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"
48 #include "inode-map.h"
51 #include "delalloc-space.h"
52 #include "block-group.h"
53 #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;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_special_inode_operations;
71 static const struct inode_operations btrfs_file_inode_operations;
72 static const struct address_space_operations btrfs_aops;
73 static const struct file_operations btrfs_dir_file_operations;
74 static const struct extent_io_ops btrfs_extent_io_ops;
76 static struct kmem_cache *btrfs_inode_cachep;
77 struct kmem_cache *btrfs_trans_handle_cachep;
78 struct kmem_cache *btrfs_path_cachep;
79 struct kmem_cache *btrfs_free_space_cachep;
80 struct kmem_cache *btrfs_free_space_bitmap_cachep;
82 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
83 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
84 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
85 static noinline int cow_file_range(struct btrfs_inode *inode,
86 struct page *locked_page,
87 u64 start, u64 end, int *page_started,
88 unsigned long *nr_written, int unlock);
89 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
90 u64 len, u64 orig_start, u64 block_start,
91 u64 block_len, u64 orig_block_len,
92 u64 ram_bytes, int compress_type,
95 static void __endio_write_update_ordered(struct btrfs_inode *inode,
96 const u64 offset, const u64 bytes,
100 * Cleanup all submitted ordered extents in specified range to handle errors
101 * from the btrfs_run_delalloc_range() callback.
103 * NOTE: caller must ensure that when an error happens, it can not call
104 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
105 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
106 * to be released, which we want to happen only when finishing the ordered
107 * extent (btrfs_finish_ordered_io()).
109 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
110 struct page *locked_page,
111 u64 offset, u64 bytes)
113 unsigned long index = offset >> PAGE_SHIFT;
114 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
115 u64 page_start = page_offset(locked_page);
116 u64 page_end = page_start + PAGE_SIZE - 1;
120 while (index <= end_index) {
121 page = find_get_page(inode->vfs_inode.i_mapping, index);
125 ClearPagePrivate2(page);
130 * In case this page belongs to the delalloc range being instantiated
131 * then skip it, since the first page of a range is going to be
132 * properly cleaned up by the caller of run_delalloc_range
134 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
139 return __endio_write_update_ordered(inode, offset, bytes, false);
142 static int btrfs_dirty_inode(struct inode *inode);
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
147 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152 struct inode *inode, struct inode *dir,
153 const struct qstr *qstr)
157 err = btrfs_init_acl(trans, inode, dir);
159 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
164 * this does all the hard work for inserting an inline extent into
165 * the btree. The caller should have done a btrfs_drop_extents so that
166 * no overlapping inline items exist in the btree
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169 struct btrfs_path *path, int extent_inserted,
170 struct btrfs_root *root, struct inode *inode,
171 u64 start, size_t size, size_t compressed_size,
173 struct page **compressed_pages)
175 struct extent_buffer *leaf;
176 struct page *page = NULL;
179 struct btrfs_file_extent_item *ei;
181 size_t cur_size = size;
182 unsigned long offset;
184 ASSERT((compressed_size > 0 && compressed_pages) ||
185 (compressed_size == 0 && !compressed_pages));
187 if (compressed_size && compressed_pages)
188 cur_size = compressed_size;
190 inode_add_bytes(inode, size);
192 if (!extent_inserted) {
193 struct btrfs_key key;
196 key.objectid = btrfs_ino(BTRFS_I(inode));
198 key.type = BTRFS_EXTENT_DATA_KEY;
200 datasize = btrfs_file_extent_calc_inline_size(cur_size);
201 path->leave_spinning = 1;
202 ret = btrfs_insert_empty_item(trans, root, path, &key,
207 leaf = path->nodes[0];
208 ei = btrfs_item_ptr(leaf, path->slots[0],
209 struct btrfs_file_extent_item);
210 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
211 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
212 btrfs_set_file_extent_encryption(leaf, ei, 0);
213 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
214 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
215 ptr = btrfs_file_extent_inline_start(ei);
217 if (compress_type != BTRFS_COMPRESS_NONE) {
220 while (compressed_size > 0) {
221 cpage = compressed_pages[i];
222 cur_size = min_t(unsigned long, compressed_size,
225 kaddr = kmap_atomic(cpage);
226 write_extent_buffer(leaf, kaddr, ptr, cur_size);
227 kunmap_atomic(kaddr);
231 compressed_size -= cur_size;
233 btrfs_set_file_extent_compression(leaf, ei,
236 page = find_get_page(inode->i_mapping,
237 start >> PAGE_SHIFT);
238 btrfs_set_file_extent_compression(leaf, ei, 0);
239 kaddr = kmap_atomic(page);
240 offset = offset_in_page(start);
241 write_extent_buffer(leaf, kaddr + offset, ptr, size);
242 kunmap_atomic(kaddr);
245 btrfs_mark_buffer_dirty(leaf);
246 btrfs_release_path(path);
249 * We align size to sectorsize for inline extents just for simplicity
252 size = ALIGN(size, root->fs_info->sectorsize);
253 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
258 * we're an inline extent, so nobody can
259 * extend the file past i_size without locking
260 * a page we already have locked.
262 * We must do any isize and inode updates
263 * before we unlock the pages. Otherwise we
264 * could end up racing with unlink.
266 BTRFS_I(inode)->disk_i_size = inode->i_size;
267 ret = btrfs_update_inode(trans, root, inode);
275 * conditionally insert an inline extent into the file. This
276 * does the checks required to make sure the data is small enough
277 * to fit as an inline extent.
279 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
280 u64 end, size_t compressed_size,
282 struct page **compressed_pages)
284 struct btrfs_root *root = inode->root;
285 struct btrfs_fs_info *fs_info = root->fs_info;
286 struct btrfs_trans_handle *trans;
287 u64 isize = i_size_read(&inode->vfs_inode);
288 u64 actual_end = min(end + 1, isize);
289 u64 inline_len = actual_end - start;
290 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
291 u64 data_len = inline_len;
293 struct btrfs_path *path;
294 int extent_inserted = 0;
295 u32 extent_item_size;
298 data_len = compressed_size;
301 actual_end > fs_info->sectorsize ||
302 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
304 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
306 data_len > fs_info->max_inline) {
310 path = btrfs_alloc_path();
314 trans = btrfs_join_transaction(root);
316 btrfs_free_path(path);
317 return PTR_ERR(trans);
319 trans->block_rsv = &inode->block_rsv;
321 if (compressed_size && compressed_pages)
322 extent_item_size = btrfs_file_extent_calc_inline_size(
325 extent_item_size = btrfs_file_extent_calc_inline_size(
328 ret = __btrfs_drop_extents(trans, root, inode, path, start, aligned_end,
329 NULL, 1, 1, extent_item_size,
332 btrfs_abort_transaction(trans, ret);
336 if (isize > actual_end)
337 inline_len = min_t(u64, isize, actual_end);
338 ret = insert_inline_extent(trans, path, extent_inserted,
339 root, &inode->vfs_inode, start,
340 inline_len, compressed_size,
341 compress_type, compressed_pages);
342 if (ret && ret != -ENOSPC) {
343 btrfs_abort_transaction(trans, ret);
345 } else if (ret == -ENOSPC) {
350 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
351 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
354 * Don't forget to free the reserved space, as for inlined extent
355 * it won't count as data extent, free them directly here.
356 * And at reserve time, it's always aligned to page size, so
357 * just free one page here.
359 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
360 btrfs_free_path(path);
361 btrfs_end_transaction(trans);
365 struct async_extent {
370 unsigned long nr_pages;
372 struct list_head list;
377 struct page *locked_page;
380 unsigned int write_flags;
381 struct list_head extents;
382 struct cgroup_subsys_state *blkcg_css;
383 struct btrfs_work work;
388 /* Number of chunks in flight; must be first in the structure */
390 struct async_chunk chunks[];
393 static noinline int add_async_extent(struct async_chunk *cow,
394 u64 start, u64 ram_size,
397 unsigned long nr_pages,
400 struct async_extent *async_extent;
402 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
403 BUG_ON(!async_extent); /* -ENOMEM */
404 async_extent->start = start;
405 async_extent->ram_size = ram_size;
406 async_extent->compressed_size = compressed_size;
407 async_extent->pages = pages;
408 async_extent->nr_pages = nr_pages;
409 async_extent->compress_type = compress_type;
410 list_add_tail(&async_extent->list, &cow->extents);
415 * Check if the inode has flags compatible with compression
417 static inline bool inode_can_compress(struct btrfs_inode *inode)
419 if (inode->flags & BTRFS_INODE_NODATACOW ||
420 inode->flags & BTRFS_INODE_NODATASUM)
426 * Check if the inode needs to be submitted to compression, based on mount
427 * options, defragmentation, properties or heuristics.
429 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
432 struct btrfs_fs_info *fs_info = inode->root->fs_info;
434 if (!inode_can_compress(inode)) {
435 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
436 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
441 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
444 if (inode->defrag_compress)
446 /* bad compression ratios */
447 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
449 if (btrfs_test_opt(fs_info, COMPRESS) ||
450 inode->flags & BTRFS_INODE_COMPRESS ||
451 inode->prop_compress)
452 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
456 static inline void inode_should_defrag(struct btrfs_inode *inode,
457 u64 start, u64 end, u64 num_bytes, u64 small_write)
459 /* If this is a small write inside eof, kick off a defrag */
460 if (num_bytes < small_write &&
461 (start > 0 || end + 1 < inode->disk_i_size))
462 btrfs_add_inode_defrag(NULL, inode);
466 * we create compressed extents in two phases. The first
467 * phase compresses a range of pages that have already been
468 * locked (both pages and state bits are locked).
470 * This is done inside an ordered work queue, and the compression
471 * is spread across many cpus. The actual IO submission is step
472 * two, and the ordered work queue takes care of making sure that
473 * happens in the same order things were put onto the queue by
474 * writepages and friends.
476 * If this code finds it can't get good compression, it puts an
477 * entry onto the work queue to write the uncompressed bytes. This
478 * makes sure that both compressed inodes and uncompressed inodes
479 * are written in the same order that the flusher thread sent them
482 static noinline int compress_file_range(struct async_chunk *async_chunk)
484 struct inode *inode = async_chunk->inode;
485 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
486 u64 blocksize = fs_info->sectorsize;
487 u64 start = async_chunk->start;
488 u64 end = async_chunk->end;
492 struct page **pages = NULL;
493 unsigned long nr_pages;
494 unsigned long total_compressed = 0;
495 unsigned long total_in = 0;
498 int compress_type = fs_info->compress_type;
499 int compressed_extents = 0;
502 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
506 * We need to save i_size before now because it could change in between
507 * us evaluating the size and assigning it. This is because we lock and
508 * unlock the page in truncate and fallocate, and then modify the i_size
511 * The barriers are to emulate READ_ONCE, remove that once i_size_read
515 i_size = i_size_read(inode);
517 actual_end = min_t(u64, i_size, end + 1);
520 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
521 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
522 nr_pages = min_t(unsigned long, nr_pages,
523 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
526 * we don't want to send crud past the end of i_size through
527 * compression, that's just a waste of CPU time. So, if the
528 * end of the file is before the start of our current
529 * requested range of bytes, we bail out to the uncompressed
530 * cleanup code that can deal with all of this.
532 * It isn't really the fastest way to fix things, but this is a
533 * very uncommon corner.
535 if (actual_end <= start)
536 goto cleanup_and_bail_uncompressed;
538 total_compressed = actual_end - start;
541 * skip compression for a small file range(<=blocksize) that
542 * isn't an inline extent, since it doesn't save disk space at all.
544 if (total_compressed <= blocksize &&
545 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
546 goto cleanup_and_bail_uncompressed;
548 total_compressed = min_t(unsigned long, total_compressed,
549 BTRFS_MAX_UNCOMPRESSED);
554 * we do compression for mount -o compress and when the
555 * inode has not been flagged as nocompress. This flag can
556 * change at any time if we discover bad compression ratios.
558 if (inode_need_compress(BTRFS_I(inode), start, end)) {
560 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
562 /* just bail out to the uncompressed code */
567 if (BTRFS_I(inode)->defrag_compress)
568 compress_type = BTRFS_I(inode)->defrag_compress;
569 else if (BTRFS_I(inode)->prop_compress)
570 compress_type = BTRFS_I(inode)->prop_compress;
573 * we need to call clear_page_dirty_for_io on each
574 * page in the range. Otherwise applications with the file
575 * mmap'd can wander in and change the page contents while
576 * we are compressing them.
578 * If the compression fails for any reason, we set the pages
579 * dirty again later on.
581 * Note that the remaining part is redirtied, the start pointer
582 * has moved, the end is the original one.
585 extent_range_clear_dirty_for_io(inode, start, end);
589 /* Compression level is applied here and only here */
590 ret = btrfs_compress_pages(
591 compress_type | (fs_info->compress_level << 4),
592 inode->i_mapping, start,
599 unsigned long offset = offset_in_page(total_compressed);
600 struct page *page = pages[nr_pages - 1];
603 /* zero the tail end of the last page, we might be
604 * sending it down to disk
607 kaddr = kmap_atomic(page);
608 memset(kaddr + offset, 0,
610 kunmap_atomic(kaddr);
617 /* lets try to make an inline extent */
618 if (ret || total_in < actual_end) {
619 /* we didn't compress the entire range, try
620 * to make an uncompressed inline extent.
622 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
623 0, BTRFS_COMPRESS_NONE,
626 /* try making a compressed inline extent */
627 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
629 compress_type, pages);
632 unsigned long clear_flags = EXTENT_DELALLOC |
633 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
634 EXTENT_DO_ACCOUNTING;
635 unsigned long page_error_op;
637 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
640 * inline extent creation worked or returned error,
641 * we don't need to create any more async work items.
642 * Unlock and free up our temp pages.
644 * We use DO_ACCOUNTING here because we need the
645 * delalloc_release_metadata to be done _after_ we drop
646 * our outstanding extent for clearing delalloc for this
649 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
659 * Ensure we only free the compressed pages if we have
660 * them allocated, as we can still reach here with
661 * inode_need_compress() == false.
664 for (i = 0; i < nr_pages; i++) {
665 WARN_ON(pages[i]->mapping);
676 * we aren't doing an inline extent round the compressed size
677 * up to a block size boundary so the allocator does sane
680 total_compressed = ALIGN(total_compressed, blocksize);
683 * one last check to make sure the compression is really a
684 * win, compare the page count read with the blocks on disk,
685 * compression must free at least one sector size
687 total_in = ALIGN(total_in, PAGE_SIZE);
688 if (total_compressed + blocksize <= total_in) {
689 compressed_extents++;
692 * The async work queues will take care of doing actual
693 * allocation on disk for these compressed pages, and
694 * will submit them to the elevator.
696 add_async_extent(async_chunk, start, total_in,
697 total_compressed, pages, nr_pages,
700 if (start + total_in < end) {
706 return compressed_extents;
711 * the compression code ran but failed to make things smaller,
712 * free any pages it allocated and our page pointer array
714 for (i = 0; i < nr_pages; i++) {
715 WARN_ON(pages[i]->mapping);
720 total_compressed = 0;
723 /* flag the file so we don't compress in the future */
724 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
725 !(BTRFS_I(inode)->prop_compress)) {
726 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
729 cleanup_and_bail_uncompressed:
731 * No compression, but we still need to write the pages in the file
732 * we've been given so far. redirty the locked page if it corresponds
733 * to our extent and set things up for the async work queue to run
734 * cow_file_range to do the normal delalloc dance.
736 if (async_chunk->locked_page &&
737 (page_offset(async_chunk->locked_page) >= start &&
738 page_offset(async_chunk->locked_page)) <= end) {
739 __set_page_dirty_nobuffers(async_chunk->locked_page);
740 /* unlocked later on in the async handlers */
744 extent_range_redirty_for_io(inode, start, end);
745 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
746 BTRFS_COMPRESS_NONE);
747 compressed_extents++;
749 return compressed_extents;
752 static void free_async_extent_pages(struct async_extent *async_extent)
756 if (!async_extent->pages)
759 for (i = 0; i < async_extent->nr_pages; i++) {
760 WARN_ON(async_extent->pages[i]->mapping);
761 put_page(async_extent->pages[i]);
763 kfree(async_extent->pages);
764 async_extent->nr_pages = 0;
765 async_extent->pages = NULL;
769 * phase two of compressed writeback. This is the ordered portion
770 * of the code, which only gets called in the order the work was
771 * queued. We walk all the async extents created by compress_file_range
772 * and send them down to the disk.
774 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
776 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
777 struct btrfs_fs_info *fs_info = inode->root->fs_info;
778 struct async_extent *async_extent;
780 struct btrfs_key ins;
781 struct extent_map *em;
782 struct btrfs_root *root = inode->root;
783 struct extent_io_tree *io_tree = &inode->io_tree;
787 while (!list_empty(&async_chunk->extents)) {
788 async_extent = list_entry(async_chunk->extents.next,
789 struct async_extent, list);
790 list_del(&async_extent->list);
793 lock_extent(io_tree, async_extent->start,
794 async_extent->start + async_extent->ram_size - 1);
795 /* did the compression code fall back to uncompressed IO? */
796 if (!async_extent->pages) {
797 int page_started = 0;
798 unsigned long nr_written = 0;
800 /* allocate blocks */
801 ret = cow_file_range(inode, async_chunk->locked_page,
803 async_extent->start +
804 async_extent->ram_size - 1,
805 &page_started, &nr_written, 0);
810 * if page_started, cow_file_range inserted an
811 * inline extent and took care of all the unlocking
812 * and IO for us. Otherwise, we need to submit
813 * all those pages down to the drive.
815 if (!page_started && !ret)
816 extent_write_locked_range(&inode->vfs_inode,
818 async_extent->start +
819 async_extent->ram_size - 1,
821 else if (ret && async_chunk->locked_page)
822 unlock_page(async_chunk->locked_page);
828 ret = btrfs_reserve_extent(root, async_extent->ram_size,
829 async_extent->compressed_size,
830 async_extent->compressed_size,
831 0, alloc_hint, &ins, 1, 1);
833 free_async_extent_pages(async_extent);
835 if (ret == -ENOSPC) {
836 unlock_extent(io_tree, async_extent->start,
837 async_extent->start +
838 async_extent->ram_size - 1);
841 * we need to redirty the pages if we decide to
842 * fallback to uncompressed IO, otherwise we
843 * will not submit these pages down to lower
846 extent_range_redirty_for_io(&inode->vfs_inode,
848 async_extent->start +
849 async_extent->ram_size - 1);
856 * here we're doing allocation and writeback of the
859 em = create_io_em(inode, async_extent->start,
860 async_extent->ram_size, /* len */
861 async_extent->start, /* orig_start */
862 ins.objectid, /* block_start */
863 ins.offset, /* block_len */
864 ins.offset, /* orig_block_len */
865 async_extent->ram_size, /* ram_bytes */
866 async_extent->compress_type,
867 BTRFS_ORDERED_COMPRESSED);
869 /* ret value is not necessary due to void function */
870 goto out_free_reserve;
873 ret = btrfs_add_ordered_extent_compress(inode,
876 async_extent->ram_size,
878 BTRFS_ORDERED_COMPRESSED,
879 async_extent->compress_type);
881 btrfs_drop_extent_cache(inode, async_extent->start,
882 async_extent->start +
883 async_extent->ram_size - 1, 0);
884 goto out_free_reserve;
886 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
889 * clear dirty, set writeback and unlock the pages.
891 extent_clear_unlock_delalloc(inode, async_extent->start,
892 async_extent->start +
893 async_extent->ram_size - 1,
894 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
895 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
897 if (btrfs_submit_compressed_write(inode, async_extent->start,
898 async_extent->ram_size,
900 ins.offset, async_extent->pages,
901 async_extent->nr_pages,
902 async_chunk->write_flags,
903 async_chunk->blkcg_css)) {
904 struct page *p = async_extent->pages[0];
905 const u64 start = async_extent->start;
906 const u64 end = start + async_extent->ram_size - 1;
908 p->mapping = inode->vfs_inode.i_mapping;
909 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
912 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
915 free_async_extent_pages(async_extent);
917 alloc_hint = ins.objectid + ins.offset;
923 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
924 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
926 extent_clear_unlock_delalloc(inode, async_extent->start,
927 async_extent->start +
928 async_extent->ram_size - 1,
929 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
930 EXTENT_DELALLOC_NEW |
931 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
932 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
933 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
935 free_async_extent_pages(async_extent);
940 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
943 struct extent_map_tree *em_tree = &inode->extent_tree;
944 struct extent_map *em;
947 read_lock(&em_tree->lock);
948 em = search_extent_mapping(em_tree, start, num_bytes);
951 * if block start isn't an actual block number then find the
952 * first block in this inode and use that as a hint. If that
953 * block is also bogus then just don't worry about it.
955 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
957 em = search_extent_mapping(em_tree, 0, 0);
958 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
959 alloc_hint = em->block_start;
963 alloc_hint = em->block_start;
967 read_unlock(&em_tree->lock);
973 * when extent_io.c finds a delayed allocation range in the file,
974 * the call backs end up in this code. The basic idea is to
975 * allocate extents on disk for the range, and create ordered data structs
976 * in ram to track those extents.
978 * locked_page is the page that writepage had locked already. We use
979 * it to make sure we don't do extra locks or unlocks.
981 * *page_started is set to one if we unlock locked_page and do everything
982 * required to start IO on it. It may be clean and already done with
985 static noinline int cow_file_range(struct btrfs_inode *inode,
986 struct page *locked_page,
987 u64 start, u64 end, int *page_started,
988 unsigned long *nr_written, int unlock)
990 struct btrfs_root *root = inode->root;
991 struct btrfs_fs_info *fs_info = root->fs_info;
994 unsigned long ram_size;
995 u64 cur_alloc_size = 0;
997 u64 blocksize = fs_info->sectorsize;
998 struct btrfs_key ins;
999 struct extent_map *em;
1000 unsigned clear_bits;
1001 unsigned long page_ops;
1002 bool extent_reserved = false;
1005 if (btrfs_is_free_space_inode(inode)) {
1011 num_bytes = ALIGN(end - start + 1, blocksize);
1012 num_bytes = max(blocksize, num_bytes);
1013 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1015 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1018 /* lets try to make an inline extent */
1019 ret = cow_file_range_inline(inode, start, end, 0,
1020 BTRFS_COMPRESS_NONE, NULL);
1023 * We use DO_ACCOUNTING here because we need the
1024 * delalloc_release_metadata to be run _after_ we drop
1025 * our outstanding extent for clearing delalloc for this
1028 extent_clear_unlock_delalloc(inode, start, end, NULL,
1029 EXTENT_LOCKED | EXTENT_DELALLOC |
1030 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1031 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1032 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1033 PAGE_END_WRITEBACK);
1034 *nr_written = *nr_written +
1035 (end - start + PAGE_SIZE) / PAGE_SIZE;
1038 } else if (ret < 0) {
1043 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1044 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1047 * Relocation relies on the relocated extents to have exactly the same
1048 * size as the original extents. Normally writeback for relocation data
1049 * extents follows a NOCOW path because relocation preallocates the
1050 * extents. However, due to an operation such as scrub turning a block
1051 * group to RO mode, it may fallback to COW mode, so we must make sure
1052 * an extent allocated during COW has exactly the requested size and can
1053 * not be split into smaller extents, otherwise relocation breaks and
1054 * fails during the stage where it updates the bytenr of file extent
1057 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1058 min_alloc_size = num_bytes;
1060 min_alloc_size = fs_info->sectorsize;
1062 while (num_bytes > 0) {
1063 cur_alloc_size = num_bytes;
1064 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1065 min_alloc_size, 0, alloc_hint,
1069 cur_alloc_size = ins.offset;
1070 extent_reserved = true;
1072 ram_size = ins.offset;
1073 em = create_io_em(inode, start, ins.offset, /* len */
1074 start, /* orig_start */
1075 ins.objectid, /* block_start */
1076 ins.offset, /* block_len */
1077 ins.offset, /* orig_block_len */
1078 ram_size, /* ram_bytes */
1079 BTRFS_COMPRESS_NONE, /* compress_type */
1080 BTRFS_ORDERED_REGULAR /* type */);
1085 free_extent_map(em);
1087 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1088 ram_size, cur_alloc_size, 0);
1090 goto out_drop_extent_cache;
1092 if (root->root_key.objectid ==
1093 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1094 ret = btrfs_reloc_clone_csums(inode, start,
1097 * Only drop cache here, and process as normal.
1099 * We must not allow extent_clear_unlock_delalloc()
1100 * at out_unlock label to free meta of this ordered
1101 * extent, as its meta should be freed by
1102 * btrfs_finish_ordered_io().
1104 * So we must continue until @start is increased to
1105 * skip current ordered extent.
1108 btrfs_drop_extent_cache(inode, start,
1109 start + ram_size - 1, 0);
1112 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1114 /* we're not doing compressed IO, don't unlock the first
1115 * page (which the caller expects to stay locked), don't
1116 * clear any dirty bits and don't set any writeback bits
1118 * Do set the Private2 bit so we know this page was properly
1119 * setup for writepage
1121 page_ops = unlock ? PAGE_UNLOCK : 0;
1122 page_ops |= PAGE_SET_PRIVATE2;
1124 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1126 EXTENT_LOCKED | EXTENT_DELALLOC,
1128 if (num_bytes < cur_alloc_size)
1131 num_bytes -= cur_alloc_size;
1132 alloc_hint = ins.objectid + ins.offset;
1133 start += cur_alloc_size;
1134 extent_reserved = false;
1137 * btrfs_reloc_clone_csums() error, since start is increased
1138 * extent_clear_unlock_delalloc() at out_unlock label won't
1139 * free metadata of current ordered extent, we're OK to exit.
1147 out_drop_extent_cache:
1148 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1150 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1151 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1153 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1154 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1155 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1158 * If we reserved an extent for our delalloc range (or a subrange) and
1159 * failed to create the respective ordered extent, then it means that
1160 * when we reserved the extent we decremented the extent's size from
1161 * the data space_info's bytes_may_use counter and incremented the
1162 * space_info's bytes_reserved counter by the same amount. We must make
1163 * sure extent_clear_unlock_delalloc() does not try to decrement again
1164 * the data space_info's bytes_may_use counter, therefore we do not pass
1165 * it the flag EXTENT_CLEAR_DATA_RESV.
1167 if (extent_reserved) {
1168 extent_clear_unlock_delalloc(inode, start,
1169 start + cur_alloc_size - 1,
1173 start += cur_alloc_size;
1177 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1178 clear_bits | EXTENT_CLEAR_DATA_RESV,
1184 * work queue call back to started compression on a file and pages
1186 static noinline void async_cow_start(struct btrfs_work *work)
1188 struct async_chunk *async_chunk;
1189 int compressed_extents;
1191 async_chunk = container_of(work, struct async_chunk, work);
1193 compressed_extents = compress_file_range(async_chunk);
1194 if (compressed_extents == 0) {
1195 btrfs_add_delayed_iput(async_chunk->inode);
1196 async_chunk->inode = NULL;
1201 * work queue call back to submit previously compressed pages
1203 static noinline void async_cow_submit(struct btrfs_work *work)
1205 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1207 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1208 unsigned long nr_pages;
1210 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1213 /* atomic_sub_return implies a barrier */
1214 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1216 cond_wake_up_nomb(&fs_info->async_submit_wait);
1219 * ->inode could be NULL if async_chunk_start has failed to compress,
1220 * in which case we don't have anything to submit, yet we need to
1221 * always adjust ->async_delalloc_pages as its paired with the init
1222 * happening in cow_file_range_async
1224 if (async_chunk->inode)
1225 submit_compressed_extents(async_chunk);
1228 static noinline void async_cow_free(struct btrfs_work *work)
1230 struct async_chunk *async_chunk;
1232 async_chunk = container_of(work, struct async_chunk, work);
1233 if (async_chunk->inode)
1234 btrfs_add_delayed_iput(async_chunk->inode);
1235 if (async_chunk->blkcg_css)
1236 css_put(async_chunk->blkcg_css);
1238 * Since the pointer to 'pending' is at the beginning of the array of
1239 * async_chunk's, freeing it ensures the whole array has been freed.
1241 if (atomic_dec_and_test(async_chunk->pending))
1242 kvfree(async_chunk->pending);
1245 static int cow_file_range_async(struct btrfs_inode *inode,
1246 struct writeback_control *wbc,
1247 struct page *locked_page,
1248 u64 start, u64 end, int *page_started,
1249 unsigned long *nr_written)
1251 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1252 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1253 struct async_cow *ctx;
1254 struct async_chunk *async_chunk;
1255 unsigned long nr_pages;
1257 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1259 bool should_compress;
1261 const unsigned int write_flags = wbc_to_write_flags(wbc);
1263 unlock_extent(&inode->io_tree, start, end);
1265 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1266 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1268 should_compress = false;
1270 should_compress = true;
1273 nofs_flag = memalloc_nofs_save();
1274 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1275 memalloc_nofs_restore(nofs_flag);
1278 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1279 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1280 EXTENT_DO_ACCOUNTING;
1281 unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1282 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
1285 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1286 clear_bits, page_ops);
1290 async_chunk = ctx->chunks;
1291 atomic_set(&ctx->num_chunks, num_chunks);
1293 for (i = 0; i < num_chunks; i++) {
1294 if (should_compress)
1295 cur_end = min(end, start + SZ_512K - 1);
1300 * igrab is called higher up in the call chain, take only the
1301 * lightweight reference for the callback lifetime
1303 ihold(&inode->vfs_inode);
1304 async_chunk[i].pending = &ctx->num_chunks;
1305 async_chunk[i].inode = &inode->vfs_inode;
1306 async_chunk[i].start = start;
1307 async_chunk[i].end = cur_end;
1308 async_chunk[i].write_flags = write_flags;
1309 INIT_LIST_HEAD(&async_chunk[i].extents);
1312 * The locked_page comes all the way from writepage and its
1313 * the original page we were actually given. As we spread
1314 * this large delalloc region across multiple async_chunk
1315 * structs, only the first struct needs a pointer to locked_page
1317 * This way we don't need racey decisions about who is supposed
1322 * Depending on the compressibility, the pages might or
1323 * might not go through async. We want all of them to
1324 * be accounted against wbc once. Let's do it here
1325 * before the paths diverge. wbc accounting is used
1326 * only for foreign writeback detection and doesn't
1327 * need full accuracy. Just account the whole thing
1328 * against the first page.
1330 wbc_account_cgroup_owner(wbc, locked_page,
1332 async_chunk[i].locked_page = locked_page;
1335 async_chunk[i].locked_page = NULL;
1338 if (blkcg_css != blkcg_root_css) {
1340 async_chunk[i].blkcg_css = blkcg_css;
1342 async_chunk[i].blkcg_css = NULL;
1345 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1346 async_cow_submit, async_cow_free);
1348 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1349 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1351 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1353 *nr_written += nr_pages;
1354 start = cur_end + 1;
1360 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1361 u64 bytenr, u64 num_bytes)
1364 struct btrfs_ordered_sum *sums;
1367 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1368 bytenr + num_bytes - 1, &list, 0);
1369 if (ret == 0 && list_empty(&list))
1372 while (!list_empty(&list)) {
1373 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1374 list_del(&sums->list);
1382 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1383 const u64 start, const u64 end,
1384 int *page_started, unsigned long *nr_written)
1386 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1387 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1388 BTRFS_DATA_RELOC_TREE_OBJECTID);
1389 const u64 range_bytes = end + 1 - start;
1390 struct extent_io_tree *io_tree = &inode->io_tree;
1391 u64 range_start = start;
1395 * If EXTENT_NORESERVE is set it means that when the buffered write was
1396 * made we had not enough available data space and therefore we did not
1397 * reserve data space for it, since we though we could do NOCOW for the
1398 * respective file range (either there is prealloc extent or the inode
1399 * has the NOCOW bit set).
1401 * However when we need to fallback to COW mode (because for example the
1402 * block group for the corresponding extent was turned to RO mode by a
1403 * scrub or relocation) we need to do the following:
1405 * 1) We increment the bytes_may_use counter of the data space info.
1406 * If COW succeeds, it allocates a new data extent and after doing
1407 * that it decrements the space info's bytes_may_use counter and
1408 * increments its bytes_reserved counter by the same amount (we do
1409 * this at btrfs_add_reserved_bytes()). So we need to increment the
1410 * bytes_may_use counter to compensate (when space is reserved at
1411 * buffered write time, the bytes_may_use counter is incremented);
1413 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1414 * that if the COW path fails for any reason, it decrements (through
1415 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1416 * data space info, which we incremented in the step above.
1418 * If we need to fallback to cow and the inode corresponds to a free
1419 * space cache inode or an inode of the data relocation tree, we must
1420 * also increment bytes_may_use of the data space_info for the same
1421 * reason. Space caches and relocated data extents always get a prealloc
1422 * extent for them, however scrub or balance may have set the block
1423 * group that contains that extent to RO mode and therefore force COW
1424 * when starting writeback.
1426 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1427 EXTENT_NORESERVE, 0);
1428 if (count > 0 || is_space_ino || is_reloc_ino) {
1430 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1431 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1433 if (is_space_ino || is_reloc_ino)
1434 bytes = range_bytes;
1436 spin_lock(&sinfo->lock);
1437 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1438 spin_unlock(&sinfo->lock);
1441 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1445 return cow_file_range(inode, locked_page, start, end, page_started,
1450 * when nowcow writeback call back. This checks for snapshots or COW copies
1451 * of the extents that exist in the file, and COWs the file as required.
1453 * If no cow copies or snapshots exist, we write directly to the existing
1456 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1457 struct page *locked_page,
1458 const u64 start, const u64 end,
1459 int *page_started, int force,
1460 unsigned long *nr_written)
1462 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1463 struct btrfs_root *root = inode->root;
1464 struct btrfs_path *path;
1465 u64 cow_start = (u64)-1;
1466 u64 cur_offset = start;
1468 bool check_prev = true;
1469 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1470 u64 ino = btrfs_ino(inode);
1472 u64 disk_bytenr = 0;
1474 path = btrfs_alloc_path();
1476 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1477 EXTENT_LOCKED | EXTENT_DELALLOC |
1478 EXTENT_DO_ACCOUNTING |
1479 EXTENT_DEFRAG, PAGE_UNLOCK |
1481 PAGE_SET_WRITEBACK |
1482 PAGE_END_WRITEBACK);
1487 struct btrfs_key found_key;
1488 struct btrfs_file_extent_item *fi;
1489 struct extent_buffer *leaf;
1499 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1505 * If there is no extent for our range when doing the initial
1506 * search, then go back to the previous slot as it will be the
1507 * one containing the search offset
1509 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1510 leaf = path->nodes[0];
1511 btrfs_item_key_to_cpu(leaf, &found_key,
1512 path->slots[0] - 1);
1513 if (found_key.objectid == ino &&
1514 found_key.type == BTRFS_EXTENT_DATA_KEY)
1519 /* Go to next leaf if we have exhausted the current one */
1520 leaf = path->nodes[0];
1521 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1522 ret = btrfs_next_leaf(root, path);
1524 if (cow_start != (u64)-1)
1525 cur_offset = cow_start;
1530 leaf = path->nodes[0];
1533 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535 /* Didn't find anything for our INO */
1536 if (found_key.objectid > ino)
1539 * Keep searching until we find an EXTENT_ITEM or there are no
1540 * more extents for this inode
1542 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1543 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1548 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1549 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1550 found_key.offset > end)
1554 * If the found extent starts after requested offset, then
1555 * adjust extent_end to be right before this extent begins
1557 if (found_key.offset > cur_offset) {
1558 extent_end = found_key.offset;
1564 * Found extent which begins before our range and potentially
1567 fi = btrfs_item_ptr(leaf, path->slots[0],
1568 struct btrfs_file_extent_item);
1569 extent_type = btrfs_file_extent_type(leaf, fi);
1571 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1572 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1573 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1574 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1575 extent_offset = btrfs_file_extent_offset(leaf, fi);
1576 extent_end = found_key.offset +
1577 btrfs_file_extent_num_bytes(leaf, fi);
1579 btrfs_file_extent_disk_num_bytes(leaf, fi);
1581 * If the extent we got ends before our current offset,
1582 * skip to the next extent.
1584 if (extent_end <= cur_offset) {
1589 if (disk_bytenr == 0)
1591 /* Skip compressed/encrypted/encoded extents */
1592 if (btrfs_file_extent_compression(leaf, fi) ||
1593 btrfs_file_extent_encryption(leaf, fi) ||
1594 btrfs_file_extent_other_encoding(leaf, fi))
1597 * If extent is created before the last volume's snapshot
1598 * this implies the extent is shared, hence we can't do
1599 * nocow. This is the same check as in
1600 * btrfs_cross_ref_exist but without calling
1601 * btrfs_search_slot.
1603 if (!freespace_inode &&
1604 btrfs_file_extent_generation(leaf, fi) <=
1605 btrfs_root_last_snapshot(&root->root_item))
1607 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1609 /* If extent is RO, we must COW it */
1610 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1612 ret = btrfs_cross_ref_exist(root, ino,
1614 extent_offset, disk_bytenr, false);
1617 * ret could be -EIO if the above fails to read
1621 if (cow_start != (u64)-1)
1622 cur_offset = cow_start;
1626 WARN_ON_ONCE(freespace_inode);
1629 disk_bytenr += extent_offset;
1630 disk_bytenr += cur_offset - found_key.offset;
1631 num_bytes = min(end + 1, extent_end) - cur_offset;
1633 * If there are pending snapshots for this root, we
1634 * fall into common COW way
1636 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1639 * force cow if csum exists in the range.
1640 * this ensure that csum for a given extent are
1641 * either valid or do not exist.
1643 ret = csum_exist_in_range(fs_info, disk_bytenr,
1647 * ret could be -EIO if the above fails to read
1651 if (cow_start != (u64)-1)
1652 cur_offset = cow_start;
1655 WARN_ON_ONCE(freespace_inode);
1658 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1661 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1662 extent_end = found_key.offset + ram_bytes;
1663 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1664 /* Skip extents outside of our requested range */
1665 if (extent_end <= start) {
1670 /* If this triggers then we have a memory corruption */
1675 * If nocow is false then record the beginning of the range
1676 * that needs to be COWed
1679 if (cow_start == (u64)-1)
1680 cow_start = cur_offset;
1681 cur_offset = extent_end;
1682 if (cur_offset > end)
1688 btrfs_release_path(path);
1691 * COW range from cow_start to found_key.offset - 1. As the key
1692 * will contain the beginning of the first extent that can be
1693 * NOCOW, following one which needs to be COW'ed
1695 if (cow_start != (u64)-1) {
1696 ret = fallback_to_cow(inode, locked_page,
1697 cow_start, found_key.offset - 1,
1698 page_started, nr_written);
1701 cow_start = (u64)-1;
1704 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1705 u64 orig_start = found_key.offset - extent_offset;
1706 struct extent_map *em;
1708 em = create_io_em(inode, cur_offset, num_bytes,
1710 disk_bytenr, /* block_start */
1711 num_bytes, /* block_len */
1712 disk_num_bytes, /* orig_block_len */
1713 ram_bytes, BTRFS_COMPRESS_NONE,
1714 BTRFS_ORDERED_PREALLOC);
1719 free_extent_map(em);
1720 ret = btrfs_add_ordered_extent(inode, cur_offset,
1721 disk_bytenr, num_bytes,
1723 BTRFS_ORDERED_PREALLOC);
1725 btrfs_drop_extent_cache(inode, cur_offset,
1726 cur_offset + num_bytes - 1,
1731 ret = btrfs_add_ordered_extent(inode, cur_offset,
1732 disk_bytenr, num_bytes,
1734 BTRFS_ORDERED_NOCOW);
1740 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1743 if (root->root_key.objectid ==
1744 BTRFS_DATA_RELOC_TREE_OBJECTID)
1746 * Error handled later, as we must prevent
1747 * extent_clear_unlock_delalloc() in error handler
1748 * from freeing metadata of created ordered extent.
1750 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1753 extent_clear_unlock_delalloc(inode, cur_offset,
1754 cur_offset + num_bytes - 1,
1755 locked_page, EXTENT_LOCKED |
1757 EXTENT_CLEAR_DATA_RESV,
1758 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1760 cur_offset = extent_end;
1763 * btrfs_reloc_clone_csums() error, now we're OK to call error
1764 * handler, as metadata for created ordered extent will only
1765 * be freed by btrfs_finish_ordered_io().
1769 if (cur_offset > end)
1772 btrfs_release_path(path);
1774 if (cur_offset <= end && cow_start == (u64)-1)
1775 cow_start = cur_offset;
1777 if (cow_start != (u64)-1) {
1779 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1780 page_started, nr_written);
1787 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1789 if (ret && cur_offset < end)
1790 extent_clear_unlock_delalloc(inode, cur_offset, end,
1791 locked_page, EXTENT_LOCKED |
1792 EXTENT_DELALLOC | EXTENT_DEFRAG |
1793 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1795 PAGE_SET_WRITEBACK |
1796 PAGE_END_WRITEBACK);
1797 btrfs_free_path(path);
1801 static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end)
1804 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1805 !(inode->flags & BTRFS_INODE_PREALLOC))
1809 * @defrag_bytes is a hint value, no spinlock held here,
1810 * if is not zero, it means the file is defragging.
1811 * Force cow if given extent needs to be defragged.
1813 if (inode->defrag_bytes &&
1814 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL))
1821 * Function to process delayed allocation (create CoW) for ranges which are
1822 * being touched for the first time.
1824 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1825 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1826 struct writeback_control *wbc)
1829 int force_cow = need_force_cow(inode, start, end);
1831 if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1832 ret = run_delalloc_nocow(inode, locked_page, start, end,
1833 page_started, 1, nr_written);
1834 } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1835 ret = run_delalloc_nocow(inode, locked_page, start, end,
1836 page_started, 0, nr_written);
1837 } else if (!inode_can_compress(inode) ||
1838 !inode_need_compress(inode, start, end)) {
1839 ret = cow_file_range(inode, locked_page, start, end,
1840 page_started, nr_written, 1);
1842 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1843 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1844 page_started, nr_written);
1847 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1852 void btrfs_split_delalloc_extent(struct inode *inode,
1853 struct extent_state *orig, u64 split)
1857 /* not delalloc, ignore it */
1858 if (!(orig->state & EXTENT_DELALLOC))
1861 size = orig->end - orig->start + 1;
1862 if (size > BTRFS_MAX_EXTENT_SIZE) {
1867 * See the explanation in btrfs_merge_delalloc_extent, the same
1868 * applies here, just in reverse.
1870 new_size = orig->end - split + 1;
1871 num_extents = count_max_extents(new_size);
1872 new_size = split - orig->start;
1873 num_extents += count_max_extents(new_size);
1874 if (count_max_extents(size) >= num_extents)
1878 spin_lock(&BTRFS_I(inode)->lock);
1879 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1880 spin_unlock(&BTRFS_I(inode)->lock);
1884 * Handle merged delayed allocation extents so we can keep track of new extents
1885 * that are just merged onto old extents, such as when we are doing sequential
1886 * writes, so we can properly account for the metadata space we'll need.
1888 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1889 struct extent_state *other)
1891 u64 new_size, old_size;
1894 /* not delalloc, ignore it */
1895 if (!(other->state & EXTENT_DELALLOC))
1898 if (new->start > other->start)
1899 new_size = new->end - other->start + 1;
1901 new_size = other->end - new->start + 1;
1903 /* we're not bigger than the max, unreserve the space and go */
1904 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1905 spin_lock(&BTRFS_I(inode)->lock);
1906 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1907 spin_unlock(&BTRFS_I(inode)->lock);
1912 * We have to add up either side to figure out how many extents were
1913 * accounted for before we merged into one big extent. If the number of
1914 * extents we accounted for is <= the amount we need for the new range
1915 * then we can return, otherwise drop. Think of it like this
1919 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1920 * need 2 outstanding extents, on one side we have 1 and the other side
1921 * we have 1 so they are == and we can return. But in this case
1923 * [MAX_SIZE+4k][MAX_SIZE+4k]
1925 * Each range on their own accounts for 2 extents, but merged together
1926 * they are only 3 extents worth of accounting, so we need to drop in
1929 old_size = other->end - other->start + 1;
1930 num_extents = count_max_extents(old_size);
1931 old_size = new->end - new->start + 1;
1932 num_extents += count_max_extents(old_size);
1933 if (count_max_extents(new_size) >= num_extents)
1936 spin_lock(&BTRFS_I(inode)->lock);
1937 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1938 spin_unlock(&BTRFS_I(inode)->lock);
1941 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1942 struct inode *inode)
1944 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1946 spin_lock(&root->delalloc_lock);
1947 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1948 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1949 &root->delalloc_inodes);
1950 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1951 &BTRFS_I(inode)->runtime_flags);
1952 root->nr_delalloc_inodes++;
1953 if (root->nr_delalloc_inodes == 1) {
1954 spin_lock(&fs_info->delalloc_root_lock);
1955 BUG_ON(!list_empty(&root->delalloc_root));
1956 list_add_tail(&root->delalloc_root,
1957 &fs_info->delalloc_roots);
1958 spin_unlock(&fs_info->delalloc_root_lock);
1961 spin_unlock(&root->delalloc_lock);
1965 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1966 struct btrfs_inode *inode)
1968 struct btrfs_fs_info *fs_info = root->fs_info;
1970 if (!list_empty(&inode->delalloc_inodes)) {
1971 list_del_init(&inode->delalloc_inodes);
1972 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1973 &inode->runtime_flags);
1974 root->nr_delalloc_inodes--;
1975 if (!root->nr_delalloc_inodes) {
1976 ASSERT(list_empty(&root->delalloc_inodes));
1977 spin_lock(&fs_info->delalloc_root_lock);
1978 BUG_ON(list_empty(&root->delalloc_root));
1979 list_del_init(&root->delalloc_root);
1980 spin_unlock(&fs_info->delalloc_root_lock);
1985 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1986 struct btrfs_inode *inode)
1988 spin_lock(&root->delalloc_lock);
1989 __btrfs_del_delalloc_inode(root, inode);
1990 spin_unlock(&root->delalloc_lock);
1994 * Properly track delayed allocation bytes in the inode and to maintain the
1995 * list of inodes that have pending delalloc work to be done.
1997 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2000 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2002 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2005 * set_bit and clear bit hooks normally require _irqsave/restore
2006 * but in this case, we are only testing for the DELALLOC
2007 * bit, which is only set or cleared with irqs on
2009 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2010 struct btrfs_root *root = BTRFS_I(inode)->root;
2011 u64 len = state->end + 1 - state->start;
2012 u32 num_extents = count_max_extents(len);
2013 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2015 spin_lock(&BTRFS_I(inode)->lock);
2016 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2017 spin_unlock(&BTRFS_I(inode)->lock);
2019 /* For sanity tests */
2020 if (btrfs_is_testing(fs_info))
2023 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2024 fs_info->delalloc_batch);
2025 spin_lock(&BTRFS_I(inode)->lock);
2026 BTRFS_I(inode)->delalloc_bytes += len;
2027 if (*bits & EXTENT_DEFRAG)
2028 BTRFS_I(inode)->defrag_bytes += len;
2029 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2030 &BTRFS_I(inode)->runtime_flags))
2031 btrfs_add_delalloc_inodes(root, inode);
2032 spin_unlock(&BTRFS_I(inode)->lock);
2035 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2036 (*bits & EXTENT_DELALLOC_NEW)) {
2037 spin_lock(&BTRFS_I(inode)->lock);
2038 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2040 spin_unlock(&BTRFS_I(inode)->lock);
2045 * Once a range is no longer delalloc this function ensures that proper
2046 * accounting happens.
2048 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2049 struct extent_state *state, unsigned *bits)
2051 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2052 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2053 u64 len = state->end + 1 - state->start;
2054 u32 num_extents = count_max_extents(len);
2056 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2057 spin_lock(&inode->lock);
2058 inode->defrag_bytes -= len;
2059 spin_unlock(&inode->lock);
2063 * set_bit and clear bit hooks normally require _irqsave/restore
2064 * but in this case, we are only testing for the DELALLOC
2065 * bit, which is only set or cleared with irqs on
2067 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2068 struct btrfs_root *root = inode->root;
2069 bool do_list = !btrfs_is_free_space_inode(inode);
2071 spin_lock(&inode->lock);
2072 btrfs_mod_outstanding_extents(inode, -num_extents);
2073 spin_unlock(&inode->lock);
2076 * We don't reserve metadata space for space cache inodes so we
2077 * don't need to call delalloc_release_metadata if there is an
2080 if (*bits & EXTENT_CLEAR_META_RESV &&
2081 root != fs_info->tree_root)
2082 btrfs_delalloc_release_metadata(inode, len, false);
2084 /* For sanity tests. */
2085 if (btrfs_is_testing(fs_info))
2088 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2089 do_list && !(state->state & EXTENT_NORESERVE) &&
2090 (*bits & EXTENT_CLEAR_DATA_RESV))
2091 btrfs_free_reserved_data_space_noquota(fs_info, len);
2093 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2094 fs_info->delalloc_batch);
2095 spin_lock(&inode->lock);
2096 inode->delalloc_bytes -= len;
2097 if (do_list && inode->delalloc_bytes == 0 &&
2098 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2099 &inode->runtime_flags))
2100 btrfs_del_delalloc_inode(root, inode);
2101 spin_unlock(&inode->lock);
2104 if ((state->state & EXTENT_DELALLOC_NEW) &&
2105 (*bits & EXTENT_DELALLOC_NEW)) {
2106 spin_lock(&inode->lock);
2107 ASSERT(inode->new_delalloc_bytes >= len);
2108 inode->new_delalloc_bytes -= len;
2109 spin_unlock(&inode->lock);
2114 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2115 * in a chunk's stripe. This function ensures that bios do not span a
2118 * @page - The page we are about to add to the bio
2119 * @size - size we want to add to the bio
2120 * @bio - bio we want to ensure is smaller than a stripe
2121 * @bio_flags - flags of the bio
2123 * return 1 if page cannot be added to the bio
2124 * return 0 if page can be added to the bio
2125 * return error otherwise
2127 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2128 unsigned long bio_flags)
2130 struct inode *inode = page->mapping->host;
2131 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2132 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
2136 struct btrfs_io_geometry geom;
2138 if (bio_flags & EXTENT_BIO_COMPRESSED)
2141 length = bio->bi_iter.bi_size;
2142 map_length = length;
2143 ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length,
2148 if (geom.len < length + size)
2154 * in order to insert checksums into the metadata in large chunks,
2155 * we wait until bio submission time. All the pages in the bio are
2156 * checksummed and sums are attached onto the ordered extent record.
2158 * At IO completion time the cums attached on the ordered extent record
2159 * are inserted into the btree
2161 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
2164 struct inode *inode = private_data;
2166 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2170 * extent_io.c submission hook. This does the right thing for csum calculation
2171 * on write, or reading the csums from the tree before a read.
2173 * Rules about async/sync submit,
2174 * a) read: sync submit
2176 * b) write without checksum: sync submit
2178 * c) write with checksum:
2179 * c-1) if bio is issued by fsync: sync submit
2180 * (sync_writers != 0)
2182 * c-2) if root is reloc root: sync submit
2183 * (only in case of buffered IO)
2185 * c-3) otherwise: async submit
2187 static blk_status_t btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
2189 unsigned long bio_flags)
2192 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2193 struct btrfs_root *root = BTRFS_I(inode)->root;
2194 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2195 blk_status_t ret = 0;
2197 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2199 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2201 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2202 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2204 if (bio_op(bio) != REQ_OP_WRITE) {
2205 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2209 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2210 ret = btrfs_submit_compressed_read(inode, bio,
2214 } else if (!skip_sum) {
2215 ret = btrfs_lookup_bio_sums(inode, bio, (u64)-1, NULL);
2220 } else if (async && !skip_sum) {
2221 /* csum items have already been cloned */
2222 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2224 /* we're doing a write, do the async checksumming */
2225 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2226 0, inode, btrfs_submit_bio_start);
2228 } else if (!skip_sum) {
2229 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2235 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2239 bio->bi_status = ret;
2246 * given a list of ordered sums record them in the inode. This happens
2247 * at IO completion time based on sums calculated at bio submission time.
2249 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2250 struct inode *inode, struct list_head *list)
2252 struct btrfs_ordered_sum *sum;
2255 list_for_each_entry(sum, list, list) {
2256 trans->adding_csums = true;
2257 ret = btrfs_csum_file_blocks(trans,
2258 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2259 trans->adding_csums = false;
2266 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2267 unsigned int extra_bits,
2268 struct extent_state **cached_state)
2270 WARN_ON(PAGE_ALIGNED(end));
2271 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2275 /* see btrfs_writepage_start_hook for details on why this is required */
2276 struct btrfs_writepage_fixup {
2278 struct inode *inode;
2279 struct btrfs_work work;
2282 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2284 struct btrfs_writepage_fixup *fixup;
2285 struct btrfs_ordered_extent *ordered;
2286 struct extent_state *cached_state = NULL;
2287 struct extent_changeset *data_reserved = NULL;
2289 struct btrfs_inode *inode;
2293 bool free_delalloc_space = true;
2295 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2297 inode = BTRFS_I(fixup->inode);
2298 page_start = page_offset(page);
2299 page_end = page_offset(page) + PAGE_SIZE - 1;
2302 * This is similar to page_mkwrite, we need to reserve the space before
2303 * we take the page lock.
2305 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2311 * Before we queued this fixup, we took a reference on the page.
2312 * page->mapping may go NULL, but it shouldn't be moved to a different
2315 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2317 * Unfortunately this is a little tricky, either
2319 * 1) We got here and our page had already been dealt with and
2320 * we reserved our space, thus ret == 0, so we need to just
2321 * drop our space reservation and bail. This can happen the
2322 * first time we come into the fixup worker, or could happen
2323 * while waiting for the ordered extent.
2324 * 2) Our page was already dealt with, but we happened to get an
2325 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2326 * this case we obviously don't have anything to release, but
2327 * because the page was already dealt with we don't want to
2328 * mark the page with an error, so make sure we're resetting
2329 * ret to 0. This is why we have this check _before_ the ret
2330 * check, because we do not want to have a surprise ENOSPC
2331 * when the page was already properly dealt with.
2334 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2335 btrfs_delalloc_release_space(inode, data_reserved,
2336 page_start, PAGE_SIZE,
2344 * We can't mess with the page state unless it is locked, so now that
2345 * it is locked bail if we failed to make our space reservation.
2350 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2352 /* already ordered? We're done */
2353 if (PagePrivate2(page))
2356 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2358 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2361 btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1);
2362 btrfs_put_ordered_extent(ordered);
2366 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2372 * Everything went as planned, we're now the owner of a dirty page with
2373 * delayed allocation bits set and space reserved for our COW
2376 * The page was dirty when we started, nothing should have cleaned it.
2378 BUG_ON(!PageDirty(page));
2379 free_delalloc_space = false;
2381 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2382 if (free_delalloc_space)
2383 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2385 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2390 * We hit ENOSPC or other errors. Update the mapping and page
2391 * to reflect the errors and clean the page.
2393 mapping_set_error(page->mapping, ret);
2394 end_extent_writepage(page, ret, page_start, page_end);
2395 clear_page_dirty_for_io(page);
2398 ClearPageChecked(page);
2402 extent_changeset_free(data_reserved);
2404 * As a precaution, do a delayed iput in case it would be the last iput
2405 * that could need flushing space. Recursing back to fixup worker would
2408 btrfs_add_delayed_iput(&inode->vfs_inode);
2412 * There are a few paths in the higher layers of the kernel that directly
2413 * set the page dirty bit without asking the filesystem if it is a
2414 * good idea. This causes problems because we want to make sure COW
2415 * properly happens and the data=ordered rules are followed.
2417 * In our case any range that doesn't have the ORDERED bit set
2418 * hasn't been properly setup for IO. We kick off an async process
2419 * to fix it up. The async helper will wait for ordered extents, set
2420 * the delalloc bit and make it safe to write the page.
2422 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2424 struct inode *inode = page->mapping->host;
2425 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2426 struct btrfs_writepage_fixup *fixup;
2428 /* this page is properly in the ordered list */
2429 if (TestClearPagePrivate2(page))
2433 * PageChecked is set below when we create a fixup worker for this page,
2434 * don't try to create another one if we're already PageChecked()
2436 * The extent_io writepage code will redirty the page if we send back
2439 if (PageChecked(page))
2442 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2447 * We are already holding a reference to this inode from
2448 * write_cache_pages. We need to hold it because the space reservation
2449 * takes place outside of the page lock, and we can't trust
2450 * page->mapping outside of the page lock.
2453 SetPageChecked(page);
2455 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2457 fixup->inode = inode;
2458 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2463 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2464 struct btrfs_inode *inode, u64 file_pos,
2465 struct btrfs_file_extent_item *stack_fi,
2466 u64 qgroup_reserved)
2468 struct btrfs_root *root = inode->root;
2469 struct btrfs_path *path;
2470 struct extent_buffer *leaf;
2471 struct btrfs_key ins;
2472 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2473 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2474 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2475 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2476 int extent_inserted = 0;
2479 path = btrfs_alloc_path();
2484 * we may be replacing one extent in the tree with another.
2485 * The new extent is pinned in the extent map, and we don't want
2486 * to drop it from the cache until it is completely in the btree.
2488 * So, tell btrfs_drop_extents to leave this extent in the cache.
2489 * the caller is expected to unpin it and allow it to be merged
2492 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2493 file_pos + num_bytes, NULL, 0,
2494 1, sizeof(*stack_fi), &extent_inserted);
2498 if (!extent_inserted) {
2499 ins.objectid = btrfs_ino(inode);
2500 ins.offset = file_pos;
2501 ins.type = BTRFS_EXTENT_DATA_KEY;
2503 path->leave_spinning = 1;
2504 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2509 leaf = path->nodes[0];
2510 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2511 write_extent_buffer(leaf, stack_fi,
2512 btrfs_item_ptr_offset(leaf, path->slots[0]),
2513 sizeof(struct btrfs_file_extent_item));
2515 btrfs_mark_buffer_dirty(leaf);
2516 btrfs_release_path(path);
2518 inode_add_bytes(&inode->vfs_inode, num_bytes);
2520 ins.objectid = disk_bytenr;
2521 ins.offset = disk_num_bytes;
2522 ins.type = BTRFS_EXTENT_ITEM_KEY;
2524 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2528 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2529 file_pos, qgroup_reserved, &ins);
2531 btrfs_free_path(path);
2536 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2539 struct btrfs_block_group *cache;
2541 cache = btrfs_lookup_block_group(fs_info, start);
2544 spin_lock(&cache->lock);
2545 cache->delalloc_bytes -= len;
2546 spin_unlock(&cache->lock);
2548 btrfs_put_block_group(cache);
2551 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2552 struct inode *inode,
2553 struct btrfs_ordered_extent *oe)
2555 struct btrfs_file_extent_item stack_fi;
2558 memset(&stack_fi, 0, sizeof(stack_fi));
2559 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2560 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2561 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2562 oe->disk_num_bytes);
2563 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2564 logical_len = oe->truncated_len;
2566 logical_len = oe->num_bytes;
2567 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2568 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2569 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2570 /* Encryption and other encoding is reserved and all 0 */
2572 return insert_reserved_file_extent(trans, BTRFS_I(inode), oe->file_offset,
2573 &stack_fi, oe->qgroup_rsv);
2577 * As ordered data IO finishes, this gets called so we can finish
2578 * an ordered extent if the range of bytes in the file it covers are
2581 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2583 struct inode *inode = ordered_extent->inode;
2584 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2585 struct btrfs_root *root = BTRFS_I(inode)->root;
2586 struct btrfs_trans_handle *trans = NULL;
2587 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2588 struct extent_state *cached_state = NULL;
2590 int compress_type = 0;
2592 u64 logical_len = ordered_extent->num_bytes;
2593 bool freespace_inode;
2594 bool truncated = false;
2595 bool range_locked = false;
2596 bool clear_new_delalloc_bytes = false;
2597 bool clear_reserved_extent = true;
2598 unsigned int clear_bits;
2600 start = ordered_extent->file_offset;
2601 end = start + ordered_extent->num_bytes - 1;
2603 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2604 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2605 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2606 clear_new_delalloc_bytes = true;
2608 freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
2610 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2615 btrfs_free_io_failure_record(BTRFS_I(inode), start, end);
2617 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2619 logical_len = ordered_extent->truncated_len;
2620 /* Truncated the entire extent, don't bother adding */
2625 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2626 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2628 btrfs_inode_safe_disk_i_size_write(inode, 0);
2629 if (freespace_inode)
2630 trans = btrfs_join_transaction_spacecache(root);
2632 trans = btrfs_join_transaction(root);
2633 if (IS_ERR(trans)) {
2634 ret = PTR_ERR(trans);
2638 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2639 ret = btrfs_update_inode_fallback(trans, root, inode);
2640 if (ret) /* -ENOMEM or corruption */
2641 btrfs_abort_transaction(trans, ret);
2645 range_locked = true;
2646 lock_extent_bits(io_tree, start, end, &cached_state);
2648 if (freespace_inode)
2649 trans = btrfs_join_transaction_spacecache(root);
2651 trans = btrfs_join_transaction(root);
2652 if (IS_ERR(trans)) {
2653 ret = PTR_ERR(trans);
2658 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2660 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2661 compress_type = ordered_extent->compress_type;
2662 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2663 BUG_ON(compress_type);
2664 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2665 ordered_extent->file_offset,
2666 ordered_extent->file_offset +
2669 BUG_ON(root == fs_info->tree_root);
2670 ret = insert_ordered_extent_file_extent(trans, inode,
2673 clear_reserved_extent = false;
2674 btrfs_release_delalloc_bytes(fs_info,
2675 ordered_extent->disk_bytenr,
2676 ordered_extent->disk_num_bytes);
2679 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2680 ordered_extent->file_offset,
2681 ordered_extent->num_bytes, trans->transid);
2683 btrfs_abort_transaction(trans, ret);
2687 ret = add_pending_csums(trans, inode, &ordered_extent->list);
2689 btrfs_abort_transaction(trans, ret);
2693 btrfs_inode_safe_disk_i_size_write(inode, 0);
2694 ret = btrfs_update_inode_fallback(trans, root, inode);
2695 if (ret) { /* -ENOMEM or corruption */
2696 btrfs_abort_transaction(trans, ret);
2701 clear_bits = EXTENT_DEFRAG;
2703 clear_bits |= EXTENT_LOCKED;
2704 if (clear_new_delalloc_bytes)
2705 clear_bits |= EXTENT_DELALLOC_NEW;
2706 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits,
2707 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2711 btrfs_end_transaction(trans);
2713 if (ret || truncated) {
2714 u64 unwritten_start = start;
2717 unwritten_start += logical_len;
2718 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2720 /* Drop the cache for the part of the extent we didn't write. */
2721 btrfs_drop_extent_cache(BTRFS_I(inode), unwritten_start, end, 0);
2724 * If the ordered extent had an IOERR or something else went
2725 * wrong we need to return the space for this ordered extent
2726 * back to the allocator. We only free the extent in the
2727 * truncated case if we didn't write out the extent at all.
2729 * If we made it past insert_reserved_file_extent before we
2730 * errored out then we don't need to do this as the accounting
2731 * has already been done.
2733 if ((ret || !logical_len) &&
2734 clear_reserved_extent &&
2735 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2736 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2738 * Discard the range before returning it back to the
2741 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2742 btrfs_discard_extent(fs_info,
2743 ordered_extent->disk_bytenr,
2744 ordered_extent->disk_num_bytes,
2746 btrfs_free_reserved_extent(fs_info,
2747 ordered_extent->disk_bytenr,
2748 ordered_extent->disk_num_bytes, 1);
2753 * This needs to be done to make sure anybody waiting knows we are done
2754 * updating everything for this ordered extent.
2756 btrfs_remove_ordered_extent(inode, ordered_extent);
2759 btrfs_put_ordered_extent(ordered_extent);
2760 /* once for the tree */
2761 btrfs_put_ordered_extent(ordered_extent);
2766 static void finish_ordered_fn(struct btrfs_work *work)
2768 struct btrfs_ordered_extent *ordered_extent;
2769 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2770 btrfs_finish_ordered_io(ordered_extent);
2773 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2774 u64 end, int uptodate)
2776 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2777 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2778 struct btrfs_ordered_extent *ordered_extent = NULL;
2779 struct btrfs_workqueue *wq;
2781 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2783 ClearPagePrivate2(page);
2784 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2785 end - start + 1, uptodate))
2788 if (btrfs_is_free_space_inode(inode))
2789 wq = fs_info->endio_freespace_worker;
2791 wq = fs_info->endio_write_workers;
2793 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2794 btrfs_queue_work(wq, &ordered_extent->work);
2797 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
2798 int icsum, struct page *page, int pgoff, u64 start,
2801 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2802 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2804 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2806 u8 csum[BTRFS_CSUM_SIZE];
2808 csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size;
2810 kaddr = kmap_atomic(page);
2811 shash->tfm = fs_info->csum_shash;
2813 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
2815 if (memcmp(csum, csum_expected, csum_size))
2818 kunmap_atomic(kaddr);
2821 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
2822 io_bio->mirror_num);
2824 btrfs_dev_stat_inc_and_print(io_bio->device,
2825 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2826 memset(kaddr + pgoff, 1, len);
2827 flush_dcache_page(page);
2828 kunmap_atomic(kaddr);
2833 * when reads are done, we need to check csums to verify the data is correct
2834 * if there's a match, we allow the bio to finish. If not, the code in
2835 * extent_io.c will try to find good copies for us.
2837 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2838 u64 phy_offset, struct page *page,
2839 u64 start, u64 end, int mirror)
2841 size_t offset = start - page_offset(page);
2842 struct inode *inode = page->mapping->host;
2843 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2844 struct btrfs_root *root = BTRFS_I(inode)->root;
2846 if (PageChecked(page)) {
2847 ClearPageChecked(page);
2851 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2854 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2855 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2856 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
2860 phy_offset >>= inode->i_sb->s_blocksize_bits;
2861 return check_data_csum(inode, io_bio, phy_offset, page, offset, start,
2862 (size_t)(end - start + 1));
2866 * btrfs_add_delayed_iput - perform a delayed iput on @inode
2868 * @inode: The inode we want to perform iput on
2870 * This function uses the generic vfs_inode::i_count to track whether we should
2871 * just decrement it (in case it's > 1) or if this is the last iput then link
2872 * the inode to the delayed iput machinery. Delayed iputs are processed at
2873 * transaction commit time/superblock commit/cleaner kthread.
2875 void btrfs_add_delayed_iput(struct inode *inode)
2877 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2878 struct btrfs_inode *binode = BTRFS_I(inode);
2880 if (atomic_add_unless(&inode->i_count, -1, 1))
2883 atomic_inc(&fs_info->nr_delayed_iputs);
2884 spin_lock(&fs_info->delayed_iput_lock);
2885 ASSERT(list_empty(&binode->delayed_iput));
2886 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
2887 spin_unlock(&fs_info->delayed_iput_lock);
2888 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
2889 wake_up_process(fs_info->cleaner_kthread);
2892 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
2893 struct btrfs_inode *inode)
2895 list_del_init(&inode->delayed_iput);
2896 spin_unlock(&fs_info->delayed_iput_lock);
2897 iput(&inode->vfs_inode);
2898 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
2899 wake_up(&fs_info->delayed_iputs_wait);
2900 spin_lock(&fs_info->delayed_iput_lock);
2903 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
2904 struct btrfs_inode *inode)
2906 if (!list_empty(&inode->delayed_iput)) {
2907 spin_lock(&fs_info->delayed_iput_lock);
2908 if (!list_empty(&inode->delayed_iput))
2909 run_delayed_iput_locked(fs_info, inode);
2910 spin_unlock(&fs_info->delayed_iput_lock);
2914 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
2917 spin_lock(&fs_info->delayed_iput_lock);
2918 while (!list_empty(&fs_info->delayed_iputs)) {
2919 struct btrfs_inode *inode;
2921 inode = list_first_entry(&fs_info->delayed_iputs,
2922 struct btrfs_inode, delayed_iput);
2923 run_delayed_iput_locked(fs_info, inode);
2925 spin_unlock(&fs_info->delayed_iput_lock);
2929 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
2930 * @fs_info - the fs_info for this fs
2931 * @return - EINTR if we were killed, 0 if nothing's pending
2933 * This will wait on any delayed iputs that are currently running with KILLABLE
2934 * set. Once they are all done running we will return, unless we are killed in
2935 * which case we return EINTR. This helps in user operations like fallocate etc
2936 * that might get blocked on the iputs.
2938 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
2940 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
2941 atomic_read(&fs_info->nr_delayed_iputs) == 0);
2948 * This creates an orphan entry for the given inode in case something goes wrong
2949 * in the middle of an unlink.
2951 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
2952 struct btrfs_inode *inode)
2956 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
2957 if (ret && ret != -EEXIST) {
2958 btrfs_abort_transaction(trans, ret);
2966 * We have done the delete so we can go ahead and remove the orphan item for
2967 * this particular inode.
2969 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
2970 struct btrfs_inode *inode)
2972 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
2976 * this cleans up any orphans that may be left on the list from the last use
2979 int btrfs_orphan_cleanup(struct btrfs_root *root)
2981 struct btrfs_fs_info *fs_info = root->fs_info;
2982 struct btrfs_path *path;
2983 struct extent_buffer *leaf;
2984 struct btrfs_key key, found_key;
2985 struct btrfs_trans_handle *trans;
2986 struct inode *inode;
2987 u64 last_objectid = 0;
2988 int ret = 0, nr_unlink = 0;
2990 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2993 path = btrfs_alloc_path();
2998 path->reada = READA_BACK;
3000 key.objectid = BTRFS_ORPHAN_OBJECTID;
3001 key.type = BTRFS_ORPHAN_ITEM_KEY;
3002 key.offset = (u64)-1;
3005 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3010 * if ret == 0 means we found what we were searching for, which
3011 * is weird, but possible, so only screw with path if we didn't
3012 * find the key and see if we have stuff that matches
3016 if (path->slots[0] == 0)
3021 /* pull out the item */
3022 leaf = path->nodes[0];
3023 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3025 /* make sure the item matches what we want */
3026 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3028 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3031 /* release the path since we're done with it */
3032 btrfs_release_path(path);
3035 * this is where we are basically btrfs_lookup, without the
3036 * crossing root thing. we store the inode number in the
3037 * offset of the orphan item.
3040 if (found_key.offset == last_objectid) {
3042 "Error removing orphan entry, stopping orphan cleanup");
3047 last_objectid = found_key.offset;
3049 found_key.objectid = found_key.offset;
3050 found_key.type = BTRFS_INODE_ITEM_KEY;
3051 found_key.offset = 0;
3052 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3053 ret = PTR_ERR_OR_ZERO(inode);
3054 if (ret && ret != -ENOENT)
3057 if (ret == -ENOENT && root == fs_info->tree_root) {
3058 struct btrfs_root *dead_root;
3059 int is_dead_root = 0;
3062 * this is an orphan in the tree root. Currently these
3063 * could come from 2 sources:
3064 * a) a snapshot deletion in progress
3065 * b) a free space cache inode
3066 * We need to distinguish those two, as the snapshot
3067 * orphan must not get deleted.
3068 * find_dead_roots already ran before us, so if this
3069 * is a snapshot deletion, we should find the root
3070 * in the fs_roots radix tree.
3073 spin_lock(&fs_info->fs_roots_radix_lock);
3074 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3075 (unsigned long)found_key.objectid);
3076 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3078 spin_unlock(&fs_info->fs_roots_radix_lock);
3081 /* prevent this orphan from being found again */
3082 key.offset = found_key.objectid - 1;
3089 * If we have an inode with links, there are a couple of
3090 * possibilities. Old kernels (before v3.12) used to create an
3091 * orphan item for truncate indicating that there were possibly
3092 * extent items past i_size that needed to be deleted. In v3.12,
3093 * truncate was changed to update i_size in sync with the extent
3094 * items, but the (useless) orphan item was still created. Since
3095 * v4.18, we don't create the orphan item for truncate at all.
3097 * So, this item could mean that we need to do a truncate, but
3098 * only if this filesystem was last used on a pre-v3.12 kernel
3099 * and was not cleanly unmounted. The odds of that are quite
3100 * slim, and it's a pain to do the truncate now, so just delete
3103 * It's also possible that this orphan item was supposed to be
3104 * deleted but wasn't. The inode number may have been reused,
3105 * but either way, we can delete the orphan item.
3107 if (ret == -ENOENT || inode->i_nlink) {
3110 trans = btrfs_start_transaction(root, 1);
3111 if (IS_ERR(trans)) {
3112 ret = PTR_ERR(trans);
3115 btrfs_debug(fs_info, "auto deleting %Lu",
3116 found_key.objectid);
3117 ret = btrfs_del_orphan_item(trans, root,
3118 found_key.objectid);
3119 btrfs_end_transaction(trans);
3127 /* this will do delete_inode and everything for us */
3130 /* release the path since we're done with it */
3131 btrfs_release_path(path);
3133 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3135 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3136 trans = btrfs_join_transaction(root);
3138 btrfs_end_transaction(trans);
3142 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3146 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3147 btrfs_free_path(path);
3152 * very simple check to peek ahead in the leaf looking for xattrs. If we
3153 * don't find any xattrs, we know there can't be any acls.
3155 * slot is the slot the inode is in, objectid is the objectid of the inode
3157 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3158 int slot, u64 objectid,
3159 int *first_xattr_slot)
3161 u32 nritems = btrfs_header_nritems(leaf);
3162 struct btrfs_key found_key;
3163 static u64 xattr_access = 0;
3164 static u64 xattr_default = 0;
3167 if (!xattr_access) {
3168 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3169 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3170 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3171 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3175 *first_xattr_slot = -1;
3176 while (slot < nritems) {
3177 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3179 /* we found a different objectid, there must not be acls */
3180 if (found_key.objectid != objectid)
3183 /* we found an xattr, assume we've got an acl */
3184 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3185 if (*first_xattr_slot == -1)
3186 *first_xattr_slot = slot;
3187 if (found_key.offset == xattr_access ||
3188 found_key.offset == xattr_default)
3193 * we found a key greater than an xattr key, there can't
3194 * be any acls later on
3196 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3203 * it goes inode, inode backrefs, xattrs, extents,
3204 * so if there are a ton of hard links to an inode there can
3205 * be a lot of backrefs. Don't waste time searching too hard,
3206 * this is just an optimization
3211 /* we hit the end of the leaf before we found an xattr or
3212 * something larger than an xattr. We have to assume the inode
3215 if (*first_xattr_slot == -1)
3216 *first_xattr_slot = slot;
3221 * read an inode from the btree into the in-memory inode
3223 static int btrfs_read_locked_inode(struct inode *inode,
3224 struct btrfs_path *in_path)
3226 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3227 struct btrfs_path *path = in_path;
3228 struct extent_buffer *leaf;
3229 struct btrfs_inode_item *inode_item;
3230 struct btrfs_root *root = BTRFS_I(inode)->root;
3231 struct btrfs_key location;
3236 bool filled = false;
3237 int first_xattr_slot;
3239 ret = btrfs_fill_inode(inode, &rdev);
3244 path = btrfs_alloc_path();
3249 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3251 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3253 if (path != in_path)
3254 btrfs_free_path(path);
3258 leaf = path->nodes[0];
3263 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3264 struct btrfs_inode_item);
3265 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3266 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3267 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3268 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3269 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3270 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3271 round_up(i_size_read(inode), fs_info->sectorsize));
3273 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3274 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3276 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3277 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3279 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3280 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3282 BTRFS_I(inode)->i_otime.tv_sec =
3283 btrfs_timespec_sec(leaf, &inode_item->otime);
3284 BTRFS_I(inode)->i_otime.tv_nsec =
3285 btrfs_timespec_nsec(leaf, &inode_item->otime);
3287 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3288 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3289 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3291 inode_set_iversion_queried(inode,
3292 btrfs_inode_sequence(leaf, inode_item));
3293 inode->i_generation = BTRFS_I(inode)->generation;
3295 rdev = btrfs_inode_rdev(leaf, inode_item);
3297 BTRFS_I(inode)->index_cnt = (u64)-1;
3298 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3302 * If we were modified in the current generation and evicted from memory
3303 * and then re-read we need to do a full sync since we don't have any
3304 * idea about which extents were modified before we were evicted from
3307 * This is required for both inode re-read from disk and delayed inode
3308 * in delayed_nodes_tree.
3310 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3311 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3312 &BTRFS_I(inode)->runtime_flags);
3315 * We don't persist the id of the transaction where an unlink operation
3316 * against the inode was last made. So here we assume the inode might
3317 * have been evicted, and therefore the exact value of last_unlink_trans
3318 * lost, and set it to last_trans to avoid metadata inconsistencies
3319 * between the inode and its parent if the inode is fsync'ed and the log
3320 * replayed. For example, in the scenario:
3323 * ln mydir/foo mydir/bar
3326 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3327 * xfs_io -c fsync mydir/foo
3329 * mount fs, triggers fsync log replay
3331 * We must make sure that when we fsync our inode foo we also log its
3332 * parent inode, otherwise after log replay the parent still has the
3333 * dentry with the "bar" name but our inode foo has a link count of 1
3334 * and doesn't have an inode ref with the name "bar" anymore.
3336 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3337 * but it guarantees correctness at the expense of occasional full
3338 * transaction commits on fsync if our inode is a directory, or if our
3339 * inode is not a directory, logging its parent unnecessarily.
3341 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3344 * Same logic as for last_unlink_trans. We don't persist the generation
3345 * of the last transaction where this inode was used for a reflink
3346 * operation, so after eviction and reloading the inode we must be
3347 * pessimistic and assume the last transaction that modified the inode.
3349 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3352 if (inode->i_nlink != 1 ||
3353 path->slots[0] >= btrfs_header_nritems(leaf))
3356 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3357 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3360 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3361 if (location.type == BTRFS_INODE_REF_KEY) {
3362 struct btrfs_inode_ref *ref;
3364 ref = (struct btrfs_inode_ref *)ptr;
3365 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3366 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3367 struct btrfs_inode_extref *extref;
3369 extref = (struct btrfs_inode_extref *)ptr;
3370 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3375 * try to precache a NULL acl entry for files that don't have
3376 * any xattrs or acls
3378 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3379 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3380 if (first_xattr_slot != -1) {
3381 path->slots[0] = first_xattr_slot;
3382 ret = btrfs_load_inode_props(inode, path);
3385 "error loading props for ino %llu (root %llu): %d",
3386 btrfs_ino(BTRFS_I(inode)),
3387 root->root_key.objectid, ret);
3389 if (path != in_path)
3390 btrfs_free_path(path);
3393 cache_no_acl(inode);
3395 switch (inode->i_mode & S_IFMT) {
3397 inode->i_mapping->a_ops = &btrfs_aops;
3398 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3399 inode->i_fop = &btrfs_file_operations;
3400 inode->i_op = &btrfs_file_inode_operations;
3403 inode->i_fop = &btrfs_dir_file_operations;
3404 inode->i_op = &btrfs_dir_inode_operations;
3407 inode->i_op = &btrfs_symlink_inode_operations;
3408 inode_nohighmem(inode);
3409 inode->i_mapping->a_ops = &btrfs_aops;
3412 inode->i_op = &btrfs_special_inode_operations;
3413 init_special_inode(inode, inode->i_mode, rdev);
3417 btrfs_sync_inode_flags_to_i_flags(inode);
3422 * given a leaf and an inode, copy the inode fields into the leaf
3424 static void fill_inode_item(struct btrfs_trans_handle *trans,
3425 struct extent_buffer *leaf,
3426 struct btrfs_inode_item *item,
3427 struct inode *inode)
3429 struct btrfs_map_token token;
3431 btrfs_init_map_token(&token, leaf);
3433 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3434 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3435 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3436 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3437 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3439 btrfs_set_token_timespec_sec(&token, &item->atime,
3440 inode->i_atime.tv_sec);
3441 btrfs_set_token_timespec_nsec(&token, &item->atime,
3442 inode->i_atime.tv_nsec);
3444 btrfs_set_token_timespec_sec(&token, &item->mtime,
3445 inode->i_mtime.tv_sec);
3446 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3447 inode->i_mtime.tv_nsec);
3449 btrfs_set_token_timespec_sec(&token, &item->ctime,
3450 inode->i_ctime.tv_sec);
3451 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3452 inode->i_ctime.tv_nsec);
3454 btrfs_set_token_timespec_sec(&token, &item->otime,
3455 BTRFS_I(inode)->i_otime.tv_sec);
3456 btrfs_set_token_timespec_nsec(&token, &item->otime,
3457 BTRFS_I(inode)->i_otime.tv_nsec);
3459 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3460 btrfs_set_token_inode_generation(&token, item,
3461 BTRFS_I(inode)->generation);
3462 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3463 btrfs_set_token_inode_transid(&token, item, trans->transid);
3464 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3465 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3466 btrfs_set_token_inode_block_group(&token, item, 0);
3470 * copy everything in the in-memory inode into the btree.
3472 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3473 struct btrfs_root *root, struct inode *inode)
3475 struct btrfs_inode_item *inode_item;
3476 struct btrfs_path *path;
3477 struct extent_buffer *leaf;
3480 path = btrfs_alloc_path();
3484 path->leave_spinning = 1;
3485 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3493 leaf = path->nodes[0];
3494 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3495 struct btrfs_inode_item);
3497 fill_inode_item(trans, leaf, inode_item, inode);
3498 btrfs_mark_buffer_dirty(leaf);
3499 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3502 btrfs_free_path(path);
3507 * copy everything in the in-memory inode into the btree.
3509 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3510 struct btrfs_root *root, struct inode *inode)
3512 struct btrfs_fs_info *fs_info = root->fs_info;
3516 * If the inode is a free space inode, we can deadlock during commit
3517 * if we put it into the delayed code.
3519 * The data relocation inode should also be directly updated
3522 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3523 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3524 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3525 btrfs_update_root_times(trans, root);
3527 ret = btrfs_delayed_update_inode(trans, root, inode);
3529 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3533 return btrfs_update_inode_item(trans, root, inode);
3536 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3537 struct btrfs_root *root,
3538 struct inode *inode)
3542 ret = btrfs_update_inode(trans, root, inode);
3544 return btrfs_update_inode_item(trans, root, inode);
3549 * unlink helper that gets used here in inode.c and in the tree logging
3550 * recovery code. It remove a link in a directory with a given name, and
3551 * also drops the back refs in the inode to the directory
3553 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3554 struct btrfs_root *root,
3555 struct btrfs_inode *dir,
3556 struct btrfs_inode *inode,
3557 const char *name, int name_len)
3559 struct btrfs_fs_info *fs_info = root->fs_info;
3560 struct btrfs_path *path;
3562 struct btrfs_dir_item *di;
3564 u64 ino = btrfs_ino(inode);
3565 u64 dir_ino = btrfs_ino(dir);
3567 path = btrfs_alloc_path();
3573 path->leave_spinning = 1;
3574 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3575 name, name_len, -1);
3576 if (IS_ERR_OR_NULL(di)) {
3577 ret = di ? PTR_ERR(di) : -ENOENT;
3580 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3583 btrfs_release_path(path);
3586 * If we don't have dir index, we have to get it by looking up
3587 * the inode ref, since we get the inode ref, remove it directly,
3588 * it is unnecessary to do delayed deletion.
3590 * But if we have dir index, needn't search inode ref to get it.
3591 * Since the inode ref is close to the inode item, it is better
3592 * that we delay to delete it, and just do this deletion when
3593 * we update the inode item.
3595 if (inode->dir_index) {
3596 ret = btrfs_delayed_delete_inode_ref(inode);
3598 index = inode->dir_index;
3603 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3607 "failed to delete reference to %.*s, inode %llu parent %llu",
3608 name_len, name, ino, dir_ino);
3609 btrfs_abort_transaction(trans, ret);
3613 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3615 btrfs_abort_transaction(trans, ret);
3619 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3621 if (ret != 0 && ret != -ENOENT) {
3622 btrfs_abort_transaction(trans, ret);
3626 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3631 btrfs_abort_transaction(trans, ret);
3634 * If we have a pending delayed iput we could end up with the final iput
3635 * being run in btrfs-cleaner context. If we have enough of these built
3636 * up we can end up burning a lot of time in btrfs-cleaner without any
3637 * way to throttle the unlinks. Since we're currently holding a ref on
3638 * the inode we can run the delayed iput here without any issues as the
3639 * final iput won't be done until after we drop the ref we're currently
3642 btrfs_run_delayed_iput(fs_info, inode);
3644 btrfs_free_path(path);
3648 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3649 inode_inc_iversion(&inode->vfs_inode);
3650 inode_inc_iversion(&dir->vfs_inode);
3651 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3652 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3653 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3658 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3659 struct btrfs_root *root,
3660 struct btrfs_inode *dir, struct btrfs_inode *inode,
3661 const char *name, int name_len)
3664 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3666 drop_nlink(&inode->vfs_inode);
3667 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3673 * helper to start transaction for unlink and rmdir.
3675 * unlink and rmdir are special in btrfs, they do not always free space, so
3676 * if we cannot make our reservations the normal way try and see if there is
3677 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3678 * allow the unlink to occur.
3680 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3682 struct btrfs_root *root = BTRFS_I(dir)->root;
3685 * 1 for the possible orphan item
3686 * 1 for the dir item
3687 * 1 for the dir index
3688 * 1 for the inode ref
3691 return btrfs_start_transaction_fallback_global_rsv(root, 5);
3694 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3696 struct btrfs_root *root = BTRFS_I(dir)->root;
3697 struct btrfs_trans_handle *trans;
3698 struct inode *inode = d_inode(dentry);
3701 trans = __unlink_start_trans(dir);
3703 return PTR_ERR(trans);
3705 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
3708 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3709 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3710 dentry->d_name.len);
3714 if (inode->i_nlink == 0) {
3715 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3721 btrfs_end_transaction(trans);
3722 btrfs_btree_balance_dirty(root->fs_info);
3726 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3727 struct inode *dir, struct dentry *dentry)
3729 struct btrfs_root *root = BTRFS_I(dir)->root;
3730 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
3731 struct btrfs_path *path;
3732 struct extent_buffer *leaf;
3733 struct btrfs_dir_item *di;
3734 struct btrfs_key key;
3735 const char *name = dentry->d_name.name;
3736 int name_len = dentry->d_name.len;
3740 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3742 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
3743 objectid = inode->root->root_key.objectid;
3744 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3745 objectid = inode->location.objectid;
3751 path = btrfs_alloc_path();
3755 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3756 name, name_len, -1);
3757 if (IS_ERR_OR_NULL(di)) {
3758 ret = di ? PTR_ERR(di) : -ENOENT;
3762 leaf = path->nodes[0];
3763 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3764 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3765 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3767 btrfs_abort_transaction(trans, ret);
3770 btrfs_release_path(path);
3773 * This is a placeholder inode for a subvolume we didn't have a
3774 * reference to at the time of the snapshot creation. In the meantime
3775 * we could have renamed the real subvol link into our snapshot, so
3776 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
3777 * Instead simply lookup the dir_index_item for this entry so we can
3778 * remove it. Otherwise we know we have a ref to the root and we can
3779 * call btrfs_del_root_ref, and it _shouldn't_ fail.
3781 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3782 di = btrfs_search_dir_index_item(root, path, dir_ino,
3784 if (IS_ERR_OR_NULL(di)) {
3789 btrfs_abort_transaction(trans, ret);
3793 leaf = path->nodes[0];
3794 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3796 btrfs_release_path(path);
3798 ret = btrfs_del_root_ref(trans, objectid,
3799 root->root_key.objectid, dir_ino,
3800 &index, name, name_len);
3802 btrfs_abort_transaction(trans, ret);
3807 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
3809 btrfs_abort_transaction(trans, ret);
3813 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
3814 inode_inc_iversion(dir);
3815 dir->i_mtime = dir->i_ctime = current_time(dir);
3816 ret = btrfs_update_inode_fallback(trans, root, dir);
3818 btrfs_abort_transaction(trans, ret);
3820 btrfs_free_path(path);
3825 * Helper to check if the subvolume references other subvolumes or if it's
3828 static noinline int may_destroy_subvol(struct btrfs_root *root)
3830 struct btrfs_fs_info *fs_info = root->fs_info;
3831 struct btrfs_path *path;
3832 struct btrfs_dir_item *di;
3833 struct btrfs_key key;
3837 path = btrfs_alloc_path();
3841 /* Make sure this root isn't set as the default subvol */
3842 dir_id = btrfs_super_root_dir(fs_info->super_copy);
3843 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
3844 dir_id, "default", 7, 0);
3845 if (di && !IS_ERR(di)) {
3846 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
3847 if (key.objectid == root->root_key.objectid) {
3850 "deleting default subvolume %llu is not allowed",
3854 btrfs_release_path(path);
3857 key.objectid = root->root_key.objectid;
3858 key.type = BTRFS_ROOT_REF_KEY;
3859 key.offset = (u64)-1;
3861 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3867 if (path->slots[0] > 0) {
3869 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3870 if (key.objectid == root->root_key.objectid &&
3871 key.type == BTRFS_ROOT_REF_KEY)
3875 btrfs_free_path(path);
3879 /* Delete all dentries for inodes belonging to the root */
3880 static void btrfs_prune_dentries(struct btrfs_root *root)
3882 struct btrfs_fs_info *fs_info = root->fs_info;
3883 struct rb_node *node;
3884 struct rb_node *prev;
3885 struct btrfs_inode *entry;
3886 struct inode *inode;
3889 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3890 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3892 spin_lock(&root->inode_lock);
3894 node = root->inode_tree.rb_node;
3898 entry = rb_entry(node, struct btrfs_inode, rb_node);
3900 if (objectid < btrfs_ino(entry))
3901 node = node->rb_left;
3902 else if (objectid > btrfs_ino(entry))
3903 node = node->rb_right;
3909 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3910 if (objectid <= btrfs_ino(entry)) {
3914 prev = rb_next(prev);
3918 entry = rb_entry(node, struct btrfs_inode, rb_node);
3919 objectid = btrfs_ino(entry) + 1;
3920 inode = igrab(&entry->vfs_inode);
3922 spin_unlock(&root->inode_lock);
3923 if (atomic_read(&inode->i_count) > 1)
3924 d_prune_aliases(inode);
3926 * btrfs_drop_inode will have it removed from the inode
3927 * cache when its usage count hits zero.
3931 spin_lock(&root->inode_lock);
3935 if (cond_resched_lock(&root->inode_lock))
3938 node = rb_next(node);
3940 spin_unlock(&root->inode_lock);
3943 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
3945 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
3946 struct btrfs_root *root = BTRFS_I(dir)->root;
3947 struct inode *inode = d_inode(dentry);
3948 struct btrfs_root *dest = BTRFS_I(inode)->root;
3949 struct btrfs_trans_handle *trans;
3950 struct btrfs_block_rsv block_rsv;
3956 * Don't allow to delete a subvolume with send in progress. This is
3957 * inside the inode lock so the error handling that has to drop the bit
3958 * again is not run concurrently.
3960 spin_lock(&dest->root_item_lock);
3961 if (dest->send_in_progress) {
3962 spin_unlock(&dest->root_item_lock);
3964 "attempt to delete subvolume %llu during send",
3965 dest->root_key.objectid);
3968 root_flags = btrfs_root_flags(&dest->root_item);
3969 btrfs_set_root_flags(&dest->root_item,
3970 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
3971 spin_unlock(&dest->root_item_lock);
3973 down_write(&fs_info->subvol_sem);
3975 err = may_destroy_subvol(dest);
3979 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
3981 * One for dir inode,
3982 * two for dir entries,
3983 * two for root ref/backref.
3985 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
3989 trans = btrfs_start_transaction(root, 0);
3990 if (IS_ERR(trans)) {
3991 err = PTR_ERR(trans);
3994 trans->block_rsv = &block_rsv;
3995 trans->bytes_reserved = block_rsv.size;
3997 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
3999 ret = btrfs_unlink_subvol(trans, dir, dentry);
4002 btrfs_abort_transaction(trans, ret);
4006 btrfs_record_root_in_trans(trans, dest);
4008 memset(&dest->root_item.drop_progress, 0,
4009 sizeof(dest->root_item.drop_progress));
4010 dest->root_item.drop_level = 0;
4011 btrfs_set_root_refs(&dest->root_item, 0);
4013 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4014 ret = btrfs_insert_orphan_item(trans,
4016 dest->root_key.objectid);
4018 btrfs_abort_transaction(trans, ret);
4024 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4025 BTRFS_UUID_KEY_SUBVOL,
4026 dest->root_key.objectid);
4027 if (ret && ret != -ENOENT) {
4028 btrfs_abort_transaction(trans, ret);
4032 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4033 ret = btrfs_uuid_tree_remove(trans,
4034 dest->root_item.received_uuid,
4035 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4036 dest->root_key.objectid);
4037 if (ret && ret != -ENOENT) {
4038 btrfs_abort_transaction(trans, ret);
4044 free_anon_bdev(dest->anon_dev);
4047 trans->block_rsv = NULL;
4048 trans->bytes_reserved = 0;
4049 ret = btrfs_end_transaction(trans);
4052 inode->i_flags |= S_DEAD;
4054 btrfs_subvolume_release_metadata(root, &block_rsv);
4056 up_write(&fs_info->subvol_sem);
4058 spin_lock(&dest->root_item_lock);
4059 root_flags = btrfs_root_flags(&dest->root_item);
4060 btrfs_set_root_flags(&dest->root_item,
4061 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4062 spin_unlock(&dest->root_item_lock);
4064 d_invalidate(dentry);
4065 btrfs_prune_dentries(dest);
4066 ASSERT(dest->send_in_progress == 0);
4069 if (dest->ino_cache_inode) {
4070 iput(dest->ino_cache_inode);
4071 dest->ino_cache_inode = NULL;
4078 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4080 struct inode *inode = d_inode(dentry);
4082 struct btrfs_root *root = BTRFS_I(dir)->root;
4083 struct btrfs_trans_handle *trans;
4084 u64 last_unlink_trans;
4086 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4088 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4089 return btrfs_delete_subvolume(dir, dentry);
4091 trans = __unlink_start_trans(dir);
4093 return PTR_ERR(trans);
4095 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4096 err = btrfs_unlink_subvol(trans, dir, dentry);
4100 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4104 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4106 /* now the directory is empty */
4107 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4108 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4109 dentry->d_name.len);
4111 btrfs_i_size_write(BTRFS_I(inode), 0);
4113 * Propagate the last_unlink_trans value of the deleted dir to
4114 * its parent directory. This is to prevent an unrecoverable
4115 * log tree in the case we do something like this:
4117 * 2) create snapshot under dir foo
4118 * 3) delete the snapshot
4121 * 6) fsync foo or some file inside foo
4123 if (last_unlink_trans >= trans->transid)
4124 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4127 btrfs_end_transaction(trans);
4128 btrfs_btree_balance_dirty(root->fs_info);
4134 * Return this if we need to call truncate_block for the last bit of the
4137 #define NEED_TRUNCATE_BLOCK 1
4140 * this can truncate away extent items, csum items and directory items.
4141 * It starts at a high offset and removes keys until it can't find
4142 * any higher than new_size
4144 * csum items that cross the new i_size are truncated to the new size
4147 * min_type is the minimum key type to truncate down to. If set to 0, this
4148 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4150 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4151 struct btrfs_root *root,
4152 struct inode *inode,
4153 u64 new_size, u32 min_type)
4155 struct btrfs_fs_info *fs_info = root->fs_info;
4156 struct btrfs_path *path;
4157 struct extent_buffer *leaf;
4158 struct btrfs_file_extent_item *fi;
4159 struct btrfs_key key;
4160 struct btrfs_key found_key;
4161 u64 extent_start = 0;
4162 u64 extent_num_bytes = 0;
4163 u64 extent_offset = 0;
4165 u64 last_size = new_size;
4166 u32 found_type = (u8)-1;
4169 int pending_del_nr = 0;
4170 int pending_del_slot = 0;
4171 int extent_type = -1;
4173 u64 ino = btrfs_ino(BTRFS_I(inode));
4174 u64 bytes_deleted = 0;
4175 bool be_nice = false;
4176 bool should_throttle = false;
4177 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4178 struct extent_state *cached_state = NULL;
4180 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4183 * For non-free space inodes and non-shareable roots, we want to back
4184 * off from time to time. This means all inodes in subvolume roots,
4185 * reloc roots, and data reloc roots.
4187 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4188 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4191 path = btrfs_alloc_path();
4194 path->reada = READA_BACK;
4196 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4197 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, (u64)-1,
4201 * We want to drop from the next block forward in case this
4202 * new size is not block aligned since we will be keeping the
4203 * last block of the extent just the way it is.
4205 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4206 fs_info->sectorsize),
4211 * This function is also used to drop the items in the log tree before
4212 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4213 * it is used to drop the logged items. So we shouldn't kill the delayed
4216 if (min_type == 0 && root == BTRFS_I(inode)->root)
4217 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4220 key.offset = (u64)-1;
4225 * with a 16K leaf size and 128MB extents, you can actually queue
4226 * up a huge file in a single leaf. Most of the time that
4227 * bytes_deleted is > 0, it will be huge by the time we get here
4229 if (be_nice && bytes_deleted > SZ_32M &&
4230 btrfs_should_end_transaction(trans)) {
4235 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4241 /* there are no items in the tree for us to truncate, we're
4244 if (path->slots[0] == 0)
4250 u64 clear_start = 0, clear_len = 0;
4253 leaf = path->nodes[0];
4254 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4255 found_type = found_key.type;
4257 if (found_key.objectid != ino)
4260 if (found_type < min_type)
4263 item_end = found_key.offset;
4264 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4265 fi = btrfs_item_ptr(leaf, path->slots[0],
4266 struct btrfs_file_extent_item);
4267 extent_type = btrfs_file_extent_type(leaf, fi);
4268 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4270 btrfs_file_extent_num_bytes(leaf, fi);
4272 trace_btrfs_truncate_show_fi_regular(
4273 BTRFS_I(inode), leaf, fi,
4275 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4276 item_end += btrfs_file_extent_ram_bytes(leaf,
4279 trace_btrfs_truncate_show_fi_inline(
4280 BTRFS_I(inode), leaf, fi, path->slots[0],
4285 if (found_type > min_type) {
4288 if (item_end < new_size)
4290 if (found_key.offset >= new_size)
4296 /* FIXME, shrink the extent if the ref count is only 1 */
4297 if (found_type != BTRFS_EXTENT_DATA_KEY)
4300 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4303 clear_start = found_key.offset;
4304 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4306 u64 orig_num_bytes =
4307 btrfs_file_extent_num_bytes(leaf, fi);
4308 extent_num_bytes = ALIGN(new_size -
4310 fs_info->sectorsize);
4311 clear_start = ALIGN(new_size, fs_info->sectorsize);
4312 btrfs_set_file_extent_num_bytes(leaf, fi,
4314 num_dec = (orig_num_bytes -
4316 if (test_bit(BTRFS_ROOT_SHAREABLE,
4319 inode_sub_bytes(inode, num_dec);
4320 btrfs_mark_buffer_dirty(leaf);
4323 btrfs_file_extent_disk_num_bytes(leaf,
4325 extent_offset = found_key.offset -
4326 btrfs_file_extent_offset(leaf, fi);
4328 /* FIXME blocksize != 4096 */
4329 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4330 if (extent_start != 0) {
4332 if (test_bit(BTRFS_ROOT_SHAREABLE,
4334 inode_sub_bytes(inode, num_dec);
4337 clear_len = num_dec;
4338 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4340 * we can't truncate inline items that have had
4344 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4345 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4346 btrfs_file_extent_compression(leaf, fi) == 0) {
4347 u32 size = (u32)(new_size - found_key.offset);
4349 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4350 size = btrfs_file_extent_calc_inline_size(size);
4351 btrfs_truncate_item(path, size, 1);
4352 } else if (!del_item) {
4354 * We have to bail so the last_size is set to
4355 * just before this extent.
4357 ret = NEED_TRUNCATE_BLOCK;
4361 * Inline extents are special, we just treat
4362 * them as a full sector worth in the file
4363 * extent tree just for simplicity sake.
4365 clear_len = fs_info->sectorsize;
4368 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4369 inode_sub_bytes(inode, item_end + 1 - new_size);
4373 * We use btrfs_truncate_inode_items() to clean up log trees for
4374 * multiple fsyncs, and in this case we don't want to clear the
4375 * file extent range because it's just the log.
4377 if (root == BTRFS_I(inode)->root) {
4378 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
4379 clear_start, clear_len);
4381 btrfs_abort_transaction(trans, ret);
4387 last_size = found_key.offset;
4389 last_size = new_size;
4391 if (!pending_del_nr) {
4392 /* no pending yet, add ourselves */
4393 pending_del_slot = path->slots[0];
4395 } else if (pending_del_nr &&
4396 path->slots[0] + 1 == pending_del_slot) {
4397 /* hop on the pending chunk */
4399 pending_del_slot = path->slots[0];
4406 should_throttle = false;
4409 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4410 struct btrfs_ref ref = { 0 };
4412 bytes_deleted += extent_num_bytes;
4414 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4415 extent_start, extent_num_bytes, 0);
4416 ref.real_root = root->root_key.objectid;
4417 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4418 ino, extent_offset);
4419 ret = btrfs_free_extent(trans, &ref);
4421 btrfs_abort_transaction(trans, ret);
4425 if (btrfs_should_throttle_delayed_refs(trans))
4426 should_throttle = true;
4430 if (found_type == BTRFS_INODE_ITEM_KEY)
4433 if (path->slots[0] == 0 ||
4434 path->slots[0] != pending_del_slot ||
4436 if (pending_del_nr) {
4437 ret = btrfs_del_items(trans, root, path,
4441 btrfs_abort_transaction(trans, ret);
4446 btrfs_release_path(path);
4449 * We can generate a lot of delayed refs, so we need to
4450 * throttle every once and a while and make sure we're
4451 * adding enough space to keep up with the work we are
4452 * generating. Since we hold a transaction here we
4453 * can't flush, and we don't want to FLUSH_LIMIT because
4454 * we could have generated too many delayed refs to
4455 * actually allocate, so just bail if we're short and
4456 * let the normal reservation dance happen higher up.
4458 if (should_throttle) {
4459 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4460 BTRFS_RESERVE_NO_FLUSH);
4472 if (ret >= 0 && pending_del_nr) {
4475 err = btrfs_del_items(trans, root, path, pending_del_slot,
4478 btrfs_abort_transaction(trans, err);
4482 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4483 ASSERT(last_size >= new_size);
4484 if (!ret && last_size > new_size)
4485 last_size = new_size;
4486 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4487 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start,
4488 (u64)-1, &cached_state);
4491 btrfs_free_path(path);
4496 * btrfs_truncate_block - read, zero a chunk and write a block
4497 * @inode - inode that we're zeroing
4498 * @from - the offset to start zeroing
4499 * @len - the length to zero, 0 to zero the entire range respective to the
4501 * @front - zero up to the offset instead of from the offset on
4503 * This will find the block for the "from" offset and cow the block and zero the
4504 * part we want to zero. This is used with truncate and hole punching.
4506 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4509 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4510 struct address_space *mapping = inode->i_mapping;
4511 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4512 struct btrfs_ordered_extent *ordered;
4513 struct extent_state *cached_state = NULL;
4514 struct extent_changeset *data_reserved = NULL;
4516 bool only_release_metadata = false;
4517 u32 blocksize = fs_info->sectorsize;
4518 pgoff_t index = from >> PAGE_SHIFT;
4519 unsigned offset = from & (blocksize - 1);
4521 gfp_t mask = btrfs_alloc_write_mask(mapping);
4522 size_t write_bytes = blocksize;
4527 if (IS_ALIGNED(offset, blocksize) &&
4528 (!len || IS_ALIGNED(len, blocksize)))
4531 block_start = round_down(from, blocksize);
4532 block_end = block_start + blocksize - 1;
4534 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved,
4535 block_start, blocksize);
4537 if (btrfs_check_nocow_lock(BTRFS_I(inode), block_start,
4538 &write_bytes) > 0) {
4539 /* For nocow case, no need to reserve data space */
4540 only_release_metadata = true;
4545 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), blocksize);
4547 if (!only_release_metadata)
4548 btrfs_free_reserved_data_space(BTRFS_I(inode),
4549 data_reserved, block_start, blocksize);
4553 page = find_or_create_page(mapping, index, mask);
4555 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4556 block_start, blocksize, true);
4557 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4562 if (!PageUptodate(page)) {
4563 ret = btrfs_readpage(NULL, page);
4565 if (page->mapping != mapping) {
4570 if (!PageUptodate(page)) {
4575 wait_on_page_writeback(page);
4577 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4578 set_page_extent_mapped(page);
4580 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), block_start);
4582 unlock_extent_cached(io_tree, block_start, block_end,
4586 btrfs_start_ordered_extent(inode, ordered, 1);
4587 btrfs_put_ordered_extent(ordered);
4591 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4592 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4593 0, 0, &cached_state);
4595 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), block_start, block_end, 0,
4598 unlock_extent_cached(io_tree, block_start, block_end,
4603 if (offset != blocksize) {
4605 len = blocksize - offset;
4608 memset(kaddr + (block_start - page_offset(page)),
4611 memset(kaddr + (block_start - page_offset(page)) + offset,
4613 flush_dcache_page(page);
4616 ClearPageChecked(page);
4617 set_page_dirty(page);
4618 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4620 if (only_release_metadata)
4621 set_extent_bit(&BTRFS_I(inode)->io_tree, block_start,
4622 block_end, EXTENT_NORESERVE, NULL, NULL,
4627 if (only_release_metadata)
4628 btrfs_delalloc_release_metadata(BTRFS_I(inode),
4631 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4632 block_start, blocksize, true);
4634 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4638 if (only_release_metadata)
4639 btrfs_check_nocow_unlock(BTRFS_I(inode));
4640 extent_changeset_free(data_reserved);
4644 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4645 u64 offset, u64 len)
4647 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4648 struct btrfs_trans_handle *trans;
4652 * Still need to make sure the inode looks like it's been updated so
4653 * that any holes get logged if we fsync.
4655 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4656 BTRFS_I(inode)->last_trans = fs_info->generation;
4657 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4658 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4663 * 1 - for the one we're dropping
4664 * 1 - for the one we're adding
4665 * 1 - for updating the inode.
4667 trans = btrfs_start_transaction(root, 3);
4669 return PTR_ERR(trans);
4671 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4673 btrfs_abort_transaction(trans, ret);
4674 btrfs_end_transaction(trans);
4678 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4679 offset, 0, 0, len, 0, len, 0, 0, 0);
4681 btrfs_abort_transaction(trans, ret);
4683 btrfs_update_inode(trans, root, inode);
4684 btrfs_end_transaction(trans);
4689 * This function puts in dummy file extents for the area we're creating a hole
4690 * for. So if we are truncating this file to a larger size we need to insert
4691 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4692 * the range between oldsize and size
4694 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4696 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4697 struct btrfs_root *root = BTRFS_I(inode)->root;
4698 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4699 struct extent_map *em = NULL;
4700 struct extent_state *cached_state = NULL;
4701 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4702 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4703 u64 block_end = ALIGN(size, fs_info->sectorsize);
4710 * If our size started in the middle of a block we need to zero out the
4711 * rest of the block before we expand the i_size, otherwise we could
4712 * expose stale data.
4714 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4718 if (size <= hole_start)
4721 btrfs_lock_and_flush_ordered_range(BTRFS_I(inode), hole_start,
4722 block_end - 1, &cached_state);
4723 cur_offset = hole_start;
4725 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4726 block_end - cur_offset);
4732 last_byte = min(extent_map_end(em), block_end);
4733 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4734 hole_size = last_byte - cur_offset;
4736 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4737 struct extent_map *hole_em;
4739 err = maybe_insert_hole(root, inode, cur_offset,
4744 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4745 cur_offset, hole_size);
4749 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4750 cur_offset + hole_size - 1, 0);
4751 hole_em = alloc_extent_map();
4753 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4754 &BTRFS_I(inode)->runtime_flags);
4757 hole_em->start = cur_offset;
4758 hole_em->len = hole_size;
4759 hole_em->orig_start = cur_offset;
4761 hole_em->block_start = EXTENT_MAP_HOLE;
4762 hole_em->block_len = 0;
4763 hole_em->orig_block_len = 0;
4764 hole_em->ram_bytes = hole_size;
4765 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4766 hole_em->generation = fs_info->generation;
4769 write_lock(&em_tree->lock);
4770 err = add_extent_mapping(em_tree, hole_em, 1);
4771 write_unlock(&em_tree->lock);
4774 btrfs_drop_extent_cache(BTRFS_I(inode),
4779 free_extent_map(hole_em);
4781 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4782 cur_offset, hole_size);
4787 free_extent_map(em);
4789 cur_offset = last_byte;
4790 if (cur_offset >= block_end)
4793 free_extent_map(em);
4794 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4798 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4800 struct btrfs_root *root = BTRFS_I(inode)->root;
4801 struct btrfs_trans_handle *trans;
4802 loff_t oldsize = i_size_read(inode);
4803 loff_t newsize = attr->ia_size;
4804 int mask = attr->ia_valid;
4808 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4809 * special case where we need to update the times despite not having
4810 * these flags set. For all other operations the VFS set these flags
4811 * explicitly if it wants a timestamp update.
4813 if (newsize != oldsize) {
4814 inode_inc_iversion(inode);
4815 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4816 inode->i_ctime = inode->i_mtime =
4817 current_time(inode);
4820 if (newsize > oldsize) {
4822 * Don't do an expanding truncate while snapshotting is ongoing.
4823 * This is to ensure the snapshot captures a fully consistent
4824 * state of this file - if the snapshot captures this expanding
4825 * truncation, it must capture all writes that happened before
4828 btrfs_drew_write_lock(&root->snapshot_lock);
4829 ret = btrfs_cont_expand(inode, oldsize, newsize);
4831 btrfs_drew_write_unlock(&root->snapshot_lock);
4835 trans = btrfs_start_transaction(root, 1);
4836 if (IS_ERR(trans)) {
4837 btrfs_drew_write_unlock(&root->snapshot_lock);
4838 return PTR_ERR(trans);
4841 i_size_write(inode, newsize);
4842 btrfs_inode_safe_disk_i_size_write(inode, 0);
4843 pagecache_isize_extended(inode, oldsize, newsize);
4844 ret = btrfs_update_inode(trans, root, inode);
4845 btrfs_drew_write_unlock(&root->snapshot_lock);
4846 btrfs_end_transaction(trans);
4850 * We're truncating a file that used to have good data down to
4851 * zero. Make sure it gets into the ordered flush list so that
4852 * any new writes get down to disk quickly.
4855 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4856 &BTRFS_I(inode)->runtime_flags);
4858 truncate_setsize(inode, newsize);
4860 /* Disable nonlocked read DIO to avoid the endless truncate */
4861 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
4862 inode_dio_wait(inode);
4863 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
4865 ret = btrfs_truncate(inode, newsize == oldsize);
4866 if (ret && inode->i_nlink) {
4870 * Truncate failed, so fix up the in-memory size. We
4871 * adjusted disk_i_size down as we removed extents, so
4872 * wait for disk_i_size to be stable and then update the
4873 * in-memory size to match.
4875 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
4878 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4885 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4887 struct inode *inode = d_inode(dentry);
4888 struct btrfs_root *root = BTRFS_I(inode)->root;
4891 if (btrfs_root_readonly(root))
4894 err = setattr_prepare(dentry, attr);
4898 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4899 err = btrfs_setsize(inode, attr);
4904 if (attr->ia_valid) {
4905 setattr_copy(inode, attr);
4906 inode_inc_iversion(inode);
4907 err = btrfs_dirty_inode(inode);
4909 if (!err && attr->ia_valid & ATTR_MODE)
4910 err = posix_acl_chmod(inode, inode->i_mode);
4917 * While truncating the inode pages during eviction, we get the VFS calling
4918 * btrfs_invalidatepage() against each page of the inode. This is slow because
4919 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4920 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4921 * extent_state structures over and over, wasting lots of time.
4923 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4924 * those expensive operations on a per page basis and do only the ordered io
4925 * finishing, while we release here the extent_map and extent_state structures,
4926 * without the excessive merging and splitting.
4928 static void evict_inode_truncate_pages(struct inode *inode)
4930 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4931 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4932 struct rb_node *node;
4934 ASSERT(inode->i_state & I_FREEING);
4935 truncate_inode_pages_final(&inode->i_data);
4937 write_lock(&map_tree->lock);
4938 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
4939 struct extent_map *em;
4941 node = rb_first_cached(&map_tree->map);
4942 em = rb_entry(node, struct extent_map, rb_node);
4943 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4944 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4945 remove_extent_mapping(map_tree, em);
4946 free_extent_map(em);
4947 if (need_resched()) {
4948 write_unlock(&map_tree->lock);
4950 write_lock(&map_tree->lock);
4953 write_unlock(&map_tree->lock);
4956 * Keep looping until we have no more ranges in the io tree.
4957 * We can have ongoing bios started by readahead that have
4958 * their endio callback (extent_io.c:end_bio_extent_readpage)
4959 * still in progress (unlocked the pages in the bio but did not yet
4960 * unlocked the ranges in the io tree). Therefore this means some
4961 * ranges can still be locked and eviction started because before
4962 * submitting those bios, which are executed by a separate task (work
4963 * queue kthread), inode references (inode->i_count) were not taken
4964 * (which would be dropped in the end io callback of each bio).
4965 * Therefore here we effectively end up waiting for those bios and
4966 * anyone else holding locked ranges without having bumped the inode's
4967 * reference count - if we don't do it, when they access the inode's
4968 * io_tree to unlock a range it may be too late, leading to an
4969 * use-after-free issue.
4971 spin_lock(&io_tree->lock);
4972 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4973 struct extent_state *state;
4974 struct extent_state *cached_state = NULL;
4977 unsigned state_flags;
4979 node = rb_first(&io_tree->state);
4980 state = rb_entry(node, struct extent_state, rb_node);
4981 start = state->start;
4983 state_flags = state->state;
4984 spin_unlock(&io_tree->lock);
4986 lock_extent_bits(io_tree, start, end, &cached_state);
4989 * If still has DELALLOC flag, the extent didn't reach disk,
4990 * and its reserved space won't be freed by delayed_ref.
4991 * So we need to free its reserved space here.
4992 * (Refer to comment in btrfs_invalidatepage, case 2)
4994 * Note, end is the bytenr of last byte, so we need + 1 here.
4996 if (state_flags & EXTENT_DELALLOC)
4997 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5000 clear_extent_bit(io_tree, start, end,
5001 EXTENT_LOCKED | EXTENT_DELALLOC |
5002 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5006 spin_lock(&io_tree->lock);
5008 spin_unlock(&io_tree->lock);
5011 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5012 struct btrfs_block_rsv *rsv)
5014 struct btrfs_fs_info *fs_info = root->fs_info;
5015 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5016 struct btrfs_trans_handle *trans;
5017 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5021 * Eviction should be taking place at some place safe because of our
5022 * delayed iputs. However the normal flushing code will run delayed
5023 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5025 * We reserve the delayed_refs_extra here again because we can't use
5026 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5027 * above. We reserve our extra bit here because we generate a ton of
5028 * delayed refs activity by truncating.
5030 * If we cannot make our reservation we'll attempt to steal from the
5031 * global reserve, because we really want to be able to free up space.
5033 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5034 BTRFS_RESERVE_FLUSH_EVICT);
5037 * Try to steal from the global reserve if there is space for
5040 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5041 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5043 "could not allocate space for delete; will truncate on mount");
5044 return ERR_PTR(-ENOSPC);
5046 delayed_refs_extra = 0;
5049 trans = btrfs_join_transaction(root);
5053 if (delayed_refs_extra) {
5054 trans->block_rsv = &fs_info->trans_block_rsv;
5055 trans->bytes_reserved = delayed_refs_extra;
5056 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5057 delayed_refs_extra, 1);
5062 void btrfs_evict_inode(struct inode *inode)
5064 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5065 struct btrfs_trans_handle *trans;
5066 struct btrfs_root *root = BTRFS_I(inode)->root;
5067 struct btrfs_block_rsv *rsv;
5070 trace_btrfs_inode_evict(inode);
5077 evict_inode_truncate_pages(inode);
5079 if (inode->i_nlink &&
5080 ((btrfs_root_refs(&root->root_item) != 0 &&
5081 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5082 btrfs_is_free_space_inode(BTRFS_I(inode))))
5085 if (is_bad_inode(inode))
5088 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5090 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5093 if (inode->i_nlink > 0) {
5094 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5095 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5099 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5103 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5106 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5109 btrfs_i_size_write(BTRFS_I(inode), 0);
5112 trans = evict_refill_and_join(root, rsv);
5116 trans->block_rsv = rsv;
5118 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5119 trans->block_rsv = &fs_info->trans_block_rsv;
5120 btrfs_end_transaction(trans);
5121 btrfs_btree_balance_dirty(fs_info);
5122 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5129 * Errors here aren't a big deal, it just means we leave orphan items in
5130 * the tree. They will be cleaned up on the next mount. If the inode
5131 * number gets reused, cleanup deletes the orphan item without doing
5132 * anything, and unlink reuses the existing orphan item.
5134 * If it turns out that we are dropping too many of these, we might want
5135 * to add a mechanism for retrying these after a commit.
5137 trans = evict_refill_and_join(root, rsv);
5138 if (!IS_ERR(trans)) {
5139 trans->block_rsv = rsv;
5140 btrfs_orphan_del(trans, BTRFS_I(inode));
5141 trans->block_rsv = &fs_info->trans_block_rsv;
5142 btrfs_end_transaction(trans);
5145 if (!(root == fs_info->tree_root ||
5146 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5147 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5150 btrfs_free_block_rsv(fs_info, rsv);
5153 * If we didn't successfully delete, the orphan item will still be in
5154 * the tree and we'll retry on the next mount. Again, we might also want
5155 * to retry these periodically in the future.
5157 btrfs_remove_delayed_node(BTRFS_I(inode));
5162 * Return the key found in the dir entry in the location pointer, fill @type
5163 * with BTRFS_FT_*, and return 0.
5165 * If no dir entries were found, returns -ENOENT.
5166 * If found a corrupted location in dir entry, returns -EUCLEAN.
5168 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5169 struct btrfs_key *location, u8 *type)
5171 const char *name = dentry->d_name.name;
5172 int namelen = dentry->d_name.len;
5173 struct btrfs_dir_item *di;
5174 struct btrfs_path *path;
5175 struct btrfs_root *root = BTRFS_I(dir)->root;
5178 path = btrfs_alloc_path();
5182 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5184 if (IS_ERR_OR_NULL(di)) {
5185 ret = di ? PTR_ERR(di) : -ENOENT;
5189 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5190 if (location->type != BTRFS_INODE_ITEM_KEY &&
5191 location->type != BTRFS_ROOT_ITEM_KEY) {
5193 btrfs_warn(root->fs_info,
5194 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5195 __func__, name, btrfs_ino(BTRFS_I(dir)),
5196 location->objectid, location->type, location->offset);
5199 *type = btrfs_dir_type(path->nodes[0], di);
5201 btrfs_free_path(path);
5206 * when we hit a tree root in a directory, the btrfs part of the inode
5207 * needs to be changed to reflect the root directory of the tree root. This
5208 * is kind of like crossing a mount point.
5210 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5212 struct dentry *dentry,
5213 struct btrfs_key *location,
5214 struct btrfs_root **sub_root)
5216 struct btrfs_path *path;
5217 struct btrfs_root *new_root;
5218 struct btrfs_root_ref *ref;
5219 struct extent_buffer *leaf;
5220 struct btrfs_key key;
5224 path = btrfs_alloc_path();
5231 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5232 key.type = BTRFS_ROOT_REF_KEY;
5233 key.offset = location->objectid;
5235 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5242 leaf = path->nodes[0];
5243 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5244 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5245 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5248 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5249 (unsigned long)(ref + 1),
5250 dentry->d_name.len);
5254 btrfs_release_path(path);
5256 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5257 if (IS_ERR(new_root)) {
5258 err = PTR_ERR(new_root);
5262 *sub_root = new_root;
5263 location->objectid = btrfs_root_dirid(&new_root->root_item);
5264 location->type = BTRFS_INODE_ITEM_KEY;
5265 location->offset = 0;
5268 btrfs_free_path(path);
5272 static void inode_tree_add(struct inode *inode)
5274 struct btrfs_root *root = BTRFS_I(inode)->root;
5275 struct btrfs_inode *entry;
5277 struct rb_node *parent;
5278 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5279 u64 ino = btrfs_ino(BTRFS_I(inode));
5281 if (inode_unhashed(inode))
5284 spin_lock(&root->inode_lock);
5285 p = &root->inode_tree.rb_node;
5288 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5290 if (ino < btrfs_ino(entry))
5291 p = &parent->rb_left;
5292 else if (ino > btrfs_ino(entry))
5293 p = &parent->rb_right;
5295 WARN_ON(!(entry->vfs_inode.i_state &
5296 (I_WILL_FREE | I_FREEING)));
5297 rb_replace_node(parent, new, &root->inode_tree);
5298 RB_CLEAR_NODE(parent);
5299 spin_unlock(&root->inode_lock);
5303 rb_link_node(new, parent, p);
5304 rb_insert_color(new, &root->inode_tree);
5305 spin_unlock(&root->inode_lock);
5308 static void inode_tree_del(struct btrfs_inode *inode)
5310 struct btrfs_root *root = inode->root;
5313 spin_lock(&root->inode_lock);
5314 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5315 rb_erase(&inode->rb_node, &root->inode_tree);
5316 RB_CLEAR_NODE(&inode->rb_node);
5317 empty = RB_EMPTY_ROOT(&root->inode_tree);
5319 spin_unlock(&root->inode_lock);
5321 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5322 spin_lock(&root->inode_lock);
5323 empty = RB_EMPTY_ROOT(&root->inode_tree);
5324 spin_unlock(&root->inode_lock);
5326 btrfs_add_dead_root(root);
5331 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5333 struct btrfs_iget_args *args = p;
5335 inode->i_ino = args->ino;
5336 BTRFS_I(inode)->location.objectid = args->ino;
5337 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5338 BTRFS_I(inode)->location.offset = 0;
5339 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5340 BUG_ON(args->root && !BTRFS_I(inode)->root);
5344 static int btrfs_find_actor(struct inode *inode, void *opaque)
5346 struct btrfs_iget_args *args = opaque;
5348 return args->ino == BTRFS_I(inode)->location.objectid &&
5349 args->root == BTRFS_I(inode)->root;
5352 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5353 struct btrfs_root *root)
5355 struct inode *inode;
5356 struct btrfs_iget_args args;
5357 unsigned long hashval = btrfs_inode_hash(ino, root);
5362 inode = iget5_locked(s, hashval, btrfs_find_actor,
5363 btrfs_init_locked_inode,
5369 * Get an inode object given its inode number and corresponding root.
5370 * Path can be preallocated to prevent recursing back to iget through
5371 * allocator. NULL is also valid but may require an additional allocation
5374 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5375 struct btrfs_root *root, struct btrfs_path *path)
5377 struct inode *inode;
5379 inode = btrfs_iget_locked(s, ino, root);
5381 return ERR_PTR(-ENOMEM);
5383 if (inode->i_state & I_NEW) {
5386 ret = btrfs_read_locked_inode(inode, path);
5388 inode_tree_add(inode);
5389 unlock_new_inode(inode);
5393 * ret > 0 can come from btrfs_search_slot called by
5394 * btrfs_read_locked_inode, this means the inode item
5399 inode = ERR_PTR(ret);
5406 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5408 return btrfs_iget_path(s, ino, root, NULL);
5411 static struct inode *new_simple_dir(struct super_block *s,
5412 struct btrfs_key *key,
5413 struct btrfs_root *root)
5415 struct inode *inode = new_inode(s);
5418 return ERR_PTR(-ENOMEM);
5420 BTRFS_I(inode)->root = btrfs_grab_root(root);
5421 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5422 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5424 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5426 * We only need lookup, the rest is read-only and there's no inode
5427 * associated with the dentry
5429 inode->i_op = &simple_dir_inode_operations;
5430 inode->i_opflags &= ~IOP_XATTR;
5431 inode->i_fop = &simple_dir_operations;
5432 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5433 inode->i_mtime = current_time(inode);
5434 inode->i_atime = inode->i_mtime;
5435 inode->i_ctime = inode->i_mtime;
5436 BTRFS_I(inode)->i_otime = inode->i_mtime;
5441 static inline u8 btrfs_inode_type(struct inode *inode)
5444 * Compile-time asserts that generic FT_* types still match
5447 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5448 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5449 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5450 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5451 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5452 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5453 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5454 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5456 return fs_umode_to_ftype(inode->i_mode);
5459 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5461 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5462 struct inode *inode;
5463 struct btrfs_root *root = BTRFS_I(dir)->root;
5464 struct btrfs_root *sub_root = root;
5465 struct btrfs_key location;
5469 if (dentry->d_name.len > BTRFS_NAME_LEN)
5470 return ERR_PTR(-ENAMETOOLONG);
5472 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5474 return ERR_PTR(ret);
5476 if (location.type == BTRFS_INODE_ITEM_KEY) {
5477 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5481 /* Do extra check against inode mode with di_type */
5482 if (btrfs_inode_type(inode) != di_type) {
5484 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5485 inode->i_mode, btrfs_inode_type(inode),
5488 return ERR_PTR(-EUCLEAN);
5493 ret = fixup_tree_root_location(fs_info, dir, dentry,
5494 &location, &sub_root);
5497 inode = ERR_PTR(ret);
5499 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5501 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5503 if (root != sub_root)
5504 btrfs_put_root(sub_root);
5506 if (!IS_ERR(inode) && root != sub_root) {
5507 down_read(&fs_info->cleanup_work_sem);
5508 if (!sb_rdonly(inode->i_sb))
5509 ret = btrfs_orphan_cleanup(sub_root);
5510 up_read(&fs_info->cleanup_work_sem);
5513 inode = ERR_PTR(ret);
5520 static int btrfs_dentry_delete(const struct dentry *dentry)
5522 struct btrfs_root *root;
5523 struct inode *inode = d_inode(dentry);
5525 if (!inode && !IS_ROOT(dentry))
5526 inode = d_inode(dentry->d_parent);
5529 root = BTRFS_I(inode)->root;
5530 if (btrfs_root_refs(&root->root_item) == 0)
5533 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5539 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5542 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5544 if (inode == ERR_PTR(-ENOENT))
5546 return d_splice_alias(inode, dentry);
5550 * All this infrastructure exists because dir_emit can fault, and we are holding
5551 * the tree lock when doing readdir. For now just allocate a buffer and copy
5552 * our information into that, and then dir_emit from the buffer. This is
5553 * similar to what NFS does, only we don't keep the buffer around in pagecache
5554 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5555 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5558 static int btrfs_opendir(struct inode *inode, struct file *file)
5560 struct btrfs_file_private *private;
5562 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5565 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5566 if (!private->filldir_buf) {
5570 file->private_data = private;
5581 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5584 struct dir_entry *entry = addr;
5585 char *name = (char *)(entry + 1);
5587 ctx->pos = get_unaligned(&entry->offset);
5588 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5589 get_unaligned(&entry->ino),
5590 get_unaligned(&entry->type)))
5592 addr += sizeof(struct dir_entry) +
5593 get_unaligned(&entry->name_len);
5599 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5601 struct inode *inode = file_inode(file);
5602 struct btrfs_root *root = BTRFS_I(inode)->root;
5603 struct btrfs_file_private *private = file->private_data;
5604 struct btrfs_dir_item *di;
5605 struct btrfs_key key;
5606 struct btrfs_key found_key;
5607 struct btrfs_path *path;
5609 struct list_head ins_list;
5610 struct list_head del_list;
5612 struct extent_buffer *leaf;
5619 struct btrfs_key location;
5621 if (!dir_emit_dots(file, ctx))
5624 path = btrfs_alloc_path();
5628 addr = private->filldir_buf;
5629 path->reada = READA_FORWARD;
5631 INIT_LIST_HEAD(&ins_list);
5632 INIT_LIST_HEAD(&del_list);
5633 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5636 key.type = BTRFS_DIR_INDEX_KEY;
5637 key.offset = ctx->pos;
5638 key.objectid = btrfs_ino(BTRFS_I(inode));
5640 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5645 struct dir_entry *entry;
5647 leaf = path->nodes[0];
5648 slot = path->slots[0];
5649 if (slot >= btrfs_header_nritems(leaf)) {
5650 ret = btrfs_next_leaf(root, path);
5658 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5660 if (found_key.objectid != key.objectid)
5662 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5664 if (found_key.offset < ctx->pos)
5666 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5668 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5669 name_len = btrfs_dir_name_len(leaf, di);
5670 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5672 btrfs_release_path(path);
5673 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5676 addr = private->filldir_buf;
5683 put_unaligned(name_len, &entry->name_len);
5684 name_ptr = (char *)(entry + 1);
5685 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5687 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5689 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5690 put_unaligned(location.objectid, &entry->ino);
5691 put_unaligned(found_key.offset, &entry->offset);
5693 addr += sizeof(struct dir_entry) + name_len;
5694 total_len += sizeof(struct dir_entry) + name_len;
5698 btrfs_release_path(path);
5700 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5704 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5709 * Stop new entries from being returned after we return the last
5712 * New directory entries are assigned a strictly increasing
5713 * offset. This means that new entries created during readdir
5714 * are *guaranteed* to be seen in the future by that readdir.
5715 * This has broken buggy programs which operate on names as
5716 * they're returned by readdir. Until we re-use freed offsets
5717 * we have this hack to stop new entries from being returned
5718 * under the assumption that they'll never reach this huge
5721 * This is being careful not to overflow 32bit loff_t unless the
5722 * last entry requires it because doing so has broken 32bit apps
5725 if (ctx->pos >= INT_MAX)
5726 ctx->pos = LLONG_MAX;
5733 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5734 btrfs_free_path(path);
5739 * This is somewhat expensive, updating the tree every time the
5740 * inode changes. But, it is most likely to find the inode in cache.
5741 * FIXME, needs more benchmarking...there are no reasons other than performance
5742 * to keep or drop this code.
5744 static int btrfs_dirty_inode(struct inode *inode)
5746 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5747 struct btrfs_root *root = BTRFS_I(inode)->root;
5748 struct btrfs_trans_handle *trans;
5751 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5754 trans = btrfs_join_transaction(root);
5756 return PTR_ERR(trans);
5758 ret = btrfs_update_inode(trans, root, inode);
5759 if (ret && ret == -ENOSPC) {
5760 /* whoops, lets try again with the full transaction */
5761 btrfs_end_transaction(trans);
5762 trans = btrfs_start_transaction(root, 1);
5764 return PTR_ERR(trans);
5766 ret = btrfs_update_inode(trans, root, inode);
5768 btrfs_end_transaction(trans);
5769 if (BTRFS_I(inode)->delayed_node)
5770 btrfs_balance_delayed_items(fs_info);
5776 * This is a copy of file_update_time. We need this so we can return error on
5777 * ENOSPC for updating the inode in the case of file write and mmap writes.
5779 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5782 struct btrfs_root *root = BTRFS_I(inode)->root;
5783 bool dirty = flags & ~S_VERSION;
5785 if (btrfs_root_readonly(root))
5788 if (flags & S_VERSION)
5789 dirty |= inode_maybe_inc_iversion(inode, dirty);
5790 if (flags & S_CTIME)
5791 inode->i_ctime = *now;
5792 if (flags & S_MTIME)
5793 inode->i_mtime = *now;
5794 if (flags & S_ATIME)
5795 inode->i_atime = *now;
5796 return dirty ? btrfs_dirty_inode(inode) : 0;
5800 * find the highest existing sequence number in a directory
5801 * and then set the in-memory index_cnt variable to reflect
5802 * free sequence numbers
5804 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5806 struct btrfs_root *root = inode->root;
5807 struct btrfs_key key, found_key;
5808 struct btrfs_path *path;
5809 struct extent_buffer *leaf;
5812 key.objectid = btrfs_ino(inode);
5813 key.type = BTRFS_DIR_INDEX_KEY;
5814 key.offset = (u64)-1;
5816 path = btrfs_alloc_path();
5820 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5823 /* FIXME: we should be able to handle this */
5829 * MAGIC NUMBER EXPLANATION:
5830 * since we search a directory based on f_pos we have to start at 2
5831 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5832 * else has to start at 2
5834 if (path->slots[0] == 0) {
5835 inode->index_cnt = 2;
5841 leaf = path->nodes[0];
5842 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5844 if (found_key.objectid != btrfs_ino(inode) ||
5845 found_key.type != BTRFS_DIR_INDEX_KEY) {
5846 inode->index_cnt = 2;
5850 inode->index_cnt = found_key.offset + 1;
5852 btrfs_free_path(path);
5857 * helper to find a free sequence number in a given directory. This current
5858 * code is very simple, later versions will do smarter things in the btree
5860 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
5864 if (dir->index_cnt == (u64)-1) {
5865 ret = btrfs_inode_delayed_dir_index_count(dir);
5867 ret = btrfs_set_inode_index_count(dir);
5873 *index = dir->index_cnt;
5879 static int btrfs_insert_inode_locked(struct inode *inode)
5881 struct btrfs_iget_args args;
5883 args.ino = BTRFS_I(inode)->location.objectid;
5884 args.root = BTRFS_I(inode)->root;
5886 return insert_inode_locked4(inode,
5887 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5888 btrfs_find_actor, &args);
5892 * Inherit flags from the parent inode.
5894 * Currently only the compression flags and the cow flags are inherited.
5896 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
5903 flags = BTRFS_I(dir)->flags;
5905 if (flags & BTRFS_INODE_NOCOMPRESS) {
5906 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
5907 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
5908 } else if (flags & BTRFS_INODE_COMPRESS) {
5909 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
5910 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
5913 if (flags & BTRFS_INODE_NODATACOW) {
5914 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
5915 if (S_ISREG(inode->i_mode))
5916 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5919 btrfs_sync_inode_flags_to_i_flags(inode);
5922 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5923 struct btrfs_root *root,
5925 const char *name, int name_len,
5926 u64 ref_objectid, u64 objectid,
5927 umode_t mode, u64 *index)
5929 struct btrfs_fs_info *fs_info = root->fs_info;
5930 struct inode *inode;
5931 struct btrfs_inode_item *inode_item;
5932 struct btrfs_key *location;
5933 struct btrfs_path *path;
5934 struct btrfs_inode_ref *ref;
5935 struct btrfs_key key[2];
5937 int nitems = name ? 2 : 1;
5939 unsigned int nofs_flag;
5942 path = btrfs_alloc_path();
5944 return ERR_PTR(-ENOMEM);
5946 nofs_flag = memalloc_nofs_save();
5947 inode = new_inode(fs_info->sb);
5948 memalloc_nofs_restore(nofs_flag);
5950 btrfs_free_path(path);
5951 return ERR_PTR(-ENOMEM);
5955 * O_TMPFILE, set link count to 0, so that after this point,
5956 * we fill in an inode item with the correct link count.
5959 set_nlink(inode, 0);
5962 * we have to initialize this early, so we can reclaim the inode
5963 * number if we fail afterwards in this function.
5965 inode->i_ino = objectid;
5968 trace_btrfs_inode_request(dir);
5970 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
5972 btrfs_free_path(path);
5974 return ERR_PTR(ret);
5980 * index_cnt is ignored for everything but a dir,
5981 * btrfs_set_inode_index_count has an explanation for the magic
5984 BTRFS_I(inode)->index_cnt = 2;
5985 BTRFS_I(inode)->dir_index = *index;
5986 BTRFS_I(inode)->root = btrfs_grab_root(root);
5987 BTRFS_I(inode)->generation = trans->transid;
5988 inode->i_generation = BTRFS_I(inode)->generation;
5991 * We could have gotten an inode number from somebody who was fsynced
5992 * and then removed in this same transaction, so let's just set full
5993 * sync since it will be a full sync anyway and this will blow away the
5994 * old info in the log.
5996 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5998 key[0].objectid = objectid;
5999 key[0].type = BTRFS_INODE_ITEM_KEY;
6002 sizes[0] = sizeof(struct btrfs_inode_item);
6006 * Start new inodes with an inode_ref. This is slightly more
6007 * efficient for small numbers of hard links since they will
6008 * be packed into one item. Extended refs will kick in if we
6009 * add more hard links than can fit in the ref item.
6011 key[1].objectid = objectid;
6012 key[1].type = BTRFS_INODE_REF_KEY;
6013 key[1].offset = ref_objectid;
6015 sizes[1] = name_len + sizeof(*ref);
6018 location = &BTRFS_I(inode)->location;
6019 location->objectid = objectid;
6020 location->offset = 0;
6021 location->type = BTRFS_INODE_ITEM_KEY;
6023 ret = btrfs_insert_inode_locked(inode);
6029 path->leave_spinning = 1;
6030 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6034 inode_init_owner(inode, dir, mode);
6035 inode_set_bytes(inode, 0);
6037 inode->i_mtime = current_time(inode);
6038 inode->i_atime = inode->i_mtime;
6039 inode->i_ctime = inode->i_mtime;
6040 BTRFS_I(inode)->i_otime = inode->i_mtime;
6042 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6043 struct btrfs_inode_item);
6044 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6045 sizeof(*inode_item));
6046 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6049 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6050 struct btrfs_inode_ref);
6051 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6052 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6053 ptr = (unsigned long)(ref + 1);
6054 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6057 btrfs_mark_buffer_dirty(path->nodes[0]);
6058 btrfs_free_path(path);
6060 btrfs_inherit_iflags(inode, dir);
6062 if (S_ISREG(mode)) {
6063 if (btrfs_test_opt(fs_info, NODATASUM))
6064 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6065 if (btrfs_test_opt(fs_info, NODATACOW))
6066 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6067 BTRFS_INODE_NODATASUM;
6070 inode_tree_add(inode);
6072 trace_btrfs_inode_new(inode);
6073 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6075 btrfs_update_root_times(trans, root);
6077 ret = btrfs_inode_inherit_props(trans, inode, dir);
6080 "error inheriting props for ino %llu (root %llu): %d",
6081 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6086 discard_new_inode(inode);
6089 BTRFS_I(dir)->index_cnt--;
6090 btrfs_free_path(path);
6091 return ERR_PTR(ret);
6095 * utility function to add 'inode' into 'parent_inode' with
6096 * a give name and a given sequence number.
6097 * if 'add_backref' is true, also insert a backref from the
6098 * inode to the parent directory.
6100 int btrfs_add_link(struct btrfs_trans_handle *trans,
6101 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6102 const char *name, int name_len, int add_backref, u64 index)
6105 struct btrfs_key key;
6106 struct btrfs_root *root = parent_inode->root;
6107 u64 ino = btrfs_ino(inode);
6108 u64 parent_ino = btrfs_ino(parent_inode);
6110 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6111 memcpy(&key, &inode->root->root_key, sizeof(key));
6114 key.type = BTRFS_INODE_ITEM_KEY;
6118 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6119 ret = btrfs_add_root_ref(trans, key.objectid,
6120 root->root_key.objectid, parent_ino,
6121 index, name, name_len);
6122 } else if (add_backref) {
6123 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6127 /* Nothing to clean up yet */
6131 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6132 btrfs_inode_type(&inode->vfs_inode), index);
6133 if (ret == -EEXIST || ret == -EOVERFLOW)
6136 btrfs_abort_transaction(trans, ret);
6140 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6142 inode_inc_iversion(&parent_inode->vfs_inode);
6144 * If we are replaying a log tree, we do not want to update the mtime
6145 * and ctime of the parent directory with the current time, since the
6146 * log replay procedure is responsible for setting them to their correct
6147 * values (the ones it had when the fsync was done).
6149 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6150 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6152 parent_inode->vfs_inode.i_mtime = now;
6153 parent_inode->vfs_inode.i_ctime = now;
6155 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6157 btrfs_abort_transaction(trans, ret);
6161 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6164 err = btrfs_del_root_ref(trans, key.objectid,
6165 root->root_key.objectid, parent_ino,
6166 &local_index, name, name_len);
6168 btrfs_abort_transaction(trans, err);
6169 } else if (add_backref) {
6173 err = btrfs_del_inode_ref(trans, root, name, name_len,
6174 ino, parent_ino, &local_index);
6176 btrfs_abort_transaction(trans, err);
6179 /* Return the original error code */
6183 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6184 struct btrfs_inode *dir, struct dentry *dentry,
6185 struct btrfs_inode *inode, int backref, u64 index)
6187 int err = btrfs_add_link(trans, dir, inode,
6188 dentry->d_name.name, dentry->d_name.len,
6195 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6196 umode_t mode, dev_t rdev)
6198 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6199 struct btrfs_trans_handle *trans;
6200 struct btrfs_root *root = BTRFS_I(dir)->root;
6201 struct inode *inode = NULL;
6207 * 2 for inode item and ref
6209 * 1 for xattr if selinux is on
6211 trans = btrfs_start_transaction(root, 5);
6213 return PTR_ERR(trans);
6215 err = btrfs_find_free_ino(root, &objectid);
6219 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6220 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6222 if (IS_ERR(inode)) {
6223 err = PTR_ERR(inode);
6229 * If the active LSM wants to access the inode during
6230 * d_instantiate it needs these. Smack checks to see
6231 * if the filesystem supports xattrs by looking at the
6234 inode->i_op = &btrfs_special_inode_operations;
6235 init_special_inode(inode, inode->i_mode, rdev);
6237 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6241 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6246 btrfs_update_inode(trans, root, inode);
6247 d_instantiate_new(dentry, inode);
6250 btrfs_end_transaction(trans);
6251 btrfs_btree_balance_dirty(fs_info);
6253 inode_dec_link_count(inode);
6254 discard_new_inode(inode);
6259 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6260 umode_t mode, bool excl)
6262 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6263 struct btrfs_trans_handle *trans;
6264 struct btrfs_root *root = BTRFS_I(dir)->root;
6265 struct inode *inode = NULL;
6271 * 2 for inode item and ref
6273 * 1 for xattr if selinux is on
6275 trans = btrfs_start_transaction(root, 5);
6277 return PTR_ERR(trans);
6279 err = btrfs_find_free_ino(root, &objectid);
6283 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6284 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6286 if (IS_ERR(inode)) {
6287 err = PTR_ERR(inode);
6292 * If the active LSM wants to access the inode during
6293 * d_instantiate it needs these. Smack checks to see
6294 * if the filesystem supports xattrs by looking at the
6297 inode->i_fop = &btrfs_file_operations;
6298 inode->i_op = &btrfs_file_inode_operations;
6299 inode->i_mapping->a_ops = &btrfs_aops;
6301 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6305 err = btrfs_update_inode(trans, root, inode);
6309 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6314 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6315 d_instantiate_new(dentry, inode);
6318 btrfs_end_transaction(trans);
6320 inode_dec_link_count(inode);
6321 discard_new_inode(inode);
6323 btrfs_btree_balance_dirty(fs_info);
6327 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6328 struct dentry *dentry)
6330 struct btrfs_trans_handle *trans = NULL;
6331 struct btrfs_root *root = BTRFS_I(dir)->root;
6332 struct inode *inode = d_inode(old_dentry);
6333 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6338 /* do not allow sys_link's with other subvols of the same device */
6339 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6342 if (inode->i_nlink >= BTRFS_LINK_MAX)
6345 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6350 * 2 items for inode and inode ref
6351 * 2 items for dir items
6352 * 1 item for parent inode
6353 * 1 item for orphan item deletion if O_TMPFILE
6355 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6356 if (IS_ERR(trans)) {
6357 err = PTR_ERR(trans);
6362 /* There are several dir indexes for this inode, clear the cache. */
6363 BTRFS_I(inode)->dir_index = 0ULL;
6365 inode_inc_iversion(inode);
6366 inode->i_ctime = current_time(inode);
6368 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6370 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6376 struct dentry *parent = dentry->d_parent;
6378 err = btrfs_update_inode(trans, root, inode);
6381 if (inode->i_nlink == 1) {
6383 * If new hard link count is 1, it's a file created
6384 * with open(2) O_TMPFILE flag.
6386 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6390 d_instantiate(dentry, inode);
6391 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6396 btrfs_end_transaction(trans);
6398 inode_dec_link_count(inode);
6401 btrfs_btree_balance_dirty(fs_info);
6405 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6407 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6408 struct inode *inode = NULL;
6409 struct btrfs_trans_handle *trans;
6410 struct btrfs_root *root = BTRFS_I(dir)->root;
6416 * 2 items for inode and ref
6417 * 2 items for dir items
6418 * 1 for xattr if selinux is on
6420 trans = btrfs_start_transaction(root, 5);
6422 return PTR_ERR(trans);
6424 err = btrfs_find_free_ino(root, &objectid);
6428 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6429 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6430 S_IFDIR | mode, &index);
6431 if (IS_ERR(inode)) {
6432 err = PTR_ERR(inode);
6437 /* these must be set before we unlock the inode */
6438 inode->i_op = &btrfs_dir_inode_operations;
6439 inode->i_fop = &btrfs_dir_file_operations;
6441 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6445 btrfs_i_size_write(BTRFS_I(inode), 0);
6446 err = btrfs_update_inode(trans, root, inode);
6450 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6451 dentry->d_name.name,
6452 dentry->d_name.len, 0, index);
6456 d_instantiate_new(dentry, inode);
6459 btrfs_end_transaction(trans);
6461 inode_dec_link_count(inode);
6462 discard_new_inode(inode);
6464 btrfs_btree_balance_dirty(fs_info);
6468 static noinline int uncompress_inline(struct btrfs_path *path,
6470 size_t pg_offset, u64 extent_offset,
6471 struct btrfs_file_extent_item *item)
6474 struct extent_buffer *leaf = path->nodes[0];
6477 unsigned long inline_size;
6481 WARN_ON(pg_offset != 0);
6482 compress_type = btrfs_file_extent_compression(leaf, item);
6483 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6484 inline_size = btrfs_file_extent_inline_item_len(leaf,
6485 btrfs_item_nr(path->slots[0]));
6486 tmp = kmalloc(inline_size, GFP_NOFS);
6489 ptr = btrfs_file_extent_inline_start(item);
6491 read_extent_buffer(leaf, tmp, ptr, inline_size);
6493 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6494 ret = btrfs_decompress(compress_type, tmp, page,
6495 extent_offset, inline_size, max_size);
6498 * decompression code contains a memset to fill in any space between the end
6499 * of the uncompressed data and the end of max_size in case the decompressed
6500 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6501 * the end of an inline extent and the beginning of the next block, so we
6502 * cover that region here.
6505 if (max_size + pg_offset < PAGE_SIZE) {
6506 char *map = kmap(page);
6507 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6515 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6516 * @inode: file to search in
6517 * @page: page to read extent data into if the extent is inline
6518 * @pg_offset: offset into @page to copy to
6519 * @start: file offset
6520 * @len: length of range starting at @start
6522 * This returns the first &struct extent_map which overlaps with the given
6523 * range, reading it from the B-tree and caching it if necessary. Note that
6524 * there may be more extents which overlap the given range after the returned
6527 * If @page is not NULL and the extent is inline, this also reads the extent
6528 * data directly into the page and marks the extent up to date in the io_tree.
6530 * Return: ERR_PTR on error, non-NULL extent_map on success.
6532 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6533 struct page *page, size_t pg_offset,
6536 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6538 u64 extent_start = 0;
6540 u64 objectid = btrfs_ino(inode);
6541 int extent_type = -1;
6542 struct btrfs_path *path = NULL;
6543 struct btrfs_root *root = inode->root;
6544 struct btrfs_file_extent_item *item;
6545 struct extent_buffer *leaf;
6546 struct btrfs_key found_key;
6547 struct extent_map *em = NULL;
6548 struct extent_map_tree *em_tree = &inode->extent_tree;
6549 struct extent_io_tree *io_tree = &inode->io_tree;
6551 read_lock(&em_tree->lock);
6552 em = lookup_extent_mapping(em_tree, start, len);
6553 read_unlock(&em_tree->lock);
6556 if (em->start > start || em->start + em->len <= start)
6557 free_extent_map(em);
6558 else if (em->block_start == EXTENT_MAP_INLINE && page)
6559 free_extent_map(em);
6563 em = alloc_extent_map();
6568 em->start = EXTENT_MAP_HOLE;
6569 em->orig_start = EXTENT_MAP_HOLE;
6571 em->block_len = (u64)-1;
6573 path = btrfs_alloc_path();
6579 /* Chances are we'll be called again, so go ahead and do readahead */
6580 path->reada = READA_FORWARD;
6583 * Unless we're going to uncompress the inline extent, no sleep would
6586 path->leave_spinning = 1;
6588 path->recurse = btrfs_is_free_space_inode(inode);
6590 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6593 } else if (ret > 0) {
6594 if (path->slots[0] == 0)
6600 leaf = path->nodes[0];
6601 item = btrfs_item_ptr(leaf, path->slots[0],
6602 struct btrfs_file_extent_item);
6603 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6604 if (found_key.objectid != objectid ||
6605 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6607 * If we backup past the first extent we want to move forward
6608 * and see if there is an extent in front of us, otherwise we'll
6609 * say there is a hole for our whole search range which can
6616 extent_type = btrfs_file_extent_type(leaf, item);
6617 extent_start = found_key.offset;
6618 extent_end = btrfs_file_extent_end(path);
6619 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6620 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6621 /* Only regular file could have regular/prealloc extent */
6622 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6625 "regular/prealloc extent found for non-regular inode %llu",
6629 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6631 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6632 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6637 if (start >= extent_end) {
6639 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6640 ret = btrfs_next_leaf(root, path);
6646 leaf = path->nodes[0];
6648 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6649 if (found_key.objectid != objectid ||
6650 found_key.type != BTRFS_EXTENT_DATA_KEY)
6652 if (start + len <= found_key.offset)
6654 if (start > found_key.offset)
6657 /* New extent overlaps with existing one */
6659 em->orig_start = start;
6660 em->len = found_key.offset - start;
6661 em->block_start = EXTENT_MAP_HOLE;
6665 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6667 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6668 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6670 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6674 size_t extent_offset;
6680 size = btrfs_file_extent_ram_bytes(leaf, item);
6681 extent_offset = page_offset(page) + pg_offset - extent_start;
6682 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6683 size - extent_offset);
6684 em->start = extent_start + extent_offset;
6685 em->len = ALIGN(copy_size, fs_info->sectorsize);
6686 em->orig_block_len = em->len;
6687 em->orig_start = em->start;
6688 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6690 btrfs_set_path_blocking(path);
6691 if (!PageUptodate(page)) {
6692 if (btrfs_file_extent_compression(leaf, item) !=
6693 BTRFS_COMPRESS_NONE) {
6694 ret = uncompress_inline(path, page, pg_offset,
6695 extent_offset, item);
6700 read_extent_buffer(leaf, map + pg_offset, ptr,
6702 if (pg_offset + copy_size < PAGE_SIZE) {
6703 memset(map + pg_offset + copy_size, 0,
6704 PAGE_SIZE - pg_offset -
6709 flush_dcache_page(page);
6711 set_extent_uptodate(io_tree, em->start,
6712 extent_map_end(em) - 1, NULL, GFP_NOFS);
6717 em->orig_start = start;
6719 em->block_start = EXTENT_MAP_HOLE;
6722 btrfs_release_path(path);
6723 if (em->start > start || extent_map_end(em) <= start) {
6725 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6726 em->start, em->len, start, len);
6731 write_lock(&em_tree->lock);
6732 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6733 write_unlock(&em_tree->lock);
6735 btrfs_free_path(path);
6737 trace_btrfs_get_extent(root, inode, em);
6740 free_extent_map(em);
6741 return ERR_PTR(ret);
6746 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6749 struct extent_map *em;
6750 struct extent_map *hole_em = NULL;
6751 u64 delalloc_start = start;
6757 em = btrfs_get_extent(inode, NULL, 0, start, len);
6761 * If our em maps to:
6763 * - a pre-alloc extent,
6764 * there might actually be delalloc bytes behind it.
6766 if (em->block_start != EXTENT_MAP_HOLE &&
6767 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6772 /* check to see if we've wrapped (len == -1 or similar) */
6781 /* ok, we didn't find anything, lets look for delalloc */
6782 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6783 end, len, EXTENT_DELALLOC, 1);
6784 delalloc_end = delalloc_start + delalloc_len;
6785 if (delalloc_end < delalloc_start)
6786 delalloc_end = (u64)-1;
6789 * We didn't find anything useful, return the original results from
6792 if (delalloc_start > end || delalloc_end <= start) {
6799 * Adjust the delalloc_start to make sure it doesn't go backwards from
6800 * the start they passed in
6802 delalloc_start = max(start, delalloc_start);
6803 delalloc_len = delalloc_end - delalloc_start;
6805 if (delalloc_len > 0) {
6808 const u64 hole_end = extent_map_end(hole_em);
6810 em = alloc_extent_map();
6818 * When btrfs_get_extent can't find anything it returns one
6821 * Make sure what it found really fits our range, and adjust to
6822 * make sure it is based on the start from the caller
6824 if (hole_end <= start || hole_em->start > end) {
6825 free_extent_map(hole_em);
6828 hole_start = max(hole_em->start, start);
6829 hole_len = hole_end - hole_start;
6832 if (hole_em && delalloc_start > hole_start) {
6834 * Our hole starts before our delalloc, so we have to
6835 * return just the parts of the hole that go until the
6838 em->len = min(hole_len, delalloc_start - hole_start);
6839 em->start = hole_start;
6840 em->orig_start = hole_start;
6842 * Don't adjust block start at all, it is fixed at
6845 em->block_start = hole_em->block_start;
6846 em->block_len = hole_len;
6847 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6848 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6851 * Hole is out of passed range or it starts after
6854 em->start = delalloc_start;
6855 em->len = delalloc_len;
6856 em->orig_start = delalloc_start;
6857 em->block_start = EXTENT_MAP_DELALLOC;
6858 em->block_len = delalloc_len;
6865 free_extent_map(hole_em);
6867 free_extent_map(em);
6868 return ERR_PTR(err);
6873 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6876 const u64 orig_start,
6877 const u64 block_start,
6878 const u64 block_len,
6879 const u64 orig_block_len,
6880 const u64 ram_bytes,
6883 struct extent_map *em = NULL;
6886 if (type != BTRFS_ORDERED_NOCOW) {
6887 em = create_io_em(inode, start, len, orig_start, block_start,
6888 block_len, orig_block_len, ram_bytes,
6889 BTRFS_COMPRESS_NONE, /* compress_type */
6894 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
6898 free_extent_map(em);
6899 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
6908 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6911 struct btrfs_root *root = inode->root;
6912 struct btrfs_fs_info *fs_info = root->fs_info;
6913 struct extent_map *em;
6914 struct btrfs_key ins;
6918 alloc_hint = get_extent_allocation_hint(inode, start, len);
6919 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6920 0, alloc_hint, &ins, 1, 1);
6922 return ERR_PTR(ret);
6924 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
6925 ins.objectid, ins.offset, ins.offset,
6926 ins.offset, BTRFS_ORDERED_REGULAR);
6927 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6929 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
6936 * Check if we can do nocow write into the range [@offset, @offset + @len)
6938 * @offset: File offset
6939 * @len: The length to write, will be updated to the nocow writeable
6941 * @orig_start: (optional) Return the original file offset of the file extent
6942 * @orig_len: (optional) Return the original on-disk length of the file extent
6943 * @ram_bytes: (optional) Return the ram_bytes of the file extent
6944 * @strict: if true, omit optimizations that might force us into unnecessary
6945 * cow. e.g., don't trust generation number.
6947 * This function will flush ordered extents in the range to ensure proper
6948 * nocow checks for (nowait == false) case.
6951 * >0 and update @len if we can do nocow write
6952 * 0 if we can't do nocow write
6953 * <0 if error happened
6955 * NOTE: This only checks the file extents, caller is responsible to wait for
6956 * any ordered extents.
6958 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6959 u64 *orig_start, u64 *orig_block_len,
6960 u64 *ram_bytes, bool strict)
6962 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6963 struct btrfs_path *path;
6965 struct extent_buffer *leaf;
6966 struct btrfs_root *root = BTRFS_I(inode)->root;
6967 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6968 struct btrfs_file_extent_item *fi;
6969 struct btrfs_key key;
6976 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6978 path = btrfs_alloc_path();
6982 ret = btrfs_lookup_file_extent(NULL, root, path,
6983 btrfs_ino(BTRFS_I(inode)), offset, 0);
6987 slot = path->slots[0];
6990 /* can't find the item, must cow */
6997 leaf = path->nodes[0];
6998 btrfs_item_key_to_cpu(leaf, &key, slot);
6999 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7000 key.type != BTRFS_EXTENT_DATA_KEY) {
7001 /* not our file or wrong item type, must cow */
7005 if (key.offset > offset) {
7006 /* Wrong offset, must cow */
7010 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7011 found_type = btrfs_file_extent_type(leaf, fi);
7012 if (found_type != BTRFS_FILE_EXTENT_REG &&
7013 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7014 /* not a regular extent, must cow */
7018 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7021 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7022 if (extent_end <= offset)
7025 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7026 if (disk_bytenr == 0)
7029 if (btrfs_file_extent_compression(leaf, fi) ||
7030 btrfs_file_extent_encryption(leaf, fi) ||
7031 btrfs_file_extent_other_encoding(leaf, fi))
7035 * Do the same check as in btrfs_cross_ref_exist but without the
7036 * unnecessary search.
7039 (btrfs_file_extent_generation(leaf, fi) <=
7040 btrfs_root_last_snapshot(&root->root_item)))
7043 backref_offset = btrfs_file_extent_offset(leaf, fi);
7046 *orig_start = key.offset - backref_offset;
7047 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7048 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7051 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7054 num_bytes = min(offset + *len, extent_end) - offset;
7055 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7058 range_end = round_up(offset + num_bytes,
7059 root->fs_info->sectorsize) - 1;
7060 ret = test_range_bit(io_tree, offset, range_end,
7061 EXTENT_DELALLOC, 0, NULL);
7068 btrfs_release_path(path);
7071 * look for other files referencing this extent, if we
7072 * find any we must cow
7075 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7076 key.offset - backref_offset, disk_bytenr,
7084 * adjust disk_bytenr and num_bytes to cover just the bytes
7085 * in this extent we are about to write. If there
7086 * are any csums in that range we have to cow in order
7087 * to keep the csums correct
7089 disk_bytenr += backref_offset;
7090 disk_bytenr += offset - key.offset;
7091 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7094 * all of the above have passed, it is safe to overwrite this extent
7100 btrfs_free_path(path);
7104 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7105 struct extent_state **cached_state, bool writing)
7107 struct btrfs_ordered_extent *ordered;
7111 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7114 * We're concerned with the entire range that we're going to be
7115 * doing DIO to, so we need to make sure there's no ordered
7116 * extents in this range.
7118 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7119 lockend - lockstart + 1);
7122 * We need to make sure there are no buffered pages in this
7123 * range either, we could have raced between the invalidate in
7124 * generic_file_direct_write and locking the extent. The
7125 * invalidate needs to happen so that reads after a write do not
7129 (!writing || !filemap_range_has_page(inode->i_mapping,
7130 lockstart, lockend)))
7133 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7138 * If we are doing a DIO read and the ordered extent we
7139 * found is for a buffered write, we can not wait for it
7140 * to complete and retry, because if we do so we can
7141 * deadlock with concurrent buffered writes on page
7142 * locks. This happens only if our DIO read covers more
7143 * than one extent map, if at this point has already
7144 * created an ordered extent for a previous extent map
7145 * and locked its range in the inode's io tree, and a
7146 * concurrent write against that previous extent map's
7147 * range and this range started (we unlock the ranges
7148 * in the io tree only when the bios complete and
7149 * buffered writes always lock pages before attempting
7150 * to lock range in the io tree).
7153 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7154 btrfs_start_ordered_extent(inode, ordered, 1);
7157 btrfs_put_ordered_extent(ordered);
7160 * We could trigger writeback for this range (and wait
7161 * for it to complete) and then invalidate the pages for
7162 * this range (through invalidate_inode_pages2_range()),
7163 * but that can lead us to a deadlock with a concurrent
7164 * call to readahead (a buffered read or a defrag call
7165 * triggered a readahead) on a page lock due to an
7166 * ordered dio extent we created before but did not have
7167 * yet a corresponding bio submitted (whence it can not
7168 * complete), which makes readahead wait for that
7169 * ordered extent to complete while holding a lock on
7184 /* The callers of this must take lock_extent() */
7185 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7186 u64 len, u64 orig_start, u64 block_start,
7187 u64 block_len, u64 orig_block_len,
7188 u64 ram_bytes, int compress_type,
7191 struct extent_map_tree *em_tree;
7192 struct extent_map *em;
7195 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7196 type == BTRFS_ORDERED_COMPRESSED ||
7197 type == BTRFS_ORDERED_NOCOW ||
7198 type == BTRFS_ORDERED_REGULAR);
7200 em_tree = &inode->extent_tree;
7201 em = alloc_extent_map();
7203 return ERR_PTR(-ENOMEM);
7206 em->orig_start = orig_start;
7208 em->block_len = block_len;
7209 em->block_start = block_start;
7210 em->orig_block_len = orig_block_len;
7211 em->ram_bytes = ram_bytes;
7212 em->generation = -1;
7213 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7214 if (type == BTRFS_ORDERED_PREALLOC) {
7215 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7216 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7217 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7218 em->compress_type = compress_type;
7222 btrfs_drop_extent_cache(inode, em->start,
7223 em->start + em->len - 1, 0);
7224 write_lock(&em_tree->lock);
7225 ret = add_extent_mapping(em_tree, em, 1);
7226 write_unlock(&em_tree->lock);
7228 * The caller has taken lock_extent(), who could race with us
7231 } while (ret == -EEXIST);
7234 free_extent_map(em);
7235 return ERR_PTR(ret);
7238 /* em got 2 refs now, callers needs to do free_extent_map once. */
7243 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7244 struct inode *inode,
7245 struct btrfs_dio_data *dio_data,
7248 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7249 struct extent_map *em = *map;
7253 * We don't allocate a new extent in the following cases
7255 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7257 * 2) The extent is marked as PREALLOC. We're good to go here and can
7258 * just use the extent.
7261 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7262 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7263 em->block_start != EXTENT_MAP_HOLE)) {
7265 u64 block_start, orig_start, orig_block_len, ram_bytes;
7267 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7268 type = BTRFS_ORDERED_PREALLOC;
7270 type = BTRFS_ORDERED_NOCOW;
7271 len = min(len, em->len - (start - em->start));
7272 block_start = em->block_start + (start - em->start);
7274 if (can_nocow_extent(inode, start, &len, &orig_start,
7275 &orig_block_len, &ram_bytes, false) == 1 &&
7276 btrfs_inc_nocow_writers(fs_info, block_start)) {
7277 struct extent_map *em2;
7279 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7280 orig_start, block_start,
7281 len, orig_block_len,
7283 btrfs_dec_nocow_writers(fs_info, block_start);
7284 if (type == BTRFS_ORDERED_PREALLOC) {
7285 free_extent_map(em);
7289 if (em2 && IS_ERR(em2)) {
7294 * For inode marked NODATACOW or extent marked PREALLOC,
7295 * use the existing or preallocated extent, so does not
7296 * need to adjust btrfs_space_info's bytes_may_use.
7298 btrfs_free_reserved_data_space_noquota(fs_info, len);
7303 /* this will cow the extent */
7304 free_extent_map(em);
7305 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7311 len = min(len, em->len - (start - em->start));
7315 * Need to update the i_size under the extent lock so buffered
7316 * readers will get the updated i_size when we unlock.
7318 if (start + len > i_size_read(inode))
7319 i_size_write(inode, start + len);
7321 dio_data->reserve -= len;
7326 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7327 loff_t length, unsigned int flags, struct iomap *iomap,
7328 struct iomap *srcmap)
7330 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7331 struct extent_map *em;
7332 struct extent_state *cached_state = NULL;
7333 struct btrfs_dio_data *dio_data = NULL;
7334 u64 lockstart, lockend;
7335 const bool write = !!(flags & IOMAP_WRITE);
7338 bool unlock_extents = false;
7339 bool sync = (current->journal_info == BTRFS_DIO_SYNC_STUB);
7342 * We used current->journal_info here to see if we were sync, but
7343 * there's a lot of tests in the enospc machinery to not do flushing if
7344 * we have a journal_info set, so we need to clear this out and re-set
7347 ASSERT(current->journal_info == NULL ||
7348 current->journal_info == BTRFS_DIO_SYNC_STUB);
7349 current->journal_info = NULL;
7352 len = min_t(u64, len, fs_info->sectorsize);
7355 lockend = start + len - 1;
7358 * The generic stuff only does filemap_write_and_wait_range, which
7359 * isn't enough if we've written compressed pages to this area, so we
7360 * need to flush the dirty pages again to make absolutely sure that any
7361 * outstanding dirty pages are on disk.
7363 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7364 &BTRFS_I(inode)->runtime_flags)) {
7365 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7366 start + length - 1);
7371 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7375 dio_data->sync = sync;
7376 dio_data->length = length;
7378 dio_data->reserve = round_up(length, fs_info->sectorsize);
7379 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7380 &dio_data->data_reserved,
7381 start, dio_data->reserve);
7383 extent_changeset_free(dio_data->data_reserved);
7388 iomap->private = dio_data;
7392 * If this errors out it's because we couldn't invalidate pagecache for
7393 * this range and we need to fallback to buffered.
7395 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7400 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7407 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7408 * io. INLINE is special, and we could probably kludge it in here, but
7409 * it's still buffered so for safety lets just fall back to the generic
7412 * For COMPRESSED we _have_ to read the entire extent in so we can
7413 * decompress it, so there will be buffering required no matter what we
7414 * do, so go ahead and fallback to buffered.
7416 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7417 * to buffered IO. Don't blame me, this is the price we pay for using
7420 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7421 em->block_start == EXTENT_MAP_INLINE) {
7422 free_extent_map(em);
7427 len = min(len, em->len - (start - em->start));
7429 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7433 unlock_extents = true;
7434 /* Recalc len in case the new em is smaller than requested */
7435 len = min(len, em->len - (start - em->start));
7438 * We need to unlock only the end area that we aren't using.
7439 * The rest is going to be unlocked by the endio routine.
7441 lockstart = start + len;
7442 if (lockstart < lockend)
7443 unlock_extents = true;
7447 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7448 lockstart, lockend, &cached_state);
7450 free_extent_state(cached_state);
7453 * Translate extent map information to iomap.
7454 * We trim the extents (and move the addr) even though iomap code does
7455 * that, since we have locked only the parts we are performing I/O in.
7457 if ((em->block_start == EXTENT_MAP_HOLE) ||
7458 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7459 iomap->addr = IOMAP_NULL_ADDR;
7460 iomap->type = IOMAP_HOLE;
7462 iomap->addr = em->block_start + (start - em->start);
7463 iomap->type = IOMAP_MAPPED;
7465 iomap->offset = start;
7466 iomap->bdev = fs_info->fs_devices->latest_bdev;
7467 iomap->length = len;
7469 free_extent_map(em);
7474 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7478 btrfs_delalloc_release_space(BTRFS_I(inode),
7479 dio_data->data_reserved, start,
7480 dio_data->reserve, true);
7481 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7482 extent_changeset_free(dio_data->data_reserved);
7488 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7489 ssize_t written, unsigned int flags, struct iomap *iomap)
7492 struct btrfs_dio_data *dio_data = iomap->private;
7493 size_t submitted = dio_data->submitted;
7494 const bool write = !!(flags & IOMAP_WRITE);
7496 if (!write && (iomap->type == IOMAP_HOLE)) {
7497 /* If reading from a hole, unlock and return */
7498 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7502 if (submitted < length) {
7504 length -= submitted;
7506 __endio_write_update_ordered(BTRFS_I(inode), pos,
7509 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7515 if (dio_data->reserve)
7516 btrfs_delalloc_release_space(BTRFS_I(inode),
7517 dio_data->data_reserved, pos,
7518 dio_data->reserve, true);
7519 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7520 extent_changeset_free(dio_data->data_reserved);
7524 * We're all done, we can re-set the current->journal_info now safely
7527 if (dio_data->sync) {
7528 ASSERT(current->journal_info == NULL);
7529 current->journal_info = BTRFS_DIO_SYNC_STUB;
7532 iomap->private = NULL;
7537 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7540 * This implies a barrier so that stores to dio_bio->bi_status before
7541 * this and loads of dio_bio->bi_status after this are fully ordered.
7543 if (!refcount_dec_and_test(&dip->refs))
7546 if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
7547 __endio_write_update_ordered(BTRFS_I(dip->inode),
7548 dip->logical_offset,
7550 !dip->dio_bio->bi_status);
7552 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7553 dip->logical_offset,
7554 dip->logical_offset + dip->bytes - 1);
7557 bio_endio(dip->dio_bio);
7561 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7563 unsigned long bio_flags)
7565 struct btrfs_dio_private *dip = bio->bi_private;
7566 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7569 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7571 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7575 refcount_inc(&dip->refs);
7576 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7578 refcount_dec(&dip->refs);
7582 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7583 struct btrfs_io_bio *io_bio,
7584 const bool uptodate)
7586 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7587 const u32 sectorsize = fs_info->sectorsize;
7588 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7589 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7590 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7591 struct bio_vec bvec;
7592 struct bvec_iter iter;
7593 u64 start = io_bio->logical;
7595 blk_status_t err = BLK_STS_OK;
7597 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7598 unsigned int i, nr_sectors, pgoff;
7600 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7601 pgoff = bvec.bv_offset;
7602 for (i = 0; i < nr_sectors; i++) {
7603 ASSERT(pgoff < PAGE_SIZE);
7605 (!csum || !check_data_csum(inode, io_bio, icsum,
7606 bvec.bv_page, pgoff,
7607 start, sectorsize))) {
7608 clean_io_failure(fs_info, failure_tree, io_tree,
7609 start, bvec.bv_page,
7610 btrfs_ino(BTRFS_I(inode)),
7613 blk_status_t status;
7615 status = btrfs_submit_read_repair(inode,
7617 start - io_bio->logical,
7618 bvec.bv_page, pgoff,
7620 start + sectorsize - 1,
7622 submit_dio_repair_bio);
7626 start += sectorsize;
7628 pgoff += sectorsize;
7634 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7635 const u64 offset, const u64 bytes,
7636 const bool uptodate)
7638 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7639 struct btrfs_ordered_extent *ordered = NULL;
7640 struct btrfs_workqueue *wq;
7641 u64 ordered_offset = offset;
7642 u64 ordered_bytes = bytes;
7645 if (btrfs_is_free_space_inode(inode))
7646 wq = fs_info->endio_freespace_worker;
7648 wq = fs_info->endio_write_workers;
7650 while (ordered_offset < offset + bytes) {
7651 last_offset = ordered_offset;
7652 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7656 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7658 btrfs_queue_work(wq, &ordered->work);
7661 * If btrfs_dec_test_ordered_pending does not find any ordered
7662 * extent in the range, we can exit.
7664 if (ordered_offset == last_offset)
7667 * Our bio might span multiple ordered extents. In this case
7668 * we keep going until we have accounted the whole dio.
7670 if (ordered_offset < offset + bytes) {
7671 ordered_bytes = offset + bytes - ordered_offset;
7677 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
7678 struct bio *bio, u64 offset)
7680 struct inode *inode = private_data;
7682 return btrfs_csum_one_bio(BTRFS_I(inode), bio, offset, 1);
7685 static void btrfs_end_dio_bio(struct bio *bio)
7687 struct btrfs_dio_private *dip = bio->bi_private;
7688 blk_status_t err = bio->bi_status;
7691 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7692 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7693 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7695 (unsigned long long)bio->bi_iter.bi_sector,
7696 bio->bi_iter.bi_size, err);
7698 if (bio_op(bio) == REQ_OP_READ) {
7699 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7704 dip->dio_bio->bi_status = err;
7707 btrfs_dio_private_put(dip);
7710 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7711 struct inode *inode, u64 file_offset, int async_submit)
7713 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7714 struct btrfs_dio_private *dip = bio->bi_private;
7715 bool write = bio_op(bio) == REQ_OP_WRITE;
7718 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7720 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7723 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7728 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7731 if (write && async_submit) {
7732 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
7734 btrfs_submit_bio_start_direct_io);
7738 * If we aren't doing async submit, calculate the csum of the
7741 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
7747 csum_offset = file_offset - dip->logical_offset;
7748 csum_offset >>= inode->i_sb->s_blocksize_bits;
7749 csum_offset *= btrfs_super_csum_size(fs_info->super_copy);
7750 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
7753 ret = btrfs_map_bio(fs_info, bio, 0);
7759 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7760 * or ordered extents whether or not we submit any bios.
7762 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7763 struct inode *inode,
7766 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7767 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7769 struct btrfs_dio_private *dip;
7771 dip_size = sizeof(*dip);
7772 if (!write && csum) {
7773 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7774 const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
7777 nblocks = dio_bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
7778 dip_size += csum_size * nblocks;
7781 dip = kzalloc(dip_size, GFP_NOFS);
7786 dip->logical_offset = file_offset;
7787 dip->bytes = dio_bio->bi_iter.bi_size;
7788 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7789 dip->dio_bio = dio_bio;
7790 refcount_set(&dip->refs, 1);
7794 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
7795 struct bio *dio_bio, loff_t file_offset)
7797 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7798 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7799 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7800 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7801 BTRFS_BLOCK_GROUP_RAID56_MASK);
7802 struct btrfs_dio_private *dip;
7805 int async_submit = 0;
7807 int clone_offset = 0;
7810 blk_status_t status;
7811 struct btrfs_io_geometry geom;
7812 struct btrfs_dio_data *dio_data = iomap->private;
7814 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7817 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7818 file_offset + dio_bio->bi_iter.bi_size - 1);
7820 dio_bio->bi_status = BLK_STS_RESOURCE;
7822 return BLK_QC_T_NONE;
7825 if (!write && csum) {
7827 * Load the csums up front to reduce csum tree searches and
7828 * contention when submitting bios.
7830 status = btrfs_lookup_bio_sums(inode, dio_bio, file_offset,
7832 if (status != BLK_STS_OK)
7836 start_sector = dio_bio->bi_iter.bi_sector;
7837 submit_len = dio_bio->bi_iter.bi_size;
7840 ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio),
7841 start_sector << 9, submit_len,
7844 status = errno_to_blk_status(ret);
7847 ASSERT(geom.len <= INT_MAX);
7849 clone_len = min_t(int, submit_len, geom.len);
7852 * This will never fail as it's passing GPF_NOFS and
7853 * the allocation is backed by btrfs_bioset.
7855 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7856 bio->bi_private = dip;
7857 bio->bi_end_io = btrfs_end_dio_bio;
7858 btrfs_io_bio(bio)->logical = file_offset;
7860 ASSERT(submit_len >= clone_len);
7861 submit_len -= clone_len;
7864 * Increase the count before we submit the bio so we know
7865 * the end IO handler won't happen before we increase the
7866 * count. Otherwise, the dip might get freed before we're
7867 * done setting it up.
7869 * We transfer the initial reference to the last bio, so we
7870 * don't need to increment the reference count for the last one.
7872 if (submit_len > 0) {
7873 refcount_inc(&dip->refs);
7875 * If we are submitting more than one bio, submit them
7876 * all asynchronously. The exception is RAID 5 or 6, as
7877 * asynchronous checksums make it difficult to collect
7878 * full stripe writes.
7884 status = btrfs_submit_dio_bio(bio, inode, file_offset,
7889 refcount_dec(&dip->refs);
7893 dio_data->submitted += clone_len;
7894 clone_offset += clone_len;
7895 start_sector += clone_len >> 9;
7896 file_offset += clone_len;
7897 } while (submit_len > 0);
7898 return BLK_QC_T_NONE;
7901 dip->dio_bio->bi_status = status;
7902 btrfs_dio_private_put(dip);
7903 return BLK_QC_T_NONE;
7906 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
7907 const struct iov_iter *iter, loff_t offset)
7911 unsigned int blocksize_mask = fs_info->sectorsize - 1;
7912 ssize_t retval = -EINVAL;
7914 if (offset & blocksize_mask)
7917 if (iov_iter_alignment(iter) & blocksize_mask)
7920 /* If this is a write we don't need to check anymore */
7921 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
7924 * Check to make sure we don't have duplicate iov_base's in this
7925 * iovec, if so return EINVAL, otherwise we'll get csum errors
7926 * when reading back.
7928 for (seg = 0; seg < iter->nr_segs; seg++) {
7929 for (i = seg + 1; i < iter->nr_segs; i++) {
7930 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7939 static inline int btrfs_maybe_fsync_end_io(struct kiocb *iocb, ssize_t size,
7940 int error, unsigned flags)
7943 * Now if we're still in the context of our submitter we know we can't
7944 * safely run generic_write_sync(), so clear our flag here so that the
7945 * caller knows to follow up with a sync.
7947 if (current->journal_info == BTRFS_DIO_SYNC_STUB) {
7948 current->journal_info = NULL;
7956 iocb->ki_flags |= IOCB_DSYNC;
7957 return generic_write_sync(iocb, size);
7963 static const struct iomap_ops btrfs_dio_iomap_ops = {
7964 .iomap_begin = btrfs_dio_iomap_begin,
7965 .iomap_end = btrfs_dio_iomap_end,
7968 static const struct iomap_dio_ops btrfs_dio_ops = {
7969 .submit_io = btrfs_submit_direct,
7972 static const struct iomap_dio_ops btrfs_sync_dops = {
7973 .submit_io = btrfs_submit_direct,
7974 .end_io = btrfs_maybe_fsync_end_io,
7977 ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
7979 struct file *file = iocb->ki_filp;
7980 struct inode *inode = file->f_mapping->host;
7981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7982 struct extent_changeset *data_reserved = NULL;
7983 loff_t offset = iocb->ki_pos;
7985 bool relock = false;
7988 if (check_direct_IO(fs_info, iter, offset))
7991 count = iov_iter_count(iter);
7992 if (iov_iter_rw(iter) == WRITE) {
7994 * If the write DIO is beyond the EOF, we need update
7995 * the isize, but it is protected by i_mutex. So we can
7996 * not unlock the i_mutex at this case.
7998 if (offset + count <= inode->i_size) {
7999 inode_unlock(inode);
8002 down_read(&BTRFS_I(inode)->dio_sem);
8006 * We have are actually a sync iocb, so we need our fancy endio to know
8007 * if we need to sync.
8009 if (current->journal_info)
8010 ret = iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops,
8011 &btrfs_sync_dops, is_sync_kiocb(iocb));
8013 ret = iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops,
8014 &btrfs_dio_ops, is_sync_kiocb(iocb));
8016 if (ret == -ENOTBLK)
8019 if (iov_iter_rw(iter) == WRITE)
8020 up_read(&BTRFS_I(inode)->dio_sem);
8025 extent_changeset_free(data_reserved);
8029 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8034 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8038 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8041 int btrfs_readpage(struct file *file, struct page *page)
8043 return extent_read_full_page(page, btrfs_get_extent, 0);
8046 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8048 struct inode *inode = page->mapping->host;
8051 if (current->flags & PF_MEMALLOC) {
8052 redirty_page_for_writepage(wbc, page);
8058 * If we are under memory pressure we will call this directly from the
8059 * VM, we need to make sure we have the inode referenced for the ordered
8060 * extent. If not just return like we didn't do anything.
8062 if (!igrab(inode)) {
8063 redirty_page_for_writepage(wbc, page);
8064 return AOP_WRITEPAGE_ACTIVATE;
8066 ret = extent_write_full_page(page, wbc);
8067 btrfs_add_delayed_iput(inode);
8071 static int btrfs_writepages(struct address_space *mapping,
8072 struct writeback_control *wbc)
8074 return extent_writepages(mapping, wbc);
8077 static void btrfs_readahead(struct readahead_control *rac)
8079 extent_readahead(rac);
8082 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8084 int ret = try_release_extent_mapping(page, gfp_flags);
8086 detach_page_private(page);
8090 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8092 if (PageWriteback(page) || PageDirty(page))
8094 return __btrfs_releasepage(page, gfp_flags);
8097 #ifdef CONFIG_MIGRATION
8098 static int btrfs_migratepage(struct address_space *mapping,
8099 struct page *newpage, struct page *page,
8100 enum migrate_mode mode)
8104 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8105 if (ret != MIGRATEPAGE_SUCCESS)
8108 if (page_has_private(page))
8109 attach_page_private(newpage, detach_page_private(page));
8111 if (PagePrivate2(page)) {
8112 ClearPagePrivate2(page);
8113 SetPagePrivate2(newpage);
8116 if (mode != MIGRATE_SYNC_NO_COPY)
8117 migrate_page_copy(newpage, page);
8119 migrate_page_states(newpage, page);
8120 return MIGRATEPAGE_SUCCESS;
8124 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8125 unsigned int length)
8127 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8128 struct extent_io_tree *tree = &inode->io_tree;
8129 struct btrfs_ordered_extent *ordered;
8130 struct extent_state *cached_state = NULL;
8131 u64 page_start = page_offset(page);
8132 u64 page_end = page_start + PAGE_SIZE - 1;
8135 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8138 * we have the page locked, so new writeback can't start,
8139 * and the dirty bit won't be cleared while we are here.
8141 * Wait for IO on this page so that we can safely clear
8142 * the PagePrivate2 bit and do ordered accounting
8144 wait_on_page_writeback(page);
8147 btrfs_releasepage(page, GFP_NOFS);
8151 if (!inode_evicting)
8152 lock_extent_bits(tree, page_start, page_end, &cached_state);
8155 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8158 ordered->file_offset + ordered->num_bytes - 1);
8160 * IO on this page will never be started, so we need
8161 * to account for any ordered extents now
8163 if (!inode_evicting)
8164 clear_extent_bit(tree, start, end,
8165 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8166 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8167 EXTENT_DEFRAG, 1, 0, &cached_state);
8169 * whoever cleared the private bit is responsible
8170 * for the finish_ordered_io
8172 if (TestClearPagePrivate2(page)) {
8173 struct btrfs_ordered_inode_tree *tree;
8176 tree = &inode->ordered_tree;
8178 spin_lock_irq(&tree->lock);
8179 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8180 new_len = start - ordered->file_offset;
8181 if (new_len < ordered->truncated_len)
8182 ordered->truncated_len = new_len;
8183 spin_unlock_irq(&tree->lock);
8185 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8187 end - start + 1, 1))
8188 btrfs_finish_ordered_io(ordered);
8190 btrfs_put_ordered_extent(ordered);
8191 if (!inode_evicting) {
8192 cached_state = NULL;
8193 lock_extent_bits(tree, start, end,
8198 if (start < page_end)
8203 * Qgroup reserved space handler
8204 * Page here will be either
8205 * 1) Already written to disk or ordered extent already submitted
8206 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8207 * Qgroup will be handled by its qgroup_record then.
8208 * btrfs_qgroup_free_data() call will do nothing here.
8210 * 2) Not written to disk yet
8211 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8212 * bit of its io_tree, and free the qgroup reserved data space.
8213 * Since the IO will never happen for this page.
8215 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8216 if (!inode_evicting) {
8217 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8218 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8219 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8222 __btrfs_releasepage(page, GFP_NOFS);
8225 ClearPageChecked(page);
8226 detach_page_private(page);
8230 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8231 * called from a page fault handler when a page is first dirtied. Hence we must
8232 * be careful to check for EOF conditions here. We set the page up correctly
8233 * for a written page which means we get ENOSPC checking when writing into
8234 * holes and correct delalloc and unwritten extent mapping on filesystems that
8235 * support these features.
8237 * We are not allowed to take the i_mutex here so we have to play games to
8238 * protect against truncate races as the page could now be beyond EOF. Because
8239 * truncate_setsize() writes the inode size before removing pages, once we have
8240 * the page lock we can determine safely if the page is beyond EOF. If it is not
8241 * beyond EOF, then the page is guaranteed safe against truncation until we
8244 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8246 struct page *page = vmf->page;
8247 struct inode *inode = file_inode(vmf->vma->vm_file);
8248 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8249 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8250 struct btrfs_ordered_extent *ordered;
8251 struct extent_state *cached_state = NULL;
8252 struct extent_changeset *data_reserved = NULL;
8254 unsigned long zero_start;
8264 reserved_space = PAGE_SIZE;
8266 sb_start_pagefault(inode->i_sb);
8267 page_start = page_offset(page);
8268 page_end = page_start + PAGE_SIZE - 1;
8272 * Reserving delalloc space after obtaining the page lock can lead to
8273 * deadlock. For example, if a dirty page is locked by this function
8274 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8275 * dirty page write out, then the btrfs_writepage() function could
8276 * end up waiting indefinitely to get a lock on the page currently
8277 * being processed by btrfs_page_mkwrite() function.
8279 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8280 page_start, reserved_space);
8282 ret2 = file_update_time(vmf->vma->vm_file);
8286 ret = vmf_error(ret2);
8292 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8295 size = i_size_read(inode);
8297 if ((page->mapping != inode->i_mapping) ||
8298 (page_start >= size)) {
8299 /* page got truncated out from underneath us */
8302 wait_on_page_writeback(page);
8304 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8305 set_page_extent_mapped(page);
8308 * we can't set the delalloc bits if there are pending ordered
8309 * extents. Drop our locks and wait for them to finish
8311 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8314 unlock_extent_cached(io_tree, page_start, page_end,
8317 btrfs_start_ordered_extent(inode, ordered, 1);
8318 btrfs_put_ordered_extent(ordered);
8322 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8323 reserved_space = round_up(size - page_start,
8324 fs_info->sectorsize);
8325 if (reserved_space < PAGE_SIZE) {
8326 end = page_start + reserved_space - 1;
8327 btrfs_delalloc_release_space(BTRFS_I(inode),
8328 data_reserved, page_start,
8329 PAGE_SIZE - reserved_space, true);
8334 * page_mkwrite gets called when the page is firstly dirtied after it's
8335 * faulted in, but write(2) could also dirty a page and set delalloc
8336 * bits, thus in this case for space account reason, we still need to
8337 * clear any delalloc bits within this page range since we have to
8338 * reserve data&meta space before lock_page() (see above comments).
8340 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8341 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8342 EXTENT_DEFRAG, 0, 0, &cached_state);
8344 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8347 unlock_extent_cached(io_tree, page_start, page_end,
8349 ret = VM_FAULT_SIGBUS;
8353 /* page is wholly or partially inside EOF */
8354 if (page_start + PAGE_SIZE > size)
8355 zero_start = offset_in_page(size);
8357 zero_start = PAGE_SIZE;
8359 if (zero_start != PAGE_SIZE) {
8361 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8362 flush_dcache_page(page);
8365 ClearPageChecked(page);
8366 set_page_dirty(page);
8367 SetPageUptodate(page);
8369 BTRFS_I(inode)->last_trans = fs_info->generation;
8370 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8371 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8373 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8375 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8376 sb_end_pagefault(inode->i_sb);
8377 extent_changeset_free(data_reserved);
8378 return VM_FAULT_LOCKED;
8383 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8384 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8385 reserved_space, (ret != 0));
8387 sb_end_pagefault(inode->i_sb);
8388 extent_changeset_free(data_reserved);
8392 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8394 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8395 struct btrfs_root *root = BTRFS_I(inode)->root;
8396 struct btrfs_block_rsv *rsv;
8398 struct btrfs_trans_handle *trans;
8399 u64 mask = fs_info->sectorsize - 1;
8400 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8402 if (!skip_writeback) {
8403 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8410 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8411 * things going on here:
8413 * 1) We need to reserve space to update our inode.
8415 * 2) We need to have something to cache all the space that is going to
8416 * be free'd up by the truncate operation, but also have some slack
8417 * space reserved in case it uses space during the truncate (thank you
8418 * very much snapshotting).
8420 * And we need these to be separate. The fact is we can use a lot of
8421 * space doing the truncate, and we have no earthly idea how much space
8422 * we will use, so we need the truncate reservation to be separate so it
8423 * doesn't end up using space reserved for updating the inode. We also
8424 * need to be able to stop the transaction and start a new one, which
8425 * means we need to be able to update the inode several times, and we
8426 * have no idea of knowing how many times that will be, so we can't just
8427 * reserve 1 item for the entirety of the operation, so that has to be
8428 * done separately as well.
8430 * So that leaves us with
8432 * 1) rsv - for the truncate reservation, which we will steal from the
8433 * transaction reservation.
8434 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8435 * updating the inode.
8437 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8440 rsv->size = min_size;
8444 * 1 for the truncate slack space
8445 * 1 for updating the inode.
8447 trans = btrfs_start_transaction(root, 2);
8448 if (IS_ERR(trans)) {
8449 ret = PTR_ERR(trans);
8453 /* Migrate the slack space for the truncate to our reserve */
8454 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8459 * So if we truncate and then write and fsync we normally would just
8460 * write the extents that changed, which is a problem if we need to
8461 * first truncate that entire inode. So set this flag so we write out
8462 * all of the extents in the inode to the sync log so we're completely
8465 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8466 trans->block_rsv = rsv;
8469 ret = btrfs_truncate_inode_items(trans, root, inode,
8471 BTRFS_EXTENT_DATA_KEY);
8472 trans->block_rsv = &fs_info->trans_block_rsv;
8473 if (ret != -ENOSPC && ret != -EAGAIN)
8476 ret = btrfs_update_inode(trans, root, inode);
8480 btrfs_end_transaction(trans);
8481 btrfs_btree_balance_dirty(fs_info);
8483 trans = btrfs_start_transaction(root, 2);
8484 if (IS_ERR(trans)) {
8485 ret = PTR_ERR(trans);
8490 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8491 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8492 rsv, min_size, false);
8493 BUG_ON(ret); /* shouldn't happen */
8494 trans->block_rsv = rsv;
8498 * We can't call btrfs_truncate_block inside a trans handle as we could
8499 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8500 * we've truncated everything except the last little bit, and can do
8501 * btrfs_truncate_block and then update the disk_i_size.
8503 if (ret == NEED_TRUNCATE_BLOCK) {
8504 btrfs_end_transaction(trans);
8505 btrfs_btree_balance_dirty(fs_info);
8507 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
8510 trans = btrfs_start_transaction(root, 1);
8511 if (IS_ERR(trans)) {
8512 ret = PTR_ERR(trans);
8515 btrfs_inode_safe_disk_i_size_write(inode, 0);
8521 trans->block_rsv = &fs_info->trans_block_rsv;
8522 ret2 = btrfs_update_inode(trans, root, inode);
8526 ret2 = btrfs_end_transaction(trans);
8529 btrfs_btree_balance_dirty(fs_info);
8532 btrfs_free_block_rsv(fs_info, rsv);
8538 * create a new subvolume directory/inode (helper for the ioctl).
8540 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8541 struct btrfs_root *new_root,
8542 struct btrfs_root *parent_root,
8545 struct inode *inode;
8549 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8550 new_dirid, new_dirid,
8551 S_IFDIR | (~current_umask() & S_IRWXUGO),
8554 return PTR_ERR(inode);
8555 inode->i_op = &btrfs_dir_inode_operations;
8556 inode->i_fop = &btrfs_dir_file_operations;
8558 set_nlink(inode, 1);
8559 btrfs_i_size_write(BTRFS_I(inode), 0);
8560 unlock_new_inode(inode);
8562 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8564 btrfs_err(new_root->fs_info,
8565 "error inheriting subvolume %llu properties: %d",
8566 new_root->root_key.objectid, err);
8568 err = btrfs_update_inode(trans, new_root, inode);
8574 struct inode *btrfs_alloc_inode(struct super_block *sb)
8576 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8577 struct btrfs_inode *ei;
8578 struct inode *inode;
8580 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8587 ei->last_sub_trans = 0;
8588 ei->logged_trans = 0;
8589 ei->delalloc_bytes = 0;
8590 ei->new_delalloc_bytes = 0;
8591 ei->defrag_bytes = 0;
8592 ei->disk_i_size = 0;
8595 ei->index_cnt = (u64)-1;
8597 ei->last_unlink_trans = 0;
8598 ei->last_reflink_trans = 0;
8599 ei->last_log_commit = 0;
8601 spin_lock_init(&ei->lock);
8602 ei->outstanding_extents = 0;
8603 if (sb->s_magic != BTRFS_TEST_MAGIC)
8604 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8605 BTRFS_BLOCK_RSV_DELALLOC);
8606 ei->runtime_flags = 0;
8607 ei->prop_compress = BTRFS_COMPRESS_NONE;
8608 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8610 ei->delayed_node = NULL;
8612 ei->i_otime.tv_sec = 0;
8613 ei->i_otime.tv_nsec = 0;
8615 inode = &ei->vfs_inode;
8616 extent_map_tree_init(&ei->extent_tree);
8617 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8618 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8619 IO_TREE_INODE_IO_FAILURE, inode);
8620 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8621 IO_TREE_INODE_FILE_EXTENT, inode);
8622 ei->io_tree.track_uptodate = true;
8623 ei->io_failure_tree.track_uptodate = true;
8624 atomic_set(&ei->sync_writers, 0);
8625 mutex_init(&ei->log_mutex);
8626 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8627 INIT_LIST_HEAD(&ei->delalloc_inodes);
8628 INIT_LIST_HEAD(&ei->delayed_iput);
8629 RB_CLEAR_NODE(&ei->rb_node);
8630 init_rwsem(&ei->dio_sem);
8635 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8636 void btrfs_test_destroy_inode(struct inode *inode)
8638 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8639 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8643 void btrfs_free_inode(struct inode *inode)
8645 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8648 void btrfs_destroy_inode(struct inode *inode)
8650 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8651 struct btrfs_ordered_extent *ordered;
8652 struct btrfs_root *root = BTRFS_I(inode)->root;
8654 WARN_ON(!hlist_empty(&inode->i_dentry));
8655 WARN_ON(inode->i_data.nrpages);
8656 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
8657 WARN_ON(BTRFS_I(inode)->block_rsv.size);
8658 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8659 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8660 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
8661 WARN_ON(BTRFS_I(inode)->csum_bytes);
8662 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8665 * This can happen where we create an inode, but somebody else also
8666 * created the same inode and we need to destroy the one we already
8673 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
8679 "found ordered extent %llu %llu on inode cleanup",
8680 ordered->file_offset, ordered->num_bytes);
8681 btrfs_remove_ordered_extent(inode, ordered);
8682 btrfs_put_ordered_extent(ordered);
8683 btrfs_put_ordered_extent(ordered);
8686 btrfs_qgroup_check_reserved_leak(BTRFS_I(inode));
8687 inode_tree_del(BTRFS_I(inode));
8688 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8689 btrfs_inode_clear_file_extent_range(BTRFS_I(inode), 0, (u64)-1);
8690 btrfs_put_root(BTRFS_I(inode)->root);
8693 int btrfs_drop_inode(struct inode *inode)
8695 struct btrfs_root *root = BTRFS_I(inode)->root;
8700 /* the snap/subvol tree is on deleting */
8701 if (btrfs_root_refs(&root->root_item) == 0)
8704 return generic_drop_inode(inode);
8707 static void init_once(void *foo)
8709 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8711 inode_init_once(&ei->vfs_inode);
8714 void __cold btrfs_destroy_cachep(void)
8717 * Make sure all delayed rcu free inodes are flushed before we
8721 kmem_cache_destroy(btrfs_inode_cachep);
8722 kmem_cache_destroy(btrfs_trans_handle_cachep);
8723 kmem_cache_destroy(btrfs_path_cachep);
8724 kmem_cache_destroy(btrfs_free_space_cachep);
8725 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8728 int __init btrfs_init_cachep(void)
8730 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8731 sizeof(struct btrfs_inode), 0,
8732 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8734 if (!btrfs_inode_cachep)
8737 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8738 sizeof(struct btrfs_trans_handle), 0,
8739 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8740 if (!btrfs_trans_handle_cachep)
8743 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8744 sizeof(struct btrfs_path), 0,
8745 SLAB_MEM_SPREAD, NULL);
8746 if (!btrfs_path_cachep)
8749 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8750 sizeof(struct btrfs_free_space), 0,
8751 SLAB_MEM_SPREAD, NULL);
8752 if (!btrfs_free_space_cachep)
8755 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8756 PAGE_SIZE, PAGE_SIZE,
8757 SLAB_RED_ZONE, NULL);
8758 if (!btrfs_free_space_bitmap_cachep)
8763 btrfs_destroy_cachep();
8767 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8768 u32 request_mask, unsigned int flags)
8771 struct inode *inode = d_inode(path->dentry);
8772 u32 blocksize = inode->i_sb->s_blocksize;
8773 u32 bi_flags = BTRFS_I(inode)->flags;
8775 stat->result_mask |= STATX_BTIME;
8776 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8777 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8778 if (bi_flags & BTRFS_INODE_APPEND)
8779 stat->attributes |= STATX_ATTR_APPEND;
8780 if (bi_flags & BTRFS_INODE_COMPRESS)
8781 stat->attributes |= STATX_ATTR_COMPRESSED;
8782 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8783 stat->attributes |= STATX_ATTR_IMMUTABLE;
8784 if (bi_flags & BTRFS_INODE_NODUMP)
8785 stat->attributes |= STATX_ATTR_NODUMP;
8787 stat->attributes_mask |= (STATX_ATTR_APPEND |
8788 STATX_ATTR_COMPRESSED |
8789 STATX_ATTR_IMMUTABLE |
8792 generic_fillattr(inode, stat);
8793 stat->dev = BTRFS_I(inode)->root->anon_dev;
8795 spin_lock(&BTRFS_I(inode)->lock);
8796 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8797 spin_unlock(&BTRFS_I(inode)->lock);
8798 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8799 ALIGN(delalloc_bytes, blocksize)) >> 9;
8803 static int btrfs_rename_exchange(struct inode *old_dir,
8804 struct dentry *old_dentry,
8805 struct inode *new_dir,
8806 struct dentry *new_dentry)
8808 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8809 struct btrfs_trans_handle *trans;
8810 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8811 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8812 struct inode *new_inode = new_dentry->d_inode;
8813 struct inode *old_inode = old_dentry->d_inode;
8814 struct timespec64 ctime = current_time(old_inode);
8815 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8816 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8821 bool root_log_pinned = false;
8822 bool dest_log_pinned = false;
8824 /* we only allow rename subvolume link between subvolumes */
8825 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8828 /* close the race window with snapshot create/destroy ioctl */
8829 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8830 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8831 down_read(&fs_info->subvol_sem);
8834 * We want to reserve the absolute worst case amount of items. So if
8835 * both inodes are subvols and we need to unlink them then that would
8836 * require 4 item modifications, but if they are both normal inodes it
8837 * would require 5 item modifications, so we'll assume their normal
8838 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8839 * should cover the worst case number of items we'll modify.
8841 trans = btrfs_start_transaction(root, 12);
8842 if (IS_ERR(trans)) {
8843 ret = PTR_ERR(trans);
8848 btrfs_record_root_in_trans(trans, dest);
8851 * We need to find a free sequence number both in the source and
8852 * in the destination directory for the exchange.
8854 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8857 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8861 BTRFS_I(old_inode)->dir_index = 0ULL;
8862 BTRFS_I(new_inode)->dir_index = 0ULL;
8864 /* Reference for the source. */
8865 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8866 /* force full log commit if subvolume involved. */
8867 btrfs_set_log_full_commit(trans);
8869 btrfs_pin_log_trans(root);
8870 root_log_pinned = true;
8871 ret = btrfs_insert_inode_ref(trans, dest,
8872 new_dentry->d_name.name,
8873 new_dentry->d_name.len,
8875 btrfs_ino(BTRFS_I(new_dir)),
8881 /* And now for the dest. */
8882 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8883 /* force full log commit if subvolume involved. */
8884 btrfs_set_log_full_commit(trans);
8886 btrfs_pin_log_trans(dest);
8887 dest_log_pinned = true;
8888 ret = btrfs_insert_inode_ref(trans, root,
8889 old_dentry->d_name.name,
8890 old_dentry->d_name.len,
8892 btrfs_ino(BTRFS_I(old_dir)),
8898 /* Update inode version and ctime/mtime. */
8899 inode_inc_iversion(old_dir);
8900 inode_inc_iversion(new_dir);
8901 inode_inc_iversion(old_inode);
8902 inode_inc_iversion(new_inode);
8903 old_dir->i_ctime = old_dir->i_mtime = ctime;
8904 new_dir->i_ctime = new_dir->i_mtime = ctime;
8905 old_inode->i_ctime = ctime;
8906 new_inode->i_ctime = ctime;
8908 if (old_dentry->d_parent != new_dentry->d_parent) {
8909 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8910 BTRFS_I(old_inode), 1);
8911 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8912 BTRFS_I(new_inode), 1);
8915 /* src is a subvolume */
8916 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8917 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
8918 } else { /* src is an inode */
8919 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
8920 BTRFS_I(old_dentry->d_inode),
8921 old_dentry->d_name.name,
8922 old_dentry->d_name.len);
8924 ret = btrfs_update_inode(trans, root, old_inode);
8927 btrfs_abort_transaction(trans, ret);
8931 /* dest is a subvolume */
8932 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8933 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
8934 } else { /* dest is an inode */
8935 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
8936 BTRFS_I(new_dentry->d_inode),
8937 new_dentry->d_name.name,
8938 new_dentry->d_name.len);
8940 ret = btrfs_update_inode(trans, dest, new_inode);
8943 btrfs_abort_transaction(trans, ret);
8947 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8948 new_dentry->d_name.name,
8949 new_dentry->d_name.len, 0, old_idx);
8951 btrfs_abort_transaction(trans, ret);
8955 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8956 old_dentry->d_name.name,
8957 old_dentry->d_name.len, 0, new_idx);
8959 btrfs_abort_transaction(trans, ret);
8963 if (old_inode->i_nlink == 1)
8964 BTRFS_I(old_inode)->dir_index = old_idx;
8965 if (new_inode->i_nlink == 1)
8966 BTRFS_I(new_inode)->dir_index = new_idx;
8968 if (root_log_pinned) {
8969 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
8970 new_dentry->d_parent);
8971 btrfs_end_log_trans(root);
8972 root_log_pinned = false;
8974 if (dest_log_pinned) {
8975 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
8976 old_dentry->d_parent);
8977 btrfs_end_log_trans(dest);
8978 dest_log_pinned = false;
8982 * If we have pinned a log and an error happened, we unpin tasks
8983 * trying to sync the log and force them to fallback to a transaction
8984 * commit if the log currently contains any of the inodes involved in
8985 * this rename operation (to ensure we do not persist a log with an
8986 * inconsistent state for any of these inodes or leading to any
8987 * inconsistencies when replayed). If the transaction was aborted, the
8988 * abortion reason is propagated to userspace when attempting to commit
8989 * the transaction. If the log does not contain any of these inodes, we
8990 * allow the tasks to sync it.
8992 if (ret && (root_log_pinned || dest_log_pinned)) {
8993 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
8994 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
8995 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
8997 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
8998 btrfs_set_log_full_commit(trans);
9000 if (root_log_pinned) {
9001 btrfs_end_log_trans(root);
9002 root_log_pinned = false;
9004 if (dest_log_pinned) {
9005 btrfs_end_log_trans(dest);
9006 dest_log_pinned = false;
9009 ret2 = btrfs_end_transaction(trans);
9010 ret = ret ? ret : ret2;
9012 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9013 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9014 up_read(&fs_info->subvol_sem);
9019 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9020 struct btrfs_root *root,
9022 struct dentry *dentry)
9025 struct inode *inode;
9029 ret = btrfs_find_free_ino(root, &objectid);
9033 inode = btrfs_new_inode(trans, root, dir,
9034 dentry->d_name.name,
9036 btrfs_ino(BTRFS_I(dir)),
9038 S_IFCHR | WHITEOUT_MODE,
9041 if (IS_ERR(inode)) {
9042 ret = PTR_ERR(inode);
9046 inode->i_op = &btrfs_special_inode_operations;
9047 init_special_inode(inode, inode->i_mode,
9050 ret = btrfs_init_inode_security(trans, inode, dir,
9055 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9056 BTRFS_I(inode), 0, index);
9060 ret = btrfs_update_inode(trans, root, inode);
9062 unlock_new_inode(inode);
9064 inode_dec_link_count(inode);
9070 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9071 struct inode *new_dir, struct dentry *new_dentry,
9074 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9075 struct btrfs_trans_handle *trans;
9076 unsigned int trans_num_items;
9077 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9078 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9079 struct inode *new_inode = d_inode(new_dentry);
9080 struct inode *old_inode = d_inode(old_dentry);
9084 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9085 bool log_pinned = false;
9087 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9090 /* we only allow rename subvolume link between subvolumes */
9091 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9094 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9095 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9098 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9099 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9103 /* check for collisions, even if the name isn't there */
9104 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9105 new_dentry->d_name.name,
9106 new_dentry->d_name.len);
9109 if (ret == -EEXIST) {
9111 * eexist without a new_inode */
9112 if (WARN_ON(!new_inode)) {
9116 /* maybe -EOVERFLOW */
9123 * we're using rename to replace one file with another. Start IO on it
9124 * now so we don't add too much work to the end of the transaction
9126 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9127 filemap_flush(old_inode->i_mapping);
9129 /* close the racy window with snapshot create/destroy ioctl */
9130 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9131 down_read(&fs_info->subvol_sem);
9133 * We want to reserve the absolute worst case amount of items. So if
9134 * both inodes are subvols and we need to unlink them then that would
9135 * require 4 item modifications, but if they are both normal inodes it
9136 * would require 5 item modifications, so we'll assume they are normal
9137 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9138 * should cover the worst case number of items we'll modify.
9139 * If our rename has the whiteout flag, we need more 5 units for the
9140 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9141 * when selinux is enabled).
9143 trans_num_items = 11;
9144 if (flags & RENAME_WHITEOUT)
9145 trans_num_items += 5;
9146 trans = btrfs_start_transaction(root, trans_num_items);
9147 if (IS_ERR(trans)) {
9148 ret = PTR_ERR(trans);
9153 btrfs_record_root_in_trans(trans, dest);
9155 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9159 BTRFS_I(old_inode)->dir_index = 0ULL;
9160 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9161 /* force full log commit if subvolume involved. */
9162 btrfs_set_log_full_commit(trans);
9164 btrfs_pin_log_trans(root);
9166 ret = btrfs_insert_inode_ref(trans, dest,
9167 new_dentry->d_name.name,
9168 new_dentry->d_name.len,
9170 btrfs_ino(BTRFS_I(new_dir)), index);
9175 inode_inc_iversion(old_dir);
9176 inode_inc_iversion(new_dir);
9177 inode_inc_iversion(old_inode);
9178 old_dir->i_ctime = old_dir->i_mtime =
9179 new_dir->i_ctime = new_dir->i_mtime =
9180 old_inode->i_ctime = current_time(old_dir);
9182 if (old_dentry->d_parent != new_dentry->d_parent)
9183 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9184 BTRFS_I(old_inode), 1);
9186 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9187 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9189 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9190 BTRFS_I(d_inode(old_dentry)),
9191 old_dentry->d_name.name,
9192 old_dentry->d_name.len);
9194 ret = btrfs_update_inode(trans, root, old_inode);
9197 btrfs_abort_transaction(trans, ret);
9202 inode_inc_iversion(new_inode);
9203 new_inode->i_ctime = current_time(new_inode);
9204 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9205 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9206 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9207 BUG_ON(new_inode->i_nlink == 0);
9209 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9210 BTRFS_I(d_inode(new_dentry)),
9211 new_dentry->d_name.name,
9212 new_dentry->d_name.len);
9214 if (!ret && new_inode->i_nlink == 0)
9215 ret = btrfs_orphan_add(trans,
9216 BTRFS_I(d_inode(new_dentry)));
9218 btrfs_abort_transaction(trans, ret);
9223 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9224 new_dentry->d_name.name,
9225 new_dentry->d_name.len, 0, index);
9227 btrfs_abort_transaction(trans, ret);
9231 if (old_inode->i_nlink == 1)
9232 BTRFS_I(old_inode)->dir_index = index;
9235 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9236 new_dentry->d_parent);
9237 btrfs_end_log_trans(root);
9241 if (flags & RENAME_WHITEOUT) {
9242 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9246 btrfs_abort_transaction(trans, ret);
9252 * If we have pinned the log and an error happened, we unpin tasks
9253 * trying to sync the log and force them to fallback to a transaction
9254 * commit if the log currently contains any of the inodes involved in
9255 * this rename operation (to ensure we do not persist a log with an
9256 * inconsistent state for any of these inodes or leading to any
9257 * inconsistencies when replayed). If the transaction was aborted, the
9258 * abortion reason is propagated to userspace when attempting to commit
9259 * the transaction. If the log does not contain any of these inodes, we
9260 * allow the tasks to sync it.
9262 if (ret && log_pinned) {
9263 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9264 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9265 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9267 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9268 btrfs_set_log_full_commit(trans);
9270 btrfs_end_log_trans(root);
9273 ret2 = btrfs_end_transaction(trans);
9274 ret = ret ? ret : ret2;
9276 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9277 up_read(&fs_info->subvol_sem);
9282 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9283 struct inode *new_dir, struct dentry *new_dentry,
9286 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9289 if (flags & RENAME_EXCHANGE)
9290 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9293 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9296 struct btrfs_delalloc_work {
9297 struct inode *inode;
9298 struct completion completion;
9299 struct list_head list;
9300 struct btrfs_work work;
9303 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9305 struct btrfs_delalloc_work *delalloc_work;
9306 struct inode *inode;
9308 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9310 inode = delalloc_work->inode;
9311 filemap_flush(inode->i_mapping);
9312 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9313 &BTRFS_I(inode)->runtime_flags))
9314 filemap_flush(inode->i_mapping);
9317 complete(&delalloc_work->completion);
9320 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9322 struct btrfs_delalloc_work *work;
9324 work = kmalloc(sizeof(*work), GFP_NOFS);
9328 init_completion(&work->completion);
9329 INIT_LIST_HEAD(&work->list);
9330 work->inode = inode;
9331 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9337 * some fairly slow code that needs optimization. This walks the list
9338 * of all the inodes with pending delalloc and forces them to disk.
9340 static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot)
9342 struct btrfs_inode *binode;
9343 struct inode *inode;
9344 struct btrfs_delalloc_work *work, *next;
9345 struct list_head works;
9346 struct list_head splice;
9349 INIT_LIST_HEAD(&works);
9350 INIT_LIST_HEAD(&splice);
9352 mutex_lock(&root->delalloc_mutex);
9353 spin_lock(&root->delalloc_lock);
9354 list_splice_init(&root->delalloc_inodes, &splice);
9355 while (!list_empty(&splice)) {
9356 binode = list_entry(splice.next, struct btrfs_inode,
9359 list_move_tail(&binode->delalloc_inodes,
9360 &root->delalloc_inodes);
9361 inode = igrab(&binode->vfs_inode);
9363 cond_resched_lock(&root->delalloc_lock);
9366 spin_unlock(&root->delalloc_lock);
9369 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9370 &binode->runtime_flags);
9371 work = btrfs_alloc_delalloc_work(inode);
9377 list_add_tail(&work->list, &works);
9378 btrfs_queue_work(root->fs_info->flush_workers,
9380 if (*nr != U64_MAX) {
9386 spin_lock(&root->delalloc_lock);
9388 spin_unlock(&root->delalloc_lock);
9391 list_for_each_entry_safe(work, next, &works, list) {
9392 list_del_init(&work->list);
9393 wait_for_completion(&work->completion);
9397 if (!list_empty(&splice)) {
9398 spin_lock(&root->delalloc_lock);
9399 list_splice_tail(&splice, &root->delalloc_inodes);
9400 spin_unlock(&root->delalloc_lock);
9402 mutex_unlock(&root->delalloc_mutex);
9406 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9408 struct btrfs_fs_info *fs_info = root->fs_info;
9411 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9414 return start_delalloc_inodes(root, &nr, true);
9417 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr)
9419 struct btrfs_root *root;
9420 struct list_head splice;
9423 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9426 INIT_LIST_HEAD(&splice);
9428 mutex_lock(&fs_info->delalloc_root_mutex);
9429 spin_lock(&fs_info->delalloc_root_lock);
9430 list_splice_init(&fs_info->delalloc_roots, &splice);
9431 while (!list_empty(&splice) && nr) {
9432 root = list_first_entry(&splice, struct btrfs_root,
9434 root = btrfs_grab_root(root);
9436 list_move_tail(&root->delalloc_root,
9437 &fs_info->delalloc_roots);
9438 spin_unlock(&fs_info->delalloc_root_lock);
9440 ret = start_delalloc_inodes(root, &nr, false);
9441 btrfs_put_root(root);
9444 spin_lock(&fs_info->delalloc_root_lock);
9446 spin_unlock(&fs_info->delalloc_root_lock);
9450 if (!list_empty(&splice)) {
9451 spin_lock(&fs_info->delalloc_root_lock);
9452 list_splice_tail(&splice, &fs_info->delalloc_roots);
9453 spin_unlock(&fs_info->delalloc_root_lock);
9455 mutex_unlock(&fs_info->delalloc_root_mutex);
9459 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9460 const char *symname)
9462 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9463 struct btrfs_trans_handle *trans;
9464 struct btrfs_root *root = BTRFS_I(dir)->root;
9465 struct btrfs_path *path;
9466 struct btrfs_key key;
9467 struct inode *inode = NULL;
9474 struct btrfs_file_extent_item *ei;
9475 struct extent_buffer *leaf;
9477 name_len = strlen(symname);
9478 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9479 return -ENAMETOOLONG;
9482 * 2 items for inode item and ref
9483 * 2 items for dir items
9484 * 1 item for updating parent inode item
9485 * 1 item for the inline extent item
9486 * 1 item for xattr if selinux is on
9488 trans = btrfs_start_transaction(root, 7);
9490 return PTR_ERR(trans);
9492 err = btrfs_find_free_ino(root, &objectid);
9496 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9497 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9498 objectid, S_IFLNK|S_IRWXUGO, &index);
9499 if (IS_ERR(inode)) {
9500 err = PTR_ERR(inode);
9506 * If the active LSM wants to access the inode during
9507 * d_instantiate it needs these. Smack checks to see
9508 * if the filesystem supports xattrs by looking at the
9511 inode->i_fop = &btrfs_file_operations;
9512 inode->i_op = &btrfs_file_inode_operations;
9513 inode->i_mapping->a_ops = &btrfs_aops;
9514 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9516 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9520 path = btrfs_alloc_path();
9525 key.objectid = btrfs_ino(BTRFS_I(inode));
9527 key.type = BTRFS_EXTENT_DATA_KEY;
9528 datasize = btrfs_file_extent_calc_inline_size(name_len);
9529 err = btrfs_insert_empty_item(trans, root, path, &key,
9532 btrfs_free_path(path);
9535 leaf = path->nodes[0];
9536 ei = btrfs_item_ptr(leaf, path->slots[0],
9537 struct btrfs_file_extent_item);
9538 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9539 btrfs_set_file_extent_type(leaf, ei,
9540 BTRFS_FILE_EXTENT_INLINE);
9541 btrfs_set_file_extent_encryption(leaf, ei, 0);
9542 btrfs_set_file_extent_compression(leaf, ei, 0);
9543 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9544 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9546 ptr = btrfs_file_extent_inline_start(ei);
9547 write_extent_buffer(leaf, symname, ptr, name_len);
9548 btrfs_mark_buffer_dirty(leaf);
9549 btrfs_free_path(path);
9551 inode->i_op = &btrfs_symlink_inode_operations;
9552 inode_nohighmem(inode);
9553 inode_set_bytes(inode, name_len);
9554 btrfs_i_size_write(BTRFS_I(inode), name_len);
9555 err = btrfs_update_inode(trans, root, inode);
9557 * Last step, add directory indexes for our symlink inode. This is the
9558 * last step to avoid extra cleanup of these indexes if an error happens
9562 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9563 BTRFS_I(inode), 0, index);
9567 d_instantiate_new(dentry, inode);
9570 btrfs_end_transaction(trans);
9572 inode_dec_link_count(inode);
9573 discard_new_inode(inode);
9575 btrfs_btree_balance_dirty(fs_info);
9579 static int insert_prealloc_file_extent(struct btrfs_trans_handle *trans,
9580 struct inode *inode, struct btrfs_key *ins,
9583 struct btrfs_file_extent_item stack_fi;
9584 u64 start = ins->objectid;
9585 u64 len = ins->offset;
9588 memset(&stack_fi, 0, sizeof(stack_fi));
9590 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9591 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9592 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9593 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9594 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9595 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9596 /* Encryption and other encoding is reserved and all 0 */
9598 ret = btrfs_qgroup_release_data(BTRFS_I(inode), file_offset, len);
9601 return insert_reserved_file_extent(trans, BTRFS_I(inode), file_offset,
9604 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9605 u64 start, u64 num_bytes, u64 min_size,
9606 loff_t actual_len, u64 *alloc_hint,
9607 struct btrfs_trans_handle *trans)
9609 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9610 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9611 struct extent_map *em;
9612 struct btrfs_root *root = BTRFS_I(inode)->root;
9613 struct btrfs_key ins;
9614 u64 cur_offset = start;
9615 u64 clear_offset = start;
9618 u64 last_alloc = (u64)-1;
9620 bool own_trans = true;
9621 u64 end = start + num_bytes - 1;
9625 while (num_bytes > 0) {
9627 trans = btrfs_start_transaction(root, 3);
9628 if (IS_ERR(trans)) {
9629 ret = PTR_ERR(trans);
9634 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9635 cur_bytes = max(cur_bytes, min_size);
9637 * If we are severely fragmented we could end up with really
9638 * small allocations, so if the allocator is returning small
9639 * chunks lets make its job easier by only searching for those
9642 cur_bytes = min(cur_bytes, last_alloc);
9643 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9644 min_size, 0, *alloc_hint, &ins, 1, 0);
9647 btrfs_end_transaction(trans);
9652 * We've reserved this space, and thus converted it from
9653 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9654 * from here on out we will only need to clear our reservation
9655 * for the remaining unreserved area, so advance our
9656 * clear_offset by our extent size.
9658 clear_offset += ins.offset;
9659 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9661 last_alloc = ins.offset;
9662 ret = insert_prealloc_file_extent(trans, inode, &ins, cur_offset);
9664 btrfs_free_reserved_extent(fs_info, ins.objectid,
9666 btrfs_abort_transaction(trans, ret);
9668 btrfs_end_transaction(trans);
9672 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9673 cur_offset + ins.offset -1, 0);
9675 em = alloc_extent_map();
9677 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9678 &BTRFS_I(inode)->runtime_flags);
9682 em->start = cur_offset;
9683 em->orig_start = cur_offset;
9684 em->len = ins.offset;
9685 em->block_start = ins.objectid;
9686 em->block_len = ins.offset;
9687 em->orig_block_len = ins.offset;
9688 em->ram_bytes = ins.offset;
9689 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9690 em->generation = trans->transid;
9693 write_lock(&em_tree->lock);
9694 ret = add_extent_mapping(em_tree, em, 1);
9695 write_unlock(&em_tree->lock);
9698 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9699 cur_offset + ins.offset - 1,
9702 free_extent_map(em);
9704 num_bytes -= ins.offset;
9705 cur_offset += ins.offset;
9706 *alloc_hint = ins.objectid + ins.offset;
9708 inode_inc_iversion(inode);
9709 inode->i_ctime = current_time(inode);
9710 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9711 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9712 (actual_len > inode->i_size) &&
9713 (cur_offset > inode->i_size)) {
9714 if (cur_offset > actual_len)
9715 i_size = actual_len;
9717 i_size = cur_offset;
9718 i_size_write(inode, i_size);
9719 btrfs_inode_safe_disk_i_size_write(inode, 0);
9722 ret = btrfs_update_inode(trans, root, inode);
9725 btrfs_abort_transaction(trans, ret);
9727 btrfs_end_transaction(trans);
9732 btrfs_end_transaction(trans);
9734 if (clear_offset < end)
9735 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9736 end - clear_offset + 1);
9740 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9741 u64 start, u64 num_bytes, u64 min_size,
9742 loff_t actual_len, u64 *alloc_hint)
9744 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9745 min_size, actual_len, alloc_hint,
9749 int btrfs_prealloc_file_range_trans(struct inode *inode,
9750 struct btrfs_trans_handle *trans, int mode,
9751 u64 start, u64 num_bytes, u64 min_size,
9752 loff_t actual_len, u64 *alloc_hint)
9754 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9755 min_size, actual_len, alloc_hint, trans);
9758 static int btrfs_set_page_dirty(struct page *page)
9760 return __set_page_dirty_nobuffers(page);
9763 static int btrfs_permission(struct inode *inode, int mask)
9765 struct btrfs_root *root = BTRFS_I(inode)->root;
9766 umode_t mode = inode->i_mode;
9768 if (mask & MAY_WRITE &&
9769 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9770 if (btrfs_root_readonly(root))
9772 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9775 return generic_permission(inode, mask);
9778 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9780 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9781 struct btrfs_trans_handle *trans;
9782 struct btrfs_root *root = BTRFS_I(dir)->root;
9783 struct inode *inode = NULL;
9789 * 5 units required for adding orphan entry
9791 trans = btrfs_start_transaction(root, 5);
9793 return PTR_ERR(trans);
9795 ret = btrfs_find_free_ino(root, &objectid);
9799 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9800 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9801 if (IS_ERR(inode)) {
9802 ret = PTR_ERR(inode);
9807 inode->i_fop = &btrfs_file_operations;
9808 inode->i_op = &btrfs_file_inode_operations;
9810 inode->i_mapping->a_ops = &btrfs_aops;
9811 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9813 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9817 ret = btrfs_update_inode(trans, root, inode);
9820 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9825 * We set number of links to 0 in btrfs_new_inode(), and here we set
9826 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9829 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9831 set_nlink(inode, 1);
9832 d_tmpfile(dentry, inode);
9833 unlock_new_inode(inode);
9834 mark_inode_dirty(inode);
9836 btrfs_end_transaction(trans);
9838 discard_new_inode(inode);
9839 btrfs_btree_balance_dirty(fs_info);
9843 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9845 struct inode *inode = tree->private_data;
9846 unsigned long index = start >> PAGE_SHIFT;
9847 unsigned long end_index = end >> PAGE_SHIFT;
9850 while (index <= end_index) {
9851 page = find_get_page(inode->i_mapping, index);
9852 ASSERT(page); /* Pages should be in the extent_io_tree */
9853 set_page_writeback(page);
9861 * Add an entry indicating a block group or device which is pinned by a
9862 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9863 * negative errno on failure.
9865 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
9866 bool is_block_group)
9868 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9869 struct btrfs_swapfile_pin *sp, *entry;
9871 struct rb_node *parent = NULL;
9873 sp = kmalloc(sizeof(*sp), GFP_NOFS);
9878 sp->is_block_group = is_block_group;
9880 spin_lock(&fs_info->swapfile_pins_lock);
9881 p = &fs_info->swapfile_pins.rb_node;
9884 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
9885 if (sp->ptr < entry->ptr ||
9886 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
9888 } else if (sp->ptr > entry->ptr ||
9889 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
9890 p = &(*p)->rb_right;
9892 spin_unlock(&fs_info->swapfile_pins_lock);
9897 rb_link_node(&sp->node, parent, p);
9898 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
9899 spin_unlock(&fs_info->swapfile_pins_lock);
9903 /* Free all of the entries pinned by this swapfile. */
9904 static void btrfs_free_swapfile_pins(struct inode *inode)
9906 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9907 struct btrfs_swapfile_pin *sp;
9908 struct rb_node *node, *next;
9910 spin_lock(&fs_info->swapfile_pins_lock);
9911 node = rb_first(&fs_info->swapfile_pins);
9913 next = rb_next(node);
9914 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
9915 if (sp->inode == inode) {
9916 rb_erase(&sp->node, &fs_info->swapfile_pins);
9917 if (sp->is_block_group)
9918 btrfs_put_block_group(sp->ptr);
9923 spin_unlock(&fs_info->swapfile_pins_lock);
9926 struct btrfs_swap_info {
9932 unsigned long nr_pages;
9936 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
9937 struct btrfs_swap_info *bsi)
9939 unsigned long nr_pages;
9940 u64 first_ppage, first_ppage_reported, next_ppage;
9943 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
9944 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
9945 PAGE_SIZE) >> PAGE_SHIFT;
9947 if (first_ppage >= next_ppage)
9949 nr_pages = next_ppage - first_ppage;
9951 first_ppage_reported = first_ppage;
9952 if (bsi->start == 0)
9953 first_ppage_reported++;
9954 if (bsi->lowest_ppage > first_ppage_reported)
9955 bsi->lowest_ppage = first_ppage_reported;
9956 if (bsi->highest_ppage < (next_ppage - 1))
9957 bsi->highest_ppage = next_ppage - 1;
9959 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
9962 bsi->nr_extents += ret;
9963 bsi->nr_pages += nr_pages;
9967 static void btrfs_swap_deactivate(struct file *file)
9969 struct inode *inode = file_inode(file);
9971 btrfs_free_swapfile_pins(inode);
9972 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
9975 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
9978 struct inode *inode = file_inode(file);
9979 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9980 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9981 struct extent_state *cached_state = NULL;
9982 struct extent_map *em = NULL;
9983 struct btrfs_device *device = NULL;
9984 struct btrfs_swap_info bsi = {
9985 .lowest_ppage = (sector_t)-1ULL,
9992 * If the swap file was just created, make sure delalloc is done. If the
9993 * file changes again after this, the user is doing something stupid and
9994 * we don't really care.
9996 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10001 * The inode is locked, so these flags won't change after we check them.
10003 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10004 btrfs_warn(fs_info, "swapfile must not be compressed");
10007 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10008 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10011 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10012 btrfs_warn(fs_info, "swapfile must not be checksummed");
10017 * Balance or device remove/replace/resize can move stuff around from
10018 * under us. The exclop protection makes sure they aren't running/won't
10019 * run concurrently while we are mapping the swap extents, and
10020 * fs_info->swapfile_pins prevents them from running while the swap
10021 * file is active and moving the extents. Note that this also prevents
10022 * a concurrent device add which isn't actually necessary, but it's not
10023 * really worth the trouble to allow it.
10025 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10026 btrfs_warn(fs_info,
10027 "cannot activate swapfile while exclusive operation is running");
10031 * Snapshots can create extents which require COW even if NODATACOW is
10032 * set. We use this counter to prevent snapshots. We must increment it
10033 * before walking the extents because we don't want a concurrent
10034 * snapshot to run after we've already checked the extents.
10036 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10038 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10040 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10042 while (start < isize) {
10043 u64 logical_block_start, physical_block_start;
10044 struct btrfs_block_group *bg;
10045 u64 len = isize - start;
10047 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10053 if (em->block_start == EXTENT_MAP_HOLE) {
10054 btrfs_warn(fs_info, "swapfile must not have holes");
10058 if (em->block_start == EXTENT_MAP_INLINE) {
10060 * It's unlikely we'll ever actually find ourselves
10061 * here, as a file small enough to fit inline won't be
10062 * big enough to store more than the swap header, but in
10063 * case something changes in the future, let's catch it
10064 * here rather than later.
10066 btrfs_warn(fs_info, "swapfile must not be inline");
10070 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10071 btrfs_warn(fs_info, "swapfile must not be compressed");
10076 logical_block_start = em->block_start + (start - em->start);
10077 len = min(len, em->len - (start - em->start));
10078 free_extent_map(em);
10081 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10087 btrfs_warn(fs_info,
10088 "swapfile must not be copy-on-write");
10093 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10099 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10100 btrfs_warn(fs_info,
10101 "swapfile must have single data profile");
10106 if (device == NULL) {
10107 device = em->map_lookup->stripes[0].dev;
10108 ret = btrfs_add_swapfile_pin(inode, device, false);
10113 } else if (device != em->map_lookup->stripes[0].dev) {
10114 btrfs_warn(fs_info, "swapfile must be on one device");
10119 physical_block_start = (em->map_lookup->stripes[0].physical +
10120 (logical_block_start - em->start));
10121 len = min(len, em->len - (logical_block_start - em->start));
10122 free_extent_map(em);
10125 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10127 btrfs_warn(fs_info,
10128 "could not find block group containing swapfile");
10133 ret = btrfs_add_swapfile_pin(inode, bg, true);
10135 btrfs_put_block_group(bg);
10142 if (bsi.block_len &&
10143 bsi.block_start + bsi.block_len == physical_block_start) {
10144 bsi.block_len += len;
10146 if (bsi.block_len) {
10147 ret = btrfs_add_swap_extent(sis, &bsi);
10152 bsi.block_start = physical_block_start;
10153 bsi.block_len = len;
10160 ret = btrfs_add_swap_extent(sis, &bsi);
10163 if (!IS_ERR_OR_NULL(em))
10164 free_extent_map(em);
10166 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10169 btrfs_swap_deactivate(file);
10171 btrfs_exclop_finish(fs_info);
10177 sis->bdev = device->bdev;
10178 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10179 sis->max = bsi.nr_pages;
10180 sis->pages = bsi.nr_pages - 1;
10181 sis->highest_bit = bsi.nr_pages - 1;
10182 return bsi.nr_extents;
10185 static void btrfs_swap_deactivate(struct file *file)
10189 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10192 return -EOPNOTSUPP;
10196 static const struct inode_operations btrfs_dir_inode_operations = {
10197 .getattr = btrfs_getattr,
10198 .lookup = btrfs_lookup,
10199 .create = btrfs_create,
10200 .unlink = btrfs_unlink,
10201 .link = btrfs_link,
10202 .mkdir = btrfs_mkdir,
10203 .rmdir = btrfs_rmdir,
10204 .rename = btrfs_rename2,
10205 .symlink = btrfs_symlink,
10206 .setattr = btrfs_setattr,
10207 .mknod = btrfs_mknod,
10208 .listxattr = btrfs_listxattr,
10209 .permission = btrfs_permission,
10210 .get_acl = btrfs_get_acl,
10211 .set_acl = btrfs_set_acl,
10212 .update_time = btrfs_update_time,
10213 .tmpfile = btrfs_tmpfile,
10216 static const struct file_operations btrfs_dir_file_operations = {
10217 .llseek = generic_file_llseek,
10218 .read = generic_read_dir,
10219 .iterate_shared = btrfs_real_readdir,
10220 .open = btrfs_opendir,
10221 .unlocked_ioctl = btrfs_ioctl,
10222 #ifdef CONFIG_COMPAT
10223 .compat_ioctl = btrfs_compat_ioctl,
10225 .release = btrfs_release_file,
10226 .fsync = btrfs_sync_file,
10229 static const struct extent_io_ops btrfs_extent_io_ops = {
10230 /* mandatory callbacks */
10231 .submit_bio_hook = btrfs_submit_bio_hook,
10232 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10236 * btrfs doesn't support the bmap operation because swapfiles
10237 * use bmap to make a mapping of extents in the file. They assume
10238 * these extents won't change over the life of the file and they
10239 * use the bmap result to do IO directly to the drive.
10241 * the btrfs bmap call would return logical addresses that aren't
10242 * suitable for IO and they also will change frequently as COW
10243 * operations happen. So, swapfile + btrfs == corruption.
10245 * For now we're avoiding this by dropping bmap.
10247 static const struct address_space_operations btrfs_aops = {
10248 .readpage = btrfs_readpage,
10249 .writepage = btrfs_writepage,
10250 .writepages = btrfs_writepages,
10251 .readahead = btrfs_readahead,
10252 .direct_IO = noop_direct_IO,
10253 .invalidatepage = btrfs_invalidatepage,
10254 .releasepage = btrfs_releasepage,
10255 #ifdef CONFIG_MIGRATION
10256 .migratepage = btrfs_migratepage,
10258 .set_page_dirty = btrfs_set_page_dirty,
10259 .error_remove_page = generic_error_remove_page,
10260 .swap_activate = btrfs_swap_activate,
10261 .swap_deactivate = btrfs_swap_deactivate,
10264 static const struct inode_operations btrfs_file_inode_operations = {
10265 .getattr = btrfs_getattr,
10266 .setattr = btrfs_setattr,
10267 .listxattr = btrfs_listxattr,
10268 .permission = btrfs_permission,
10269 .fiemap = btrfs_fiemap,
10270 .get_acl = btrfs_get_acl,
10271 .set_acl = btrfs_set_acl,
10272 .update_time = btrfs_update_time,
10274 static const struct inode_operations btrfs_special_inode_operations = {
10275 .getattr = btrfs_getattr,
10276 .setattr = btrfs_setattr,
10277 .permission = btrfs_permission,
10278 .listxattr = btrfs_listxattr,
10279 .get_acl = btrfs_get_acl,
10280 .set_acl = btrfs_set_acl,
10281 .update_time = btrfs_update_time,
10283 static const struct inode_operations btrfs_symlink_inode_operations = {
10284 .get_link = page_get_link,
10285 .getattr = btrfs_getattr,
10286 .setattr = btrfs_setattr,
10287 .permission = btrfs_permission,
10288 .listxattr = btrfs_listxattr,
10289 .update_time = btrfs_update_time,
10292 const struct dentry_operations btrfs_dentry_operations = {
10293 .d_delete = btrfs_dentry_delete,