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);
1220 free_async_extent_pages(async_extent);
1222 * Here we used to try again by going back to non-compressed
1223 * path for ENOSPC. But we can't reserve space even for
1224 * compressed size, how could it work for uncompressed size
1225 * which requires larger size? So here we directly go error
1231 /* Here we're doing allocation and writeback of the compressed pages */
1232 em = create_io_em(inode, start,
1233 async_extent->ram_size, /* len */
1234 start, /* orig_start */
1235 ins.objectid, /* block_start */
1236 ins.offset, /* block_len */
1237 ins.offset, /* orig_block_len */
1238 async_extent->ram_size, /* ram_bytes */
1239 async_extent->compress_type,
1240 BTRFS_ORDERED_COMPRESSED);
1243 goto out_free_reserve;
1245 free_extent_map(em);
1247 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1248 async_extent->ram_size, /* num_bytes */
1249 async_extent->ram_size, /* ram_bytes */
1250 ins.objectid, /* disk_bytenr */
1251 ins.offset, /* disk_num_bytes */
1253 1 << BTRFS_ORDERED_COMPRESSED,
1254 async_extent->compress_type);
1255 if (IS_ERR(ordered)) {
1256 btrfs_drop_extent_map_range(inode, start, end, false);
1257 ret = PTR_ERR(ordered);
1258 goto out_free_reserve;
1260 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1262 /* Clear dirty, set writeback and unlock the pages. */
1263 extent_clear_unlock_delalloc(inode, start, end,
1264 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1265 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1266 btrfs_submit_compressed_write(ordered,
1267 async_extent->pages, /* compressed_pages */
1268 async_extent->nr_pages,
1269 async_chunk->write_flags, true);
1270 *alloc_hint = ins.objectid + ins.offset;
1272 if (async_chunk->blkcg_css)
1273 kthread_associate_blkcg(NULL);
1274 kfree(async_extent);
1278 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1279 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1281 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1282 extent_clear_unlock_delalloc(inode, start, end,
1283 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1284 EXTENT_DELALLOC_NEW |
1285 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1286 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1287 PAGE_END_WRITEBACK);
1288 free_async_extent_pages(async_extent);
1293 * Phase two of compressed writeback. This is the ordered portion of the code,
1294 * which only gets called in the order the work was queued. We walk all the
1295 * async extents created by compress_file_range and send them down to the disk.
1297 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1299 struct btrfs_inode *inode = async_chunk->inode;
1300 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1301 struct async_extent *async_extent;
1305 while (!list_empty(&async_chunk->extents)) {
1309 async_extent = list_entry(async_chunk->extents.next,
1310 struct async_extent, list);
1311 list_del(&async_extent->list);
1312 extent_start = async_extent->start;
1313 ram_size = async_extent->ram_size;
1315 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1317 btrfs_debug(fs_info,
1318 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1319 inode->root->root_key.objectid,
1320 btrfs_ino(inode), extent_start, ram_size, ret);
1324 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1327 struct extent_map_tree *em_tree = &inode->extent_tree;
1328 struct extent_map *em;
1331 read_lock(&em_tree->lock);
1332 em = search_extent_mapping(em_tree, start, num_bytes);
1335 * if block start isn't an actual block number then find the
1336 * first block in this inode and use that as a hint. If that
1337 * block is also bogus then just don't worry about it.
1339 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1340 free_extent_map(em);
1341 em = search_extent_mapping(em_tree, 0, 0);
1342 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1343 alloc_hint = em->block_start;
1345 free_extent_map(em);
1347 alloc_hint = em->block_start;
1348 free_extent_map(em);
1351 read_unlock(&em_tree->lock);
1357 * when extent_io.c finds a delayed allocation range in the file,
1358 * the call backs end up in this code. The basic idea is to
1359 * allocate extents on disk for the range, and create ordered data structs
1360 * in ram to track those extents.
1362 * locked_page is the page that writepage had locked already. We use
1363 * it to make sure we don't do extra locks or unlocks.
1365 * *page_started is set to one if we unlock locked_page and do everything
1366 * required to start IO on it. It may be clean and already done with
1367 * IO when we return.
1369 * When unlock == 1, we unlock the pages in successfully allocated regions.
1370 * When unlock == 0, we leave them locked for writing them out.
1372 * However, we unlock all the pages except @locked_page in case of failure.
1374 * In summary, page locking state will be as follow:
1376 * - page_started == 1 (return value)
1377 * - All the pages are unlocked. IO is started.
1378 * - Note that this can happen only on success
1380 * - All the pages except @locked_page are unlocked in any case
1382 * - On success, all the pages are locked for writing out them
1383 * - On failure, all the pages except @locked_page are unlocked
1385 * When a failure happens in the second or later iteration of the
1386 * while-loop, the ordered extents created in previous iterations are kept
1387 * intact. So, the caller must clean them up by calling
1388 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1391 static noinline int cow_file_range(struct btrfs_inode *inode,
1392 struct page *locked_page,
1393 u64 start, u64 end, int *page_started,
1394 unsigned long *nr_written, int unlock,
1397 struct btrfs_root *root = inode->root;
1398 struct btrfs_fs_info *fs_info = root->fs_info;
1400 u64 orig_start = start;
1402 unsigned long ram_size;
1403 u64 cur_alloc_size = 0;
1405 u64 blocksize = fs_info->sectorsize;
1406 struct btrfs_key ins;
1407 struct extent_map *em;
1408 unsigned clear_bits;
1409 unsigned long page_ops;
1410 bool extent_reserved = false;
1413 if (btrfs_is_free_space_inode(inode)) {
1418 num_bytes = ALIGN(end - start + 1, blocksize);
1419 num_bytes = max(blocksize, num_bytes);
1420 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1422 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1425 * Due to the page size limit, for subpage we can only trigger the
1426 * writeback for the dirty sectors of page, that means data writeback
1427 * is doing more writeback than what we want.
1429 * This is especially unexpected for some call sites like fallocate,
1430 * where we only increase i_size after everything is done.
1431 * This means we can trigger inline extent even if we didn't want to.
1432 * So here we skip inline extent creation completely.
1434 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1435 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1438 /* lets try to make an inline extent */
1439 ret = cow_file_range_inline(inode, actual_end, 0,
1440 BTRFS_COMPRESS_NONE, NULL, false);
1443 * We use DO_ACCOUNTING here because we need the
1444 * delalloc_release_metadata to be run _after_ we drop
1445 * our outstanding extent for clearing delalloc for this
1448 extent_clear_unlock_delalloc(inode, start, end,
1450 EXTENT_LOCKED | EXTENT_DELALLOC |
1451 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1452 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1453 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1454 *nr_written = *nr_written +
1455 (end - start + PAGE_SIZE) / PAGE_SIZE;
1458 * locked_page is locked by the caller of
1459 * writepage_delalloc(), not locked by
1460 * __process_pages_contig().
1462 * We can't let __process_pages_contig() to unlock it,
1463 * as it doesn't have any subpage::writers recorded.
1465 * Here we manually unlock the page, since the caller
1466 * can't use page_started to determine if it's an
1467 * inline extent or a compressed extent.
1469 unlock_page(locked_page);
1471 } else if (ret < 0) {
1476 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1479 * Relocation relies on the relocated extents to have exactly the same
1480 * size as the original extents. Normally writeback for relocation data
1481 * extents follows a NOCOW path because relocation preallocates the
1482 * extents. However, due to an operation such as scrub turning a block
1483 * group to RO mode, it may fallback to COW mode, so we must make sure
1484 * an extent allocated during COW has exactly the requested size and can
1485 * not be split into smaller extents, otherwise relocation breaks and
1486 * fails during the stage where it updates the bytenr of file extent
1489 if (btrfs_is_data_reloc_root(root))
1490 min_alloc_size = num_bytes;
1492 min_alloc_size = fs_info->sectorsize;
1494 while (num_bytes > 0) {
1495 struct btrfs_ordered_extent *ordered;
1497 cur_alloc_size = num_bytes;
1498 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1499 min_alloc_size, 0, alloc_hint,
1503 cur_alloc_size = ins.offset;
1504 extent_reserved = true;
1506 ram_size = ins.offset;
1507 em = create_io_em(inode, start, ins.offset, /* len */
1508 start, /* orig_start */
1509 ins.objectid, /* block_start */
1510 ins.offset, /* block_len */
1511 ins.offset, /* orig_block_len */
1512 ram_size, /* ram_bytes */
1513 BTRFS_COMPRESS_NONE, /* compress_type */
1514 BTRFS_ORDERED_REGULAR /* type */);
1519 free_extent_map(em);
1521 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1522 ram_size, ins.objectid, cur_alloc_size,
1523 0, 1 << BTRFS_ORDERED_REGULAR,
1524 BTRFS_COMPRESS_NONE);
1525 if (IS_ERR(ordered)) {
1526 ret = PTR_ERR(ordered);
1527 goto out_drop_extent_cache;
1530 if (btrfs_is_data_reloc_root(root)) {
1531 ret = btrfs_reloc_clone_csums(ordered);
1534 * Only drop cache here, and process as normal.
1536 * We must not allow extent_clear_unlock_delalloc()
1537 * at out_unlock label to free meta of this ordered
1538 * extent, as its meta should be freed by
1539 * btrfs_finish_ordered_io().
1541 * So we must continue until @start is increased to
1542 * skip current ordered extent.
1545 btrfs_drop_extent_map_range(inode, start,
1546 start + ram_size - 1,
1549 btrfs_put_ordered_extent(ordered);
1551 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1554 * We're not doing compressed IO, don't unlock the first page
1555 * (which the caller expects to stay locked), don't clear any
1556 * dirty bits and don't set any writeback bits
1558 * Do set the Ordered (Private2) bit so we know this page was
1559 * properly setup for writepage.
1561 page_ops = unlock ? PAGE_UNLOCK : 0;
1562 page_ops |= PAGE_SET_ORDERED;
1564 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1566 EXTENT_LOCKED | EXTENT_DELALLOC,
1568 if (num_bytes < cur_alloc_size)
1571 num_bytes -= cur_alloc_size;
1572 alloc_hint = ins.objectid + ins.offset;
1573 start += cur_alloc_size;
1574 extent_reserved = false;
1577 * btrfs_reloc_clone_csums() error, since start is increased
1578 * extent_clear_unlock_delalloc() at out_unlock label won't
1579 * free metadata of current ordered extent, we're OK to exit.
1587 out_drop_extent_cache:
1588 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1590 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1591 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1594 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1595 * caller to write out the successfully allocated region and retry.
1597 if (done_offset && ret == -EAGAIN) {
1598 if (orig_start < start)
1599 *done_offset = start - 1;
1601 *done_offset = start;
1603 } else if (ret == -EAGAIN) {
1604 /* Convert to -ENOSPC since the caller cannot retry. */
1609 * Now, we have three regions to clean up:
1611 * |-------(1)----|---(2)---|-------------(3)----------|
1612 * `- orig_start `- start `- start + cur_alloc_size `- end
1614 * We process each region below.
1617 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1618 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1619 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1622 * For the range (1). We have already instantiated the ordered extents
1623 * for this region. They are cleaned up by
1624 * btrfs_cleanup_ordered_extents() in e.g,
1625 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1626 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1627 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1630 * However, in case of unlock == 0, we still need to unlock the pages
1631 * (except @locked_page) to ensure all the pages are unlocked.
1633 if (!unlock && orig_start < start) {
1635 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1636 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1637 locked_page, 0, page_ops);
1641 * For the range (2). If we reserved an extent for our delalloc range
1642 * (or a subrange) and failed to create the respective ordered extent,
1643 * then it means that when we reserved the extent we decremented the
1644 * extent's size from the data space_info's bytes_may_use counter and
1645 * incremented the space_info's bytes_reserved counter by the same
1646 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1647 * to decrement again the data space_info's bytes_may_use counter,
1648 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1650 if (extent_reserved) {
1651 extent_clear_unlock_delalloc(inode, start,
1652 start + cur_alloc_size - 1,
1656 start += cur_alloc_size;
1662 * For the range (3). We never touched the region. In addition to the
1663 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1664 * space_info's bytes_may_use counter, reserved in
1665 * btrfs_check_data_free_space().
1667 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1668 clear_bits | EXTENT_CLEAR_DATA_RESV,
1674 * work queue call back to started compression on a file and pages
1676 static noinline void async_cow_start(struct btrfs_work *work)
1678 struct async_chunk *async_chunk;
1679 int compressed_extents;
1681 async_chunk = container_of(work, struct async_chunk, work);
1683 compressed_extents = compress_file_range(async_chunk);
1684 if (compressed_extents == 0) {
1685 btrfs_add_delayed_iput(async_chunk->inode);
1686 async_chunk->inode = NULL;
1691 * work queue call back to submit previously compressed pages
1693 static noinline void async_cow_submit(struct btrfs_work *work)
1695 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1697 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1698 unsigned long nr_pages;
1700 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1704 * ->inode could be NULL if async_chunk_start has failed to compress,
1705 * in which case we don't have anything to submit, yet we need to
1706 * always adjust ->async_delalloc_pages as its paired with the init
1707 * happening in run_delalloc_compressed
1709 if (async_chunk->inode)
1710 submit_compressed_extents(async_chunk);
1712 /* atomic_sub_return implies a barrier */
1713 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1715 cond_wake_up_nomb(&fs_info->async_submit_wait);
1718 static noinline void async_cow_free(struct btrfs_work *work)
1720 struct async_chunk *async_chunk;
1721 struct async_cow *async_cow;
1723 async_chunk = container_of(work, struct async_chunk, work);
1724 if (async_chunk->inode)
1725 btrfs_add_delayed_iput(async_chunk->inode);
1726 if (async_chunk->blkcg_css)
1727 css_put(async_chunk->blkcg_css);
1729 async_cow = async_chunk->async_cow;
1730 if (atomic_dec_and_test(&async_cow->num_chunks))
1734 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1735 struct writeback_control *wbc,
1736 struct page *locked_page,
1737 u64 start, u64 end, int *page_started,
1738 unsigned long *nr_written)
1740 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1741 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1742 struct async_cow *ctx;
1743 struct async_chunk *async_chunk;
1744 unsigned long nr_pages;
1745 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1748 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1750 nofs_flag = memalloc_nofs_save();
1751 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1752 memalloc_nofs_restore(nofs_flag);
1756 unlock_extent(&inode->io_tree, start, end, NULL);
1757 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1759 async_chunk = ctx->chunks;
1760 atomic_set(&ctx->num_chunks, num_chunks);
1762 for (i = 0; i < num_chunks; i++) {
1763 u64 cur_end = min(end, start + SZ_512K - 1);
1766 * igrab is called higher up in the call chain, take only the
1767 * lightweight reference for the callback lifetime
1769 ihold(&inode->vfs_inode);
1770 async_chunk[i].async_cow = ctx;
1771 async_chunk[i].inode = inode;
1772 async_chunk[i].start = start;
1773 async_chunk[i].end = cur_end;
1774 async_chunk[i].write_flags = write_flags;
1775 INIT_LIST_HEAD(&async_chunk[i].extents);
1778 * The locked_page comes all the way from writepage and its
1779 * the original page we were actually given. As we spread
1780 * this large delalloc region across multiple async_chunk
1781 * structs, only the first struct needs a pointer to locked_page
1783 * This way we don't need racey decisions about who is supposed
1788 * Depending on the compressibility, the pages might or
1789 * might not go through async. We want all of them to
1790 * be accounted against wbc once. Let's do it here
1791 * before the paths diverge. wbc accounting is used
1792 * only for foreign writeback detection and doesn't
1793 * need full accuracy. Just account the whole thing
1794 * against the first page.
1796 wbc_account_cgroup_owner(wbc, locked_page,
1798 async_chunk[i].locked_page = locked_page;
1801 async_chunk[i].locked_page = NULL;
1804 if (blkcg_css != blkcg_root_css) {
1806 async_chunk[i].blkcg_css = blkcg_css;
1807 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1809 async_chunk[i].blkcg_css = NULL;
1812 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1813 async_cow_submit, async_cow_free);
1815 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1816 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1818 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1820 *nr_written += nr_pages;
1821 start = cur_end + 1;
1827 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1828 struct page *locked_page, u64 start,
1829 u64 end, int *page_started,
1830 unsigned long *nr_written,
1831 struct writeback_control *wbc)
1833 u64 done_offset = end;
1835 bool locked_page_done = false;
1837 while (start <= end) {
1838 ret = cow_file_range(inode, locked_page, start, end, page_started,
1839 nr_written, 0, &done_offset);
1840 if (ret && ret != -EAGAIN)
1843 if (*page_started) {
1851 if (done_offset == start) {
1852 wait_on_bit_io(&inode->root->fs_info->flags,
1853 BTRFS_FS_NEED_ZONE_FINISH,
1854 TASK_UNINTERRUPTIBLE);
1858 if (!locked_page_done) {
1859 __set_page_dirty_nobuffers(locked_page);
1860 account_page_redirty(locked_page);
1862 locked_page_done = true;
1863 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1865 start = done_offset + 1;
1873 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1874 u64 bytenr, u64 num_bytes, bool nowait)
1876 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1877 struct btrfs_ordered_sum *sums;
1881 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1883 if (ret == 0 && list_empty(&list))
1886 while (!list_empty(&list)) {
1887 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1888 list_del(&sums->list);
1896 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1897 const u64 start, const u64 end,
1898 int *page_started, unsigned long *nr_written)
1900 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1901 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1902 const u64 range_bytes = end + 1 - start;
1903 struct extent_io_tree *io_tree = &inode->io_tree;
1904 u64 range_start = start;
1908 * If EXTENT_NORESERVE is set it means that when the buffered write was
1909 * made we had not enough available data space and therefore we did not
1910 * reserve data space for it, since we though we could do NOCOW for the
1911 * respective file range (either there is prealloc extent or the inode
1912 * has the NOCOW bit set).
1914 * However when we need to fallback to COW mode (because for example the
1915 * block group for the corresponding extent was turned to RO mode by a
1916 * scrub or relocation) we need to do the following:
1918 * 1) We increment the bytes_may_use counter of the data space info.
1919 * If COW succeeds, it allocates a new data extent and after doing
1920 * that it decrements the space info's bytes_may_use counter and
1921 * increments its bytes_reserved counter by the same amount (we do
1922 * this at btrfs_add_reserved_bytes()). So we need to increment the
1923 * bytes_may_use counter to compensate (when space is reserved at
1924 * buffered write time, the bytes_may_use counter is incremented);
1926 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1927 * that if the COW path fails for any reason, it decrements (through
1928 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1929 * data space info, which we incremented in the step above.
1931 * If we need to fallback to cow and the inode corresponds to a free
1932 * space cache inode or an inode of the data relocation tree, we must
1933 * also increment bytes_may_use of the data space_info for the same
1934 * reason. Space caches and relocated data extents always get a prealloc
1935 * extent for them, however scrub or balance may have set the block
1936 * group that contains that extent to RO mode and therefore force COW
1937 * when starting writeback.
1939 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1940 EXTENT_NORESERVE, 0, NULL);
1941 if (count > 0 || is_space_ino || is_reloc_ino) {
1943 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1944 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1946 if (is_space_ino || is_reloc_ino)
1947 bytes = range_bytes;
1949 spin_lock(&sinfo->lock);
1950 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1951 spin_unlock(&sinfo->lock);
1954 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1958 return cow_file_range(inode, locked_page, start, end, page_started,
1959 nr_written, 1, NULL);
1962 struct can_nocow_file_extent_args {
1965 /* Start file offset of the range we want to NOCOW. */
1967 /* End file offset (inclusive) of the range we want to NOCOW. */
1969 bool writeback_path;
1972 * Free the path passed to can_nocow_file_extent() once it's not needed
1977 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1982 /* Number of bytes that can be written to in NOCOW mode. */
1987 * Check if we can NOCOW the file extent that the path points to.
1988 * This function may return with the path released, so the caller should check
1989 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1991 * Returns: < 0 on error
1992 * 0 if we can not NOCOW
1995 static int can_nocow_file_extent(struct btrfs_path *path,
1996 struct btrfs_key *key,
1997 struct btrfs_inode *inode,
1998 struct can_nocow_file_extent_args *args)
2000 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
2001 struct extent_buffer *leaf = path->nodes[0];
2002 struct btrfs_root *root = inode->root;
2003 struct btrfs_file_extent_item *fi;
2008 bool nowait = path->nowait;
2010 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2011 extent_type = btrfs_file_extent_type(leaf, fi);
2013 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
2016 /* Can't access these fields unless we know it's not an inline extent. */
2017 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2018 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
2019 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
2021 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
2022 extent_type == BTRFS_FILE_EXTENT_REG)
2026 * If the extent was created before the generation where the last snapshot
2027 * for its subvolume was created, then this implies the extent is shared,
2028 * hence we must COW.
2030 if (!args->strict &&
2031 btrfs_file_extent_generation(leaf, fi) <=
2032 btrfs_root_last_snapshot(&root->root_item))
2035 /* An explicit hole, must COW. */
2036 if (args->disk_bytenr == 0)
2039 /* Compressed/encrypted/encoded extents must be COWed. */
2040 if (btrfs_file_extent_compression(leaf, fi) ||
2041 btrfs_file_extent_encryption(leaf, fi) ||
2042 btrfs_file_extent_other_encoding(leaf, fi))
2045 extent_end = btrfs_file_extent_end(path);
2048 * The following checks can be expensive, as they need to take other
2049 * locks and do btree or rbtree searches, so release the path to avoid
2050 * blocking other tasks for too long.
2052 btrfs_release_path(path);
2054 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
2055 key->offset - args->extent_offset,
2056 args->disk_bytenr, args->strict, path);
2057 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2061 if (args->free_path) {
2063 * We don't need the path anymore, plus through the
2064 * csum_exist_in_range() call below we will end up allocating
2065 * another path. So free the path to avoid unnecessary extra
2068 btrfs_free_path(path);
2072 /* If there are pending snapshots for this root, we must COW. */
2073 if (args->writeback_path && !is_freespace_inode &&
2074 atomic_read(&root->snapshot_force_cow))
2077 args->disk_bytenr += args->extent_offset;
2078 args->disk_bytenr += args->start - key->offset;
2079 args->num_bytes = min(args->end + 1, extent_end) - args->start;
2082 * Force COW if csums exist in the range. This ensures that csums for a
2083 * given extent are either valid or do not exist.
2085 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
2087 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2093 if (args->free_path && path)
2094 btrfs_free_path(path);
2096 return ret < 0 ? ret : can_nocow;
2100 * when nowcow writeback call back. This checks for snapshots or COW copies
2101 * of the extents that exist in the file, and COWs the file as required.
2103 * If no cow copies or snapshots exist, we write directly to the existing
2106 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2107 struct page *locked_page,
2108 const u64 start, const u64 end,
2110 unsigned long *nr_written)
2112 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2113 struct btrfs_root *root = inode->root;
2114 struct btrfs_path *path;
2115 u64 cow_start = (u64)-1;
2116 u64 cur_offset = start;
2118 bool check_prev = true;
2119 u64 ino = btrfs_ino(inode);
2120 struct btrfs_block_group *bg;
2122 struct can_nocow_file_extent_args nocow_args = { 0 };
2124 path = btrfs_alloc_path();
2126 extent_clear_unlock_delalloc(inode, start, end, locked_page,
2127 EXTENT_LOCKED | EXTENT_DELALLOC |
2128 EXTENT_DO_ACCOUNTING |
2129 EXTENT_DEFRAG, PAGE_UNLOCK |
2130 PAGE_START_WRITEBACK |
2131 PAGE_END_WRITEBACK);
2135 nocow_args.end = end;
2136 nocow_args.writeback_path = true;
2139 struct btrfs_ordered_extent *ordered;
2140 struct btrfs_key found_key;
2141 struct btrfs_file_extent_item *fi;
2142 struct extent_buffer *leaf;
2151 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2157 * If there is no extent for our range when doing the initial
2158 * search, then go back to the previous slot as it will be the
2159 * one containing the search offset
2161 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2162 leaf = path->nodes[0];
2163 btrfs_item_key_to_cpu(leaf, &found_key,
2164 path->slots[0] - 1);
2165 if (found_key.objectid == ino &&
2166 found_key.type == BTRFS_EXTENT_DATA_KEY)
2171 /* Go to next leaf if we have exhausted the current one */
2172 leaf = path->nodes[0];
2173 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2174 ret = btrfs_next_leaf(root, path);
2176 if (cow_start != (u64)-1)
2177 cur_offset = cow_start;
2182 leaf = path->nodes[0];
2185 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2187 /* Didn't find anything for our INO */
2188 if (found_key.objectid > ino)
2191 * Keep searching until we find an EXTENT_ITEM or there are no
2192 * more extents for this inode
2194 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2195 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2200 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2201 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2202 found_key.offset > end)
2206 * If the found extent starts after requested offset, then
2207 * adjust extent_end to be right before this extent begins
2209 if (found_key.offset > cur_offset) {
2210 extent_end = found_key.offset;
2216 * Found extent which begins before our range and potentially
2219 fi = btrfs_item_ptr(leaf, path->slots[0],
2220 struct btrfs_file_extent_item);
2221 extent_type = btrfs_file_extent_type(leaf, fi);
2222 /* If this is triggered then we have a memory corruption. */
2223 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2224 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2228 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2229 extent_end = btrfs_file_extent_end(path);
2232 * If the extent we got ends before our current offset, skip to
2235 if (extent_end <= cur_offset) {
2240 nocow_args.start = cur_offset;
2241 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2243 if (cow_start != (u64)-1)
2244 cur_offset = cow_start;
2246 } else if (ret == 0) {
2251 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2256 * If nocow is false then record the beginning of the range
2257 * that needs to be COWed
2260 if (cow_start == (u64)-1)
2261 cow_start = cur_offset;
2262 cur_offset = extent_end;
2263 if (cur_offset > end)
2265 if (!path->nodes[0])
2272 * COW range from cow_start to found_key.offset - 1. As the key
2273 * will contain the beginning of the first extent that can be
2274 * NOCOW, following one which needs to be COW'ed
2276 if (cow_start != (u64)-1) {
2277 ret = fallback_to_cow(inode, locked_page,
2278 cow_start, found_key.offset - 1,
2279 page_started, nr_written);
2282 cow_start = (u64)-1;
2285 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2286 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2288 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2289 struct extent_map *em;
2291 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2293 nocow_args.disk_bytenr, /* block_start */
2294 nocow_args.num_bytes, /* block_len */
2295 nocow_args.disk_num_bytes, /* orig_block_len */
2296 ram_bytes, BTRFS_COMPRESS_NONE,
2297 BTRFS_ORDERED_PREALLOC);
2302 free_extent_map(em);
2305 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2306 nocow_args.num_bytes, nocow_args.num_bytes,
2307 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2309 ? (1 << BTRFS_ORDERED_PREALLOC)
2310 : (1 << BTRFS_ORDERED_NOCOW),
2311 BTRFS_COMPRESS_NONE);
2312 if (IS_ERR(ordered)) {
2314 btrfs_drop_extent_map_range(inode, cur_offset,
2317 ret = PTR_ERR(ordered);
2322 btrfs_dec_nocow_writers(bg);
2326 if (btrfs_is_data_reloc_root(root))
2328 * Error handled later, as we must prevent
2329 * extent_clear_unlock_delalloc() in error handler
2330 * from freeing metadata of created ordered extent.
2332 ret = btrfs_reloc_clone_csums(ordered);
2333 btrfs_put_ordered_extent(ordered);
2335 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2336 locked_page, EXTENT_LOCKED |
2338 EXTENT_CLEAR_DATA_RESV,
2339 PAGE_UNLOCK | PAGE_SET_ORDERED);
2341 cur_offset = extent_end;
2344 * btrfs_reloc_clone_csums() error, now we're OK to call error
2345 * handler, as metadata for created ordered extent will only
2346 * be freed by btrfs_finish_ordered_io().
2350 if (cur_offset > end)
2353 btrfs_release_path(path);
2355 if (cur_offset <= end && cow_start == (u64)-1)
2356 cow_start = cur_offset;
2358 if (cow_start != (u64)-1) {
2360 ret = fallback_to_cow(inode, locked_page, cow_start, end,
2361 page_started, nr_written);
2368 btrfs_dec_nocow_writers(bg);
2370 if (ret && cur_offset < end)
2371 extent_clear_unlock_delalloc(inode, cur_offset, end,
2372 locked_page, EXTENT_LOCKED |
2373 EXTENT_DELALLOC | EXTENT_DEFRAG |
2374 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2375 PAGE_START_WRITEBACK |
2376 PAGE_END_WRITEBACK);
2377 btrfs_free_path(path);
2381 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2383 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2384 if (inode->defrag_bytes &&
2385 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2394 * Function to process delayed allocation (create CoW) for ranges which are
2395 * being touched for the first time.
2397 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2398 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2399 struct writeback_control *wbc)
2402 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2405 * The range must cover part of the @locked_page, or the returned
2406 * @page_started can confuse the caller.
2408 ASSERT(!(end <= page_offset(locked_page) ||
2409 start >= page_offset(locked_page) + PAGE_SIZE));
2411 if (should_nocow(inode, start, end)) {
2413 * Normally on a zoned device we're only doing COW writes, but
2414 * in case of relocation on a zoned filesystem we have taken
2415 * precaution, that we're only writing sequentially. It's safe
2416 * to use run_delalloc_nocow() here, like for regular
2417 * preallocated inodes.
2419 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2420 ret = run_delalloc_nocow(inode, locked_page, start, end,
2421 page_started, nr_written);
2425 if (btrfs_inode_can_compress(inode) &&
2426 inode_need_compress(inode, start, end) &&
2427 run_delalloc_compressed(inode, wbc, locked_page, start,
2428 end, page_started, nr_written))
2432 ret = run_delalloc_zoned(inode, locked_page, start, end,
2433 page_started, nr_written, wbc);
2435 ret = cow_file_range(inode, locked_page, start, end,
2436 page_started, nr_written, 1, NULL);
2441 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2446 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2447 struct extent_state *orig, u64 split)
2449 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2452 /* not delalloc, ignore it */
2453 if (!(orig->state & EXTENT_DELALLOC))
2456 size = orig->end - orig->start + 1;
2457 if (size > fs_info->max_extent_size) {
2462 * See the explanation in btrfs_merge_delalloc_extent, the same
2463 * applies here, just in reverse.
2465 new_size = orig->end - split + 1;
2466 num_extents = count_max_extents(fs_info, new_size);
2467 new_size = split - orig->start;
2468 num_extents += count_max_extents(fs_info, new_size);
2469 if (count_max_extents(fs_info, size) >= num_extents)
2473 spin_lock(&inode->lock);
2474 btrfs_mod_outstanding_extents(inode, 1);
2475 spin_unlock(&inode->lock);
2479 * Handle merged delayed allocation extents so we can keep track of new extents
2480 * that are just merged onto old extents, such as when we are doing sequential
2481 * writes, so we can properly account for the metadata space we'll need.
2483 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2484 struct extent_state *other)
2486 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2487 u64 new_size, old_size;
2490 /* not delalloc, ignore it */
2491 if (!(other->state & EXTENT_DELALLOC))
2494 if (new->start > other->start)
2495 new_size = new->end - other->start + 1;
2497 new_size = other->end - new->start + 1;
2499 /* we're not bigger than the max, unreserve the space and go */
2500 if (new_size <= fs_info->max_extent_size) {
2501 spin_lock(&inode->lock);
2502 btrfs_mod_outstanding_extents(inode, -1);
2503 spin_unlock(&inode->lock);
2508 * We have to add up either side to figure out how many extents were
2509 * accounted for before we merged into one big extent. If the number of
2510 * extents we accounted for is <= the amount we need for the new range
2511 * then we can return, otherwise drop. Think of it like this
2515 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2516 * need 2 outstanding extents, on one side we have 1 and the other side
2517 * we have 1 so they are == and we can return. But in this case
2519 * [MAX_SIZE+4k][MAX_SIZE+4k]
2521 * Each range on their own accounts for 2 extents, but merged together
2522 * they are only 3 extents worth of accounting, so we need to drop in
2525 old_size = other->end - other->start + 1;
2526 num_extents = count_max_extents(fs_info, old_size);
2527 old_size = new->end - new->start + 1;
2528 num_extents += count_max_extents(fs_info, old_size);
2529 if (count_max_extents(fs_info, new_size) >= num_extents)
2532 spin_lock(&inode->lock);
2533 btrfs_mod_outstanding_extents(inode, -1);
2534 spin_unlock(&inode->lock);
2537 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2538 struct btrfs_inode *inode)
2540 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2542 spin_lock(&root->delalloc_lock);
2543 if (list_empty(&inode->delalloc_inodes)) {
2544 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2545 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2546 root->nr_delalloc_inodes++;
2547 if (root->nr_delalloc_inodes == 1) {
2548 spin_lock(&fs_info->delalloc_root_lock);
2549 BUG_ON(!list_empty(&root->delalloc_root));
2550 list_add_tail(&root->delalloc_root,
2551 &fs_info->delalloc_roots);
2552 spin_unlock(&fs_info->delalloc_root_lock);
2555 spin_unlock(&root->delalloc_lock);
2558 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2559 struct btrfs_inode *inode)
2561 struct btrfs_fs_info *fs_info = root->fs_info;
2563 if (!list_empty(&inode->delalloc_inodes)) {
2564 list_del_init(&inode->delalloc_inodes);
2565 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2566 &inode->runtime_flags);
2567 root->nr_delalloc_inodes--;
2568 if (!root->nr_delalloc_inodes) {
2569 ASSERT(list_empty(&root->delalloc_inodes));
2570 spin_lock(&fs_info->delalloc_root_lock);
2571 BUG_ON(list_empty(&root->delalloc_root));
2572 list_del_init(&root->delalloc_root);
2573 spin_unlock(&fs_info->delalloc_root_lock);
2578 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2579 struct btrfs_inode *inode)
2581 spin_lock(&root->delalloc_lock);
2582 __btrfs_del_delalloc_inode(root, inode);
2583 spin_unlock(&root->delalloc_lock);
2587 * Properly track delayed allocation bytes in the inode and to maintain the
2588 * list of inodes that have pending delalloc work to be done.
2590 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2593 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2595 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2598 * set_bit and clear bit hooks normally require _irqsave/restore
2599 * but in this case, we are only testing for the DELALLOC
2600 * bit, which is only set or cleared with irqs on
2602 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2603 struct btrfs_root *root = inode->root;
2604 u64 len = state->end + 1 - state->start;
2605 u32 num_extents = count_max_extents(fs_info, len);
2606 bool do_list = !btrfs_is_free_space_inode(inode);
2608 spin_lock(&inode->lock);
2609 btrfs_mod_outstanding_extents(inode, num_extents);
2610 spin_unlock(&inode->lock);
2612 /* For sanity tests */
2613 if (btrfs_is_testing(fs_info))
2616 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2617 fs_info->delalloc_batch);
2618 spin_lock(&inode->lock);
2619 inode->delalloc_bytes += len;
2620 if (bits & EXTENT_DEFRAG)
2621 inode->defrag_bytes += len;
2622 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2623 &inode->runtime_flags))
2624 btrfs_add_delalloc_inodes(root, inode);
2625 spin_unlock(&inode->lock);
2628 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2629 (bits & EXTENT_DELALLOC_NEW)) {
2630 spin_lock(&inode->lock);
2631 inode->new_delalloc_bytes += state->end + 1 - state->start;
2632 spin_unlock(&inode->lock);
2637 * Once a range is no longer delalloc this function ensures that proper
2638 * accounting happens.
2640 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2641 struct extent_state *state, u32 bits)
2643 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2644 u64 len = state->end + 1 - state->start;
2645 u32 num_extents = count_max_extents(fs_info, len);
2647 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2648 spin_lock(&inode->lock);
2649 inode->defrag_bytes -= len;
2650 spin_unlock(&inode->lock);
2654 * set_bit and clear bit hooks normally require _irqsave/restore
2655 * but in this case, we are only testing for the DELALLOC
2656 * bit, which is only set or cleared with irqs on
2658 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2659 struct btrfs_root *root = inode->root;
2660 bool do_list = !btrfs_is_free_space_inode(inode);
2662 spin_lock(&inode->lock);
2663 btrfs_mod_outstanding_extents(inode, -num_extents);
2664 spin_unlock(&inode->lock);
2667 * We don't reserve metadata space for space cache inodes so we
2668 * don't need to call delalloc_release_metadata if there is an
2671 if (bits & EXTENT_CLEAR_META_RESV &&
2672 root != fs_info->tree_root)
2673 btrfs_delalloc_release_metadata(inode, len, false);
2675 /* For sanity tests. */
2676 if (btrfs_is_testing(fs_info))
2679 if (!btrfs_is_data_reloc_root(root) &&
2680 do_list && !(state->state & EXTENT_NORESERVE) &&
2681 (bits & EXTENT_CLEAR_DATA_RESV))
2682 btrfs_free_reserved_data_space_noquota(fs_info, len);
2684 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2685 fs_info->delalloc_batch);
2686 spin_lock(&inode->lock);
2687 inode->delalloc_bytes -= len;
2688 if (do_list && inode->delalloc_bytes == 0 &&
2689 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2690 &inode->runtime_flags))
2691 btrfs_del_delalloc_inode(root, inode);
2692 spin_unlock(&inode->lock);
2695 if ((state->state & EXTENT_DELALLOC_NEW) &&
2696 (bits & EXTENT_DELALLOC_NEW)) {
2697 spin_lock(&inode->lock);
2698 ASSERT(inode->new_delalloc_bytes >= len);
2699 inode->new_delalloc_bytes -= len;
2700 if (bits & EXTENT_ADD_INODE_BYTES)
2701 inode_add_bytes(&inode->vfs_inode, len);
2702 spin_unlock(&inode->lock);
2706 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2707 struct btrfs_ordered_extent *ordered)
2709 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2710 u64 len = bbio->bio.bi_iter.bi_size;
2711 struct btrfs_ordered_extent *new;
2714 /* Must always be called for the beginning of an ordered extent. */
2715 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2718 /* No need to split if the ordered extent covers the entire bio. */
2719 if (ordered->disk_num_bytes == len) {
2720 refcount_inc(&ordered->refs);
2721 bbio->ordered = ordered;
2726 * Don't split the extent_map for NOCOW extents, as we're writing into
2727 * a pre-existing one.
2729 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2730 ret = split_extent_map(bbio->inode, bbio->file_offset,
2731 ordered->num_bytes, len,
2732 ordered->disk_bytenr);
2737 new = btrfs_split_ordered_extent(ordered, len);
2739 return PTR_ERR(new);
2740 bbio->ordered = new;
2745 * given a list of ordered sums record them in the inode. This happens
2746 * at IO completion time based on sums calculated at bio submission time.
2748 static int add_pending_csums(struct btrfs_trans_handle *trans,
2749 struct list_head *list)
2751 struct btrfs_ordered_sum *sum;
2752 struct btrfs_root *csum_root = NULL;
2755 list_for_each_entry(sum, list, list) {
2756 trans->adding_csums = true;
2758 csum_root = btrfs_csum_root(trans->fs_info,
2760 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2761 trans->adding_csums = false;
2768 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2771 struct extent_state **cached_state)
2773 u64 search_start = start;
2774 const u64 end = start + len - 1;
2776 while (search_start < end) {
2777 const u64 search_len = end - search_start + 1;
2778 struct extent_map *em;
2782 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2786 if (em->block_start != EXTENT_MAP_HOLE)
2790 if (em->start < search_start)
2791 em_len -= search_start - em->start;
2792 if (em_len > search_len)
2793 em_len = search_len;
2795 ret = set_extent_bit(&inode->io_tree, search_start,
2796 search_start + em_len - 1,
2797 EXTENT_DELALLOC_NEW, cached_state);
2799 search_start = extent_map_end(em);
2800 free_extent_map(em);
2807 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2808 unsigned int extra_bits,
2809 struct extent_state **cached_state)
2811 WARN_ON(PAGE_ALIGNED(end));
2813 if (start >= i_size_read(&inode->vfs_inode) &&
2814 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2816 * There can't be any extents following eof in this case so just
2817 * set the delalloc new bit for the range directly.
2819 extra_bits |= EXTENT_DELALLOC_NEW;
2823 ret = btrfs_find_new_delalloc_bytes(inode, start,
2830 return set_extent_bit(&inode->io_tree, start, end,
2831 EXTENT_DELALLOC | extra_bits, cached_state);
2834 /* see btrfs_writepage_start_hook for details on why this is required */
2835 struct btrfs_writepage_fixup {
2837 struct btrfs_inode *inode;
2838 struct btrfs_work work;
2841 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2843 struct btrfs_writepage_fixup *fixup;
2844 struct btrfs_ordered_extent *ordered;
2845 struct extent_state *cached_state = NULL;
2846 struct extent_changeset *data_reserved = NULL;
2848 struct btrfs_inode *inode;
2852 bool free_delalloc_space = true;
2854 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2856 inode = fixup->inode;
2857 page_start = page_offset(page);
2858 page_end = page_offset(page) + PAGE_SIZE - 1;
2861 * This is similar to page_mkwrite, we need to reserve the space before
2862 * we take the page lock.
2864 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2870 * Before we queued this fixup, we took a reference on the page.
2871 * page->mapping may go NULL, but it shouldn't be moved to a different
2874 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2876 * Unfortunately this is a little tricky, either
2878 * 1) We got here and our page had already been dealt with and
2879 * we reserved our space, thus ret == 0, so we need to just
2880 * drop our space reservation and bail. This can happen the
2881 * first time we come into the fixup worker, or could happen
2882 * while waiting for the ordered extent.
2883 * 2) Our page was already dealt with, but we happened to get an
2884 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2885 * this case we obviously don't have anything to release, but
2886 * because the page was already dealt with we don't want to
2887 * mark the page with an error, so make sure we're resetting
2888 * ret to 0. This is why we have this check _before_ the ret
2889 * check, because we do not want to have a surprise ENOSPC
2890 * when the page was already properly dealt with.
2893 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2894 btrfs_delalloc_release_space(inode, data_reserved,
2895 page_start, PAGE_SIZE,
2903 * We can't mess with the page state unless it is locked, so now that
2904 * it is locked bail if we failed to make our space reservation.
2909 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2911 /* already ordered? We're done */
2912 if (PageOrdered(page))
2915 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2917 unlock_extent(&inode->io_tree, page_start, page_end,
2920 btrfs_start_ordered_extent(ordered);
2921 btrfs_put_ordered_extent(ordered);
2925 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2931 * Everything went as planned, we're now the owner of a dirty page with
2932 * delayed allocation bits set and space reserved for our COW
2935 * The page was dirty when we started, nothing should have cleaned it.
2937 BUG_ON(!PageDirty(page));
2938 free_delalloc_space = false;
2940 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2941 if (free_delalloc_space)
2942 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2944 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2948 * We hit ENOSPC or other errors. Update the mapping and page
2949 * to reflect the errors and clean the page.
2951 mapping_set_error(page->mapping, ret);
2952 end_extent_writepage(page, ret, page_start, page_end);
2953 clear_page_dirty_for_io(page);
2955 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2959 extent_changeset_free(data_reserved);
2961 * As a precaution, do a delayed iput in case it would be the last iput
2962 * that could need flushing space. Recursing back to fixup worker would
2965 btrfs_add_delayed_iput(inode);
2969 * There are a few paths in the higher layers of the kernel that directly
2970 * set the page dirty bit without asking the filesystem if it is a
2971 * good idea. This causes problems because we want to make sure COW
2972 * properly happens and the data=ordered rules are followed.
2974 * In our case any range that doesn't have the ORDERED bit set
2975 * hasn't been properly setup for IO. We kick off an async process
2976 * to fix it up. The async helper will wait for ordered extents, set
2977 * the delalloc bit and make it safe to write the page.
2979 int btrfs_writepage_cow_fixup(struct page *page)
2981 struct inode *inode = page->mapping->host;
2982 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2983 struct btrfs_writepage_fixup *fixup;
2985 /* This page has ordered extent covering it already */
2986 if (PageOrdered(page))
2990 * PageChecked is set below when we create a fixup worker for this page,
2991 * don't try to create another one if we're already PageChecked()
2993 * The extent_io writepage code will redirty the page if we send back
2996 if (PageChecked(page))
2999 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
3004 * We are already holding a reference to this inode from
3005 * write_cache_pages. We need to hold it because the space reservation
3006 * takes place outside of the page lock, and we can't trust
3007 * page->mapping outside of the page lock.
3010 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3012 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3014 fixup->inode = BTRFS_I(inode);
3015 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3020 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3021 struct btrfs_inode *inode, u64 file_pos,
3022 struct btrfs_file_extent_item *stack_fi,
3023 const bool update_inode_bytes,
3024 u64 qgroup_reserved)
3026 struct btrfs_root *root = inode->root;
3027 const u64 sectorsize = root->fs_info->sectorsize;
3028 struct btrfs_path *path;
3029 struct extent_buffer *leaf;
3030 struct btrfs_key ins;
3031 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3032 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3033 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3034 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3035 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3036 struct btrfs_drop_extents_args drop_args = { 0 };
3039 path = btrfs_alloc_path();
3044 * we may be replacing one extent in the tree with another.
3045 * The new extent is pinned in the extent map, and we don't want
3046 * to drop it from the cache until it is completely in the btree.
3048 * So, tell btrfs_drop_extents to leave this extent in the cache.
3049 * the caller is expected to unpin it and allow it to be merged
3052 drop_args.path = path;
3053 drop_args.start = file_pos;
3054 drop_args.end = file_pos + num_bytes;
3055 drop_args.replace_extent = true;
3056 drop_args.extent_item_size = sizeof(*stack_fi);
3057 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3061 if (!drop_args.extent_inserted) {
3062 ins.objectid = btrfs_ino(inode);
3063 ins.offset = file_pos;
3064 ins.type = BTRFS_EXTENT_DATA_KEY;
3066 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3071 leaf = path->nodes[0];
3072 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3073 write_extent_buffer(leaf, stack_fi,
3074 btrfs_item_ptr_offset(leaf, path->slots[0]),
3075 sizeof(struct btrfs_file_extent_item));
3077 btrfs_mark_buffer_dirty(leaf);
3078 btrfs_release_path(path);
3081 * If we dropped an inline extent here, we know the range where it is
3082 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3083 * number of bytes only for that range containing the inline extent.
3084 * The remaining of the range will be processed when clearning the
3085 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3087 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3088 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3090 inline_size = drop_args.bytes_found - inline_size;
3091 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3092 drop_args.bytes_found -= inline_size;
3093 num_bytes -= sectorsize;
3096 if (update_inode_bytes)
3097 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3099 ins.objectid = disk_bytenr;
3100 ins.offset = disk_num_bytes;
3101 ins.type = BTRFS_EXTENT_ITEM_KEY;
3103 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3107 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3109 qgroup_reserved, &ins);
3111 btrfs_free_path(path);
3116 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3119 struct btrfs_block_group *cache;
3121 cache = btrfs_lookup_block_group(fs_info, start);
3124 spin_lock(&cache->lock);
3125 cache->delalloc_bytes -= len;
3126 spin_unlock(&cache->lock);
3128 btrfs_put_block_group(cache);
3131 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3132 struct btrfs_ordered_extent *oe)
3134 struct btrfs_file_extent_item stack_fi;
3135 bool update_inode_bytes;
3136 u64 num_bytes = oe->num_bytes;
3137 u64 ram_bytes = oe->ram_bytes;
3139 memset(&stack_fi, 0, sizeof(stack_fi));
3140 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3141 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3142 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3143 oe->disk_num_bytes);
3144 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3145 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3146 num_bytes = oe->truncated_len;
3147 ram_bytes = num_bytes;
3149 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3150 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3151 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3152 /* Encryption and other encoding is reserved and all 0 */
3155 * For delalloc, when completing an ordered extent we update the inode's
3156 * bytes when clearing the range in the inode's io tree, so pass false
3157 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3158 * except if the ordered extent was truncated.
3160 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3161 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3162 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3164 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3165 oe->file_offset, &stack_fi,
3166 update_inode_bytes, oe->qgroup_rsv);
3170 * As ordered data IO finishes, this gets called so we can finish
3171 * an ordered extent if the range of bytes in the file it covers are
3174 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3176 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3177 struct btrfs_root *root = inode->root;
3178 struct btrfs_fs_info *fs_info = root->fs_info;
3179 struct btrfs_trans_handle *trans = NULL;
3180 struct extent_io_tree *io_tree = &inode->io_tree;
3181 struct extent_state *cached_state = NULL;
3183 int compress_type = 0;
3185 u64 logical_len = ordered_extent->num_bytes;
3186 bool freespace_inode;
3187 bool truncated = false;
3188 bool clear_reserved_extent = true;
3189 unsigned int clear_bits = EXTENT_DEFRAG;
3191 start = ordered_extent->file_offset;
3192 end = start + ordered_extent->num_bytes - 1;
3194 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3195 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3196 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3197 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3198 clear_bits |= EXTENT_DELALLOC_NEW;
3200 freespace_inode = btrfs_is_free_space_inode(inode);
3201 if (!freespace_inode)
3202 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3204 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3209 if (btrfs_is_zoned(fs_info))
3210 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3211 ordered_extent->disk_num_bytes);
3213 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3215 logical_len = ordered_extent->truncated_len;
3216 /* Truncated the entire extent, don't bother adding */
3221 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3222 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3224 btrfs_inode_safe_disk_i_size_write(inode, 0);
3225 if (freespace_inode)
3226 trans = btrfs_join_transaction_spacecache(root);
3228 trans = btrfs_join_transaction(root);
3229 if (IS_ERR(trans)) {
3230 ret = PTR_ERR(trans);
3234 trans->block_rsv = &inode->block_rsv;
3235 ret = btrfs_update_inode_fallback(trans, root, inode);
3236 if (ret) /* -ENOMEM or corruption */
3237 btrfs_abort_transaction(trans, ret);
3241 clear_bits |= EXTENT_LOCKED;
3242 lock_extent(io_tree, start, end, &cached_state);
3244 if (freespace_inode)
3245 trans = btrfs_join_transaction_spacecache(root);
3247 trans = btrfs_join_transaction(root);
3248 if (IS_ERR(trans)) {
3249 ret = PTR_ERR(trans);
3254 trans->block_rsv = &inode->block_rsv;
3256 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3257 compress_type = ordered_extent->compress_type;
3258 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3259 BUG_ON(compress_type);
3260 ret = btrfs_mark_extent_written(trans, inode,
3261 ordered_extent->file_offset,
3262 ordered_extent->file_offset +
3264 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3265 ordered_extent->disk_num_bytes);
3267 BUG_ON(root == fs_info->tree_root);
3268 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3270 clear_reserved_extent = false;
3271 btrfs_release_delalloc_bytes(fs_info,
3272 ordered_extent->disk_bytenr,
3273 ordered_extent->disk_num_bytes);
3276 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3277 ordered_extent->num_bytes, trans->transid);
3279 btrfs_abort_transaction(trans, ret);
3283 ret = add_pending_csums(trans, &ordered_extent->list);
3285 btrfs_abort_transaction(trans, ret);
3290 * If this is a new delalloc range, clear its new delalloc flag to
3291 * update the inode's number of bytes. This needs to be done first
3292 * before updating the inode item.
3294 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3295 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3296 clear_extent_bit(&inode->io_tree, start, end,
3297 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3300 btrfs_inode_safe_disk_i_size_write(inode, 0);
3301 ret = btrfs_update_inode_fallback(trans, root, inode);
3302 if (ret) { /* -ENOMEM or corruption */
3303 btrfs_abort_transaction(trans, ret);
3308 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3312 btrfs_end_transaction(trans);
3314 if (ret || truncated) {
3315 u64 unwritten_start = start;
3318 * If we failed to finish this ordered extent for any reason we
3319 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3320 * extent, and mark the inode with the error if it wasn't
3321 * already set. Any error during writeback would have already
3322 * set the mapping error, so we need to set it if we're the ones
3323 * marking this ordered extent as failed.
3325 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3326 &ordered_extent->flags))
3327 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3330 unwritten_start += logical_len;
3331 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3333 /* Drop extent maps for the part of the extent we didn't write. */
3334 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3337 * If the ordered extent had an IOERR or something else went
3338 * wrong we need to return the space for this ordered extent
3339 * back to the allocator. We only free the extent in the
3340 * truncated case if we didn't write out the extent at all.
3342 * If we made it past insert_reserved_file_extent before we
3343 * errored out then we don't need to do this as the accounting
3344 * has already been done.
3346 if ((ret || !logical_len) &&
3347 clear_reserved_extent &&
3348 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3349 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3351 * Discard the range before returning it back to the
3354 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3355 btrfs_discard_extent(fs_info,
3356 ordered_extent->disk_bytenr,
3357 ordered_extent->disk_num_bytes,
3359 btrfs_free_reserved_extent(fs_info,
3360 ordered_extent->disk_bytenr,
3361 ordered_extent->disk_num_bytes, 1);
3366 * This needs to be done to make sure anybody waiting knows we are done
3367 * updating everything for this ordered extent.
3369 btrfs_remove_ordered_extent(inode, ordered_extent);
3372 btrfs_put_ordered_extent(ordered_extent);
3373 /* once for the tree */
3374 btrfs_put_ordered_extent(ordered_extent);
3379 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3381 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3382 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3383 btrfs_finish_ordered_zoned(ordered);
3384 return btrfs_finish_one_ordered(ordered);
3387 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3388 struct page *page, u64 start,
3389 u64 end, bool uptodate)
3391 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3393 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate);
3397 * Verify the checksum for a single sector without any extra action that depend
3398 * on the type of I/O.
3400 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3401 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3403 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3406 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3408 shash->tfm = fs_info->csum_shash;
3410 kaddr = kmap_local_page(page) + pgoff;
3411 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3412 kunmap_local(kaddr);
3414 if (memcmp(csum, csum_expected, fs_info->csum_size))
3420 * Verify the checksum of a single data sector.
3422 * @bbio: btrfs_io_bio which contains the csum
3423 * @dev: device the sector is on
3424 * @bio_offset: offset to the beginning of the bio (in bytes)
3425 * @bv: bio_vec to check
3427 * Check if the checksum on a data block is valid. When a checksum mismatch is
3428 * detected, report the error and fill the corrupted range with zero.
3430 * Return %true if the sector is ok or had no checksum to start with, else %false.
3432 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3433 u32 bio_offset, struct bio_vec *bv)
3435 struct btrfs_inode *inode = bbio->inode;
3436 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3437 u64 file_offset = bbio->file_offset + bio_offset;
3438 u64 end = file_offset + bv->bv_len - 1;
3440 u8 csum[BTRFS_CSUM_SIZE];
3442 ASSERT(bv->bv_len == fs_info->sectorsize);
3447 if (btrfs_is_data_reloc_root(inode->root) &&
3448 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3450 /* Skip the range without csum for data reloc inode */
3451 clear_extent_bits(&inode->io_tree, file_offset, end,
3456 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3458 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3464 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3467 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3473 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3475 * @inode: The inode we want to perform iput on
3477 * This function uses the generic vfs_inode::i_count to track whether we should
3478 * just decrement it (in case it's > 1) or if this is the last iput then link
3479 * the inode to the delayed iput machinery. Delayed iputs are processed at
3480 * transaction commit time/superblock commit/cleaner kthread.
3482 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3484 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3486 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3489 atomic_inc(&fs_info->nr_delayed_iputs);
3490 spin_lock(&fs_info->delayed_iput_lock);
3491 ASSERT(list_empty(&inode->delayed_iput));
3492 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3493 spin_unlock(&fs_info->delayed_iput_lock);
3494 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3495 wake_up_process(fs_info->cleaner_kthread);
3498 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3499 struct btrfs_inode *inode)
3501 list_del_init(&inode->delayed_iput);
3502 spin_unlock(&fs_info->delayed_iput_lock);
3503 iput(&inode->vfs_inode);
3504 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3505 wake_up(&fs_info->delayed_iputs_wait);
3506 spin_lock(&fs_info->delayed_iput_lock);
3509 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3510 struct btrfs_inode *inode)
3512 if (!list_empty(&inode->delayed_iput)) {
3513 spin_lock(&fs_info->delayed_iput_lock);
3514 if (!list_empty(&inode->delayed_iput))
3515 run_delayed_iput_locked(fs_info, inode);
3516 spin_unlock(&fs_info->delayed_iput_lock);
3520 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3523 spin_lock(&fs_info->delayed_iput_lock);
3524 while (!list_empty(&fs_info->delayed_iputs)) {
3525 struct btrfs_inode *inode;
3527 inode = list_first_entry(&fs_info->delayed_iputs,
3528 struct btrfs_inode, delayed_iput);
3529 run_delayed_iput_locked(fs_info, inode);
3530 cond_resched_lock(&fs_info->delayed_iput_lock);
3532 spin_unlock(&fs_info->delayed_iput_lock);
3536 * Wait for flushing all delayed iputs
3538 * @fs_info: the filesystem
3540 * This will wait on any delayed iputs that are currently running with KILLABLE
3541 * set. Once they are all done running we will return, unless we are killed in
3542 * which case we return EINTR. This helps in user operations like fallocate etc
3543 * that might get blocked on the iputs.
3545 * Return EINTR if we were killed, 0 if nothing's pending
3547 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3549 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3550 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3557 * This creates an orphan entry for the given inode in case something goes wrong
3558 * in the middle of an unlink.
3560 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3561 struct btrfs_inode *inode)
3565 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3566 if (ret && ret != -EEXIST) {
3567 btrfs_abort_transaction(trans, ret);
3575 * We have done the delete so we can go ahead and remove the orphan item for
3576 * this particular inode.
3578 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3579 struct btrfs_inode *inode)
3581 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3585 * this cleans up any orphans that may be left on the list from the last use
3588 int btrfs_orphan_cleanup(struct btrfs_root *root)
3590 struct btrfs_fs_info *fs_info = root->fs_info;
3591 struct btrfs_path *path;
3592 struct extent_buffer *leaf;
3593 struct btrfs_key key, found_key;
3594 struct btrfs_trans_handle *trans;
3595 struct inode *inode;
3596 u64 last_objectid = 0;
3597 int ret = 0, nr_unlink = 0;
3599 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3602 path = btrfs_alloc_path();
3607 path->reada = READA_BACK;
3609 key.objectid = BTRFS_ORPHAN_OBJECTID;
3610 key.type = BTRFS_ORPHAN_ITEM_KEY;
3611 key.offset = (u64)-1;
3614 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3619 * if ret == 0 means we found what we were searching for, which
3620 * is weird, but possible, so only screw with path if we didn't
3621 * find the key and see if we have stuff that matches
3625 if (path->slots[0] == 0)
3630 /* pull out the item */
3631 leaf = path->nodes[0];
3632 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3634 /* make sure the item matches what we want */
3635 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3637 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3640 /* release the path since we're done with it */
3641 btrfs_release_path(path);
3644 * this is where we are basically btrfs_lookup, without the
3645 * crossing root thing. we store the inode number in the
3646 * offset of the orphan item.
3649 if (found_key.offset == last_objectid) {
3651 "Error removing orphan entry, stopping orphan cleanup");
3656 last_objectid = found_key.offset;
3658 found_key.objectid = found_key.offset;
3659 found_key.type = BTRFS_INODE_ITEM_KEY;
3660 found_key.offset = 0;
3661 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3662 ret = PTR_ERR_OR_ZERO(inode);
3663 if (ret && ret != -ENOENT)
3666 if (ret == -ENOENT && root == fs_info->tree_root) {
3667 struct btrfs_root *dead_root;
3668 int is_dead_root = 0;
3671 * This is an orphan in the tree root. Currently these
3672 * could come from 2 sources:
3673 * a) a root (snapshot/subvolume) deletion in progress
3674 * b) a free space cache inode
3675 * We need to distinguish those two, as the orphan item
3676 * for a root must not get deleted before the deletion
3677 * of the snapshot/subvolume's tree completes.
3679 * btrfs_find_orphan_roots() ran before us, which has
3680 * found all deleted roots and loaded them into
3681 * fs_info->fs_roots_radix. So here we can find if an
3682 * orphan item corresponds to a deleted root by looking
3683 * up the root from that radix tree.
3686 spin_lock(&fs_info->fs_roots_radix_lock);
3687 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3688 (unsigned long)found_key.objectid);
3689 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3691 spin_unlock(&fs_info->fs_roots_radix_lock);
3694 /* prevent this orphan from being found again */
3695 key.offset = found_key.objectid - 1;
3702 * If we have an inode with links, there are a couple of
3705 * 1. We were halfway through creating fsverity metadata for the
3706 * file. In that case, the orphan item represents incomplete
3707 * fsverity metadata which must be cleaned up with
3708 * btrfs_drop_verity_items and deleting the orphan item.
3710 * 2. Old kernels (before v3.12) used to create an
3711 * orphan item for truncate indicating that there were possibly
3712 * extent items past i_size that needed to be deleted. In v3.12,
3713 * truncate was changed to update i_size in sync with the extent
3714 * items, but the (useless) orphan item was still created. Since
3715 * v4.18, we don't create the orphan item for truncate at all.
3717 * So, this item could mean that we need to do a truncate, but
3718 * only if this filesystem was last used on a pre-v3.12 kernel
3719 * and was not cleanly unmounted. The odds of that are quite
3720 * slim, and it's a pain to do the truncate now, so just delete
3723 * It's also possible that this orphan item was supposed to be
3724 * deleted but wasn't. The inode number may have been reused,
3725 * but either way, we can delete the orphan item.
3727 if (ret == -ENOENT || inode->i_nlink) {
3729 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3734 trans = btrfs_start_transaction(root, 1);
3735 if (IS_ERR(trans)) {
3736 ret = PTR_ERR(trans);
3740 btrfs_debug(fs_info, "auto deleting %Lu",
3741 found_key.objectid);
3742 ret = btrfs_del_orphan_item(trans, root,
3743 found_key.objectid);
3744 btrfs_end_transaction(trans);
3754 /* this will do delete_inode and everything for us */
3757 /* release the path since we're done with it */
3758 btrfs_release_path(path);
3760 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3761 trans = btrfs_join_transaction(root);
3763 btrfs_end_transaction(trans);
3767 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3771 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3772 btrfs_free_path(path);
3777 * very simple check to peek ahead in the leaf looking for xattrs. If we
3778 * don't find any xattrs, we know there can't be any acls.
3780 * slot is the slot the inode is in, objectid is the objectid of the inode
3782 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3783 int slot, u64 objectid,
3784 int *first_xattr_slot)
3786 u32 nritems = btrfs_header_nritems(leaf);
3787 struct btrfs_key found_key;
3788 static u64 xattr_access = 0;
3789 static u64 xattr_default = 0;
3792 if (!xattr_access) {
3793 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3794 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3795 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3796 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3800 *first_xattr_slot = -1;
3801 while (slot < nritems) {
3802 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3804 /* we found a different objectid, there must not be acls */
3805 if (found_key.objectid != objectid)
3808 /* we found an xattr, assume we've got an acl */
3809 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3810 if (*first_xattr_slot == -1)
3811 *first_xattr_slot = slot;
3812 if (found_key.offset == xattr_access ||
3813 found_key.offset == xattr_default)
3818 * we found a key greater than an xattr key, there can't
3819 * be any acls later on
3821 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3828 * it goes inode, inode backrefs, xattrs, extents,
3829 * so if there are a ton of hard links to an inode there can
3830 * be a lot of backrefs. Don't waste time searching too hard,
3831 * this is just an optimization
3836 /* we hit the end of the leaf before we found an xattr or
3837 * something larger than an xattr. We have to assume the inode
3840 if (*first_xattr_slot == -1)
3841 *first_xattr_slot = slot;
3846 * read an inode from the btree into the in-memory inode
3848 static int btrfs_read_locked_inode(struct inode *inode,
3849 struct btrfs_path *in_path)
3851 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3852 struct btrfs_path *path = in_path;
3853 struct extent_buffer *leaf;
3854 struct btrfs_inode_item *inode_item;
3855 struct btrfs_root *root = BTRFS_I(inode)->root;
3856 struct btrfs_key location;
3861 bool filled = false;
3862 int first_xattr_slot;
3864 ret = btrfs_fill_inode(inode, &rdev);
3869 path = btrfs_alloc_path();
3874 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3876 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3878 if (path != in_path)
3879 btrfs_free_path(path);
3883 leaf = path->nodes[0];
3888 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3889 struct btrfs_inode_item);
3890 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3891 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3892 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3893 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3894 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3895 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3896 round_up(i_size_read(inode), fs_info->sectorsize));
3898 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3899 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3901 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3902 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3904 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3905 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3907 BTRFS_I(inode)->i_otime.tv_sec =
3908 btrfs_timespec_sec(leaf, &inode_item->otime);
3909 BTRFS_I(inode)->i_otime.tv_nsec =
3910 btrfs_timespec_nsec(leaf, &inode_item->otime);
3912 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3913 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3914 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3916 inode_set_iversion_queried(inode,
3917 btrfs_inode_sequence(leaf, inode_item));
3918 inode->i_generation = BTRFS_I(inode)->generation;
3920 rdev = btrfs_inode_rdev(leaf, inode_item);
3922 BTRFS_I(inode)->index_cnt = (u64)-1;
3923 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3924 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3928 * If we were modified in the current generation and evicted from memory
3929 * and then re-read we need to do a full sync since we don't have any
3930 * idea about which extents were modified before we were evicted from
3933 * This is required for both inode re-read from disk and delayed inode
3934 * in delayed_nodes_tree.
3936 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3937 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3938 &BTRFS_I(inode)->runtime_flags);
3941 * We don't persist the id of the transaction where an unlink operation
3942 * against the inode was last made. So here we assume the inode might
3943 * have been evicted, and therefore the exact value of last_unlink_trans
3944 * lost, and set it to last_trans to avoid metadata inconsistencies
3945 * between the inode and its parent if the inode is fsync'ed and the log
3946 * replayed. For example, in the scenario:
3949 * ln mydir/foo mydir/bar
3952 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3953 * xfs_io -c fsync mydir/foo
3955 * mount fs, triggers fsync log replay
3957 * We must make sure that when we fsync our inode foo we also log its
3958 * parent inode, otherwise after log replay the parent still has the
3959 * dentry with the "bar" name but our inode foo has a link count of 1
3960 * and doesn't have an inode ref with the name "bar" anymore.
3962 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3963 * but it guarantees correctness at the expense of occasional full
3964 * transaction commits on fsync if our inode is a directory, or if our
3965 * inode is not a directory, logging its parent unnecessarily.
3967 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3970 * Same logic as for last_unlink_trans. We don't persist the generation
3971 * of the last transaction where this inode was used for a reflink
3972 * operation, so after eviction and reloading the inode we must be
3973 * pessimistic and assume the last transaction that modified the inode.
3975 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3978 if (inode->i_nlink != 1 ||
3979 path->slots[0] >= btrfs_header_nritems(leaf))
3982 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3983 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3986 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3987 if (location.type == BTRFS_INODE_REF_KEY) {
3988 struct btrfs_inode_ref *ref;
3990 ref = (struct btrfs_inode_ref *)ptr;
3991 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3992 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3993 struct btrfs_inode_extref *extref;
3995 extref = (struct btrfs_inode_extref *)ptr;
3996 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4001 * try to precache a NULL acl entry for files that don't have
4002 * any xattrs or acls
4004 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4005 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4006 if (first_xattr_slot != -1) {
4007 path->slots[0] = first_xattr_slot;
4008 ret = btrfs_load_inode_props(inode, path);
4011 "error loading props for ino %llu (root %llu): %d",
4012 btrfs_ino(BTRFS_I(inode)),
4013 root->root_key.objectid, ret);
4015 if (path != in_path)
4016 btrfs_free_path(path);
4019 cache_no_acl(inode);
4021 switch (inode->i_mode & S_IFMT) {
4023 inode->i_mapping->a_ops = &btrfs_aops;
4024 inode->i_fop = &btrfs_file_operations;
4025 inode->i_op = &btrfs_file_inode_operations;
4028 inode->i_fop = &btrfs_dir_file_operations;
4029 inode->i_op = &btrfs_dir_inode_operations;
4032 inode->i_op = &btrfs_symlink_inode_operations;
4033 inode_nohighmem(inode);
4034 inode->i_mapping->a_ops = &btrfs_aops;
4037 inode->i_op = &btrfs_special_inode_operations;
4038 init_special_inode(inode, inode->i_mode, rdev);
4042 btrfs_sync_inode_flags_to_i_flags(inode);
4047 * given a leaf and an inode, copy the inode fields into the leaf
4049 static void fill_inode_item(struct btrfs_trans_handle *trans,
4050 struct extent_buffer *leaf,
4051 struct btrfs_inode_item *item,
4052 struct inode *inode)
4054 struct btrfs_map_token token;
4057 btrfs_init_map_token(&token, leaf);
4059 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4060 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4061 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4062 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4063 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4065 btrfs_set_token_timespec_sec(&token, &item->atime,
4066 inode->i_atime.tv_sec);
4067 btrfs_set_token_timespec_nsec(&token, &item->atime,
4068 inode->i_atime.tv_nsec);
4070 btrfs_set_token_timespec_sec(&token, &item->mtime,
4071 inode->i_mtime.tv_sec);
4072 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4073 inode->i_mtime.tv_nsec);
4075 btrfs_set_token_timespec_sec(&token, &item->ctime,
4076 inode->i_ctime.tv_sec);
4077 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4078 inode->i_ctime.tv_nsec);
4080 btrfs_set_token_timespec_sec(&token, &item->otime,
4081 BTRFS_I(inode)->i_otime.tv_sec);
4082 btrfs_set_token_timespec_nsec(&token, &item->otime,
4083 BTRFS_I(inode)->i_otime.tv_nsec);
4085 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4086 btrfs_set_token_inode_generation(&token, item,
4087 BTRFS_I(inode)->generation);
4088 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4089 btrfs_set_token_inode_transid(&token, item, trans->transid);
4090 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4091 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4092 BTRFS_I(inode)->ro_flags);
4093 btrfs_set_token_inode_flags(&token, item, flags);
4094 btrfs_set_token_inode_block_group(&token, item, 0);
4098 * copy everything in the in-memory inode into the btree.
4100 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4101 struct btrfs_root *root,
4102 struct btrfs_inode *inode)
4104 struct btrfs_inode_item *inode_item;
4105 struct btrfs_path *path;
4106 struct extent_buffer *leaf;
4109 path = btrfs_alloc_path();
4113 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4120 leaf = path->nodes[0];
4121 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4122 struct btrfs_inode_item);
4124 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4125 btrfs_mark_buffer_dirty(leaf);
4126 btrfs_set_inode_last_trans(trans, inode);
4129 btrfs_free_path(path);
4134 * copy everything in the in-memory inode into the btree.
4136 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4137 struct btrfs_root *root,
4138 struct btrfs_inode *inode)
4140 struct btrfs_fs_info *fs_info = root->fs_info;
4144 * If the inode is a free space inode, we can deadlock during commit
4145 * if we put it into the delayed code.
4147 * The data relocation inode should also be directly updated
4150 if (!btrfs_is_free_space_inode(inode)
4151 && !btrfs_is_data_reloc_root(root)
4152 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4153 btrfs_update_root_times(trans, root);
4155 ret = btrfs_delayed_update_inode(trans, root, inode);
4157 btrfs_set_inode_last_trans(trans, inode);
4161 return btrfs_update_inode_item(trans, root, inode);
4164 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4165 struct btrfs_root *root, struct btrfs_inode *inode)
4169 ret = btrfs_update_inode(trans, root, inode);
4171 return btrfs_update_inode_item(trans, root, inode);
4176 * unlink helper that gets used here in inode.c and in the tree logging
4177 * recovery code. It remove a link in a directory with a given name, and
4178 * also drops the back refs in the inode to the directory
4180 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4181 struct btrfs_inode *dir,
4182 struct btrfs_inode *inode,
4183 const struct fscrypt_str *name,
4184 struct btrfs_rename_ctx *rename_ctx)
4186 struct btrfs_root *root = dir->root;
4187 struct btrfs_fs_info *fs_info = root->fs_info;
4188 struct btrfs_path *path;
4190 struct btrfs_dir_item *di;
4192 u64 ino = btrfs_ino(inode);
4193 u64 dir_ino = btrfs_ino(dir);
4195 path = btrfs_alloc_path();
4201 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4202 if (IS_ERR_OR_NULL(di)) {
4203 ret = di ? PTR_ERR(di) : -ENOENT;
4206 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4209 btrfs_release_path(path);
4212 * If we don't have dir index, we have to get it by looking up
4213 * the inode ref, since we get the inode ref, remove it directly,
4214 * it is unnecessary to do delayed deletion.
4216 * But if we have dir index, needn't search inode ref to get it.
4217 * Since the inode ref is close to the inode item, it is better
4218 * that we delay to delete it, and just do this deletion when
4219 * we update the inode item.
4221 if (inode->dir_index) {
4222 ret = btrfs_delayed_delete_inode_ref(inode);
4224 index = inode->dir_index;
4229 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4232 "failed to delete reference to %.*s, inode %llu parent %llu",
4233 name->len, name->name, ino, dir_ino);
4234 btrfs_abort_transaction(trans, ret);
4239 rename_ctx->index = index;
4241 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4243 btrfs_abort_transaction(trans, ret);
4248 * If we are in a rename context, we don't need to update anything in the
4249 * log. That will be done later during the rename by btrfs_log_new_name().
4250 * Besides that, doing it here would only cause extra unnecessary btree
4251 * operations on the log tree, increasing latency for applications.
4254 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4255 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4259 * If we have a pending delayed iput we could end up with the final iput
4260 * being run in btrfs-cleaner context. If we have enough of these built
4261 * up we can end up burning a lot of time in btrfs-cleaner without any
4262 * way to throttle the unlinks. Since we're currently holding a ref on
4263 * the inode we can run the delayed iput here without any issues as the
4264 * final iput won't be done until after we drop the ref we're currently
4267 btrfs_run_delayed_iput(fs_info, inode);
4269 btrfs_free_path(path);
4273 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4274 inode_inc_iversion(&inode->vfs_inode);
4275 inode_inc_iversion(&dir->vfs_inode);
4276 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4277 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4278 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4279 ret = btrfs_update_inode(trans, root, dir);
4284 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4285 struct btrfs_inode *dir, struct btrfs_inode *inode,
4286 const struct fscrypt_str *name)
4290 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4292 drop_nlink(&inode->vfs_inode);
4293 ret = btrfs_update_inode(trans, inode->root, inode);
4299 * helper to start transaction for unlink and rmdir.
4301 * unlink and rmdir are special in btrfs, they do not always free space, so
4302 * if we cannot make our reservations the normal way try and see if there is
4303 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4304 * allow the unlink to occur.
4306 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4308 struct btrfs_root *root = dir->root;
4310 return btrfs_start_transaction_fallback_global_rsv(root,
4311 BTRFS_UNLINK_METADATA_UNITS);
4314 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4316 struct btrfs_trans_handle *trans;
4317 struct inode *inode = d_inode(dentry);
4319 struct fscrypt_name fname;
4321 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4325 /* This needs to handle no-key deletions later on */
4327 trans = __unlink_start_trans(BTRFS_I(dir));
4328 if (IS_ERR(trans)) {
4329 ret = PTR_ERR(trans);
4333 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4336 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4341 if (inode->i_nlink == 0) {
4342 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4348 btrfs_end_transaction(trans);
4349 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4351 fscrypt_free_filename(&fname);
4355 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4356 struct btrfs_inode *dir, struct dentry *dentry)
4358 struct btrfs_root *root = dir->root;
4359 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4360 struct btrfs_path *path;
4361 struct extent_buffer *leaf;
4362 struct btrfs_dir_item *di;
4363 struct btrfs_key key;
4367 u64 dir_ino = btrfs_ino(dir);
4368 struct fscrypt_name fname;
4370 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4374 /* This needs to handle no-key deletions later on */
4376 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4377 objectid = inode->root->root_key.objectid;
4378 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4379 objectid = inode->location.objectid;
4382 fscrypt_free_filename(&fname);
4386 path = btrfs_alloc_path();
4392 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4393 &fname.disk_name, -1);
4394 if (IS_ERR_OR_NULL(di)) {
4395 ret = di ? PTR_ERR(di) : -ENOENT;
4399 leaf = path->nodes[0];
4400 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4401 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4402 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4404 btrfs_abort_transaction(trans, ret);
4407 btrfs_release_path(path);
4410 * This is a placeholder inode for a subvolume we didn't have a
4411 * reference to at the time of the snapshot creation. In the meantime
4412 * we could have renamed the real subvol link into our snapshot, so
4413 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4414 * Instead simply lookup the dir_index_item for this entry so we can
4415 * remove it. Otherwise we know we have a ref to the root and we can
4416 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4418 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4419 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4420 if (IS_ERR_OR_NULL(di)) {
4425 btrfs_abort_transaction(trans, ret);
4429 leaf = path->nodes[0];
4430 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4432 btrfs_release_path(path);
4434 ret = btrfs_del_root_ref(trans, objectid,
4435 root->root_key.objectid, dir_ino,
4436 &index, &fname.disk_name);
4438 btrfs_abort_transaction(trans, ret);
4443 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4445 btrfs_abort_transaction(trans, ret);
4449 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4450 inode_inc_iversion(&dir->vfs_inode);
4451 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4452 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4453 ret = btrfs_update_inode_fallback(trans, root, dir);
4455 btrfs_abort_transaction(trans, ret);
4457 btrfs_free_path(path);
4458 fscrypt_free_filename(&fname);
4463 * Helper to check if the subvolume references other subvolumes or if it's
4466 static noinline int may_destroy_subvol(struct btrfs_root *root)
4468 struct btrfs_fs_info *fs_info = root->fs_info;
4469 struct btrfs_path *path;
4470 struct btrfs_dir_item *di;
4471 struct btrfs_key key;
4472 struct fscrypt_str name = FSTR_INIT("default", 7);
4476 path = btrfs_alloc_path();
4480 /* Make sure this root isn't set as the default subvol */
4481 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4482 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4484 if (di && !IS_ERR(di)) {
4485 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4486 if (key.objectid == root->root_key.objectid) {
4489 "deleting default subvolume %llu is not allowed",
4493 btrfs_release_path(path);
4496 key.objectid = root->root_key.objectid;
4497 key.type = BTRFS_ROOT_REF_KEY;
4498 key.offset = (u64)-1;
4500 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4506 if (path->slots[0] > 0) {
4508 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4509 if (key.objectid == root->root_key.objectid &&
4510 key.type == BTRFS_ROOT_REF_KEY)
4514 btrfs_free_path(path);
4518 /* Delete all dentries for inodes belonging to the root */
4519 static void btrfs_prune_dentries(struct btrfs_root *root)
4521 struct btrfs_fs_info *fs_info = root->fs_info;
4522 struct rb_node *node;
4523 struct rb_node *prev;
4524 struct btrfs_inode *entry;
4525 struct inode *inode;
4528 if (!BTRFS_FS_ERROR(fs_info))
4529 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4531 spin_lock(&root->inode_lock);
4533 node = root->inode_tree.rb_node;
4537 entry = rb_entry(node, struct btrfs_inode, rb_node);
4539 if (objectid < btrfs_ino(entry))
4540 node = node->rb_left;
4541 else if (objectid > btrfs_ino(entry))
4542 node = node->rb_right;
4548 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4549 if (objectid <= btrfs_ino(entry)) {
4553 prev = rb_next(prev);
4557 entry = rb_entry(node, struct btrfs_inode, rb_node);
4558 objectid = btrfs_ino(entry) + 1;
4559 inode = igrab(&entry->vfs_inode);
4561 spin_unlock(&root->inode_lock);
4562 if (atomic_read(&inode->i_count) > 1)
4563 d_prune_aliases(inode);
4565 * btrfs_drop_inode will have it removed from the inode
4566 * cache when its usage count hits zero.
4570 spin_lock(&root->inode_lock);
4574 if (cond_resched_lock(&root->inode_lock))
4577 node = rb_next(node);
4579 spin_unlock(&root->inode_lock);
4582 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4584 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4585 struct btrfs_root *root = dir->root;
4586 struct inode *inode = d_inode(dentry);
4587 struct btrfs_root *dest = BTRFS_I(inode)->root;
4588 struct btrfs_trans_handle *trans;
4589 struct btrfs_block_rsv block_rsv;
4594 * Don't allow to delete a subvolume with send in progress. This is
4595 * inside the inode lock so the error handling that has to drop the bit
4596 * again is not run concurrently.
4598 spin_lock(&dest->root_item_lock);
4599 if (dest->send_in_progress) {
4600 spin_unlock(&dest->root_item_lock);
4602 "attempt to delete subvolume %llu during send",
4603 dest->root_key.objectid);
4606 if (atomic_read(&dest->nr_swapfiles)) {
4607 spin_unlock(&dest->root_item_lock);
4609 "attempt to delete subvolume %llu with active swapfile",
4610 root->root_key.objectid);
4613 root_flags = btrfs_root_flags(&dest->root_item);
4614 btrfs_set_root_flags(&dest->root_item,
4615 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4616 spin_unlock(&dest->root_item_lock);
4618 down_write(&fs_info->subvol_sem);
4620 ret = may_destroy_subvol(dest);
4624 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4626 * One for dir inode,
4627 * two for dir entries,
4628 * two for root ref/backref.
4630 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4634 trans = btrfs_start_transaction(root, 0);
4635 if (IS_ERR(trans)) {
4636 ret = PTR_ERR(trans);
4639 trans->block_rsv = &block_rsv;
4640 trans->bytes_reserved = block_rsv.size;
4642 btrfs_record_snapshot_destroy(trans, dir);
4644 ret = btrfs_unlink_subvol(trans, dir, dentry);
4646 btrfs_abort_transaction(trans, ret);
4650 ret = btrfs_record_root_in_trans(trans, dest);
4652 btrfs_abort_transaction(trans, ret);
4656 memset(&dest->root_item.drop_progress, 0,
4657 sizeof(dest->root_item.drop_progress));
4658 btrfs_set_root_drop_level(&dest->root_item, 0);
4659 btrfs_set_root_refs(&dest->root_item, 0);
4661 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4662 ret = btrfs_insert_orphan_item(trans,
4664 dest->root_key.objectid);
4666 btrfs_abort_transaction(trans, ret);
4671 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4672 BTRFS_UUID_KEY_SUBVOL,
4673 dest->root_key.objectid);
4674 if (ret && ret != -ENOENT) {
4675 btrfs_abort_transaction(trans, ret);
4678 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4679 ret = btrfs_uuid_tree_remove(trans,
4680 dest->root_item.received_uuid,
4681 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4682 dest->root_key.objectid);
4683 if (ret && ret != -ENOENT) {
4684 btrfs_abort_transaction(trans, ret);
4689 free_anon_bdev(dest->anon_dev);
4692 trans->block_rsv = NULL;
4693 trans->bytes_reserved = 0;
4694 ret = btrfs_end_transaction(trans);
4695 inode->i_flags |= S_DEAD;
4697 btrfs_subvolume_release_metadata(root, &block_rsv);
4699 up_write(&fs_info->subvol_sem);
4701 spin_lock(&dest->root_item_lock);
4702 root_flags = btrfs_root_flags(&dest->root_item);
4703 btrfs_set_root_flags(&dest->root_item,
4704 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4705 spin_unlock(&dest->root_item_lock);
4707 d_invalidate(dentry);
4708 btrfs_prune_dentries(dest);
4709 ASSERT(dest->send_in_progress == 0);
4715 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4717 struct inode *inode = d_inode(dentry);
4718 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4720 struct btrfs_trans_handle *trans;
4721 u64 last_unlink_trans;
4722 struct fscrypt_name fname;
4724 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4726 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4727 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4729 "extent tree v2 doesn't support snapshot deletion yet");
4732 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4735 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4739 /* This needs to handle no-key deletions later on */
4741 trans = __unlink_start_trans(BTRFS_I(dir));
4742 if (IS_ERR(trans)) {
4743 err = PTR_ERR(trans);
4747 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4748 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4752 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4756 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4758 /* now the directory is empty */
4759 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4762 btrfs_i_size_write(BTRFS_I(inode), 0);
4764 * Propagate the last_unlink_trans value of the deleted dir to
4765 * its parent directory. This is to prevent an unrecoverable
4766 * log tree in the case we do something like this:
4768 * 2) create snapshot under dir foo
4769 * 3) delete the snapshot
4772 * 6) fsync foo or some file inside foo
4774 if (last_unlink_trans >= trans->transid)
4775 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4778 btrfs_end_transaction(trans);
4780 btrfs_btree_balance_dirty(fs_info);
4781 fscrypt_free_filename(&fname);
4787 * btrfs_truncate_block - read, zero a chunk and write a block
4788 * @inode - inode that we're zeroing
4789 * @from - the offset to start zeroing
4790 * @len - the length to zero, 0 to zero the entire range respective to the
4792 * @front - zero up to the offset instead of from the offset on
4794 * This will find the block for the "from" offset and cow the block and zero the
4795 * part we want to zero. This is used with truncate and hole punching.
4797 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4800 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4801 struct address_space *mapping = inode->vfs_inode.i_mapping;
4802 struct extent_io_tree *io_tree = &inode->io_tree;
4803 struct btrfs_ordered_extent *ordered;
4804 struct extent_state *cached_state = NULL;
4805 struct extent_changeset *data_reserved = NULL;
4806 bool only_release_metadata = false;
4807 u32 blocksize = fs_info->sectorsize;
4808 pgoff_t index = from >> PAGE_SHIFT;
4809 unsigned offset = from & (blocksize - 1);
4811 gfp_t mask = btrfs_alloc_write_mask(mapping);
4812 size_t write_bytes = blocksize;
4817 if (IS_ALIGNED(offset, blocksize) &&
4818 (!len || IS_ALIGNED(len, blocksize)))
4821 block_start = round_down(from, blocksize);
4822 block_end = block_start + blocksize - 1;
4824 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4827 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4828 /* For nocow case, no need to reserve data space */
4829 only_release_metadata = true;
4834 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4836 if (!only_release_metadata)
4837 btrfs_free_reserved_data_space(inode, data_reserved,
4838 block_start, blocksize);
4842 page = find_or_create_page(mapping, index, mask);
4844 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4846 btrfs_delalloc_release_extents(inode, blocksize);
4850 ret = set_page_extent_mapped(page);
4854 if (!PageUptodate(page)) {
4855 ret = btrfs_read_folio(NULL, page_folio(page));
4857 if (page->mapping != mapping) {
4862 if (!PageUptodate(page)) {
4867 wait_on_page_writeback(page);
4869 lock_extent(io_tree, block_start, block_end, &cached_state);
4871 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4873 unlock_extent(io_tree, block_start, block_end, &cached_state);
4876 btrfs_start_ordered_extent(ordered);
4877 btrfs_put_ordered_extent(ordered);
4881 clear_extent_bit(&inode->io_tree, block_start, block_end,
4882 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4885 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4888 unlock_extent(io_tree, block_start, block_end, &cached_state);
4892 if (offset != blocksize) {
4894 len = blocksize - offset;
4896 memzero_page(page, (block_start - page_offset(page)),
4899 memzero_page(page, (block_start - page_offset(page)) + offset,
4902 btrfs_page_clear_checked(fs_info, page, block_start,
4903 block_end + 1 - block_start);
4904 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4905 unlock_extent(io_tree, block_start, block_end, &cached_state);
4907 if (only_release_metadata)
4908 set_extent_bit(&inode->io_tree, block_start, block_end,
4909 EXTENT_NORESERVE, NULL);
4913 if (only_release_metadata)
4914 btrfs_delalloc_release_metadata(inode, blocksize, true);
4916 btrfs_delalloc_release_space(inode, data_reserved,
4917 block_start, blocksize, true);
4919 btrfs_delalloc_release_extents(inode, blocksize);
4923 if (only_release_metadata)
4924 btrfs_check_nocow_unlock(inode);
4925 extent_changeset_free(data_reserved);
4929 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4930 u64 offset, u64 len)
4932 struct btrfs_fs_info *fs_info = root->fs_info;
4933 struct btrfs_trans_handle *trans;
4934 struct btrfs_drop_extents_args drop_args = { 0 };
4938 * If NO_HOLES is enabled, we don't need to do anything.
4939 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4940 * or btrfs_update_inode() will be called, which guarantee that the next
4941 * fsync will know this inode was changed and needs to be logged.
4943 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4947 * 1 - for the one we're dropping
4948 * 1 - for the one we're adding
4949 * 1 - for updating the inode.
4951 trans = btrfs_start_transaction(root, 3);
4953 return PTR_ERR(trans);
4955 drop_args.start = offset;
4956 drop_args.end = offset + len;
4957 drop_args.drop_cache = true;
4959 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4961 btrfs_abort_transaction(trans, ret);
4962 btrfs_end_transaction(trans);
4966 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4968 btrfs_abort_transaction(trans, ret);
4970 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4971 btrfs_update_inode(trans, root, inode);
4973 btrfs_end_transaction(trans);
4978 * This function puts in dummy file extents for the area we're creating a hole
4979 * for. So if we are truncating this file to a larger size we need to insert
4980 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4981 * the range between oldsize and size
4983 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4985 struct btrfs_root *root = inode->root;
4986 struct btrfs_fs_info *fs_info = root->fs_info;
4987 struct extent_io_tree *io_tree = &inode->io_tree;
4988 struct extent_map *em = NULL;
4989 struct extent_state *cached_state = NULL;
4990 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4991 u64 block_end = ALIGN(size, fs_info->sectorsize);
4998 * If our size started in the middle of a block we need to zero out the
4999 * rest of the block before we expand the i_size, otherwise we could
5000 * expose stale data.
5002 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5006 if (size <= hole_start)
5009 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5011 cur_offset = hole_start;
5013 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5014 block_end - cur_offset);
5020 last_byte = min(extent_map_end(em), block_end);
5021 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5022 hole_size = last_byte - cur_offset;
5024 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5025 struct extent_map *hole_em;
5027 err = maybe_insert_hole(root, inode, cur_offset,
5032 err = btrfs_inode_set_file_extent_range(inode,
5033 cur_offset, hole_size);
5037 hole_em = alloc_extent_map();
5039 btrfs_drop_extent_map_range(inode, cur_offset,
5040 cur_offset + hole_size - 1,
5042 btrfs_set_inode_full_sync(inode);
5045 hole_em->start = cur_offset;
5046 hole_em->len = hole_size;
5047 hole_em->orig_start = cur_offset;
5049 hole_em->block_start = EXTENT_MAP_HOLE;
5050 hole_em->block_len = 0;
5051 hole_em->orig_block_len = 0;
5052 hole_em->ram_bytes = hole_size;
5053 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5054 hole_em->generation = fs_info->generation;
5056 err = btrfs_replace_extent_map_range(inode, hole_em, true);
5057 free_extent_map(hole_em);
5059 err = btrfs_inode_set_file_extent_range(inode,
5060 cur_offset, hole_size);
5065 free_extent_map(em);
5067 cur_offset = last_byte;
5068 if (cur_offset >= block_end)
5071 free_extent_map(em);
5072 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5076 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5078 struct btrfs_root *root = BTRFS_I(inode)->root;
5079 struct btrfs_trans_handle *trans;
5080 loff_t oldsize = i_size_read(inode);
5081 loff_t newsize = attr->ia_size;
5082 int mask = attr->ia_valid;
5086 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5087 * special case where we need to update the times despite not having
5088 * these flags set. For all other operations the VFS set these flags
5089 * explicitly if it wants a timestamp update.
5091 if (newsize != oldsize) {
5092 inode_inc_iversion(inode);
5093 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5094 inode->i_mtime = current_time(inode);
5095 inode->i_ctime = inode->i_mtime;
5099 if (newsize > oldsize) {
5101 * Don't do an expanding truncate while snapshotting is ongoing.
5102 * This is to ensure the snapshot captures a fully consistent
5103 * state of this file - if the snapshot captures this expanding
5104 * truncation, it must capture all writes that happened before
5107 btrfs_drew_write_lock(&root->snapshot_lock);
5108 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5110 btrfs_drew_write_unlock(&root->snapshot_lock);
5114 trans = btrfs_start_transaction(root, 1);
5115 if (IS_ERR(trans)) {
5116 btrfs_drew_write_unlock(&root->snapshot_lock);
5117 return PTR_ERR(trans);
5120 i_size_write(inode, newsize);
5121 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5122 pagecache_isize_extended(inode, oldsize, newsize);
5123 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5124 btrfs_drew_write_unlock(&root->snapshot_lock);
5125 btrfs_end_transaction(trans);
5127 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5129 if (btrfs_is_zoned(fs_info)) {
5130 ret = btrfs_wait_ordered_range(inode,
5131 ALIGN(newsize, fs_info->sectorsize),
5138 * We're truncating a file that used to have good data down to
5139 * zero. Make sure any new writes to the file get on disk
5143 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5144 &BTRFS_I(inode)->runtime_flags);
5146 truncate_setsize(inode, newsize);
5148 inode_dio_wait(inode);
5150 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5151 if (ret && inode->i_nlink) {
5155 * Truncate failed, so fix up the in-memory size. We
5156 * adjusted disk_i_size down as we removed extents, so
5157 * wait for disk_i_size to be stable and then update the
5158 * in-memory size to match.
5160 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5163 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5170 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5173 struct inode *inode = d_inode(dentry);
5174 struct btrfs_root *root = BTRFS_I(inode)->root;
5177 if (btrfs_root_readonly(root))
5180 err = setattr_prepare(idmap, dentry, attr);
5184 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5185 err = btrfs_setsize(inode, attr);
5190 if (attr->ia_valid) {
5191 setattr_copy(idmap, inode, attr);
5192 inode_inc_iversion(inode);
5193 err = btrfs_dirty_inode(BTRFS_I(inode));
5195 if (!err && attr->ia_valid & ATTR_MODE)
5196 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5203 * While truncating the inode pages during eviction, we get the VFS
5204 * calling btrfs_invalidate_folio() against each folio of the inode. This
5205 * is slow because the calls to btrfs_invalidate_folio() result in a
5206 * huge amount of calls to lock_extent() and clear_extent_bit(),
5207 * which keep merging and splitting extent_state structures over and over,
5208 * wasting lots of time.
5210 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5211 * skip all those expensive operations on a per folio basis and do only
5212 * the ordered io finishing, while we release here the extent_map and
5213 * extent_state structures, without the excessive merging and splitting.
5215 static void evict_inode_truncate_pages(struct inode *inode)
5217 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5218 struct rb_node *node;
5220 ASSERT(inode->i_state & I_FREEING);
5221 truncate_inode_pages_final(&inode->i_data);
5223 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5226 * Keep looping until we have no more ranges in the io tree.
5227 * We can have ongoing bios started by readahead that have
5228 * their endio callback (extent_io.c:end_bio_extent_readpage)
5229 * still in progress (unlocked the pages in the bio but did not yet
5230 * unlocked the ranges in the io tree). Therefore this means some
5231 * ranges can still be locked and eviction started because before
5232 * submitting those bios, which are executed by a separate task (work
5233 * queue kthread), inode references (inode->i_count) were not taken
5234 * (which would be dropped in the end io callback of each bio).
5235 * Therefore here we effectively end up waiting for those bios and
5236 * anyone else holding locked ranges without having bumped the inode's
5237 * reference count - if we don't do it, when they access the inode's
5238 * io_tree to unlock a range it may be too late, leading to an
5239 * use-after-free issue.
5241 spin_lock(&io_tree->lock);
5242 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5243 struct extent_state *state;
5244 struct extent_state *cached_state = NULL;
5247 unsigned state_flags;
5249 node = rb_first(&io_tree->state);
5250 state = rb_entry(node, struct extent_state, rb_node);
5251 start = state->start;
5253 state_flags = state->state;
5254 spin_unlock(&io_tree->lock);
5256 lock_extent(io_tree, start, end, &cached_state);
5259 * If still has DELALLOC flag, the extent didn't reach disk,
5260 * and its reserved space won't be freed by delayed_ref.
5261 * So we need to free its reserved space here.
5262 * (Refer to comment in btrfs_invalidate_folio, case 2)
5264 * Note, end is the bytenr of last byte, so we need + 1 here.
5266 if (state_flags & EXTENT_DELALLOC)
5267 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5270 clear_extent_bit(io_tree, start, end,
5271 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5275 spin_lock(&io_tree->lock);
5277 spin_unlock(&io_tree->lock);
5280 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5281 struct btrfs_block_rsv *rsv)
5283 struct btrfs_fs_info *fs_info = root->fs_info;
5284 struct btrfs_trans_handle *trans;
5285 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5289 * Eviction should be taking place at some place safe because of our
5290 * delayed iputs. However the normal flushing code will run delayed
5291 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5293 * We reserve the delayed_refs_extra here again because we can't use
5294 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5295 * above. We reserve our extra bit here because we generate a ton of
5296 * delayed refs activity by truncating.
5298 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5299 * if we fail to make this reservation we can re-try without the
5300 * delayed_refs_extra so we can make some forward progress.
5302 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5303 BTRFS_RESERVE_FLUSH_EVICT);
5305 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5306 BTRFS_RESERVE_FLUSH_EVICT);
5309 "could not allocate space for delete; will truncate on mount");
5310 return ERR_PTR(-ENOSPC);
5312 delayed_refs_extra = 0;
5315 trans = btrfs_join_transaction(root);
5319 if (delayed_refs_extra) {
5320 trans->block_rsv = &fs_info->trans_block_rsv;
5321 trans->bytes_reserved = delayed_refs_extra;
5322 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5323 delayed_refs_extra, true);
5328 void btrfs_evict_inode(struct inode *inode)
5330 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5331 struct btrfs_trans_handle *trans;
5332 struct btrfs_root *root = BTRFS_I(inode)->root;
5333 struct btrfs_block_rsv *rsv = NULL;
5336 trace_btrfs_inode_evict(inode);
5339 fsverity_cleanup_inode(inode);
5344 evict_inode_truncate_pages(inode);
5346 if (inode->i_nlink &&
5347 ((btrfs_root_refs(&root->root_item) != 0 &&
5348 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5349 btrfs_is_free_space_inode(BTRFS_I(inode))))
5352 if (is_bad_inode(inode))
5355 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5358 if (inode->i_nlink > 0) {
5359 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5360 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5365 * This makes sure the inode item in tree is uptodate and the space for
5366 * the inode update is released.
5368 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5373 * This drops any pending insert or delete operations we have for this
5374 * inode. We could have a delayed dir index deletion queued up, but
5375 * we're removing the inode completely so that'll be taken care of in
5378 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5380 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5383 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5384 rsv->failfast = true;
5386 btrfs_i_size_write(BTRFS_I(inode), 0);
5389 struct btrfs_truncate_control control = {
5390 .inode = BTRFS_I(inode),
5391 .ino = btrfs_ino(BTRFS_I(inode)),
5396 trans = evict_refill_and_join(root, rsv);
5400 trans->block_rsv = rsv;
5402 ret = btrfs_truncate_inode_items(trans, root, &control);
5403 trans->block_rsv = &fs_info->trans_block_rsv;
5404 btrfs_end_transaction(trans);
5406 * We have not added new delayed items for our inode after we
5407 * have flushed its delayed items, so no need to throttle on
5408 * delayed items. However we have modified extent buffers.
5410 btrfs_btree_balance_dirty_nodelay(fs_info);
5411 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5418 * Errors here aren't a big deal, it just means we leave orphan items in
5419 * the tree. They will be cleaned up on the next mount. If the inode
5420 * number gets reused, cleanup deletes the orphan item without doing
5421 * anything, and unlink reuses the existing orphan item.
5423 * If it turns out that we are dropping too many of these, we might want
5424 * to add a mechanism for retrying these after a commit.
5426 trans = evict_refill_and_join(root, rsv);
5427 if (!IS_ERR(trans)) {
5428 trans->block_rsv = rsv;
5429 btrfs_orphan_del(trans, BTRFS_I(inode));
5430 trans->block_rsv = &fs_info->trans_block_rsv;
5431 btrfs_end_transaction(trans);
5435 btrfs_free_block_rsv(fs_info, rsv);
5437 * If we didn't successfully delete, the orphan item will still be in
5438 * the tree and we'll retry on the next mount. Again, we might also want
5439 * to retry these periodically in the future.
5441 btrfs_remove_delayed_node(BTRFS_I(inode));
5442 fsverity_cleanup_inode(inode);
5447 * Return the key found in the dir entry in the location pointer, fill @type
5448 * with BTRFS_FT_*, and return 0.
5450 * If no dir entries were found, returns -ENOENT.
5451 * If found a corrupted location in dir entry, returns -EUCLEAN.
5453 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5454 struct btrfs_key *location, u8 *type)
5456 struct btrfs_dir_item *di;
5457 struct btrfs_path *path;
5458 struct btrfs_root *root = dir->root;
5460 struct fscrypt_name fname;
5462 path = btrfs_alloc_path();
5466 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5470 * fscrypt_setup_filename() should never return a positive value, but
5471 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5475 /* This needs to handle no-key deletions later on */
5477 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5478 &fname.disk_name, 0);
5479 if (IS_ERR_OR_NULL(di)) {
5480 ret = di ? PTR_ERR(di) : -ENOENT;
5484 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5485 if (location->type != BTRFS_INODE_ITEM_KEY &&
5486 location->type != BTRFS_ROOT_ITEM_KEY) {
5488 btrfs_warn(root->fs_info,
5489 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5490 __func__, fname.disk_name.name, btrfs_ino(dir),
5491 location->objectid, location->type, location->offset);
5494 *type = btrfs_dir_ftype(path->nodes[0], di);
5496 fscrypt_free_filename(&fname);
5497 btrfs_free_path(path);
5502 * when we hit a tree root in a directory, the btrfs part of the inode
5503 * needs to be changed to reflect the root directory of the tree root. This
5504 * is kind of like crossing a mount point.
5506 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5507 struct btrfs_inode *dir,
5508 struct dentry *dentry,
5509 struct btrfs_key *location,
5510 struct btrfs_root **sub_root)
5512 struct btrfs_path *path;
5513 struct btrfs_root *new_root;
5514 struct btrfs_root_ref *ref;
5515 struct extent_buffer *leaf;
5516 struct btrfs_key key;
5519 struct fscrypt_name fname;
5521 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5525 path = btrfs_alloc_path();
5532 key.objectid = dir->root->root_key.objectid;
5533 key.type = BTRFS_ROOT_REF_KEY;
5534 key.offset = location->objectid;
5536 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5543 leaf = path->nodes[0];
5544 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5545 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5546 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5549 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5550 (unsigned long)(ref + 1), fname.disk_name.len);
5554 btrfs_release_path(path);
5556 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5557 if (IS_ERR(new_root)) {
5558 err = PTR_ERR(new_root);
5562 *sub_root = new_root;
5563 location->objectid = btrfs_root_dirid(&new_root->root_item);
5564 location->type = BTRFS_INODE_ITEM_KEY;
5565 location->offset = 0;
5568 btrfs_free_path(path);
5569 fscrypt_free_filename(&fname);
5573 static void inode_tree_add(struct btrfs_inode *inode)
5575 struct btrfs_root *root = inode->root;
5576 struct btrfs_inode *entry;
5578 struct rb_node *parent;
5579 struct rb_node *new = &inode->rb_node;
5580 u64 ino = btrfs_ino(inode);
5582 if (inode_unhashed(&inode->vfs_inode))
5585 spin_lock(&root->inode_lock);
5586 p = &root->inode_tree.rb_node;
5589 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5591 if (ino < btrfs_ino(entry))
5592 p = &parent->rb_left;
5593 else if (ino > btrfs_ino(entry))
5594 p = &parent->rb_right;
5596 WARN_ON(!(entry->vfs_inode.i_state &
5597 (I_WILL_FREE | I_FREEING)));
5598 rb_replace_node(parent, new, &root->inode_tree);
5599 RB_CLEAR_NODE(parent);
5600 spin_unlock(&root->inode_lock);
5604 rb_link_node(new, parent, p);
5605 rb_insert_color(new, &root->inode_tree);
5606 spin_unlock(&root->inode_lock);
5609 static void inode_tree_del(struct btrfs_inode *inode)
5611 struct btrfs_root *root = inode->root;
5614 spin_lock(&root->inode_lock);
5615 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5616 rb_erase(&inode->rb_node, &root->inode_tree);
5617 RB_CLEAR_NODE(&inode->rb_node);
5618 empty = RB_EMPTY_ROOT(&root->inode_tree);
5620 spin_unlock(&root->inode_lock);
5622 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5623 spin_lock(&root->inode_lock);
5624 empty = RB_EMPTY_ROOT(&root->inode_tree);
5625 spin_unlock(&root->inode_lock);
5627 btrfs_add_dead_root(root);
5632 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5634 struct btrfs_iget_args *args = p;
5636 inode->i_ino = args->ino;
5637 BTRFS_I(inode)->location.objectid = args->ino;
5638 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5639 BTRFS_I(inode)->location.offset = 0;
5640 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5641 BUG_ON(args->root && !BTRFS_I(inode)->root);
5643 if (args->root && args->root == args->root->fs_info->tree_root &&
5644 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5645 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5646 &BTRFS_I(inode)->runtime_flags);
5650 static int btrfs_find_actor(struct inode *inode, void *opaque)
5652 struct btrfs_iget_args *args = opaque;
5654 return args->ino == BTRFS_I(inode)->location.objectid &&
5655 args->root == BTRFS_I(inode)->root;
5658 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5659 struct btrfs_root *root)
5661 struct inode *inode;
5662 struct btrfs_iget_args args;
5663 unsigned long hashval = btrfs_inode_hash(ino, root);
5668 inode = iget5_locked(s, hashval, btrfs_find_actor,
5669 btrfs_init_locked_inode,
5675 * Get an inode object given its inode number and corresponding root.
5676 * Path can be preallocated to prevent recursing back to iget through
5677 * allocator. NULL is also valid but may require an additional allocation
5680 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5681 struct btrfs_root *root, struct btrfs_path *path)
5683 struct inode *inode;
5685 inode = btrfs_iget_locked(s, ino, root);
5687 return ERR_PTR(-ENOMEM);
5689 if (inode->i_state & I_NEW) {
5692 ret = btrfs_read_locked_inode(inode, path);
5694 inode_tree_add(BTRFS_I(inode));
5695 unlock_new_inode(inode);
5699 * ret > 0 can come from btrfs_search_slot called by
5700 * btrfs_read_locked_inode, this means the inode item
5705 inode = ERR_PTR(ret);
5712 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5714 return btrfs_iget_path(s, ino, root, NULL);
5717 static struct inode *new_simple_dir(struct super_block *s,
5718 struct btrfs_key *key,
5719 struct btrfs_root *root)
5721 struct inode *inode = new_inode(s);
5724 return ERR_PTR(-ENOMEM);
5726 BTRFS_I(inode)->root = btrfs_grab_root(root);
5727 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5728 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5730 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5732 * We only need lookup, the rest is read-only and there's no inode
5733 * associated with the dentry
5735 inode->i_op = &simple_dir_inode_operations;
5736 inode->i_opflags &= ~IOP_XATTR;
5737 inode->i_fop = &simple_dir_operations;
5738 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5739 inode->i_mtime = current_time(inode);
5740 inode->i_atime = inode->i_mtime;
5741 inode->i_ctime = inode->i_mtime;
5742 BTRFS_I(inode)->i_otime = inode->i_mtime;
5747 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5748 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5749 static_assert(BTRFS_FT_DIR == FT_DIR);
5750 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5751 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5752 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5753 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5754 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5756 static inline u8 btrfs_inode_type(struct inode *inode)
5758 return fs_umode_to_ftype(inode->i_mode);
5761 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5763 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5764 struct inode *inode;
5765 struct btrfs_root *root = BTRFS_I(dir)->root;
5766 struct btrfs_root *sub_root = root;
5767 struct btrfs_key location;
5771 if (dentry->d_name.len > BTRFS_NAME_LEN)
5772 return ERR_PTR(-ENAMETOOLONG);
5774 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5776 return ERR_PTR(ret);
5778 if (location.type == BTRFS_INODE_ITEM_KEY) {
5779 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5783 /* Do extra check against inode mode with di_type */
5784 if (btrfs_inode_type(inode) != di_type) {
5786 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5787 inode->i_mode, btrfs_inode_type(inode),
5790 return ERR_PTR(-EUCLEAN);
5795 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5796 &location, &sub_root);
5799 inode = ERR_PTR(ret);
5801 inode = new_simple_dir(dir->i_sb, &location, root);
5803 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5804 btrfs_put_root(sub_root);
5809 down_read(&fs_info->cleanup_work_sem);
5810 if (!sb_rdonly(inode->i_sb))
5811 ret = btrfs_orphan_cleanup(sub_root);
5812 up_read(&fs_info->cleanup_work_sem);
5815 inode = ERR_PTR(ret);
5822 static int btrfs_dentry_delete(const struct dentry *dentry)
5824 struct btrfs_root *root;
5825 struct inode *inode = d_inode(dentry);
5827 if (!inode && !IS_ROOT(dentry))
5828 inode = d_inode(dentry->d_parent);
5831 root = BTRFS_I(inode)->root;
5832 if (btrfs_root_refs(&root->root_item) == 0)
5835 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5841 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5844 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5846 if (inode == ERR_PTR(-ENOENT))
5848 return d_splice_alias(inode, dentry);
5852 * All this infrastructure exists because dir_emit can fault, and we are holding
5853 * the tree lock when doing readdir. For now just allocate a buffer and copy
5854 * our information into that, and then dir_emit from the buffer. This is
5855 * similar to what NFS does, only we don't keep the buffer around in pagecache
5856 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5857 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5860 static int btrfs_opendir(struct inode *inode, struct file *file)
5862 struct btrfs_file_private *private;
5864 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5867 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5868 if (!private->filldir_buf) {
5872 file->private_data = private;
5883 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5886 struct dir_entry *entry = addr;
5887 char *name = (char *)(entry + 1);
5889 ctx->pos = get_unaligned(&entry->offset);
5890 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5891 get_unaligned(&entry->ino),
5892 get_unaligned(&entry->type)))
5894 addr += sizeof(struct dir_entry) +
5895 get_unaligned(&entry->name_len);
5901 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5903 struct inode *inode = file_inode(file);
5904 struct btrfs_root *root = BTRFS_I(inode)->root;
5905 struct btrfs_file_private *private = file->private_data;
5906 struct btrfs_dir_item *di;
5907 struct btrfs_key key;
5908 struct btrfs_key found_key;
5909 struct btrfs_path *path;
5911 struct list_head ins_list;
5912 struct list_head del_list;
5919 struct btrfs_key location;
5921 if (!dir_emit_dots(file, ctx))
5924 path = btrfs_alloc_path();
5928 addr = private->filldir_buf;
5929 path->reada = READA_FORWARD;
5931 INIT_LIST_HEAD(&ins_list);
5932 INIT_LIST_HEAD(&del_list);
5933 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5936 key.type = BTRFS_DIR_INDEX_KEY;
5937 key.offset = ctx->pos;
5938 key.objectid = btrfs_ino(BTRFS_I(inode));
5940 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5941 struct dir_entry *entry;
5942 struct extent_buffer *leaf = path->nodes[0];
5945 if (found_key.objectid != key.objectid)
5947 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5949 if (found_key.offset < ctx->pos)
5951 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5953 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5954 name_len = btrfs_dir_name_len(leaf, di);
5955 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5957 btrfs_release_path(path);
5958 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5961 addr = private->filldir_buf;
5967 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5969 name_ptr = (char *)(entry + 1);
5970 read_extent_buffer(leaf, name_ptr,
5971 (unsigned long)(di + 1), name_len);
5972 put_unaligned(name_len, &entry->name_len);
5973 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5974 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5975 put_unaligned(location.objectid, &entry->ino);
5976 put_unaligned(found_key.offset, &entry->offset);
5978 addr += sizeof(struct dir_entry) + name_len;
5979 total_len += sizeof(struct dir_entry) + name_len;
5981 /* Catch error encountered during iteration */
5985 btrfs_release_path(path);
5987 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5991 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5996 * Stop new entries from being returned after we return the last
5999 * New directory entries are assigned a strictly increasing
6000 * offset. This means that new entries created during readdir
6001 * are *guaranteed* to be seen in the future by that readdir.
6002 * This has broken buggy programs which operate on names as
6003 * they're returned by readdir. Until we re-use freed offsets
6004 * we have this hack to stop new entries from being returned
6005 * under the assumption that they'll never reach this huge
6008 * This is being careful not to overflow 32bit loff_t unless the
6009 * last entry requires it because doing so has broken 32bit apps
6012 if (ctx->pos >= INT_MAX)
6013 ctx->pos = LLONG_MAX;
6020 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6021 btrfs_free_path(path);
6026 * This is somewhat expensive, updating the tree every time the
6027 * inode changes. But, it is most likely to find the inode in cache.
6028 * FIXME, needs more benchmarking...there are no reasons other than performance
6029 * to keep or drop this code.
6031 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6033 struct btrfs_root *root = inode->root;
6034 struct btrfs_fs_info *fs_info = root->fs_info;
6035 struct btrfs_trans_handle *trans;
6038 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6041 trans = btrfs_join_transaction(root);
6043 return PTR_ERR(trans);
6045 ret = btrfs_update_inode(trans, root, inode);
6046 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6047 /* whoops, lets try again with the full transaction */
6048 btrfs_end_transaction(trans);
6049 trans = btrfs_start_transaction(root, 1);
6051 return PTR_ERR(trans);
6053 ret = btrfs_update_inode(trans, root, inode);
6055 btrfs_end_transaction(trans);
6056 if (inode->delayed_node)
6057 btrfs_balance_delayed_items(fs_info);
6063 * This is a copy of file_update_time. We need this so we can return error on
6064 * ENOSPC for updating the inode in the case of file write and mmap writes.
6066 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6069 struct btrfs_root *root = BTRFS_I(inode)->root;
6070 bool dirty = flags & ~S_VERSION;
6072 if (btrfs_root_readonly(root))
6075 if (flags & S_VERSION)
6076 dirty |= inode_maybe_inc_iversion(inode, dirty);
6077 if (flags & S_CTIME)
6078 inode->i_ctime = *now;
6079 if (flags & S_MTIME)
6080 inode->i_mtime = *now;
6081 if (flags & S_ATIME)
6082 inode->i_atime = *now;
6083 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6087 * find the highest existing sequence number in a directory
6088 * and then set the in-memory index_cnt variable to reflect
6089 * free sequence numbers
6091 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6093 struct btrfs_root *root = inode->root;
6094 struct btrfs_key key, found_key;
6095 struct btrfs_path *path;
6096 struct extent_buffer *leaf;
6099 key.objectid = btrfs_ino(inode);
6100 key.type = BTRFS_DIR_INDEX_KEY;
6101 key.offset = (u64)-1;
6103 path = btrfs_alloc_path();
6107 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6110 /* FIXME: we should be able to handle this */
6115 if (path->slots[0] == 0) {
6116 inode->index_cnt = BTRFS_DIR_START_INDEX;
6122 leaf = path->nodes[0];
6123 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6125 if (found_key.objectid != btrfs_ino(inode) ||
6126 found_key.type != BTRFS_DIR_INDEX_KEY) {
6127 inode->index_cnt = BTRFS_DIR_START_INDEX;
6131 inode->index_cnt = found_key.offset + 1;
6133 btrfs_free_path(path);
6138 * helper to find a free sequence number in a given directory. This current
6139 * code is very simple, later versions will do smarter things in the btree
6141 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6145 if (dir->index_cnt == (u64)-1) {
6146 ret = btrfs_inode_delayed_dir_index_count(dir);
6148 ret = btrfs_set_inode_index_count(dir);
6154 *index = dir->index_cnt;
6160 static int btrfs_insert_inode_locked(struct inode *inode)
6162 struct btrfs_iget_args args;
6164 args.ino = BTRFS_I(inode)->location.objectid;
6165 args.root = BTRFS_I(inode)->root;
6167 return insert_inode_locked4(inode,
6168 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6169 btrfs_find_actor, &args);
6172 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6173 unsigned int *trans_num_items)
6175 struct inode *dir = args->dir;
6176 struct inode *inode = args->inode;
6179 if (!args->orphan) {
6180 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6186 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6188 fscrypt_free_filename(&args->fname);
6192 /* 1 to add inode item */
6193 *trans_num_items = 1;
6194 /* 1 to add compression property */
6195 if (BTRFS_I(dir)->prop_compress)
6196 (*trans_num_items)++;
6197 /* 1 to add default ACL xattr */
6198 if (args->default_acl)
6199 (*trans_num_items)++;
6200 /* 1 to add access ACL xattr */
6202 (*trans_num_items)++;
6203 #ifdef CONFIG_SECURITY
6204 /* 1 to add LSM xattr */
6205 if (dir->i_security)
6206 (*trans_num_items)++;
6209 /* 1 to add orphan item */
6210 (*trans_num_items)++;
6214 * 1 to add dir index
6215 * 1 to update parent inode item
6217 * No need for 1 unit for the inode ref item because it is
6218 * inserted in a batch together with the inode item at
6219 * btrfs_create_new_inode().
6221 *trans_num_items += 3;
6226 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6228 posix_acl_release(args->acl);
6229 posix_acl_release(args->default_acl);
6230 fscrypt_free_filename(&args->fname);
6234 * Inherit flags from the parent inode.
6236 * Currently only the compression flags and the cow flags are inherited.
6238 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6244 if (flags & BTRFS_INODE_NOCOMPRESS) {
6245 inode->flags &= ~BTRFS_INODE_COMPRESS;
6246 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6247 } else if (flags & BTRFS_INODE_COMPRESS) {
6248 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6249 inode->flags |= BTRFS_INODE_COMPRESS;
6252 if (flags & BTRFS_INODE_NODATACOW) {
6253 inode->flags |= BTRFS_INODE_NODATACOW;
6254 if (S_ISREG(inode->vfs_inode.i_mode))
6255 inode->flags |= BTRFS_INODE_NODATASUM;
6258 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6261 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6262 struct btrfs_new_inode_args *args)
6264 struct inode *dir = args->dir;
6265 struct inode *inode = args->inode;
6266 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6267 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6268 struct btrfs_root *root;
6269 struct btrfs_inode_item *inode_item;
6270 struct btrfs_key *location;
6271 struct btrfs_path *path;
6273 struct btrfs_inode_ref *ref;
6274 struct btrfs_key key[2];
6276 struct btrfs_item_batch batch;
6280 path = btrfs_alloc_path();
6285 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6286 root = BTRFS_I(inode)->root;
6288 ret = btrfs_get_free_objectid(root, &objectid);
6291 inode->i_ino = objectid;
6295 * O_TMPFILE, set link count to 0, so that after this point, we
6296 * fill in an inode item with the correct link count.
6298 set_nlink(inode, 0);
6300 trace_btrfs_inode_request(dir);
6302 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6306 /* index_cnt is ignored for everything but a dir. */
6307 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6308 BTRFS_I(inode)->generation = trans->transid;
6309 inode->i_generation = BTRFS_I(inode)->generation;
6312 * Subvolumes don't inherit flags from their parent directory.
6313 * Originally this was probably by accident, but we probably can't
6314 * change it now without compatibility issues.
6317 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6319 if (S_ISREG(inode->i_mode)) {
6320 if (btrfs_test_opt(fs_info, NODATASUM))
6321 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6322 if (btrfs_test_opt(fs_info, NODATACOW))
6323 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6324 BTRFS_INODE_NODATASUM;
6327 location = &BTRFS_I(inode)->location;
6328 location->objectid = objectid;
6329 location->offset = 0;
6330 location->type = BTRFS_INODE_ITEM_KEY;
6332 ret = btrfs_insert_inode_locked(inode);
6335 BTRFS_I(dir)->index_cnt--;
6340 * We could have gotten an inode number from somebody who was fsynced
6341 * and then removed in this same transaction, so let's just set full
6342 * sync since it will be a full sync anyway and this will blow away the
6343 * old info in the log.
6345 btrfs_set_inode_full_sync(BTRFS_I(inode));
6347 key[0].objectid = objectid;
6348 key[0].type = BTRFS_INODE_ITEM_KEY;
6351 sizes[0] = sizeof(struct btrfs_inode_item);
6353 if (!args->orphan) {
6355 * Start new inodes with an inode_ref. This is slightly more
6356 * efficient for small numbers of hard links since they will
6357 * be packed into one item. Extended refs will kick in if we
6358 * add more hard links than can fit in the ref item.
6360 key[1].objectid = objectid;
6361 key[1].type = BTRFS_INODE_REF_KEY;
6363 key[1].offset = objectid;
6364 sizes[1] = 2 + sizeof(*ref);
6366 key[1].offset = btrfs_ino(BTRFS_I(dir));
6367 sizes[1] = name->len + sizeof(*ref);
6371 batch.keys = &key[0];
6372 batch.data_sizes = &sizes[0];
6373 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6374 batch.nr = args->orphan ? 1 : 2;
6375 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6377 btrfs_abort_transaction(trans, ret);
6381 inode->i_mtime = current_time(inode);
6382 inode->i_atime = inode->i_mtime;
6383 inode->i_ctime = inode->i_mtime;
6384 BTRFS_I(inode)->i_otime = inode->i_mtime;
6387 * We're going to fill the inode item now, so at this point the inode
6388 * must be fully initialized.
6391 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6392 struct btrfs_inode_item);
6393 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6394 sizeof(*inode_item));
6395 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6397 if (!args->orphan) {
6398 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6399 struct btrfs_inode_ref);
6400 ptr = (unsigned long)(ref + 1);
6402 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6403 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6404 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6406 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6408 btrfs_set_inode_ref_index(path->nodes[0], ref,
6409 BTRFS_I(inode)->dir_index);
6410 write_extent_buffer(path->nodes[0], name->name, ptr,
6415 btrfs_mark_buffer_dirty(path->nodes[0]);
6417 * We don't need the path anymore, plus inheriting properties, adding
6418 * ACLs, security xattrs, orphan item or adding the link, will result in
6419 * allocating yet another path. So just free our path.
6421 btrfs_free_path(path);
6425 struct inode *parent;
6428 * Subvolumes inherit properties from their parent subvolume,
6429 * not the directory they were created in.
6431 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6432 BTRFS_I(dir)->root);
6433 if (IS_ERR(parent)) {
6434 ret = PTR_ERR(parent);
6436 ret = btrfs_inode_inherit_props(trans, inode, parent);
6440 ret = btrfs_inode_inherit_props(trans, inode, dir);
6444 "error inheriting props for ino %llu (root %llu): %d",
6445 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6450 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6453 if (!args->subvol) {
6454 ret = btrfs_init_inode_security(trans, args);
6456 btrfs_abort_transaction(trans, ret);
6461 inode_tree_add(BTRFS_I(inode));
6463 trace_btrfs_inode_new(inode);
6464 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6466 btrfs_update_root_times(trans, root);
6469 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6471 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6472 0, BTRFS_I(inode)->dir_index);
6475 btrfs_abort_transaction(trans, ret);
6483 * discard_new_inode() calls iput(), but the caller owns the reference
6487 discard_new_inode(inode);
6489 btrfs_free_path(path);
6494 * utility function to add 'inode' into 'parent_inode' with
6495 * a give name and a given sequence number.
6496 * if 'add_backref' is true, also insert a backref from the
6497 * inode to the parent directory.
6499 int btrfs_add_link(struct btrfs_trans_handle *trans,
6500 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6501 const struct fscrypt_str *name, int add_backref, u64 index)
6504 struct btrfs_key key;
6505 struct btrfs_root *root = parent_inode->root;
6506 u64 ino = btrfs_ino(inode);
6507 u64 parent_ino = btrfs_ino(parent_inode);
6509 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6510 memcpy(&key, &inode->root->root_key, sizeof(key));
6513 key.type = BTRFS_INODE_ITEM_KEY;
6517 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6518 ret = btrfs_add_root_ref(trans, key.objectid,
6519 root->root_key.objectid, parent_ino,
6521 } else if (add_backref) {
6522 ret = btrfs_insert_inode_ref(trans, root, name,
6523 ino, parent_ino, index);
6526 /* Nothing to clean up yet */
6530 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6531 btrfs_inode_type(&inode->vfs_inode), index);
6532 if (ret == -EEXIST || ret == -EOVERFLOW)
6535 btrfs_abort_transaction(trans, ret);
6539 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6541 inode_inc_iversion(&parent_inode->vfs_inode);
6543 * If we are replaying a log tree, we do not want to update the mtime
6544 * and ctime of the parent directory with the current time, since the
6545 * log replay procedure is responsible for setting them to their correct
6546 * values (the ones it had when the fsync was done).
6548 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6549 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6551 parent_inode->vfs_inode.i_mtime = now;
6552 parent_inode->vfs_inode.i_ctime = now;
6554 ret = btrfs_update_inode(trans, root, parent_inode);
6556 btrfs_abort_transaction(trans, ret);
6560 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6563 err = btrfs_del_root_ref(trans, key.objectid,
6564 root->root_key.objectid, parent_ino,
6565 &local_index, name);
6567 btrfs_abort_transaction(trans, err);
6568 } else if (add_backref) {
6572 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6575 btrfs_abort_transaction(trans, err);
6578 /* Return the original error code */
6582 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6583 struct inode *inode)
6585 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6586 struct btrfs_root *root = BTRFS_I(dir)->root;
6587 struct btrfs_new_inode_args new_inode_args = {
6592 unsigned int trans_num_items;
6593 struct btrfs_trans_handle *trans;
6596 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6600 trans = btrfs_start_transaction(root, trans_num_items);
6601 if (IS_ERR(trans)) {
6602 err = PTR_ERR(trans);
6603 goto out_new_inode_args;
6606 err = btrfs_create_new_inode(trans, &new_inode_args);
6608 d_instantiate_new(dentry, inode);
6610 btrfs_end_transaction(trans);
6611 btrfs_btree_balance_dirty(fs_info);
6613 btrfs_new_inode_args_destroy(&new_inode_args);
6620 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6621 struct dentry *dentry, umode_t mode, dev_t rdev)
6623 struct inode *inode;
6625 inode = new_inode(dir->i_sb);
6628 inode_init_owner(idmap, inode, dir, mode);
6629 inode->i_op = &btrfs_special_inode_operations;
6630 init_special_inode(inode, inode->i_mode, rdev);
6631 return btrfs_create_common(dir, dentry, inode);
6634 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6635 struct dentry *dentry, umode_t mode, bool excl)
6637 struct inode *inode;
6639 inode = new_inode(dir->i_sb);
6642 inode_init_owner(idmap, inode, dir, mode);
6643 inode->i_fop = &btrfs_file_operations;
6644 inode->i_op = &btrfs_file_inode_operations;
6645 inode->i_mapping->a_ops = &btrfs_aops;
6646 return btrfs_create_common(dir, dentry, inode);
6649 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6650 struct dentry *dentry)
6652 struct btrfs_trans_handle *trans = NULL;
6653 struct btrfs_root *root = BTRFS_I(dir)->root;
6654 struct inode *inode = d_inode(old_dentry);
6655 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6656 struct fscrypt_name fname;
6661 /* do not allow sys_link's with other subvols of the same device */
6662 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6665 if (inode->i_nlink >= BTRFS_LINK_MAX)
6668 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6672 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6677 * 2 items for inode and inode ref
6678 * 2 items for dir items
6679 * 1 item for parent inode
6680 * 1 item for orphan item deletion if O_TMPFILE
6682 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6683 if (IS_ERR(trans)) {
6684 err = PTR_ERR(trans);
6689 /* There are several dir indexes for this inode, clear the cache. */
6690 BTRFS_I(inode)->dir_index = 0ULL;
6692 inode_inc_iversion(inode);
6693 inode->i_ctime = current_time(inode);
6695 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6697 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6698 &fname.disk_name, 1, index);
6703 struct dentry *parent = dentry->d_parent;
6705 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6708 if (inode->i_nlink == 1) {
6710 * If new hard link count is 1, it's a file created
6711 * with open(2) O_TMPFILE flag.
6713 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6717 d_instantiate(dentry, inode);
6718 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6722 fscrypt_free_filename(&fname);
6724 btrfs_end_transaction(trans);
6726 inode_dec_link_count(inode);
6729 btrfs_btree_balance_dirty(fs_info);
6733 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6734 struct dentry *dentry, umode_t mode)
6736 struct inode *inode;
6738 inode = new_inode(dir->i_sb);
6741 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6742 inode->i_op = &btrfs_dir_inode_operations;
6743 inode->i_fop = &btrfs_dir_file_operations;
6744 return btrfs_create_common(dir, dentry, inode);
6747 static noinline int uncompress_inline(struct btrfs_path *path,
6749 struct btrfs_file_extent_item *item)
6752 struct extent_buffer *leaf = path->nodes[0];
6755 unsigned long inline_size;
6759 compress_type = btrfs_file_extent_compression(leaf, item);
6760 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6761 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6762 tmp = kmalloc(inline_size, GFP_NOFS);
6765 ptr = btrfs_file_extent_inline_start(item);
6767 read_extent_buffer(leaf, tmp, ptr, inline_size);
6769 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6770 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6773 * decompression code contains a memset to fill in any space between the end
6774 * of the uncompressed data and the end of max_size in case the decompressed
6775 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6776 * the end of an inline extent and the beginning of the next block, so we
6777 * cover that region here.
6780 if (max_size < PAGE_SIZE)
6781 memzero_page(page, max_size, PAGE_SIZE - max_size);
6786 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6789 struct btrfs_file_extent_item *fi;
6793 if (!page || PageUptodate(page))
6796 ASSERT(page_offset(page) == 0);
6798 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6799 struct btrfs_file_extent_item);
6800 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6801 return uncompress_inline(path, page, fi);
6803 copy_size = min_t(u64, PAGE_SIZE,
6804 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6805 kaddr = kmap_local_page(page);
6806 read_extent_buffer(path->nodes[0], kaddr,
6807 btrfs_file_extent_inline_start(fi), copy_size);
6808 kunmap_local(kaddr);
6809 if (copy_size < PAGE_SIZE)
6810 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6815 * Lookup the first extent overlapping a range in a file.
6817 * @inode: file to search in
6818 * @page: page to read extent data into if the extent is inline
6819 * @pg_offset: offset into @page to copy to
6820 * @start: file offset
6821 * @len: length of range starting at @start
6823 * Return the first &struct extent_map which overlaps the given range, reading
6824 * it from the B-tree and caching it if necessary. Note that there may be more
6825 * extents which overlap the given range after the returned extent_map.
6827 * If @page is not NULL and the extent is inline, this also reads the extent
6828 * data directly into the page and marks the extent up to date in the io_tree.
6830 * Return: ERR_PTR on error, non-NULL extent_map on success.
6832 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6833 struct page *page, size_t pg_offset,
6836 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6838 u64 extent_start = 0;
6840 u64 objectid = btrfs_ino(inode);
6841 int extent_type = -1;
6842 struct btrfs_path *path = NULL;
6843 struct btrfs_root *root = inode->root;
6844 struct btrfs_file_extent_item *item;
6845 struct extent_buffer *leaf;
6846 struct btrfs_key found_key;
6847 struct extent_map *em = NULL;
6848 struct extent_map_tree *em_tree = &inode->extent_tree;
6850 read_lock(&em_tree->lock);
6851 em = lookup_extent_mapping(em_tree, start, len);
6852 read_unlock(&em_tree->lock);
6855 if (em->start > start || em->start + em->len <= start)
6856 free_extent_map(em);
6857 else if (em->block_start == EXTENT_MAP_INLINE && page)
6858 free_extent_map(em);
6862 em = alloc_extent_map();
6867 em->start = EXTENT_MAP_HOLE;
6868 em->orig_start = EXTENT_MAP_HOLE;
6870 em->block_len = (u64)-1;
6872 path = btrfs_alloc_path();
6878 /* Chances are we'll be called again, so go ahead and do readahead */
6879 path->reada = READA_FORWARD;
6882 * The same explanation in load_free_space_cache applies here as well,
6883 * we only read when we're loading the free space cache, and at that
6884 * point the commit_root has everything we need.
6886 if (btrfs_is_free_space_inode(inode)) {
6887 path->search_commit_root = 1;
6888 path->skip_locking = 1;
6891 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6894 } else if (ret > 0) {
6895 if (path->slots[0] == 0)
6901 leaf = path->nodes[0];
6902 item = btrfs_item_ptr(leaf, path->slots[0],
6903 struct btrfs_file_extent_item);
6904 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6905 if (found_key.objectid != objectid ||
6906 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6908 * If we backup past the first extent we want to move forward
6909 * and see if there is an extent in front of us, otherwise we'll
6910 * say there is a hole for our whole search range which can
6917 extent_type = btrfs_file_extent_type(leaf, item);
6918 extent_start = found_key.offset;
6919 extent_end = btrfs_file_extent_end(path);
6920 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6921 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6922 /* Only regular file could have regular/prealloc extent */
6923 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6926 "regular/prealloc extent found for non-regular inode %llu",
6930 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6932 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6933 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6938 if (start >= extent_end) {
6940 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6941 ret = btrfs_next_leaf(root, path);
6947 leaf = path->nodes[0];
6949 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6950 if (found_key.objectid != objectid ||
6951 found_key.type != BTRFS_EXTENT_DATA_KEY)
6953 if (start + len <= found_key.offset)
6955 if (start > found_key.offset)
6958 /* New extent overlaps with existing one */
6960 em->orig_start = start;
6961 em->len = found_key.offset - start;
6962 em->block_start = EXTENT_MAP_HOLE;
6966 btrfs_extent_item_to_extent_map(inode, path, item, em);
6968 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6969 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6971 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6973 * Inline extent can only exist at file offset 0. This is
6974 * ensured by tree-checker and inline extent creation path.
6975 * Thus all members representing file offsets should be zero.
6977 ASSERT(pg_offset == 0);
6978 ASSERT(extent_start == 0);
6979 ASSERT(em->start == 0);
6982 * btrfs_extent_item_to_extent_map() should have properly
6983 * initialized em members already.
6985 * Other members are not utilized for inline extents.
6987 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6988 ASSERT(em->len == fs_info->sectorsize);
6990 ret = read_inline_extent(inode, path, page);
6997 em->orig_start = start;
6999 em->block_start = EXTENT_MAP_HOLE;
7002 btrfs_release_path(path);
7003 if (em->start > start || extent_map_end(em) <= start) {
7005 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7006 em->start, em->len, start, len);
7011 write_lock(&em_tree->lock);
7012 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7013 write_unlock(&em_tree->lock);
7015 btrfs_free_path(path);
7017 trace_btrfs_get_extent(root, inode, em);
7020 free_extent_map(em);
7021 return ERR_PTR(ret);
7026 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7027 struct btrfs_dio_data *dio_data,
7030 const u64 orig_start,
7031 const u64 block_start,
7032 const u64 block_len,
7033 const u64 orig_block_len,
7034 const u64 ram_bytes,
7037 struct extent_map *em = NULL;
7038 struct btrfs_ordered_extent *ordered;
7040 if (type != BTRFS_ORDERED_NOCOW) {
7041 em = create_io_em(inode, start, len, orig_start, block_start,
7042 block_len, orig_block_len, ram_bytes,
7043 BTRFS_COMPRESS_NONE, /* compress_type */
7048 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7049 block_start, block_len, 0,
7051 (1 << BTRFS_ORDERED_DIRECT),
7052 BTRFS_COMPRESS_NONE);
7053 if (IS_ERR(ordered)) {
7055 free_extent_map(em);
7056 btrfs_drop_extent_map_range(inode, start,
7057 start + len - 1, false);
7059 em = ERR_CAST(ordered);
7061 ASSERT(!dio_data->ordered);
7062 dio_data->ordered = ordered;
7069 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7070 struct btrfs_dio_data *dio_data,
7073 struct btrfs_root *root = inode->root;
7074 struct btrfs_fs_info *fs_info = root->fs_info;
7075 struct extent_map *em;
7076 struct btrfs_key ins;
7080 alloc_hint = get_extent_allocation_hint(inode, start, len);
7081 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7082 0, alloc_hint, &ins, 1, 1);
7084 return ERR_PTR(ret);
7086 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7087 ins.objectid, ins.offset, ins.offset,
7088 ins.offset, BTRFS_ORDERED_REGULAR);
7089 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7091 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7097 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7099 struct btrfs_block_group *block_group;
7100 bool readonly = false;
7102 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7103 if (!block_group || block_group->ro)
7106 btrfs_put_block_group(block_group);
7111 * Check if we can do nocow write into the range [@offset, @offset + @len)
7113 * @offset: File offset
7114 * @len: The length to write, will be updated to the nocow writeable
7116 * @orig_start: (optional) Return the original file offset of the file extent
7117 * @orig_len: (optional) Return the original on-disk length of the file extent
7118 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7119 * @strict: if true, omit optimizations that might force us into unnecessary
7120 * cow. e.g., don't trust generation number.
7123 * >0 and update @len if we can do nocow write
7124 * 0 if we can't do nocow write
7125 * <0 if error happened
7127 * NOTE: This only checks the file extents, caller is responsible to wait for
7128 * any ordered extents.
7130 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7131 u64 *orig_start, u64 *orig_block_len,
7132 u64 *ram_bytes, bool nowait, bool strict)
7134 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7135 struct can_nocow_file_extent_args nocow_args = { 0 };
7136 struct btrfs_path *path;
7138 struct extent_buffer *leaf;
7139 struct btrfs_root *root = BTRFS_I(inode)->root;
7140 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7141 struct btrfs_file_extent_item *fi;
7142 struct btrfs_key key;
7145 path = btrfs_alloc_path();
7148 path->nowait = nowait;
7150 ret = btrfs_lookup_file_extent(NULL, root, path,
7151 btrfs_ino(BTRFS_I(inode)), offset, 0);
7156 if (path->slots[0] == 0) {
7157 /* can't find the item, must cow */
7164 leaf = path->nodes[0];
7165 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7166 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7167 key.type != BTRFS_EXTENT_DATA_KEY) {
7168 /* not our file or wrong item type, must cow */
7172 if (key.offset > offset) {
7173 /* Wrong offset, must cow */
7177 if (btrfs_file_extent_end(path) <= offset)
7180 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7181 found_type = btrfs_file_extent_type(leaf, fi);
7183 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7185 nocow_args.start = offset;
7186 nocow_args.end = offset + *len - 1;
7187 nocow_args.strict = strict;
7188 nocow_args.free_path = true;
7190 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7191 /* can_nocow_file_extent() has freed the path. */
7195 /* Treat errors as not being able to NOCOW. */
7201 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7204 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7205 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7208 range_end = round_up(offset + nocow_args.num_bytes,
7209 root->fs_info->sectorsize) - 1;
7210 ret = test_range_bit(io_tree, offset, range_end,
7211 EXTENT_DELALLOC, 0, NULL);
7219 *orig_start = key.offset - nocow_args.extent_offset;
7221 *orig_block_len = nocow_args.disk_num_bytes;
7223 *len = nocow_args.num_bytes;
7226 btrfs_free_path(path);
7230 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7231 struct extent_state **cached_state,
7232 unsigned int iomap_flags)
7234 const bool writing = (iomap_flags & IOMAP_WRITE);
7235 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7236 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7237 struct btrfs_ordered_extent *ordered;
7242 if (!try_lock_extent(io_tree, lockstart, lockend,
7246 lock_extent(io_tree, lockstart, lockend, cached_state);
7249 * We're concerned with the entire range that we're going to be
7250 * doing DIO to, so we need to make sure there's no ordered
7251 * extents in this range.
7253 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7254 lockend - lockstart + 1);
7257 * We need to make sure there are no buffered pages in this
7258 * range either, we could have raced between the invalidate in
7259 * generic_file_direct_write and locking the extent. The
7260 * invalidate needs to happen so that reads after a write do not
7264 (!writing || !filemap_range_has_page(inode->i_mapping,
7265 lockstart, lockend)))
7268 unlock_extent(io_tree, lockstart, lockend, cached_state);
7272 btrfs_put_ordered_extent(ordered);
7277 * If we are doing a DIO read and the ordered extent we
7278 * found is for a buffered write, we can not wait for it
7279 * to complete and retry, because if we do so we can
7280 * deadlock with concurrent buffered writes on page
7281 * locks. This happens only if our DIO read covers more
7282 * than one extent map, if at this point has already
7283 * created an ordered extent for a previous extent map
7284 * and locked its range in the inode's io tree, and a
7285 * concurrent write against that previous extent map's
7286 * range and this range started (we unlock the ranges
7287 * in the io tree only when the bios complete and
7288 * buffered writes always lock pages before attempting
7289 * to lock range in the io tree).
7292 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7293 btrfs_start_ordered_extent(ordered);
7295 ret = nowait ? -EAGAIN : -ENOTBLK;
7296 btrfs_put_ordered_extent(ordered);
7299 * We could trigger writeback for this range (and wait
7300 * for it to complete) and then invalidate the pages for
7301 * this range (through invalidate_inode_pages2_range()),
7302 * but that can lead us to a deadlock with a concurrent
7303 * call to readahead (a buffered read or a defrag call
7304 * triggered a readahead) on a page lock due to an
7305 * ordered dio extent we created before but did not have
7306 * yet a corresponding bio submitted (whence it can not
7307 * complete), which makes readahead wait for that
7308 * ordered extent to complete while holding a lock on
7311 ret = nowait ? -EAGAIN : -ENOTBLK;
7323 /* The callers of this must take lock_extent() */
7324 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7325 u64 len, u64 orig_start, u64 block_start,
7326 u64 block_len, u64 orig_block_len,
7327 u64 ram_bytes, int compress_type,
7330 struct extent_map *em;
7333 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7334 type == BTRFS_ORDERED_COMPRESSED ||
7335 type == BTRFS_ORDERED_NOCOW ||
7336 type == BTRFS_ORDERED_REGULAR);
7338 em = alloc_extent_map();
7340 return ERR_PTR(-ENOMEM);
7343 em->orig_start = orig_start;
7345 em->block_len = block_len;
7346 em->block_start = block_start;
7347 em->orig_block_len = orig_block_len;
7348 em->ram_bytes = ram_bytes;
7349 em->generation = -1;
7350 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7351 if (type == BTRFS_ORDERED_PREALLOC) {
7352 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7353 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7354 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7355 em->compress_type = compress_type;
7358 ret = btrfs_replace_extent_map_range(inode, em, true);
7360 free_extent_map(em);
7361 return ERR_PTR(ret);
7364 /* em got 2 refs now, callers needs to do free_extent_map once. */
7369 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7370 struct inode *inode,
7371 struct btrfs_dio_data *dio_data,
7372 u64 start, u64 *lenp,
7373 unsigned int iomap_flags)
7375 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7376 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7377 struct extent_map *em = *map;
7379 u64 block_start, orig_start, orig_block_len, ram_bytes;
7380 struct btrfs_block_group *bg;
7381 bool can_nocow = false;
7382 bool space_reserved = false;
7388 * We don't allocate a new extent in the following cases
7390 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7392 * 2) The extent is marked as PREALLOC. We're good to go here and can
7393 * just use the extent.
7396 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7397 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7398 em->block_start != EXTENT_MAP_HOLE)) {
7399 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7400 type = BTRFS_ORDERED_PREALLOC;
7402 type = BTRFS_ORDERED_NOCOW;
7403 len = min(len, em->len - (start - em->start));
7404 block_start = em->block_start + (start - em->start);
7406 if (can_nocow_extent(inode, start, &len, &orig_start,
7407 &orig_block_len, &ram_bytes, false, false) == 1) {
7408 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7416 struct extent_map *em2;
7418 /* We can NOCOW, so only need to reserve metadata space. */
7419 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7422 /* Our caller expects us to free the input extent map. */
7423 free_extent_map(em);
7425 btrfs_dec_nocow_writers(bg);
7426 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7430 space_reserved = true;
7432 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7433 orig_start, block_start,
7434 len, orig_block_len,
7436 btrfs_dec_nocow_writers(bg);
7437 if (type == BTRFS_ORDERED_PREALLOC) {
7438 free_extent_map(em);
7448 dio_data->nocow_done = true;
7450 /* Our caller expects us to free the input extent map. */
7451 free_extent_map(em);
7460 * If we could not allocate data space before locking the file
7461 * range and we can't do a NOCOW write, then we have to fail.
7463 if (!dio_data->data_space_reserved) {
7469 * We have to COW and we have already reserved data space before,
7470 * so now we reserve only metadata.
7472 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7476 space_reserved = true;
7478 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7484 len = min(len, em->len - (start - em->start));
7486 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7487 prev_len - len, true);
7491 * We have created our ordered extent, so we can now release our reservation
7492 * for an outstanding extent.
7494 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7497 * Need to update the i_size under the extent lock so buffered
7498 * readers will get the updated i_size when we unlock.
7500 if (start + len > i_size_read(inode))
7501 i_size_write(inode, start + len);
7503 if (ret && space_reserved) {
7504 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7505 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7511 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7512 loff_t length, unsigned int flags, struct iomap *iomap,
7513 struct iomap *srcmap)
7515 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7516 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7517 struct extent_map *em;
7518 struct extent_state *cached_state = NULL;
7519 struct btrfs_dio_data *dio_data = iter->private;
7520 u64 lockstart, lockend;
7521 const bool write = !!(flags & IOMAP_WRITE);
7524 const u64 data_alloc_len = length;
7525 bool unlock_extents = false;
7528 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7529 * we're NOWAIT we may submit a bio for a partial range and return
7530 * EIOCBQUEUED, which would result in an errant short read.
7532 * The best way to handle this would be to allow for partial completions
7533 * of iocb's, so we could submit the partial bio, return and fault in
7534 * the rest of the pages, and then submit the io for the rest of the
7535 * range. However we don't have that currently, so simply return
7536 * -EAGAIN at this point so that the normal path is used.
7538 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7542 * Cap the size of reads to that usually seen in buffered I/O as we need
7543 * to allocate a contiguous array for the checksums.
7546 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7549 lockend = start + len - 1;
7552 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7553 * enough if we've written compressed pages to this area, so we need to
7554 * flush the dirty pages again to make absolutely sure that any
7555 * outstanding dirty pages are on disk - the first flush only starts
7556 * compression on the data, while keeping the pages locked, so by the
7557 * time the second flush returns we know bios for the compressed pages
7558 * were submitted and finished, and the pages no longer under writeback.
7560 * If we have a NOWAIT request and we have any pages in the range that
7561 * are locked, likely due to compression still in progress, we don't want
7562 * to block on page locks. We also don't want to block on pages marked as
7563 * dirty or under writeback (same as for the non-compression case).
7564 * iomap_dio_rw() did the same check, but after that and before we got
7565 * here, mmap'ed writes may have happened or buffered reads started
7566 * (readpage() and readahead(), which lock pages), as we haven't locked
7567 * the file range yet.
7569 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7570 &BTRFS_I(inode)->runtime_flags)) {
7571 if (flags & IOMAP_NOWAIT) {
7572 if (filemap_range_needs_writeback(inode->i_mapping,
7573 lockstart, lockend))
7576 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7577 start + length - 1);
7583 memset(dio_data, 0, sizeof(*dio_data));
7586 * We always try to allocate data space and must do it before locking
7587 * the file range, to avoid deadlocks with concurrent writes to the same
7588 * range if the range has several extents and the writes don't expand the
7589 * current i_size (the inode lock is taken in shared mode). If we fail to
7590 * allocate data space here we continue and later, after locking the
7591 * file range, we fail with ENOSPC only if we figure out we can not do a
7594 if (write && !(flags & IOMAP_NOWAIT)) {
7595 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7596 &dio_data->data_reserved,
7597 start, data_alloc_len, false);
7599 dio_data->data_space_reserved = true;
7600 else if (ret && !(BTRFS_I(inode)->flags &
7601 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7606 * If this errors out it's because we couldn't invalidate pagecache for
7607 * this range and we need to fallback to buffered IO, or we are doing a
7608 * NOWAIT read/write and we need to block.
7610 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7614 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7621 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7622 * io. INLINE is special, and we could probably kludge it in here, but
7623 * it's still buffered so for safety lets just fall back to the generic
7626 * For COMPRESSED we _have_ to read the entire extent in so we can
7627 * decompress it, so there will be buffering required no matter what we
7628 * do, so go ahead and fallback to buffered.
7630 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7631 * to buffered IO. Don't blame me, this is the price we pay for using
7634 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7635 em->block_start == EXTENT_MAP_INLINE) {
7636 free_extent_map(em);
7638 * If we are in a NOWAIT context, return -EAGAIN in order to
7639 * fallback to buffered IO. This is not only because we can
7640 * block with buffered IO (no support for NOWAIT semantics at
7641 * the moment) but also to avoid returning short reads to user
7642 * space - this happens if we were able to read some data from
7643 * previous non-compressed extents and then when we fallback to
7644 * buffered IO, at btrfs_file_read_iter() by calling
7645 * filemap_read(), we fail to fault in pages for the read buffer,
7646 * in which case filemap_read() returns a short read (the number
7647 * of bytes previously read is > 0, so it does not return -EFAULT).
7649 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7653 len = min(len, em->len - (start - em->start));
7656 * If we have a NOWAIT request and the range contains multiple extents
7657 * (or a mix of extents and holes), then we return -EAGAIN to make the
7658 * caller fallback to a context where it can do a blocking (without
7659 * NOWAIT) request. This way we avoid doing partial IO and returning
7660 * success to the caller, which is not optimal for writes and for reads
7661 * it can result in unexpected behaviour for an application.
7663 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7664 * iomap_dio_rw(), we can end up returning less data then what the caller
7665 * asked for, resulting in an unexpected, and incorrect, short read.
7666 * That is, the caller asked to read N bytes and we return less than that,
7667 * which is wrong unless we are crossing EOF. This happens if we get a
7668 * page fault error when trying to fault in pages for the buffer that is
7669 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7670 * have previously submitted bios for other extents in the range, in
7671 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7672 * those bios have completed by the time we get the page fault error,
7673 * which we return back to our caller - we should only return EIOCBQUEUED
7674 * after we have submitted bios for all the extents in the range.
7676 if ((flags & IOMAP_NOWAIT) && len < length) {
7677 free_extent_map(em);
7683 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7684 start, &len, flags);
7687 unlock_extents = true;
7688 /* Recalc len in case the new em is smaller than requested */
7689 len = min(len, em->len - (start - em->start));
7690 if (dio_data->data_space_reserved) {
7692 u64 release_len = 0;
7694 if (dio_data->nocow_done) {
7695 release_offset = start;
7696 release_len = data_alloc_len;
7697 } else if (len < data_alloc_len) {
7698 release_offset = start + len;
7699 release_len = data_alloc_len - len;
7702 if (release_len > 0)
7703 btrfs_free_reserved_data_space(BTRFS_I(inode),
7704 dio_data->data_reserved,
7710 * We need to unlock only the end area that we aren't using.
7711 * The rest is going to be unlocked by the endio routine.
7713 lockstart = start + len;
7714 if (lockstart < lockend)
7715 unlock_extents = true;
7719 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7722 free_extent_state(cached_state);
7725 * Translate extent map information to iomap.
7726 * We trim the extents (and move the addr) even though iomap code does
7727 * that, since we have locked only the parts we are performing I/O in.
7729 if ((em->block_start == EXTENT_MAP_HOLE) ||
7730 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7731 iomap->addr = IOMAP_NULL_ADDR;
7732 iomap->type = IOMAP_HOLE;
7734 iomap->addr = em->block_start + (start - em->start);
7735 iomap->type = IOMAP_MAPPED;
7737 iomap->offset = start;
7738 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7739 iomap->length = len;
7740 free_extent_map(em);
7745 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7748 if (dio_data->data_space_reserved) {
7749 btrfs_free_reserved_data_space(BTRFS_I(inode),
7750 dio_data->data_reserved,
7751 start, data_alloc_len);
7752 extent_changeset_free(dio_data->data_reserved);
7758 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7759 ssize_t written, unsigned int flags, struct iomap *iomap)
7761 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7762 struct btrfs_dio_data *dio_data = iter->private;
7763 size_t submitted = dio_data->submitted;
7764 const bool write = !!(flags & IOMAP_WRITE);
7767 if (!write && (iomap->type == IOMAP_HOLE)) {
7768 /* If reading from a hole, unlock and return */
7769 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7774 if (submitted < length) {
7776 length -= submitted;
7778 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7779 pos, length, false);
7781 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7782 pos + length - 1, NULL);
7786 btrfs_put_ordered_extent(dio_data->ordered);
7787 dio_data->ordered = NULL;
7791 extent_changeset_free(dio_data->data_reserved);
7795 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7797 struct btrfs_dio_private *dip =
7798 container_of(bbio, struct btrfs_dio_private, bbio);
7799 struct btrfs_inode *inode = bbio->inode;
7800 struct bio *bio = &bbio->bio;
7802 if (bio->bi_status) {
7803 btrfs_warn(inode->root->fs_info,
7804 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7805 btrfs_ino(inode), bio->bi_opf,
7806 dip->file_offset, dip->bytes, bio->bi_status);
7809 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7810 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7811 dip->file_offset, dip->bytes,
7814 unlock_extent(&inode->io_tree, dip->file_offset,
7815 dip->file_offset + dip->bytes - 1, NULL);
7818 bbio->bio.bi_private = bbio->private;
7819 iomap_dio_bio_end_io(bio);
7822 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7825 struct btrfs_bio *bbio = btrfs_bio(bio);
7826 struct btrfs_dio_private *dip =
7827 container_of(bbio, struct btrfs_dio_private, bbio);
7828 struct btrfs_dio_data *dio_data = iter->private;
7830 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7831 btrfs_dio_end_io, bio->bi_private);
7832 bbio->inode = BTRFS_I(iter->inode);
7833 bbio->file_offset = file_offset;
7835 dip->file_offset = file_offset;
7836 dip->bytes = bio->bi_iter.bi_size;
7838 dio_data->submitted += bio->bi_iter.bi_size;
7841 * Check if we are doing a partial write. If we are, we need to split
7842 * the ordered extent to match the submitted bio. Hang on to the
7843 * remaining unfinishable ordered_extent in dio_data so that it can be
7844 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7845 * remaining pages is blocked on the outstanding ordered extent.
7847 if (iter->flags & IOMAP_WRITE) {
7850 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7852 bbio->bio.bi_status = errno_to_blk_status(ret);
7853 btrfs_dio_end_io(bbio);
7858 btrfs_submit_bio(bbio, 0);
7861 static const struct iomap_ops btrfs_dio_iomap_ops = {
7862 .iomap_begin = btrfs_dio_iomap_begin,
7863 .iomap_end = btrfs_dio_iomap_end,
7866 static const struct iomap_dio_ops btrfs_dio_ops = {
7867 .submit_io = btrfs_dio_submit_io,
7868 .bio_set = &btrfs_dio_bioset,
7871 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7873 struct btrfs_dio_data data = { 0 };
7875 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7876 IOMAP_DIO_PARTIAL, &data, done_before);
7879 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7882 struct btrfs_dio_data data = { 0 };
7884 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7885 IOMAP_DIO_PARTIAL, &data, done_before);
7888 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7893 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7898 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7899 * file range (0 to LLONG_MAX), but that is not enough if we have
7900 * compression enabled. The first filemap_fdatawrite_range() only kicks
7901 * in the compression of data (in an async thread) and will return
7902 * before the compression is done and writeback is started. A second
7903 * filemap_fdatawrite_range() is needed to wait for the compression to
7904 * complete and writeback to start. We also need to wait for ordered
7905 * extents to complete, because our fiemap implementation uses mainly
7906 * file extent items to list the extents, searching for extent maps
7907 * only for file ranges with holes or prealloc extents to figure out
7908 * if we have delalloc in those ranges.
7910 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7911 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7916 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7919 static int btrfs_writepages(struct address_space *mapping,
7920 struct writeback_control *wbc)
7922 return extent_writepages(mapping, wbc);
7925 static void btrfs_readahead(struct readahead_control *rac)
7927 extent_readahead(rac);
7931 * For release_folio() and invalidate_folio() we have a race window where
7932 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7933 * If we continue to release/invalidate the page, we could cause use-after-free
7934 * for subpage spinlock. So this function is to spin and wait for subpage
7937 static void wait_subpage_spinlock(struct page *page)
7939 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7940 struct btrfs_subpage *subpage;
7942 if (!btrfs_is_subpage(fs_info, page))
7945 ASSERT(PagePrivate(page) && page->private);
7946 subpage = (struct btrfs_subpage *)page->private;
7949 * This may look insane as we just acquire the spinlock and release it,
7950 * without doing anything. But we just want to make sure no one is
7951 * still holding the subpage spinlock.
7952 * And since the page is not dirty nor writeback, and we have page
7953 * locked, the only possible way to hold a spinlock is from the endio
7954 * function to clear page writeback.
7956 * Here we just acquire the spinlock so that all existing callers
7957 * should exit and we're safe to release/invalidate the page.
7959 spin_lock_irq(&subpage->lock);
7960 spin_unlock_irq(&subpage->lock);
7963 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7965 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7968 wait_subpage_spinlock(&folio->page);
7969 clear_page_extent_mapped(&folio->page);
7974 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7976 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7978 return __btrfs_release_folio(folio, gfp_flags);
7981 #ifdef CONFIG_MIGRATION
7982 static int btrfs_migrate_folio(struct address_space *mapping,
7983 struct folio *dst, struct folio *src,
7984 enum migrate_mode mode)
7986 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7988 if (ret != MIGRATEPAGE_SUCCESS)
7991 if (folio_test_ordered(src)) {
7992 folio_clear_ordered(src);
7993 folio_set_ordered(dst);
7996 return MIGRATEPAGE_SUCCESS;
7999 #define btrfs_migrate_folio NULL
8002 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8005 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8006 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8007 struct extent_io_tree *tree = &inode->io_tree;
8008 struct extent_state *cached_state = NULL;
8009 u64 page_start = folio_pos(folio);
8010 u64 page_end = page_start + folio_size(folio) - 1;
8012 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8015 * We have folio locked so no new ordered extent can be created on this
8016 * page, nor bio can be submitted for this folio.
8018 * But already submitted bio can still be finished on this folio.
8019 * Furthermore, endio function won't skip folio which has Ordered
8020 * (Private2) already cleared, so it's possible for endio and
8021 * invalidate_folio to do the same ordered extent accounting twice
8024 * So here we wait for any submitted bios to finish, so that we won't
8025 * do double ordered extent accounting on the same folio.
8027 folio_wait_writeback(folio);
8028 wait_subpage_spinlock(&folio->page);
8031 * For subpage case, we have call sites like
8032 * btrfs_punch_hole_lock_range() which passes range not aligned to
8034 * If the range doesn't cover the full folio, we don't need to and
8035 * shouldn't clear page extent mapped, as folio->private can still
8036 * record subpage dirty bits for other part of the range.
8038 * For cases that invalidate the full folio even the range doesn't
8039 * cover the full folio, like invalidating the last folio, we're
8040 * still safe to wait for ordered extent to finish.
8042 if (!(offset == 0 && length == folio_size(folio))) {
8043 btrfs_release_folio(folio, GFP_NOFS);
8047 if (!inode_evicting)
8048 lock_extent(tree, page_start, page_end, &cached_state);
8051 while (cur < page_end) {
8052 struct btrfs_ordered_extent *ordered;
8055 u32 extra_flags = 0;
8057 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8058 page_end + 1 - cur);
8060 range_end = page_end;
8062 * No ordered extent covering this range, we are safe
8063 * to delete all extent states in the range.
8065 extra_flags = EXTENT_CLEAR_ALL_BITS;
8068 if (ordered->file_offset > cur) {
8070 * There is a range between [cur, oe->file_offset) not
8071 * covered by any ordered extent.
8072 * We are safe to delete all extent states, and handle
8073 * the ordered extent in the next iteration.
8075 range_end = ordered->file_offset - 1;
8076 extra_flags = EXTENT_CLEAR_ALL_BITS;
8080 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8082 ASSERT(range_end + 1 - cur < U32_MAX);
8083 range_len = range_end + 1 - cur;
8084 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8086 * If Ordered (Private2) is cleared, it means endio has
8087 * already been executed for the range.
8088 * We can't delete the extent states as
8089 * btrfs_finish_ordered_io() may still use some of them.
8093 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8096 * IO on this page will never be started, so we need to account
8097 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8098 * here, must leave that up for the ordered extent completion.
8100 * This will also unlock the range for incoming
8101 * btrfs_finish_ordered_io().
8103 if (!inode_evicting)
8104 clear_extent_bit(tree, cur, range_end,
8106 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8107 EXTENT_DEFRAG, &cached_state);
8109 spin_lock_irq(&inode->ordered_tree.lock);
8110 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8111 ordered->truncated_len = min(ordered->truncated_len,
8112 cur - ordered->file_offset);
8113 spin_unlock_irq(&inode->ordered_tree.lock);
8116 * If the ordered extent has finished, we're safe to delete all
8117 * the extent states of the range, otherwise
8118 * btrfs_finish_ordered_io() will get executed by endio for
8119 * other pages, so we can't delete extent states.
8121 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8122 cur, range_end + 1 - cur)) {
8123 btrfs_finish_ordered_io(ordered);
8125 * The ordered extent has finished, now we're again
8126 * safe to delete all extent states of the range.
8128 extra_flags = EXTENT_CLEAR_ALL_BITS;
8132 btrfs_put_ordered_extent(ordered);
8134 * Qgroup reserved space handler
8135 * Sector(s) here will be either:
8137 * 1) Already written to disk or bio already finished
8138 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8139 * Qgroup will be handled by its qgroup_record then.
8140 * btrfs_qgroup_free_data() call will do nothing here.
8142 * 2) Not written to disk yet
8143 * Then btrfs_qgroup_free_data() call will clear the
8144 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8145 * reserved data space.
8146 * Since the IO will never happen for this page.
8148 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8149 if (!inode_evicting) {
8150 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8151 EXTENT_DELALLOC | EXTENT_UPTODATE |
8152 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8153 extra_flags, &cached_state);
8155 cur = range_end + 1;
8158 * We have iterated through all ordered extents of the page, the page
8159 * should not have Ordered (Private2) anymore, or the above iteration
8160 * did something wrong.
8162 ASSERT(!folio_test_ordered(folio));
8163 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8164 if (!inode_evicting)
8165 __btrfs_release_folio(folio, GFP_NOFS);
8166 clear_page_extent_mapped(&folio->page);
8170 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8171 * called from a page fault handler when a page is first dirtied. Hence we must
8172 * be careful to check for EOF conditions here. We set the page up correctly
8173 * for a written page which means we get ENOSPC checking when writing into
8174 * holes and correct delalloc and unwritten extent mapping on filesystems that
8175 * support these features.
8177 * We are not allowed to take the i_mutex here so we have to play games to
8178 * protect against truncate races as the page could now be beyond EOF. Because
8179 * truncate_setsize() writes the inode size before removing pages, once we have
8180 * the page lock we can determine safely if the page is beyond EOF. If it is not
8181 * beyond EOF, then the page is guaranteed safe against truncation until we
8184 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8186 struct page *page = vmf->page;
8187 struct inode *inode = file_inode(vmf->vma->vm_file);
8188 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8189 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8190 struct btrfs_ordered_extent *ordered;
8191 struct extent_state *cached_state = NULL;
8192 struct extent_changeset *data_reserved = NULL;
8193 unsigned long zero_start;
8203 reserved_space = PAGE_SIZE;
8205 sb_start_pagefault(inode->i_sb);
8206 page_start = page_offset(page);
8207 page_end = page_start + PAGE_SIZE - 1;
8211 * Reserving delalloc space after obtaining the page lock can lead to
8212 * deadlock. For example, if a dirty page is locked by this function
8213 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8214 * dirty page write out, then the btrfs_writepages() function could
8215 * end up waiting indefinitely to get a lock on the page currently
8216 * being processed by btrfs_page_mkwrite() function.
8218 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8219 page_start, reserved_space);
8221 ret2 = file_update_time(vmf->vma->vm_file);
8225 ret = vmf_error(ret2);
8231 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8233 down_read(&BTRFS_I(inode)->i_mmap_lock);
8235 size = i_size_read(inode);
8237 if ((page->mapping != inode->i_mapping) ||
8238 (page_start >= size)) {
8239 /* page got truncated out from underneath us */
8242 wait_on_page_writeback(page);
8244 lock_extent(io_tree, page_start, page_end, &cached_state);
8245 ret2 = set_page_extent_mapped(page);
8247 ret = vmf_error(ret2);
8248 unlock_extent(io_tree, page_start, page_end, &cached_state);
8253 * we can't set the delalloc bits if there are pending ordered
8254 * extents. Drop our locks and wait for them to finish
8256 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8259 unlock_extent(io_tree, page_start, page_end, &cached_state);
8261 up_read(&BTRFS_I(inode)->i_mmap_lock);
8262 btrfs_start_ordered_extent(ordered);
8263 btrfs_put_ordered_extent(ordered);
8267 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8268 reserved_space = round_up(size - page_start,
8269 fs_info->sectorsize);
8270 if (reserved_space < PAGE_SIZE) {
8271 end = page_start + reserved_space - 1;
8272 btrfs_delalloc_release_space(BTRFS_I(inode),
8273 data_reserved, page_start,
8274 PAGE_SIZE - reserved_space, true);
8279 * page_mkwrite gets called when the page is firstly dirtied after it's
8280 * faulted in, but write(2) could also dirty a page and set delalloc
8281 * bits, thus in this case for space account reason, we still need to
8282 * clear any delalloc bits within this page range since we have to
8283 * reserve data&meta space before lock_page() (see above comments).
8285 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8286 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8287 EXTENT_DEFRAG, &cached_state);
8289 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8292 unlock_extent(io_tree, page_start, page_end, &cached_state);
8293 ret = VM_FAULT_SIGBUS;
8297 /* page is wholly or partially inside EOF */
8298 if (page_start + PAGE_SIZE > size)
8299 zero_start = offset_in_page(size);
8301 zero_start = PAGE_SIZE;
8303 if (zero_start != PAGE_SIZE)
8304 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8306 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8307 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8308 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8310 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8312 unlock_extent(io_tree, page_start, page_end, &cached_state);
8313 up_read(&BTRFS_I(inode)->i_mmap_lock);
8315 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8316 sb_end_pagefault(inode->i_sb);
8317 extent_changeset_free(data_reserved);
8318 return VM_FAULT_LOCKED;
8322 up_read(&BTRFS_I(inode)->i_mmap_lock);
8324 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8325 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8326 reserved_space, (ret != 0));
8328 sb_end_pagefault(inode->i_sb);
8329 extent_changeset_free(data_reserved);
8333 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8335 struct btrfs_truncate_control control = {
8337 .ino = btrfs_ino(inode),
8338 .min_type = BTRFS_EXTENT_DATA_KEY,
8339 .clear_extent_range = true,
8341 struct btrfs_root *root = inode->root;
8342 struct btrfs_fs_info *fs_info = root->fs_info;
8343 struct btrfs_block_rsv *rsv;
8345 struct btrfs_trans_handle *trans;
8346 u64 mask = fs_info->sectorsize - 1;
8347 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8349 if (!skip_writeback) {
8350 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8351 inode->vfs_inode.i_size & (~mask),
8358 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8359 * things going on here:
8361 * 1) We need to reserve space to update our inode.
8363 * 2) We need to have something to cache all the space that is going to
8364 * be free'd up by the truncate operation, but also have some slack
8365 * space reserved in case it uses space during the truncate (thank you
8366 * very much snapshotting).
8368 * And we need these to be separate. The fact is we can use a lot of
8369 * space doing the truncate, and we have no earthly idea how much space
8370 * we will use, so we need the truncate reservation to be separate so it
8371 * doesn't end up using space reserved for updating the inode. We also
8372 * need to be able to stop the transaction and start a new one, which
8373 * means we need to be able to update the inode several times, and we
8374 * have no idea of knowing how many times that will be, so we can't just
8375 * reserve 1 item for the entirety of the operation, so that has to be
8376 * done separately as well.
8378 * So that leaves us with
8380 * 1) rsv - for the truncate reservation, which we will steal from the
8381 * transaction reservation.
8382 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8383 * updating the inode.
8385 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8388 rsv->size = min_size;
8389 rsv->failfast = true;
8392 * 1 for the truncate slack space
8393 * 1 for updating the inode.
8395 trans = btrfs_start_transaction(root, 2);
8396 if (IS_ERR(trans)) {
8397 ret = PTR_ERR(trans);
8401 /* Migrate the slack space for the truncate to our reserve */
8402 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8405 * We have reserved 2 metadata units when we started the transaction and
8406 * min_size matches 1 unit, so this should never fail, but if it does,
8407 * it's not critical we just fail truncation.
8410 btrfs_end_transaction(trans);
8414 trans->block_rsv = rsv;
8417 struct extent_state *cached_state = NULL;
8418 const u64 new_size = inode->vfs_inode.i_size;
8419 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8421 control.new_size = new_size;
8422 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8424 * We want to drop from the next block forward in case this new
8425 * size is not block aligned since we will be keeping the last
8426 * block of the extent just the way it is.
8428 btrfs_drop_extent_map_range(inode,
8429 ALIGN(new_size, fs_info->sectorsize),
8432 ret = btrfs_truncate_inode_items(trans, root, &control);
8434 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8435 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8437 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8439 trans->block_rsv = &fs_info->trans_block_rsv;
8440 if (ret != -ENOSPC && ret != -EAGAIN)
8443 ret = btrfs_update_inode(trans, root, inode);
8447 btrfs_end_transaction(trans);
8448 btrfs_btree_balance_dirty(fs_info);
8450 trans = btrfs_start_transaction(root, 2);
8451 if (IS_ERR(trans)) {
8452 ret = PTR_ERR(trans);
8457 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8458 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8459 rsv, min_size, false);
8461 * We have reserved 2 metadata units when we started the
8462 * transaction and min_size matches 1 unit, so this should never
8463 * fail, but if it does, it's not critical we just fail truncation.
8468 trans->block_rsv = rsv;
8472 * We can't call btrfs_truncate_block inside a trans handle as we could
8473 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8474 * know we've truncated everything except the last little bit, and can
8475 * do btrfs_truncate_block and then update the disk_i_size.
8477 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8478 btrfs_end_transaction(trans);
8479 btrfs_btree_balance_dirty(fs_info);
8481 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8484 trans = btrfs_start_transaction(root, 1);
8485 if (IS_ERR(trans)) {
8486 ret = PTR_ERR(trans);
8489 btrfs_inode_safe_disk_i_size_write(inode, 0);
8495 trans->block_rsv = &fs_info->trans_block_rsv;
8496 ret2 = btrfs_update_inode(trans, root, inode);
8500 ret2 = btrfs_end_transaction(trans);
8503 btrfs_btree_balance_dirty(fs_info);
8506 btrfs_free_block_rsv(fs_info, rsv);
8508 * So if we truncate and then write and fsync we normally would just
8509 * write the extents that changed, which is a problem if we need to
8510 * first truncate that entire inode. So set this flag so we write out
8511 * all of the extents in the inode to the sync log so we're completely
8514 * If no extents were dropped or trimmed we don't need to force the next
8515 * fsync to truncate all the inode's items from the log and re-log them
8516 * all. This means the truncate operation did not change the file size,
8517 * or changed it to a smaller size but there was only an implicit hole
8518 * between the old i_size and the new i_size, and there were no prealloc
8519 * extents beyond i_size to drop.
8521 if (control.extents_found > 0)
8522 btrfs_set_inode_full_sync(inode);
8527 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8530 struct inode *inode;
8532 inode = new_inode(dir->i_sb);
8535 * Subvolumes don't inherit the sgid bit or the parent's gid if
8536 * the parent's sgid bit is set. This is probably a bug.
8538 inode_init_owner(idmap, inode, NULL,
8539 S_IFDIR | (~current_umask() & S_IRWXUGO));
8540 inode->i_op = &btrfs_dir_inode_operations;
8541 inode->i_fop = &btrfs_dir_file_operations;
8546 struct inode *btrfs_alloc_inode(struct super_block *sb)
8548 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8549 struct btrfs_inode *ei;
8550 struct inode *inode;
8552 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8559 ei->last_sub_trans = 0;
8560 ei->logged_trans = 0;
8561 ei->delalloc_bytes = 0;
8562 ei->new_delalloc_bytes = 0;
8563 ei->defrag_bytes = 0;
8564 ei->disk_i_size = 0;
8568 ei->index_cnt = (u64)-1;
8570 ei->last_unlink_trans = 0;
8571 ei->last_reflink_trans = 0;
8572 ei->last_log_commit = 0;
8574 spin_lock_init(&ei->lock);
8575 ei->outstanding_extents = 0;
8576 if (sb->s_magic != BTRFS_TEST_MAGIC)
8577 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8578 BTRFS_BLOCK_RSV_DELALLOC);
8579 ei->runtime_flags = 0;
8580 ei->prop_compress = BTRFS_COMPRESS_NONE;
8581 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8583 ei->delayed_node = NULL;
8585 ei->i_otime.tv_sec = 0;
8586 ei->i_otime.tv_nsec = 0;
8588 inode = &ei->vfs_inode;
8589 extent_map_tree_init(&ei->extent_tree);
8590 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8591 ei->io_tree.inode = ei;
8592 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8593 IO_TREE_INODE_FILE_EXTENT);
8594 mutex_init(&ei->log_mutex);
8595 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8596 INIT_LIST_HEAD(&ei->delalloc_inodes);
8597 INIT_LIST_HEAD(&ei->delayed_iput);
8598 RB_CLEAR_NODE(&ei->rb_node);
8599 init_rwsem(&ei->i_mmap_lock);
8604 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8605 void btrfs_test_destroy_inode(struct inode *inode)
8607 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8608 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8612 void btrfs_free_inode(struct inode *inode)
8614 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8617 void btrfs_destroy_inode(struct inode *vfs_inode)
8619 struct btrfs_ordered_extent *ordered;
8620 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8621 struct btrfs_root *root = inode->root;
8622 bool freespace_inode;
8624 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8625 WARN_ON(vfs_inode->i_data.nrpages);
8626 WARN_ON(inode->block_rsv.reserved);
8627 WARN_ON(inode->block_rsv.size);
8628 WARN_ON(inode->outstanding_extents);
8629 if (!S_ISDIR(vfs_inode->i_mode)) {
8630 WARN_ON(inode->delalloc_bytes);
8631 WARN_ON(inode->new_delalloc_bytes);
8633 WARN_ON(inode->csum_bytes);
8634 WARN_ON(inode->defrag_bytes);
8637 * This can happen where we create an inode, but somebody else also
8638 * created the same inode and we need to destroy the one we already
8645 * If this is a free space inode do not take the ordered extents lockdep
8648 freespace_inode = btrfs_is_free_space_inode(inode);
8651 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8655 btrfs_err(root->fs_info,
8656 "found ordered extent %llu %llu on inode cleanup",
8657 ordered->file_offset, ordered->num_bytes);
8659 if (!freespace_inode)
8660 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8662 btrfs_remove_ordered_extent(inode, ordered);
8663 btrfs_put_ordered_extent(ordered);
8664 btrfs_put_ordered_extent(ordered);
8667 btrfs_qgroup_check_reserved_leak(inode);
8668 inode_tree_del(inode);
8669 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8670 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8671 btrfs_put_root(inode->root);
8674 int btrfs_drop_inode(struct inode *inode)
8676 struct btrfs_root *root = BTRFS_I(inode)->root;
8681 /* the snap/subvol tree is on deleting */
8682 if (btrfs_root_refs(&root->root_item) == 0)
8685 return generic_drop_inode(inode);
8688 static void init_once(void *foo)
8690 struct btrfs_inode *ei = foo;
8692 inode_init_once(&ei->vfs_inode);
8695 void __cold btrfs_destroy_cachep(void)
8698 * Make sure all delayed rcu free inodes are flushed before we
8702 bioset_exit(&btrfs_dio_bioset);
8703 kmem_cache_destroy(btrfs_inode_cachep);
8706 int __init btrfs_init_cachep(void)
8708 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8709 sizeof(struct btrfs_inode), 0,
8710 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8712 if (!btrfs_inode_cachep)
8715 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8716 offsetof(struct btrfs_dio_private, bbio.bio),
8722 btrfs_destroy_cachep();
8726 static int btrfs_getattr(struct mnt_idmap *idmap,
8727 const struct path *path, struct kstat *stat,
8728 u32 request_mask, unsigned int flags)
8732 struct inode *inode = d_inode(path->dentry);
8733 u32 blocksize = inode->i_sb->s_blocksize;
8734 u32 bi_flags = BTRFS_I(inode)->flags;
8735 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8737 stat->result_mask |= STATX_BTIME;
8738 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8739 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8740 if (bi_flags & BTRFS_INODE_APPEND)
8741 stat->attributes |= STATX_ATTR_APPEND;
8742 if (bi_flags & BTRFS_INODE_COMPRESS)
8743 stat->attributes |= STATX_ATTR_COMPRESSED;
8744 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8745 stat->attributes |= STATX_ATTR_IMMUTABLE;
8746 if (bi_flags & BTRFS_INODE_NODUMP)
8747 stat->attributes |= STATX_ATTR_NODUMP;
8748 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8749 stat->attributes |= STATX_ATTR_VERITY;
8751 stat->attributes_mask |= (STATX_ATTR_APPEND |
8752 STATX_ATTR_COMPRESSED |
8753 STATX_ATTR_IMMUTABLE |
8756 generic_fillattr(idmap, inode, stat);
8757 stat->dev = BTRFS_I(inode)->root->anon_dev;
8759 spin_lock(&BTRFS_I(inode)->lock);
8760 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8761 inode_bytes = inode_get_bytes(inode);
8762 spin_unlock(&BTRFS_I(inode)->lock);
8763 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8764 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8768 static int btrfs_rename_exchange(struct inode *old_dir,
8769 struct dentry *old_dentry,
8770 struct inode *new_dir,
8771 struct dentry *new_dentry)
8773 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8774 struct btrfs_trans_handle *trans;
8775 unsigned int trans_num_items;
8776 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8777 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8778 struct inode *new_inode = new_dentry->d_inode;
8779 struct inode *old_inode = old_dentry->d_inode;
8780 struct timespec64 ctime = current_time(old_inode);
8781 struct btrfs_rename_ctx old_rename_ctx;
8782 struct btrfs_rename_ctx new_rename_ctx;
8783 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8784 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8789 bool need_abort = false;
8790 struct fscrypt_name old_fname, new_fname;
8791 struct fscrypt_str *old_name, *new_name;
8794 * For non-subvolumes allow exchange only within one subvolume, in the
8795 * same inode namespace. Two subvolumes (represented as directory) can
8796 * be exchanged as they're a logical link and have a fixed inode number.
8799 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8800 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8803 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8807 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8809 fscrypt_free_filename(&old_fname);
8813 old_name = &old_fname.disk_name;
8814 new_name = &new_fname.disk_name;
8816 /* close the race window with snapshot create/destroy ioctl */
8817 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8818 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8819 down_read(&fs_info->subvol_sem);
8823 * 1 to remove old dir item
8824 * 1 to remove old dir index
8825 * 1 to add new dir item
8826 * 1 to add new dir index
8827 * 1 to update parent inode
8829 * If the parents are the same, we only need to account for one
8831 trans_num_items = (old_dir == new_dir ? 9 : 10);
8832 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8834 * 1 to remove old root ref
8835 * 1 to remove old root backref
8836 * 1 to add new root ref
8837 * 1 to add new root backref
8839 trans_num_items += 4;
8842 * 1 to update inode item
8843 * 1 to remove old inode ref
8844 * 1 to add new inode ref
8846 trans_num_items += 3;
8848 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8849 trans_num_items += 4;
8851 trans_num_items += 3;
8852 trans = btrfs_start_transaction(root, trans_num_items);
8853 if (IS_ERR(trans)) {
8854 ret = PTR_ERR(trans);
8859 ret = btrfs_record_root_in_trans(trans, dest);
8865 * We need to find a free sequence number both in the source and
8866 * in the destination directory for the exchange.
8868 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8871 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8875 BTRFS_I(old_inode)->dir_index = 0ULL;
8876 BTRFS_I(new_inode)->dir_index = 0ULL;
8878 /* Reference for the source. */
8879 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8880 /* force full log commit if subvolume involved. */
8881 btrfs_set_log_full_commit(trans);
8883 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8884 btrfs_ino(BTRFS_I(new_dir)),
8891 /* And now for the dest. */
8892 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8893 /* force full log commit if subvolume involved. */
8894 btrfs_set_log_full_commit(trans);
8896 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8897 btrfs_ino(BTRFS_I(old_dir)),
8901 btrfs_abort_transaction(trans, ret);
8906 /* Update inode version and ctime/mtime. */
8907 inode_inc_iversion(old_dir);
8908 inode_inc_iversion(new_dir);
8909 inode_inc_iversion(old_inode);
8910 inode_inc_iversion(new_inode);
8911 old_dir->i_mtime = ctime;
8912 old_dir->i_ctime = ctime;
8913 new_dir->i_mtime = ctime;
8914 new_dir->i_ctime = ctime;
8915 old_inode->i_ctime = ctime;
8916 new_inode->i_ctime = ctime;
8918 if (old_dentry->d_parent != new_dentry->d_parent) {
8919 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8920 BTRFS_I(old_inode), true);
8921 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8922 BTRFS_I(new_inode), true);
8925 /* src is a subvolume */
8926 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8927 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8928 } else { /* src is an inode */
8929 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8930 BTRFS_I(old_dentry->d_inode),
8931 old_name, &old_rename_ctx);
8933 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8936 btrfs_abort_transaction(trans, ret);
8940 /* dest is a subvolume */
8941 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8942 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8943 } else { /* dest is an inode */
8944 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8945 BTRFS_I(new_dentry->d_inode),
8946 new_name, &new_rename_ctx);
8948 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8951 btrfs_abort_transaction(trans, ret);
8955 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8956 new_name, 0, old_idx);
8958 btrfs_abort_transaction(trans, ret);
8962 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8963 old_name, 0, new_idx);
8965 btrfs_abort_transaction(trans, ret);
8969 if (old_inode->i_nlink == 1)
8970 BTRFS_I(old_inode)->dir_index = old_idx;
8971 if (new_inode->i_nlink == 1)
8972 BTRFS_I(new_inode)->dir_index = new_idx;
8975 * Now pin the logs of the roots. We do it to ensure that no other task
8976 * can sync the logs while we are in progress with the rename, because
8977 * that could result in an inconsistency in case any of the inodes that
8978 * are part of this rename operation were logged before.
8980 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8981 btrfs_pin_log_trans(root);
8982 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8983 btrfs_pin_log_trans(dest);
8985 /* Do the log updates for all inodes. */
8986 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8987 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8988 old_rename_ctx.index, new_dentry->d_parent);
8989 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8990 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8991 new_rename_ctx.index, old_dentry->d_parent);
8993 /* Now unpin the logs. */
8994 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8995 btrfs_end_log_trans(root);
8996 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8997 btrfs_end_log_trans(dest);
8999 ret2 = btrfs_end_transaction(trans);
9000 ret = ret ? ret : ret2;
9002 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9003 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9004 up_read(&fs_info->subvol_sem);
9006 fscrypt_free_filename(&new_fname);
9007 fscrypt_free_filename(&old_fname);
9011 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9014 struct inode *inode;
9016 inode = new_inode(dir->i_sb);
9018 inode_init_owner(idmap, inode, dir,
9019 S_IFCHR | WHITEOUT_MODE);
9020 inode->i_op = &btrfs_special_inode_operations;
9021 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9026 static int btrfs_rename(struct mnt_idmap *idmap,
9027 struct inode *old_dir, struct dentry *old_dentry,
9028 struct inode *new_dir, struct dentry *new_dentry,
9031 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9032 struct btrfs_new_inode_args whiteout_args = {
9034 .dentry = old_dentry,
9036 struct btrfs_trans_handle *trans;
9037 unsigned int trans_num_items;
9038 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9039 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9040 struct inode *new_inode = d_inode(new_dentry);
9041 struct inode *old_inode = d_inode(old_dentry);
9042 struct btrfs_rename_ctx rename_ctx;
9046 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9047 struct fscrypt_name old_fname, new_fname;
9049 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9052 /* we only allow rename subvolume link between subvolumes */
9053 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9056 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9057 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9060 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9061 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9064 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9068 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9070 fscrypt_free_filename(&old_fname);
9074 /* check for collisions, even if the name isn't there */
9075 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9077 if (ret == -EEXIST) {
9079 * eexist without a new_inode */
9080 if (WARN_ON(!new_inode)) {
9081 goto out_fscrypt_names;
9084 /* maybe -EOVERFLOW */
9085 goto out_fscrypt_names;
9091 * we're using rename to replace one file with another. Start IO on it
9092 * now so we don't add too much work to the end of the transaction
9094 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9095 filemap_flush(old_inode->i_mapping);
9097 if (flags & RENAME_WHITEOUT) {
9098 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9099 if (!whiteout_args.inode) {
9101 goto out_fscrypt_names;
9103 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9105 goto out_whiteout_inode;
9107 /* 1 to update the old parent inode. */
9108 trans_num_items = 1;
9111 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9112 /* Close the race window with snapshot create/destroy ioctl */
9113 down_read(&fs_info->subvol_sem);
9115 * 1 to remove old root ref
9116 * 1 to remove old root backref
9117 * 1 to add new root ref
9118 * 1 to add new root backref
9120 trans_num_items += 4;
9124 * 1 to remove old inode ref
9125 * 1 to add new inode ref
9127 trans_num_items += 3;
9130 * 1 to remove old dir item
9131 * 1 to remove old dir index
9132 * 1 to add new dir item
9133 * 1 to add new dir index
9135 trans_num_items += 4;
9136 /* 1 to update new parent inode if it's not the same as the old parent */
9137 if (new_dir != old_dir)
9142 * 1 to remove inode ref
9143 * 1 to remove dir item
9144 * 1 to remove dir index
9145 * 1 to possibly add orphan item
9147 trans_num_items += 5;
9149 trans = btrfs_start_transaction(root, trans_num_items);
9150 if (IS_ERR(trans)) {
9151 ret = PTR_ERR(trans);
9156 ret = btrfs_record_root_in_trans(trans, dest);
9161 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9165 BTRFS_I(old_inode)->dir_index = 0ULL;
9166 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9167 /* force full log commit if subvolume involved. */
9168 btrfs_set_log_full_commit(trans);
9170 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9171 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9177 inode_inc_iversion(old_dir);
9178 inode_inc_iversion(new_dir);
9179 inode_inc_iversion(old_inode);
9180 old_dir->i_mtime = current_time(old_dir);
9181 old_dir->i_ctime = old_dir->i_mtime;
9182 new_dir->i_mtime = old_dir->i_mtime;
9183 new_dir->i_ctime = old_dir->i_mtime;
9184 old_inode->i_ctime = old_dir->i_mtime;
9186 if (old_dentry->d_parent != new_dentry->d_parent)
9187 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9188 BTRFS_I(old_inode), true);
9190 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9191 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9193 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9194 BTRFS_I(d_inode(old_dentry)),
9195 &old_fname.disk_name, &rename_ctx);
9197 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9200 btrfs_abort_transaction(trans, ret);
9205 inode_inc_iversion(new_inode);
9206 new_inode->i_ctime = current_time(new_inode);
9207 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9208 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9209 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9210 BUG_ON(new_inode->i_nlink == 0);
9212 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9213 BTRFS_I(d_inode(new_dentry)),
9214 &new_fname.disk_name);
9216 if (!ret && new_inode->i_nlink == 0)
9217 ret = btrfs_orphan_add(trans,
9218 BTRFS_I(d_inode(new_dentry)));
9220 btrfs_abort_transaction(trans, ret);
9225 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9226 &new_fname.disk_name, 0, index);
9228 btrfs_abort_transaction(trans, ret);
9232 if (old_inode->i_nlink == 1)
9233 BTRFS_I(old_inode)->dir_index = index;
9235 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9236 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9237 rename_ctx.index, new_dentry->d_parent);
9239 if (flags & RENAME_WHITEOUT) {
9240 ret = btrfs_create_new_inode(trans, &whiteout_args);
9242 btrfs_abort_transaction(trans, ret);
9245 unlock_new_inode(whiteout_args.inode);
9246 iput(whiteout_args.inode);
9247 whiteout_args.inode = NULL;
9251 ret2 = btrfs_end_transaction(trans);
9252 ret = ret ? ret : ret2;
9254 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9255 up_read(&fs_info->subvol_sem);
9256 if (flags & RENAME_WHITEOUT)
9257 btrfs_new_inode_args_destroy(&whiteout_args);
9259 if (flags & RENAME_WHITEOUT)
9260 iput(whiteout_args.inode);
9262 fscrypt_free_filename(&old_fname);
9263 fscrypt_free_filename(&new_fname);
9267 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9268 struct dentry *old_dentry, struct inode *new_dir,
9269 struct dentry *new_dentry, unsigned int flags)
9273 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9276 if (flags & RENAME_EXCHANGE)
9277 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9280 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9283 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9288 struct btrfs_delalloc_work {
9289 struct inode *inode;
9290 struct completion completion;
9291 struct list_head list;
9292 struct btrfs_work work;
9295 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9297 struct btrfs_delalloc_work *delalloc_work;
9298 struct inode *inode;
9300 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9302 inode = delalloc_work->inode;
9303 filemap_flush(inode->i_mapping);
9304 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9305 &BTRFS_I(inode)->runtime_flags))
9306 filemap_flush(inode->i_mapping);
9309 complete(&delalloc_work->completion);
9312 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9314 struct btrfs_delalloc_work *work;
9316 work = kmalloc(sizeof(*work), GFP_NOFS);
9320 init_completion(&work->completion);
9321 INIT_LIST_HEAD(&work->list);
9322 work->inode = inode;
9323 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9329 * some fairly slow code that needs optimization. This walks the list
9330 * of all the inodes with pending delalloc and forces them to disk.
9332 static int start_delalloc_inodes(struct btrfs_root *root,
9333 struct writeback_control *wbc, bool snapshot,
9334 bool in_reclaim_context)
9336 struct btrfs_inode *binode;
9337 struct inode *inode;
9338 struct btrfs_delalloc_work *work, *next;
9339 struct list_head works;
9340 struct list_head splice;
9342 bool full_flush = wbc->nr_to_write == LONG_MAX;
9344 INIT_LIST_HEAD(&works);
9345 INIT_LIST_HEAD(&splice);
9347 mutex_lock(&root->delalloc_mutex);
9348 spin_lock(&root->delalloc_lock);
9349 list_splice_init(&root->delalloc_inodes, &splice);
9350 while (!list_empty(&splice)) {
9351 binode = list_entry(splice.next, struct btrfs_inode,
9354 list_move_tail(&binode->delalloc_inodes,
9355 &root->delalloc_inodes);
9357 if (in_reclaim_context &&
9358 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9361 inode = igrab(&binode->vfs_inode);
9363 cond_resched_lock(&root->delalloc_lock);
9366 spin_unlock(&root->delalloc_lock);
9369 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9370 &binode->runtime_flags);
9372 work = btrfs_alloc_delalloc_work(inode);
9378 list_add_tail(&work->list, &works);
9379 btrfs_queue_work(root->fs_info->flush_workers,
9382 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9383 btrfs_add_delayed_iput(BTRFS_I(inode));
9384 if (ret || wbc->nr_to_write <= 0)
9388 spin_lock(&root->delalloc_lock);
9390 spin_unlock(&root->delalloc_lock);
9393 list_for_each_entry_safe(work, next, &works, list) {
9394 list_del_init(&work->list);
9395 wait_for_completion(&work->completion);
9399 if (!list_empty(&splice)) {
9400 spin_lock(&root->delalloc_lock);
9401 list_splice_tail(&splice, &root->delalloc_inodes);
9402 spin_unlock(&root->delalloc_lock);
9404 mutex_unlock(&root->delalloc_mutex);
9408 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9410 struct writeback_control wbc = {
9411 .nr_to_write = LONG_MAX,
9412 .sync_mode = WB_SYNC_NONE,
9414 .range_end = LLONG_MAX,
9416 struct btrfs_fs_info *fs_info = root->fs_info;
9418 if (BTRFS_FS_ERROR(fs_info))
9421 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9424 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9425 bool in_reclaim_context)
9427 struct writeback_control wbc = {
9429 .sync_mode = WB_SYNC_NONE,
9431 .range_end = LLONG_MAX,
9433 struct btrfs_root *root;
9434 struct list_head splice;
9437 if (BTRFS_FS_ERROR(fs_info))
9440 INIT_LIST_HEAD(&splice);
9442 mutex_lock(&fs_info->delalloc_root_mutex);
9443 spin_lock(&fs_info->delalloc_root_lock);
9444 list_splice_init(&fs_info->delalloc_roots, &splice);
9445 while (!list_empty(&splice)) {
9447 * Reset nr_to_write here so we know that we're doing a full
9451 wbc.nr_to_write = LONG_MAX;
9453 root = list_first_entry(&splice, struct btrfs_root,
9455 root = btrfs_grab_root(root);
9457 list_move_tail(&root->delalloc_root,
9458 &fs_info->delalloc_roots);
9459 spin_unlock(&fs_info->delalloc_root_lock);
9461 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9462 btrfs_put_root(root);
9463 if (ret < 0 || wbc.nr_to_write <= 0)
9465 spin_lock(&fs_info->delalloc_root_lock);
9467 spin_unlock(&fs_info->delalloc_root_lock);
9471 if (!list_empty(&splice)) {
9472 spin_lock(&fs_info->delalloc_root_lock);
9473 list_splice_tail(&splice, &fs_info->delalloc_roots);
9474 spin_unlock(&fs_info->delalloc_root_lock);
9476 mutex_unlock(&fs_info->delalloc_root_mutex);
9480 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9481 struct dentry *dentry, const char *symname)
9483 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9484 struct btrfs_trans_handle *trans;
9485 struct btrfs_root *root = BTRFS_I(dir)->root;
9486 struct btrfs_path *path;
9487 struct btrfs_key key;
9488 struct inode *inode;
9489 struct btrfs_new_inode_args new_inode_args = {
9493 unsigned int trans_num_items;
9498 struct btrfs_file_extent_item *ei;
9499 struct extent_buffer *leaf;
9501 name_len = strlen(symname);
9502 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9503 return -ENAMETOOLONG;
9505 inode = new_inode(dir->i_sb);
9508 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9509 inode->i_op = &btrfs_symlink_inode_operations;
9510 inode_nohighmem(inode);
9511 inode->i_mapping->a_ops = &btrfs_aops;
9512 btrfs_i_size_write(BTRFS_I(inode), name_len);
9513 inode_set_bytes(inode, name_len);
9515 new_inode_args.inode = inode;
9516 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9519 /* 1 additional item for the inline extent */
9522 trans = btrfs_start_transaction(root, trans_num_items);
9523 if (IS_ERR(trans)) {
9524 err = PTR_ERR(trans);
9525 goto out_new_inode_args;
9528 err = btrfs_create_new_inode(trans, &new_inode_args);
9532 path = btrfs_alloc_path();
9535 btrfs_abort_transaction(trans, err);
9536 discard_new_inode(inode);
9540 key.objectid = btrfs_ino(BTRFS_I(inode));
9542 key.type = BTRFS_EXTENT_DATA_KEY;
9543 datasize = btrfs_file_extent_calc_inline_size(name_len);
9544 err = btrfs_insert_empty_item(trans, root, path, &key,
9547 btrfs_abort_transaction(trans, err);
9548 btrfs_free_path(path);
9549 discard_new_inode(inode);
9553 leaf = path->nodes[0];
9554 ei = btrfs_item_ptr(leaf, path->slots[0],
9555 struct btrfs_file_extent_item);
9556 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9557 btrfs_set_file_extent_type(leaf, ei,
9558 BTRFS_FILE_EXTENT_INLINE);
9559 btrfs_set_file_extent_encryption(leaf, ei, 0);
9560 btrfs_set_file_extent_compression(leaf, ei, 0);
9561 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9562 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9564 ptr = btrfs_file_extent_inline_start(ei);
9565 write_extent_buffer(leaf, symname, ptr, name_len);
9566 btrfs_mark_buffer_dirty(leaf);
9567 btrfs_free_path(path);
9569 d_instantiate_new(dentry, inode);
9572 btrfs_end_transaction(trans);
9573 btrfs_btree_balance_dirty(fs_info);
9575 btrfs_new_inode_args_destroy(&new_inode_args);
9582 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9583 struct btrfs_trans_handle *trans_in,
9584 struct btrfs_inode *inode,
9585 struct btrfs_key *ins,
9588 struct btrfs_file_extent_item stack_fi;
9589 struct btrfs_replace_extent_info extent_info;
9590 struct btrfs_trans_handle *trans = trans_in;
9591 struct btrfs_path *path;
9592 u64 start = ins->objectid;
9593 u64 len = ins->offset;
9594 int qgroup_released;
9597 memset(&stack_fi, 0, sizeof(stack_fi));
9599 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9600 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9601 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9602 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9603 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9604 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9605 /* Encryption and other encoding is reserved and all 0 */
9607 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9608 if (qgroup_released < 0)
9609 return ERR_PTR(qgroup_released);
9612 ret = insert_reserved_file_extent(trans, inode,
9613 file_offset, &stack_fi,
9614 true, qgroup_released);
9620 extent_info.disk_offset = start;
9621 extent_info.disk_len = len;
9622 extent_info.data_offset = 0;
9623 extent_info.data_len = len;
9624 extent_info.file_offset = file_offset;
9625 extent_info.extent_buf = (char *)&stack_fi;
9626 extent_info.is_new_extent = true;
9627 extent_info.update_times = true;
9628 extent_info.qgroup_reserved = qgroup_released;
9629 extent_info.insertions = 0;
9631 path = btrfs_alloc_path();
9637 ret = btrfs_replace_file_extents(inode, path, file_offset,
9638 file_offset + len - 1, &extent_info,
9640 btrfs_free_path(path);
9647 * We have released qgroup data range at the beginning of the function,
9648 * and normally qgroup_released bytes will be freed when committing
9650 * But if we error out early, we have to free what we have released
9651 * or we leak qgroup data reservation.
9653 btrfs_qgroup_free_refroot(inode->root->fs_info,
9654 inode->root->root_key.objectid, qgroup_released,
9655 BTRFS_QGROUP_RSV_DATA);
9656 return ERR_PTR(ret);
9659 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9660 u64 start, u64 num_bytes, u64 min_size,
9661 loff_t actual_len, u64 *alloc_hint,
9662 struct btrfs_trans_handle *trans)
9664 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9665 struct extent_map *em;
9666 struct btrfs_root *root = BTRFS_I(inode)->root;
9667 struct btrfs_key ins;
9668 u64 cur_offset = start;
9669 u64 clear_offset = start;
9672 u64 last_alloc = (u64)-1;
9674 bool own_trans = true;
9675 u64 end = start + num_bytes - 1;
9679 while (num_bytes > 0) {
9680 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9681 cur_bytes = max(cur_bytes, min_size);
9683 * If we are severely fragmented we could end up with really
9684 * small allocations, so if the allocator is returning small
9685 * chunks lets make its job easier by only searching for those
9688 cur_bytes = min(cur_bytes, last_alloc);
9689 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9690 min_size, 0, *alloc_hint, &ins, 1, 0);
9695 * We've reserved this space, and thus converted it from
9696 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9697 * from here on out we will only need to clear our reservation
9698 * for the remaining unreserved area, so advance our
9699 * clear_offset by our extent size.
9701 clear_offset += ins.offset;
9703 last_alloc = ins.offset;
9704 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9707 * Now that we inserted the prealloc extent we can finally
9708 * decrement the number of reservations in the block group.
9709 * If we did it before, we could race with relocation and have
9710 * relocation miss the reserved extent, making it fail later.
9712 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9713 if (IS_ERR(trans)) {
9714 ret = PTR_ERR(trans);
9715 btrfs_free_reserved_extent(fs_info, ins.objectid,
9720 em = alloc_extent_map();
9722 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9723 cur_offset + ins.offset - 1, false);
9724 btrfs_set_inode_full_sync(BTRFS_I(inode));
9728 em->start = cur_offset;
9729 em->orig_start = cur_offset;
9730 em->len = ins.offset;
9731 em->block_start = ins.objectid;
9732 em->block_len = ins.offset;
9733 em->orig_block_len = ins.offset;
9734 em->ram_bytes = ins.offset;
9735 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9736 em->generation = trans->transid;
9738 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9739 free_extent_map(em);
9741 num_bytes -= ins.offset;
9742 cur_offset += ins.offset;
9743 *alloc_hint = ins.objectid + ins.offset;
9745 inode_inc_iversion(inode);
9746 inode->i_ctime = current_time(inode);
9747 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9748 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9749 (actual_len > inode->i_size) &&
9750 (cur_offset > inode->i_size)) {
9751 if (cur_offset > actual_len)
9752 i_size = actual_len;
9754 i_size = cur_offset;
9755 i_size_write(inode, i_size);
9756 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9759 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9762 btrfs_abort_transaction(trans, ret);
9764 btrfs_end_transaction(trans);
9769 btrfs_end_transaction(trans);
9773 if (clear_offset < end)
9774 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9775 end - clear_offset + 1);
9779 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9780 u64 start, u64 num_bytes, u64 min_size,
9781 loff_t actual_len, u64 *alloc_hint)
9783 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9784 min_size, actual_len, alloc_hint,
9788 int btrfs_prealloc_file_range_trans(struct inode *inode,
9789 struct btrfs_trans_handle *trans, int mode,
9790 u64 start, u64 num_bytes, u64 min_size,
9791 loff_t actual_len, u64 *alloc_hint)
9793 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9794 min_size, actual_len, alloc_hint, trans);
9797 static int btrfs_permission(struct mnt_idmap *idmap,
9798 struct inode *inode, int mask)
9800 struct btrfs_root *root = BTRFS_I(inode)->root;
9801 umode_t mode = inode->i_mode;
9803 if (mask & MAY_WRITE &&
9804 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9805 if (btrfs_root_readonly(root))
9807 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9810 return generic_permission(idmap, inode, mask);
9813 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9814 struct file *file, umode_t mode)
9816 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9817 struct btrfs_trans_handle *trans;
9818 struct btrfs_root *root = BTRFS_I(dir)->root;
9819 struct inode *inode;
9820 struct btrfs_new_inode_args new_inode_args = {
9822 .dentry = file->f_path.dentry,
9825 unsigned int trans_num_items;
9828 inode = new_inode(dir->i_sb);
9831 inode_init_owner(idmap, inode, dir, mode);
9832 inode->i_fop = &btrfs_file_operations;
9833 inode->i_op = &btrfs_file_inode_operations;
9834 inode->i_mapping->a_ops = &btrfs_aops;
9836 new_inode_args.inode = inode;
9837 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9841 trans = btrfs_start_transaction(root, trans_num_items);
9842 if (IS_ERR(trans)) {
9843 ret = PTR_ERR(trans);
9844 goto out_new_inode_args;
9847 ret = btrfs_create_new_inode(trans, &new_inode_args);
9850 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9851 * set it to 1 because d_tmpfile() will issue a warning if the count is
9854 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9856 set_nlink(inode, 1);
9859 d_tmpfile(file, inode);
9860 unlock_new_inode(inode);
9861 mark_inode_dirty(inode);
9864 btrfs_end_transaction(trans);
9865 btrfs_btree_balance_dirty(fs_info);
9867 btrfs_new_inode_args_destroy(&new_inode_args);
9871 return finish_open_simple(file, ret);
9874 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9876 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9877 unsigned long index = start >> PAGE_SHIFT;
9878 unsigned long end_index = end >> PAGE_SHIFT;
9882 ASSERT(end + 1 - start <= U32_MAX);
9883 len = end + 1 - start;
9884 while (index <= end_index) {
9885 page = find_get_page(inode->vfs_inode.i_mapping, index);
9886 ASSERT(page); /* Pages should be in the extent_io_tree */
9888 btrfs_page_set_writeback(fs_info, page, start, len);
9894 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9897 switch (compress_type) {
9898 case BTRFS_COMPRESS_NONE:
9899 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9900 case BTRFS_COMPRESS_ZLIB:
9901 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9902 case BTRFS_COMPRESS_LZO:
9904 * The LZO format depends on the sector size. 64K is the maximum
9905 * sector size that we support.
9907 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9909 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9910 (fs_info->sectorsize_bits - 12);
9911 case BTRFS_COMPRESS_ZSTD:
9912 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9918 static ssize_t btrfs_encoded_read_inline(
9920 struct iov_iter *iter, u64 start,
9922 struct extent_state **cached_state,
9923 u64 extent_start, size_t count,
9924 struct btrfs_ioctl_encoded_io_args *encoded,
9927 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9928 struct btrfs_root *root = inode->root;
9929 struct btrfs_fs_info *fs_info = root->fs_info;
9930 struct extent_io_tree *io_tree = &inode->io_tree;
9931 struct btrfs_path *path;
9932 struct extent_buffer *leaf;
9933 struct btrfs_file_extent_item *item;
9939 path = btrfs_alloc_path();
9944 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9948 /* The extent item disappeared? */
9953 leaf = path->nodes[0];
9954 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9956 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9957 ptr = btrfs_file_extent_inline_start(item);
9959 encoded->len = min_t(u64, extent_start + ram_bytes,
9960 inode->vfs_inode.i_size) - iocb->ki_pos;
9961 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9962 btrfs_file_extent_compression(leaf, item));
9965 encoded->compression = ret;
9966 if (encoded->compression) {
9969 inline_size = btrfs_file_extent_inline_item_len(leaf,
9971 if (inline_size > count) {
9975 count = inline_size;
9976 encoded->unencoded_len = ram_bytes;
9977 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9979 count = min_t(u64, count, encoded->len);
9980 encoded->len = count;
9981 encoded->unencoded_len = count;
9982 ptr += iocb->ki_pos - extent_start;
9985 tmp = kmalloc(count, GFP_NOFS);
9990 read_extent_buffer(leaf, tmp, ptr, count);
9991 btrfs_release_path(path);
9992 unlock_extent(io_tree, start, lockend, cached_state);
9993 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9996 ret = copy_to_iter(tmp, count, iter);
10001 btrfs_free_path(path);
10005 struct btrfs_encoded_read_private {
10006 wait_queue_head_t wait;
10008 blk_status_t status;
10011 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10013 struct btrfs_encoded_read_private *priv = bbio->private;
10015 if (bbio->bio.bi_status) {
10017 * The memory barrier implied by the atomic_dec_return() here
10018 * pairs with the memory barrier implied by the
10019 * atomic_dec_return() or io_wait_event() in
10020 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10021 * write is observed before the load of status in
10022 * btrfs_encoded_read_regular_fill_pages().
10024 WRITE_ONCE(priv->status, bbio->bio.bi_status);
10026 if (!atomic_dec_return(&priv->pending))
10027 wake_up(&priv->wait);
10028 bio_put(&bbio->bio);
10031 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10032 u64 file_offset, u64 disk_bytenr,
10033 u64 disk_io_size, struct page **pages)
10035 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10036 struct btrfs_encoded_read_private priv = {
10037 .pending = ATOMIC_INIT(1),
10039 unsigned long i = 0;
10040 struct btrfs_bio *bbio;
10042 init_waitqueue_head(&priv.wait);
10044 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10045 btrfs_encoded_read_endio, &priv);
10046 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10047 bbio->inode = inode;
10050 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10052 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10053 atomic_inc(&priv.pending);
10054 btrfs_submit_bio(bbio, 0);
10056 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10057 btrfs_encoded_read_endio, &priv);
10058 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10059 bbio->inode = inode;
10064 disk_bytenr += bytes;
10065 disk_io_size -= bytes;
10066 } while (disk_io_size);
10068 atomic_inc(&priv.pending);
10069 btrfs_submit_bio(bbio, 0);
10071 if (atomic_dec_return(&priv.pending))
10072 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10073 /* See btrfs_encoded_read_endio() for ordering. */
10074 return blk_status_to_errno(READ_ONCE(priv.status));
10077 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10078 struct iov_iter *iter,
10079 u64 start, u64 lockend,
10080 struct extent_state **cached_state,
10081 u64 disk_bytenr, u64 disk_io_size,
10082 size_t count, bool compressed,
10085 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10086 struct extent_io_tree *io_tree = &inode->io_tree;
10087 struct page **pages;
10088 unsigned long nr_pages, i;
10090 size_t page_offset;
10093 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10094 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10097 ret = btrfs_alloc_page_array(nr_pages, pages);
10103 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10104 disk_io_size, pages);
10108 unlock_extent(io_tree, start, lockend, cached_state);
10109 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10116 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10117 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10120 while (cur < count) {
10121 size_t bytes = min_t(size_t, count - cur,
10122 PAGE_SIZE - page_offset);
10124 if (copy_page_to_iter(pages[i], page_offset, bytes,
10135 for (i = 0; i < nr_pages; i++) {
10137 __free_page(pages[i]);
10143 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10144 struct btrfs_ioctl_encoded_io_args *encoded)
10146 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10147 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10148 struct extent_io_tree *io_tree = &inode->io_tree;
10150 size_t count = iov_iter_count(iter);
10151 u64 start, lockend, disk_bytenr, disk_io_size;
10152 struct extent_state *cached_state = NULL;
10153 struct extent_map *em;
10154 bool unlocked = false;
10156 file_accessed(iocb->ki_filp);
10158 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10160 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10161 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10164 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10166 * We don't know how long the extent containing iocb->ki_pos is, but if
10167 * it's compressed we know that it won't be longer than this.
10169 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10172 struct btrfs_ordered_extent *ordered;
10174 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10175 lockend - start + 1);
10177 goto out_unlock_inode;
10178 lock_extent(io_tree, start, lockend, &cached_state);
10179 ordered = btrfs_lookup_ordered_range(inode, start,
10180 lockend - start + 1);
10183 btrfs_put_ordered_extent(ordered);
10184 unlock_extent(io_tree, start, lockend, &cached_state);
10188 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10191 goto out_unlock_extent;
10194 if (em->block_start == EXTENT_MAP_INLINE) {
10195 u64 extent_start = em->start;
10198 * For inline extents we get everything we need out of the
10201 free_extent_map(em);
10203 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10204 &cached_state, extent_start,
10205 count, encoded, &unlocked);
10210 * We only want to return up to EOF even if the extent extends beyond
10213 encoded->len = min_t(u64, extent_map_end(em),
10214 inode->vfs_inode.i_size) - iocb->ki_pos;
10215 if (em->block_start == EXTENT_MAP_HOLE ||
10216 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10217 disk_bytenr = EXTENT_MAP_HOLE;
10218 count = min_t(u64, count, encoded->len);
10219 encoded->len = count;
10220 encoded->unencoded_len = count;
10221 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10222 disk_bytenr = em->block_start;
10224 * Bail if the buffer isn't large enough to return the whole
10225 * compressed extent.
10227 if (em->block_len > count) {
10231 disk_io_size = em->block_len;
10232 count = em->block_len;
10233 encoded->unencoded_len = em->ram_bytes;
10234 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10235 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10236 em->compress_type);
10239 encoded->compression = ret;
10241 disk_bytenr = em->block_start + (start - em->start);
10242 if (encoded->len > count)
10243 encoded->len = count;
10245 * Don't read beyond what we locked. This also limits the page
10246 * allocations that we'll do.
10248 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10249 count = start + disk_io_size - iocb->ki_pos;
10250 encoded->len = count;
10251 encoded->unencoded_len = count;
10252 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10254 free_extent_map(em);
10257 if (disk_bytenr == EXTENT_MAP_HOLE) {
10258 unlock_extent(io_tree, start, lockend, &cached_state);
10259 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10261 ret = iov_iter_zero(count, iter);
10265 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10266 &cached_state, disk_bytenr,
10267 disk_io_size, count,
10268 encoded->compression,
10274 iocb->ki_pos += encoded->len;
10276 free_extent_map(em);
10279 unlock_extent(io_tree, start, lockend, &cached_state);
10282 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10286 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10287 const struct btrfs_ioctl_encoded_io_args *encoded)
10289 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10290 struct btrfs_root *root = inode->root;
10291 struct btrfs_fs_info *fs_info = root->fs_info;
10292 struct extent_io_tree *io_tree = &inode->io_tree;
10293 struct extent_changeset *data_reserved = NULL;
10294 struct extent_state *cached_state = NULL;
10295 struct btrfs_ordered_extent *ordered;
10299 u64 num_bytes, ram_bytes, disk_num_bytes;
10300 unsigned long nr_pages, i;
10301 struct page **pages;
10302 struct btrfs_key ins;
10303 bool extent_reserved = false;
10304 struct extent_map *em;
10307 switch (encoded->compression) {
10308 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10309 compression = BTRFS_COMPRESS_ZLIB;
10311 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10312 compression = BTRFS_COMPRESS_ZSTD;
10314 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10315 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10316 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10317 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10318 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10319 /* The sector size must match for LZO. */
10320 if (encoded->compression -
10321 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10322 fs_info->sectorsize_bits)
10324 compression = BTRFS_COMPRESS_LZO;
10329 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10332 orig_count = iov_iter_count(from);
10334 /* The extent size must be sane. */
10335 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10336 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10340 * The compressed data must be smaller than the decompressed data.
10342 * It's of course possible for data to compress to larger or the same
10343 * size, but the buffered I/O path falls back to no compression for such
10344 * data, and we don't want to break any assumptions by creating these
10347 * Note that this is less strict than the current check we have that the
10348 * compressed data must be at least one sector smaller than the
10349 * decompressed data. We only want to enforce the weaker requirement
10350 * from old kernels that it is at least one byte smaller.
10352 if (orig_count >= encoded->unencoded_len)
10355 /* The extent must start on a sector boundary. */
10356 start = iocb->ki_pos;
10357 if (!IS_ALIGNED(start, fs_info->sectorsize))
10361 * The extent must end on a sector boundary. However, we allow a write
10362 * which ends at or extends i_size to have an unaligned length; we round
10363 * up the extent size and set i_size to the unaligned end.
10365 if (start + encoded->len < inode->vfs_inode.i_size &&
10366 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10369 /* Finally, the offset in the unencoded data must be sector-aligned. */
10370 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10373 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10374 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10375 end = start + num_bytes - 1;
10378 * If the extent cannot be inline, the compressed data on disk must be
10379 * sector-aligned. For convenience, we extend it with zeroes if it
10382 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10383 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10384 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10387 for (i = 0; i < nr_pages; i++) {
10388 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10391 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10396 kaddr = kmap_local_page(pages[i]);
10397 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10398 kunmap_local(kaddr);
10402 if (bytes < PAGE_SIZE)
10403 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10404 kunmap_local(kaddr);
10408 struct btrfs_ordered_extent *ordered;
10410 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10413 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10414 start >> PAGE_SHIFT,
10415 end >> PAGE_SHIFT);
10418 lock_extent(io_tree, start, end, &cached_state);
10419 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10421 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10424 btrfs_put_ordered_extent(ordered);
10425 unlock_extent(io_tree, start, end, &cached_state);
10430 * We don't use the higher-level delalloc space functions because our
10431 * num_bytes and disk_num_bytes are different.
10433 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10436 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10438 goto out_free_data_space;
10439 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10442 goto out_qgroup_free_data;
10444 /* Try an inline extent first. */
10445 if (start == 0 && encoded->unencoded_len == encoded->len &&
10446 encoded->unencoded_offset == 0) {
10447 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10448 compression, pages, true);
10452 goto out_delalloc_release;
10456 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10457 disk_num_bytes, 0, 0, &ins, 1, 1);
10459 goto out_delalloc_release;
10460 extent_reserved = true;
10462 em = create_io_em(inode, start, num_bytes,
10463 start - encoded->unencoded_offset, ins.objectid,
10464 ins.offset, ins.offset, ram_bytes, compression,
10465 BTRFS_ORDERED_COMPRESSED);
10468 goto out_free_reserved;
10470 free_extent_map(em);
10472 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10473 ins.objectid, ins.offset,
10474 encoded->unencoded_offset,
10475 (1 << BTRFS_ORDERED_ENCODED) |
10476 (1 << BTRFS_ORDERED_COMPRESSED),
10478 if (IS_ERR(ordered)) {
10479 btrfs_drop_extent_map_range(inode, start, end, false);
10480 ret = PTR_ERR(ordered);
10481 goto out_free_reserved;
10483 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10485 if (start + encoded->len > inode->vfs_inode.i_size)
10486 i_size_write(&inode->vfs_inode, start + encoded->len);
10488 unlock_extent(io_tree, start, end, &cached_state);
10490 btrfs_delalloc_release_extents(inode, num_bytes);
10492 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10497 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10498 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10499 out_delalloc_release:
10500 btrfs_delalloc_release_extents(inode, num_bytes);
10501 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10502 out_qgroup_free_data:
10504 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10505 out_free_data_space:
10507 * If btrfs_reserve_extent() succeeded, then we already decremented
10510 if (!extent_reserved)
10511 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10513 unlock_extent(io_tree, start, end, &cached_state);
10515 for (i = 0; i < nr_pages; i++) {
10517 __free_page(pages[i]);
10522 iocb->ki_pos += encoded->len;
10528 * Add an entry indicating a block group or device which is pinned by a
10529 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10530 * negative errno on failure.
10532 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10533 bool is_block_group)
10535 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10536 struct btrfs_swapfile_pin *sp, *entry;
10537 struct rb_node **p;
10538 struct rb_node *parent = NULL;
10540 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10545 sp->is_block_group = is_block_group;
10546 sp->bg_extent_count = 1;
10548 spin_lock(&fs_info->swapfile_pins_lock);
10549 p = &fs_info->swapfile_pins.rb_node;
10552 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10553 if (sp->ptr < entry->ptr ||
10554 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10555 p = &(*p)->rb_left;
10556 } else if (sp->ptr > entry->ptr ||
10557 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10558 p = &(*p)->rb_right;
10560 if (is_block_group)
10561 entry->bg_extent_count++;
10562 spin_unlock(&fs_info->swapfile_pins_lock);
10567 rb_link_node(&sp->node, parent, p);
10568 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10569 spin_unlock(&fs_info->swapfile_pins_lock);
10573 /* Free all of the entries pinned by this swapfile. */
10574 static void btrfs_free_swapfile_pins(struct inode *inode)
10576 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10577 struct btrfs_swapfile_pin *sp;
10578 struct rb_node *node, *next;
10580 spin_lock(&fs_info->swapfile_pins_lock);
10581 node = rb_first(&fs_info->swapfile_pins);
10583 next = rb_next(node);
10584 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10585 if (sp->inode == inode) {
10586 rb_erase(&sp->node, &fs_info->swapfile_pins);
10587 if (sp->is_block_group) {
10588 btrfs_dec_block_group_swap_extents(sp->ptr,
10589 sp->bg_extent_count);
10590 btrfs_put_block_group(sp->ptr);
10596 spin_unlock(&fs_info->swapfile_pins_lock);
10599 struct btrfs_swap_info {
10605 unsigned long nr_pages;
10609 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10610 struct btrfs_swap_info *bsi)
10612 unsigned long nr_pages;
10613 unsigned long max_pages;
10614 u64 first_ppage, first_ppage_reported, next_ppage;
10618 * Our swapfile may have had its size extended after the swap header was
10619 * written. In that case activating the swapfile should not go beyond
10620 * the max size set in the swap header.
10622 if (bsi->nr_pages >= sis->max)
10625 max_pages = sis->max - bsi->nr_pages;
10626 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10627 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10629 if (first_ppage >= next_ppage)
10631 nr_pages = next_ppage - first_ppage;
10632 nr_pages = min(nr_pages, max_pages);
10634 first_ppage_reported = first_ppage;
10635 if (bsi->start == 0)
10636 first_ppage_reported++;
10637 if (bsi->lowest_ppage > first_ppage_reported)
10638 bsi->lowest_ppage = first_ppage_reported;
10639 if (bsi->highest_ppage < (next_ppage - 1))
10640 bsi->highest_ppage = next_ppage - 1;
10642 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10645 bsi->nr_extents += ret;
10646 bsi->nr_pages += nr_pages;
10650 static void btrfs_swap_deactivate(struct file *file)
10652 struct inode *inode = file_inode(file);
10654 btrfs_free_swapfile_pins(inode);
10655 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10658 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10661 struct inode *inode = file_inode(file);
10662 struct btrfs_root *root = BTRFS_I(inode)->root;
10663 struct btrfs_fs_info *fs_info = root->fs_info;
10664 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10665 struct extent_state *cached_state = NULL;
10666 struct extent_map *em = NULL;
10667 struct btrfs_device *device = NULL;
10668 struct btrfs_swap_info bsi = {
10669 .lowest_ppage = (sector_t)-1ULL,
10676 * If the swap file was just created, make sure delalloc is done. If the
10677 * file changes again after this, the user is doing something stupid and
10678 * we don't really care.
10680 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10685 * The inode is locked, so these flags won't change after we check them.
10687 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10688 btrfs_warn(fs_info, "swapfile must not be compressed");
10691 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10692 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10695 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10696 btrfs_warn(fs_info, "swapfile must not be checksummed");
10701 * Balance or device remove/replace/resize can move stuff around from
10702 * under us. The exclop protection makes sure they aren't running/won't
10703 * run concurrently while we are mapping the swap extents, and
10704 * fs_info->swapfile_pins prevents them from running while the swap
10705 * file is active and moving the extents. Note that this also prevents
10706 * a concurrent device add which isn't actually necessary, but it's not
10707 * really worth the trouble to allow it.
10709 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10710 btrfs_warn(fs_info,
10711 "cannot activate swapfile while exclusive operation is running");
10716 * Prevent snapshot creation while we are activating the swap file.
10717 * We do not want to race with snapshot creation. If snapshot creation
10718 * already started before we bumped nr_swapfiles from 0 to 1 and
10719 * completes before the first write into the swap file after it is
10720 * activated, than that write would fallback to COW.
10722 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10723 btrfs_exclop_finish(fs_info);
10724 btrfs_warn(fs_info,
10725 "cannot activate swapfile because snapshot creation is in progress");
10729 * Snapshots can create extents which require COW even if NODATACOW is
10730 * set. We use this counter to prevent snapshots. We must increment it
10731 * before walking the extents because we don't want a concurrent
10732 * snapshot to run after we've already checked the extents.
10734 * It is possible that subvolume is marked for deletion but still not
10735 * removed yet. To prevent this race, we check the root status before
10736 * activating the swapfile.
10738 spin_lock(&root->root_item_lock);
10739 if (btrfs_root_dead(root)) {
10740 spin_unlock(&root->root_item_lock);
10742 btrfs_exclop_finish(fs_info);
10743 btrfs_warn(fs_info,
10744 "cannot activate swapfile because subvolume %llu is being deleted",
10745 root->root_key.objectid);
10748 atomic_inc(&root->nr_swapfiles);
10749 spin_unlock(&root->root_item_lock);
10751 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10753 lock_extent(io_tree, 0, isize - 1, &cached_state);
10755 while (start < isize) {
10756 u64 logical_block_start, physical_block_start;
10757 struct btrfs_block_group *bg;
10758 u64 len = isize - start;
10760 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10766 if (em->block_start == EXTENT_MAP_HOLE) {
10767 btrfs_warn(fs_info, "swapfile must not have holes");
10771 if (em->block_start == EXTENT_MAP_INLINE) {
10773 * It's unlikely we'll ever actually find ourselves
10774 * here, as a file small enough to fit inline won't be
10775 * big enough to store more than the swap header, but in
10776 * case something changes in the future, let's catch it
10777 * here rather than later.
10779 btrfs_warn(fs_info, "swapfile must not be inline");
10783 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10784 btrfs_warn(fs_info, "swapfile must not be compressed");
10789 logical_block_start = em->block_start + (start - em->start);
10790 len = min(len, em->len - (start - em->start));
10791 free_extent_map(em);
10794 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10800 btrfs_warn(fs_info,
10801 "swapfile must not be copy-on-write");
10806 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10812 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10813 btrfs_warn(fs_info,
10814 "swapfile must have single data profile");
10819 if (device == NULL) {
10820 device = em->map_lookup->stripes[0].dev;
10821 ret = btrfs_add_swapfile_pin(inode, device, false);
10826 } else if (device != em->map_lookup->stripes[0].dev) {
10827 btrfs_warn(fs_info, "swapfile must be on one device");
10832 physical_block_start = (em->map_lookup->stripes[0].physical +
10833 (logical_block_start - em->start));
10834 len = min(len, em->len - (logical_block_start - em->start));
10835 free_extent_map(em);
10838 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10840 btrfs_warn(fs_info,
10841 "could not find block group containing swapfile");
10846 if (!btrfs_inc_block_group_swap_extents(bg)) {
10847 btrfs_warn(fs_info,
10848 "block group for swapfile at %llu is read-only%s",
10850 atomic_read(&fs_info->scrubs_running) ?
10851 " (scrub running)" : "");
10852 btrfs_put_block_group(bg);
10857 ret = btrfs_add_swapfile_pin(inode, bg, true);
10859 btrfs_put_block_group(bg);
10866 if (bsi.block_len &&
10867 bsi.block_start + bsi.block_len == physical_block_start) {
10868 bsi.block_len += len;
10870 if (bsi.block_len) {
10871 ret = btrfs_add_swap_extent(sis, &bsi);
10876 bsi.block_start = physical_block_start;
10877 bsi.block_len = len;
10884 ret = btrfs_add_swap_extent(sis, &bsi);
10887 if (!IS_ERR_OR_NULL(em))
10888 free_extent_map(em);
10890 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10893 btrfs_swap_deactivate(file);
10895 btrfs_drew_write_unlock(&root->snapshot_lock);
10897 btrfs_exclop_finish(fs_info);
10903 sis->bdev = device->bdev;
10904 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10905 sis->max = bsi.nr_pages;
10906 sis->pages = bsi.nr_pages - 1;
10907 sis->highest_bit = bsi.nr_pages - 1;
10908 return bsi.nr_extents;
10911 static void btrfs_swap_deactivate(struct file *file)
10915 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10918 return -EOPNOTSUPP;
10923 * Update the number of bytes used in the VFS' inode. When we replace extents in
10924 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10925 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10926 * always get a correct value.
10928 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10929 const u64 add_bytes,
10930 const u64 del_bytes)
10932 if (add_bytes == del_bytes)
10935 spin_lock(&inode->lock);
10937 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10939 inode_add_bytes(&inode->vfs_inode, add_bytes);
10940 spin_unlock(&inode->lock);
10944 * Verify that there are no ordered extents for a given file range.
10946 * @inode: The target inode.
10947 * @start: Start offset of the file range, should be sector size aligned.
10948 * @end: End offset (inclusive) of the file range, its value +1 should be
10949 * sector size aligned.
10951 * This should typically be used for cases where we locked an inode's VFS lock in
10952 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10953 * we have flushed all delalloc in the range, we have waited for all ordered
10954 * extents in the range to complete and finally we have locked the file range in
10955 * the inode's io_tree.
10957 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10959 struct btrfs_root *root = inode->root;
10960 struct btrfs_ordered_extent *ordered;
10962 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10965 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10967 btrfs_err(root->fs_info,
10968 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10969 start, end, btrfs_ino(inode), root->root_key.objectid,
10970 ordered->file_offset,
10971 ordered->file_offset + ordered->num_bytes - 1);
10972 btrfs_put_ordered_extent(ordered);
10975 ASSERT(ordered == NULL);
10978 static const struct inode_operations btrfs_dir_inode_operations = {
10979 .getattr = btrfs_getattr,
10980 .lookup = btrfs_lookup,
10981 .create = btrfs_create,
10982 .unlink = btrfs_unlink,
10983 .link = btrfs_link,
10984 .mkdir = btrfs_mkdir,
10985 .rmdir = btrfs_rmdir,
10986 .rename = btrfs_rename2,
10987 .symlink = btrfs_symlink,
10988 .setattr = btrfs_setattr,
10989 .mknod = btrfs_mknod,
10990 .listxattr = btrfs_listxattr,
10991 .permission = btrfs_permission,
10992 .get_inode_acl = btrfs_get_acl,
10993 .set_acl = btrfs_set_acl,
10994 .update_time = btrfs_update_time,
10995 .tmpfile = btrfs_tmpfile,
10996 .fileattr_get = btrfs_fileattr_get,
10997 .fileattr_set = btrfs_fileattr_set,
11000 static const struct file_operations btrfs_dir_file_operations = {
11001 .llseek = generic_file_llseek,
11002 .read = generic_read_dir,
11003 .iterate_shared = btrfs_real_readdir,
11004 .open = btrfs_opendir,
11005 .unlocked_ioctl = btrfs_ioctl,
11006 #ifdef CONFIG_COMPAT
11007 .compat_ioctl = btrfs_compat_ioctl,
11009 .release = btrfs_release_file,
11010 .fsync = btrfs_sync_file,
11014 * btrfs doesn't support the bmap operation because swapfiles
11015 * use bmap to make a mapping of extents in the file. They assume
11016 * these extents won't change over the life of the file and they
11017 * use the bmap result to do IO directly to the drive.
11019 * the btrfs bmap call would return logical addresses that aren't
11020 * suitable for IO and they also will change frequently as COW
11021 * operations happen. So, swapfile + btrfs == corruption.
11023 * For now we're avoiding this by dropping bmap.
11025 static const struct address_space_operations btrfs_aops = {
11026 .read_folio = btrfs_read_folio,
11027 .writepages = btrfs_writepages,
11028 .readahead = btrfs_readahead,
11029 .invalidate_folio = btrfs_invalidate_folio,
11030 .release_folio = btrfs_release_folio,
11031 .migrate_folio = btrfs_migrate_folio,
11032 .dirty_folio = filemap_dirty_folio,
11033 .error_remove_page = generic_error_remove_page,
11034 .swap_activate = btrfs_swap_activate,
11035 .swap_deactivate = btrfs_swap_deactivate,
11038 static const struct inode_operations btrfs_file_inode_operations = {
11039 .getattr = btrfs_getattr,
11040 .setattr = btrfs_setattr,
11041 .listxattr = btrfs_listxattr,
11042 .permission = btrfs_permission,
11043 .fiemap = btrfs_fiemap,
11044 .get_inode_acl = btrfs_get_acl,
11045 .set_acl = btrfs_set_acl,
11046 .update_time = btrfs_update_time,
11047 .fileattr_get = btrfs_fileattr_get,
11048 .fileattr_set = btrfs_fileattr_set,
11050 static const struct inode_operations btrfs_special_inode_operations = {
11051 .getattr = btrfs_getattr,
11052 .setattr = btrfs_setattr,
11053 .permission = btrfs_permission,
11054 .listxattr = btrfs_listxattr,
11055 .get_inode_acl = btrfs_get_acl,
11056 .set_acl = btrfs_set_acl,
11057 .update_time = btrfs_update_time,
11059 static const struct inode_operations btrfs_symlink_inode_operations = {
11060 .get_link = page_get_link,
11061 .getattr = btrfs_getattr,
11062 .setattr = btrfs_setattr,
11063 .permission = btrfs_permission,
11064 .listxattr = btrfs_listxattr,
11065 .update_time = btrfs_update_time,
11068 const struct dentry_operations btrfs_dentry_operations = {
11069 .d_delete = btrfs_dentry_delete,