1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
9 #include <linux/blk-cgroup.h>
10 #include <linux/file.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/xattr.h>
21 #include <linux/posix_acl.h>
22 #include <linux/falloc.h>
23 #include <linux/slab.h>
24 #include <linux/ratelimit.h>
25 #include <linux/btrfs.h>
26 #include <linux/blkdev.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/uio.h>
29 #include <linux/magic.h>
30 #include <linux/iversion.h>
31 #include <linux/swap.h>
32 #include <linux/migrate.h>
33 #include <linux/sched/mm.h>
34 #include <linux/iomap.h>
35 #include <asm/unaligned.h>
36 #include <linux/fsverity.h>
40 #include "transaction.h"
41 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "ordered-data.h"
47 #include "compression.h"
49 #include "free-space-cache.h"
52 #include "delalloc-space.h"
53 #include "block-group.h"
54 #include "space-info.h"
57 #include "inode-item.h"
59 #include "accessors.h"
60 #include "extent-tree.h"
61 #include "root-tree.h"
64 #include "file-item.h"
65 #include "uuid-tree.h"
69 #include "relocation.h"
75 struct btrfs_iget_args {
77 struct btrfs_root *root;
80 struct btrfs_dio_data {
82 struct extent_changeset *data_reserved;
83 struct btrfs_ordered_extent *ordered;
84 bool data_space_reserved;
88 struct btrfs_dio_private {
93 /* This must be last */
94 struct btrfs_bio bbio;
97 static struct bio_set btrfs_dio_bioset;
99 struct btrfs_rename_ctx {
100 /* Output field. Stores the index number of the old directory entry. */
105 * Used by data_reloc_print_warning_inode() to pass needed info for filename
106 * resolution and output of error message.
108 struct data_reloc_warn {
109 struct btrfs_path path;
110 struct btrfs_fs_info *fs_info;
111 u64 extent_item_size;
116 static const struct inode_operations btrfs_dir_inode_operations;
117 static const struct inode_operations btrfs_symlink_inode_operations;
118 static const struct inode_operations btrfs_special_inode_operations;
119 static const struct inode_operations btrfs_file_inode_operations;
120 static const struct address_space_operations btrfs_aops;
121 static const struct file_operations btrfs_dir_file_operations;
123 static struct kmem_cache *btrfs_inode_cachep;
125 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
126 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
127 static noinline int cow_file_range(struct btrfs_inode *inode,
128 struct page *locked_page,
129 u64 start, u64 end, int *page_started,
130 unsigned long *nr_written, int unlock,
132 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
133 u64 len, u64 orig_start, u64 block_start,
134 u64 block_len, u64 orig_block_len,
135 u64 ram_bytes, int compress_type,
138 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
139 u64 root, void *warn_ctx)
141 struct data_reloc_warn *warn = warn_ctx;
142 struct btrfs_fs_info *fs_info = warn->fs_info;
143 struct extent_buffer *eb;
144 struct btrfs_inode_item *inode_item;
145 struct inode_fs_paths *ipath = NULL;
146 struct btrfs_root *local_root;
147 struct btrfs_key key;
148 unsigned int nofs_flag;
152 local_root = btrfs_get_fs_root(fs_info, root, true);
153 if (IS_ERR(local_root)) {
154 ret = PTR_ERR(local_root);
158 /* This makes the path point to (inum INODE_ITEM ioff). */
160 key.type = BTRFS_INODE_ITEM_KEY;
163 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
165 btrfs_put_root(local_root);
166 btrfs_release_path(&warn->path);
170 eb = warn->path.nodes[0];
171 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
172 nlink = btrfs_inode_nlink(eb, inode_item);
173 btrfs_release_path(&warn->path);
175 nofs_flag = memalloc_nofs_save();
176 ipath = init_ipath(4096, local_root, &warn->path);
177 memalloc_nofs_restore(nofs_flag);
179 btrfs_put_root(local_root);
180 ret = PTR_ERR(ipath);
183 * -ENOMEM, not a critical error, just output an generic error
187 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
188 warn->logical, warn->mirror_num, root, inum, offset);
191 ret = paths_from_inode(inum, ipath);
196 * We deliberately ignore the bit ipath might have been too small to
197 * hold all of the paths here
199 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
201 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
202 warn->logical, warn->mirror_num, root, inum, offset,
203 fs_info->sectorsize, nlink,
204 (char *)(unsigned long)ipath->fspath->val[i]);
207 btrfs_put_root(local_root);
213 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
214 warn->logical, warn->mirror_num, root, inum, offset, ret);
221 * Do extra user-friendly error output (e.g. lookup all the affected files).
223 * Return true if we succeeded doing the backref lookup.
224 * Return false if such lookup failed, and has to fallback to the old error message.
226 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
227 const u8 *csum, const u8 *csum_expected,
230 struct btrfs_fs_info *fs_info = inode->root->fs_info;
231 struct btrfs_path path = { 0 };
232 struct btrfs_key found_key = { 0 };
233 struct extent_buffer *eb;
234 struct btrfs_extent_item *ei;
235 const u32 csum_size = fs_info->csum_size;
241 mutex_lock(&fs_info->reloc_mutex);
242 logical = btrfs_get_reloc_bg_bytenr(fs_info);
243 mutex_unlock(&fs_info->reloc_mutex);
245 if (logical == U64_MAX) {
246 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
247 btrfs_warn_rl(fs_info,
248 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
249 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
250 CSUM_FMT_VALUE(csum_size, csum),
251 CSUM_FMT_VALUE(csum_size, csum_expected),
257 btrfs_warn_rl(fs_info,
258 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
259 inode->root->root_key.objectid,
260 btrfs_ino(inode), file_off, logical,
261 CSUM_FMT_VALUE(csum_size, csum),
262 CSUM_FMT_VALUE(csum_size, csum_expected),
265 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
267 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
272 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
273 item_size = btrfs_item_size(eb, path.slots[0]);
274 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
275 unsigned long ptr = 0;
280 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
281 item_size, &ref_root,
284 btrfs_warn_rl(fs_info,
285 "failed to resolve tree backref for logical %llu: %d",
292 btrfs_warn_rl(fs_info,
293 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
295 (ref_level ? "node" : "leaf"),
296 ref_level, ref_root);
298 btrfs_release_path(&path);
300 struct btrfs_backref_walk_ctx ctx = { 0 };
301 struct data_reloc_warn reloc_warn = { 0 };
303 btrfs_release_path(&path);
305 ctx.bytenr = found_key.objectid;
306 ctx.extent_item_pos = logical - found_key.objectid;
307 ctx.fs_info = fs_info;
309 reloc_warn.logical = logical;
310 reloc_warn.extent_item_size = found_key.offset;
311 reloc_warn.mirror_num = mirror_num;
312 reloc_warn.fs_info = fs_info;
314 iterate_extent_inodes(&ctx, true,
315 data_reloc_print_warning_inode, &reloc_warn);
319 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
320 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
322 struct btrfs_root *root = inode->root;
323 const u32 csum_size = root->fs_info->csum_size;
325 /* For data reloc tree, it's better to do a backref lookup instead. */
326 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
327 return print_data_reloc_error(inode, logical_start, csum,
328 csum_expected, mirror_num);
330 /* Output without objectid, which is more meaningful */
331 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
332 btrfs_warn_rl(root->fs_info,
333 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
334 root->root_key.objectid, btrfs_ino(inode),
336 CSUM_FMT_VALUE(csum_size, csum),
337 CSUM_FMT_VALUE(csum_size, csum_expected),
340 btrfs_warn_rl(root->fs_info,
341 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
342 root->root_key.objectid, btrfs_ino(inode),
344 CSUM_FMT_VALUE(csum_size, csum),
345 CSUM_FMT_VALUE(csum_size, csum_expected),
351 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
353 * ilock_flags can have the following bit set:
355 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
356 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
358 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
360 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
362 if (ilock_flags & BTRFS_ILOCK_SHARED) {
363 if (ilock_flags & BTRFS_ILOCK_TRY) {
364 if (!inode_trylock_shared(&inode->vfs_inode))
369 inode_lock_shared(&inode->vfs_inode);
371 if (ilock_flags & BTRFS_ILOCK_TRY) {
372 if (!inode_trylock(&inode->vfs_inode))
377 inode_lock(&inode->vfs_inode);
379 if (ilock_flags & BTRFS_ILOCK_MMAP)
380 down_write(&inode->i_mmap_lock);
385 * btrfs_inode_unlock - unock inode i_rwsem
387 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
388 * to decide whether the lock acquired is shared or exclusive.
390 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
392 if (ilock_flags & BTRFS_ILOCK_MMAP)
393 up_write(&inode->i_mmap_lock);
394 if (ilock_flags & BTRFS_ILOCK_SHARED)
395 inode_unlock_shared(&inode->vfs_inode);
397 inode_unlock(&inode->vfs_inode);
401 * Cleanup all submitted ordered extents in specified range to handle errors
402 * from the btrfs_run_delalloc_range() callback.
404 * NOTE: caller must ensure that when an error happens, it can not call
405 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
406 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
407 * to be released, which we want to happen only when finishing the ordered
408 * extent (btrfs_finish_ordered_io()).
410 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
411 struct page *locked_page,
412 u64 offset, u64 bytes)
414 unsigned long index = offset >> PAGE_SHIFT;
415 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
416 u64 page_start = 0, page_end = 0;
420 page_start = page_offset(locked_page);
421 page_end = page_start + PAGE_SIZE - 1;
424 while (index <= end_index) {
426 * For locked page, we will call end_extent_writepage() on it
427 * in run_delalloc_range() for the error handling. That
428 * end_extent_writepage() function will call
429 * btrfs_mark_ordered_io_finished() to clear page Ordered and
430 * run the ordered extent accounting.
432 * Here we can't just clear the Ordered bit, or
433 * btrfs_mark_ordered_io_finished() would skip the accounting
434 * for the page range, and the ordered extent will never finish.
436 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
440 page = find_get_page(inode->vfs_inode.i_mapping, index);
446 * Here we just clear all Ordered bits for every page in the
447 * range, then btrfs_mark_ordered_io_finished() will handle
448 * the ordered extent accounting for the range.
450 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
456 /* The locked page covers the full range, nothing needs to be done */
457 if (bytes + offset <= page_start + PAGE_SIZE)
460 * In case this page belongs to the delalloc range being
461 * instantiated then skip it, since the first page of a range is
462 * going to be properly cleaned up by the caller of
465 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
466 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
467 offset = page_offset(locked_page) + PAGE_SIZE;
471 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
474 static int btrfs_dirty_inode(struct btrfs_inode *inode);
476 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
477 struct btrfs_new_inode_args *args)
481 if (args->default_acl) {
482 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
488 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
492 if (!args->default_acl && !args->acl)
493 cache_no_acl(args->inode);
494 return btrfs_xattr_security_init(trans, args->inode, args->dir,
495 &args->dentry->d_name);
499 * this does all the hard work for inserting an inline extent into
500 * the btree. The caller should have done a btrfs_drop_extents so that
501 * no overlapping inline items exist in the btree
503 static int insert_inline_extent(struct btrfs_trans_handle *trans,
504 struct btrfs_path *path,
505 struct btrfs_inode *inode, bool extent_inserted,
506 size_t size, size_t compressed_size,
508 struct page **compressed_pages,
511 struct btrfs_root *root = inode->root;
512 struct extent_buffer *leaf;
513 struct page *page = NULL;
516 struct btrfs_file_extent_item *ei;
518 size_t cur_size = size;
521 ASSERT((compressed_size > 0 && compressed_pages) ||
522 (compressed_size == 0 && !compressed_pages));
524 if (compressed_size && compressed_pages)
525 cur_size = compressed_size;
527 if (!extent_inserted) {
528 struct btrfs_key key;
531 key.objectid = btrfs_ino(inode);
533 key.type = BTRFS_EXTENT_DATA_KEY;
535 datasize = btrfs_file_extent_calc_inline_size(cur_size);
536 ret = btrfs_insert_empty_item(trans, root, path, &key,
541 leaf = path->nodes[0];
542 ei = btrfs_item_ptr(leaf, path->slots[0],
543 struct btrfs_file_extent_item);
544 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
545 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
546 btrfs_set_file_extent_encryption(leaf, ei, 0);
547 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
548 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
549 ptr = btrfs_file_extent_inline_start(ei);
551 if (compress_type != BTRFS_COMPRESS_NONE) {
554 while (compressed_size > 0) {
555 cpage = compressed_pages[i];
556 cur_size = min_t(unsigned long, compressed_size,
559 kaddr = kmap_local_page(cpage);
560 write_extent_buffer(leaf, kaddr, ptr, cur_size);
565 compressed_size -= cur_size;
567 btrfs_set_file_extent_compression(leaf, ei,
570 page = find_get_page(inode->vfs_inode.i_mapping, 0);
571 btrfs_set_file_extent_compression(leaf, ei, 0);
572 kaddr = kmap_local_page(page);
573 write_extent_buffer(leaf, kaddr, ptr, size);
577 btrfs_mark_buffer_dirty(leaf);
578 btrfs_release_path(path);
581 * We align size to sectorsize for inline extents just for simplicity
584 ret = btrfs_inode_set_file_extent_range(inode, 0,
585 ALIGN(size, root->fs_info->sectorsize));
590 * We're an inline extent, so nobody can extend the file past i_size
591 * without locking a page we already have locked.
593 * We must do any i_size and inode updates before we unlock the pages.
594 * Otherwise we could end up racing with unlink.
596 i_size = i_size_read(&inode->vfs_inode);
597 if (update_i_size && size > i_size) {
598 i_size_write(&inode->vfs_inode, size);
601 inode->disk_i_size = i_size;
609 * conditionally insert an inline extent into the file. This
610 * does the checks required to make sure the data is small enough
611 * to fit as an inline extent.
613 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
614 size_t compressed_size,
616 struct page **compressed_pages,
619 struct btrfs_drop_extents_args drop_args = { 0 };
620 struct btrfs_root *root = inode->root;
621 struct btrfs_fs_info *fs_info = root->fs_info;
622 struct btrfs_trans_handle *trans;
623 u64 data_len = (compressed_size ?: size);
625 struct btrfs_path *path;
628 * We can create an inline extent if it ends at or beyond the current
629 * i_size, is no larger than a sector (decompressed), and the (possibly
630 * compressed) data fits in a leaf and the configured maximum inline
633 if (size < i_size_read(&inode->vfs_inode) ||
634 size > fs_info->sectorsize ||
635 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
636 data_len > fs_info->max_inline)
639 path = btrfs_alloc_path();
643 trans = btrfs_join_transaction(root);
645 btrfs_free_path(path);
646 return PTR_ERR(trans);
648 trans->block_rsv = &inode->block_rsv;
650 drop_args.path = path;
652 drop_args.end = fs_info->sectorsize;
653 drop_args.drop_cache = true;
654 drop_args.replace_extent = true;
655 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
656 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
658 btrfs_abort_transaction(trans, ret);
662 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
663 size, compressed_size, compress_type,
664 compressed_pages, update_i_size);
665 if (ret && ret != -ENOSPC) {
666 btrfs_abort_transaction(trans, ret);
668 } else if (ret == -ENOSPC) {
673 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
674 ret = btrfs_update_inode(trans, root, inode);
675 if (ret && ret != -ENOSPC) {
676 btrfs_abort_transaction(trans, ret);
678 } else if (ret == -ENOSPC) {
683 btrfs_set_inode_full_sync(inode);
686 * Don't forget to free the reserved space, as for inlined extent
687 * it won't count as data extent, free them directly here.
688 * And at reserve time, it's always aligned to page size, so
689 * just free one page here.
691 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
692 btrfs_free_path(path);
693 btrfs_end_transaction(trans);
697 struct async_extent {
702 unsigned long nr_pages;
704 struct list_head list;
708 struct btrfs_inode *inode;
709 struct page *locked_page;
712 blk_opf_t write_flags;
713 struct list_head extents;
714 struct cgroup_subsys_state *blkcg_css;
715 struct btrfs_work work;
716 struct async_cow *async_cow;
721 struct async_chunk chunks[];
724 static noinline int add_async_extent(struct async_chunk *cow,
725 u64 start, u64 ram_size,
728 unsigned long nr_pages,
731 struct async_extent *async_extent;
733 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
734 BUG_ON(!async_extent); /* -ENOMEM */
735 async_extent->start = start;
736 async_extent->ram_size = ram_size;
737 async_extent->compressed_size = compressed_size;
738 async_extent->pages = pages;
739 async_extent->nr_pages = nr_pages;
740 async_extent->compress_type = compress_type;
741 list_add_tail(&async_extent->list, &cow->extents);
746 * Check if the inode needs to be submitted to compression, based on mount
747 * options, defragmentation, properties or heuristics.
749 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
752 struct btrfs_fs_info *fs_info = inode->root->fs_info;
754 if (!btrfs_inode_can_compress(inode)) {
755 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
756 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
761 * Special check for subpage.
763 * We lock the full page then run each delalloc range in the page, thus
764 * for the following case, we will hit some subpage specific corner case:
767 * | |///////| |///////|
770 * In above case, both range A and range B will try to unlock the full
771 * page [0, 64K), causing the one finished later will have page
772 * unlocked already, triggering various page lock requirement BUG_ON()s.
774 * So here we add an artificial limit that subpage compression can only
775 * if the range is fully page aligned.
777 * In theory we only need to ensure the first page is fully covered, but
778 * the tailing partial page will be locked until the full compression
779 * finishes, delaying the write of other range.
781 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
782 * first to prevent any submitted async extent to unlock the full page.
783 * By this, we can ensure for subpage case that only the last async_cow
784 * will unlock the full page.
786 if (fs_info->sectorsize < PAGE_SIZE) {
787 if (!PAGE_ALIGNED(start) ||
788 !PAGE_ALIGNED(end + 1))
793 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
796 if (inode->defrag_compress)
798 /* bad compression ratios */
799 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
801 if (btrfs_test_opt(fs_info, COMPRESS) ||
802 inode->flags & BTRFS_INODE_COMPRESS ||
803 inode->prop_compress)
804 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
808 static inline void inode_should_defrag(struct btrfs_inode *inode,
809 u64 start, u64 end, u64 num_bytes, u32 small_write)
811 /* If this is a small write inside eof, kick off a defrag */
812 if (num_bytes < small_write &&
813 (start > 0 || end + 1 < inode->disk_i_size))
814 btrfs_add_inode_defrag(NULL, inode, small_write);
818 * we create compressed extents in two phases. The first
819 * phase compresses a range of pages that have already been
820 * locked (both pages and state bits are locked).
822 * This is done inside an ordered work queue, and the compression
823 * is spread across many cpus. The actual IO submission is step
824 * two, and the ordered work queue takes care of making sure that
825 * happens in the same order things were put onto the queue by
826 * writepages and friends.
828 * If this code finds it can't get good compression, it puts an
829 * entry onto the work queue to write the uncompressed bytes. This
830 * makes sure that both compressed inodes and uncompressed inodes
831 * are written in the same order that the flusher thread sent them
834 static noinline int compress_file_range(struct async_chunk *async_chunk)
836 struct btrfs_inode *inode = async_chunk->inode;
837 struct btrfs_fs_info *fs_info = inode->root->fs_info;
838 struct address_space *mapping = inode->vfs_inode.i_mapping;
839 u64 blocksize = fs_info->sectorsize;
840 u64 start = async_chunk->start;
841 u64 end = async_chunk->end;
845 struct page **pages = NULL;
846 unsigned long nr_pages;
847 unsigned long total_compressed = 0;
848 unsigned long total_in = 0;
851 int compress_type = fs_info->compress_type;
852 int compressed_extents = 0;
855 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
858 * We need to save i_size before now because it could change in between
859 * us evaluating the size and assigning it. This is because we lock and
860 * unlock the page in truncate and fallocate, and then modify the i_size
863 * The barriers are to emulate READ_ONCE, remove that once i_size_read
867 i_size = i_size_read(&inode->vfs_inode);
869 actual_end = min_t(u64, i_size, end + 1);
872 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
873 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
876 * we don't want to send crud past the end of i_size through
877 * compression, that's just a waste of CPU time. So, if the
878 * end of the file is before the start of our current
879 * requested range of bytes, we bail out to the uncompressed
880 * cleanup code that can deal with all of this.
882 * It isn't really the fastest way to fix things, but this is a
883 * very uncommon corner.
885 if (actual_end <= start)
886 goto cleanup_and_bail_uncompressed;
888 total_compressed = actual_end - start;
891 * Skip compression for a small file range(<=blocksize) that
892 * isn't an inline extent, since it doesn't save disk space at all.
894 if (total_compressed <= blocksize &&
895 (start > 0 || end + 1 < inode->disk_i_size))
896 goto cleanup_and_bail_uncompressed;
899 * For subpage case, we require full page alignment for the sector
901 * Thus we must also check against @actual_end, not just @end.
903 if (blocksize < PAGE_SIZE) {
904 if (!PAGE_ALIGNED(start) ||
905 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
906 goto cleanup_and_bail_uncompressed;
909 total_compressed = min_t(unsigned long, total_compressed,
910 BTRFS_MAX_UNCOMPRESSED);
915 * we do compression for mount -o compress and when the
916 * inode has not been flagged as nocompress. This flag can
917 * change at any time if we discover bad compression ratios.
919 if (inode_need_compress(inode, start, end)) {
921 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
923 /* just bail out to the uncompressed code */
928 if (inode->defrag_compress)
929 compress_type = inode->defrag_compress;
930 else if (inode->prop_compress)
931 compress_type = inode->prop_compress;
934 * we need to call clear_page_dirty_for_io on each
935 * page in the range. Otherwise applications with the file
936 * mmap'd can wander in and change the page contents while
937 * we are compressing them.
939 * If the compression fails for any reason, we set the pages
940 * dirty again later on.
942 * Note that the remaining part is redirtied, the start pointer
943 * has moved, the end is the original one.
946 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
950 /* Compression level is applied here and only here */
951 ret = btrfs_compress_pages(
952 compress_type | (fs_info->compress_level << 4),
960 unsigned long offset = offset_in_page(total_compressed);
961 struct page *page = pages[nr_pages - 1];
963 /* zero the tail end of the last page, we might be
964 * sending it down to disk
967 memzero_page(page, offset, PAGE_SIZE - offset);
973 * Check cow_file_range() for why we don't even try to create inline
974 * extent for subpage case.
976 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
977 /* lets try to make an inline extent */
978 if (ret || total_in < actual_end) {
979 /* we didn't compress the entire range, try
980 * to make an uncompressed inline extent.
982 ret = cow_file_range_inline(inode, actual_end,
983 0, BTRFS_COMPRESS_NONE,
986 /* try making a compressed inline extent */
987 ret = cow_file_range_inline(inode, actual_end,
989 compress_type, pages,
993 unsigned long clear_flags = EXTENT_DELALLOC |
994 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
995 EXTENT_DO_ACCOUNTING;
998 mapping_set_error(mapping, -EIO);
1001 * inline extent creation worked or returned error,
1002 * we don't need to create any more async work items.
1003 * Unlock and free up our temp pages.
1005 * We use DO_ACCOUNTING here because we need the
1006 * delalloc_release_metadata to be done _after_ we drop
1007 * our outstanding extent for clearing delalloc for this
1010 extent_clear_unlock_delalloc(inode, start, end,
1014 PAGE_START_WRITEBACK |
1015 PAGE_END_WRITEBACK);
1018 * Ensure we only free the compressed pages if we have
1019 * them allocated, as we can still reach here with
1020 * inode_need_compress() == false.
1023 for (i = 0; i < nr_pages; i++) {
1024 WARN_ON(pages[i]->mapping);
1033 if (will_compress) {
1035 * we aren't doing an inline extent round the compressed size
1036 * up to a block size boundary so the allocator does sane
1039 total_compressed = ALIGN(total_compressed, blocksize);
1042 * one last check to make sure the compression is really a
1043 * win, compare the page count read with the blocks on disk,
1044 * compression must free at least one sector size
1046 total_in = round_up(total_in, fs_info->sectorsize);
1047 if (total_compressed + blocksize <= total_in) {
1048 compressed_extents++;
1051 * The async work queues will take care of doing actual
1052 * allocation on disk for these compressed pages, and
1053 * will submit them to the elevator.
1055 add_async_extent(async_chunk, start, total_in,
1056 total_compressed, pages, nr_pages,
1059 if (start + total_in < end) {
1065 return compressed_extents;
1070 * the compression code ran but failed to make things smaller,
1071 * free any pages it allocated and our page pointer array
1073 for (i = 0; i < nr_pages; i++) {
1074 WARN_ON(pages[i]->mapping);
1079 total_compressed = 0;
1082 /* flag the file so we don't compress in the future */
1083 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
1084 !(inode->prop_compress)) {
1085 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1088 cleanup_and_bail_uncompressed:
1090 * No compression, but we still need to write the pages in the file
1091 * we've been given so far. redirty the locked page if it corresponds
1092 * to our extent and set things up for the async work queue to run
1093 * cow_file_range to do the normal delalloc dance.
1095 if (async_chunk->locked_page &&
1096 (page_offset(async_chunk->locked_page) >= start &&
1097 page_offset(async_chunk->locked_page)) <= end) {
1098 __set_page_dirty_nobuffers(async_chunk->locked_page);
1099 /* unlocked later on in the async handlers */
1103 extent_range_redirty_for_io(&inode->vfs_inode, start, end);
1104 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1105 BTRFS_COMPRESS_NONE);
1106 compressed_extents++;
1108 return compressed_extents;
1111 static void free_async_extent_pages(struct async_extent *async_extent)
1115 if (!async_extent->pages)
1118 for (i = 0; i < async_extent->nr_pages; i++) {
1119 WARN_ON(async_extent->pages[i]->mapping);
1120 put_page(async_extent->pages[i]);
1122 kfree(async_extent->pages);
1123 async_extent->nr_pages = 0;
1124 async_extent->pages = NULL;
1127 static int submit_uncompressed_range(struct btrfs_inode *inode,
1128 struct async_extent *async_extent,
1129 struct page *locked_page)
1131 u64 start = async_extent->start;
1132 u64 end = async_extent->start + async_extent->ram_size - 1;
1133 unsigned long nr_written = 0;
1134 int page_started = 0;
1136 struct writeback_control wbc = {
1137 .sync_mode = WB_SYNC_ALL,
1138 .range_start = start,
1140 .no_cgroup_owner = 1,
1144 * Call cow_file_range() to run the delalloc range directly, since we
1145 * won't go to NOCOW or async path again.
1147 * Also we call cow_file_range() with @unlock_page == 0, so that we
1148 * can directly submit them without interruption.
1150 ret = cow_file_range(inode, locked_page, start, end, &page_started,
1151 &nr_written, 0, NULL);
1152 /* Inline extent inserted, page gets unlocked and everything is done */
1157 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1159 const u64 page_start = page_offset(locked_page);
1160 const u64 page_end = page_start + PAGE_SIZE - 1;
1162 set_page_writeback(locked_page);
1163 end_page_writeback(locked_page);
1164 end_extent_writepage(locked_page, ret, page_start, page_end);
1165 unlock_page(locked_page);
1170 /* All pages will be unlocked, including @locked_page */
1171 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1172 ret = extent_write_locked_range(&inode->vfs_inode, start, end, &wbc);
1173 wbc_detach_inode(&wbc);
1177 static int submit_one_async_extent(struct btrfs_inode *inode,
1178 struct async_chunk *async_chunk,
1179 struct async_extent *async_extent,
1182 struct extent_io_tree *io_tree = &inode->io_tree;
1183 struct btrfs_root *root = inode->root;
1184 struct btrfs_fs_info *fs_info = root->fs_info;
1185 struct btrfs_ordered_extent *ordered;
1186 struct btrfs_key ins;
1187 struct page *locked_page = NULL;
1188 struct extent_map *em;
1190 u64 start = async_extent->start;
1191 u64 end = async_extent->start + async_extent->ram_size - 1;
1193 if (async_chunk->blkcg_css)
1194 kthread_associate_blkcg(async_chunk->blkcg_css);
1197 * If async_chunk->locked_page is in the async_extent range, we need to
1200 if (async_chunk->locked_page) {
1201 u64 locked_page_start = page_offset(async_chunk->locked_page);
1202 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1204 if (!(start >= locked_page_end || end <= locked_page_start))
1205 locked_page = async_chunk->locked_page;
1207 lock_extent(io_tree, start, end, NULL);
1209 /* We have fall back to uncompressed write */
1210 if (!async_extent->pages) {
1211 ret = submit_uncompressed_range(inode, async_extent, locked_page);
1215 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1216 async_extent->compressed_size,
1217 async_extent->compressed_size,
1218 0, *alloc_hint, &ins, 1, 1);
1221 * Here we used to try again by going back to non-compressed
1222 * path for ENOSPC. But we can't reserve space even for
1223 * compressed size, how could it work for uncompressed size
1224 * which requires larger size? So here we directly go error
1230 /* Here we're doing allocation and writeback of the compressed pages */
1231 em = create_io_em(inode, start,
1232 async_extent->ram_size, /* len */
1233 start, /* orig_start */
1234 ins.objectid, /* block_start */
1235 ins.offset, /* block_len */
1236 ins.offset, /* orig_block_len */
1237 async_extent->ram_size, /* ram_bytes */
1238 async_extent->compress_type,
1239 BTRFS_ORDERED_COMPRESSED);
1242 goto out_free_reserve;
1244 free_extent_map(em);
1246 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1247 async_extent->ram_size, /* num_bytes */
1248 async_extent->ram_size, /* ram_bytes */
1249 ins.objectid, /* disk_bytenr */
1250 ins.offset, /* disk_num_bytes */
1252 1 << BTRFS_ORDERED_COMPRESSED,
1253 async_extent->compress_type);
1254 if (IS_ERR(ordered)) {
1255 btrfs_drop_extent_map_range(inode, start, end, false);
1256 ret = PTR_ERR(ordered);
1257 goto out_free_reserve;
1259 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1261 /* Clear dirty, set writeback and unlock the pages. */
1262 extent_clear_unlock_delalloc(inode, start, end,
1263 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1264 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1265 btrfs_submit_compressed_write(ordered,
1266 async_extent->pages, /* compressed_pages */
1267 async_extent->nr_pages,
1268 async_chunk->write_flags, true);
1269 *alloc_hint = ins.objectid + ins.offset;
1271 if (async_chunk->blkcg_css)
1272 kthread_associate_blkcg(NULL);
1273 kfree(async_extent);
1277 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1278 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1280 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1281 extent_clear_unlock_delalloc(inode, start, end,
1282 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1283 EXTENT_DELALLOC_NEW |
1284 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1285 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1286 PAGE_END_WRITEBACK);
1287 free_async_extent_pages(async_extent);
1292 * Phase two of compressed writeback. This is the ordered portion of the code,
1293 * which only gets called in the order the work was queued. We walk all the
1294 * async extents created by compress_file_range and send them down to the disk.
1296 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1298 struct btrfs_inode *inode = async_chunk->inode;
1299 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1300 struct async_extent *async_extent;
1304 while (!list_empty(&async_chunk->extents)) {
1308 async_extent = list_entry(async_chunk->extents.next,
1309 struct async_extent, list);
1310 list_del(&async_extent->list);
1311 extent_start = async_extent->start;
1312 ram_size = async_extent->ram_size;
1314 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1316 btrfs_debug(fs_info,
1317 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1318 inode->root->root_key.objectid,
1319 btrfs_ino(inode), extent_start, ram_size, ret);
1323 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1326 struct extent_map_tree *em_tree = &inode->extent_tree;
1327 struct extent_map *em;
1330 read_lock(&em_tree->lock);
1331 em = search_extent_mapping(em_tree, start, num_bytes);
1334 * if block start isn't an actual block number then find the
1335 * first block in this inode and use that as a hint. If that
1336 * block is also bogus then just don't worry about it.
1338 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1339 free_extent_map(em);
1340 em = search_extent_mapping(em_tree, 0, 0);
1341 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1342 alloc_hint = em->block_start;
1344 free_extent_map(em);
1346 alloc_hint = em->block_start;
1347 free_extent_map(em);
1350 read_unlock(&em_tree->lock);
1356 * when extent_io.c finds a delayed allocation range in the file,
1357 * the call backs end up in this code. The basic idea is to
1358 * allocate extents on disk for the range, and create ordered data structs
1359 * in ram to track those extents.
1361 * locked_page is the page that writepage had locked already. We use
1362 * it to make sure we don't do extra locks or unlocks.
1364 * *page_started is set to one if we unlock locked_page and do everything
1365 * required to start IO on it. It may be clean and already done with
1366 * IO when we return.
1368 * When unlock == 1, we unlock the pages in successfully allocated regions.
1369 * When unlock == 0, we leave them locked for writing them out.
1371 * However, we unlock all the pages except @locked_page in case of failure.
1373 * In summary, page locking state will be as follow:
1375 * - page_started == 1 (return value)
1376 * - All the pages are unlocked. IO is started.
1377 * - Note that this can happen only on success
1379 * - All the pages except @locked_page are unlocked in any case
1381 * - On success, all the pages are locked for writing out them
1382 * - On failure, all the pages except @locked_page are unlocked
1384 * When a failure happens in the second or later iteration of the
1385 * while-loop, the ordered extents created in previous iterations are kept
1386 * intact. So, the caller must clean them up by calling
1387 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1390 static noinline int cow_file_range(struct btrfs_inode *inode,
1391 struct page *locked_page,
1392 u64 start, u64 end, int *page_started,
1393 unsigned long *nr_written, int unlock,
1396 struct btrfs_root *root = inode->root;
1397 struct btrfs_fs_info *fs_info = root->fs_info;
1399 u64 orig_start = start;
1401 unsigned long ram_size;
1402 u64 cur_alloc_size = 0;
1404 u64 blocksize = fs_info->sectorsize;
1405 struct btrfs_key ins;
1406 struct extent_map *em;
1407 unsigned clear_bits;
1408 unsigned long page_ops;
1409 bool extent_reserved = false;
1412 if (btrfs_is_free_space_inode(inode)) {
1417 num_bytes = ALIGN(end - start + 1, blocksize);
1418 num_bytes = max(blocksize, num_bytes);
1419 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1421 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1424 * Due to the page size limit, for subpage we can only trigger the
1425 * writeback for the dirty sectors of page, that means data writeback
1426 * is doing more writeback than what we want.
1428 * This is especially unexpected for some call sites like fallocate,
1429 * where we only increase i_size after everything is done.
1430 * This means we can trigger inline extent even if we didn't want to.
1431 * So here we skip inline extent creation completely.
1433 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1434 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1437 /* lets try to make an inline extent */
1438 ret = cow_file_range_inline(inode, actual_end, 0,
1439 BTRFS_COMPRESS_NONE, NULL, false);
1442 * We use DO_ACCOUNTING here because we need the
1443 * delalloc_release_metadata to be run _after_ we drop
1444 * our outstanding extent for clearing delalloc for this
1447 extent_clear_unlock_delalloc(inode, start, end,
1449 EXTENT_LOCKED | EXTENT_DELALLOC |
1450 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1451 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1452 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1453 *nr_written = *nr_written +
1454 (end - start + PAGE_SIZE) / PAGE_SIZE;
1457 * locked_page is locked by the caller of
1458 * writepage_delalloc(), not locked by
1459 * __process_pages_contig().
1461 * We can't let __process_pages_contig() to unlock it,
1462 * as it doesn't have any subpage::writers recorded.
1464 * Here we manually unlock the page, since the caller
1465 * can't use page_started to determine if it's an
1466 * inline extent or a compressed extent.
1468 unlock_page(locked_page);
1470 } else if (ret < 0) {
1475 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1478 * Relocation relies on the relocated extents to have exactly the same
1479 * size as the original extents. Normally writeback for relocation data
1480 * extents follows a NOCOW path because relocation preallocates the
1481 * extents. However, due to an operation such as scrub turning a block
1482 * group to RO mode, it may fallback to COW mode, so we must make sure
1483 * an extent allocated during COW has exactly the requested size and can
1484 * not be split into smaller extents, otherwise relocation breaks and
1485 * fails during the stage where it updates the bytenr of file extent
1488 if (btrfs_is_data_reloc_root(root))
1489 min_alloc_size = num_bytes;
1491 min_alloc_size = fs_info->sectorsize;
1493 while (num_bytes > 0) {
1494 struct btrfs_ordered_extent *ordered;
1496 cur_alloc_size = num_bytes;
1497 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1498 min_alloc_size, 0, alloc_hint,
1502 cur_alloc_size = ins.offset;
1503 extent_reserved = true;
1505 ram_size = ins.offset;
1506 em = create_io_em(inode, start, ins.offset, /* len */
1507 start, /* orig_start */
1508 ins.objectid, /* block_start */
1509 ins.offset, /* block_len */
1510 ins.offset, /* orig_block_len */
1511 ram_size, /* ram_bytes */
1512 BTRFS_COMPRESS_NONE, /* compress_type */
1513 BTRFS_ORDERED_REGULAR /* type */);
1518 free_extent_map(em);
1520 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1521 ram_size, ins.objectid, cur_alloc_size,
1522 0, 1 << BTRFS_ORDERED_REGULAR,
1523 BTRFS_COMPRESS_NONE);
1524 if (IS_ERR(ordered)) {
1525 ret = PTR_ERR(ordered);
1526 goto out_drop_extent_cache;
1529 if (btrfs_is_data_reloc_root(root)) {
1530 ret = btrfs_reloc_clone_csums(ordered);
1533 * Only drop cache here, and process as normal.
1535 * We must not allow extent_clear_unlock_delalloc()
1536 * at out_unlock label to free meta of this ordered
1537 * extent, as its meta should be freed by
1538 * btrfs_finish_ordered_io().
1540 * So we must continue until @start is increased to
1541 * skip current ordered extent.
1544 btrfs_drop_extent_map_range(inode, start,
1545 start + ram_size - 1,
1548 btrfs_put_ordered_extent(ordered);
1550 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1553 * We're not doing compressed IO, don't unlock the first page
1554 * (which the caller expects to stay locked), don't clear any
1555 * dirty bits and don't set any writeback bits
1557 * Do set the Ordered (Private2) bit so we know this page was
1558 * properly setup for writepage.
1560 page_ops = unlock ? PAGE_UNLOCK : 0;
1561 page_ops |= PAGE_SET_ORDERED;
1563 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1565 EXTENT_LOCKED | EXTENT_DELALLOC,
1567 if (num_bytes < cur_alloc_size)
1570 num_bytes -= cur_alloc_size;
1571 alloc_hint = ins.objectid + ins.offset;
1572 start += cur_alloc_size;
1573 extent_reserved = false;
1576 * btrfs_reloc_clone_csums() error, since start is increased
1577 * extent_clear_unlock_delalloc() at out_unlock label won't
1578 * free metadata of current ordered extent, we're OK to exit.
1586 out_drop_extent_cache:
1587 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1589 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1590 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1593 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1594 * caller to write out the successfully allocated region and retry.
1596 if (done_offset && ret == -EAGAIN) {
1597 if (orig_start < start)
1598 *done_offset = start - 1;
1600 *done_offset = start;
1602 } else if (ret == -EAGAIN) {
1603 /* Convert to -ENOSPC since the caller cannot retry. */
1608 * Now, we have three regions to clean up:
1610 * |-------(1)----|---(2)---|-------------(3)----------|
1611 * `- orig_start `- start `- start + cur_alloc_size `- end
1613 * We process each region below.
1616 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1617 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1618 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1621 * For the range (1). We have already instantiated the ordered extents
1622 * for this region. They are cleaned up by
1623 * btrfs_cleanup_ordered_extents() in e.g,
1624 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1625 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1626 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1629 * However, in case of unlock == 0, we still need to unlock the pages
1630 * (except @locked_page) to ensure all the pages are unlocked.
1632 if (!unlock && orig_start < start) {
1634 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1635 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1636 locked_page, 0, page_ops);
1640 * For the range (2). If we reserved an extent for our delalloc range
1641 * (or a subrange) and failed to create the respective ordered extent,
1642 * then it means that when we reserved the extent we decremented the
1643 * extent's size from the data space_info's bytes_may_use counter and
1644 * incremented the space_info's bytes_reserved counter by the same
1645 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1646 * to decrement again the data space_info's bytes_may_use counter,
1647 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1649 if (extent_reserved) {
1650 extent_clear_unlock_delalloc(inode, start,
1651 start + cur_alloc_size - 1,
1655 start += cur_alloc_size;
1659 * For the range (3). We never touched the region. In addition to the
1660 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1661 * space_info's bytes_may_use counter, reserved in
1662 * btrfs_check_data_free_space().
1665 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1666 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1667 clear_bits, page_ops);
1673 * work queue call back to started compression on a file and pages
1675 static noinline void async_cow_start(struct btrfs_work *work)
1677 struct async_chunk *async_chunk;
1678 int compressed_extents;
1680 async_chunk = container_of(work, struct async_chunk, work);
1682 compressed_extents = compress_file_range(async_chunk);
1683 if (compressed_extents == 0) {
1684 btrfs_add_delayed_iput(async_chunk->inode);
1685 async_chunk->inode = NULL;
1690 * work queue call back to submit previously compressed pages
1692 static noinline void async_cow_submit(struct btrfs_work *work)
1694 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1696 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1697 unsigned long nr_pages;
1699 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1703 * ->inode could be NULL if async_chunk_start has failed to compress,
1704 * in which case we don't have anything to submit, yet we need to
1705 * always adjust ->async_delalloc_pages as its paired with the init
1706 * happening in run_delalloc_compressed
1708 if (async_chunk->inode)
1709 submit_compressed_extents(async_chunk);
1711 /* atomic_sub_return implies a barrier */
1712 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1714 cond_wake_up_nomb(&fs_info->async_submit_wait);
1717 static noinline void async_cow_free(struct btrfs_work *work)
1719 struct async_chunk *async_chunk;
1720 struct async_cow *async_cow;
1722 async_chunk = container_of(work, struct async_chunk, work);
1723 if (async_chunk->inode)
1724 btrfs_add_delayed_iput(async_chunk->inode);
1725 if (async_chunk->blkcg_css)
1726 css_put(async_chunk->blkcg_css);
1728 async_cow = async_chunk->async_cow;
1729 if (atomic_dec_and_test(&async_cow->num_chunks))
1733 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1734 struct writeback_control *wbc,
1735 struct page *locked_page,
1736 u64 start, u64 end, int *page_started,
1737 unsigned long *nr_written)
1739 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1740 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1741 struct async_cow *ctx;
1742 struct async_chunk *async_chunk;
1743 unsigned long nr_pages;
1744 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1747 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1749 nofs_flag = memalloc_nofs_save();
1750 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1751 memalloc_nofs_restore(nofs_flag);
1755 unlock_extent(&inode->io_tree, start, end, NULL);
1756 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1758 async_chunk = ctx->chunks;
1759 atomic_set(&ctx->num_chunks, num_chunks);
1761 for (i = 0; i < num_chunks; i++) {
1762 u64 cur_end = min(end, start + SZ_512K - 1);
1765 * igrab is called higher up in the call chain, take only the
1766 * lightweight reference for the callback lifetime
1768 ihold(&inode->vfs_inode);
1769 async_chunk[i].async_cow = ctx;
1770 async_chunk[i].inode = inode;
1771 async_chunk[i].start = start;
1772 async_chunk[i].end = cur_end;
1773 async_chunk[i].write_flags = write_flags;
1774 INIT_LIST_HEAD(&async_chunk[i].extents);
1777 * The locked_page comes all the way from writepage and its
1778 * the original page we were actually given. As we spread
1779 * this large delalloc region across multiple async_chunk
1780 * structs, only the first struct needs a pointer to locked_page
1782 * This way we don't need racey decisions about who is supposed
1787 * Depending on the compressibility, the pages might or
1788 * might not go through async. We want all of them to
1789 * be accounted against wbc once. Let's do it here
1790 * before the paths diverge. wbc accounting is used
1791 * only for foreign writeback detection and doesn't
1792 * need full accuracy. Just account the whole thing
1793 * against the first page.
1795 wbc_account_cgroup_owner(wbc, locked_page,
1797 async_chunk[i].locked_page = locked_page;
1800 async_chunk[i].locked_page = NULL;
1803 if (blkcg_css != blkcg_root_css) {
1805 async_chunk[i].blkcg_css = blkcg_css;
1806 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1808 async_chunk[i].blkcg_css = NULL;
1811 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1812 async_cow_submit, async_cow_free);
1814 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1815 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1817 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1819 *nr_written += nr_pages;
1820 start = cur_end + 1;
1826 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1827 struct page *locked_page, u64 start,
1828 u64 end, int *page_started,
1829 unsigned long *nr_written,
1830 struct writeback_control *wbc)
1832 u64 done_offset = end;
1834 bool locked_page_done = false;
1836 while (start <= end) {
1837 ret = cow_file_range(inode, locked_page, start, end, page_started,
1838 nr_written, 0, &done_offset);
1839 if (ret && ret != -EAGAIN)
1842 if (*page_started) {
1850 if (done_offset == start) {
1851 wait_on_bit_io(&inode->root->fs_info->flags,
1852 BTRFS_FS_NEED_ZONE_FINISH,
1853 TASK_UNINTERRUPTIBLE);
1857 if (!locked_page_done) {
1858 __set_page_dirty_nobuffers(locked_page);
1859 account_page_redirty(locked_page);
1861 locked_page_done = true;
1862 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1864 start = done_offset + 1;
1872 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1873 u64 bytenr, u64 num_bytes, bool nowait)
1875 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1876 struct btrfs_ordered_sum *sums;
1880 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1882 if (ret == 0 && list_empty(&list))
1885 while (!list_empty(&list)) {
1886 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1887 list_del(&sums->list);
1895 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1896 const u64 start, const u64 end,
1897 int *page_started, unsigned long *nr_written)
1899 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1900 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1901 const u64 range_bytes = end + 1 - start;
1902 struct extent_io_tree *io_tree = &inode->io_tree;
1903 u64 range_start = start;
1907 * If EXTENT_NORESERVE is set it means that when the buffered write was
1908 * made we had not enough available data space and therefore we did not
1909 * reserve data space for it, since we though we could do NOCOW for the
1910 * respective file range (either there is prealloc extent or the inode
1911 * has the NOCOW bit set).
1913 * However when we need to fallback to COW mode (because for example the
1914 * block group for the corresponding extent was turned to RO mode by a
1915 * scrub or relocation) we need to do the following:
1917 * 1) We increment the bytes_may_use counter of the data space info.
1918 * If COW succeeds, it allocates a new data extent and after doing
1919 * that it decrements the space info's bytes_may_use counter and
1920 * increments its bytes_reserved counter by the same amount (we do
1921 * this at btrfs_add_reserved_bytes()). So we need to increment the
1922 * bytes_may_use counter to compensate (when space is reserved at
1923 * buffered write time, the bytes_may_use counter is incremented);
1925 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1926 * that if the COW path fails for any reason, it decrements (through
1927 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1928 * data space info, which we incremented in the step above.
1930 * If we need to fallback to cow and the inode corresponds to a free
1931 * space cache inode or an inode of the data relocation tree, we must
1932 * also increment bytes_may_use of the data space_info for the same
1933 * reason. Space caches and relocated data extents always get a prealloc
1934 * extent for them, however scrub or balance may have set the block
1935 * group that contains that extent to RO mode and therefore force COW
1936 * when starting writeback.
1938 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1939 EXTENT_NORESERVE, 0, NULL);
1940 if (count > 0 || is_space_ino || is_reloc_ino) {
1942 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1943 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1945 if (is_space_ino || is_reloc_ino)
1946 bytes = range_bytes;
1948 spin_lock(&sinfo->lock);
1949 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1950 spin_unlock(&sinfo->lock);
1953 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1957 return cow_file_range(inode, locked_page, start, end, page_started,
1958 nr_written, 1, NULL);
1961 struct can_nocow_file_extent_args {
1964 /* Start file offset of the range we want to NOCOW. */
1966 /* End file offset (inclusive) of the range we want to NOCOW. */
1968 bool writeback_path;
1971 * Free the path passed to can_nocow_file_extent() once it's not needed
1976 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1981 /* Number of bytes that can be written to in NOCOW mode. */
1986 * Check if we can NOCOW the file extent that the path points to.
1987 * This function may return with the path released, so the caller should check
1988 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1990 * Returns: < 0 on error
1991 * 0 if we can not NOCOW
1994 static int can_nocow_file_extent(struct btrfs_path *path,
1995 struct btrfs_key *key,
1996 struct btrfs_inode *inode,
1997 struct can_nocow_file_extent_args *args)
1999 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
2000 struct extent_buffer *leaf = path->nodes[0];
2001 struct btrfs_root *root = inode->root;
2002 struct btrfs_file_extent_item *fi;
2007 bool nowait = path->nowait;
2009 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2010 extent_type = btrfs_file_extent_type(leaf, fi);
2012 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
2015 /* Can't access these fields unless we know it's not an inline extent. */
2016 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2017 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
2018 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
2020 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
2021 extent_type == BTRFS_FILE_EXTENT_REG)
2025 * If the extent was created before the generation where the last snapshot
2026 * for its subvolume was created, then this implies the extent is shared,
2027 * hence we must COW.
2029 if (!args->strict &&
2030 btrfs_file_extent_generation(leaf, fi) <=
2031 btrfs_root_last_snapshot(&root->root_item))
2034 /* An explicit hole, must COW. */
2035 if (args->disk_bytenr == 0)
2038 /* Compressed/encrypted/encoded extents must be COWed. */
2039 if (btrfs_file_extent_compression(leaf, fi) ||
2040 btrfs_file_extent_encryption(leaf, fi) ||
2041 btrfs_file_extent_other_encoding(leaf, fi))
2044 extent_end = btrfs_file_extent_end(path);
2047 * The following checks can be expensive, as they need to take other
2048 * locks and do btree or rbtree searches, so release the path to avoid
2049 * blocking other tasks for too long.
2051 btrfs_release_path(path);
2053 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
2054 key->offset - args->extent_offset,
2055 args->disk_bytenr, args->strict, path);
2056 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2060 if (args->free_path) {
2062 * We don't need the path anymore, plus through the
2063 * csum_exist_in_range() call below we will end up allocating
2064 * another path. So free the path to avoid unnecessary extra
2067 btrfs_free_path(path);
2071 /* If there are pending snapshots for this root, we must COW. */
2072 if (args->writeback_path && !is_freespace_inode &&
2073 atomic_read(&root->snapshot_force_cow))
2076 args->disk_bytenr += args->extent_offset;
2077 args->disk_bytenr += args->start - key->offset;
2078 args->num_bytes = min(args->end + 1, extent_end) - args->start;
2081 * Force COW if csums exist in the range. This ensures that csums for a
2082 * given extent are either valid or do not exist.
2084 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
2086 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2092 if (args->free_path && path)
2093 btrfs_free_path(path);
2095 return ret < 0 ? ret : can_nocow;
2099 * when nowcow writeback call back. This checks for snapshots or COW copies
2100 * of the extents that exist in the file, and COWs the file as required.
2102 * If no cow copies or snapshots exist, we write directly to the existing
2105 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2106 struct page *locked_page,
2107 const u64 start, const u64 end,
2109 unsigned long *nr_written)
2111 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2112 struct btrfs_root *root = inode->root;
2113 struct btrfs_path *path;
2114 u64 cow_start = (u64)-1;
2115 u64 cur_offset = start;
2117 bool check_prev = true;
2118 u64 ino = btrfs_ino(inode);
2119 struct btrfs_block_group *bg;
2121 struct can_nocow_file_extent_args nocow_args = { 0 };
2123 path = btrfs_alloc_path();
2125 extent_clear_unlock_delalloc(inode, start, end, locked_page,
2126 EXTENT_LOCKED | EXTENT_DELALLOC |
2127 EXTENT_DO_ACCOUNTING |
2128 EXTENT_DEFRAG, PAGE_UNLOCK |
2129 PAGE_START_WRITEBACK |
2130 PAGE_END_WRITEBACK);
2134 nocow_args.end = end;
2135 nocow_args.writeback_path = true;
2138 struct btrfs_ordered_extent *ordered;
2139 struct btrfs_key found_key;
2140 struct btrfs_file_extent_item *fi;
2141 struct extent_buffer *leaf;
2150 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2156 * If there is no extent for our range when doing the initial
2157 * search, then go back to the previous slot as it will be the
2158 * one containing the search offset
2160 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2161 leaf = path->nodes[0];
2162 btrfs_item_key_to_cpu(leaf, &found_key,
2163 path->slots[0] - 1);
2164 if (found_key.objectid == ino &&
2165 found_key.type == BTRFS_EXTENT_DATA_KEY)
2170 /* Go to next leaf if we have exhausted the current one */
2171 leaf = path->nodes[0];
2172 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2173 ret = btrfs_next_leaf(root, path);
2175 if (cow_start != (u64)-1)
2176 cur_offset = cow_start;
2181 leaf = path->nodes[0];
2184 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2186 /* Didn't find anything for our INO */
2187 if (found_key.objectid > ino)
2190 * Keep searching until we find an EXTENT_ITEM or there are no
2191 * more extents for this inode
2193 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2194 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2199 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2200 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2201 found_key.offset > end)
2205 * If the found extent starts after requested offset, then
2206 * adjust extent_end to be right before this extent begins
2208 if (found_key.offset > cur_offset) {
2209 extent_end = found_key.offset;
2215 * Found extent which begins before our range and potentially
2218 fi = btrfs_item_ptr(leaf, path->slots[0],
2219 struct btrfs_file_extent_item);
2220 extent_type = btrfs_file_extent_type(leaf, fi);
2221 /* If this is triggered then we have a memory corruption. */
2222 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2223 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2227 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2228 extent_end = btrfs_file_extent_end(path);
2231 * If the extent we got ends before our current offset, skip to
2234 if (extent_end <= cur_offset) {
2239 nocow_args.start = cur_offset;
2240 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2242 if (cow_start != (u64)-1)
2243 cur_offset = cow_start;
2245 } else if (ret == 0) {
2250 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2255 * If nocow is false then record the beginning of the range
2256 * that needs to be COWed
2259 if (cow_start == (u64)-1)
2260 cow_start = cur_offset;
2261 cur_offset = extent_end;
2262 if (cur_offset > end)
2264 if (!path->nodes[0])
2271 * COW range from cow_start to found_key.offset - 1. As the key
2272 * will contain the beginning of the first extent that can be
2273 * NOCOW, following one which needs to be COW'ed
2275 if (cow_start != (u64)-1) {
2276 ret = fallback_to_cow(inode, locked_page,
2277 cow_start, found_key.offset - 1,
2278 page_started, nr_written);
2281 cow_start = (u64)-1;
2284 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2285 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2287 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2288 struct extent_map *em;
2290 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2292 nocow_args.disk_bytenr, /* block_start */
2293 nocow_args.num_bytes, /* block_len */
2294 nocow_args.disk_num_bytes, /* orig_block_len */
2295 ram_bytes, BTRFS_COMPRESS_NONE,
2296 BTRFS_ORDERED_PREALLOC);
2301 free_extent_map(em);
2304 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2305 nocow_args.num_bytes, nocow_args.num_bytes,
2306 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2308 ? (1 << BTRFS_ORDERED_PREALLOC)
2309 : (1 << BTRFS_ORDERED_NOCOW),
2310 BTRFS_COMPRESS_NONE);
2311 if (IS_ERR(ordered)) {
2313 btrfs_drop_extent_map_range(inode, cur_offset,
2316 ret = PTR_ERR(ordered);
2321 btrfs_dec_nocow_writers(bg);
2325 if (btrfs_is_data_reloc_root(root))
2327 * Error handled later, as we must prevent
2328 * extent_clear_unlock_delalloc() in error handler
2329 * from freeing metadata of created ordered extent.
2331 ret = btrfs_reloc_clone_csums(ordered);
2332 btrfs_put_ordered_extent(ordered);
2334 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2335 locked_page, EXTENT_LOCKED |
2337 EXTENT_CLEAR_DATA_RESV,
2338 PAGE_UNLOCK | PAGE_SET_ORDERED);
2340 cur_offset = extent_end;
2343 * btrfs_reloc_clone_csums() error, now we're OK to call error
2344 * handler, as metadata for created ordered extent will only
2345 * be freed by btrfs_finish_ordered_io().
2349 if (cur_offset > end)
2352 btrfs_release_path(path);
2354 if (cur_offset <= end && cow_start == (u64)-1)
2355 cow_start = cur_offset;
2357 if (cow_start != (u64)-1) {
2359 ret = fallback_to_cow(inode, locked_page, cow_start, end,
2360 page_started, nr_written);
2367 btrfs_dec_nocow_writers(bg);
2369 if (ret && cur_offset < end)
2370 extent_clear_unlock_delalloc(inode, cur_offset, end,
2371 locked_page, EXTENT_LOCKED |
2372 EXTENT_DELALLOC | EXTENT_DEFRAG |
2373 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2374 PAGE_START_WRITEBACK |
2375 PAGE_END_WRITEBACK);
2376 btrfs_free_path(path);
2380 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2382 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2383 if (inode->defrag_bytes &&
2384 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2393 * Function to process delayed allocation (create CoW) for ranges which are
2394 * being touched for the first time.
2396 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2397 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2398 struct writeback_control *wbc)
2401 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2404 * The range must cover part of the @locked_page, or the returned
2405 * @page_started can confuse the caller.
2407 ASSERT(!(end <= page_offset(locked_page) ||
2408 start >= page_offset(locked_page) + PAGE_SIZE));
2410 if (should_nocow(inode, start, end)) {
2412 * Normally on a zoned device we're only doing COW writes, but
2413 * in case of relocation on a zoned filesystem we have taken
2414 * precaution, that we're only writing sequentially. It's safe
2415 * to use run_delalloc_nocow() here, like for regular
2416 * preallocated inodes.
2418 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2419 ret = run_delalloc_nocow(inode, locked_page, start, end,
2420 page_started, nr_written);
2424 if (btrfs_inode_can_compress(inode) &&
2425 inode_need_compress(inode, start, end) &&
2426 run_delalloc_compressed(inode, wbc, locked_page, start,
2427 end, page_started, nr_written))
2431 ret = run_delalloc_zoned(inode, locked_page, start, end,
2432 page_started, nr_written, wbc);
2434 ret = cow_file_range(inode, locked_page, start, end,
2435 page_started, nr_written, 1, NULL);
2440 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2445 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2446 struct extent_state *orig, u64 split)
2448 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2451 /* not delalloc, ignore it */
2452 if (!(orig->state & EXTENT_DELALLOC))
2455 size = orig->end - orig->start + 1;
2456 if (size > fs_info->max_extent_size) {
2461 * See the explanation in btrfs_merge_delalloc_extent, the same
2462 * applies here, just in reverse.
2464 new_size = orig->end - split + 1;
2465 num_extents = count_max_extents(fs_info, new_size);
2466 new_size = split - orig->start;
2467 num_extents += count_max_extents(fs_info, new_size);
2468 if (count_max_extents(fs_info, size) >= num_extents)
2472 spin_lock(&inode->lock);
2473 btrfs_mod_outstanding_extents(inode, 1);
2474 spin_unlock(&inode->lock);
2478 * Handle merged delayed allocation extents so we can keep track of new extents
2479 * that are just merged onto old extents, such as when we are doing sequential
2480 * writes, so we can properly account for the metadata space we'll need.
2482 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2483 struct extent_state *other)
2485 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2486 u64 new_size, old_size;
2489 /* not delalloc, ignore it */
2490 if (!(other->state & EXTENT_DELALLOC))
2493 if (new->start > other->start)
2494 new_size = new->end - other->start + 1;
2496 new_size = other->end - new->start + 1;
2498 /* we're not bigger than the max, unreserve the space and go */
2499 if (new_size <= fs_info->max_extent_size) {
2500 spin_lock(&inode->lock);
2501 btrfs_mod_outstanding_extents(inode, -1);
2502 spin_unlock(&inode->lock);
2507 * We have to add up either side to figure out how many extents were
2508 * accounted for before we merged into one big extent. If the number of
2509 * extents we accounted for is <= the amount we need for the new range
2510 * then we can return, otherwise drop. Think of it like this
2514 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2515 * need 2 outstanding extents, on one side we have 1 and the other side
2516 * we have 1 so they are == and we can return. But in this case
2518 * [MAX_SIZE+4k][MAX_SIZE+4k]
2520 * Each range on their own accounts for 2 extents, but merged together
2521 * they are only 3 extents worth of accounting, so we need to drop in
2524 old_size = other->end - other->start + 1;
2525 num_extents = count_max_extents(fs_info, old_size);
2526 old_size = new->end - new->start + 1;
2527 num_extents += count_max_extents(fs_info, old_size);
2528 if (count_max_extents(fs_info, new_size) >= num_extents)
2531 spin_lock(&inode->lock);
2532 btrfs_mod_outstanding_extents(inode, -1);
2533 spin_unlock(&inode->lock);
2536 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2537 struct btrfs_inode *inode)
2539 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2541 spin_lock(&root->delalloc_lock);
2542 if (list_empty(&inode->delalloc_inodes)) {
2543 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2544 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2545 root->nr_delalloc_inodes++;
2546 if (root->nr_delalloc_inodes == 1) {
2547 spin_lock(&fs_info->delalloc_root_lock);
2548 BUG_ON(!list_empty(&root->delalloc_root));
2549 list_add_tail(&root->delalloc_root,
2550 &fs_info->delalloc_roots);
2551 spin_unlock(&fs_info->delalloc_root_lock);
2554 spin_unlock(&root->delalloc_lock);
2557 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2558 struct btrfs_inode *inode)
2560 struct btrfs_fs_info *fs_info = root->fs_info;
2562 if (!list_empty(&inode->delalloc_inodes)) {
2563 list_del_init(&inode->delalloc_inodes);
2564 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2565 &inode->runtime_flags);
2566 root->nr_delalloc_inodes--;
2567 if (!root->nr_delalloc_inodes) {
2568 ASSERT(list_empty(&root->delalloc_inodes));
2569 spin_lock(&fs_info->delalloc_root_lock);
2570 BUG_ON(list_empty(&root->delalloc_root));
2571 list_del_init(&root->delalloc_root);
2572 spin_unlock(&fs_info->delalloc_root_lock);
2577 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2578 struct btrfs_inode *inode)
2580 spin_lock(&root->delalloc_lock);
2581 __btrfs_del_delalloc_inode(root, inode);
2582 spin_unlock(&root->delalloc_lock);
2586 * Properly track delayed allocation bytes in the inode and to maintain the
2587 * list of inodes that have pending delalloc work to be done.
2589 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2592 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2594 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2597 * set_bit and clear bit hooks normally require _irqsave/restore
2598 * but in this case, we are only testing for the DELALLOC
2599 * bit, which is only set or cleared with irqs on
2601 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2602 struct btrfs_root *root = inode->root;
2603 u64 len = state->end + 1 - state->start;
2604 u32 num_extents = count_max_extents(fs_info, len);
2605 bool do_list = !btrfs_is_free_space_inode(inode);
2607 spin_lock(&inode->lock);
2608 btrfs_mod_outstanding_extents(inode, num_extents);
2609 spin_unlock(&inode->lock);
2611 /* For sanity tests */
2612 if (btrfs_is_testing(fs_info))
2615 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2616 fs_info->delalloc_batch);
2617 spin_lock(&inode->lock);
2618 inode->delalloc_bytes += len;
2619 if (bits & EXTENT_DEFRAG)
2620 inode->defrag_bytes += len;
2621 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2622 &inode->runtime_flags))
2623 btrfs_add_delalloc_inodes(root, inode);
2624 spin_unlock(&inode->lock);
2627 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2628 (bits & EXTENT_DELALLOC_NEW)) {
2629 spin_lock(&inode->lock);
2630 inode->new_delalloc_bytes += state->end + 1 - state->start;
2631 spin_unlock(&inode->lock);
2636 * Once a range is no longer delalloc this function ensures that proper
2637 * accounting happens.
2639 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2640 struct extent_state *state, u32 bits)
2642 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2643 u64 len = state->end + 1 - state->start;
2644 u32 num_extents = count_max_extents(fs_info, len);
2646 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2647 spin_lock(&inode->lock);
2648 inode->defrag_bytes -= len;
2649 spin_unlock(&inode->lock);
2653 * set_bit and clear bit hooks normally require _irqsave/restore
2654 * but in this case, we are only testing for the DELALLOC
2655 * bit, which is only set or cleared with irqs on
2657 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2658 struct btrfs_root *root = inode->root;
2659 bool do_list = !btrfs_is_free_space_inode(inode);
2661 spin_lock(&inode->lock);
2662 btrfs_mod_outstanding_extents(inode, -num_extents);
2663 spin_unlock(&inode->lock);
2666 * We don't reserve metadata space for space cache inodes so we
2667 * don't need to call delalloc_release_metadata if there is an
2670 if (bits & EXTENT_CLEAR_META_RESV &&
2671 root != fs_info->tree_root)
2672 btrfs_delalloc_release_metadata(inode, len, false);
2674 /* For sanity tests. */
2675 if (btrfs_is_testing(fs_info))
2678 if (!btrfs_is_data_reloc_root(root) &&
2679 do_list && !(state->state & EXTENT_NORESERVE) &&
2680 (bits & EXTENT_CLEAR_DATA_RESV))
2681 btrfs_free_reserved_data_space_noquota(fs_info, len);
2683 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2684 fs_info->delalloc_batch);
2685 spin_lock(&inode->lock);
2686 inode->delalloc_bytes -= len;
2687 if (do_list && inode->delalloc_bytes == 0 &&
2688 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2689 &inode->runtime_flags))
2690 btrfs_del_delalloc_inode(root, inode);
2691 spin_unlock(&inode->lock);
2694 if ((state->state & EXTENT_DELALLOC_NEW) &&
2695 (bits & EXTENT_DELALLOC_NEW)) {
2696 spin_lock(&inode->lock);
2697 ASSERT(inode->new_delalloc_bytes >= len);
2698 inode->new_delalloc_bytes -= len;
2699 if (bits & EXTENT_ADD_INODE_BYTES)
2700 inode_add_bytes(&inode->vfs_inode, len);
2701 spin_unlock(&inode->lock);
2705 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2706 struct btrfs_ordered_extent *ordered)
2708 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2709 u64 len = bbio->bio.bi_iter.bi_size;
2710 struct btrfs_ordered_extent *new;
2713 /* Must always be called for the beginning of an ordered extent. */
2714 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2717 /* No need to split if the ordered extent covers the entire bio. */
2718 if (ordered->disk_num_bytes == len) {
2719 refcount_inc(&ordered->refs);
2720 bbio->ordered = ordered;
2725 * Don't split the extent_map for NOCOW extents, as we're writing into
2726 * a pre-existing one.
2728 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2729 ret = split_extent_map(bbio->inode, bbio->file_offset,
2730 ordered->num_bytes, len,
2731 ordered->disk_bytenr);
2736 new = btrfs_split_ordered_extent(ordered, len);
2738 return PTR_ERR(new);
2739 bbio->ordered = new;
2744 * given a list of ordered sums record them in the inode. This happens
2745 * at IO completion time based on sums calculated at bio submission time.
2747 static int add_pending_csums(struct btrfs_trans_handle *trans,
2748 struct list_head *list)
2750 struct btrfs_ordered_sum *sum;
2751 struct btrfs_root *csum_root = NULL;
2754 list_for_each_entry(sum, list, list) {
2755 trans->adding_csums = true;
2757 csum_root = btrfs_csum_root(trans->fs_info,
2759 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2760 trans->adding_csums = false;
2767 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2770 struct extent_state **cached_state)
2772 u64 search_start = start;
2773 const u64 end = start + len - 1;
2775 while (search_start < end) {
2776 const u64 search_len = end - search_start + 1;
2777 struct extent_map *em;
2781 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2785 if (em->block_start != EXTENT_MAP_HOLE)
2789 if (em->start < search_start)
2790 em_len -= search_start - em->start;
2791 if (em_len > search_len)
2792 em_len = search_len;
2794 ret = set_extent_bit(&inode->io_tree, search_start,
2795 search_start + em_len - 1,
2796 EXTENT_DELALLOC_NEW, cached_state);
2798 search_start = extent_map_end(em);
2799 free_extent_map(em);
2806 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2807 unsigned int extra_bits,
2808 struct extent_state **cached_state)
2810 WARN_ON(PAGE_ALIGNED(end));
2812 if (start >= i_size_read(&inode->vfs_inode) &&
2813 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2815 * There can't be any extents following eof in this case so just
2816 * set the delalloc new bit for the range directly.
2818 extra_bits |= EXTENT_DELALLOC_NEW;
2822 ret = btrfs_find_new_delalloc_bytes(inode, start,
2829 return set_extent_bit(&inode->io_tree, start, end,
2830 EXTENT_DELALLOC | extra_bits, cached_state);
2833 /* see btrfs_writepage_start_hook for details on why this is required */
2834 struct btrfs_writepage_fixup {
2836 struct btrfs_inode *inode;
2837 struct btrfs_work work;
2840 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2842 struct btrfs_writepage_fixup *fixup;
2843 struct btrfs_ordered_extent *ordered;
2844 struct extent_state *cached_state = NULL;
2845 struct extent_changeset *data_reserved = NULL;
2847 struct btrfs_inode *inode;
2851 bool free_delalloc_space = true;
2853 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2855 inode = fixup->inode;
2856 page_start = page_offset(page);
2857 page_end = page_offset(page) + PAGE_SIZE - 1;
2860 * This is similar to page_mkwrite, we need to reserve the space before
2861 * we take the page lock.
2863 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2869 * Before we queued this fixup, we took a reference on the page.
2870 * page->mapping may go NULL, but it shouldn't be moved to a different
2873 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2875 * Unfortunately this is a little tricky, either
2877 * 1) We got here and our page had already been dealt with and
2878 * we reserved our space, thus ret == 0, so we need to just
2879 * drop our space reservation and bail. This can happen the
2880 * first time we come into the fixup worker, or could happen
2881 * while waiting for the ordered extent.
2882 * 2) Our page was already dealt with, but we happened to get an
2883 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2884 * this case we obviously don't have anything to release, but
2885 * because the page was already dealt with we don't want to
2886 * mark the page with an error, so make sure we're resetting
2887 * ret to 0. This is why we have this check _before_ the ret
2888 * check, because we do not want to have a surprise ENOSPC
2889 * when the page was already properly dealt with.
2892 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2893 btrfs_delalloc_release_space(inode, data_reserved,
2894 page_start, PAGE_SIZE,
2902 * We can't mess with the page state unless it is locked, so now that
2903 * it is locked bail if we failed to make our space reservation.
2908 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2910 /* already ordered? We're done */
2911 if (PageOrdered(page))
2914 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2916 unlock_extent(&inode->io_tree, page_start, page_end,
2919 btrfs_start_ordered_extent(ordered);
2920 btrfs_put_ordered_extent(ordered);
2924 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2930 * Everything went as planned, we're now the owner of a dirty page with
2931 * delayed allocation bits set and space reserved for our COW
2934 * The page was dirty when we started, nothing should have cleaned it.
2936 BUG_ON(!PageDirty(page));
2937 free_delalloc_space = false;
2939 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2940 if (free_delalloc_space)
2941 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2943 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2947 * We hit ENOSPC or other errors. Update the mapping and page
2948 * to reflect the errors and clean the page.
2950 mapping_set_error(page->mapping, ret);
2951 end_extent_writepage(page, ret, page_start, page_end);
2952 clear_page_dirty_for_io(page);
2954 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2958 extent_changeset_free(data_reserved);
2960 * As a precaution, do a delayed iput in case it would be the last iput
2961 * that could need flushing space. Recursing back to fixup worker would
2964 btrfs_add_delayed_iput(inode);
2968 * There are a few paths in the higher layers of the kernel that directly
2969 * set the page dirty bit without asking the filesystem if it is a
2970 * good idea. This causes problems because we want to make sure COW
2971 * properly happens and the data=ordered rules are followed.
2973 * In our case any range that doesn't have the ORDERED bit set
2974 * hasn't been properly setup for IO. We kick off an async process
2975 * to fix it up. The async helper will wait for ordered extents, set
2976 * the delalloc bit and make it safe to write the page.
2978 int btrfs_writepage_cow_fixup(struct page *page)
2980 struct inode *inode = page->mapping->host;
2981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2982 struct btrfs_writepage_fixup *fixup;
2984 /* This page has ordered extent covering it already */
2985 if (PageOrdered(page))
2989 * PageChecked is set below when we create a fixup worker for this page,
2990 * don't try to create another one if we're already PageChecked()
2992 * The extent_io writepage code will redirty the page if we send back
2995 if (PageChecked(page))
2998 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
3003 * We are already holding a reference to this inode from
3004 * write_cache_pages. We need to hold it because the space reservation
3005 * takes place outside of the page lock, and we can't trust
3006 * page->mapping outside of the page lock.
3009 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3011 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3013 fixup->inode = BTRFS_I(inode);
3014 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3019 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3020 struct btrfs_inode *inode, u64 file_pos,
3021 struct btrfs_file_extent_item *stack_fi,
3022 const bool update_inode_bytes,
3023 u64 qgroup_reserved)
3025 struct btrfs_root *root = inode->root;
3026 const u64 sectorsize = root->fs_info->sectorsize;
3027 struct btrfs_path *path;
3028 struct extent_buffer *leaf;
3029 struct btrfs_key ins;
3030 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3031 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3032 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3033 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3034 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3035 struct btrfs_drop_extents_args drop_args = { 0 };
3038 path = btrfs_alloc_path();
3043 * we may be replacing one extent in the tree with another.
3044 * The new extent is pinned in the extent map, and we don't want
3045 * to drop it from the cache until it is completely in the btree.
3047 * So, tell btrfs_drop_extents to leave this extent in the cache.
3048 * the caller is expected to unpin it and allow it to be merged
3051 drop_args.path = path;
3052 drop_args.start = file_pos;
3053 drop_args.end = file_pos + num_bytes;
3054 drop_args.replace_extent = true;
3055 drop_args.extent_item_size = sizeof(*stack_fi);
3056 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3060 if (!drop_args.extent_inserted) {
3061 ins.objectid = btrfs_ino(inode);
3062 ins.offset = file_pos;
3063 ins.type = BTRFS_EXTENT_DATA_KEY;
3065 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3070 leaf = path->nodes[0];
3071 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3072 write_extent_buffer(leaf, stack_fi,
3073 btrfs_item_ptr_offset(leaf, path->slots[0]),
3074 sizeof(struct btrfs_file_extent_item));
3076 btrfs_mark_buffer_dirty(leaf);
3077 btrfs_release_path(path);
3080 * If we dropped an inline extent here, we know the range where it is
3081 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3082 * number of bytes only for that range containing the inline extent.
3083 * The remaining of the range will be processed when clearning the
3084 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3086 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3087 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3089 inline_size = drop_args.bytes_found - inline_size;
3090 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3091 drop_args.bytes_found -= inline_size;
3092 num_bytes -= sectorsize;
3095 if (update_inode_bytes)
3096 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3098 ins.objectid = disk_bytenr;
3099 ins.offset = disk_num_bytes;
3100 ins.type = BTRFS_EXTENT_ITEM_KEY;
3102 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3106 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3108 qgroup_reserved, &ins);
3110 btrfs_free_path(path);
3115 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3118 struct btrfs_block_group *cache;
3120 cache = btrfs_lookup_block_group(fs_info, start);
3123 spin_lock(&cache->lock);
3124 cache->delalloc_bytes -= len;
3125 spin_unlock(&cache->lock);
3127 btrfs_put_block_group(cache);
3130 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3131 struct btrfs_ordered_extent *oe)
3133 struct btrfs_file_extent_item stack_fi;
3134 bool update_inode_bytes;
3135 u64 num_bytes = oe->num_bytes;
3136 u64 ram_bytes = oe->ram_bytes;
3138 memset(&stack_fi, 0, sizeof(stack_fi));
3139 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3140 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3141 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3142 oe->disk_num_bytes);
3143 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3144 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3145 num_bytes = oe->truncated_len;
3146 ram_bytes = num_bytes;
3148 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3149 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3150 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3151 /* Encryption and other encoding is reserved and all 0 */
3154 * For delalloc, when completing an ordered extent we update the inode's
3155 * bytes when clearing the range in the inode's io tree, so pass false
3156 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3157 * except if the ordered extent was truncated.
3159 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3160 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3161 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3163 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3164 oe->file_offset, &stack_fi,
3165 update_inode_bytes, oe->qgroup_rsv);
3169 * As ordered data IO finishes, this gets called so we can finish
3170 * an ordered extent if the range of bytes in the file it covers are
3173 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3175 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3176 struct btrfs_root *root = inode->root;
3177 struct btrfs_fs_info *fs_info = root->fs_info;
3178 struct btrfs_trans_handle *trans = NULL;
3179 struct extent_io_tree *io_tree = &inode->io_tree;
3180 struct extent_state *cached_state = NULL;
3182 int compress_type = 0;
3184 u64 logical_len = ordered_extent->num_bytes;
3185 bool freespace_inode;
3186 bool truncated = false;
3187 bool clear_reserved_extent = true;
3188 unsigned int clear_bits = EXTENT_DEFRAG;
3190 start = ordered_extent->file_offset;
3191 end = start + ordered_extent->num_bytes - 1;
3193 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3194 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3195 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3196 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3197 clear_bits |= EXTENT_DELALLOC_NEW;
3199 freespace_inode = btrfs_is_free_space_inode(inode);
3200 if (!freespace_inode)
3201 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3203 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3208 if (btrfs_is_zoned(fs_info))
3209 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3210 ordered_extent->disk_num_bytes);
3212 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3214 logical_len = ordered_extent->truncated_len;
3215 /* Truncated the entire extent, don't bother adding */
3220 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3221 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3223 btrfs_inode_safe_disk_i_size_write(inode, 0);
3224 if (freespace_inode)
3225 trans = btrfs_join_transaction_spacecache(root);
3227 trans = btrfs_join_transaction(root);
3228 if (IS_ERR(trans)) {
3229 ret = PTR_ERR(trans);
3233 trans->block_rsv = &inode->block_rsv;
3234 ret = btrfs_update_inode_fallback(trans, root, inode);
3235 if (ret) /* -ENOMEM or corruption */
3236 btrfs_abort_transaction(trans, ret);
3240 clear_bits |= EXTENT_LOCKED;
3241 lock_extent(io_tree, start, end, &cached_state);
3243 if (freespace_inode)
3244 trans = btrfs_join_transaction_spacecache(root);
3246 trans = btrfs_join_transaction(root);
3247 if (IS_ERR(trans)) {
3248 ret = PTR_ERR(trans);
3253 trans->block_rsv = &inode->block_rsv;
3255 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3256 compress_type = ordered_extent->compress_type;
3257 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3258 BUG_ON(compress_type);
3259 ret = btrfs_mark_extent_written(trans, inode,
3260 ordered_extent->file_offset,
3261 ordered_extent->file_offset +
3263 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3264 ordered_extent->disk_num_bytes);
3266 BUG_ON(root == fs_info->tree_root);
3267 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3269 clear_reserved_extent = false;
3270 btrfs_release_delalloc_bytes(fs_info,
3271 ordered_extent->disk_bytenr,
3272 ordered_extent->disk_num_bytes);
3275 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3276 ordered_extent->num_bytes, trans->transid);
3278 btrfs_abort_transaction(trans, ret);
3282 ret = add_pending_csums(trans, &ordered_extent->list);
3284 btrfs_abort_transaction(trans, ret);
3289 * If this is a new delalloc range, clear its new delalloc flag to
3290 * update the inode's number of bytes. This needs to be done first
3291 * before updating the inode item.
3293 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3294 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3295 clear_extent_bit(&inode->io_tree, start, end,
3296 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3299 btrfs_inode_safe_disk_i_size_write(inode, 0);
3300 ret = btrfs_update_inode_fallback(trans, root, inode);
3301 if (ret) { /* -ENOMEM or corruption */
3302 btrfs_abort_transaction(trans, ret);
3307 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3311 btrfs_end_transaction(trans);
3313 if (ret || truncated) {
3314 u64 unwritten_start = start;
3317 * If we failed to finish this ordered extent for any reason we
3318 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3319 * extent, and mark the inode with the error if it wasn't
3320 * already set. Any error during writeback would have already
3321 * set the mapping error, so we need to set it if we're the ones
3322 * marking this ordered extent as failed.
3324 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3325 &ordered_extent->flags))
3326 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3329 unwritten_start += logical_len;
3330 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3332 /* Drop extent maps for the part of the extent we didn't write. */
3333 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3336 * If the ordered extent had an IOERR or something else went
3337 * wrong we need to return the space for this ordered extent
3338 * back to the allocator. We only free the extent in the
3339 * truncated case if we didn't write out the extent at all.
3341 * If we made it past insert_reserved_file_extent before we
3342 * errored out then we don't need to do this as the accounting
3343 * has already been done.
3345 if ((ret || !logical_len) &&
3346 clear_reserved_extent &&
3347 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3348 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3350 * Discard the range before returning it back to the
3353 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3354 btrfs_discard_extent(fs_info,
3355 ordered_extent->disk_bytenr,
3356 ordered_extent->disk_num_bytes,
3358 btrfs_free_reserved_extent(fs_info,
3359 ordered_extent->disk_bytenr,
3360 ordered_extent->disk_num_bytes, 1);
3365 * This needs to be done to make sure anybody waiting knows we are done
3366 * updating everything for this ordered extent.
3368 btrfs_remove_ordered_extent(inode, ordered_extent);
3371 btrfs_put_ordered_extent(ordered_extent);
3372 /* once for the tree */
3373 btrfs_put_ordered_extent(ordered_extent);
3378 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3380 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3381 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3382 btrfs_finish_ordered_zoned(ordered);
3383 return btrfs_finish_one_ordered(ordered);
3386 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3387 struct page *page, u64 start,
3388 u64 end, bool uptodate)
3390 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3392 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate);
3396 * Verify the checksum for a single sector without any extra action that depend
3397 * on the type of I/O.
3399 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3400 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3402 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3405 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3407 shash->tfm = fs_info->csum_shash;
3409 kaddr = kmap_local_page(page) + pgoff;
3410 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3411 kunmap_local(kaddr);
3413 if (memcmp(csum, csum_expected, fs_info->csum_size))
3419 * Verify the checksum of a single data sector.
3421 * @bbio: btrfs_io_bio which contains the csum
3422 * @dev: device the sector is on
3423 * @bio_offset: offset to the beginning of the bio (in bytes)
3424 * @bv: bio_vec to check
3426 * Check if the checksum on a data block is valid. When a checksum mismatch is
3427 * detected, report the error and fill the corrupted range with zero.
3429 * Return %true if the sector is ok or had no checksum to start with, else %false.
3431 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3432 u32 bio_offset, struct bio_vec *bv)
3434 struct btrfs_inode *inode = bbio->inode;
3435 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3436 u64 file_offset = bbio->file_offset + bio_offset;
3437 u64 end = file_offset + bv->bv_len - 1;
3439 u8 csum[BTRFS_CSUM_SIZE];
3441 ASSERT(bv->bv_len == fs_info->sectorsize);
3446 if (btrfs_is_data_reloc_root(inode->root) &&
3447 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3449 /* Skip the range without csum for data reloc inode */
3450 clear_extent_bits(&inode->io_tree, file_offset, end,
3455 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3457 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3463 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3466 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3472 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3474 * @inode: The inode we want to perform iput on
3476 * This function uses the generic vfs_inode::i_count to track whether we should
3477 * just decrement it (in case it's > 1) or if this is the last iput then link
3478 * the inode to the delayed iput machinery. Delayed iputs are processed at
3479 * transaction commit time/superblock commit/cleaner kthread.
3481 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3483 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3484 unsigned long flags;
3486 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3489 atomic_inc(&fs_info->nr_delayed_iputs);
3491 * Need to be irq safe here because we can be called from either an irq
3492 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3495 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3496 ASSERT(list_empty(&inode->delayed_iput));
3497 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3498 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3499 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3500 wake_up_process(fs_info->cleaner_kthread);
3503 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3504 struct btrfs_inode *inode)
3506 list_del_init(&inode->delayed_iput);
3507 spin_unlock_irq(&fs_info->delayed_iput_lock);
3508 iput(&inode->vfs_inode);
3509 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3510 wake_up(&fs_info->delayed_iputs_wait);
3511 spin_lock_irq(&fs_info->delayed_iput_lock);
3514 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3515 struct btrfs_inode *inode)
3517 if (!list_empty(&inode->delayed_iput)) {
3518 spin_lock_irq(&fs_info->delayed_iput_lock);
3519 if (!list_empty(&inode->delayed_iput))
3520 run_delayed_iput_locked(fs_info, inode);
3521 spin_unlock_irq(&fs_info->delayed_iput_lock);
3525 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3528 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3529 * calls btrfs_add_delayed_iput() and that needs to lock
3530 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3531 * prevent a deadlock.
3533 spin_lock_irq(&fs_info->delayed_iput_lock);
3534 while (!list_empty(&fs_info->delayed_iputs)) {
3535 struct btrfs_inode *inode;
3537 inode = list_first_entry(&fs_info->delayed_iputs,
3538 struct btrfs_inode, delayed_iput);
3539 run_delayed_iput_locked(fs_info, inode);
3540 if (need_resched()) {
3541 spin_unlock_irq(&fs_info->delayed_iput_lock);
3543 spin_lock_irq(&fs_info->delayed_iput_lock);
3546 spin_unlock_irq(&fs_info->delayed_iput_lock);
3550 * Wait for flushing all delayed iputs
3552 * @fs_info: the filesystem
3554 * This will wait on any delayed iputs that are currently running with KILLABLE
3555 * set. Once they are all done running we will return, unless we are killed in
3556 * which case we return EINTR. This helps in user operations like fallocate etc
3557 * that might get blocked on the iputs.
3559 * Return EINTR if we were killed, 0 if nothing's pending
3561 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3563 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3564 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3571 * This creates an orphan entry for the given inode in case something goes wrong
3572 * in the middle of an unlink.
3574 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3575 struct btrfs_inode *inode)
3579 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3580 if (ret && ret != -EEXIST) {
3581 btrfs_abort_transaction(trans, ret);
3589 * We have done the delete so we can go ahead and remove the orphan item for
3590 * this particular inode.
3592 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3593 struct btrfs_inode *inode)
3595 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3599 * this cleans up any orphans that may be left on the list from the last use
3602 int btrfs_orphan_cleanup(struct btrfs_root *root)
3604 struct btrfs_fs_info *fs_info = root->fs_info;
3605 struct btrfs_path *path;
3606 struct extent_buffer *leaf;
3607 struct btrfs_key key, found_key;
3608 struct btrfs_trans_handle *trans;
3609 struct inode *inode;
3610 u64 last_objectid = 0;
3611 int ret = 0, nr_unlink = 0;
3613 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3616 path = btrfs_alloc_path();
3621 path->reada = READA_BACK;
3623 key.objectid = BTRFS_ORPHAN_OBJECTID;
3624 key.type = BTRFS_ORPHAN_ITEM_KEY;
3625 key.offset = (u64)-1;
3628 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3633 * if ret == 0 means we found what we were searching for, which
3634 * is weird, but possible, so only screw with path if we didn't
3635 * find the key and see if we have stuff that matches
3639 if (path->slots[0] == 0)
3644 /* pull out the item */
3645 leaf = path->nodes[0];
3646 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3648 /* make sure the item matches what we want */
3649 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3651 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3654 /* release the path since we're done with it */
3655 btrfs_release_path(path);
3658 * this is where we are basically btrfs_lookup, without the
3659 * crossing root thing. we store the inode number in the
3660 * offset of the orphan item.
3663 if (found_key.offset == last_objectid) {
3665 "Error removing orphan entry, stopping orphan cleanup");
3670 last_objectid = found_key.offset;
3672 found_key.objectid = found_key.offset;
3673 found_key.type = BTRFS_INODE_ITEM_KEY;
3674 found_key.offset = 0;
3675 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3676 if (IS_ERR(inode)) {
3677 ret = PTR_ERR(inode);
3683 if (!inode && root == fs_info->tree_root) {
3684 struct btrfs_root *dead_root;
3685 int is_dead_root = 0;
3688 * This is an orphan in the tree root. Currently these
3689 * could come from 2 sources:
3690 * a) a root (snapshot/subvolume) deletion in progress
3691 * b) a free space cache inode
3692 * We need to distinguish those two, as the orphan item
3693 * for a root must not get deleted before the deletion
3694 * of the snapshot/subvolume's tree completes.
3696 * btrfs_find_orphan_roots() ran before us, which has
3697 * found all deleted roots and loaded them into
3698 * fs_info->fs_roots_radix. So here we can find if an
3699 * orphan item corresponds to a deleted root by looking
3700 * up the root from that radix tree.
3703 spin_lock(&fs_info->fs_roots_radix_lock);
3704 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3705 (unsigned long)found_key.objectid);
3706 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3708 spin_unlock(&fs_info->fs_roots_radix_lock);
3711 /* prevent this orphan from being found again */
3712 key.offset = found_key.objectid - 1;
3719 * If we have an inode with links, there are a couple of
3722 * 1. We were halfway through creating fsverity metadata for the
3723 * file. In that case, the orphan item represents incomplete
3724 * fsverity metadata which must be cleaned up with
3725 * btrfs_drop_verity_items and deleting the orphan item.
3727 * 2. Old kernels (before v3.12) used to create an
3728 * orphan item for truncate indicating that there were possibly
3729 * extent items past i_size that needed to be deleted. In v3.12,
3730 * truncate was changed to update i_size in sync with the extent
3731 * items, but the (useless) orphan item was still created. Since
3732 * v4.18, we don't create the orphan item for truncate at all.
3734 * So, this item could mean that we need to do a truncate, but
3735 * only if this filesystem was last used on a pre-v3.12 kernel
3736 * and was not cleanly unmounted. The odds of that are quite
3737 * slim, and it's a pain to do the truncate now, so just delete
3740 * It's also possible that this orphan item was supposed to be
3741 * deleted but wasn't. The inode number may have been reused,
3742 * but either way, we can delete the orphan item.
3744 if (!inode || inode->i_nlink) {
3746 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3752 trans = btrfs_start_transaction(root, 1);
3753 if (IS_ERR(trans)) {
3754 ret = PTR_ERR(trans);
3757 btrfs_debug(fs_info, "auto deleting %Lu",
3758 found_key.objectid);
3759 ret = btrfs_del_orphan_item(trans, root,
3760 found_key.objectid);
3761 btrfs_end_transaction(trans);
3769 /* this will do delete_inode and everything for us */
3772 /* release the path since we're done with it */
3773 btrfs_release_path(path);
3775 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3776 trans = btrfs_join_transaction(root);
3778 btrfs_end_transaction(trans);
3782 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3786 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3787 btrfs_free_path(path);
3792 * very simple check to peek ahead in the leaf looking for xattrs. If we
3793 * don't find any xattrs, we know there can't be any acls.
3795 * slot is the slot the inode is in, objectid is the objectid of the inode
3797 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3798 int slot, u64 objectid,
3799 int *first_xattr_slot)
3801 u32 nritems = btrfs_header_nritems(leaf);
3802 struct btrfs_key found_key;
3803 static u64 xattr_access = 0;
3804 static u64 xattr_default = 0;
3807 if (!xattr_access) {
3808 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3809 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3810 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3811 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3815 *first_xattr_slot = -1;
3816 while (slot < nritems) {
3817 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3819 /* we found a different objectid, there must not be acls */
3820 if (found_key.objectid != objectid)
3823 /* we found an xattr, assume we've got an acl */
3824 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3825 if (*first_xattr_slot == -1)
3826 *first_xattr_slot = slot;
3827 if (found_key.offset == xattr_access ||
3828 found_key.offset == xattr_default)
3833 * we found a key greater than an xattr key, there can't
3834 * be any acls later on
3836 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3843 * it goes inode, inode backrefs, xattrs, extents,
3844 * so if there are a ton of hard links to an inode there can
3845 * be a lot of backrefs. Don't waste time searching too hard,
3846 * this is just an optimization
3851 /* we hit the end of the leaf before we found an xattr or
3852 * something larger than an xattr. We have to assume the inode
3855 if (*first_xattr_slot == -1)
3856 *first_xattr_slot = slot;
3861 * read an inode from the btree into the in-memory inode
3863 static int btrfs_read_locked_inode(struct inode *inode,
3864 struct btrfs_path *in_path)
3866 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3867 struct btrfs_path *path = in_path;
3868 struct extent_buffer *leaf;
3869 struct btrfs_inode_item *inode_item;
3870 struct btrfs_root *root = BTRFS_I(inode)->root;
3871 struct btrfs_key location;
3876 bool filled = false;
3877 int first_xattr_slot;
3879 ret = btrfs_fill_inode(inode, &rdev);
3884 path = btrfs_alloc_path();
3889 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3891 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3893 if (path != in_path)
3894 btrfs_free_path(path);
3898 leaf = path->nodes[0];
3903 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3904 struct btrfs_inode_item);
3905 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3906 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3907 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3908 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3909 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3910 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3911 round_up(i_size_read(inode), fs_info->sectorsize));
3913 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3914 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3916 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3917 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3919 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3920 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3922 BTRFS_I(inode)->i_otime.tv_sec =
3923 btrfs_timespec_sec(leaf, &inode_item->otime);
3924 BTRFS_I(inode)->i_otime.tv_nsec =
3925 btrfs_timespec_nsec(leaf, &inode_item->otime);
3927 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3928 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3929 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3931 inode_set_iversion_queried(inode,
3932 btrfs_inode_sequence(leaf, inode_item));
3933 inode->i_generation = BTRFS_I(inode)->generation;
3935 rdev = btrfs_inode_rdev(leaf, inode_item);
3937 BTRFS_I(inode)->index_cnt = (u64)-1;
3938 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3939 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3943 * If we were modified in the current generation and evicted from memory
3944 * and then re-read we need to do a full sync since we don't have any
3945 * idea about which extents were modified before we were evicted from
3948 * This is required for both inode re-read from disk and delayed inode
3949 * in delayed_nodes_tree.
3951 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3952 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3953 &BTRFS_I(inode)->runtime_flags);
3956 * We don't persist the id of the transaction where an unlink operation
3957 * against the inode was last made. So here we assume the inode might
3958 * have been evicted, and therefore the exact value of last_unlink_trans
3959 * lost, and set it to last_trans to avoid metadata inconsistencies
3960 * between the inode and its parent if the inode is fsync'ed and the log
3961 * replayed. For example, in the scenario:
3964 * ln mydir/foo mydir/bar
3967 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3968 * xfs_io -c fsync mydir/foo
3970 * mount fs, triggers fsync log replay
3972 * We must make sure that when we fsync our inode foo we also log its
3973 * parent inode, otherwise after log replay the parent still has the
3974 * dentry with the "bar" name but our inode foo has a link count of 1
3975 * and doesn't have an inode ref with the name "bar" anymore.
3977 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3978 * but it guarantees correctness at the expense of occasional full
3979 * transaction commits on fsync if our inode is a directory, or if our
3980 * inode is not a directory, logging its parent unnecessarily.
3982 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3985 * Same logic as for last_unlink_trans. We don't persist the generation
3986 * of the last transaction where this inode was used for a reflink
3987 * operation, so after eviction and reloading the inode we must be
3988 * pessimistic and assume the last transaction that modified the inode.
3990 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3993 if (inode->i_nlink != 1 ||
3994 path->slots[0] >= btrfs_header_nritems(leaf))
3997 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3998 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
4001 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4002 if (location.type == BTRFS_INODE_REF_KEY) {
4003 struct btrfs_inode_ref *ref;
4005 ref = (struct btrfs_inode_ref *)ptr;
4006 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
4007 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
4008 struct btrfs_inode_extref *extref;
4010 extref = (struct btrfs_inode_extref *)ptr;
4011 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4016 * try to precache a NULL acl entry for files that don't have
4017 * any xattrs or acls
4019 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4020 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4021 if (first_xattr_slot != -1) {
4022 path->slots[0] = first_xattr_slot;
4023 ret = btrfs_load_inode_props(inode, path);
4026 "error loading props for ino %llu (root %llu): %d",
4027 btrfs_ino(BTRFS_I(inode)),
4028 root->root_key.objectid, ret);
4030 if (path != in_path)
4031 btrfs_free_path(path);
4034 cache_no_acl(inode);
4036 switch (inode->i_mode & S_IFMT) {
4038 inode->i_mapping->a_ops = &btrfs_aops;
4039 inode->i_fop = &btrfs_file_operations;
4040 inode->i_op = &btrfs_file_inode_operations;
4043 inode->i_fop = &btrfs_dir_file_operations;
4044 inode->i_op = &btrfs_dir_inode_operations;
4047 inode->i_op = &btrfs_symlink_inode_operations;
4048 inode_nohighmem(inode);
4049 inode->i_mapping->a_ops = &btrfs_aops;
4052 inode->i_op = &btrfs_special_inode_operations;
4053 init_special_inode(inode, inode->i_mode, rdev);
4057 btrfs_sync_inode_flags_to_i_flags(inode);
4062 * given a leaf and an inode, copy the inode fields into the leaf
4064 static void fill_inode_item(struct btrfs_trans_handle *trans,
4065 struct extent_buffer *leaf,
4066 struct btrfs_inode_item *item,
4067 struct inode *inode)
4069 struct btrfs_map_token token;
4072 btrfs_init_map_token(&token, leaf);
4074 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4075 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4076 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4077 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4078 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4080 btrfs_set_token_timespec_sec(&token, &item->atime,
4081 inode->i_atime.tv_sec);
4082 btrfs_set_token_timespec_nsec(&token, &item->atime,
4083 inode->i_atime.tv_nsec);
4085 btrfs_set_token_timespec_sec(&token, &item->mtime,
4086 inode->i_mtime.tv_sec);
4087 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4088 inode->i_mtime.tv_nsec);
4090 btrfs_set_token_timespec_sec(&token, &item->ctime,
4091 inode->i_ctime.tv_sec);
4092 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4093 inode->i_ctime.tv_nsec);
4095 btrfs_set_token_timespec_sec(&token, &item->otime,
4096 BTRFS_I(inode)->i_otime.tv_sec);
4097 btrfs_set_token_timespec_nsec(&token, &item->otime,
4098 BTRFS_I(inode)->i_otime.tv_nsec);
4100 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4101 btrfs_set_token_inode_generation(&token, item,
4102 BTRFS_I(inode)->generation);
4103 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4104 btrfs_set_token_inode_transid(&token, item, trans->transid);
4105 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4106 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4107 BTRFS_I(inode)->ro_flags);
4108 btrfs_set_token_inode_flags(&token, item, flags);
4109 btrfs_set_token_inode_block_group(&token, item, 0);
4113 * copy everything in the in-memory inode into the btree.
4115 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4116 struct btrfs_root *root,
4117 struct btrfs_inode *inode)
4119 struct btrfs_inode_item *inode_item;
4120 struct btrfs_path *path;
4121 struct extent_buffer *leaf;
4124 path = btrfs_alloc_path();
4128 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4135 leaf = path->nodes[0];
4136 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4137 struct btrfs_inode_item);
4139 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4140 btrfs_mark_buffer_dirty(leaf);
4141 btrfs_set_inode_last_trans(trans, inode);
4144 btrfs_free_path(path);
4149 * copy everything in the in-memory inode into the btree.
4151 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4152 struct btrfs_root *root,
4153 struct btrfs_inode *inode)
4155 struct btrfs_fs_info *fs_info = root->fs_info;
4159 * If the inode is a free space inode, we can deadlock during commit
4160 * if we put it into the delayed code.
4162 * The data relocation inode should also be directly updated
4165 if (!btrfs_is_free_space_inode(inode)
4166 && !btrfs_is_data_reloc_root(root)
4167 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4168 btrfs_update_root_times(trans, root);
4170 ret = btrfs_delayed_update_inode(trans, root, inode);
4172 btrfs_set_inode_last_trans(trans, inode);
4176 return btrfs_update_inode_item(trans, root, inode);
4179 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4180 struct btrfs_root *root, struct btrfs_inode *inode)
4184 ret = btrfs_update_inode(trans, root, inode);
4186 return btrfs_update_inode_item(trans, root, inode);
4191 * unlink helper that gets used here in inode.c and in the tree logging
4192 * recovery code. It remove a link in a directory with a given name, and
4193 * also drops the back refs in the inode to the directory
4195 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4196 struct btrfs_inode *dir,
4197 struct btrfs_inode *inode,
4198 const struct fscrypt_str *name,
4199 struct btrfs_rename_ctx *rename_ctx)
4201 struct btrfs_root *root = dir->root;
4202 struct btrfs_fs_info *fs_info = root->fs_info;
4203 struct btrfs_path *path;
4205 struct btrfs_dir_item *di;
4207 u64 ino = btrfs_ino(inode);
4208 u64 dir_ino = btrfs_ino(dir);
4210 path = btrfs_alloc_path();
4216 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4217 if (IS_ERR_OR_NULL(di)) {
4218 ret = di ? PTR_ERR(di) : -ENOENT;
4221 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4224 btrfs_release_path(path);
4227 * If we don't have dir index, we have to get it by looking up
4228 * the inode ref, since we get the inode ref, remove it directly,
4229 * it is unnecessary to do delayed deletion.
4231 * But if we have dir index, needn't search inode ref to get it.
4232 * Since the inode ref is close to the inode item, it is better
4233 * that we delay to delete it, and just do this deletion when
4234 * we update the inode item.
4236 if (inode->dir_index) {
4237 ret = btrfs_delayed_delete_inode_ref(inode);
4239 index = inode->dir_index;
4244 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4247 "failed to delete reference to %.*s, inode %llu parent %llu",
4248 name->len, name->name, ino, dir_ino);
4249 btrfs_abort_transaction(trans, ret);
4254 rename_ctx->index = index;
4256 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4258 btrfs_abort_transaction(trans, ret);
4263 * If we are in a rename context, we don't need to update anything in the
4264 * log. That will be done later during the rename by btrfs_log_new_name().
4265 * Besides that, doing it here would only cause extra unnecessary btree
4266 * operations on the log tree, increasing latency for applications.
4269 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4270 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4274 * If we have a pending delayed iput we could end up with the final iput
4275 * being run in btrfs-cleaner context. If we have enough of these built
4276 * up we can end up burning a lot of time in btrfs-cleaner without any
4277 * way to throttle the unlinks. Since we're currently holding a ref on
4278 * the inode we can run the delayed iput here without any issues as the
4279 * final iput won't be done until after we drop the ref we're currently
4282 btrfs_run_delayed_iput(fs_info, inode);
4284 btrfs_free_path(path);
4288 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4289 inode_inc_iversion(&inode->vfs_inode);
4290 inode_inc_iversion(&dir->vfs_inode);
4291 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4292 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4293 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4294 ret = btrfs_update_inode(trans, root, dir);
4299 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4300 struct btrfs_inode *dir, struct btrfs_inode *inode,
4301 const struct fscrypt_str *name)
4305 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4307 drop_nlink(&inode->vfs_inode);
4308 ret = btrfs_update_inode(trans, inode->root, inode);
4314 * helper to start transaction for unlink and rmdir.
4316 * unlink and rmdir are special in btrfs, they do not always free space, so
4317 * if we cannot make our reservations the normal way try and see if there is
4318 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4319 * allow the unlink to occur.
4321 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4323 struct btrfs_root *root = dir->root;
4325 return btrfs_start_transaction_fallback_global_rsv(root,
4326 BTRFS_UNLINK_METADATA_UNITS);
4329 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4331 struct btrfs_trans_handle *trans;
4332 struct inode *inode = d_inode(dentry);
4334 struct fscrypt_name fname;
4336 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4340 /* This needs to handle no-key deletions later on */
4342 trans = __unlink_start_trans(BTRFS_I(dir));
4343 if (IS_ERR(trans)) {
4344 ret = PTR_ERR(trans);
4348 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4351 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4356 if (inode->i_nlink == 0) {
4357 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4363 btrfs_end_transaction(trans);
4364 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4366 fscrypt_free_filename(&fname);
4370 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4371 struct btrfs_inode *dir, struct dentry *dentry)
4373 struct btrfs_root *root = dir->root;
4374 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4375 struct btrfs_path *path;
4376 struct extent_buffer *leaf;
4377 struct btrfs_dir_item *di;
4378 struct btrfs_key key;
4382 u64 dir_ino = btrfs_ino(dir);
4383 struct fscrypt_name fname;
4385 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4389 /* This needs to handle no-key deletions later on */
4391 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4392 objectid = inode->root->root_key.objectid;
4393 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4394 objectid = inode->location.objectid;
4397 fscrypt_free_filename(&fname);
4401 path = btrfs_alloc_path();
4407 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4408 &fname.disk_name, -1);
4409 if (IS_ERR_OR_NULL(di)) {
4410 ret = di ? PTR_ERR(di) : -ENOENT;
4414 leaf = path->nodes[0];
4415 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4416 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4417 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4419 btrfs_abort_transaction(trans, ret);
4422 btrfs_release_path(path);
4425 * This is a placeholder inode for a subvolume we didn't have a
4426 * reference to at the time of the snapshot creation. In the meantime
4427 * we could have renamed the real subvol link into our snapshot, so
4428 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4429 * Instead simply lookup the dir_index_item for this entry so we can
4430 * remove it. Otherwise we know we have a ref to the root and we can
4431 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4433 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4434 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4435 if (IS_ERR_OR_NULL(di)) {
4440 btrfs_abort_transaction(trans, ret);
4444 leaf = path->nodes[0];
4445 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4447 btrfs_release_path(path);
4449 ret = btrfs_del_root_ref(trans, objectid,
4450 root->root_key.objectid, dir_ino,
4451 &index, &fname.disk_name);
4453 btrfs_abort_transaction(trans, ret);
4458 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4460 btrfs_abort_transaction(trans, ret);
4464 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4465 inode_inc_iversion(&dir->vfs_inode);
4466 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4467 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4468 ret = btrfs_update_inode_fallback(trans, root, dir);
4470 btrfs_abort_transaction(trans, ret);
4472 btrfs_free_path(path);
4473 fscrypt_free_filename(&fname);
4478 * Helper to check if the subvolume references other subvolumes or if it's
4481 static noinline int may_destroy_subvol(struct btrfs_root *root)
4483 struct btrfs_fs_info *fs_info = root->fs_info;
4484 struct btrfs_path *path;
4485 struct btrfs_dir_item *di;
4486 struct btrfs_key key;
4487 struct fscrypt_str name = FSTR_INIT("default", 7);
4491 path = btrfs_alloc_path();
4495 /* Make sure this root isn't set as the default subvol */
4496 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4497 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4499 if (di && !IS_ERR(di)) {
4500 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4501 if (key.objectid == root->root_key.objectid) {
4504 "deleting default subvolume %llu is not allowed",
4508 btrfs_release_path(path);
4511 key.objectid = root->root_key.objectid;
4512 key.type = BTRFS_ROOT_REF_KEY;
4513 key.offset = (u64)-1;
4515 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4521 if (path->slots[0] > 0) {
4523 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4524 if (key.objectid == root->root_key.objectid &&
4525 key.type == BTRFS_ROOT_REF_KEY)
4529 btrfs_free_path(path);
4533 /* Delete all dentries for inodes belonging to the root */
4534 static void btrfs_prune_dentries(struct btrfs_root *root)
4536 struct btrfs_fs_info *fs_info = root->fs_info;
4537 struct rb_node *node;
4538 struct rb_node *prev;
4539 struct btrfs_inode *entry;
4540 struct inode *inode;
4543 if (!BTRFS_FS_ERROR(fs_info))
4544 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4546 spin_lock(&root->inode_lock);
4548 node = root->inode_tree.rb_node;
4552 entry = rb_entry(node, struct btrfs_inode, rb_node);
4554 if (objectid < btrfs_ino(entry))
4555 node = node->rb_left;
4556 else if (objectid > btrfs_ino(entry))
4557 node = node->rb_right;
4563 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4564 if (objectid <= btrfs_ino(entry)) {
4568 prev = rb_next(prev);
4572 entry = rb_entry(node, struct btrfs_inode, rb_node);
4573 objectid = btrfs_ino(entry) + 1;
4574 inode = igrab(&entry->vfs_inode);
4576 spin_unlock(&root->inode_lock);
4577 if (atomic_read(&inode->i_count) > 1)
4578 d_prune_aliases(inode);
4580 * btrfs_drop_inode will have it removed from the inode
4581 * cache when its usage count hits zero.
4585 spin_lock(&root->inode_lock);
4589 if (cond_resched_lock(&root->inode_lock))
4592 node = rb_next(node);
4594 spin_unlock(&root->inode_lock);
4597 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4599 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4600 struct btrfs_root *root = dir->root;
4601 struct inode *inode = d_inode(dentry);
4602 struct btrfs_root *dest = BTRFS_I(inode)->root;
4603 struct btrfs_trans_handle *trans;
4604 struct btrfs_block_rsv block_rsv;
4609 * Don't allow to delete a subvolume with send in progress. This is
4610 * inside the inode lock so the error handling that has to drop the bit
4611 * again is not run concurrently.
4613 spin_lock(&dest->root_item_lock);
4614 if (dest->send_in_progress) {
4615 spin_unlock(&dest->root_item_lock);
4617 "attempt to delete subvolume %llu during send",
4618 dest->root_key.objectid);
4621 if (atomic_read(&dest->nr_swapfiles)) {
4622 spin_unlock(&dest->root_item_lock);
4624 "attempt to delete subvolume %llu with active swapfile",
4625 root->root_key.objectid);
4628 root_flags = btrfs_root_flags(&dest->root_item);
4629 btrfs_set_root_flags(&dest->root_item,
4630 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4631 spin_unlock(&dest->root_item_lock);
4633 down_write(&fs_info->subvol_sem);
4635 ret = may_destroy_subvol(dest);
4639 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4641 * One for dir inode,
4642 * two for dir entries,
4643 * two for root ref/backref.
4645 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4649 trans = btrfs_start_transaction(root, 0);
4650 if (IS_ERR(trans)) {
4651 ret = PTR_ERR(trans);
4654 trans->block_rsv = &block_rsv;
4655 trans->bytes_reserved = block_rsv.size;
4657 btrfs_record_snapshot_destroy(trans, dir);
4659 ret = btrfs_unlink_subvol(trans, dir, dentry);
4661 btrfs_abort_transaction(trans, ret);
4665 ret = btrfs_record_root_in_trans(trans, dest);
4667 btrfs_abort_transaction(trans, ret);
4671 memset(&dest->root_item.drop_progress, 0,
4672 sizeof(dest->root_item.drop_progress));
4673 btrfs_set_root_drop_level(&dest->root_item, 0);
4674 btrfs_set_root_refs(&dest->root_item, 0);
4676 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4677 ret = btrfs_insert_orphan_item(trans,
4679 dest->root_key.objectid);
4681 btrfs_abort_transaction(trans, ret);
4686 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4687 BTRFS_UUID_KEY_SUBVOL,
4688 dest->root_key.objectid);
4689 if (ret && ret != -ENOENT) {
4690 btrfs_abort_transaction(trans, ret);
4693 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4694 ret = btrfs_uuid_tree_remove(trans,
4695 dest->root_item.received_uuid,
4696 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4697 dest->root_key.objectid);
4698 if (ret && ret != -ENOENT) {
4699 btrfs_abort_transaction(trans, ret);
4704 free_anon_bdev(dest->anon_dev);
4707 trans->block_rsv = NULL;
4708 trans->bytes_reserved = 0;
4709 ret = btrfs_end_transaction(trans);
4710 inode->i_flags |= S_DEAD;
4712 btrfs_subvolume_release_metadata(root, &block_rsv);
4714 up_write(&fs_info->subvol_sem);
4716 spin_lock(&dest->root_item_lock);
4717 root_flags = btrfs_root_flags(&dest->root_item);
4718 btrfs_set_root_flags(&dest->root_item,
4719 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4720 spin_unlock(&dest->root_item_lock);
4722 d_invalidate(dentry);
4723 btrfs_prune_dentries(dest);
4724 ASSERT(dest->send_in_progress == 0);
4730 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4732 struct inode *inode = d_inode(dentry);
4733 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4735 struct btrfs_trans_handle *trans;
4736 u64 last_unlink_trans;
4737 struct fscrypt_name fname;
4739 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4741 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4742 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4744 "extent tree v2 doesn't support snapshot deletion yet");
4747 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4750 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4754 /* This needs to handle no-key deletions later on */
4756 trans = __unlink_start_trans(BTRFS_I(dir));
4757 if (IS_ERR(trans)) {
4758 err = PTR_ERR(trans);
4762 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4763 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4767 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4771 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4773 /* now the directory is empty */
4774 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4777 btrfs_i_size_write(BTRFS_I(inode), 0);
4779 * Propagate the last_unlink_trans value of the deleted dir to
4780 * its parent directory. This is to prevent an unrecoverable
4781 * log tree in the case we do something like this:
4783 * 2) create snapshot under dir foo
4784 * 3) delete the snapshot
4787 * 6) fsync foo or some file inside foo
4789 if (last_unlink_trans >= trans->transid)
4790 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4793 btrfs_end_transaction(trans);
4795 btrfs_btree_balance_dirty(fs_info);
4796 fscrypt_free_filename(&fname);
4802 * btrfs_truncate_block - read, zero a chunk and write a block
4803 * @inode - inode that we're zeroing
4804 * @from - the offset to start zeroing
4805 * @len - the length to zero, 0 to zero the entire range respective to the
4807 * @front - zero up to the offset instead of from the offset on
4809 * This will find the block for the "from" offset and cow the block and zero the
4810 * part we want to zero. This is used with truncate and hole punching.
4812 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4815 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4816 struct address_space *mapping = inode->vfs_inode.i_mapping;
4817 struct extent_io_tree *io_tree = &inode->io_tree;
4818 struct btrfs_ordered_extent *ordered;
4819 struct extent_state *cached_state = NULL;
4820 struct extent_changeset *data_reserved = NULL;
4821 bool only_release_metadata = false;
4822 u32 blocksize = fs_info->sectorsize;
4823 pgoff_t index = from >> PAGE_SHIFT;
4824 unsigned offset = from & (blocksize - 1);
4826 gfp_t mask = btrfs_alloc_write_mask(mapping);
4827 size_t write_bytes = blocksize;
4832 if (IS_ALIGNED(offset, blocksize) &&
4833 (!len || IS_ALIGNED(len, blocksize)))
4836 block_start = round_down(from, blocksize);
4837 block_end = block_start + blocksize - 1;
4839 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4842 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4843 /* For nocow case, no need to reserve data space */
4844 only_release_metadata = true;
4849 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4851 if (!only_release_metadata)
4852 btrfs_free_reserved_data_space(inode, data_reserved,
4853 block_start, blocksize);
4857 page = find_or_create_page(mapping, index, mask);
4859 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4861 btrfs_delalloc_release_extents(inode, blocksize);
4866 if (!PageUptodate(page)) {
4867 ret = btrfs_read_folio(NULL, page_folio(page));
4869 if (page->mapping != mapping) {
4874 if (!PageUptodate(page)) {
4881 * We unlock the page after the io is completed and then re-lock it
4882 * above. release_folio() could have come in between that and cleared
4883 * PagePrivate(), but left the page in the mapping. Set the page mapped
4884 * here to make sure it's properly set for the subpage stuff.
4886 ret = set_page_extent_mapped(page);
4890 wait_on_page_writeback(page);
4892 lock_extent(io_tree, block_start, block_end, &cached_state);
4894 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4896 unlock_extent(io_tree, block_start, block_end, &cached_state);
4899 btrfs_start_ordered_extent(ordered);
4900 btrfs_put_ordered_extent(ordered);
4904 clear_extent_bit(&inode->io_tree, block_start, block_end,
4905 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4908 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4911 unlock_extent(io_tree, block_start, block_end, &cached_state);
4915 if (offset != blocksize) {
4917 len = blocksize - offset;
4919 memzero_page(page, (block_start - page_offset(page)),
4922 memzero_page(page, (block_start - page_offset(page)) + offset,
4925 btrfs_page_clear_checked(fs_info, page, block_start,
4926 block_end + 1 - block_start);
4927 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4928 unlock_extent(io_tree, block_start, block_end, &cached_state);
4930 if (only_release_metadata)
4931 set_extent_bit(&inode->io_tree, block_start, block_end,
4932 EXTENT_NORESERVE, NULL);
4936 if (only_release_metadata)
4937 btrfs_delalloc_release_metadata(inode, blocksize, true);
4939 btrfs_delalloc_release_space(inode, data_reserved,
4940 block_start, blocksize, true);
4942 btrfs_delalloc_release_extents(inode, blocksize);
4946 if (only_release_metadata)
4947 btrfs_check_nocow_unlock(inode);
4948 extent_changeset_free(data_reserved);
4952 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4953 u64 offset, u64 len)
4955 struct btrfs_fs_info *fs_info = root->fs_info;
4956 struct btrfs_trans_handle *trans;
4957 struct btrfs_drop_extents_args drop_args = { 0 };
4961 * If NO_HOLES is enabled, we don't need to do anything.
4962 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4963 * or btrfs_update_inode() will be called, which guarantee that the next
4964 * fsync will know this inode was changed and needs to be logged.
4966 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4970 * 1 - for the one we're dropping
4971 * 1 - for the one we're adding
4972 * 1 - for updating the inode.
4974 trans = btrfs_start_transaction(root, 3);
4976 return PTR_ERR(trans);
4978 drop_args.start = offset;
4979 drop_args.end = offset + len;
4980 drop_args.drop_cache = true;
4982 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4984 btrfs_abort_transaction(trans, ret);
4985 btrfs_end_transaction(trans);
4989 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4991 btrfs_abort_transaction(trans, ret);
4993 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4994 btrfs_update_inode(trans, root, inode);
4996 btrfs_end_transaction(trans);
5001 * This function puts in dummy file extents for the area we're creating a hole
5002 * for. So if we are truncating this file to a larger size we need to insert
5003 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5004 * the range between oldsize and size
5006 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5008 struct btrfs_root *root = inode->root;
5009 struct btrfs_fs_info *fs_info = root->fs_info;
5010 struct extent_io_tree *io_tree = &inode->io_tree;
5011 struct extent_map *em = NULL;
5012 struct extent_state *cached_state = NULL;
5013 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5014 u64 block_end = ALIGN(size, fs_info->sectorsize);
5021 * If our size started in the middle of a block we need to zero out the
5022 * rest of the block before we expand the i_size, otherwise we could
5023 * expose stale data.
5025 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5029 if (size <= hole_start)
5032 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5034 cur_offset = hole_start;
5036 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5037 block_end - cur_offset);
5043 last_byte = min(extent_map_end(em), block_end);
5044 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5045 hole_size = last_byte - cur_offset;
5047 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5048 struct extent_map *hole_em;
5050 err = maybe_insert_hole(root, inode, cur_offset,
5055 err = btrfs_inode_set_file_extent_range(inode,
5056 cur_offset, hole_size);
5060 hole_em = alloc_extent_map();
5062 btrfs_drop_extent_map_range(inode, cur_offset,
5063 cur_offset + hole_size - 1,
5065 btrfs_set_inode_full_sync(inode);
5068 hole_em->start = cur_offset;
5069 hole_em->len = hole_size;
5070 hole_em->orig_start = cur_offset;
5072 hole_em->block_start = EXTENT_MAP_HOLE;
5073 hole_em->block_len = 0;
5074 hole_em->orig_block_len = 0;
5075 hole_em->ram_bytes = hole_size;
5076 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5077 hole_em->generation = fs_info->generation;
5079 err = btrfs_replace_extent_map_range(inode, hole_em, true);
5080 free_extent_map(hole_em);
5082 err = btrfs_inode_set_file_extent_range(inode,
5083 cur_offset, hole_size);
5088 free_extent_map(em);
5090 cur_offset = last_byte;
5091 if (cur_offset >= block_end)
5094 free_extent_map(em);
5095 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5099 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5101 struct btrfs_root *root = BTRFS_I(inode)->root;
5102 struct btrfs_trans_handle *trans;
5103 loff_t oldsize = i_size_read(inode);
5104 loff_t newsize = attr->ia_size;
5105 int mask = attr->ia_valid;
5109 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5110 * special case where we need to update the times despite not having
5111 * these flags set. For all other operations the VFS set these flags
5112 * explicitly if it wants a timestamp update.
5114 if (newsize != oldsize) {
5115 inode_inc_iversion(inode);
5116 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5117 inode->i_mtime = current_time(inode);
5118 inode->i_ctime = inode->i_mtime;
5122 if (newsize > oldsize) {
5124 * Don't do an expanding truncate while snapshotting is ongoing.
5125 * This is to ensure the snapshot captures a fully consistent
5126 * state of this file - if the snapshot captures this expanding
5127 * truncation, it must capture all writes that happened before
5130 btrfs_drew_write_lock(&root->snapshot_lock);
5131 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5133 btrfs_drew_write_unlock(&root->snapshot_lock);
5137 trans = btrfs_start_transaction(root, 1);
5138 if (IS_ERR(trans)) {
5139 btrfs_drew_write_unlock(&root->snapshot_lock);
5140 return PTR_ERR(trans);
5143 i_size_write(inode, newsize);
5144 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5145 pagecache_isize_extended(inode, oldsize, newsize);
5146 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5147 btrfs_drew_write_unlock(&root->snapshot_lock);
5148 btrfs_end_transaction(trans);
5150 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5152 if (btrfs_is_zoned(fs_info)) {
5153 ret = btrfs_wait_ordered_range(inode,
5154 ALIGN(newsize, fs_info->sectorsize),
5161 * We're truncating a file that used to have good data down to
5162 * zero. Make sure any new writes to the file get on disk
5166 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5167 &BTRFS_I(inode)->runtime_flags);
5169 truncate_setsize(inode, newsize);
5171 inode_dio_wait(inode);
5173 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5174 if (ret && inode->i_nlink) {
5178 * Truncate failed, so fix up the in-memory size. We
5179 * adjusted disk_i_size down as we removed extents, so
5180 * wait for disk_i_size to be stable and then update the
5181 * in-memory size to match.
5183 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5186 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5193 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5196 struct inode *inode = d_inode(dentry);
5197 struct btrfs_root *root = BTRFS_I(inode)->root;
5200 if (btrfs_root_readonly(root))
5203 err = setattr_prepare(idmap, dentry, attr);
5207 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5208 err = btrfs_setsize(inode, attr);
5213 if (attr->ia_valid) {
5214 setattr_copy(idmap, inode, attr);
5215 inode_inc_iversion(inode);
5216 err = btrfs_dirty_inode(BTRFS_I(inode));
5218 if (!err && attr->ia_valid & ATTR_MODE)
5219 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5226 * While truncating the inode pages during eviction, we get the VFS
5227 * calling btrfs_invalidate_folio() against each folio of the inode. This
5228 * is slow because the calls to btrfs_invalidate_folio() result in a
5229 * huge amount of calls to lock_extent() and clear_extent_bit(),
5230 * which keep merging and splitting extent_state structures over and over,
5231 * wasting lots of time.
5233 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5234 * skip all those expensive operations on a per folio basis and do only
5235 * the ordered io finishing, while we release here the extent_map and
5236 * extent_state structures, without the excessive merging and splitting.
5238 static void evict_inode_truncate_pages(struct inode *inode)
5240 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5241 struct rb_node *node;
5243 ASSERT(inode->i_state & I_FREEING);
5244 truncate_inode_pages_final(&inode->i_data);
5246 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5249 * Keep looping until we have no more ranges in the io tree.
5250 * We can have ongoing bios started by readahead that have
5251 * their endio callback (extent_io.c:end_bio_extent_readpage)
5252 * still in progress (unlocked the pages in the bio but did not yet
5253 * unlocked the ranges in the io tree). Therefore this means some
5254 * ranges can still be locked and eviction started because before
5255 * submitting those bios, which are executed by a separate task (work
5256 * queue kthread), inode references (inode->i_count) were not taken
5257 * (which would be dropped in the end io callback of each bio).
5258 * Therefore here we effectively end up waiting for those bios and
5259 * anyone else holding locked ranges without having bumped the inode's
5260 * reference count - if we don't do it, when they access the inode's
5261 * io_tree to unlock a range it may be too late, leading to an
5262 * use-after-free issue.
5264 spin_lock(&io_tree->lock);
5265 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5266 struct extent_state *state;
5267 struct extent_state *cached_state = NULL;
5270 unsigned state_flags;
5272 node = rb_first(&io_tree->state);
5273 state = rb_entry(node, struct extent_state, rb_node);
5274 start = state->start;
5276 state_flags = state->state;
5277 spin_unlock(&io_tree->lock);
5279 lock_extent(io_tree, start, end, &cached_state);
5282 * If still has DELALLOC flag, the extent didn't reach disk,
5283 * and its reserved space won't be freed by delayed_ref.
5284 * So we need to free its reserved space here.
5285 * (Refer to comment in btrfs_invalidate_folio, case 2)
5287 * Note, end is the bytenr of last byte, so we need + 1 here.
5289 if (state_flags & EXTENT_DELALLOC)
5290 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5293 clear_extent_bit(io_tree, start, end,
5294 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5298 spin_lock(&io_tree->lock);
5300 spin_unlock(&io_tree->lock);
5303 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5304 struct btrfs_block_rsv *rsv)
5306 struct btrfs_fs_info *fs_info = root->fs_info;
5307 struct btrfs_trans_handle *trans;
5308 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5312 * Eviction should be taking place at some place safe because of our
5313 * delayed iputs. However the normal flushing code will run delayed
5314 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5316 * We reserve the delayed_refs_extra here again because we can't use
5317 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5318 * above. We reserve our extra bit here because we generate a ton of
5319 * delayed refs activity by truncating.
5321 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5322 * if we fail to make this reservation we can re-try without the
5323 * delayed_refs_extra so we can make some forward progress.
5325 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5326 BTRFS_RESERVE_FLUSH_EVICT);
5328 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5329 BTRFS_RESERVE_FLUSH_EVICT);
5332 "could not allocate space for delete; will truncate on mount");
5333 return ERR_PTR(-ENOSPC);
5335 delayed_refs_extra = 0;
5338 trans = btrfs_join_transaction(root);
5342 if (delayed_refs_extra) {
5343 trans->block_rsv = &fs_info->trans_block_rsv;
5344 trans->bytes_reserved = delayed_refs_extra;
5345 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5346 delayed_refs_extra, true);
5351 void btrfs_evict_inode(struct inode *inode)
5353 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5354 struct btrfs_trans_handle *trans;
5355 struct btrfs_root *root = BTRFS_I(inode)->root;
5356 struct btrfs_block_rsv *rsv = NULL;
5359 trace_btrfs_inode_evict(inode);
5362 fsverity_cleanup_inode(inode);
5367 evict_inode_truncate_pages(inode);
5369 if (inode->i_nlink &&
5370 ((btrfs_root_refs(&root->root_item) != 0 &&
5371 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5372 btrfs_is_free_space_inode(BTRFS_I(inode))))
5375 if (is_bad_inode(inode))
5378 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5381 if (inode->i_nlink > 0) {
5382 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5383 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5388 * This makes sure the inode item in tree is uptodate and the space for
5389 * the inode update is released.
5391 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5396 * This drops any pending insert or delete operations we have for this
5397 * inode. We could have a delayed dir index deletion queued up, but
5398 * we're removing the inode completely so that'll be taken care of in
5401 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5403 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5406 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5407 rsv->failfast = true;
5409 btrfs_i_size_write(BTRFS_I(inode), 0);
5412 struct btrfs_truncate_control control = {
5413 .inode = BTRFS_I(inode),
5414 .ino = btrfs_ino(BTRFS_I(inode)),
5419 trans = evict_refill_and_join(root, rsv);
5423 trans->block_rsv = rsv;
5425 ret = btrfs_truncate_inode_items(trans, root, &control);
5426 trans->block_rsv = &fs_info->trans_block_rsv;
5427 btrfs_end_transaction(trans);
5429 * We have not added new delayed items for our inode after we
5430 * have flushed its delayed items, so no need to throttle on
5431 * delayed items. However we have modified extent buffers.
5433 btrfs_btree_balance_dirty_nodelay(fs_info);
5434 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5441 * Errors here aren't a big deal, it just means we leave orphan items in
5442 * the tree. They will be cleaned up on the next mount. If the inode
5443 * number gets reused, cleanup deletes the orphan item without doing
5444 * anything, and unlink reuses the existing orphan item.
5446 * If it turns out that we are dropping too many of these, we might want
5447 * to add a mechanism for retrying these after a commit.
5449 trans = evict_refill_and_join(root, rsv);
5450 if (!IS_ERR(trans)) {
5451 trans->block_rsv = rsv;
5452 btrfs_orphan_del(trans, BTRFS_I(inode));
5453 trans->block_rsv = &fs_info->trans_block_rsv;
5454 btrfs_end_transaction(trans);
5458 btrfs_free_block_rsv(fs_info, rsv);
5460 * If we didn't successfully delete, the orphan item will still be in
5461 * the tree and we'll retry on the next mount. Again, we might also want
5462 * to retry these periodically in the future.
5464 btrfs_remove_delayed_node(BTRFS_I(inode));
5465 fsverity_cleanup_inode(inode);
5470 * Return the key found in the dir entry in the location pointer, fill @type
5471 * with BTRFS_FT_*, and return 0.
5473 * If no dir entries were found, returns -ENOENT.
5474 * If found a corrupted location in dir entry, returns -EUCLEAN.
5476 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5477 struct btrfs_key *location, u8 *type)
5479 struct btrfs_dir_item *di;
5480 struct btrfs_path *path;
5481 struct btrfs_root *root = dir->root;
5483 struct fscrypt_name fname;
5485 path = btrfs_alloc_path();
5489 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5493 * fscrypt_setup_filename() should never return a positive value, but
5494 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5498 /* This needs to handle no-key deletions later on */
5500 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5501 &fname.disk_name, 0);
5502 if (IS_ERR_OR_NULL(di)) {
5503 ret = di ? PTR_ERR(di) : -ENOENT;
5507 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5508 if (location->type != BTRFS_INODE_ITEM_KEY &&
5509 location->type != BTRFS_ROOT_ITEM_KEY) {
5511 btrfs_warn(root->fs_info,
5512 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5513 __func__, fname.disk_name.name, btrfs_ino(dir),
5514 location->objectid, location->type, location->offset);
5517 *type = btrfs_dir_ftype(path->nodes[0], di);
5519 fscrypt_free_filename(&fname);
5520 btrfs_free_path(path);
5525 * when we hit a tree root in a directory, the btrfs part of the inode
5526 * needs to be changed to reflect the root directory of the tree root. This
5527 * is kind of like crossing a mount point.
5529 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5530 struct btrfs_inode *dir,
5531 struct dentry *dentry,
5532 struct btrfs_key *location,
5533 struct btrfs_root **sub_root)
5535 struct btrfs_path *path;
5536 struct btrfs_root *new_root;
5537 struct btrfs_root_ref *ref;
5538 struct extent_buffer *leaf;
5539 struct btrfs_key key;
5542 struct fscrypt_name fname;
5544 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5548 path = btrfs_alloc_path();
5555 key.objectid = dir->root->root_key.objectid;
5556 key.type = BTRFS_ROOT_REF_KEY;
5557 key.offset = location->objectid;
5559 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5566 leaf = path->nodes[0];
5567 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5568 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5569 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5572 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5573 (unsigned long)(ref + 1), fname.disk_name.len);
5577 btrfs_release_path(path);
5579 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5580 if (IS_ERR(new_root)) {
5581 err = PTR_ERR(new_root);
5585 *sub_root = new_root;
5586 location->objectid = btrfs_root_dirid(&new_root->root_item);
5587 location->type = BTRFS_INODE_ITEM_KEY;
5588 location->offset = 0;
5591 btrfs_free_path(path);
5592 fscrypt_free_filename(&fname);
5596 static void inode_tree_add(struct btrfs_inode *inode)
5598 struct btrfs_root *root = inode->root;
5599 struct btrfs_inode *entry;
5601 struct rb_node *parent;
5602 struct rb_node *new = &inode->rb_node;
5603 u64 ino = btrfs_ino(inode);
5605 if (inode_unhashed(&inode->vfs_inode))
5608 spin_lock(&root->inode_lock);
5609 p = &root->inode_tree.rb_node;
5612 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5614 if (ino < btrfs_ino(entry))
5615 p = &parent->rb_left;
5616 else if (ino > btrfs_ino(entry))
5617 p = &parent->rb_right;
5619 WARN_ON(!(entry->vfs_inode.i_state &
5620 (I_WILL_FREE | I_FREEING)));
5621 rb_replace_node(parent, new, &root->inode_tree);
5622 RB_CLEAR_NODE(parent);
5623 spin_unlock(&root->inode_lock);
5627 rb_link_node(new, parent, p);
5628 rb_insert_color(new, &root->inode_tree);
5629 spin_unlock(&root->inode_lock);
5632 static void inode_tree_del(struct btrfs_inode *inode)
5634 struct btrfs_root *root = inode->root;
5637 spin_lock(&root->inode_lock);
5638 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5639 rb_erase(&inode->rb_node, &root->inode_tree);
5640 RB_CLEAR_NODE(&inode->rb_node);
5641 empty = RB_EMPTY_ROOT(&root->inode_tree);
5643 spin_unlock(&root->inode_lock);
5645 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5646 spin_lock(&root->inode_lock);
5647 empty = RB_EMPTY_ROOT(&root->inode_tree);
5648 spin_unlock(&root->inode_lock);
5650 btrfs_add_dead_root(root);
5655 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5657 struct btrfs_iget_args *args = p;
5659 inode->i_ino = args->ino;
5660 BTRFS_I(inode)->location.objectid = args->ino;
5661 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5662 BTRFS_I(inode)->location.offset = 0;
5663 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5664 BUG_ON(args->root && !BTRFS_I(inode)->root);
5666 if (args->root && args->root == args->root->fs_info->tree_root &&
5667 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5668 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5669 &BTRFS_I(inode)->runtime_flags);
5673 static int btrfs_find_actor(struct inode *inode, void *opaque)
5675 struct btrfs_iget_args *args = opaque;
5677 return args->ino == BTRFS_I(inode)->location.objectid &&
5678 args->root == BTRFS_I(inode)->root;
5681 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5682 struct btrfs_root *root)
5684 struct inode *inode;
5685 struct btrfs_iget_args args;
5686 unsigned long hashval = btrfs_inode_hash(ino, root);
5691 inode = iget5_locked(s, hashval, btrfs_find_actor,
5692 btrfs_init_locked_inode,
5698 * Get an inode object given its inode number and corresponding root.
5699 * Path can be preallocated to prevent recursing back to iget through
5700 * allocator. NULL is also valid but may require an additional allocation
5703 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5704 struct btrfs_root *root, struct btrfs_path *path)
5706 struct inode *inode;
5708 inode = btrfs_iget_locked(s, ino, root);
5710 return ERR_PTR(-ENOMEM);
5712 if (inode->i_state & I_NEW) {
5715 ret = btrfs_read_locked_inode(inode, path);
5717 inode_tree_add(BTRFS_I(inode));
5718 unlock_new_inode(inode);
5722 * ret > 0 can come from btrfs_search_slot called by
5723 * btrfs_read_locked_inode, this means the inode item
5728 inode = ERR_PTR(ret);
5735 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5737 return btrfs_iget_path(s, ino, root, NULL);
5740 static struct inode *new_simple_dir(struct super_block *s,
5741 struct btrfs_key *key,
5742 struct btrfs_root *root)
5744 struct inode *inode = new_inode(s);
5747 return ERR_PTR(-ENOMEM);
5749 BTRFS_I(inode)->root = btrfs_grab_root(root);
5750 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5751 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5753 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5755 * We only need lookup, the rest is read-only and there's no inode
5756 * associated with the dentry
5758 inode->i_op = &simple_dir_inode_operations;
5759 inode->i_opflags &= ~IOP_XATTR;
5760 inode->i_fop = &simple_dir_operations;
5761 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5762 inode->i_mtime = current_time(inode);
5763 inode->i_atime = inode->i_mtime;
5764 inode->i_ctime = inode->i_mtime;
5765 BTRFS_I(inode)->i_otime = inode->i_mtime;
5770 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5771 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5772 static_assert(BTRFS_FT_DIR == FT_DIR);
5773 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5774 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5775 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5776 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5777 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5779 static inline u8 btrfs_inode_type(struct inode *inode)
5781 return fs_umode_to_ftype(inode->i_mode);
5784 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5786 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5787 struct inode *inode;
5788 struct btrfs_root *root = BTRFS_I(dir)->root;
5789 struct btrfs_root *sub_root = root;
5790 struct btrfs_key location;
5794 if (dentry->d_name.len > BTRFS_NAME_LEN)
5795 return ERR_PTR(-ENAMETOOLONG);
5797 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5799 return ERR_PTR(ret);
5801 if (location.type == BTRFS_INODE_ITEM_KEY) {
5802 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5806 /* Do extra check against inode mode with di_type */
5807 if (btrfs_inode_type(inode) != di_type) {
5809 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5810 inode->i_mode, btrfs_inode_type(inode),
5813 return ERR_PTR(-EUCLEAN);
5818 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5819 &location, &sub_root);
5822 inode = ERR_PTR(ret);
5824 inode = new_simple_dir(dir->i_sb, &location, root);
5826 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5827 btrfs_put_root(sub_root);
5832 down_read(&fs_info->cleanup_work_sem);
5833 if (!sb_rdonly(inode->i_sb))
5834 ret = btrfs_orphan_cleanup(sub_root);
5835 up_read(&fs_info->cleanup_work_sem);
5838 inode = ERR_PTR(ret);
5845 static int btrfs_dentry_delete(const struct dentry *dentry)
5847 struct btrfs_root *root;
5848 struct inode *inode = d_inode(dentry);
5850 if (!inode && !IS_ROOT(dentry))
5851 inode = d_inode(dentry->d_parent);
5854 root = BTRFS_I(inode)->root;
5855 if (btrfs_root_refs(&root->root_item) == 0)
5858 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5864 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5867 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5869 if (inode == ERR_PTR(-ENOENT))
5871 return d_splice_alias(inode, dentry);
5875 * Find the highest existing sequence number in a directory and then set the
5876 * in-memory index_cnt variable to the first free sequence number.
5878 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5880 struct btrfs_root *root = inode->root;
5881 struct btrfs_key key, found_key;
5882 struct btrfs_path *path;
5883 struct extent_buffer *leaf;
5886 key.objectid = btrfs_ino(inode);
5887 key.type = BTRFS_DIR_INDEX_KEY;
5888 key.offset = (u64)-1;
5890 path = btrfs_alloc_path();
5894 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5897 /* FIXME: we should be able to handle this */
5902 if (path->slots[0] == 0) {
5903 inode->index_cnt = BTRFS_DIR_START_INDEX;
5909 leaf = path->nodes[0];
5910 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5912 if (found_key.objectid != btrfs_ino(inode) ||
5913 found_key.type != BTRFS_DIR_INDEX_KEY) {
5914 inode->index_cnt = BTRFS_DIR_START_INDEX;
5918 inode->index_cnt = found_key.offset + 1;
5920 btrfs_free_path(path);
5924 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5926 if (dir->index_cnt == (u64)-1) {
5929 ret = btrfs_inode_delayed_dir_index_count(dir);
5931 ret = btrfs_set_inode_index_count(dir);
5937 *index = dir->index_cnt;
5943 * All this infrastructure exists because dir_emit can fault, and we are holding
5944 * the tree lock when doing readdir. For now just allocate a buffer and copy
5945 * our information into that, and then dir_emit from the buffer. This is
5946 * similar to what NFS does, only we don't keep the buffer around in pagecache
5947 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5948 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5951 static int btrfs_opendir(struct inode *inode, struct file *file)
5953 struct btrfs_file_private *private;
5957 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5961 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5964 private->last_index = last_index;
5965 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5966 if (!private->filldir_buf) {
5970 file->private_data = private;
5981 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5984 struct dir_entry *entry = addr;
5985 char *name = (char *)(entry + 1);
5987 ctx->pos = get_unaligned(&entry->offset);
5988 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5989 get_unaligned(&entry->ino),
5990 get_unaligned(&entry->type)))
5992 addr += sizeof(struct dir_entry) +
5993 get_unaligned(&entry->name_len);
5999 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
6001 struct inode *inode = file_inode(file);
6002 struct btrfs_root *root = BTRFS_I(inode)->root;
6003 struct btrfs_file_private *private = file->private_data;
6004 struct btrfs_dir_item *di;
6005 struct btrfs_key key;
6006 struct btrfs_key found_key;
6007 struct btrfs_path *path;
6009 struct list_head ins_list;
6010 struct list_head del_list;
6017 struct btrfs_key location;
6019 if (!dir_emit_dots(file, ctx))
6022 path = btrfs_alloc_path();
6026 addr = private->filldir_buf;
6027 path->reada = READA_FORWARD;
6029 INIT_LIST_HEAD(&ins_list);
6030 INIT_LIST_HEAD(&del_list);
6031 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
6032 &ins_list, &del_list);
6035 key.type = BTRFS_DIR_INDEX_KEY;
6036 key.offset = ctx->pos;
6037 key.objectid = btrfs_ino(BTRFS_I(inode));
6039 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
6040 struct dir_entry *entry;
6041 struct extent_buffer *leaf = path->nodes[0];
6044 if (found_key.objectid != key.objectid)
6046 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6048 if (found_key.offset < ctx->pos)
6050 if (found_key.offset > private->last_index)
6052 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6054 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
6055 name_len = btrfs_dir_name_len(leaf, di);
6056 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6058 btrfs_release_path(path);
6059 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6062 addr = private->filldir_buf;
6068 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
6070 name_ptr = (char *)(entry + 1);
6071 read_extent_buffer(leaf, name_ptr,
6072 (unsigned long)(di + 1), name_len);
6073 put_unaligned(name_len, &entry->name_len);
6074 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
6075 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6076 put_unaligned(location.objectid, &entry->ino);
6077 put_unaligned(found_key.offset, &entry->offset);
6079 addr += sizeof(struct dir_entry) + name_len;
6080 total_len += sizeof(struct dir_entry) + name_len;
6082 /* Catch error encountered during iteration */
6086 btrfs_release_path(path);
6088 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6092 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6097 * Stop new entries from being returned after we return the last
6100 * New directory entries are assigned a strictly increasing
6101 * offset. This means that new entries created during readdir
6102 * are *guaranteed* to be seen in the future by that readdir.
6103 * This has broken buggy programs which operate on names as
6104 * they're returned by readdir. Until we re-use freed offsets
6105 * we have this hack to stop new entries from being returned
6106 * under the assumption that they'll never reach this huge
6109 * This is being careful not to overflow 32bit loff_t unless the
6110 * last entry requires it because doing so has broken 32bit apps
6113 if (ctx->pos >= INT_MAX)
6114 ctx->pos = LLONG_MAX;
6121 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6122 btrfs_free_path(path);
6127 * This is somewhat expensive, updating the tree every time the
6128 * inode changes. But, it is most likely to find the inode in cache.
6129 * FIXME, needs more benchmarking...there are no reasons other than performance
6130 * to keep or drop this code.
6132 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6134 struct btrfs_root *root = inode->root;
6135 struct btrfs_fs_info *fs_info = root->fs_info;
6136 struct btrfs_trans_handle *trans;
6139 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6142 trans = btrfs_join_transaction(root);
6144 return PTR_ERR(trans);
6146 ret = btrfs_update_inode(trans, root, inode);
6147 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6148 /* whoops, lets try again with the full transaction */
6149 btrfs_end_transaction(trans);
6150 trans = btrfs_start_transaction(root, 1);
6152 return PTR_ERR(trans);
6154 ret = btrfs_update_inode(trans, root, inode);
6156 btrfs_end_transaction(trans);
6157 if (inode->delayed_node)
6158 btrfs_balance_delayed_items(fs_info);
6164 * This is a copy of file_update_time. We need this so we can return error on
6165 * ENOSPC for updating the inode in the case of file write and mmap writes.
6167 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6170 struct btrfs_root *root = BTRFS_I(inode)->root;
6171 bool dirty = flags & ~S_VERSION;
6173 if (btrfs_root_readonly(root))
6176 if (flags & S_VERSION)
6177 dirty |= inode_maybe_inc_iversion(inode, dirty);
6178 if (flags & S_CTIME)
6179 inode->i_ctime = *now;
6180 if (flags & S_MTIME)
6181 inode->i_mtime = *now;
6182 if (flags & S_ATIME)
6183 inode->i_atime = *now;
6184 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6188 * helper to find a free sequence number in a given directory. This current
6189 * code is very simple, later versions will do smarter things in the btree
6191 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6195 if (dir->index_cnt == (u64)-1) {
6196 ret = btrfs_inode_delayed_dir_index_count(dir);
6198 ret = btrfs_set_inode_index_count(dir);
6204 *index = dir->index_cnt;
6210 static int btrfs_insert_inode_locked(struct inode *inode)
6212 struct btrfs_iget_args args;
6214 args.ino = BTRFS_I(inode)->location.objectid;
6215 args.root = BTRFS_I(inode)->root;
6217 return insert_inode_locked4(inode,
6218 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6219 btrfs_find_actor, &args);
6222 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6223 unsigned int *trans_num_items)
6225 struct inode *dir = args->dir;
6226 struct inode *inode = args->inode;
6229 if (!args->orphan) {
6230 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6236 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6238 fscrypt_free_filename(&args->fname);
6242 /* 1 to add inode item */
6243 *trans_num_items = 1;
6244 /* 1 to add compression property */
6245 if (BTRFS_I(dir)->prop_compress)
6246 (*trans_num_items)++;
6247 /* 1 to add default ACL xattr */
6248 if (args->default_acl)
6249 (*trans_num_items)++;
6250 /* 1 to add access ACL xattr */
6252 (*trans_num_items)++;
6253 #ifdef CONFIG_SECURITY
6254 /* 1 to add LSM xattr */
6255 if (dir->i_security)
6256 (*trans_num_items)++;
6259 /* 1 to add orphan item */
6260 (*trans_num_items)++;
6264 * 1 to add dir index
6265 * 1 to update parent inode item
6267 * No need for 1 unit for the inode ref item because it is
6268 * inserted in a batch together with the inode item at
6269 * btrfs_create_new_inode().
6271 *trans_num_items += 3;
6276 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6278 posix_acl_release(args->acl);
6279 posix_acl_release(args->default_acl);
6280 fscrypt_free_filename(&args->fname);
6284 * Inherit flags from the parent inode.
6286 * Currently only the compression flags and the cow flags are inherited.
6288 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6294 if (flags & BTRFS_INODE_NOCOMPRESS) {
6295 inode->flags &= ~BTRFS_INODE_COMPRESS;
6296 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6297 } else if (flags & BTRFS_INODE_COMPRESS) {
6298 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6299 inode->flags |= BTRFS_INODE_COMPRESS;
6302 if (flags & BTRFS_INODE_NODATACOW) {
6303 inode->flags |= BTRFS_INODE_NODATACOW;
6304 if (S_ISREG(inode->vfs_inode.i_mode))
6305 inode->flags |= BTRFS_INODE_NODATASUM;
6308 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6311 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6312 struct btrfs_new_inode_args *args)
6314 struct inode *dir = args->dir;
6315 struct inode *inode = args->inode;
6316 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6317 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6318 struct btrfs_root *root;
6319 struct btrfs_inode_item *inode_item;
6320 struct btrfs_key *location;
6321 struct btrfs_path *path;
6323 struct btrfs_inode_ref *ref;
6324 struct btrfs_key key[2];
6326 struct btrfs_item_batch batch;
6330 path = btrfs_alloc_path();
6335 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6336 root = BTRFS_I(inode)->root;
6338 ret = btrfs_get_free_objectid(root, &objectid);
6341 inode->i_ino = objectid;
6345 * O_TMPFILE, set link count to 0, so that after this point, we
6346 * fill in an inode item with the correct link count.
6348 set_nlink(inode, 0);
6350 trace_btrfs_inode_request(dir);
6352 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6356 /* index_cnt is ignored for everything but a dir. */
6357 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6358 BTRFS_I(inode)->generation = trans->transid;
6359 inode->i_generation = BTRFS_I(inode)->generation;
6362 * Subvolumes don't inherit flags from their parent directory.
6363 * Originally this was probably by accident, but we probably can't
6364 * change it now without compatibility issues.
6367 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6369 if (S_ISREG(inode->i_mode)) {
6370 if (btrfs_test_opt(fs_info, NODATASUM))
6371 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6372 if (btrfs_test_opt(fs_info, NODATACOW))
6373 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6374 BTRFS_INODE_NODATASUM;
6377 location = &BTRFS_I(inode)->location;
6378 location->objectid = objectid;
6379 location->offset = 0;
6380 location->type = BTRFS_INODE_ITEM_KEY;
6382 ret = btrfs_insert_inode_locked(inode);
6385 BTRFS_I(dir)->index_cnt--;
6390 * We could have gotten an inode number from somebody who was fsynced
6391 * and then removed in this same transaction, so let's just set full
6392 * sync since it will be a full sync anyway and this will blow away the
6393 * old info in the log.
6395 btrfs_set_inode_full_sync(BTRFS_I(inode));
6397 key[0].objectid = objectid;
6398 key[0].type = BTRFS_INODE_ITEM_KEY;
6401 sizes[0] = sizeof(struct btrfs_inode_item);
6403 if (!args->orphan) {
6405 * Start new inodes with an inode_ref. This is slightly more
6406 * efficient for small numbers of hard links since they will
6407 * be packed into one item. Extended refs will kick in if we
6408 * add more hard links than can fit in the ref item.
6410 key[1].objectid = objectid;
6411 key[1].type = BTRFS_INODE_REF_KEY;
6413 key[1].offset = objectid;
6414 sizes[1] = 2 + sizeof(*ref);
6416 key[1].offset = btrfs_ino(BTRFS_I(dir));
6417 sizes[1] = name->len + sizeof(*ref);
6421 batch.keys = &key[0];
6422 batch.data_sizes = &sizes[0];
6423 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6424 batch.nr = args->orphan ? 1 : 2;
6425 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6427 btrfs_abort_transaction(trans, ret);
6431 inode->i_mtime = current_time(inode);
6432 inode->i_atime = inode->i_mtime;
6433 inode->i_ctime = inode->i_mtime;
6434 BTRFS_I(inode)->i_otime = inode->i_mtime;
6437 * We're going to fill the inode item now, so at this point the inode
6438 * must be fully initialized.
6441 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6442 struct btrfs_inode_item);
6443 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6444 sizeof(*inode_item));
6445 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6447 if (!args->orphan) {
6448 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6449 struct btrfs_inode_ref);
6450 ptr = (unsigned long)(ref + 1);
6452 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6453 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6454 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6456 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6458 btrfs_set_inode_ref_index(path->nodes[0], ref,
6459 BTRFS_I(inode)->dir_index);
6460 write_extent_buffer(path->nodes[0], name->name, ptr,
6465 btrfs_mark_buffer_dirty(path->nodes[0]);
6467 * We don't need the path anymore, plus inheriting properties, adding
6468 * ACLs, security xattrs, orphan item or adding the link, will result in
6469 * allocating yet another path. So just free our path.
6471 btrfs_free_path(path);
6475 struct inode *parent;
6478 * Subvolumes inherit properties from their parent subvolume,
6479 * not the directory they were created in.
6481 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6482 BTRFS_I(dir)->root);
6483 if (IS_ERR(parent)) {
6484 ret = PTR_ERR(parent);
6486 ret = btrfs_inode_inherit_props(trans, inode, parent);
6490 ret = btrfs_inode_inherit_props(trans, inode, dir);
6494 "error inheriting props for ino %llu (root %llu): %d",
6495 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6500 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6503 if (!args->subvol) {
6504 ret = btrfs_init_inode_security(trans, args);
6506 btrfs_abort_transaction(trans, ret);
6511 inode_tree_add(BTRFS_I(inode));
6513 trace_btrfs_inode_new(inode);
6514 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6516 btrfs_update_root_times(trans, root);
6519 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6521 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6522 0, BTRFS_I(inode)->dir_index);
6525 btrfs_abort_transaction(trans, ret);
6533 * discard_new_inode() calls iput(), but the caller owns the reference
6537 discard_new_inode(inode);
6539 btrfs_free_path(path);
6544 * utility function to add 'inode' into 'parent_inode' with
6545 * a give name and a given sequence number.
6546 * if 'add_backref' is true, also insert a backref from the
6547 * inode to the parent directory.
6549 int btrfs_add_link(struct btrfs_trans_handle *trans,
6550 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6551 const struct fscrypt_str *name, int add_backref, u64 index)
6554 struct btrfs_key key;
6555 struct btrfs_root *root = parent_inode->root;
6556 u64 ino = btrfs_ino(inode);
6557 u64 parent_ino = btrfs_ino(parent_inode);
6559 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6560 memcpy(&key, &inode->root->root_key, sizeof(key));
6563 key.type = BTRFS_INODE_ITEM_KEY;
6567 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6568 ret = btrfs_add_root_ref(trans, key.objectid,
6569 root->root_key.objectid, parent_ino,
6571 } else if (add_backref) {
6572 ret = btrfs_insert_inode_ref(trans, root, name,
6573 ino, parent_ino, index);
6576 /* Nothing to clean up yet */
6580 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6581 btrfs_inode_type(&inode->vfs_inode), index);
6582 if (ret == -EEXIST || ret == -EOVERFLOW)
6585 btrfs_abort_transaction(trans, ret);
6589 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6591 inode_inc_iversion(&parent_inode->vfs_inode);
6593 * If we are replaying a log tree, we do not want to update the mtime
6594 * and ctime of the parent directory with the current time, since the
6595 * log replay procedure is responsible for setting them to their correct
6596 * values (the ones it had when the fsync was done).
6598 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6599 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6601 parent_inode->vfs_inode.i_mtime = now;
6602 parent_inode->vfs_inode.i_ctime = now;
6604 ret = btrfs_update_inode(trans, root, parent_inode);
6606 btrfs_abort_transaction(trans, ret);
6610 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6613 err = btrfs_del_root_ref(trans, key.objectid,
6614 root->root_key.objectid, parent_ino,
6615 &local_index, name);
6617 btrfs_abort_transaction(trans, err);
6618 } else if (add_backref) {
6622 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6625 btrfs_abort_transaction(trans, err);
6628 /* Return the original error code */
6632 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6633 struct inode *inode)
6635 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6636 struct btrfs_root *root = BTRFS_I(dir)->root;
6637 struct btrfs_new_inode_args new_inode_args = {
6642 unsigned int trans_num_items;
6643 struct btrfs_trans_handle *trans;
6646 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6650 trans = btrfs_start_transaction(root, trans_num_items);
6651 if (IS_ERR(trans)) {
6652 err = PTR_ERR(trans);
6653 goto out_new_inode_args;
6656 err = btrfs_create_new_inode(trans, &new_inode_args);
6658 d_instantiate_new(dentry, inode);
6660 btrfs_end_transaction(trans);
6661 btrfs_btree_balance_dirty(fs_info);
6663 btrfs_new_inode_args_destroy(&new_inode_args);
6670 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6671 struct dentry *dentry, umode_t mode, dev_t rdev)
6673 struct inode *inode;
6675 inode = new_inode(dir->i_sb);
6678 inode_init_owner(idmap, inode, dir, mode);
6679 inode->i_op = &btrfs_special_inode_operations;
6680 init_special_inode(inode, inode->i_mode, rdev);
6681 return btrfs_create_common(dir, dentry, inode);
6684 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6685 struct dentry *dentry, umode_t mode, bool excl)
6687 struct inode *inode;
6689 inode = new_inode(dir->i_sb);
6692 inode_init_owner(idmap, inode, dir, mode);
6693 inode->i_fop = &btrfs_file_operations;
6694 inode->i_op = &btrfs_file_inode_operations;
6695 inode->i_mapping->a_ops = &btrfs_aops;
6696 return btrfs_create_common(dir, dentry, inode);
6699 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6700 struct dentry *dentry)
6702 struct btrfs_trans_handle *trans = NULL;
6703 struct btrfs_root *root = BTRFS_I(dir)->root;
6704 struct inode *inode = d_inode(old_dentry);
6705 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6706 struct fscrypt_name fname;
6711 /* do not allow sys_link's with other subvols of the same device */
6712 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6715 if (inode->i_nlink >= BTRFS_LINK_MAX)
6718 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6722 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6727 * 2 items for inode and inode ref
6728 * 2 items for dir items
6729 * 1 item for parent inode
6730 * 1 item for orphan item deletion if O_TMPFILE
6732 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6733 if (IS_ERR(trans)) {
6734 err = PTR_ERR(trans);
6739 /* There are several dir indexes for this inode, clear the cache. */
6740 BTRFS_I(inode)->dir_index = 0ULL;
6742 inode_inc_iversion(inode);
6743 inode->i_ctime = current_time(inode);
6745 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6747 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6748 &fname.disk_name, 1, index);
6753 struct dentry *parent = dentry->d_parent;
6755 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6758 if (inode->i_nlink == 1) {
6760 * If new hard link count is 1, it's a file created
6761 * with open(2) O_TMPFILE flag.
6763 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6767 d_instantiate(dentry, inode);
6768 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6772 fscrypt_free_filename(&fname);
6774 btrfs_end_transaction(trans);
6776 inode_dec_link_count(inode);
6779 btrfs_btree_balance_dirty(fs_info);
6783 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6784 struct dentry *dentry, umode_t mode)
6786 struct inode *inode;
6788 inode = new_inode(dir->i_sb);
6791 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6792 inode->i_op = &btrfs_dir_inode_operations;
6793 inode->i_fop = &btrfs_dir_file_operations;
6794 return btrfs_create_common(dir, dentry, inode);
6797 static noinline int uncompress_inline(struct btrfs_path *path,
6799 struct btrfs_file_extent_item *item)
6802 struct extent_buffer *leaf = path->nodes[0];
6805 unsigned long inline_size;
6809 compress_type = btrfs_file_extent_compression(leaf, item);
6810 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6811 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6812 tmp = kmalloc(inline_size, GFP_NOFS);
6815 ptr = btrfs_file_extent_inline_start(item);
6817 read_extent_buffer(leaf, tmp, ptr, inline_size);
6819 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6820 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6823 * decompression code contains a memset to fill in any space between the end
6824 * of the uncompressed data and the end of max_size in case the decompressed
6825 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6826 * the end of an inline extent and the beginning of the next block, so we
6827 * cover that region here.
6830 if (max_size < PAGE_SIZE)
6831 memzero_page(page, max_size, PAGE_SIZE - max_size);
6836 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6839 struct btrfs_file_extent_item *fi;
6843 if (!page || PageUptodate(page))
6846 ASSERT(page_offset(page) == 0);
6848 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6849 struct btrfs_file_extent_item);
6850 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6851 return uncompress_inline(path, page, fi);
6853 copy_size = min_t(u64, PAGE_SIZE,
6854 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6855 kaddr = kmap_local_page(page);
6856 read_extent_buffer(path->nodes[0], kaddr,
6857 btrfs_file_extent_inline_start(fi), copy_size);
6858 kunmap_local(kaddr);
6859 if (copy_size < PAGE_SIZE)
6860 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6865 * Lookup the first extent overlapping a range in a file.
6867 * @inode: file to search in
6868 * @page: page to read extent data into if the extent is inline
6869 * @pg_offset: offset into @page to copy to
6870 * @start: file offset
6871 * @len: length of range starting at @start
6873 * Return the first &struct extent_map which overlaps the given range, reading
6874 * it from the B-tree and caching it if necessary. Note that there may be more
6875 * extents which overlap the given range after the returned extent_map.
6877 * If @page is not NULL and the extent is inline, this also reads the extent
6878 * data directly into the page and marks the extent up to date in the io_tree.
6880 * Return: ERR_PTR on error, non-NULL extent_map on success.
6882 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6883 struct page *page, size_t pg_offset,
6886 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6888 u64 extent_start = 0;
6890 u64 objectid = btrfs_ino(inode);
6891 int extent_type = -1;
6892 struct btrfs_path *path = NULL;
6893 struct btrfs_root *root = inode->root;
6894 struct btrfs_file_extent_item *item;
6895 struct extent_buffer *leaf;
6896 struct btrfs_key found_key;
6897 struct extent_map *em = NULL;
6898 struct extent_map_tree *em_tree = &inode->extent_tree;
6900 read_lock(&em_tree->lock);
6901 em = lookup_extent_mapping(em_tree, start, len);
6902 read_unlock(&em_tree->lock);
6905 if (em->start > start || em->start + em->len <= start)
6906 free_extent_map(em);
6907 else if (em->block_start == EXTENT_MAP_INLINE && page)
6908 free_extent_map(em);
6912 em = alloc_extent_map();
6917 em->start = EXTENT_MAP_HOLE;
6918 em->orig_start = EXTENT_MAP_HOLE;
6920 em->block_len = (u64)-1;
6922 path = btrfs_alloc_path();
6928 /* Chances are we'll be called again, so go ahead and do readahead */
6929 path->reada = READA_FORWARD;
6932 * The same explanation in load_free_space_cache applies here as well,
6933 * we only read when we're loading the free space cache, and at that
6934 * point the commit_root has everything we need.
6936 if (btrfs_is_free_space_inode(inode)) {
6937 path->search_commit_root = 1;
6938 path->skip_locking = 1;
6941 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6944 } else if (ret > 0) {
6945 if (path->slots[0] == 0)
6951 leaf = path->nodes[0];
6952 item = btrfs_item_ptr(leaf, path->slots[0],
6953 struct btrfs_file_extent_item);
6954 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6955 if (found_key.objectid != objectid ||
6956 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6958 * If we backup past the first extent we want to move forward
6959 * and see if there is an extent in front of us, otherwise we'll
6960 * say there is a hole for our whole search range which can
6967 extent_type = btrfs_file_extent_type(leaf, item);
6968 extent_start = found_key.offset;
6969 extent_end = btrfs_file_extent_end(path);
6970 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6971 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6972 /* Only regular file could have regular/prealloc extent */
6973 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6976 "regular/prealloc extent found for non-regular inode %llu",
6980 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6982 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6983 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6988 if (start >= extent_end) {
6990 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6991 ret = btrfs_next_leaf(root, path);
6997 leaf = path->nodes[0];
6999 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7000 if (found_key.objectid != objectid ||
7001 found_key.type != BTRFS_EXTENT_DATA_KEY)
7003 if (start + len <= found_key.offset)
7005 if (start > found_key.offset)
7008 /* New extent overlaps with existing one */
7010 em->orig_start = start;
7011 em->len = found_key.offset - start;
7012 em->block_start = EXTENT_MAP_HOLE;
7016 btrfs_extent_item_to_extent_map(inode, path, item, em);
7018 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7019 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7021 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7023 * Inline extent can only exist at file offset 0. This is
7024 * ensured by tree-checker and inline extent creation path.
7025 * Thus all members representing file offsets should be zero.
7027 ASSERT(pg_offset == 0);
7028 ASSERT(extent_start == 0);
7029 ASSERT(em->start == 0);
7032 * btrfs_extent_item_to_extent_map() should have properly
7033 * initialized em members already.
7035 * Other members are not utilized for inline extents.
7037 ASSERT(em->block_start == EXTENT_MAP_INLINE);
7038 ASSERT(em->len == fs_info->sectorsize);
7040 ret = read_inline_extent(inode, path, page);
7047 em->orig_start = start;
7049 em->block_start = EXTENT_MAP_HOLE;
7052 btrfs_release_path(path);
7053 if (em->start > start || extent_map_end(em) <= start) {
7055 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7056 em->start, em->len, start, len);
7061 write_lock(&em_tree->lock);
7062 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7063 write_unlock(&em_tree->lock);
7065 btrfs_free_path(path);
7067 trace_btrfs_get_extent(root, inode, em);
7070 free_extent_map(em);
7071 return ERR_PTR(ret);
7076 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7077 struct btrfs_dio_data *dio_data,
7080 const u64 orig_start,
7081 const u64 block_start,
7082 const u64 block_len,
7083 const u64 orig_block_len,
7084 const u64 ram_bytes,
7087 struct extent_map *em = NULL;
7088 struct btrfs_ordered_extent *ordered;
7090 if (type != BTRFS_ORDERED_NOCOW) {
7091 em = create_io_em(inode, start, len, orig_start, block_start,
7092 block_len, orig_block_len, ram_bytes,
7093 BTRFS_COMPRESS_NONE, /* compress_type */
7098 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7099 block_start, block_len, 0,
7101 (1 << BTRFS_ORDERED_DIRECT),
7102 BTRFS_COMPRESS_NONE);
7103 if (IS_ERR(ordered)) {
7105 free_extent_map(em);
7106 btrfs_drop_extent_map_range(inode, start,
7107 start + len - 1, false);
7109 em = ERR_CAST(ordered);
7111 ASSERT(!dio_data->ordered);
7112 dio_data->ordered = ordered;
7119 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7120 struct btrfs_dio_data *dio_data,
7123 struct btrfs_root *root = inode->root;
7124 struct btrfs_fs_info *fs_info = root->fs_info;
7125 struct extent_map *em;
7126 struct btrfs_key ins;
7130 alloc_hint = get_extent_allocation_hint(inode, start, len);
7131 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7132 0, alloc_hint, &ins, 1, 1);
7134 return ERR_PTR(ret);
7136 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7137 ins.objectid, ins.offset, ins.offset,
7138 ins.offset, BTRFS_ORDERED_REGULAR);
7139 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7141 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7147 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7149 struct btrfs_block_group *block_group;
7150 bool readonly = false;
7152 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7153 if (!block_group || block_group->ro)
7156 btrfs_put_block_group(block_group);
7161 * Check if we can do nocow write into the range [@offset, @offset + @len)
7163 * @offset: File offset
7164 * @len: The length to write, will be updated to the nocow writeable
7166 * @orig_start: (optional) Return the original file offset of the file extent
7167 * @orig_len: (optional) Return the original on-disk length of the file extent
7168 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7169 * @strict: if true, omit optimizations that might force us into unnecessary
7170 * cow. e.g., don't trust generation number.
7173 * >0 and update @len if we can do nocow write
7174 * 0 if we can't do nocow write
7175 * <0 if error happened
7177 * NOTE: This only checks the file extents, caller is responsible to wait for
7178 * any ordered extents.
7180 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7181 u64 *orig_start, u64 *orig_block_len,
7182 u64 *ram_bytes, bool nowait, bool strict)
7184 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7185 struct can_nocow_file_extent_args nocow_args = { 0 };
7186 struct btrfs_path *path;
7188 struct extent_buffer *leaf;
7189 struct btrfs_root *root = BTRFS_I(inode)->root;
7190 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7191 struct btrfs_file_extent_item *fi;
7192 struct btrfs_key key;
7195 path = btrfs_alloc_path();
7198 path->nowait = nowait;
7200 ret = btrfs_lookup_file_extent(NULL, root, path,
7201 btrfs_ino(BTRFS_I(inode)), offset, 0);
7206 if (path->slots[0] == 0) {
7207 /* can't find the item, must cow */
7214 leaf = path->nodes[0];
7215 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7216 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7217 key.type != BTRFS_EXTENT_DATA_KEY) {
7218 /* not our file or wrong item type, must cow */
7222 if (key.offset > offset) {
7223 /* Wrong offset, must cow */
7227 if (btrfs_file_extent_end(path) <= offset)
7230 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7231 found_type = btrfs_file_extent_type(leaf, fi);
7233 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7235 nocow_args.start = offset;
7236 nocow_args.end = offset + *len - 1;
7237 nocow_args.strict = strict;
7238 nocow_args.free_path = true;
7240 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7241 /* can_nocow_file_extent() has freed the path. */
7245 /* Treat errors as not being able to NOCOW. */
7251 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7254 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7255 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7258 range_end = round_up(offset + nocow_args.num_bytes,
7259 root->fs_info->sectorsize) - 1;
7260 ret = test_range_bit(io_tree, offset, range_end,
7261 EXTENT_DELALLOC, 0, NULL);
7269 *orig_start = key.offset - nocow_args.extent_offset;
7271 *orig_block_len = nocow_args.disk_num_bytes;
7273 *len = nocow_args.num_bytes;
7276 btrfs_free_path(path);
7280 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7281 struct extent_state **cached_state,
7282 unsigned int iomap_flags)
7284 const bool writing = (iomap_flags & IOMAP_WRITE);
7285 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7286 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7287 struct btrfs_ordered_extent *ordered;
7292 if (!try_lock_extent(io_tree, lockstart, lockend,
7296 lock_extent(io_tree, lockstart, lockend, cached_state);
7299 * We're concerned with the entire range that we're going to be
7300 * doing DIO to, so we need to make sure there's no ordered
7301 * extents in this range.
7303 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7304 lockend - lockstart + 1);
7307 * We need to make sure there are no buffered pages in this
7308 * range either, we could have raced between the invalidate in
7309 * generic_file_direct_write and locking the extent. The
7310 * invalidate needs to happen so that reads after a write do not
7314 (!writing || !filemap_range_has_page(inode->i_mapping,
7315 lockstart, lockend)))
7318 unlock_extent(io_tree, lockstart, lockend, cached_state);
7322 btrfs_put_ordered_extent(ordered);
7327 * If we are doing a DIO read and the ordered extent we
7328 * found is for a buffered write, we can not wait for it
7329 * to complete and retry, because if we do so we can
7330 * deadlock with concurrent buffered writes on page
7331 * locks. This happens only if our DIO read covers more
7332 * than one extent map, if at this point has already
7333 * created an ordered extent for a previous extent map
7334 * and locked its range in the inode's io tree, and a
7335 * concurrent write against that previous extent map's
7336 * range and this range started (we unlock the ranges
7337 * in the io tree only when the bios complete and
7338 * buffered writes always lock pages before attempting
7339 * to lock range in the io tree).
7342 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7343 btrfs_start_ordered_extent(ordered);
7345 ret = nowait ? -EAGAIN : -ENOTBLK;
7346 btrfs_put_ordered_extent(ordered);
7349 * We could trigger writeback for this range (and wait
7350 * for it to complete) and then invalidate the pages for
7351 * this range (through invalidate_inode_pages2_range()),
7352 * but that can lead us to a deadlock with a concurrent
7353 * call to readahead (a buffered read or a defrag call
7354 * triggered a readahead) on a page lock due to an
7355 * ordered dio extent we created before but did not have
7356 * yet a corresponding bio submitted (whence it can not
7357 * complete), which makes readahead wait for that
7358 * ordered extent to complete while holding a lock on
7361 ret = nowait ? -EAGAIN : -ENOTBLK;
7373 /* The callers of this must take lock_extent() */
7374 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7375 u64 len, u64 orig_start, u64 block_start,
7376 u64 block_len, u64 orig_block_len,
7377 u64 ram_bytes, int compress_type,
7380 struct extent_map *em;
7383 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7384 type == BTRFS_ORDERED_COMPRESSED ||
7385 type == BTRFS_ORDERED_NOCOW ||
7386 type == BTRFS_ORDERED_REGULAR);
7388 em = alloc_extent_map();
7390 return ERR_PTR(-ENOMEM);
7393 em->orig_start = orig_start;
7395 em->block_len = block_len;
7396 em->block_start = block_start;
7397 em->orig_block_len = orig_block_len;
7398 em->ram_bytes = ram_bytes;
7399 em->generation = -1;
7400 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7401 if (type == BTRFS_ORDERED_PREALLOC) {
7402 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7403 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7404 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7405 em->compress_type = compress_type;
7408 ret = btrfs_replace_extent_map_range(inode, em, true);
7410 free_extent_map(em);
7411 return ERR_PTR(ret);
7414 /* em got 2 refs now, callers needs to do free_extent_map once. */
7419 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7420 struct inode *inode,
7421 struct btrfs_dio_data *dio_data,
7422 u64 start, u64 *lenp,
7423 unsigned int iomap_flags)
7425 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7426 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7427 struct extent_map *em = *map;
7429 u64 block_start, orig_start, orig_block_len, ram_bytes;
7430 struct btrfs_block_group *bg;
7431 bool can_nocow = false;
7432 bool space_reserved = false;
7438 * We don't allocate a new extent in the following cases
7440 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7442 * 2) The extent is marked as PREALLOC. We're good to go here and can
7443 * just use the extent.
7446 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7447 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7448 em->block_start != EXTENT_MAP_HOLE)) {
7449 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7450 type = BTRFS_ORDERED_PREALLOC;
7452 type = BTRFS_ORDERED_NOCOW;
7453 len = min(len, em->len - (start - em->start));
7454 block_start = em->block_start + (start - em->start);
7456 if (can_nocow_extent(inode, start, &len, &orig_start,
7457 &orig_block_len, &ram_bytes, false, false) == 1) {
7458 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7466 struct extent_map *em2;
7468 /* We can NOCOW, so only need to reserve metadata space. */
7469 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7472 /* Our caller expects us to free the input extent map. */
7473 free_extent_map(em);
7475 btrfs_dec_nocow_writers(bg);
7476 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7480 space_reserved = true;
7482 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7483 orig_start, block_start,
7484 len, orig_block_len,
7486 btrfs_dec_nocow_writers(bg);
7487 if (type == BTRFS_ORDERED_PREALLOC) {
7488 free_extent_map(em);
7498 dio_data->nocow_done = true;
7500 /* Our caller expects us to free the input extent map. */
7501 free_extent_map(em);
7510 * If we could not allocate data space before locking the file
7511 * range and we can't do a NOCOW write, then we have to fail.
7513 if (!dio_data->data_space_reserved) {
7519 * We have to COW and we have already reserved data space before,
7520 * so now we reserve only metadata.
7522 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7526 space_reserved = true;
7528 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7534 len = min(len, em->len - (start - em->start));
7536 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7537 prev_len - len, true);
7541 * We have created our ordered extent, so we can now release our reservation
7542 * for an outstanding extent.
7544 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7547 * Need to update the i_size under the extent lock so buffered
7548 * readers will get the updated i_size when we unlock.
7550 if (start + len > i_size_read(inode))
7551 i_size_write(inode, start + len);
7553 if (ret && space_reserved) {
7554 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7555 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7561 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7562 loff_t length, unsigned int flags, struct iomap *iomap,
7563 struct iomap *srcmap)
7565 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7566 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7567 struct extent_map *em;
7568 struct extent_state *cached_state = NULL;
7569 struct btrfs_dio_data *dio_data = iter->private;
7570 u64 lockstart, lockend;
7571 const bool write = !!(flags & IOMAP_WRITE);
7574 const u64 data_alloc_len = length;
7575 bool unlock_extents = false;
7578 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7579 * we're NOWAIT we may submit a bio for a partial range and return
7580 * EIOCBQUEUED, which would result in an errant short read.
7582 * The best way to handle this would be to allow for partial completions
7583 * of iocb's, so we could submit the partial bio, return and fault in
7584 * the rest of the pages, and then submit the io for the rest of the
7585 * range. However we don't have that currently, so simply return
7586 * -EAGAIN at this point so that the normal path is used.
7588 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7592 * Cap the size of reads to that usually seen in buffered I/O as we need
7593 * to allocate a contiguous array for the checksums.
7596 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7599 lockend = start + len - 1;
7602 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7603 * enough if we've written compressed pages to this area, so we need to
7604 * flush the dirty pages again to make absolutely sure that any
7605 * outstanding dirty pages are on disk - the first flush only starts
7606 * compression on the data, while keeping the pages locked, so by the
7607 * time the second flush returns we know bios for the compressed pages
7608 * were submitted and finished, and the pages no longer under writeback.
7610 * If we have a NOWAIT request and we have any pages in the range that
7611 * are locked, likely due to compression still in progress, we don't want
7612 * to block on page locks. We also don't want to block on pages marked as
7613 * dirty or under writeback (same as for the non-compression case).
7614 * iomap_dio_rw() did the same check, but after that and before we got
7615 * here, mmap'ed writes may have happened or buffered reads started
7616 * (readpage() and readahead(), which lock pages), as we haven't locked
7617 * the file range yet.
7619 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7620 &BTRFS_I(inode)->runtime_flags)) {
7621 if (flags & IOMAP_NOWAIT) {
7622 if (filemap_range_needs_writeback(inode->i_mapping,
7623 lockstart, lockend))
7626 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7627 start + length - 1);
7633 memset(dio_data, 0, sizeof(*dio_data));
7636 * We always try to allocate data space and must do it before locking
7637 * the file range, to avoid deadlocks with concurrent writes to the same
7638 * range if the range has several extents and the writes don't expand the
7639 * current i_size (the inode lock is taken in shared mode). If we fail to
7640 * allocate data space here we continue and later, after locking the
7641 * file range, we fail with ENOSPC only if we figure out we can not do a
7644 if (write && !(flags & IOMAP_NOWAIT)) {
7645 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7646 &dio_data->data_reserved,
7647 start, data_alloc_len, false);
7649 dio_data->data_space_reserved = true;
7650 else if (ret && !(BTRFS_I(inode)->flags &
7651 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7656 * If this errors out it's because we couldn't invalidate pagecache for
7657 * this range and we need to fallback to buffered IO, or we are doing a
7658 * NOWAIT read/write and we need to block.
7660 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7664 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7671 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7672 * io. INLINE is special, and we could probably kludge it in here, but
7673 * it's still buffered so for safety lets just fall back to the generic
7676 * For COMPRESSED we _have_ to read the entire extent in so we can
7677 * decompress it, so there will be buffering required no matter what we
7678 * do, so go ahead and fallback to buffered.
7680 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7681 * to buffered IO. Don't blame me, this is the price we pay for using
7684 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7685 em->block_start == EXTENT_MAP_INLINE) {
7686 free_extent_map(em);
7688 * If we are in a NOWAIT context, return -EAGAIN in order to
7689 * fallback to buffered IO. This is not only because we can
7690 * block with buffered IO (no support for NOWAIT semantics at
7691 * the moment) but also to avoid returning short reads to user
7692 * space - this happens if we were able to read some data from
7693 * previous non-compressed extents and then when we fallback to
7694 * buffered IO, at btrfs_file_read_iter() by calling
7695 * filemap_read(), we fail to fault in pages for the read buffer,
7696 * in which case filemap_read() returns a short read (the number
7697 * of bytes previously read is > 0, so it does not return -EFAULT).
7699 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7703 len = min(len, em->len - (start - em->start));
7706 * If we have a NOWAIT request and the range contains multiple extents
7707 * (or a mix of extents and holes), then we return -EAGAIN to make the
7708 * caller fallback to a context where it can do a blocking (without
7709 * NOWAIT) request. This way we avoid doing partial IO and returning
7710 * success to the caller, which is not optimal for writes and for reads
7711 * it can result in unexpected behaviour for an application.
7713 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7714 * iomap_dio_rw(), we can end up returning less data then what the caller
7715 * asked for, resulting in an unexpected, and incorrect, short read.
7716 * That is, the caller asked to read N bytes and we return less than that,
7717 * which is wrong unless we are crossing EOF. This happens if we get a
7718 * page fault error when trying to fault in pages for the buffer that is
7719 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7720 * have previously submitted bios for other extents in the range, in
7721 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7722 * those bios have completed by the time we get the page fault error,
7723 * which we return back to our caller - we should only return EIOCBQUEUED
7724 * after we have submitted bios for all the extents in the range.
7726 if ((flags & IOMAP_NOWAIT) && len < length) {
7727 free_extent_map(em);
7733 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7734 start, &len, flags);
7737 unlock_extents = true;
7738 /* Recalc len in case the new em is smaller than requested */
7739 len = min(len, em->len - (start - em->start));
7740 if (dio_data->data_space_reserved) {
7742 u64 release_len = 0;
7744 if (dio_data->nocow_done) {
7745 release_offset = start;
7746 release_len = data_alloc_len;
7747 } else if (len < data_alloc_len) {
7748 release_offset = start + len;
7749 release_len = data_alloc_len - len;
7752 if (release_len > 0)
7753 btrfs_free_reserved_data_space(BTRFS_I(inode),
7754 dio_data->data_reserved,
7760 * We need to unlock only the end area that we aren't using.
7761 * The rest is going to be unlocked by the endio routine.
7763 lockstart = start + len;
7764 if (lockstart < lockend)
7765 unlock_extents = true;
7769 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7772 free_extent_state(cached_state);
7775 * Translate extent map information to iomap.
7776 * We trim the extents (and move the addr) even though iomap code does
7777 * that, since we have locked only the parts we are performing I/O in.
7779 if ((em->block_start == EXTENT_MAP_HOLE) ||
7780 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7781 iomap->addr = IOMAP_NULL_ADDR;
7782 iomap->type = IOMAP_HOLE;
7784 iomap->addr = em->block_start + (start - em->start);
7785 iomap->type = IOMAP_MAPPED;
7787 iomap->offset = start;
7788 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7789 iomap->length = len;
7790 free_extent_map(em);
7795 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7798 if (dio_data->data_space_reserved) {
7799 btrfs_free_reserved_data_space(BTRFS_I(inode),
7800 dio_data->data_reserved,
7801 start, data_alloc_len);
7802 extent_changeset_free(dio_data->data_reserved);
7808 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7809 ssize_t written, unsigned int flags, struct iomap *iomap)
7811 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7812 struct btrfs_dio_data *dio_data = iter->private;
7813 size_t submitted = dio_data->submitted;
7814 const bool write = !!(flags & IOMAP_WRITE);
7817 if (!write && (iomap->type == IOMAP_HOLE)) {
7818 /* If reading from a hole, unlock and return */
7819 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7824 if (submitted < length) {
7826 length -= submitted;
7828 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7829 pos, length, false);
7831 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7832 pos + length - 1, NULL);
7836 btrfs_put_ordered_extent(dio_data->ordered);
7837 dio_data->ordered = NULL;
7841 extent_changeset_free(dio_data->data_reserved);
7845 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7847 struct btrfs_dio_private *dip =
7848 container_of(bbio, struct btrfs_dio_private, bbio);
7849 struct btrfs_inode *inode = bbio->inode;
7850 struct bio *bio = &bbio->bio;
7852 if (bio->bi_status) {
7853 btrfs_warn(inode->root->fs_info,
7854 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7855 btrfs_ino(inode), bio->bi_opf,
7856 dip->file_offset, dip->bytes, bio->bi_status);
7859 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7860 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7861 dip->file_offset, dip->bytes,
7864 unlock_extent(&inode->io_tree, dip->file_offset,
7865 dip->file_offset + dip->bytes - 1, NULL);
7868 bbio->bio.bi_private = bbio->private;
7869 iomap_dio_bio_end_io(bio);
7872 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7875 struct btrfs_bio *bbio = btrfs_bio(bio);
7876 struct btrfs_dio_private *dip =
7877 container_of(bbio, struct btrfs_dio_private, bbio);
7878 struct btrfs_dio_data *dio_data = iter->private;
7880 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7881 btrfs_dio_end_io, bio->bi_private);
7882 bbio->inode = BTRFS_I(iter->inode);
7883 bbio->file_offset = file_offset;
7885 dip->file_offset = file_offset;
7886 dip->bytes = bio->bi_iter.bi_size;
7888 dio_data->submitted += bio->bi_iter.bi_size;
7891 * Check if we are doing a partial write. If we are, we need to split
7892 * the ordered extent to match the submitted bio. Hang on to the
7893 * remaining unfinishable ordered_extent in dio_data so that it can be
7894 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7895 * remaining pages is blocked on the outstanding ordered extent.
7897 if (iter->flags & IOMAP_WRITE) {
7900 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7902 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7903 file_offset, dip->bytes,
7905 bio->bi_status = errno_to_blk_status(ret);
7906 iomap_dio_bio_end_io(bio);
7911 btrfs_submit_bio(bbio, 0);
7914 static const struct iomap_ops btrfs_dio_iomap_ops = {
7915 .iomap_begin = btrfs_dio_iomap_begin,
7916 .iomap_end = btrfs_dio_iomap_end,
7919 static const struct iomap_dio_ops btrfs_dio_ops = {
7920 .submit_io = btrfs_dio_submit_io,
7921 .bio_set = &btrfs_dio_bioset,
7924 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7926 struct btrfs_dio_data data = { 0 };
7928 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7929 IOMAP_DIO_PARTIAL, &data, done_before);
7932 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7935 struct btrfs_dio_data data = { 0 };
7937 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7938 IOMAP_DIO_PARTIAL, &data, done_before);
7941 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7946 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7951 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7952 * file range (0 to LLONG_MAX), but that is not enough if we have
7953 * compression enabled. The first filemap_fdatawrite_range() only kicks
7954 * in the compression of data (in an async thread) and will return
7955 * before the compression is done and writeback is started. A second
7956 * filemap_fdatawrite_range() is needed to wait for the compression to
7957 * complete and writeback to start. We also need to wait for ordered
7958 * extents to complete, because our fiemap implementation uses mainly
7959 * file extent items to list the extents, searching for extent maps
7960 * only for file ranges with holes or prealloc extents to figure out
7961 * if we have delalloc in those ranges.
7963 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7964 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7969 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7972 static int btrfs_writepages(struct address_space *mapping,
7973 struct writeback_control *wbc)
7975 return extent_writepages(mapping, wbc);
7978 static void btrfs_readahead(struct readahead_control *rac)
7980 extent_readahead(rac);
7984 * For release_folio() and invalidate_folio() we have a race window where
7985 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7986 * If we continue to release/invalidate the page, we could cause use-after-free
7987 * for subpage spinlock. So this function is to spin and wait for subpage
7990 static void wait_subpage_spinlock(struct page *page)
7992 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7993 struct btrfs_subpage *subpage;
7995 if (!btrfs_is_subpage(fs_info, page))
7998 ASSERT(PagePrivate(page) && page->private);
7999 subpage = (struct btrfs_subpage *)page->private;
8002 * This may look insane as we just acquire the spinlock and release it,
8003 * without doing anything. But we just want to make sure no one is
8004 * still holding the subpage spinlock.
8005 * And since the page is not dirty nor writeback, and we have page
8006 * locked, the only possible way to hold a spinlock is from the endio
8007 * function to clear page writeback.
8009 * Here we just acquire the spinlock so that all existing callers
8010 * should exit and we're safe to release/invalidate the page.
8012 spin_lock_irq(&subpage->lock);
8013 spin_unlock_irq(&subpage->lock);
8016 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8018 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8021 wait_subpage_spinlock(&folio->page);
8022 clear_page_extent_mapped(&folio->page);
8027 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8029 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8031 return __btrfs_release_folio(folio, gfp_flags);
8034 #ifdef CONFIG_MIGRATION
8035 static int btrfs_migrate_folio(struct address_space *mapping,
8036 struct folio *dst, struct folio *src,
8037 enum migrate_mode mode)
8039 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8041 if (ret != MIGRATEPAGE_SUCCESS)
8044 if (folio_test_ordered(src)) {
8045 folio_clear_ordered(src);
8046 folio_set_ordered(dst);
8049 return MIGRATEPAGE_SUCCESS;
8052 #define btrfs_migrate_folio NULL
8055 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8058 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8059 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8060 struct extent_io_tree *tree = &inode->io_tree;
8061 struct extent_state *cached_state = NULL;
8062 u64 page_start = folio_pos(folio);
8063 u64 page_end = page_start + folio_size(folio) - 1;
8065 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8068 * We have folio locked so no new ordered extent can be created on this
8069 * page, nor bio can be submitted for this folio.
8071 * But already submitted bio can still be finished on this folio.
8072 * Furthermore, endio function won't skip folio which has Ordered
8073 * (Private2) already cleared, so it's possible for endio and
8074 * invalidate_folio to do the same ordered extent accounting twice
8077 * So here we wait for any submitted bios to finish, so that we won't
8078 * do double ordered extent accounting on the same folio.
8080 folio_wait_writeback(folio);
8081 wait_subpage_spinlock(&folio->page);
8084 * For subpage case, we have call sites like
8085 * btrfs_punch_hole_lock_range() which passes range not aligned to
8087 * If the range doesn't cover the full folio, we don't need to and
8088 * shouldn't clear page extent mapped, as folio->private can still
8089 * record subpage dirty bits for other part of the range.
8091 * For cases that invalidate the full folio even the range doesn't
8092 * cover the full folio, like invalidating the last folio, we're
8093 * still safe to wait for ordered extent to finish.
8095 if (!(offset == 0 && length == folio_size(folio))) {
8096 btrfs_release_folio(folio, GFP_NOFS);
8100 if (!inode_evicting)
8101 lock_extent(tree, page_start, page_end, &cached_state);
8104 while (cur < page_end) {
8105 struct btrfs_ordered_extent *ordered;
8108 u32 extra_flags = 0;
8110 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8111 page_end + 1 - cur);
8113 range_end = page_end;
8115 * No ordered extent covering this range, we are safe
8116 * to delete all extent states in the range.
8118 extra_flags = EXTENT_CLEAR_ALL_BITS;
8121 if (ordered->file_offset > cur) {
8123 * There is a range between [cur, oe->file_offset) not
8124 * covered by any ordered extent.
8125 * We are safe to delete all extent states, and handle
8126 * the ordered extent in the next iteration.
8128 range_end = ordered->file_offset - 1;
8129 extra_flags = EXTENT_CLEAR_ALL_BITS;
8133 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8135 ASSERT(range_end + 1 - cur < U32_MAX);
8136 range_len = range_end + 1 - cur;
8137 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8139 * If Ordered (Private2) is cleared, it means endio has
8140 * already been executed for the range.
8141 * We can't delete the extent states as
8142 * btrfs_finish_ordered_io() may still use some of them.
8146 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8149 * IO on this page will never be started, so we need to account
8150 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8151 * here, must leave that up for the ordered extent completion.
8153 * This will also unlock the range for incoming
8154 * btrfs_finish_ordered_io().
8156 if (!inode_evicting)
8157 clear_extent_bit(tree, cur, range_end,
8159 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8160 EXTENT_DEFRAG, &cached_state);
8162 spin_lock_irq(&inode->ordered_tree.lock);
8163 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8164 ordered->truncated_len = min(ordered->truncated_len,
8165 cur - ordered->file_offset);
8166 spin_unlock_irq(&inode->ordered_tree.lock);
8169 * If the ordered extent has finished, we're safe to delete all
8170 * the extent states of the range, otherwise
8171 * btrfs_finish_ordered_io() will get executed by endio for
8172 * other pages, so we can't delete extent states.
8174 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8175 cur, range_end + 1 - cur)) {
8176 btrfs_finish_ordered_io(ordered);
8178 * The ordered extent has finished, now we're again
8179 * safe to delete all extent states of the range.
8181 extra_flags = EXTENT_CLEAR_ALL_BITS;
8185 btrfs_put_ordered_extent(ordered);
8187 * Qgroup reserved space handler
8188 * Sector(s) here will be either:
8190 * 1) Already written to disk or bio already finished
8191 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8192 * Qgroup will be handled by its qgroup_record then.
8193 * btrfs_qgroup_free_data() call will do nothing here.
8195 * 2) Not written to disk yet
8196 * Then btrfs_qgroup_free_data() call will clear the
8197 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8198 * reserved data space.
8199 * Since the IO will never happen for this page.
8201 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8202 if (!inode_evicting) {
8203 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8204 EXTENT_DELALLOC | EXTENT_UPTODATE |
8205 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8206 extra_flags, &cached_state);
8208 cur = range_end + 1;
8211 * We have iterated through all ordered extents of the page, the page
8212 * should not have Ordered (Private2) anymore, or the above iteration
8213 * did something wrong.
8215 ASSERT(!folio_test_ordered(folio));
8216 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8217 if (!inode_evicting)
8218 __btrfs_release_folio(folio, GFP_NOFS);
8219 clear_page_extent_mapped(&folio->page);
8223 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8224 * called from a page fault handler when a page is first dirtied. Hence we must
8225 * be careful to check for EOF conditions here. We set the page up correctly
8226 * for a written page which means we get ENOSPC checking when writing into
8227 * holes and correct delalloc and unwritten extent mapping on filesystems that
8228 * support these features.
8230 * We are not allowed to take the i_mutex here so we have to play games to
8231 * protect against truncate races as the page could now be beyond EOF. Because
8232 * truncate_setsize() writes the inode size before removing pages, once we have
8233 * the page lock we can determine safely if the page is beyond EOF. If it is not
8234 * beyond EOF, then the page is guaranteed safe against truncation until we
8237 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8239 struct page *page = vmf->page;
8240 struct inode *inode = file_inode(vmf->vma->vm_file);
8241 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8242 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8243 struct btrfs_ordered_extent *ordered;
8244 struct extent_state *cached_state = NULL;
8245 struct extent_changeset *data_reserved = NULL;
8246 unsigned long zero_start;
8256 reserved_space = PAGE_SIZE;
8258 sb_start_pagefault(inode->i_sb);
8259 page_start = page_offset(page);
8260 page_end = page_start + PAGE_SIZE - 1;
8264 * Reserving delalloc space after obtaining the page lock can lead to
8265 * deadlock. For example, if a dirty page is locked by this function
8266 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8267 * dirty page write out, then the btrfs_writepages() function could
8268 * end up waiting indefinitely to get a lock on the page currently
8269 * being processed by btrfs_page_mkwrite() function.
8271 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8272 page_start, reserved_space);
8274 ret2 = file_update_time(vmf->vma->vm_file);
8278 ret = vmf_error(ret2);
8284 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8286 down_read(&BTRFS_I(inode)->i_mmap_lock);
8288 size = i_size_read(inode);
8290 if ((page->mapping != inode->i_mapping) ||
8291 (page_start >= size)) {
8292 /* page got truncated out from underneath us */
8295 wait_on_page_writeback(page);
8297 lock_extent(io_tree, page_start, page_end, &cached_state);
8298 ret2 = set_page_extent_mapped(page);
8300 ret = vmf_error(ret2);
8301 unlock_extent(io_tree, page_start, page_end, &cached_state);
8306 * we can't set the delalloc bits if there are pending ordered
8307 * extents. Drop our locks and wait for them to finish
8309 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8312 unlock_extent(io_tree, page_start, page_end, &cached_state);
8314 up_read(&BTRFS_I(inode)->i_mmap_lock);
8315 btrfs_start_ordered_extent(ordered);
8316 btrfs_put_ordered_extent(ordered);
8320 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8321 reserved_space = round_up(size - page_start,
8322 fs_info->sectorsize);
8323 if (reserved_space < PAGE_SIZE) {
8324 end = page_start + reserved_space - 1;
8325 btrfs_delalloc_release_space(BTRFS_I(inode),
8326 data_reserved, page_start,
8327 PAGE_SIZE - reserved_space, true);
8332 * page_mkwrite gets called when the page is firstly dirtied after it's
8333 * faulted in, but write(2) could also dirty a page and set delalloc
8334 * bits, thus in this case for space account reason, we still need to
8335 * clear any delalloc bits within this page range since we have to
8336 * reserve data&meta space before lock_page() (see above comments).
8338 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8339 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8340 EXTENT_DEFRAG, &cached_state);
8342 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8345 unlock_extent(io_tree, page_start, page_end, &cached_state);
8346 ret = VM_FAULT_SIGBUS;
8350 /* page is wholly or partially inside EOF */
8351 if (page_start + PAGE_SIZE > size)
8352 zero_start = offset_in_page(size);
8354 zero_start = PAGE_SIZE;
8356 if (zero_start != PAGE_SIZE)
8357 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8359 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8360 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8361 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8363 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8365 unlock_extent(io_tree, page_start, page_end, &cached_state);
8366 up_read(&BTRFS_I(inode)->i_mmap_lock);
8368 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8369 sb_end_pagefault(inode->i_sb);
8370 extent_changeset_free(data_reserved);
8371 return VM_FAULT_LOCKED;
8375 up_read(&BTRFS_I(inode)->i_mmap_lock);
8377 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8378 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8379 reserved_space, (ret != 0));
8381 sb_end_pagefault(inode->i_sb);
8382 extent_changeset_free(data_reserved);
8386 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8388 struct btrfs_truncate_control control = {
8390 .ino = btrfs_ino(inode),
8391 .min_type = BTRFS_EXTENT_DATA_KEY,
8392 .clear_extent_range = true,
8394 struct btrfs_root *root = inode->root;
8395 struct btrfs_fs_info *fs_info = root->fs_info;
8396 struct btrfs_block_rsv *rsv;
8398 struct btrfs_trans_handle *trans;
8399 u64 mask = fs_info->sectorsize - 1;
8400 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8402 if (!skip_writeback) {
8403 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8404 inode->vfs_inode.i_size & (~mask),
8411 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8412 * things going on here:
8414 * 1) We need to reserve space to update our inode.
8416 * 2) We need to have something to cache all the space that is going to
8417 * be free'd up by the truncate operation, but also have some slack
8418 * space reserved in case it uses space during the truncate (thank you
8419 * very much snapshotting).
8421 * And we need these to be separate. The fact is we can use a lot of
8422 * space doing the truncate, and we have no earthly idea how much space
8423 * we will use, so we need the truncate reservation to be separate so it
8424 * doesn't end up using space reserved for updating the inode. We also
8425 * need to be able to stop the transaction and start a new one, which
8426 * means we need to be able to update the inode several times, and we
8427 * have no idea of knowing how many times that will be, so we can't just
8428 * reserve 1 item for the entirety of the operation, so that has to be
8429 * done separately as well.
8431 * So that leaves us with
8433 * 1) rsv - for the truncate reservation, which we will steal from the
8434 * transaction reservation.
8435 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8436 * updating the inode.
8438 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8441 rsv->size = min_size;
8442 rsv->failfast = true;
8445 * 1 for the truncate slack space
8446 * 1 for updating the inode.
8448 trans = btrfs_start_transaction(root, 2);
8449 if (IS_ERR(trans)) {
8450 ret = PTR_ERR(trans);
8454 /* Migrate the slack space for the truncate to our reserve */
8455 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8458 * We have reserved 2 metadata units when we started the transaction and
8459 * min_size matches 1 unit, so this should never fail, but if it does,
8460 * it's not critical we just fail truncation.
8463 btrfs_end_transaction(trans);
8467 trans->block_rsv = rsv;
8470 struct extent_state *cached_state = NULL;
8471 const u64 new_size = inode->vfs_inode.i_size;
8472 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8474 control.new_size = new_size;
8475 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8477 * We want to drop from the next block forward in case this new
8478 * size is not block aligned since we will be keeping the last
8479 * block of the extent just the way it is.
8481 btrfs_drop_extent_map_range(inode,
8482 ALIGN(new_size, fs_info->sectorsize),
8485 ret = btrfs_truncate_inode_items(trans, root, &control);
8487 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8488 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8490 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8492 trans->block_rsv = &fs_info->trans_block_rsv;
8493 if (ret != -ENOSPC && ret != -EAGAIN)
8496 ret = btrfs_update_inode(trans, root, inode);
8500 btrfs_end_transaction(trans);
8501 btrfs_btree_balance_dirty(fs_info);
8503 trans = btrfs_start_transaction(root, 2);
8504 if (IS_ERR(trans)) {
8505 ret = PTR_ERR(trans);
8510 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8511 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8512 rsv, min_size, false);
8514 * We have reserved 2 metadata units when we started the
8515 * transaction and min_size matches 1 unit, so this should never
8516 * fail, but if it does, it's not critical we just fail truncation.
8521 trans->block_rsv = rsv;
8525 * We can't call btrfs_truncate_block inside a trans handle as we could
8526 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8527 * know we've truncated everything except the last little bit, and can
8528 * do btrfs_truncate_block and then update the disk_i_size.
8530 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8531 btrfs_end_transaction(trans);
8532 btrfs_btree_balance_dirty(fs_info);
8534 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8537 trans = btrfs_start_transaction(root, 1);
8538 if (IS_ERR(trans)) {
8539 ret = PTR_ERR(trans);
8542 btrfs_inode_safe_disk_i_size_write(inode, 0);
8548 trans->block_rsv = &fs_info->trans_block_rsv;
8549 ret2 = btrfs_update_inode(trans, root, inode);
8553 ret2 = btrfs_end_transaction(trans);
8556 btrfs_btree_balance_dirty(fs_info);
8559 btrfs_free_block_rsv(fs_info, rsv);
8561 * So if we truncate and then write and fsync we normally would just
8562 * write the extents that changed, which is a problem if we need to
8563 * first truncate that entire inode. So set this flag so we write out
8564 * all of the extents in the inode to the sync log so we're completely
8567 * If no extents were dropped or trimmed we don't need to force the next
8568 * fsync to truncate all the inode's items from the log and re-log them
8569 * all. This means the truncate operation did not change the file size,
8570 * or changed it to a smaller size but there was only an implicit hole
8571 * between the old i_size and the new i_size, and there were no prealloc
8572 * extents beyond i_size to drop.
8574 if (control.extents_found > 0)
8575 btrfs_set_inode_full_sync(inode);
8580 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8583 struct inode *inode;
8585 inode = new_inode(dir->i_sb);
8588 * Subvolumes don't inherit the sgid bit or the parent's gid if
8589 * the parent's sgid bit is set. This is probably a bug.
8591 inode_init_owner(idmap, inode, NULL,
8592 S_IFDIR | (~current_umask() & S_IRWXUGO));
8593 inode->i_op = &btrfs_dir_inode_operations;
8594 inode->i_fop = &btrfs_dir_file_operations;
8599 struct inode *btrfs_alloc_inode(struct super_block *sb)
8601 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8602 struct btrfs_inode *ei;
8603 struct inode *inode;
8605 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8612 ei->last_sub_trans = 0;
8613 ei->logged_trans = 0;
8614 ei->delalloc_bytes = 0;
8615 ei->new_delalloc_bytes = 0;
8616 ei->defrag_bytes = 0;
8617 ei->disk_i_size = 0;
8621 ei->index_cnt = (u64)-1;
8623 ei->last_unlink_trans = 0;
8624 ei->last_reflink_trans = 0;
8625 ei->last_log_commit = 0;
8627 spin_lock_init(&ei->lock);
8628 ei->outstanding_extents = 0;
8629 if (sb->s_magic != BTRFS_TEST_MAGIC)
8630 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8631 BTRFS_BLOCK_RSV_DELALLOC);
8632 ei->runtime_flags = 0;
8633 ei->prop_compress = BTRFS_COMPRESS_NONE;
8634 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8636 ei->delayed_node = NULL;
8638 ei->i_otime.tv_sec = 0;
8639 ei->i_otime.tv_nsec = 0;
8641 inode = &ei->vfs_inode;
8642 extent_map_tree_init(&ei->extent_tree);
8643 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8644 ei->io_tree.inode = ei;
8645 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8646 IO_TREE_INODE_FILE_EXTENT);
8647 mutex_init(&ei->log_mutex);
8648 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8649 INIT_LIST_HEAD(&ei->delalloc_inodes);
8650 INIT_LIST_HEAD(&ei->delayed_iput);
8651 RB_CLEAR_NODE(&ei->rb_node);
8652 init_rwsem(&ei->i_mmap_lock);
8657 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8658 void btrfs_test_destroy_inode(struct inode *inode)
8660 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8661 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8665 void btrfs_free_inode(struct inode *inode)
8667 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8670 void btrfs_destroy_inode(struct inode *vfs_inode)
8672 struct btrfs_ordered_extent *ordered;
8673 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8674 struct btrfs_root *root = inode->root;
8675 bool freespace_inode;
8677 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8678 WARN_ON(vfs_inode->i_data.nrpages);
8679 WARN_ON(inode->block_rsv.reserved);
8680 WARN_ON(inode->block_rsv.size);
8681 WARN_ON(inode->outstanding_extents);
8682 if (!S_ISDIR(vfs_inode->i_mode)) {
8683 WARN_ON(inode->delalloc_bytes);
8684 WARN_ON(inode->new_delalloc_bytes);
8686 WARN_ON(inode->csum_bytes);
8687 WARN_ON(inode->defrag_bytes);
8690 * This can happen where we create an inode, but somebody else also
8691 * created the same inode and we need to destroy the one we already
8698 * If this is a free space inode do not take the ordered extents lockdep
8701 freespace_inode = btrfs_is_free_space_inode(inode);
8704 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8708 btrfs_err(root->fs_info,
8709 "found ordered extent %llu %llu on inode cleanup",
8710 ordered->file_offset, ordered->num_bytes);
8712 if (!freespace_inode)
8713 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8715 btrfs_remove_ordered_extent(inode, ordered);
8716 btrfs_put_ordered_extent(ordered);
8717 btrfs_put_ordered_extent(ordered);
8720 btrfs_qgroup_check_reserved_leak(inode);
8721 inode_tree_del(inode);
8722 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8723 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8724 btrfs_put_root(inode->root);
8727 int btrfs_drop_inode(struct inode *inode)
8729 struct btrfs_root *root = BTRFS_I(inode)->root;
8734 /* the snap/subvol tree is on deleting */
8735 if (btrfs_root_refs(&root->root_item) == 0)
8738 return generic_drop_inode(inode);
8741 static void init_once(void *foo)
8743 struct btrfs_inode *ei = foo;
8745 inode_init_once(&ei->vfs_inode);
8748 void __cold btrfs_destroy_cachep(void)
8751 * Make sure all delayed rcu free inodes are flushed before we
8755 bioset_exit(&btrfs_dio_bioset);
8756 kmem_cache_destroy(btrfs_inode_cachep);
8759 int __init btrfs_init_cachep(void)
8761 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8762 sizeof(struct btrfs_inode), 0,
8763 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8765 if (!btrfs_inode_cachep)
8768 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8769 offsetof(struct btrfs_dio_private, bbio.bio),
8775 btrfs_destroy_cachep();
8779 static int btrfs_getattr(struct mnt_idmap *idmap,
8780 const struct path *path, struct kstat *stat,
8781 u32 request_mask, unsigned int flags)
8785 struct inode *inode = d_inode(path->dentry);
8786 u32 blocksize = inode->i_sb->s_blocksize;
8787 u32 bi_flags = BTRFS_I(inode)->flags;
8788 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8790 stat->result_mask |= STATX_BTIME;
8791 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8792 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8793 if (bi_flags & BTRFS_INODE_APPEND)
8794 stat->attributes |= STATX_ATTR_APPEND;
8795 if (bi_flags & BTRFS_INODE_COMPRESS)
8796 stat->attributes |= STATX_ATTR_COMPRESSED;
8797 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8798 stat->attributes |= STATX_ATTR_IMMUTABLE;
8799 if (bi_flags & BTRFS_INODE_NODUMP)
8800 stat->attributes |= STATX_ATTR_NODUMP;
8801 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8802 stat->attributes |= STATX_ATTR_VERITY;
8804 stat->attributes_mask |= (STATX_ATTR_APPEND |
8805 STATX_ATTR_COMPRESSED |
8806 STATX_ATTR_IMMUTABLE |
8809 generic_fillattr(idmap, inode, stat);
8810 stat->dev = BTRFS_I(inode)->root->anon_dev;
8812 spin_lock(&BTRFS_I(inode)->lock);
8813 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8814 inode_bytes = inode_get_bytes(inode);
8815 spin_unlock(&BTRFS_I(inode)->lock);
8816 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8817 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8821 static int btrfs_rename_exchange(struct inode *old_dir,
8822 struct dentry *old_dentry,
8823 struct inode *new_dir,
8824 struct dentry *new_dentry)
8826 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8827 struct btrfs_trans_handle *trans;
8828 unsigned int trans_num_items;
8829 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8830 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8831 struct inode *new_inode = new_dentry->d_inode;
8832 struct inode *old_inode = old_dentry->d_inode;
8833 struct timespec64 ctime = current_time(old_inode);
8834 struct btrfs_rename_ctx old_rename_ctx;
8835 struct btrfs_rename_ctx new_rename_ctx;
8836 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8837 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8842 bool need_abort = false;
8843 struct fscrypt_name old_fname, new_fname;
8844 struct fscrypt_str *old_name, *new_name;
8847 * For non-subvolumes allow exchange only within one subvolume, in the
8848 * same inode namespace. Two subvolumes (represented as directory) can
8849 * be exchanged as they're a logical link and have a fixed inode number.
8852 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8853 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8856 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8860 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8862 fscrypt_free_filename(&old_fname);
8866 old_name = &old_fname.disk_name;
8867 new_name = &new_fname.disk_name;
8869 /* close the race window with snapshot create/destroy ioctl */
8870 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8871 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8872 down_read(&fs_info->subvol_sem);
8876 * 1 to remove old dir item
8877 * 1 to remove old dir index
8878 * 1 to add new dir item
8879 * 1 to add new dir index
8880 * 1 to update parent inode
8882 * If the parents are the same, we only need to account for one
8884 trans_num_items = (old_dir == new_dir ? 9 : 10);
8885 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8887 * 1 to remove old root ref
8888 * 1 to remove old root backref
8889 * 1 to add new root ref
8890 * 1 to add new root backref
8892 trans_num_items += 4;
8895 * 1 to update inode item
8896 * 1 to remove old inode ref
8897 * 1 to add new inode ref
8899 trans_num_items += 3;
8901 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8902 trans_num_items += 4;
8904 trans_num_items += 3;
8905 trans = btrfs_start_transaction(root, trans_num_items);
8906 if (IS_ERR(trans)) {
8907 ret = PTR_ERR(trans);
8912 ret = btrfs_record_root_in_trans(trans, dest);
8918 * We need to find a free sequence number both in the source and
8919 * in the destination directory for the exchange.
8921 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8924 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8928 BTRFS_I(old_inode)->dir_index = 0ULL;
8929 BTRFS_I(new_inode)->dir_index = 0ULL;
8931 /* Reference for the source. */
8932 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8933 /* force full log commit if subvolume involved. */
8934 btrfs_set_log_full_commit(trans);
8936 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8937 btrfs_ino(BTRFS_I(new_dir)),
8944 /* And now for the dest. */
8945 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8946 /* force full log commit if subvolume involved. */
8947 btrfs_set_log_full_commit(trans);
8949 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8950 btrfs_ino(BTRFS_I(old_dir)),
8954 btrfs_abort_transaction(trans, ret);
8959 /* Update inode version and ctime/mtime. */
8960 inode_inc_iversion(old_dir);
8961 inode_inc_iversion(new_dir);
8962 inode_inc_iversion(old_inode);
8963 inode_inc_iversion(new_inode);
8964 old_dir->i_mtime = ctime;
8965 old_dir->i_ctime = ctime;
8966 new_dir->i_mtime = ctime;
8967 new_dir->i_ctime = ctime;
8968 old_inode->i_ctime = ctime;
8969 new_inode->i_ctime = ctime;
8971 if (old_dentry->d_parent != new_dentry->d_parent) {
8972 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8973 BTRFS_I(old_inode), true);
8974 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8975 BTRFS_I(new_inode), true);
8978 /* src is a subvolume */
8979 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8980 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8981 } else { /* src is an inode */
8982 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8983 BTRFS_I(old_dentry->d_inode),
8984 old_name, &old_rename_ctx);
8986 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8989 btrfs_abort_transaction(trans, ret);
8993 /* dest is a subvolume */
8994 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8995 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8996 } else { /* dest is an inode */
8997 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8998 BTRFS_I(new_dentry->d_inode),
8999 new_name, &new_rename_ctx);
9001 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9004 btrfs_abort_transaction(trans, ret);
9008 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9009 new_name, 0, old_idx);
9011 btrfs_abort_transaction(trans, ret);
9015 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9016 old_name, 0, new_idx);
9018 btrfs_abort_transaction(trans, ret);
9022 if (old_inode->i_nlink == 1)
9023 BTRFS_I(old_inode)->dir_index = old_idx;
9024 if (new_inode->i_nlink == 1)
9025 BTRFS_I(new_inode)->dir_index = new_idx;
9028 * Now pin the logs of the roots. We do it to ensure that no other task
9029 * can sync the logs while we are in progress with the rename, because
9030 * that could result in an inconsistency in case any of the inodes that
9031 * are part of this rename operation were logged before.
9033 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9034 btrfs_pin_log_trans(root);
9035 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9036 btrfs_pin_log_trans(dest);
9038 /* Do the log updates for all inodes. */
9039 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9040 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9041 old_rename_ctx.index, new_dentry->d_parent);
9042 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9043 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9044 new_rename_ctx.index, old_dentry->d_parent);
9046 /* Now unpin the logs. */
9047 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9048 btrfs_end_log_trans(root);
9049 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9050 btrfs_end_log_trans(dest);
9052 ret2 = btrfs_end_transaction(trans);
9053 ret = ret ? ret : ret2;
9055 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9056 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9057 up_read(&fs_info->subvol_sem);
9059 fscrypt_free_filename(&new_fname);
9060 fscrypt_free_filename(&old_fname);
9064 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9067 struct inode *inode;
9069 inode = new_inode(dir->i_sb);
9071 inode_init_owner(idmap, inode, dir,
9072 S_IFCHR | WHITEOUT_MODE);
9073 inode->i_op = &btrfs_special_inode_operations;
9074 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9079 static int btrfs_rename(struct mnt_idmap *idmap,
9080 struct inode *old_dir, struct dentry *old_dentry,
9081 struct inode *new_dir, struct dentry *new_dentry,
9084 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9085 struct btrfs_new_inode_args whiteout_args = {
9087 .dentry = old_dentry,
9089 struct btrfs_trans_handle *trans;
9090 unsigned int trans_num_items;
9091 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9092 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9093 struct inode *new_inode = d_inode(new_dentry);
9094 struct inode *old_inode = d_inode(old_dentry);
9095 struct btrfs_rename_ctx rename_ctx;
9099 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9100 struct fscrypt_name old_fname, new_fname;
9102 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9105 /* we only allow rename subvolume link between subvolumes */
9106 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9109 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9110 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9113 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9114 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9117 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9121 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9123 fscrypt_free_filename(&old_fname);
9127 /* check for collisions, even if the name isn't there */
9128 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9130 if (ret == -EEXIST) {
9132 * eexist without a new_inode */
9133 if (WARN_ON(!new_inode)) {
9134 goto out_fscrypt_names;
9137 /* maybe -EOVERFLOW */
9138 goto out_fscrypt_names;
9144 * we're using rename to replace one file with another. Start IO on it
9145 * now so we don't add too much work to the end of the transaction
9147 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9148 filemap_flush(old_inode->i_mapping);
9150 if (flags & RENAME_WHITEOUT) {
9151 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9152 if (!whiteout_args.inode) {
9154 goto out_fscrypt_names;
9156 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9158 goto out_whiteout_inode;
9160 /* 1 to update the old parent inode. */
9161 trans_num_items = 1;
9164 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9165 /* Close the race window with snapshot create/destroy ioctl */
9166 down_read(&fs_info->subvol_sem);
9168 * 1 to remove old root ref
9169 * 1 to remove old root backref
9170 * 1 to add new root ref
9171 * 1 to add new root backref
9173 trans_num_items += 4;
9177 * 1 to remove old inode ref
9178 * 1 to add new inode ref
9180 trans_num_items += 3;
9183 * 1 to remove old dir item
9184 * 1 to remove old dir index
9185 * 1 to add new dir item
9186 * 1 to add new dir index
9188 trans_num_items += 4;
9189 /* 1 to update new parent inode if it's not the same as the old parent */
9190 if (new_dir != old_dir)
9195 * 1 to remove inode ref
9196 * 1 to remove dir item
9197 * 1 to remove dir index
9198 * 1 to possibly add orphan item
9200 trans_num_items += 5;
9202 trans = btrfs_start_transaction(root, trans_num_items);
9203 if (IS_ERR(trans)) {
9204 ret = PTR_ERR(trans);
9209 ret = btrfs_record_root_in_trans(trans, dest);
9214 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9218 BTRFS_I(old_inode)->dir_index = 0ULL;
9219 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9220 /* force full log commit if subvolume involved. */
9221 btrfs_set_log_full_commit(trans);
9223 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9224 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9230 inode_inc_iversion(old_dir);
9231 inode_inc_iversion(new_dir);
9232 inode_inc_iversion(old_inode);
9233 old_dir->i_mtime = current_time(old_dir);
9234 old_dir->i_ctime = old_dir->i_mtime;
9235 new_dir->i_mtime = old_dir->i_mtime;
9236 new_dir->i_ctime = old_dir->i_mtime;
9237 old_inode->i_ctime = old_dir->i_mtime;
9239 if (old_dentry->d_parent != new_dentry->d_parent)
9240 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9241 BTRFS_I(old_inode), true);
9243 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9244 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9246 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9247 BTRFS_I(d_inode(old_dentry)),
9248 &old_fname.disk_name, &rename_ctx);
9250 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9253 btrfs_abort_transaction(trans, ret);
9258 inode_inc_iversion(new_inode);
9259 new_inode->i_ctime = current_time(new_inode);
9260 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9261 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9262 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9263 BUG_ON(new_inode->i_nlink == 0);
9265 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9266 BTRFS_I(d_inode(new_dentry)),
9267 &new_fname.disk_name);
9269 if (!ret && new_inode->i_nlink == 0)
9270 ret = btrfs_orphan_add(trans,
9271 BTRFS_I(d_inode(new_dentry)));
9273 btrfs_abort_transaction(trans, ret);
9278 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9279 &new_fname.disk_name, 0, index);
9281 btrfs_abort_transaction(trans, ret);
9285 if (old_inode->i_nlink == 1)
9286 BTRFS_I(old_inode)->dir_index = index;
9288 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9289 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9290 rename_ctx.index, new_dentry->d_parent);
9292 if (flags & RENAME_WHITEOUT) {
9293 ret = btrfs_create_new_inode(trans, &whiteout_args);
9295 btrfs_abort_transaction(trans, ret);
9298 unlock_new_inode(whiteout_args.inode);
9299 iput(whiteout_args.inode);
9300 whiteout_args.inode = NULL;
9304 ret2 = btrfs_end_transaction(trans);
9305 ret = ret ? ret : ret2;
9307 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9308 up_read(&fs_info->subvol_sem);
9309 if (flags & RENAME_WHITEOUT)
9310 btrfs_new_inode_args_destroy(&whiteout_args);
9312 if (flags & RENAME_WHITEOUT)
9313 iput(whiteout_args.inode);
9315 fscrypt_free_filename(&old_fname);
9316 fscrypt_free_filename(&new_fname);
9320 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9321 struct dentry *old_dentry, struct inode *new_dir,
9322 struct dentry *new_dentry, unsigned int flags)
9326 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9329 if (flags & RENAME_EXCHANGE)
9330 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9333 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9336 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9341 struct btrfs_delalloc_work {
9342 struct inode *inode;
9343 struct completion completion;
9344 struct list_head list;
9345 struct btrfs_work work;
9348 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9350 struct btrfs_delalloc_work *delalloc_work;
9351 struct inode *inode;
9353 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9355 inode = delalloc_work->inode;
9356 filemap_flush(inode->i_mapping);
9357 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9358 &BTRFS_I(inode)->runtime_flags))
9359 filemap_flush(inode->i_mapping);
9362 complete(&delalloc_work->completion);
9365 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9367 struct btrfs_delalloc_work *work;
9369 work = kmalloc(sizeof(*work), GFP_NOFS);
9373 init_completion(&work->completion);
9374 INIT_LIST_HEAD(&work->list);
9375 work->inode = inode;
9376 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9382 * some fairly slow code that needs optimization. This walks the list
9383 * of all the inodes with pending delalloc and forces them to disk.
9385 static int start_delalloc_inodes(struct btrfs_root *root,
9386 struct writeback_control *wbc, bool snapshot,
9387 bool in_reclaim_context)
9389 struct btrfs_inode *binode;
9390 struct inode *inode;
9391 struct btrfs_delalloc_work *work, *next;
9392 struct list_head works;
9393 struct list_head splice;
9395 bool full_flush = wbc->nr_to_write == LONG_MAX;
9397 INIT_LIST_HEAD(&works);
9398 INIT_LIST_HEAD(&splice);
9400 mutex_lock(&root->delalloc_mutex);
9401 spin_lock(&root->delalloc_lock);
9402 list_splice_init(&root->delalloc_inodes, &splice);
9403 while (!list_empty(&splice)) {
9404 binode = list_entry(splice.next, struct btrfs_inode,
9407 list_move_tail(&binode->delalloc_inodes,
9408 &root->delalloc_inodes);
9410 if (in_reclaim_context &&
9411 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9414 inode = igrab(&binode->vfs_inode);
9416 cond_resched_lock(&root->delalloc_lock);
9419 spin_unlock(&root->delalloc_lock);
9422 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9423 &binode->runtime_flags);
9425 work = btrfs_alloc_delalloc_work(inode);
9431 list_add_tail(&work->list, &works);
9432 btrfs_queue_work(root->fs_info->flush_workers,
9435 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9436 btrfs_add_delayed_iput(BTRFS_I(inode));
9437 if (ret || wbc->nr_to_write <= 0)
9441 spin_lock(&root->delalloc_lock);
9443 spin_unlock(&root->delalloc_lock);
9446 list_for_each_entry_safe(work, next, &works, list) {
9447 list_del_init(&work->list);
9448 wait_for_completion(&work->completion);
9452 if (!list_empty(&splice)) {
9453 spin_lock(&root->delalloc_lock);
9454 list_splice_tail(&splice, &root->delalloc_inodes);
9455 spin_unlock(&root->delalloc_lock);
9457 mutex_unlock(&root->delalloc_mutex);
9461 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9463 struct writeback_control wbc = {
9464 .nr_to_write = LONG_MAX,
9465 .sync_mode = WB_SYNC_NONE,
9467 .range_end = LLONG_MAX,
9469 struct btrfs_fs_info *fs_info = root->fs_info;
9471 if (BTRFS_FS_ERROR(fs_info))
9474 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9477 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9478 bool in_reclaim_context)
9480 struct writeback_control wbc = {
9482 .sync_mode = WB_SYNC_NONE,
9484 .range_end = LLONG_MAX,
9486 struct btrfs_root *root;
9487 struct list_head splice;
9490 if (BTRFS_FS_ERROR(fs_info))
9493 INIT_LIST_HEAD(&splice);
9495 mutex_lock(&fs_info->delalloc_root_mutex);
9496 spin_lock(&fs_info->delalloc_root_lock);
9497 list_splice_init(&fs_info->delalloc_roots, &splice);
9498 while (!list_empty(&splice)) {
9500 * Reset nr_to_write here so we know that we're doing a full
9504 wbc.nr_to_write = LONG_MAX;
9506 root = list_first_entry(&splice, struct btrfs_root,
9508 root = btrfs_grab_root(root);
9510 list_move_tail(&root->delalloc_root,
9511 &fs_info->delalloc_roots);
9512 spin_unlock(&fs_info->delalloc_root_lock);
9514 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9515 btrfs_put_root(root);
9516 if (ret < 0 || wbc.nr_to_write <= 0)
9518 spin_lock(&fs_info->delalloc_root_lock);
9520 spin_unlock(&fs_info->delalloc_root_lock);
9524 if (!list_empty(&splice)) {
9525 spin_lock(&fs_info->delalloc_root_lock);
9526 list_splice_tail(&splice, &fs_info->delalloc_roots);
9527 spin_unlock(&fs_info->delalloc_root_lock);
9529 mutex_unlock(&fs_info->delalloc_root_mutex);
9533 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9534 struct dentry *dentry, const char *symname)
9536 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9537 struct btrfs_trans_handle *trans;
9538 struct btrfs_root *root = BTRFS_I(dir)->root;
9539 struct btrfs_path *path;
9540 struct btrfs_key key;
9541 struct inode *inode;
9542 struct btrfs_new_inode_args new_inode_args = {
9546 unsigned int trans_num_items;
9551 struct btrfs_file_extent_item *ei;
9552 struct extent_buffer *leaf;
9554 name_len = strlen(symname);
9555 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9556 return -ENAMETOOLONG;
9558 inode = new_inode(dir->i_sb);
9561 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9562 inode->i_op = &btrfs_symlink_inode_operations;
9563 inode_nohighmem(inode);
9564 inode->i_mapping->a_ops = &btrfs_aops;
9565 btrfs_i_size_write(BTRFS_I(inode), name_len);
9566 inode_set_bytes(inode, name_len);
9568 new_inode_args.inode = inode;
9569 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9572 /* 1 additional item for the inline extent */
9575 trans = btrfs_start_transaction(root, trans_num_items);
9576 if (IS_ERR(trans)) {
9577 err = PTR_ERR(trans);
9578 goto out_new_inode_args;
9581 err = btrfs_create_new_inode(trans, &new_inode_args);
9585 path = btrfs_alloc_path();
9588 btrfs_abort_transaction(trans, err);
9589 discard_new_inode(inode);
9593 key.objectid = btrfs_ino(BTRFS_I(inode));
9595 key.type = BTRFS_EXTENT_DATA_KEY;
9596 datasize = btrfs_file_extent_calc_inline_size(name_len);
9597 err = btrfs_insert_empty_item(trans, root, path, &key,
9600 btrfs_abort_transaction(trans, err);
9601 btrfs_free_path(path);
9602 discard_new_inode(inode);
9606 leaf = path->nodes[0];
9607 ei = btrfs_item_ptr(leaf, path->slots[0],
9608 struct btrfs_file_extent_item);
9609 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9610 btrfs_set_file_extent_type(leaf, ei,
9611 BTRFS_FILE_EXTENT_INLINE);
9612 btrfs_set_file_extent_encryption(leaf, ei, 0);
9613 btrfs_set_file_extent_compression(leaf, ei, 0);
9614 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9615 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9617 ptr = btrfs_file_extent_inline_start(ei);
9618 write_extent_buffer(leaf, symname, ptr, name_len);
9619 btrfs_mark_buffer_dirty(leaf);
9620 btrfs_free_path(path);
9622 d_instantiate_new(dentry, inode);
9625 btrfs_end_transaction(trans);
9626 btrfs_btree_balance_dirty(fs_info);
9628 btrfs_new_inode_args_destroy(&new_inode_args);
9635 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9636 struct btrfs_trans_handle *trans_in,
9637 struct btrfs_inode *inode,
9638 struct btrfs_key *ins,
9641 struct btrfs_file_extent_item stack_fi;
9642 struct btrfs_replace_extent_info extent_info;
9643 struct btrfs_trans_handle *trans = trans_in;
9644 struct btrfs_path *path;
9645 u64 start = ins->objectid;
9646 u64 len = ins->offset;
9647 int qgroup_released;
9650 memset(&stack_fi, 0, sizeof(stack_fi));
9652 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9653 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9654 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9655 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9656 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9657 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9658 /* Encryption and other encoding is reserved and all 0 */
9660 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9661 if (qgroup_released < 0)
9662 return ERR_PTR(qgroup_released);
9665 ret = insert_reserved_file_extent(trans, inode,
9666 file_offset, &stack_fi,
9667 true, qgroup_released);
9673 extent_info.disk_offset = start;
9674 extent_info.disk_len = len;
9675 extent_info.data_offset = 0;
9676 extent_info.data_len = len;
9677 extent_info.file_offset = file_offset;
9678 extent_info.extent_buf = (char *)&stack_fi;
9679 extent_info.is_new_extent = true;
9680 extent_info.update_times = true;
9681 extent_info.qgroup_reserved = qgroup_released;
9682 extent_info.insertions = 0;
9684 path = btrfs_alloc_path();
9690 ret = btrfs_replace_file_extents(inode, path, file_offset,
9691 file_offset + len - 1, &extent_info,
9693 btrfs_free_path(path);
9700 * We have released qgroup data range at the beginning of the function,
9701 * and normally qgroup_released bytes will be freed when committing
9703 * But if we error out early, we have to free what we have released
9704 * or we leak qgroup data reservation.
9706 btrfs_qgroup_free_refroot(inode->root->fs_info,
9707 inode->root->root_key.objectid, qgroup_released,
9708 BTRFS_QGROUP_RSV_DATA);
9709 return ERR_PTR(ret);
9712 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9713 u64 start, u64 num_bytes, u64 min_size,
9714 loff_t actual_len, u64 *alloc_hint,
9715 struct btrfs_trans_handle *trans)
9717 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9718 struct extent_map *em;
9719 struct btrfs_root *root = BTRFS_I(inode)->root;
9720 struct btrfs_key ins;
9721 u64 cur_offset = start;
9722 u64 clear_offset = start;
9725 u64 last_alloc = (u64)-1;
9727 bool own_trans = true;
9728 u64 end = start + num_bytes - 1;
9732 while (num_bytes > 0) {
9733 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9734 cur_bytes = max(cur_bytes, min_size);
9736 * If we are severely fragmented we could end up with really
9737 * small allocations, so if the allocator is returning small
9738 * chunks lets make its job easier by only searching for those
9741 cur_bytes = min(cur_bytes, last_alloc);
9742 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9743 min_size, 0, *alloc_hint, &ins, 1, 0);
9748 * We've reserved this space, and thus converted it from
9749 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9750 * from here on out we will only need to clear our reservation
9751 * for the remaining unreserved area, so advance our
9752 * clear_offset by our extent size.
9754 clear_offset += ins.offset;
9756 last_alloc = ins.offset;
9757 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9760 * Now that we inserted the prealloc extent we can finally
9761 * decrement the number of reservations in the block group.
9762 * If we did it before, we could race with relocation and have
9763 * relocation miss the reserved extent, making it fail later.
9765 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9766 if (IS_ERR(trans)) {
9767 ret = PTR_ERR(trans);
9768 btrfs_free_reserved_extent(fs_info, ins.objectid,
9773 em = alloc_extent_map();
9775 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9776 cur_offset + ins.offset - 1, false);
9777 btrfs_set_inode_full_sync(BTRFS_I(inode));
9781 em->start = cur_offset;
9782 em->orig_start = cur_offset;
9783 em->len = ins.offset;
9784 em->block_start = ins.objectid;
9785 em->block_len = ins.offset;
9786 em->orig_block_len = ins.offset;
9787 em->ram_bytes = ins.offset;
9788 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9789 em->generation = trans->transid;
9791 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9792 free_extent_map(em);
9794 num_bytes -= ins.offset;
9795 cur_offset += ins.offset;
9796 *alloc_hint = ins.objectid + ins.offset;
9798 inode_inc_iversion(inode);
9799 inode->i_ctime = current_time(inode);
9800 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9801 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9802 (actual_len > inode->i_size) &&
9803 (cur_offset > inode->i_size)) {
9804 if (cur_offset > actual_len)
9805 i_size = actual_len;
9807 i_size = cur_offset;
9808 i_size_write(inode, i_size);
9809 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9812 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9815 btrfs_abort_transaction(trans, ret);
9817 btrfs_end_transaction(trans);
9822 btrfs_end_transaction(trans);
9826 if (clear_offset < end)
9827 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9828 end - clear_offset + 1);
9832 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9833 u64 start, u64 num_bytes, u64 min_size,
9834 loff_t actual_len, u64 *alloc_hint)
9836 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9837 min_size, actual_len, alloc_hint,
9841 int btrfs_prealloc_file_range_trans(struct inode *inode,
9842 struct btrfs_trans_handle *trans, int mode,
9843 u64 start, u64 num_bytes, u64 min_size,
9844 loff_t actual_len, u64 *alloc_hint)
9846 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9847 min_size, actual_len, alloc_hint, trans);
9850 static int btrfs_permission(struct mnt_idmap *idmap,
9851 struct inode *inode, int mask)
9853 struct btrfs_root *root = BTRFS_I(inode)->root;
9854 umode_t mode = inode->i_mode;
9856 if (mask & MAY_WRITE &&
9857 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9858 if (btrfs_root_readonly(root))
9860 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9863 return generic_permission(idmap, inode, mask);
9866 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9867 struct file *file, umode_t mode)
9869 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9870 struct btrfs_trans_handle *trans;
9871 struct btrfs_root *root = BTRFS_I(dir)->root;
9872 struct inode *inode;
9873 struct btrfs_new_inode_args new_inode_args = {
9875 .dentry = file->f_path.dentry,
9878 unsigned int trans_num_items;
9881 inode = new_inode(dir->i_sb);
9884 inode_init_owner(idmap, inode, dir, mode);
9885 inode->i_fop = &btrfs_file_operations;
9886 inode->i_op = &btrfs_file_inode_operations;
9887 inode->i_mapping->a_ops = &btrfs_aops;
9889 new_inode_args.inode = inode;
9890 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9894 trans = btrfs_start_transaction(root, trans_num_items);
9895 if (IS_ERR(trans)) {
9896 ret = PTR_ERR(trans);
9897 goto out_new_inode_args;
9900 ret = btrfs_create_new_inode(trans, &new_inode_args);
9903 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9904 * set it to 1 because d_tmpfile() will issue a warning if the count is
9907 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9909 set_nlink(inode, 1);
9912 d_tmpfile(file, inode);
9913 unlock_new_inode(inode);
9914 mark_inode_dirty(inode);
9917 btrfs_end_transaction(trans);
9918 btrfs_btree_balance_dirty(fs_info);
9920 btrfs_new_inode_args_destroy(&new_inode_args);
9924 return finish_open_simple(file, ret);
9927 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9929 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9930 unsigned long index = start >> PAGE_SHIFT;
9931 unsigned long end_index = end >> PAGE_SHIFT;
9935 ASSERT(end + 1 - start <= U32_MAX);
9936 len = end + 1 - start;
9937 while (index <= end_index) {
9938 page = find_get_page(inode->vfs_inode.i_mapping, index);
9939 ASSERT(page); /* Pages should be in the extent_io_tree */
9941 btrfs_page_set_writeback(fs_info, page, start, len);
9947 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9950 switch (compress_type) {
9951 case BTRFS_COMPRESS_NONE:
9952 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9953 case BTRFS_COMPRESS_ZLIB:
9954 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9955 case BTRFS_COMPRESS_LZO:
9957 * The LZO format depends on the sector size. 64K is the maximum
9958 * sector size that we support.
9960 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9962 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9963 (fs_info->sectorsize_bits - 12);
9964 case BTRFS_COMPRESS_ZSTD:
9965 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9971 static ssize_t btrfs_encoded_read_inline(
9973 struct iov_iter *iter, u64 start,
9975 struct extent_state **cached_state,
9976 u64 extent_start, size_t count,
9977 struct btrfs_ioctl_encoded_io_args *encoded,
9980 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9981 struct btrfs_root *root = inode->root;
9982 struct btrfs_fs_info *fs_info = root->fs_info;
9983 struct extent_io_tree *io_tree = &inode->io_tree;
9984 struct btrfs_path *path;
9985 struct extent_buffer *leaf;
9986 struct btrfs_file_extent_item *item;
9992 path = btrfs_alloc_path();
9997 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
10001 /* The extent item disappeared? */
10006 leaf = path->nodes[0];
10007 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10009 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10010 ptr = btrfs_file_extent_inline_start(item);
10012 encoded->len = min_t(u64, extent_start + ram_bytes,
10013 inode->vfs_inode.i_size) - iocb->ki_pos;
10014 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10015 btrfs_file_extent_compression(leaf, item));
10018 encoded->compression = ret;
10019 if (encoded->compression) {
10020 size_t inline_size;
10022 inline_size = btrfs_file_extent_inline_item_len(leaf,
10024 if (inline_size > count) {
10028 count = inline_size;
10029 encoded->unencoded_len = ram_bytes;
10030 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10032 count = min_t(u64, count, encoded->len);
10033 encoded->len = count;
10034 encoded->unencoded_len = count;
10035 ptr += iocb->ki_pos - extent_start;
10038 tmp = kmalloc(count, GFP_NOFS);
10043 read_extent_buffer(leaf, tmp, ptr, count);
10044 btrfs_release_path(path);
10045 unlock_extent(io_tree, start, lockend, cached_state);
10046 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10049 ret = copy_to_iter(tmp, count, iter);
10054 btrfs_free_path(path);
10058 struct btrfs_encoded_read_private {
10059 wait_queue_head_t wait;
10061 blk_status_t status;
10064 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10066 struct btrfs_encoded_read_private *priv = bbio->private;
10068 if (bbio->bio.bi_status) {
10070 * The memory barrier implied by the atomic_dec_return() here
10071 * pairs with the memory barrier implied by the
10072 * atomic_dec_return() or io_wait_event() in
10073 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10074 * write is observed before the load of status in
10075 * btrfs_encoded_read_regular_fill_pages().
10077 WRITE_ONCE(priv->status, bbio->bio.bi_status);
10079 if (!atomic_dec_return(&priv->pending))
10080 wake_up(&priv->wait);
10081 bio_put(&bbio->bio);
10084 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10085 u64 file_offset, u64 disk_bytenr,
10086 u64 disk_io_size, struct page **pages)
10088 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10089 struct btrfs_encoded_read_private priv = {
10090 .pending = ATOMIC_INIT(1),
10092 unsigned long i = 0;
10093 struct btrfs_bio *bbio;
10095 init_waitqueue_head(&priv.wait);
10097 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10098 btrfs_encoded_read_endio, &priv);
10099 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10100 bbio->inode = inode;
10103 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10105 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10106 atomic_inc(&priv.pending);
10107 btrfs_submit_bio(bbio, 0);
10109 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10110 btrfs_encoded_read_endio, &priv);
10111 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10112 bbio->inode = inode;
10117 disk_bytenr += bytes;
10118 disk_io_size -= bytes;
10119 } while (disk_io_size);
10121 atomic_inc(&priv.pending);
10122 btrfs_submit_bio(bbio, 0);
10124 if (atomic_dec_return(&priv.pending))
10125 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10126 /* See btrfs_encoded_read_endio() for ordering. */
10127 return blk_status_to_errno(READ_ONCE(priv.status));
10130 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10131 struct iov_iter *iter,
10132 u64 start, u64 lockend,
10133 struct extent_state **cached_state,
10134 u64 disk_bytenr, u64 disk_io_size,
10135 size_t count, bool compressed,
10138 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10139 struct extent_io_tree *io_tree = &inode->io_tree;
10140 struct page **pages;
10141 unsigned long nr_pages, i;
10143 size_t page_offset;
10146 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10147 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10150 ret = btrfs_alloc_page_array(nr_pages, pages);
10156 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10157 disk_io_size, pages);
10161 unlock_extent(io_tree, start, lockend, cached_state);
10162 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10169 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10170 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10173 while (cur < count) {
10174 size_t bytes = min_t(size_t, count - cur,
10175 PAGE_SIZE - page_offset);
10177 if (copy_page_to_iter(pages[i], page_offset, bytes,
10188 for (i = 0; i < nr_pages; i++) {
10190 __free_page(pages[i]);
10196 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10197 struct btrfs_ioctl_encoded_io_args *encoded)
10199 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10200 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10201 struct extent_io_tree *io_tree = &inode->io_tree;
10203 size_t count = iov_iter_count(iter);
10204 u64 start, lockend, disk_bytenr, disk_io_size;
10205 struct extent_state *cached_state = NULL;
10206 struct extent_map *em;
10207 bool unlocked = false;
10209 file_accessed(iocb->ki_filp);
10211 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10213 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10214 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10217 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10219 * We don't know how long the extent containing iocb->ki_pos is, but if
10220 * it's compressed we know that it won't be longer than this.
10222 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10225 struct btrfs_ordered_extent *ordered;
10227 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10228 lockend - start + 1);
10230 goto out_unlock_inode;
10231 lock_extent(io_tree, start, lockend, &cached_state);
10232 ordered = btrfs_lookup_ordered_range(inode, start,
10233 lockend - start + 1);
10236 btrfs_put_ordered_extent(ordered);
10237 unlock_extent(io_tree, start, lockend, &cached_state);
10241 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10244 goto out_unlock_extent;
10247 if (em->block_start == EXTENT_MAP_INLINE) {
10248 u64 extent_start = em->start;
10251 * For inline extents we get everything we need out of the
10254 free_extent_map(em);
10256 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10257 &cached_state, extent_start,
10258 count, encoded, &unlocked);
10263 * We only want to return up to EOF even if the extent extends beyond
10266 encoded->len = min_t(u64, extent_map_end(em),
10267 inode->vfs_inode.i_size) - iocb->ki_pos;
10268 if (em->block_start == EXTENT_MAP_HOLE ||
10269 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10270 disk_bytenr = EXTENT_MAP_HOLE;
10271 count = min_t(u64, count, encoded->len);
10272 encoded->len = count;
10273 encoded->unencoded_len = count;
10274 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10275 disk_bytenr = em->block_start;
10277 * Bail if the buffer isn't large enough to return the whole
10278 * compressed extent.
10280 if (em->block_len > count) {
10284 disk_io_size = em->block_len;
10285 count = em->block_len;
10286 encoded->unencoded_len = em->ram_bytes;
10287 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10288 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10289 em->compress_type);
10292 encoded->compression = ret;
10294 disk_bytenr = em->block_start + (start - em->start);
10295 if (encoded->len > count)
10296 encoded->len = count;
10298 * Don't read beyond what we locked. This also limits the page
10299 * allocations that we'll do.
10301 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10302 count = start + disk_io_size - iocb->ki_pos;
10303 encoded->len = count;
10304 encoded->unencoded_len = count;
10305 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10307 free_extent_map(em);
10310 if (disk_bytenr == EXTENT_MAP_HOLE) {
10311 unlock_extent(io_tree, start, lockend, &cached_state);
10312 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10314 ret = iov_iter_zero(count, iter);
10318 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10319 &cached_state, disk_bytenr,
10320 disk_io_size, count,
10321 encoded->compression,
10327 iocb->ki_pos += encoded->len;
10329 free_extent_map(em);
10332 unlock_extent(io_tree, start, lockend, &cached_state);
10335 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10339 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10340 const struct btrfs_ioctl_encoded_io_args *encoded)
10342 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10343 struct btrfs_root *root = inode->root;
10344 struct btrfs_fs_info *fs_info = root->fs_info;
10345 struct extent_io_tree *io_tree = &inode->io_tree;
10346 struct extent_changeset *data_reserved = NULL;
10347 struct extent_state *cached_state = NULL;
10348 struct btrfs_ordered_extent *ordered;
10352 u64 num_bytes, ram_bytes, disk_num_bytes;
10353 unsigned long nr_pages, i;
10354 struct page **pages;
10355 struct btrfs_key ins;
10356 bool extent_reserved = false;
10357 struct extent_map *em;
10360 switch (encoded->compression) {
10361 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10362 compression = BTRFS_COMPRESS_ZLIB;
10364 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10365 compression = BTRFS_COMPRESS_ZSTD;
10367 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10368 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10369 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10370 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10371 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10372 /* The sector size must match for LZO. */
10373 if (encoded->compression -
10374 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10375 fs_info->sectorsize_bits)
10377 compression = BTRFS_COMPRESS_LZO;
10382 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10385 orig_count = iov_iter_count(from);
10387 /* The extent size must be sane. */
10388 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10389 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10393 * The compressed data must be smaller than the decompressed data.
10395 * It's of course possible for data to compress to larger or the same
10396 * size, but the buffered I/O path falls back to no compression for such
10397 * data, and we don't want to break any assumptions by creating these
10400 * Note that this is less strict than the current check we have that the
10401 * compressed data must be at least one sector smaller than the
10402 * decompressed data. We only want to enforce the weaker requirement
10403 * from old kernels that it is at least one byte smaller.
10405 if (orig_count >= encoded->unencoded_len)
10408 /* The extent must start on a sector boundary. */
10409 start = iocb->ki_pos;
10410 if (!IS_ALIGNED(start, fs_info->sectorsize))
10414 * The extent must end on a sector boundary. However, we allow a write
10415 * which ends at or extends i_size to have an unaligned length; we round
10416 * up the extent size and set i_size to the unaligned end.
10418 if (start + encoded->len < inode->vfs_inode.i_size &&
10419 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10422 /* Finally, the offset in the unencoded data must be sector-aligned. */
10423 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10426 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10427 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10428 end = start + num_bytes - 1;
10431 * If the extent cannot be inline, the compressed data on disk must be
10432 * sector-aligned. For convenience, we extend it with zeroes if it
10435 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10436 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10437 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10440 for (i = 0; i < nr_pages; i++) {
10441 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10444 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10449 kaddr = kmap_local_page(pages[i]);
10450 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10451 kunmap_local(kaddr);
10455 if (bytes < PAGE_SIZE)
10456 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10457 kunmap_local(kaddr);
10461 struct btrfs_ordered_extent *ordered;
10463 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10466 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10467 start >> PAGE_SHIFT,
10468 end >> PAGE_SHIFT);
10471 lock_extent(io_tree, start, end, &cached_state);
10472 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10474 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10477 btrfs_put_ordered_extent(ordered);
10478 unlock_extent(io_tree, start, end, &cached_state);
10483 * We don't use the higher-level delalloc space functions because our
10484 * num_bytes and disk_num_bytes are different.
10486 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10489 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10491 goto out_free_data_space;
10492 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10495 goto out_qgroup_free_data;
10497 /* Try an inline extent first. */
10498 if (start == 0 && encoded->unencoded_len == encoded->len &&
10499 encoded->unencoded_offset == 0) {
10500 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10501 compression, pages, true);
10505 goto out_delalloc_release;
10509 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10510 disk_num_bytes, 0, 0, &ins, 1, 1);
10512 goto out_delalloc_release;
10513 extent_reserved = true;
10515 em = create_io_em(inode, start, num_bytes,
10516 start - encoded->unencoded_offset, ins.objectid,
10517 ins.offset, ins.offset, ram_bytes, compression,
10518 BTRFS_ORDERED_COMPRESSED);
10521 goto out_free_reserved;
10523 free_extent_map(em);
10525 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10526 ins.objectid, ins.offset,
10527 encoded->unencoded_offset,
10528 (1 << BTRFS_ORDERED_ENCODED) |
10529 (1 << BTRFS_ORDERED_COMPRESSED),
10531 if (IS_ERR(ordered)) {
10532 btrfs_drop_extent_map_range(inode, start, end, false);
10533 ret = PTR_ERR(ordered);
10534 goto out_free_reserved;
10536 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10538 if (start + encoded->len > inode->vfs_inode.i_size)
10539 i_size_write(&inode->vfs_inode, start + encoded->len);
10541 unlock_extent(io_tree, start, end, &cached_state);
10543 btrfs_delalloc_release_extents(inode, num_bytes);
10545 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10550 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10551 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10552 out_delalloc_release:
10553 btrfs_delalloc_release_extents(inode, num_bytes);
10554 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10555 out_qgroup_free_data:
10557 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10558 out_free_data_space:
10560 * If btrfs_reserve_extent() succeeded, then we already decremented
10563 if (!extent_reserved)
10564 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10566 unlock_extent(io_tree, start, end, &cached_state);
10568 for (i = 0; i < nr_pages; i++) {
10570 __free_page(pages[i]);
10575 iocb->ki_pos += encoded->len;
10581 * Add an entry indicating a block group or device which is pinned by a
10582 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10583 * negative errno on failure.
10585 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10586 bool is_block_group)
10588 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10589 struct btrfs_swapfile_pin *sp, *entry;
10590 struct rb_node **p;
10591 struct rb_node *parent = NULL;
10593 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10598 sp->is_block_group = is_block_group;
10599 sp->bg_extent_count = 1;
10601 spin_lock(&fs_info->swapfile_pins_lock);
10602 p = &fs_info->swapfile_pins.rb_node;
10605 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10606 if (sp->ptr < entry->ptr ||
10607 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10608 p = &(*p)->rb_left;
10609 } else if (sp->ptr > entry->ptr ||
10610 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10611 p = &(*p)->rb_right;
10613 if (is_block_group)
10614 entry->bg_extent_count++;
10615 spin_unlock(&fs_info->swapfile_pins_lock);
10620 rb_link_node(&sp->node, parent, p);
10621 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10622 spin_unlock(&fs_info->swapfile_pins_lock);
10626 /* Free all of the entries pinned by this swapfile. */
10627 static void btrfs_free_swapfile_pins(struct inode *inode)
10629 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10630 struct btrfs_swapfile_pin *sp;
10631 struct rb_node *node, *next;
10633 spin_lock(&fs_info->swapfile_pins_lock);
10634 node = rb_first(&fs_info->swapfile_pins);
10636 next = rb_next(node);
10637 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10638 if (sp->inode == inode) {
10639 rb_erase(&sp->node, &fs_info->swapfile_pins);
10640 if (sp->is_block_group) {
10641 btrfs_dec_block_group_swap_extents(sp->ptr,
10642 sp->bg_extent_count);
10643 btrfs_put_block_group(sp->ptr);
10649 spin_unlock(&fs_info->swapfile_pins_lock);
10652 struct btrfs_swap_info {
10658 unsigned long nr_pages;
10662 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10663 struct btrfs_swap_info *bsi)
10665 unsigned long nr_pages;
10666 unsigned long max_pages;
10667 u64 first_ppage, first_ppage_reported, next_ppage;
10671 * Our swapfile may have had its size extended after the swap header was
10672 * written. In that case activating the swapfile should not go beyond
10673 * the max size set in the swap header.
10675 if (bsi->nr_pages >= sis->max)
10678 max_pages = sis->max - bsi->nr_pages;
10679 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10680 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10682 if (first_ppage >= next_ppage)
10684 nr_pages = next_ppage - first_ppage;
10685 nr_pages = min(nr_pages, max_pages);
10687 first_ppage_reported = first_ppage;
10688 if (bsi->start == 0)
10689 first_ppage_reported++;
10690 if (bsi->lowest_ppage > first_ppage_reported)
10691 bsi->lowest_ppage = first_ppage_reported;
10692 if (bsi->highest_ppage < (next_ppage - 1))
10693 bsi->highest_ppage = next_ppage - 1;
10695 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10698 bsi->nr_extents += ret;
10699 bsi->nr_pages += nr_pages;
10703 static void btrfs_swap_deactivate(struct file *file)
10705 struct inode *inode = file_inode(file);
10707 btrfs_free_swapfile_pins(inode);
10708 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10711 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10714 struct inode *inode = file_inode(file);
10715 struct btrfs_root *root = BTRFS_I(inode)->root;
10716 struct btrfs_fs_info *fs_info = root->fs_info;
10717 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10718 struct extent_state *cached_state = NULL;
10719 struct extent_map *em = NULL;
10720 struct btrfs_device *device = NULL;
10721 struct btrfs_swap_info bsi = {
10722 .lowest_ppage = (sector_t)-1ULL,
10729 * If the swap file was just created, make sure delalloc is done. If the
10730 * file changes again after this, the user is doing something stupid and
10731 * we don't really care.
10733 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10738 * The inode is locked, so these flags won't change after we check them.
10740 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10741 btrfs_warn(fs_info, "swapfile must not be compressed");
10744 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10745 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10748 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10749 btrfs_warn(fs_info, "swapfile must not be checksummed");
10754 * Balance or device remove/replace/resize can move stuff around from
10755 * under us. The exclop protection makes sure they aren't running/won't
10756 * run concurrently while we are mapping the swap extents, and
10757 * fs_info->swapfile_pins prevents them from running while the swap
10758 * file is active and moving the extents. Note that this also prevents
10759 * a concurrent device add which isn't actually necessary, but it's not
10760 * really worth the trouble to allow it.
10762 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10763 btrfs_warn(fs_info,
10764 "cannot activate swapfile while exclusive operation is running");
10769 * Prevent snapshot creation while we are activating the swap file.
10770 * We do not want to race with snapshot creation. If snapshot creation
10771 * already started before we bumped nr_swapfiles from 0 to 1 and
10772 * completes before the first write into the swap file after it is
10773 * activated, than that write would fallback to COW.
10775 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10776 btrfs_exclop_finish(fs_info);
10777 btrfs_warn(fs_info,
10778 "cannot activate swapfile because snapshot creation is in progress");
10782 * Snapshots can create extents which require COW even if NODATACOW is
10783 * set. We use this counter to prevent snapshots. We must increment it
10784 * before walking the extents because we don't want a concurrent
10785 * snapshot to run after we've already checked the extents.
10787 * It is possible that subvolume is marked for deletion but still not
10788 * removed yet. To prevent this race, we check the root status before
10789 * activating the swapfile.
10791 spin_lock(&root->root_item_lock);
10792 if (btrfs_root_dead(root)) {
10793 spin_unlock(&root->root_item_lock);
10795 btrfs_exclop_finish(fs_info);
10796 btrfs_warn(fs_info,
10797 "cannot activate swapfile because subvolume %llu is being deleted",
10798 root->root_key.objectid);
10801 atomic_inc(&root->nr_swapfiles);
10802 spin_unlock(&root->root_item_lock);
10804 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10806 lock_extent(io_tree, 0, isize - 1, &cached_state);
10808 while (start < isize) {
10809 u64 logical_block_start, physical_block_start;
10810 struct btrfs_block_group *bg;
10811 u64 len = isize - start;
10813 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10819 if (em->block_start == EXTENT_MAP_HOLE) {
10820 btrfs_warn(fs_info, "swapfile must not have holes");
10824 if (em->block_start == EXTENT_MAP_INLINE) {
10826 * It's unlikely we'll ever actually find ourselves
10827 * here, as a file small enough to fit inline won't be
10828 * big enough to store more than the swap header, but in
10829 * case something changes in the future, let's catch it
10830 * here rather than later.
10832 btrfs_warn(fs_info, "swapfile must not be inline");
10836 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10837 btrfs_warn(fs_info, "swapfile must not be compressed");
10842 logical_block_start = em->block_start + (start - em->start);
10843 len = min(len, em->len - (start - em->start));
10844 free_extent_map(em);
10847 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10853 btrfs_warn(fs_info,
10854 "swapfile must not be copy-on-write");
10859 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10865 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10866 btrfs_warn(fs_info,
10867 "swapfile must have single data profile");
10872 if (device == NULL) {
10873 device = em->map_lookup->stripes[0].dev;
10874 ret = btrfs_add_swapfile_pin(inode, device, false);
10879 } else if (device != em->map_lookup->stripes[0].dev) {
10880 btrfs_warn(fs_info, "swapfile must be on one device");
10885 physical_block_start = (em->map_lookup->stripes[0].physical +
10886 (logical_block_start - em->start));
10887 len = min(len, em->len - (logical_block_start - em->start));
10888 free_extent_map(em);
10891 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10893 btrfs_warn(fs_info,
10894 "could not find block group containing swapfile");
10899 if (!btrfs_inc_block_group_swap_extents(bg)) {
10900 btrfs_warn(fs_info,
10901 "block group for swapfile at %llu is read-only%s",
10903 atomic_read(&fs_info->scrubs_running) ?
10904 " (scrub running)" : "");
10905 btrfs_put_block_group(bg);
10910 ret = btrfs_add_swapfile_pin(inode, bg, true);
10912 btrfs_put_block_group(bg);
10919 if (bsi.block_len &&
10920 bsi.block_start + bsi.block_len == physical_block_start) {
10921 bsi.block_len += len;
10923 if (bsi.block_len) {
10924 ret = btrfs_add_swap_extent(sis, &bsi);
10929 bsi.block_start = physical_block_start;
10930 bsi.block_len = len;
10937 ret = btrfs_add_swap_extent(sis, &bsi);
10940 if (!IS_ERR_OR_NULL(em))
10941 free_extent_map(em);
10943 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10946 btrfs_swap_deactivate(file);
10948 btrfs_drew_write_unlock(&root->snapshot_lock);
10950 btrfs_exclop_finish(fs_info);
10956 sis->bdev = device->bdev;
10957 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10958 sis->max = bsi.nr_pages;
10959 sis->pages = bsi.nr_pages - 1;
10960 sis->highest_bit = bsi.nr_pages - 1;
10961 return bsi.nr_extents;
10964 static void btrfs_swap_deactivate(struct file *file)
10968 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10971 return -EOPNOTSUPP;
10976 * Update the number of bytes used in the VFS' inode. When we replace extents in
10977 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10978 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10979 * always get a correct value.
10981 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10982 const u64 add_bytes,
10983 const u64 del_bytes)
10985 if (add_bytes == del_bytes)
10988 spin_lock(&inode->lock);
10990 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10992 inode_add_bytes(&inode->vfs_inode, add_bytes);
10993 spin_unlock(&inode->lock);
10997 * Verify that there are no ordered extents for a given file range.
10999 * @inode: The target inode.
11000 * @start: Start offset of the file range, should be sector size aligned.
11001 * @end: End offset (inclusive) of the file range, its value +1 should be
11002 * sector size aligned.
11004 * This should typically be used for cases where we locked an inode's VFS lock in
11005 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11006 * we have flushed all delalloc in the range, we have waited for all ordered
11007 * extents in the range to complete and finally we have locked the file range in
11008 * the inode's io_tree.
11010 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11012 struct btrfs_root *root = inode->root;
11013 struct btrfs_ordered_extent *ordered;
11015 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11018 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11020 btrfs_err(root->fs_info,
11021 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11022 start, end, btrfs_ino(inode), root->root_key.objectid,
11023 ordered->file_offset,
11024 ordered->file_offset + ordered->num_bytes - 1);
11025 btrfs_put_ordered_extent(ordered);
11028 ASSERT(ordered == NULL);
11031 static const struct inode_operations btrfs_dir_inode_operations = {
11032 .getattr = btrfs_getattr,
11033 .lookup = btrfs_lookup,
11034 .create = btrfs_create,
11035 .unlink = btrfs_unlink,
11036 .link = btrfs_link,
11037 .mkdir = btrfs_mkdir,
11038 .rmdir = btrfs_rmdir,
11039 .rename = btrfs_rename2,
11040 .symlink = btrfs_symlink,
11041 .setattr = btrfs_setattr,
11042 .mknod = btrfs_mknod,
11043 .listxattr = btrfs_listxattr,
11044 .permission = btrfs_permission,
11045 .get_inode_acl = btrfs_get_acl,
11046 .set_acl = btrfs_set_acl,
11047 .update_time = btrfs_update_time,
11048 .tmpfile = btrfs_tmpfile,
11049 .fileattr_get = btrfs_fileattr_get,
11050 .fileattr_set = btrfs_fileattr_set,
11053 static const struct file_operations btrfs_dir_file_operations = {
11054 .llseek = generic_file_llseek,
11055 .read = generic_read_dir,
11056 .iterate_shared = btrfs_real_readdir,
11057 .open = btrfs_opendir,
11058 .unlocked_ioctl = btrfs_ioctl,
11059 #ifdef CONFIG_COMPAT
11060 .compat_ioctl = btrfs_compat_ioctl,
11062 .release = btrfs_release_file,
11063 .fsync = btrfs_sync_file,
11067 * btrfs doesn't support the bmap operation because swapfiles
11068 * use bmap to make a mapping of extents in the file. They assume
11069 * these extents won't change over the life of the file and they
11070 * use the bmap result to do IO directly to the drive.
11072 * the btrfs bmap call would return logical addresses that aren't
11073 * suitable for IO and they also will change frequently as COW
11074 * operations happen. So, swapfile + btrfs == corruption.
11076 * For now we're avoiding this by dropping bmap.
11078 static const struct address_space_operations btrfs_aops = {
11079 .read_folio = btrfs_read_folio,
11080 .writepages = btrfs_writepages,
11081 .readahead = btrfs_readahead,
11082 .invalidate_folio = btrfs_invalidate_folio,
11083 .release_folio = btrfs_release_folio,
11084 .migrate_folio = btrfs_migrate_folio,
11085 .dirty_folio = filemap_dirty_folio,
11086 .error_remove_page = generic_error_remove_page,
11087 .swap_activate = btrfs_swap_activate,
11088 .swap_deactivate = btrfs_swap_deactivate,
11091 static const struct inode_operations btrfs_file_inode_operations = {
11092 .getattr = btrfs_getattr,
11093 .setattr = btrfs_setattr,
11094 .listxattr = btrfs_listxattr,
11095 .permission = btrfs_permission,
11096 .fiemap = btrfs_fiemap,
11097 .get_inode_acl = btrfs_get_acl,
11098 .set_acl = btrfs_set_acl,
11099 .update_time = btrfs_update_time,
11100 .fileattr_get = btrfs_fileattr_get,
11101 .fileattr_set = btrfs_fileattr_set,
11103 static const struct inode_operations btrfs_special_inode_operations = {
11104 .getattr = btrfs_getattr,
11105 .setattr = btrfs_setattr,
11106 .permission = btrfs_permission,
11107 .listxattr = btrfs_listxattr,
11108 .get_inode_acl = btrfs_get_acl,
11109 .set_acl = btrfs_set_acl,
11110 .update_time = btrfs_update_time,
11112 static const struct inode_operations btrfs_symlink_inode_operations = {
11113 .get_link = page_get_link,
11114 .getattr = btrfs_getattr,
11115 .setattr = btrfs_setattr,
11116 .permission = btrfs_permission,
11117 .listxattr = btrfs_listxattr,
11118 .update_time = btrfs_update_time,
11121 const struct dentry_operations btrfs_dentry_operations = {
11122 .d_delete = btrfs_dentry_delete,