3000 btrfs_inode_safe_disk_i_size_write(inode, 0);
3001 if (freespace_inode)
3002 trans = btrfs_join_transaction_spacecache(root);
3004 trans = btrfs_join_transaction(root);
3005 if (IS_ERR(trans)) {
3006 ret = PTR_ERR(trans);
3010 trans->block_rsv = &inode->block_rsv;
3011 ret = btrfs_update_inode_fallback(trans, root, inode);
3012 if (ret) /* -ENOMEM or corruption */
3013 btrfs_abort_transaction(trans, ret);
3017 clear_bits |= EXTENT_LOCKED;
3018 lock_extent_bits(io_tree, start, end, &cached_state);
3020 if (freespace_inode)
3021 trans = btrfs_join_transaction_spacecache(root);
3023 trans = btrfs_join_transaction(root);
3024 if (IS_ERR(trans)) {
3025 ret = PTR_ERR(trans);
3030 trans->block_rsv = &inode->block_rsv;
3032 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3033 compress_type = ordered_extent->compress_type;
3034 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3035 BUG_ON(compress_type);
3036 ret = btrfs_mark_extent_written(trans, inode,
3037 ordered_extent->file_offset,
3038 ordered_extent->file_offset +
3041 BUG_ON(root == fs_info->tree_root);
3042 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3044 clear_reserved_extent = false;
3045 btrfs_release_delalloc_bytes(fs_info,
3046 ordered_extent->disk_bytenr,
3047 ordered_extent->disk_num_bytes);
3050 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3051 ordered_extent->num_bytes, trans->transid);
3053 btrfs_abort_transaction(trans, ret);
3057 ret = add_pending_csums(trans, &ordered_extent->list);
3059 btrfs_abort_transaction(trans, ret);
3064 * If this is a new delalloc range, clear its new delalloc flag to
3065 * update the inode's number of bytes. This needs to be done first
3066 * before updating the inode item.
3068 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3069 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3070 clear_extent_bit(&inode->io_tree, start, end,
3071 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3072 0, 0, &cached_state);
3074 btrfs_inode_safe_disk_i_size_write(inode, 0);
3075 ret = btrfs_update_inode_fallback(trans, root, inode);
3076 if (ret) { /* -ENOMEM or corruption */
3077 btrfs_abort_transaction(trans, ret);
3082 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3083 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
3087 btrfs_end_transaction(trans);
3089 if (ret || truncated) {
3090 u64 unwritten_start = start;
3093 * If we failed to finish this ordered extent for any reason we
3094 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3095 * extent, and mark the inode with the error if it wasn't
3096 * already set. Any error during writeback would have already
3097 * set the mapping error, so we need to set it if we're the ones
3098 * marking this ordered extent as failed.
3100 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3101 &ordered_extent->flags))
3102 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3105 unwritten_start += logical_len;
3106 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3108 /* Drop the cache for the part of the extent we didn't write. */
3109 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
3112 * If the ordered extent had an IOERR or something else went
3113 * wrong we need to return the space for this ordered extent
3114 * back to the allocator. We only free the extent in the
3115 * truncated case if we didn't write out the extent at all.
3117 * If we made it past insert_reserved_file_extent before we
3118 * errored out then we don't need to do this as the accounting
3119 * has already been done.
3121 if ((ret || !logical_len) &&
3122 clear_reserved_extent &&
3123 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3124 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3126 * Discard the range before returning it back to the
3129 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3130 btrfs_discard_extent(fs_info,
3131 ordered_extent->disk_bytenr,
3132 ordered_extent->disk_num_bytes,
3134 btrfs_free_reserved_extent(fs_info,
3135 ordered_extent->disk_bytenr,
3136 ordered_extent->disk_num_bytes, 1);
3141 * This needs to be done to make sure anybody waiting knows we are done
3142 * updating everything for this ordered extent.
3144 btrfs_remove_ordered_extent(inode, ordered_extent);
3147 btrfs_put_ordered_extent(ordered_extent);
3148 /* once for the tree */
3149 btrfs_put_ordered_extent(ordered_extent);
3154 static void finish_ordered_fn(struct btrfs_work *work)
3156 struct btrfs_ordered_extent *ordered_extent;
3157 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3158 btrfs_finish_ordered_io(ordered_extent);
3161 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
3162 u64 end, int uptodate)
3164 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
3165 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3166 struct btrfs_ordered_extent *ordered_extent = NULL;
3167 struct btrfs_workqueue *wq;
3169 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3171 ClearPagePrivate2(page);
3172 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3173 end - start + 1, uptodate))
3176 if (btrfs_is_free_space_inode(inode))
3177 wq = fs_info->endio_freespace_worker;
3179 wq = fs_info->endio_write_workers;
3181 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
3182 btrfs_queue_work(wq, &ordered_extent->work);
3186 * check_data_csum - verify checksum of one sector of uncompressed data
3188 * @io_bio: btrfs_io_bio which contains the csum
3189 * @bio_offset: offset to the beginning of the bio (in bytes)
3190 * @page: page where is the data to be verified
3191 * @pgoff: offset inside the page
3192 * @start: logical offset in the file
3194 * The length of such check is always one sector size.
3196 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
3197 u32 bio_offset, struct page *page, u32 pgoff,
3200 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3201 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3203 u32 len = fs_info->sectorsize;
3204 const u32 csum_size = fs_info->csum_size;
3205 unsigned int offset_sectors;
3207 u8 csum[BTRFS_CSUM_SIZE];
3209 ASSERT(pgoff + len <= PAGE_SIZE);
3211 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
3212 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
3214 kaddr = kmap_atomic(page);
3215 shash->tfm = fs_info->csum_shash;
3217 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
3219 if (memcmp(csum, csum_expected, csum_size))
3222 kunmap_atomic(kaddr);
3225 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3226 io_bio->mirror_num);
3228 btrfs_dev_stat_inc_and_print(io_bio->device,
3229 BTRFS_DEV_STAT_CORRUPTION_ERRS);
3230 memset(kaddr + pgoff, 1, len);
3231 flush_dcache_page(page);
3232 kunmap_atomic(kaddr);
3237 * When reads are done, we need to check csums to verify the data is correct.
3238 * if there's a match, we allow the bio to finish. If not, the code in
3239 * extent_io.c will try to find good copies for us.
3241 * @bio_offset: offset to the beginning of the bio (in bytes)
3242 * @start: file offset of the range start
3243 * @end: file offset of the range end (inclusive)
3245 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3246 struct page *page, u64 start, u64 end)
3248 struct inode *inode = page->mapping->host;
3249 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3250 struct btrfs_root *root = BTRFS_I(inode)->root;
3251 const u32 sectorsize = root->fs_info->sectorsize;
3254 if (PageChecked(page)) {
3255 ClearPageChecked(page);
3259 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3262 if (!root->fs_info->csum_root)
3265 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3266 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3267 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3271 ASSERT(page_offset(page) <= start &&
3272 end <= page_offset(page) + PAGE_SIZE - 1);
3273 for (pg_off = offset_in_page(start);
3274 pg_off < offset_in_page(end);
3275 pg_off += sectorsize, bio_offset += sectorsize) {
3278 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off,
3279 page_offset(page) + pg_off);
3287 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3289 * @inode: The inode we want to perform iput on
3291 * This function uses the generic vfs_inode::i_count to track whether we should
3292 * just decrement it (in case it's > 1) or if this is the last iput then link
3293 * the inode to the delayed iput machinery. Delayed iputs are processed at
3294 * transaction commit time/superblock commit/cleaner kthread.
3296 void btrfs_add_delayed_iput(struct inode *inode)
3298 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3299 struct btrfs_inode *binode = BTRFS_I(inode);
3301 if (atomic_add_unless(&inode->i_count, -1, 1))
3304 atomic_inc(&fs_info->nr_delayed_iputs);
3305 spin_lock(&fs_info->delayed_iput_lock);
3306 ASSERT(list_empty(&binode->delayed_iput));
3307 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3308 spin_unlock(&fs_info->delayed_iput_lock);
3309 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3310 wake_up_process(fs_info->cleaner_kthread);
3313 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3314 struct btrfs_inode *inode)
3316 list_del_init(&inode->delayed_iput);
3317 spin_unlock(&fs_info->delayed_iput_lock);
3318 iput(&inode->vfs_inode);
3319 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3320 wake_up(&fs_info->delayed_iputs_wait);
3321 spin_lock(&fs_info->delayed_iput_lock);
3324 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3325 struct btrfs_inode *inode)
3327 if (!list_empty(&inode->delayed_iput)) {
3328 spin_lock(&fs_info->delayed_iput_lock);
3329 if (!list_empty(&inode->delayed_iput))
3330 run_delayed_iput_locked(fs_info, inode);
3331 spin_unlock(&fs_info->delayed_iput_lock);
3335 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3338 spin_lock(&fs_info->delayed_iput_lock);
3339 while (!list_empty(&fs_info->delayed_iputs)) {
3340 struct btrfs_inode *inode;
3342 inode = list_first_entry(&fs_info->delayed_iputs,
3343 struct btrfs_inode, delayed_iput);
3344 run_delayed_iput_locked(fs_info, inode);
3345 cond_resched_lock(&fs_info->delayed_iput_lock);
3347 spin_unlock(&fs_info->delayed_iput_lock);
3351 * Wait for flushing all delayed iputs
3353 * @fs_info: the filesystem
3355 * This will wait on any delayed iputs that are currently running with KILLABLE
3356 * set. Once they are all done running we will return, unless we are killed in
3357 * which case we return EINTR. This helps in user operations like fallocate etc
3358 * that might get blocked on the iputs.
3360 * Return EINTR if we were killed, 0 if nothing's pending
3362 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3364 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3365 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3372 * This creates an orphan entry for the given inode in case something goes wrong
3373 * in the middle of an unlink.
3375 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3376 struct btrfs_inode *inode)
3380 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3381 if (ret && ret != -EEXIST) {
3382 btrfs_abort_transaction(trans, ret);
3390 * We have done the delete so we can go ahead and remove the orphan item for
3391 * this particular inode.
3393 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3394 struct btrfs_inode *inode)
3396 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3400 * this cleans up any orphans that may be left on the list from the last use
3403 int btrfs_orphan_cleanup(struct btrfs_root *root)
3405 struct btrfs_fs_info *fs_info = root->fs_info;
3406 struct btrfs_path *path;
3407 struct extent_buffer *leaf;
3408 struct btrfs_key key, found_key;
3409 struct btrfs_trans_handle *trans;
3410 struct inode *inode;
3411 u64 last_objectid = 0;
3412 int ret = 0, nr_unlink = 0;
3414 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3417 path = btrfs_alloc_path();
3422 path->reada = READA_BACK;
3424 key.objectid = BTRFS_ORPHAN_OBJECTID;
3425 key.type = BTRFS_ORPHAN_ITEM_KEY;
3426 key.offset = (u64)-1;
3429 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3434 * if ret == 0 means we found what we were searching for, which
3435 * is weird, but possible, so only screw with path if we didn't
3436 * find the key and see if we have stuff that matches
3440 if (path->slots[0] == 0)
3445 /* pull out the item */
3446 leaf = path->nodes[0];
3447 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3449 /* make sure the item matches what we want */
3450 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3452 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3455 /* release the path since we're done with it */
3456 btrfs_release_path(path);
3459 * this is where we are basically btrfs_lookup, without the
3460 * crossing root thing. we store the inode number in the
3461 * offset of the orphan item.
3464 if (found_key.offset == last_objectid) {
3466 "Error removing orphan entry, stopping orphan cleanup");
3471 last_objectid = found_key.offset;
3473 found_key.objectid = found_key.offset;
3474 found_key.type = BTRFS_INODE_ITEM_KEY;
3475 found_key.offset = 0;
3476 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3477 ret = PTR_ERR_OR_ZERO(inode);
3478 if (ret && ret != -ENOENT)
3481 if (ret == -ENOENT && root == fs_info->tree_root) {
3482 struct btrfs_root *dead_root;
3483 int is_dead_root = 0;
3486 * This is an orphan in the tree root. Currently these
3487 * could come from 2 sources:
3488 * a) a root (snapshot/subvolume) deletion in progress
3489 * b) a free space cache inode
3490 * We need to distinguish those two, as the orphan item
3491 * for a root must not get deleted before the deletion
3492 * of the snapshot/subvolume's tree completes.
3494 * btrfs_find_orphan_roots() ran before us, which has
3495 * found all deleted roots and loaded them into
3496 * fs_info->fs_roots_radix. So here we can find if an
3497 * orphan item corresponds to a deleted root by looking
3498 * up the root from that radix tree.
3501 spin_lock(&fs_info->fs_roots_radix_lock);
3502 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3503 (unsigned long)found_key.objectid);
3504 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3506 spin_unlock(&fs_info->fs_roots_radix_lock);
3509 /* prevent this orphan from being found again */
3510 key.offset = found_key.objectid - 1;
3517 * If we have an inode with links, there are a couple of
3518 * possibilities. Old kernels (before v3.12) used to create an
3519 * orphan item for truncate indicating that there were possibly
3520 * extent items past i_size that needed to be deleted. In v3.12,
3521 * truncate was changed to update i_size in sync with the extent
3522 * items, but the (useless) orphan item was still created. Since
3523 * v4.18, we don't create the orphan item for truncate at all.
3525 * So, this item could mean that we need to do a truncate, but
3526 * only if this filesystem was last used on a pre-v3.12 kernel
3527 * and was not cleanly unmounted. The odds of that are quite
3528 * slim, and it's a pain to do the truncate now, so just delete
3531 * It's also possible that this orphan item was supposed to be
3532 * deleted but wasn't. The inode number may have been reused,
3533 * but either way, we can delete the orphan item.
3535 if (ret == -ENOENT || inode->i_nlink) {
3538 trans = btrfs_start_transaction(root, 1);
3539 if (IS_ERR(trans)) {
3540 ret = PTR_ERR(trans);
3543 btrfs_debug(fs_info, "auto deleting %Lu",
3544 found_key.objectid);
3545 ret = btrfs_del_orphan_item(trans, root,
3546 found_key.objectid);
3547 btrfs_end_transaction(trans);
3555 /* this will do delete_inode and everything for us */
3558 /* release the path since we're done with it */
3559 btrfs_release_path(path);
3561 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3563 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3564 trans = btrfs_join_transaction(root);
3566 btrfs_end_transaction(trans);
3570 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3574 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3575 btrfs_free_path(path);
3580 * very simple check to peek ahead in the leaf looking for xattrs. If we
3581 * don't find any xattrs, we know there can't be any acls.
3583 * slot is the slot the inode is in, objectid is the objectid of the inode
3585 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3586 int slot, u64 objectid,
3587 int *first_xattr_slot)
3589 u32 nritems = btrfs_header_nritems(leaf);
3590 struct btrfs_key found_key;
3591 static u64 xattr_access = 0;
3592 static u64 xattr_default = 0;
3595 if (!xattr_access) {
3596 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3597 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3598 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3599 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3603 *first_xattr_slot = -1;
3604 while (slot < nritems) {
3605 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3607 /* we found a different objectid, there must not be acls */
3608 if (found_key.objectid != objectid)
3611 /* we found an xattr, assume we've got an acl */
3612 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3613 if (*first_xattr_slot == -1)
3614 *first_xattr_slot = slot;
3615 if (found_key.offset == xattr_access ||
3616 found_key.offset == xattr_default)
3621 * we found a key greater than an xattr key, there can't
3622 * be any acls later on
3624 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3631 * it goes inode, inode backrefs, xattrs, extents,
3632 * so if there are a ton of hard links to an inode there can
3633 * be a lot of backrefs. Don't waste time searching too hard,
3634 * this is just an optimization
3639 /* we hit the end of the leaf before we found an xattr or
3640 * something larger than an xattr. We have to assume the inode
3643 if (*first_xattr_slot == -1)
3644 *first_xattr_slot = slot;
3649 * read an inode from the btree into the in-memory inode
3651 static int btrfs_read_locked_inode(struct inode *inode,
3652 struct btrfs_path *in_path)
3654 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3655 struct btrfs_path *path = in_path;
3656 struct extent_buffer *leaf;
3657 struct btrfs_inode_item *inode_item;
3658 struct btrfs_root *root = BTRFS_I(inode)->root;
3659 struct btrfs_key location;
3664 bool filled = false;
3665 int first_xattr_slot;
3667 ret = btrfs_fill_inode(inode, &rdev);
3672 path = btrfs_alloc_path();
3677 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3679 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3681 if (path != in_path)
3682 btrfs_free_path(path);
3686 leaf = path->nodes[0];
3691 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3692 struct btrfs_inode_item);
3693 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3694 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3695 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3696 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3697 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3698 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3699 round_up(i_size_read(inode), fs_info->sectorsize));
3701 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3702 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3704 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3705 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3707 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3708 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3710 BTRFS_I(inode)->i_otime.tv_sec =
3711 btrfs_timespec_sec(leaf, &inode_item->otime);
3712 BTRFS_I(inode)->i_otime.tv_nsec =
3713 btrfs_timespec_nsec(leaf, &inode_item->otime);
3715 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3716 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3717 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3719 inode_set_iversion_queried(inode,
3720 btrfs_inode_sequence(leaf, inode_item));
3721 inode->i_generation = BTRFS_I(inode)->generation;
3723 rdev = btrfs_inode_rdev(leaf, inode_item);
3725 BTRFS_I(inode)->index_cnt = (u64)-1;
3726 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3730 * If we were modified in the current generation and evicted from memory
3731 * and then re-read we need to do a full sync since we don't have any
3732 * idea about which extents were modified before we were evicted from
3735 * This is required for both inode re-read from disk and delayed inode
3736 * in delayed_nodes_tree.
3738 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3739 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3740 &BTRFS_I(inode)->runtime_flags);
3743 * We don't persist the id of the transaction where an unlink operation
3744 * against the inode was last made. So here we assume the inode might
3745 * have been evicted, and therefore the exact value of last_unlink_trans
3746 * lost, and set it to last_trans to avoid metadata inconsistencies
3747 * between the inode and its parent if the inode is fsync'ed and the log
3748 * replayed. For example, in the scenario:
3751 * ln mydir/foo mydir/bar
3754 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3755 * xfs_io -c fsync mydir/foo
3757 * mount fs, triggers fsync log replay
3759 * We must make sure that when we fsync our inode foo we also log its
3760 * parent inode, otherwise after log replay the parent still has the
3761 * dentry with the "bar" name but our inode foo has a link count of 1
3762 * and doesn't have an inode ref with the name "bar" anymore.
3764 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3765 * but it guarantees correctness at the expense of occasional full
3766 * transaction commits on fsync if our inode is a directory, or if our
3767 * inode is not a directory, logging its parent unnecessarily.
3769 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3772 * Same logic as for last_unlink_trans. We don't persist the generation
3773 * of the last transaction where this inode was used for a reflink
3774 * operation, so after eviction and reloading the inode we must be
3775 * pessimistic and assume the last transaction that modified the inode.
3777 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3780 if (inode->i_nlink != 1 ||
3781 path->slots[0] >= btrfs_header_nritems(leaf))
3784 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3785 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3788 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3789 if (location.type == BTRFS_INODE_REF_KEY) {
3790 struct btrfs_inode_ref *ref;
3792 ref = (struct btrfs_inode_ref *)ptr;
3793 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3794 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3795 struct btrfs_inode_extref *extref;
3797 extref = (struct btrfs_inode_extref *)ptr;
3798 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3803 * try to precache a NULL acl entry for files that don't have
3804 * any xattrs or acls
3806 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3807 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3808 if (first_xattr_slot != -1) {
3809 path->slots[0] = first_xattr_slot;
3810 ret = btrfs_load_inode_props(inode, path);
3813 "error loading props for ino %llu (root %llu): %d",
3814 btrfs_ino(BTRFS_I(inode)),
3815 root->root_key.objectid, ret);
3817 if (path != in_path)
3818 btrfs_free_path(path);
3821 cache_no_acl(inode);
3823 switch (inode->i_mode & S_IFMT) {
3825 inode->i_mapping->a_ops = &btrfs_aops;
3826 inode->i_fop = &btrfs_file_operations;
3827 inode->i_op = &btrfs_file_inode_operations;
3830 inode->i_fop = &btrfs_dir_file_operations;
3831 inode->i_op = &btrfs_dir_inode_operations;
3834 inode->i_op = &btrfs_symlink_inode_operations;
3835 inode_nohighmem(inode);
3836 inode->i_mapping->a_ops = &btrfs_aops;
3839 inode->i_op = &btrfs_special_inode_operations;
3840 init_special_inode(inode, inode->i_mode, rdev);
3844 btrfs_sync_inode_flags_to_i_flags(inode);
3849 * given a leaf and an inode, copy the inode fields into the leaf
3851 static void fill_inode_item(struct btrfs_trans_handle *trans,
3852 struct extent_buffer *leaf,
3853 struct btrfs_inode_item *item,
3854 struct inode *inode)
3856 struct btrfs_map_token token;
3858 btrfs_init_map_token(&token, leaf);
3860 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3861 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3862 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3863 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3864 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3866 btrfs_set_token_timespec_sec(&token, &item->atime,
3867 inode->i_atime.tv_sec);
3868 btrfs_set_token_timespec_nsec(&token, &item->atime,
3869 inode->i_atime.tv_nsec);
3871 btrfs_set_token_timespec_sec(&token, &item->mtime,
3872 inode->i_mtime.tv_sec);
3873 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3874 inode->i_mtime.tv_nsec);
3876 btrfs_set_token_timespec_sec(&token, &item->ctime,
3877 inode->i_ctime.tv_sec);
3878 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3879 inode->i_ctime.tv_nsec);
3881 btrfs_set_token_timespec_sec(&token, &item->otime,
3882 BTRFS_I(inode)->i_otime.tv_sec);
3883 btrfs_set_token_timespec_nsec(&token, &item->otime,
3884 BTRFS_I(inode)->i_otime.tv_nsec);
3886 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3887 btrfs_set_token_inode_generation(&token, item,
3888 BTRFS_I(inode)->generation);
3889 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3890 btrfs_set_token_inode_transid(&token, item, trans->transid);
3891 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3892 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3893 btrfs_set_token_inode_block_group(&token, item, 0);
3897 * copy everything in the in-memory inode into the btree.
3899 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3900 struct btrfs_root *root,
3901 struct btrfs_inode *inode)
3903 struct btrfs_inode_item *inode_item;
3904 struct btrfs_path *path;
3905 struct extent_buffer *leaf;
3908 path = btrfs_alloc_path();
3912 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3919 leaf = path->nodes[0];
3920 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3921 struct btrfs_inode_item);
3923 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3924 btrfs_mark_buffer_dirty(leaf);
3925 btrfs_set_inode_last_trans(trans, inode);
3928 btrfs_free_path(path);
3933 * copy everything in the in-memory inode into the btree.
3935 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3936 struct btrfs_root *root,
3937 struct btrfs_inode *inode)
3939 struct btrfs_fs_info *fs_info = root->fs_info;
3943 * If the inode is a free space inode, we can deadlock during commit
3944 * if we put it into the delayed code.
3946 * The data relocation inode should also be directly updated
3949 if (!btrfs_is_free_space_inode(inode)
3950 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3951 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3952 btrfs_update_root_times(trans, root);
3954 ret = btrfs_delayed_update_inode(trans, root, inode);
3956 btrfs_set_inode_last_trans(trans, inode);
3960 return btrfs_update_inode_item(trans, root, inode);
3963 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3964 struct btrfs_root *root, struct btrfs_inode *inode)
3968 ret = btrfs_update_inode(trans, root, inode);
3970 return btrfs_update_inode_item(trans, root, inode);
3975 * unlink helper that gets used here in inode.c and in the tree logging
3976 * recovery code. It remove a link in a directory with a given name, and
3977 * also drops the back refs in the inode to the directory
3979 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3980 struct btrfs_root *root,
3981 struct btrfs_inode *dir,
3982 struct btrfs_inode *inode,
3983 const char *name, int name_len)
3985 struct btrfs_fs_info *fs_info = root->fs_info;
3986 struct btrfs_path *path;
3988 struct btrfs_dir_item *di;
3990 u64 ino = btrfs_ino(inode);
3991 u64 dir_ino = btrfs_ino(dir);
3993 path = btrfs_alloc_path();
3999 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4000 name, name_len, -1);
4001 if (IS_ERR_OR_NULL(di)) {
4002 ret = di ? PTR_ERR(di) : -ENOENT;
4005 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4008 btrfs_release_path(path);
4011 * If we don't have dir index, we have to get it by looking up
4012 * the inode ref, since we get the inode ref, remove it directly,
4013 * it is unnecessary to do delayed deletion.
4015 * But if we have dir index, needn't search inode ref to get it.
4016 * Since the inode ref is close to the inode item, it is better
4017 * that we delay to delete it, and just do this deletion when
4018 * we update the inode item.
4020 if (inode->dir_index) {
4021 ret = btrfs_delayed_delete_inode_ref(inode);
4023 index = inode->dir_index;
4028 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4032 "failed to delete reference to %.*s, inode %llu parent %llu",
4033 name_len, name, ino, dir_ino);
4034 btrfs_abort_transaction(trans, ret);
4038 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4040 btrfs_abort_transaction(trans, ret);
4044 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4046 if (ret != 0 && ret != -ENOENT) {
4047 btrfs_abort_transaction(trans, ret);
4051 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4056 btrfs_abort_transaction(trans, ret);
4059 * If we have a pending delayed iput we could end up with the final iput
4060 * being run in btrfs-cleaner context. If we have enough of these built
4061 * up we can end up burning a lot of time in btrfs-cleaner without any
4062 * way to throttle the unlinks. Since we're currently holding a ref on
4063 * the inode we can run the delayed iput here without any issues as the
4064 * final iput won't be done until after we drop the ref we're currently
4067 btrfs_run_delayed_iput(fs_info, inode);
4069 btrfs_free_path(path);
4073 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4074 inode_inc_iversion(&inode->vfs_inode);
4075 inode_inc_iversion(&dir->vfs_inode);
4076 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4077 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4078 ret = btrfs_update_inode(trans, root, dir);
4083 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4084 struct btrfs_root *root,
4085 struct btrfs_inode *dir, struct btrfs_inode *inode,
4086 const char *name, int name_len)
4089 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4091 drop_nlink(&inode->vfs_inode);
4092 ret = btrfs_update_inode(trans, root, inode);
4098 * helper to start transaction for unlink and rmdir.
4100 * unlink and rmdir are special in btrfs, they do not always free space, so
4101 * if we cannot make our reservations the normal way try and see if there is
4102 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4103 * allow the unlink to occur.
4105 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4107 struct btrfs_root *root = BTRFS_I(dir)->root;
4110 * 1 for the possible orphan item
4111 * 1 for the dir item
4112 * 1 for the dir index
4113 * 1 for the inode ref
4116 return btrfs_start_transaction_fallback_global_rsv(root, 5);
4119 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4121 struct btrfs_root *root = BTRFS_I(dir)->root;
4122 struct btrfs_trans_handle *trans;
4123 struct inode *inode = d_inode(dentry);
4126 trans = __unlink_start_trans(dir);
4128 return PTR_ERR(trans);
4130 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4133 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4134 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4135 dentry->d_name.len);
4139 if (inode->i_nlink == 0) {
4140 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4146 btrfs_end_transaction(trans);
4147 btrfs_btree_balance_dirty(root->fs_info);
4151 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4152 struct inode *dir, struct dentry *dentry)
4154 struct btrfs_root *root = BTRFS_I(dir)->root;
4155 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4156 struct btrfs_path *path;
4157 struct extent_buffer *leaf;
4158 struct btrfs_dir_item *di;
4159 struct btrfs_key key;
4160 const char *name = dentry->d_name.name;
4161 int name_len = dentry->d_name.len;
4165 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4167 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4168 objectid = inode->root->root_key.objectid;
4169 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4170 objectid = inode->location.objectid;
4176 path = btrfs_alloc_path();
4180 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4181 name, name_len, -1);
4182 if (IS_ERR_OR_NULL(di)) {
4183 ret = di ? PTR_ERR(di) : -ENOENT;
4187 leaf = path->nodes[0];
4188 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4189 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4190 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4192 btrfs_abort_transaction(trans, ret);
4195 btrfs_release_path(path);
4198 * This is a placeholder inode for a subvolume we didn't have a
4199 * reference to at the time of the snapshot creation. In the meantime
4200 * we could have renamed the real subvol link into our snapshot, so
4201 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
4202 * Instead simply lookup the dir_index_item for this entry so we can
4203 * remove it. Otherwise we know we have a ref to the root and we can
4204 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4206 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4207 di = btrfs_search_dir_index_item(root, path, dir_ino,
4209 if (IS_ERR_OR_NULL(di)) {
4214 btrfs_abort_transaction(trans, ret);
4218 leaf = path->nodes[0];
4219 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4221 btrfs_release_path(path);
4223 ret = btrfs_del_root_ref(trans, objectid,
4224 root->root_key.objectid, dir_ino,
4225 &index, name, name_len);
4227 btrfs_abort_transaction(trans, ret);
4232 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4234 btrfs_abort_transaction(trans, ret);
4238 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4239 inode_inc_iversion(dir);
4240 dir->i_mtime = dir->i_ctime = current_time(dir);
4241 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
4243 btrfs_abort_transaction(trans, ret);
4245 btrfs_free_path(path);
4250 * Helper to check if the subvolume references other subvolumes or if it's
4253 static noinline int may_destroy_subvol(struct btrfs_root *root)
4255 struct btrfs_fs_info *fs_info = root->fs_info;
4256 struct btrfs_path *path;
4257 struct btrfs_dir_item *di;
4258 struct btrfs_key key;
4262 path = btrfs_alloc_path();
4266 /* Make sure this root isn't set as the default subvol */
4267 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4268 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4269 dir_id, "default", 7, 0);
4270 if (di && !IS_ERR(di)) {
4271 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4272 if (key.objectid == root->root_key.objectid) {
4275 "deleting default subvolume %llu is not allowed",
4279 btrfs_release_path(path);
4282 key.objectid = root->root_key.objectid;
4283 key.type = BTRFS_ROOT_REF_KEY;
4284 key.offset = (u64)-1;
4286 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4292 if (path->slots[0] > 0) {
4294 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4295 if (key.objectid == root->root_key.objectid &&
4296 key.type == BTRFS_ROOT_REF_KEY)
4300 btrfs_free_path(path);
4304 /* Delete all dentries for inodes belonging to the root */
4305 static void btrfs_prune_dentries(struct btrfs_root *root)
4307 struct btrfs_fs_info *fs_info = root->fs_info;
4308 struct rb_node *node;
4309 struct rb_node *prev;
4310 struct btrfs_inode *entry;
4311 struct inode *inode;
4314 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4315 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4317 spin_lock(&root->inode_lock);
4319 node = root->inode_tree.rb_node;
4323 entry = rb_entry(node, struct btrfs_inode, rb_node);
4325 if (objectid < btrfs_ino(entry))
4326 node = node->rb_left;
4327 else if (objectid > btrfs_ino(entry))
4328 node = node->rb_right;
4334 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4335 if (objectid <= btrfs_ino(entry)) {
4339 prev = rb_next(prev);
4343 entry = rb_entry(node, struct btrfs_inode, rb_node);
4344 objectid = btrfs_ino(entry) + 1;
4345 inode = igrab(&entry->vfs_inode);
4347 spin_unlock(&root->inode_lock);
4348 if (atomic_read(&inode->i_count) > 1)
4349 d_prune_aliases(inode);
4351 * btrfs_drop_inode will have it removed from the inode
4352 * cache when its usage count hits zero.
4356 spin_lock(&root->inode_lock);
4360 if (cond_resched_lock(&root->inode_lock))
4363 node = rb_next(node);
4365 spin_unlock(&root->inode_lock);
4368 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4370 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4371 struct btrfs_root *root = BTRFS_I(dir)->root;
4372 struct inode *inode = d_inode(dentry);
4373 struct btrfs_root *dest = BTRFS_I(inode)->root;
4374 struct btrfs_trans_handle *trans;
4375 struct btrfs_block_rsv block_rsv;
4380 * Don't allow to delete a subvolume with send in progress. This is
4381 * inside the inode lock so the error handling that has to drop the bit
4382 * again is not run concurrently.
4384 spin_lock(&dest->root_item_lock);
4385 if (dest->send_in_progress) {
4386 spin_unlock(&dest->root_item_lock);
4388 "attempt to delete subvolume %llu during send",
4389 dest->root_key.objectid);
4392 root_flags = btrfs_root_flags(&dest->root_item);
4393 btrfs_set_root_flags(&dest->root_item,
4394 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4395 spin_unlock(&dest->root_item_lock);
4397 down_write(&fs_info->subvol_sem);
4399 ret = may_destroy_subvol(dest);
4403 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4405 * One for dir inode,
4406 * two for dir entries,
4407 * two for root ref/backref.
4409 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4413 trans = btrfs_start_transaction(root, 0);
4414 if (IS_ERR(trans)) {
4415 ret = PTR_ERR(trans);
4418 trans->block_rsv = &block_rsv;
4419 trans->bytes_reserved = block_rsv.size;
4421 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4423 ret = btrfs_unlink_subvol(trans, dir, dentry);
4425 btrfs_abort_transaction(trans, ret);
4429 ret = btrfs_record_root_in_trans(trans, dest);
4431 btrfs_abort_transaction(trans, ret);
4435 memset(&dest->root_item.drop_progress, 0,
4436 sizeof(dest->root_item.drop_progress));
4437 btrfs_set_root_drop_level(&dest->root_item, 0);
4438 btrfs_set_root_refs(&dest->root_item, 0);
4440 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4441 ret = btrfs_insert_orphan_item(trans,
4443 dest->root_key.objectid);
4445 btrfs_abort_transaction(trans, ret);
4450 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4451 BTRFS_UUID_KEY_SUBVOL,
4452 dest->root_key.objectid);
4453 if (ret && ret != -ENOENT) {
4454 btrfs_abort_transaction(trans, ret);
4457 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4458 ret = btrfs_uuid_tree_remove(trans,
4459 dest->root_item.received_uuid,
4460 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4461 dest->root_key.objectid);
4462 if (ret && ret != -ENOENT) {
4463 btrfs_abort_transaction(trans, ret);
4468 free_anon_bdev(dest->anon_dev);
4471 trans->block_rsv = NULL;
4472 trans->bytes_reserved = 0;
4473 ret = btrfs_end_transaction(trans);
4474 inode->i_flags |= S_DEAD;
4476 btrfs_subvolume_release_metadata(root, &block_rsv);
4478 up_write(&fs_info->subvol_sem);
4480 spin_lock(&dest->root_item_lock);
4481 root_flags = btrfs_root_flags(&dest->root_item);
4482 btrfs_set_root_flags(&dest->root_item,
4483 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4484 spin_unlock(&dest->root_item_lock);
4486 d_invalidate(dentry);
4487 btrfs_prune_dentries(dest);
4488 ASSERT(dest->send_in_progress == 0);
4494 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4496 struct inode *inode = d_inode(dentry);
4498 struct btrfs_root *root = BTRFS_I(dir)->root;
4499 struct btrfs_trans_handle *trans;
4500 u64 last_unlink_trans;
4502 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4504 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4505 return btrfs_delete_subvolume(dir, dentry);
4507 trans = __unlink_start_trans(dir);
4509 return PTR_ERR(trans);
4511 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4512 err = btrfs_unlink_subvol(trans, dir, dentry);
4516 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4520 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4522 /* now the directory is empty */
4523 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4524 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4525 dentry->d_name.len);
4527 btrfs_i_size_write(BTRFS_I(inode), 0);
4529 * Propagate the last_unlink_trans value of the deleted dir to
4530 * its parent directory. This is to prevent an unrecoverable
4531 * log tree in the case we do something like this:
4533 * 2) create snapshot under dir foo
4534 * 3) delete the snapshot
4537 * 6) fsync foo or some file inside foo
4539 if (last_unlink_trans >= trans->transid)
4540 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4543 btrfs_end_transaction(trans);
4544 btrfs_btree_balance_dirty(root->fs_info);
4550 * Return this if we need to call truncate_block for the last bit of the
4553 #define NEED_TRUNCATE_BLOCK 1
4556 * this can truncate away extent items, csum items and directory items.
4557 * It starts at a high offset and removes keys until it can't find
4558 * any higher than new_size
4560 * csum items that cross the new i_size are truncated to the new size
4563 * min_type is the minimum key type to truncate down to. If set to 0, this
4564 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4566 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4567 struct btrfs_root *root,
4568 struct btrfs_inode *inode,
4569 u64 new_size, u32 min_type)
4571 struct btrfs_fs_info *fs_info = root->fs_info;
4572 struct btrfs_path *path;
4573 struct extent_buffer *leaf;
4574 struct btrfs_file_extent_item *fi;
4575 struct btrfs_key key;
4576 struct btrfs_key found_key;
4577 u64 extent_start = 0;
4578 u64 extent_num_bytes = 0;
4579 u64 extent_offset = 0;
4581 u64 last_size = new_size;
4582 u32 found_type = (u8)-1;
4585 int pending_del_nr = 0;
4586 int pending_del_slot = 0;
4587 int extent_type = -1;
4589 u64 ino = btrfs_ino(inode);
4590 u64 bytes_deleted = 0;
4591 bool be_nice = false;
4592 bool should_throttle = false;
4593 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4594 struct extent_state *cached_state = NULL;
4596 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4599 * For non-free space inodes and non-shareable roots, we want to back
4600 * off from time to time. This means all inodes in subvolume roots,
4601 * reloc roots, and data reloc roots.
4603 if (!btrfs_is_free_space_inode(inode) &&
4604 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4607 path = btrfs_alloc_path();
4610 path->reada = READA_BACK;
4612 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4613 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4617 * We want to drop from the next block forward in case this
4618 * new size is not block aligned since we will be keeping the
4619 * last block of the extent just the way it is.
4621 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4622 fs_info->sectorsize),
4627 * This function is also used to drop the items in the log tree before
4628 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4629 * it is used to drop the logged items. So we shouldn't kill the delayed
4632 if (min_type == 0 && root == inode->root)
4633 btrfs_kill_delayed_inode_items(inode);
4636 key.offset = (u64)-1;
4641 * with a 16K leaf size and 128MB extents, you can actually queue
4642 * up a huge file in a single leaf. Most of the time that
4643 * bytes_deleted is > 0, it will be huge by the time we get here
4645 if (be_nice && bytes_deleted > SZ_32M &&
4646 btrfs_should_end_transaction(trans)) {
4651 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4657 /* there are no items in the tree for us to truncate, we're
4660 if (path->slots[0] == 0)
4666 u64 clear_start = 0, clear_len = 0;
4669 leaf = path->nodes[0];
4670 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4671 found_type = found_key.type;
4673 if (found_key.objectid != ino)
4676 if (found_type < min_type)
4679 item_end = found_key.offset;
4680 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4681 fi = btrfs_item_ptr(leaf, path->slots[0],
4682 struct btrfs_file_extent_item);
4683 extent_type = btrfs_file_extent_type(leaf, fi);
4684 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4686 btrfs_file_extent_num_bytes(leaf, fi);
4688 trace_btrfs_truncate_show_fi_regular(
4689 inode, leaf, fi, found_key.offset);
4690 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4691 item_end += btrfs_file_extent_ram_bytes(leaf,
4694 trace_btrfs_truncate_show_fi_inline(
4695 inode, leaf, fi, path->slots[0],
4700 if (found_type > min_type) {
4703 if (item_end < new_size)
4705 if (found_key.offset >= new_size)
4711 /* FIXME, shrink the extent if the ref count is only 1 */
4712 if (found_type != BTRFS_EXTENT_DATA_KEY)
4715 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4718 clear_start = found_key.offset;
4719 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4721 u64 orig_num_bytes =
4722 btrfs_file_extent_num_bytes(leaf, fi);
4723 extent_num_bytes = ALIGN(new_size -
4725 fs_info->sectorsize);
4726 clear_start = ALIGN(new_size, fs_info->sectorsize);
4727 btrfs_set_file_extent_num_bytes(leaf, fi,
4729 num_dec = (orig_num_bytes -
4731 if (test_bit(BTRFS_ROOT_SHAREABLE,
4734 inode_sub_bytes(&inode->vfs_inode,
4736 btrfs_mark_buffer_dirty(leaf);
4739 btrfs_file_extent_disk_num_bytes(leaf,
4741 extent_offset = found_key.offset -
4742 btrfs_file_extent_offset(leaf, fi);
4744 /* FIXME blocksize != 4096 */
4745 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4746 if (extent_start != 0) {
4748 if (test_bit(BTRFS_ROOT_SHAREABLE,
4750 inode_sub_bytes(&inode->vfs_inode,
4754 clear_len = num_dec;
4755 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4757 * we can't truncate inline items that have had
4761 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4762 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4763 btrfs_file_extent_compression(leaf, fi) == 0) {
4764 u32 size = (u32)(new_size - found_key.offset);
4766 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4767 size = btrfs_file_extent_calc_inline_size(size);
4768 btrfs_truncate_item(path, size, 1);
4769 } else if (!del_item) {
4771 * We have to bail so the last_size is set to
4772 * just before this extent.
4774 ret = NEED_TRUNCATE_BLOCK;
4778 * Inline extents are special, we just treat
4779 * them as a full sector worth in the file
4780 * extent tree just for simplicity sake.
4782 clear_len = fs_info->sectorsize;
4785 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4786 inode_sub_bytes(&inode->vfs_inode,
4787 item_end + 1 - new_size);
4791 * We use btrfs_truncate_inode_items() to clean up log trees for
4792 * multiple fsyncs, and in this case we don't want to clear the
4793 * file extent range because it's just the log.
4795 if (root == inode->root) {
4796 ret = btrfs_inode_clear_file_extent_range(inode,
4797 clear_start, clear_len);
4799 btrfs_abort_transaction(trans, ret);
4805 last_size = found_key.offset;
4807 last_size = new_size;
4809 if (!pending_del_nr) {
4810 /* no pending yet, add ourselves */
4811 pending_del_slot = path->slots[0];
4813 } else if (pending_del_nr &&
4814 path->slots[0] + 1 == pending_del_slot) {
4815 /* hop on the pending chunk */
4817 pending_del_slot = path->slots[0];
4824 should_throttle = false;
4827 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4828 struct btrfs_ref ref = { 0 };
4830 bytes_deleted += extent_num_bytes;
4832 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4833 extent_start, extent_num_bytes, 0);
4834 ref.real_root = root->root_key.objectid;
4835 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4836 ino, extent_offset);
4837 ret = btrfs_free_extent(trans, &ref);
4839 btrfs_abort_transaction(trans, ret);
4843 if (btrfs_should_throttle_delayed_refs(trans))
4844 should_throttle = true;
4848 if (found_type == BTRFS_INODE_ITEM_KEY)
4851 if (path->slots[0] == 0 ||
4852 path->slots[0] != pending_del_slot ||
4854 if (pending_del_nr) {
4855 ret = btrfs_del_items(trans, root, path,
4859 btrfs_abort_transaction(trans, ret);
4864 btrfs_release_path(path);
4867 * We can generate a lot of delayed refs, so we need to
4868 * throttle every once and a while and make sure we're
4869 * adding enough space to keep up with the work we are
4870 * generating. Since we hold a transaction here we
4871 * can't flush, and we don't want to FLUSH_LIMIT because
4872 * we could have generated too many delayed refs to
4873 * actually allocate, so just bail if we're short and
4874 * let the normal reservation dance happen higher up.
4876 if (should_throttle) {
4877 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4878 BTRFS_RESERVE_NO_FLUSH);
4890 if (ret >= 0 && pending_del_nr) {
4893 err = btrfs_del_items(trans, root, path, pending_del_slot,
4896 btrfs_abort_transaction(trans, err);
4900 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4901 ASSERT(last_size >= new_size);
4902 if (!ret && last_size > new_size)
4903 last_size = new_size;
4904 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4905 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4909 btrfs_free_path(path);
4914 * btrfs_truncate_block - read, zero a chunk and write a block
4915 * @inode - inode that we're zeroing
4916 * @from - the offset to start zeroing
4917 * @len - the length to zero, 0 to zero the entire range respective to the
4919 * @front - zero up to the offset instead of from the offset on
4921 * This will find the block for the "from" offset and cow the block and zero the
4922 * part we want to zero. This is used with truncate and hole punching.
4924 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4927 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4928 struct address_space *mapping = inode->vfs_inode.i_mapping;
4929 struct extent_io_tree *io_tree = &inode->io_tree;
4930 struct btrfs_ordered_extent *ordered;
4931 struct extent_state *cached_state = NULL;
4932 struct extent_changeset *data_reserved = NULL;
4933 bool only_release_metadata = false;
4934 u32 blocksize = fs_info->sectorsize;
4935 pgoff_t index = from >> PAGE_SHIFT;
4936 unsigned offset = from & (blocksize - 1);
4938 gfp_t mask = btrfs_alloc_write_mask(mapping);
4939 size_t write_bytes = blocksize;
4944 if (IS_ALIGNED(offset, blocksize) &&
4945 (!len || IS_ALIGNED(len, blocksize)))
4948 block_start = round_down(from, blocksize);
4949 block_end = block_start + blocksize - 1;
4951 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4954 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4955 /* For nocow case, no need to reserve data space */
4956 only_release_metadata = true;
4961 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4963 if (!only_release_metadata)
4964 btrfs_free_reserved_data_space(inode, data_reserved,
4965 block_start, blocksize);
4969 page = find_or_create_page(mapping, index, mask);
4971 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4973 btrfs_delalloc_release_extents(inode, blocksize);
4977 ret = set_page_extent_mapped(page);
4981 if (!PageUptodate(page)) {
4982 ret = btrfs_readpage(NULL, page);
4984 if (page->mapping != mapping) {
4989 if (!PageUptodate(page)) {
4994 wait_on_page_writeback(page);
4996 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4998 ordered = btrfs_lookup_ordered_extent(inode, block_start);
5000 unlock_extent_cached(io_tree, block_start, block_end,
5004 btrfs_start_ordered_extent(ordered, 1);
5005 btrfs_put_ordered_extent(ordered);
5009 clear_extent_bit(&inode->io_tree, block_start, block_end,
5010 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
5011 0, 0, &cached_state);
5013 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
5016 unlock_extent_cached(io_tree, block_start, block_end,
5021 if (offset != blocksize) {
5023 len = blocksize - offset;
5025 memzero_page(page, (block_start - page_offset(page)),
5028 memzero_page(page, (block_start - page_offset(page)) + offset,
5030 flush_dcache_page(page);
5032 ClearPageChecked(page);
5033 set_page_dirty(page);
5034 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
5036 if (only_release_metadata)
5037 set_extent_bit(&inode->io_tree, block_start, block_end,
5038 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
5042 if (only_release_metadata)
5043 btrfs_delalloc_release_metadata(inode, blocksize, true);
5045 btrfs_delalloc_release_space(inode, data_reserved,
5046 block_start, blocksize, true);
5048 btrfs_delalloc_release_extents(inode, blocksize);
5052 if (only_release_metadata)
5053 btrfs_check_nocow_unlock(inode);
5054 extent_changeset_free(data_reserved);
5058 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
5059 u64 offset, u64 len)
5061 struct btrfs_fs_info *fs_info = root->fs_info;
5062 struct btrfs_trans_handle *trans;
5063 struct btrfs_drop_extents_args drop_args = { 0 };
5067 * Still need to make sure the inode looks like it's been updated so
5068 * that any holes get logged if we fsync.
5070 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
5071 inode->last_trans = fs_info->generation;
5072 inode->last_sub_trans = root->log_transid;
5073 inode->last_log_commit = root->last_log_commit;
5078 * 1 - for the one we're dropping
5079 * 1 - for the one we're adding
5080 * 1 - for updating the inode.
5082 trans = btrfs_start_transaction(root, 3);
5084 return PTR_ERR(trans);
5086 drop_args.start = offset;
5087 drop_args.end = offset + len;
5088 drop_args.drop_cache = true;
5090 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
5092 btrfs_abort_transaction(trans, ret);
5093 btrfs_end_transaction(trans);
5097 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
5098 offset, 0, 0, len, 0, len, 0, 0, 0);
5100 btrfs_abort_transaction(trans, ret);
5102 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
5103 btrfs_update_inode(trans, root, inode);
5105 btrfs_end_transaction(trans);
5110 * This function puts in dummy file extents for the area we're creating a hole
5111 * for. So if we are truncating this file to a larger size we need to insert
5112 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5113 * the range between oldsize and size
5115 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5117 struct btrfs_root *root = inode->root;
5118 struct btrfs_fs_info *fs_info = root->fs_info;
5119 struct extent_io_tree *io_tree = &inode->io_tree;
5120 struct extent_map *em = NULL;
5121 struct extent_state *cached_state = NULL;
5122 struct extent_map_tree *em_tree = &inode->extent_tree;
5123 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5124 u64 block_end = ALIGN(size, fs_info->sectorsize);
5131 * If our size started in the middle of a block we need to zero out the
5132 * rest of the block before we expand the i_size, otherwise we could
5133 * expose stale data.
5135 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5139 if (size <= hole_start)
5142 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5144 cur_offset = hole_start;
5146 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5147 block_end - cur_offset);
5153 last_byte = min(extent_map_end(em), block_end);
5154 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5155 hole_size = last_byte - cur_offset;
5157 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5158 struct extent_map *hole_em;
5160 err = maybe_insert_hole(root, inode, cur_offset,
5165 err = btrfs_inode_set_file_extent_range(inode,
5166 cur_offset, hole_size);
5170 btrfs_drop_extent_cache(inode, cur_offset,
5171 cur_offset + hole_size - 1, 0);
5172 hole_em = alloc_extent_map();
5174 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5175 &inode->runtime_flags);
5178 hole_em->start = cur_offset;
5179 hole_em->len = hole_size;
5180 hole_em->orig_start = cur_offset;
5182 hole_em->block_start = EXTENT_MAP_HOLE;
5183 hole_em->block_len = 0;
5184 hole_em->orig_block_len = 0;
5185 hole_em->ram_bytes = hole_size;
5186 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5187 hole_em->generation = fs_info->generation;
5190 write_lock(&em_tree->lock);
5191 err = add_extent_mapping(em_tree, hole_em, 1);
5192 write_unlock(&em_tree->lock);
5195 btrfs_drop_extent_cache(inode, cur_offset,
5199 free_extent_map(hole_em);
5201 err = btrfs_inode_set_file_extent_range(inode,
5202 cur_offset, hole_size);
5207 free_extent_map(em);
5209 cur_offset = last_byte;
5210 if (cur_offset >= block_end)
5213 free_extent_map(em);
5214 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5218 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5220 struct btrfs_root *root = BTRFS_I(inode)->root;
5221 struct btrfs_trans_handle *trans;
5222 loff_t oldsize = i_size_read(inode);
5223 loff_t newsize = attr->ia_size;
5224 int mask = attr->ia_valid;
5228 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5229 * special case where we need to update the times despite not having
5230 * these flags set. For all other operations the VFS set these flags
5231 * explicitly if it wants a timestamp update.
5233 if (newsize != oldsize) {
5234 inode_inc_iversion(inode);
5235 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5236 inode->i_ctime = inode->i_mtime =
5237 current_time(inode);
5240 if (newsize > oldsize) {
5242 * Don't do an expanding truncate while snapshotting is ongoing.
5243 * This is to ensure the snapshot captures a fully consistent
5244 * state of this file - if the snapshot captures this expanding
5245 * truncation, it must capture all writes that happened before
5248 btrfs_drew_write_lock(&root->snapshot_lock);
5249 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5251 btrfs_drew_write_unlock(&root->snapshot_lock);
5255 trans = btrfs_start_transaction(root, 1);
5256 if (IS_ERR(trans)) {
5257 btrfs_drew_write_unlock(&root->snapshot_lock);
5258 return PTR_ERR(trans);
5261 i_size_write(inode, newsize);
5262 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5263 pagecache_isize_extended(inode, oldsize, newsize);
5264 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5265 btrfs_drew_write_unlock(&root->snapshot_lock);
5266 btrfs_end_transaction(trans);
5268 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5270 if (btrfs_is_zoned(fs_info)) {
5271 ret = btrfs_wait_ordered_range(inode,
5272 ALIGN(newsize, fs_info->sectorsize),
5279 * We're truncating a file that used to have good data down to
5280 * zero. Make sure any new writes to the file get on disk
5284 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5285 &BTRFS_I(inode)->runtime_flags);
5287 truncate_setsize(inode, newsize);
5289 inode_dio_wait(inode);
5291 ret = btrfs_truncate(inode, newsize == oldsize);
5292 if (ret && inode->i_nlink) {
5296 * Truncate failed, so fix up the in-memory size. We
5297 * adjusted disk_i_size down as we removed extents, so
5298 * wait for disk_i_size to be stable and then update the
5299 * in-memory size to match.
5301 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5304 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5311 static int btrfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5314 struct inode *inode = d_inode(dentry);
5315 struct btrfs_root *root = BTRFS_I(inode)->root;
5318 if (btrfs_root_readonly(root))
5321 err = setattr_prepare(&init_user_ns, dentry, attr);
5325 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5326 err = btrfs_setsize(inode, attr);
5331 if (attr->ia_valid) {
5332 setattr_copy(&init_user_ns, inode, attr);
5333 inode_inc_iversion(inode);
5334 err = btrfs_dirty_inode(inode);
5336 if (!err && attr->ia_valid & ATTR_MODE)
5337 err = posix_acl_chmod(&init_user_ns, inode,
5345 * While truncating the inode pages during eviction, we get the VFS calling
5346 * btrfs_invalidatepage() against each page of the inode. This is slow because
5347 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5348 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5349 * extent_state structures over and over, wasting lots of time.
5351 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5352 * those expensive operations on a per page basis and do only the ordered io
5353 * finishing, while we release here the extent_map and extent_state structures,
5354 * without the excessive merging and splitting.
5356 static void evict_inode_truncate_pages(struct inode *inode)
5358 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5359 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5360 struct rb_node *node;
5362 ASSERT(inode->i_state & I_FREEING);
5363 truncate_inode_pages_final(&inode->i_data);
5365 write_lock(&map_tree->lock);
5366 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5367 struct extent_map *em;
5369 node = rb_first_cached(&map_tree->map);
5370 em = rb_entry(node, struct extent_map, rb_node);
5371 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5372 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5373 remove_extent_mapping(map_tree, em);
5374 free_extent_map(em);
5375 if (need_resched()) {
5376 write_unlock(&map_tree->lock);
5378 write_lock(&map_tree->lock);
5381 write_unlock(&map_tree->lock);
5384 * Keep looping until we have no more ranges in the io tree.
5385 * We can have ongoing bios started by readahead that have
5386 * their endio callback (extent_io.c:end_bio_extent_readpage)
5387 * still in progress (unlocked the pages in the bio but did not yet
5388 * unlocked the ranges in the io tree). Therefore this means some
5389 * ranges can still be locked and eviction started because before
5390 * submitting those bios, which are executed by a separate task (work
5391 * queue kthread), inode references (inode->i_count) were not taken
5392 * (which would be dropped in the end io callback of each bio).
5393 * Therefore here we effectively end up waiting for those bios and
5394 * anyone else holding locked ranges without having bumped the inode's
5395 * reference count - if we don't do it, when they access the inode's
5396 * io_tree to unlock a range it may be too late, leading to an
5397 * use-after-free issue.
5399 spin_lock(&io_tree->lock);
5400 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5401 struct extent_state *state;
5402 struct extent_state *cached_state = NULL;
5405 unsigned state_flags;
5407 node = rb_first(&io_tree->state);
5408 state = rb_entry(node, struct extent_state, rb_node);
5409 start = state->start;
5411 state_flags = state->state;
5412 spin_unlock(&io_tree->lock);
5414 lock_extent_bits(io_tree, start, end, &cached_state);
5417 * If still has DELALLOC flag, the extent didn't reach disk,
5418 * and its reserved space won't be freed by delayed_ref.
5419 * So we need to free its reserved space here.
5420 * (Refer to comment in btrfs_invalidatepage, case 2)
5422 * Note, end is the bytenr of last byte, so we need + 1 here.
5424 if (state_flags & EXTENT_DELALLOC)
5425 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5428 clear_extent_bit(io_tree, start, end,
5429 EXTENT_LOCKED | EXTENT_DELALLOC |
5430 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5434 spin_lock(&io_tree->lock);
5436 spin_unlock(&io_tree->lock);
5439 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5440 struct btrfs_block_rsv *rsv)
5442 struct btrfs_fs_info *fs_info = root->fs_info;
5443 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5444 struct btrfs_trans_handle *trans;
5445 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5449 * Eviction should be taking place at some place safe because of our
5450 * delayed iputs. However the normal flushing code will run delayed
5451 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5453 * We reserve the delayed_refs_extra here again because we can't use
5454 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5455 * above. We reserve our extra bit here because we generate a ton of
5456 * delayed refs activity by truncating.
5458 * If we cannot make our reservation we'll attempt to steal from the
5459 * global reserve, because we really want to be able to free up space.
5461 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5462 BTRFS_RESERVE_FLUSH_EVICT);
5465 * Try to steal from the global reserve if there is space for
5468 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5469 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5471 "could not allocate space for delete; will truncate on mount");
5472 return ERR_PTR(-ENOSPC);
5474 delayed_refs_extra = 0;
5477 trans = btrfs_join_transaction(root);
5481 if (delayed_refs_extra) {
5482 trans->block_rsv = &fs_info->trans_block_rsv;
5483 trans->bytes_reserved = delayed_refs_extra;
5484 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5485 delayed_refs_extra, 1);
5490 void btrfs_evict_inode(struct inode *inode)
5492 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5493 struct btrfs_trans_handle *trans;
5494 struct btrfs_root *root = BTRFS_I(inode)->root;
5495 struct btrfs_block_rsv *rsv;
5498 trace_btrfs_inode_evict(inode);
5505 evict_inode_truncate_pages(inode);
5507 if (inode->i_nlink &&
5508 ((btrfs_root_refs(&root->root_item) != 0 &&
5509 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5510 btrfs_is_free_space_inode(BTRFS_I(inode))))
5513 if (is_bad_inode(inode))
5516 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5518 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5521 if (inode->i_nlink > 0) {
5522 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5523 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5527 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5531 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5534 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5537 btrfs_i_size_write(BTRFS_I(inode), 0);
5540 trans = evict_refill_and_join(root, rsv);
5544 trans->block_rsv = rsv;
5546 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5548 trans->block_rsv = &fs_info->trans_block_rsv;
5549 btrfs_end_transaction(trans);
5550 btrfs_btree_balance_dirty(fs_info);
5551 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5558 * Errors here aren't a big deal, it just means we leave orphan items in
5559 * the tree. They will be cleaned up on the next mount. If the inode
5560 * number gets reused, cleanup deletes the orphan item without doing
5561 * anything, and unlink reuses the existing orphan item.
5563 * If it turns out that we are dropping too many of these, we might want
5564 * to add a mechanism for retrying these after a commit.
5566 trans = evict_refill_and_join(root, rsv);
5567 if (!IS_ERR(trans)) {
5568 trans->block_rsv = rsv;
5569 btrfs_orphan_del(trans, BTRFS_I(inode));
5570 trans->block_rsv = &fs_info->trans_block_rsv;
5571 btrfs_end_transaction(trans);
5575 btrfs_free_block_rsv(fs_info, rsv);
5578 * If we didn't successfully delete, the orphan item will still be in
5579 * the tree and we'll retry on the next mount. Again, we might also want
5580 * to retry these periodically in the future.
5582 btrfs_remove_delayed_node(BTRFS_I(inode));
5587 * Return the key found in the dir entry in the location pointer, fill @type
5588 * with BTRFS_FT_*, and return 0.
5590 * If no dir entries were found, returns -ENOENT.
5591 * If found a corrupted location in dir entry, returns -EUCLEAN.
5593 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5594 struct btrfs_key *location, u8 *type)
5596 const char *name = dentry->d_name.name;
5597 int namelen = dentry->d_name.len;
5598 struct btrfs_dir_item *di;
5599 struct btrfs_path *path;
5600 struct btrfs_root *root = BTRFS_I(dir)->root;
5603 path = btrfs_alloc_path();
5607 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5609 if (IS_ERR_OR_NULL(di)) {
5610 ret = di ? PTR_ERR(di) : -ENOENT;
5614 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5615 if (location->type != BTRFS_INODE_ITEM_KEY &&
5616 location->type != BTRFS_ROOT_ITEM_KEY) {
5618 btrfs_warn(root->fs_info,
5619 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5620 __func__, name, btrfs_ino(BTRFS_I(dir)),
5621 location->objectid, location->type, location->offset);
5624 *type = btrfs_dir_type(path->nodes[0], di);
5626 btrfs_free_path(path);
5631 * when we hit a tree root in a directory, the btrfs part of the inode
5632 * needs to be changed to reflect the root directory of the tree root. This
5633 * is kind of like crossing a mount point.
5635 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5637 struct dentry *dentry,
5638 struct btrfs_key *location,
5639 struct btrfs_root **sub_root)
5641 struct btrfs_path *path;
5642 struct btrfs_root *new_root;
5643 struct btrfs_root_ref *ref;
5644 struct extent_buffer *leaf;
5645 struct btrfs_key key;
5649 path = btrfs_alloc_path();
5656 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5657 key.type = BTRFS_ROOT_REF_KEY;
5658 key.offset = location->objectid;
5660 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5667 leaf = path->nodes[0];
5668 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5669 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5670 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5673 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5674 (unsigned long)(ref + 1),
5675 dentry->d_name.len);
5679 btrfs_release_path(path);
5681 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5682 if (IS_ERR(new_root)) {
5683 err = PTR_ERR(new_root);
5687 *sub_root = new_root;
5688 location->objectid = btrfs_root_dirid(&new_root->root_item);
5689 location->type = BTRFS_INODE_ITEM_KEY;
5690 location->offset = 0;
5693 btrfs_free_path(path);
5697 static void inode_tree_add(struct inode *inode)
5699 struct btrfs_root *root = BTRFS_I(inode)->root;
5700 struct btrfs_inode *entry;
5702 struct rb_node *parent;
5703 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5704 u64 ino = btrfs_ino(BTRFS_I(inode));
5706 if (inode_unhashed(inode))
5709 spin_lock(&root->inode_lock);
5710 p = &root->inode_tree.rb_node;
5713 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5715 if (ino < btrfs_ino(entry))
5716 p = &parent->rb_left;
5717 else if (ino > btrfs_ino(entry))
5718 p = &parent->rb_right;
5720 WARN_ON(!(entry->vfs_inode.i_state &
5721 (I_WILL_FREE | I_FREEING)));
5722 rb_replace_node(parent, new, &root->inode_tree);
5723 RB_CLEAR_NODE(parent);
5724 spin_unlock(&root->inode_lock);
5728 rb_link_node(new, parent, p);
5729 rb_insert_color(new, &root->inode_tree);
5730 spin_unlock(&root->inode_lock);
5733 static void inode_tree_del(struct btrfs_inode *inode)
5735 struct btrfs_root *root = inode->root;
5738 spin_lock(&root->inode_lock);
5739 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5740 rb_erase(&inode->rb_node, &root->inode_tree);
5741 RB_CLEAR_NODE(&inode->rb_node);
5742 empty = RB_EMPTY_ROOT(&root->inode_tree);
5744 spin_unlock(&root->inode_lock);
5746 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5747 spin_lock(&root->inode_lock);
5748 empty = RB_EMPTY_ROOT(&root->inode_tree);
5749 spin_unlock(&root->inode_lock);
5751 btrfs_add_dead_root(root);
5756 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5758 struct btrfs_iget_args *args = p;
5760 inode->i_ino = args->ino;
5761 BTRFS_I(inode)->location.objectid = args->ino;
5762 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5763 BTRFS_I(inode)->location.offset = 0;
5764 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5765 BUG_ON(args->root && !BTRFS_I(inode)->root);
5769 static int btrfs_find_actor(struct inode *inode, void *opaque)
5771 struct btrfs_iget_args *args = opaque;
5773 return args->ino == BTRFS_I(inode)->location.objectid &&
5774 args->root == BTRFS_I(inode)->root;
5777 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5778 struct btrfs_root *root)
5780 struct inode *inode;
5781 struct btrfs_iget_args args;
5782 unsigned long hashval = btrfs_inode_hash(ino, root);
5787 inode = iget5_locked(s, hashval, btrfs_find_actor,
5788 btrfs_init_locked_inode,
5794 * Get an inode object given its inode number and corresponding root.
5795 * Path can be preallocated to prevent recursing back to iget through
5796 * allocator. NULL is also valid but may require an additional allocation
5799 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5800 struct btrfs_root *root, struct btrfs_path *path)
5802 struct inode *inode;
5804 inode = btrfs_iget_locked(s, ino, root);
5806 return ERR_PTR(-ENOMEM);
5808 if (inode->i_state & I_NEW) {
5811 ret = btrfs_read_locked_inode(inode, path);
5813 inode_tree_add(inode);
5814 unlock_new_inode(inode);
5818 * ret > 0 can come from btrfs_search_slot called by
5819 * btrfs_read_locked_inode, this means the inode item
5824 inode = ERR_PTR(ret);
5831 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5833 return btrfs_iget_path(s, ino, root, NULL);
5836 static struct inode *new_simple_dir(struct super_block *s,
5837 struct btrfs_key *key,
5838 struct btrfs_root *root)
5840 struct inode *inode = new_inode(s);
5843 return ERR_PTR(-ENOMEM);
5845 BTRFS_I(inode)->root = btrfs_grab_root(root);
5846 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5847 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5849 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5851 * We only need lookup, the rest is read-only and there's no inode
5852 * associated with the dentry
5854 inode->i_op = &simple_dir_inode_operations;
5855 inode->i_opflags &= ~IOP_XATTR;
5856 inode->i_fop = &simple_dir_operations;
5857 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5858 inode->i_mtime = current_time(inode);
5859 inode->i_atime = inode->i_mtime;
5860 inode->i_ctime = inode->i_mtime;
5861 BTRFS_I(inode)->i_otime = inode->i_mtime;
5866 static inline u8 btrfs_inode_type(struct inode *inode)
5869 * Compile-time asserts that generic FT_* types still match
5872 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5873 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5874 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5875 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5876 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5877 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5878 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5879 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5881 return fs_umode_to_ftype(inode->i_mode);
5884 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5886 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5887 struct inode *inode;
5888 struct btrfs_root *root = BTRFS_I(dir)->root;
5889 struct btrfs_root *sub_root = root;
5890 struct btrfs_key location;
5894 if (dentry->d_name.len > BTRFS_NAME_LEN)
5895 return ERR_PTR(-ENAMETOOLONG);
5897 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5899 return ERR_PTR(ret);
5901 if (location.type == BTRFS_INODE_ITEM_KEY) {
5902 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5906 /* Do extra check against inode mode with di_type */
5907 if (btrfs_inode_type(inode) != di_type) {
5909 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5910 inode->i_mode, btrfs_inode_type(inode),
5913 return ERR_PTR(-EUCLEAN);
5918 ret = fixup_tree_root_location(fs_info, dir, dentry,
5919 &location, &sub_root);
5922 inode = ERR_PTR(ret);
5924 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5926 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5928 if (root != sub_root)
5929 btrfs_put_root(sub_root);
5931 if (!IS_ERR(inode) && root != sub_root) {
5932 down_read(&fs_info->cleanup_work_sem);
5933 if (!sb_rdonly(inode->i_sb))
5934 ret = btrfs_orphan_cleanup(sub_root);
5935 up_read(&fs_info->cleanup_work_sem);
5938 inode = ERR_PTR(ret);
5945 static int btrfs_dentry_delete(const struct dentry *dentry)
5947 struct btrfs_root *root;
5948 struct inode *inode = d_inode(dentry);
5950 if (!inode && !IS_ROOT(dentry))
5951 inode = d_inode(dentry->d_parent);
5954 root = BTRFS_I(inode)->root;
5955 if (btrfs_root_refs(&root->root_item) == 0)
5958 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5964 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5967 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5969 if (inode == ERR_PTR(-ENOENT))
5971 return d_splice_alias(inode, dentry);
5975 * All this infrastructure exists because dir_emit can fault, and we are holding
5976 * the tree lock when doing readdir. For now just allocate a buffer and copy
5977 * our information into that, and then dir_emit from the buffer. This is
5978 * similar to what NFS does, only we don't keep the buffer around in pagecache
5979 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5980 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5983 static int btrfs_opendir(struct inode *inode, struct file *file)
5985 struct btrfs_file_private *private;
5987 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5990 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5991 if (!private->filldir_buf) {
5995 file->private_data = private;
6006 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
6009 struct dir_entry *entry = addr;
6010 char *name = (char *)(entry + 1);
6012 ctx->pos = get_unaligned(&entry->offset);
6013 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
6014 get_unaligned(&entry->ino),
6015 get_unaligned(&entry->type)))
6017 addr += sizeof(struct dir_entry) +
6018 get_unaligned(&entry->name_len);
6024 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
6026 struct inode *inode = file_inode(file);
6027 struct btrfs_root *root = BTRFS_I(inode)->root;
6028 struct btrfs_file_private *private = file->private_data;
6029 struct btrfs_dir_item *di;
6030 struct btrfs_key key;
6031 struct btrfs_key found_key;
6032 struct btrfs_path *path;
6034 struct list_head ins_list;
6035 struct list_head del_list;
6037 struct extent_buffer *leaf;
6044 struct btrfs_key location;
6046 if (!dir_emit_dots(file, ctx))
6049 path = btrfs_alloc_path();
6053 addr = private->filldir_buf;
6054 path->reada = READA_FORWARD;
6056 INIT_LIST_HEAD(&ins_list);
6057 INIT_LIST_HEAD(&del_list);
6058 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
6061 key.type = BTRFS_DIR_INDEX_KEY;
6062 key.offset = ctx->pos;
6063 key.objectid = btrfs_ino(BTRFS_I(inode));
6065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6070 struct dir_entry *entry;
6072 leaf = path->nodes[0];
6073 slot = path->slots[0];
6074 if (slot >= btrfs_header_nritems(leaf)) {
6075 ret = btrfs_next_leaf(root, path);
6083 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6085 if (found_key.objectid != key.objectid)
6087 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6089 if (found_key.offset < ctx->pos)
6091 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6093 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6094 name_len = btrfs_dir_name_len(leaf, di);
6095 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6097 btrfs_release_path(path);
6098 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6101 addr = private->filldir_buf;
6108 put_unaligned(name_len, &entry->name_len);
6109 name_ptr = (char *)(entry + 1);
6110 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6112 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
6114 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6115 put_unaligned(location.objectid, &entry->ino);
6116 put_unaligned(found_key.offset, &entry->offset);
6118 addr += sizeof(struct dir_entry) + name_len;
6119 total_len += sizeof(struct dir_entry) + name_len;
6123 btrfs_release_path(path);
6125 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6129 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6134 * Stop new entries from being returned after we return the last
6137 * New directory entries are assigned a strictly increasing
6138 * offset. This means that new entries created during readdir
6139 * are *guaranteed* to be seen in the future by that readdir.
6140 * This has broken buggy programs which operate on names as
6141 * they're returned by readdir. Until we re-use freed offsets
6142 * we have this hack to stop new entries from being returned
6143 * under the assumption that they'll never reach this huge
6146 * This is being careful not to overflow 32bit loff_t unless the
6147 * last entry requires it because doing so has broken 32bit apps
6150 if (ctx->pos >= INT_MAX)
6151 ctx->pos = LLONG_MAX;
6158 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6159 btrfs_free_path(path);
6164 * This is somewhat expensive, updating the tree every time the
6165 * inode changes. But, it is most likely to find the inode in cache.
6166 * FIXME, needs more benchmarking...there are no reasons other than performance
6167 * to keep or drop this code.
6169 static int btrfs_dirty_inode(struct inode *inode)
6171 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6172 struct btrfs_root *root = BTRFS_I(inode)->root;
6173 struct btrfs_trans_handle *trans;
6176 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6179 trans = btrfs_join_transaction(root);
6181 return PTR_ERR(trans);
6183 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6184 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6185 /* whoops, lets try again with the full transaction */
6186 btrfs_end_transaction(trans);
6187 trans = btrfs_start_transaction(root, 1);
6189 return PTR_ERR(trans);
6191 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
6193 btrfs_end_transaction(trans);
6194 if (BTRFS_I(inode)->delayed_node)
6195 btrfs_balance_delayed_items(fs_info);
6201 * This is a copy of file_update_time. We need this so we can return error on
6202 * ENOSPC for updating the inode in the case of file write and mmap writes.
6204 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6207 struct btrfs_root *root = BTRFS_I(inode)->root;
6208 bool dirty = flags & ~S_VERSION;
6210 if (btrfs_root_readonly(root))
6213 if (flags & S_VERSION)
6214 dirty |= inode_maybe_inc_iversion(inode, dirty);
6215 if (flags & S_CTIME)
6216 inode->i_ctime = *now;
6217 if (flags & S_MTIME)
6218 inode->i_mtime = *now;
6219 if (flags & S_ATIME)
6220 inode->i_atime = *now;
6221 return dirty ? btrfs_dirty_inode(inode) : 0;
6225 * find the highest existing sequence number in a directory
6226 * and then set the in-memory index_cnt variable to reflect
6227 * free sequence numbers
6229 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6231 struct btrfs_root *root = inode->root;
6232 struct btrfs_key key, found_key;
6233 struct btrfs_path *path;
6234 struct extent_buffer *leaf;
6237 key.objectid = btrfs_ino(inode);
6238 key.type = BTRFS_DIR_INDEX_KEY;
6239 key.offset = (u64)-1;
6241 path = btrfs_alloc_path();
6245 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6248 /* FIXME: we should be able to handle this */
6254 * MAGIC NUMBER EXPLANATION:
6255 * since we search a directory based on f_pos we have to start at 2
6256 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6257 * else has to start at 2
6259 if (path->slots[0] == 0) {
6260 inode->index_cnt = 2;
6266 leaf = path->nodes[0];
6267 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6269 if (found_key.objectid != btrfs_ino(inode) ||
6270 found_key.type != BTRFS_DIR_INDEX_KEY) {
6271 inode->index_cnt = 2;
6275 inode->index_cnt = found_key.offset + 1;
6277 btrfs_free_path(path);
6282 * helper to find a free sequence number in a given directory. This current
6283 * code is very simple, later versions will do smarter things in the btree
6285 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6289 if (dir->index_cnt == (u64)-1) {
6290 ret = btrfs_inode_delayed_dir_index_count(dir);
6292 ret = btrfs_set_inode_index_count(dir);
6298 *index = dir->index_cnt;
6304 static int btrfs_insert_inode_locked(struct inode *inode)
6306 struct btrfs_iget_args args;
6308 args.ino = BTRFS_I(inode)->location.objectid;
6309 args.root = BTRFS_I(inode)->root;
6311 return insert_inode_locked4(inode,
6312 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6313 btrfs_find_actor, &args);
6317 * Inherit flags from the parent inode.
6319 * Currently only the compression flags and the cow flags are inherited.
6321 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6328 flags = BTRFS_I(dir)->flags;
6330 if (flags & BTRFS_INODE_NOCOMPRESS) {
6331 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6332 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6333 } else if (flags & BTRFS_INODE_COMPRESS) {
6334 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6335 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6338 if (flags & BTRFS_INODE_NODATACOW) {
6339 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6340 if (S_ISREG(inode->i_mode))
6341 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6344 btrfs_sync_inode_flags_to_i_flags(inode);
6347 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6348 struct btrfs_root *root,
6350 const char *name, int name_len,
6351 u64 ref_objectid, u64 objectid,
6352 umode_t mode, u64 *index)
6354 struct btrfs_fs_info *fs_info = root->fs_info;
6355 struct inode *inode;
6356 struct btrfs_inode_item *inode_item;
6357 struct btrfs_key *location;
6358 struct btrfs_path *path;
6359 struct btrfs_inode_ref *ref;
6360 struct btrfs_key key[2];
6362 int nitems = name ? 2 : 1;
6364 unsigned int nofs_flag;
6367 path = btrfs_alloc_path();
6369 return ERR_PTR(-ENOMEM);
6371 nofs_flag = memalloc_nofs_save();
6372 inode = new_inode(fs_info->sb);
6373 memalloc_nofs_restore(nofs_flag);
6375 btrfs_free_path(path);
6376 return ERR_PTR(-ENOMEM);
6380 * O_TMPFILE, set link count to 0, so that after this point,
6381 * we fill in an inode item with the correct link count.
6384 set_nlink(inode, 0);
6387 * we have to initialize this early, so we can reclaim the inode
6388 * number if we fail afterwards in this function.
6390 inode->i_ino = objectid;
6393 trace_btrfs_inode_request(dir);
6395 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6397 btrfs_free_path(path);
6399 return ERR_PTR(ret);
6405 * index_cnt is ignored for everything but a dir,
6406 * btrfs_set_inode_index_count has an explanation for the magic
6409 BTRFS_I(inode)->index_cnt = 2;
6410 BTRFS_I(inode)->dir_index = *index;
6411 BTRFS_I(inode)->root = btrfs_grab_root(root);
6412 BTRFS_I(inode)->generation = trans->transid;
6413 inode->i_generation = BTRFS_I(inode)->generation;
6416 * We could have gotten an inode number from somebody who was fsynced
6417 * and then removed in this same transaction, so let's just set full
6418 * sync since it will be a full sync anyway and this will blow away the
6419 * old info in the log.
6421 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6423 key[0].objectid = objectid;
6424 key[0].type = BTRFS_INODE_ITEM_KEY;
6427 sizes[0] = sizeof(struct btrfs_inode_item);
6431 * Start new inodes with an inode_ref. This is slightly more
6432 * efficient for small numbers of hard links since they will
6433 * be packed into one item. Extended refs will kick in if we
6434 * add more hard links than can fit in the ref item.
6436 key[1].objectid = objectid;
6437 key[1].type = BTRFS_INODE_REF_KEY;
6438 key[1].offset = ref_objectid;
6440 sizes[1] = name_len + sizeof(*ref);
6443 location = &BTRFS_I(inode)->location;
6444 location->objectid = objectid;
6445 location->offset = 0;
6446 location->type = BTRFS_INODE_ITEM_KEY;
6448 ret = btrfs_insert_inode_locked(inode);
6454 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6458 inode_init_owner(&init_user_ns, inode, dir, mode);
6459 inode_set_bytes(inode, 0);
6461 inode->i_mtime = current_time(inode);
6462 inode->i_atime = inode->i_mtime;
6463 inode->i_ctime = inode->i_mtime;
6464 BTRFS_I(inode)->i_otime = inode->i_mtime;
6466 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6467 struct btrfs_inode_item);
6468 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6469 sizeof(*inode_item));
6470 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6473 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6474 struct btrfs_inode_ref);
6475 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6476 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6477 ptr = (unsigned long)(ref + 1);
6478 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6481 btrfs_mark_buffer_dirty(path->nodes[0]);
6482 btrfs_free_path(path);
6484 btrfs_inherit_iflags(inode, dir);
6486 if (S_ISREG(mode)) {
6487 if (btrfs_test_opt(fs_info, NODATASUM))
6488 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6489 if (btrfs_test_opt(fs_info, NODATACOW))
6490 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6491 BTRFS_INODE_NODATASUM;
6494 inode_tree_add(inode);
6496 trace_btrfs_inode_new(inode);
6497 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6499 btrfs_update_root_times(trans, root);
6501 ret = btrfs_inode_inherit_props(trans, inode, dir);
6504 "error inheriting props for ino %llu (root %llu): %d",
6505 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6510 discard_new_inode(inode);
6513 BTRFS_I(dir)->index_cnt--;
6514 btrfs_free_path(path);
6515 return ERR_PTR(ret);
6519 * utility function to add 'inode' into 'parent_inode' with
6520 * a give name and a given sequence number.
6521 * if 'add_backref' is true, also insert a backref from the
6522 * inode to the parent directory.
6524 int btrfs_add_link(struct btrfs_trans_handle *trans,
6525 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6526 const char *name, int name_len, int add_backref, u64 index)
6529 struct btrfs_key key;
6530 struct btrfs_root *root = parent_inode->root;
6531 u64 ino = btrfs_ino(inode);
6532 u64 parent_ino = btrfs_ino(parent_inode);
6534 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6535 memcpy(&key, &inode->root->root_key, sizeof(key));
6538 key.type = BTRFS_INODE_ITEM_KEY;
6542 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6543 ret = btrfs_add_root_ref(trans, key.objectid,
6544 root->root_key.objectid, parent_ino,
6545 index, name, name_len);
6546 } else if (add_backref) {
6547 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6551 /* Nothing to clean up yet */
6555 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6556 btrfs_inode_type(&inode->vfs_inode), index);
6557 if (ret == -EEXIST || ret == -EOVERFLOW)
6560 btrfs_abort_transaction(trans, ret);
6564 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6566 inode_inc_iversion(&parent_inode->vfs_inode);
6568 * If we are replaying a log tree, we do not want to update the mtime
6569 * and ctime of the parent directory with the current time, since the
6570 * log replay procedure is responsible for setting them to their correct
6571 * values (the ones it had when the fsync was done).
6573 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6574 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6576 parent_inode->vfs_inode.i_mtime = now;
6577 parent_inode->vfs_inode.i_ctime = now;
6579 ret = btrfs_update_inode(trans, root, parent_inode);
6581 btrfs_abort_transaction(trans, ret);
6585 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6588 err = btrfs_del_root_ref(trans, key.objectid,
6589 root->root_key.objectid, parent_ino,
6590 &local_index, name, name_len);
6592 btrfs_abort_transaction(trans, err);
6593 } else if (add_backref) {
6597 err = btrfs_del_inode_ref(trans, root, name, name_len,
6598 ino, parent_ino, &local_index);
6600 btrfs_abort_transaction(trans, err);
6603 /* Return the original error code */
6607 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6608 struct btrfs_inode *dir, struct dentry *dentry,
6609 struct btrfs_inode *inode, int backref, u64 index)
6611 int err = btrfs_add_link(trans, dir, inode,
6612 dentry->d_name.name, dentry->d_name.len,
6619 static int btrfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
6620 struct dentry *dentry, umode_t mode, dev_t rdev)
6622 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6623 struct btrfs_trans_handle *trans;
6624 struct btrfs_root *root = BTRFS_I(dir)->root;
6625 struct inode *inode = NULL;
6631 * 2 for inode item and ref
6633 * 1 for xattr if selinux is on
6635 trans = btrfs_start_transaction(root, 5);
6637 return PTR_ERR(trans);
6639 err = btrfs_get_free_objectid(root, &objectid);
6643 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6644 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6646 if (IS_ERR(inode)) {
6647 err = PTR_ERR(inode);
6653 * If the active LSM wants to access the inode during
6654 * d_instantiate it needs these. Smack checks to see
6655 * if the filesystem supports xattrs by looking at the
6658 inode->i_op = &btrfs_special_inode_operations;
6659 init_special_inode(inode, inode->i_mode, rdev);
6661 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6665 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6670 btrfs_update_inode(trans, root, BTRFS_I(inode));
6671 d_instantiate_new(dentry, inode);
6674 btrfs_end_transaction(trans);
6675 btrfs_btree_balance_dirty(fs_info);
6677 inode_dec_link_count(inode);
6678 discard_new_inode(inode);
6683 static int btrfs_create(struct user_namespace *mnt_userns, struct inode *dir,
6684 struct dentry *dentry, umode_t mode, bool excl)
6686 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6687 struct btrfs_trans_handle *trans;
6688 struct btrfs_root *root = BTRFS_I(dir)->root;
6689 struct inode *inode = NULL;
6695 * 2 for inode item and ref
6697 * 1 for xattr if selinux is on
6699 trans = btrfs_start_transaction(root, 5);
6701 return PTR_ERR(trans);
6703 err = btrfs_get_free_objectid(root, &objectid);
6707 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6708 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6710 if (IS_ERR(inode)) {
6711 err = PTR_ERR(inode);
6716 * If the active LSM wants to access the inode during
6717 * d_instantiate it needs these. Smack checks to see
6718 * if the filesystem supports xattrs by looking at the
6721 inode->i_fop = &btrfs_file_operations;
6722 inode->i_op = &btrfs_file_inode_operations;
6723 inode->i_mapping->a_ops = &btrfs_aops;
6725 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6729 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6733 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6738 d_instantiate_new(dentry, inode);
6741 btrfs_end_transaction(trans);
6743 inode_dec_link_count(inode);
6744 discard_new_inode(inode);
6746 btrfs_btree_balance_dirty(fs_info);
6750 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6751 struct dentry *dentry)
6753 struct btrfs_trans_handle *trans = NULL;
6754 struct btrfs_root *root = BTRFS_I(dir)->root;
6755 struct inode *inode = d_inode(old_dentry);
6756 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6761 /* do not allow sys_link's with other subvols of the same device */
6762 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6765 if (inode->i_nlink >= BTRFS_LINK_MAX)
6768 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6773 * 2 items for inode and inode ref
6774 * 2 items for dir items
6775 * 1 item for parent inode
6776 * 1 item for orphan item deletion if O_TMPFILE
6778 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6779 if (IS_ERR(trans)) {
6780 err = PTR_ERR(trans);
6785 /* There are several dir indexes for this inode, clear the cache. */
6786 BTRFS_I(inode)->dir_index = 0ULL;
6788 inode_inc_iversion(inode);
6789 inode->i_ctime = current_time(inode);
6791 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6793 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6799 struct dentry *parent = dentry->d_parent;
6801 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6804 if (inode->i_nlink == 1) {
6806 * If new hard link count is 1, it's a file created
6807 * with open(2) O_TMPFILE flag.
6809 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6813 d_instantiate(dentry, inode);
6814 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6819 btrfs_end_transaction(trans);
6821 inode_dec_link_count(inode);
6824 btrfs_btree_balance_dirty(fs_info);
6828 static int btrfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
6829 struct dentry *dentry, umode_t mode)
6831 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6832 struct inode *inode = NULL;
6833 struct btrfs_trans_handle *trans;
6834 struct btrfs_root *root = BTRFS_I(dir)->root;
6840 * 2 items for inode and ref
6841 * 2 items for dir items
6842 * 1 for xattr if selinux is on
6844 trans = btrfs_start_transaction(root, 5);
6846 return PTR_ERR(trans);
6848 err = btrfs_get_free_objectid(root, &objectid);
6852 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6853 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6854 S_IFDIR | mode, &index);
6855 if (IS_ERR(inode)) {
6856 err = PTR_ERR(inode);
6861 /* these must be set before we unlock the inode */
6862 inode->i_op = &btrfs_dir_inode_operations;
6863 inode->i_fop = &btrfs_dir_file_operations;
6865 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6869 btrfs_i_size_write(BTRFS_I(inode), 0);
6870 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6874 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6875 dentry->d_name.name,
6876 dentry->d_name.len, 0, index);
6880 d_instantiate_new(dentry, inode);
6883 btrfs_end_transaction(trans);
6885 inode_dec_link_count(inode);
6886 discard_new_inode(inode);
6888 btrfs_btree_balance_dirty(fs_info);
6892 static noinline int uncompress_inline(struct btrfs_path *path,
6894 size_t pg_offset, u64 extent_offset,
6895 struct btrfs_file_extent_item *item)
6898 struct extent_buffer *leaf = path->nodes[0];
6901 unsigned long inline_size;
6905 WARN_ON(pg_offset != 0);
6906 compress_type = btrfs_file_extent_compression(leaf, item);
6907 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6908 inline_size = btrfs_file_extent_inline_item_len(leaf,
6909 btrfs_item_nr(path->slots[0]));
6910 tmp = kmalloc(inline_size, GFP_NOFS);
6913 ptr = btrfs_file_extent_inline_start(item);
6915 read_extent_buffer(leaf, tmp, ptr, inline_size);
6917 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6918 ret = btrfs_decompress(compress_type, tmp, page,
6919 extent_offset, inline_size, max_size);
6922 * decompression code contains a memset to fill in any space between the end
6923 * of the uncompressed data and the end of max_size in case the decompressed
6924 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6925 * the end of an inline extent and the beginning of the next block, so we
6926 * cover that region here.
6929 if (max_size + pg_offset < PAGE_SIZE)
6930 memzero_page(page, pg_offset + max_size,
6931 PAGE_SIZE - max_size - pg_offset);
6937 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6938 * @inode: file to search in
6939 * @page: page to read extent data into if the extent is inline
6940 * @pg_offset: offset into @page to copy to
6941 * @start: file offset
6942 * @len: length of range starting at @start
6944 * This returns the first &struct extent_map which overlaps with the given
6945 * range, reading it from the B-tree and caching it if necessary. Note that
6946 * there may be more extents which overlap the given range after the returned
6949 * If @page is not NULL and the extent is inline, this also reads the extent
6950 * data directly into the page and marks the extent up to date in the io_tree.
6952 * Return: ERR_PTR on error, non-NULL extent_map on success.
6954 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6955 struct page *page, size_t pg_offset,
6958 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6960 u64 extent_start = 0;
6962 u64 objectid = btrfs_ino(inode);
6963 int extent_type = -1;
6964 struct btrfs_path *path = NULL;
6965 struct btrfs_root *root = inode->root;
6966 struct btrfs_file_extent_item *item;
6967 struct extent_buffer *leaf;
6968 struct btrfs_key found_key;
6969 struct extent_map *em = NULL;
6970 struct extent_map_tree *em_tree = &inode->extent_tree;
6971 struct extent_io_tree *io_tree = &inode->io_tree;
6973 read_lock(&em_tree->lock);
6974 em = lookup_extent_mapping(em_tree, start, len);
6975 read_unlock(&em_tree->lock);
6978 if (em->start > start || em->start + em->len <= start)
6979 free_extent_map(em);
6980 else if (em->block_start == EXTENT_MAP_INLINE && page)
6981 free_extent_map(em);
6985 em = alloc_extent_map();
6990 em->start = EXTENT_MAP_HOLE;
6991 em->orig_start = EXTENT_MAP_HOLE;
6993 em->block_len = (u64)-1;
6995 path = btrfs_alloc_path();
7001 /* Chances are we'll be called again, so go ahead and do readahead */
7002 path->reada = READA_FORWARD;
7005 * The same explanation in load_free_space_cache applies here as well,
7006 * we only read when we're loading the free space cache, and at that
7007 * point the commit_root has everything we need.
7009 if (btrfs_is_free_space_inode(inode)) {
7010 path->search_commit_root = 1;
7011 path->skip_locking = 1;
7014 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
7017 } else if (ret > 0) {
7018 if (path->slots[0] == 0)
7024 leaf = path->nodes[0];
7025 item = btrfs_item_ptr(leaf, path->slots[0],
7026 struct btrfs_file_extent_item);
7027 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7028 if (found_key.objectid != objectid ||
7029 found_key.type != BTRFS_EXTENT_DATA_KEY) {
7031 * If we backup past the first extent we want to move forward
7032 * and see if there is an extent in front of us, otherwise we'll
7033 * say there is a hole for our whole search range which can
7040 extent_type = btrfs_file_extent_type(leaf, item);
7041 extent_start = found_key.offset;
7042 extent_end = btrfs_file_extent_end(path);
7043 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7044 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7045 /* Only regular file could have regular/prealloc extent */
7046 if (!S_ISREG(inode->vfs_inode.i_mode)) {
7049 "regular/prealloc extent found for non-regular inode %llu",
7053 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7055 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7056 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7061 if (start >= extent_end) {
7063 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7064 ret = btrfs_next_leaf(root, path);
7070 leaf = path->nodes[0];
7072 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7073 if (found_key.objectid != objectid ||
7074 found_key.type != BTRFS_EXTENT_DATA_KEY)
7076 if (start + len <= found_key.offset)
7078 if (start > found_key.offset)
7081 /* New extent overlaps with existing one */
7083 em->orig_start = start;
7084 em->len = found_key.offset - start;
7085 em->block_start = EXTENT_MAP_HOLE;
7089 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
7091 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7092 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7094 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7098 size_t extent_offset;
7104 size = btrfs_file_extent_ram_bytes(leaf, item);
7105 extent_offset = page_offset(page) + pg_offset - extent_start;
7106 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7107 size - extent_offset);
7108 em->start = extent_start + extent_offset;
7109 em->len = ALIGN(copy_size, fs_info->sectorsize);
7110 em->orig_block_len = em->len;
7111 em->orig_start = em->start;
7112 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7114 if (!PageUptodate(page)) {
7115 if (btrfs_file_extent_compression(leaf, item) !=
7116 BTRFS_COMPRESS_NONE) {
7117 ret = uncompress_inline(path, page, pg_offset,
7118 extent_offset, item);
7122 map = kmap_local_page(page);
7123 read_extent_buffer(leaf, map + pg_offset, ptr,
7125 if (pg_offset + copy_size < PAGE_SIZE) {
7126 memset(map + pg_offset + copy_size, 0,
7127 PAGE_SIZE - pg_offset -
7132 flush_dcache_page(page);
7134 set_extent_uptodate(io_tree, em->start,
7135 extent_map_end(em) - 1, NULL, GFP_NOFS);
7140 em->orig_start = start;
7142 em->block_start = EXTENT_MAP_HOLE;
7145 btrfs_release_path(path);
7146 if (em->start > start || extent_map_end(em) <= start) {
7148 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7149 em->start, em->len, start, len);
7154 write_lock(&em_tree->lock);
7155 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7156 write_unlock(&em_tree->lock);
7158 btrfs_free_path(path);
7160 trace_btrfs_get_extent(root, inode, em);
7163 free_extent_map(em);
7164 return ERR_PTR(ret);
7169 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7172 struct extent_map *em;
7173 struct extent_map *hole_em = NULL;
7174 u64 delalloc_start = start;
7180 em = btrfs_get_extent(inode, NULL, 0, start, len);
7184 * If our em maps to:
7186 * - a pre-alloc extent,
7187 * there might actually be delalloc bytes behind it.
7189 if (em->block_start != EXTENT_MAP_HOLE &&
7190 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7195 /* check to see if we've wrapped (len == -1 or similar) */
7204 /* ok, we didn't find anything, lets look for delalloc */
7205 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
7206 end, len, EXTENT_DELALLOC, 1);
7207 delalloc_end = delalloc_start + delalloc_len;
7208 if (delalloc_end < delalloc_start)
7209 delalloc_end = (u64)-1;
7212 * We didn't find anything useful, return the original results from
7215 if (delalloc_start > end || delalloc_end <= start) {
7222 * Adjust the delalloc_start to make sure it doesn't go backwards from
7223 * the start they passed in
7225 delalloc_start = max(start, delalloc_start);
7226 delalloc_len = delalloc_end - delalloc_start;
7228 if (delalloc_len > 0) {
7231 const u64 hole_end = extent_map_end(hole_em);
7233 em = alloc_extent_map();
7241 * When btrfs_get_extent can't find anything it returns one
7244 * Make sure what it found really fits our range, and adjust to
7245 * make sure it is based on the start from the caller
7247 if (hole_end <= start || hole_em->start > end) {
7248 free_extent_map(hole_em);
7251 hole_start = max(hole_em->start, start);
7252 hole_len = hole_end - hole_start;
7255 if (hole_em && delalloc_start > hole_start) {
7257 * Our hole starts before our delalloc, so we have to
7258 * return just the parts of the hole that go until the
7261 em->len = min(hole_len, delalloc_start - hole_start);
7262 em->start = hole_start;
7263 em->orig_start = hole_start;
7265 * Don't adjust block start at all, it is fixed at
7268 em->block_start = hole_em->block_start;
7269 em->block_len = hole_len;
7270 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7271 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7274 * Hole is out of passed range or it starts after
7277 em->start = delalloc_start;
7278 em->len = delalloc_len;
7279 em->orig_start = delalloc_start;
7280 em->block_start = EXTENT_MAP_DELALLOC;
7281 em->block_len = delalloc_len;
7288 free_extent_map(hole_em);
7290 free_extent_map(em);
7291 return ERR_PTR(err);
7296 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7299 const u64 orig_start,
7300 const u64 block_start,
7301 const u64 block_len,
7302 const u64 orig_block_len,
7303 const u64 ram_bytes,
7306 struct extent_map *em = NULL;
7309 if (type != BTRFS_ORDERED_NOCOW) {
7310 em = create_io_em(inode, start, len, orig_start, block_start,
7311 block_len, orig_block_len, ram_bytes,
7312 BTRFS_COMPRESS_NONE, /* compress_type */
7317 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7321 free_extent_map(em);
7322 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7331 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7334 struct btrfs_root *root = inode->root;
7335 struct btrfs_fs_info *fs_info = root->fs_info;
7336 struct extent_map *em;
7337 struct btrfs_key ins;
7341 alloc_hint = get_extent_allocation_hint(inode, start, len);
7342 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7343 0, alloc_hint, &ins, 1, 1);
7345 return ERR_PTR(ret);
7347 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7348 ins.objectid, ins.offset, ins.offset,
7349 ins.offset, BTRFS_ORDERED_REGULAR);
7350 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7352 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7358 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7360 struct btrfs_block_group *block_group;
7361 bool readonly = false;
7363 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7364 if (!block_group || block_group->ro)
7367 btrfs_put_block_group(block_group);
7372 * Check if we can do nocow write into the range [@offset, @offset + @len)
7374 * @offset: File offset
7375 * @len: The length to write, will be updated to the nocow writeable
7377 * @orig_start: (optional) Return the original file offset of the file extent
7378 * @orig_len: (optional) Return the original on-disk length of the file extent
7379 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7380 * @strict: if true, omit optimizations that might force us into unnecessary
7381 * cow. e.g., don't trust generation number.
7384 * >0 and update @len if we can do nocow write
7385 * 0 if we can't do nocow write
7386 * <0 if error happened
7388 * NOTE: This only checks the file extents, caller is responsible to wait for
7389 * any ordered extents.
7391 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7392 u64 *orig_start, u64 *orig_block_len,
7393 u64 *ram_bytes, bool strict)
7395 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7396 struct btrfs_path *path;
7398 struct extent_buffer *leaf;
7399 struct btrfs_root *root = BTRFS_I(inode)->root;
7400 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7401 struct btrfs_file_extent_item *fi;
7402 struct btrfs_key key;
7409 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7411 path = btrfs_alloc_path();
7415 ret = btrfs_lookup_file_extent(NULL, root, path,
7416 btrfs_ino(BTRFS_I(inode)), offset, 0);
7420 slot = path->slots[0];
7423 /* can't find the item, must cow */
7430 leaf = path->nodes[0];
7431 btrfs_item_key_to_cpu(leaf, &key, slot);
7432 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7433 key.type != BTRFS_EXTENT_DATA_KEY) {
7434 /* not our file or wrong item type, must cow */
7438 if (key.offset > offset) {
7439 /* Wrong offset, must cow */
7443 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7444 found_type = btrfs_file_extent_type(leaf, fi);
7445 if (found_type != BTRFS_FILE_EXTENT_REG &&
7446 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7447 /* not a regular extent, must cow */
7451 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7454 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7455 if (extent_end <= offset)
7458 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7459 if (disk_bytenr == 0)
7462 if (btrfs_file_extent_compression(leaf, fi) ||
7463 btrfs_file_extent_encryption(leaf, fi) ||
7464 btrfs_file_extent_other_encoding(leaf, fi))
7468 * Do the same check as in btrfs_cross_ref_exist but without the
7469 * unnecessary search.
7472 (btrfs_file_extent_generation(leaf, fi) <=
7473 btrfs_root_last_snapshot(&root->root_item)))
7476 backref_offset = btrfs_file_extent_offset(leaf, fi);
7479 *orig_start = key.offset - backref_offset;
7480 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7481 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7484 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7487 num_bytes = min(offset + *len, extent_end) - offset;
7488 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7491 range_end = round_up(offset + num_bytes,
7492 root->fs_info->sectorsize) - 1;
7493 ret = test_range_bit(io_tree, offset, range_end,
7494 EXTENT_DELALLOC, 0, NULL);
7501 btrfs_release_path(path);
7504 * look for other files referencing this extent, if we
7505 * find any we must cow
7508 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7509 key.offset - backref_offset, disk_bytenr,
7517 * adjust disk_bytenr and num_bytes to cover just the bytes
7518 * in this extent we are about to write. If there
7519 * are any csums in that range we have to cow in order
7520 * to keep the csums correct
7522 disk_bytenr += backref_offset;
7523 disk_bytenr += offset - key.offset;
7524 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7527 * all of the above have passed, it is safe to overwrite this extent
7533 btrfs_free_path(path);
7537 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7538 struct extent_state **cached_state, bool writing)
7540 struct btrfs_ordered_extent *ordered;
7544 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7547 * We're concerned with the entire range that we're going to be
7548 * doing DIO to, so we need to make sure there's no ordered
7549 * extents in this range.
7551 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7552 lockend - lockstart + 1);
7555 * We need to make sure there are no buffered pages in this
7556 * range either, we could have raced between the invalidate in
7557 * generic_file_direct_write and locking the extent. The
7558 * invalidate needs to happen so that reads after a write do not
7562 (!writing || !filemap_range_has_page(inode->i_mapping,
7563 lockstart, lockend)))
7566 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7571 * If we are doing a DIO read and the ordered extent we
7572 * found is for a buffered write, we can not wait for it
7573 * to complete and retry, because if we do so we can
7574 * deadlock with concurrent buffered writes on page
7575 * locks. This happens only if our DIO read covers more
7576 * than one extent map, if at this point has already
7577 * created an ordered extent for a previous extent map
7578 * and locked its range in the inode's io tree, and a
7579 * concurrent write against that previous extent map's
7580 * range and this range started (we unlock the ranges
7581 * in the io tree only when the bios complete and
7582 * buffered writes always lock pages before attempting
7583 * to lock range in the io tree).
7586 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7587 btrfs_start_ordered_extent(ordered, 1);
7590 btrfs_put_ordered_extent(ordered);
7593 * We could trigger writeback for this range (and wait
7594 * for it to complete) and then invalidate the pages for
7595 * this range (through invalidate_inode_pages2_range()),
7596 * but that can lead us to a deadlock with a concurrent
7597 * call to readahead (a buffered read or a defrag call
7598 * triggered a readahead) on a page lock due to an
7599 * ordered dio extent we created before but did not have
7600 * yet a corresponding bio submitted (whence it can not
7601 * complete), which makes readahead wait for that
7602 * ordered extent to complete while holding a lock on
7617 /* The callers of this must take lock_extent() */
7618 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7619 u64 len, u64 orig_start, u64 block_start,
7620 u64 block_len, u64 orig_block_len,
7621 u64 ram_bytes, int compress_type,
7624 struct extent_map_tree *em_tree;
7625 struct extent_map *em;
7628 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7629 type == BTRFS_ORDERED_COMPRESSED ||
7630 type == BTRFS_ORDERED_NOCOW ||
7631 type == BTRFS_ORDERED_REGULAR);
7633 em_tree = &inode->extent_tree;
7634 em = alloc_extent_map();
7636 return ERR_PTR(-ENOMEM);
7639 em->orig_start = orig_start;
7641 em->block_len = block_len;
7642 em->block_start = block_start;
7643 em->orig_block_len = orig_block_len;
7644 em->ram_bytes = ram_bytes;
7645 em->generation = -1;
7646 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7647 if (type == BTRFS_ORDERED_PREALLOC) {
7648 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7649 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7650 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7651 em->compress_type = compress_type;
7655 btrfs_drop_extent_cache(inode, em->start,
7656 em->start + em->len - 1, 0);
7657 write_lock(&em_tree->lock);
7658 ret = add_extent_mapping(em_tree, em, 1);
7659 write_unlock(&em_tree->lock);
7661 * The caller has taken lock_extent(), who could race with us
7664 } while (ret == -EEXIST);
7667 free_extent_map(em);
7668 return ERR_PTR(ret);
7671 /* em got 2 refs now, callers needs to do free_extent_map once. */
7676 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7677 struct inode *inode,
7678 struct btrfs_dio_data *dio_data,
7681 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7682 struct extent_map *em = *map;
7686 * We don't allocate a new extent in the following cases
7688 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7690 * 2) The extent is marked as PREALLOC. We're good to go here and can
7691 * just use the extent.
7694 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7695 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7696 em->block_start != EXTENT_MAP_HOLE)) {
7698 u64 block_start, orig_start, orig_block_len, ram_bytes;
7700 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7701 type = BTRFS_ORDERED_PREALLOC;
7703 type = BTRFS_ORDERED_NOCOW;
7704 len = min(len, em->len - (start - em->start));
7705 block_start = em->block_start + (start - em->start);
7707 if (can_nocow_extent(inode, start, &len, &orig_start,
7708 &orig_block_len, &ram_bytes, false) == 1 &&
7709 btrfs_inc_nocow_writers(fs_info, block_start)) {
7710 struct extent_map *em2;
7712 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7713 orig_start, block_start,
7714 len, orig_block_len,
7716 btrfs_dec_nocow_writers(fs_info, block_start);
7717 if (type == BTRFS_ORDERED_PREALLOC) {
7718 free_extent_map(em);
7722 if (em2 && IS_ERR(em2)) {
7727 * For inode marked NODATACOW or extent marked PREALLOC,
7728 * use the existing or preallocated extent, so does not
7729 * need to adjust btrfs_space_info's bytes_may_use.
7731 btrfs_free_reserved_data_space_noquota(fs_info, len);
7736 /* this will cow the extent */
7737 free_extent_map(em);
7738 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7744 len = min(len, em->len - (start - em->start));
7748 * Need to update the i_size under the extent lock so buffered
7749 * readers will get the updated i_size when we unlock.
7751 if (start + len > i_size_read(inode))
7752 i_size_write(inode, start + len);
7754 dio_data->reserve -= len;
7759 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7760 loff_t length, unsigned int flags, struct iomap *iomap,
7761 struct iomap *srcmap)
7763 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7764 struct extent_map *em;
7765 struct extent_state *cached_state = NULL;
7766 struct btrfs_dio_data *dio_data = NULL;
7767 u64 lockstart, lockend;
7768 const bool write = !!(flags & IOMAP_WRITE);
7771 bool unlock_extents = false;
7774 len = min_t(u64, len, fs_info->sectorsize);
7777 lockend = start + len - 1;
7780 * The generic stuff only does filemap_write_and_wait_range, which
7781 * isn't enough if we've written compressed pages to this area, so we
7782 * need to flush the dirty pages again to make absolutely sure that any
7783 * outstanding dirty pages are on disk.
7785 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7786 &BTRFS_I(inode)->runtime_flags)) {
7787 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7788 start + length - 1);
7793 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7797 dio_data->length = length;
7799 dio_data->reserve = round_up(length, fs_info->sectorsize);
7800 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7801 &dio_data->data_reserved,
7802 start, dio_data->reserve);
7804 extent_changeset_free(dio_data->data_reserved);
7809 iomap->private = dio_data;
7813 * If this errors out it's because we couldn't invalidate pagecache for
7814 * this range and we need to fallback to buffered.
7816 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7821 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7828 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7829 * io. INLINE is special, and we could probably kludge it in here, but
7830 * it's still buffered so for safety lets just fall back to the generic
7833 * For COMPRESSED we _have_ to read the entire extent in so we can
7834 * decompress it, so there will be buffering required no matter what we
7835 * do, so go ahead and fallback to buffered.
7837 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7838 * to buffered IO. Don't blame me, this is the price we pay for using
7841 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7842 em->block_start == EXTENT_MAP_INLINE) {
7843 free_extent_map(em);
7848 len = min(len, em->len - (start - em->start));
7850 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7854 unlock_extents = true;
7855 /* Recalc len in case the new em is smaller than requested */
7856 len = min(len, em->len - (start - em->start));
7859 * We need to unlock only the end area that we aren't using.
7860 * The rest is going to be unlocked by the endio routine.
7862 lockstart = start + len;
7863 if (lockstart < lockend)
7864 unlock_extents = true;
7868 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7869 lockstart, lockend, &cached_state);
7871 free_extent_state(cached_state);
7874 * Translate extent map information to iomap.
7875 * We trim the extents (and move the addr) even though iomap code does
7876 * that, since we have locked only the parts we are performing I/O in.
7878 if ((em->block_start == EXTENT_MAP_HOLE) ||
7879 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7880 iomap->addr = IOMAP_NULL_ADDR;
7881 iomap->type = IOMAP_HOLE;
7883 iomap->addr = em->block_start + (start - em->start);
7884 iomap->type = IOMAP_MAPPED;
7886 iomap->offset = start;
7887 iomap->bdev = fs_info->fs_devices->latest_bdev;
7888 iomap->length = len;
7890 if (write && btrfs_use_zone_append(BTRFS_I(inode), em->block_start))
7891 iomap->flags |= IOMAP_F_ZONE_APPEND;
7893 free_extent_map(em);
7898 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7902 btrfs_delalloc_release_space(BTRFS_I(inode),
7903 dio_data->data_reserved, start,
7904 dio_data->reserve, true);
7905 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7906 extent_changeset_free(dio_data->data_reserved);
7912 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7913 ssize_t written, unsigned int flags, struct iomap *iomap)
7916 struct btrfs_dio_data *dio_data = iomap->private;
7917 size_t submitted = dio_data->submitted;
7918 const bool write = !!(flags & IOMAP_WRITE);
7920 if (!write && (iomap->type == IOMAP_HOLE)) {
7921 /* If reading from a hole, unlock and return */
7922 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7926 if (submitted < length) {
7928 length -= submitted;
7930 __endio_write_update_ordered(BTRFS_I(inode), pos,
7933 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7939 if (dio_data->reserve)
7940 btrfs_delalloc_release_space(BTRFS_I(inode),
7941 dio_data->data_reserved, pos,
7942 dio_data->reserve, true);
7943 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7944 extent_changeset_free(dio_data->data_reserved);
7948 iomap->private = NULL;
7953 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7956 * This implies a barrier so that stores to dio_bio->bi_status before
7957 * this and loads of dio_bio->bi_status after this are fully ordered.
7959 if (!refcount_dec_and_test(&dip->refs))
7962 if (btrfs_op(dip->dio_bio) == BTRFS_MAP_WRITE) {
7963 __endio_write_update_ordered(BTRFS_I(dip->inode),
7964 dip->logical_offset,
7966 !dip->dio_bio->bi_status);
7968 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7969 dip->logical_offset,
7970 dip->logical_offset + dip->bytes - 1);
7973 bio_endio(dip->dio_bio);
7977 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7979 unsigned long bio_flags)
7981 struct btrfs_dio_private *dip = bio->bi_private;
7982 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7985 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7987 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7991 refcount_inc(&dip->refs);
7992 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7994 refcount_dec(&dip->refs);
7998 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7999 struct btrfs_io_bio *io_bio,
8000 const bool uptodate)
8002 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
8003 const u32 sectorsize = fs_info->sectorsize;
8004 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
8005 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8006 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8007 struct bio_vec bvec;
8008 struct bvec_iter iter;
8009 u64 start = io_bio->logical;
8011 blk_status_t err = BLK_STS_OK;
8013 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
8014 unsigned int i, nr_sectors, pgoff;
8016 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8017 pgoff = bvec.bv_offset;
8018 for (i = 0; i < nr_sectors; i++) {
8019 ASSERT(pgoff < PAGE_SIZE);
8021 (!csum || !check_data_csum(inode, io_bio,
8022 bio_offset, bvec.bv_page,
8024 clean_io_failure(fs_info, failure_tree, io_tree,
8025 start, bvec.bv_page,
8026 btrfs_ino(BTRFS_I(inode)),
8029 blk_status_t status;
8031 ASSERT((start - io_bio->logical) < UINT_MAX);
8032 status = btrfs_submit_read_repair(inode,
8034 start - io_bio->logical,
8035 bvec.bv_page, pgoff,
8037 start + sectorsize - 1,
8039 submit_dio_repair_bio);
8043 start += sectorsize;
8044 ASSERT(bio_offset + sectorsize > bio_offset);
8045 bio_offset += sectorsize;
8046 pgoff += sectorsize;
8052 static void __endio_write_update_ordered(struct btrfs_inode *inode,
8053 const u64 offset, const u64 bytes,
8054 const bool uptodate)
8056 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8057 struct btrfs_ordered_extent *ordered = NULL;
8058 struct btrfs_workqueue *wq;
8059 u64 ordered_offset = offset;
8060 u64 ordered_bytes = bytes;
8063 if (btrfs_is_free_space_inode(inode))
8064 wq = fs_info->endio_freespace_worker;
8066 wq = fs_info->endio_write_workers;
8068 while (ordered_offset < offset + bytes) {
8069 last_offset = ordered_offset;
8070 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8074 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
8076 btrfs_queue_work(wq, &ordered->work);
8079 /* No ordered extent found in the range, exit */
8080 if (ordered_offset == last_offset)
8083 * Our bio might span multiple ordered extents. In this case
8084 * we keep going until we have accounted the whole dio.
8086 if (ordered_offset < offset + bytes) {
8087 ordered_bytes = offset + bytes - ordered_offset;
8093 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
8095 u64 dio_file_offset)
8097 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
8100 static void btrfs_end_dio_bio(struct bio *bio)
8102 struct btrfs_dio_private *dip = bio->bi_private;
8103 blk_status_t err = bio->bi_status;
8106 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8107 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8108 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8109 bio->bi_opf, bio->bi_iter.bi_sector,
8110 bio->bi_iter.bi_size, err);
8112 if (bio_op(bio) == REQ_OP_READ) {
8113 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
8118 dip->dio_bio->bi_status = err;
8120 btrfs_record_physical_zoned(dip->inode, dip->logical_offset, bio);
8123 btrfs_dio_private_put(dip);
8126 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8127 struct inode *inode, u64 file_offset, int async_submit)
8129 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8130 struct btrfs_dio_private *dip = bio->bi_private;
8131 bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
8134 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8136 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8139 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8144 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8147 if (write && async_submit) {
8148 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
8149 btrfs_submit_bio_start_direct_io);
8153 * If we aren't doing async submit, calculate the csum of the
8156 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
8162 csum_offset = file_offset - dip->logical_offset;
8163 csum_offset >>= fs_info->sectorsize_bits;
8164 csum_offset *= fs_info->csum_size;
8165 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
8168 ret = btrfs_map_bio(fs_info, bio, 0);
8174 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
8175 * or ordered extents whether or not we submit any bios.
8177 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
8178 struct inode *inode,
8181 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8182 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
8184 struct btrfs_dio_private *dip;
8186 dip_size = sizeof(*dip);
8187 if (!write && csum) {
8188 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8191 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
8192 dip_size += fs_info->csum_size * nblocks;
8195 dip = kzalloc(dip_size, GFP_NOFS);
8200 dip->logical_offset = file_offset;
8201 dip->bytes = dio_bio->bi_iter.bi_size;
8202 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
8203 dip->dio_bio = dio_bio;
8204 refcount_set(&dip->refs, 1);
8208 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
8209 struct bio *dio_bio, loff_t file_offset)
8211 const bool write = (btrfs_op(dio_bio) == BTRFS_MAP_WRITE);
8212 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8213 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
8214 BTRFS_BLOCK_GROUP_RAID56_MASK);
8215 struct btrfs_dio_private *dip;
8218 int async_submit = 0;
8220 int clone_offset = 0;
8224 blk_status_t status;
8225 struct btrfs_io_geometry geom;
8226 struct btrfs_dio_data *dio_data = iomap->private;
8227 struct extent_map *em = NULL;
8229 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
8232 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8233 file_offset + dio_bio->bi_iter.bi_size - 1);
8235 dio_bio->bi_status = BLK_STS_RESOURCE;
8237 return BLK_QC_T_NONE;
8242 * Load the csums up front to reduce csum tree searches and
8243 * contention when submitting bios.
8245 * If we have csums disabled this will do nothing.
8247 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
8248 if (status != BLK_STS_OK)
8252 start_sector = dio_bio->bi_iter.bi_sector;
8253 submit_len = dio_bio->bi_iter.bi_size;
8256 logical = start_sector << 9;
8257 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
8259 status = errno_to_blk_status(PTR_ERR(em));
8263 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
8264 logical, submit_len, &geom);
8266 status = errno_to_blk_status(ret);
8269 ASSERT(geom.len <= INT_MAX);
8271 clone_len = min_t(int, submit_len, geom.len);
8274 * This will never fail as it's passing GPF_NOFS and
8275 * the allocation is backed by btrfs_bioset.
8277 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
8278 bio->bi_private = dip;
8279 bio->bi_end_io = btrfs_end_dio_bio;
8280 btrfs_io_bio(bio)->logical = file_offset;
8282 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
8283 status = extract_ordered_extent(BTRFS_I(inode), bio,
8291 ASSERT(submit_len >= clone_len);
8292 submit_len -= clone_len;
8295 * Increase the count before we submit the bio so we know
8296 * the end IO handler won't happen before we increase the
8297 * count. Otherwise, the dip might get freed before we're
8298 * done setting it up.
8300 * We transfer the initial reference to the last bio, so we
8301 * don't need to increment the reference count for the last one.
8303 if (submit_len > 0) {
8304 refcount_inc(&dip->refs);
8306 * If we are submitting more than one bio, submit them
8307 * all asynchronously. The exception is RAID 5 or 6, as
8308 * asynchronous checksums make it difficult to collect
8309 * full stripe writes.
8315 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8320 refcount_dec(&dip->refs);
8324 dio_data->submitted += clone_len;
8325 clone_offset += clone_len;
8326 start_sector += clone_len >> 9;
8327 file_offset += clone_len;
8329 free_extent_map(em);
8330 } while (submit_len > 0);
8331 return BLK_QC_T_NONE;
8334 free_extent_map(em);
8336 dip->dio_bio->bi_status = status;
8337 btrfs_dio_private_put(dip);
8339 return BLK_QC_T_NONE;
8342 const struct iomap_ops btrfs_dio_iomap_ops = {
8343 .iomap_begin = btrfs_dio_iomap_begin,
8344 .iomap_end = btrfs_dio_iomap_end,
8347 const struct iomap_dio_ops btrfs_dio_ops = {
8348 .submit_io = btrfs_submit_direct,
8351 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8356 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8360 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8363 int btrfs_readpage(struct file *file, struct page *page)
8365 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8366 u64 start = page_offset(page);
8367 u64 end = start + PAGE_SIZE - 1;
8368 unsigned long bio_flags = 0;
8369 struct bio *bio = NULL;
8372 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8374 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8376 ret = submit_one_bio(bio, 0, bio_flags);
8380 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8382 struct inode *inode = page->mapping->host;
8385 if (current->flags & PF_MEMALLOC) {
8386 redirty_page_for_writepage(wbc, page);
8392 * If we are under memory pressure we will call this directly from the
8393 * VM, we need to make sure we have the inode referenced for the ordered
8394 * extent. If not just return like we didn't do anything.
8396 if (!igrab(inode)) {
8397 redirty_page_for_writepage(wbc, page);
8398 return AOP_WRITEPAGE_ACTIVATE;
8400 ret = extent_write_full_page(page, wbc);
8401 btrfs_add_delayed_iput(inode);
8405 static int btrfs_writepages(struct address_space *mapping,
8406 struct writeback_control *wbc)
8408 return extent_writepages(mapping, wbc);
8411 static void btrfs_readahead(struct readahead_control *rac)
8413 extent_readahead(rac);
8416 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8418 int ret = try_release_extent_mapping(page, gfp_flags);
8420 clear_page_extent_mapped(page);
8424 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8426 if (PageWriteback(page) || PageDirty(page))
8428 return __btrfs_releasepage(page, gfp_flags);
8431 #ifdef CONFIG_MIGRATION
8432 static int btrfs_migratepage(struct address_space *mapping,
8433 struct page *newpage, struct page *page,
8434 enum migrate_mode mode)
8438 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8439 if (ret != MIGRATEPAGE_SUCCESS)
8442 if (page_has_private(page))
8443 attach_page_private(newpage, detach_page_private(page));
8445 if (PagePrivate2(page)) {
8446 ClearPagePrivate2(page);
8447 SetPagePrivate2(newpage);
8450 if (mode != MIGRATE_SYNC_NO_COPY)
8451 migrate_page_copy(newpage, page);
8453 migrate_page_states(newpage, page);
8454 return MIGRATEPAGE_SUCCESS;
8458 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8459 unsigned int length)
8461 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8462 struct extent_io_tree *tree = &inode->io_tree;
8463 struct btrfs_ordered_extent *ordered;
8464 struct extent_state *cached_state = NULL;
8465 u64 page_start = page_offset(page);
8466 u64 page_end = page_start + PAGE_SIZE - 1;
8469 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8470 bool found_ordered = false;
8471 bool completed_ordered = false;
8474 * we have the page locked, so new writeback can't start,
8475 * and the dirty bit won't be cleared while we are here.
8477 * Wait for IO on this page so that we can safely clear
8478 * the PagePrivate2 bit and do ordered accounting
8480 wait_on_page_writeback(page);
8483 * For subpage case, we have call sites like
8484 * btrfs_punch_hole_lock_range() which passes range not aligned to
8486 * If the range doesn't cover the full page, we don't need to and
8487 * shouldn't clear page extent mapped, as page->private can still
8488 * record subpage dirty bits for other part of the range.
8490 * For cases that can invalidate the full even the range doesn't
8491 * cover the full page, like invalidating the last page, we're
8492 * still safe to wait for ordered extent to finish.
8494 if (!(offset == 0 && length == PAGE_SIZE)) {
8495 btrfs_releasepage(page, GFP_NOFS);
8499 if (!inode_evicting)
8500 lock_extent_bits(tree, page_start, page_end, &cached_state);
8504 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8506 found_ordered = true;
8508 ordered->file_offset + ordered->num_bytes - 1);
8510 * IO on this page will never be started, so we need to account
8511 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8512 * here, must leave that up for the ordered extent completion.
8514 if (!inode_evicting)
8515 clear_extent_bit(tree, start, end,
8517 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8518 EXTENT_DEFRAG, 1, 0, &cached_state);
8520 * whoever cleared the private bit is responsible
8521 * for the finish_ordered_io
8523 if (TestClearPagePrivate2(page)) {
8524 spin_lock_irq(&inode->ordered_tree.lock);
8525 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8526 ordered->truncated_len = min(ordered->truncated_len,
8527 start - ordered->file_offset);
8528 spin_unlock_irq(&inode->ordered_tree.lock);
8530 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8532 end - start + 1, 1)) {
8533 btrfs_finish_ordered_io(ordered);
8534 completed_ordered = true;
8537 btrfs_put_ordered_extent(ordered);
8538 if (!inode_evicting) {
8539 cached_state = NULL;
8540 lock_extent_bits(tree, start, end,
8545 if (start < page_end)
8550 * Qgroup reserved space handler
8551 * Page here will be either
8552 * 1) Already written to disk or ordered extent already submitted
8553 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8554 * Qgroup will be handled by its qgroup_record then.
8555 * btrfs_qgroup_free_data() call will do nothing here.
8557 * 2) Not written to disk yet
8558 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8559 * bit of its io_tree, and free the qgroup reserved data space.
8560 * Since the IO will never happen for this page.
8562 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8563 if (!inode_evicting) {
8567 * If there's an ordered extent for this range and we have not
8568 * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set
8569 * in the range for the ordered extent completion. We must also
8570 * not delete the range, otherwise we would lose that bit (and
8571 * any other bits set in the range). Make sure EXTENT_UPTODATE
8572 * is cleared if we don't delete, otherwise it can lead to
8573 * corruptions if the i_size is extented later.
8575 if (found_ordered && !completed_ordered)
8577 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8578 EXTENT_DELALLOC | EXTENT_UPTODATE |
8579 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8580 delete, &cached_state);
8582 __btrfs_releasepage(page, GFP_NOFS);
8585 ClearPageChecked(page);
8586 clear_page_extent_mapped(page);
8590 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8591 * called from a page fault handler when a page is first dirtied. Hence we must
8592 * be careful to check for EOF conditions here. We set the page up correctly
8593 * for a written page which means we get ENOSPC checking when writing into
8594 * holes and correct delalloc and unwritten extent mapping on filesystems that
8595 * support these features.
8597 * We are not allowed to take the i_mutex here so we have to play games to
8598 * protect against truncate races as the page could now be beyond EOF. Because
8599 * truncate_setsize() writes the inode size before removing pages, once we have
8600 * the page lock we can determine safely if the page is beyond EOF. If it is not
8601 * beyond EOF, then the page is guaranteed safe against truncation until we
8604 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8606 struct page *page = vmf->page;
8607 struct inode *inode = file_inode(vmf->vma->vm_file);
8608 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8609 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8610 struct btrfs_ordered_extent *ordered;
8611 struct extent_state *cached_state = NULL;
8612 struct extent_changeset *data_reserved = NULL;
8613 unsigned long zero_start;
8623 reserved_space = PAGE_SIZE;
8625 sb_start_pagefault(inode->i_sb);
8626 page_start = page_offset(page);
8627 page_end = page_start + PAGE_SIZE - 1;
8631 * Reserving delalloc space after obtaining the page lock can lead to
8632 * deadlock. For example, if a dirty page is locked by this function
8633 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8634 * dirty page write out, then the btrfs_writepage() function could
8635 * end up waiting indefinitely to get a lock on the page currently
8636 * being processed by btrfs_page_mkwrite() function.
8638 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8639 page_start, reserved_space);
8641 ret2 = file_update_time(vmf->vma->vm_file);
8645 ret = vmf_error(ret2);
8651 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8653 down_read(&BTRFS_I(inode)->i_mmap_lock);
8655 size = i_size_read(inode);
8657 if ((page->mapping != inode->i_mapping) ||
8658 (page_start >= size)) {
8659 /* page got truncated out from underneath us */
8662 wait_on_page_writeback(page);
8664 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8665 ret2 = set_page_extent_mapped(page);
8667 ret = vmf_error(ret2);
8668 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8673 * we can't set the delalloc bits if there are pending ordered
8674 * extents. Drop our locks and wait for them to finish
8676 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8679 unlock_extent_cached(io_tree, page_start, page_end,
8682 up_read(&BTRFS_I(inode)->i_mmap_lock);
8683 btrfs_start_ordered_extent(ordered, 1);
8684 btrfs_put_ordered_extent(ordered);
8688 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8689 reserved_space = round_up(size - page_start,
8690 fs_info->sectorsize);
8691 if (reserved_space < PAGE_SIZE) {
8692 end = page_start + reserved_space - 1;
8693 btrfs_delalloc_release_space(BTRFS_I(inode),
8694 data_reserved, page_start,
8695 PAGE_SIZE - reserved_space, true);
8700 * page_mkwrite gets called when the page is firstly dirtied after it's
8701 * faulted in, but write(2) could also dirty a page and set delalloc
8702 * bits, thus in this case for space account reason, we still need to
8703 * clear any delalloc bits within this page range since we have to
8704 * reserve data&meta space before lock_page() (see above comments).
8706 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8707 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8708 EXTENT_DEFRAG, 0, 0, &cached_state);
8710 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8713 unlock_extent_cached(io_tree, page_start, page_end,
8715 ret = VM_FAULT_SIGBUS;
8719 /* page is wholly or partially inside EOF */
8720 if (page_start + PAGE_SIZE > size)
8721 zero_start = offset_in_page(size);
8723 zero_start = PAGE_SIZE;
8725 if (zero_start != PAGE_SIZE) {
8726 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8727 flush_dcache_page(page);
8729 ClearPageChecked(page);
8730 set_page_dirty(page);
8731 SetPageUptodate(page);
8733 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8735 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8736 up_read(&BTRFS_I(inode)->i_mmap_lock);
8738 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8739 sb_end_pagefault(inode->i_sb);
8740 extent_changeset_free(data_reserved);
8741 return VM_FAULT_LOCKED;
8745 up_read(&BTRFS_I(inode)->i_mmap_lock);
8747 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8748 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8749 reserved_space, (ret != 0));
8751 sb_end_pagefault(inode->i_sb);
8752 extent_changeset_free(data_reserved);
8756 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8758 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8759 struct btrfs_root *root = BTRFS_I(inode)->root;
8760 struct btrfs_block_rsv *rsv;
8762 struct btrfs_trans_handle *trans;
8763 u64 mask = fs_info->sectorsize - 1;
8764 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8766 if (!skip_writeback) {
8767 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8774 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8775 * things going on here:
8777 * 1) We need to reserve space to update our inode.
8779 * 2) We need to have something to cache all the space that is going to
8780 * be free'd up by the truncate operation, but also have some slack
8781 * space reserved in case it uses space during the truncate (thank you
8782 * very much snapshotting).
8784 * And we need these to be separate. The fact is we can use a lot of
8785 * space doing the truncate, and we have no earthly idea how much space
8786 * we will use, so we need the truncate reservation to be separate so it
8787 * doesn't end up using space reserved for updating the inode. We also
8788 * need to be able to stop the transaction and start a new one, which
8789 * means we need to be able to update the inode several times, and we
8790 * have no idea of knowing how many times that will be, so we can't just
8791 * reserve 1 item for the entirety of the operation, so that has to be
8792 * done separately as well.
8794 * So that leaves us with
8796 * 1) rsv - for the truncate reservation, which we will steal from the
8797 * transaction reservation.
8798 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8799 * updating the inode.
8801 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8804 rsv->size = min_size;
8808 * 1 for the truncate slack space
8809 * 1 for updating the inode.
8811 trans = btrfs_start_transaction(root, 2);
8812 if (IS_ERR(trans)) {
8813 ret = PTR_ERR(trans);
8817 /* Migrate the slack space for the truncate to our reserve */
8818 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8823 * So if we truncate and then write and fsync we normally would just
8824 * write the extents that changed, which is a problem if we need to
8825 * first truncate that entire inode. So set this flag so we write out
8826 * all of the extents in the inode to the sync log so we're completely
8829 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8830 trans->block_rsv = rsv;
8833 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8835 BTRFS_EXTENT_DATA_KEY);
8836 trans->block_rsv = &fs_info->trans_block_rsv;
8837 if (ret != -ENOSPC && ret != -EAGAIN)
8840 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8844 btrfs_end_transaction(trans);
8845 btrfs_btree_balance_dirty(fs_info);
8847 trans = btrfs_start_transaction(root, 2);
8848 if (IS_ERR(trans)) {
8849 ret = PTR_ERR(trans);
8854 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8855 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8856 rsv, min_size, false);
8857 BUG_ON(ret); /* shouldn't happen */
8858 trans->block_rsv = rsv;
8862 * We can't call btrfs_truncate_block inside a trans handle as we could
8863 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8864 * we've truncated everything except the last little bit, and can do
8865 * btrfs_truncate_block and then update the disk_i_size.
8867 if (ret == NEED_TRUNCATE_BLOCK) {
8868 btrfs_end_transaction(trans);
8869 btrfs_btree_balance_dirty(fs_info);
8871 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8874 trans = btrfs_start_transaction(root, 1);
8875 if (IS_ERR(trans)) {
8876 ret = PTR_ERR(trans);
8879 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8885 trans->block_rsv = &fs_info->trans_block_rsv;
8886 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8890 ret2 = btrfs_end_transaction(trans);
8893 btrfs_btree_balance_dirty(fs_info);
8896 btrfs_free_block_rsv(fs_info, rsv);
8902 * create a new subvolume directory/inode (helper for the ioctl).
8904 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8905 struct btrfs_root *new_root,
8906 struct btrfs_root *parent_root)
8908 struct inode *inode;
8913 err = btrfs_get_free_objectid(new_root, &ino);
8917 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8918 S_IFDIR | (~current_umask() & S_IRWXUGO),
8921 return PTR_ERR(inode);
8922 inode->i_op = &btrfs_dir_inode_operations;
8923 inode->i_fop = &btrfs_dir_file_operations;
8925 set_nlink(inode, 1);
8926 btrfs_i_size_write(BTRFS_I(inode), 0);
8927 unlock_new_inode(inode);
8929 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8931 btrfs_err(new_root->fs_info,
8932 "error inheriting subvolume %llu properties: %d",
8933 new_root->root_key.objectid, err);
8935 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8941 struct inode *btrfs_alloc_inode(struct super_block *sb)
8943 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8944 struct btrfs_inode *ei;
8945 struct inode *inode;
8947 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8954 ei->last_sub_trans = 0;
8955 ei->logged_trans = 0;
8956 ei->delalloc_bytes = 0;
8957 ei->new_delalloc_bytes = 0;
8958 ei->defrag_bytes = 0;
8959 ei->disk_i_size = 0;
8962 ei->index_cnt = (u64)-1;
8964 ei->last_unlink_trans = 0;
8965 ei->last_reflink_trans = 0;
8966 ei->last_log_commit = 0;
8968 spin_lock_init(&ei->lock);
8969 ei->outstanding_extents = 0;
8970 if (sb->s_magic != BTRFS_TEST_MAGIC)
8971 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8972 BTRFS_BLOCK_RSV_DELALLOC);
8973 ei->runtime_flags = 0;
8974 ei->prop_compress = BTRFS_COMPRESS_NONE;
8975 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8977 ei->delayed_node = NULL;
8979 ei->i_otime.tv_sec = 0;
8980 ei->i_otime.tv_nsec = 0;
8982 inode = &ei->vfs_inode;
8983 extent_map_tree_init(&ei->extent_tree);
8984 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8985 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8986 IO_TREE_INODE_IO_FAILURE, inode);
8987 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8988 IO_TREE_INODE_FILE_EXTENT, inode);
8989 ei->io_tree.track_uptodate = true;
8990 ei->io_failure_tree.track_uptodate = true;
8991 atomic_set(&ei->sync_writers, 0);
8992 mutex_init(&ei->log_mutex);
8993 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8994 INIT_LIST_HEAD(&ei->delalloc_inodes);
8995 INIT_LIST_HEAD(&ei->delayed_iput);
8996 RB_CLEAR_NODE(&ei->rb_node);
8997 init_rwsem(&ei->i_mmap_lock);
9002 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9003 void btrfs_test_destroy_inode(struct inode *inode)
9005 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9006 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9010 void btrfs_free_inode(struct inode *inode)
9012 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9015 void btrfs_destroy_inode(struct inode *vfs_inode)
9017 struct btrfs_ordered_extent *ordered;
9018 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
9019 struct btrfs_root *root = inode->root;
9021 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
9022 WARN_ON(vfs_inode->i_data.nrpages);
9023 WARN_ON(inode->block_rsv.reserved);
9024 WARN_ON(inode->block_rsv.size);
9025 WARN_ON(inode->outstanding_extents);
9026 WARN_ON(inode->delalloc_bytes);
9027 WARN_ON(inode->new_delalloc_bytes);
9028 WARN_ON(inode->csum_bytes);
9029 WARN_ON(inode->defrag_bytes);
9032 * This can happen where we create an inode, but somebody else also
9033 * created the same inode and we need to destroy the one we already
9040 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9044 btrfs_err(root->fs_info,
9045 "found ordered extent %llu %llu on inode cleanup",
9046 ordered->file_offset, ordered->num_bytes);
9047 btrfs_remove_ordered_extent(inode, ordered);
9048 btrfs_put_ordered_extent(ordered);
9049 btrfs_put_ordered_extent(ordered);
9052 btrfs_qgroup_check_reserved_leak(inode);
9053 inode_tree_del(inode);
9054 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9055 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
9056 btrfs_put_root(inode->root);
9059 int btrfs_drop_inode(struct inode *inode)
9061 struct btrfs_root *root = BTRFS_I(inode)->root;
9066 /* the snap/subvol tree is on deleting */
9067 if (btrfs_root_refs(&root->root_item) == 0)
9070 return generic_drop_inode(inode);
9073 static void init_once(void *foo)
9075 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9077 inode_init_once(&ei->vfs_inode);
9080 void __cold btrfs_destroy_cachep(void)
9083 * Make sure all delayed rcu free inodes are flushed before we
9087 kmem_cache_destroy(btrfs_inode_cachep);
9088 kmem_cache_destroy(btrfs_trans_handle_cachep);
9089 kmem_cache_destroy(btrfs_path_cachep);
9090 kmem_cache_destroy(btrfs_free_space_cachep);
9091 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
9094 int __init btrfs_init_cachep(void)
9096 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9097 sizeof(struct btrfs_inode), 0,
9098 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9100 if (!btrfs_inode_cachep)
9103 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9104 sizeof(struct btrfs_trans_handle), 0,
9105 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9106 if (!btrfs_trans_handle_cachep)
9109 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9110 sizeof(struct btrfs_path), 0,
9111 SLAB_MEM_SPREAD, NULL);
9112 if (!btrfs_path_cachep)
9115 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9116 sizeof(struct btrfs_free_space), 0,
9117 SLAB_MEM_SPREAD, NULL);
9118 if (!btrfs_free_space_cachep)
9121 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
9122 PAGE_SIZE, PAGE_SIZE,
9123 SLAB_MEM_SPREAD, NULL);
9124 if (!btrfs_free_space_bitmap_cachep)
9129 btrfs_destroy_cachep();
9133 static int btrfs_getattr(struct user_namespace *mnt_userns,
9134 const struct path *path, struct kstat *stat,
9135 u32 request_mask, unsigned int flags)
9139 struct inode *inode = d_inode(path->dentry);
9140 u32 blocksize = inode->i_sb->s_blocksize;
9141 u32 bi_flags = BTRFS_I(inode)->flags;
9143 stat->result_mask |= STATX_BTIME;
9144 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9145 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9146 if (bi_flags & BTRFS_INODE_APPEND)
9147 stat->attributes |= STATX_ATTR_APPEND;
9148 if (bi_flags & BTRFS_INODE_COMPRESS)
9149 stat->attributes |= STATX_ATTR_COMPRESSED;
9150 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9151 stat->attributes |= STATX_ATTR_IMMUTABLE;
9152 if (bi_flags & BTRFS_INODE_NODUMP)
9153 stat->attributes |= STATX_ATTR_NODUMP;
9155 stat->attributes_mask |= (STATX_ATTR_APPEND |
9156 STATX_ATTR_COMPRESSED |
9157 STATX_ATTR_IMMUTABLE |
9160 generic_fillattr(&init_user_ns, inode, stat);
9161 stat->dev = BTRFS_I(inode)->root->anon_dev;
9163 spin_lock(&BTRFS_I(inode)->lock);
9164 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9165 inode_bytes = inode_get_bytes(inode);
9166 spin_unlock(&BTRFS_I(inode)->lock);
9167 stat->blocks = (ALIGN(inode_bytes, blocksize) +
9168 ALIGN(delalloc_bytes, blocksize)) >> 9;
9172 static int btrfs_rename_exchange(struct inode *old_dir,
9173 struct dentry *old_dentry,
9174 struct inode *new_dir,
9175 struct dentry *new_dentry)
9177 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9178 struct btrfs_trans_handle *trans;
9179 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9180 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9181 struct inode *new_inode = new_dentry->d_inode;
9182 struct inode *old_inode = old_dentry->d_inode;
9183 struct timespec64 ctime = current_time(old_inode);
9184 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9185 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9190 bool root_log_pinned = false;
9191 bool dest_log_pinned = false;
9192 bool need_abort = false;
9195 * For non-subvolumes allow exchange only within one subvolume, in the
9196 * same inode namespace. Two subvolumes (represented as directory) can
9197 * be exchanged as they're a logical link and have a fixed inode number.
9200 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
9201 new_ino != BTRFS_FIRST_FREE_OBJECTID))
9204 /* close the race window with snapshot create/destroy ioctl */
9205 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
9206 new_ino == BTRFS_FIRST_FREE_OBJECTID)
9207 down_read(&fs_info->subvol_sem);
9210 * We want to reserve the absolute worst case amount of items. So if
9211 * both inodes are subvols and we need to unlink them then that would
9212 * require 4 item modifications, but if they are both normal inodes it
9213 * would require 5 item modifications, so we'll assume their normal
9214 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9215 * should cover the worst case number of items we'll modify.
9217 trans = btrfs_start_transaction(root, 12);
9218 if (IS_ERR(trans)) {
9219 ret = PTR_ERR(trans);
9224 ret = btrfs_record_root_in_trans(trans, dest);
9230 * We need to find a free sequence number both in the source and
9231 * in the destination directory for the exchange.
9233 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9236 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9240 BTRFS_I(old_inode)->dir_index = 0ULL;
9241 BTRFS_I(new_inode)->dir_index = 0ULL;
9243 /* Reference for the source. */
9244 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9245 /* force full log commit if subvolume involved. */
9246 btrfs_set_log_full_commit(trans);
9248 btrfs_pin_log_trans(root);
9249 root_log_pinned = true;
9250 ret = btrfs_insert_inode_ref(trans, dest,
9251 new_dentry->d_name.name,
9252 new_dentry->d_name.len,
9254 btrfs_ino(BTRFS_I(new_dir)),
9261 /* And now for the dest. */
9262 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9263 /* force full log commit if subvolume involved. */
9264 btrfs_set_log_full_commit(trans);
9266 btrfs_pin_log_trans(dest);
9267 dest_log_pinned = true;
9268 ret = btrfs_insert_inode_ref(trans, root,
9269 old_dentry->d_name.name,
9270 old_dentry->d_name.len,
9272 btrfs_ino(BTRFS_I(old_dir)),
9276 btrfs_abort_transaction(trans, ret);
9281 /* Update inode version and ctime/mtime. */
9282 inode_inc_iversion(old_dir);
9283 inode_inc_iversion(new_dir);
9284 inode_inc_iversion(old_inode);
9285 inode_inc_iversion(new_inode);
9286 old_dir->i_ctime = old_dir->i_mtime = ctime;
9287 new_dir->i_ctime = new_dir->i_mtime = ctime;
9288 old_inode->i_ctime = ctime;
9289 new_inode->i_ctime = ctime;
9291 if (old_dentry->d_parent != new_dentry->d_parent) {
9292 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9293 BTRFS_I(old_inode), 1);
9294 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9295 BTRFS_I(new_inode), 1);
9298 /* src is a subvolume */
9299 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9300 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9301 } else { /* src is an inode */
9302 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9303 BTRFS_I(old_dentry->d_inode),
9304 old_dentry->d_name.name,
9305 old_dentry->d_name.len);
9307 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9310 btrfs_abort_transaction(trans, ret);
9314 /* dest is a subvolume */
9315 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9316 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9317 } else { /* dest is an inode */
9318 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9319 BTRFS_I(new_dentry->d_inode),
9320 new_dentry->d_name.name,
9321 new_dentry->d_name.len);
9323 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9326 btrfs_abort_transaction(trans, ret);
9330 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9331 new_dentry->d_name.name,
9332 new_dentry->d_name.len, 0, old_idx);
9334 btrfs_abort_transaction(trans, ret);
9338 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9339 old_dentry->d_name.name,
9340 old_dentry->d_name.len, 0, new_idx);
9342 btrfs_abort_transaction(trans, ret);
9346 if (old_inode->i_nlink == 1)
9347 BTRFS_I(old_inode)->dir_index = old_idx;
9348 if (new_inode->i_nlink == 1)
9349 BTRFS_I(new_inode)->dir_index = new_idx;
9351 if (root_log_pinned) {
9352 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9353 new_dentry->d_parent);
9354 btrfs_end_log_trans(root);
9355 root_log_pinned = false;
9357 if (dest_log_pinned) {
9358 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9359 old_dentry->d_parent);
9360 btrfs_end_log_trans(dest);
9361 dest_log_pinned = false;
9365 * If we have pinned a log and an error happened, we unpin tasks
9366 * trying to sync the log and force them to fallback to a transaction
9367 * commit if the log currently contains any of the inodes involved in
9368 * this rename operation (to ensure we do not persist a log with an
9369 * inconsistent state for any of these inodes or leading to any
9370 * inconsistencies when replayed). If the transaction was aborted, the
9371 * abortion reason is propagated to userspace when attempting to commit
9372 * the transaction. If the log does not contain any of these inodes, we
9373 * allow the tasks to sync it.
9375 if (ret && (root_log_pinned || dest_log_pinned)) {
9376 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9377 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9378 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9380 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9381 btrfs_set_log_full_commit(trans);
9383 if (root_log_pinned) {
9384 btrfs_end_log_trans(root);
9385 root_log_pinned = false;
9387 if (dest_log_pinned) {
9388 btrfs_end_log_trans(dest);
9389 dest_log_pinned = false;
9392 ret2 = btrfs_end_transaction(trans);
9393 ret = ret ? ret : ret2;
9395 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9396 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9397 up_read(&fs_info->subvol_sem);
9402 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9403 struct btrfs_root *root,
9405 struct dentry *dentry)
9408 struct inode *inode;
9412 ret = btrfs_get_free_objectid(root, &objectid);
9416 inode = btrfs_new_inode(trans, root, dir,
9417 dentry->d_name.name,
9419 btrfs_ino(BTRFS_I(dir)),
9421 S_IFCHR | WHITEOUT_MODE,
9424 if (IS_ERR(inode)) {
9425 ret = PTR_ERR(inode);
9429 inode->i_op = &btrfs_special_inode_operations;
9430 init_special_inode(inode, inode->i_mode,
9433 ret = btrfs_init_inode_security(trans, inode, dir,
9438 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9439 BTRFS_I(inode), 0, index);
9443 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9445 unlock_new_inode(inode);
9447 inode_dec_link_count(inode);
9453 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9454 struct inode *new_dir, struct dentry *new_dentry,
9457 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9458 struct btrfs_trans_handle *trans;
9459 unsigned int trans_num_items;
9460 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9461 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9462 struct inode *new_inode = d_inode(new_dentry);
9463 struct inode *old_inode = d_inode(old_dentry);
9467 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9468 bool log_pinned = false;
9470 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9473 /* we only allow rename subvolume link between subvolumes */
9474 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9477 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9478 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9481 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9482 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9486 /* check for collisions, even if the name isn't there */
9487 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9488 new_dentry->d_name.name,
9489 new_dentry->d_name.len);
9492 if (ret == -EEXIST) {
9494 * eexist without a new_inode */
9495 if (WARN_ON(!new_inode)) {
9499 /* maybe -EOVERFLOW */
9506 * we're using rename to replace one file with another. Start IO on it
9507 * now so we don't add too much work to the end of the transaction
9509 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9510 filemap_flush(old_inode->i_mapping);
9512 /* close the racy window with snapshot create/destroy ioctl */
9513 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9514 down_read(&fs_info->subvol_sem);
9516 * We want to reserve the absolute worst case amount of items. So if
9517 * both inodes are subvols and we need to unlink them then that would
9518 * require 4 item modifications, but if they are both normal inodes it
9519 * would require 5 item modifications, so we'll assume they are normal
9520 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9521 * should cover the worst case number of items we'll modify.
9522 * If our rename has the whiteout flag, we need more 5 units for the
9523 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9524 * when selinux is enabled).
9526 trans_num_items = 11;
9527 if (flags & RENAME_WHITEOUT)
9528 trans_num_items += 5;
9529 trans = btrfs_start_transaction(root, trans_num_items);
9530 if (IS_ERR(trans)) {
9531 ret = PTR_ERR(trans);
9536 ret = btrfs_record_root_in_trans(trans, dest);
9541 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9545 BTRFS_I(old_inode)->dir_index = 0ULL;
9546 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9547 /* force full log commit if subvolume involved. */
9548 btrfs_set_log_full_commit(trans);
9550 btrfs_pin_log_trans(root);
9552 ret = btrfs_insert_inode_ref(trans, dest,
9553 new_dentry->d_name.name,
9554 new_dentry->d_name.len,
9556 btrfs_ino(BTRFS_I(new_dir)), index);
9561 inode_inc_iversion(old_dir);
9562 inode_inc_iversion(new_dir);
9563 inode_inc_iversion(old_inode);
9564 old_dir->i_ctime = old_dir->i_mtime =
9565 new_dir->i_ctime = new_dir->i_mtime =
9566 old_inode->i_ctime = current_time(old_dir);
9568 if (old_dentry->d_parent != new_dentry->d_parent)
9569 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9570 BTRFS_I(old_inode), 1);
9572 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9573 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9575 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9576 BTRFS_I(d_inode(old_dentry)),
9577 old_dentry->d_name.name,
9578 old_dentry->d_name.len);
9580 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9583 btrfs_abort_transaction(trans, ret);
9588 inode_inc_iversion(new_inode);
9589 new_inode->i_ctime = current_time(new_inode);
9590 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9591 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9592 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9593 BUG_ON(new_inode->i_nlink == 0);
9595 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9596 BTRFS_I(d_inode(new_dentry)),
9597 new_dentry->d_name.name,
9598 new_dentry->d_name.len);
9600 if (!ret && new_inode->i_nlink == 0)
9601 ret = btrfs_orphan_add(trans,
9602 BTRFS_I(d_inode(new_dentry)));
9604 btrfs_abort_transaction(trans, ret);
9609 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9610 new_dentry->d_name.name,
9611 new_dentry->d_name.len, 0, index);
9613 btrfs_abort_transaction(trans, ret);
9617 if (old_inode->i_nlink == 1)
9618 BTRFS_I(old_inode)->dir_index = index;
9621 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9622 new_dentry->d_parent);
9623 btrfs_end_log_trans(root);
9627 if (flags & RENAME_WHITEOUT) {
9628 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9632 btrfs_abort_transaction(trans, ret);
9638 * If we have pinned the log and an error happened, we unpin tasks
9639 * trying to sync the log and force them to fallback to a transaction
9640 * commit if the log currently contains any of the inodes involved in
9641 * this rename operation (to ensure we do not persist a log with an
9642 * inconsistent state for any of these inodes or leading to any
9643 * inconsistencies when replayed). If the transaction was aborted, the
9644 * abortion reason is propagated to userspace when attempting to commit
9645 * the transaction. If the log does not contain any of these inodes, we
9646 * allow the tasks to sync it.
9648 if (ret && log_pinned) {
9649 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9650 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9651 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9653 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9654 btrfs_set_log_full_commit(trans);
9656 btrfs_end_log_trans(root);
9659 ret2 = btrfs_end_transaction(trans);
9660 ret = ret ? ret : ret2;
9662 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9663 up_read(&fs_info->subvol_sem);
9668 static int btrfs_rename2(struct user_namespace *mnt_userns, struct inode *old_dir,
9669 struct dentry *old_dentry, struct inode *new_dir,
9670 struct dentry *new_dentry, unsigned int flags)
9672 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9675 if (flags & RENAME_EXCHANGE)
9676 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9679 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9682 struct btrfs_delalloc_work {
9683 struct inode *inode;
9684 struct completion completion;
9685 struct list_head list;
9686 struct btrfs_work work;
9689 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9691 struct btrfs_delalloc_work *delalloc_work;
9692 struct inode *inode;
9694 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9696 inode = delalloc_work->inode;
9697 filemap_flush(inode->i_mapping);
9698 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9699 &BTRFS_I(inode)->runtime_flags))
9700 filemap_flush(inode->i_mapping);
9703 complete(&delalloc_work->completion);
9706 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9708 struct btrfs_delalloc_work *work;
9710 work = kmalloc(sizeof(*work), GFP_NOFS);
9714 init_completion(&work->completion);
9715 INIT_LIST_HEAD(&work->list);
9716 work->inode = inode;
9717 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9723 * some fairly slow code that needs optimization. This walks the list
9724 * of all the inodes with pending delalloc and forces them to disk.
9726 static int start_delalloc_inodes(struct btrfs_root *root,
9727 struct writeback_control *wbc, bool snapshot,
9728 bool in_reclaim_context)
9730 struct btrfs_inode *binode;
9731 struct inode *inode;
9732 struct btrfs_delalloc_work *work, *next;
9733 struct list_head works;
9734 struct list_head splice;
9736 bool full_flush = wbc->nr_to_write == LONG_MAX;
9738 INIT_LIST_HEAD(&works);
9739 INIT_LIST_HEAD(&splice);
9741 mutex_lock(&root->delalloc_mutex);
9742 spin_lock(&root->delalloc_lock);
9743 list_splice_init(&root->delalloc_inodes, &splice);
9744 while (!list_empty(&splice)) {
9745 binode = list_entry(splice.next, struct btrfs_inode,
9748 list_move_tail(&binode->delalloc_inodes,
9749 &root->delalloc_inodes);
9751 if (in_reclaim_context &&
9752 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9755 inode = igrab(&binode->vfs_inode);
9757 cond_resched_lock(&root->delalloc_lock);
9760 spin_unlock(&root->delalloc_lock);
9763 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9764 &binode->runtime_flags);
9766 work = btrfs_alloc_delalloc_work(inode);
9772 list_add_tail(&work->list, &works);
9773 btrfs_queue_work(root->fs_info->flush_workers,
9776 ret = sync_inode(inode, wbc);
9778 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9779 &BTRFS_I(inode)->runtime_flags))
9780 ret = sync_inode(inode, wbc);
9781 btrfs_add_delayed_iput(inode);
9782 if (ret || wbc->nr_to_write <= 0)
9786 spin_lock(&root->delalloc_lock);
9788 spin_unlock(&root->delalloc_lock);
9791 list_for_each_entry_safe(work, next, &works, list) {
9792 list_del_init(&work->list);
9793 wait_for_completion(&work->completion);
9797 if (!list_empty(&splice)) {
9798 spin_lock(&root->delalloc_lock);
9799 list_splice_tail(&splice, &root->delalloc_inodes);
9800 spin_unlock(&root->delalloc_lock);
9802 mutex_unlock(&root->delalloc_mutex);
9806 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9808 struct writeback_control wbc = {
9809 .nr_to_write = LONG_MAX,
9810 .sync_mode = WB_SYNC_NONE,
9812 .range_end = LLONG_MAX,
9814 struct btrfs_fs_info *fs_info = root->fs_info;
9816 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9819 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9822 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9823 bool in_reclaim_context)
9825 struct writeback_control wbc = {
9827 .sync_mode = WB_SYNC_NONE,
9829 .range_end = LLONG_MAX,
9831 struct btrfs_root *root;
9832 struct list_head splice;
9835 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9838 INIT_LIST_HEAD(&splice);
9840 mutex_lock(&fs_info->delalloc_root_mutex);
9841 spin_lock(&fs_info->delalloc_root_lock);
9842 list_splice_init(&fs_info->delalloc_roots, &splice);
9843 while (!list_empty(&splice)) {
9845 * Reset nr_to_write here so we know that we're doing a full
9849 wbc.nr_to_write = LONG_MAX;
9851 root = list_first_entry(&splice, struct btrfs_root,
9853 root = btrfs_grab_root(root);
9855 list_move_tail(&root->delalloc_root,
9856 &fs_info->delalloc_roots);
9857 spin_unlock(&fs_info->delalloc_root_lock);
9859 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9860 btrfs_put_root(root);
9861 if (ret < 0 || wbc.nr_to_write <= 0)
9863 spin_lock(&fs_info->delalloc_root_lock);
9865 spin_unlock(&fs_info->delalloc_root_lock);
9869 if (!list_empty(&splice)) {
9870 spin_lock(&fs_info->delalloc_root_lock);
9871 list_splice_tail(&splice, &fs_info->delalloc_roots);
9872 spin_unlock(&fs_info->delalloc_root_lock);
9874 mutex_unlock(&fs_info->delalloc_root_mutex);
9878 static int btrfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
9879 struct dentry *dentry, const char *symname)
9881 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9882 struct btrfs_trans_handle *trans;
9883 struct btrfs_root *root = BTRFS_I(dir)->root;
9884 struct btrfs_path *path;
9885 struct btrfs_key key;
9886 struct inode *inode = NULL;
9893 struct btrfs_file_extent_item *ei;
9894 struct extent_buffer *leaf;
9896 name_len = strlen(symname);
9897 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9898 return -ENAMETOOLONG;
9901 * 2 items for inode item and ref
9902 * 2 items for dir items
9903 * 1 item for updating parent inode item
9904 * 1 item for the inline extent item
9905 * 1 item for xattr if selinux is on
9907 trans = btrfs_start_transaction(root, 7);
9909 return PTR_ERR(trans);
9911 err = btrfs_get_free_objectid(root, &objectid);
9915 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9916 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9917 objectid, S_IFLNK|S_IRWXUGO, &index);
9918 if (IS_ERR(inode)) {
9919 err = PTR_ERR(inode);
9925 * If the active LSM wants to access the inode during
9926 * d_instantiate it needs these. Smack checks to see
9927 * if the filesystem supports xattrs by looking at the
9930 inode->i_fop = &btrfs_file_operations;
9931 inode->i_op = &btrfs_file_inode_operations;
9932 inode->i_mapping->a_ops = &btrfs_aops;
9934 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9938 path = btrfs_alloc_path();
9943 key.objectid = btrfs_ino(BTRFS_I(inode));
9945 key.type = BTRFS_EXTENT_DATA_KEY;
9946 datasize = btrfs_file_extent_calc_inline_size(name_len);
9947 err = btrfs_insert_empty_item(trans, root, path, &key,
9950 btrfs_free_path(path);
9953 leaf = path->nodes[0];
9954 ei = btrfs_item_ptr(leaf, path->slots[0],
9955 struct btrfs_file_extent_item);
9956 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9957 btrfs_set_file_extent_type(leaf, ei,
9958 BTRFS_FILE_EXTENT_INLINE);
9959 btrfs_set_file_extent_encryption(leaf, ei, 0);
9960 btrfs_set_file_extent_compression(leaf, ei, 0);
9961 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9962 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9964 ptr = btrfs_file_extent_inline_start(ei);
9965 write_extent_buffer(leaf, symname, ptr, name_len);
9966 btrfs_mark_buffer_dirty(leaf);
9967 btrfs_free_path(path);
9969 inode->i_op = &btrfs_symlink_inode_operations;
9970 inode_nohighmem(inode);
9971 inode_set_bytes(inode, name_len);
9972 btrfs_i_size_write(BTRFS_I(inode), name_len);
9973 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9975 * Last step, add directory indexes for our symlink inode. This is the
9976 * last step to avoid extra cleanup of these indexes if an error happens
9980 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9981 BTRFS_I(inode), 0, index);
9985 d_instantiate_new(dentry, inode);
9988 btrfs_end_transaction(trans);
9990 inode_dec_link_count(inode);
9991 discard_new_inode(inode);
9993 btrfs_btree_balance_dirty(fs_info);
9997 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9998 struct btrfs_trans_handle *trans_in,
9999 struct btrfs_inode *inode,
10000 struct btrfs_key *ins,
10003 struct btrfs_file_extent_item stack_fi;
10004 struct btrfs_replace_extent_info extent_info;
10005 struct btrfs_trans_handle *trans = trans_in;
10006 struct btrfs_path *path;
10007 u64 start = ins->objectid;
10008 u64 len = ins->offset;
10009 int qgroup_released;
10012 memset(&stack_fi, 0, sizeof(stack_fi));
10014 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
10015 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
10016 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
10017 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
10018 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
10019 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
10020 /* Encryption and other encoding is reserved and all 0 */
10022 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
10023 if (qgroup_released < 0)
10024 return ERR_PTR(qgroup_released);
10027 ret = insert_reserved_file_extent(trans, inode,
10028 file_offset, &stack_fi,
10029 true, qgroup_released);
10035 extent_info.disk_offset = start;
10036 extent_info.disk_len = len;
10037 extent_info.data_offset = 0;
10038 extent_info.data_len = len;
10039 extent_info.file_offset = file_offset;
10040 extent_info.extent_buf = (char *)&stack_fi;
10041 extent_info.is_new_extent = true;
10042 extent_info.qgroup_reserved = qgroup_released;
10043 extent_info.insertions = 0;
10045 path = btrfs_alloc_path();
10051 ret = btrfs_replace_file_extents(inode, path, file_offset,
10052 file_offset + len - 1, &extent_info,
10054 btrfs_free_path(path);
10061 * We have released qgroup data range at the beginning of the function,
10062 * and normally qgroup_released bytes will be freed when committing
10064 * But if we error out early, we have to free what we have released
10065 * or we leak qgroup data reservation.
10067 btrfs_qgroup_free_refroot(inode->root->fs_info,
10068 inode->root->root_key.objectid, qgroup_released,
10069 BTRFS_QGROUP_RSV_DATA);
10070 return ERR_PTR(ret);
10073 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10074 u64 start, u64 num_bytes, u64 min_size,
10075 loff_t actual_len, u64 *alloc_hint,
10076 struct btrfs_trans_handle *trans)
10078 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10079 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10080 struct extent_map *em;
10081 struct btrfs_root *root = BTRFS_I(inode)->root;
10082 struct btrfs_key ins;
10083 u64 cur_offset = start;
10084 u64 clear_offset = start;
10087 u64 last_alloc = (u64)-1;
10089 bool own_trans = true;
10090 u64 end = start + num_bytes - 1;
10094 while (num_bytes > 0) {
10095 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10096 cur_bytes = max(cur_bytes, min_size);
10098 * If we are severely fragmented we could end up with really
10099 * small allocations, so if the allocator is returning small
10100 * chunks lets make its job easier by only searching for those
10103 cur_bytes = min(cur_bytes, last_alloc);
10104 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10105 min_size, 0, *alloc_hint, &ins, 1, 0);
10110 * We've reserved this space, and thus converted it from
10111 * ->bytes_may_use to ->bytes_reserved. Any error that happens
10112 * from here on out we will only need to clear our reservation
10113 * for the remaining unreserved area, so advance our
10114 * clear_offset by our extent size.
10116 clear_offset += ins.offset;
10118 last_alloc = ins.offset;
10119 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
10122 * Now that we inserted the prealloc extent we can finally
10123 * decrement the number of reservations in the block group.
10124 * If we did it before, we could race with relocation and have
10125 * relocation miss the reserved extent, making it fail later.
10127 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10128 if (IS_ERR(trans)) {
10129 ret = PTR_ERR(trans);
10130 btrfs_free_reserved_extent(fs_info, ins.objectid,
10135 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10136 cur_offset + ins.offset -1, 0);
10138 em = alloc_extent_map();
10140 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10141 &BTRFS_I(inode)->runtime_flags);
10145 em->start = cur_offset;
10146 em->orig_start = cur_offset;
10147 em->len = ins.offset;
10148 em->block_start = ins.objectid;
10149 em->block_len = ins.offset;
10150 em->orig_block_len = ins.offset;
10151 em->ram_bytes = ins.offset;
10152 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10153 em->generation = trans->transid;
10156 write_lock(&em_tree->lock);
10157 ret = add_extent_mapping(em_tree, em, 1);
10158 write_unlock(&em_tree->lock);
10159 if (ret != -EEXIST)
10161 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10162 cur_offset + ins.offset - 1,
10165 free_extent_map(em);
10167 num_bytes -= ins.offset;
10168 cur_offset += ins.offset;
10169 *alloc_hint = ins.objectid + ins.offset;
10171 inode_inc_iversion(inode);
10172 inode->i_ctime = current_time(inode);
10173 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10174 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10175 (actual_len > inode->i_size) &&
10176 (cur_offset > inode->i_size)) {
10177 if (cur_offset > actual_len)
10178 i_size = actual_len;
10180 i_size = cur_offset;
10181 i_size_write(inode, i_size);
10182 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
10185 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10188 btrfs_abort_transaction(trans, ret);
10190 btrfs_end_transaction(trans);
10195 btrfs_end_transaction(trans);
10199 if (clear_offset < end)
10200 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
10201 end - clear_offset + 1);
10205 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10206 u64 start, u64 num_bytes, u64 min_size,
10207 loff_t actual_len, u64 *alloc_hint)
10209 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10210 min_size, actual_len, alloc_hint,
10214 int btrfs_prealloc_file_range_trans(struct inode *inode,
10215 struct btrfs_trans_handle *trans, int mode,
10216 u64 start, u64 num_bytes, u64 min_size,
10217 loff_t actual_len, u64 *alloc_hint)
10219 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10220 min_size, actual_len, alloc_hint, trans);
10223 static int btrfs_set_page_dirty(struct page *page)
10225 return __set_page_dirty_nobuffers(page);
10228 static int btrfs_permission(struct user_namespace *mnt_userns,
10229 struct inode *inode, int mask)
10231 struct btrfs_root *root = BTRFS_I(inode)->root;
10232 umode_t mode = inode->i_mode;
10234 if (mask & MAY_WRITE &&
10235 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10236 if (btrfs_root_readonly(root))
10238 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10241 return generic_permission(&init_user_ns, inode, mask);
10244 static int btrfs_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
10245 struct dentry *dentry, umode_t mode)
10247 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10248 struct btrfs_trans_handle *trans;
10249 struct btrfs_root *root = BTRFS_I(dir)->root;
10250 struct inode *inode = NULL;
10256 * 5 units required for adding orphan entry
10258 trans = btrfs_start_transaction(root, 5);
10260 return PTR_ERR(trans);
10262 ret = btrfs_get_free_objectid(root, &objectid);
10266 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10267 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10268 if (IS_ERR(inode)) {
10269 ret = PTR_ERR(inode);
10274 inode->i_fop = &btrfs_file_operations;
10275 inode->i_op = &btrfs_file_inode_operations;
10277 inode->i_mapping->a_ops = &btrfs_aops;
10279 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10283 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
10286 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10291 * We set number of links to 0 in btrfs_new_inode(), and here we set
10292 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10295 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10297 set_nlink(inode, 1);
10298 d_tmpfile(dentry, inode);
10299 unlock_new_inode(inode);
10300 mark_inode_dirty(inode);
10302 btrfs_end_transaction(trans);
10304 discard_new_inode(inode);
10305 btrfs_btree_balance_dirty(fs_info);
10309 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10311 struct inode *inode = tree->private_data;
10312 unsigned long index = start >> PAGE_SHIFT;
10313 unsigned long end_index = end >> PAGE_SHIFT;
10316 while (index <= end_index) {
10317 page = find_get_page(inode->i_mapping, index);
10318 ASSERT(page); /* Pages should be in the extent_io_tree */
10319 set_page_writeback(page);
10327 * Add an entry indicating a block group or device which is pinned by a
10328 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10329 * negative errno on failure.
10331 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10332 bool is_block_group)
10334 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10335 struct btrfs_swapfile_pin *sp, *entry;
10336 struct rb_node **p;
10337 struct rb_node *parent = NULL;
10339 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10344 sp->is_block_group = is_block_group;
10345 sp->bg_extent_count = 1;
10347 spin_lock(&fs_info->swapfile_pins_lock);
10348 p = &fs_info->swapfile_pins.rb_node;
10351 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10352 if (sp->ptr < entry->ptr ||
10353 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10354 p = &(*p)->rb_left;
10355 } else if (sp->ptr > entry->ptr ||
10356 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10357 p = &(*p)->rb_right;
10359 if (is_block_group)
10360 entry->bg_extent_count++;
10361 spin_unlock(&fs_info->swapfile_pins_lock);
10366 rb_link_node(&sp->node, parent, p);
10367 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10368 spin_unlock(&fs_info->swapfile_pins_lock);
10372 /* Free all of the entries pinned by this swapfile. */
10373 static void btrfs_free_swapfile_pins(struct inode *inode)
10375 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10376 struct btrfs_swapfile_pin *sp;
10377 struct rb_node *node, *next;
10379 spin_lock(&fs_info->swapfile_pins_lock);
10380 node = rb_first(&fs_info->swapfile_pins);
10382 next = rb_next(node);
10383 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10384 if (sp->inode == inode) {
10385 rb_erase(&sp->node, &fs_info->swapfile_pins);
10386 if (sp->is_block_group) {
10387 btrfs_dec_block_group_swap_extents(sp->ptr,
10388 sp->bg_extent_count);
10389 btrfs_put_block_group(sp->ptr);
10395 spin_unlock(&fs_info->swapfile_pins_lock);
10398 struct btrfs_swap_info {
10404 unsigned long nr_pages;
10408 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10409 struct btrfs_swap_info *bsi)
10411 unsigned long nr_pages;
10412 u64 first_ppage, first_ppage_reported, next_ppage;
10415 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10416 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10417 PAGE_SIZE) >> PAGE_SHIFT;
10419 if (first_ppage >= next_ppage)
10421 nr_pages = next_ppage - first_ppage;
10423 first_ppage_reported = first_ppage;
10424 if (bsi->start == 0)
10425 first_ppage_reported++;
10426 if (bsi->lowest_ppage > first_ppage_reported)
10427 bsi->lowest_ppage = first_ppage_reported;
10428 if (bsi->highest_ppage < (next_ppage - 1))
10429 bsi->highest_ppage = next_ppage - 1;
10431 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10434 bsi->nr_extents += ret;
10435 bsi->nr_pages += nr_pages;
10439 static void btrfs_swap_deactivate(struct file *file)
10441 struct inode *inode = file_inode(file);
10443 btrfs_free_swapfile_pins(inode);
10444 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10447 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10450 struct inode *inode = file_inode(file);
10451 struct btrfs_root *root = BTRFS_I(inode)->root;
10452 struct btrfs_fs_info *fs_info = root->fs_info;
10453 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10454 struct extent_state *cached_state = NULL;
10455 struct extent_map *em = NULL;
10456 struct btrfs_device *device = NULL;
10457 struct btrfs_swap_info bsi = {
10458 .lowest_ppage = (sector_t)-1ULL,
10465 * If the swap file was just created, make sure delalloc is done. If the
10466 * file changes again after this, the user is doing something stupid and
10467 * we don't really care.
10469 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10474 * The inode is locked, so these flags won't change after we check them.
10476 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10477 btrfs_warn(fs_info, "swapfile must not be compressed");
10480 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10481 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10484 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10485 btrfs_warn(fs_info, "swapfile must not be checksummed");
10490 * Balance or device remove/replace/resize can move stuff around from
10491 * under us. The exclop protection makes sure they aren't running/won't
10492 * run concurrently while we are mapping the swap extents, and
10493 * fs_info->swapfile_pins prevents them from running while the swap
10494 * file is active and moving the extents. Note that this also prevents
10495 * a concurrent device add which isn't actually necessary, but it's not
10496 * really worth the trouble to allow it.
10498 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10499 btrfs_warn(fs_info,
10500 "cannot activate swapfile while exclusive operation is running");
10505 * Prevent snapshot creation while we are activating the swap file.
10506 * We do not want to race with snapshot creation. If snapshot creation
10507 * already started before we bumped nr_swapfiles from 0 to 1 and
10508 * completes before the first write into the swap file after it is
10509 * activated, than that write would fallback to COW.
10511 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10512 btrfs_exclop_finish(fs_info);
10513 btrfs_warn(fs_info,
10514 "cannot activate swapfile because snapshot creation is in progress");
10518 * Snapshots can create extents which require COW even if NODATACOW is
10519 * set. We use this counter to prevent snapshots. We must increment it
10520 * before walking the extents because we don't want a concurrent
10521 * snapshot to run after we've already checked the extents.
10523 atomic_inc(&root->nr_swapfiles);
10525 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10527 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10529 while (start < isize) {
10530 u64 logical_block_start, physical_block_start;
10531 struct btrfs_block_group *bg;
10532 u64 len = isize - start;
10534 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10540 if (em->block_start == EXTENT_MAP_HOLE) {
10541 btrfs_warn(fs_info, "swapfile must not have holes");
10545 if (em->block_start == EXTENT_MAP_INLINE) {
10547 * It's unlikely we'll ever actually find ourselves
10548 * here, as a file small enough to fit inline won't be
10549 * big enough to store more than the swap header, but in
10550 * case something changes in the future, let's catch it
10551 * here rather than later.
10553 btrfs_warn(fs_info, "swapfile must not be inline");
10557 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10558 btrfs_warn(fs_info, "swapfile must not be compressed");
10563 logical_block_start = em->block_start + (start - em->start);
10564 len = min(len, em->len - (start - em->start));
10565 free_extent_map(em);
10568 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10574 btrfs_warn(fs_info,
10575 "swapfile must not be copy-on-write");
10580 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10586 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10587 btrfs_warn(fs_info,
10588 "swapfile must have single data profile");
10593 if (device == NULL) {
10594 device = em->map_lookup->stripes[0].dev;
10595 ret = btrfs_add_swapfile_pin(inode, device, false);
10600 } else if (device != em->map_lookup->stripes[0].dev) {
10601 btrfs_warn(fs_info, "swapfile must be on one device");
10606 physical_block_start = (em->map_lookup->stripes[0].physical +
10607 (logical_block_start - em->start));
10608 len = min(len, em->len - (logical_block_start - em->start));
10609 free_extent_map(em);
10612 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10614 btrfs_warn(fs_info,
10615 "could not find block group containing swapfile");
10620 if (!btrfs_inc_block_group_swap_extents(bg)) {
10621 btrfs_warn(fs_info,
10622 "block group for swapfile at %llu is read-only%s",
10624 atomic_read(&fs_info->scrubs_running) ?
10625 " (scrub running)" : "");
10626 btrfs_put_block_group(bg);
10631 ret = btrfs_add_swapfile_pin(inode, bg, true);
10633 btrfs_put_block_group(bg);
10640 if (bsi.block_len &&
10641 bsi.block_start + bsi.block_len == physical_block_start) {
10642 bsi.block_len += len;
10644 if (bsi.block_len) {
10645 ret = btrfs_add_swap_extent(sis, &bsi);
10650 bsi.block_start = physical_block_start;
10651 bsi.block_len = len;
10658 ret = btrfs_add_swap_extent(sis, &bsi);
10661 if (!IS_ERR_OR_NULL(em))
10662 free_extent_map(em);
10664 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10667 btrfs_swap_deactivate(file);
10669 btrfs_drew_write_unlock(&root->snapshot_lock);
10671 btrfs_exclop_finish(fs_info);
10677 sis->bdev = device->bdev;
10678 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10679 sis->max = bsi.nr_pages;
10680 sis->pages = bsi.nr_pages - 1;
10681 sis->highest_bit = bsi.nr_pages - 1;
10682 return bsi.nr_extents;
10685 static void btrfs_swap_deactivate(struct file *file)
10689 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10692 return -EOPNOTSUPP;
10697 * Update the number of bytes used in the VFS' inode. When we replace extents in
10698 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10699 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10700 * always get a correct value.
10702 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10703 const u64 add_bytes,
10704 const u64 del_bytes)
10706 if (add_bytes == del_bytes)
10709 spin_lock(&inode->lock);
10711 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10713 inode_add_bytes(&inode->vfs_inode, add_bytes);
10714 spin_unlock(&inode->lock);
10717 static const struct inode_operations btrfs_dir_inode_operations = {
10718 .getattr = btrfs_getattr,
10719 .lookup = btrfs_lookup,
10720 .create = btrfs_create,
10721 .unlink = btrfs_unlink,
10722 .link = btrfs_link,
10723 .mkdir = btrfs_mkdir,
10724 .rmdir = btrfs_rmdir,
10725 .rename = btrfs_rename2,
10726 .symlink = btrfs_symlink,
10727 .setattr = btrfs_setattr,
10728 .mknod = btrfs_mknod,
10729 .listxattr = btrfs_listxattr,
10730 .permission = btrfs_permission,
10731 .get_acl = btrfs_get_acl,
10732 .set_acl = btrfs_set_acl,
10733 .update_time = btrfs_update_time,
10734 .tmpfile = btrfs_tmpfile,
10735 .fileattr_get = btrfs_fileattr_get,
10736 .fileattr_set = btrfs_fileattr_set,
10739 static const struct file_operations btrfs_dir_file_operations = {
10740 .llseek = generic_file_llseek,
10741 .read = generic_read_dir,
10742 .iterate_shared = btrfs_real_readdir,
10743 .open = btrfs_opendir,
10744 .unlocked_ioctl = btrfs_ioctl,
10745 #ifdef CONFIG_COMPAT
10746 .compat_ioctl = btrfs_compat_ioctl,
10748 .release = btrfs_release_file,
10749 .fsync = btrfs_sync_file,
10753 * btrfs doesn't support the bmap operation because swapfiles
10754 * use bmap to make a mapping of extents in the file. They assume
10755 * these extents won't change over the life of the file and they
10756 * use the bmap result to do IO directly to the drive.
10758 * the btrfs bmap call would return logical addresses that aren't
10759 * suitable for IO and they also will change frequently as COW
10760 * operations happen. So, swapfile + btrfs == corruption.
10762 * For now we're avoiding this by dropping bmap.
10764 static const struct address_space_operations btrfs_aops = {
10765 .readpage = btrfs_readpage,
10766 .writepage = btrfs_writepage,
10767 .writepages = btrfs_writepages,
10768 .readahead = btrfs_readahead,
10769 .direct_IO = noop_direct_IO,
10770 .invalidatepage = btrfs_invalidatepage,
10771 .releasepage = btrfs_releasepage,
10772 #ifdef CONFIG_MIGRATION
10773 .migratepage = btrfs_migratepage,
10775 .set_page_dirty = btrfs_set_page_dirty,
10776 .error_remove_page = generic_error_remove_page,
10777 .swap_activate = btrfs_swap_activate,
10778 .swap_deactivate = btrfs_swap_deactivate,
10781 static const struct inode_operations btrfs_file_inode_operations = {
10782 .getattr = btrfs_getattr,
10783 .setattr = btrfs_setattr,
10784 .listxattr = btrfs_listxattr,
10785 .permission = btrfs_permission,
10786 .fiemap = btrfs_fiemap,
10787 .get_acl = btrfs_get_acl,
10788 .set_acl = btrfs_set_acl,
10789 .update_time = btrfs_update_time,
10790 .fileattr_get = btrfs_fileattr_get,
10791 .fileattr_set = btrfs_fileattr_set,
10793 static const struct inode_operations btrfs_special_inode_operations = {
10794 .getattr = btrfs_getattr,
10795 .setattr = btrfs_setattr,
10796 .permission = btrfs_permission,
10797 .listxattr = btrfs_listxattr,
10798 .get_acl = btrfs_get_acl,
10799 .set_acl = btrfs_set_acl,
10800 .update_time = btrfs_update_time,
10802 static const struct inode_operations btrfs_symlink_inode_operations = {
10803 .get_link = page_get_link,
10804 .getattr = btrfs_getattr,
10805 .setattr = btrfs_setattr,
10806 .permission = btrfs_permission,
10807 .listxattr = btrfs_listxattr,
10808 .update_time = btrfs_update_time,
10811 const struct dentry_operations btrfs_dentry_operations = {
10812 .d_delete = btrfs_dentry_delete,